From d2206ce18efe9e16f05b329e54ddc2fb331824c5 Mon Sep 17 00:00:00 2001 From: Jimmy Song Date: Wed, 20 Dec 2017 18:52:00 +0800 Subject: [PATCH] add kubernetes SIGs picture --- appendix/debug-kubernetes-services.html | 2 +- appendix/docker-best-practice.html | 2 +- appendix/index.html | 2 +- appendix/issues.html | 2 +- appendix/kubernetes-1.7-changelog.html | 2 +- appendix/kubernetes-1.8-changelog.html | 2 +- appendix/kubernetes-1.9-changelog.html | 2 +- appendix/kubernetes-changelog.html | 2 +- appendix/material-share.html | 2 +- appendix/tricks.html | 2 +- ...bernetes-and-cloud-native-app-overview.html | 2 +- concepts/aggregated-api-server.html | 2 +- concepts/annotation.html | 2 +- concepts/apiservice.html | 2 +- concepts/concepts.html | 2 +- concepts/configmap-hot-update.html | 2 +- concepts/configmap.html | 2 +- concepts/cronjob.html | 2 +- concepts/custom-metrics-hpa.html | 2 +- concepts/custom-resource.html | 2 +- concepts/daemonset.html | 2 +- concepts/deployment.html | 2 +- concepts/garbage-collection.html | 2 +- concepts/horizontal-pod-autoscaling.html | 2 +- concepts/index.html | 2 +- concepts/ingress.html | 2 +- concepts/init-containers.html | 2 +- concepts/job.html | 2 +- concepts/label.html | 2 +- concepts/namespace.html | 2 +- concepts/network-policy.html | 2 +- concepts/node.html | 2 +- concepts/objects.html | 2 +- concepts/pod-hook.html | 2 +- concepts/pod-lifecycle.html | 2 +- concepts/pod-overview.html | 2 +- concepts/pod-preset.html | 2 +- concepts/pod-security-policy.html | 2 +- concepts/pod.html | 2 +- concepts/replicaset.html | 2 +- concepts/secret.html | 2 +- concepts/service.html | 2 +- concepts/serviceaccount.html | 2 +- concepts/statefulset.html | 2 +- concepts/traefik-ingress-controller.html | 2 +- concepts/volume.html | 2 +- develop/client-go-sample.html | 2 +- develop/contribute.html | 2 +- develop/developing-environment.html | 2 +- develop/index.html | 2 +- develop/minikube.html | 2 +- develop/sigs-and-working-group.html | 5 +++-- develop/testing.html | 2 +- guide/access-cluster.html | 2 +- guide/access-kubernetes-cluster.html | 2 +- ...netes-pods-from-outside-of-the-cluster.html | 2 +- ...pplication-development-deployment-flow.html | 2 +- guide/auth-with-kubeconfig-or-token.html | 2 +- ...uthenticate-across-clusters-kubeconfig.html | 2 +- guide/authentication.html | 2 +- ...carbin-mobile-dashboard-for-kubernetes.html | 2 +- guide/cluster-security-management.html | 2 +- guide/command-usage.html | 2 +- guide/configure-liveness-readiness-probes.html | 2 +- guide/configure-pod-service-account.html | 2 +- ...onnecting-to-applications-port-forward.html | 2 +- guide/deploy-applications-in-kubernetes.html | 2 +- guide/docker-cli-to-kubectl.html | 2 +- guide/index.html | 2 +- guide/ip-masq-agent.html | 2 +- guide/kubectl-cheatsheet.html | 2 +- ...ectl-user-authentication-authorization.html | 2 +- .../kubelet-authentication-authorization.html | 2 +- guide/kubernetes-desktop-client.html | 2 +- guide/managing-tls-in-a-cluster.html | 2 +- guide/migrating-hadoop-yarn-to-kubernetes.html | 2 +- guide/rbac.html | 2 +- guide/resource-configuration.html | 2 +- guide/resource-quota-management.html | 2 +- guide/secret-configuration.html | 2 +- guide/service-access-application-cluster.html | 2 +- guide/tls-bootstrapping.html | 2 +- ...sing-etcdctl-to-access-kubernetes-data.html | 2 +- guide/using-kubectl.html | 2 +- guide/using-statefulset.html | 2 +- ...oud-native-architecture-with-kubernetes.png | Bin 0 -> 682401 bytes images/creating-kubernetes-native-app.jpg | Bin 0 -> 324356 bytes images/kubernetes-sigs.jpg | Bin 0 -> 422071 bytes index.html | 2 +- practice/app-log-collection.html | 2 +- practice/cephfs.html | 2 +- practice/ci-cd.html | 2 +- practice/configuration-best-practice.html | 2 +- practice/create-kubeconfig.html | 2 +- practice/create-private-charts-repo.html | 2 +- practice/create-tls-and-secret-key.html | 2 +- practice/dashboard-addon-installation.html | 2 +- practice/dashboard-upgrade.html | 2 +- practice/data-persistence-problem.html | 2 +- practice/distributed-load-test.html | 2 +- practice/drone-ci-cd.html | 2 +- practice/edge-node-configuration.html | 2 +- practice/efk-addon-installation.html | 2 +- practice/etcd-cluster-installation.html | 2 +- practice/flannel-installation.html | 2 +- practice/glusterfs.html | 2 +- practice/heapster-addon-installation.html | 2 +- practice/heapster.html | 2 +- practice/helm.html | 2 +- practice/index.html | 2 +- practice/install-kubernetes-on-centos.html | 2 +- practice/jenkins-ci-cd.html | 2 +- practice/kubectl-installation.html | 2 +- practice/kubedns-addon-installation.html | 2 +- .../manage-compute-resources-container.html | 2 +- practice/manually-upgrade.html | 2 +- practice/master-installation.html | 2 +- practice/monitor.html | 2 +- practice/monitoring.html | 2 +- .../network-and-cluster-perfermance-test.html | 2 +- practice/nginx-ingress-installation.html | 2 +- practice/node-installation.html | 2 +- practice/operation.html | 2 +- practice/prometheus.html | 2 +- .../service-discovery-and-loadbalancing.html | 2 +- practice/service-rolling-update.html | 2 +- practice/services-management-tool.html | 2 +- ...tainers-using-glusterfs-with-openshift.html | 2 +- practice/storage.html | 2 +- practice/traefik-ingress-installation.html | 2 +- practice/update-and-upgrade.html | 2 +- .../using-ceph-for-persistent-storage.html | 2 +- ...using-glusterfs-for-persistent-storage.html | 2 +- .../using-heapster-to-get-object-metrics.html | 2 +- ...ometheus-to-monitor-kuberentes-cluster.html | 2 +- search_plus_index.json | 2 +- usecases/big-data.html | 2 +- usecases/conduit-installation.html | 2 +- usecases/conduit-overview.html | 2 +- usecases/conduit.html | 2 +- usecases/configuring-request-routing.html | 2 +- usecases/edge-computing.html | 2 +- usecases/faas.html | 2 +- usecases/index.html | 2 +- usecases/install-and-expand-istio-mesh.html | 2 +- usecases/integrating-vms.html | 2 +- usecases/istio-installation.html | 2 +- usecases/istio.html | 2 +- usecases/linkerd-user-guide.html | 2 +- usecases/linkerd.html | 2 +- usecases/microservices.html | 2 +- usecases/openfaas.html | 2 +- ...spark-with-kubernetes-native-scheduler.html | 2 +- usecases/serverless.html | 2 +- .../service-discovery-in-microservices.html | 2 +- usecases/service-mesh.html | 2 +- usecases/spark-standalone-on-kubernetes.html | 2 +- usecases/understanding-serverless.html | 2 +- 158 files changed, 157 insertions(+), 156 deletions(-) create mode 100644 images/building-cloud-native-architecture-with-kubernetes.png create mode 100644 images/creating-kubernetes-native-app.jpg create mode 100644 images/kubernetes-sigs.jpg diff --git a/appendix/debug-kubernetes-services.html b/appendix/debug-kubernetes-services.html index 84cac9d78..caf688bbc 100644 --- a/appendix/debug-kubernetes-services.html +++ b/appendix/debug-kubernetes-services.html @@ -2810,7 +2810,7 @@ kubectl top pod POD_NAME --containers diff --git a/appendix/docker-best-practice.html b/appendix/docker-best-practice.html index 145250ca1..0af120333 100644 --- a/appendix/docker-best-practice.html +++ b/appendix/docker-best-practice.html @@ -2858,7 +2858,7 @@ diff --git a/appendix/index.html b/appendix/index.html index 0b49ec17f..3bb2cd407 100644 --- a/appendix/index.html +++ b/appendix/index.html @@ -2787,7 +2787,7 @@ diff --git a/appendix/issues.html b/appendix/issues.html index 9f6c3aac3..37e9d539b 100644 --- a/appendix/issues.html +++ b/appendix/issues.html @@ -2849,7 +2849,7 @@ kubectl patch deploy --namespace kube-system tiller-deploy -p diff --git a/appendix/kubernetes-1.7-changelog.html b/appendix/kubernetes-1.7-changelog.html index dd049f102..5329e06d0 100644 --- a/appendix/kubernetes-1.7-changelog.html +++ b/appendix/kubernetes-1.7-changelog.html @@ -2816,7 +2816,7 @@ diff --git a/appendix/kubernetes-1.8-changelog.html b/appendix/kubernetes-1.8-changelog.html index d48c06824..79449c01b 100644 --- a/appendix/kubernetes-1.8-changelog.html +++ b/appendix/kubernetes-1.8-changelog.html @@ -2793,7 +2793,7 @@ diff --git a/appendix/kubernetes-1.9-changelog.html b/appendix/kubernetes-1.9-changelog.html index b0dce3628..546312a59 100644 --- a/appendix/kubernetes-1.9-changelog.html +++ b/appendix/kubernetes-1.9-changelog.html @@ -2793,7 +2793,7 @@ diff --git a/appendix/kubernetes-changelog.html b/appendix/kubernetes-changelog.html index 60e5e1ab6..32c8e557f 100644 --- a/appendix/kubernetes-changelog.html +++ b/appendix/kubernetes-changelog.html @@ -2786,7 +2786,7 @@ diff --git a/appendix/material-share.html b/appendix/material-share.html index f29c046c3..d589d6eb6 100644 --- a/appendix/material-share.html +++ b/appendix/material-share.html @@ -2849,7 +2849,7 @@ diff --git a/appendix/tricks.html b/appendix/tricks.html index f7c5412be..3710991b1 100644 --- a/appendix/tricks.html +++ b/appendix/tricks.html @@ -2948,7 +2948,7 @@ diff --git a/cloud-native/kubernetes-and-cloud-native-app-overview.html b/cloud-native/kubernetes-and-cloud-native-app-overview.html index c4a40f826..49efe9533 100644 --- a/cloud-native/kubernetes-and-cloud-native-app-overview.html +++ b/cloud-native/kubernetes-and-cloud-native-app-overview.html @@ -3131,7 +3131,7 @@ diff --git a/concepts/aggregated-api-server.html b/concepts/aggregated-api-server.html index cd17d9a36..6cfbf43ba 100644 --- a/concepts/aggregated-api-server.html +++ b/concepts/aggregated-api-server.html @@ -2797,7 +2797,7 @@ diff --git a/concepts/annotation.html b/concepts/annotation.html index 2e2da9759..da6a01f64 100644 --- a/concepts/annotation.html +++ b/concepts/annotation.html @@ -2843,7 +2843,7 @@ diff --git a/concepts/apiservice.html b/concepts/apiservice.html index 0082bf9b5..0671cbd69 100644 --- a/concepts/apiservice.html +++ b/concepts/apiservice.html @@ -2889,7 +2889,7 @@ v1 diff --git a/concepts/concepts.html b/concepts/concepts.html index 4ede714a3..86cb7df6b 100644 --- a/concepts/concepts.html +++ b/concepts/concepts.html @@ -2860,7 +2860,7 @@ diff --git a/concepts/configmap-hot-update.html b/concepts/configmap-hot-update.html index 35359417a..4687dd677 100644 --- a/concepts/configmap-hot-update.html +++ b/concepts/configmap-hot-update.html @@ -3084,7 +3084,7 @@ DEBUG diff --git a/concepts/configmap.html b/concepts/configmap.html index 58a21517d..846bbc0a7 100644 --- a/concepts/configmap.html +++ b/concepts/configmap.html @@ -3065,7 +3065,7 @@ log_level=INFO diff --git a/concepts/cronjob.html b/concepts/cronjob.html index 33f3ca581..c780c7add 100644 --- a/concepts/cronjob.html +++ b/concepts/cronjob.html @@ -2872,7 +2872,7 @@ job "hello-1202039034" deleted diff --git a/concepts/custom-metrics-hpa.html b/concepts/custom-metrics-hpa.html index 28e92c4d2..50d37b9ea 100644 --- a/concepts/custom-metrics-hpa.html +++ b/concepts/custom-metrics-hpa.html @@ -2829,7 +2829,7 @@ diff --git a/concepts/custom-resource.html b/concepts/custom-resource.html index 7ea55c095..f9b5a49f9 100644 --- a/concepts/custom-resource.html +++ b/concepts/custom-resource.html @@ -2889,7 +2889,7 @@ diff --git a/concepts/daemonset.html b/concepts/daemonset.html index 0c21cd8ee..d4b75a4aa 100644 --- a/concepts/daemonset.html +++ b/concepts/daemonset.html @@ -2851,7 +2851,7 @@ diff --git a/concepts/deployment.html b/concepts/deployment.html index d0b4fee8e..fb6c394ac 100644 --- a/concepts/deployment.html +++ b/concepts/deployment.html @@ -3323,7 +3323,7 @@ $ echo $? diff --git a/concepts/garbage-collection.html b/concepts/garbage-collection.html index 45490ea04..179c2b29f 100644 --- a/concepts/garbage-collection.html +++ b/concepts/garbage-collection.html @@ -2868,7 +2868,7 @@ curl -X DELETE localhost:8080/apis/extensions/v1beta1/namespaces/default/replica diff --git a/concepts/horizontal-pod-autoscaling.html b/concepts/horizontal-pod-autoscaling.html index 4aaa95c0b..65f9aaccf 100644 --- a/concepts/horizontal-pod-autoscaling.html +++ b/concepts/horizontal-pod-autoscaling.html @@ -2890,7 +2890,7 @@ kubectl delete hpa diff --git a/concepts/index.html b/concepts/index.html index 75e4bdbad..a54a03796 100644 --- a/concepts/index.html +++ b/concepts/index.html @@ -2847,7 +2847,7 @@ Kubernetes 还提供完善的管理] diff --git a/concepts/ingress.html b/concepts/ingress.html index 4899960ce..2737a2fa6 100644 --- a/concepts/ingress.html +++ b/concepts/ingress.html @@ -3000,7 +3000,7 @@ NAME RULE BACKEND ADDRESS diff --git a/concepts/init-containers.html b/concepts/init-containers.html index e6a76e993..f4552abd8 100644 --- a/concepts/init-containers.html +++ b/concepts/init-containers.html @@ -2974,7 +2974,7 @@ myapp-pod 1/1 Running 0 9m diff --git a/concepts/job.html b/concepts/job.html index 2ae8781f4..4e79ae557 100644 --- a/concepts/job.html +++ b/concepts/job.html @@ -2812,7 +2812,7 @@ $ kubectl logs $pods diff --git a/concepts/label.html b/concepts/label.html index 9e3d14144..aef7460b2 100644 --- a/concepts/label.html +++ b/concepts/label.html @@ -2857,7 +2857,7 @@ $ kubectl get pods -l &apo diff --git a/concepts/namespace.html b/concepts/namespace.html index 0e60afc14..18d46d2ba 100644 --- a/concepts/namespace.html +++ b/concepts/namespace.html @@ -2788,7 +2788,7 @@ diff --git a/concepts/network-policy.html b/concepts/network-policy.html index 50feaff02..761195df7 100644 --- a/concepts/network-policy.html +++ b/concepts/network-policy.html @@ -2841,7 +2841,7 @@ diff --git a/concepts/node.html b/concepts/node.html index 05dc4b968..7ee93007f 100644 --- a/concepts/node.html +++ b/concepts/node.html @@ -2811,7 +2811,7 @@ diff --git a/concepts/objects.html b/concepts/objects.html index 2667f4abd..6f66a215b 100644 --- a/concepts/objects.html +++ b/concepts/objects.html @@ -2878,7 +2878,7 @@ diff --git a/concepts/pod-hook.html b/concepts/pod-hook.html index a319acf5c..b3598ff61 100644 --- a/concepts/pod-hook.html +++ b/concepts/pod-hook.html @@ -2810,7 +2810,7 @@ diff --git a/concepts/pod-lifecycle.html b/concepts/pod-lifecycle.html index bd10cfb4b..381801518 100644 --- a/concepts/pod-lifecycle.html +++ b/concepts/pod-lifecycle.html @@ -2933,7 +2933,7 @@ diff --git a/concepts/pod-overview.html b/concepts/pod-overview.html index cc9b0ef16..cace902ec 100644 --- a/concepts/pod-overview.html +++ b/concepts/pod-overview.html @@ -2823,7 +2823,7 @@ diff --git a/concepts/pod-preset.html b/concepts/pod-preset.html index 7c73dbbb6..a5290c813 100644 --- a/concepts/pod-preset.html +++ b/concepts/pod-preset.html @@ -2816,7 +2816,7 @@ diff --git a/concepts/pod-security-policy.html b/concepts/pod-security-policy.html index 9aa77f88e..040f0f39d 100644 --- a/concepts/pod-security-policy.html +++ b/concepts/pod-security-policy.html @@ -2984,7 +2984,7 @@ podsecuritypolicy "permissive" delete diff --git a/concepts/pod.html b/concepts/pod.html index 9564a3163..9fe375170 100644 --- a/concepts/pod.html +++ b/concepts/pod.html @@ -2864,7 +2864,7 @@ diff --git a/concepts/replicaset.html b/concepts/replicaset.html index 08f9b9120..7a667bd85 100644 --- a/concepts/replicaset.html +++ b/concepts/replicaset.html @@ -2828,7 +2828,7 @@ diff --git a/concepts/secret.html b/concepts/secret.html index 3c74a60d7..b5d7f7f97 100644 --- a/concepts/secret.html +++ b/concepts/secret.html @@ -2906,7 +2906,7 @@ token diff --git a/concepts/service.html b/concepts/service.html index b132b8763..6ea12f804 100644 --- a/concepts/service.html +++ b/concepts/service.html @@ -3111,7 +3111,7 @@ iptables 代理不会隐藏 Kubernetes 集&#x diff --git a/concepts/serviceaccount.html b/concepts/serviceaccount.html index 25cb128ce..dcfeea181 100644 --- a/concepts/serviceaccount.html +++ b/concepts/serviceaccount.html @@ -2927,7 +2927,7 @@ serviceaccounts/default diff --git a/concepts/statefulset.html b/concepts/statefulset.html index 3495b4c11..a3d8b50bc 100644 --- a/concepts/statefulset.html +++ b/concepts/statefulset.html @@ -3246,7 +3246,7 @@ zk-0 10.254.98.14 <nodes> 2181:31693/TCP 5m diff --git a/concepts/traefik-ingress-controller.html b/concepts/traefik-ingress-controller.html index 13e80d2bc..8432fc411 100644 --- a/concepts/traefik-ingress-controller.html +++ b/concepts/traefik-ingress-controller.html @@ -2832,7 +2832,7 @@ diff --git a/concepts/volume.html b/concepts/volume.html index f76b59440..ce3d99720 100644 --- a/concepts/volume.html +++ b/concepts/volume.html @@ -2933,7 +2933,7 @@ diff --git a/develop/client-go-sample.html b/develop/client-go-sample.html index 3a58bd76c..fff0b796f 100644 --- a/develop/client-go-sample.html +++ b/develop/client-go-sample.html @@ -3018,7 +3018,7 @@ Events: diff --git a/develop/contribute.html b/develop/contribute.html index 614bec71f..935547486 100644 --- a/develop/contribute.html +++ b/develop/contribute.html @@ -2786,7 +2786,7 @@ diff --git a/develop/developing-environment.html b/develop/developing-environment.html index ebfc1b94f..3ec62a2fd 100644 --- a/develop/developing-environment.html +++ b/develop/developing-environment.html @@ -2799,7 +2799,7 @@ diff --git a/develop/index.html b/develop/index.html index a37cc0cd4..716feb336 100644 --- a/develop/index.html +++ b/develop/index.html @@ -2779,7 +2779,7 @@ diff --git a/develop/minikube.html b/develop/minikube.html index 4476e90e3..a5c5045b8 100644 --- a/develop/minikube.html +++ b/develop/minikube.html @@ -2816,7 +2816,7 @@ Kubectl is now configured to use the cluster. diff --git a/develop/sigs-and-working-group.html b/develop/sigs-and-working-group.html index e23da0ea7..53d5b18a9 100644 --- a/develop/sigs-and-working-group.html +++ b/develop/sigs-and-working-group.html @@ -2735,6 +2735,7 @@

SIG和工作组

Kubernetes的社区是以SIG（Special Interest Group特别兴趣小组）和工作组的形式组织起来的，每个工作组都会定期召开视频会议。

所有的SIG和工作组都使用slack和邮件列表沟通。

+
Kubernetes SIG
图片 - Kubernetes SIG

主要SIG列表

  • api-machinery：所有API级别的功能，包括了API server、API注册和发现、通用的API CRUD语义，准入控制，编码/解码，转换，默认值，持久化层（etcd），OpenAPI，第三方资源，垃圾手机和客户端库的方方面面。
    • @@ -2778,7 +2779,7 @@

详细信息请参考 https://github.com/kubernetes/community/blob/master/sig-list.md

Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: -2017-12-03 13:11:06 +2017-12-20 18:50:16
@@ -2822,7 +2823,7 @@ diff --git a/develop/testing.html b/develop/testing.html index 31b7d347d..21eadad61 100644 --- a/develop/testing.html +++ b/develop/testing.html @@ -2893,7 +2893,7 @@ make test_e2e_node TEST_ARGS= diff --git a/guide/access-cluster.html b/guide/access-cluster.html index 302aae617..4491bde45 100644 --- a/guide/access-cluster.html +++ b/guide/access-cluster.html @@ -2989,7 +2989,7 @@ $ curl $APISERVER/api --header diff --git a/guide/access-kubernetes-cluster.html b/guide/access-kubernetes-cluster.html index 48727f09c..32c58783c 100644 --- a/guide/access-kubernetes-cluster.html +++ b/guide/access-kubernetes-cluster.html @@ -2785,7 +2785,7 @@ diff --git a/guide/accessing-kubernetes-pods-from-outside-of-the-cluster.html b/guide/accessing-kubernetes-pods-from-outside-of-the-cluster.html index 2717c6f38..2d1fec06a 100644 --- a/guide/accessing-kubernetes-pods-from-outside-of-the-cluster.html +++ b/guide/accessing-kubernetes-pods-from-outside-of-the-cluster.html @@ -2901,7 +2901,7 @@ influxdb 10.97.121.42 10.13.242.236 8086:30051/TCP 39s diff --git a/guide/application-development-deployment-flow.html b/guide/application-development-deployment-flow.html index 7ef9ad915..a315b12c0 100644 --- a/guide/application-development-deployment-flow.html +++ b/guide/application-development-deployment-flow.html @@ -2780,7 +2780,7 @@ diff --git a/guide/auth-with-kubeconfig-or-token.html b/guide/auth-with-kubeconfig-or-token.html index beafd69d1..8a660764c 100644 --- a/guide/auth-with-kubeconfig-or-token.html +++ b/guide/auth-with-kubeconfig-or-token.html @@ -2851,7 +2851,7 @@ ca.crt: 1310 bytes diff --git a/guide/authenticate-across-clusters-kubeconfig.html b/guide/authenticate-across-clusters-kubeconfig.html index 7db0696d4..1b09082d8 100644 --- a/guide/authenticate-across-clusters-kubeconfig.html +++ b/guide/authenticate-across-clusters-kubeconfig.html @@ -3004,7 +3004,7 @@ $ kubectl config use-context federal-context diff --git a/guide/authentication.html b/guide/authentication.html index 349ee59fe..d55d0984d 100644 --- a/guide/authentication.html +++ b/guide/authentication.html @@ -3280,7 +3280,7 @@ tar xzf easy-rsa.tar.gz diff --git a/guide/carbin-mobile-dashboard-for-kubernetes.html b/guide/carbin-mobile-dashboard-for-kubernetes.html index 205be4540..3ef3290bd 100644 --- a/guide/carbin-mobile-dashboard-for-kubernetes.html +++ b/guide/carbin-mobile-dashboard-for-kubernetes.html @@ -2791,7 +2791,7 @@ diff --git a/guide/cluster-security-management.html b/guide/cluster-security-management.html index d29895fc1..34367cdf3 100644 --- a/guide/cluster-security-management.html +++ b/guide/cluster-security-management.html @@ -2779,7 +2779,7 @@ diff --git a/guide/command-usage.html b/guide/command-usage.html index 2f68b9fcb..9901190f8 100644 --- a/guide/command-usage.html +++ b/guide/command-usage.html @@ -2779,7 +2779,7 @@ diff --git a/guide/configure-liveness-readiness-probes.html b/guide/configure-liveness-readiness-probes.html index cb3a944f6..b6e4db934 100644 --- a/guide/configure-liveness-readiness-probes.html +++ b/guide/configure-liveness-readiness-probes.html @@ -2968,7 +2968,7 @@ the Container has been restarted:

diff --git a/guide/configure-pod-service-account.html b/guide/configure-pod-service-account.html index d037c4139..ebb23013d 100644 --- a/guide/configure-pod-service-account.html +++ b/guide/configure-pod-service-account.html @@ -2926,7 +2926,7 @@ serviceaccounts/default diff --git a/guide/connecting-to-applications-port-forward.html b/guide/connecting-to-applications-port-forward.html index 1dfdf3ba0..eb6471dc8 100644 --- a/guide/connecting-to-applications-port-forward.html +++ b/guide/connecting-to-applications-port-forward.html @@ -2818,7 +2818,7 @@ kubectl get pods redis-master --template='{{(index (index .spec.containers diff --git a/guide/deploy-applications-in-kubernetes.html b/guide/deploy-applications-in-kubernetes.html index efda72f85..7cc892de0 100644 --- a/guide/deploy-applications-in-kubernetes.html +++ b/guide/deploy-applications-in-kubernetes.html @@ -2818,7 +2818,7 @@ diff --git a/guide/docker-cli-to-kubectl.html b/guide/docker-cli-to-kubectl.html index 944b5d071..ce5d160a4 100644 --- a/guide/docker-cli-to-kubectl.html +++ b/guide/docker-cli-to-kubectl.html @@ -2955,7 +2955,7 @@ InfluxDB is running at https://108.59.85.141/api/v1/namespaces/kube-system/servi diff --git a/guide/index.html b/guide/index.html index 50b20e166..baa5eb0bd 100644 --- a/guide/index.html +++ b/guide/index.html @@ -2783,7 +2783,7 @@ diff --git a/guide/ip-masq-agent.html b/guide/ip-masq-agent.html index f1eabf943..2854ce9c6 100644 --- a/guide/ip-masq-agent.html +++ b/guide/ip-masq-agent.html @@ -2843,7 +2843,7 @@ MASQUERADE all -- anywhere anywhere /* ip-masq-agent: diff --git a/guide/kubectl-cheatsheet.html b/guide/kubectl-cheatsheet.html index 80a5c15a1..a1ca20f0a 100644 --- a/guide/kubectl-cheatsheet.html +++ b/guide/kubectl-cheatsheet.html @@ -3185,7 +3185,7 @@ $ kubectl taint nodes foo dedicated=special-user:NoSchedule diff --git a/guide/kubectl-user-authentication-authorization.html b/guide/kubectl-user-authentication-authorization.html index aa7c1a928..b7fd0dd96 100644 --- a/guide/kubectl-user-authentication-authorization.html +++ b/guide/kubectl-user-authentication-authorization.html @@ -2877,7 +2877,7 @@ No resources found. diff --git a/guide/kubelet-authentication-authorization.html b/guide/kubelet-authentication-authorization.html index 2c8d88638..782fd5e30 100644 --- a/guide/kubelet-authentication-authorization.html +++ b/guide/kubelet-authentication-authorization.html @@ -2891,7 +2891,7 @@ diff --git a/guide/kubernetes-desktop-client.html b/guide/kubernetes-desktop-client.html index 9eb3f4733..51b0c73c7 100644 --- a/guide/kubernetes-desktop-client.html +++ b/guide/kubernetes-desktop-client.html @@ -2788,7 +2788,7 @@ diff --git a/guide/managing-tls-in-a-cluster.html b/guide/managing-tls-in-a-cluster.html index a1cc50106..48e08a419 100644 --- a/guide/managing-tls-in-a-cluster.html +++ b/guide/managing-tls-in-a-cluster.html @@ -2869,7 +2869,7 @@ my-svc.my-namespace 10m yourname@example.com Approved,Issued diff --git a/guide/migrating-hadoop-yarn-to-kubernetes.html b/guide/migrating-hadoop-yarn-to-kubernetes.html index 7318efa71..40d890dc8 100644 --- a/guide/migrating-hadoop-yarn-to-kubernetes.html +++ b/guide/migrating-hadoop-yarn-to-kubernetes.html @@ -2949,7 +2949,7 @@ kubectl create configmap spark-config \ diff --git a/guide/rbac.html b/guide/rbac.html index 077156c86..e2cc8e7e8 100644 --- a/guide/rbac.html +++ b/guide/rbac.html @@ -3306,7 +3306,7 @@ diff --git a/guide/resource-configuration.html b/guide/resource-configuration.html index c96c2dabb..1918b9248 100644 --- a/guide/resource-configuration.html +++ b/guide/resource-configuration.html @@ -2779,7 +2779,7 @@ diff --git a/guide/resource-quota-management.html b/guide/resource-quota-management.html index b42204eaa..8abdc6673 100644 --- a/guide/resource-quota-management.html +++ b/guide/resource-quota-management.html @@ -2851,7 +2851,7 @@ diff --git a/guide/secret-configuration.html b/guide/secret-configuration.html index ec1c2d351..e6f99c7b5 100644 --- a/guide/secret-configuration.html +++ b/guide/secret-configuration.html @@ -3200,7 +3200,7 @@ secret "test-db-secret" created diff --git a/guide/service-access-application-cluster.html b/guide/service-access-application-cluster.html index 2efd4bbbe..48ca3f4ed 100644 --- a/guide/service-access-application-cluster.html +++ b/guide/service-access-application-cluster.html @@ -2849,7 +2849,7 @@ kubectl describe replicasets diff --git a/guide/tls-bootstrapping.html b/guide/tls-bootstrapping.html index 82903cdd7..8a46c093b 100644 --- a/guide/tls-bootstrapping.html +++ b/guide/tls-bootstrapping.html @@ -2896,7 +2896,7 @@ diff --git a/guide/using-etcdctl-to-access-kubernetes-data.html b/guide/using-etcdctl-to-access-kubernetes-data.html index 9e36b9932..cd850e01a 100644 --- a/guide/using-etcdctl-to-access-kubernetes-data.html +++ b/guide/using-etcdctl-to-access-kubernetes-data.html @@ -2885,7 +2885,7 @@ diff --git a/guide/using-kubectl.html b/guide/using-kubectl.html index b6a62cc00..4c34959e9 100644 --- a/guide/using-kubectl.html +++ b/guide/using-kubectl.html @@ -2806,7 +2806,7 @@ diff --git a/guide/using-statefulset.html b/guide/using-statefulset.html index ccedc13bf..be155152f 100644 --- a/guide/using-statefulset.html +++ b/guide/using-statefulset.html @@ -3073,7 +3073,7 @@ sed -i s'/zookeeper.connect=localhost:2181/zookee diff --git a/images/building-cloud-native-architecture-with-kubernetes.png b/images/building-cloud-native-architecture-with-kubernetes.png new file mode 100644 index 0000000000000000000000000000000000000000..c2dd825aa5fbd6c190c8796f687d9a23a5576343 GIT binary patch literal 682401 zcmeFZWmH^Sur7)Pf(IvfLU4DtU?DicT>}JnmxkaLoZu4NT^e_HcW8pUYjZnW&Xzas z-GAQC!&nSfFKSxNS@u<}4pESkL`5P%f`WoVmHHs21O@dn6AJ2uQmevlut^#j=qu_aq%CqBiy>WE_m_ zjLdHZk;ur%`0b59^D2o+{LvipN#L!SqoXY^6O)UJ3!@7gqm8{O6AKRy4-+#h6DunN zgo44r&DznxmBHG9{P#}&>_^PR!N}g+*3sO?n(SA<28K3HjskDr{u=0ietw^)qxt8* zMzVJJV_A>|GX44w6AL3V)Bp4hY0CfWDX)@=gN>!rukKZ>%^d|<`F|t+_v61t`@NN@ zjg_sviGu^AnIOj>P5%Du-}LhK<|dH!{TfD))iAgv%W* zH@c&OdQ;wGW==r5w84-07gG7A-c|fSzAK@e8S`|nHGqkR!YgQu|GfIKV6a`nn~av6 z)7jNWu>Z3K;60VFSAN0x>w@REWHA46eY1gkce=F{T-fp-gk&%;7IlvwQB!CD3-a7_+g#qmWB-KKN)o=0bx{YMA9PNe{o;o6vBP%N9|&VL;MYF7yNk-U*!{uf7sgX7cDkFg!?aqUcmhq zN5#n2UVc~~sK@x^A^u;qNa+yn`(?nu{O>XS-(&iJm+JpLrvIx=|9{t}3NbdQs{d9| zNa5|*_Vme-0aF-7J*T4ME*}6_dMp3dxBza5*6s#mfBKgtXne2#Wbb)~+3Jn#(DuFn zo-L|F#B_-~vf|^AeQv;)pnsom_7~7$@r!jU*ebKI-2QH9u1edSV_VW;IQ| zW1#68&9U3+;a;b&IY?r!X8gCx7Otky*GQmuO+MN!GWy5NzOBJ=mUbkTuXj;Zd;!cw zC({nUjw(WOJc_~y%Bl%)kz)Dx=+y6Es@7?KOsMMpLx+>;y|i2kS9=EFopdH#ou%;W zbGN#11(Lptpb{ia`xl$cX5q{_K24YK6^w@b)4FVtLIdk?vGn@FTyi``J7<}a1M{sr zHf`RPEn((Oz4-TJ%+ew3W*@UtDgTF(B-2Cu`hGUX=CqBk!b#u;S#~z;dj{xdn_&X7#7n6!Z%;LH#RS^z0FGWYQ=00O1*Dndg@$f?t!1gmcvrK(Wt;BZl*=n}~2SA^##yMuiyCkVRs_aYbVpwRkIb6(#6aII0ntE0&Z^}mH2|yNQF{2o~#DA(nY`t z(m%KXvBckm;4TR%eR@AK{9{J}pQU^k!`)^Khj?XNab|GLZ|on7Z?7F3cSD_ZCVcSW z4($<$#ap|>a7PcMKt~7+^80Or zgkQld(N)?LtCsW>Ihjo%`9Ea&$YWEvHj#Y*d?f$Ff!f1l7IL!WBSv_+H=B9+?+L$3 zgc46%{8FM5C_csP0rNSc@3CB=^^Nr+yPVR*-XyvJyy30E{4={FseXw=!f1C5I?M43 z`eZ<8L}jjyJ=HZIPK$G-K)DgdE5P;#{Xt9q>o5C}XSzz-U#paAY{DXsDjf61C#yHMFH#}a>TWXZE)cGw7>Q-Sg zdHUk5cA0)^_<=h!8-sZ`P3ZpsYBsW)ouEcKofppPkl_Jl`%((x3lWbsAL$4hv&->z z^P%8B96ilf@Wfm%gP7gT+C3xEOU6IH0LZ2JOX{7qU+?-6GM%G2@Hm4qS{~YUgua+b zBirs0hL)^fXrX-ldg7I;nfdOOpM z@=ylI%Kx@SBh@@bSQ6a$(dBo9qkhOo; zJujY@pB=`FC`o_FvJ9s>RhG9j^Z0YwpVp$o>ijZZ!2Hvd{Eo5iFy!3-Sdgn}wCNe% zcr5i9e8OKZGxhpj0OW77VLBuuAfsBlax3L=l!5<@Nn3BC>Nc*OqGt3sRERZ}Vm#Gt z`k<&fG+T&87w|Gz7#b_{YD}fA>$357C}K6D)YlDLVeZ5OSEvmkWys&z?fR{@C$jZ3 zdit$^1S(>(^d2jDg$Z35SU~q)`g5a;LK!;MW}^$a6LBlLoPa07Ov=NIC2gO>Ne|~m z7M`6)+9Z)jX+=~@hu9Z*o&G`|Y5-M!plAeSAO4ED0FQ6lpDYtbdI#^++lW$BjSeJA zbPTSWvsS07aCmP-f~2TL`foH4a1z(KKsTjWfJYcvsS&%v8 zr!DpTNDuQ?Aqno-3tnZW;l^~*70Kk$226PD&R$ntgU4gW1gb4JRf|*82P@I5he}mf z%+x0?FH>;u@p_)?ml(CoS2-^0!<74k=C5`)_Hj03O3?Y9Gn>4psUFiRtZL{l3NdQ+32s_Om!-Y&|UXua#tB=SZdDQ{&us( zB{*bkK~YZMBbLbw33+J3?NpimVJMhj(d;12{KtHGS^FYt zbX5?F<&Te()SPpRu@X~L^7s98mHVSArW?nkRLdIqpFEcBdAuE=<(?5&Nl6Q%kY}m% zqHdleR2imh0QPW{71rd|4OsGzjJAQ=vg!(?l>^nMSHt}uh?pnm^>rL1w@v|g$a3p6 z1zBS=V|w6PX&8BM#CJ`|G5MQV$qKjE9i5}jcF$EXK5Cm_)nrB&Z|RDN_e5w1h2lZ1 zlM&4F$3rePBM|P`0H%PQz%vOWPwoVtQ~$0+`P)(?`%(dO4#CoHX|)YWpL2U#wB;*u zZI?0^O2rZ@i*eS&hT^lk=?^g>YwJQxaDyoYGHK}V*T2_Cytm!#5SiSzPVT%5wnHIt zA0Q0Ai0U&B0tEsUJ{ieP`kCGlfL>JEz869D+CZ@|x~VCsq*J*2`_v)#BVVD^>ON|+ z$F_Fw1zOd;RQe@ZfFE#-cXHD|-K4Y8oW_S=YWw7$P3#2bpL`I_GM_?XdY-Wy0|mAT zVrbMkk5R|q!1&)AK4a)gsXYsRnQlf7{Ye5iig!Ja_)tCg(QaFDcM_xTn3M+#2|(%i z%e)RipaZEd5^-xsthu{wYx`Vm z89&tkr+eF;B+k3>)xSLorOT~PeDt&}NKdYdt|C#E1q0&XwYGgkDqPwx{->KHy!wio zQNoLIOg9#MNHNR5Asp@_W7QE^{ zquVSue{ocZ$6$s8qb-)#*j^G4-2`+Do(i55Jd?8@U^ThbhpDrQ`oZfdlWL{eE87t` z5)Joo6f$~7E2|L}>-)3$@{Kw$TFBq17e&5-G=n#m*Dw|a6n9A?(G<-d1~!oJuu zwnQo?z|-J9ZeKT%%xymLQKTDscCnkD@kPMP2p1AO%zd(CCD`;EsAKVl*L`}q#>m`N z@6EDR2_!S$T;ZIE? ztJ!+0d*?op9NT-0bnk?MgyB(#L}IQ}^24uK=&VnRRWn2*dZ7?QQ^67~9Rdb1#MN_9 zoeDDBb1`icIq5SHy_D!Z1n`yJFg8{W(l&FEl>2;v4;z@jfo`e%WOLy?4F2png{m^+ zdLr!r%#o8xdi)08yh1^6KTH|Zwy;3Fh7xQt`{KFp2cW(LUM}Vk<`P}D3$%U5$2(>1 zDncvrFQ=3TYke$G1-XQ8w4=F+bUb2Z$yq1j8=EcqAuBFyV9w4c#6Qi=rKS~8RGcnK zF-IgW2^fBMtX9DV zhd}Nd%5tn~y-qo==B_Q4RdZgF@K9cX?cwAg&zO2R{u>;W zU7VDy(;;DjE=1C0QlMbUW04lF7P#1Fv#bx(9g)wI^l?uTpA-MY8vrT`+2Z@6*X$jo zCGjq%N+JfzfhCwSJbBl*PGt==$DVN7i7uVNIvqL^k8dbsxLDM@ zneO5>PUNtu1q2^06kud9AqOwHL`E*ZjDEE2UBr!Ewn3jsTV(2du28Gns?6aNW9ODe zbS&RZ&yRdie=hjFGtX<}L`k_X})G#cj@C!qBF7UKA2X>fo0YQsg)?Wh6Rp#dD# zaPz%@%?=%W=fLK%^36=OWbJEfUnXy}5GvN%4YD@D5&I-V2WjJ`^F^UuCpuF*sm-Zg zcz0WR2(ot|PZ&shlV-4TE-ffS*BA^QG)w}*i5|Gi7FzCUrF(7V)^ARzk*lP6p^!XV zg?_nfoG(S|3832S(2{juW$AwonG-Oy3v++u=;k)MRO=-~wo%~H@oU*cbd}L~X1khB zx!urFubnmwi`1dl+f+xtsOiy;rk!{-#8XX9lNnj4_`6a)PPnWEH)} zY4>EcMNx%$Y0nPy1?}lvQIFn_$C+dodmTHPrpwWDH8I_YG=jf@vUeiBZ6u*X<5?Ct z^>qdK{(bmtX^N6Y?IhF|5x5MBO1KkFe#&cA5z&54mmRHCgrXUZdIpnxNjEr7vollERnMCpghEMkNVJFa4GK?K$0=w<$gKQPzdg72!pAzDS{`$iG-_gJ zV`7c0OhYLqoQ7$|p2_=Jze<^@If^NN;||E;Zq{l1X$N1!`a1h7Vgu26#?A=lW6%wi z4Avt|v4eml%bYwDS~>^k&j`Gpbg7%9^8sS?jXEFkbmI?kX+w!YI)Pf{{owbW*#v19 zvV6ehy7;(tp?dwb6TeL&SOzL@Cp}M=t`W?4oo-GWOu{PMCYh>^qq45IoT(0$gEg-y zGm-2!JJxr45T8A$YTHE#;?!1bI}okeF%1x&&*8kiqw@W_^L!Pl3roEp<1sYn11$q2 zb+1429#G4#Y-og9m9<}NB4k#@Tes9u@!h$5vWygGdxw9fRyK59`+(WMb9Xa*uzj*I zr$!-#-?){Sx0#1#4-89Tsg8vxm0q$0jkc;7y8d?qjO``tK<_u0uKO$+%w6|uq!p}| zoZt7~w7GvzIs9sZx8Ol+Ap1TAgl9BXgP6&NAL_!@qlzh_9y#9$z? z??ZbpZHA9W{g<#Oxt;=}$_Ic%R+ii+r5y#K70tbw(j;B!TbOHJBvJJCJQa0bC z)6#DZ%i!DwOK^1Uz|r_0d}Lr^yGQUY#^;Kv3(@ZPxB0iv|s;&ykny zE}Gq#PB!Kg>S7}osMJo8>hA1E;p-7SYKNu_`aIViI3nG^D|uw8v2x%pEbeHYToQ&C=Y8<%{3y7v5xN(y$@9k>a>)dE ziO3XgD(Dj(@(U^q*YoW_u)>idI?Iui+&{mNKH_iG{4jJ9AG5;5_W zNL=t>IPutfez5fAT9+f|PiuH^b0`ELVz+=v89G~fI2)9VmYYPMKPemWN1Fm-qv}2G zk4`*nBv)6Sq)V6>tb+!Y$A89_kB~p6432zybP-6d^{!kk_Qs=$ASv*E)ET!+@Gwnd zL3bqiDQGAUB3#5LR+Rr}5G&U-s`cZu-KC#+kfH~A1o!%;GoE&C%ky3T0b2d zB@jO#XivvK-E@`Hdo|T|cA}woXC*o#%y(n{DZ+7w`w4`^!63~jB%q;6Y1{RJs%hfG z*6c&}G%>Fy@J?w=`dy#PBz;b^H;+I}dOAAK8*`eO)6-s4jHf_x)Z-wpAW0l4J=?A8 zoj!6ZQg;>sKv2%Mn#=Bu{mWH*#p|I}sj{F;i~@W~Xkp~f$bw~sLPn3a_P zeN}7`&SqKXGly#}B7&#n{p4?t?DD*Qus$InX~*?-?4wa?z{IuHNCJL75rNh_S+~^( zuHIC1mkYPDh8aB9bnmH#CHt-@M&_j-Kw6A32Z$b#IqivG5aPGQv^Qu17@Q%M`>t!T zySY_iKRB}PW(=Ypf442sNXkpto+}*sh-GE=N?>mbHtr(M&E;&3o!LpeVbg*|vYE zX2BU*F@%XocvN+F^t_oqnWz$1rp3%@!7jb&)x}>B3g<8Mp~7$-tj_|}0}F6((O8o{ zK+I4e)=(z0Ae2TV{P%YYFOq(uiQy&Av{Yt?jrNgyoYkwJm)5qt%^;9xDQchyBsHr) zuta|yHi_F`uZTKA6h%>e5z@Fw`&YOeAmxiT<$e^7Ca_6wV+{Ybjsqbh%)z?DJ3q@} zO+wk_afFUh1UNeX1m$_d#r-qLY%X9)p4?e*^xxpnKXZRGvZ~7ESbB=gLFuQ+V;%wcIlp+BS>R zUl^Pr7rF(~d|rEI+UPdQK1tu2X-Ca9)0%SK=)`KN+5=c!xt}moN)(1Lf>Uj~1=FoU z9k%cJTAU|ISR_{L!m^pVGjxzv1(r|Sk*(RgUItvu30uNC$L>HXG?f1Voe-|@J-x5DN>`h-yk%)Qw!uEJXt%*YD9f2DsL<`IybbETeA2^bg#dZSgqAq3z8i>!h6^8Imn-XzNQ^ zF4-sdEmt=35#aNSvBp>OEA~7Ma%t~urPE1uziqiwCOQ-Q4#~+TL<>eQuQ$E7A8USU ztnZUDN3?G4Z^0Bgk3C7|bMNYElGS}(_9D>bOHj920LX>Ssn4lbl}0M_sl%rHtSl`sw?AMZzQ9G>IA4vCYP@F+-8*(5@$S-1U5^X3!ATZ)+UgzhW$T*rcvnGZUN@fF z0C&`(_>Jlg960*ia}$T-8LolMJ(f_{Ugud66Sx_Ujrwp9v05hQsJ*?GLM1-xnz;8c zI@7Ms*E7_xp`CGhu~sh#^ zfqZry&U+F5qn+P{*hbcSg;>%wPdOl&w|Fhk=NT!8@O8dxf3aq9BD`iaOhScv13Np9 z(l``A*dQ^`(-*hlk#FqQ_Pu!YqZ!*4dthtUvMf^rYpb2KGT}+7!<(M!9b!}GI(d=? zpa7tn#U~>MO^|bMn0Z_2?z3}3;VNDo*Z-V9- zgg^?9>}*bW==N=S8R&w*$e_?rP^VQ(gNZ^qlMrzy^ z35FwfWh1*tPZcqa!R~EaY_3>ggH0$QXMB#4Yd{wd$feAwH`=<$7=u!}W-E#`Qj zN>Q<1FmA;wB>O7CyFApxA zm!QsSBso}?Zde}3kAzz?+VZ{&z3wCVjk(7#%dCS-&g2b{{RW8TS8M)Vz@-WVzTD)a5$=0o7)aT&zNGk6=k%Xd23)zI?I)^X^J+q{4jIlU`B{_~V~-;>WV z1;`lGk2&=F<#hSo=!3K5&oLj>bmk<$vCyEy&KMQlMd+3m*BJG7W2ukpxuZwqOGRw0 zuHxr45l%C07v{L?Y$(Pl;7) zTo)`(Xs!3~<=&e-M97+AU6WV4emOT?gc-?7X0Jq&Fp0)U#W?vcSU5e#ygQw?qgm5uO8TO5%Q;aDjG20N zsL$-k3AnYN!ePCIt06;+l{41OZ76xHv4h#Wy z^hZq#mJ^)CRWW#PdS7!H?E?7Rr_#2pOp1Hc!_#3SWFHfs{SO2hKJ)boMCkNo>LhAM z=ft>({Q{;5B82PZ8a6zQR_y0ZY}Hox8n&Mq5o^}x&5k~*Gba{2%5B}l%@%@=rP3>s zmGNTuB{OL063QYqmuy{Bq|#aRtF{7e@zPSdf5GtMRPls<$oyc6&rOq>KnxwerXFL4 z81ia5Hy<-3=P_As5LL7?sSK2?aO+Ll*7r@l++bFRHw~p2J%KdU%oZ4Cy>m4i*;wQ- z=qYE1d;@Ultpc|Fy9K@Wsto|c8HSrCNfGiIi(&e41du~(X_nwU5|EOEF??ef<41i> z96(17d;VtR=W0*ja%QvAzTpBxY1wqm^Vsn%Cv9cTLqZG609A;+UZOA?p@@6({+870=vRj!HilE0j(hRd zR?1y}H!pADt_4Xl1S1zNQBt?h;h${envNC8QyKMW+Az_n_6h*B z4PsiS4Fv<2rAq(+($HZT@w20NUDYq&JYS$kwMDM!SO{Vah_-dZM6KZm$PYQ>=h6q_ zkJ?GgIYJ;H?%yCF35o;_gz6Z&f!`pY8csI|1k~wy+rs68Imwq11SpyW;aLFIj)hl}!or3N%rtRjbhf z7(8Q$NM^;`w(ikW7uy5z8 zYnoB1B1%c^V~xW`8g6u&A4kh$>L-fG{FW$YsN>PAx(9ISz8M_#?I>?k(M?JHeFQA3 z&ZnN1feqV4<@jwrb;#LI)hrlC7=-xqpT0vc3?k$11TyjsN|xXeh5`fpwVpZ?;9b(U{OMMxHl+?IpW&9(?A%FDE& zz8eS{Y3{wt5ux`dfvv!7PmUtC^H7rdGGg1wvf5Tx6__zJvQpZ=#E1+TE|4S~n!Lgb zxz36cIY{LF5GTGjb`v^M45tCXeLzmO=-l9tso}_5OZCu^;>s5OX4iqXLqHwzL=%>} zTHmVK+KUxui4-}B6!Qt*Zxv!^=7~?t+sdE3AyD-5xy?wrsLnkA>DzC$Cr}T^dx-qU z)*G+=v#clZJ5GwiiYu4d$TQ7HW8xzwQe_4&T>y8DizMAo1T8@v{0i%5K_2P7IrR_n zKQMN5Z>+f;ZMc%(Zr%*^{`3$OBQ-w2!2K9Y>=SIT+lKuv-OTms$_ssgX$q*X7n`Df&VqmP6 zB(ki%7Kp`S?6F?K5WTMa?1i=!-`>?bcDt-AxVFDivG!=5qYVG|xaShF<|GhQ)|H^5 z&Q9}<*pnv0$hmK0O};J97LCzb#j3B@M4cbUp{;##FNuppYc9o9HTJ72@wlnWlGOKv~cUE@zPZSsq<_{6*#fX@0h1H zH5FDK{*Nwa8)o~s5g5ER9IcHF%==&Or2Ucj%S`io%~Ui^?9J2VH*FRO2E-z0gWF@{ zuhCzdHD@uZi{($(0``G=F)h2X+_>}e8}6G-B;%eQBC!PRA1nW zPpyq&WJVv8b#_Q?mP=ahG=7-QJ!1q#i6kmVs@dEVYCrHNA3pB&HEg0J%F2oLqQM3T zOHSrUig*jBYE4b^*aqw%%Y{!VFFKv&Ykg400!Q$_P1s7nVe5#IhveD93!7Nb z%we;IbvD=pqfLJ%O0uEp?08?dZ(L=F*6hBwFDDjKEd)cbTaq5OXS05%bv9q?3X&slvCStMl2dg0NjdyX zlpz=n(X4&teaQ8Oofqmngeo}TO(W3b$Xph}XcLYALe4y^%2(RTO)X1pk`$#rh~^FW zs6YF@d`H_fd?AhEZZJBLjVJ}kzAeUy{`47F?~Wu@UN>lM^PSruv8@x{8tB@_lc3=J z9I0mJG$N+mD<**%pu`;52e}#l$i#mFXIc1D$kVlBl#-+wd5BSC0_bN^}%Y&M4i)D6q9PwwIjuk*bipeE_sw~>aMHt@Fwjy>ca`hVM~msIm&~Q^Y|_J|bO9uX=cq|MBtMK6_*)T_DoAr4zI!CG@x^)#nHfGReTHF_g|tLu4B&isI7sTrC5 z>Q>G4TrDhNyL^fz-{ig75P^&vD$^6ZJoptTqdnm#&3@=ojKT}mk#@W&P@7P%k4^Vw zeJ2osq7J-4I{75IxsE)M9gpFvz8(5qy&gNulP!AeYr{F@^@<;q%uaQQpT~4s8nAL! zLI@jnB+r@cB+d^|n|K|+C8o*sHt2W`3OmH&34QqfrH|s&WutQL)#jwgWc#&4`9sOq zlQhiQP5Wo$y(9QB&T**Sx`*bv@zu8y$d_e3VMn~+pi=_3G_I^FPVWAx_rv|M5( znSRJ7q01hRG2T-(gLu(uGEA?7?M3Zjjkq+r$ahG0 zj`lAP`J&IY0Xs=gn92FKsGYhG+zO%)SC2}E)a~hMPgvtIB;AMf`01O)_Z^ahwxqXm z4jBgiE$nexN~=iC(^rPxfSt9m86eRK92Z!liT0d;gIzb2h%a&F%<8?NMzcvShZqFn zQ=NJJE}$=654gJsG{~{L8)1IXQ>93 z#ax+Lc#KNk-(7G{M1_6zOPo3zG-XPGl%pyp-Yc}Y`HWzXM;Y(eOkgBXM(=$OnC^S` zk*1|i5KZ~!nFY(SqNV0{LpS}yh243o^XciPyZH)y6(?^$)CSau_JmpSaMe`UB7PYWY$d)6*e|uCNp@bV zrm;SvIe%chISd{0l;GE*v%QS(MOHnv9U;!#+zd=<-I0J5XB1kImLLpk6Z}a`8Tz!x zQXb%W{=}+>D3?GlH`<({v*mV59kr~|hTyUwg-!V&whXCZtkuf?8h!EX!6)t~(Im66 z52-V#63Y1U2<^>AdTd1%5VDQHLB3P7=w;7YY#qNnbXxM&Z#1dPQ<*4Q;Ef$8sOw;xHwr>y@fn{8ew0`3kW_ zE1V)$p@pJWKvoLl{uuBkC{MP~kZtt=LHCYhb{MAUlN$D%b3RxqnaD~vbjPL>q-Mym zQfnpO5G_BKR&ny(Y-(1y#f^fq94h4p#KBaUhika$Qfa$*W$Pa+Eq8{NroKk9`uaxE z{Una5H-BA29~2GQW};K8|GC0^qhwwR!(e;Xp58>HKr z7Udd3Nb;~!t4N4vCz<&3N$S}CTz zMg92P>3K7@1MwAB6*td?wX@<`(+S?^XB;X8WILCF(5&t*5Et0 zkMz)yich*6f4er`o>d_A`B}S>9oy{Mj{ky0s2xEUC5cInb@sS^>~=3h1d&Rr;Tk=& z^#E=)+TBqq##vJP+~EUzG{jfbpu+uuj5bKWS~W?KatI?+ypMv3*VJxgv-Gtw_Ner3 zCj%82Q3*s3uELEZ@$qOs#^;8icw|Xd%W4_~Z(d=q(YnpN;ZXe(jetl86#$1#p%v5F zDOAVy*56<2>T(V1X>!5Yx4)=vV?_v5%2#1$S4G>91$_g(SzLk=Ufr}_PQ$bTKxV$s zv0O@00P!3mUE+V8RGL?rZzejx~K0;ln(J@!oLldbrJZ`j$uk6Wjw6XH}?owy#UM@PQL zLn}3RWZ(fO0b6oI@zL`mDdV&3mNUuRcQ1HUn>qMrZ~($L3VJ<%(YB~WAsN;A%SH)= zdr+5dvJS#JI+truhf%%W5h5l#*J}r8ZW%k+k?LOET~yNCK0UqHR0hsermG{2$Apdt zICd4&LOpLBZVoq8ysts`a5;3GgQ5A)+}B^U_X_gm@sn7A-f6GXd9G~B3nkXwxRge} zOS;EWSx<;fWYgRK+~bOUU7)DJzd>LVn^hgj>f?}$^9bB_{1SzQNhco?dNf5GbbO@a zdGzTtJym1+pkM6) zv6DZ3x04@IxBFQ93GDg8YCjjYxaQR|RLXvitXu0`mDCD(tPWVRthF9vzdp26UG9Uo zuhCm>$DQrwqG{AP@>=uYS<>W?`D^=;659Qk6t&D{w3Gq?IpZ$!HAl&;S&8jv8wWe5 zqofKt7+L|5*?E{aDHYx{U+5HCR)}WdUAdk2)LidrQ$6{v zc~Kb_-BE3g((Vlo?`}^|)Z1Pzaq1fx37Vg%76rcyhLlj-u8x>#)DOO{-wypatbw_M zVZ+F;)A7+rV8Dqm>ytWPQh+&X0w_;}}j~i+U5Z?LQ2d z9_^uq1Dm-GuH3@~@$yMSx~}0#kGEs5E)q7|YyjIFa@=S7)vA1FDg6ep&wTcJ{9rBZ zTjUcdWp^FECn|hF{2gK&)l?4`wpo&8ZR35SDfrfcj@x>pjm6!rHi?5<5_P4TNaDwV zfQBdV2iC>cwoQ}thYheI(lCQKHH#Z!?oXeZ6-?_l-u+}{*BPEcUr0;6;|5xDqc0az z{=Cc#=-6F3Q(QraHEOcPN^%)ycHHxgKVWtMU=Wb{Fr14}esQ;VsBd(CR9SXqF?&6g zo5S_eF@GE0dl#*FURi{utiO{cNSwhKIN&G7T`MO96V+@uIk}NUI9kp&Za3B~HlJ=W z3cqsCvjHk++1Iu`C~OJT-?Cn>qqE@D2eaC?y&YsL%YBLPG^f@R7q|IMoy2QK+F_K1 zn(hkCzmJi-!y$e<&J85q;UJm2MHNh{d(%6#wC;@u=$rwQS3bY2u?lFEW-%&Lba|xo zf4o(Yvp9cc`_{z8vs#*7A1{~}SYJ}p%Ls&;D&k$~4q!MvJt=97WoTO6M(>BHJ(r#Jw%}~OPkUfCqIqm_eSWG@>{{~<9X&dG zKp>AF_rS-~dO9ctFY?v&5YK@I*Y3wSydKAdI(&5-_G~cX19UPwp!Kd`=cZ1cUdy$% zg5!tB6jCh|Loa+HAqdbL?0s!m`U{3Q{f%4|X(QMzpO-6iZ`BVINHcY`9ydwGy!})O zJSmXaOK+c?8jhE5=c)yNvt>39Hw^PTF!JFc@<_KJRPHS$oAJH&Z%pvX-81#ucN;Hy zHwXi?s1#yBVdtW~HT%ow$Cp&6Cfz4Nrd$HX@T0hMBP(|0?-0kEtC`XLxS<*HY{r&x8 zG7jkE{t_^5R?#$g+{}2&6pr^wUMDXlyNZbEnaIL4r$3UmZT|g`Grm1LqZf=?YiL&k zOXtg)J;Pc{i&`_IGt7vu1ykYNIHt=Hq&|Mgq}|p*0OzCO{xtz9-LTefi?+@hH3hAs zy3VkQGrp6cI^vN)(WO}m?XC+}YDx&AHH>I$cTu@?HcTo4N)ZF7;?1ST?iz0KXDsKN z_bWc+eV9ER(vI53x_JgpU4Y^WWQxe;!6dLA6iusfJ2u>;HqUH5y@y6Y8$*5L#ll89!~4LOsd9FIphj?QqF5$8d%vXK@a@7j2)iZ z;tUF4EE}mQULPcm+xHIpq2+N`HWC_+^rl}g+%FkO2ITUp2nJ3Wv1;qn5C=u zSudl!rOVN%S+V&Joz0eJWh8nlcdaAkCePL_(HfRjE zODR;C;^}LuCj#5f zx#B@1_3fn4RazJGF$QQM{9xh=)5bN6XqZ#bS!y9c~4Wh+g>IQ{_30ne+H z!!V6-LnummZ*9-D6Fjt*!R4BIusF72(t83YNd!rvdpe(E$v*zEXux zf?TY&n7)rLyE;9B3QYoglTzKdfY$dUyZ1~c>pPbR`688HeI%#R<6N19h}rboqnOz7 zn1oPsf+TW_%8AAo)(>vhe~5;|)C%MDg(Q?k|)M<(GiwrmB(*@Eq?r6|w!yVB9AxRQ0fzy%nKAomDROF;VaZ6(B! z9>9Jo;5_=~UqkuJnF;ImbID?{)-aDy!i5OA&1tv9aP5IbRR?naL0tJ-RG*JON7zLn z)}{TbiSJW&e#sjn1)`G!iafTAHa{VqPzH;Jdf8RDWa{|ZO7ADkYu>juizR`7WkZEU zFp{1a!575JFI<(1#y+Ipyt;=!#oUBiZbphqi&{7l-`FZoNJDE+Lauy9&mKkKS7#?F zUk&fUQfQ?3eqvaR4|~(&hMxUUR{5el70}1TuDcjW6y`uGCZ%BsL;UMqMqxvk<+K0~ zr(sPqZ74{>OK_!q#-}OGyr^B)qT#u(YDT?smcMEA$;f%z*{Dm&1JS$)zOOXVI(gpP zI;>ON`MC91PfD)}(_b#!c$1aurftA|-TU211Xs#&iLPl$zh=Ss?chsrw?ue?1*#cbwnXR1Bg9PkMM z#@D;Uuf{b{nE$B>-^&P|boNX=%Aacp80rAqm(vX9#(gNXa){)3SeMaR&27ARb=9nM zZ@?_u@X(9reI6&_+*dCZZ1R*4KdbQhmt;~&S+bnk8zAvtUzc7B`51d$#?^8(pBP}u z@i>LRjIB~F&aIg`9bgscq|60b^VXTIs<~bNc#~1ie}k)iD~(;*^^eyhc|?8DgRC?{ zxNLe7(E;gBp8_#V*T1giXkAv%hS<+28;p9muE(jXnfs&nL8}>Vy$(6g6>M|I6>KBL zA{DtyHV#>G?(46%yI(V$npWMIKxW<{2JSFxpXI5{QgQ+!e-Bvg1KZhbK#v*XnipXw zAT_lunW5=de;}{@w9xx}&@#eRnkHyCb?%B2RSWaF(W0_QlX084wZId-WzO9kGH0M1 zzEe_*ZKyc<-W|53$EMb7h>&n%s^e3N{JR;|zfjzM66oW|i zhLV6!?;Y~7rPd1xKRz-2nd>2gk(KhT<2cyas7*ON8p?Yu4p@3Wbhr*6Jf(kPtvfZK z$~rNt`Pdxq)_%@+)t8>1)kI{o97iwtQz9G$kNg}WEBfhO>mMZ}&h&sDI=fkUa+OKK zrA@BQlfX^R(AAw@`6wd0PpHQkS%0ROMv{b2^*)avzZd8E>}_tNk?7Mx(44s9C9D}a z6bgSF3HJCAPeE0otcoq`$wsA_E%A#nzhA-tEJ)ZnmEMD8YBWkH;e`mMwF01dDG(fP z{7^uxD-*d3EjZV5X-_xJ)Hyi}7Y3%!A9x)DO@ zZT;^SsflyHo=~Gtn>VdFNJ?{ z8)E#JU{Vs+SM&@zg68WOck}qXp(r@%n-I~ixe7+QWnQ<`j%E6f=Y_TP2lh&9>bJMf z|5TvAB{NXWLV_6l+Isl-JWL3N@jJUU$WE6mn2ksTzhl$vGs&-~KX3L|&RcL@9)@0} zoXv10-J+_NMHC(2C1x1`3)p!70D1x2l9{hwkO1=rlkX8Z#eFVz!1J0fkGzEyZ@{nT zH?fb{Q3MbSE(oT#CItC~ttfh1-U6s@cJef-|2*d7{D{hy?Llq5&Gm}46!a~|@mF@lnmPtVYqz3t-ips%lNmU{mCcloL z@a)U+;GNnw&|%7HZ+=e0mqp6X$iO6G-2&m9~mc|G&e3Z07?Fxl|$T&n;Dsm7# zz3UZSin{jYh9x+l(T`}At%VDC&4$doKUqpzeG@(D?pa7s57qK*Qaq5OuFKxH=Mxvr zPA*j!gw9&bWcZ`#4$%8*@{K}o2vf3fy|JAeyFb1W`!LFoJS`_QGg6(`BiK;EJ-bsO$!?}?gx(PZC7t_ zcXKV4158F`+ffhByhCaJ$Hqfsq3ua{cdvpG z@Hr^EQZiVqX8d0&_aAgmghEU1R#LF>;wDI2+W%YU+7>Pp1%z+t!ftNkwZr;4r#M|k z{dJtg#Jsl*3)u_+U=w16QppLr^c2_xWrLNA;q zW4Z&SUexyC!sMd&0leBTF|ru7kLwqazR`xa(i$(bF~jHUPWW4}RKVYFkkqzAWh9Cc1%>k--g(X!%tf?(eXpJT?VoMf%T&#d{{4 zD+o5|e~4Eq<@f&|(~l4VXXw2(O`kq@+bTWa%5KF(S}&|OJhi3xL8E>YAB6QPYqe~d zmzw2eo`U9Tw}Z2jb0w|kbF0oKRM&}%lbo{NHp*6SOkJ3AjhiwDo@*#j{2nvGb(7b8 z=M8S2wuX^ck0y-%BIiIvny7ICH}KLJ1`%_Ybemg=5t9FNyJM+Y{X2g8gX^Ve3|loq zSlC?)db>%`>+oPr!tcw;fg_09^t52)3v$8ygATS?ZgL=z#$za5H>ZOE(~jJ*>j%W- zNI$4F@jrAhP&~U3MQ_8XJ?@$yIsx;&y&~N=;ubU~zr`)9=BTMltw9H9o7ErWR$Ve; z%D7ytZE6D?c}aePNErw!rucs&dN^LdHHUs~{5i>XlC)d{&pV&nXLa&#dd4f-ITsnM z)>vY^;c%zf3POS>DZdX?b-~!726GqK0wp-|urNUTJ?bm-{Q~)HGqO=67ptyf`yFCe z%sOI)b2FbBNMV#}G^;oRvO$kilfz;4YZq7(hVJfJk zm~ZyT-SKf)3ERuMod5d?U2EPYCU{13%ia{0fl6?BCD=dgEZPOkYTk-vH24#Bvs<UPO7px}(knh3o^`LpuOOyuW46bg^;O!b%r!@fCIN zP$u#(>i=Ql|G-Oh9~jJ$c$kiowaoK3T-E(6yM9X?OuI)Ov<_#RVxLC4m51mV=c$a_ zDJiW9HqRSR9xg6&fSVv84au@6qIj=NVW|ABkLbFOLKkCIHEhE$H{vl;_?Xo^g~hl~ zyiSBXYd59~bM|h7McyMrl0DmCLo>&i{iH_ah!PRgpEv4z0s(TBj9bP$=lPCy7EV+n z%;niufoid|oC8C)vC1|2whN04|Hf)f412ku0jWeJZ^RXhUC$yxxd@$KJ2tZPhNV}K zk;2XJ2fVuOKAHpn0-i1?v;@PYI{C%Pkc}00mGbTv+ZuaaLX-jHqXt9>``h8!pP)i( zHD@H_ZU|$12tK>uRAd^&-qtpH{-Fs6<{<%I_0zdtSc0{&PBQ=HZvQ5KsnG%-e17R{ zESASs9-GQ#hFYtfX$+uhXcdZ<%OGY+C+0)?5%sIiHNoKn?GoF@+xl(}qzg@1M(-9* z7nvjz3d|ye@PJ5%UKdm7KbjHf$VP9Y0tzC*3?(PvCJ%rJWr{h509Hs0lUj~MYS~RY zp0kAsJ*+APXeKg2S$|in6PHj2c-8e(i{1fda7#+YQi%&NHmWx!YvWTqxXyz>c)Q+1 z8Axp6Pfo}K-t-?Q)10Vuw6VP7BLZborYr3_#TFZ|o|; zb`c$miKU|!y~XjMorOHd|K_-VrS31v08$6%dRozBlFpyS8wuk9z0cb2#7S3a^5mf-xiV+-MCQ!LxqCTl}X~t$}!&0T3z2dh%^j zvB|!|#LreRy|OR-k)hL7uTsi*cKRfEBwcD9D>!X}K@ggzvsp;>u{P8?t+J_CRY=xw z5#5<>kT`w3u-q;%hW9tIsRo%Xo#!L#;1L!N+?XI)${Oihb|IowDv4B0)NilbrcHCA zs$^L#O5!(H=jNoKKu3C3Js*Q*o?q$o6=d}%MHJYXZ0ZY5}_Uzq|M(B6>PAeyuxc-g6inb*3$F7+&7vsLB}+l(#O6eHFy zfsGeYq(g6&>Tg{}%p`rz7EeF0N-6eaP!?V%ty%bWCTtYXMhEzNuEr2g)!T+8b5YqB z<*#>Xli}j7`iFU~>~)Y{1Hr!3v>Sz{3#<=zp;Z&gf9t7#=iTl|61z3)CiGD7D4hm; z*h0-UwhN9{jnb0hzw%{k-_T5L^MlT#D!rASt_W)b+tCL|TiGdOg84r2pJ<91#m!H&?0VnlaXYD_1O59=i#mv|J?SwY4nTw)tG5B6XkB0eMY=zsj_bILv?;NYKg zn(x#);G*PboKGFZ)kEv4KOKmkzr!V_u&SmUVFmgfy-EsmGBeFqK_*@Ga)uYt zW#SuLl9;K>y+1qN??i(bHx>Sp$*(`K#|3nLuRcsL+W>XU>M5mvPEvnbD*~xNupW3Z zPrm+!@J1r32}xzvT~lqzuuD|_2Dhdf+_g+qUIHFp=J__ zU)lF))agp&h^_TvQm=~2J57;A^TU1~zeAVGk+>c{?61ISB$gxY>LprY{7&KA*O0{j z-=pTs1p)_fQEM=#+_h_=)6$*F55LxU(Tr24qy#NC3n|)mhK4~mU@_N2;8EV-iBa0@ zCjJ4&2mIVGd!SU)z;8W}%Ue)HE+*+lx2Fphuc#GQ9ZAXMuxdlq;Bb6!HyGtADXa{s zPBh|~ny<*=K0TGKA?u*b_>M-mxD-TPyG?~wK6Zq?-74+2eHuAzfFJ{f+j0fyLRTkM zJIsLdtSXz5NJ72bu3Tb`C{6V5HP>vmJquw}VARltrQ(FqK>;+h}(v zgA+F$LqP$8Y~DfD+ozzWCM(3G1BJDWXi6~9i5tPt-^mR|j2E87f4~4?x~;J*za?Sl zw3%1>W6&sS3AX~-d%ssLOp`xOHP(w_Fk-L+XP`Prb)}`78!aW>XbF>?E z0u=s-u{tv{dYM84XL`)}b|l~5)T(J!G*=qso0%UrvO}**l&kmfus=U8gN{}yaB8@< z4#3}owC=cG@ctxb%+yLb337cgw*RvfKr%9UtTMVrYuc1QGGhv-Yw>M=Mj*>NtszOS z;u~NRxzwpD0a08KRUh0_U<7oybjjg?e+~NudtOSEiKCgp*7qGvf2`xJikc*(Q%h|t zX{Q+Wr*FZuM92@j4s2VrSl{7M@tRO(h>Nu_ay}8USs+VwAa`f&TA9N zY+jXFT9#R_(NWqMFG6Nus*(SRGKcH8BZ^EE!*nUc3nKP5M@uUs<4D2UhVFDdqOlD^ zd^YJ6Bax7~Nr#FgMlGWJqso2U8h7!^K^-9&dn;>2b^ygmGkToo)+9{SHfU$aP(%WX zd~hmmAo+`-I z!eBXbLFP;J%-f;b+s_*YGiicA3o9NO$$NPQ^jFt4cNTM#Lp zt4a(rX}h$&$4mAq`t|P8hn+eHreSfU>JFXb_u7NhqH$RAM{y86z?X9dYj9 zN~7vph5!~f@(JhQsf^ilB%pqHM!!y8M_ZbKNLdX_V^9P?g}?loV}l_LI|c`xHPlo# zN~%N)NnO2_A{PUVfJBMHA9<~}#Pz7TV^=X}WGg-q_tKngXgs5OPF4(y15xHMd~oPf zrm62xhBQ4|WLYauKIky6#p0mJk&6d4Wu(?hmcEe2-c*63Yx8UUZsG*0{HKjMAhLqM z>5aCfmK*2K-1@w1n4lBXKM+iKeR(TI&DXNQ>rbPc5A!~2mTSMA7UI07$=u4HtKSBU zQP}*EZOgf}bHX;)Zu$X@7Q86iKdY9J<`x^Eki;h2H_JZb2Bmo)!c_(Oe|Op!=qUog ziLTUf)r=0nEN=_RL^CAzACB_MRjrvuQdpuEN<7T^a~99+-N>%0mGrtILLKU1+VP^} z{pmu#6kARo2%VjU}FkRZI#< zRW52BpvpnAiT$KvWfM|`W{s#WW2BqG-Nht>vBa>V! zl|V}&Xj|tasfMjm%_|2q-?AYGmRi{qTC3IV=E%caOX3aqt@l3sP;SPOEJ?^I9xt}^ zq_;GkIU1t*Q23w(iSVEa<-j8(8RDc)m8A?)lD#n?IgL-K=0U}7sJ1(d4ohb=vJFb> zRfmsgvk+o^Q?m@V5Ngy>kzkzJVpW7MGT!hQn^hc?GCNS-nWi=wr=!ccjsxwhW}-q} z;b;gxbgmr7YY_gb*hv7!_%mMRF>)_>L84Xq=U`SER-_mwKMp!mW%VnMeZX;Q8uvBe9noJbw;tJB7y}a%9zQIWt@>$=V~y6f{ia#H*@3ap+c1K+@|sFvggE*) z;4PXnR^V69#`AQ`_l~R9mKD07kz_wra*W0qY4W#K?Ff~9EsSzqR+oj*0xdS%68MTH zp)~n6M&-7}C)>O>CT|9YdKr(fu-mjIG~4CrT4&1rbOcpK_G*IV0;$W!bC;9ZJli2z z9nA(awIH@_F?rDKFohGT$Qik1Y&E~ix{~8czaJEo)I}>87n=RKRyMianEOmdGQBbb z6be?t*5IX7mSt2HrA-UUYC*r0=ABE07g!Tm7o2~~B-A*URy0_@l)ygLEG9VPjn&bvl}1!zVYrTlN`*>l zqA6`qQc=>`@kF**0gNp1vjg?j)|Yxw%*=)()()5p3%@YHg~?k!D9r@~v_Lm5F4V9Nb~A-Z_`P z*hn)q>QkOhESBkVvQ%~d4{{9qb~|3~#7V?>{^e{J1Wf5=Spk5ilZMs0?Al(=IDo0q z296V$aqXjnK@bxAvHeYhMWd)r0NXOOS~?|={M1(^`%A_pLHYSvq(H(}XdhZxF>ohW z?Pk3jziC$BfNwM9U!luwdUMMvkawmdp7Tg(xhXcE`g`^bN?WmF!^GzU=F9zk$hRcn z4|FoC>({xZ`?kh>3gutch};Idy;$fFomD%TO1SJ38ttwhqX}hAHXHhDYvntv0}o8% zGnAB!s$}ydC#vIJ6;4iJ)nzp(Va`4qCD+Tzb-y&KW?4HDD=sz*WuhL=UtG=?tHh@^ zD=0LQPuSJgREry}B@|8#l`s|%HSa{sw=*k^G}D-gltt{rEJFtU@e^=MISh2jl{ISY z(Ge?fQYlq5SXl?y*vXl+FcZ{O8u#lqO0<07v~K0wa|RWDk!^rcll7afYSq(F zS}Ulksw#_(1e?g2kdv&6aUwo4cbE%p% zgMwG{fNVBW?_wa-DV0P~W?);#A%kWm_wMTYvkH_8(vx&boFwb(?Et6|+-?8rr zqo7%QzM4IJkg!EY+}}f{%DNMTaaQVOhP=NNPmS}o)wb}#H_9G@dg-DgF`2RV)k9IJL8IeWdfI!ni&7g^7^as8hI3ga?)2r9%k2H3DTH3Ife*xxUn{JhDd;Oq zosKQ7pk1`G=xWR#N=@)mtvZ`)yk7=+Z)>txiIKXbFhFkSzqvkU+O4t{FRvC|#oOW+ zM+uQIlTI`XF+zi7DigZLXmc?7J+Qi3I+0U)L)^9svk-rk!FMq3V6a?)wE&w3Eee+@ zt?QhIDKYy~#2{iwVp8T$`D(7_g?)WTqr0Q^z+h}F8-~ld%>p>>YWDJViiQ}cKi$Ps zySqAN9$CB+Kpm~R`Z|)+XpKc>=hNh4eo=Knc@|{t;bc^DFrpZBQ>srJQI1tSD92px z2~4i0NSA_kr#@!~HN}U`@irGdcR=~~@8A0mxjQJU*YOGURy_g|!Gm(;db5$@@{Gu4 zY+$3SIj`Oa1J!>98R_*$E3$FxeGP6|Je?54K43nlHeJ4{E2@}hh0;4m#@)gh{3Hgz z9QaaG(3MkD4b!!@!sK=pmz9-s(pF~gv=)kj%^8Xyfx=L5!lNeh%rR#;x|TS+|1EJZ z@o4pEEow7$PxYWY*=YT}tHyf&l2O&^h|v{f7zSmtctD5!es`u;vg@)3ho+-0>rp>% zGwLkcmS9GRgaicz1_l5CTLi!HO?Unl?>sg81w`Vp-Wb0)Gp7}NXuUly(#wScjEB|} zBB@6|p8GX386poc0${VYVlutrUo=7KS%#*5`$bfg{!z^kd%=4$g*EUUI4_6I=l=}3 zy1BQH*B7z+OWANvDtLt_t|lXWFApZ<4DvafL#IFi$_THJb$NDry3uO;caAp$f9A*; zYe%J;14x9=OQ^~Vc-SMn1ir)4BI80x!i11^8w%WmzH(E6Z8pgdQ=t`WX>)9HWm8Yf zM~A@|epopa-<}Z}XSdzX7Zx^Hs?vuZJ>mF{d?(3PXr}L%8Hppjjp~5H%jbk>(Wni= zcU-Q8;4u|(5GM|;5a{XCjvS~nT?Ju5`o$waS&dr@Yg&$oOF^vs%{R~@`XwMe@K*`f z5m6k^gd_YK+c49wzN?1(#Nyqh@fso%M8+}T!4sm_IDfpW>;PB|{0#!ZQzZP zIIpYuDu<>g$yC>?WX6_3MOZCU&HX6~=`CjRo&O<wVpm!3`cjkVfTS&q!ke13$v+9X2b;_oa#CSDm&w zRlhvK?FatU|JuNNHib?4k>$|&w_cm;%4!7tv_|uCekgEcM*21FB2Lzw=v53wg3#^^ z$=4G|GC~qaBVrBupT8YwSmC(6V6QwBK$e1yPXJSXTP zn!p`mnaz$$Q#d$$JUmT4mTH6NS$L@=I+$KX;V8f#m_>yrLkbtZ!8?I&143r<&pR+pvUHS-_=+5NBT9u7e74waP1q{J6?m$j_q6w2y z+ySq(_ptFRa$|o*BIw1SSiEx#8c~EA;4cfUvx*wqMsO%znKEy_x!O>3iJq80|8UxQ zg|_60)h*;nFQ)ikGywoUuWe4K)$3%{H7InNvuN?hFYqYlMs>Z3ybe*{H1mD}K!Cmi z0)!0E0N6Lv>c4AbO}!dyn^F9gDC5mV9qBuE;(SRZxvE^d3G*#!GATm}ekIp6ZEs;! zF$d#58}?v9$tN{J-xdj~e>_7{AXN8|;yZ9;>%)`=+qrW%En_BAdjMqt?S7z$hMgs2 zeIg#(moX2Gua8Y{26oAyGUQY#Y1i=cM2H0QNH&l;`8hxBhuJ3Y-`vU7tg(@)F-Ho~ zk$4gdIJ`w7fx8mq^U~5%1GyxriLv+2I{dzkM<6l9MCw2SsJME821cP!f(I&<15W33x13SOL>mk1%b~xBKM)&?cjJO$44XjgohI0kL z=`o!$XP|}{5yD&xF8_=K`vWBhx+j_lfhga~Zj?{D66xDUwDu@1&xRs|TB8d6o{bX% zaW=Dvz=c4{oeW<%77p${q@mq6ZL1}4oBeqeUxX?U?PP~D-Tve)7Gwwx5?c%UtDizk z;u=DdC012NF&#+tvG*+x1&K=V3I`6O9149@iNV#aXZzLw5j=Znen{e=JX?UOaGD+; za58*Vb6{6qq`k@z9u8FJ=KIB+nk!<&I5#Rdo;PS%TiS~76Tg2WCg(NO_uhVF()i43 zmC4bB&A*TiN=PNN6j^4D8f8QzS13#bSW@QqXWm+Fa+~;@y9&?~Qiy=Vbk43w3p@Mo zl5ziz1gV5S4SMob102dO==hB|a6II;J^owk8}v015(f`}eci4P-(~%}v8m&Hl)7rY z*?xB$Hw4EK=!;X-dN?;2H9PkmzWJdfYI_d&UFlM< zbPwABOOMJVc2)evGr=npS0sv6b6eKkn#)X3U1Xj{SP7jxST|&KC{U@HzOE)%Dro;Z z3jhjknugE@ipWJEMg(ox<;wkbuH!xGxAXei%15$+B8|ih2e;PPuUmwm6&d(`B(w9r zx&4P?YZ6SQr863Ds8kfv-yG{11Ys- zb@X8+2J~T7Qs<{a%Z<(rYn+WXmq!HnPS5w3y0X(9QbB-lA+7?^Arwa7kKXYFa)xE3 zDKv;G>rm%fq@a-pB8Os-?C4N!=r+4ER%U3P!+QAq99 z$jEbTsX5LXmd={voP1++l93{wvoMtU4pTGy1&k$>li;#H433_jx;8l((;DSguVhAT z8w;8NA(RzfAM;#eim(7acIMKr5f>s!rXNvNHMxe*k(J^OHV9|3$^*F(nUNPcnUOIc zotY2*l}4OmlP2JCpuBN@M_jS7;pDr2MorARRLet`*pl^^t|HJQjIH$qADPrevGHFs zzeFJ?L7rUTv8y{xsTm`b7Xzz+N-EP z;QbXZzQZGK0tZ_X$V(5Pu+<)+t z9qNp4*xqH;L$c$ut7#@_Q;!Rc^IuR)h?EQl43xflY9!*lRWomPiM^j1QhIuK?d4Rw zP363wO4NI5oqIc2nC+=J13lr8td#M2FCcV{%%g6+D#Te%5z>+qZ1xw;t=IRZszsyN zI;2@I>DiKTKrpa(!q*l~5?^!oxLj*-oEx4-mB$uCl}5qB&}t!3IAE5?KOnV3q_97a z!}qA!Rfb1Y3(wCL7Os(Q?6u#GHgp98kIMbY30del3pX|@Fk_@N&t|xiTAB*6=20y4 zz4s_~=nmp#03oto8mtF`a{H3;Jk;$I>((z;X=a-iL{{1^Mqx(DCct0qDda){MaQ9eS zOf|0))_W+)m;gE*OX#h*G@!%y5b1Krua^AaG$y%lu=}xjd=n(o4!nmJr#3jQ(nTn~ z9~vkC0wnjp>p}Fs$_=9f%sKXeV?r%igIh(8n&v0UKJyf4nO)_WpSboO3p(FQowTM= zEh){FnU9=Ow#Mh)OLuJ9_lUq|A_oqE@m-e-@9-L)!zonyLu3{{eC7eghY2!qs+4?x zv;?Vk#U*wH&QnOZo(&9L1`N{(7e)^E_#MK3%?z@iQJ07qxVO z83ycP0%Q4Wrbm^lSq(%Bta^?rg`N$B54Khm@aa*Ska>Ss@7E-P%H2KQi#t^nCho!~ zvuS9MVeQ!0t~{bUB1%g=m04S+m@k?#&2(n|B2$ZDrnc_tO>kmBU?aYAWe`4w5hZPb zL1>m@(u;i>NM|E*8WPpzzv5GG2M`yx-)QPPI!Xb*Q= zzF?^;S`Q_((yR*r>t5F*fdn|;MUxP3lIYv8j3jP(!=FVW zO45}1&V#fCh8~R#PuqPjfS{?HYJ9OSg0<;Qegt_%S^j2lrlg{?V2LKWl*`Je4*5jm zKO3lw4M3Oi&*Nz*{j;MIUv7+O71?8m?;H=g?kCZ^n*w`i0-AVc8C&Lx%?Y9lwzZ$tVPt9$(J4>vPlHrb99V~sjZ`|c=A zn5n23A%hGU&De`ubZTm9z-gnoAA8>k%0Sxjy4I0xRndGS4+wJw-Lr)L>@dcrvrizE zT<*!C!bS_4o)AuM_Cic#5nbJ4YzC2O$Qklz{q!^{%M(k1bF_wQ8?|Syw7nTm!BR&s zMi)>VS>k^~GY@%q@2HbpQq(j{=zgJ~kFh0E*WF#bH_%mVUwG z2pb1)x8}c(0@L;LA?Wnkw$ktNYIz4&&=kqi9`k zjDf4oVKfxO-H&61mks}bCjnn5e$`augsPN}3m*6T7Pl^3rK6y-E%905v}5+inMjO_ zQF}9UW?Jj~+br>S)f>UoSfi&q|`67F5T9 zMfW2|kLSqYT0*5Xka*vE`_i7XW`PKe{VWR#j@2Cx zqIWyYuw9a4sUhV8`k5r%T^>)gyWlAouh{RGL&Zjy*YzA&-{S6P@4q~4--so?3K4#- zvX_m= zHhw4SoDN4{idd_OuA_aws6xsc{M}6>P}mm_SEj)uK0Gug9wcq^y?rvAK;f{95?Au zI4m= zzTs=wct|JQ1uA~F0ow5e=(lwZ^*u)uGeK!PGj(&UuQzb0><9%nvR+O(q(kK;@*`yN(LJ_e15>c z+|WMPK7NpFJmt(jCw0Yryf}Sa(f=*D1za4i($BH`sFpG-Ty%YAY+et}O1#s29Kgll zGpO))2J_j0-XkWQ#2;=xIlL@(9a0wD+?elQKB4M7*cM>4dCQcE`Aqa4eV+IfZ1yN- zr142#x+xvo#jfX2r{*)sR&*4LgrB#>snXzRz2|q4yN1HfJ?PD*7Tj&<&31W3h+U}- zx|T3W{C#3?J@dPs#8-TnM{r*ol6;TiM9kgyvduy4@)Fu-0H9enbNoHJfjxfB+T&6G zz7E&)cfvL}==i&yaFZ8Z7X1#W^Wn#({dU!5URl@h`fRiLiQ3!nqQkF8>u2RCtMwa# zXgTAf!tZk~-m-+MOBu zJ2<)l&goXOp>bw27GPo=>oBH=?LLE2??FG_Iv?%*5T{Tv=cFa%-tC~%N=VcVa0q}hjv*&5hmkqvUF{3BF zYRkqss!#85CtX-y(py+dI6Ad(sNk9^aTrfFmt^aw-eU8dPjKqzvqxJ}q!DIkV(cq- z8rK|>EQH)80Ue3yIk9S;fhHarjvfH3Lfi%=*)c}P+GRZ#VA(7@e&xOV8b#%=HX0Jd zoW&|6BitG0cIHMa=%o)3p28E9xT5Fo_SHahf_H}8^>1XPU_UZ`NteztL=b(8CrdlK z9b(qd+TBy{_ZQ$6^9U#@GJ3y0qH9+q(|0X1Cjzd zzEvm+8#@)=S!s6FUQ8^c{Dze?K0Y5Tsd&Cny0=T>h&lzqq~a##aww z*W^k}DLPUf54>-bzBccyl==CB-$u&ZnxGTizbf!ZOEBdGI(K_>a39y^cz%}0Zl%1@ z(E|QFSujW+&wS;F!K8#6$JDlyIHxcE_{)Xij{Z~Hv^^zroiDg)w@OK!`^-B@atiD*nUKYx8aC09HZEP%1(FO443_YGlW$@fZVz?W@i*c z5c#qBJmt?e-PO}4=W4{J zD$J)7$PH5DBC`lHT(&{+fI>hDp(Z%tTfjasu3aYkU+>vb?Kj~|lkSd&qK6q%=1^4o z)(ZM)KN@c9@j&q@{e1T8-LB_zL&6BjX74^?aNH$~pUh6WXsYwa;mH}vc9iDB4hL0^XgNqk&bz2 zyP!3f<+-AoH|us|dtU?o+Q&l9e+#?B6d7bI{P#1`!gLCY!XhR9qh?Emyk+F={m>fBu}JZ()5FYG1K2U{g> zcG$Hz@_M*}=0g3FXVt;JsBBHh_=f#u7N_Uxn@>jhvmw_#E}CU3H>t?fe=7azZ%DrV z+lzk>mOu<5%p)#yV7;dZHkd40;*fZ%&oDDu7F6ckOv5cW#39v1?TN662@2wtknr5j zttid4j`AF@I3*2xBKfd)KxXw z_N^cwHOiKnK8@xx@9zxeXytB<&NZjzKho9Y#A5SRt?Nv~Sz-mxc53gw4)Gwug)@G3 zj`A@nsg0c`R7K`b0B{!o43xS$^Gw%dlOpdL_Bl=~uy~mB@+4Jnx}y*^6t``BBJSAA zGSCQ$xjBO8>Wk6uej8ZIhrGO6bDop%+e#ry&3%p`?q<8?@kerIj9efPWvG;5u~HJt z>nPbC(n5GE?23w&2>mhQv;`Yp%lXg+8!8G}6=%e6bhO}*rd_h)jKj3jniJDMY!{dz zBiR_r|4jkg9i^IGvtMERcA}PW#!iC>WS5)A5~$!yv~evz6;-UT!R(|Jg)0_#TpscS znT;frj6Dx1>zC{yq(Y42y`oFRBX;?zbRQnG+(l0%^>TwAG81dxA*sIKw?DD(2q67d zj2qic0iO1Yu-wlQj#nS$NS~k{!Q6BBvEykUX2n&y}CX zQ)Dw*+m*GBN5g)<*)L?>or(1PDJ8Y2GubPptX~)m*O7wog6F66nfBsuxGP?mBZ_kF zy$K%rI2qtv91;hV1KnCYx=(SJ$Lm3nV;@N?8Fg%|jOkVrtp;*bX=;p2a6gfIG@7>0 zm1c9s_P6Oj#PP~(yV-g)SQll+D_7dp;;+tDB~?|nR#x%#6m0>w+ThjMDp#s$lQ2*N}Og7DK+`E-v4<2MlXT17|o|vl-ZiD`=jY`3_uXvlEZ}Kf{FaZlWPCl zUc)6V_8hYJ0l8j&m@ZR7p?C=b3%dcA!4I#u47?aK^KD)vuw6GhRIj~4#dSgfuq9Le zmtGFX*S#jO8g4Zm>F*t-d-w=l%UPQvG^=@9x7I+oR=-&Do4fn0a6^|Zn~MMyryVK; zSSK@~X8fNqy)1Fme0>dg*IwLx_F=x=HAHmUjN6)?XZU$L%D~i)Tkn^riSjMzIT`+3 zYPwvdC3M_B;R5;}#MJ`O1W4a1xQlffI=b_G;eCVG3F@+Gmy^ecZ z%$T*#DkQ%kBawN{%!wm=F!p$(ksQBk7W1V&avSyC=gx$UQcpG2sx9>Y`Lm!-k|}fH z=(uMa()rTc!p_$mJ%Ai&#Jb!7&wxlpne#OS@o6Q$?29G8u=lfc^%2>gJX4})MRS}v zmWjZc%nV#qcyfXZyYJCa{^ziwC@I;>RUmi=gt#V99Y6}!T9b|k$P}^GB9K%eJzXS? z2h~U_-TJIxAcpA6x*fj@umW=>v!)K95Eh=%X+sJts4dY@%cqcCtt_lb`P;xN4_%7$ z_5^i)HU7&Ea!y8C0u0&^LH)S~?STP8+>vP*n?qO16oMO_A3Rw(-fkM7!quudJWldNWUW*cJQA`f4~VM#^rH`rG2x1> z2klCa&={Nf&H&+d!e0}FF|dIUh$PqxZf`9C{+pOHXoY*Z$d#>_Npz&m0Q?7NXwdHQ z)%laB_Fd^JLQhPC_{x2u(NVq-LTAaJdK0(I7o~I8nh>v3) z+#HXs3j_HtU=YvsRhdwB~BtkjDcW>`T45yECBdkzZg4T9P~5;M2ij2j}$ zMkTpewTUF?mI+4^f57iZhwp3!`d}3){5j^PFSE{f`x0@1tjXNuqe&kZR`?=u zR*G5jJjtm0o)z3?cMj!41Vw`%Pk>0A~H$Z^SfLAeK|k3*8Mr2ED1#|_N?@6^sHyi z$g|Fj=VsAggho__)O7i_UhS^&Q@&>HdkN*a9{X)(;!>SRH{60(;i8gwQ&M2dskPsf zO8@dV$IbS!%*!FZ^?x2TY=ZI^jSmB~j~&e=>`r!hpQw+=Y>R@u1xe&_IA#l9hpCk6 z`I4rQEO{TGzy>dC&8wja!bOp$V1U1!_M|z4!o;qdPHh4<%`Lr}>+*1Wqp5}`Sw?#o zu&hqZP-|hpFF@ps+48s>5};R@6gzb6$<^|nN*Mv{#!!AJ3wx`nd$j`*)T%7B=9Lwd ze}+&?t-nh1VFD5E$aA+yQ_kU^HY|mrKU!w(cT=>TNEggr%eK_uMv2(wk|yvq>_Ezo z1(67!?D8f;>&cTg<`)}D68sbhKF76Fyc)m{yi!&&7$=5u8C!4r@PVmNKf&0ue=uN! zwM#s=2>!*7`MUMUc!e3a4o!B80KO|GTgi`_DOk~;Wp>wCD+d+Gi9rflM2xGL{_9Bx z^Fl$LhTqAMh-ujfR3X5Iv5?cMfUBg+KO8O!HYA!sLw0Lcdxa&hwu~e_hE;ln??#c- zU~|BaFGz&^oF`=3&h#?!$HE}of+4IgF@3mQU|gNREzKsh%bQX?6o27Qe=XT zDg>LrXeXy|Py2b0gNs{b<5;f+_T1^ZT z{xDx?JD&fJjK<^IAVm6~#t;&eGzf^RId!3fO`&>-G~>4^{`OMR*W_WTwJaDgL?~ai z_U!hwmIi(f1|<3I(_b{lLPVeAsh z95B=f*(pB#h=v+KW;EOIsR1)}`)oN_%RMmqbbnh@cJy6Oi6;B$vc2h-51syfoA0UNzBI~*R$ni5LFnQ zzZzWMK_S0K3GlG{WB?@!5e&&U*eC>y2@LBNnU61!Jd$cRBBcwT!_G%;$v?%3J3k=A z4b-F6f3!N;C`yYn8&ChM2XXGoAf?MynA&N)FP04vtD{;uzo?S#dD9vvyJ2@wQk418 zJ~zpAe&;yF@t!^>$MYH=H_5$Q&n&xM?LM;e5*liDbQYQ*2oKDhTG7Mq6&$p3Uv}yalK_3RfLKF`aj%cM8=?B94Wy zb=HPVQ5Y`Bye8<(ufRp<>dUG`ojo~?$ibrRU==7Keb3x1x%-WpSsd$*3`l41cpRQG z@siB-YAxJ?DbHP?&|A%poUp@KA*9x-sa5OG8~qAayk7>6~)v)yng zG;bmV0WFQ$jEBTL{rfp{Z>8e(A@c6AW)+q_+?oUlk!f3cG=;#92Pv2{F=eb!XRLd{ zg%?%}sg;U9CGeeyr;hgQlv;I2j_896_Mr#WPS!P*{B7 z_kVXlcvQEgo^ohZ?Qg{24-pj*fuDU-d$4*hp6)9e&u@AJ-?dPO_IcwUd1MzfXf?b1 zHlGJa+59qppQ&t;1(aOR*`W?65gSG?cbIJjm=p;NJnokRao-Q(86 z#P7m9jz7q6g6Vl8HoeS`=bKy7_(b4YpP8F_3?{ZWe|;?Z=xdsVRt-*kMgP6#AUIf{ zBm2ePb{Jv3`}OUO)>WQ!E_A2^u$xEdTURKCwqHTX@8I|oBPq*dtF)KZ<4W6E_R5&d zHxlgIvjJal7jP3V!H>kQ8BDq&e$OO`wvnKfhG_lIjSsVJ0Z$mPNG^a?^VhWn(R4%* zFyrspQXz1EXfy_QP1d%$J3JowpoBeTc*i)HAhHJ26w(2&{Fk6@T$8z+Y%AvsOtqKF zBEMIdd70JpwA@Aq<5Vm3>YN5(I&u=PLe-d1D!xpSigh~L?eAv?Q5P#+d(6o@kHDw+ z562HHG&-1KeySI4V;80oNN~t@l#9tIO_&f@)}f%G1;L8|LndNKqY|`bkB_e zojF42TV7vzw^EaXQGN5>VB8X>{iRv`&`ZJ=K)4PW9WE>yzr@KBh5i3=^-kfH23yy5 zY`bHlW81dX9ox27Y#S@KZCf3yW7|o`$)CO7{hj=0>!=RaHJ++jbKYZAz>@(|@Ud}X zhMO}ukU7uGa*2~-9F7Z53s$&Hh!3Izf-WK^tUVX^rV-i&!nNOoY4|RIDt$M%SzOOhzqL`39Uz)m6Z{tmEAWK5G2vJd`JNME1h7 zVB19!g5BX$U_2^#z-8kk#SCXs=kenDD>Y%-jE)!k`58~WU8<4`Wx@b7Lrq`uSV{j$ zO274LusIjAipvMk-R5ycl5M)m_W*@WM=(LCL$&+_@s5;^JLc%HGhnSNveHn#d-X@5 zYPpqwefAK8LgdBrc1h%8M<3CHV3I9{j*D;+(IP)$CCi>s00PW8Frel^6ohaP`kwBokn#AFIUNZElA|z;q%j%{u7L_UK}t-`>c4H;J6i8P%^H3U-nvj!#g! zauC>Yrb>~4)P_E^slBXY<+85(Fu0wH6lZH!;K5;R_E@4BgEsrVO(b*d|0~u1O}sy@ zWq#bz$#scB9@8#G|J^g2XP9&qt){cU!{vDjq4qhu#x!wl=Bn*bxxD8d{%@1>K*D+D znzCWr$9pJjWdu4Hz%!%|jNHSAt8DESCI%rE3^%pSCX?eM;`7b-ld(tUgEa4!+T-Y2 z*lp-*m#X-yOXTl6_QU}9{7J?D(}za}(exbEaDtG59V)fffFbi}s{8lbff#sgAY-zD zHqdiP^lx8@6N$L_9*N!W`}iKjc6v=oVJ@-;N?|Hc|7^waIPx|fjq+IODqC23+%79Z zO{Zn#o`m|^I0R=^$daX#GSqF#FiGZHs4IrfDl#?c_&q;flQ?Zw+YEhGUB$to!xQH4 z$>6@sJOp0Nq5{kNPq5yd_h7vNja>g?hw8uCWjzrBvKgY7zD=0a_dl*su(DB4ni5TTr0^zcP=!t0jTOr3iW_c5aFI1{#m z9Ed9%OeXO!K z(8w6?@W_O;_ZBnu=qE)QQ}Zn1od4^l-E4^DWk(3(0GZ2i0-!Tt9mvH4p-Z9)&RdL< zNc+LP`7!t;=>gqbHjz7+%_6w_%GFAImoo)|IcIRMAO6l=Nkl%B*I%R;@T5Rp=#tG# z9kwYg({2^V(L+Tg#@ljgsTbt@~#Bx2sc)+UdI=h6|9^e%p68MY{mLwJLM4|9Oy4 z;(3x0{9?4)7RByO(*VkZ2QtieiRamteji^_ai5`1r(Jjj%&H-~PEScNcf^+YI7U;J zj?^gWvJiJ}&Mod_lah5k`yq#YQ1(!cm$!+_HCunGa|n{962VurIneoLgLY~$so`%y z%nG?UnqWN$dL4Q;Q;dV7(I;jBu!W0_e^1)9L zSrLTMPzM`DNc^_7_g2)Gb=TalI8(!^fD>o5`_Vy6dPsBn`D-xGahPG z<Tg7oRWwvq=ZBV<+_V{8WRlEEeR12!fjbZWY?$G zk?ic6iW1ww>=+!b6WnZ+bi$2Up_FUd@Nuqor#^$gM_43GXinHvIsg{f@K8dM(pA57 z_N7cFfpQZ70MKDTUn<1ch21MPkO3{$5+lNSYnx%5XSy2L zt|yf#A`{;^tE}yLIm#+s>L@Qi{r(5nzc1P}Lp%!uwiPKaG^jVXs$8g!J`^dO|BIE0Oip7yr)@Y&GL7HdWcQ+sXO#^=**|dS{010a7GKO~dq2>9KfPb9 zGVr1IU)H};-*;jbKFct2>YOSDZM2jt+K&wc{Oh8iLhn7_311I$B{#{7p$eUKO$opL z?o#>gsriw~>r}d!QEK|stf*ujo@`Xbz3tMTjy|DuAfn5$Y`_1om(4n1Kb~(doog>e z^UzL!02`$Nt4pJATXYVP*b%2{5+!Kh8Kn@jeK}V=-EsR$d|J@+ekjmws% ztS`?y&5kK|g(9TwU+rAdt+u~F1G2Dz!IUmnv^ySLLx@v{S5Vqf+Oc+>ljpc;9PCMy z>vki64?-SnQnmQCmjq);Tyfkp`wd?rPlVdGfIRBv^}yByoMr<$sAspZ8mWDSnvAQ0Rz z$-W3O8Binq>0xI6$-C)n1VS)|KYfQx@#)y%@Ip~ayezD!O1MAB6zsI>tNBg?C=7&m zMFxCd#72_k*zS$kO(It)s_0Nf#?z0OGJ*~cq@o%mxtU0G(03#H;&;ymd2uNbC=xDL zEm?Un$}RRLF}j8crx0yW`rum~0`9aG1mgr{DRMIe2MwOASt<>6Zk*cyJ9kJ2stH+p z#pB@)K~mh4sr2{r;GLX^7Fd7HFk-<%))fFfQ0hXLeaF87DB$kPL2w$)4FdWmcR1lH z+Z~u42JesZP$gV!ersTkx<$KCDz6&=l9So9k9+F9fGXiEa1QYGNt_oqyKQzHSqeso zsJm?C0iDm;cn~|0m;XW8YX6I})&JQrh}PS^Y)mzh9TU)c))Z{cPlk)MCw32y=D33s zuq#=1sc3G3lL2xpO%!ZplLQdkx^MeF(YJ0mt?0V*-}HeWXFLDNMx@}u`fugb7$Ibf z*?l1QdT+d8{)G~z>-j1D@_9>koo{~Uxd@A%4vgQSNpsm*+y*iXYtv{#9%fA+jy4Su zHS#8;M4bZdRoWY?6uXkHe3J<`Gl`&M8wH%uzd?jG0=K{2^!jW|l$EK;R#&!@=zb{S z80dcWx?j_Kb84*o;7LmmRLs@?Y4EOCXIPLl3NRQ^^52qmyhgD}5QI>xZogsLOT{!_ z1l_71&x$kVR~(3`q91|*Uc!auyjY2-+=HF1Hl^Ed4|s>K40I+AEoPGD-33Hi{UUju zNSfSli^lMsA>KccUjK!i96il#nNu3O4@8te6@UDWeaRk%09L~6%=O>ph%6@cImPs4 zDzQ=999>Htw>&;xDXX#MT`CcW8|DSVRqexmSce>4Ip`0~)njVA`3j;%ad5CJlZ$YZ zRY)YF@&|7148nmCGcg+=VaP9#D#bAwg7bD5znLvJHa3{ZG9&(btuT#BtI;dbBuAN` z9CQxhg8|Fild=h!w5J}7L$isLV8&USBS|N27G2_ad>|M*B$fhO1H%(srEq|J5lF9x zQSpNXaMt?FsR&x94&IJ5{G4}SYmN}3D8n9XTC#v}2xx&dm1(OfL&ysrJB1|%wFaL?1P{jY!tla}mH~#B5Ld<4l$rfW95iLkAEWVyB>!=>O1uo1 zY@1cil8c2zjf*%Z)B}#!Dy?R$!b6mV1vm_I@OPQ1;2}#iC&Ci?EbN{#rO4OjXQs{G zAA3)}!G zWK(r~ZijncmumYSrs_oJbpDIZ;hRECyrdUou}|%u<|I!X9IKFf=xM+VHrg^iri!Cc za|~%xs0mB5h6svM5ux_dne*o~WnO)4QM*Bl_D!x;=MqpDbi9Nw)8f0TO82l1DM>?waKm$91BAt4g=}1O1-0 zfA(4(qJz}_`yNROen40E#DDJhn(y+Yx-;DUcdIA)>A~|aK0$!bs;>TDv&@sv`t=r7 z_pYzYT|1LJK~G8<^C*nRI#G&!JJ3ylOJOKSG;CxW-!1f;_rccpI+0c;MM<}YFcAB> z+9fJN%r(!3CkBBEvFyJVD7ddK8Tm!F&y%@#1<46APyD>ThAc&T#aw4M z=@fSvccRjJz?I@S@7lo5(*pvZj8;{}*8PvFCbhCqR9Jl*wfYKImiMv%_(DZTWq+PP9Sl_s)J zEa$|?PjU2Mlcdmw({wiwI{ zz4bw1Oh=(&gBeSCC&{G!HT3UYZKSHgI#QFf<)6#&9wyca%V_lIKKEopdx#CUo7jT- zk3S3V9%}CNZ2_V-B-s&dXnq9L>RwGP6&*D-NUW!)muF{J0a49bba031XP}S^lf`5P z)QGet!#=Tm!5Cx1fu5j$m_TXczI2~iEJ&uXJ6Nd-d6l{Rf;3wLctlinTHY%iLfrOL ziJABzf@xb$ciDe8-$JHsNP+h<_-wGSQ?jS-ME zrc^OFb&OBPr;(NJJV5Av`#M9?iU%eC4T)mRh-oN4!qYS~X%VwXB7(LVb$zrDZ`lk z)9;mxeR0ZV&%@lFS{z2h#Uc=`nAFkYRwrfZ^cQ8u!$!>85j@QLGQZ2ywnvZ`!qG0X zCHXWEy!TDkYv;|FW}KOQIRBdlB}U=%J{s3CGsTS+*M&xL^^wWxcUIt_=l4VoJ7~Yz zd}!cQHQkl=`=1na`!z7TK)NkvISRQC%VAlLlys zeNghZ_p-3y!i>{YTUpNb7>H2I!MC)ot*Qj@veRJAxIGfs*eaH?(vTt2&&*RqWm(~p zV!Fz(kSQV!Vw$qTnbEpSeR5>{dqS?Ytg0A%SoO|=juK=e zne|QfP3XhYZ`w+GV(|oa8r^5-fLoHkl}tPXlwq~X_&UK;!9*d!_b`B4U66$#_vTo6 z2799BT$rf_ad0Sm!6}3MJPVwO8r7p#GzHB3WQS6ryvE|Wy{l3jIa=Y4Cev;lqLM2SzEcu#JoI_f zLET|9Nn)|`aveCO4hJ_fNSS+-?ckeTRwAVw;u5AAueeF`?{7Ct*ax(;ssRaHLKYGWZDt zqiLfmY-x8+uCOn8NKWGhrrw*3X0CjP5LrV)0sNBb_XhukXfH+V-NpLsy~LV9yH^|9 z(2G_r{P&G98Ib-KpL?_)``Z(DsVnm!ew?T^lsfzSFw(7ut^Y$8MBDKnUqLJ?SkxJ} z&gfv4=W|$VofQ>ZTO4d3`X8he&Z?pudLK1%4-{L|N#G4%_pRy6RNz;5zhfW7={Il* z43p_;gpxeD^DhnsTZc-E2#(^L>^9%XoU$cwML5+VQI2=tqa(uk~f*zv+HA z%`s`&OBUKmCsQ&@<2`cDh=G0PuV$NtMdn^OnPMul<=y;8^Y@Ld`_sz*a6A$2cb=3a zHlStGWxe$Dwv!sMRHE*e0(U=fb#2c)ta!+Izc`Hsoa`r|?e|D=`pBY;E^nvbo9}I% z7y8yin44~CLDm5D;IKj^-_qnaKRfde;E>HB@Elo9+5te4_9dA(3D(a7+8i?>d3QI|_^0E~&xc-!32z?ml zA*Y@H0tzpzZDp4<3|&!^4fYixOFH8fGK}JkE9);tIel37SGjSn4c6fE=ve4+N}2W@ zbo^ujkYJ7i_~8bVbJdjsea5Z=sQX+&BZ6o_u_Wy%y8)!lMVd-gk?Rd5KP4eZ>FqqB z$?u8Zj}PAvcoD*c>@V<<{@m`IB9J7L>rnoXLoTCq;B31omJYN9J>Tz01vs!lI>G8K zc!;yUBn&j9-a)8l*otzh>TyVK{oA1SGhFo3-D|72erImf1--ya9; zVrXFv;Y8B%h_PqV4U;*FM~!F33dv!H2O+L_#-1W025wmEsp+dbo&kdBFE624&5#@u zr9w`iuaZgAQ{jK0bQ;m>^vor}Gmtd$4jai$HGFr2se8KH-e-A+CrHUq(qaCy*WXRh zGIn0DoJ>(xkWIuvfnl1f9!gBu!oRCS!6{RSSh1g4)AS_lQ6=<5dJ=dTfHxRkOV1aa){;Y?O&_{TGvTh}aXa~( zEVccP!{3mB@_(w`Vc-Ji|b3N;s_3oe-imYdwVljKP}R{KW8V;3UD1Pnj;AV>+3G58GsZY$`2 z!TMgENt^vtfQ2}M1cQ&+$*;k}Jt36T=j8qyD?8ZjvK@E2gLE3R@;u{Tog9EomJkaC z)?@n?_o?wYj(xI&CH5oZUwj?$w<9>n_j`WH;Zo3lK@5WD((i#t%i$9PDKG(LYjt^w z-;ggpyJmr@gr!P=^1e*atpSvbSn$Lpou>HD0y+h>um8EtAHjpgZ~7dJyqokb2kZbyp#_=;)UHwyDHweP ze9w^t8qYyo9~!EArEaiyuZ9eM~@kZ1ySN(ct|VW4+zv zc7ZL-kYU_Dd)gfbDPiU4Kqwk@xf^y(IjX7&bUtcOTc#TO4x>X)Ah(7!_v?OO>f7Dh zsi=KvnvQrU@)6@?&_Lq_U$5}+0~iDrZa*2Ko2nfbQ@8v*+#pNp3s_pE=_0T*HB38f za$#iE1KFh<^ldjzGuScSXve~4}93mL^@fyCli()DnWY6+Rdu!Lu%2NDDj z{)Vn`g)%Kw77i;Vqr}{R*s_#;gb*@pUIYeG2;Qj}N5afJVmW0lMW9~9h{1T|0dH*a z@6ZDisU@Ks5RN$M#DU1jDDH-6WWNnJ9rP|3boq-zx~>_Kq5+?B>u{s=Nl=P?*$%Ao zs)`6PQY9~OjPCLlv$aU}aUPhgf&-5^3)0Ydfrj86F}Z+U4i37>ru0)~b;dMNPW%Gc zJ9-o`DHiG>6L~dj8wGbnq+`C?j(iSW#a|739*GrWJwfVmNdvHHv>r_Sddd4z7vLy> z^fovRszjlkIv5@IO;^q>kHW~X6sdb?P(Jc=#?mvj8&^Q7O!k@!A5Y}ojCRIPv5c5& z;xgw~KdB6=aZ6*g*I5j~1S`hkXUi}kDn(-;2ypAWkHtyO`@I_WUE_S|$^T`4O#Oz? zpwD06oGr=W>lfr3cq{Pdc=oLzy3{Mmz6OLJH^84L#yO#r2j8QXhxd7wItvxf_;Tt8 z5P@m>WCLHvwfF`XVwxco-3@US(4X_A=h4?Gxw-$)K}4JJqGvhX46! zakF_HqHGbjh{3t2%bR&I0kVpjiPzEf-AmuUFzkcfRr&7~lqEnKFk3Ft&=Ag;ZU&10 z#0J&A{Gv4(>bCIHrmAhnmL!rY=xgm!O6&3aa0|w@? z0AWYHDvayANwo)cBsQ5BMk9U)f85V9z>rhgNEE-{2>&SgV-A07|9->Hxu}Ez$?z99 z6U+5-8&ka*vr`j*DCCm;@?QBM|5Ot+(V!*DYP*+cHMR6as61-~Cab z^6}04;n~l8>An9|Jn6Ql6f@4`^Sw8BnF)INdEH@h*CF|MQU7a_E*8G#Dpv4N*!r3q zp+CsWz#$6uN%7O&zh24&0_+|c6h{wBEl~)wjS1I5V4^$}W_<+7#oC={%vY<_2y0qkd zf|FB1$8I~6AqVzhm*J*FO{$kltDnpCR#s9&>MIH3;D?3{g=>qi9Za_d&VHmWM% zK3_~wpdk%anObTq1ql-tP3M7td2?EYXFC_?M;I?qeY`K^G`8a{i~ zq3R7t9WG!&yOj`haM0aDAmR_30Dd}2$_5Qt4%F<%^6*nw4JaSHP*IA-?=IC13zT{k z#L5`f@&SqqpO=UX9i^Him0{~{*X|@D1`=P)?qr{L~s&~4yJK{LaUR|FQ z=nJ*_m>IiUvHzoE(TPD41so}tdNMM%HSt|_Qt!%V?~9c~~r z;rizggD7&jB|L)mt3J%Q-QoUf^^tI<_5T{jSM22}+=>%74zIWWxS&2o;6i4i2rDU; zFK*D^_&q_CDT!wta>+my|BPbS=rbaBJd$t3}^qr`?s#a1%_nhuy&G&6steX zo-cX^x9vG-f?Tdy{UH-OwS(uVU0Z(=<9b~iAN!riXfE`$-Zre42t}(hJrC2k_wCy# zdc+9^AGvdD3onDD6$*FMLN-h&LB3G*@w>?WL-1>z=nk3BJDut_A@&~M)_q6&%X($@ z%P+g*U%j~N5iHgMFo^Y9;(-{2O9kRt7#9t+PI@pt*Da@Jw;v%=1UG=4ctKi#odP{3 zLd%s6P4lA+tQc0ON?XC}c?j|R^K);dR;zv086kx%kBeD14MVVIGP340)8uF#GX)d7 z`zjD#@8c)#kAO~ADps2ViJ$wwjCIvR7|wyp$Q07;0%ku|gJQOxqQkd<(#T{D2m>@y zXjA_{b(vHCC9Qz&8wGv4W#B*O0^1>tGVXfcWP^ARQ3c81K`z3vTk*?TlzInaM&rA26_%|xzdd)BUpC5;9wl+=|C@#B)9uP z$7+a26w5sm2;d{XYZR!TqUwVKxTBg&QuhVJk%a90{`S_RJ^z4z`kP3;U-WWq>Zs{( zl)jH_Mx50AU36`GzR_j-hMv`GyVl)lZZWj9|46$$a-Jrwj(8h7{2jKWS5)Ck-0iv2 z1K7pisw*q2sIj$h&~4Mc$Aqzeu8v|sk#$BWo#s?t`{{6!LX)(E!Wegnx~svFJ_h%F zN1cI$SnMDsp9!;ALbe#kEwf7&{8PUYC{(F~W$!$a9 zT}b%a=ZF5`{Vd2d6Z=XEkB6Bijla$@%J|Va(U0c9INu-?40Ac2^8=OsZ#zX8kcI%u z*+K3mBI~qJpB77w3_N8?c3ar}!}Xm~=*{$+942}ILhlUy8vi%CT~FjtpW;=%<`=Cn zFN<9u%!RAwecT*V_o+GshtbBXWUqtU3d`f0Tgym{c)7=5UkgA<@i~o4`R|;__NV9R z!8f`%5%mqjZjmY(Fdbh)l(I8(nNK2-`3yAS%4_}jItlA6_VyR5UO%NLDvV(a9n65G zjY3xy21)26`b>H7*H8HkERho)<4FlkS^;>XLm=yoGlu%)cJ*-4qNF225N{Yvy2Qsn z8H^mAFkfg`5vQHAM$-XtYJn@hqpul?NxNAsKL#O$hEr72&xS+$u_^O8U_e(pn?htX;%hR z5r#w^t7yzo1SG84r_~$YZQcMeR5lRppsr{(SHx&I;b2%HGc>LUDuLDi(veSS$u9qf zLZM;=vt#g$^e>Alep95ye&7{-X-yMW{ssvt0(~>4*~6JPAy5zX(C$G<4kFs7tJ0kG zohRFb%ceBi%Ho*OpzQ`J_&RC3fSznq@;wKGkT3z#V-LBno4VCB`BEnOPl&%PKZ&W4h@U>R=-ggU&Fy@TGJ0H=;}+R! zb`N?zj`V>bfU!581z^cSJsytiItP$F?wuy{;q50cr;*CK4REfSXlJ2;ho{S z&Mf-6UjuD{#vF{l+kKX-jLP84w{u_q%9Mr;?~q_g8keZWTm@nN{J@W(} zn7Pxd_S%~}ZXoL^Alv`>!sA;VUfG-1Y^dd!ZIW-ckq_REobEi0LtfzYo6r=L%RAY6 zO7(O?9b$t3OCkZOTpEy8VUR5+feVD5`crXpnKeDzh&->kC628xU+V#{&UsWF)5zNA zT%ML7&~e^5pToIeuwtWEh%)er6#E#n^78F4FC`p6O#fuIsRcD4Zt8Z zBaeQ7=IrpsTLqjbt84uc2Jsj?Rr}jAA3Ja`x~RG(nrT1wnWy+;B}fJqT?AU36kz)$ z%rn461$)3t9k{^mz$fkrw~YWeZ>i%5Ck*bb=MSuHxtRusIHDngYG1G{ z4LII`_3UzDG+2epiQZH0dkDT~$WGDGR-vCY1L7sAK>|-7q!tQ>6N^mbf5&LlXf-m} zu9CH9*PbPa4HuMn%}8<9MrSG1ATfZlBL*vt++)@1?U!Np;GYw09sO)bt%|T#akf2_ zvT7?~p`7%S$Ly6J6V%tm8!FIo*w;)S-4%4~HF5b^aM-=rtsZ7oz6ZVOY4 zH+SCK%aZ?G`T5Z#1^xi`ItftW5=)DLEKlBhCS@50ebZ;?WKR z^mV4}eE~bcQk~ZvxYD$}M1#ptpsY}BU zf*y1cuxROx;WD@61#?XO;D{&+%-HH}eLAdN=;<4SB0_)08J-1s;~fnBxOnetbrD8C z*ZNi=7evakz$kb5s0hW?)mJ2W!~IMqU8Adt%YNPOi92j?WJ`aBgucFx937dc zo3ev(4AUAVq?_`jP7iX2-qTU+B=S)VgS z4ku8r`DP!lF+m<3vmeIg8E(Dd?-D)E7CLpO71mG{IdDEMCxmt3fjukT0mir(A9 zf@07PptdD6G<4i5%K5*{tN{tB!%;+2m*?Vup3_m;;9XEbT1%b7(H+49IOPuOz2c>E z&l%YtO^Q?KT!O=5$UtC$duQGHIYVjXwof{0-Y_WrcLEENs_oQi((-JreGx^BapWjr zpzhl^_UYvHF*eMoJoVSc(nF`c5_&Hf=g$zQnmbZz@4l;mm#kb%tQD5(H#e+N-Zrp5GTur47lXL~Rq)Oqqy=Ml z8P1-`w@F3c!kfUMDQ~?K{%5n;&zs&m`Ri}@L^m5Bm*iwhl}-OLwPMc)h0rKt0jsq_}(;G%FvI5@Af4A^B3? zB?`a@r-cVt#|PFT*9id12|5i@L+C71!ZTcqT47}I;`Hb-(t}`swh%=&5q`x6) zv!ZcU{`UPCO#fWnCEsHOMa2_o0vDvT_0kH3oW`ftvt@$I>yN9CO=!Z+o@0j?&=zbg=D(sOXh$Rtu%Z3)4f-8^Zw~9 zHcfVd(M4pkeA6xHHOM_@XV&{ug1`IznQ&W`=u1ukdFxe)82nGL4y%oFT8jg1y!&kGXiPS1Rq5kw1tY1+Tb@*}|s$a=-AHF=NdW zTHy~TeXiHxn;D7|-F&1!#%s%V0ji|a#b2X>DxFu#rMuROo(%$ds9MAj26cocD!5z1t=rxHYdepK%#Tfule_XYmKr0p`7YvSIFY(FqVrpbPE z&RwRWfH;<4c9pNOTUT4#a4Y{W3qap%ZL&rQ>1%cxJ+9wcEYxExTH6!a^QFhi!s0MeZ#{faB8JZrBfXw4*Tgz$>9xObe;w#; z_XONz?7gIt^)uW&1LZBRd<}P^Tv}8;JL&e6&PzYz3*xMFNP!3cdIRyRw5t}#D!F~OjVBU zoZP`ZgpwN+mnXDSt`W!2%C=KM_$k%qYSZxPX?Yf04FNl|=%JMr2=sJ&a~`zji0@&D zR#~ADT3WT-)3Ul-&2P9y`uF7Kni3ZiN661|lseh6y%4UP`ooZxAw{P83=1rNx%WbBIs!SIfjJ$ zf%?GNL(R#dN}pzNv4lNx%0{+^KA92QO4pBBmc8mDY_Vt1>LC25ht**~oq$e)c0(71 zyJb`Vj3~rGSl8r)tzRR3Z6gv{PS8EZ^AAA_pWVT;aiSvoHv-p>OAL0wFiRkIDW8M; zYdi9f8UC^gwU)NFX^Z6t#V$sMA+LX`Buw{R3D2%rjwR%>-E--m$ja2v1@HJ-Kin!T z6;|!jc~#8C&n^gSA2|#24!-eIVGQOt)AUrv`P3mS0kT-|10La%q7qS^4KsQzkcH6o zgaN7pH!|UDbhhZwh7s#sEu%}N#W=^t;{$DU1)Lf9*gg`|Y-Y?*6Mt5DNgZICJ{6be z-UyuraFi>-CxZ`tsLs%2xxt*LYKHK&#?X7g1uDxYUl~3@(kawEpKqsn%jMoQ>uqJL z%S*i1n9>ZNTzIfvcQ80^|GWS*G+}?$TC@7%?6f_)#ZYv$=rf@~UNWI_I@J6p95Zx( z;;nqee5J?HpfrhV0;AF@fd?awz$d=xEf+hZu10o2t=J(SCH{tAHXIJ>X)^(o z^D7(NJn974wb{^WD_PWPqVf4{iet)@8IvcL+U(fDh~(q0YxV25oOQ31{5oyZ?NEjz zU~^3FtLY9F*U+=uH9e>|kSM|PH{cd>%lEfVfD%Gvd8ps|l7P5dv^BB!ewdRipB2pk zl1awWNr*PfYBgQ zjGJs=li+|u@f#HJ0=+-t6dr9C5SbN8XA#~p374n8KlLBazG&&6&iWQ$bHiT6yjHgH z^k4Nyo?%Y&6}ViOIl_}T@Dlx{IAi)K!~*&}v4pO5=BI*Q5g*8kkoHp#S`Jc5QP1%` zUi~Qb)~qqzqZP%M^0aBhl#R$t0Lr`IknF;iX~b6SqBNq&a|W0L)C%CHqX{vy0ROui z&pa?iH18H_zY1j?*=X>jN9ER{oiH4b21(c8a2muJczC{nd{EjM@-J79&gWJDHWTiH# z1W$T9E!b_;1h~vZx@i{OR+}W#X-7EU%}vOaMwJ=&J(L8LmvNiK%5q{<(im&RG8E2` z2CejX`-=(u-yKw^;(r)b5GOZ~p7h6?F3w0Zb$gdf{Y)|$X zjMNruCTc8#Z41vsx#gRR<+*;XzxHFS%%jeLsj_?QvlY`+9+-+k`{L>Cf?=xzo;ouuvAPx!aHB9VeZV+XG`Z@@nSEi?0I^`k>KHz)4Q^8BV8SK$ z+Q*kg{9746UqFUtNvxidDl(JgDd_mO<3h`P-7@tr;I7t1tI;}u_dtLIQMdXk$!3D6zer@MJASCAuYI!+qFDW*#&fn}G$T>QC#g@jr@5 zxebJ4!R?p7g)=MvD;FC@K(vpu8r;5*us>Fy{8ImaRouKw7{tcU&O;6nUwBk()9WMD z!fa}cE_LU!eIDta`sS+AJAw1}{>CX|^q2G9@h^M(oGCVTXBk7ZibWAzvV{VCi_Wg3 z@=K2nBAjV(D>!&9B2DzVNP=0J(6c^q@u^GM=!*%CQ20ygc5dt}U#g(upDgK}bN3Ey zdM1Rv@WP#q3+Z~RJyCA9T1h4T180hv{N;wT$y=kh^**A0yLeeBTfL5;+ zp}*JYiAPR?@VO|qg{bihf;FKS1y2w5L&c>41uOZ`dy=NWuf>i5f5BW2YCYR$pudyyGG zTc%Mk)uep!S(|kyg?E$EO*&+a8ui@2CxG~@VM;Yc)n~%}u&W+@*sV$*QV3+D1o|8% zm!{H9Vro=q6=B$+yGTVCA5|33I1vvJrD71|axTDXTd%&gFVs;>*wD$J7y)SCZ}ss* zv2?Zin+R=E=#kDD;itiOqkg9)nT*|8LyYfeDkjmGvTcZ{oK}{eq#|t7tFTT;4+Lhk zK_%v-dIV>k3?fkB=LSf+$Wo~18fgrfUXgM@TFmxk* zQo<`2rt{@-kl)#+;IP$>NBgm#5K=(#YdS2@gBLk0X{$I;cGU0no6sb2PdA0Kbh2?W z8olzk)9SCEzUt2pf<}&um48I4(cxXyq;3a&53?mP7LFJG*Ay`~XI1Ix;_@?}(6?YD z_F_|z-Lq^t{byRGG+X~!@P{GJP|XinlVNn*bXzLp@p7;z$j5}l((~Y7PgBV+UIP?R z@l%iJZa_QH6XG<6xNuIVI!cS2bxfZRNBk(G8t4sy09_iTqdAC zzt5oep!YmngQ{tbw?Nx@=VZG3Q~lCYbEyn-zyY+@U&%k`)7LIe{GhJ;DtP7Pb)`Q) z2|uH}uk*oiqVNID_gb_5`CuJQbcFRME1DBIEF*rK2CSO0li!8bcRr5Q|;=hx^v(zYqgF+>K)ja9~d zsXNI=Eb>m8WJ)w+2@m<~41nHqZ{k3ku+YywRRHI22$tWKvT20LLaVAs$6r&+;Z-z} za&LhlQ64NJgg8aNqe}oW`FLk1@G)&vnYAS_{r=~O!&RH*FQ^@REH;$5)qcjL)FS-PKs>B`8QIVkkQ{3@ z?)d#^EMJTRpG;99++U`>lgZ^|SS4h?N0UjH%D z|J{t&tKfCE`7Gy?=uxG=wClA138b=x!`}C^o1hRGf2;Gpp^bB3T6?qOV1jd>I`)VYWB|3zm31qmu6{0v^{D_gb{i3_&C*Q4NBB_ zYmK^jd*dz4hU5>{P?9WifX7^Yi*}qgerX-+wk4N;5iVQ7`jc&|(w6Is8SOH>Cf}BI zvA5;*pb5|4xOjZ)Wn3pwWd6c!z7yQGlP>$Y2Uy5;*b0KFXqWl@4szo!UG-L#xG2!@ zbK&ruQoX(exqVtzy^4rhyONi{?@j|j++nC=mms}838-WTl7cK&6^*@dQRQ*!>2z^H z@PnkP;_=zM&yp7$-Wpj|)iudou+%gvV?E(1hKbmA67u3jKg^RPpm@QJ(VHsFWYbcg zvZAZrD8z{vb2fN??{*38Cz{bo(-FGHCfi;S=Zwl4LpqRkS=C1m({@>GMz*#tO(Qa~_;4GHsu;?~4YN68&|!t!EWSKiGGlt(Cp< zzwO`uXjN&n+i1Id*}vcai)PN9lI;j^PlLtaM^$7WBPJV&n;?nFJi%8eXMEvqK=j%a z5FqnwGQ@G0{kH9v8;mB32-BdvbeX2tzi2Clwp3}+e$B6*gU_|N!#OL!J;&8`B6vSr zfio;gq`TuGOIzeE9m(~pwW1=qLp^h~Ru(Pf-8eoyCXt0_orr=&gM~v<_yy&c`K~NY z{`FBgYYiWZiTwArW6j;v?r(#|&oMbPwCY8?Y_~aS_e){sg=WqqMjqK!%$P*}6n1hi z!h)tf@wqH=FrAv#*J|V5O)Zrb#Z&f2s;VhYvtP*c$YS|4Ot@WRszF7zb*iaWnTF&Q z@}h&KRd#aOa|?B+TIRl`WvJzZYvr3lBm^Bzy2>=O1AGHb&@aQNIN>;DVHDND9byaGLI;NMZi!T4vP5z|@Lev?*P>Aob z+rqcXF;v=xf?kCkbDg!;llR2trwC}EL*6YDa6O&<_`CBpD}uy#JB;@B_qrfA*EYZw z<$H}@_fu7M9qITIt-;HLTD5xKP>G?|ZfV|sIa>bjPa#7s6yC#4<`;-tP*h%GX}YQ< z)t!EsOMmT;sjqVV&z&@dmuoZNQtf|KFg4>ZNw7`sm4eyxBIbYHyQ{8i3nnlB%e_%! z@xA~Adz9I>E1hqX4_jw^IO%pM0?FqH6*X^pICV$Z-{_YSK9rFq zwo+*tF`DXML>p4K-_N*raWGKaLi}ph@W# z1c`cC-k7)WtSp?KGh>s=%x9b!GR>Mks>ts{po`GlG^UUushDT@S%_14ni{z2aDUbm z&|rUaH&s<^^95EvZ*tW}q|9(tR(G`tI#!y?k@H5dG6bJsq(W35!Z}&#o+9kRklPVF zx*JQ+;j7ewC2H>TQcM{RSPwp{xA&-HK&8%+8x)ed8*y%%Kt&n}*h2fBUrDISdy6nEQW6li1S`^Li&}=!=-Pe6@%>^G99{q1yNl8f@m8_E? zlm)(blLntGJQ{VnFK7a{Xu#BqjqT?V&ZmcmDcmRDZSbxOkUQTI!MD2qI=E(`c@2;? z)xi_2x}z?ozvn_Y+z|N`D1Y5-oM%BT@%S4c)J}@{z}c@rFcTAa{eFHhsCMuZhe%O2 z@kL=90(i?%Ij(W!McX;)Yb#N-nki($V0! z<_b zHt;a9T$CBH67p>>pD zB4@Hv!%$PRVXM4Uc^uvH6|S@P@9k(whs9MnTV()~QuL_mvB%(qsD6)84L5S+_s1O| z!~7g#Vu_c~ge~Ho?HKr;XK+Bet_9irBA2pqIx#a?!|Mw_IP}8}PApW)C>y+0v1}xS z2@#;3tFCiPG9o_9e}`f7gY4r#JlIN@k-}=t&_kMh@*cECJ6u*)mhW|p(<0|+XkudGnd-vb14eL-$7%0-!n3os z1c6YW^Ixalf2>5G|Dc=*?{B6Q*)NUn!Unvk&x@l=8edWbt9AaLNJ~rYG&uv;$qwTQ z{pQ;*<>zY=YO~sOzmebm7A6w8yxRn(5=bTk zj5#hqlLDx~Us;ExHuAJWUHitzYTBeNH{2#IHPsKfDlbXnC9{n*igtOb)~!dTD=yw7 z<%88PvKY)2pU7a5ufbsl6RgBb$Ca&ySnw$BSJfsrqmwDEDp%7}jlIx7VCj8MO2~4w z0=xwxSRTVJWYqD*8Nm|_sHPLCOYf7}BkRY>2%Mf@#14ySdlpOa5vlt#QM8E@&Md|| zO^wGF$$jNfD~>g7oHaDeOVGClzn@ZM63#}POl#s)LV$HHdFm3}*oC_rlxc$)XE52e zX1A82aF@w!vmL{Slwq0Bj;~}E*X_@>aL?D+)BgNoqPMPja)}S46T|FWJ{;;(!4!i7 zl!Bw--14GajXmaB?el+0xIpll4g*r~TpFRgRH39Ow0IIq8*hN&#rYO-nh@B%&^{@S$@S$u{x9b7_Z*gb*m?r49W> zKT^Ke=}~Fs7px-+WOJ%i&JVpWR3Ea`sA#kO1z}F2;KXv5O&C;NE6))(^ddn99~AgJ z5b*!JFZjAm2#_t9Szmh_jR2JndAf86cmwJV`R@g7drtVSZ~eY5D7AzEM$LV9Y4y8j z%e6W#h!>`~ww9LA#gq4*!GFsB`*~#vo{t4h#~55LIa}+|_5MiO=_^!7_!ho?pIU0; zbNN*}u@y61=fYgVQN;h(=zj7`CXPV6fd&|>EBDROuvRrhli~7RrLpK)i#JKpdF*|k z@pIqc{g=U5{bz;>ntcc|8;XVg@G^|G2(F|<+(BK9Y8%1#VqV>{8u)W$QC@E9Bm}4k zUga$c&oWN#L?sBTh6PqmrLMk<)b1DwiBUr`{0$lkaX{fJ7WoxSBU*5RUd}6=IeRqz zX|5s{1*lGiIPf&2l8_G5ww3R6pY{MoeGY^Uh%467jhjSv>Qk0O-PxqzWJN5NgZ+7? zmZ6UPVKVyKHJO0`Kh#d#E=7d~UT-f87wn95r*-`DySUB@OWBgpl#Ti^V)mA7A_!{` z;EAEZPGYE>T)RYxC|$+jexI*|_w#C2FdLE0+nlZncU%MuUp3C?=y6=t{#RocNrm=- zN^6w1^$W1L&$>ZFi(YLvw=A1Ck>g2x&|iWeaeQrM%zWhgYX=lU=t*N$0-s{2DXrH8 zZPJ|V`Jcd0Q(Lq?fR9B=zLlcDI72f!C=TQ>Om5^B#Y)rK$wzvW4cVMdk*d{VwG;ji z%Je+eijalAHTm36(>aCD!Unis>e$ceORP-TJnjdlM$Qi2VZqi(m`u}FMZxd>q?3rF z&M#XMecUsQ5AM~%olzSdlx`(u8M%l((fJV;5AZTB$$T&GJtk7-gpWR=B&JcAh^H>f z6819Txt?v?K+MPsyv| zYM!+o2lK`7=4Cd5{~=1jcYiDi(Aiwh^8Aa@|CVR_7E$gd7nZ2k;J+f2!22K2_fU}8 zp^H~u*!3z(1!!y)b7_pcB-|?hJW5z(6;Ks;({KI6m3nAi)b9E)I#9XRK-#X=eiRG6 zgpWOoW}gJywbQ9I0@GG^Lx|pqSI)tRo(HBdbn*j?Kq5u%n~g%r@m@7}olJf8Y%lh- zz+*%PdRcktXQ(p20dF=BQaRT~IaaAOZ;^-YVM!-AZ4MOb7ur&T?FS84ILt5AnRH?^ z!j}CwijTvzuNMu_LFHrg&n9JrORk%ZRJm+O##5I5rb{V;(d(^!s37Q?@L4 z?^itb2Uf-jCMO z7wE8I+O=&yK=Pi)T~<~$*B7>+aC}^9SP+T-dAH!}rQmC7=6&jJmwQZfbViiWa~CAJ z*^QrBUfggf9Dxx?g*$X#wxuAQ@%?|;&ubv!9G)Gp7>L?31OALdu&^*Gr&?IGLgHH->_#^! zn|=i?C!W)+!LQO_9;(b|{RLp3|7yL6zQjD3ZQKtZAIg;qTxe3FhN1sSc;T>!e3VE@ z50;Z=&O6p+3>P8Dd|f(0p`_!Yk&TWYAV@A|_L599EjBHZ%zQr1M5#D{r{sO2sVaaj zF%v|Box3Yj&ZD&xugcWg%+x54F&Vxo|x(#B3%aK@2^|3OWmVDmI+-y7YR;V|GcJG|pJ5`)A`>8K5i5HQA0cU~8%d39#DLf@g8 zi#N6R<0__6|L(#cmmF{pMgH*iRF}-#G)vyGG1D;9)A)-9@JUOMB{LBu+t4I@`7N^D zz@zm zwDOjnk75pl$c(_)=o>p{t;3G?mWV5pmcm?VS0~AvsI1d3}%mwkuu{*akE=rzK}{bXnv_Di`q)O6|BfIHBhl4b_*92sz|eyN9SDwfk+{1 zkAn8=@a-V{nd+q8Mqz8{ui``tIOC2ZaPCU{&f>K#!C+b>^ONyVost7Mvmzt5wX2{V zs}o{l-;W(WkK61v#sat|5i0W9dyeS)Z{YgBDJjvURg?j=xTc?mBoPo1>kRtyOCq%m zqxx>b7=3SJ40v5n6;F6i{Uu}8SK1H1tU^_GfEUag6f@Kgw^!(cRWiuzoVgI-;R zl-$d3Jo{7pvbX>7^jZTnY(f(@`!K!LvdtBveeyW&F_QI56RsC>?9}WQq#MfA_fAv& ztv>z~gQ7FU>{e}T0vF=5zsqK~`DC(5Mf`jo75F?;2=M-WK|882+Z!excp@XdKGhZ? zL1T1AO1%s(R-0Yt=Tj>0D_aQ6r)IemUj_S~2!r}9BmTl@f_Hw@N4zo8lN8&ABS-xF zuRBJ`n0)d%bLGM}EnDU0aASrjdVHDGjyjQZCe~OLReUStL>{0FHMu}r|GYb#zVXg2 zy%VYks&)zeN;G{kB)sQ-Ds}f0zBL7BfbdBuy3|tlr~ZQG#zuFs<_+mD-$fmiGSU4H zirnSfa1q|Cc!vTxr0*bCL*&zCR&U~gJ&^TH+;>reF#8ZWDbOvY$4JNb#VP&kF^4?2 zu`tfT-)cO3_8@&SgXX`B8=F~;mz7Ub9xCFI93MhRq?4Q6D`Rks zY^lp+;(3yV>J)Doz0^o$vpF^sphk3=B_ZJ0%O;U&7F_&B3BTCH<6w@{#SIB{X|sW; zm2fM_TABgol_W0O8XM3kCLSgz^E>~61wReBliGVXztS8rO_ibQBGBS#rY1^vbJCEM zyMp{Z;l%p`h}do@5bMC+WD*Zk_uSEDe;i#{vC4BN{FBPPLl<`bBiruB00~pZ9v%1X zwrh7#0XI+rk17HXA}KQN$F5KlP*)M*;wR}k_FiurJD-gSezEBFz{zRm>)^n6ftbRy zms`$QoA>ov|25Gc`MtoQ%?=n5an*b6cTKB*3oAGVLpUJXv9Pn#&)cJ~y&|bo{Xd*! zbfA-QyHeijnPh?hw<)j1s$Qy$&~K}9^{ib!j1z|$hVvMImSXg)`AXmS@+gWr`-o>_O*?65^<(vah5fqt!Ym(_dqRf8wE4&(N^_j-9h)J_#)> z{XmsNnRFMt9GFD>mn1M=D?-*1j$cxhG2|(;$X#?iIqdOFaDRBvyT8NV(riwZ|HP;Y z-Y_~V^35v=E^EWk9Lsh^g+4f$7z`De$q>dVSek-%%7j$)JM%;VGxdVHA$CqD`NMru z(@7v#eD5s6l!!MoG!t7|2ZRpA1e%o2=t{V*@b8I}q8aMTzcsWyN^QmXL6kNmA0%1? zD{;NexRToO2I`o_%7e>_^7iZFA{KdGm^q!S*e1x+qGM-RfDS~skU~?ijsjlw1 z(_o{qp|sA2=7E%GV~1%tdDxrP+zq=@2!w|UQPkMS`B8iDyW%C-s%;58S{=hGtsM0W zgo3A-q@xEmoeN z7ycO({uw~?eXd>(DFw!7M7OhA%@WIEp-|Qhq_0I$6t>SGxwzi?{`Ou2w|-`^sPzPB z$o4l@lShO95HtM~JpCu?=cY%{uoH#%IJ&iJp0c0Tz!p5=43~4ePB$5ke%u^g`mB|a zz3;;zCVrO$IGVp-;%dq(2^wPxWVE`}O#?frih1FDs_Cy4WeATjL(k-Pd_y~-9XYgD!<%2mU~}4mtfiYN5K$i0u%N>r&N~MB^McB_5MerY*{r#;G8ELC=NhB^qqsl&Cs;;dwNSS|75r?%}A11GH1vut96(WM{Me=HXwXWn#4`? zH>s(@l?I`aN6p^nc3ic(bDb)8$XXGDyo9EK?@#V^pauTxVp|i5HB?SGYX22BODS_Y zEEW(aiA}NhsS7H6#c*Th8c%9tHbFHpu*yxluw4L%i-fwEt}%>Ft!z}TBKzT#O;)=a z3d>ACDyAQj3iW|p?V>(%7m2mj>FRuCzE1DnDHZK}DbL=w1_jl&&eWrPp;DO4)?~hKvlSRPOx=lSFlrdyH)X@aq_D%RC zLwD>tucgOqE1g`Zd_%UyWb+Ka&*2!#Sk6-rFthkIp699q7+gGVwz4f{;q{eLvY@i{ ze>5`xS)jlMlnZOxgDb4JPn_bYqh~=kTcb5^>Q%@U2jv)4sOu$R7#L6ie*0|EbCzw< z#x7rF-CdmU?LlM{6KFf!RDT_@iYWfpI}ntMldHT zQv?+1#aMk6JKWQL-%-|x75v}m;B%M+X=2q<-r1ZEQ;cXYon@;} z!Q+)4dpjJ)&dwNBCo�Xok}2g#8>(wo6PTqL$w%Xxj{(G!tEIn8729vdgxZHo+#g z_Eg>1^?tf1#>#fUj@!DInCa=$)VU4fPc9V2GM@&9>UlSYwR(vLpYWq0@Pl``UVF!8?s1R ziYN0x(M+~XbFsmjYVZm=wjF5PuPsgICEN1{^f7<~bIt%3OJ+uZFWhP6-f^yY=4K*l z0XDr2pl{E4Ypz?68V2g;$h*!J{fXQHp%&Dl@uWM!s14g6D_m1k?8aH>3d>y9X?ZiV z_;sKC++fh0MaM)oBc)tomL{=yH050>HXYMk6Y5p!Q+zG_mEou zVrV&?Fa0tjxyu_=Oa8PxYh%~It*%#=_$wV>A~p~GQsku>@JPyhvY!=18d0XwuE!gC z!o%85co;&ZM^?(%=n%XmOdbhM(Hado?(y7Wu(or*9ES&|;P6p@Wxq!+n*1|q_vv^} z2al)zX%s5igiu#Txdm-9$_N3s?K_Q;SNn;jT<7#s4uZ?(A`D*zQ_SHFlT=28p*mM2c%5p_>Z?KZ#fQb`>}Nx2e`mkvUDbve7E)kG zTQ=DuVU@mCNs#Qz0!>E_u6St+kHio$wFhhF#YjW>?UDzZu+!-E@qP(;QYUKQY%6@Bc*_zaT{#|TY``ayEB_C=mP8>`kvp^>j-Tg!a zE#Dc(?|-YG|I;!6TzQ9~={SQW3iW+{+|f!C=+ZTyk-E~JH{p>#&MMnim?r-*Fz5fQ zrAzLzhg> zE$_#E64wVx)W(>7dZ>udBW#I#t=IG2`2uY$C9m}fnICCC$!_~a}K(Nq(xc})j@ zv482CF=aPNO6r7?2N7{6jEr~iRl}4(3>|odv6U%`iS#tmZFo=}${+I}YM^?u4p|jX zf?%=&;ib)D$ewsS+6YwcL|RPhH{m7QUyoflDwxDblxks_snzp#T0v1&5)72*`1V$N{j0GEl(t!V_L~PwXaMV0GH&9>NXjT2dXbKaRO7<>^t@#>nNyh!68G5zMez`Ee z@wjwXR{Ix+GYbLJL>lYc7Gs63&-0X${Fu>!7^5ipEq7+cysumwHDVk*oMlin1lfb^`xE4^a(VSXBEF_bENY1%1Y($ zec6}RIXX8u=8ZXRr@x_#NexnQVOU^5EqG2Y-}PM2?bBFn16M}+^wPYxD;pd)A|`Y z8dn1XO#|skug~2tL(LQ$mT=okZxwQq_+`>mU%BD2NCmb6LWSo-nuXC_n6>55a!_K= zNUK_vmo)H1^FF@yg;gK`t4^pie3mA@R~duYGrF{F@{(f=+WdQ&ZDJ$BEGu=a9&)Rk zix}=TizY&P5MU}@ArfXSw@C_WJm%1+TfI^!?U|;m81;_6>|xh=_lyoNxSW=@i$XWW zR31&|7vYNw2wo2ytM=N=OiF{k1G!adrg+Ix;hZx*-YChge>of3Ps=j>p8F(L2K2*>o@YNvXUGfB#GqHys4hg=`MjiCqbm?@s#p1Tl97 zI{v9-13iiSHuu%o$PAs9t272|c6tI{)%<}44d9zIPTMYgaghS5qd?5b>0*!r*J}S6 z1x6Kw+Q^7*m)=kj8DD`_WRqtGp(vft!smOfmn(K=>iLe&B4TU0|ZeLq=glQ z7=8EPm^&_iuDNE0a%-XEQz?crP9ZJ6xP|@>%G<|n96ri0@SkR131qFo*!8_~De!r} z9d!cF^@;$W2cKE{RyX?pO~HRR``s>zm>_jHRbIC7De3#USdiyWMj70RbCx^Cylpx zNH4V?d{3!j$IQ+ax4Sd^gKcSQ_2VmQRkm{QTuVunAH7n0W?>Vex;a%E^|hXycAFhd zn_=sqQt;kbwnK7WDK{UoJTry7rP7GGsUv5k2H6|%<2-0MQF2O{6qF%4G)-tn$-itA z*~LQxQJz9!8Im0l6LNlH1Arb+Xz(tG31Z3<2a z(lEJX+(8cARIz!kAM77(H;P_q$18v4MeUX=D(U56or!niHfRQ8t*#~-XB=LXIcDF zxL`txUvZCn@D!#X%^(5jfv>NM!m>YO*e_3u(AKTaE~&W}2ckP91JN_$L>_NQSURi& z*5QhMu;#II7*YI2nIBb7&XyNvP(8GPFhv;AC&D}ge1bceKK-zGPQJozOE~=2%LTa zmn;64ha$jg2l(SHzrS0Fq>4OD_>>9UzRyg`*sC@FNI*pZRnYe9a=S2KFtsHkB-U%C zdz;*VHy{%1t$e*;wMq@Z36*3{bm);Z&<%}>$?mDg2KCgp^#FbRo2j>}fE_|1kg zd;#H^o3D`fw?0@%gDKmdK5zXk8N7Wb=igA!KEd+Ezl9kj16PsHKb^lzXL7JWMH|V* z^Yc>vl7r?K;y0E(dY2aMr}R7h8ko+|Q}??%Yxq8Q(xOf_Eu>aItYU`f!jW>w_$nG# z1|}iRo|aQZqO)X;B*(9DP8zX#6e%37bl#$zlBz3jBsNZQtu|}T(S`g{h>l-4t5GC0 zp3-tOcqM9Ly&|&rDN$Q_LSr3~T4w=+`QymusHpsXOHsendzzXkC~3jAa(8!e6?fw! z3IVTPr?;^p@|9AY^Z^b(Dhs*Y0WVw5Q{N%V$SD0Vnw3=DQS4eX@=~4EnuMxRDNh#c z*SWsN4;#JjeLIrz``{0-u3{>+(F>2^b0(5q^-c?$XB{c&o%5|6z(D&oKAK@KE`;2~ zcAQKTGgB z6JfuC`hwz1{Nr#a74aZ8t{dTcCm5>1B|5^9d2#YUMQ3G7%pzEU-$rHXHvAcm&`JwDq=Uqf1bJ6c?;l-WY*WLhg2(kyD)gL1Y`zct}mt z2;@WeAl8#)AM=CZXX_Ms4iGf;;x8O)3a+^4@OpXku52<|xy30jTK$M!gCv5jj-|j6F*>vkQdiF8$oJo%h+!_aJ(ZtQe#f25X!tIu=e;*~f z%n|1rCiwqp_lK*q`E8WPlO$f$@w$)I5ADkIk{5A&VFG!Grc1yj0&x3O!D0Ljb;6yo zz{CFk+kpNf$3WDO80U6p}RK#H!3N?=jNhE+oQD3k(A}iN;y% z&?bqinvO_?)*89{fzcFujk4(fHw&P+eacbKX|SjywQ@-t7TbDoqu&; zVZ(lB;yZM~!?PpgUB&wKi}(@O)Ye3Q4-9aHIdFuB49gfV6Xdo#aC&w9USW@GW8j5n z1!1z$sMl}lL)aP;x`h&B?o-*$4l08Wa%YKRcfA-dX#S6r-~WpPccf(zNn{O$yZeyN zM0kY2edH`1!pH?q4m{NZp!|f(Q_8tNc<~{iPE-ErOvJ{LZKUs1h)KSCMY8^8wfUAS zP5HAyB2th4E-7K)*0ql1PYpMxk-@*Vd6HqFf?Dw>KUkHxK zoS;R==GDrlvsw8=-?v?$-DGXX|B3BE$H&(KYLv+Pl!WCN#1=1?o8XLB=I{UTRs0AW za3=GN~b&qV`qjydehm zz9m<;Xvj~tV2%mpMgH|-3LN98Vjb7de1EOazP&iyZ{~t!N-#{t`rPo(4eL@!ceMOu zt8gYHfPS?UI%EiEL&@g}>a1}Tb{vg`L}l=ni97E-53%0&>>>}^msYkn*a3=JbWm+^ zc-!?wun111VJV5u9Op-vFQtVC^vGoVy+h3aoC`b{edM7+i2a>v<&Koio+)cQDQpmJ zGs6Wmlni-`^ci!ty2hhtN0PoeLb%S}1QmTwMNIH`uv4s5q_DRlV*^hjf-!}6UJac> zenC6|v6@GDRg{n*!*r;DZX9tsBN%vd zZo$C(EHU;*hIh1i{JWnBNJs@eurA8rr$7KkK;S-WCh%n(?2h~<3{fWViO$p!xoBye z^8A00Xbm(85yvnw;5nG%GefRav}ROqB~-0%zEoO3>c97wmDKs*csVe9N@rMyIeT=S zQ5=lFVdn3WtiyPzUv;VG|8Zjjbs=FQ&=&BQbcJRkU8;2cF%=WkH`q8uzooYnm(wP9 z;evP>lcR($wa+}azy!t|0C^lIcV)+s6yV6FvWE>9aUs9v%;Ijg)W+o)?ceY*e(5Nm~jX~yXe6Vd37iUmqK~XR-t>H$K zr(DA%VcBr4rNUvrtnb~Ao)!V)?RmF_h8cs`Me8@W?KK2XfYxDYmxJp9{R?pKNZnA< z8xn8E6h#s95;aZvN!U^|(d(bSUmk+;7N0I3>ke!?x_ZXWAZc{nkgl|ul1{UN(cc<7*R#$RCCQWf+I*>35 z38z8}1-q<>B2&Q!MlYP)49y;UL~b?b=A&Aqkld`~2LvXpcz2=( zkLluuFaSYQA|=UkFe>};84k*=d2aYKB=B$FQN@4iyApnX<0tWt43*2_7@V!q?|qG> zY+9b3MgHYnSm^ETy-QYzP|w#GT;>8dXFBIh_pVuNp685+R{I0$wP1DDGRJiTNwP$0e_PVU)0` zju{mSB$S3-2${Gp7y;CpWpSF+&^*uoDJK);mzZ)lkYW#i6-Q<|jPyB7k|2&N)M;qr z!+?DM2VTM%JpVNKJTmybq;x6bDj3^h3)F7-;4F6*@P|M_K~YByn}OQdC-D9yh@b5c zZ_douD*q$*-^q5vHR1+d7(%rRWxP6N0R~2Ht8MIMCm5#_-T&z&O5YrP+&aN4YhOKn7Qw!QsK zFS7kW_p8m_xHWZ+_R1&t-raFq#7MheJH8#~PdYNu;oSZjPz@0i=T|RZnBvGjY?{(q zovZqI&ezE>oBD41t@T!PFd3G;GAC914A=)F-Y9wk#3C+%8z#5TwRow0HvgR)HRJn591e2ks;EoMTaA-$R zKqWe$-tvdjtvXTb5y8LXGZp6?4s zIR#a|-T`UeYLiuMPXnr-7wTPf+GPdHQ2=VI%W#XWrRWs151dtXyebqwTpRrF$uP1Z zS>vqa%IT+&4#uY9B$M;G@I2##VRH|3i(*r>Ftzxul_mpT3GP=asCe97=mfFuvo38m0;14 zXXG+BNzu(8}V}&CZfp9P)hapLhYNglC+csx)DY7q^ywG{im0P{f zW3>T7`&`7T67?8;Bw9z@zs$3}2qKCN?BvuGKE52j28tD+*pg@Glj5n@xg!^g8Se|XaD~o!1|Ir^VkIVeFmKOQecuZzHj}U5QjW-{ zSs`>CQo@Ezi_e_1KPaq!FKdz+FFN6sz6OU1Q?YF4wfxP)BQ-j5P@r}Sh2x~6_-o?m zjj{wIgm%Tgqss<6O&a<-_;V<1>nyjNZ%>yX>MKv8WIHa*=_lz;Kb=c9?NWg4A?Y+( z>^xZ%fvZw-XtSj|#!%IcBAAL*V>s%lq@k{C2{IJNqVRVdHq})JMV3)&9Q#-v%Q8v! zbR|g&SJ2Y5%d)Z@?XX!#ywjlb+tW<9QhSbA=}9V5t7yqr3RQx&sc84}c)76FY@)yH zdO51xNw5Hb*bim8boI=|t;Wg;u!v&-2g?A~<+APO9qk0rxm^)@M?0Ci@LpUAE2c)r zA)|kh;|8El_e&|XO+}x85U?W`$_!5MtF zpaRX3D8mf@^fWskR~{uEXRZ;I#KpuWmhjI&Bu!eK6P(Xt_*|9aMlj4N$Xm!A%xHnz z_FyEg|3Ri79NlBv>txGg2cdD1zsv29`m!HTUHF^a@;Wy&^SK*RAr%=Nj@oL}4r2`nd=n5T0rAyD1&3_FE`H!e7GwB;-@^$AD?N+5Y zTQ!ih39MqNn+8MuM^v$*>!T(3{>1GYXCUroW^@K|Prv0_(rA4jOka=XzYTCaR&*uY zoUtUNWS?&^@^Kn`ikr^)GYz(M95nK*G|GQW<|aJ^eIR^*E_2B8o;GFVl@kT~8buSX zlOp1zx@AvJ&;IBMvK;AxgMfg>Q4D6lhTCmi%~MC@gr`HUqS1YHd`#k+#ix@LXe53w zXv}7;=Fk!LbE5T6-9^{Cv)ToKxN^3#kkZ)q>Fde#f3fwAVVQ+p+i9|G+qOMTc2iBZ zZB4dovW=-}vaQLsZM(ib&x`ND`|GY>_fh-W>so7_!YW9_gRTk`!?f5`>_C2K#HSfB zY7Y8Cj{#Nnmw&iSPu*(a!gXr6=9nJ7#=cX$<}83n;YwK+4$fvyORE85>z0YAv560u zM`0HmtvqHZ$3S5YJClNoL~0;J)t2M_#8Kg}qjgoe6-;_XNso@)`kJ(qw}|+)GPg9n z8phUC2`2zn6#Tk}Krwl4V{UtWa5z(2d;VI0#SWGku%jgC%+ynbFjE{BexWK%S-DknVQrWEknp?Pf21vqw*# zezM}NY!nwAg^W}*)Ualzvz|x)W#qR)b$FV=>h3G*Q|9J@9z@vszpA$TsborE zX_v5mhG6Foz71rs=RD7rn4&PB8I&A>e1e>mXM&!=D2EgH3 zd{zDCibScgZYX%;k1No`?0U7a^Hwnjr&qOVqUvu3DYHYhQ7HJKgu#gHAxWO-e=#Oh z)W*Lx8>RL9Q*twbH1389Ra0oan9M=EoM+jW>JI_14;6`kIHqrG+8Qxb{RtpCv4tI1 z?pWYFaZdcK#6~>VpI71qNWX?37jJ*w;h94oi%yO;1Nm}==%Pq*SI(d*OiQppazrcG(mNi`T$+!{#v!|J?RTqgkxWjEGb zByyQcBjOZ?1wp<;_FwoXN>G)MB^z{6i$grwd^?Y|s$K<6(0`3a5`JYIK|<;|vIWh@zxZgS* z9EUIFkQ8n%tvVPpKXu%_TzA!wWmO$;e52Z27@;g7rCfcr@}!P@L4>#BRNCxU$;U}c zog$5mEqu+5Zr@Xnt_ofpYz=NL_9>h;E^ab|c@Aacwy|>K@mZrHrzrpyl_Bx=dqTck z$^=N<3uoZDXjk3=-~fmTLH+jqGJgiw(JV`@84K*zjNFvBq$meXwC|R082mob>ueX5 zDeHg~O62J(rLnEZ>7Y{3G!gXl{pI3f6^ZI9)bBCWT>Vr`B-A?una%M;oU2sEy_;xUU> zXPT0GtZX()FU-?ylSJy4aEFV*=ejB#@S`I<&j`KEXxp{VJ(zVQ@gce0HrtfG-2@x?y>z?_DGNs_A=Wu+vAZ0Z=~yVfud|Pjn2l--JtxF5=cEpu~@hNwf+fn?>-?$_IO5CBi4YQ%Gk{0 zrnaB`%#CMu>Q^4YirK_Bt#_13i+fa&V zG>C(kh2jZ?B0&nKsKPQu4Q8}dGAOT_W0Q95Xh9U!RyRT#*Cn6N2vjMhzkTPnfcQ+t z9=I42{oNob=tvP^kyJ~y94(%Mkr@F~bYIn~$X0;$?DuIx+?bncnG{B75~!gd8YK;d zjjR$8jLf9U43G&i42>ON?D=v!$#X;9EClh}VlM8?6C;Rce^P(K>YLKh@sj7ng46>( zt8C+nJ%t%=lsuWT(wFJVrx>N7^M<5dDNqIHm4XDsHmyb0udSU=xmxyZ2?v<`=QrFi zL$OJ}MLTVA>T8(DW>m=(6BFZt5=LXw*!ZMhB|U}ERSpZmCY-&zTw~5<0dzI8^787m$um@z z1xav4>et@3JWew_!d{0<0r|IPz$ybqI9 z40!oDbp~$NXaQi)yLq96qW!E^5)h@z5s;gMb~F*GeTkcS&b zDp3d{ZATJJ9(0>+-O$E+^Re1&XJ&SAX6wq;nlX0X%x`+J%JZ=5)XDa;zVE%kQIGh`w~%Avrpn0_rpITDifL^{g&4bY}g`CW77H)h0C)4 zGNRjV#;uSt4)wk~fA=RQhK2_F{~t93B$IA1gtk3l;p2Tr^AWMCYYo&a$4&4m`E#!@ zD^n`U=H(Irq3toMPI}WGSq-08Ec`g!Z#lKJFfkx>Anc9AlHGU1?`sY+9AtNRsbKq{ znNc~z$9s6j7z^oNPy3ZPT8zc%zqa0Hg*=gX%~Y5q&22H@s-K4KAMOUD}tV_`%7wgQos=HUa7!Uy|#OV)}b=Ub2f>zTe(3yV|Z`GE(_ZQjbw?H@nP&%Q-Dk0Gz`rvNXZ-uB5TkQ{vWu;B~cW$XdAC~E? z{71&K!vm2@G8IAY!Yejje zmUCsKm7#j|w6SFP)On~yBIe4tCEVE0)!Da$$Y}U7f`lR^Eu9dCuL48mH+;GR*oFv7 zB1GuX<&(P3x*$x}9;tJk0=|0{m=Vs{N_V?fT(~%r9Qcyy&WmVBIap zwI;3)JN|r+%s(^<{qFCa#%30y@d$MvycuiI8}eiWeM=+3=D0i5rN5a}&TQ9q72+4y zm8&ApQ)Dyd>+x)z_MI!J`7doI;ET3Li~E(KE5T5;Wdht_9>;1BrXn&ZQHP_ho<|DgGP`RVMO}JuK=|FR^F(X$41p#LCpOZ(Pp5sbs>Mo!mA=^bktWU6*U-dd*h*!eqC!NTI`=kbw{9t*gU%r9}USNYO~?3UUN z$&pl*O%klk)ZaDv(U5{kmCRuis?&x<%?U^?a)Z!~#Yl}po#d%atbRj})>bB=87O&M zTw>UxvPfZ|C?k0b{XM(tihA<|CD}64q*$$}q zbA}AjXW8X6$q%%9Vxfo>%tZ*c{^X89I8)1-aKS-CzhYpo`~6dmHL zq`z6b#wY@p%MRJ`h`s&W0;6`FMbqXi_2QBdTO6wI}8aWUde#jUAU&T9mv z9zO{(cyxe3_9+rE1}Ey_jkTrX&B3!o3SN=JlMYNrZHjR%J5xt|O~p&5TgkKBTPFB=M!g?=aLm0fpxDsKf371N9;a0&@NVT12dLuSmk z!1fKILE5<{iWyO63RT23kHYUJ_avlQb(V&0{Jg>6BU||!P3G6j_Q>=#D0n0kVn0G^ zEvdPrKusJiMAJ9DMCE9{sW^!9GQ#5&Qw*%`$ywfJ^FnhKZmp#8e#)pa z!M(a;KRX+DyL(4RuDQ7+<1=@9)noM3K2j3CQLMQ$XWZ#vLq1IWg7MuHrHnie_A(=p zks1PjHoM2P-^Bxl@%hure5p2?jo%W*iePM!Z{;a5T_Uz!;_V3}VnYz6qlmapXL8ih z21?7X>0P{?yK|^pdJ%ZwY(x3>J8=E%3u9j>Av zkZi5HE}iOpe0==ZtsutNrcke-I?9iH^1x@}cURL#(j)32+H}|x)Re4o_%Q80kzH%7 z_sacmf%f_eO-*bkv6xo*AA`jMu|yj%-=T0T1t1ixw3&BS1;g0=f(Zwfq@|V!Ohwj2 z)a%@S}yZ?>ZR>oO**8CEbF%nU}hNSMmx_GPTwC*m` z?%e5eyD*E`dIw*zo233##k7>?!$r44;J4fveG$POMn#*Ep7|Ngi*eR_AV}xI`b)_u z*zX#bo3`r^VZgSu)uQb;K0b)A({ymKYT5U1o1^_?p{1a$s?}dxJcfIc>DKGc4kGI4 z(h6`O#n!uiUw)Tm9&73yLow;gKXl8c#FYB7m3i<&ZlE;yJ*G?X!H->ec`yM;=&_VN!wkPI5^Kc9caG zV~A~>40$SPx5i$dZaH0O%vWZ=C#o;=fKD8qURhN&WsH|b8ZyH@TUy$Q8M94J_JDAd zI|wuYBMdD!-|v@_&b`{SdXKK=9{k zDXFr2czKcX9eBb*M1fMvkr)i3Ki{6m`A!ZPpcjH>NXV1xX|zs*^7(CeR>qAh(X}89 z4Dhj>h{-*OxA=Gn3FCT5Z;2#@AjxmeL~g#km2)Mn;Xe<9?6M~AjNaF*M_vLPxISqEd|s?;nWGT-9hNLUR)ss`AR-2Lp3m}ks+{V0 zJ7P%jOVH|=W2;;Y4O3+UC`i%ghysI`$Cr9T)m~1JT*9yN#sBG||F47Ti48g-vebKW zKkGPETEBpDd`G1^=K1Ecn{Cll0vUrCy{4r!oA;8DXjzlpKHBnPW;sTBcMfMcPjTFp zsJ7~I=z4JcIihXRFEsvL%^!XEUp2`qN~6&~New=)76P5pCq3dh@DjpZAL5-dffBYMYH z9kGlgTUO%O9u50P9!Fk3{n7JXM(Z};&a~4UCO&f`%pX(BJIEW28!s0om5xuYA(I)$ zp#2yaXaiA1wY!OGbQ-og9oETscBrxB5nGzMyp9%`#!IB@sCi5HqJg*LufDQBu-C`T z8?yQ{LNubE1^R2nDeG8P;$r%%3<4viq6v8%|8$ju8psXAL&{5GAK?lsaX}Gcis0mR zucWTn@!k!%lq7GNvg+FBU*L8toLkm>jc*B)x0Es`qTR0jE=7+d+4p|QEXQGI;0cOvDwLKO zoe~+L4p3)!tN)2H->u?AhMA3Dsv0?NbllLcjVD`33-zdAgJX;nVn7yFA)bx~FRTef zC*K{l>s_w5>aHrN_!TeTN8;j)5s&pBRnsO+SdYrxA6N&@j=z*)z^Sek6N2F`Y&@1( zZc_I@l(aVUd=dI}{t@hXPN>!>G6+haR^WTXVD_GIyc<~s+N--^BZSSkvfhAV zBjsY;Sw`h4Z*9}G!}%X09f>Cu=dzDZMX^H-LPWDA+Nm-XR1onTIP5^@R2Wc z+m<)ilj+GWCn-f%^OQ3umNOao_s82q17TU?opqP$fn4|m`# z@HJ!wS-%XX67u>n3PvEKY9OOjF#^>LYhZ7ROm!ZmL)NTqFYy`UB&$|9px93F4iR)o zot@3mPT{u{#6lBjtxrp*?w9NF7Ho3G`<`7+xaKI&q^Sub`y(UB%AW6JW&{dxX9jja z+1~{hFiKlh(@3W9QhVab}U8(EzMdQhfM%)5Om`>9hW@YFdQ@2o>49d7f9>w^Uy=+V30Mn+0_@ zkVh6BsFLRAA{^DlSWzl+6HvB3Q|G=T!XwL+S^jRQ7D*N_xGKezHCExbz7pq*FI_e` z?Q$rfbX`pa27hLge2LBdhLw_awFE=OgsK z?EX6<;c`T96WZ0x5Ul?gI&qnis32z#@Tc&{1x;Nze)9<+ML&VPD;y4kmJt z27AwWRSzKNw{~KM?vtF=pl)C*K$UzUVIyQs&h&ye$1v#BYwGV6j=p~W-qM8=GpKO+ zCtDz6T8=T{oWa>gBn6$rC`DNpo(g(ke0;Ck!o% zT~|C(ZDiP1k<}OZT<>za(3;hBT}7iSk{gL^?{n5B)O2Tk*(`?Dq{P~%i7W=zE4-Fk z`HRM;=fXN1mt78FnMTmvDD$v4J$V>mbcUniETrD0zM}WuG`Xt&6oEE%T8c_4Oak^( z(4#qIpkk&a6+byTfxZ;oPJeH2kr>*XQRPT{xpOIPx$>0ZTSMHI8zy5I4+)ods6{k= zQma8PPBW$}3P_gn3@jZ8Rw803sdpBBqb_A)0j9i%MbmF~c5{6fU7x!U1hIMB(zV~U zu1+>SzZlbo3wuo51;eFAkPE>Z3oKwCu!~iZ3XxpglhDYfzVfSo!>6N-i6!JLY@=6F z%OZ9{%|w2HyAQpjM5c@g4l$^d|9m=N)7&^KV9~!X*`57VR$#pvI#wStYn3!Pm8nWv z4AZ3(qGM5>FRn&D8x~NjOBI{WqRXJG+iKsvTtqi-TM5HuOm?KhZWH<|O*;KYR_3Px z`IJq(>>sMC6p?f)C{zgfD0Iq7y!NwU-8+M2D7|xCv}$T)kgADapusJNBh7%G7L6=6 z!{ab`|0oSgSpCQcx+_LeY!Bdli2D0CKOa}nOw?`ozPK=grur4!?2om+UXNnYYi<$+ zx$(R|?$Chk5}i5|JqM}U*>yY>ml*7gBm?_jzi_edahH&di)CeH9>D&$fqGskAf8ML z5sxz}(;5K*VO__P;0<6f9wK*sf_VVs_BM+(-73?)1n0P!L~b@y`4a2X%xZvND&tNt zc?uQU;iW~r@o)m*%)b%f!1JFyW=D`Pm*z{2X!p3rPy2 z>6Gr+dK2aC<%OuiYWYmTUiypvl}_RkMofztX+-p9n%W;UPin{f^zIOj?fly*22_^s*m%#9Fg`hZ zG{N+533)VTbb2PH*ce=uO!t{E0FptVhH@{o{lQG-{8f+UAjwH-BPTcU4Xb}l01nhj+)nWSXLA`JPhNysKSR{70= z-3>*a($B1S7H^t~_E50~`pXg+ky<|rM~#jgzM5y(@9B`J<(fR(z4Yi&b&krP!#33j zX!VIQ+7{bAzq%x@NT7S9Ej~M_Y2D>^Z;{Md3K!8bQQqKrEUIe3NP{r6$y4LUhT=?j zD_72C*dJkVNktuoLMJ6BXSvL4{f)cM<7zuYqj6j^NzHN^=K*l#pE~@Ip)hV#AZ{q{r$-u3T^LRNpD~K=F@^P@LlW3n?6G>?owor>vyfFy8(!00fhb> zHACJQ2JFWDVLue*<7sJC!bxN(<}R~T8$|YSNzYg;FL;-^u0~%Ug8i?Bj ze-1Sh&b3u?VqIULM1@=Q9FsLhLAHhMz{l%Km4l^F5ZH#dTEK9~fS?`4v;U?nB;gr* z2%@eR2@ec zPzqoCZ@y1^fRA@OQ1Aejs)GH=3sZgFM`P#PS;LC07oe~Ozb2E6dK`D_00*);FK%h* z$AMkHj40=Ex4X<{%BH(!*ddNxO+Dh$1SX@-BjEB}$q`Tmm}$czKEV*8qb@>x`ox4v zO+35(6@q-t;~)ztWFKGKmW=}aWv7b&N~7Le_!O;XN)$c-i#Q!!-FAsDz`X8W`qweb zR!RBa+e1(h{Z`3bJ!Jl8#4~O&xpnO--AYR?w+n<`rKW``65W=zlZ-_9uL`S{0*;Aa z1(BrIFfqH*Qsm3cE`IH|?R($z&2Q-yazA1gJHP%?Q?9-jRA+G4o#R3xKW-{#_|tXg z-Itd4*~>;mmJtrD9SwAK)%T;pp@S=ZgbB$yxH>@;AJI@fsLy6uL_z3;cR$b4zB4q^ zB{~axyMdPTSjQIpfTo|`cK9en;7>cW!4|Q2N z#767sOY{tcGdW=VpT!*lGFw#$SS4flsOjD|{i$Ezmedg?2bI7d>33j);I~B_8)bM! z(8)dQ^CPZ#|IA#I!$5fPp;BNG5>SDvL+7V!u$m;J#?Ql(Ww9unSaFuhB~e1(T0ji_ zyu179ML0cXz=WCjU7iKcj-Zg>2H$I3WsZjd{$KzX`Wi zw3Ka?<1$aD|4NM*X!Q-ofMcLNZ*v!-3H|-EDvKejHf3C`vs0KM`z+HrfWuxK%Zt=u z;_Ua&h)bJuo55vEV9_r(*URP&i(&$hKV0T(_@9|8jc`Jg{_GN!Yh$=-AlgQaN1;w?}vM?%=tw30kmtQ`ziys z*sje**UuK&_w&=wq|8cLQfBReGvSuXtI*io&8PV#e(5o$)VaZ)YQ3zq^vx@v(8u&w z=!ZThG78MB&rvFkxp6OTBH#m}_8j~Q>5!x!_#!ebif9VSHhj+_PR;n}hg5N)Oob6^ zg9E{Hib7pQWd2Pu&`1RNKGfQU6DTn!dqpq+xE2Yr?m-QKe+z^h<-H6PNI9&kTgCVDq#BMx@}59UkTUKbSfmw+mu;%ilbP+rn+!!dQ4; zPV0c<)%kL=I@!4$L8mW8%?dF#V<+uFb*zzzv1TxmeKa*TWFudU%S)4J_LU6-U)dgJ1aF}KMoz})l%>L0@4EccDywPykf3Uc zkJss2gS6eO?wIn6k9u_cG2dG7&%-lQh{rj5KfY^QTP}+9R`Of_g-v>R6%dsjR0i3r z&cw;V-fD1G`wYBjXY(lHArrUg*Qxe=B+pEKov-M3)ky2~$Aot^D(PR~YWL4fzxOAA=lUQcGo-ZCAMKv2 z>$>zQW(&A+I`lXoh6`c^WyY|3!vD5WZ}%&)Q8vgL$V$w-#{7Y6S`F4XG|R9E_a{rw zf?>5jz*@v-a0b%ZuJes^c+7RhhO4HZU*@jQkcFQ7@d(g%JOSeR0Dg8nvh_-vl`X;X zOrF0)TuM-7-!J;*rMnOW(gNi9W%sL26C)!s*~q!3u>ichil)F{>$oSAdoD;iW*2+= z>0g03;Bs3bos5Mh#E9mR%Cs2s6Z5%DAbpZe}hXo`6@zdqiBx|7zV8er{kI%9QvJN`2^k#9!nXd4oBRSlIh zteK;-jWB=WGVfsy<%}}o3Jg9hPeT}SBGka$#KWiX(VS7rBDMe!v z+j%aiG0l+C>+M@m!h^X18q+xy*)!hEi%i! zw{3U+H>?$S#CW8WInab{3I!#lVja2aOv-dZM~IVIc}7*WtzG0U$&Oy@CwZ;a>UKf} z$Ch7R|G1b_Jmw+L%~qvqIqA8X>RLvx)Gsw|@)eFSMHLt}ip|aVsidjt>KE$OEqts7 z&v$LDr#V9eoFRk$0EvYk%7{%|T}ZD^SQ~MPIe9QU{IrAPCU8s%g}23BO}EyxP!GGm z?K>fHxeElW>C72mWJ_iH&YP1|_cs=HT3PO7>NNkQaMHO!wFj?@sIUdVub2!2>9_v7Uj&*?I|#E?ph>(Wjn=qT;D3BPw17ww zzGvS4)V{T67ac?#5}1oq{`e=IW7PfijkmQXmIcU}>(~mx%f$&Jy9$Fc;>m(Y;+92o=vk?sZ)! z>e^nq?G9(fmV7 zy2x*~-xg~Uq!Sp>OK1!-Xns2^32z^pQN8Xiu~9@F{?O8G6q_Q4zWoG2j6^B>LpXPM zPD4+w7XO!ZeG{GcGy8WVS*vJz9AGpB_@5tZ4ub@<7f0%@gV-=x!W4KBF!8sQ_2Tve z*O&;Vutg!`f-lyF=3iIu=8?d_{NrHlMv`bI@q%GV_-kq;mU|UVg2tH}?F}`e{SCZS z4d5qW!Y|~xw;)JcU`w!LBHDjLbD-#M0G0-bLa+9YCm7?&DE~349)d&Pc6Q!XglL9l z7h_JSe>vFmsnB)As~=>t1zI&3bh~yzuf6=qH6fLW4Wqg0g^p%MJ8wLPC4QGtiSZt{ zDiuaGHL(WAPZdqFXP!0bA1S|D*}vd@Tr=R6$~X7XS}mQ_CYz2ri{zFdn|IguX@@wR zp%eDnzDC!MRrzhFS*#=KXtXD*P##M19Y>R`-R%^_ob+80<4wmrgcS#-rxy$>zlF>l zIxRRd!sNz|4HrtH>G=61MtqwbwQKJ}1vkCf!t8z!|LPISpo)%fpj*V(r9xA+6!z-)y=>@Pipvrb91*KC>&HJjr zM}u}q+G%R&0Ig;be|8!9>DiutLC0xq-RnXwY#8u_lp$>J7v7>#;sY20!H(QJQEZ!` zujEoBYdv9l;QLcxfxe)AL}R;*1>y*g$RmGAS0L7r=wW`CKBT6IKy9GM{~IWPBzUO% ziADWEUXBWcEO6+z8Acyitq8+0S4BF2~Nj|50Yh z;N4g)mY&u4LD2FB%kjaHY@MQ#FDBDKA4QgWEcnqjrNyb2U+)h!k0H-d;f;TENm&7y5CyEXXg-Xr$)+=tq`md&z23R;ZvMgXf z@V!#z;eYk2?H#3D5tmk0TvwKrY`h+SWs`Z(6Aq-K#po`v;t6zN)&QB zfW@aAfJ}nq&XZbvg=m&YFRb7DI@?8@DWaw}2fW9PfJXv9Fs{vp2Y6{VSZhU&lzE7e zj{pWD zfuzf1wp5ADY&?keWwu1{(^wrqdn#Xc2MdP4PC-`|tPbdKleP1`K{NzS2$>OuX!~p* zz)$km_=|lV=%z%H3!0q{!0M?sFW_-lAwmT42mZNJi1;L8)rm2MUUXXBJN~9 z{f`&GzgKZPNntOYN3l+>j75fqDxKZeE#;FtDK#IskAagv5rtyUv$c7V?}fBbE~YJf zgJoJ~0=L?CyLz)NMdeRibbM7#qi6*vY8*NQ4{|Pl8dX0ZrR-@ZF1B?)P1Vka{w&?s zYqXa09^Gkza^qQSeif|n?n()k$bTT&^wF4Qs}ky%W4B#z8{XGl-47x=Sg%!L{ZJXj zP7jbTa9XX(=xP3-8-pg58ll&aujsoOAJrY5%UvStao2%yljza=jrFbM$kKAZg=Yfi z!u$Qu^DG0xO9IIXPeeb~I_}k(v0$@qT!Pr|d?Sl*mvn|4pZk;KN7&RKmL~DU@86c# zT3v4*IMzNWEkhYRo6u32qc1&3t(6u+@*QTC3|@u9u-V?c9eW5^7I5Y&exSL92(r`O zLKc$QC0xU!lPc=z#WH1UCMv8#4hrWCVG9a-=wDu=6@b^XdEOoZt}xq!p?V~cr3Fik z#Hjj)gg&=Q=+oR1PqSqzd6AF^$A30znR!pvAbnW9?yZ}kXjk=lc52|5(U>*kgli}l z|DYnP#&LZ-P5714fkS~KPErA@&kK3JGjSIpa5(k613;{{KZ^|m9kkP>nOlh5(jb*< z3F5ld6RZjerZ%ZtXNs`r;g1LsibRA+CMwE#T3KBUY5ZHHCasA`)S&4Tvg#nd6TJ4N zuT!2e@z=j;Q4N7@0atNih%DaPEBE^FK{J5+oB6M5rFv+<9DEO5-K@^$kBQR$H2f@p zA$5aaGJa9nJ8|e8(GYw=4F(8< zzRuZ<^9_5RlI*R##s~3U-W>nF;OnK_7CP)3@eZGQKAAgS&nXX4r7DIh(HDIpzAJc{ ztA>JL8h ?*IDbk6*!cjGgag@rj$a7n$j5YdPH6;!}%y_Dy>;-dX{%eI;IX$Ohgw zgE7}bF#eXKdQo2ap-$?G!wpBuD{zkIL(_9s9|5E+Mapxf9M%F24ymO9&WRq$o7(}| zo0Zdz_SQYgyy6@We0W^O?wvdX=jT|t8TOj@WN(cI1FQn!0d`=eZ23i-(%l`hX@4VZ z<-z`@peik+(nBzv3il{2Qb&+{9s^HFPscUJ-W`&R zyUKuKw@w^O7is8%mH}I_MiEI8+s6rk37Efa-Noy#jQ_x zEG1ePZ)7ilcya}Kvl-!svF{F8ATYlC<8e>Ip)784<~Kf(eqmy_i=;&PrhvKkH!{Ur zcbEwu`{0EugVSbvMT5N$tmHj>dczX}VjUy|I2#|T$0dG*@v^5sHOi&@esWeLY))^f z;A^XiWzv1%;0jf-g^#nAt=>EYU%Ul><<%}~`A%?j3AqY0Rf^kK!F$JAKT%7IFO59M z7)_p9SN+PLwvpbPu)pduSf~L0s;Dzs@bYPNYJ-pzxgBQ2nGJ! zV@Jq3x_9*_{y{4aJ88AkxcFt0^4Lx1QC?fQX;Mh+=@hANmXl$WwRa3IQ97oa9qNc4Vty45*_&T>;4Fd9zTR))P|Ub}DdBbUH*;yd|6!+? zBY-~jStH=!{DTmX`#=W5;RxJMD&ALsvy|n^e7fD%Wf4&a%T6l!w=@;eL^vVDSHrbg zq1#C@8qQ|#C&y#H=Tzt@Kfvd$Aa_6K{kEiToENUPtIRL|wviAtfXknx6L%$TkhUhE ziZ^;7bHZCrPN49Hg=_tKmv8xPChs^tOP?%r?#ry)L8u+S27{-W_AkCuibN&2b@k5I zU(YiSbI)p7d2a)!+fhF%;q2{93d(zL3(kx>=3)379Zk|0p>Ns22H*&}9n1BiMKJb% zPyKq8AloL~GAcNaL}b$N*Xte7*a~W+y{Y)>KAW1hOR#49lh3(67!)6O>PG+61;n~L zw`avxGQGH-(jd+zVMc&$-dUnIQDNMakelpWV@W3<^FM}=zaz;l9|ZNxQ1$~bLNT1b z_P{Zbr(OVA-d>%%0P_;Nusq+dX7G8opSt$q+TzF!;FKicRz z41SCq+{<1dre#)HE>yhuR;XEnyCDjSuAjCX!c zR&)Ip{s;X(m-_!gN45z9fw7~3fad#kGtOr>6}0fp3jrsVjOE%ijy7tA+&eT3zvKNI zwih}!)!I47f=*z!YSJz?HwRw4kp>CzWkXnDF*z?<#Z>;UGv~`aso5MR6{1cBm)Y@C z5K+TR#A7Hot;AxwAKCXDX1$ho4^~xFb}BA-UZ3jGm^19murzq7GE3xJCTT4&gRqg| zvpe1koz^Voi#f7Z(NEM7RWx43&YCn7>oV=^pwWGw?;je|&(07d+W(HSRnS8j#*IC6 zBmceGd4>nO4qq~}EwcW*WPZj2`@5q2Y;q5D+0VeL+|ug;U8Yc$*af!Cct>ICzJTcM za0G}k+)emW-G0B^z~TI)Mk@pGBP=gBy8(EwTF=93#}^0aXavGR{Alh@ptJesb9k4v znipFxc-cWIHQK(!dtPkNW*w8GOyh|0u|kzs_!itP&}DtH zTQPfOXLqH_iY_4-TQc!e6A+!NoVnZLTXuHLDdl}aHE`pN$Q*+E_p$e%hTvOY2z}^t z0Br<3wu4`dfJ{}P-8iurf%`>KRxg0P!PFWZqz|~LW^vl!7DmNojR~IcE9ofIQuH0Y$uuZVi58j2bjm60ICIyn!?(`#h0_A69cppLLdCvz6V)u zrNzCzNWfG`CSclLR3#u1$DDO4b$m(Qfcnr6 zfKHi@dXa>J!qWtOJVA&kZBdSwmzV##Qw{@bUl#yoE!e4-BJ>DyVQM{Pwd-?+9Yko; z{{=KDWN6&M^(}iT+R#SNJy;zer+^jTpGc;TA$d^bd$=cO-ZAJG-6s1jU=JycLF9VO z02=~u2=@P^vAHc8K=t{fuqdyCPQ{UkA#1m}U(F~8As~i}Z{cNC@SF0M1&u z9}WXJX7*wp`pEk%lr=|IdM{sHT%k(d2EE7Kq?e0bLIvbyP5B4L^Ce-+H)CqhE2wmP?qg6Vab+L;R$A#=%JD3M} zNN2Z#c@0+7<#}{?+T=G#9&)@gt&4T>oHJK1m>gpwTuj|U>0UzY8l5muPTru=J9O`T zGx|mUF|T{VgL)*MvXq*T{MYynuxNpsVYS*W2b&4FOhF;wwb9(oR;+%u6SKW0An{*< zpfG^4tX@T8GFe--{r*5Bvi~Hc?CBFoE#7D_{i4 zkPVnMc%4;IU?}|dcPs7wX(H-g7-7v&{b|h3cSnw$K$wn}cDcr&7ZW%NtE_gUyaMof z^V|aGO*_Oo`@_a7IHVnu#*!#_chhyo9Mhru#8uqR;WDv~AA*aSe~tw2iuYR63( zAqcNqPk%+0j5v2;UdJ}*cY{fG0Q`~{EwN(p0wrkC4|-wfxUjI=RN&|y__CDZ4zR$Z z*5>Db?^w{M=CZTB&IAHyY(L%}9;(yFfno1Yf#}C~IL6BggjN#Wpq~CHDvbkre?r2Gii#flfCxHHQ8Q`J0o2C_gD673SBHGb z_l=JizdxRr-B3a=OB1N1V#qLkfwfSeVTmTADLz1zp_;+v1Rng_!A_=VGYg#Bst9i%xdS_WOiA>3tlz5<>UNgbd2o{X!V`oqUrBnZL;d1RMTeL}wTYHKbuWML6GcL=WW@J@s7Y2*0K1kWHjiFtI? zr@oEf$aA&^kJ;*+WVY%JiYCP)ak5OBU5{~u2%oSGWrzY8{~h+P2hP) zX5QjdaZS0Dbr@m$-^Z;Z(BIqeKFf=S4+Y{`0IYB0GMm$Sky!|JAW=SrmMtt5Ka^Vu z{hN3gP7a@2b@-frA~P9feCpNa3x?>Inh30*0H{nAQqE0UeE>0c^k9+y#)|#PE{*tCwBY2v--+XT0)8)v7y>};b9wG?c1&mnE zh}ef`h$UVFFL9(%ov#5PVM*v*ZG>fu=`0}q2wF^IaNBYRJU<$8>P5t@1I(H;?0y z#|6icK^iDx2bHjZip42s!?UA$JV7Gh8r$M8ErQ9!1+^K({dP1>8NPG{UWLT*@fop$ zNWE85L!CQFmJAeI1dsbR5njn&prJB?93u#t3JAOIQ?!ke65Ap>3&-eL@x9@v@cCI_ zaNuX*0U;@;tkLI&1ZPT3k*CSaWo&NkPSO~_TuW<1U1TLCW#wyNa!29Vw(xIw0_=C; z#)RYXApHOj>h2_Aa0tgrlx}b~ad;>U=eFj67#8ropm0UUfmRuDL?1s1Ol(R`M&;f! z0FTVZ&gzozF zG>bw34ol;Hek3C+dOX!>=T6`4C4UWLq-PXj0yWnXn8zh7k%dAsXkTEX*uIC%v#}I( z_I{(wpYd^xY85`BE++9CIEE~(kuqG{vqu^Xn`C51LC=+zd*_pCy4n5ubh4#ty!(Yp z#J6|;6D+feekc5gz>;8u;NJP_mkRSTQol7!JfzJ+pIJA%#*o;R{K|jP*Z!;8_U~^i zWMG9_WapsoA^aBPVr2F~f_NlQxCm}S;ccZXVs`|zW$3n(+#PbpQ4YlbNVg7J3ow~b zYC$PLNQ@bAa=6!>%;6P&{6D_FF|5*f-8;9kZ8zDrn`}Wetft(tSJYI2df0 zQS4%Z?mU4N@%H?09`m+p^j=_$u>~{n79^~c71|CzNMRLeNRYnB18pYe3O^j!hv<## z{Vg7kd#?K~`|9xseNR$XSS&~YS0T0K>F$Vt)h?J;tY-~h=gLTvHobrD7Nn&@1SXm> z=1B1fJQtzvkIZRnqy75eAXn4Uvg`{k%j+-ie!BVrXBw16?|VIdaku5rjXsCc13IYi z8o%q&$bdoI^Q*=c&)p;yaB~_!x$Tf|oENB^PT$}$LeT9L`A!7 zGVBhfNTMojFRVr>9Cz`k^uqXnLMu!k@Z8>yn&yjNUmxc@qOZ2=Mp+XBcpS`%f zruB_XCkfj~I;NR_Aoeiiw_x7ib}&?`#iC76yN=^hv;})hO1Y|hgqo@Y_Vy* zDN#^>>UrvV(QA#s?2~juM_NJC^+t4ZP6P2wr&Hm@jBu6+`)BIWRd1&8@fR0Vn@m?cX29%>WTwueI9` z3l16s9@hq^|5d2e;tdy^BhL-txP&4?~g2T>IlN^P5-?%%Q87DVc6S%8#mC;LTYi?Ku&O9$=T!G znj;ss&H8B-fJ4E$4XUJw@}l!_N0-7CXKxZ|u1hi<$O+cQn7d=}L_Xi65UveGV!R=w z3*8eC-Eqvawhn#Oi^WA@d~gQkt+PM}t^)gFVY=U>f;|y?*2w-ofXx(8N@Jdjt9MSn zvxUnEx~hfzrOjO!0liAP+N4%REOGojZ1H@ zaotbDn6K+-+YmQK@K2A-UT@fuNHt=wRiI{YJdO2QN-B+qtOb(7#O5au9}xwU6mi@G zvF$*1maGzYA1WI}-)5NlgW$be5QK-w|%rk zdys;!=TA7;*KU)hi)F^HAFKH-Zx@E3n4te5;mMOj3+M?L{Q-h_UVY(wV?K#Gu0NMN zBbPVHY#%Zz%1qo0lf*+3J#g;`#q=XOhwOn!tasCaN7z-+xRtfPLlx&hbw2QhuVh;x zz({0BeX9+mSnLR*f4qUa3V9gJWQ;=v6& zXiy`LjMn$FgHMeRa?-dB^w;-2L(8<=Js7`VcHAjqTo~_<+QE=Oy1Q)$!X{nC_5Kl2 zU~5X=urSGqp^Jdkd;)2$2_x)n>;>u+f?I~C0~l3!LQtP%aj!yN#@=^b$U1i_YajjzGZML{XP|^OBpY%+lg%|o6mT+#-g9qZN&S1 zZ#e&DNv9I0!tP_o10Vc=C~bGW_O)O4HQ#TMGv6t&d;BO_EKzih?mKKxV@5EU_i6uas?n#LHMa|=%&Yg1HwYod? zgOayl+%K8D%vQIJ=J$GSzqB<|cZTf4*5d`GS5o2I2o^_%)+=AZd#57y6xDteIgIJJ zWhB&RC77=8K6sG!K*VrlNLwXb;z!>iaPeDXe#`w|E^oUd!1ihvD1_6%`=l5oXXvd( zEf-1mUOxj-LqRGm=33k8Mmivo%sKlcPw=@tNb{`MWQDe%whcVS;1YdAW=SYRY6xHc z4kE>G>iUuU>*H1jqj8dtV&zS8Jc(t+VKC7JpyW|2#Jq?BVA@oaBtGY?>-VSxF+RN< zAbcQ2Ci|}kIn>6+hBS91L<*f(wE$Fd)el;Xz;$A!ZPXwt{YvV5j826#=UH z<|7DMdIiA@kT0L~&Hg|vLF0iHx{k9BgAf+dSi3J-CP_NQg31^HE}-(9JYI}Fnxe9i zv*$rdzO>mnRjLQFV^B?mB_yk~QSva1G0jCNfkUc6k~R4S8G<<>auQ4r!1I8E?X*y# z@ptwRWcdYC$F&$l-|{kh0l2|490Tz8_~jP~G`s{iddt`6jZ+|Exl4*nI_K+% zFfIkNsWI5*pUk6^^L`>S;HtB0WmpA*gQjP!hSF=I71l-cE~ghWClE2zGK*8L%Ev!H z=Q){NI{ikzS=`(Z!+4Je@O=W|i%*8@uGi^+YxQY9NW!~t{H5yYLyeapNIpfqa_TF$ z+blF4{+aFC_cs7Uwq(H72bVA~N&C<-1VrNTP9XN1}yBzwD zNTijjRR+;yuhs~&*p_N@)f}T%64J#;SlMUohp9h{V4TKI_y)ta`<+qX{W@$Qi$=2H z<$)xCs}3MCUkM^!QV*?8bF_4QuGWE+RC|`E@j+@Qhzcn@M>`c$LJqG|<{ADlxjg&&VNNL<%7$nzWTJWNYxYA{s;WY}h~62I&Z)>HO3f-dxC_ zt^I`8JAb^mDuh9~D6P}G{1=Y+pHf<*c4@F%81#cAIVLh02$I$n6bEl9tIa=?d}j;| zxqsm>sPZ%(EZ=ohF_3UcK_&5p(Nzl*sWjK0Zw93f69`6Yf--yuLIa#<_6PueKoAY- z3<1mvY=sZR`EmPKq`22rx*CNyEUFY%LlMW*harwt1?lhm7z_gIR$}&!X~HwNaIQu( z<64t_IIE$3 z!($Dp0HLPO)v5{BhB;_}LXxxw4E0T3x^YRDz%r8sl@gx)JfhXMjE8**_k61gk6S`A zz&-dbU?^Cc)%roeKG0`rSAQNG!gi5Oc3CDy3N;?@Cho5)TI77G`<)Y!Jl0teRNQLk zfU`%V5%YTV-ocGpie&?2g6>RfcA#Th#7+leG6j76cWJPrMXOn}Sx8NmWH4aaVawsT zM={BV#s_}>_U>gZ1@OrR&z;X|8ECF^|+8aW1JWnVztFM7H<(RjiZN5m+2dAS>fSfJCdu~ z0C<|p6HGyhf(KO9bbXH9E;zj?Fdt(6Z}Y)LE{kUPsq^#9E86Y8t~TOd{Jz*%f9Y-; zP#l~cwffmRYCb!yv-1OfIH_Jn5S0pj;U=m#au$?Z5KR9IR#Wp^sst?N2@3dP@YyE~ zR+E#$xX&))eb#`&G*hZ`rx6&XhcPOqBpqB5Pn4_^mEp|wh!E?`!`t@_1PDJ=a5hiTab*r@w%`hy*S*d`g``0H%?_`<;dm4% z17M#zlTMwHsJ%VoCGaZub#uYPLs<-;l@OUhf>hGB@ll$K>hg)3%OtQ*xLV<4ng<4T zR3ackn1-MWAsH{mibe?GFNcE5%!4IfW`rW6+FXX@vkKSDni4?^jQdMJC0JH9JMZQi z8PeuUY|DjnttIz?5bHnHt(!IxmDo!MlSwoQ9G>%6WaJgiWeobV0(RxEiwsN`$Jv(Ya*&x>K}a5(C&VXTJ;h>vR%@nps`{uYw`~UX=$p&TQh|9$Mx!; z*5cIOU2L2PI*>_xUKh-PqyzGIXR!7#6A}KFp4P;!`|Se0X17bkhW6Vsxl?i>Z8rN8 z5?AxJi?wsYn+tw-#J}pzgJ>K}-xI$NxB(F$(pd;hqznUp2GWd~yC|dKSHUR4dANK+ z(%~e&*U;4@AhI(7qe`pZ2 zO7IsQY?h}J6c4McG_fq_sLI2@SKOR`GG$b>HsHY)(N8S^2}}@= zxTVP<=(Ba8?EyqClv@*oBaOVRHuWj`t&h5BcdCD25kUss+!YQ@3xNAE@g~y%-E^6) zu=@2#VdqS(>h$T-UEQ$x*A^GLBl?}i^BH~+D-qpLrSP*N@Bc_z&dn;RkVXx<- ztDk*L)zFWIAzp;G;Gp~MWJlD70+gZcM_68Kx?4kwEa6V|O9)L+$}6ACWW;-U{=$~` zU3K;y+nPw+Xqo41xwgDObr(4Qlj>(rOcNNy&Q0=mM}aUJ?>Q(MGJ8{#+Rp~({CAh8)UhBeeF?C?UfXhvk6`ntFiXD>K6 zQ&?t>=W|;O=DWll6>p&a-Q|K=>1N>ZH!e4FnksDi*5ke#L!P%Cxkv=ol zPWkc@~RE_DB$?_OEG8XJs%I>tUaYe$_Vhv5K; zBSotCg{$*^Pq(t651)lpOR$HDxi<2yH3Yd_H|}=z+3Sq!Ylq!L-)TxaB8i~$^0kUH zTYq^SbwE>~GA`LA*g8S;x7(QPoqK%N6R^z8Ds1W*#NHrNdf+Z;uz%FVE#^~MUX%Wl8nX% zk#;=fz>tlVj3XUDj$*!+dP&^dPj+iisyfO%?9>kT!ls}#4EV)E(|jNeuVKe5zM$+^ zHN6bb?|?uwD9BfMB$uFCEs=|BIB8Ur33Ggg5!@Pbi^m)hO13=?H;VJ>?JL)X<9yRJ z$`MMbM|9XHjVIItIShjugQ3`#05K)kR(C|b*$nUg2-JsO#SI~Mw<^2I$$ijl;9otL z0XZw5WJ1VL7>*Iaj6W}KsKou^b0Fpv{GxT0&lUS31&yzQ6=px85(D#0{yjL1G>zsn z7k74OES5^bgc-qX8waJb3OP~ycTB3s;}AI)Bq=JQ?DrT03dtRVlg;3#>7C0X{;mbu za?=8{vflnL@e+Z-W!c`vi>o%5Q46&ooA#fKl-eUzPEY_jlS-sXO3;s9VClRw%zP;w z*Mg>fO{IfXSxqfHjhr464%y}qJQ`67MJhw4nndsd)spmCQazAT(U=$q34=O?RxcMY zRb6e^W;FvlrghHY5WXQDw_040F)Qrk0P;a4_Edt;A^9t?3p!8>5(|H~H(fS5`#d80 zKXm}$qItO#BD>yY-~ReKxwlzBIhmkd7R^-Xp>66nj1L4C>V-t6fY)s-h!4FTrGc$+ z=vRa10m-CnyWfA;l6dVWh9cmo>YjiE8XF+nUL5Zs?ynC!@&4zY+11Ehulwg7e{Ued zWCqouV7Ic;R?R6FV7Rb%SV4=chxeJlW^qtc)(d3|S*J#eY`YXM#`soK4H*1L01Xez z7hbD{x-aK0;lxe`8jZzs1RjgVLmbB;r7Siaj0uhp@$USL=uJ@*AykKri6))Yjn*#y zyN|8aD__7Ol{-;UiLotXh7de$zaf299>#$tW&AkCW87_r%rz4` zOcK!6Xk^9lb(^-p8DF%K4CvNq*H=P5!&5Oj2-68!ob2r8MZ`4Jk9K&-v2&1w%vDc~ zpNV_2Y9bj4)Dh4`hr_WJ9C9N1fV3_;;v(K~KnU0LSPe3pL#4L$2$395TY>O{Y89}txZ0{ahsG3N^u*NoTcH1}eNHkI1Q(GPp zi^zZ)oyf(ksf*8C$Q2JS`CBj7IR6%`{+qG4E>u->b_;D4Z0s1wYn*k6F^+e$_!tmF zT{KIkK@z?A=?vqb<8Cx*37ava-G7dbzKMfSRxIDDvn6Wq7ER#fJVf52HS9URaw)Y| zTj+Yybg$u}xuqF_k2JC9E%-+sJNUupuMGsGfeZ zd4}njjr$#5Ci{!(#n2W@U#ZxPqRS4| zPN&?@-1{3EdX>QTjCgi-&2z#(RPN6QnxN?IUv8|LsK;bbMrzt{pjUFNdjFW&u&bk{ zCXUjS$77=VOKkXexv<;7?7%CszA$7gx6g8NXoc}GeSRG_g2$9|_*&N293wp*q~dm{ z#(Y*3>JezigiNeff5v&n&;&v%z#(A}L1Pui%~6dht)&`X;xo`#eM$?L+=({M17tE1 z#?G$22AvwZjPDL(N1s=ru$=o?{~Q~Aim)PMIV^=93j-^J4}~e_~?#R7uNBUyC?cFe4+7BqMTQ=BfxOp~$nln%vlD=8SNZg*dLq zgK0`-UOGE^e&{AnuBIFiHXEYlX|6-Gtn1ubi)iV%BZ zWJ^eu{IBBXf78$BqM`KxQCRf%fEd zzPRij3{4a8W*4L%NVGtrN)QJWqIwCHeWiW)`N5rf~GL z;|y{4wl+4?rea>G7GGdEXLfIyO6E3x)WaWgR*QWinvgJpEbnz$aidF?ZTN1-r$eYj zI_hDSAz%Lqp-^0kt=vq*OGlsx|I{`eJzf{7*_~dK*-*gKphO5RK1F^})S?bKrO#?0 zZ2(;|PhCgQ>N+8RAs9C0+qZHVl>xe6tGuZ#5jyo3^5U~_{Po(1@@`PE?^Ba2^M}TR zcJp+N00U#ccZxAwsTa@urIVIJ#-ilMlLOAeV$5^JCR)wP`7>KB;nw3q?!{W4xahQqO}}c6^8#=1O8iA;0F(WQ{c98(PRR57QtLTle};tta>F_ zI%~buzhM6CQ%Jt<350tP+~mDGoh}<)b);1Q-6y(YU|fAplq^u!v`$=SOSOg|UPN{X zam?fglDH#F-zVv11gSi)V~j44Qwd}};vrrpvLc}9w~^v4Ma9+s(qIOMI2{CyK&#z4 zr^v6NrG+sdj*Dt5szNHYRbAwkXy5D_U#1y zMKWw9GeZ9KNs-w$M_2usN=x>Wq5?$ahB6m`1fX6DclW)txMvE>11_FA!L3MPZbkDC zY4VRq+L?2NrwP0B@vtg86}v)J61&y=S7{Q#PPd!(h!pHfS~PYBB6rs%JP%m<{2|)f zWc9@Tb(#JFnIUQ)l0jGUbT?{(C{!D9yKhVhkj?#Kd^vPxrQ&U23CWSm9wAOdrDRF= z0CTtbI5{^~QIG(WQi&J|(?n82i&1&8z>clKOqMA-o!4QvDKKR;hE7ak%Vy&qqjV>k z`v2Mlpbl}G2`XgU8b_W874&*{?bqhR`WF+o$u8I*j}fBG0IhKYEWQ)eUzbngp{TQ5cNwNh(2z;ei4$jObJsfx(mYmnQ7s`By^My)d4h1eHij2%>c6UUTWEN4cRM6cHl zG7-AysgUC(zWiR}4!LDkdZH`Q6GHIK+kD{^M=0xUatH~*>ua3U2t{dLpH;>l>qIQi z>_*k+((N#qa1B`obxcu0i1AS!eJ}=dy^IYwS8<~{H+dzdll>%pIK6u9B3f+5;H3sK z;C56~$B>A$OdG;UXP7#9jSV-wOluUl>zR#>@Ux8h1#qQS zD^e24f3^Mp+5p;tgOdSUz?5m2!s%ek*}qnlj4OYAPt?8bs$rTQ#^2;NNPf^V>||=% z+pMvyiN8>8p#ZoELV)Caet|_~gyB}`pfPwlaOk0hJ3e{D#{`>$$_xX$9iZ?CP>P>8 zB3sa>FjTgBhAgo^KlWj7s8)k(;i6>*?vEz35#}Isn?QlP%I2`F7AiBa8VqKd+^Vyhrd5N@Q;`h-lSh3R*9*eS; zHnzKPx4FI8oPHIj{MIe=ejo!uJ%5G+xha0-$?zvxix!2<9G%A*gt63#Sn2a$;Cx;d z&mWBs)YGZvmVR48cWrY7XPw@h@q(W>7~SQnd(rz+Q2~}lP1#VC-=MVVbxhK=kt6GS zQ<0>=(M`@7XvXruATYHfoi$mK(X_GAtMiZQ58DhnpE~(-c)K019vnJP^M4w+gKo`- zO~03QKTs5L>xU(90~Gz#jpBuJ1UG3>6$Bq7T-atY!=nKVzimjtYA%YHWn+lwm%=88 zm3JKut<5KKAtC~=p(973!bC)6EBDTbN11(i?llafJRPl*X4Y!t4{Hik{j36>PeP}j zl;aWXQ8}agMdu9!41T~zv|=oq$hmle}X`Fvf;+Pb=G3h;bE-i_kvRXv3sRu zIyZ#(_EW)urU9_t1rW&hDOVnc*a-wwON}&x3@?Ni$Ye{T ztoX4oM0g=$a#8^!`GW2tLRkfMV-h7|>{2<|*^;GVo;*23{2t0=_yqRHlAehT$9>LS zyb#0U9wfVctU{QWNLH`U{WV$?IwmZ{91Q}4%4U4a2*cZAbEs(5bb$P@AS$ucJt!uz z+{@F*qa44>@?k!?MhX=UAuiqGHwRBf03byHRS2Rls@!K%cn= zT0e8JZSLZIcJ=hN^FfeDsF|O(LL%w)7|gs+gymBjKm!6xR4CVtMI-{E{!n5h@JGbi z(oNW&#Ypl_9VPrNB~2V;JLKJ9i5k-6AD`}zM+~Ks71!#YWtMp!EV6YHmW4>ctLQ}N zMF@KEV%gh98S<@jLoUqP1QHunnn0D40EffgOVqcxMe}^N+zv^_lMu0iA6_S(bIbPe z{Jq5U>J>K!bXeu6$?!fuQ-X;e8j! z??JzS$QiGV1{*IYPBMYmDFrgG{L|U~m!r-A;b;@8)o%D7SpZNBy{@jzwHOQYTd(b> zT__cXPb3R>U-M;*{4*N_b`dmAA-A;NwuN2(Y%&T@w+wzaV9 zipq4-O0Lf(Z$Zu9WHPH@BL%Bb=N5TsIe7OW(e{39#R$eeQuqmnmWXg1#Kp{|7>~@KJLK4DxR=9Ar{NF2pmWGtnJ9<~xWDlk4Y$?88$X*LHcIHM88{)HUZJ zwva$RY-a?#V6=n=E1QUXDJf~xE<~6w9E!?%zrB%IVDMHvU`&654IS;@DLJ3Mdv@T& z!Kx}AqiR!~m*p{sgoqG9dZH}rBo+l{?Q?2qs}U#i`mF}iHQHgWW;Q%+57k%t+dd^2 zygnPE_KNnEMuN$f5gGr~x0Fp$VrKrhso$fKqS4SS<4EohC%>B|jE?|)i%g-7>DR@2$ zUCE^XLj)3{9m&2#%xi{iqIaobK|jLdG+HVIdf$A}(Q~PxN}PxD`=Bh^^tSeGfsEiW zy|7z!g<%1hO7=oyys*c~db+nJkPf34?uZ^OR^($k*~q<%02o(ivOfW*Q%+xhV&s`b z`S1o7sm`q6-O1%X?AkwjQRAUI5m`kyN$h1&wb}8t=-?A`WzrAcu$%}viFNkj@Ig@- z^&dKmG32x@O|1>O`-OWLb&|M9nZ3n%fR?a2kr#1&PPo06P?Hf#!&D|fXxcrM1(p>aa@s*L zBRAfiv`U2jk*KK9wPmO$P~-YyFi$cbq*2PhF%9~>&y2e>W1ci()3OIiKO=l?Has9Q zQ(I9#i~DiYq^~0Tz34U}`je&;+6+J55dG5FwX7bd(89%H9N`8hFYl^dAMrv5Zln@c z0wkcofqj_bp_%NbnSe5v|Sz6v*;`D$m!;=?>%feAOEVfpyX>n z3%BhlqiBy4E)CgyP32r}+ zA~fBg6QlYue~}(2v?BfK$xG(Du_+!#e4DSVbUt6xEex{|*4%lqUP)Xbz#`(>K?S#_ zd5uOfOJu^dWG)!)!Dxu1KSOjJ?L07u9BcY|z7C<&fN#T{#Mr!Gxm=ss3~a?9B2~() z<^8!v8FQ$TnH{st?AvEKS?Ym;sD*{SDS>8>8`YmJ=qY$1X|RysnRO1aT091*JtxQJ zn$y@>0V&2yb^PS+5V-{+o$A3x^s|U1mDmz=N;8n58qUuH2-3Me1!zB6KUv*6Bo?P_ z!w4m=x`;I=`VWaqB*WvVipo=3r`BOpV{Sw(u2iixIPZkDFR9JQuy8;Rg@~!Q&ECYj z`NSKf`4w=kQWIwo)Z`{h8G&z!LmQP_mB?O_CoY}i6(v{NCU1a-5b>)b+ECr&eh2r% zBQ1!*$231Ir*je1+~esg6gwECqQn)esJM?7pli?-vN|}?UX9n%B*=c}*d=p7@l!bX zX-uV7d-P(gYhe_9LXOC%qk%KKpIFFDc;!aFL|~L($B%i;{iUh6|7(ZvhDKlb%OFj|X@ ziIJN{$`Ky~n{{}iIXI~jZr2f&h!K5(A$i^HS4uik=}b*oAf4#bT|aS;eD!#~x5m#b zFK6LYQ)YFSGF`%q36vm{Ur(^v1q^a@)$5_oQA9rojcunq*QsTeqJD`>b?Tuf%}HdC zGDnjH7wHYJu3;oQUzr9@D^*HyJySe5gF*LGO(ZC@IZ%<5N{op6viC^dLBdg5yBE3Y zpjROK$b5)<7q+!UPYFMazpgV>h0`_AP3>REnDHAIyGVtcFV`7D>>B%JAjgQ!lJQDh|PIplz!C2qDP%D~_ z&n^F0f$VaV$J(g3!vc!Nj;m`TOSmMOr{ZxI7#>RU_fR>^9zsopeKYj+cwvCK>|}xd z2pjofAhmW?jd%8@c7{@hm&}tmSi3tD>dV2dO=zk?=(OTU!K;UcpwI3A!qJq;Lr)@> zHAVP`pP1xn)!IAS&s&#O!T1fg;r9zt7Jzz-x88%MzBaga~IA#vwgYB5u@l zhr^uJ=H#DX>5q$k^lO^8z%jq-d5>8y*NZ-G)WB76l;i(-vhA{(S0Bx zYXLG^qa=2Pvva%nI4B~Vie;qkW^D3N7g_U~yMQw=3@s?cOAhV6!`E5DDrsS_Zp}JZ z*vH1)8C(NtmtaHJDhtP*Vw?!Nif0l%3`gdkuW#2(KZ7<(R%c+oEJ)n|s&DF>P7+L9 zG<-#8!~Jrh;hD!`L>=~{vUp!25EWCOM4_1F5p~=yfkn|E65qlK&UiuIkLTrjet|x1 z=PaIMPoavL{?6517W><$*2;N=?}s_z+MYn;^n^eKEIb_UQZFZOBRhA z-pzX-W%SQrZ_hgJPoeR1=|1~8p<{^)j$KR{wYz0U_xjBVN5dADN*>S57CLk@{`c1LKM(ic#Hu`>6S==WrNg?s!yBg-CE zDF^bvgWvb(^ewHeKhC2w_J4a5Cwl;{4HSxyzms>KDA{l1TcS8zgFu)ti)}zC-|#0H zv$B3}5kx{kZ%l7-6a7dTF;{f*FHUG@*P^Zv2tnP1U1GEdMYtsq0%B1!H+58ssp3Ch zI4m86WHY3k&!I0P)rLpmLa1?tQuU|=9?`lR~$prAvUi1))uJELk0v1z~CuNsK zd03*l6n=ScH)pt~8kALCkM)uLRyp{7^25a_2A_W8gb61ilC(4-H}QA4=)_yO<<<|a z!Cwu05K8k_&T8)mH#n)^bWLL$z}nKZjNEz8H{D5TDOUf#@4Eu%eczCGWCL^9ak=!X zeU3EL+WNg$!TAp3qUkI>KU|^o8uNT(sPS^yo9Tmq)}W>b*m`k z(dzO(U%z?Upp~~l#kk zC+I29hKUK1#aOR=uaZ7~YKq7+yIAT=@e0T*s0#evU{m2jH$~7XWRsc55l`u?0%J%~ zQq#!qGSz!2BO=mvX7ojw@AqugSYJu$kjJ?E;;H|^uL~W`fvT~B4t(czUE$cJ=W2t7 z+(*0RiYwJlO$SB4uP%XMEHby`ZN)y#$NH&)emA@60$YSuN}lt!o5v6f<<&aErrrPT zr2h+q)RBUQ8lS<~iWpdL{oC+k?zH>K(q?LH#La1J99J<04t7)JeIF<5s&?w;S&BOK=6(=rO z(MZnN>{ne45{?TX2zHIfJcguYe!_5X=*RouOYU!2S*`#Vlhhr|jP@ijXH4v1DVp02 zk%LA{z$6#V83k1g%Dotu)lQuC7=c2!mLwpkAnKJZsk0$MxmIOc0YjuUK)lOg*xgbI7bSLmQWn{l>iTcr-~&)nG_z?X5213(}>f zd3f_@NnF!z_W{~@bG|nl%;Hm;7Gr4C06dL`8)oT8OT_7Db}Q@Y7O8e-Kc^g;Z{3a6 zM*@#>Zh~JP&bJ!dc~{kC{I|dH9DFm@n7?btDfvxFhBSrfmbw1br4+xB^M?9Q?Nw0A zStY5+=(AN$ZOBX;rO*BN%7Qy|yu*lUE$taV-+BD%LUJ#FzDJP&ZQp{L+W z-6{#nc-X7BLBrSdR<_3w@euT2vZP~Dhy`Ad2MAjM8Pdpo>Q#-*NAjv4MUDT+=cDzFke-SPUk}bH>d&9Aasrtsrgh zP&aek?3>(7B{wIUV@aElmQRisGvh7O(L4rSAD6&n?Q%0dTGgug@Mbn0+ByTDGd97s zv|Sf6WI}I;pMe=4!yoHyJC)N(0n%~&a-up61RBJ*eS8YiW&6kCm z!F*0O@NsuG_p0Br#Ykiz70W`Nr2a*uaLv*%U?Yx1(L0zI>XO)#x4j~gWRY3DD)x3; z%z_x+)%z{TseSrVNRa=5VMZuexfpW63Sxa|X}WG&3y_b45TTw-|cj8S@~;LPu0VDxSsAf9pQ4Zg+^ zUakAD2=YH&4#2s;8-RPLc*9leclpw5arg3U=1Z9 zIy-hE>=5PYq`Fodvqfky4(SVm55iAz@j}2WLH|@PKga1Wv3^)EimO*k1V2{Bz=hlt zMp22872r8ghVtl5u%pm!Scvq$k)@|a|2b={UDKEeI6y6%$THXNznQ)b!%UVf3)vpU z_`#sEP8+#Wa%8B)R`i?u45EBmWurK>u~ZQPaErfH=!^vou8CK+WIXkYLA+x>`4cR5 z4A9!wm&R&Q*T?h1=dm}uRY|c@^i;CB-*cUly)ij;JZAY5;vQw_CM_P#R5Almp7IVv z3^%@Xj$cD#heQld(zysH*{-qbxde`>6>$p;7hQnV7j+gEYywOoj=4oq3NR7~28`eG z-Za{!ItkKoM1rZsba5I6pZ{c^|LSc^`woW%qSu?`r&V-BMOyMk`Wd1>m&((nv`09v z!|$kq>4Y__5b_upXa>`2hF`6}1v+{i`(3qv{pdZ>MIb``LxFtO zS!G080<$APk9L$&I}39_Zk6)Afs7v?+cYK1MP0YddruYDpIz>#@^txJUz*t-4A1O5 z+_C^uA{S#30G+BLfdt!#^?x@7AShA+YM*E~IdgpR+G#U;jkbdv?7|!d?jpu+Qpebk^1V5KxJBZZ1ukvl)`RLoj9x z!y#M6E?149rbY~5f!1S(8ACD+q>hv%17o8AxYz&AH^7Dkum06YE(s}4%8aMULE>98 zeVNw908VFR@7zoFMb6--q*NJSUo@G6OBDLuZNsrUEI{0*nR;F^_2A8bjgJre8T`+b zT`sTN@iov^qDB!FvlE;_%hl{YE;tj;+}n3<>fPs(UX_`zR{uGbV3bYZDK5 zqa!Hef=%^>56epw>) zt3*89L%}XL08Q24)SL2pRMgO`+04G2m&2M(QCJKh9s0ZZQ0BjMG2e z7k~Mv%?5yW&{3@WNw()#AFtuyOB{d2x3#auFt*<}T)F$@J8XW>oI1W0Jx`K&=$dky z!tj?!f&eA3F8{(vvM)#(xZ)T3uxJqdtVN0;`lFHGBpDdu5+8-!vk75e2ei+EQH0&L z)~;M&Huya!eb-)y-`|AaJaG;W36kfa#bPV$7G5DVMWd`f%LoTwZKePm^_aX9;6Zx9 z{KJ*1-z&G;%xmC~2#?t@DW}b*g%+f4rgn>41nXc)38F41ue?7neuOm0^z)2IK@r=t z9EUq>+)aN;&cm1Eq=9lZo2%X)@DBG!^7 zFe#A{ac2F509p9si>?Dv`!p|AZ$;eOWdl8;Vx;^pVK?vZA2&OljMXMV2KDUsyf3yK zBai#Vf{F0?M)lS-gCl;E%XOMseGeuvBd$JaYXN7_DXyfc|Nnb@{9v2EM7 zC${aR!-+MqZQHhO+o#|E-si(U``zbbuYT6ER`;rP-E~*}>MAr25d1Z zD62uIurE*BjFe8rd^%gt435}*GTrPdt6K9KFja6#S&#d?fb}UerSSj z20(68Z%JQ6HZYP{HiAHqNPge>{)<4Nw7*C0cl);eHY(+A&ZjwTv(Z&$+tbx569>3j z4Xf)85SFEuYSnv>tfG5-b9a>owJbzm64kmAMbxiKDLn0^`kUru#R^#I7d9NiHkdd#GU&5 z@n#kM`qENYR@OgyMr>iA$t?e#Z&5f4GymR8)AKZlWhA)mo;HmbK5kkAxF&O$kW6Gi zAP?$IkQQH3+YjhzGnWrSp@hb0>{RrMnV6a&`+;;kyg`_AB4R)X1SlirB}24Jgwgk! z>qaFM4xWSZ@0thn`^W~xZixc_WMtrQxgC9`UaTqa9|W@CXC|mErm>*9shFXLR4wAD z1nR0+jV(T1d9H0dP^9s&K$IoJm5!@OYfs zbgZik{Q1UKyXTi^M9|f9RR+++aABUVy@<`$oGTuv5HWr`+}CeHL6AK8kx;9Rq~<~E z{3^h^aWhJG23eRGl|u$|K=qG2H74?Xw$ z5(dcRIYVcI%}DIg8&Vwf->4B`Z2Hk`?q5|aG^+j>U|*khI&3l9h{U5bq5OsgAz^v7 zlb70>d$gv17>L#YV4~nOIDpn>$QYt2!eU&9Y+j%do|qkp zXKGV+AYoVsNWSKH2ow2=U+rwwMi zxA^hlcW;A$pU@DV>I2n~p8bSw1i-77NHMrXmY%lm@x`((d=RN2i=D~(yN62p5);mV z>i>{||8cbd7MMi-qYBTxS8ZzErzvB7S9U&cb+Tr%0E68IbR4U-$3$TNV594`Nm<+X zVs8Zz_%i^85FQ`zc9CDTieG{=eevH-c(oS7Xi>ZZ578dn75+FhIc@k zRz(GkOMWS;yGEL_Y#gdxPhi^KqM&O-3hIZGrRAF@`YVJBzv9|9K00v_HI?&)N=5Vf zH&1yofs1N73>e2!f^(Lo!7GC5=A5(ni_ z?r~65Af^FWK}X z&h%}5R6>}ET1A@4dL2aiI@mAWhw@!-E*@ii-rP;$zxC07Y5;#b#O9sd;Xo|+$9oM- zokCO=h=p)CQP@70VLWK-s>81`eHd30CstSkwB~~v-NwMb{s00i24g#lKF&%B8)jbd z*-7&uk~_48+1|YTyV;zX8}Xn7!txtM6=ur@0m5@lgaIzsOHQ}zn3_Wh!{W)))%!-u z=X9wQ(-FyfqpeShFmNd;ljP{=XgO(V=0Q=!^1qN(uigR8sqCJsiriTuP*8+H3D+>o zApivK9H1iiRZk6YI*r?J5fdz?MAD`FVmV_g@>M%ezzD3Mm6P6kq%eVVW z9Zwg%;|y^jw#ar8Xve9Pq<{cv$(`8A!Y(e+A>!l1*(hUf1^`WNX$}Iqs;Wn?Ikdfr zT~X1?6)Lgm?2pxGU}^a}0IQ;C;(E@BQw+-kBc0K7N;O>3;3pC5P(C5=T&E5wSu1KZa9n?-FfOtr{_^+ zDsG$W`7&<8F&H$7j5uH8({^t4rn7Q;<{=% zyKFnoE}~R;Q(an2UR1>)7Z zmgQR3!hC{77bS03&BtJoC#e666FU}`7^dz&!~-d)Tk?3m3kTI-?x=PgH_DTtP>8rD za9?z{tiO{2FSY+X^e?cTE&V^tji)xfnSN2{*#wPS{& z`6|m34wsXer|x6+Jy;3G5!omo!bPUz!yn?dKtqMpc3atuzWQYY~R7b!O4lmYb_^o z{3R7k%b-#k6P0`iG|BI>e83&hBus_b8WhyJH!il!bLBn$MSr#1Z0Xd$hUAfWOqN!{tYvn zy+yewcme3Nd04E$TM1f){hnV3fBk^PCVUjq>3JLxYcHh(;yS43=thHECuxNdLHqs` z5%0nVetjyuaK_^qcRetU4~awNg|JC9r{n~e z*{lpsWo*o!*cDQdQqil&y{Cp{Gn~(jt~OfQ<+I8@dwQ{%x=nZx_^;CtLoSmv^`0^K zYPL!uJC$6nCH0=~%ZTLS)Uj_QfLj~J&qAwJ3wCd&gA}dCqB;GdU`GK zzu5^B{!LtwyyJ3THzcr0_MfBle`nX-HX(kbZno(#htK}6$3qJ}Z}Y$3jUJBii;b%< zrP@cKpMBlfiaQV97KXlZ>&~k8Ija8&>E;*6?ZqNk+?T%5hN<PL_m(~Km@Lkl&L`e=yE@UC>{8EzoU2z7QTy<$#S0V14Io>7`1z62 z84z|Xi{@S9>-va!R3vcbvra=Z1wk z;0>{fn9>lIzI$G?R7jt2!Vbm#^iVh>u$Dr&-4MKY z+CeM^+^4qP_IR}nX0XR0n;e%y5z7PlmrAPEI~4wsW)$X5m}~6P_gd$H;j-vJ{Cw%% z{O>Q;S)8+?Od9NN8MjX1apEAHU%R(70_}pzzaW@IGVdH>6lU_p4HQhBO`kX$iNKLI zO>y&!D|N&R}w@Ln(sq10%DB4n)tZ$&<#|J3pVz=y3z^!G( zR1d~M3K&AHjTo>_RKo=hw0$%}wzD8kE6S+j>`agvV*iFJn$@mvPkUBF?rj&|U#a~H ze_m_{_A&l<*VmuzJ4o_&%Ld)Pd>%x5btC}(#U$yLQKY0y_)g`rDo-(?dZhl>c2KaIRx^P|h zUNUETR7Pdha(7+$2_mS|BEY^0JD2y?_yb}xlk+6$Qs`;@;P9n!22f*~&%tYN>i*#R zc`S#XvL+j~f8>#)uYkZ^9&_l&{^)(DLu5T~6@rx{7hXL*W4Xuz+m9D%tknub5#?XYxns|Mp1>b&&h zsgU>VDLzg^(2C#vZxR*>g?lAU8tw|aaad+F&z%mzw^?lz%Ri$6>L{$~5KreB5<}?| z6v(WF?9Hx6HN;a1cNTArfX&tTxdmEk;QKR=2x-wJix5d>h` zV)sV0jpZIA{_OOJH8Vc+TjnUSNy-ome-_!@qaf8tR$STtr6Cvp(U2zUkdB7r*GsvQ zKKSCT&z94bukHIbdKb4F-(!+gw_`4S*ZCd@TV>dSH3Ukl+XrI5yVoM0(;}abCEt(T zfygDk(IlRw%y~7ucJC+6ibeXbTYBGJ9{Bks)4X-*7PvllsCY!Vg$zAPdJ~z^B2Ay(7q_zh@`oq?L3F#(;``gog3@?LBKD0TjA%v*^)J_Lq1m9NlG0Y| zd*=ri{P}vDu!0_A2~erN3{c0+K%%yb4MKKa53ge z&x^`g=O3$8AFD+^=Sml{o*(P!$7kTYN>xFlF}Pg3pEt3-z-$}tqHA7pVkmdsT zQ!=MOv++bz?kE)Z$CuC|tX|cIk+z#vwZ`p=F0v!H{1Fa z(rzzf-n5qezE6%1^qO*Q-@_H6zS$G0)6` zvMdGdkl8gs`QG9=Hu22dW%+cc3PTHSb~R(#n`AsA3x7~Vu;WES<$-Emx->f_&M^0g z&{VNZs;y{nAL2u!i$gO|m;?jY;KT`qQ=WQ#Ikb*)^s~bi&m8P$L#wu!(BT^ZeIiAL zYfKG7Vhb0=K_M>HVvaC1s)Sx=Z4B36P)dpsz2sqCG;ac_i@E82PmcR@(MXKcK7!;nN!*E^Z`i+0*4gJ}$fQOTNz7BwPc~AQ4pMGQ6 z8{hu>9eGyE6-=(j0LP(zd1?> z23||n4(r3O{F-10f}!Y}%{~B>6ze`KtEtz`UqhG}+2^#pZ1~re900jN*r%uytGV4e zb8NbBjU%CZeMZ5(JI}2F|BFf+z6WGrj<52zLp6Wto`^tO9V;K-)`io^!Cq_DaBal; zjbU!$^Y?aa>lfTHr}FcLU%o=)7@=#@33S*)Ax+px1{)bDzO)%(B&;9^9Eac=+;Wmt z6{Rb(u8h|@a^`|=BFp|mg$wZT#wo=vXu6LH25~*e){{fz7RJ_pso^O3Ccy>m-rwdYzlSJZ)mZPUi!}f z6W25>96+UnQ?qs>Z6C+wGqCl z^D-C1w42yTfr=~fLP|WBh@SVkDvcJ|!5Hy>RG{t7{wX?Ebb(h5fRIOJzyJF z1#=?-6pm*{blT!~4)P|(XDg)=rl{rScOXri!z!yxyt(H=TrTXlognv7ioD@j?uT49bCM`94^+x!3B@d=(xU}B zMp5YOdfVGq#;T9VFo$N%2ZVSAo%hN#5cgEXag_M@uwI=Spl2R@$DOu&=WS_2`#I%6 z)hm6y+7okT>z-*xrJl*{oFAju8VQ|A^{qDQv-zL(ZddV(L7r_DfmAh<5YbXBWgpDq zcJX-Yo+t3Vgu1Z{|0}cHu85APBs(qJmo8>#GLv8rK%6W{R-@^hq4fg`q5jr&6;vno zg2vY3Erpe$*5keBggljl_7%Jyc01HTI|D5R+d*y|ZYY0-hV_Ew6OZzZSR zJ&D+#`nacq>FphLi`;mnQrA+=mp<;Oj>Vz&*SE4)h_Y+_0#V?C<@|OhDv?>nO>FEg zsJP4a=w~cAW}MakqjtT$NKkFkG*qG^4|{?CVmm6j`pU4d3N@edQmjuAL5Y z+ZVX9jqjsY$zth#EUP6skZz{?dNQ<uXIE9HIs(6z4&rb6no+(5*`vA_w9fq*#E~Oq_I*m>(-7GIwq>HXBx+$?c>zXPdW& zhnF#Pj3^dz{>f~bqzB2I0xz#*ZUjn+&aL2sViN59b8(!pzaFVYR!Y1adFTQ`hm5_! zNC=j`*2-1Oofi9h)UW7`WTBiO$EyK-*E6TDm z@bRYdHyyl1d(|J|Pi*^1!_b!pUV^>q3Dn!PQJIL+D&?s=bQqtztLP}+fja@8fY$o4(u zJCnTi)^W&N(x zrQtTgO(HKhSrYZM!a#T85e}idYZ9p3A4TIZh!M%i*81n}(v0NZ0?uDH4et=8AE+Dy zQhoxkFh(vW1UexQD`kIGhSM2dmPWr^P0qY0pIu8|n_51WzE6N;Pi1R;N7}Ur>RZG4b!OB^%^RjkjOsX47 zgM7$=JU2W`RASjv>th{-@V4o;%s5T$#x_GYAvMeoGXI%)5{sC69A;AcAWITRL#!as z@;$Rtzbh0U4DCPZHcuWA{R|RRipABJLW9O`(!?skrPrhxhRp9iV^B1u&eyTw=#(tVeB;T%PJmb$^W zV9$750un*Ym!d+@tE>h;iWO^>6qY!cMk2PPL59e!N64BkM%=yL{ao-4rKNOXpQab9 z_uK1?cACyH`P4WqzSh};<n=fb#-1voFDPs%Jl4m309vm-1Zlm6)IgG4AzUUV^+NTKR3mWLku^>ps@jkCzfV5 z6>Y34#=V|9dB%2-c+3>dOgle8pybKg!ORXu_h$#ANuUvPNBX~~WoLIRUsUEU=3@9v z-g!|J69rETeC)>WP*^g*d5goz#ou5|9Z4mdEt0bS{=jn^(RcWUPY*&~e)xl$^e z$Z7M!)rzPiSvcm{@qGS*)3ZM2YQZx(i+^NFkvFV2BU7Nw79QxsZ8TZ`Zp>ic>7KC} z=4p1{=WQE|Bu;$u3Ah!?o?g1tgs8t@{b4@i7D%9+$DRaO*W z&71wECUV8W^6e@jEB~Y@{AtugJ=QbeDL2Z*PD=#up2zTL5<3P{->I_-6;0$ zGcbS))lh#}%i3ARTnQmVSfMWAa0-X#<^1H1S|XLhK$l&`j6aG7*KS2RvnxG;axEZh^p^EqBd}KwWNq3B-9bQLoAI!`7EM(K*o+60G^8s#Y zMo>{gJeOJ!Q#;mFkv0vnCyqDVXbwKFnuT{$BAy@uftB!#fbu8)#C@20VKxZ^vm?yi zE<&^*lw?R%qJO+13E?4e&kzzJgXDUOtq3;OP(vQY5n8gDnb}WwrIuAHYXdhO;mANg zvKWA!^DESB_5>!KZmTT~TT@WRY#DaI%j}>U$L^W!pKq+{-xff?@^Gut_Ilhj^2O`w z@l|7u*K22mWJiZkV){Tic1rgeczj1-M=yg9goyCCte@i*>n%sWex159WhUA!|J0~s&?86*JqkV zON$ZGH4w6Y>fN*4H_8CdB?u5+0I^hVI;DKc?Y*QpvzuwCUl0SQxEG1t{wuK#TKnq? zvIbtP$mKqI?ZFle-De^eH`#l=f$q4`q$(d5J$Nq;exv)01{z64mG)bJ zmbL@cC)!AxhE<3$oV)KIMBVv~%YR2cyDRv&51&tL2PDrk{wpS;{l0B_Q6Mzh5$>Y2 zRSH_OF&>iu&{@SIbf1Ei(4Q_cyq`KCyl->v*H*)?;zh`=5;aZ{Bxt&r&p-!;#W%3S z!wW?yD)a7@$O*ug)bKvW>Kx0W!%P0!#n(e(0}df8$7q14fXKJjvwVt5#)8~_*x>IX z4L5T?`Xz!$``Rc);Xa=*)v+y;!Hp@9i5d1h|bc0mKIh&E3n3!i^$-Mvz%s@5SnK8vp`#dU9eZMH7^@Sv$|MGtm6@E)074q7gQ()2pIgSvAmoASf>r`Zuf zpuCgVCWXM0w!$a9{NUxWE7vBou~j$==iLfw{tE8DXD-bP|L{Aidf&Oa{^`t;34Z(h z5e#t;G>9D3ej3ed&LWfWjAID{4(b;jsp5I>^k;2w4GniCBpW272G%#Njv=wxS-O33 zZ;e3=bWd6NfPdc=GuRc*9Cv^1Kl#n=Omo-U%-XuBJAhZaqK)69?@`0wPd$x_fW6ew zn&L7K3(1IyPsMK@LG6Q2cOO1pDaYS{plM^hW{j}ph*Y<)a{T(;DPWYTb_3F$Sb3BH zKSS?N^b`yH)q~aqA|N-XXf#rwickG4vQL?bP9Sot=cw2VShumeO)40eL!ZiOb5nJX z=ML{VCJsXOpvj?zMmLC{TWk7_UP)cqd0!7vV1}G;^{j?1O@*eXRiO(3&5>ft(M6mL1WP_RSRnd7&q&d zp!c-x20@$tB8w=jsZNr=EA(jLY7bmkwr6fjrya3#n<_$*WdN?x8G1=lUX8O8gzoPS zx$19%zn-KNe#Mbz8nP>&9`-#~-T#H%QYqZ4PCXfB$C-&OU*gl!eJz-UTA#U6xcnQ4 z600;jHta49!PB`gCx8ltzODYSB!?9OY#U=*2i^?-cFv^` zCqYWvGaHTUD{fd481%G+ko77w`2-_a$ubjD&VZ?&0CnxKwE6p z;eFg<&}$9%>6x7jvkPq=oaOe&4b1lt?3yYLS3i+r!U1E1GGHwR3Qo$hJ~@H~HT9 zH&;!B3b=SNJYNs3{KsxwPnYP}Tv)_TgE+1YFur5c4wl`gXJc9m;r&p`y4&RuM?IhO z)Vd?H{ALzEic#qJ+iiNWa>1#JRUywpCS?f1)yn+0g*N`JKY66=!@P@76l4{WCBQ8* z$E4<1yXYJy9V_cKo=f3LEPRmHq0qxBVAKhDt&~dQtSEFUnhk9iOz8AT$F^7VF&^4# z_82j&ePPpj{tP52i-P)1jA!%>aysnZhrOw?zARCef-hmz$PS`*bm5zWQ%f)3JE-Qk zw;DW`o}bydk(^R`_NmJ9+v;cj7#%(EPnpd1@7SR=DFcc`MMVcKS&|L~ujP#+qIX{K zTCGAi*r|s&{O$ar(U8M!_yub7{pq^lZS1SU54AlzF9M;xHo(*uJ-^>W#WB%m63466SctmjPk6$8`LZ8${9YmPTKB&w>aDqdAV#dq;UOi!SyRn+uO?t7DD`b?YfAd0h)vFy6k zYIS!}s~zo4k5^GP10Hj2+~8>Ie6d1Ygy;DwKJ(|8qjUqmpLl{%=Cxz9amLhlnq@GI zbP$B6YMhrjm?^!_;KZ~%7{t4wWFfj#FhG2lNN~M`Xf7&dISeOAD6JDA4~#0WSQ33i zb{%J>u@6dCImlG%O@q>6h$pSkf(aIRiL?l=#bWxT+?h+R>v~-shmgZQeXkU=#E;Lb z8qMRqw#-r?>3%AU&;ETsT-J4gygTzMG1c4zC%f60v8&$)*$2WTIh*X|eo{xEVn*hU znOc7*$YB-qhb7kIGdf=}hwN_7*b!TsJUxDgri@qP>+G<(AK}}3vTiikCi6*XzuI^x zr3L7W2ps$}rKxdBMb_#!D%u7G2a6rI<+NPqDvCe!R!5v0`5#Wq6xa>j(eHq6T6C_= zf&#C{`1Nt!1@q?~b19LB#;Xjq0p!fdrDPAeg&}1lrs_s5VhQkkGMPBkaB#V;Z`kZtrME!wJ`0 zrVkvVZ}h!7+c847!Zi@37EVXVyKb1C${QiB{)_Kiww*d4`>uRH?E{IY7<_P2C-M|| z=5;jTL$qIWQ8Jk_*06O;W|p`L_H3B?elrdKVB`9$yyt(z7k(n!GNHf#+jL5qryKqU zt8j+o_u)UP#>K|g($SHL7rv(}*v}ZE$UEbI*xpOX+gQQo3mA;m+5NSrOb05ra#ewV zQ08QSlQq-lNk%6wMT5+Y+fr#GH>#jts z^YI+Uef`+Oay#$ub*l-V)0pPDSzt^YHXWYm3*;RMWt)nE;Wt@vLTc`Pf!!FCrdGXH z&O29)M54r%8l%ve@|<7s!$xeCpa!toBLt*VpuMCwM9JV;D2s7e||Ui6NL z2r+8BVlO3l*(fc22A7U;pa+U8c=(kE;>hHkKx$i6~G$d_?Hpqe+o zC4Ikb9Mw2(QsGfa<*RB}ni1A643uqCBfHANSxF$NFr7ALy_G;5o1|Ax+&5Tttdp|+ zkfP44;)*6OEQ~et$FjrsD4kOfs6jij5-54+Zyv=d&RwnlzFu&qol=;o&PTX0Doqzb zUXr97Zx_9Z2zs@`vu7H}5Ox;_hCm*h$FsMK-6gtg%@o;<4f`D>EH4weG)+zZdEM*f zr7Cnvx0o;z;7S=Ut?QUZa0u8DJ)#|U-DxSnhj3nqBJ{;EpX))yqjx{L!;s+9>2kM9 z;zNwwyuB)0Z7Lmv&1G;d=HR)SNS}|f{x@O`XvGlK9g=X6#lHvDd;y$lQ4%6GkNsFLjEJzSPaUDl{D5?f8`s= zTA~P!&smYwy`?s1C`$;_N(Rhb0qty&9^GAo=6#}EtAGWweUtD4+$mr zPu*IObuK%*2s(OV<5rz$%@JeM1`|<*qd^SSwau)e50qvXGbj>j1I-OI8klvuYFKmn znRyvm*l+-g_4q54BLd7nA&8$s?AV6_j@{X~FF?_y%t)t@R@=ay@}@(Bq$9UKMrWt) za)2DUlEtf&CXVANl{F=Nyo61HL9S!7Hw=KF+MKECTOW$4+SX~Dir035bTK($fw}T- z-{$2Dm*ot`E#=7%N{BAfn%jGo#9u&edR-UNj(m(urt7q&;2@ZBZNm4;{^~}m0VO7( z3Eu1>Kfq<^nvb2xNS-;>`(a*g_AXT_wdz-ODVkgF z1b$39R|)RMzgLbbF~47!wHz)s<^r>=f2c@p7!U-)pLrTD&3yxeqN^CZLYp8EnDx-R z7#bJ9H3L+n(32ZO7`8#W0Xr9v!-63$W2Wv(hGg zl}q>04q@N>82ZDfs!3ItZ`v3z4Sd<*E8iI+Uv9^ zeQ-K6VoJizE@ouup=5*WK61O0BNwEH!@`2~kV=InJZX6GbgO((_?X|3%?adg+vlA* zXQMoYLM!evj5<#fp^bfy?25!CYRh}O?J*{&?+!a~SE;5#=6k^Gx z_n0TWcfz?2N-(U)lo{3qr(QM3JfliuZviP2*HbGpnOktS|E@b#5#1q82P)XB;Qy^V z*Orjo2!F_$A9973kw&8e+3SO;5ZIcY51zx6bYS=il(!k!1BK+^(mwNkcl%e#o!Abh zlh+I63?m1!lk`EvdUgcONf@{ydEMmQ8psP1&uW{4W2(-H49f(CPnSl_9uctMSAc1G zlru>yvf;mCqvLg8OGl^+W_U^j;M;4Q?(q)9!RZcA|j}>-m)x>%%^SOV`)0NHmp*cSQT9}VhG?O3a--%?G}FKth~B;dU0yFg9;#X zw-!x@jryLnbHY(PsnSx4(wBa=K9({}uA2*8wH>*K*Vo{d9Xfw`05MZ1Cp(XG=vluz zmH#A$9?{cG2>gDw>M6#f$%YK|rbc+i;Nnkk5(J_cZB%i63c=PMk)G-1Yh{ytj|l=h zrc1hhg??Y`;t07?2g1$_Zc?gEb7Gw68FB%T0-|rw86a6uKSH>ZbN=$I)-K>{uM7D^ z5{LZUDhe?|%D2-5p})TC??a z3!@tvD%-`eAHG;+l;wzrx!Hbu=n~f-ULGTsb9l%l>MrQ!rrfxHC|B79XKsR5lMXRl zNZIaGdH(LNMQc!w;Wc;E>0VY#to3h$AZ)1ra3D%LvTYc^EPi{DNFNkunICMDAjXKQ zn)r-4{Ff{fooHrY;1uew7_WNxt>!Kk%={OOy`SXyET7E77{y5pF**49_Be*>xl=R2 zln(1QNT|{zC@iz%$4sVz7xak$3jmjo?A9pqM_d-JcWh0Mynv{(@!`N+v#bo18e$C* zQPxj)T!VC5`Cs#pfy8?PJb``?4xy;TyOdBNWZBMQg)g@A>2bdpG%H3*HbUX9TK6jL zoUALiT(M|T7jFqjEM#J$EEovpqL$c*KZ>~ps* zGH{?tqDD-GS5sX?Tchj_fxGrLpEm8R$A1zpBl9@mW1hzG?f)2~0DDHz&2kR^v9Pp(e@d-z#k^7D!|fxy65T2e#3n zGiT}RiltHTJ$9N*VTJ?jayanJN&e7^f@2yiMFnc&|Dq;*N1^WH27yJAU`QXX`)D}i zS##yT(vN6vHQ0U@yV`-%YuYI1#{zoHHUC72dkQ!d0uJeyo-XHc@>c!!LHrEQlg)zu z(9J2~jykvC3}egJBU=-6EgZ4H`=HVTJX^os#9fH;Z5)wVV{CETa`zn6E`7vHPK9Bq z1ejR{{_f3i5&!qn&sxVZi`J|f zfQ*%G=MdD5{RZpn?Yp5cdman`2ta-jT^5#jAN_JLhECr#nVm0QQCGj>3z4qvXYUY$`KDR0njim8sUh)`YlcE zU6Hcitm0$D^}&FjF9`!-2NI+B=wbZ?5fsSWdQA(*n7V}D$Il=V>ehPD0>6_Pf!EH9c$@kh~gR&iz;RAr7BWq zg*-+l?08Dt;JNBuL#?CQ2CF6O)%dH^Vi6=*s2{}Pqr~@EtYpjI2-A-_Q3*1_8p6Xd z0cRXx0$sgl6sMQ<3evt}8yuy)OH!BM#P(6kEawF_!PLogCtGz-J|K0}cwqiHT6VF6 z3q#4e^rKP&I|Wx40ps6iqmuL0^*I{mT!c=+XF}@6WlZdP4>7iQK3Ti z+A>W4V6X~!D@&^m{TW>D$PZwh}MG;+#T zRCqjI{u2*-vxPeQjFt(0)ncI-HxU!Cr)+!M$P^voQC{bbIokr`Gcv30#6tCBtp_wc)H#{+PEqE2+YyRMZGNSql%BMQC za5FN-&zF|BiHd}5>KrHAyNEqw-|rl!>bo)|LJd?TgSO|ECOQTj=ZN>(b(3E2T!6VJ zP`gDJhm=G%_gM7SkpdUE1;MM#2}LY& zbq$H^YzTu1$oE4cqG4gfp>i-cdJs9T2%))KPIP)#XfW#RwY%vbedMtN(IF=XIOE#A z3`a++YpXla`%bM4xmt`Jct(Xf9>X{j4NcBTV;2j?2r%R=u*ooyf#0QE#f~DPu~Tf+ zQ_7rR7dgUUn2H&#i!eH%a79%x>TalmKhY|oooTJ(_YUtww2l+TO)yi4&eZ$myCrQs z!XHsURyN`-T*0U-tv9PZrT1!4rJC?L+R5%+6mT2>Q0$3_tPb!dsC9g)KV?`gRcQFB zwt!1e*LCaBd+)7lCk_e>u|eS+dTAZxawmc(cWDIV%$u3(Z{53^<{O){%2_Hhf5FOR zv}hogit>t+pHKG74>7G&-beDU#0Bw;_Ff0g%(8FH&o!8g#IruPlsj68r_D+bS63s( zRjsbn_bD4NBRw5g3qQqcrF&{;CDHtHHufFhAruYZPUrT%?dIGN$v?iiFFFa+@D`_P z;6iHc zd%Sv#{DUOBGjWVb5Gu2Bn0oIxEV;E{l-?Dt31DLQCblHJ2)qq3#9{RvQsuD(@EV-U;r|I6o z60iHlKnf}s+CWT-2k(Gk&kq>zErCEx4p4&ARv28uCBfZW{sDJ7pbIxhWwLOBTVqWI z)&hq^#ss<4JDA5ZMp#NaKpECZ#nYnZs-oWwrdC*F7jepCMJIQeWJirrf^k-KQWY)- zgLgiSA-Fi*;3dAmz^Ue-qSVPLnYvQVT%S)A7Y0 zY$x{cOQoS9v6Kbesi}Vag-#Fp@I+$!>V>e-RCY#0v%E1(>jX^fvnd8xcV7VENw^94 z&Z3Ofe+8eP==M3}HQnd17;mlRLTM4tg~gTolqYkWqf7BUB2X{&TOmOOuA_f@?KM<+ z(-v}1d0#qS0e4}aWVs8>W20GE6l+O{9!-?H^jPVV73kWp66+R@F-a`vrHE0|tJ+9b zk%u%XP|S?#=&0R!Y2_e_?P|=Lfq5d1KB7Sl6|;jo@Qj^4X%%# z0^vp@);!>AS$6-<6rdQ|utQJ_x*KHEydV!1jSdsj%mb_q+I5XTb1 zmC+!fl&F;}GpAIhPM2B1qGG-l;iwxD9#3PPE;WqnH?qDB8aTs>d?@jtTw3~N!YYow zj_kIV%fs?cPlJyl29L{a_w$~u7Oiqf$*xL7ZAxstA*XsSY#3 zY+vqSgP?fye{_h0ip~4{94JX@RD_e)C*%_Z-nW4quAt}i*8OlZLJB)F7!QH3*K)?~ zo6Ysk+1eZ%)Sgk08~70rUk?3_Qiyj2!3}DJ+Vs%YDq+ms48&Ro?_&)oJE!|Psa@Q3 zPC@g-+rE*TS?ebXZ&0+MDhK~wGO#qvb|j)zndO^_zQuG-bt|9iuHaW>BDWc43tMT? z;4<|{cK9KY`g=(9Wf(**Q+5j|lBj{q90$%Oqt^6{RBV@t*M|1$<3=34ruI5|c$#Pc z3AA`Xz7QbedJRb5W0wtN2{@;~aBR=Qzo{A)F z#L!<}_?%Wdgq=M>xr2-Uf1(8w-;w_LI+y*inr*$==HweK5?&2Vo)gLt8p{uh)!q+h zKY{v|FGu*Z-2N!hG+raTDPmUr6Ms$x|T)E0k$SE$FOa zSWN%{WvFyY206wVQc4_#KYmC`oOKfGHoqI+WVmF20Vi0fGTH*&ucx|5%eHaI;l8xs z!p+)iy1ZUw5~74dy-K*v+x*8h_6*J@0=CSIQKmw)JV|;PDEDBik6RNAxEvY~l2W zgv!K`GX40)AHCnU804CRUMBP2zDhz65Spr-M)#D^exsg@+<(8))tPj0&gakusq3p~bs>q1{fY&iMQybq%&7jg&mXAoMK4e?Gv6v{_!C?U*B#$P}o`D1f&?A5} zKrk`1gV-NjbZ8+GEw16O;t5H`np%bPMuP4=v0OBbqfHp1Oynmh*^kREgQ|s;1#?n& ztPC;$%IO@ICljuBKD~ka0|c|()RYn1?{vpdG#;abQgrDa+&81z(BrBuBb127F5j;`=1^M>Yrb|7K)wCHJ$Cat- zpPs4eR&@L=A6+?uI>*vtRtFyu9$~CfH2Ja2-21c!ROI}T8*NI6M>GU%V`4@XZ5#Gg z8m$)A;(?|Ehfp)yvU0i>nc6Eb*A3yZa-3%)4!sq66N5zc(nFUptE1?-)EI8`bqTU+ z!gu6~6jzUlhmJ6>*9A0FG{>Z)czzW>^>A;Pn>^teZ|qy&F59;ss$4cbt~xuUV!i^I zJa^N#?u+0fc1FA8`1_uu#~Tj{p`bZ=d8Ke=Zy-%B)*4qx$ICUT{;zkPf#ElSc&@x} zTkgCh{<zy+dm%kF0qTDV|2N7GW*&daY-gRz=pl1=tzVA! z-asfbf+u!922+JN&TzPz=xI~OC~_2WGP%EzbXD%QNb}_k$KRYbnN(i(W_Iw z!%SLS5F46|5^0Il42qsm2R2g}aDRl;E2;sD=`h&nC9wQk=k6n4yzg6342#45|Y`Sx;&Rn4t2ZKEH9gTB*wE<>e4{AQDhTIPT@$=5Mz5 zi>aZpF`wR)^n8wB7=BX(ax4SgigiRR4yW(1zdOm#-=e#_{o=^PeD1dgl2*~4%aV9T zc8$DiK89~Vi7Ue6BK)fD7K>jPQNE2J0W=az;T4(U$y${Fp&h43;aU1ACxWsfkLK<6 zvygRIfI3G%FxILS2J9UV^=a5}6L96k<;$k53pGlh17 z1tBKKNpM1J(+vUR8pO$mH4!pLG*qT2qko5Hs~(^W!TX00N)* zdH4P6?)&Hd_sf?3sH4#S{{Cs8#>CdgrAzlNqKz4rUJS16wN6a3DzenSk^2_X zDPWLY6aDUZTlsd@^UA;N2W5oX@hqB$8blTmxGN+GQaulo2QVZOU|IOCQ<*-sdwSV~ zF5)ey4u59!fQO3D%e$ULulYJ*`sr?}?NPPk;3j$uYmhY5Ax`2+ssqGGgy|0KA)%v< zS+U$l6Pi-7xX``(9_GmAqNBu3fCtCiL_(mCFfXxS^J(Oa8dJ2>-4?5JzMs9iixw8g zNJ`h2pVgBd=m}9 zls5so1M;MEWFpf)=Cvv(nxuh)NZ_AC4=l=MxI-4(K1)dt((m^>H%35x*P;b8LxQHc zSaPBgq{~62;sA7VTeLEzpxhB6HqGR5%r|Jps|&ZaAz6Vv@Tn0#tSBf*u0n;Wob(xn z@lsxM2rL4ysOvM<{c9HcrrItpT}8Q?Dg@>+$yDD^WJ>=yGmX`qy5fN0K}Y)Xa%0Dw zbNwm&(9`un`=DbHm?!#C9p?T}bA-Pt;>NbIxJK+F9Wxk)L6F?llXq!mP0MLESd23W zMzLJ^T2#HNV`xU^Niq&P3#K1rz9{-G+<{I@&ncNlt!-)l_xjBsrhI#5^To683#fpl zqg)f(1+jUn?r!J^uR(I>M8xaz&^2KK}NI z`oU5~yv=FlvF(*iscXd$z(nBPF{%8qFep!T`)R?E_PZ7YK?qrSh8 z(S~dl2k`w2c|4zcGF`UZrUkSOcw`Kpo&%fgP4l(j=ZuZb2JFIp69pT-PB4>otoA#D`)j$n4=_9!@mqieg%tQ??w_|G%fZH_0 zMi43~no}fM>a@5 zZPO?+()vPPv+*(SF$#`LA0=mYN}mrgBXphAs06D-Pcx`;ejn2}p~b`=yZ=~Pp^+}B z=u+A51AM4=_WC5pqvHsPV?t5lIfY7u+aN1B;)3&{VP-kiHDki_J4zt(UXOw-@xB6Q zc4TT$Qj|(xGF62N;JTzv5>0kt-p3RZbW4=>V!OLrtvt5aL+{3Nslxn&_O^!Ajufk* zV}mFB0cFhf$zXa8^FJ8nAGBxt8$yHeIuED{zb=e|b-?&1sD-Xb;*p8(Zt|Cc*b7rNp+Sb_C`V_N&MSk`a^_60> zbubn}Iu797LTQ4ows)lH7yc?n4Iv}BV?1+I!%Gm4s=~$|xCzKe$}gdd%A{@z(ILBW zuL-kdc`tE${9zo}jocAV;?WLY zfg`Z)Z)*)8Nn_VGH;){_QnBM<>rI_Wduw@Knw)5%v8BnMFx*&E@F4!sgD0atRZq_~ zBSnI>G6uFi-irt^Cgn5&974f!1%04p1BoA?=#kx zyW%C8gv_Xp7peK4-mNdLD{)Y`*%Cz|q(!Cx?_5J6a!%{?H7Q;>`H!}FgF(_e&}qIl zxqbuCKVVzDAVf|>&Bb&@!Ls$`=e#W!dAc&1A5IceAkEK>8+!Qktk$eK(WrWsABdhG z`9XvlZ{^Pb&_anY=-`(Ni@g4;uH!a1SLhbuY1RLb;chKzM)_+%!~x-M7Gx+DnCMLi zK`9N0JYX}E1aiV662w(6-BWwM%`2w6xIfM>9hR_TwE7zj?dIcN3!0oy0f3i#@S)?d#*G#Dmc0$Wt5rSx^KsJ>f}=(WbX!E zl&z<;1jdsX{P)tnlbTBcGdx6K+gVk6QI_|%*OT>Dhvs~Ya_lSxp08U-kEPyS+CLEh z*SRP`eCROErbuvC0f_*w z%O}{arrVU>K5sYAKQfvy&^&jqRfv59=3*9a%%MLx^+zHl(P9pD%}$LK#_`G1Y>Q-l zo{7J1k3I#$X6~+V|Ls{Rx{zv42z*Gn@{@g6HvUEv8FnC@ds|0Gyu-%8puGqxf6%@M z;h|JvJt7<4+T`BwM6Z}%*9fL`Uk%C`<+mKeFm>LHid;1afIj6Vl5`@8FG}q!Oy9S` z^o+mlG5ocMOSKlEh#T?N>y5{2hq_&v+*QMkD8>@0k7L!2Ae7Qc$WKq(cRl`Sc%$R% z-rlwidS)&xWN{_Db}}3IepHzDTi2oSi?ND+AGVqXjw=9L@+A1J7aiC;LeKF~+GM?! zjyw>7z;v^5<1f^9PQOiGmR;kjIvtO?hC~i3_QV35O_p3+&wh=#yDGhIgY?d~V5+ys z*OU#|j^n{UbmsMQlNW*ESWdssPfrjmTVqkVN_aO6urgF?6$x?sNc3Cd=xRLE&n{tf zLk2OPA}^>e4!Idy^m>^;sO?|_6YQ*YVxaidndh>Ri0CxzwjRLxunGg5%RGKp#Z6;- z@OBE{e0n%vVb9V1F`PIEjF7=Tv|pyzZ8ZNoCrAJeRV$rQj3Mr6p}$%qx6KTd$qsgh z4+`s@s7S=0yK%r@)PzGUj7bIMU>gCK^)kT9&yAK<;XHh4#-`SMIEUH7A@e3)>7Reo zn?E}HY3suwz=C%G)bC>zNjl~yxIY~Z;1slm8i~q$$aR}Ao2?tk_Bu)9!li?+$VjZ? z_Vtw-eoy@Es%kusR#cGM65VRK*+VA!ry62qBL{l;xJ9dEWq!V&?iQ6JzH`m$ee9MY zaOs=GL1C*c8cV=++vjYwcLE6Vdz_7B|CUN|U>JycOw7o?0PuO8&kpn1r1={s&A6+* z?U4Ppk=%LyYufv&`F@m_;Lf7#f=_NT))H+nKU%ogUc)nLuvVM@CDgcEGaK?zLtIw_w!T0<( zG%|yFy$@jK&MEB-TN{UsjwoJ1s^IGoliiB~nm3mm&rC%Vasv=9AROo!2!?^<{*J%W z$u!Dm;#7Zzb(0>5Mim+db7&lw;Gn0jE}pC8cPk_x0Fu`?p5S^KRtB0iT~7BpF2hGX z5RSlnHi?E}|B_N8#7raPFQmUHdS#HIzo%Tj>n1Z{R`VKMK#eIz6=(w2Y!v886NR5^ zki5QzOms$!V3>D-L4bpj5Z96~(oRdL6~(88T$$lxct7x&|G{Y*nj5n@JA3)6M-cYJ zvF*To1nbXhDN!129jYK$(m4Rs3=SdNzTIJX*O?tprihBDVKy76!@QguEpSZJoh~pk zT`lA&G#qb&qbq|0%*B@M(l?u}3E>?eT9A4k=5onZ6w%&CX8n0cSZxmg6FnSA3roEM*PukB#E>~AKEpU z?k>@AeN0Sl3M`dwpqh?4L%g)1QfFosLcek|m52swXKTm{2Lhh{KaBV26{&Hc4HFaW z1o&Ym9&>*YT8JV?h`j}YFFPb-RUB5W`a?_brH{BtCor_@u&5b~Hiei&DiiI`>mT{AP zJJ=f7m7H`{>8PjC#t;GB?U!w5YwPQ7lQ)JJ7Z-yGPqFOGTuv`lPHz{@{b!y)d}v=* zfJ^v9Z<-iv!{h53`Rf`eMYJp`Cg@VG;En?tqY)YYg7>L0rTdfkAHr%v44O61)%`L# z>8&aIRVwc#RnPbTqQo46(|T<{J@A`o^< zmu+ltI%92rFDm}5)!erc{8__k0vw_WqC%Jnh6*IOR~@$g z_VM{nG}o6SML-hE_b812jcAAnB;thNW-+Bh3b@ijkUxohmn-SI8^_ z>wU@WfqF@8WQalI9}Rw zNEw->Q zZ%^Iy_FB5`iW7Remov}uezJqa&~6c&BMddSR8#tD=H+x(;Q&=R@%bUfRD#9&ZBMi` z3bp{zdNN8QK^ZZ*_FpdIj#PKC66KTTeX4vXh|MLF#$J*$|At!EP#52w#1kgQ+O~-lJ$&!)z1n}tqoSTlv#_EqA zXu%mN2e~^(UD^K?jc-4r33&3>OlN!kVN>>+74c4U|7|JigFt?hY~Mt=Tr;RYJpZTi z^&IRYL0+SYuJ6o?_AB>)J=za4=6RTqQ^KjiCip?LwOf$wpBf7|EoZ*MK(eIo|7budE($y#0HX-7v7bVthK(X*=pMV;IpcpaI@=VUq;qebip=5} z0a=T#K}lCF<}b%iI6w8$Sd*q=iE81Za$BU)s2@GIzP_wiad7*Ln;t?Y6knfTk-jk- z5cVCv{$fw8VzNO1L%xyuPB6H;Mu#Pt%v}*}5&v2Ftml9_z!0TTCF~FtdSH)AlHTKZ z_p_L%FiUlOuhxKG!m5PueYutyH}O}M>%ny_S$<^Dz5^-J6^v2L69X82syW)eEj_Ag za!=BV6DgP-qApWo2J=wKpctXrh`Nj!GuA3RBv-Bb2SQF`lmvj|gS>3P1TBzOpCg_? z2}^zzUza{-E$(29u|umG*~sDfIn@7w`FKrQsr`W4jqaiKU;qQI^@JriXa`8p8uO!1 z#cD2~`7`u@N`=U57k1XjU11dIUoDga!yx?D6JqBcLd=OYsvE=~DvMCO(B~&KgV!7r z_lGQ&5CBjBPZ6`LD`08rb7s_DdpbxgXJSmqt+WRzKA67LXuVNMdtN6vPVhTI&76gq zfQLx&%VZA_mOs~_Uh=%nAVP}x;^u$+3mx@!&Rs<`J~88(v4NfT7KFavQ}Y-1qRY)l z{I^JOcZKm&VrL>w=Zo_y=O9m#fg8;^V>AX*J(!ykr2y3Z3@`2k1cD56hg$(QpHmWL z;+eGr4LXv{rR%l~Xx{@4O2@7HSW#RmifA(rJWNdi-d*NWl(BHOPw67&`g?-yr$G}l z`C60pIwIGG?eq1nIMrqF_jiLlHt(15pWYZpmeEz>UP=EqYI{FbxpbCm9!GyYN2^!s zZEh^${E9=$-#_xa`5gyCU|;Yl3886%SgYlaK=95%ma) z)K`JxM1N(#7S^>L4Nc_!Yehhz_i-hKvxw+`qD78`#a_smQ=!B3k$6PExK{z=wNI`^ zPB4rf9y)$bX zJI7!oUSH$e=zui7+&n8f)`Gy;k6?s6qpOCqp4^X*4}o;f@KhMo@V$XS+Cq&O1IoXb z;LuORJYi92m=ty_IJRU6p8WHzL?XlVl;A(qYUWX4-3(=ll&HjaGL>ku>3XbkR;>ml zOEa{jO^AALh-O%^S?Tf^bTj`_U?00&e5__=KFr37@zmo$bL-Yy{U-M;6hp%qu-Eqx zojNC_JLzN2VaV9iq$J4$b7tdjg>F^EoWT3ZUp3zZSFkA&(M>pj;izST#M}@uA}t#qHo_}E$P~-|s2cN~L3gVy zY*Y6}z4d6!Z<|utC0V(PR#4MUiTV^4B`r17@9ry^5eMaVFLUmvkz@Cag->g2(~UEl zn&uun%P6Z!{7dwR4)+&`sP1;zhsp# zun*XB$Yuz2-eq2O{P;23KLbN=3(<8Q$iyyS`uOl5bWMK^YJT&t2M0F&cn#ez>D_KO z;Xf;&ek=g;ZXF$Ks2}?z({@^Is|txifsc|o8+F7iO(Kam-U{EaXB=ICslRXNIJvbe2R(Y z*vu#@`(W=`Qm9hw32%)cGiQeWdh9WP=d>NoeyB!Dk7avF^;B2o-k}eSrjvwF8*5Ae zEzv}MINtnwp($(J%-8-A<5eqz4Te^)mQVZyBbiI+wVu92zVVhBh?)Z%DI zMVRcELZ$Qa2eNhRfyMT4AmO7VeId4vxsUPA0n^ncFTawsRW!T;yk7>iUH_@uGXl#! zDm=|e7(3-pYJFS3W!7)^4m`y=We4^exji;5E7CNIJkq~R3_Jjf>D1BlGoks6#jZ}R z3M&nzlEdw^`*z;%nS57o8B8SER3u8aC)A=?5+=zrg5eEWx3o&X0>#oziYdoQoc-Jg8jcYF?{ zqX*!N&Oilj_~3czVqCXD6pHuU4BJ-ah+-v{Bk)S6Dlg0PcrKL}%Mwrrogc9JOMCWnvru@N#kQ6_rEkcl%AC_GHM=xhHni^Fey`gUoBvGF@F;(fw_bmU z{V8zoZ{$xEs65;j44bB@N)_4T1^^+`6+Yjv?MDPc>TYM++hyPv8K$1c$_@$~+ymLP zvo`NTK)x6LHn;2b&Vq>O!}GDvr|D&+WN=EN;Bzms9PPqCE^W^_88a*EZKaJ@`Kpb> zu4lcd>Hi)yJ=qK|Dk_>)l1KL6%Ld448M2D?<^S4SvvINtOIuIMn?^C3;9Hp``@3Pw zt80ZQ4zT88hx3(M5fQS;S^&i{1t@Rl%1{6lB;Sqhbx#r#u9tK~-ZAEYB7PA!k)mr2 zGs`?!Lc|s4%w&EjGT8}XJ|2!ZH2StYticr^kE;Q84Lwrx_hxf60`ib9;b5UOpL#`| zM>0;PKogY^#!x<01b%)3+Jc?cP}cpuE0z2sI_b&w+cY+TX-%VUXj_Rjf&Ge?JUW*9Odlg9ojHJVtv0yuoOB;2^D-|-2+p;gBQ zsCHca9^(h3yu|aw?aT-G@Kyit^QrzOzjXYA`Z?U-HQ!}&(biN2Bw(#eZ79aZvJGlu z-bBjS8H?SONsh+Jtt*PTm!oQ+!0cSa)R;NXNX*+1RexN)1?`x(bl1D7pt8*z<{|{H z@NK+%4g@JxupCByL;i+N7{PwPanOlha3micwP{!P+Od@pk1?^gX!1LMc7;|}PU^JQ zfbU_ZAkB0m)6uY=gDyF#ip9fwn^AyFmT%|t2`T#XY;*nVCHQH{qA^^Jq5E)ba$0Y> zxlHL#hp?+ba4N!J9g|1r4LdrN>}0g*xyxE0?mhw+A|!&`zDe4IG=uVd!#c027@lbp zRc3t>feHdwHEdiNy-kMr_&po@k@lM69dl&-%CA3Vll2^1T2*q2eI*wQaZCB8<8-(S zvwtH>RH3E}u|Zmm$(>_K+7%vG15zE^EOu+#4bw4XBoM`IfUx#}i1&Q@mLq3s>j01# zR_^oG?DGaBMsB%uzilpA5RWTah5~5>CP<(aN?AZAP4|`8Rk{?{luj{x-XZb(4zw&= z`(!$o%|?rzP)KFq0`U1W>O*9Gy&oH~R6vcup>8n-UBmrtwy^NyIQrL$7UlLA>r_4tIf%^;(K^)z8r z6Lzb?MJd5c`rxSM=RN6mE~|#>mIdnOUI%K1&bek&aLi5w*kV3U74N7fTceV zN_p0enIh20pFrM*Xnhb;Wc06jD^viqjm=77NHhz2vI**UIk-UK!paZewnDqxKpA%6 zz@1i>zyyG<6zc8F&mopcowC_&E4P7fm*U`H7BPCnkni!g&evkq(f>82qP5wNx{@Qu zU?QOK$<7)MeuY=F&+AfVjXxR>TR}6 z;q5&+nj~dc4-1!8{KqvtqvW6Q z5nZUzHgP443?<^0U}TMawNEEsx88MkCNH6YY5ci4<^2K_@+A=}=&ze8YvL}Wa>e6B zHK55zXfxrHK@U&I^L1KX=Lv+*U>my7i4K8OQX!R|S z=&)0v^qW7L{{v`;OA#5!cS=>hMTEQ+y0vX5-gA z(OU#in-F8An?IQ=gM=Um!Fb8dx2pF9D-~Gp zF!N6!TKVnp4cG5C1^_;QdbB{k@Xbih$Q(ZE1<$>T188QM+Dt>FD--8MX>kFzbBlTp zzH0B&hoMsrKwHPwfmruMi3l)u5lp7Ht zvM*V^;vYew2)X9ja9k~OKL?G0MIp^^|05~ie=Q6MyKGS$2F-?G>C4aD09}ZKd}76F zuqVH@P>=c*s;g$?5K=#LTno=c$t$~%n4&3!X0Qg{Yvd8tAHC6`KEj98T&@^SHeDBL zW!$cYmWw_xt*gjuEri6wMnlOv@52EG59f=Lrt4|J+e}FG`geDBg2Whb`ok(~cj`A5 zZCg-onat2U-RY3yS=fsd&VnNs_cYplgH1}#BMBMZi&41=xCts_T|?==6dL;|4#)2OG2X@JkO&j!r`FP zud~#qVT0dwFgHY`j5eYxR{r;-#MZ(}IA`~9%{t-ae4Dp0U8sH6AS z535~1hNf#>KvP2Fv&YGW(3gW(0*L>4Fjn^3>8{t zm?^ve>;iUlLvhH#rawQgz67I;r&s)F3-APp1U*y22mM0%E?bPYySzHIb2Bk|`%zGc zSiT*E6x11yqHPtFt5vrBElN+YYWQQ3rO-_LaU$LgBnj1eL|xwJ4kLs8qt_MjXk3<4 z@$bTJzf8FlY5r87hnau?Ll%7ghQ6bm{ko89=IPtCda84?Q5^@#`-7ngSn+C5Q1yA0 zAdvg=g@8nB;f@e3lm8(|rZ6a*kWM&ZHnBep*Y*8Kc@Z`yLn!<2bQj#jG9ED^sB~A| z76&tvb$M;4Kl(k8hy+}zZ9=aag|;tl9WtBJwO-10^F_#yK&3o1*%^4**T3n6f$Jvh z5tf|KsQjL|feJ5Vu7jRn$*6Ygyy#3{deCVb)M)I}*r2*0BG`LE5gLjc+%q&vco`GU zsgQ-|_{act)*3Fp_YX;Gt$@V~Kp9gGE#bFZuKL;Kd03dNH#Nso;b#1Dy#s?s%KoR1 zKSZ9N)Mx+#k}H1axchn3X<1M59(O?la{6OX-e*paHZP2}N!3$cji23ujAt7l8{~)3 zh$qDkcBAY5w=f)=NerG-?x2QL*C5{9aQF^vogUQrGOFbdwP|Z3!6yfWUt4^8bfh+o zBl&YEYrmf+bNMr*MSxlY+O!0Z@%RRDs5BwnzBlW!m3=a64?WYG$oQyQpmYJL(m-ZP z>4DC2v(C1ISF)>|-c(Qa7B^+6yhAEGMu~GBWj$RP@R}-8dLzNTBkD6BRRabFqJ^O$ zqF?BJRscNc@E0Rf;7!bn_iJc7F+5?BLAe z(X!~}BRJ}HbkpbayqxjE@`)S=x$;^ojwkku^~TsqZ09cpC2r9#o%Rmk^9$_Jx4SR6 zOUb)I70;~HZbl<^`gnZ+GOFQW%cG!iB$6c)iT^S>;1vVgKZo=~m>+vjB>afvlgENj zJbr=yp!vY_-A$H<0TWmPObnO`goGg=J&ec6aWc(x;(V*~O>^2EivX7TCJ5a3Ujvvg zq}ux8{7qzB!f6Pgva6hRn^0cem4?bO?WaQa?NQ$Q79FJs$#W41PIWl`oihoPtW-n- z-ku9X0S3_E3er@9ISr1G2>WIpRkRu{F!9`Avg#W>%gUw%^iX%ofnIrXzhtjvh#7Sq zf>8b#ELf9GNjAt2j%SFOiH6d(}OWUpN;!G`k1mb0dR3;v_^ z0hfy*f2hMXG8ilvH6Jo43Pbi`EO^7v7JT<0Aw4n|V089F)mK{>R_HKS-$`F?Rq(r?MJ-VrLM-i0!sp*9N&R;EM|tMEeIz9^|rhizq56-q}G zhDV6$8rCU~g?E2jk7W5>d;X_JAhk`dF-0;&bWZ~s5S9#KJ=HugW5I#QtF!gsUXxk9hLOo#Szz4I;0!-s>P zHD%mmkRWyvF$sS?OPwBjJki;e%*tP5yLZm0tw-rsm#;0STgVV&8V`F-U-C$zm4*6j zN9YG4=#Lpuql*KhVUdKwoU91c6#>}%KQKH%nOq#`aH{uL3J6S8*^2^e0*D4#Y2h4A zG|k?+fna$Od49JR61yqNbEHse345PW>*+7M%9@Y@BIr_k+Ft>wFctAY<_PK!x5=`( zL^}1XZMiyJo=%6H*Bwd96|`R0gVE5Cv3|!fW#&#%Sr+fy+6XS$__EaY3w_)1wUaP3 zd81&(HWq5nA^c4y3^;lAX?O?rmm`0+iz+$T1^&$Bm`#(WZ8MkdRa>s!ZxIwX7jN<% zUypCywHGLAxvLSq9lfv;Fx})TQyLat9?P?j*&AP90(WB^gF1lsTX%ypMOYfQCbxrh z(yUWA3#1^BL=?ef!TVffmU!jdMDTC&c~N;3D}6HGuY15*?|`4g@4dwQFQf8m1oCt? zERD&f6ztL@rnB7U7J-r!ldr}ppa$vx+)Io??vk22AoEGpaAeA=4|ccwk=(xCw&gX{ z--SyYctn5QR(^1`madyvRwhy%POAkkeY;8*FcJ0PRF~)VF~kl#qx<$3wg|`6gPJKld8)s+y%nQI4*YyB}msn|X z_<(e>KLgw`Ndj6wi?ZSdN4zb6-cUNJQ~gkdagk&8D1z#)`!#&(V<-qDVP+L3cgB#!X)kGr0Caz+1mkmvhK#H(N_q)pSVThsDEB{JKY~1mVbVY zsdu5(BOkil)IRkXhq0`?$VcIU9yv_lNiC+}HbiV+&YuN{3vqi8_2TR;N;oX$p1L3v zt5;u1hMvZEubscgkGx`6yH7Ajhp-S;mOU<>PtMQtM#>k<%N-{yNOu0RTduE|;?h`h zU&)MDyw2Pl-tu@8_`c>XOubGWgX^DS4}m2W6T`x3eBpP2IDE5wwcTo>xc=2~na7#o zLE4e>N~0{E^=J+|05LTY!E`PUB>r$ql6gmvLVW{ge(2Fnxm6tc!wlE^h4Yw3I|Cn& zEYq*v{i?dqK5?89WK0><$6kU}G|%D^=^tHI;K(+NP6!OC2*#~jzSVP@LVrybXF_Lm5-WV>bS*o z>dN{-GD_`ZfrR3c4I+mQ-*UXSo7BEdhg#4p2dY!R6v+vt*!aAy+jy+6JD;!Q!YwAr zsXCGtC}I1hkw^LAgl^BNDDY>T*3y6{3HaP#`+!P111TK)wTSQq^)F+5>8@>_xnw;7 z5&a?(Ge^1tnW(*GpFlkC7|8oJAt16ZmOaMM75^7Q<`Ayd9PJ7=8Anx9|D7iLwhHH% zKSzNad-ZPXAWA)y!)CP({(Pze#ua}SiaxAEQM>_9GyUHw6^lumE_(YYg|_f91nU+; zR10WXE%DWp3hweG2HbM|KvH)H#Sv3aaQE*l(0J;%5(9oJr0F5BF%SZ_l{(XTU19=t z_BwJSe)sxy``ope(vj3}FCJ1XBfnx)+v8N)(LFMxQs1-1B}%mW) z*T+UM6y>}GLw4f_am+b&X_3sXOkb} z1PdL=$)iTMqIlQ)D6=4V!33xKOxO@le$!0pJ|S5)5vqG59U}OQAo*IH@`e6%T`ILd zEQfiL0oZvb?Ya2aq&K+pMW;<6+9)e|IZ$*N7+^^#+WxM_CiCnfV|!2 zE>(}ah6+!}M#OMnVL)A5&Fm#zA=mwJ2>1*g8sXZnWh!_uQtSA62?t=ygjXrFpAo{o`AeP_J4+Aw! zu#eADBsLVQ_59bvMYr2zG#<+duAhF+7((6xt&+|IJX0HkL?TDvy8hl5+4Dy&pAKpY z^1HFUlfn+Gxd_4x+&MfEX2^3NM+cBznm}ybf;gpHYY4emH=3;|;75Lq2c&dl*3+XaeOhh5W z5BDAUsgXL=gG|Uu`Yu~?qr$79m{v1BXt^K8wrriUYxE;uQadG%Cb=E=Z!>R5?zV@U6M0zB2k zU__TT9jZqfq0=h}TaC2iHl$a@M#s|b6`PVv8z3vr8S#R{0f{(es5rSvOO$nGRibn} z^RuB^uhsv;;m_9lmx=jg8b`J3Ae}8bB-iyxTMzzj?M#bLislBYPil(#E%mkjZpp<& z=yAE5gn;Ean#wQ+Ixfle>?&WU+&1%S@mPz}?pxwjz?6r90 znj{yuG2Y6&4EyRUPUB-fxngImI_|jaZF+o_I-fPupdSVMzR1In55Mh>|8}?OQ!NX) zOB8=fjT--X2bV<0j~wTbqJV=@*S!T_0YQ7?XK8I0y}9`n`;4yI|}O$n-WEK`PD8j zzC8oi1YL3eK$z}F%kIZh^=>>oC|X0M!z0bf2#wwWmQMCkvbEkN&}zprByzZO3`gum z5cKq7m#(|aXQs?DHD$2twhZ@@)lDAO3^_mdUN(F}n$g{`Z$mS#f^VzSCOqWrrk1|S zJ9%l?2}M(C%I&}T@LAh*fb{@)7wrR~Rs0m}H9YyXYXP`9KDl= zxNy)P;Wlo8f>bzw46AlA=NHRY?;uv-o*k9bZrH{`?M+``oB%>3@NzAbMjVYTxd^65 zJfo2$zeLl_os?vZfQn*b!5Yf1qiR!Z7I3@QM96hL{f6t;4kJ+AiIw_hb z3z;H;PKAEh=mZ7Nzz~YN+ZXiit}580-R~^SKJ3x|n+3aY7c{kfK%ZMyOGw3AvV2oi zHV7|{l^7skoII9aB%(WSLR5D`fa(czs;~qpcd!}kz1pDRK> z3n=K}t52IE#XomxTk2d#gbRvb0+=RGtqqq^LYLn(XE2FxOt>EDNv_$2oyx{}{UeC^>)LX;9A3l7D&3&E)b*mYE%|H>B zY1RXMlSwSQG&!?&ZX_1oD-s0=>67 zrtx9?#REIL6aZL!z{NdR-`AF%?UMv%bp(rY5nhX<7XV`D&E^yI?jH=al2d~Aj2uzE z^*qF-=k>4{1B&pI!U-b|1P-)68Mw)Q_y-L!TjU0a=nNBoUI(u%UF(^3;ic)yP>C?j zmjVdg2ZSL>fLHm+&L)Dws?jqNKbR?~I(xJ}{$_vD5U6^1w3lt1x25KG zq?&f@fE?$#D{8}mN>N1Bt)D0Ve_Xv|c&1y^1RC2;CicV=+qP|cf(a&^*tTuk6Wg|J zTj$B%`#tBo&fnnAx_kBNuCA&^&GBUye$LN7xfRSuo|nBXKQ?AZ#y<_ydo#TKVIcp7 za&m24M=rJ>uy6~XsI10&D5l|xWJ&}aZs;Vm*!E)LKFC$M&6hQ=9adKTs}=eh%#&nq zxJbg?>n1zK)C2ts7y}ai({f?lKV1jlhR}c%hYhfZ=E9(!XP9r+A~GXd^4y(ztS!}| zu7*Vz8pa#;BF6a|?8_{oUXVwex_%7f2)_4msP=H4rXtUgU4Ry{$EcrdDM=mQL**Ef zf+>}J>rO^iji8iP2*S>^*^^(YspovH9!L#1xQSs{vXd$oP}6N!rUnTsJ~m~M;b-DS zd9kI9;SQ82#>SS76V~r$Jm9itw2kdsu_>7MEfR1KK%mp3G${`0SBg))r>qQ^7Oz0K z{eoow?(_fAFtD@>rrUP1Y=9Wewtqf&tt2Z{8)w$YJn;D)v#n|b=J4L4*Ic*+QvIeY zsRWt`OMA!Ojg|JjPp*py`iTe(1IgZTBoW1SSb1yYZmOv9Oi_S07YogG{&=llqiL5d zzO}+QhHrhE!-tc>9EfAMY?XZ#1e-@pd>ENzkUh$j!4&N+p>=#{bQmq zKa+|ZQN&LybT#dh+lHI0Fbi9GB%YDXgsp>HeeLt4BpU`B1(B@O#meOSn>XwJhqX}w zsxPwy!r(SVB);otqh5#0XXGb7&e7m^EL3OB@N{^I)t}2!n@M!fQhBe>qPBk9;iwQd z{AHwyCK5bWrRf{e6yLRDX3&5TCn)-3^eqP5Iq$U)Ofqs<%QO}+8$<1dNTW(o zfis#fRY9V_$}TQNE6sOSK(CR?g!#g#0MazWhA<1Lw-Er{uF!j1+0R-9)^e4GTTHa1 zY)koT^(&r>naugT4=5zVY|1a<8~rT*h#0Cv8X6tepBv>g5#&zmR6=z`@$&x)5YC`rpMzf-YhFt z@8@wdq6X`Ku7$o2N%ZktAp==T8nZ&W33LM11DSdGm3IIY&48g;p}F>Z5Yfg8%_=5= z;f;V3xhZz7HZ~ctlOPc*J8Tg@)cl=PW=6$FejnNpx6KIIZ9Mwx%GXrzEqH!?QaI}!jwz1@~)ujewM=#b@hmEHR z3tD|1OtS-ej9GVM|4L*NGmRW%H)oUz`hm|&k@cVIi_0by0+yFyeu%j1pD&P??@=GU zmm6*8=#T5qi0%Y02UAlU*%L~zLh-3#xs90qQXAOx-fA=VTA~-ky~oJ8YCP9aVu3|E zF*19|Flr9{Wn&^gYhWj3;vPiPH{2ceI`}$1JvTd6v22YSxwsQhdq#ye(m_@7;3!9j zL=%$z8m=n+#A<%;MUaT3PjesFU(0QDpP@`zf0a3Nz##Sd$N=TU@cl%+Ttiu9YpDgA zAQ!t(UN_TxMhv61>J@oWGdyRP1kSaCrZ8?bXmXV_Blc0V zSn4bl*Ti=@mcC9kmTI)aZ3UIOAOR<Ye>LQ{omrwst_4sMZrfMroemuoCs#TQKr+xoB;S4^Hh`({$>9Y0{ z%cd((64@N&81z3iO=A0|9bl$X^Zq-6-;dV2qE()#9SzeuZoh~+%# zx+#g&NU9;eg;ij+L)Ju|V8*BiX76xV)9v!{l1K92dM ztE%gozb3un<2U`F5thI&D2qHVvIR?)p6z<7M&+n#uAT_vPO6fSlw**0m5^&JQ^h9% z8w5xr93=M3E?#TDvij4;pnh?+|J28Nrq6UxlX(PsSivgY zjz>-u$E1%+6>X?my0dajc;3xvC=K$I>?zL&BpWdh<^+;@r=h_WU<>1pD30b>MBZvc zm8lll7Fdu;pE{@Fomc|wkvi)Mno6*5j9-p&^Ig>!ZOBy_GIg#A<+fPnb_;P+O9l-U zeOYcP!F7-(imxi_Nu<1>|yHl^;Lddd@uQ#j&A)dJ{PE?>sjdqVf59#~n41rWG7Xra&8^PLb3YRk~mTV$2|2Y2$uOh0fwpZbRcmwAG zlzvd3InY(2MPej+(YL!;3eTlb!Qq-6?G zIN0Za!)QN+a&M5xdc4mP^ca|~=zfueEIfd7PFV1!n&kEIG6*si`nPGlj6X#WQcqc! zh+g<rqVEy$tWx@nE1!`aJghP%IQxASZW=|Vz?8k;@tcjnkr z<+!W3poh=y1-*8s!;UQ8me0zUD}A^&{cjfiH&qR88}W3Xo!7Z+SpC#+Rw)JtN^M*g z-l^-aE`^_H)b@FOFPEP0-uLTzp4)2f5y{Oz?ZkhO^VIeS2ThVlRx7*q_POUFwY}~B zx*t6@SOJ{wbGHzIKTdI3nbmMC+3aP?3#m$0mJNWqu9UYW-Ps_)Vj*8pvOkWplyM-} z$D|q<;M+jF6IluY3vLoTje1`XMbeS0#1QhY?+)F10-xfK@CS_2kbamjy5Op zUsnHzFZiz~XEGhgh|?;-rlRo_K;%fM^4OQsy@@rXJ4_px#r(awU*I$uy;vuI8i~0Rs{Y-peLLw*%=>eW%>(RW3(;Pb573T?iTlz|wz&7> z8Qy3}h2oNri~VV(;)*!5)SyI*vHbl#h~-E&UpF4JMRft)k?nFVxO1sC9d)GeeQu+f zm5uFaO{C(kis2S9;&zc^imA+*N}1}ord}(_KW%w7GCK(fW;Mor`0jkb*Wvr}^Ls|9 z)qm~4Z+QWG4c0d`JM)?6>nqWS_ zG%V}NpF;cheu6);CII!USIv?C139wqfG===U_$&i28RX{HW(_|-@6g^4D+TPOc(ZF zA5RRFug~E`qys@)2DmA9oFJGCkLA#rv-Axcv;Pct|N9tr21Hi?vfNw7rN?0aOCbk7 z`~ub&S_bqGU%^kaHyS9pow~ojN>GlC(CwtJw)nYBMTr_tZ;7q@c~` zCK>e@DdkpK`lN~zo;sJGt8Cdt0L0+8dWRi(ztF1@+I{}bcE_A_xZD@5;ZIlSC~}FMlwzFEmI(qdd%(IhDRl#^7k)1hC^dd$3G?-1 zJ-O_8HfYaQ;rsyvEBLpPg<{qsOSxeI+OyRr#F+q+P=`MW7Vp=r`{6F=!P_0NVjjeqIqA0)Wef%{#2jX}PO zfr$YA)37H4y_iEwae{xp%^{-CLP^Q9&BGfDUlQ$jo*t|twBLo+|IK4v;Gta=O&cRa zUn5g1uFOH#0DExf+i|wtxO0FW_9QQTd#dG zmVrokyaPMpQCwHUZV~OnhNHrx2|A*x;gX##0mpMrP@|X4Rk6ZVgt&sRKVWU96&hdu z03qF0t>C_zj6EgRWAsp)b9~_=;$x@N(Nk_7q~3JLQ|rY>*=+*;WMluQR{i_l(5Ba| zSsP*mj+d5cLtKr#I7R0C(he*_O+gJ4sfs|)Tl6TLr=h=(t|Z>%QR&`jrVvtg7o&qH zh>I|j!@x2sUp+&l&zyqF(B|*M4DUNJZ{cUs=%U7)WWehIVC0pKSgV*??mNa)#s4pN9q?^gqky*TEmtIrA>N`fE8=FYSvKMBEuPrf1B zJ@kKIj2f7$4712=RR}rSKQ>2zx4S}V7JPum%Gb>HC+&LMe}DMza4SUw2>y(#mbq@7 zCVC+Rv1+GHT2B`ZwaD;1oy^r;AK3x&gxt+i?K+@rmG42n_ll->H;t9ZQIUiyyNT^A z)!?Rk2{M)94)@hUD(urt>hZ-FweT96>AicZ>(zPf6<{g9i3$R>9$YL|J*NPM*e(?| zw@c)hN;aN1MXQlxQBde){Oc*B7j(b3?Ne;CbW`W=ORYAYB<(C?Sq;0VPizUqM@}i;ZxeD2xm#cbA1?*2dgZCS!m>&`C93 zH6frn;xihrLvN&tA`}!iEBhVkaD-9r{Pza%uHNMlbO7cysp+T@amJJ>kQ{5>-;3;w z#kuvO`PB*S-ASgs>mvU%WauMG}$k22%cCh56_g_|HkO6bOX0iGz!48`L zF;7spgVOF&iX(T$SX(a!U--W%5HO8|2bJ4GxRF2ZYjL!w?Dy+*f$5_t%+o1L>EX_-c0XPRVY!nCuM{=Yf^!0@mCi+Avr(LjhaEBMvd)cxrK z=Vr0+HN&_dSP%HWx8kc#5ZJ;6$eUOp>@yRRV@z(>Uut9TGeGz8UImmo$5^j(N3K_SGjVPM#jPePrQ<>!JzLkV-+ zI7^RCwm_9dkk*mV0s~=Mvy+L#zx#D38jq&ZSV`}vKL>i;{u*R@{CVyPShbYr;F|Kh zF}v}6v&M90aQ?|*mgTtMUDeR%JSbU?Ru}7Q6hIPi_6`~NIU6$#ni*HLESUK5*x#bZ z_r7qw601T9OX2^@LL}-ScpzSmx2;^s2Z-|cAkePgIJPreuIcTz1)t|{eML=|&r|#@ zo0Giu#-E?>GuO+l;o+qbO_`S#e&vDj+>S9#@;t81Tbg!az%7f%_dnuwP-m^a~NjR^Q@`uP+D<@xT8PhArf2 zTwlPD*@po3`yG9=r zhd143+58H#cJJ}f^s|YH;#9`uIxbcQm99lZVd2L*GNd^YFW$H8GU(}b`W0`jS}B#( z5Cf7jvWm~2H)DB^on~D(_-BtF$GeXodUcbqLXLUXzzZUgrFb_X8`GM$=jHKUcOUP& zB3Xgvs%I{JA?d7$Cz8$q==zv)W`C<_^{*-)IyPTBni)^uf!YLXGIdhV|2Ry&Zs`NO zuT~x3GSo^8+}v<{509qvR(J!+x_|P1K2LrgMSW1VPq373k)9N`foNx@ytG-%KT>k< zkWkeG{S`VvN`oAMtium?-f%mmrOZl6NnwB+{=?mO{M^g;Ce$U6Ich}`H6XCK*;e^j zzKG%*G}3mj$HhPG#$tK$KCdS18&22)2zQPKnhH_eV3n z*X69Y;Vl2aZKW$`>LUI>%&e>rSG|a{$YU<#6cpFn4__1%Ilg!1xyTRPfcxL~^J)eG zcJP3?wva@Db$FVBgOc?;Dqpjg}R%V*($N${-eS4&k4KybDSym4;MPB zg(m+*`~N~2!lx|=sP)zRVe$ieTE05vXbVTx^zHi(~Fe@7gJ~JEyCqi6n@pz?31z&2WdM(pf5dK+}$3+cE-$cu{Wq~ZR z#)}nM<+|IX>vIx5UwajrF&3PUkYy3o3qsK5kDN!J^=<8=bPpF6Ww$}!B(pzG+3Zpo zbdz=sJn_ErDXFP5*x8PR3SSC5RVws{HCtyfwiRloKe)UeSU;`*nGbIxdSjK5(?TeZ zr2YIG@{LVr4@WAZmxXcdjwlz&z;>Qi7LLd|9O-Y{VENZ||0n*kA6x9q@be%X|2e*^ zNGR*=oM9yHOVyh22x~s+yDk^#2-~&k1pPiuYWIJZFTOqG-=I~$LxtL zt!s9N+qvnDmq~grKA)c+KChM@809SKMZla(mIO6y&8{PJX5>7!cnCP^r9dDQ z$RYh<(Ipz7){40ov-C#S-p}T!A2+=}mE?FtrZQ}Ko^1R+PfSlQ%tYj37645$uH5DHz2@K; z)J9p@)y~e&$;k=&zD5;>4A1|UijO=YuwgT0TwcPvAbcu9yD;!m!^0XY+@@2#u>bc- zx(L6bSUVJ`)uJbTEwpB$fTgjirpC@dq|dY;!yQ}0skU!Kl;)7a`N7*mj4#NK+jjqA z)!>$`h}`S(ZWYC3P~l?f7aH!5;Ashsd8i6s7HPB1@z0*S;!^+W8Y_d{D2hq1tq;f` zo73*@n?r*2R|@*9l7I(EYE5)si5d`LSl^!RqEH5wCj+Cn{3m>5MH^E=T$^ujnQm`G ze7`R{3b=)iFPA}lpYoL6)l5{T%jhZlC6*g)(wL(U*(J2p)sxfG+NRk#(JpHdlFIcu zt{;$(qTMYdCw`|s|gEt z-Kxbnpdc`oMhL~v9~M;Hp;<@AI8=`PU3Caa z1Q|lYSa&m0p6P4?P`=!*SXE7fRn&Pr?}C2?=62w=i2;t^_K^NjEt3lmr;ki zobvuCVy~nVs8HQk-$|+<=1Y&E%1NJuI;rO(%^ymcEhtiA6~E=q0c$%qBHn8|W$0j{ z|M&*sBNGbrs96rLf0o|NTqkjpW)}G|J@v6A3{YR~{dv~Nx9>EF=&i;cU6K|cOr;+o;eoMh6&4Lzng(i=tEvs4{_k4ci+EsFt z`n%8lQe@5~@AXy`&o8?<035K;OSoxA4|SR=!yo)Z$e`Ve=k=`aOKapVqiEYa^X)qj zEJ%p28SDFe5LPuwYFBfN<;w5HM@Qr!Ya9%nw!~Y95+r=8;K(C7LT=j&w|lE3%$61& zBn~jj7XTGF_d2KLt33hO=^g4H+g1uHjr^0-u)rGeL0$~mCD-4CtFR_)IauaV`!+@JO>f31~TnxZ@{wOE-yRs`lu zpr(e}+D+dt|E+4h3uQ^?CI}SyVq+1&X=cXU8BWiT_ z0RwCP_JQk(ZYNEE;fPr7ut@e~ZGM8s|NT%_^zNqv-~v1-J`| zQa9L^L9sTiZ8cCuV^~P}-ZDo^U08BCX;ybto+sFJYCpKdi}6-lJry}xHK0$7}N+d)3s`DpyrDEC@njxeUs0gFaroWo^1YP3ss5L-o7_@UEzyjQp8m0@Id?4~y)E z#|5keNM<{n$k+wa?fiRR@e>cQJtHRwzm-XoNF7zYP_ZPxcbLC;iv+>f7|?)XLc1kI z=o0kIUzU+hon0lVAcyJu@-o_N{;pDsy^znobP-m63%dfQilK z-BBk(M~{5P698ooem@PQar<~a{IS#R^v#B7kIYv`uYx=Y5CdNip7nPyw=6#`>9E#< ziMZo%fi0cDKVBIN85@RG<$+tjlXON4+rDx=o*0l$lB%(uz%kiKgREkZ3DKv0pr5UF zyze?$6hVT{rLxn3^c*@W{zq%-FPNi1&WKPFJQX>*KrTuolNt8Z-wkr~t0uyARjH?y zzs5q8HzMlg6Ny)H(rKI^6&NMKpcHgKGL9IyV10sw3|oB36DFwLq&dN|h@|CZli|=R zQkHU{mb;NH&4tEAF)0@AC3+D=4f-P^()*+#JLN;CE%cE8WQV(Q8qR)-dprT*hRIf; z3x*z|ih~Vjh6%6}c)}=?8Ou{@>VEsdxa$(SZeFi~uzu2nUJfT}b`F=1nGcz&eg=KO7zBzB&1T0pIw$-_ud-$;)SWG|#^#EQf z6xN5&jt&(CN#Jp{)$Nx&C(rcUK-vM`cG^MBsnw*%JvWib^SV#({oBhcQ)rL$YeNQL z0~@Cq8=pwD4L~HJ!)FoSKaQu2TAn~Rst7x*SVULd(V#Q=;=cK++g?wEejiPsde-E1 z+7z6UKVD5&0-WO1xbc1_ESPt5+TU0H)%&JF0RxOo^E*7U+emxcZ0azhN4MRYatvOq zzTyuF%=Pn3vbJAW7`}U!r_QwhpRin`1B)N1%sXCysjdteOIVpJk5B~3y)4XLa0{!| zl?>w)Iq;|>+Qev_60p?M`VmYny{{~a*>jQILy>he*B_dp-smv1@WC)G^Q*z3m4lWx zHY2koD`(siYVQcN47HurdLox3|2;iOvVV=r;a$RZtKJ4ul+Dgwi1@mhN-Q6^ug zRM};UPc%J}3MJ)tpk?p7``Ue1h# zf|Yq_^>w)NBR@C9!Dj9@F%?;iDefc+hwne!wmZ0!mW_~>sGukT6Idb)uL+Eup%FuB zvV52(v7oGlGlWOcNAgygj>Y@KmP<5*2l#1m(;}q^wNCq!$I-t$iYsiY8+^)ZOzKM- zcLN|ltMxIpxvpqk(hkF;;JF!DqY0MDw42vHEn;_5BZx%O1yDRF=646iB_k`w5I?pM zcb;JX`g2}MJ-By1x>u;4ehEascurM5-+iKws6w!=oOZZmD149jD*;+^<||y>b5k#| zoFQmFU<>=ZBL&8~;gK)JMQ>rWx#l9j5N+r!82#Uy5o#T8+u-Zh z&d*m=RMbs?lHB|0${<{4U~mvQ5Z^)+j0uyLNRXH|B;f*+Tt*(W!3Xqg;!6_scE0&R zYjgZp|L?8N=Dm(^cZ*?EweQRWr*UEbmw@;68nAd}5Z@h6&ZR!rir;k|%)MBczDj*{ z-dT~Ud=>G13K2L4BOHdqo)jC~?I>L^rnY)PE*{=dFp z8fj~K3ccht{Pg0-zo5)kC1F>ZS(DB?-OBpf8Fk>+@|oGx(9f?h^#H4fkMN2vW?P4y zL@ThlRPl8bx1qWSWI1mR;yv-j6~(0I;QUnK$LqhYjyqY3%GHr}&vw4*pHI`b$}b0u zckuHbvkgJ&jcd=ONfxUPT(%5+ig{)vA1kRXZX}aTFgQ#I-Gd8AsBi8Du2Xk8z)Q-M zihRXcci6#^n??o2J4=Ai+G39`>IhI|F8v4P&CQG11yAP6n^@-*XMHd|E_8@-V`pP* z=`{{dkPFF2ZPp|TPO^bYv^f-#q0l1{1)#)R(O7__>2cPbO4`Hr0s>cs6(ILyopmy3 z(BB=2Isx6iwIx{apZy{J$bH;L$qh*%Y`Hbz3mng=;6X8-*c#6?)FuM|CO@`;t&JI= zb5qsY8BQL8&gjD1qBQiMC5rZ9kncK11wyNmkPu<+FQKcWAIX>SFTV-wEw%jz?ug~^ zRv+iysWObR6&}wkegT!mBYGYQ$?-n#>AlbE4R6_K*pYrd^L-u`WqIKgg85-#5w7`K z90cZh#O84OLivDMvZhT@kkVKTiSo2LpSMo`WDR$Mq7`DtmeH(!tK)lRBK0BhdG)EB zYKNMe0CbdzqD699UdI5Tu={z#@v((EYU*qVv6Yq8mtyC-fZh}Ohu4?eHXT?fIE86y zt#3Y2x9^k*iZ%SZFWxjW;=R!M*u=!&>w#~2?%8CJ{|E}67<3-!|N1!9p1~;ax}oq! zna%x@7IpQIbA7%|ex{E{dEMp7OD>@hK|@F*iKnAwn5$`VO)u*`1?Xijc62mPEXc6_ zhY2j9LES}Oy?k%{-nUULc+=z8K~n|4!r;<7VgUyDe0E@s438d=zns6l?crSi+mGS# z1L>esaf{#wOIFFCi}$L{1%+VNlpB=@3E<`FLVX*i`)pus*8x@vTdfN-fs|;OBu+aT zTJe`}Q&2$RXr69iKjGoBi11v8UP3!#^{Fq{6Hu!9*5)Tg3*GZDL~hBy?GP<*(e)?j z_lP8ANBr-D+LDgo?+rQD4$;9YfDTrL*pBSG$};$Qk8D#;)mX^K)*dpdWn4 zzgAGrfp&LIf19B+FazxX>lS(GWN%ZGUx7OK`9yG>v=QQ;jQgJSd81d&*~1U%9OMJkB zTnZ~Da%G)%G8bk6bRn$tAV=-3npSx2!n6s3m$|w*V8Ur}Lfn+BlyV|sKauA&TMS5) z!-cmam78rFP|s0dxVGu!n|5h}3VtsC)x066g?ZJ+t(6F9QNLmbJ3Cg%%8OQ7#p;aC zS`itpx{KNaI9fbD&{cjGWTmIt8+=y^6zQk5R44(~1rf<3EJ8{BQCNyflq-)Et-k0j zNkRdhE1pw5`g!kli|Y&Yw)l zHF3Vbq-(p6Ixo6uc#)+Zh(xfY`unZd{d1oObhAKI#6KYiB4l3g@Uf6|VN>l2SiSh| z2H=bPWl@Io31P_31gTsi!(Uls%e!g^<=l|YEUY*gDOH#4w&uWF2Qi3Z0bASzlJ?5pL-yD%wzxo_bh0+cY4Sic^BS=h5>Paqanz-OhsH+?m zoZR}3APM22-YM*L9D9OBGcEUdCugJv=EjNms-}FRx~h``0J?DGTN`aBHqa4{WmLTm z%)B5xQ9q8lf=bO59wTjgtnU+BjWfo8=Fdf%CZuXpWd4^x+o66z{%+MDc-G>MYUhxa zmHimbdV{vr{H->93DKlp^xYQ8)rEM9i`k||>kE*bkXpH00aT>Tw@}%r$MRgvm4#M% zDo{5&1h557V;cbzA;uOW5{~G#gAYiJsh*lCy$c=zrrn`FXtshVcx~6B|OmK%9C?|kfe&6 zgCnE-&8;8}xRMNLw_pv~nS#l4HFN`ZA}D26b|Te{(Se4uRT&Y%Kk4tHPLc%mdGJ@m67xp_-fbu?|XqM?7~#L%+4{CJCGA-rnR8^+(SGjBN{2JjCm3 z(DUM^-Zd3^8It-H4{(GVYYzU^;;tJ<+)E5FpR0bbvSel@Rykz}|9z|{p1^pOfIcBBn#gdUxPJi( z8OuyEQ;~um8m;sgOa|?f25SP2eB_fgJ14j3^U&>_oW8V~n zgrj#Fm4tJdg&!z_J+}6$OCdgJG{}gOtnLv7t;@e`+)xcZ<~v{1nsY!an>O5F!ORud z=F%#(5??q3UTDn8!+F%kmT3+D2k_~~5l_7aPfvEsti(hRvLcS7Qpj$dCOih+9!?q+ zwj<{=G6J$G3CI}5uBVFB1kBHav>Hsc_+sWYBKq-n(r(xhqX5`&u3hEbgW+C6F|l1z8Vf6YLW%^KRW1;Ks#oc%SPU1#3o zaVSGm;9f{c=UmPV)t!vcx6d95bh~BeJJi{f7&1(|Ze3#1?0&(@&B@5giKbJ9uGJ4n zT+hOyT;C>@5dO?UTxtRh0C>{2tIIdaXD& zbS3)HF8B?gh0G#MANyKJ$@lqo>k8C606q^P2_G37WS^E_0wRk3ca{=_4Kz)V7-%e$ z$Nla$S1s3OweiO>daZv3T$F#7&+w7y)U$@5+S?5MJBk=|*X_$`l~Io1Z1j&-h16ZH zXh`sO1ag+Xv)^ymg8=vR9{f_blae9;6^xaI1$+=xOz`_t>HF*BsVctLlR-1}9pkSc zu+APevJ7I1_}D^^Uqajogwe&QiBWd(`)y zyHI@lc>l3Gk*3j+r6sJxqb?uN`krYkWv>R&aj*X+d^9MAA(wsTJWtHfG_7rm4?ve{ z=r(!=k30Oa6!pF|K5rz64Ysil#X#q@j5knF(RX2I`mD(`{2y{^%La0ebZ_XIt4#J1 zpq8!`@(i8I3kCWB{ZjzRmbz3^fd~5A)9~buYHENi&a!IR9?BxlDXKZdni<)W&Q_*0 z)gEG~n08BZ0QU`UZV71#p6Gmo^qP6XmOD2yO@LyKdZRD!=miN-M6T0R<0=@CS!Q)? zSp+CZ&V|)wi@P(`ixH3vs+?i$ABs|*Hf%AnkSxE%850*2En3R3DYyNlUll9b8l{!0 z-y~>!h_orL(m(njD1fnS3M>69bLu1&_50Yc;r! zpia{2GafMOqn!ub{v6LbY8o+m-*pdaB3SaWyPS$Nuu{jq%a(lbKyn1%V~#2bq>cB? z7?qj=(VgZTz5f|2pbtw-XjB5HFMs+FshC2^F1QU_<7F1dP2P-5yJ?SjhX_3$Q#87* z9IsYI4ELNdW?KTOPSiwIeHzn{u90JMjihndCqGG;H>iSd3WK^jbD~whYf1?6w;(5UVq?Z7kB?+x)O0+Y;WpP4& z_#BFQ8~QS6KZ2ogC@d;^MkD1T@TmuJ2fG#Ur4~eiQRE1zt>(?&;MBN$fQ=0Vd-gFv zfE9ORAq~gFXg{^Ji+xqaxz-K*1C9BJ@2n zJfOet#hT8Lf-#KA3t@h6#2_?sW77UxNigLPMz~!Q!Q^x@_nle(PJH#qPP7a4CBj)u zRJ2(Br?WxJgsjd+{c_ztQuMS0Swqh$9@B3IMkBIG?nRSQ@upF0P1+rf>k+?i>i0#P zRChX?j+ZWB$>@MfvxPv?#lC4111agUUcN>NLKx!R2x!PtA(`ct%rbu;#AUtZXKFa) zsieF?__%O@Xt`@Mf=#x)-LS5I4G%8BKs;)6pQz?Tc`YzZv`#dkbL6!-WSg;)Lq#zv z^0%cE7j_sVm3|al;f*%^SdidhzAXlyu0c)^(@niD?QeY{#E|3g*vX}M?Eg&0Yk_X0 zb&loK*vY5eDC2=-(-atnfvC8_A3bjJ_$4|AkJy8&n{J+?fx2kCF5FQ&UsvRQ^iZY! zL^`!XkhuQSPjM58hKr@PZ_7e0)!FX^<+?(*$FY{wK+MGba3KOL&8Hkt^m__6V^iGhP6Riv}mwlRI~ySn|n%A;9H21lZ98HS^+D@M~E{4bO(5wU^-u*^}uXmyKvEyia3AQdze!LUbAJwuFs0H$1Hr#FAgnIc4WQTE0&$d*<}yNw_n_HFGFEJJAfvt)#2y zYz2H&I(Z+|7|dKvARvsBV6XDKg z)4lF4p08uVO_$c9-q%56H^@q-e?xVO)djjAK%aXVtJG_y(7&0+W0M0IBaaBWSXHbe zPF5^mxL%hVprCJO+Nd&{zpQOd-nuy{t1?WG>eLu)Pd-K)D|1ci)^D{!QilXCruV?*If_5P!bG(u9oX#uPGxKa{`DWu8?vGCX%-t@DQUFN-#g3h6 zr&r0Or}6N0G}o>X^84>5;4<_QgkK&;B2MPLZS)nupiYuGH_v68#(fuszjORSNtEM* zT04MA&yB@rB2mJ<1&)APVP?9Jn51fYh|&0yfYa4nuEkJAvKXzHAdTWJ%|=*|m@JYF zhix!$BF8b>xg?FX90Mky#TvORLuuBG9HU6Bmjx{N+^Gwjl6`L;3x6=xb~KnRPhY-a+Lia_7;X7yK-(_$tjxfZ1m3))mA!J z*3K3oTB{2%*_AlOZ4Ooq8~n#yn>+VVE-nSF&ZmpKZV?+!u+D0gG0V6sFi`x5`Y`Yr zsQth5EGsJZ%A}509R4<1SJ07BOD)_Ow$?^3o~YW;+EhX>&oaG#szLTl>hqWD7Q#06 zo8C8lMA6YaEjdZPt8Qj1(Q)~aEL;V%F~FdNy#7^+k}?&;|PSgx`Z<3;W|-k&YkUGW4V;2CNTs7NNSe}!}c z^3uxg4EXv;g8R>hFOu8W4=YIckuVE5XZ7Mx7KOQWL|8v6Y6H>LDi$p9E6K(+@!2}s zPq^;aZ23>w`-(8NAkprzPDvBL1^=W?hPNd2MVJo(M}%R&$>tLSpCka~g3K1c#UdFU z5K$$3O%X2VO-QN3V=KI^#_@kwWwcHZE6>hSr1McJhI_Q-_^4mOjf#wONaFQn8l3OQ zb7VC}>_P4oT&1C{*i(vQii=}%^@Z-O-wF%LtPWTX#S{?KZ}#8ifG(lEPOq!(=48f{ zdMAdH0bbr+cKtzu0B<{_O=L+ZLO6v>J)o49flj!%CRnm#Y{AT=m6oQHnV#NhwmH!o z1qtS%ZMKiokoCF-mH7<&8VBs%YV9WBC>3yJYXYY7yGY;yO%XG)EZMzajCNmy{>aggE9}SObfJ!a_>Cg{Ic#XM2=XH? zoA0;PBgru~DU7?>jf14Q^s*xtMzVSXTeFQaP4noix`nl)K_ZVrV5b{Lx zlIHKrk>y!ifEsu;Vu3J}S^7s$b5BC*9KKSM=#7$JD$+A89lP7xlIQ7R6wD_a)SE04 z`eW-?QO3q4B?m+=#bi|dzBuAHm^>$Ot)%?3Z1FXQPzB3n7>v=ZO9>kH`29mGc|7sp zreN9Ho;NK+;$X7KF;3M;rhG8vsk@C@l=TLm+OME}_@N!c!!`e|;qm-s^!Hk=1|OsD z-zEC`C>Xy1PHsXFM(86}|Er;5hMX3)t0Js+IYoS?j9jv) z8NxB4UJF=^&s+{#)(8tUJ8F=s;B}%Mp9fYi;c&VzcQtw+qLz<37dUk;T2}KEAJ*g5 za`{^P>JYkPP0+}$YEIVG=8Xc}FbUdpMg~&8%A1V`C*^&59CJiS@p&t& zG^j_<^X+uXER&gRM#Z9Lz)Y8Z0lkz>ad;PfuR-^CF$LQ;E|}w57_xidhk1}C&~#;T zUJnK^k^=b*X&sp$BGXw)9$7`JK*agOEqi z_w#L#j>`XLIB#NEdgZxe5K)$6!6FxU7&)2gwR@*^JA5Lk4gWqz2N8KGqMFKD=*>i=sb>rr$pD(g( zZYi77@X3{{K+ZXwS|B4eB`b%Le7@$e-~Q@1T%uPr<@lLl%T9DyUJ1tV%T#8m(-y1wYuL-xwF5f=GeRXzFj+oX=-lP;JZ){i>}~e^4D_DcMb?(Lw?uyuP%wnMU8Y?Tjjo zoa?T1oOHVfWYUDw8C_JqDXqosNR_ZWAXb*b%0p71o2=tkli-=qYn2_kBbKR-fg5mp zytU)iW8soI~|iehqUi1ypqE~~A{<{fuxRj^qZ zw<_Wi9b<)qo0&DH<<&wd+p4m~_Za1qw;H^^36u3EUH4|AcK6G0epr97EOKqPGuY+% z`lSBthV=wKuGm9UORe2LctXd&hg9T8n!5HbI#d|Z$OP}k_TNolFI`=YIxQe23b>3L za_oKwGG;j7H%{Wmkz7*mz)t-e)$`VKj@PU%L`~)n?Ck2}a%dm=?Y=#7PEnTXATd;K zuRD4aLy6{?%PpOk88dUyEKc{JPtFkYvVY5sCQu;w#Y&N`rntqj)z|vf+2wr0YLjEK z`o=t0$S#_vnjJ+2vt5OWcBRlq9Ap*k>{?eIL4{1vL!N7EzB8j<1F zmH*Z!TzfE{{qWH3nTay!Kydnhw`?WBe-!~H>#`;$w5f?n*XyKBlZF|@T4?n5T{Xr1 z?Oq7lIpsI-h5=d*|3lzfg+>uS{omxG0}Fxwc2PRzkCvFkb+wa)=h)aXH7H@MvA`vB zMaR^^>C#;RO7)sK55A)pzY5_Yyk6489+Px~J>qYYIt|Mg+xomjvm+x!nxRVb61>_K z0UI$L`@^KC^E5-xIP``lwod1ypgTA+fZ&?TDj}HN7lo7b)-@b6}#M^2T z1;@2k3{vM1M7?8Js2m=oqj9pPv;$DMRBxD7?5E8^T~Q>pvJv#6ogZs1FHDXc|Dej^R4?3io6e1eOZ>!{Ilgj|X-$s$o>+bt9oitS9Q%!B8k03s z98d0VN6qnuHhRqQOW=8m@u3nEE&6Iulvsuna-MJr(X!ZhBUxq2nzbfVMZ?=6-H*!c zJ_GA6GhWmATc6RKT+UqmhNwb?(R3|O>!V9N;YFsPzUzKHODY9hs3WxH3DwK-WU^h~ z1gB2DI7@T$@8x>->xAjh&=4(^k@J7$#BV)Iq-hiTF=g2fj{g}Cs(k;bjD*5g=8Gu4 zF)5juU91Flg3;D#Ia<_!$g!G@0C7ENB^+3)G+8K5lUd$dYkkw`Wp?Bm-e{5`Ht!fB zR>e=`V6y?Ib&Z~D8th3bvI-hz_?c?C87iQ8-9(4m2KRMf#`410f?~h|1PwrhoJC^RcrL<94wD&Hhf53kU4$gq9q? zgaBXGtWtvfE=89I4yOv*LTGwfRix|iJZgOP3{{h);ewo9FaS%Gyufdv?n5qo_!iI( z@B54UTz=fWE>A5MCxc}eeD62Y2N(>ty3)%^|5Dra6-oX45ENej1)YVsW;U^WY_Wd5 z&fvWr#T>SohbOu?Qu`iWNO==H+a$O>I`ld_%gCyO}5xlz~X@plL88J5p_{nshDTY*LVcT?u;LIN~1r}m_xTGx8&c<1Q87kTiQ z9M3fT9r~UO&Q#iBVX}Lo=YfSx5M)B#b@(gIjgLY~)`}#R$FlRm$;Ebq>n`T`W`C>q zJ#+ItW7n=+4RM|5zaPdD(6gjcumK58pAPqFO4e}?&EZN`@)ek#%@@^m-3|QCMUPL$ zW=#KAth2zM97@avX%QWpa2stYv97A`GmL_eyQ!T`TU0(8lD3@%15W=2KqhCzDSyDH zM&C#~XGON81N|)YKZ=@fCLkR50*~%5rm>`lx$Dt`fx+^geeru(8Dp4Hug7#yv6_lh z<<7MYshjWgp9P#lEQEU68?-A5srrm*a0g}6uwC@zX)Xk|Vap!pjQE>LDHs(#IR`l~ zunsAfv-@#XLHh{|zms+cY{(D9W#VJ$H-thORS&6<)P^-(qbiCa1N>d(Vr~WJu&C*L z2qC8l3zkq|waI-(F%9NGekp_FYqzHOw8-%HU~NC=2jp zp}d4qCz&5dxSfayO$XOWWUBB~Ru4`dx?0hgS3U?Bs?j8i=y_ETCy$8a{g$StlkTTB zN3bO`ui$(vQ$BSG_urav&DIRFDi%N73G`qkXuAm%?KfZxFqg|(owVSzGbvS$u6qWo z*Ng5%>@`ey%BG1M=)U@N7t|;7yRFW;to>Abz=EM(Q_pes@-~B#EMPq-T^tL!Jjjxm z?(n`kNoOK6sVr2kMlrD?3YQQdkceZHLJ3m0^ZSK99DJ&~DFSAJW@31riRe6u8mWE%&ASjQ!S3rqg3tkZ6cja%Gh#Jc84OD}ENw#*aaZuJTxOt+8V7 z-NMKW;u0>%q^ScjZSs_6{J5&pJB2wRk%PXPTZDtK0%sLqg<84!oZ$Tun<)$9x_%Ow zSGFWR^>|WV+W&9Q?f`y6yI^s&*a)~YdKh!KrP>aQ?zZkd4@na4?7e$xe+>fr{bUAj z=iT7Q>t#ae`DC+^`TiN)HSTeZw6z<&X~k8^d-UU1X7UpkSgZN>kr)4-l!PsC*($gi zbNhUK%**oO40qt^B(MjZH<@O8JsIM?qw6Z7;V@zSJ1UC^F?6*uXD72<-${7O@Z4CR z*hP%1*uPWnBBmV8AL0NvsAY-ij4qkxD-%0O#J+=v?umTlS;Q)5V#tN8yBu1xhk1Io z;OP2UKe0YLuwGeVfZn#|qaoCf-=8428=)K3&PI$*T1=GD6u&<&Cu4f)`sh9xz*Pfxg^q1#RE!E8Xo` zUEB@K?CP;Bgye6Cclv48`tjx!?2LtXpt*>HoJw!zTcLPe*J5d@`r@RhMh(S}1sv*g zzHa~=A1W1#rzeLDCg+E7$x&00rs=`v|L8-+?=hSqSecyEW3c59DOf&wDrqo*7K)2hc}yGC*qK?US}GWNxUjjwD?`$Rl#0;i3Qx4z?IRm$2PmTfhpfM?2m7u%|{xNt|mg{A?R3%*6HsqJRMB>Sch+k(Cu{q zyR(mh+tS6+R)p?qx|87w(LP+?R4v;Cu!cc^v$JZ<>DWGSj7Dfi$S z%YS;Y8NmS}AP1mlBa@MV4A=}ypbp1K`&?h|A5+IFriZWibj}-k8>T!&uEh(JA7^%= zpm-P`2Um|d$&8+Cyjm1Up4gIk^wObyNA_@xKpOg2dtGs0e`2J|eoVk)+qa%TUzdA6 z%xY(E%|8FwU|nR076^4`vQK6F_h6d*{=u}fc<5=u$l---0~poZ4xR20pchTQVbz>$ z`|W0Ln6>yie|oU~)ZBH3=T`l=ZGy{GlRTCw*fxrPoq7jq_lSn8b2G@TsA! zS?DUZ0UArDDg{edDD_7)eL2-M!PHHPV>WUP*C}tzxQx_pRU3lRx?Ysfo_=?Y_}?B% z1Yrh}+geUZH%|g})R|hs{>oauUtW;nN2^;c zb*v_PU~0|MQQqKa&6{)jroiD6DXIHIpoY?78FTsHmC>*|=350SiwtY($7KIHmN8M+ z->+KxGRti3sQi=xuX;%?q^S5T-L-owB%qjdF82*RwK8o1*)3*@hi&?o#D(xl1QhUF z8i*gTn6e$AyX%H;o{kqUsiCK@%uV6FKXOAg{X>vOo2N@?|3;Ki~ELMfSK|U@N_j>y1!?J!lF%%UUs!*Z2{g`mr47g!M>&;*9 zNHx~nm@;2Z)5IE;6ZjFALTWr&a`h+2s(R3Jo8x_K^fjJJo&|btpdUK?`A$#Xfcp-Hu`0{ z@v;Sl13`TyVm{F9>!rc;y5a)2Y-gMnj)WBPb^}SGR?zIV;&&!UNZHH%l_f%uU(TW1Vxj5Li_=k{EUri`)fhn{wFYJZyULfW{>EP`!`Ja z!GJl`W2-Y+kpU-hi--h|6s{N(Ccm1?YGhoz^W)ip&CAyQcQ8J98d=u((X>WCu$hV8 zHu)1AGN_IUM74F$58!vO?&sXg?3LB$n0_rHVN1BqipJp`r`ScMf0G}87D$fYGV1)k zX#DYC0`L99gN@OSwwU_^XJ810mOvz%v#8|iO5yDniI=_QPAiX(&*08sz5|}OpDxcj zP{mya%|>zm>{P}eO-+Q4V_rJTo#El(AZxxOtp=4ZGVD%g$h&Uep7?R2ZmJ^&u5aU{ zuOxhoO^%GhS?1JG(hCd7v^G~{45w74FjJ5_?dXjOl3P@CA3Ux|*UfB!o2TeK!)0!A zzar@Pl<4CiqwckyjPl+DSd{1~y4-v(=xl`6w%ri*Bl^{dQ~1R}wEA+;MJu{ly)k?X zDbR*EeI?^{>Vh|bJQ8<^Bw)u}{Z<09Q89fwk7e3Id;{kz1)}mkEU!XaqQU1yqI`<< z9%*;&7QYFo1s&Eos=3iZqq*qWCHiTO>Vh~4Hoo4Wyz!yz(!Ed+LKD{x%*cEy$i1{s zQDi79<^(cAY;sKy0rM9YuB!p}AjI_S;-NO{i56js-;M~HflfcOa}ZGo&bpI4vB}a{ zozvODz^S65-9=J(|3Fme-UiZ9`m)-~FWec5OuG#9ejGZu&IZ&aQ_OHhbFbPMwlHfc zUo)Bq6RyC}EDbT|`q24YDjCctSys8o>Wub^RG==N8z|LhgMVY?^ zBDsoc5}Rms72PfruK$+BHd2CDh^fewHpJEP-RSJiy1m+JjlAfklC z40|7)w9M>m4rks>7}1Pg2?pEH_TEJa1{tnw<1#qr3Z$3NZN6{HhVC2_rC%=rKnbDP z3PJTLHPP4JEvMVZHYa1<90rSp8mIytF)|)RrjnW3{THs2@=-yhOqb^w7qjBVD>&@! zhTjh;iqf?<{RAb7`tOHV45TG^F=t=@by#~u3)=|}Oyy~Jx0ddr^?F&RyDI<%Yq$Qu zWFPg<4~=Nz{GhhPgRa0%*(a_;pRVUGFiHOyo{a-zpO;TCZ_neIt#}8ziKeKfx=?am zzcdnZm0k>*_V!^auxB{$=dayo)sIaBVu(bkzz&(PaLH0Ho&sj4e+@Yn%1<4<+BNmG zX_@R0-4Bh2x%$xsjB;f(zDDoYKFwVm)W1{FJRP1W63~B6A5tGaIF2iAJhtD0`@pgs z@!Yc8E%B{2{}8R#c5Kv;hPX==v0x62B|5f{s3z74T;%|aR4c>6)` zDGNsXJ#IO{?-4a&bFx-=$_4H(62?Ku2a{xWQKr(7NxL_1I52^k*6(7i)gH;WZe^t@ zU^elT$q$J>UpfP{c+OeV_RE4x^Xqh6pS34F$~bI385c@xmUtQM$i|WmZa@;1w|>}{ za$B^Ll(Xkh&T%?wC)1`1unS}txW(gBGN_RMOs!3%K(hW#fmhl(e{;x4j)k%KByD!Q zo(C+7_>E(otP~mNe0y4TK)R2#&>(+akJ`2-l6!^bi=@6rBPWAGx2`TC^d}8?{uJ!qffkAalt9l4+k@6bCM1GkNSL$m z`DtY;Y>aoi)8s-qUM)Pc%G`j90)*vhw)Ff-3A-A#8%^~LBY0=@auma(zl6U4sgaGc zN==`=ymJJ2=~aLBT32($Sx|vpea8i$-RA)88JfFmDMiP8N+tg{GzJe>3Gz^|#Cq^J7v%Sv zI=SX&0$4oo{lMgCTg@Knzi0wY)SutfOLEPW@u+TGp3lUCMVrBK2bG05ju3XEcCKbV z8;XuVdK`I^3P&CPs{;&MnpffJTSYXl_tk?Fsx4olKsc`ruNR2tETN#I)_BvthDLSx ze+|*P7b^XR+wCCQ^tUoFGHB%f<9<4u%jL=!47Wvu)nVlObe(+5CTPEN6KJVT4&4!A zMweSq>mcso5jDT_@jI}`rS2yBbifou7XpPHz_0QGxfE+s+1v@}qboxF(ninB&_htS z2+_rMlx*?I-ed{#q!_=a<_${`HawMeXiga=yX&FTx{f;~k2%H~YBOq)uiRRWdm4K8 zfe~NWg(Li!oPv6`f|@~ z=#)tI{)$R2#Kx6Xv4!5re-d;=geWbVX%x#)XK-?CLOO8QAEx3#0`D5o^|^SFhUka; zU(*4J>9xnhSCc~}Yf{y_Sr*0&ldRLm`O-6qLs1}dYa;#v{Vd>A{|F<=mm>`b6tEUc zLx!_yVe|vI++>Na5Q<#2q0o4AA%T209J3*Hw5&EzOmI)IS=0O;)I(U3V$Qa%RjJPA z6Hd6jGe*z&e*cw^!VB~t# z7|JG7Db)lTAA}^T64b}u=aJdgc>`FQ+zFSyKF&OpbCwVye( z3vv8k3c-$iVi3Az%J)$TBHk|7w(12mY-m_y7ht}62(4>cUYKfc%n1bl9%-T;{(KvC zCpP&==%4no4EIB8R$zY({>JW`pAWW*EIy||0$fCuz}>MGT3Xn=A55oh2tV3ZSJgl27tbFpk;C?14o2grdEYC(cQU?l97gabzjIu) z*4(>bB8$47wW`rDusI{5h7G(^Rk_sgj-+$W zxJ#_yHm4LxXD#Dy9Ym;45AjBev~dOWByOw9Uoov>Osfgr96bKgcZp+RvstT31B;Zw z-NX*-x_Ynb0H!kL-H>&nEysCALTO5p{e6aS#}%Z(Qo>57tMzp%t~Ej>x+gj+;a`*}{8WU#wJXTT7C6H^QGB{q%RQ4h`VakacUq zYrR;HTJ$51zXjM4YB(xevpgtpFGrQUA#G*1piqG?Ly(Cuj}9>;TXLiHXTD(|JcnI= z>{7YF8s&jotfyVjI~_YH-3^4G{{XoRNn5lWGIDUgb6TZvhHjZK>`FRy|Gb()&lbb< zxju`ALuHNv?XxkoS7pfp+eQx$#-jUp=f?lf0tjs<^c(+Sbmpc*oy6@{$h={}s_S#9 zR(LlOwHZh+F}9~wEi^-3c*QckJy@dC5J<;&>bvJo>!&(ET9Jk_tr>13jl;*_(xwAl zuA(+6#9J8RDHLk~ZJhY}(ARG|PL}0KMI_pv(v}35$L-M?^4eiAwI%*q+Ag-FD@VAik{FB$fpJI@ zspiYh;yUem*bIF38Spr1&&>3wl+a!0Inwr@UcS3 zr=~DUVSt)Lenk8%tt9s6>$&rN7ns--o(#fkDng-URhhI zhTSz}q{7D*8t62S5 zix5x1#{~ISAHDTKId#utUe~5!ngVgGba*{XjM;BGj4|Be-+&*13i9n(!;J1Y%(W;z z(Nqw|4?NG|`>(2(PEEg=hiTRVeJ$kZ)QFBrtjr9^_rXG73E0c|7?k^8MoM`E$M%5# z&S+pj(P@i}XN{@PWFCqz$0z5HUqCC)RA_Ceq;t$9kJ>%e)|dVxL7Yp-5MmcDJ>)Gg zt*H)?TLd}i-=FV3F^mqmF3Kn&qtyd}2-)c3caXPl@#K9xs4WwN9Vyg8+ic8&rSkc( zce*Uo*>$m?*|qo(w=zn+BqLM8Rz*Pz8kNRZ6&)BA(|n!o@qbIO$zUZ`s4ZIT^_ z5@uO*^y1*7s}sy_rfF}eH64!K02-vdfKQ%1l3_Z|Z~ZQAAPP0U&?eh0znlGKo$((NeDcLhYFgj)9VD#2Bf26*FcY)d~Sv%avMV3&zu~@WQ)y4sO=j~4x z#D9`Jw#XmXAVsAOl@s%f_+|s`5sGU&3RbdhjMdsHOxeWAeV!S;KjlakL?d`jC0J8jcyt z9y2ij$ip-Fv#k4kA)|B%zor&pXJ}~vbd6bGIHrkf7@_=EiB4!2Qa?fkr^>itX6fn- zOCDLUtNv(#SYe&1Hbhu(K46jI}bku>v5S~|jF)|+9x$b-8JTxFfEodggX#zPye zZ+TwTuwrn z(95sq?=2O?W#Q-uncT{nS{fv{F^PuaLIaR(vtAmr@QjQbeYC{9MWheZ7Ke5x#da9? zLJ!G%+>^fyQOv*|DMYxf1xsq1u89_MSf!v2KoWpccW?rBdqqZqkoD8>V2 zF#Uh3s7-6KESHYR{D@JMp5q@)bC>5QHAuH3oi7|*>-$)nnW?dWPd4KKldFnuub6kTMwDv7@{3zN6;z z+Ds-ZyI5w)sMjNR&pRp>(k$EhvtWMHH$b}d&tFE$vRF0g232fsqO}@Jo|W!PC=WO< zgYUF76BovE{?+wTi-M4U!j?Y*KR`4l0au=PaJm-X71_%H7T|7+_{;v<-9Td^sFLKzPAbxD_s^K|9vdRbAmb>oTm{IM~Wnr zI&n!gY<)fxRN9AMV-DXEvc+Fng=|-wxHNA@VM*;O$JKX5P83bsE_6+d=LviSgENUE z!d>Z}RTVNOQ@*?A=(mhYp1^_%cCfRP^=Q$i~g=hsO0L||e{(pa*X z!=g#TSjR2hm8QKupGuX|=r48;&#`+%Ps&pDWWSyk8WSMp9+!8IrQy(g3wokA|N}Hpwf})3U7H zZ^@&f`MtwgkX^_5Q&+$$rTY+uz+(VmM(_<*kT$xT20fjfWwaw5z;d&RV!D9dIMIy9 zJD~Gl4ker0Po{e;JmY|OnYfvI&jf>Cwacq9)#+@jgXc6y>|KVh8)PduISPY5vPbL_ za4y>0tKjk7Rn2&$*#bKz2q!C!Ia|pDiPZ-3!vb`AU5euxsle~gr*#k`0+r3PZtQ&t|gP2GTH3~3~ybGx2@ zH&)k`_KN$hA1G7#?#w!qbJ_!A#HbphgrhEm9QaGmybX}cV#zzK(kW?G8yyB9<)aCX zo)jzure)>9yAMtIV!i(E|0Ch+n-El7Y*EhymAV*w7%8YEOcsKx zl6i=*54TwDkj9(_6>NlSUuLK%W1(#!WtcA(AVzpk99&OGv_i0cF(X?uHACYJ4_)6g zVBfqs&)eFsV8=)dKo z!Y`+v4onuE-lfth08hY&B=|KOjQ6&K6n%lcEwIyAT3YJk`nDRcsA%TuM{RkCzC&!V z2BJ+`Bu4wh`RUa=)~8X3$bEa#kLUWZ8{0c;!c@iD*Zpy}+}B!aGL{H8>fg={=2R!y zlJ#^5v`dO?L}pxW*QrIWm#;ig`D*t1u#qa%4$HozCO|44vXbF<_06OIzyxUm2Z|6~ z5wo{MVwDn+8BBsxdxBFE!E3cSIbB@khs zlQ<YLz%#waBs%~1 z8F05m6^Qqd@I15sbn`;;bLxtL_Z8ypu!zHy#zUm;bc{#cHa~p=JC@KOZClY@uS(ZhS|QJdAr41NI>a2~{iy$qPFu}Oa=p!J?qBLq@;=y`HG z>U38@>_IRb4m|dn^P8(#`<)|k)*2fRuU)+sRv&BYHJC8QXtX=Zn@}>a*k4it)^u5! znAdcCj@zBG9mh9awE|?};UpeLW)C}Am!_zs`q3K&JNXeZICRb-lSFoZKuCwDwQpO% zC``FxGhLI=?${^meStjXm>aoSS~)qGSCZAyZ5mfI&xJ>gzTo{eg!DWP zjA}-PaE=x8+q+y`hb<@1RMDpY_TL%gua=RjQ;7><1!qOH9grztCX_4RU4-pFydUOv z?J4^AQq(mxci|cr{JudtB&DV1S z8qN0~tqxHw_F1ZLR)WK`KN@(N^0i^=-B3-ag*QRO!C$B?0=Hk%4;I!3$+97N-}B#?cGBCz44e5~KbW zFd26tBi-ib(c~%*{!-n$O}GscUT}n165@6+>E-G%{KA6w%2+kf?#Fdf@=j>ekrH1BunPHUHBWHH!+kO$Wf>t?K#53&Y+U+xftTak8=+Xq zzU79ygOTq3I6S!T+bXYCfC}N_|U{CGZ!sZd5)Ll$u4_^nit(Vd@ zpuQH-Z?W^FImk<38yJPahAYR_hFa!dlV{CgyAJ^w(D#?mppdXokY-UA$sAXt`Ygs%{>UiHbWb-i&poA4Z}ONH+r~1-TxuVCcb- zor7iY6XiD?n-jfOJ_&v@DZs~RL)yo|S+C%!ZP|{g3@Cf8|83<-T zN80R`GR+_bA~F_+0^8I>XIgBtwG@q3HbBLa1x^r{?~c)d7fqt%*r;^5Xp%;?By;O_ zj#{l`!LhgnyCdO6FSWpr?m zYT~ncMup+lyr`vyj*I?w;voyO*s(0z?eW$ZXp+$sLrTI=cw{#1JRbG;VP2NkZEvJ> zST^Dz8JkGZ?NGxyVffQ{dV-Kj=3;_n-tXJRC|tHC<5Aiz5Fjo~eK8Zp(+{ZyCy&73 ztq~OLF@JjXk!pDy{eW}E_U3_RR@+);Y6VN_l^~)wjsK%O|HLu(<#4LNyMcagCVTI3 zmK;{?NVxFYFtMxD|1^-8lc}42)k<5kzm*#8Qr~)krv9pJKg%aiDR3e6<|7R?%mfkiLszS$GooaA8ht!hG53rSe3Sj@5|&Rv0J7(D;9woI8!?@tdHZZ z6 ztq^gHd%+B6@FD&`dGERy+Ib>+#|VFp)eJ-jf`S_PSmI=stkNNPbb%JCvj<+LDu1W0 z*K~z`ZuXrMg@K8Wn?9N)9HQAiIsA}(Dk)TVTPWc+NC1t@GbrfVU5E%Hx>c=citTPl zLCN@W09u@Lt0oS3ejunJFggn0UcQe#x=lAZ&P!&!2XLf&UFFvIkd~f78H2 zsdj^9;W|AXw;Mj!UtbA@1PT2JVSV?0_7;-QyCgFtyPG}cV95v(E7F|h3@|ES-S-hZ zdW`$4a#B{yYg}9B3j3^kjR5St;2(z12poBE1MoHnmskBM{V^6DZVh(#oHb;{1)P2Q z`c>_-=62*2)J^FQ1DyTsVbN*#Wtb%iZ~J>Qcu<-A8T}XP+AzB=ipQem&B@iYEo#b% zjR2ocQ?isTMR#q^=YQBaU?F}aZaC<-O*z{nL)rs8XJ%4dk0^G@WpgO-lGXx7f1RUw z;G(*5#ClP_qvjuZypF0w_#&*zy<3RDSe-cUn!{piZgY)w14Dm;N537*NckBOd6>Ih zc{<981e#A_216`Xo)r{AIM3->FTwfj(Z+O-5pD)QFKU^N&$RttbWay1GHxzdpT(XnlqGm&G)!g8%>u0ysMyrm(s3jer0gC`F#Uneq(qSFwv8wLyxROTuz3){wI7sJN@W_Jf0~S-`p7i zyQNE!OqGR8=Y9IYOf2gdJ--}Wa$Gh8SP(ewZdX4u5RgitT)R5YAJ*n#G zqBpvzTG1#{n<+3t`8xY*0%F|iv_M@8FvR_p1v+@}zhq^Yb5_K*IepsngG44F`kT(@ zPe2Ib+j&Ds+cFt)&4-`>u}M%tk2TaBNU zC8l|9SMudS@YL96v`HZeh8xOYMUno{PUXwd{?QJ2fEiGcAqmxQ1ymof&;YNZ?U7qQ z=CrO9!!uIo6ff(&s_%$xJ88pR${T=yFpSYU7gaS?@u~Z4pmaNeP#O@&x_a@LTk1X_mvHdIq(wJ^P~^xhHv%Ny7Pea+?~Lxc!xm4^ur+cXp3W z5Ug@bW2;j28)oPm4CG5ZRa8~D{bBzy`8Q3+RdXo2@9Ltf2x;i?_BccLwM6||SCOL# z*1dy|@z7-Qil^r1R8yi3N4n3~^u6TqhU_QR^XM}=N>ETc{3p+k;;KCz0X-j0`R6g^ zna721Mv}Flo{z({Mz6#Cq0XuiCR+dqT|YEYDrFTc5RuDHG4jhHOS`-03PXRG>TH(h~U)0XoAQ^#BlDO)uxwd9F7?KneRCgQXYhb zPV>OoI;)Dd?}La*uD3_9e^QNkKOC{zT=yXn2cH*&$N+~Z5nNY@7-p|lAZsM)39PR~ zCRTw>Sy@Mut+6{Z4nW^@%hx#iVXPBOMUWf0D_@t;zoe`Bi+?>J%=LNWklnX|A)Xwr zLt-Vg09%|rkjMv+23x-OAdg;zo2IhU9?1?J0v&HOa}(g=n^n5(pEJ$pzUNxX$jHoW zV5Hy7qIp)1VFzZaC$f=9-l7L9uW%{aKA)ypvUILrzX+Jf=5Vn)6Ir;DQkDGlJ!Ix$ zsM^C`7gZ(3ZnU-6BSfjwPlklB@t-2I9Y09wvmI%ivf}~`u}jP3^)hyS$syNRk#-pH zz@`E`+$2|Wk(Q8zl20IsPu0(06w;H6xRZjor z%gmFjSH52f3+)v`BnvNezlj{xy?^CAR=w9Lv@qGfmd`@n`moWZxx4=53AE}& zHSypVHdJRagvIy&P+-pRzMqr-z-^gT{;CWU-s)K#HSh-R3ak@OunX-?L4PV8fs+QB zdwdm=;KBQu8Pct3(7YN%^i=WxBx;a@8CRe>FhLY^mZ02R8|Q@avkBFpA|u;_n0Pr2 zv_XOUZ#A!M)|xPlu04GsHpo$fdO;Gn@OFe5o|)V^UsrB$n}M+8yU&$GV|~@cDo3mq z?Q)55gpgJbn5&g16H5@Yrk zQ~v@}&i8CY7LnSUr)I-Yk)n)Hqmrx?;UXrkDZg^)I3ph`Th?Vasu!=t9Dj8<%|T&0 z;EKPWm)7!W;n2XAFt#YKkrqEXYSBePB=MIn7CI`TqdmW?Lo(uCmy)D^%$%S~W4bqV z#>h@g$)Tgw~NB8A8VT*@P6qWufnA%+gZ1hu14{2_j#tXa6K=9`XaRcu<6X0)`B}S z?bv>4aM}TcK?@-)s z_0_fi`0BOy%HbnPku@Zs(CXG|<8^b=%ZZ>KgxKIVhoS?+Q>rq)oOz`mwX?&ncEf%cBf;bCV=a~JZeqOH-voUf|L92?CighoyGiHq zx6lD68?CPU{p38|*9lCshK0IzyM)7>d#CeV&uDSh=MUanTjSu`yuUs8fhn*?hyoFE zr{Ge}O{yCg0+6Lh$<-E7)Hmg)%-@$75oDGdT| z^5|f9Vh|o2E~*Ojek*<&h5KlF2fXxAffnNN9tQf`N|P+sVP>7n&VtzSUJRS}-L!X|{py%fD;N+3d7S}XEkxHr7$5!dJ z9l$hv3exr@F2N657O68^yeU$FhV7Evb<3Y=VO*f3advb>of9cum{@iL#1YhfG1DBM{aOHqw_OtZTJ0A_iS*RLoT?Nm~vjh}p zxztZrwYTo{(oMw8dhAuGnC$P@*7H@%S1E~z1`Q4lY$n+%i-=QYNI=S%pj#qG8e8?s z`XwAVVKmJhk4~@;z|V5lJp2LKe=6tiH&3BkY1{u%2~izpNE11DK4r=L1_2o>P5hVm zs(7&E`xX`JjuXO}e@O6qFg=$lI-Q!PAsT=(E}$6zDGFJYokh289VavwyR|bIy!7&-yqdF-1?id$$@o=kIGeg(`HHx2tET@ch&iQ5w;_21D zO@jy~CL2Qv1KDjErRv4c)NPjR8Yxko&pKgGSRXT87r>`>{`d=h)8@E*?HQ<3F7W#% zwBHLo71aOQK}&ewEdCFPpoV! zTA5~p)}?wID)za^0W>cUnsDl(LURkPuQ5H+0)_Ci~O){x?3cfUsr4 z1jNhNFX*IE3g^&aAAD^jW^-HNC)+S^|I*G?Abo3>rH zFJmdPJfcFl;#KISwcB0xPoGS(T?KwI^FjgEFudRG*~JkE0E~9F-;`H>-2^_RTBuME z4d+aA((K3tI2FUT)xGBBu`qG~^PL1F%m3_3m0b!+r#T(OUSd+uhc%qF9wpqd2U2N@ z1OFUKhQ_k1!_fMzHEZHZMMI=(k_zBNBWOLm09WM558OS6) zKOe*gl3q;j3_t{(?94F?&y?ULRJ{9&mu^qWO`7ICZBrEzv(d8Z&goQMU!zZu1N)6Q z%>3KTX)oKU+i}H^2}dq=oH*zYa{u*jU;o!Kl_7$1bUYLONMB|_N= zQWshBvt~ZqiMA>*t$Btx^B;tpV3ywpJG(LjxZ(zFyxTjS30%^DRJ>3VFhhzu0l=CD zi5BL%CZNi_HkQ1VXJ258=%har!6F@ak`$L=NR4q@?Hburxk3608EYHwobEcqEIctX ztVE@$c%$iU`+>TMSQ*-ysp!mF!qfDEak|a8WBMignM)Y{9(LO+8fgOf&5mjJw(RVEU#o-f zy5B;Qq)`>ZTN5@V4%d^DBoI|_A5dM2W*r?l$x zSg#c7v~4{F24G1H=n|`|Pbaqpi`7e#Duiku*00vD-uTfc1Ttfr1d7Sh&0;lTzJ&fE zd5$h51rHLZZcDl^e-3vnU9oC}@o0!709H?$<$D0%&H)A_`6!}MNt%#)3UPcyhUc3; z*=djnc=yx|JyTOhrAwb<%R%WZOf0;*(bWc8)!*xqA&>#hGmbw7>+&djO%|7hj7)26 zkuBmFsB7bu7lUR@>!Cmj*Gs#`G&O_sxE{rDKPYv12xr{5rLw*-a7}3%u#M`Y7C)Lx z49{=7Du2PlLrH=ys!3*4NmtL?^xgj||LC3Zr^VPBXIRzy_wMZQP?tJ=^75GLPl%m$ z)3L7iXN#SD6qziy)f?ajD_{Pad4ZTFoVbC3!JD7~^UYYvkeBQ89Hm*;<>^d=JZH6K zkywJ~LEjZ<_$szU8~)97T>LAu46?gFG&rGTZh^Lv8h@YQv5;Kxj9B|@zR2|svQZr2 zShOIK7scr9!+&ppiz-q`@9kW6kj|^jl}wjLKqhz3`Or&+iaow-?=gWoGR=nl6L5pa zW3>D}T}vc_Nqc=IlrdN9QwZlV+*^m`phM>)fzzJP0qFjccD!8miCSpLsjwCw6I(e2w0ml509N*6i^JZ9CF%7%2`r4jOk5ul6B4E(vZncGRlNqis zR(doaNvUQks!(yM2?K{vw4*%`?rkVou5M1af-_dz=R9K-w+Ifs$gljDPCVsoZ*(7E zLYHph=2}b2Al%_muq_R)T~P&OQ6|8lgp0oG@UFj5oFlkDy|xlxs;p`T_xMq?$9zC@ z_9(6909*;I7&kFNJbXh`*`w7&2zpqoMpD^?q)ynpzXpLSFzUx3){$HV(ae99d{Lvb zUKS}nENU3#RIHVgn#Hy0E{nkA-x#s+ygR(w(m_I#j$!F_@tPmD@~$Q@305=7{cX@` zV_3}sw7qKE@IEr3_dLxfpfC-KoF0%IK}$DPDHI>JeM#78(%^um^71F;B5g)ztTAae z8X-;|qHJ^#Qdk|3a?VRjl}Hf%C2L zcPG3Z;$rb~RoCc#ibs%}_0lNNp56@%|fU-CIqor_6VL8~GHLS}8pr2X+ipI|*e z%0@@e9o=3W<{soc^5d{Y$u(^|SB~H)TM(z%@Lt4B&&d>P zQYcmA<2k9PqwCF&C^Sc543WJ-IDjNVfKR^r6Dm;KOn@7TK3<&#Zh1f9{$_8WiYc(9TPT z(+Na)()tyx>bvJP(XUFSR%P}Xx&OHzx)>yrXNFa8^)DD7GOvHbky&DidT2@J*Hqp8 zP~jugcSI3YH;UZuLn$&0q6h1kA+{iz0=M)`Yf>iu@DEs-SC8o+#SKc+ch0=65|Fsu zzhCjPhc10O!exG54T7(!Sq%wE>$ypdcRK$`N0&Z`s-70pKe8Ylks8}n0{|TCplDYKLYy#6=7>Y4M6yYjgOS%Y7x=^~J=c|Gh|jO3E==uvA+`t*M)0{n&y0cN zCUe4(4(t-?`hdY|?cEE1hdQ1et|2=bs0v&Atm*yfBV@;p0X|}K zcVfR#{JtQrBDt0LUTdVl&}kcv1fL!DJ90jxGQ{&5z#qP)w$c3R$kiCn(0GHd{(3+Z9(&N@@Ts;bD;>+LSd-X#QgAl9cN2~pH{KGs~hf|?o zJEoL!Ewa6bL%%*YT`1a_=2U&?uRK4Lvk9B|6$&@L6@|E*wufkq82_|FEfhZA^p^*& zR4gE|`5Q#XX;`08j>`^woo!x!2;#!}5;A8CbYEl+eX%-rc)3SI@&_nX$I&%YJ0QIqo-#+pQ zUKzAso`XcLOR*r6sf6uJ^%Jx?+0kFGD;#3y&WbW?4?vN*#vtUnR{fEaS&gGKo)LUqo)B6cd9eASK$(MCROZoGOS(!AFOuOMPACV9hs0^VW2E>J$Sd! zUuPcMj-59<&XsS?=$IsT7ZJ13xOhM=b?@`b%_>+$cy}kWqeHx>6HYP@Vn*RiAHmtO z!ROOoVqBj?QEtMV^68NSgBT=8QnRP(n6%ttb-k&+-U6;;{Ug+MdSQ zeSe72=i6}R@8=J&#~sX4@r~vH^js3+zabiG%puNF2(gT?r)cpGyoU-QWyq_lYvpQ7= zVzA=7-8;Ncej@pDWpe_4Xb+XK=FxG9wo3gVOd|{?EuX$U4S!*Jbs0>j&{SiG)JOR- z`fAk;{kU3KW;$CpW00$%$odaJs)CT(HkY(1G5}J>G*0%8vdU*<0!^6-f!+p@f@+mB zYIfB+=|2Dbx;${)vj3htfjeN5AHGlS24*6Ii!3Qh1T4TzZVWC*NB0gGY@Ni>2oq$N zl?C*`oSsavP;HcWC8%tn)*rPrzi-vyZ>f{L^9ZQ@`dkZ#En(M@11K*kqrP$qNT{vA zxncesh&R9fM2wCNc+v=Ja(;w(I$ry4bh|7fn$nTdzva9vc&D{L#lfOrW}oKvkw%+m z%bm1wH|ZsnjiXGY^1dfcaYkl5Wu=AWZ;?=(fr-Ykd&kmBZm@W@Z49YNvpH1U6n-R- ziCw0Dd~VF*&L4O!@Fy1|Eh>X6%fu%ykbS=BpJ#T{@%;}VNOZ!VObvnEN&#mp#2^$t|8&NSyQW)cja538oqRXtb22Y9yQf7(hO%c=4 z-(L3j12WoehPNvC+`z|o15mIGthau_y^HtW5u?u+T@BeMJ)M@edzs6JXh zVa+=qbZ*>*xW0WYB_y(nvQe;bkCgpG#GcZmUsh3t|`rXK&S|1pRD7Q7u`8jHWV zyPLlK4ZbCj%mbN1U>G_Z4b6`IXxl7)h)>p(A=X-t7CO|rV*lA4ib}}UBZ1Nkx>fr| z$?PP%=;2?9`=3bL%oWPBl=gjqU*sR;$%ibtxRnYZ6^F>+l7=fm5X5Znx)&OU^nSmXfz0hliKL27(iK zryt!yA2Xe83Y9CCBOvQBGSK5t_>b}l?Oy4HJ0ig^IGw+ol*zW-s@uyp-v6wns|5x@ z!%_&^5@qoFei>q^`^Fg#u6#9_)qSG7dw`PKT!Z{Bf;K>()CI-(7$PYL{1fUUdYrR< zPsE^U?|-@^$vTfOSdXd{q91p@{o2q{u_jO}h`eAaa87eAS|*0{yLpw{9d`1#&OP`p zqh2dJ1IWO&yHpuH$bg|FF@N6px6nMvk<30aR;^wuC1kyD*mGfg8p;mYi;9Xs??*cp zaX-g|R&Wm6dLhh03BtF0^*g;!x0<_YVN4uvRDPjfjwDKv75fZMJ-Y$ zO-TlVQc5S}+EcRrZ5k`6lR9~#Rv7O1XbV+6ZA3fx^KwCg+S=4+R--=;qEn<5`*ciBVLOfh%_nUftIfI^6wFWa$PYE^agx z1qW{S(Bjx5?o)uU7Li+RMY9;k4hLp?kC+#_?5y|iZf1DNR;2->3?2GoDE{kF2`z$a zURI6VYSGIJKO)&Hy)I5~AD$%ygCVs!*sgC4@(FkuE@%0kCZ+2U&!qK`rvZ^OQ3naE zgLU`4hZR}@J~0IB6tUKnTmkfrkH?3cBKkOaY&||v-`7NRx1iK~ngiUS4Povb?fV0) z!#94mfT}gtj0k@l3hw?S-q`x0=O1IFyEC|v$R)CimoCi$X9jVQdo$#IP~~ln09Lzx zod3NhXp$liL0);*{$FTA{Pc7si7w%krrFJ7o{kyA7 zRzXGldN5`EyCq43a4H0O!IWZ3bKY$8V7vik5GMYp|m4=4x^m^?5i- zB%)bb`newJHa37932qWDtjjU@=Ui7~uuH`_6tePLS6mhY!s&rjG={oHi1_lKsR(MJS-`nX&b+9!3v@hZMXjQt zzpd_HN?o#~BM5l>OQsdZt2L_5t-QX*SJGh&>s4KFpJavwS|r?(>=eFmBZn6LHs>WG zt|5mS@{{*m+MlHmAr9>;_#M{h)1|wPm2Ocf68L~Pw?X^4sq4*PxUqcd>R~T<19>p& z#vhP-d#`(#xPOMl4bfw*8c+3mB=>J@{i?Z|DSV0hig9@O`_g7@8u+(SaO}oz1S8-{25o@Kn_DufK038of}G7(j=24 zq2gr}{fYno@2l1=7|gdwb}s#x>>ZcosbKtcI#--ON>25;#1my&L%_M0!I78+$aG%O zu&imOzt?#zdBFLAC6j%Bo`{t8pPEHyUfR+&ZWAuX^4xPM+JVS^)w=nDD`Rum2HHD5 zW&O4i)_v{%qs9FDM590%^S@)y@qOPsEtZWnUe`eF%qIFQlJ-U98kD@asVMbhJM6|Y z6HatmDNj8b@w^@;*DvufHro+WML~jRrzf$PPIK!lUIITlJkLZiio9=^_o{fAu44z` z+b?#9C`6v6TZIsp=6AWsCXGeBGaA)&3aOUo^&O3|ea{5=wCjlQ5^`UyuO+V-j33d1 zZvCqD0x6o%M$hADlbm7qQ`(UfZKM4gLdnvsqfz{Jwiv>U6e2FT$9}{b+vXWzA5v|) z&ri#xmu2oY+wcBVc=W* zfAzKfY&MU>*`s*m>q)!r!=J&pi6C~*w4%6OTHs6vDgT!y&spKP*0@>M)S5cf7O1jX zRftH%a)b$9Rm%8gKu-FS(=K5H}nO>NSaVsuqd2Ske+QFno z4V)e+G=)vGwSpgEV8*Q&7)Hr2zv-iDW(?T=Ad35Tv@8ob76_Fl@A5jzgyifQ;{1q61zx zL)|w?fxf?2ko5?^q~xpzT-={s8*hAVtnkQ;713pe(#_ac2EmDvl7;E(1+GjepYt-H zf!7a4;HNaEng;^_n2d;X_0U}t{Zi@wKqJNN7f7!T1PTB+0!IKu^tm)#BDl1 zKkIS8+T}wQW@ax64u;3hY((mBHq~gQEYa26a>@C{K*py~7Yh7tr%PP9wTq|2LVbRE zI@{~PECoF+*g$$!*es;c3G$~sMozN-?S4ca72F=s5GS-lDOtJr2b3(mGOFWqsKvr0 zi}9rf34g~D1>d&DqZDg7!|2hpF)vg09IDNT`g^p^3oSYKrq39|vg`ao^IC87REkDP zv!_$bcI$UfDil#m$-=Sn*)98DYj`8Wdk{2uYQrOwtYQe7G&uGVCA^YFDdzO6LmE!kUP~$~-nXS8@60IvOgMBr0tW6EzSiE|Dc>CjitVTkK zWtTE0g+sL-<80Kcva+gpyEmb}_vghTn<+TRnFOhUI(8;;`E#xIzFreWn>=aZGd=!0 z-o*#X)bHS{pZ9}s{u+@ko3ZR5e6{qNr$|#qXnu9-{u49%`}IlTAH&9nD}# z;}OykE$8_g9>9CaDlOC6^ZmA=i9ikPbDh8rU(5vwE%El8A%Uiz^rJlJr$9 zO;sdg6KEmVLg-{;PjL$`;<5-OI)pc;qHlenf*j|dx2s^nJH(ZTjtzH=KU%^ibX6AE7lXH|5#ShN_*vo`MOFSaJ!Q#khb zd+^#a6_VU)`)Wv(l^c_E2a|Crcmlxm87e>Ljk!{OUCuEK2ltJyK9%JVK4QFuX)-Df zoK}!-aka^QYo_w?5*e9{ug-|of>3|l4s0~Dq@BAct9@k`2TH!FM=O0i^{hXFSkfa< z0T_+R{cBon_hLS7(p(}mTx`M?Ht+keaDbZJrpeDqc)ESWlTH6BXC$||O&?e(gfz;P z+LWD@m2_iOR!!B^6T6pIjQNmR|9CC|xc*W{i&>LHP#E}A*-?Az9QWx+Yqsf}J2>7+OtP$E=^hsWc3VClO4b`2$ zTmqrP7vc$;;bu>>dx?cMxBr6-{|ir`5LAKL{dxM@$F;X8{abekG|^3WcDS&f>v%Uq zeqQ^MA8YJFZ-t`xX;(9G($!f}94o`I=cd8Y3cX3)Ou7+IZ7-+O^xNHrI%*$-k0squ zU(Q%kOKF(Z0e@q>cGv7v884Ma;{#4M(=!Bn^n{bL;6Wi$Gxfiz{aq#D6vaPO9N{yU zj&a+}4LodC4H}z&fdpM5pTwRq7d>?w7irwAKQh)%qAwd zCSSIG{K4bMpA9Jd;zi;Al`*5wag#r1s@?C|{_Exp>P%m`(h-6qSPJV_4q;^;X}RK> zc?E8N&`A#;CwPh)k^eksx5yJpq(xWYXDUx*kg+{*q3PR11u7n4$kZQ!{jEQ1n`M2r zsmbE)Oeg~)mGG^UZt?}TG}rC=ONFZ9ws_g8h|I2s%4cQD>itTYCdx2nhUz?TGWfZi zQs^T{efVd&x!bIfmDc1#?v@ip1mtpAqSC`MuO3X++TsT`3E_rP#bJ6%qrFnS zt5JOZOv5D)@%$>_@-aqiq=+m?Ynfkc0-=8+G=NvJ$l?Tob`k&Pk5a;VdiUQ@;BPlB z&lHSt{og#Milm!idpq?K#vA_w*NI}8rlTgmg8Y|xhw`(@$3O`jA_}GJO`(ULHUxdT zw<7bYK?x1m$Qaq6U6yMXw-ZBlHX7Yg)OSYBpNSn0?<5Di3PGQ0G@D^a@L=h*v-S?Z zR|fJeDcam}UkWEM=h7awqU%-%+J7r!`{I29)*5oos9c#XpU-u*!zYFH9P4D*OTr92 zTy4w|s<=AR%0hVNvqroQ5C6D}r1ti%aFtU83@Nj5|5zJHJAE(r`5t~EVEX#{nmoSt zc)qj4>$-Jl!^`zv<9nTZhApTO5HiFZu^<-P>g4)hkY~|MfFw&KV7kb}s-IUv1QDj% ztrkYYP=eaoapQhKdueOaLpUDVOWjSlXp7$-sJ?i9m5nBP#?X7m1&}I0b6E|8X&2fz z+fEg3z*WUO8SW8=O3>$jkp0Ggw&@9@5BaX~YLUC&QaHJC8uPa#=o*3!HMW`KWH?b- zdK*&@RR~l+xBHTHvYg4n+G1mH9UK{#!^J?>?s6k1<~!`Ea{t-4{uRdD_@RpX_3Knr zZ?`jU)2!vdLgm=J#U6hXDXGUIW9ynRAe^|5wDY7{Ai|M(Axe|p`Eo0(H`ryD8^G$m zzZPCig2+PZk=jFO_f?TKAa93SiC986AVOZ_kKZ+#8~QN`M|TUMTWpLd%UI8Fs@uAK zlq~7-V+D=;BLcScF*Lc=40X$)Lhhf_h&Vz$(&`vdNK*dz>Y7_9A?`X=)H&jc`LXW4 zdd(aA)k?!cdAY!u>!}se)cD=rR!3|*Twx?sVh?oqP00yzDD~m<6G*G{#x|bJdINsc z-&KW?K?bau_JuR38Iv*l|a6hY&uQU zQc|D;Z?g=UE>tctxE1){isJLhB>C3A>iD|%f%~y}LG}#}EPCzIwy(p&B8$lPA)5Eh z>(by#<(=#GKDl;y1j?749o)Em53}$Z-6NA(!Pv6b^Ip(aORepi(X}Rjv&+TlZM8X z&CrQ_Q5d#AYt|k)c`-2RLISGc7Yc8ut+R1m^b#!wrV@0x9ogt)zrF_^>GXw;GR(&Z z+kAk<5QH%ej@9q$0AHRY>Z&riF3*Y)-s1-mUPYF0THjjIU*~J9A6@OR_nCCsYlrJ@ z^**~FK|bzw6@0|F8q5u;;_6T!alH@b^ONP%u&)SYXoT%g;xG2JH5CVHL6&6U!bd(t z8#jJAr%#=u`v$4|A_===BC$?^C49DEU$ISk2l$v=Dz^lmepO=)Cl#le`ZW3lFrY7P z$SpMgvO?FWcN;HIG++8#y{9LIh`DFqvZf}xL;N)=V90pxCfSpXemeoqEC?O?hlsj9 zXL}Zig%53UJB7LSw3|+Lr_UiiK6#C$TI4Ut$X@ha{w~6NKQ)mkf4*2X+sWl!ieOzr zGmXgoXWmmPH3t3Ydh;=-Cpe%Mt{%`$yN^P&1pYirCl7g;aDw!0Cgd;p&jL}7HNO&U zmfcPrWb7!grhuok)oKqKWUX!i5FV&f+2Wq2a#ScP0bPw^0vJ_O$TcZpyD5P+)No`6 zJl|GI<{qJH9Bpa-4Z3@dm)}gXs7z-P6+aNvdMps`DY#8bG@$Mtm|V6C;4fA5+uSk@ zk<#-DSeOy}zX<)b8&t#zL!;vdKrb8j}=`&B!y2S!m3$4 z+8tXkqlhvjG#a#NH+(xg5cl+-zZ$#?F{P003efu7=I=Sd?3j zdkTI@K1!z7>9pPI-Q6|B6z{&%iaXq`Jb_0*AmDa$)Ys3`uDXZ|J2GDl_{Vx}VhU@8 zPSI){CNYd%tFhAuMLeuTU$ZH>uP&^bX1ShSd%rIQ+GM4^+p+h-7va@f{b=3u7ur?C zOg5I$vB==@eCvO}dpo}o{kZ%zgJ0QJhtpt|F3qEfh=;HHdpD_v<p-tw zwt#8^{axT7ASQ-ggwxicDSsWt*d`RvpgoF@2`+HjZN&Gy2@Oj;7_j09{pz2Z`I5DC zj;*J{ffE`RA0`c6#?j~r{Ti2#7t6S5^3nZ8Na?%)orZ*7z?( zO8MCNJS|75JtxBrXMdQSm>3Hm5^%RPC39UHTep|^l2&?JSMt&6yY|K%)3=M1!Y^Is zaxQn&1K&J*Bu&@;}{5#2&viZRMB}Zbfyx*sc2SZ}Hv3(5#XMd|t z{Ku<2MsYDU$ML_M<@6rU3Tmq{mTpKG)Stk zec!BvwbtZ2yu1;0E4LDz|t53CcICNUbm4Uob3MCRd_ylMq$}sWmW#&#sjVT@&_dRt%i20lh1Rf~`}p zsYe&@kEa;x?awVo#=&)LEN0eBgTeOQTxM*$&*I+cZDk4Mhex-Z&b{s(b0WPEmsq+wFi zO=pjl@N2e_I}Rcv;9BD2&!no>JJtS6edE%EsAW z@zQ4yt?}lmy@?vBjBT=pX<~&pF={(swCY1Ppm{SLc!DvmW!^NsylLt;0f53s%7k(t zSxpju{jI1<)&gM~M7YE!(wXQKKEl+Xb4-Ia3Y8GhSE5}-lTKs0rNy1YA$H~JlNS?Y zh*PovwnEX244ZeR=1Oioi%fk|#6V9yL4D*eVRlF9vNR|fj-HS?OzG6>z^^>}+7qW$ zNm(hpy6&1~;Lf@rI4>AOfl`xrK=OU3GE3dK?iU^T81vR^A#sLNgxcV-Y5=DCPZ_sd zKL6@ptrhkBo$?%4O+K;mxY4HpG~4=lp7~`;Z#lapTszk*=g9nFC|u>78mAg@b^k~4 zSiFJC+3~vByR3Q_bbGHz7ZqvCwGxug$fUxCx&9mc>S5dWVe;{_%Y*KDiajDg*Q<#D zd%y7@)yuVver0|r;HJ-E*n$^{-iis*7>uxXmglbgW{3(zUxC}BS?-oC>Dwx}rabNv zV^NeDTyaV5ZFYz7{0Yn&G~~BHs?lmSis%adrz8OxpU^u&d_BcK7|(pE<9(@{@H}@I za?Wuw6gd0)az38W+OJ8(T6z#p-}%^a;dyblp!1Q>`<##9u?f~z5vO4}!vpmAT#9`0 z-Rv1iHtutkw$g`GT`XtAUbSj;7C_a?bU#Z@{&?OC_x5%=PK>aHcu3nb4qqP}40V+! zHpJ}W_-ImqvB~6o-G&kM`dAzaK3{9YzL#d|2}Z<4xFY#n1%JU5U#r*h^ zu{lsl1emDs7jq@vw&EiRLi7wmla!MF9*>R5v|R&aC@=bi!YSa3c;Ivi z4MD5h)b^-O!Epsn8tq8TK%C<;$SkxS3!42TiLZHLa;a2Y;`(be_x;BUHAp$A18i+$tKBXKMH#`2*{JPU7LWKQP8*MO$B z*#-vY*>P3ee!O}%6L6Dbt4K&j72SKh_pIAsQ)>s3`mx}gu8fZsXkZkYN@;9U!-Mli zgdU~I4AL#M4I_Q6KQY*A7Fc zqZqUB;=xgMke+I+RrL8QWC+oX99ryrw&{Y-m?t}{a{CKj7B5Iu6xgmNoG8Lx8~hAG z%^w&HpdjQUX+=eFo@~>&Wb5U-SorU(VV#e#%Z>5<+GNIAJF)oE+Jm(rY+rNAipo!v zkX4joHQEEn{h;jjbcz?m%aL7iH&S0vLB8-?n#U-TV0fa#vyI=+$bF)H<5=f}6j$zf zvgC+?^Z4F>vtr(CFskX)LyznE??L$$=p52pO91_-0BNxwil$>w46Py39Yd$2Lyq;9 zEBo!#gvbT)YIEVOFBByhfuK^+NhJkhdtSaH{O8-)#w*-9z3uG%r^Ng@TnIs$dSxf8 z)nECh;+Ulya0Gu&J?9^qgOA{uoz9mUQ@1!Oo!2urIvzX68>}G24?vq8l;_2n2OsAb z^6U`@lOlB!I;pO=-cT7H?_C`PrKoj>-;xVIi`9k(YZ|H=-GdZNs%GWP%DIZ zxc<%5LLtHo?hB}laH4-6L^(I5#q4|V(h+gR9pc5IkatmgbqEUyo1@5#owY9lKW8K< zJqlq%Kxt#O>u;HV>NW!;KXIlIWU9ETA%$Za3aIZ6Gm4^0Fa3Wj3`8l;VaI$GuVI_1 zOA!X?6Tc{yL3Xh>sHX%zAb(y4D(eA7uBkXHrU4BN?z* zQ2}!R#~X=2zgDMXTyjW8vr=>AffrFawko!1Hgn=ChPG;h5SI`w!39<03}v*Lvej$& z5)2|M@Uz0}Tl2#DZNu0QM(*Sm1T&IPyf~8zkOg0^ z)}_i&C(9*_{!2mFtPm9leiF+>_AO@H6gl@~SCHzM30t#nu^LPt8x48K#KY8bvJVM& z5NdN;yQENyO+q5A;QtnMobY}1_1NBzbwqhzypV+3Zs!uNY;e(#^fUXf5c4z+0;5~2 zIuph;*|})3AgmHXE;xWcXW|;OP6xB?q=B(ooYX_*8bVo%JECWz*(~A~QobB`_<*6? z(iKKQjjD<9EWNhjLgx8o9s{_@MHwD?=E{Ffr84-V5v_{Vhnie+dXdUG*tbmS14!P- zuiyx`!;_PfRd35xp1ULjoS#yEEv*$>ETBAl8EYBMNQBtgR^XcZ*T!(u9^ zT@aYK(#~zbNT1`H;~X-Uhsw(Uq-v^K{Ul$fVawQZH8Brk%|Wa}JKfN>wC0gllWB{) z5X3jDuv9mh^bb|9vME!VNQhygx7#a?b~g#WhHsfkQquViqb5C`v~oiGe`zA_3%kUu z2814|2^4>iBB0sStk5kCh;DGgk8^0*-6#K1KqJ#jW5DKPLg=zBx%izgJQPU~D(8LD zl1=}Xt)UWh<*NUic}8s0N|N4pp2@%OdaCYLku(P27YZJy7uctyJ@DlW9)O}sQHlK3 z>R5?Nv{ME0W_O#*l57e}j2eVw&kzx}1*Zr(V!UZonF=6!f3eqmL_3_z)pZ-6bfc(j z3s@kk3FvG6$87A~AIaqWtLJR5kI28nM*`g-IEK-NF15*`KF97lr z$SdV?rS%USA1}w+vb(N}cUbccD>1Q$n z_cT~7Y*wX8JuZFle^f&LGbQ#72KQwV^ST^6bk`ZfscUQNNB$`iiHKfi1{dW*@Z9hi znQd|)UFO^OJjZ^O#OBy7B-e3g7!|P+4K(Y#6=Y6}yYj_WDIX@6>Y!LRh$M(J<1RxH zT-stpnGaT&=eE1+X*1=P(f(+8o11|2&)@IKrv|&yQWVrei}WF^c1wGGugg2O@Zd#i zw%Zv{>3UY$K36Q8L-2kCeG1%qyX+1j{j9z=_u#dtvr}7S{w&g+u6M;Y9(Zq3Cvv!+ z=bJk3+`As!Wq?=9th=qmBC?TNS(=OwL#FG9NNB(jLqFShhC94I^Sx$Wfw>!0>@S_YwK* z3Z%tv-zWj7AHF2j>buD|^_hFQAuz(Kb=o}PLEapptU4G-8Q*b7r)kqbkA|F;i?$JD zRe60RM5Y~q9}HJH{d7zIXxbdPv%MAQ?}JdPev^=%;brU2R05wGP49)yYBdY=WXkU8 zBZ+G2J-ixzK0~xCiLG0SlX~6mmbWk(GjWCOHV+t4L_`qLj66I*UThV3A|5V73o<^^|h$F&fFTB`f1d)kgeGfwNpO=-xx`1u^VH3+FJy8yZGQLpV=e2 z(MLLZ>`^#A_{4)4cI&xvbrBu?fy8nad=0IIKxSVhL=lK@!%id3&KCU;S(8i(9??e1 z*3=00y}Bh^WOlZE7_UQI(gOJ*KhetO%ro4sxhKUGVJpEW5%X8aB%<;}aPt^TcR7FQ z|DcQk`9qp3n{gbbTaPW6lX1`xGd)#Wi1z(ZZ-9ll)a_Mt?P7T2t~Qzh`Dm2*t#1M_SUsoNnz7q7@mkA{Pf1b>5sBu6Hp zk|>p|81vjO`Qb0#=3sa*B$# zBITwaiqUY4AF1l~qVn{Z69`@Vz&mYc-AP^5%J@%(1>)-E$8o?=WqE$(EI!{C2O#)F zSyks>iQA!}p+VmFt9;Su+Rj%0vn=Dux!~YnBupYsx2L`2D(#NvalnPCpUHzJp?PiV z!D*Mq41Qyb;i&Ss{KOiB6NU6jF-6x^MbpR5iAv{MXyDq^(L13}e?0U~mG-*`b=tGb z!}UwlI|7<_h4*g7fo9nQIm?23s=s8y10?Jf>Fcd*6V&2zKXyWfQ(?He0S zv31hBT-4iG`BC_V34a_N@yNxR5sr$H9tw9Qu%3piF3lQ}GvNgTH)3Drb zLF+h3wo4Ab`cq7hNyVf}pqFpEPO$?eT7inh1HkY|RxSU+P z4Bb0{o!cK>M|+JAdK`USyx-EH3k!&g|P zNvRpXc2oE4RaXdM>f5N;%-4d|nm`B7?bXtePW~Zn{B7>n+}qHew)eoVNd+O%Zf;j% z3)Ms_zG%LUip|cYjD<-vG0KKFC`9DfS#U68$;hsRAzX&qnew)!CRyX8Dy^5yvB*-U zWk>;@Q}c~QveIg6wvXwRr|Yz}=7JfL1CsF^BCKjXgp|m;Z|v{ z$xIk|p1O%7AoA1q*l4IEJA>nW8Bo<003*(nF+%iPEM^=r-vo*Eky6-9PNV&_g+m}- z0@k|HQ9K0e5G!|7B=c{z)-2wnV|c)?&Ps&;$J9ALWEOVszMgE`J=wNx83z>R-*fgqu=g+bzSp|ewLX_3RzJ|yt9ye{M3*mxh4ZaptajTe z?bM|zM4xN90%*|41~^3}VkU)6q?ZckpzpMa_JEnhh)HvzL?NJKAj4ME^TM{z)J*x6 z4%6_uvODa5U|Oh{q@^NTbuwA2%zqj6paBj8C2!>Gj;qlq+P9d~3 zc}ooS%zn_z$GkW8`bbB3t0Ib{X!>?n4RKF(f573Ty^cit@{py84Cd}C17;vEw}iJ%yV}9 zhP`I1Hz2*L89(!p{Ey?8Wmx>^-p_dFMPVbO`czMU?DT_B;UWD|w`y@PPcjuvB!a3> zX)|xmoz_#r*!^?$M_Z|&+_mux1L511&ZFKZ!6dsd@9XFL^#qAgP9BB&`*Fwql}R{q zAD>{0^Ltg@rxEws-S6jxPam&q|3@{{-bBAa4Bg~I*7t*NjdMq~ke8ugrB}HRSr*gY z_}Dhn^=boEdLEpdda`H@JuX6H#@j_;UOTWzK=a^@IAdpt@Stue3<0J!^j<7*xLZWt z;AI^z;)zEU)b0(CVLe8w^yWD33rkxVvwaI(U0}W#UHLJI7g^)p^%iV}=av4}`J!eD zX%(iaPpLNRTP9fmrimxZ$o4%s139s6eJy!K+V+tA6GnwnhdGe|i;q7#cLw7)9jMc} zDh*y<@7qZMW3gn=KFU!;pfD&#+9ikD&|bf8+8WsffYXgM;zTbC)(V8Be}M2$p{XF++yrigyS4 znt8G;80P$VAaHuiGxB}kHp<2DuyR5=BgXvT*JfH8)H3(G%9-}NDfzs}8FkwIcl7NG zV|}(3)Y$QF0`!=eA*2ngCkAxuGmd$pip@-7`{KvBp?MpRwT>gtN)s9-a}iJ`uo?FZ zt;NJK7?}Q5P1ps=3A9Dh5*LD0ox1D!B@?RAfo2bf+~Py1!1P_sM8O^k?C;%H36ubM zP3YHE+At6nUWEx&X$IyK36SR3u?J?;py+`}$LYL&n&JYEwdvT;kPrlG_sZtX3LTqQ z(-Zwq^#fK>7q6~d9v(J6K6EJ}le~#7#lTHJwi_xc$dL2t^pPY4m5z&DUdFqzNOh$R zwL?v{IG8p)g`yOL*E)4@Ky*Y~ zdH0mqt>ymxm+{xuT$|5>xA}EqEpS`=!S>iI5f)h{g1BQPUXR>F?S-;4LzdX0>iRlp zISnejdw0O{M~MxxWr@#}#D?u^txxJz1nO|z%C|n@JsdQN5%jmZCId<9`N$%{}9Ul_h#3nRKE4ZAEQ@4*F#0T z@fu=}iz(av{?;EO)&C+r7qid{Ps_Hf9v;S<{BqD#97zOhQ_&)}{Toq8++qclF!MN9 z_mNxq%oPYZL)_n*9^jjPN}WeyM5m|wyFUIAj%RnEwqqRJeiHk4Nq|14M*X!(CT`qf z=sWpqpQ);>Hug)&K-Jm&^d$e0b7Tmtf0_GFP2EFhe2Kr*#gAqi{xqY)|Hw1;r)Rxn zHSc{u|HoUCn~}AY8ym9bVNq9AijZ9ws&S%V(OF*3rz^Fi-)+X{^Jt`U2RTmrcqrCv z>jvb}xQ35lO^6BBuiQ=|dF<8W`<(i*XycuF(M{Ryc%UUnQ)mJCDSIbs1W0NT^6!pf2| zPDDU*xf>{RLk9$<6h2G&+wJls%Y$e{6Xv~J^!WV_j7#W- z?*`G99kaVtZYONTelpsQr0%D4LC`pL_`x&C#Q-W}sRw?)VMoJUxYF^qvL)ol$q;)NRT z&QPWVaLd@?d`4&1-|l5CrJgrUBBGKFmzmF5PB6k!eHT{DeLH4ITOWV*Vcy>yI9u_pQf?E`22Ufp?8KlGoK~yP7 zEeY7Dl8p)nX7SWm?Mgh+KO#)(J2wsqY=x)t^Blgg_aZtzb>BvhW9_n)ddOso{mu3v zq~nNtzu)|%iKck)$|+Guu9n`YR$ZTvBue>f$q+lJKpOthW_a~!bu}4sz;o#as_uHp z&3*09ZinWE3*2KQ#%2^;{kED&S&QYDszF6iOJu<1CwQnc=H8;2SY>T(;vxn)*&w!Si_K-2M2fZ}_wnMfqe@@*Sxp3ejaRd->`yT9K(anm zW7uj8fS1B{du7kRC|*{HGBhu+R4-SCVI#@YEV6=cjU@&&3=YD^xx$W=du{9@$f|w@ z^8BjW64xr+U55N*Vuw^iG{(fHtxTa+&zUbJ5Z7Kvntppr#K;sXpPUmFawI4GJ7HNS z7dgQG*!SBcRdkwF6RqdaBV`$e~KHWaoR1LTAx=4uNU!P?B7nUX1>Mw4s5)w;L z{|gQ2Ju)XJR>pUBf5+o+CxefFVF@S-Kt(%~xGeGpgCRU{sLEc8*udU$^YLyAJs*5icu0kUZfL(?5G{^D z?JnVt1n-R85}o3WrEX)C-}6YzX)Qg8 z+G|okM@(pCi~$18%Fz-PPcad_R4EbX2pQRr`e$ctUT@||S5gBXz4&cxT7!`@qmsy! z=osucD-DZK<>y6uXFLPCyNe=?&1wEh+Mw!HI6UXSz}=?O9K&v=zp3wmu8x4Iw8su`j9*JV=O!0Jrp@tT=|7QJ&BS2rTzCDut4Gh zQo;h0BJAouMtFWC9uEoVs;&tRGBD_kb2rC#p$f#FV|boZ#h(m18C{#}no5^}Jw%>C zc~a&aoMnbGZY~+;LCkE7;&BpaETcwLT-osLcoN08UNlI<5Kdr?Pa`twBmrNhf5-JA zbfKUs!0DR8CPM?#kC-_iod^-u?L+?i8TO=rj?(n8{oHXWkQwjSA*lG}L}`|vmEjg5 z{Pd^ecL$pbv1RFPR=>l}m2r;p1z#n8B+~o+`F48j^Fi$Av38*(wXvP={uXML|u;FRu4C z_atLv>H+^z8&3c-YLsBRC{=N=(tH)ye}p0WeO`R=WbRXJq#XU~fSv#$0W%Kv+o3i9 z>L@rCP>3dt!;{4B;^Ea<2=3ZP7udIRV9b?u=qDQc3zhe{stN;8knw$Q%>46=dus4G z8obUlmiER272J%6$yWb0YkeKC$kKiWr>I$!2_jQpT-uaypg@KtK4vcO>@ULYb8r9W z=`3P%hN{hxPPXmo=q0j&IF27xr2d29eB! z4Q&bL#@(iv+@}w-`n$|O&gGy(m0Am#yi{~RFmsY$rvpJoY>8!JMRf*CwsU>%G9_b) zes_or_re8{hKhhYZ5a*hN=O*2o}E0z0ZTGcP$!J4&rUJoxEK-7K@k$##&lPR%Mn^L z!P5>#-(1iziGj2S6mfD13hZY_Kca zV}~eEL9JZ}WS3i(TyS2%toig0UNuqRh3x%{^GNY<`{=!BUE}BYjj*j_3j3e)^YsrG zfg%GVy3BZd_!`@qcia0wMY=V|);V+&D2M!IH88{=scjvW?s=XtSY{AE-I|gvd^|M1 z1%U^73{G7PMYi!MmfgW9{7a5snHi(?MC#kuq+j5aUH3dI+uFas=@``#I-)NT7+SpB z>-v3=r4Q~(+w!Z$LLI2G`bE4IY_7gT404sqZ|pd_CG{$I-{RVlv7|g#(o**0d5V)} zdV6n)9FN|l{gp?N>NBWzaCG#6JakI!`{6I+K-0Mz5f<$JTv`-zPTPsD6;DC-7 zBrv(3Ah>{eAypF20P``6h+Yn~Tw_DWb~4XC*P(K!6ml2ha?fEO^5zBUHwKp9$8H~@ zpJ*r$#Fb26tao}^H7BHLEFSo|rpr?#vdenS14?$dAo4hd;CZ0I*YIhWo1@Km7Cv#G zAnM_FV)A3k9c4!WTjlSJF+VZS<&^_gd@tTs7DYiByAC{5;$pCwu(NiIRZA>6v*29` zb_ff1`MN4Cn2FvgBWPl8$R&-&PGyE!ZOvUs1ti4C2u28-eNIH!zNba8QBkCsHwkN< zD$>&PPhS0HfurC>5HCSVlqoorhRg4^q;^BzLi$EW+pB7u-6h0b(7Z<%kJ8~W5)y}* zAyL1z(Dsrflsk%%0hqfa95mNruI2Sn5&1*qgM+fqnk{5{^uwVIvkl3) zC&&@&9!@rYJ1xsh&5~~R)10W9;PMWWEQG~wJ4GF5jQNyR{1a>cAn~tmX?2y z!wqIp^SS;m0B&uI36l+rqVY&U%(D*%%v1$lYCw%r1WQHx{Ft2j$>haGl#>G&QJKQR zPtF+Ldby3UMobP)8XXi`JMrKW^$WwoVJAipTpAvqnieVCEyAkUC7bMy0u|MpghMfw zDzizPynPgvi9Of+xfmG58RNmCpn+8ji8Ri^C>=~7{Ov{-+#s;of|R7^3==1OpaPtv zmr|xt4lX>aKbww@`<=V&XVrp;E#2x4%9YcV4FdObWVDW&vu=clZ!p3%IM{Tw6)|k2UiHy%^2H6^{*<|^b8GhrzdqNq(mrkYctgMqp9vd_ zr5b_>LLM`{GQ0gwuU^Ww-rqumBcRu#E*T7(b*09F38+K4Od`qZhhS%@hx@ze3fIC5 z2P4k#XC;9qZbgw59~CPyY&nQ8O#*pR=icMN1e^$axz$|R_&JKFF8rzID-AsbJ)I?4 zPkXvn6LY1FdV0-l@UxM}Qj?(X6lX(a;X@Eg3?Py(YZD|=pKe4TVA>uelTSs?sbS^y z;rg}~X@|pvf-+&vZYh%H;DzC);1vbbLVQEbF3>H-$U=PXPf#f`QbqKNLI{1cMcTb_ z>=uK!(rs7${a6k4SHxW55R5}JXM zOW${NCvcE6Za6f}O!lY{8jav(C=Tl2m$m*7{T)jopJ@ZAPwRsU|yqR zUQQVKXMJ>jem^p4f<=5t=q|0UtvudJhn~nXG*TSYP#D-a_|;te_fiO&3#(+gcvs!r&T?(jp2 zOjKEx!5mcij!_DX%5xfu{-f{WR%aXPAh>t=I$?L@;1y3lOc6OW`R)Bl-&UVndwY<9 z=#2g6FUF7#kAVjpaoRjeoR~j4DX!z%Jr4QcrDbJxM!$vF*bf)q$m%PT5Q`b>vP zsk)VGUhs6o#>v6jf-)2i=St*%nh*+j)sr#@<9?AiXM1~l4ZA!RmrPR_w2T`yxL`H7 z9t}<*QOL3TUQy%vF#K+qEY#N)>~KCev(kR>4UmSi-D1)sAP0G%r9~cPt|U?9)oqK^|V$-H_gSZY$~eMlxt9 z5CsWJSD);ATg*k%|gevX560+z$F5FifTfeR$!q0;Xe*X0}|Mc=R zzU-;L-)QqyMcA+)fC^Y!gFj4eW%A5&!9&=z1RP*s(rkkmZOy95SUCv@B4Qu`A4Cr8 z_3S3a?}=j|NlJkRGI}gIm(+2WAeiqYkpibM52kQYL;8H3Qcyg>lL&I8<~(>UwYlu? z6$)$)7$vo_3!3_P-ytOkM2dmoBZ`TIBZ+`dPXzy|RRF71hztC6FT*TpZYA}iz8K^3 z-?PY!h1{ten6Ag=#Cl1Nsed7U;&<3||KF9dA^^+7j{+%ha}@L6Fo=d`{HMiai89ztp%w#00c*qan9l8PV;46w*Q%HBRZEamP!<6--X!eFaQ8%>jeWZ9(IuXN zB%I~?-O@IN!~d5BaQFR0;aO1}wN{OL%Ug5p5sg*PtxZN|yuXi)1yTzwXsosH5Ds%B zBq%7zY&f>(^W${~UF14Yqwg$z`w9ls~BWGGO@5evQ?W;qFczTL&d#B6;`hWm8}Vt@$$^ka#`H3 zax=aEo!v?kyimfMT8X;whDy2P$yMi$b2(_r`*9tv8ML&=7k*5X-eC3B5yWb2#dYu~ z=u|99?!=zi20T#egyv(2yTN>Y`BiYMQsM51f*1;o*er+!lCD_3M9D*4@9g-tE(N>}& zKjZ7SztEvy8_iQGA%7EC{c-7IsiIzuSXjm51IbR3HSC(iv1ccblSx8qTT|>1EPYoXn zgiLNcc@-|4UF$AY)rv--U3@??Rjk3SIP}Ljv6-J?FD;jGMlVC3C8ye(3zcEPm4Rhl zAL&EFlR^4DVZ~W^(v&FH+0}Zvrb``CQPhd=rNLN8B(7W<67&hdbiUpkmIgjoY4m`gjym=b+xWakV7+C*zlkz1rHIad( zm086SkbJC-TAmCR7@*r*5Uq!nwpwI|kdc~bawhm2@a+ZNy5;;a*!<_xSk+g`J$TD5ebl}Kjzv{fl?#)>3a(=G$;t}u z1g8DMC2s88=jn~#H-37jC%Ls~z*&0V=mL$fRTSNm?db6hx9{g2LC>c^N_MG)q3+Z& z{H41jw(~{6%nc#NA7x?RCuy`FZZ4al9@1!kDiN2E)jkloKp4Zd0feZcQ z`b?tKhfy?k!79GeP-{w!Kr)O0S^dr;Xn&*!KXqWFl%gmP2?39g^T)vMxS2T@(V@DS z$C&Ih;kcFjOrtreGMNE1I^rq1Jc}1uMFK;aj-Lqf(lc*_C8~!cI8@`(w;iZ@b&gey zlFLA|zAWAddLcE%PQEP00Ycn#1!^m-7CA`S)X5jjF-Q|U8h;)zuvlnVKtkc_65@0i zkq|;k62()IJGwyyT%nW{(F~{0QaxLI*q#C^;w~hz-{xo(k%q+RQC$TF0YN0bh>BUx zQIZa#Z29%@J^>AcBplY>qioa!0o6TY`7npVc)ePpZCmq_!}?ZK9|!EE`~WmsA}5Dz z>c-M8LFxIQskI|svhPdl_r(I%Xh_!w08$rtSo(A8cw2Er&_wNo$Y&rCxA$Pz?i;QF zazByr8I!y_{S9%rF)TPiNlrO6uA~;t6%_eIAPyhF*0nu#*XvH|j+9E#>Q)|5X-o6)84dEXxV`GvC!Uft8v zQ&K{SA0x&OSf`4BnL$K-+-P@!Smk9=p{P>D~L|_LV9RsSS(5*BfYJ$$TFc zwLud>=RvP6U)SR941rTOcqB?IS!e$W8Oopz@ef_^`@Y0$t(B<_ef3!h@ljK?PD{)| zIo2Jk2<@Y^KAh+R8=GOLfX7LBW+f^qR#p144p)%>CLvV4(l$AX!7FL?TU}n~=T&|` zZS8u*k$F1p3YFw87mD1U#UVLX$W%!%QCG#_Ur0ml2nr&_V-Xri@;7*jZT^Xspf?A8 z`0i2{(|%D>LU|lnwVnOWZ{rC+!MM{WCZd%nC!VTVyXp$6m^5lyVRi)Q*_Y}{ZW60O zLr&|#R8d^yM+-|gHAoZ(1oLHe(+WNACkGGrgX!#9NPL;+3VJWe)?B~j=M@-q@)&7i)iG=;B3tPUkoKkm(}LrFIv2{rF|5y~Smh6bJrw*zCaXs}Ad{$e zfJL1H&;2s^6x);$u^4M%V}j#CP(W-QNCd~X?+9=ZjA76n1|4HiK#kz~B|xaokPCGm z2RDsS)!2)1BZYc|40BM(#XX7J=F7o{g{p*(e1uvkW-LWlguFLm5_7|2jssvSoeSd7 z@Wd&D4cKu{5?&|&npo=_ zN~f~vu%F(cz^;*ggLKx#afGMN*?$(NxfBNHJpx@>^J9b*E!u+4H4GSa} zI3U%6*mu3#bqX|CSbYrD+bO!y`fFoeIEy<#jAtx-02_NmgbH!HhA~G>-S->Q!Y$zs z$=!Zc2CITQ*wBLtN172`yzMPU;rnRb4X0^^DY%k50jEgweMmc7jUSJkCa;dQ9X%5G ze=-|v(5@MFzCF@UM`cde2W#*+?|_x(IIOpdG62Yr%t&5w%VFF91?#g^yS{#vcabq4 zA0HCr4ZLJyFhdr1Gu%F+xWvQ(7cI*o13MUNL@XRmzu&oJTJDaf#aAIRVYXH>v$9Y~ zgrc)&l!8-x{ZV6XLt*P6m}(%Sk=r~%}(~U7!J}7M6Q3b z^7vc2ma+$wmsr`=i>{HAZ$%0LE!kaoIOow0gmj3?W8mFY(KTRNGB6<0l#mVZ#-^zJ z+Lvl{sZ-{V3=c7ta@Y_f0{oEc=F^O28_e?=P>?zS-%MW$K+_5kP(@f!?LWQ3(?vC|vt~lF2E+6RE6<&dEsMznI3Lj$`MT zbzL4`&aJE`gH6bSTq-Uy7r3lNKMOE1*7=b_SotV;tHW#5nF&ZURk|-#xwO#)gd&n3;62e)8 zJb_@HU?XafLXS~QL@%Ve7g`88?bj6=%m7})w^!CuIK(vN- zRu>(REBE~r#gVFDJSh9@%DX2+sD<5@JiU$S>hF=Fr2pshph0A1XBe z(=BZ?9eh%y#KW+#K7vy5{r{@NdD0z)evTi5v5IMiaV5F$0%Ds7`vx=`f5Ho|S59}_ zr*JUau7XObm2qsJ`+{j1Nn%ZxLOMtdebm%r1fMbQZ=EZR{0?YuCSG4g+lVj%YBMZX%O#|Q#t&L{ej%A49u$Jkb(9)SIq zj&L(5k2ADL8#BG1cH0;q@;zK|{rVdF?JYdn=2o^u$Rq#b8VTrcoZOLQV|T^WA&3SuAQe_RgU&1tvNgcRv>M zJGr6f81cAfZO-LHx{N2tF`^kONC8(+i2khSzBgj(GeS2HUrR}T(00ZT9^Fin$nK3b z1~~IF5vCYKMR)-#k)e=dc2bC)194!sSWok6CKIj{Kb_b;A>)35(=7m#Q9`-~{qHh` z*9RnItuXs}$TiHci`Zr()C|JTs?3sW{xWz^30|qAK5TnLJO@umB))fkfov3z$8ShR zKXrC)$@_W+c?F|;R?9|)^i(vaGS0fkE~R^`fY~jwk#?ath=^@i9JvVER>U;uw6G>0 zMIB6}gOGXl{BHe89*@(MU@-nhduUB>gn)OR^}$@LdGeS0qPRp~$ka%C;wXH2fr28H}C* z+WNr%2z_!AeXaLq(lK^vgcCMn^t(a5kEA%YV`Ym5Jcz5!^scRghsY;0Mc&vGWv-bp zyI=!o+@z$YkWqf_{sp(B1s#U|w{6t~X*D+mx0pAm^juJ-@!PGxoK8k?xKrcE>Y-r| zGP^e)_c(c^N44RYz&Ri&NaN%BAaI|+?2V@j#KTE9yiZEfi2d(X8FY%7#&qeh!3}JP zO(4iHv&#gp`yghHvw2(q-N+FcG8Dn2eEA+?1*kfBSB-;Ea1$oS!`d?fzK@shJ0U3X zc-_e8sRX9fnGnKQBOwzj5PV;;BlFkKKixzpArerBW?v-|OxLHz9KZl6{&tyd1rA{K zo`NMYVFN7kv+;->(B)l+Mt(b|D_O!H*PrKEJmB&+4ekuJDHpU@kjC)Qc1g@cp9rBG z-uY+S*vkx%OQhMCwe-w?j6_{>OiSc&z*pWt^gQvDdsCLuvew$;9zRI$UoswdMbunn z;AcG|lJ<~b)M1}cLNy>G*EYR>v17#6@VYn)1jh-p2bl1&dT*z0KDInAaJ8-nRj%H+ zkm*fohnGfrcsD;jO`_CIBTrmMM%$N2w3CGk4WKd1hT9Qutq_v!rw^<)gKoZLJS{l4 z`A#h5mIY73PN%Eb)7EnIe;9fDu!Pw*CatoXK!UI;BqGFQ66iv>8!>fu64>+3?(#Rn{c5?Mo`@e0fq9p{%q(-oO*7}laK`*hQ! z_uxdOQHdlIDZMRKg6_9|GMemB8#9YidIaL4dHZ)km0$a6Vw;aEB8%2U_&<^2l7kmu z1A((hICL|D90)OP2Nu23{^zLwQ-IOIfLobv2;9`N^GcD#5OAN-;i{~ouC>DF@JCMP zDeG>L9B3YV=h031M~L0&tEFoJKyqaX#^TbLsxH@Bk-`G9szbE0O?F=CEVmcN2IknroHo4Q)=)EJV5$PYNc6OZ;R7;Rb=1nufP9HCrU@qWR-5 zy%C5kUNAp_bX3ftBkd)T(VSl%K4{}+Y(DubyZE=KncQr%LndBOcwAQZgw(Gu7uwzK z^wj%e%FQEcS;3p7#jn?ae+*$@3Ze5AA8E0lR{^}8We;s>V~rC+-GXE^VU8u5tj!8q z9_eKDVvyIlk*|0y&UfU4WE#q#<)l*;&4}&s$Joto9zO>jPO_-d?XtDRbDsFe^mN!Y ziPp&Y5~tUET2!QPd7hJEA$9@+`M{)xCP7ADe6h*J0<9c#n31Q&UHYTCye`)HyyQe$ zvaG$v*idIyScQ;$j_iYoM&NZS=WMu)U;LO)#ZDJG3Ol{s1OO!D5VAQ@1d|X9 zOwq;(%+k$Og`w02bPYgS(4EEHPIPtcze5LKpS3Rhw|Wf`fl#k*8bxXNAm`%&yDbXm z*APTsmss*dZ9`=45TBbkkHc8L6AS1iNOGOix|lVd?od?;CB5kjMb|qBl62NS|6R%u zB?Rjy<=KQjS+Q=+Y8yLTeh-FnNX1ojy~5%(X-wf2l7qtc8zoU@wFPD2!?`!A*nJWy zOzPZi?!3RTSC@`;z0Z0pM*m|Z)8GKKIuHum8VLPOPq*0kh-SJ8(2w=`9Pkro$4)(1 zR5J>^du37IIm=8Q79P@E{;OFTui!@t3&*a4S!v75&^MZsdZ3M>y$y?FWw|YB_ax;# z>)`f20H|pYeZMfe-4yh5(+|k8?7#(7#fexjCo#h^qr!O=g|{bzMnS(S`I@`|J4D3dc{TFYp%q8cWbl1F`Ih*YeoJxM0iMApE-w63<**;4G(3!}*CDT2 zxcH#moe#Vh$~xHyQ%LJMy)hJNhwdV`R!cj))5U2GhL~&c?EWm`$;GpNuei&NAW6r6 z&klV2Qf?cTTSZhJ5&vTg!wMDoB1Q5zD921PhS8+unjB(QV0&DaiDd|27?hI*7gCJF z5byD~smIw?8s07dCNfI<0=BT(4*$$gjeu1+(di5{ zp91^oHmYGY{;@s3NnNod@D36Yicg8cL%g<4MYjknEg-Wc`|=!G8}E9 z_VLpSl!4%gxtw={5=Cc->_e`C_&Imt94H9;5$vt3s7DA0`-z|2#B+AMNULty?(gqK z(!|}mo)a((64LMsOCa*h~KYzpR z%=lO=`E-7JQD)Q2Of}Ie^h=;$_PaJ(wrp}5bFdGu|2xVl9CN>}AC{Cf1 zu|xa?*an9L5Cp8cBUJHMyfQ$REk{o@b8hmNSSf`~D_BtvsTZ-*o6x6@X6TO3h^EL3 zjkEZBssF{Y_B2$G)grXPG zWPIZjpmE%+#fbp3pMf@Jf}WTm#Je%6hxim}ut$O_f>-sb{N1pd9g~RJ4b_I(X^9)Z zl1)UQOwBRkup_

>!^^IvLc)SMJt$M;IMp%#zB;|8yy}23`d%H$avtX|JxdkD?rG zJHrfAV5HiH?mMusIBQjs1O}8B7+75uLMBB2F^Mjx=77ifH!q?4E3DV6OY53<+0vO^ zCNNMd2BW?H=00vXX6MQPmh=&IEwGzO_qY`*jnrNIsC0eZ(yZ?*60X}CtC^d$S|pQ< zC7CxX0=R~p*6OX9Iazu}yLlA&)E{~BS76!XJS1w^A;A;++?jjk5K6Xo7}J!)B?21? z;x88@^&nZB=Zwzgw{bs*th=k_-hVB1Kig+cAJ<|{~;vG)Xw{N)fT>Q-9F_TQHyrEVTeq)pMmaau2P+iR4tH};p@O6#u!<=-im zMmkkv*x)O}IBQDVI9zssFo=FIKth^P0N!nm@)uVOH7r}e`}U+P?;~gXIcMw+wIyi3 zyRRiMPS+Z`b|CD1OKIgXq5C3C9D$(2BEf2O!t?yG(lqaHu!zNW<`T$JyH5mY%n4vN zJJlA+c&>)^dso`#j{W!HMMY%9;yy`Yu>QMUaOVa}94ptfE!V6q591#LH|oQM@AxT! zsm8=M+?73d?uwV~ehnnS*|TkM65wusSvho(`SWA6Bo$pAiL= z@pnRd>T?O$jRt?O_tw|HNUbfcnN&-PD|gnQQi~P)`d!Sfo-=MT>m>MID0*?F$CeNuGurm5$u-RFe3#}ue^QPN8;E0C78!l30T zP=7Vy1Sh7EchW2e^~le5anV5WkzWeO8C-Myks6U<9Opok=B{6fs=M7qE}H@EVCE^L zWuNMtL(C!aUoVUFWlAm&oNgk_iagHl1)Zn#h`O#4F|Th>A62ASb2q@8 zBe!;kB5&JrOd{o=zpda6mM)67h7PBKxfe;!v5y8a?)#`=Zpm3yY=(+sQV>qd-uI7^ zc^SZ4nYhN3Zl$iiM*-Xv;+B{6FGee#hEiAGJuYapnyBO8Kf0MYPYHEIk$bl|NEfG24zLbzJ3qJ5aO=PFVmaeo+nh;k1 zrkO4Mt(r?)N7cUc_T~GBIbfFdYm+inXgBI#zoX>l#x>-n4{BK-b5iMuKiZEhI?`sK zAxw@wKObjqyR+l9|M;xyW7;g&8|eebrCpW(%#Bw35eLzl^uQQZOT+HU>?&fvzespk z%OxfubGxL{7UzltnD?*l#n(E0+urxti%)el>K*S{HuuBw8rWl<1Oc=@K{!uD8q8}h zdvzz%Ykx}e4FSwwQJ8^$47n9GRrnT{N>kUCtmj2c59~kx%9?rbv2)r{erK!-vJ12h z@0-?(#~*O;YnWFW+EZ7_4`gvwK&XyATw$3O?v>|(nEs9_nbt^=o3Vwoc)d&-A zm+8UuuD=k;>XalG!fZ%Dx^*VbhN`|!xA4JGL=d$#6on6%Z!g&NP#$mEhqB7zzE}6O zaST=%+~>*%20>^z+7N2^QQ%VW zZR}+v?}!Af3ESrr>cIO2UpY^Yc>}!7L&RkE7<;Aug%PbYSGC%R0uE;7R0fYbBa4E? zfHOKCOZD#AD-Liog_BZHGS)nE7fBvtHC~3@`*@;FTm0TBQ!)Jj=_Zw-&3#w8a*{V3 z<7)k@bx|$LS+Iv9k%;+HRYVt}>e2KJd7M%hJ5N5TP`u+VQ&*t}~;rj-eVj z{bEQATW@6T9jy-6HkU!TsNAzpq5h|%ZlylgL{4iynueLsbPF@FSn$zp@qzHItVVHf zCtT+tXnPmJWFq{Ruxztbz!UBM8=T#-)J=-Lfh;mYKT|IDD--?X1!OdiJYtZ$^nbch z97yU?cnAoHFGij?ZY~HUIv9vYL?q~QPm?Toy`rU}$-w^ADS!2HxS%F6usgcgKuoHX zt3Fe$KJxsqO6tVrZ(a8>S@TDtTny(1MM$`fecwVc)r>ahP*>ZUcDZfEE-DFO$K4xo zu~8Spa-X>7F-y<2hXUNa53-?1^$}U%HNcF>uW0uTAK?Br{sgoqP0Q#cvnXhcm(msqBFahM+MJd1YvodzT8s! zF4jfp!D^XQDZ(H(!Z(^i)>Lp8WWy}D&Y+*0HxU5)8&%3{ooS`oANAZxJ*gvsuSBhk zQwx*!QGuXB6W2pui7S03_XD{UjsCC=CCmKX!*;Fkx@C;XAQ2i3_k6r!{TLBQm>ZAl z_(Tg$z(*e>1Dfd)EIc;lM!=)aFtLZBY_Ot)T_BXnenIBCaN^dGeC%NGs_53jiLGkS{T3K#6d4 z7uA)=bj-W$^cVD(YdJT2z~0RU(ses2Z3^^{?aQNa7SSi*-y2(IUREgpSwD9)#(Q_l zJIqq=Rg(Q;wQ8!LwlXJL-2L9vv5Dcask{uf)%#)bznJ>!sHnp3YZ!*Ep`=Sty1P51 z1wp#IySqz3RJsum>Fx&UlJ23q8w9?~`>T(CSc^3*xX(S$$-Vc%6F8wb!B9VQq=vkA zaUzK+)t|o&oNimnBB*k+8(q{fT}omJzm741n+@9>YwyW8&=KfNc0eAk(!%;>%V7PS zk8+D0*?3vz;CDFim8#R!-28o}Qha>;%btu#aP@Bd%U!$>j@UC^bp`jN94xHs;?Lik71n zUndg4usNQFeTFS?8mn(ijB(l=>P{j2bWj2+V`W||!;7my)nA546&IZse`n((>+m$Y znns*!XG@)hUCHEbZ<$e@_Mab?F%GH7@P8hT@AxRmN>*AJ!N9o02`>HwXsva1qS!Hl zJbXBcvPymBpI|6(wHy<}KFfrKG@z(3)ytYjAsM+Hj-AZjCCz*KI=foCx|+XDN}I7` zFtJ3YLKl`6y~C#)kU$At${JJG5zPEx1M)AU*WZzps+(?>b>mc-!7X!$oRtQt!{&9X z6Dr)Xf0W8dua9b2qf+Cbs!5R?R}wH!tl0>a*IN8(V)H$H9XAWpQCEY< z8cepz0R4qE18-KqOE|%CVcU`Hqw;G;&c2cXp*a${yKgQBO_{Ek&$mBCPK*ZXL=D8!dV?N< z`LTcP?c~rFT7&Wtb|ZbA-|MI6ej+UBIptkD9+mN1J#VU+5OY;dE}4e?x|PM#;N-Hf ze020*?4AFRh@u0(#S(Y;PsQ~QkMssEn6omTERWAjf2=Re*(!IZ|0OAw%7>?BC_-}Q zYy=VX@Ut2%QW92W61jj ze_#&r@LK`^^q@#7^I^J37!X#O^ni3$w6wHTRh_@8kzOQsj38zG>oVmd1Tg^ax@lLH z&4O4%I=JGcyGTjXYFWT}D}>nLJ-ZYCybbqla1|=f{NrMlx9g#c-gP{9#ycCAdCRww zF3Cr`5YhD=4_beoCMnvdZ#UtNv$=BE^uf;&eKhzf0TxTs*(qN1=Q1z2#p+7?>sRcQS*cNp;qL|((ShDPE&uHCf z)IoV{xml@#AXoY{N0lvOmZDW*;ZifP=O37RM$|kMRhjFsABMD$I9jaiQfasudKN@a zbxL+ms*G+og>ZrGOrC%WN|e3%Zth4Puimj~|D0nWz(d z9Rp-Okkb~REp35Y>K9=yBWWcQacymX&Y3GcHwCX|E#0o&*sTwjWi-?Xpc1}G>*oFs zcLoLsm?}+3^Kd*@!xnZ-I`-grx@m6Y7EEWcR@7w8U>oC^@%1Tk&D`LK=I0fk?D@K> zETSIz&bO<~JOo5yoCpsi3>JN)KC+G0A;djO9ruv`IJQmXJk)bt#i>vWZeRbiF2pKL zw9`}JY|QK?PR(aG`>~AeU_7!tOZpMSRaUzlm`?>>xjXu@X55cMa1$gp`eaRWv@ydc z1Cfb(y(3odxz5Z_UxbJ{cFG6!^AY|AZdQv5*FP$gtv)yqQwdN9HC5}8S!bP}9j2<|LeoXHcG->^y28wD9~Sq6iD1(+VN z!|^crzEvR5_^gVDRRvYV;w%fKSs-G`(dFr-6_x$&0c)`h>G{cgdPb~)T#Vg^z*c%s zhjpc*h*L{_N(G>gG*8`6gjkO*Mnj;NKVnbm!x6`8Nxn1Y9OHV`p}P#BuwWbm{??a_ikj zOd>Vq!x&G;y`qXRkNxr;KpKQQS`bvhol`p~(ECH=jP>~wclr$H-3a5ghXbqExhDw; zO&Tdur)-IVGNU~`UO7aZ|BJz9nQQ(WJb9^!c)zW^VrExRdAiM7Tzt=|T> z){b5_X}_WrbRNDJyR})@yl}td(S$u?k`@> z=3Al(23nSNrF)gDd6%jcdxPm~8HdLqntEVrSp7vo=|AjUGH2g&yWEf~G$|Z9H3R2z zd6W-?VzJ*~n>qcu3MR`HFbYE4?rHfev8=luPD1^DW45uQqwW!_oEgQV4Ba=BX01tz z)00G5G*6FvXCIjYVG$u41c6oXoPl?Atb4C2L`ea%gJrG!Lge+!HDUBEf0bM4yNz69 zF)jTaFho!^sXP3|r6AVz^F7TKlCqI63NC}A;9mJCRWDYs%>I>l2?`FpSS+tGI5PEJ z94wpiy+``>f6=I4JwI`HV5R~AzS83Ur_?HexE^@%jgt`HTJN69mfa|?j_iu^T~h^^ z&l$Mtdrk2rR#0OOGnt;rG1TBmA89hejdTHx-xk_zNu|M~;i6_bA1}!jRd`!GPM?=| zg>M0U^$Y1X#~LTzBOW)1nu!8PAn+5{ei&Ct|w zx2P2;=+=o%xp|=z7RbgIvhzU_j#osYZ6@1{09qHfjU?`o~i|t9Hb1r>Vb5bcvR|vAc zTFHD>&%r5Gc)Rn5iDm=O4ASg+=1e?~lkKDb`nhJTE zu+YJX9d{nDj;8YjJYQzt-rkn{UYwbEMIZXy#FOeefa%kHRt1@vHe%_ z><1C(ud?^^A1%|3x@ys;5*3I7Q${GmY%Y(?kdcZt5(*wQ{};*j`%?X&>PN6x$m;SzX#WCqic?4~LN04! z20qlXfzqp6iI_8m0R}{fkQga0=Tu-y_)`c5HCr+dtM4p%Hin%St9|UN6tgC;%E$e& z=ZVANx4|KY2oEfM79S9TC7N)liU_qOFVsZ;4}}UNk%I(=X1y%4`Isk;aSbsT-OJeU ztWC@HVqY`%epdfxNvib@pjR}?9jz9V`}Dweg;dcuEMTI8ot{%J?>J(ibQNn_@U`^- zxD*~8;)!IbLgstcMU(A|t@MXNqKiyc71~jffg&`+LDw=2J?OMRgal_gl(T0YGfFl!fIMO;R#Bxr7EFT+sx?o3j=pxi7 z+R8%eu7ggl-NhPeFoxci&c$KFGx?Y4uh!DtKs}4by-td|#WsEh;{*BfLqUxMOZL%= z=vnM2Lxt-{^h&$6SiOMbZUf(&AO|M?#(x-67sTe;334pfdlz;VeDp{`89tY-p`oh> zJ9&S7TEo!kK_p-@R^!{LmgnSv+!unrPN3;KRFnUx=reCB>haK2sThn3~}sUtX5 zMrGEpeI2V$s9)5I!7Ip20;{x?m{xuV((2#h`;hc+;G0WWlAeWD2;nPz_%#Kp{E9SR z^&$T@I5TomOIK@j2ofN}ei<5dk$AluG@rxFm5^Mp>9csb{v?YAq{mx?yfDIOm!RMm8RRJ&06n(v zHz6RGWX8mZ=8kFHeKCsEI^Mqbr5wLzg%>W1+BOXa_V()zgv;MYv`*c=tjygU)&D6% z5kmc=4ShGoF4_Cr0vHtLA~i0f+rhR;c{dE|B;q!hkww4}x8`^fS~72)YoZf5hfBxj zDl<3mx}Ark#SPy!;(gyo^sechnh%L|r;GTRqm|nat4d)kc$?rjBK@_KW_%qsG_7@r z_e+XC25&!3m)*#@N#NJ_)2JwRuvK9$pFm4asLf5!>rH4$@>8zrxizYxv^_;%<^``% zHs&SW*4YjJKYlAN?5=i&kgzNb)pn967`o6Lg9?^!SU(%r(CM$p>Wp0a8%4mq{ACzp zS#o3?O}^pLqZY0DdaHEBXLZ4iydNn`pwq};4m)Vs?Ba)o)F?S0phDy0({VrQAg*}Z zi=*s&>a+Hk9B_{h-r;jU(&Qz1?#6EqB;dH8tG99y2f??4(~3_0Z(ESeIN90Rd3iO1 zEjzto^d&mlm(p4!1^Aw2Ni?j=M1WyhF;yujp3ydMrZ zQ6x%2Ko!WP`&`E)a(V8?N?m#YHc10LC~GI)W7ge(fKU0j#O9k}aHc|UobJP|U7ZYW z6_f`Olq_}z=|;+`BP7$X5k_jnZn%FWb|d0WW`)z8&svjt@z!cn=9mAeGd-M=zQQ1e z9OdO5Otin>RjpvjC(F>Ve<@F=Cu6fpb~~lRUFjGBJGI^ZRRHRulE#64i7yi1=Iaa~ z|FRE`Y(2ViBvT>{ga#-uNm=>t9AMvLj#h&i2Gx(Oe1k&9Eila3*nC)ZB9zQ#NT*X2M5gl~J(b3U6A^zj5 z-iJ&`xj8?ARNiM-A0{ChXD%*k-^KY{P0FlGmRD6dttgEueHVG{kT0?Z_Xo*n>fpoI zMSA@`I|@^Kswk@FaGkB^=V*9Wj`%A^`0u%|TcwG5MomyZz4)(M$r^5Z{8r+!p+P*h ztAW`HU6NUO=Sm)z{2ijuZkOQ)L$-f{WlV?4d^UP9%1J@c z9F7Gdo&w`jj1X$);T`Qv8Z+1(<;L0O@5g+ zZez^xp`GmVr;rm}lDCHxAJZb@`^-fidN(b`vidNzH1zLIP_qGZWUdJLHxRBIQB7o@ z$DWjtwtLeYpu)r(KhwmtGu!gKRNH@5dY8kPlU_OAXLkZ|gmF(0@G7CF&z-vzG7I=W zcUg|~6MME3oo4^uqGg!cM34CU?I4NH@?uX!$pv#IEFoW?Oxj{)r1#7i(+6=I0rK)X z)%|LhVws}4By;o?*Z2J&T!^u_NA7B@d%Ec9alrIGslLx4FeHEXIjIGJ{8QLv2Mu)Q zd+{whw4#1+aPa*6Tvx{hs2mqFCe8&fM>r5M$y)uN_+h|ZYN{~ZKXl#PxR3;USBzkj zqLg}s$7l$<6WoBhA+YWbu-msW?un+Vz$!w#Dj@Jsn(O3DI2tK>^jgyy|LY^V2VLK5X|8 z_#p8ivuR6JU_L9~y;5q1_NwG|->jwWPIr?1KE!`w%iXAbb3Q^=a{kbbyf$SnfK>;w zGf_U(`_yOf9M*pVjVZSo54gWydpz}54yR%_?fla7m@$6*rs)u|K=j1BvJ|87>CdJ$ z26g<380MLuKfX6g8lqU-4urnsx$4hrm-9dIE=LklS#D4jQO+mbH0W{mFIhd;#plDL zg-gGtrEM-z{!9%nlHCeG&*S2jWbA#%*wBltWIJ>uK;wE!U7$GjgSfi1?N*#(?xH;v zPtZ|$_wqhx(+iNdc35pCL__FOR!~sTZS$@xDvB8NIB`Q~azpc+ZL%s3u_c!<^a9fZ zlJ7>~_+w4%<#k+pk`_Vdtr6%WAyZqSIad$$5)pqIRm%kpE0ous-|h0MGiLLgBZ;d( z3*n!w_m0$EZu*DOl-Z+;WS!z2QQo`;<`;v{fAlYcgJFDGg`#OMe0fXjkC!eGN7a3N zy&kJeklpMYe4%aaJinR^N3oeJ81g@_9zQbl+;1m_Y@Rk4#n{=6Kw%#PQb88ZK`#*1vCiL7OA)NcMWEpDmK z|0epHRN&IXwjHQj|LnSx;TNT<_cPi_6ZPtr`xFJ_L-232)D?A27?3~vlmu$jZMYRH zKuhOCz4lwL1{{(aAFTDDhBR6<|8;*Ty6rv#UmhN}%KjHP2{@P{{{TEk#CjF_!M9g} z5#R*LKO_OPKzacGQD$XQ_#-?O-bvgf06(Q#`r@jyq2I!xdqE}csjO4`VgRGWQqNLH zw5)xU+4DAd@Tm7IYJijd-pcvibt$0-3HqDaWeCUj{aoX$md_s-Wei1+lLQ^%2UO$R1_JQ0f`P-4?KK>$@PDu%NpzJim@c^|RSolr&XlDSUn(hdnenta^L-;!g+lr|B2NAaD7r zXZ`-9fid_0C0GJ!94ANR=U z(%=nX$1cZ3UaqE;If7{dsI`>BJAYHCEHo6NQ+UGA;xci*S6s&(?VPSc?ZK}xcUPo& zPhnE&-_Acl>F6vxmtb#Z!1uI$Ti%{JG`5I(Vr$Rxfw;BsOZ$^+$qV|Ih5=PIZ8e%2 zIy-mQB-5+i_Om^PNl_bNy%lUBT$L9CyRQXqZaU1)KOKzkMmibdZ3EgT(xk+y(c!=> zS-oLNX7_-94oZ|VJkO`OWizNAHAgr3ABh~GCRxUEOsu)Nqh-UQtsGfJ!xH^$yW9=m zgQTN*lCR?UXcSgvv*imb!%&v8>g5h*V9%WYDI_%5;Gwqb2Cx)n$%2OT-y1;ugG3Gc z<484C?tpD2&`TMShsluolUY$9ma8(o@UDSS&oMr(Xa4A}5*?s%W^3Uu~B)|Kogp&+6H!L<@89D)b$GB=8+l{-YlJq@L?7ynaDSR?m zm~?h|!j<$f2PT;Br1ekt;q!Y?Six^76I%T3&xhyoj__LGc+t?&0Y=2r)K@AQT&ep7 zAf4Y0c;0=jmph)09mqxe1@Q}_!vfu6DOKcNb>)7@z^zwEDTM|XO?X$twt0D7)#h5D z6o#V+;_MX#iWB9xNa_z3?F~NT8}Glp3-LW*P8ZBkb^Vq>?B7*;p}p{lbUDv^vU#pC z1|;)PWbZ$5oB4K(DyChqMNM)W7$5>RPl?3n-*4!0h)#&)uKK_2KU;3_Q7WB{??BZk zdp=61?2J8@4N+7h?6Zj6`rGxA-@JSk1=$4?ivmj&Nbbf&?XaYoSb`qLJ?#}(uOcp^ z+WSeRP^06gq|9W$VD-o^8otLAksC5nZ#rS1P_DJ=R| z<%y=*#7}o0Rn(h*@CjUHM)7`WS2LWHBb*ogA1#33F$s=MG(&h0yEq+l ze7w~@px0AhhUc3W%>Xm;2M3O0Fdcf9bWGBH7Wq%_-vX0PV-A{O7URWgU5Ap@6#Nef z^In{4XDT{8?W}DJz|7?^gZeNpmU$BX+OXJQ0q626H29i=;spg-Y3GMf<56LB4<@O{ zlgR(Fp_rsy=b`KB3Cn9`OeSRVIKok1U(1^MTB{9V_Ql$k{G8g^B{0kg8dRnKl#j);M$Qz-}Ddctmd(07{(&6fX7 z%mC3JCva^nm*Z&AREXdktR#)kw(^l0-&8`8;5+(~HDsdg6?x^pSF4}p%mo{xueq8R zHa(cXEp0&lWjU?Ybs_a%qaIX^HybNBmS~um291ue@R+3h)62^lqvr2EXt$oQzNGP3 zD7DrXO<8%iNsv(e%9lG}BFo>ueFSLU#6(1uReXrQu6#0mMnA1$I<{q>#Mz+ zNrb{N*$y$fyiT#v95QtEMgGO!p<1Lz$LnDl*Pj;;PRkWl^nic z?2Z}hjpPteWK6gH%Qk{j>|4Y1j8}snt$XAQG``jC$A(tlNL5u;P3~R-n8~&D%*?R; zmmoy4z>50Axr*0X|F z5Y1-J7`q+Qc%TBj##lAS>!ABf=Y~fta8s`B(FMr=znhW{4zgV&g6|3`Fgd(bZLzM* zYtp41Dc)u@(jp|tIPgHu`Y}@-Q~*DgK~Qa?7Ho!{nEb;8j(Au}5B5Jg123?NW?n!i zvE#9&<1vK%Uru4+PJ{np0|uF(=f69r3Bmb6q`-e)6{xW*+8+))+Am3Sc!Y%P(i(sK zFb-yYoq$v1W%-;mMUZ(%I(Zgw%Qhut_qpe)+SuhDYkdwzWHxx$d{JO%b{hf5g5FpD zH=?&$A>1i3O|+pizj&BtnA@4<>o)2qZ;Vmdt$nvL`iW=?p2JW?8Ehi2!wUn4vVhWw zJLSZ^8rvOGw@ifS%y7!e+1FcC<8!BNGIsFlMgH4zI^z_zYjEPsT(;|f&#hCBk<+)B zSM0n)q(bm*>`|!P8SDF8GsEYG>~1}qMud-^dC9gW#@AUnS|>(vduSsWnaUmp&QdSiy&0H6b;JE zGD+epjK@hMaPX88hgn=F4uU$*NwBE--4D$ZHQWC9a5-fd#-o38jSel*XtCEImw+mP z>6fCtEgVkoNr^BZ*vH4metv#PJ!CK& zL-h)x&&ZvRc$x|cx5Cr3mr71#47-^2O3@hf^4I-1{#k?I+iGd*NDan^v+YzD_3Cx) z?8qW-83TvdD_n29i`Lq#d$?DyJduNP-*_rY<0^G}*Q>35S|Jex;eNm9+>p9q= z1fDL8Ee$B?-Zc4O<%K$5ygzoiDEb-@<~Ch;!M$*pV2gnd@3^mNnbByLG)=(HbK~k+ zcy$zK5O2i#=YW6x0jK zsn!1$Hb5%I0*Vvj+TYvR@RBX|7^KXIO!hQ#!4V|+nggX9`To(EYFMD^gUoQJ-SZCF z5H@iVik;HQL(@=A-j0YR5}wQXM&IjJvAyHtF!`$Ub0-QDt=X2-_NB$ehq?;CFMXC; zZ$?s6Q-O{T&6}7YUazC7{#PEJrAMHVRrq?4H6oKFTGpJtkDaaU(bnL%oP0>M)q;IB z+F{6lAiCq-_6vp7okBWBHF=umdk1&wD+8reO%fFhx0K)4_(*;Hn0uZ1Vm1l9xZR!% z94LC4guP5GceZ99;&&EgS%}AD4)K|P~viqjBfs z5~lurS6B?bbc9UN`$F3oMbzXDLC>Iy;u0%c`=f=0%T#5*9ROooUS6)>PB7dBp|t@U zPNz%5kd-kCbX|_^B%MzDD6BNuvLK_Kq}<`N@%}q#Dp%`Os`VpG^yF}6Be`Pe;IPd* z|Bzp>2G9ySZraU)qk&z=uzk{Lus`10^L%-_75m5Zg_do(yv zMvd+a>-(PwQS0yxPD(0iZt-1C5B6}q{E7z8C@c~@%Ky!YBWJy2x81_TZZ?mF71>Fg zwtWR_iUwz53rIy*r-Os8Q?PuSy{v}EpOPERlpK^uee24e zN3gRHFHG4lTPAAz;X>12VF9kWB~wn}^vINNVX+TwaUj@Y=w`!_o@{Mt%Z+lMl+ZYe zJcPVx3c0$U%{34qM=bvh^<;7&u__T7Lg1Ti1S3nc1o!!dP+573Gk|#W(2yf#Acvz0 zo*u>8_Ok-PlVMhM+w$wRGsm3=GU+JeSd1si_)E$)<+%ukRWe?l-tIu|Slf zqC#yK*}h>p-Ly>Dtz!VS)u{S%*=0Ocp)e^a30Q*3gG4_yH`ib`{DysPa>jN~)=+U; zPyX{w<-72VRX8s3#XWpp+W|=lse>I+4yh1XvG0@$+3ud^n_erhp4#5y(;l-e)NHFN zn8t9bANjGMT7?A-FLICenfC-)9PB99l5LeAy~QaGAZKgExvTwHZX1cBp)iyXGo*iK z2L5+z-_0(FKb7cBFM10W-lIdFPZ+u%0__^;@nh zl9MWr{w`5#lz$4wGEzxSQ6o)Bb}@3#;!w9f{>=>g^veVFPTIK zekA^b`v{bxNDz<##n}~y^2>uq$8*o1i0PK_Ud0(IzrFeg-1p;?97}7u^fVo`I<&u5>*?8wjvnW8wdM8 zUbYx~oh8K-i;Iir^gQ&8jg9s6NQ0{br-9CG@$cWDZ5;T&Uy(vPr^PAS4HVZap&#cd zdk`~wDDNx;=Ki&>%&fDeQS?L=U+1Hi)q}=KN_@uo=Sv%iAmS*5=X+9L{^Ec>h6lco zNIa#+vm1RN5f6k!z6C>5kAoh`?s|}oJ;Y~d3B3p@#_NTv@KhpxW6&olnoMjYIk8mEA(KOz%LU=0!(`5 z^Q2 z+axhtEk@5fxBADiP zT1Y4**ccvmq^IL4(?P`Tgf0P6_11-P2zlz)%ltig!1bBIWVZ|>H~F9-8R%^MO-p%| zr~bjd`C})#g8Hfc3TgB(+2g-A!)Wv2;#Bc->idM0tX|;h$K8K42H#jjG~@z#>QV8T z1)uLWeQu7#mPtj|otwt8gdPEGN9*HBizg97bV5d`s4s;mc-(Yp@|Y|TD_)-?!=-G= z`)HJmV{eap?JI8Y(6m)1rxK&-8aRtT^( zNj%WA3)(dr^>4{{Mo7SOdPD98O+>Vu9;5&A6--j8w#tDF>=<~;z|iyG6#HCUSb^}$ zXf3>(;HDD2`s-d+8LlRK8n;^d8_yI3IXLd|19;L<(MYVwQ}s{B(4es<*L`f>tk;%Y zB~(oM&Z5UlLt{3MRG_!`=vGxA@_Ql#i41h`gZ#xU7dnHJnh0E^9w*T79>lK<(Lx>xh^lBal;yW5JQqZ1w z=Zi#nR*+bECC7LDPwVSplv-U;jTuSAYWi4csNJhhp}B#oJs zyhQZ#M&rG^dJ5V_HFX^)ZHUu)Zg~Jox{+cbriy+hQtvJOZ+BMKQCAF7^O^-iRtl=Z zJlQ4fRxw299ow*IB4oSJC&6fvdF@F(BVXxycM0#?P$rnIjDxX;xus8l5 zUXRaCxYRBz;(myc8f9XV#|VpZFEY)w-)0%IAZ#g7|IQ@^+1B){UBj(rNq$JK#B<4C|A*1&(0^r=PeoS z@B!v}dE3R~coF{d^?4bc6^|Q4C1Z@B9%|&()F5H?D+muYJ%Q6r`m+*7m)5iIB$za`Iie6G!)N75g79u(2nP0kSYxe3kx0HeiThsqxZG7w>Q7| z#VhS=C?^O{Q>f$~T`!`=cI7M|k4_Hv=XdTk$7p#; zl{Zo%r^MT&Cczu9i0S#0OngBfL}B>2w^ke&xh5)(lnM)JW)5+cx0`So(ga)3-v)V3>0z4)Gz#^{ zpmRyvu%u`8S+1fGl7FG)nqyeWJjV*K9Egq8t%c7u`5S8N3z?Z>2q^tCPW+@?H59Nl z@KquLW8wIQ%VzeIx;ka=STr6(bWDt2C}I~N2bPiU(0;ER@Gu34ipkq@f@wn2hm)8> z^10u7`a;v@->F=KV)P?$a6xLYj@)~m6swMiyPl!xtKd}`O99Wp!_!k4RBvQ52j-tL z>=n0y4*D!QQ60R1BGTkZR8*9ZkPy)2YXaN@(;L*k$_mayc>dst$ethGKEr2bBpIQ; zVVOM@(Hzj1#V(N0#A?7I<&76Fi1fa84#N9xTy&@ND3MX#y-Md}``+82n7Bh+%kpyA z2pxnjoI9wOiD4uhK1VJtqpVM1`K?gPur*tuHA}F`mtaV)Qn8+#+y@|DnNUGr?2FhI z<}D|20H=FK;rQHrf7tFrYJ{Z(9>P!RxNW)Z)}oq`LJEBYs}*;A9&VUm%}h%=s^CP} z^QHQuQ0<;nt!r_)_JMtjoxeSSDDOgk-GCA29W6SP-X*4h__2?&u@}QDi?z~ zxaPP2{o56(C@Jx|TaPd*%}2n)84nH)2HFvbVje)~hAY7`rK6|UZm?egbb6nP3JzLb z%i&?KYHx0C!o$M}Im`o#iYqFV<>e)E@(&KIRd`vegF4pI(ugKlWdmoetgS(8-_p`h zQ2N&$1EH{bbMx~lU%$e-kO;zu!5wA*c&xkvM{@VibaZq;O=EaEq0ig&{BU@_H_Edn z=you@Gn$Trjcxq6Sq{I=E0=FydiMw$o&gP7#}ITf;?wg(0xk9%%}K-2ROc7G*H=SSt@)g8CVKrPF;*4R`nkhRV?{$2$683qU@-<@etKllYX&bOpocnl669I|x<)`>*B=l82g*pqg02me6#^L} zBg!G)+ZB(GJIhTjQ7fsM-=tVSlemnFa@C^FR5Lb1@}ze)-8)xSgyvb~<$7nI38f(l z+Kq_Gob$GJfv^T(0n1wTFHB<5jOb04s1!u~p&14wGN8&hZW@xIx-Z|%r)BbZX}kJC zq_6pBI+CI34ppN!!p3DJGmskuLSR6d$C+C2T3{(&tbvj7qk}yZi!Yz>DIuz0+z)YH zIFONpak6hdMYEV{_N6}#nCX~{l2)=QkHy94be)VW=)G`IP?A0kc)prCJ({o~Dad^n zXi~y>g%rPw6%mE$v`8K`zXUd_2msnMen8or<^QmcZ{T~LYLRtc@$yiyn;p=epWh?G zwlWz5BTJ+?Z^-|0-cO#drc^3Pg9TIw6YLUBOqPQ*?{fnbKuBf<)wqAAK$p zrs0}qO?+63t%%NGU(M^^dB4chl^Q_+#WL!Mup!*lw3_m(ys_uOw=CxCw}Hie_V&!Y1g!iK znB?ISf_D>-fFRHW94yx!Nu@m=9-dA?#@vO; zrXvR@B5%anf419*^!WPxyn`(tk#1B+|EUwmzQ^{O9HGI!h^0Jmkh4|&G6w9J0t}>p zuix^BU^R}c-N>K5m;5s2aLL#HOMzU3h7pOc^g*u-t^O{|-NH738benh;}@3ZM>(vI zuPN+?Egs%18IT4nkw2eB00F)`4e^G&5-SL?67b#l@xCgE0Wd9)1W|*>#kux}nTlYr zxFNwA#PI@Q%|NH#R&1Uio6)rx6;@$OtSQJ|Oq3Knr`6)l)O`kXre2~Elmc4}H+%f= z{<3l<2#$wEd2`*a63pic-Qk+kbvc8ICGba1gn(%wO7n*4Vcgg5Cbd|R*+M=yoe{*r z4>@93JBS$&?J1k?0(nC@5-sq?-3-`JLjJ5%t*kSZPialy(0aEX_Otv?jfoX(jt0Yz z*5X*=i10^VAtz)Xi~LkO*-eCkezKOeS?p{%7=mC^3>D~xb=_K@afx#&q;qnZYuoc^ z15B?S6A)Ina1j+VC}(w|w`jnH??D9P_q0mjcuWe{*H{Vokk=AVOd6)dGg5=-(A=u} zh`+kO^5(s;P|PG>KPG#Y5uK8KISxy-Ezez%_)_}B;`!#&xQ2RSMqmG?n!cD?#wgxy=r zH5stj?)d+zH!(W|KT-YXeE)OIy&|9-2|=A9<7G-rZlMP5=>A(5sj2y^kvn^?zU_oD zG8~=qw>J=yZ@m%Knm=Np(*nY*)J(M$Ir`UeG^k{IVNKL;o;uFI089h0JEbTb*-|HB zpx{}yd9Y957g=#}&~hUsTO%StT4x2&fD z`_|?1b+)azPALMIZc9)tk*(kF-@hGk7)g=HG>vn`)>n_qTTX(@@YsV3Lg0i{1b8F% zF=eS(`SmUA8&7;aaJ92i^`ch%qj1tdvy(=xUer@d9Va2;WJQxy6V5&0ut=3Nmq|lEV9t%5p;WR0!bR6_*wrMg|=Y!a5G*6q043S3r{ywinhYM zSD*!B)n63luwAT;LL%eR_{N7=38;k1!~|}NNd#tfA!95Dr7C^= zh{aFMT-6wl074FRSx)6~Aj8fv+Z{>W6cUp@6Y>oGE(_;gR9cFLOhKGBl3rL{jol0L z7uuf2loI>zJC7|=&U*NvcnC&zfBy$dWU?fw`v;0cVp}4|B`T>iPtX}b>d{ZDR&GRH zyICRo1uSws%1CV(oa8M|t3BPR8jbA&lrKWDkX*%&6gUD8EC=sBdFKlz#ei$i78UrS zgie&&XDX36@tRH`1!ZCA1l&@X|7u6(V`dx8$ja=##3J&)N0FQh?Zg!g+^qV3*4He~ z%OyPn%8;)?(0`wQhueh63Kf)_;`JR?OJgH04!aEQ5J)Q*RW1|y`U^F541x8)G#p}*V|Q{$k%=> zE9|nmE%bCq_Ld^;4Bsm0ctR8(o5 znLKb6`1CVvjGiW$WJlnBa3nUv2EtEk_yHq%lv#Xw0BssB8owBQ&a6wC zE6}68^gChkoAO)%_HKH{&F?APhXqaW4F`#x#c=?*wKX%idZy*y+f4pmYlOKhY#d}mS zS(NDQjThuk1`>r%q9&2Q2E&Dk0UxwEIKOoqw78D&Q%_}4%-V(qs^u|2kA(|hEkngv z2?)M->JcdUVo$w$y|F|=cRpOjSg4d;aOMhUDKLk?cDse&c5(igR(Vd5#k1yrcYks4 zn0>5%r;N&?zUER}Hol(9AR|8g;gIMz21jHaGd^6IvlA7Zcc0YeYjFz zYBe;0!ieaB;~2{@T=*cruhk~ugT3iu#_gUSOC4_l*39Q)#~8&;-9!(s<3VwGLvQH% zP%7tBGVQBg+Q!JO!7uqE<)=#As4wcw)GP$QIALOmDeK@!sk(~^-g;iF+zeE zUTQRi#})E+CVk<}s+x0R9{;%RiGYZz2Q=o;Rqkp-*Gg9#_9`na@h;9>PksZF8VNBw_ayQ3K3C`%w%g>via*3E(Hk)tX;B46MK}SG?y6q@?zw?i-9|te6@t znOWTfJcn#rY3j954W-_fNzL73zg@eVaSy4uAC_{f5#M&Xm*79RUy=~C!}M40ulZi~ z#*-#0y=N6IP9tX+$tjU>IT**5iyjb*+q>kcz( zkL_%6AYrn(WKlH7LfsG^e$-2+{FZf_ISFTwQ|2)_O-8Vd_Upu{ks*uzML&Gz1zo5- zxxZ{0nD3!!(k1U09{Z&N&x?ToVD!j=j?B3)_8OfH(Vmd zgeh)duPJs= z6h)3BXlkwJTw2QA@kvO*8YAW-x3tJQxrMT%?~nviFD8f#l}s4BhBWDO2HeK}X1Gno zZlbDC%%h+svI@ogBhR$#_2-}ACC(lZw#Ta8zyiP*l`XPC zoK_iC*2wM`AqU79Ew{c?+Y5rhr$6Hv#Dp{AQvyGe5axUwBGI|Xl2DsBenchjOk9LU z46ma*F7D(WCjPzq&iN*xu+XLH-dBv!fK!o{);w{FfNEnIW^2;-M~r>JPTssb-g)~+ zZA(sqicidZUT>9l|1czGVS(}~ApIr39Gh-{vI*c-1X$bI^)X;238}m}8bjw%=3r^-BPjCJ8e!jS_!FBD}o31PG}};|{`Fdx^kNiAQu!Fz(fCJk5oPe|GSc z!`P;s;&|wZBiEBe59fE6x_|pt!&60+iJ;>hQL2rFuh2WnHII@pOL5nw!*R|`Gr~E1 z0UT+0Zmv_W*4pbOB5x(LqOooo2VV$ytg&!I;IUkAb!EI$!09et>~o8bv$fH6xETl0 zNZv4??+2P!=O*!=?jc74foiDayVo>G6b_`BEXnHt-#Rv7&}euOJI-bLllt|epTN+1 zR3lHpD{GaQ85M%z^ksbIG;3yO_;;~hKM^TgF+~N=K?JD6Tp`(fnEu3gMSAu&Uk{mk z8fENlRGXLj#xM%Z^=Z!$KLDh&j?z3L5m?C!dd{i-`bdflV1Y-;CZQJ}!%JrCm zBNweKd_g##(uND6xXeQF<3nZ~)xg3;?vB85mK{{Ng;HJ8pvVVry&??ydjHHAYN9kP zHSDk|>TmGz?|;!wZb2e9)jyr_1E@G)-OkA9D7=fyFjGUP@9un+N#iDYhzxFDR#p~J z3+OjGu21AjNatgLG0pRN%iC@M#pk(1#rUMSonxgtVwWj*&e}Kdbc*v36%R_5C8GM@ zJt`jq`)&ti0~hkeSpOeWZxt8S@B^0U=pIC`dRP@|+XMC*^*lCa&P^asxDz1< zL>PaY+!U4Hf8xx4ODz=| zlB44^J~lAY4_#|0bsax)bz7G+@+(tCBM*uBj=DZxgMT{6A`YN)rTZ2*3 z@%F_25`69?j6FNt=0lpN1=03E&B`UdCQFz(6ZeOHzx5Ne-1(5Cll(Zm|n`fl0>3<0s7B6V=?gQ`yW4f)yDfYNbkv<)-HO z0e&Z_#+|9r5r#g{I<4tb^ScY4KC4wyPE?_;?k+w;Kzm*2GRK9@ZY0J|!}uEgBv9*P z>#`{?ydjCbjSh&7=}OtL5Vz&~cQGKK$9W?=umj|}(HErudn8@Kj}p5_lfF}xMWFxh z0}jN3up*yK;e0a7#{yqM?eCXweC7B}r+5OmGTQpMOX^~cn!|(UCIa7|9e=XzQg%pK z)Zh?Uh;h@BjK27`=R1uvM6@{RWj4F;u@c1ScJ@R`V;1lAv)Xr*SG+dm=jh%{WBTq# zem=je$82uF!{wJR)5}Pf8(+AKj0;rg^!q3^-hxr99v-oT@0XcKE znwnEN)` za6>PL>%z~%3L}%Ouq3e#1?jXeB5SBCq~+QD6s7#p;O}5#UptUwFS}SEsU|!RHB61Z5pD;w|0$EAv74O@z7eq~vMuGNXtsf1UVa=*d?} zC{%pi)=V>HO)b|YM_mEb7Lr8Tfl1N+&pK}n9iCC28%33$XF<8Ah02b4n3yV`sg1+q zM0BZc>zg3zCr=e%9!gQjFYcwPUKtBuZn_XqTq$(A!w!`tQvpNCEs zNeVj%V%4w!ELB4Xaj71n)&X1JP86}YD#NUMzx3v}7eIOrS?Ha|{r&2L zNe8_u0b+2-vT+|7{!;?5r6E@7!Cb@$_7_Nfjwqdn-tK1*R%OK3p_wo!lny~o^oy`( z@7hP&T=Bl2*~2B~DrzVL(F%$?8q=LiUfLJ`#U}>mhw16zjYGD;gD5Ke@X4y{kanj= zdgd@i72{uqoHM;5Jf0CiDBjJv=cGw~$QLmhf|Z>r=IxljS*}Qr4U&9)(iQc?*;*N3N zQBYZ6-|_*L;(!1C{XJa~@Ho?lVqDDm+0MoqcmFmp3rqyY(V#JLHMK)bz(WotS52oc zI2w|Y?7G_NVdqsDO83Nj6Fl5V^P+>NMJ^cGFazJ1hovkmb)G3?A$xD6xm(S1g)H}0 zx-ik0zobOB;c|hA&Sbh$eTs$L=T@udGzsy$8B7O{NQ>{zYCvX23;VD1QLo!Ern}83 z8w)Yls=wb)Ed#8{4?~JL4G{UCE)XpICknF;By>~T#?2I-t8DsC9wDMa*Dp@!|2Zh?-#)Vb_r;Nt zf-cWFoE~~~$Zlv6BiPW&(~A!Obz3$QV981rTRAe187HMzLOJl-8(qZZHTF7zQ&BQe zKA*{}J4*S2l`aUXTH4-1gMNW*Tuz->Z=1hoEiG)t2%ur@iC0xYoxtfC4IL4))kPS% z4Bd^oRNG<&neO6qhtcD{6>wz#bt?WCwK~1YwQ)I73wMC*j4}_|pPmjL(}mZOdBbjq zrKVczE6>VRV`^x)zUHyv2S}hq+oE@yA$7l`Qpi_3@AtE|ntsX)VMze4>+>Lozti*M zJ#ajXHlu$lkVhbNx3~QTh)O8SiHb+w8!<340`s!-T*W8AXQOZoIUv33kKk-$DO|{?=i3pY z&L|K1<49mCTC$CDTlZCm*Jl?)v44)H9;NO^?&(QcXyetNB(%kj^?oDEiVB#-+nM3u za%-F$Ufw+>*>Rk_#LvaENf#jtEQ`hR_p%50{qEtY_2e8)?`@!k27OVJHIy^<{$zZl z`s}-{g|dSpsYH#E%-ZF(9gt$q&h|sU_>1)Ql!LEyFf;03t~*8+?tvCzUGehg4NXMt z{$IPf17cVZe!?Bso8N39*WoPl8Fq{KD}WE^YVmFNbpRv-UQEhI#0ri_`5ONTW=QCu zn*+T`idJFH!@paW+LK3b5EL(7Vxljnct%bu&Qn(5K!wyOatRVfn)}}Eb=9foc9wV> z?@LJbPOVOaI?jKr&hcrLQ0BYG(bm0dQ=&=Y*A`U(T=?AWp!fjZ;kV9EO|jSnBB zE})%otngNWuc2=w3d(^;4&wt}m;G@>3Qw4-(a~f^;7AYJb$MYSO{hpT#kr-ns^nPT z$4VfB-`I#NW_*9Ocf0Q6prR7rBj?H(j8tay{`puM8;_OgQ^Z8DrImg866;W9ER*o* zw%TS|K9fKL7MdAU8!31R`q1qz&p>vKLLIy=`rK8dH5qg96aNsVqp{3&&JLC0rwTLW z*A*Ce3-02N_rwG<{HS79Vb5sRvKLkMA!>jm*VTayozg?fk#_IFI=EW3Q~D028&GN=Pzeks7G zQLa=PDyfDIMPnleCqMmZ$u>uHk=;`QWrq zNTAHCyDmWR38oMZ3PGhQP=f9v%^PgKjG*T2%VB(wzNuuc_@}D&2NS)@?iL#T>Hz=((>^-t_R^`g~3ezC1vvBi7~G>8;H1k?3xXI}p7z7?OYp zx)}`af5pd)9Xvy`Y$5vKQ2p;$SgJs#&nhfp4-}qM3a#9oj)&wX&9%OPdjk5PQAq4o^*QHZ;4zQ(xe& zWOuOi_I=Jh>ZxT-dTi;utxi%>u3&EKpIG4i6^3wjf#+SQs+ri*8_e~*NHI|aT5*pCalZ#Vx zG>Y{WkW6mWt)3})-`|PD3#Y6I73=&ntpu!u?99Yl``GOv6H=LgGl5$9LiGmn4>%N2 z_sXmQN-{;@3)qW2S!L*55AHzYQl}RWnu%m-(9Z9}Lzw;iiKo)PJ88J>z<4<|4-XIj z>C>nDqN#3iTcg8@lBDmnSi`jpfmm;X5y|2}aA-Lph&rjeQ#HFb4!79M24}r3D$hGL zE*ygT={O0Vyz_(6zcIgB81}mi)#IRTAPzTnZfE6vlvua^9;X^*EWiL-(HGh2t)lZ) z0%avKW^!x(e4Ho`$MhyQf$^}8wQR31xbi{5<7HhGWF#IP6!`hDQ2O^NAaguK+sWZ4 zBbv+kLaM|U>ru+m#St+S+oV0(tcadJZQt5$0YyWLQr&$7KMMt4m>ygRf_+h^!WLEk z#YqPSo^GqRl0)3}W9*Uataw*-3qgwYW21eo`p!J=EHev(#_~nqGr}zS8^eMI>j9jG zj3z-`ym%UG<6>CxOrO7YO$=h~A7q{8FoAdP3F9f3zI-w2`lED>MwB)EY9YX4!-!=W zjgv$?(jzF76@po70XM9p<(kKWRMQqBYV+h8N)hxdz97G{sx`soNLJwCM;AFgij$@X zrKdxzS9lYRMT81_^XEsv6`y8Jdh5o@iVwhp0xl9_69`4ZhQI+3mMnn{U)=6CK>Jt6 z+iFJrVp?}eh$!k2lnf>BSc@MK%k%%o4AtrMyab|`k(XqDZ6)jtH8P3-?_Q zf2*C)X#4Ede(}5@JoPv~+5_w*Zxh1e_`^2u3mYyh8l{f!nCFsOig*8>TIOotC9q7^ zV+bxXjRe9^q9aHhJ9dg%+DUoKIb1{Ul47`dIBMSrxTOwrWro99DDDZ7DhKzTxyBkb zv~a38omMXKZFbvVCzZbAshQJ`L68bg27mGYL%?hHad(B3;XE3>_jcPBrkGIsohl%% z!@lIbI^{>p<-UMLjsypabV8ZjukE^DZo59v&C=@-Zm$cmO1i1BTu^NYE&Ji0WbFTJ zXA~uvtMpUl=bMo?83qmBGe?6994ey<#42W#j`~7tha*O`rDkZlsmSxqPa+tEn!5y_ zb_bUBSB<{{yA0AB5Nle_aIpIzj>LXP_e^BICZCEi*|B}amc;smYTnPUsq#qhlP8WMwNL(zsS`W`QFDY5qpCf#`0|zD zG_sddTbKk#zL^>{UvtKDxu9J@;|Wt43tM!?HEveA0h{$f3)C^AJJa(UewoB1 z!iO=s2?zK5Ch4b+ZmezRm1C7!=68)vL)XzAY*B@5$oThk?jetdlKEfIjGxkP`gt*g zK5I!3JH@*tw1pXQ88?>h)AGUc{QJyEGT}%eaSj__kh>|Y6VTS*e4knoQ1fOvB^Bla z#o80CBp%pR|Zyi^Z#UdATMc76U>NK_o{?@Vp-YPJOEMG zrFOStASbD{QSbeHQU6g90(1pBRJEbdU`dm1rV;w^MVfqZcy zJ*n3msz}vULpz7?q0;c?&zp%M2!s&J@8GMJp%*|u=>n^|xbWD%vv1o=?d%lIv)~6* zg*5gjpzwR>5^#sJNtGtS!_tETd{|*@XDv_fj5;wO*&xUCD-N}2@ z`NYKlvS3WaOV@(o*kS!e;Bm~!fNu=7tOz=s7g{j!Z@7PEf4A_Pm}ndgRXbo6VyV{^ zs-|aW{tf%Rxb~mW>ZLv*s0TOg#SDF{+P-uM4lB0I^GnsQ zmo-KGq8th(Lb^MjRXg7yqQzZf02FZ|$<4V1L%+LPO2g-`dy{jbY1uWVlkd2N>7RY{yAQKwR|c52fk%O zSEJQi)!ta!7cCoYnUZ%%LE6%Re7d2*IG25jgIq+4GsNpmZu}kYJ^k;q1}8IzFR1mg zBERh;hgGqsdHUuZ@4^ye|Kh#V!VE*o`^r2ftwNoaN}-s<9D1bwOFbeGFzYO7!J$E~ zkxT4%mgnnTmXo<*?yaxMhYgoWv5Tp_8Js#nW=dh-v;Vv~8E}xGKOgKQR{Q}3EZRM% zAd~q)>)W0eX-_9)^T{QF-|ZG9lRVbcCjRI8St^2PkOSXB#8N{^lW_E7m1xxEVo-cQ zpmy;6-Q8#3+~}m4eOoeTuFY$+aix}lboB8d;9)3y6qkDITF8Zp@(gFUS$QNY61V_* zsbYy!==zK2qkING(Rp$@hk40|| z{qgL}c8ziXZx-aSFd3tRcA&h{*Q?xk`9yK?I}feIU-zrX_(dd=Y(FqH#bk1S`^PeY z`ONCp?D^Nv+K3P->@g{8j0N(6hgiIjoCB{yJwrVT_$%*8+(*Cr@*M@f|4SGuq5sRn zMh_6N6cHBAC%5$m;u99*=~7OvO~DGkk%Ew@Und7D5)GcM0#5pWnQX;Zhq&Dj))K_{Nm3?}655c`fb4~CH-*0ko&9x*;-Pz0ECy_6M?^(an1#cot zLIS(nr~d}&sg6FyPbi&Q3K3j}^OCbiF^@th^q(G92ws#F_mhwuOs=rKKyda6K_qZ4 zy1vWU8s`AQBj8#fV5i5tzchETs}mTNO|(3dva!_uZz7yy{+$hsYbNkPrFmWehv4DH z0plfnF(+Suc>A(*{%W}rKsh~paNx<#cf8;@CkE9w&@q&Zdd-lsH;qN^2{s+e4n4S9 z1>(7Kl}SodVPL7nG8C8zoAP`4zlCM6EYM6dUu8d-STKl?Fh7ASsIj$9t^K28aNe^e z4n)A^v!y?nsgH12e3&7MN6IZ$FdnUrlRoZwKIjuv0*d!Ud3;fZQ?wh6yFd zjCU+jzWJU)81w(K0DziSpbB{VW^Ol(I+Q@Pft6V}=GMq#W=UI01p}v#jxN=_))z9D zXC!pBfLP-@-oXA|Kd^7sogzEeg^lAbT!-|NOj^f__L>q}$W+V@`J&^D^;c`*_~P^! z@Vcns^kg2N-ei-`3b7y_=`EJvzUvG{W!G9ZS{g&?4=p5t&izFH3u(&%djT^L@#`}8MxBCxk731PUR!X;#uI@{jUUn-d`Xl zN@^?vz}vC165?*xjVz`HyVAf++G4KW!YCj)lo92$F?&J4+*jNb68Cp+V_x30Xl$7~ zdD%YOkB>Bs9_uF8V^lR)1JNRVmM3Ipm1)komd zF`wm0RGWLZhXdHCR#;F(Apd>;_rpOh2pKz*r0gRmh-#8z<)+cW=sIUxBx|YIm4nG~ zmVv(H_AlC*|ITi{9+U__RWumNkg+riZqBFs?yIQyYh1L5OL*9Swjop!j0ODU+?_}R zx$`f@k$P@Wy|n99MgN22{aLhTE8ShMNF-hbtBQ|i<@-r4!z>y$U5qKUFU4HEbMZPA ztXIFieC=UPIpZ_JDS(8DEY-87I+ro98TdWlAH4i2mFRU+3J7>91xDrFWYuL`eokJ% zsLr>x_?h`N_j!qE6qRu?r$|1b`|)b*`H)5wGx(Mj*r)-8gg^;RzVxqF+!+=~Jq9R9 z35ee_lrF8qVywGErIoLVuNUU%0GvWQ{`%~g#~z>;>J3^^e(Hm zNs?hU|K`=J?wEM;9cviIiG|f9JL-tHlXwbvAn(#e2yk5Xt^1pNl}$K5@h@IcfrR?o z#$Z?~rg-}WSygrzIO=N;zJwWROHvK`uxq#Su;R4-d06T zkhU^1yw^XbTI>n(!PIW@D$AfTE!C;>IVct57ZWoz77tP#b2(Djc2o$B@fc(v$Glg4 zC4VQGTwf~Va$jVzsVcpfs~!I9ieAo&p=*vvNvkXy?qi}2t)jj+LGqV=B3>avmzt_N=;6qXZ|~H z2n43L-)DOYe8E)mjdO#S>Fe!XcFr9^=AVJ>pGq%{l`l$s25h=?k9X57<;Yw&lRqhrW}8Yy7+wKwz#Cy=E*#4RI};#GZRH z3Ni<)+}bmQ;hS@&2l-1Xg~zwLu9c8l{rIm{^-Ss9l~-{%6;k|aup(qJHrucL)Boub zD(i`LH!(@k-M9;o`us;2O+((8y#E1*Jo@x$0h6H=GZRPqoKr7|RABU^VI_-$TfSL~ zl_?Q>n0xzpaSJxmni>qEfhd|ZUJ=7h3P!m9_?%S&&+4{~EGTtm@6r9hEn!zu=SgqM z`@WOu$8TAPBed)I{lPP4N>DczqUe0IU@Y6xA7$+XF5mpN%6S@Qa%wD?pgTR;uXN>r z0GTQbIEh_=0=eaeK92tP#{cQqXjmV|LJR;K)a_sP!mBI-ErvPE_PbKq$C-9_b29)^Kjr14TT>+3#@u{IKeAF%K5A%a=mPi9 zQcGB9|KUr@5z&>+udJX3zt{Sk$PlimX>I+c&{Z{yAH7*;*NdarO10JT=7;KIeMP~) zq5j=L9;-Pa3%)vwm|r`^1S|)_jR@98sZNo~A8wb%2T+K*FUU;v8G=VSNpY~P7}fT- z4r~p7WM@@ApC&wX3ml?B61RRyQ}`at)`}pq`+olW!aDxvMr9ulagxu@=V2G~xSgH3 zLDLvmlkUJ-!=NJvg zj-0#tV9-Fm&&M|J0vDgy%9V&WWMZb=myVW=3F=SdF1pC9&*?mF-tzv>w_^Z;k*@`o zJhQEo9AG}`i-JZ_`fGdHnL7*jAx>jsfMIU3Z7fE!12lR=p;#cK{{90(r*fem`C_`Fnn>9l-OKmuHWar` z_+Rm2Ej0}0eq*VUIiIXEc<}nn^Av9Fm5x*iEvaeC!Fn>xHTY+Imfk`fjU)wXvl5!L zNkbFxi|~hGZoY|JNd0g=IjV2L7P zkinFPbON}Um&=3zG{Prq@$u47Wc^^}U4_MQDnjx8~Cqfzvt4U zZ%|HgNDW(54fEMTjLY*68N#|5o~L&9q5By@89}ENHnPQV#>PqaAG)GIEZDdK2BJ2v zu$IXmUEQJ%IfMC8!b6kpQlmMjM@ElL|K7HWqY7HMMEgCkJBVyF1uXqpu`sI@Kt?Vj zg#Hm(#+-jXviA8%`5GGz}ah&8kL*!b&h z%$+&kyf;%G2toU3s@sRqUL*N0nDK5+^2qt0l2wVdK$(*pz|>9c#$o z!`~no4jk12>Pg3+W}|# z%fH1%(z&sr&~Wl1Gz-!;-ke#L`8H-A?;zEDXyXra_;QR)oHSUpnVWu(3I?x5gzjO% z^*Gk3Zuf3ijGcMVcR^rL8fB=6w0DwQq0segV6#$hkmaCm6KdsJ#Pw^rX}{lAm^Lr3 zG4cJh*w&mb{5M+gBUWV3r;n@~{j1=yTh4AT0NkpptGn4$4Fr+r=_bV=SXnJDkk}V| z69Z$K!^08MPgb4B*dQLjuM1WM;wa<{&hk0m7Yh654@8mE3K7i%N+EW;>f%Top&P7e zUxzB}|9F=x<|8S!2JE8?Mge|MCK;-szh`TxTAp{P7a?paQ$Iui;!(HXgG#Pw)X*!| zCK?`^kN+MX9uD^AN5V_H8mD!n88J1jrzp529|+7|Gkh4yb`4=Ca@0l4nK$k$ah@l>gZBQ6mF`>C3Giq;wJNQqrJCe28{Rn~vhq3~Z$$XZ=Wz0l zD1Z|MA#k1=$5xxnk8v{+;4aGzlr;dO*FoE82z;I3vpLO z5TqTT^y1t8WyTKaO@;)PA^$@)mh$^Od7~3gJ!z*j;gxz2WiCz!6rvcVF}-?F=oznXT}%^+d49& z>?mf~*<7)rYxnMYW;=s6mdV@lNxh(y<~l=1|F`lGcM!6}M=MF`0YH)eudq+R2kQNx zsTmFtss)zdWL5E~n--l$i)!HC%kvXuz~d!ACV(`-4(p3oQ0thO6prmlLo7lsE_E*g z;R1bH+>YlBIz6!&c`U|}2xJ_MnoWn}jUKlNnFv7r&Pd#hjEsntzLcj7?#Ne|*_!X) zqqmm5s{02J59np=2ND=jr){t_u^(_7WOKTqvC&vCtjD6!D_eidHxvW253@b+0`4W0 z{nz?B2JBhS%X8rBC&9hq_U!~8&6dxA%3JlD6iYxB6+a(Ge)GM>TXs}aJm!zxcs=Ri z7)yksC=8`=^41;;&6F=kNiQu;C+uwd4ESvepLW?!$?qgTVo~s1!17@05W6yk^ecu4 zS`e6LEvvoJriWu{G0Slr9nntt!J2ouftFq3!Jb_w_WgF~=&yYXFVZuUwk{f+_(aqa zjKpHp4js|Wm%u12fu+2Jo@0LB8oBCt``3@W4rrZiR#;N?R)qbaevR-_O)alx>TGuC zaH9d=fj2O%U6cEo#r`e*uf6|J7?1U6K%i7kk1U(scvqCTe=4Jc8N9pq7CKo`gxaKQO^j3ixX3!JC*z~nlLbMl{Em`&z zK>GvbNIV2FDn|SE-wca~cN}~D8KnC4K@VO)P6RR+?AdV9ol?;?`p)3H%7Ty}xZdP`t_vs;kWMk$ zEVvD-Z0y`DFCp0PO}MS;(?jPYn)td7935$0T^<=wf{{F?mStH|5cvZDewKTJJne z<5ziH+Py}H2F*W(Jtq3K1yBPoob+v^i@io;7Fv%M9`*0vV|Lc;&`*v7Op>P2KE)7+ zbYoR~XnVgmfpI{pD)hIz7R~C)j6%E$FYLX-zU72vR6Q!?;m2e|*qi6JRvlURfJUHR zPZZB9AP`~@7{GtQ1dIp#f$~8Gu>mU~ZNR?q7E>%Vkq0O%Hm{uK`=X$tVkaRgy-*23 ziGqkxPJ5iB|2fqzc z53v#dZfif<>)nTp_`<}&f$3lHkp|c(%djr;Ov@y$D?(l00k#-mZ?cX)_Z(Ug3J>kU zZAP=w{=r~W3?sy#!)UtTTslaXky7~{sbgNUo$+n&T8Rhs!;Zr|-p+gZZFy{v@8>Wz zwKoPrl`{CfJ5mZxM&e`Lf92cCY?XY|h zV4WeiJOR(X0mJr1skj#mkBz-NS##F1`jn0;REcVzXT2j~-!?k$z2g+1@`U7vL`r4x z+XX|Pt%F%%^>1!&97hlu%%1+;jF9Gd7gYD|Nh(bFLp*@PR1a;qj(W1Y3&A z(CED(r~LgB7}Y#=M@Q$DMsOJEOJ*f1naksm52-~?PDWa9J#2auoy-1Xq0fZC%K+O= zSVjzrCR|tzS=vBV#bQbtKQ4k0tWEvWN=x3qTLmvh7(#O_*9Y40fO|)tL|CU{FtFL? z1DRV@G;Go3#G;S?WgC3M1NDq&>~a+C4$|0uDvl}dVvlpuQ`6H@cQDb?izs3RZ+;F1 zmaJ{!pcLNzs{0yTL_9_sS@}5&snB1}LJ9uc!Y(Pu;bZxp_5+iy{RFa#PK#pee>n@{ zrosogoGjqq4vIGChhl?R?PGO7VpL#;(Eq8$nQC9SGDeU2Pg{DP*wEA@KYAKpvUBwl zA*CP`UV)Xkv@fXsp;iQ{@`siMxnXCA!&hG9r@;vsUxS6DRwPv%ML8Wv=1@>>u@@2C zMX72a8KFH#eE;%M`A0$2q}r0TvV?sX`LYNI2$7o^bQMzL6dScASU92-$LI;e)lj9! zqxpBdmZY8^ikWaDtTyzXR<>jBqe@^XB=U7s-hCT{;F*8Fa}^H_9LFnn@>P<=dK?A&psHp!JA_)12jQ`q(1{xf+R8Gp?Ht#jx z^j!TN{OvVLbb9SLOwik_FBphw9|sq~O%ICT z4;rOEg9mw5>J+GwvsF~fP#1gY=nmHhMD9+*$OAhA;l>S=9y^| zs#aPO!aE;0ZYfVJhoU1{0oveKOp>(XMO|=lE1sv-%-aPsdM&!FT^LKcpIF=+n820Y=W)K*aUs5% zBoK)s5J;*h)p|~@ZB^&Vr_p{Nz6BhNz56Y_n{FJ8NGt^td-R1BlCeVm zhm3~xOhb>=Z{&7Pf*3a1uIR|0KcpZd*z59t61 zq>~dj`Zv{UY~dmtThDd*sDobT2ipYFS;?$+V5k`#pov4ciTBxD$k$jzPA*#p@^Mk$ zQILSm)XVfo%=fugBMIHC-V|vd2VoPY;A6$~yKKds3{i>t0m;BviVrD^3$#>XW;U;sT7vt9J=-y$3QwPh3 zA?4tJ>O%rwxTC!(ulCMVCWnTg zGbMLJAx)^lAQYQAr0Z1Y&v>gMhl?6PLD)bn6M#7dT@S0HXrEF=k2g+=$3(J5a_z{% z1zcZU*|%;XEMwuyzkgpb0P`Q3>Jp4IFCvVI1t%)eHy1JiV#p%H!4wq1mrbsJjLnXd zul3)K-(ECN4l|X$pGi)2t}3bF+qCi+jn41LQdVH?QFiJHc3&Z@s zazD#|XeUc?-ERy))K8kqAFI{1_3ZGB#a#fz?};Eu*8k(JGA3g?K@Bqt`((U-@bI*2 z{l&cVL*NY8rt^}4pFv2qn*pC?rnRbo<$mdIido91ea3I}cRda}rTB`t(vWE+M|k?a zReFOIEQ!K&YZ#=o*E=>*5WW8cG z4DnpVAO7$M{YoqrK>hYgiDHt12K0VG%4PaYf)j$3PJ`qad%~XnORG(an-Q8I$P$CY z;h2=|dS@U7y%3OIG5nM7fMaJDw}#k~_jR>!ZNEteu|w})b_jb&(DCWX#_dFTb9>FS z(0xYOXMl~=ryyYOeN@mCDydHkX-H44^&nsUhKA;8>uIuCO1bb8TkD5$?$7kJ8q@0- z=4=gpKHC$fqW?%MCh`fR?@(@Cb#!zpHLI_fkwnY^t3=mhn_|BKH@;j<~I0CRHW zXE0vxK#P&NS8|mgydGv&=FcAu+kE>`@Ph=^)e2+EnpoV)T)IhY#9rm*E-2v}HR&SA z2(W*}{`$*HH)J5?YFlX}aeHNxFm>ff8}080tDR|I6W8$2(!F~#qojHnF7GvV<> z(Y!aN)o}9A5Gzx*Gg1#wn%SyF!YVv90Bk<-0+}zP71MZdbY!&M4>Fu{e zdB_GpyFGl54ne}-?`{8VC{9qLOr4$rMV3X{fogup0hpFm^N=Vi@Qczcce`86gm|fs z*?gU$ajmWTKk2Sd(G94SF!7j|4)V3yDY<}Z+eH08Dp}K4sq+7Ypb0?h#%8hT42>w$ zp45REK6NmK&PYL?a!{*nOODO!($O_?qBt4&{vxLE2A%zt?ul=fgFIJKvvCIoc_)s5 zE_-_euAhw}r0>WmcJ)7_;pZW(@H4~i1Ry#aKp0sBIN{~y{s zRw`tyKuf@=L3q14g&^)l>m0?AGG|>4fE>w$XJe5jfIGOrF~nZ?-Mu}4*CH#^hx{=b zs=&v`2k^82iA3#o(J2~)0k{H@mArz2=@}V1y1FE9@zw;r-{;5!z`94Y9p%d@c8x|f z2W$aHgG{1kVYjif-X*i$d&6_0L?KMZ^Z#dkGQ0{>GpM2rq)DN#>6j5}=w*`?k3q&ui2&v+_Vb#nWF4QTcBscHFs>@}*kKvU-tOEEJAmKycZ zblh#Y^~2d<2E3h?7$8)UFn8y*vr|53(zc&zvl&rcQE`;D@mSI$&>S5dZMBIoC2VMl z$4k81YotcrFVH1I>x8?D&Xg_je3@YMe3b^23Cp9sK(9zzW>$RT&^baO8`A|n^+rn`%TfS%+ zG5G`c3juaofw}s!+k!!h)i0V^d{)`AY&BZWRR{GiFGCw@3lv9Jd~5@%+DMPKM4G?G zqUkf7uV!4Vg~2U!a@LV7F)C~XqQQ_+l?sz|l5i(wqHls@k?AtQrK4~HImKx8R@l>& zsYQdt%HGoT>;K>hvRFYjr7f?yJ}l6d6!hAd#A_~t_hwuI0$szrYV5l@mIY1)9=BB1>-O4k5WptBg)dzoG&EoE$LgM6Wms< z8dBc>5fIs|0%2OW!bv4z(v=x55CKjb=N@scn+X{{I+rIE<{>dzd3mm<0dS=`kpS^b zFh@lpQpD=VpFggvpB2SCrmT3k#kE;hWel0Vrlgqh zXMI6c1>gzZamg2Wk7Q?oO(D?M9!M{U&i`_d@CyuBRjnysx-SHJ;ea(axZ%ps1jeLF zHcQiUCh~fOHpO#}hwq*Yv>|K-DxLh*i$ql|>X#y2mk)|)7PVC1ChSvVcAVx!rr5IH zTw@>Wy9kyY+Xeg425-OP{*YA9KA2sqgrMUT6|DprZ+BLI#STSM{9mQ-&xm5u>ypz+ zv+0GpMTs!PiYc(pO3QTdfKp_U`XN2tNuQ#&`mHxMXiHvFyGdWLLB`tVaXgao5l1b~ zDI~BEY0$i{BFVtj-V^JiAA6VTqWo4{|tCKovB=ahkxFAjWGep zu-=YBn@Tu1IGVMF=mfh!Og=Qe6<0+zkj`2Zkk)}Ngs~_%*J7jF&oKb;hc$491?Wj+ z0D8>XTE`X_W(rypz$y;w`UN}_#zydSlKMed(jY7?oB68VdmuG{vW(z50T|HgBYF;9 zbL;Ez1A*|FG*3XnHW6knh__Es@awmBhJ?WOfF$??dr|oH*o_u;8yC6>RlRHzf4ix< zQ)-~J5B8qyiL2sEl95Js7*X1=*B`#no%dHoGTf$o+XEs6e!+BnM zH?8HltTIX6TfwSPRYW1g%m%z>W5WVpRJFz6h*=2N{)53l$iO7*&2t7;Zc8~7`*MDp z&cz!DGUi?0&5Ew(QS>p|Ma1Wn{6&&mXjQCSH-w8(7_n+94)OuI?ny5!Dr^`zuR9tZ z4YZ_~n6MT{l^loEnus+n&NoJ;07|DZO9LhTITXecR`@vY-gLw>OP*=iJ#DCZus#3`)F@6Fy2U}+q8QVBr9YEy#1Axh(2*AIGCSTU znY!^ry$dRs1R7zy^)l49tL*P#>8Er>fh7aj7rWli- zxgEcJMf%Pj+la>?mXTrPU22%Jwftc->Cb-QA7E6$68kS^8L(gUe~3^Mq@d--_lWNL zb$_EMh6mQwx26c_Rls|+oCvWkVvQRGcJ$HP?#zXJ)Y5~0EFqQhuZOYh z##h{1*hWt-I~8B6ssl{_x6jFarKx{K5;q`0Co^fqYMnH{u}~9h3R2V#KSJDZLf5@<&Mu)RgCsnGH!g#X|y4a37d zruOlusDq--e<;`_j+SeUPaWa)qeTbl>y|jp$=$Z?Xnd%)WF$GIKtrE!GJG?Zm1;Ds znm0&9^ZO6;gcQR~6LtQ&1YtmB*)r!HE50_?op1+`vkdZy}&3Z>eFJVUm3h;)W#;YkmvK0)7j3IF|8J**Sw@uVGc-ReKS7Nf0ze zN>y^C-LWVx;t=Fy*8cV4xIH>cic3Kpg{W*Li-Uf{zJQH4WiZldIb_24@XJUz_U~fc zgM0CI2@os3rlswObAxDrb6iy+<+xUKwSna0zkfm+c@n(7Bu zBDmM;n9IGHNQC4v;oO?5S-`Pjq&3G@3)klcZDGx%(|i1*ho%$y8-(Qg(&XdMJnrAM zb!&;*S3=J;!?qZj4GyP&RaTRq<#PD>Q2G2_dAV`2H09`;FRb!N6bzMwfbYV_&_5~Z z6e07(L(7@rsW(Fc=}-n|!0G438h^+stq!?CNM*y+fS84Zgjete+oZIq3D357u2Ksi zd`jX3Qe9FUC5mDvP4pWE9t37!D3I{tc2-L1Rt5IA!xv*@j|4#jXLyQ(V)(1VL+%6X zF?=%tH)^tS_O!#Sg3980d`V=;lgsNuL$@t|$42*3)%D(*K#y$$wUY_s zrLa@WPB1wfKT*iNx$_R=IJhVT@bAGuv|3;Zb{}`Yljjuf0nl}!-+(BBO8=^o z`Xd`N^HS4iTu;!39;$|-%#7DzK?GzIQ~`(W0YQNaMoa-ldK>5-6{X7L|3lMPheh>$ zUlRiiJ#=?UBi$WJOG=I)A)s`FFm$(ciy$f8-3`*+-64YH@ACP+FaPquGjq?m=j^@K z+G};h`JulAdn_ws52bn$uA=cXhQ_BJk zyn+#kGQfY4#lrBMoSKqCfIt6twb#8v`b8=kFvXyz!Y8`?@B{QChx8AEMj*WZQ2p9h z5oDY#r2n|B9E^RAWu&8kkmY+JDnsce8(dIJ2-_W-yR_6fw6oZnV?y-n1utsdaaBui zvccGOgTAtOJ71$a<3HL|1q^@OPy8vQFtP$r zun{1Fx0m41Omn>i(YyWcSs7f$e&v3s5e9I#42YuvHeif8L{y?v9MDoFP;}g|yQ!+G zVo5LNz=3Jb9wxll14N!5LVWZM{fSm27c6-qOC3}J)hI(cG0QUm$9z-c_IPN{t| z{9^$6$EIO6fI6NVfUgY@6g#zXhdEt=zK}1oPppMXc8E26(fDq63p~Q}b7a;Y(qM`w zdRTqrn^ZCzpX%Y6O%oMg&)ll-5O_ID#=WwLxQDDK{oY&SfHI9xk|O1oID$ttYxnG} zP)acXCj`1E3UJMgt01q&8$}sKMPm(Nw4u*J4 zjn9`GVt0*rAA9iFzms~4Ie)|XoD{?w*o7J?xhv9%lHqv{4}Z3!{oI?3jVb#Er`~)p zx&6Uh@Z(j1m|xgZXb-)o=b5LM?B&NhN0|P`3dao?^$l_hL<>XLdQV_qr#6Nlk zy!WaPaG38(pV?IsNHUDht971TnL}P)vk1xOg_N5X_LeG*vzOBA1RVJgSyx@Au*J}e zTlm-0z591r`J~58?U+-4G2cvkm|IPN+8qu;fdZVuV4q5&1;@7OmRF{5nlTiY>jAtU zS^O5SAY60r^#$q%Q3yRX$P?$>*!bGSJBl)P{T98`_Qh!##bkoC!aZI9h2Im7MgJZ;<)f4Vi34Z0{5hM^kP z_?+07Tv6#4|2Ha03%Ne~0Z~D4SJl!YVsoz;x%OdRpZUC)5`t}bm*4L(a-&bDH5oHl z)2?)Z6VJ7ojyjyhvi!)Ne(NI|`?b5!CQq_l^&vQl1%hB9;|Hd!jsg$LfMp>Fsv+K< z&%LuvuE>O)t%N)f%8{`hGd{<3>X=v4X1H1zSUe9}_7x^`28Ti;%y{o=Z+Yd{`D{e# z(7er}PWOLTp?O3Qga}bw46^lqvUlLnH5Q-xvye`XeRO^F#*$>vx=hcIJOO(2d?)=1 ze|icT2Y;h3XwA4Fn#1Zm!98rBm!OqpumT zG0+vG{Rw2HKpV}4kO55<)=jHVj8Y@rNGcKnTzA3lDHt`}>&NEApB3I|Cl71Cf8Pjr zO9M7a;o)czNQK^!Zo*GViHYkOYV@yUs&El-qTz5v>G=JiZq7zmZoz5axM&1~a_a&# zqG~TgQF%Vfv`Q$7;|hG7b)d`0@Q`8r5SB>ZA36Ts<9m;|Pkb$9T-ZesGXb!d;PL2Q zF1=O+?#gv~QlPF4q=H{4BELNvyxJ4drc`oP71;l_y6t#SMb~VJ-M!}SpI?&ssi7xx zvEQNPOA5lxSX_o!>G8n97xQwJ)QR8=%1z<4VgvRW%$c0X^7@ggG~DjLktG9_y$1K~ zZ4&|*zQo!V1iELk{}z|ERR^ubXRa@X?>ppbXKNNH#!`rFtiL#}-X6!NnrT*Th8?3m zcgx~J5Co+BYJ0;n!{^1CPrw}f`TTC`Fl>d!zh8lv&Y|{{WM2_~wcjDE!F1JZgCr+($`k^V z=_fV_(knFWA0l)W3II>E)}7I4$3{!UFCm7q=< zPG#M;#ZIQw3VPxl=<$))k!9o(bxpOR;C?BiREUA*_nN-e3ks_oD*3$Tct#M4io_TP z++ZD|U?>X<*z~9&ylFX%4QtjZGWb39cz>-x0pLx=sH}_oPA_G$;(WC4APv{+xSC%{ z*;1b$9ZjL40yu;6=l?d<1gDjU2SqXvvSO-!bCELT<@<7$R8X^A_eb=ND~-bMk4bo@ z&5LhY>P;z8)dy}lp3|&{v6-M5=MQ_lJnVuZmf4v+dVW3fo7@jgKE8P|LExtbA zL)kQ55U^9}9a$RM+8pFq5LF`t3_F+o;F7^+yTKbE$_LExh^QHFRzsFoT*n$naV{>^>qdIBlg`@v)|v^g%TBI~@dHx1HZK%) z$B*KBm^s>-wUre6Bftqvv!B$giofZXv%EQI`L1M+1bA4xL!!4wWW$pGjp^|7QMpQ} zqW}hA`b#W2wI5?_-YU+?qBCwp_RW7(FjbK8oOx<*^B!1W=k$UmG5@BYJ-aAGdzrd{ zSS*_siI(%vT9=LX2`m-P%qWKO54lkw`f-zoc2e#2987MC%;!(T(HF41xK4!LMQKs} z?mos+=jG!Qdw#eA#+BZ_dE>w-cQHf^AnSM6M+){gR4VVwbil~gxg{m6HuZjDo+q*Wix=R$!?3_5tn9ySob*8k0ZZ~?AB`mR z0{JmC_l*TU_qDdqw5S~O9slx*aUkt|g2xvPNtvcpyg(v+@&v7oECd%jmkg69U z@Z=6)_D3m6Z#dt~ipf!&26{Wo4L- z2zj@+w~>(;0wOc*zfw|C0GpTPAbEfiKsHFq%>1n1>gTJbHh!_y-+3Xq`|#jxHD~f_ z2o)V+eO2`Fuf+9HBS9lkF1>rj?TnTjpJB*M-reh@2K+smsPwb+Vh4shoGnuFl1f8#DQ$Jkp1>P@J<9X0vs*Q3rLa&>A1`H?rb=$k48;M8gLkFDQM>Xv9h!7&{f$^8=m2I znUGh>&A@~1K2M?Xzw4=Ek5KQ+QqHW2##lBo?05;mX8w-4SXZ;`=qg62o1lG#p zZkR>8-zC{B)RfNMK=Zv#smpNVkG;}yeS^rX5=C8xB*R&iLRr!d^ma= z2D}q_aImK{n=_L;X7yOD7p=|j;Ew!%TQ@CUBMu+Et}il$H5&QPi}=zZRWFn&Q1G`sZg>$7gQf(Y-b^Utu=Mmm}iDai;t>#N?<;&{{ya+ z&LtzroJ5$6ApAp~IN*#+J@Gu(?}dbm0RIym=(VUpcQjo2Om$!RT3!n;9{cxPeQRd_ z;goI`PcekKn&DYQk(Err&@Z7PTKBp0ZcvWkpQsv#PYWbpNxGvHTv4a@;py#gm`t=0 z)uC$h2GZ+W8zzdc6h`8i&b+L)G|9vE@r2>I-1wS&uYM`&x0TNck&&1NdzTi;tU+GR z{xv99*9H)Q)C!oz_zamzpRd6lR(n zjc1of>L}1PA$=zgHj^VfwOoI`f>g1sIus)-?E?%VDU-AomkSw%UPzj!Y4O-S<7S^6{)<*^ENpiNDBIR z7GCXSYzTkdpI&q_MCrtP&_>HU!_V*X-b0rep8f@*)Gdjf=S+Y7a{Ro**|$SKdsLz- z(vMpX^c#e90FNBi*z2e^p-^>pW|~ic-{x?wg)%0mLTTX#-Ms#MdKV6}jVZztNhd*H zk^$OqN{XN@nImGOrjzhai6t49ors6y-a^)VY1WUUB{NNZ3b8R`GYT^z;=LA^mBZtq z;`Z2g|2({>8j+4z2-kR&@E_9yLnsJwfu4I5d`_~kd0G4RgrV`Wp*$Ec4H!=OBw)`Q z^VMX9FDgonkZ6q~Nt;UtZFz-=^3IkL)G|IUw*#D7>So{DD=O>(iD}lQ{RYw%>HKU5 zi+PU+(K!P!7mip0y2=EyOFJ=12awg~3Yy?vdw6&N!vd`HnLw-(1_v2ha&ofo4y#F6 zL<9h{F99!wlnJIWgQ43+O+Yk)mm9#TtbM&d`Phb64kNBGIlFoHtda{~BCBJuPvp)P z^8By8h|1;-+wj+0Eh85tOP30oK2&lQ%M3IGB`lwuNm>TIUeTo1@Gag>!jbJ0R^bFv zn?V&eM3_MSPcVGse7g)GSrGyOQp}GYBIJp*K4y5)a(HK1?Uxne1^;=L2X6EbdE~<& zNT@03@aQP}@t`_gz>Xu^pbgsX|6@%rEe%M1%3ORdA?l?Dm694OjMk2h@{MYTlsw%P zs;Rc#tc6u4Ia@O6P`0GR;tq4OF&2d9IoB>3l-FfRI5y@M7^#gos_4(WuU!75WOZbV z*vxaf88=@ioJ53!C%ymWo4$3!SYJ!m9{_gN;4ciNTgvnc31PuS_#oxC)$<;`=1xi9 zrh3snsPpdCZU0tR9Q zJn}i=E3w}~i_!t1pfH&WV%8rRB-9ynclH{u6Ys$6?PWR&OxMKg`hZhH%x!d;XuO>` zo7~Sp>*^;i<(qT=x5IC`+fjL#mjCr+W|n2P-`<;iy&Otc(N2EnbhYw27lZX}B!j9- z_P!qXOq^z3Pa$*K?CJ`Qjef@FTj3^v3)87kZ0TVtVEQ;Z&Ya0tpCj6DO5v$WMp;{= z`+{we!w=H3HEA}zJGtG*Z2dkOLKhk!px{P_2OmZ7mPx7PZ0oX5JurWOyi0?=_U5wq zdDG?(9Q+ENXM%z`HKG@OO0J)ehV~56mu}6ue)MxRum<8LD2W5`Ugo3e3}_GmOtRLc zAMzw+H8Qw9AY_b{=~W_7u8zo`L!EO}<>isJx>I6y`&au>`2O_f0qcO}uRuin)oM89 zDd2NL{1=j&>E430r1FT+YupylZ0_h-M)`tH3V+OAuJKlyp#P&~1(K@CjLMf#=+6jElm!q#QG6G0udGxLf4)*nRILntYXPZE?OKFE!+v zX~&10UeBm|7PLAkx+8UH*yN1L?FhtOwQ5}A6UU*mxo>T4ZOmePk|pI;qpOIA`QXNT z$kM){MS?%GGt|;242G?Lc&|DGQ4HCFws+pucT3QCnLG4xVlQ%b#WiPZKe}(P=anf~ zkT#DMr>P}HRwgC(@7euZDK}uXu`$Xicb4JjMNp!4>v~5n)QLBby%MZ}U%<7>F_VKY zPCKQ@mhi1ZlMVD z5Di2MIKu#JfyqBOMwX8qf%9P_DEs+xuU-7|C~*XT3XLS!pFo+-z3gvof1zkmXkAap zGgHlkbLI@6v4$s~sYvEuj)kBw#!EcwV#iC|e&vy>z6mU7AeLa53K%02&g9*f=KHfh zV>)veK1~=*lR)~PT)9_}6`IBaPFGFrAxmFQywY#{I)PEV+SWw7d#}0V>3ic*sLZJ|i?ANE zz%%S$y^@7zhqc`P$9f{9h>rREqev{X`5>hM9Knbbf#!2oP4wBqo~VuYCMQNs5?Xo9bzo#Rn8sATO(c-uh3$CZr?EqDhGB}#{>9}MCpc4V)w&kLSUdDDvqCJ zT|cD<20f@!M2a*su{3dw;{LkeVBDENIn<=0&n*6XrEUH2!lvW8b6l6BgBZo^B2U6; zDsyNS?Hs+Mun}PdKK;ed_rNIu5UJh=T1SftHA%=jmqKhjDgzhxb%OltnNfmxc*l1& zwZ$jwp^k)Fy5}%arlvk{{Pk?@X1vIJV7{b&j&Pvl!2iR4gA6k&T)-C{=Ey&sX4D-x!u4rmeD2KK z<2lCg?Go6Oc&Gr9xL;nhR?4i`O`MTYnFS$MicBGctglRE%=`QSo)2*=?twgU+N#`*c{ z*YhNgiRaVy6*Wy4hjIo*pIJLia{#DaARKMhR{Z> zAMIBPwUs(^LQW2%X)KZKm8;Kk;$D%1eq_?*MstQ|t(&*|E@jnXp#~T-irf>Qjm+JD zzc$OQF%?UW`z)_&M_u$P1_EI6m{*0Od7V8XB0M}!3svTLOsdY~l77ghgMQaiPge4b zc_H5dW7I(}C$qSJNPt-wrA7sbGE8ESg@GgEJTroQt=vxHh&6j%?f$eXh1E@Dy6Nwf z97+ATxWWf`f*$l>)Cf66#qd!c1Ssiqh#P<3D>p}55LHGGNoC9SCo`PH0UqPNaU6N- zSShQ+M!mINjGx6+W<|v#Cz?+TB#0ZnRl_HFkS$-a{{V3dg`8*Q#rx#djdCJkm*AxxXQZG292Dn8$wLdM-ZCelj zo~=Jq2zXqU{A%UbV{cLbr;x#OG)p!eQ64>L+j_@e@N~W`JSRMg_@EsHWj1#KW6pgV*ST}$gjvTT9MSnlpELL%W7J7 ze(heEBxz#!>2q=Ev$Ot_C#)Xp{x>!Y8K3Qe^t`0e{ooYeJ@I@x3Ol?_4~qW#r)<>O zNH}2)?ZOMaG@Gwfew6=l&d8EE&z#XWXdnI15?>UR0t!%|4KQ~a+Dv%0do{|9_GirQ z>up!bxX=D_Tf!@L%o@yMH#&sD!OPvz%lbAKw@(KW&r9t-&&_MC7yLf7AX;Sjwj8~t zJyaD}VTnievjdFP@OcSYJ**4ZpD~WU^LLaQS>;Amvs#727HA#cc}zDzM`CKB^=#(0 zl~0(UZ;YU>D9LF%piJf>%;HZktFFEZA~J>>O-M@GjcyKjSYBRU`qGRkzU*xq9u9!D z9b7o)tA7vLU-6FgjjD@f*x?CUYVwI3(<6Z6-eokpdp;Exj4ggqDDgdx9se2Na_Gz@ zqAFn!&n@w)pkvQd%N`687ryuvWn{Fj=AVool^SQl{p&WY1M7p3mRD&avFOE5^Ork; z4&l*vxK)?>1Ab%^BZw9Urn~0tve6e5$J{t*{`Cgi_i>7IO*@9%Y#?<$6nmy8O!@PP zgszq#d9?3iC$@Zxhjr|K7mHNfHG0fZH09hX#qIFAp8;30ZriI~>lGt?GxjaV*RzU_ zu4T2n)}fy*dT#j?CP~CCBV8WpHZ$aUPE*Vj715sm>)fd}0ILDO1FX21MOg^QZf1EC z@AWFY9S}#!Saq!9h!P_t56}Gmz2PjZ?Vii4;=dC(V;vO{fp|^er>@tybmD4n(J`y# zNmGN%8y*$a%ZyT>tEQ#~#9pFaJI3fjJ(1w#&Wqjc?qiY02e#NQbdom{-Gi^DCbT_G|i=@iWIEpI+aIfS+u^jvdKGyZ|4u zKfBm$V=Qv6ZmH%aiFjIa7}{3tx>%2JrAn1Ahb{ljf)HaC;n>-X>Dn015Dt?mBcBJgg%3N%WxbQlR>ok(m$|+nM z*-r=AEmYVy-H4jlwmZ5yI>735rCH`Hdvp*2e)v11tIns(LR@08Jz7Z|Ikg=@emu+^ z9LBZQGXSdF7eywpJplXu;JA6*f6l|f{g@`Ag;-TpAow+VLrIc7=PB{eC5< z;Ib$+N#yifQkg}7#)S1kGK`8MVm7_7!tr#$SaLrqIn=M*g=;xrC4UZ`9q7^qW1_zxTvb9_J?D#)%FK94z zPLaOWSj_!TlByUAc8aqvPZ6H0dI_3n@CcVNz7I!vdA@h2ZYU$mB+SIEmN4_ugp0aP)zgCZnSn(!`C6xcLKWl zO4SgQ#$;np)Wp8I_NVJ~2jrg!Ec?LF_X!}CM~Yykey^dS@$Q|!K3&k#IX5g8>Sk?Y zqn07)0C*nT3O?)#w8N0Fj}t0NxS;nx5%^DU+MgeR*%qSdAY;JJsh`J;rL{r^fxo=1 zjS?q6hv-Qh;4vE;8v#EzJ^~v->aItg+-Roi_O-9$Wdp5T@POMwWXlYVn}Hr#{~f#9 z*P-JbJb_l%^M$`zY3r%M9W6iW%I*cSzj|mNUG(C$rI99M-{%fcT&ur5jTON%yuC6A zY13Vs=^-tNqGn9)Doy*-QPbsx5Rqf%R;wichvMx^;mZK#<>wdwJ8ve#5-^Cio9GkX z2?vbViKqN`*n;^HLh2|?_eMU%Cj_kE)cRt=wm*J3T`n`If7#}Ns5FpHFgQ=pjwCr{ z@xLHsUNc+zK0sZ7I#szgIjSEKpJ>Fe&UOJIf;<-GoC$@= z1yP;_VY(vd?znm!(9L7!09Z20ndJ-aY_nrA>{rOyjdT1Og6JtFt=Hsdwp3{r;>ET* zo(=p+K+h&@# z?No46@`|O=vK{psG?H)tH>`f0ESHWqKbsA6{Bo*a)V=t0$A`!opRdA7T>K#LU)I#j zRWhB#n%jq~%dqEaHa8d7$sRc3SJ}c5;nF87x~yd4tDc8c#iRxV3mbbKcbKqJ?uD?C zlY}--5WbYc%twvH0i^7;22CXCjvg>HJ}?z!w*h|OcK>?}vU)@@F#7`d1d@Zqy`Acv zwpzPC!4B^tB*%j>eci8AXx$%rqlVJ4;N3*CLb58+^K71(Jk@mPkKF}T%Ujuw?-ZHI z$1V0y9DKwdxcy3U5VHrFu=Y8;Znt|*>Bsl3t3*PZ69iKvvNvp$*&zrtk{w>-D{GIf zkWiv3j*g2CVHAGeb^Y}Sj9rR(=6(Z|J~6<2BEY}Qw?Ul@FhfN`NGH~d^`l{CXAe3V zxV_l=-EdRz7f3wKaP6>=vQY#Jedq%msi`_+%>j~4DT$3_V*W>CBgw@ldwU*n@voi! zNH5XSQ$jJuj2<1-$a=gJsuk_6Vh85AcCWtzVr7j|aGMWsB`O(@bE-|)_DIZE?i=cY6Xb}8a`v;MyC zY~s~LT@As`v+>f?pWcG4)uyQ1S}BeFu_o8OvzrGe)ytTwEhVMuGnR3qbnf>J-Zp*KV3z+W>%6Z0Q+h5||0La!0fu%8IfS3&T2^Gao-F-vtK0bK zL`;|y8hCi@N-<9}`kp?&fQRs^NnnobC?nfW^N52-PdT_(ETo*~sU%kSWAh&BrwtZH zn!yY}N+Sop4e}HC4*3RLuis&x@8;+A?1dED=Fm#%F{IYrTS`jMcSS1D{Hr;!4Ip{} zoh3*~=mUN7N*X_sGGI|gADpiV@TfO9FgIU~2A zf*YJ;hC>q-3mRbe;(wSb5Psiz*nY!P$Mt!&3@`iZ<3x;pQ+XhiaJV{pQKx?<>9OBB zx`*6LXM296XZm5gcd$~9pS!byx-o8i-F{wh}5^k)K3h(YqaD5!uRFfkiM=#9hpbA3(r% zP-OJdoRJ~`)toyVX#DYJErw1`3NM`WWnrbEP|%_cX=+9W>`dcJEM#EtQ*jaMyQOJ1 z8i+MWFPigJ{mMN)Rl@%KW8CZ{gfdzoaXQUAWLi5hb(zclw zUvE(LT-!A9#%rh>=uH||ujAa~WY;%kL)pGk$`8sv8-D_?>D7v8Fjxcr1p2nYY$*q1 z*a7&9cJYwq6=LYnEZ<0abyy&|3V|s=V{o#nzWylVxS_1}67I23oS@4b3>Bq7I1Mny z_67;|V&nHY>d8qz5cX*o8>%+66R~{w-T?13j-Y< zIVV1E$+zO5ri7E}u6!8jLgA?MZh0IL3Af$zmykhp>8wmE78W^CwI?23$z-xAH0_*L?jPu}R?6h@AHJz4A4Rh7 z>AAl?h0BGvFBlEIjOP!?!4>|14q;wdwrf#8+kWko3qpm3%B(+xmALGsYZL8S2#i_V ze=OuGt6(!o?BbEB6W|4?pu_%GBF!akFp9&kMH14~R^W5Ig141nhStGmO9pR88H{Ek zXFjzPSWn!YonYw5HrLUXNA5&>Eku$tf{vWp|2PPJ>-HiyFrLvTi<`yXYFZ5SHAcgiV8d>bC0~6L3p$T*ggIn~wr`AFS4FlK!BI+0xtAg1? zq40%^-Ce%M#Jkss;WDW&H%=EA2vZ;A^haSl0ZhNiKhXRXdg{wq8L6d5S@Y=An3A$>p81YG|Ep11P62HVHL4Sj9ihfUV!WjK z@-chV!K5EHY;LRLMBli&AkhCsbUmZa-JWExknJzQ({32`lPe*w>29lnENB~sINJXesr1|nO84)y| ztoP zPqL#r#6`jnsshM~3^iPR-MV`)v-2artqYase(q8Be)~)HSqt9x53*}uHWceO7Guyl zC9od5thIFbxrl!sqY)weN*?%Y8aY5T8}5|UC;Fta#Ln6f$m_1#c~}%94Nu(VcC{>Tfhxx`q|Zg{Hipd*kya|7y3cPMW4jtX)yIvFzgdqGLSPNtTEx&7Arn z%{<$o8Zbp!>cf~ph0UDUk(Up1{fn_0N3@Tw|BBy8^MgK4#}&#iznw<6@}gU+PnBlV z_G;9KvDwA9p>UBc>!EgAk?rlZ4pXQ3=H3`1Ljn?XwhiIK_1py)M~shKy3sc__Yd1)lVNj34#+56!X|?@`aa zR_QTfy%?@Y!gpCgTRTon$G#709(y6mPKbySWE*c)n>|kxw|Y$k-CBO#IEPH@Xrk;2 z6OnPMW?OpQvQAdbC|9L=&j#7Z{#;|0`|mp*G{or#N+tp)%&!7Cx4r=3V}$-@UZRpK z*l3fxYKGCDSWTA}hjWF`E4rSUB`Dg))}k0-X0+<2&7v8C=U*M~5l*wrv|#l}hM9n% zAW$-&y25^=3sA%Yc?mQ~Ux5-WgiQwN8)vK$t9QwWH1YD+8-^cuZAD*g5Q#iNW#M_+p zH+939w<+EIzvdPR_oE*yOtugJZT?2TN=}Z5RDXsrm=!_8S6RGt|J~lp8d4+3o`b;4 z=f#!r&;$aJJW2;PDoIVQ;S`ex_vBWK;R_QN;WTLh4@^m&;!O#Js=g75dZ*KolJ_qpjy-bu-f!Lx{ z-YKzu|I5?1C#d_*pGPcn8FDQZDUOSftKg@k%sn4e>OD7GpBW-PLu%uq!jmMO`=Q?B zO{079o?3g=v|)VGGDBqg0DVE=?r6?Z^PfB6opp}>kYB^!LWo;_O81}cb0wRaZn7~X z^dhnT_~GXG>HRjC_YbzjHL>YSx5U@u{?6@X!_k6e#cpin(X&?Lx(yQ&_?idwg*YHI za;tS8H|Lr&YL~`aHtk(-`*~>BPgm#Mxq+LJvA|b+=f6%u5{%(5Il1xuQwY|Ppx;zFSF>Yt(-!X! z^TX{Xf?+-bYLq(UgSam*>SVy6u7%46UZPI+*1PE{Q=*Y0!_*K9dF+);H9anyz5?K8 zC^1|Jpp?3Au97a0wulXEA*k4Lv-)Y%-@P)IR$V9-%G~Nmrm=`fKXZ$_<5YRSO{=+8 z9S*|?U z1{N>r{u9LG3v)0t{>T!9^7VO(8s0GtdWcabRQkHUZYAdDy%TinslHCxg8=M?ODebS zsWjzrbH`OdxmEg!b+~K-6=1=@TjgJbJ-Fcwf~d zSE(<;Y1)79NCRYGc-WxQwD(1E77DkW{TX7vPe=n%0TV~4J#4LGrdWDpnS(NkBAsU0 z3dstaU;PX5tKGHk+L0GxHzC5PSIQ_T5)BDz+j0CMRu-&H@SOI2@@11NgXaGzf9loe z;0d{3=bMWY%aM$15S-sbgTLs!Xewtp$I+EzI-*2ESV>C^WnOV8qoN5d-R2O<(Iq^J+w zG)57e!8>p=_5ffENVHV4L_LNQU!h4eE~f~Kh+tC)Z`%t(e$PG_?CF7>-Q3O=s{3~w z%M2SQ71DajDvOH&v9>TUm^)phiH?SLC*wfJCG{`xm!68sDBw8*$R4%z^aL9)Ah!UD z3HVy4h`4?6g#mnjk!InIzL}5^?JYhlNG{Sz->W8V8#P{7opDf&J7xL*de!h>L5ET?(#C{}OxeZh_`o<*d-{5?^n`pq^mL`R2%%Q(1okK|0`IG4LP!G!VyO?G3JYl&9$1)G%qd&Z|i!&W+jQh3XSm z5i3Nun}(&|zb}BkycNhQ8XO!1(xEW9J9$FUAh=!BQi}D_4e}mV@EnJLdBySJVbk@j z0mBU0n@WGGLQ{r2Jwjg;s={oa~7rr%Qgdg`Z#rGRPNL9_NVbar~T`nnElGKOKtI zw=_j%DTyLdn)~JN30rfFD4jRZ)am2mk1`zIrXl%@V_6pT5q-|MK+*Qux4)LsK59OU zNrOdt%AH@u;dXofn>`6Cx#( zr{5hyW+9EZnx_mLnOqVGM5=}>)jQ5!A^v7lbJ6z1s}97f@PG#btwl?pf0-Lv?opMJ z?&6ky8U}yASod@kiNx8yX+IH4!STUkvbgs~^%7;euIOisag^6Kn58BO(~}z-jV0f< z7>u0kjmnfwk!_9%;EX+grF{%l#E@P`24QUV{xKZBQ@FiIFkD)7J-L_tTxx#t8mU3B%x4{F*w0syd52Y+ zHg@KuM~Ud)E-!p8f4AxzL^z;h?R(y%l?$KF@@ZQliSrv1OT5?j4gtJB(&JUnI@GDv zd{QuAKiA~=NLFDH0ddw=R)TJ)%YmqPl8%5<_eVj2KaV69o1>$XlayJG%LMU250VvJ zJq5P1SnRvN1`+&Eo7nM5YF+)5(yvgeeWKU25}%OJKMZ~;u>Dl3w6rvkYpXA+TagA&pALoS z-@ktqi?D5Gj19FcQB)s*6yWmz8LA-;xDO*^At7{2R9SnJkPKC#lf0D~O_gRGP!aBu z7xlIr2-n1CgMB_SQ%#QJ+IU*vqPzp^tK3QZxaAzBtxb_0mX`I^!i1A` zyS`5>b1de$81c#|zozbqD6>0GBExq~KgN)~ zkr={($mN=1nQnl$n1tNE=46QHxzu!G$t&ciL7haAFOjK!HD{D>-5d6c?@d+*b)lMS zyh_*X=(aO(M9z-w(fds8r58NswIxo>+ZtUT*~d&u*IaYzT2LnU*?QvW7cthtSe8aH zD@oy01?V+?!`KaqS0W8AR=bjpc)fZ?<3N=ug)VOv&JU=Y(g_~IE>NZeR}iWf^c{@n zcG%=;a&{KkbRFCG7E!dK6SSAHt#l=QZ<}3-U(jK;OQDLZWdLdSAJaHWoqeVp<&6K|-a6Hy=O5e+6^1iY*c?EU(5;@lm ze}#axONfjCII@U;ytL<^B>nf248FiRFPV{orWrwbNbuT_{4t$^LiDj&E+h2~dIeW| z!A^FOh20Wc1)k(2I$jfKultHWg-pcs;&xY>yf#RokUTvwz6Fy(50A9pn=rdFx#i2x4=Ht!%&_uIb*1osmcL0vKP7C5+jnXlawwE>)6G! z=}$S9-pS8VOi!bzMMr`&`|lWr%^@l8JA>nT8S~BGCRm0!C8JK+eZWh2M&>ZWLHLgS za%e@R(ezRZ*uIz4^{WOX&cezDv54bl;2{JrqK{oaky@}g&T*8n%RBEnm_edwL(Q21 zjU^qObAaxA{f4b)J;$v=FlEdaa2x=lY!SQRX>#$M-?w`5S;YWZXDfi?7$9+GhT8)) zLeK$(!CWLtt!+ENu?e6x*1hZw|Q|<=K!qusF9HZ$w2`a5L9|{WH5^)KbI4@CSW8}#La&cfm$8nmqWiVzznv@ z0ut8n@Z;{4&^FN!QA%>ui}Cu}T1-?l8kKpPkz%}#8N{(%=Yr9%B@0t3wg#1%ee6$J zvaoQbHL3U3qSMo#LC) zLq0UK5%VjBNOxW=fO)=6K$ z^?l5nXZok6@2&#e6>UtcMq|X`Z;naMj%dF+n8`1JRd++9<{qVGO9E!iWWJWi=>JapPRG`G28|xKFt3X z{R`(uI`iP@7L+luHj7wmkU98Klo_&dv5i_v&x*x1+bHSwZzGMBukHLam9z5r@tzaX zk2fG0*ns8UkRiJ}JnR;wk|{zyZVOq;twGPg#C{j5opTB-Nrx@CBXkd7f`k8$sj~{J z@{77X9n#(1($d`$5+dE*-J50u(hbs$bfa{)NO!k%cf)!4f9HG7x$?rt3-@~0T64`g z#&0yLrq-@{pS7>?xhfz?l0nt1uLG&4HV*CXmor0vC}VfP90hljDn7aHi*!E0uis9= zMz7;3ue8X6F5w5^fE)l83kkPEht2olIyfn~bM>_S(DJ0it8`Y_M6kMa?EYsJpN*2; z!)%%mUNO_h-S1uFMb`)YBV^TCU7yUgbgND16bH)|Kph**v6%1eFjOd9MuTqyd)NRu zx22(-k!+=KMt`oAUo6o;z=|Nf9((R-U>K``_q4#XofT70s)A&jL)Qxd6GYxcNa*w! zL)H*h!->PJ`|X1PF9zzOVQcSCGs0)A_ro%PD?1c_;CBGM5|#)iC=I+9P|S6I#_#|7 z3}hg9JzSclhIbLe&UWk&If86daWpZLWi({nnEDkEJMXvwf15W&*kDJx;G6nJVg+-e?^2Mr@+!hWl9c! zo`Ez8ZT%zGL{btCOO7q+W*HnWZkr9@QSjPORR@}SbrH>S@s<$5xJGA`z}O^S>|ECy zbwh7;mIjd>YU~UqNffPc0B@D!Xcn|Aa$4b2Ft}*tShtym5BIaHq|`JsPc1Ec(bJ#j zE$cXG2)}8w#NMOfqK*2}HO`%v{+pgSxg2@`;Op z1n(MA`?a=aANz+u6r^N76Dw;}%90{uaWb=XI0+IaP`XV_YF2_3sKdRuVF@*q_9GyvH1A`SlNM{X+j zGQUfrG#YIqnry&FP*NiO*5N3;oX6uuELXi3y;*%DjgktQ4IHk2B^H^ zMaj8QPnRlSw|_n`>I#S?DIKenvs&t%L~VU~*4!#YKQxG4d+mA*Zdn&LBqXH}Wp%kxl~k%LYHxT1J1KzkzN_;Hv;96g~r+GEY85&_;Z;cNw5LBEqQRrw0JQ-whWST_e6kQU<`gK z@G2HnTsUolMiBcW0Q~Vu0`}eldJFs-e`Dyl5WYR)yuc^mGs#Fwv)ExuRQ-YxAVoQxP@`9(b zN5lV@GYSo>2OK@s9k++m27X?r>CLR!{_Mhh!jIU<4W9=-ve2Qt(T-#~?qPYjH@LqZ zwstOYdQS8BKaBPh;kJ68*&aa_=4o!JejZ$-+nv)ZyefOEJ6owely!+Tg2uL`2%7ZU zD*gLg$h(=;JExGeEr3ICcl0!CZjFy~SesSDbG#xkS@v97l~B)4)}%$se4VOP#rZkGTe6Fm+|X%Wej{ zUnvA{ z`!H?I%X>dom5N0DmcVmN!#03fIvbzjNIqA+R{q zYVUn0QFKG8@Epz{)~t_!agFRjm(NqP9VB;RIej9eY2N+vEooZNQ8&mV({#tg_&$?e z{DQi}HGomeHiCt~ zd%+Ok=Ot`IHP-M2=$!T#N{gnElg+N8H`ha(K}G!08_G!*~dlmXXnob2gvNN zzF^?tOCv{nACuGZevEC$kjt;KK(DQj;l9ALrW9HD7~JUV01FlceMkI!28dzt5gaRW zYgyWqczeU>&MD%9n_y{_s28T*+RprEf2(&(gE(Ux*s{~e8%iw)N+&{hUoB-1=B50a z^cA{Jd4I5ESmKj@d-vXpcLhOIq#6zD>gYrgXKM5`>$@9@ zt^df()27Cdb~A0YkA}Q4Ij}~t^&)ZE{I;~#Ywhm2baB;k?RPW4NoTLIXc;8Tk#4)b2Px8)~L>t zq(v5Ps}++%%bG66I5VbbI-PEnhINfT%J99mJZU~JWLr9pl}B(ru8xF%e!aXNx_Avc z5}vx9Y#Cp*eB7_}R_#cqbk_K?LDRa@&7R+<9V*d!k;zHppw~vTJt|=!tmOjL9TL8c zxZS8RmpX&L3+C(0QoAM(F;7h>NgqD7j;;TR`ilero&l*lp%BRBMvAZ7WN-Lo@%|0_ zL?1kKZ=f^sR7_>%< zTZH{Q4Q%whZ4@$AH-m?RXIB2TT`baY3>KO8XZY*o)e@i8Ax$84bQ)FnB4rK_9TFy) z9iPZcbI>{JdZRG*ai|kz;vvS^+uqcrQ}XdUVj~R-II%FYJ*x=6>0)o`Fscve3S~9; zms1Ip+wXkPqx5J6cgxlzfHA}&IF;N<6SL@&)WRYiG*HjnN$*uVy<|@;XIrDIJ13*} zo*nuwgtCK%iQyATSh$lFMlpwrWPhccn?x>N+wQic?IPsKj=l%=LbxvrwV+sZj!h%z z{g0B%2k@%QV#mXv4a|P*6PR^Ro1}qsCk2m_c`7pf)jU6M(^waIebKX6cVQ6fyFh-k z!4gGc`OYw^*x_)R0Wd)CjRuD&UHdmJBWo!FT>}J=AWKC5b1?^^gPt!2#m>ny$-lvb zE97HBb&!7r5)BJ2IAJB;_gW60p-5BmB zopTGN2U&J^_KW1dc-|gKOmA_)T!M8~4oCR2-i(h6S;hG4t68AqrUW4XN^?**vpMQ)1Zdi(E;(U{PKS8JVq%YQs#k1M3^-A3oOsK8LyLI+YEa`>3H9@;H8k zVi$Gs8(pJzmWaxpr15=C_om!j67=-A!^KF6=8kE6^0tEQo;?pW%Ig?mqBCTPv0dyG z@$dFMw)V?h!`DQ2Qf1%HFx5bKct~pD?mSm#!WcB|klqVJ?v!|5zwHr&pZza`-qQZ7 z2e!v31P1~ks?M4SU5TOjpT5wp)jJKmDTI?itK1uL&FeHTFWUbqG&Bue$F6@LBr1n@ z0bSGG(Tqj_Fi_OJwuAO7Z^G$W?047|pk0{xka#{v#C!XI)e1CRCl>k5@8FTEBG#*Fu(d znFNj8_1&ETO?f_}qO=P5|Nm>rlRIMoP~^on&#TBigD71b$}g9!ulJF6cx(QH1$Xs4 znMWTpN4mFQ)tS=g2vf90JlT7i1CQfZvLo>a7!XhhOt52On_VeQu(Gr)^j>$(@(=wgr*qK(pjWepCfMhIvRIr$o-hZK)~w~LCsAKM_gPbEqxHmt)s60LGCC-{VFM>22kleI17|jQG0yO_ z>-!u_wS4q3a&#a*OXCYihqIIOyY)*pzT{p-x-b{URvKxgQlbT#%6qJl7V;%{;TVF>{3(CN2V-m#2^Bd_s8jG^pAJzv#uYDyi@)%_ z@-deRP#E93aG0Ykt(0n2E!1x>;o`DXxY6FmDXh{yvz3aoUoAYW3sS~DKGutzF_Zt} zwU~{O=TkG*)(|;uqdMf0+5h3&AVR)Lt)pz3j~7c$8uXowk23H6Ok z-{p|3Up_&{aq>4YHYR@#$fsDZ{B+#|LeU&hO`90*9Y~X7TmOFH=V#B&($Q$%lK}1W zU7mi2obsY(4P!ggy0LG$$!O3XO!L*~u!0WPZc*>1$3W@XTTi;xa;xx;ye9QIS;c6J zdF1bq5=+i`^mSjtCj6??(NSD0llVv&uO|mz05$MzML;HlOdR(LttPK$+JllI|}U$2A+CMuxoz2_B|IT7qfJ@{(I^=$b!hBBDE35 zm8CeX{7x+|4iqjMh|N{mF1)k&k@nFiHx_;S8_ue-4mQS~im~C-p4qj12SbZ3mOCWw znWOU+>@Czrz3Ld$qsCmf93;D@Sz9b>+^^~RZlXP@I=4j3C2pBS)C2&#}uMX+H!iPm5 zXY`M8;&fB6DK~p~{6?_dpE-xP?|^PBIDi#kF|m}fk_UGn^Cry9s9qz@!CdfO;Far{ zd@`T=NiSz&Vq#%I2a5Uk?;kLjdQA|A%@|FNi!*#qaP^R=2544ytmb9Jp}op_dMGCq z>Ne`?$r)}(ta<&$vPyto<KWAkr;c+ydT-Lq- zM@5Z+1`;O?n>HT)P068eI)fgICVaAueN8D6r*=@Kzh>i>?WH}d>cPDgs1}v{+Nrl8 zyhG;1aULnB+zSNQyR5GNe9kfc>RQC2N6M&3oj#kOwKBEmg+DTk^f*gTd!kSfR?}#n z&=bHmOzU99l38T2l&+rN_bo6v>$WipRniIQL`U8vzj+A(Kub}CR0zm6fvv5J*rBm` zE;Ug@(oj8MihinC)U5AduezN6JDIk!S>MXulvuCk*y(NB2@D~8ei(uxr|?2_XR?60 zMYbjensi2%G0=x7g~pP=-#V9Nu$80DNgtpq{>5-$t6e0vr)h#;uF$9%Wt{ZhOUixQ z&Apr3l(|OlwgX8_l9eE+kF4YeAprdqd0k!NusFcp4U9~io6#LQMyUxmO1mVW8VrG0 zfm|LSnHfmoq<9F}0K-}AlG!cG%S zD9LyLB`_^jNawTyMDxIabpmOAV^6*u@4n|{9;h3`*-f)uZM- zFrb<{bSdwsqDw2}h0&PvTb$rFSgVs|)XQ$T_AJxE8hnYJYQd`nq2I~2G9+(k&-~OK zc#+YSkFs8#twV}$Ly4W}_OB^_vqaVnD4X5|=|JKw{H|WEH&0xLrMS2ehXNBjfS9lq zfg{N-48}m_k*)~Y;0wzx*p8V_W*V*~Hg-6tymhZ7V&wO>WyXU^xiVj)pTW;y5zy-2 z;e!megW%#`oLWkM8d$Cvxl~LGqB8@}SL73$oG|>cOH-cY>VUbG*oFA<^Zh>v#t5*H z9tiYT_i6Q{&0yd}HAdP9zer}+Bt~fnK*lFuEZolNzW(!>_7%Lx&JyK;V1q9Bd4p`b z&~)Lb>l^$ne}Wj>Hd)Jlp?}mT9>RT)J_9t8bThE|*u>wDmqUb2&YsaOBK(HcZvW%e zRy^;E_nhXXdK)0~&5+?!j}S^VpCkUcj62Z!K@)j)1ZGSNZKLDk1~rC|$Kr6d_`e5% zn5%W5OAn-dPf!4}&ik+tNWB)61#!zy7YjiZWfc`Zz~~ZCZC_;}V$m8(+BXQG1j1`~ z0$4L3lY|_UI}fOGjD}3P+u8JUkuN0MA9geFxz58>05cu8ownsyl12SC+3t+!;Xk+d zTXQ3IZNwc*nkEB|vO-nvBd^rmyS|Q}Y6hI>|Cr0R6yBfxbyoKUZTX*-`Q()rOYaLP zBUhYEsv)_!2hT>ur}9ZL&MA0v3wYMmy0D4M&UMjv3@lPki8}_f<(mID5PJXBZ4fxQ zI16_s9E@dhLT8R=vd6NMZ`PfcFqin3yH2B8LDUVEs9;mq6p8<(BE99b-vuNV@J_$k zw9q*ubAQ|mi-Ulq+%v6{9e|TQcH!S(6DB_0>pa^Q$f@NsITk?1KToq|qm`zp!-+!> z*UJ63CXXZGo{W33 ztnS|6NI4z3!P4V$?Kb4G1;Gx(KPMN2X4|fj8|)h%jx;R4Q)x+|jo;*<-g`tT0hh_I;zO#y= z@6y^a+y5vq;t-HoVZL1;-aqtAQ$zm}A8`F=qF$ldJrW~yyRgY6!lxbHB_lz;=b;?B z3;P|vYig&gysBp%Z-9M*j*Ze0*%9v8tk@u_n?!`l%F0Ts(@)5p1%C%P$@E+&rB($O zfR8i=a69}dlP2{H_7uA4!_sN6P5Wf_A$V>g4Lfct5lGIm2WH+x+_qgl7J!3GL)QtG zWvN9x(<@BK@Bm3sN^%T*QUv^# z_(Vck39OgokYu!EpPKu$9do12-6LEKqn;^FJmW8Ty}r*LdYLlguN8kiAJblCY{YF3 zBi$R#r~aVcy5#t1Mv*X@kK?B3QjYOTL=pb5CfC-&7{6=HxJ@-Frnv(#x4$+Ridn7^ zYz8LOuv!v8*4)`_4Jh0VQqg#GM{C^}Z$En~G&vs^WbnqXrVrZu4#@lf6EPaAkzRkpL0Q)8SR*qCpSnl}|J3KE z+(77V#ijMr7h|?mdbVnt>h^|?j@QBzeL~!;z*QrC3m}r~D+h-&$Z48Pfv;JAj4Lqo zi0~S_|3*f9e&ypD6@S?)Z`<`rWHfX3-Oe_du<`SBb>*9XxrbesKS*w|#8>%2C>&zb z{XAQ^eZ_kq(;$oY#k*{(x!-Kg(#nfE8v*a0rZtf(xGw*{m3jXisG%0gCu%x3^VWqQ zdW6QCahW6Z+zXp7L`K+6B-PXeVnr7@g)k2{BCXE!#eUAk3S-`_vv%AtADtCtNJ_!} zn!mb|a4}c2T_sNUt2nqZ68uTz^=r$DA;)mzHk!sML8`!1Tv@!kXDS$%nZZ7dW185w zNSSI=?k}Gl3>-2fHw>BgF?&lZaB4dFau~k+jiH9Z1j}Q{G>Yu^k~GVQvO^8`5}=2? z4^H7oN6ufQ*ir!kt$5o{gLtvf?>bOk5WQJ|-Z9%{j4#E&@8P!<{k5xyXyQ9WHH65% z3kYuKS?gMZB_e0~(o)W`v6qT>MoY`(u+}!r(u(YFb8fd98|b-tdJJUQ1Rr?5+{`;2 z&n+$SbMEzX8JA3ij+7H|S3NzTJ}zF9L5U=bktyY%k;5aBvR#%zuFCIw$*Z-4w2vfa zcB4B@KkuX+-F9409iG{im!l1S=i#9^#xcJLJ}=3y`L__-DE1M;D#N8G2vsH3+nOl|w84R2mS48QeADtq7dVv`fqfUJa_{&R3` zN)H`rA5MyG2|+L)B%d1X9`N-KE0(Y{Rz=9P&F&$VMbNoPdFPyaUdCoR_P06V`PQ6l z^6ILD6E`TzCbM&QE;6oqy8#(!jQY zA-ZV{6o)?fCYGJ47?J>XY%*8O@qmuXn7UzUNn5?GuY*dLVv2~6{W=l^?p(dRbS~xQ zS~+9n^8?n$P=>-0P$ca*+_e*WZ<7qSrwTs8K+4p4pVl&g0I=r6ENPbOk4LE!#CY#pCf)KJzVD~h%=cOrrzd7sOUAA)T?i; zUOgc7nheTZW%+fA)RZS*O7ihysD)-Ay$y_sA>AlA+$vP_NX! zS@S-tH0aFc#71gEO;UQyfc8W94F7`Ngo{9G8^PY0#cuK$ciX7vtH3!wFwpb_xSzF~ zh0`r9{2&r+7v3K+2yhZ5afTu=jWFFVe;JC2Jyx86caNm23@h#Qf7iJoJ$a4W>Zz~J zk(*@1Kit%iR&RFR2~0OW`tUACsQ?sC9h)@0I)Eztx?J=3eF3lp zZ&6N_q0t57L5b@s6J*I`5nI_>ZnXZbgu>5;gIEssk2wPkbEzx{H|gbYLg)PHMha}> zjANOvJ0^4n0+m~w{j;%cUwG&KDZfA(Ogj2n4mqJ=LOC+~zh33!Nz{?6mUsiTzdJSP z&tTpO@J>i!(yH^ivDQqUN&mIY#P=PK3pF+M2cyO(%&=r6);6~3b5~b6X;pvV9{~@8 zq?X2pU9KT;Jgfe8ZZ1CanjT`L$DEdh(aN+_<|k?3SODrvN~jEtQ_$NeGX{zp@fT@0 zH3F)kYx?l}z(m^U9Chw}7^U9V5S`8^?^?h1fZ#akH;S*yMja0W4cm(N{-bYrXvx3( zdm^f0SDRLr)v2&!!4ZKvlhAUe-()QZl$k@v(PAqTjuqeZP!{VX4a3OeF2~JdL#q=u z2q;rNqe@UtU_j1QmX5qGmn*nYU;&HYRr49bcL#@2eFc!g1V2zrVPZtl%BKdz<7Qhu zy6f}3Sst3SJ>tFy+oD{NXh$9?t2GJPPrVo$bJ)*dJZA~-6`?hXy!zb$DYENd$rPLP zn53Q@KL)K}&nPlce0Vl~+`oL-?(loQc8D3vSy2L!W|k3zHnl zHCem>+z-dRWGOPqfuntb_wot2-`}S%o``B18Wu}T4Uz!d%0#AyYhKp{WtU4)yX%C= zbV@}@sNQz*6TgB#2VbdTnv%#%QdM3-%@f6bey<-1H%p`}PzYgy^(k`^B#1cb&BhK| zL}VOsQsO7nxG$rhf3P;MrrumfE+kK;l0BXzUPQ9~aPs&B#XQ{e%bXy^x8ruPb++7o zN?t?OCd=V%x9I!T@5ut!Y7IuFsn0(o2cx5te_*<@46~SWy??>?wBCm)qfQj7ra8|<~*nlz5XOrX2Grgs(uY)6}bEc*@2((?Lvhz3WPWPt_l7Ouu(I_>pAu4tcC;+~iNvJT=fpTms}N<%+r} z8Ns4MZgz^4(Dz@@7~RsM!(==|UiLgHRYz3OHI*W+DvF*N6tdW#*!u$k84jwUm1Jct zH%VRw>JRBkz1TUi`+^bgOqzhlX_M1-nApT{-p+G@6%2;pZRLgwyoq*3QD}QrZn=nF z#N=wEgz4sBKUTHFh5=;SWO@1 zk^r74oW(R+m4RvRZ+|8jKBOXm%&#$5I8RDWE~0G0bPClJuAyS>dBG9V|H}A% z9vJ!QiSPN0hzhBQQhO>+^+-e04!$1i=b_d9`{HUq)aH}%rELxc#LOVe)Ev_tv5Way zwo*E322ME>k54k+YaSI?MraH7Q)m&Z3MM!n46(@!l;yQCE;LtwmV3PYev`!FqYE`u zFDlqo=nvD%Pqfwt)PDipD&qWc0mB_|5RjIJP#VSjTabZbXznfVy6tY*fmpIma?gzx zJK&YPu3HX!U<&r4*LSj$TL*%Rel&qoaubU)+Mh4G7d%*D?7(^clD z^@p0r4k3lSV3u<}86XC$B|F>?s^VcDZu##f z%w-e&q7&p295e{5l#YXDa;rf?}=G%PJ{-^C<+TPuVSv?cI z+k-lDJD2YjzB3-=@veZHGi9QRh8|7MNVE6J&PWI|6klf~f0#yyFXFYN)2(orNu1sn zm`7a5hGGDt#NQAe!*%hK@7(kX;46)#alp^p0p1O2a0utJB}ut!PU6+d6@r?Nitfj@ zNPA0mA(+8}kI-NVqg>5kF~eu`8KPQQ@Z)nR~3n-`VKWot|Q^TO_)iqgvsx% zoy<@7Oi`$$3*>GHA$v}dlb*QM9F@$?lAdy+3P7W){d0sc=|a!(XdwJ!rRcvq^U%PYLdY-F8OD!>u>}mN=`w%1;L- zci=sC=eLU+$yJhP$@*kJgII7?l=3f{Oat%$T@dru1ZAPEbShWk1bvJ>+;iCQ`XRK# z0@cKOgd!#m;&~QFf8GW7N_PG=0^H{%)Lr$Nevbz~e#Di|ffcu)(F6n5!w(Ow1&T$4 zYwdby5B^Za_9YdIET!1V{ddo0isgDLRpN8dS`={z(ta|JDtA{hyen$o%PDQBZ(i=P zramrp_`9sESLbfmoqiKv_V+yag4xB~re7*3my}CTG^rp70oeg&VN|OPW?|i}IE47#)Wt5nl^S=i)SCs=1X^W%w_g!d8xAYkMC{X7qE8~jx zR<^!1yIPI*Bw}8UjV?eHr02QwLz>u0vL&?9{&T{bRExVCk6OrlDTcziW2q{h-_pv3 zByUydY7yNKBXY@?xT&7f=S3a`d8F_4pN}=f<9(mBxn~<}mlp5cO+Z3UmX05AdJA>L z`(*J`u{F7QxIIng>kr5l(yxOIMKu52vmrYp#ajD5BE0n;b&D9Sl&CO%*j2|K{WwNM z=86aVTo9OjOPW8GarpL|G(6PM!^eJ@Dp`;k$~O>VY_X0O6t)?n-}QyH{TlGsI0F}@ z@2Bi4LQjC1rtkl9!|HbtiETInG?>|vzIE0K+{&P%I|<#5I{0zW893?4H=CNj><5^s$P!;-q0c6M znxToH2Vy3=!ux!G$o|&ZNxi{v*@_f|8>#a1JZC&wTJP&YfRzE!*&HPwr@#CgFNqNd5I>A^^cPyce<^@a75{*Ha}Cs`4aY`EhmDO zlC=dz&wjp;-_xo@A6X=-jw(-MN{LP^HhqZeOx+P2XORnD|6$cn7RquJ8RqA*scTL0 zdpCHV2exRIc($E3#0kul;b3};O%i9Sz!85E5#xZ#;Q+QXO8(##Q$84DliR|%k9rQ(WA_WHUew7r8OJ_ za&(Y8L1yffC1$ZXGiB@w%=WbDS44+QDyAA!WR+GADQX22^HEk=4eKVm18M^LFT+zV zon9-ysc6gk_>u;JFJGz6L}4|+3T7>-%I^juqu?Rg=RwHYm-_ls-*qphQZs55bJ-+TfLi3g*Jr29xf zByF`uO&ng=`%%aq-NI1fntcSn*)UN7b(=_Cb( zbso2fpzmKNhxCAlH_7dxsupTi(>Sd#^*;zOA3LfZ#1^gn-mYl{v(l!>XBoz%a65Ae zVNRyAA32qGD1I2Ku*D$%`IUO|P(o9=;gI{j0j)A)T%L@y0%Nl6*^nWs9BWx@2mpG2 z*Fm7y%%i$Xf4e4x=rY5@1QUyP`PrXy4}L@8w|ISi0CYgtfSG`Oy_>@*N1p9vGAWVK zSN7c1GY2#S;Xf%^o@W#B#f=fn68RTP5o2Q8OLf}V^EDypR6BDnX@?prjWSM#+EF;! zsQwWgJ#mz z&ujX=OxFh6LxlY*jzLW0ld?>o*GHYfrcVWGqu>5i?kM;4+ej{^m-?ZW0Z;&K1C7$=r4ZI{|=mijk2g+@h0nc6-2 z&aTg8*qP1xT8Tm`9sy4;JrKXyay9$*G-$??LxPOD?LGT&?Nas+@~=qoE;St$aClya zbpZ&`-Y?CA#@M;|->X{=KmNZ(kkLpSAJUtF?E30Xfq~D5APW6!=&P0%pHUACh(xy* z!_wk#rnI6@`T0nK*!gvm8Zn#n6t}aL+{m&ePior@(g8zw#jAu|$s;7dh`J}-94Gdv zjBb3L=h(*L<&R)^7X{2hZU5*}-r1*FSSVoD%f-djrqIfpqvrlwjq?o0?uK&2APC_M zkGWNf{v)LmMl_kB$5tgJ>3~6^)X6#Gv`OE>Esmgh`1t1ShbRULHy>Es?!**Uxy!y1 zr+iQZebz_0;&4$Tb*d+M8r?<{t{unC?qnw^DcDnkCv_|or2w5q`_U2w$wVNOES7*{ zmm_iL7U>`5MGk&1kn{j#GCc$@fe&g$C_o#6_rV)t=UZK$mS{m=3&l`-)zeJ?0tJE* zo9JRpeyCTI&cDy22s9xPXWdoLS3M?1n~i3}xR!>y_E>_WqoeZ4K{0W(rNhL4pBrH_ z%#Jse%IVV!>A*)NG)Ne2!e#d}%P3+l39f5RF{k~7NS-#IyP}Ro%d#LKUAJ{}{!_J^vuZ(DlHf)Lo6(&rE|VXK^WHul=YP4@R-8*SfYlQ+Q#Ed_s2({o!Atd4jw2eO8<(I z^p68dcv!JGSh2V?SiuP|z>E+WD*}&CK59^|8p4ssvR&g1MKsuHM9*Ut=v383;fgpS zcRq|dc*&kS-iaR@@xXc1xu0k^u^=e>efla8^>=gBM1_1p5Xf+R?*rqq=!UhR+?Dkn zqDDF;24M!d#$c-5$G>i(`R~h|1AGAg`_0hd6jpe51X9uc3o-B{rb{2}Iq=x^1!``w zTi!McV#TkxM=mhww+)ggO8GU|E=FAOmT7D}Y)8R6tUiJ!o55{87?P7X{?|?ip=4>= zQ`R3JChBvlK|?mFA`a_X7Wm-foE%@jS<|5Z>VU4#5EvdnXu#NHE}NPc10LU*Vsg@d zcOLoP?7!BW4#D>~pJ^~eHpkZjQzFPW0s^&>-;Thjg1S4go7st+I~F6lPX}A=qD%~y z%f-_bEUIgxfZw8X@hH@{bkOu;S&A!-8DpOc#{G09Jfl?Ddc-=!jY?<1_QIqh;G0Os zF?V+_?!!@aC#o8R%2eoTVS3<<(Yg2La69#W)g&9m9XW*d_I4K`X zqXLV6)sJVelA(ue>oiwwfHqX?ydzhkDYINP^i7Xf$q>G`+aY=mz*4LEkwit>t z#hbQ;ub1Iyt}cgJO)R#343qxNFO6N0|K1mQy+8|8eD7RLX#2P=kYhCOXKe?e32%fn z-!9o0dHrOFmAEFxwvj&MFPT({pDU`#^TO)z(My#}AV(b)w%>DMY^+%HT*$Wya}?r= zquP7%cE?knw=UdPXf)F+#t#<%rW>+ZzDCTjy4}N51iXOWtBDDTjc#}{SC$NRpARZK zfFYj1OC_Ev0U=5fW(o&AtK1@5^5?%yV7o(dN9t99wY2u03uG^`2;Q;mEX6V1KPcnK^r z**!1+xx;<1bVtf-+&1@Q`l>FD{T zIUjq71#yKR-&?4dW7VKm^R#UscM@?Lnq~m)7YTT#s^-~R>*TWYaBkK9yG>V~mwcm9 zi!R8RQ&4|)`&Srz>OBD}iX#CGS zf)h#F@AHrJbKm6v?p+g9%NOj&VQCgj=UoX1NYw}7m+NU^0tCUGG(Z8- zi^5}=J~6q9r5ESfgGl!^%5FVYWZ?Ht2UNn~&WG2#PX8sVVvNh5CI!cVz&dS5^K6+P z0{ckbU(bzc#aHpObwUHr%w;eed_}18H++O*uWT8?-LjuWg3$1&Uw-}#R!hXrvHL8`w}lFm~c+HY`^zX9Cc185U2 z3O0>*l+Pqb%W>VXDnXBAn5OD-lePDDhdFT_tEJJY z?aYCRi!}ydEqY=*?4%Ek*7ryi2)^~+BbEy#nH7)E_x7>4Nln=OYziq2&0Z?2Q^h#5 zE|x)Eu87s^of}D3-94hk>3P^+9mhh3{(`;xIE6?HjI~$C8c}*5`!Blw8HB`NJ)zR( zTa_p)w-;fkiAU#`tiVLbMyFN8vzxIoGq?iMi&|Cg75+eS0l zxq=;W+8IFJEvtQHF-7}VKd;{#&29a?-IWB4N@lOzW zV}6e`fb3^lfrnZ9cgfS-IsDC!YM&j0w;O{e)AdZDL7;E=1}}0CIiq}M%zkk~0nYtU z$L3UHA^8+1yEBecjb#0_6A3&d#DT zk}p*1x2N0rhVHa7gp@=)C>Bj-pj&s|)>l(tc}zPGs)T(4-U*ZIZ?7Jb-~NQV4i%Yf z3J)!I_zL_A`V5faJCQh+hPf$bKh+QPHiU3NyilTu00Oc19#fH&ZijvhwOSU%%j10K z?ryEO>`v@RFz%POPSDF-Wo1hQMNFXC733i~DNntn8pzZ8sO9UglEpk*%~JX@D@r~Z zlow;86kHOCKUL7m^_IOYYCn=&vZ6b(xCHW|P#cb&3Bq$31>L$+MZulZbkrdt;`6$Q zE9TlK3-eM$YFeEX*Sp({oB4!yoKpFYSVa%glbxG1*T{ny2yTat?RXB-7Ohki%NRmX`nqPf{?l-n1VH|?=#v-lIH=Qj(AGYe%Q{* zxc{pQw~=Gr)R(|encuUF|v}iw()|QO4fUITy6>&D$=Zb6iZfM`{^C-Ay zO+Fuadf<^#v#uRbP8GQHk5 zc1G{jMp+wm25yd#z1>S+M`cQ94V6SHB3%&$Yf<=&0j)|T>NCY*-XTd0OvnRHfOo>i1u+dpB3P+Nvz_R^!sq8CkOQXW?0=cX%S^YiM-#i7JcsQCSP- z6Z^1f32*tl&LuDCj629-Clfwwj|g@ShdHcH+)qZ_UFB$cu1F!_W()9ilCw1#{#|>o zLI7FjXL7>mcW(b8h3LDsPtQwf6jey;3*|&80=Y|}_Wxn(Ex77x!ew1pxVt+cxVyW% zJAs9}y9IX*7Tn#P;4XpS?h;&rYe??Qx6e6u{{qHhbocD8_o=FLc})eJq-aC&p!jeT z*6~{uA(3U{)eaK_1c@PsAV25DKf6pO8sy%zqX8i+cHLGQL6`7TrmVV zn*nYyY2-?c73NwKHAn`s-^;A?u z-LBFMTy6Rs=Xw*iONhB4%AW#L+hgozvscb>*6Rz?;tYj$jr1SKBMA5qZVBap_?1~=Xx6?nr~R-;G=o%Aj~ z`|$;Qn(Dao9KV{ylc%-zzt7v1sC49aeK`nKg-DceZR0cS&OH-|nk$wIRG;9UU0p*zzp{#YPc(Z(h zdtjMW{XJ}iyvqi1V$HteJil}%gVD}a)HYj|c)dJ8QC*(BDaD7C@CJkTri2mr&;+j` z?lBddlm`kL{F@!3XM@7@z*MYqaq+MlOf7nm0Qsn3jmD~ZLTo7pnNCx~d~-ISo)NeM zv_{nUIG02);6*&W?flJamYgaTv^XK;k6{?s|GL&FDJ7PqTxbOmxf}%}dQf2|KF=y> zO_3A6JDs2C7Ijq@u61HUAUpfe_cbZYgNZZvuL#&b852-IAVaS(2M->!zh`aDce=X6xwC7zC~ym@i3 z`Gh#W25W0Oo1PmEvM(o~@~pEvXm|&}{<=a4XS+L3xcotr;j2 zl$v#$<=Io^jz;NyuM6bAXAmMRygFKtxL3;%r8BgbWOJWy7G}X2$o_3FDJm!u?E7Y^ zV0|k%LK&^Nz)E;0^DEyg#t}bB=u>Wz=rV>|m)&<7YHY3!GozPbHTm_V3((LjmLI*{ zZ$li?*#_I)$kF6*8_Zu|As#+&;BrWghhg#=@vtg^<6QRIUBAXmD)XOShb;KtaQiw5 zHG@5U)I3*^6|FW`26 zHCX{MNMx}?--31E>fr0b)9_*iaE*1VcL%A2_Sp~r0?%u_zfSlFbG-Bs*p@Ic-x=S; z&zEb5b`VIw4p4f7Cf+Lr6;HqKW)`r|x;0x6^4xO>1Qn6xGc~QkR9T$)HaCc8nO1+W z3Zg1HoCKWCmo}(QfvO^M2|6typ0z^B_+Nge+|(ybqFYL`8i8I!fA@0_Ah2oT6GljW zK<93OK#|tyz%N8Vc$a2u$T#v4X`EQ%Ys!gIIgFXV?yBsJ9^NlVgNdR5W ztx>Kr&Onmw`V}f|iQ!Ys`-V_8jJw$C;AE_2fF%~m(VZ6cLG|2a8XgEJuOWVR91SikK|3{$s9|uAb ze9M1kjEx^Ufl3tp`e?a2+S0tZp)o0|Xdf>femYbGUCNs<@?ZdrfN)_Ce~V}%Df~`(*SsI<7z_%o5+e$N z>_;e6iB6|rv!_-8bj~x zEg^Od=nR&c*XqxCkm$S=H5moA*GGf*r?H`5G{K?pJG&j$AJ{pEjtb~32e?r?8ePKF z*@x_S0$*A`fR+hQb|JgrQoh|e!TM!<<2V(|LfD=lhP6bK=r$LK^nQm>e~F-(X`ctO9v;*SP3X8M8uiO}fdIzCID^zB2CEEC~_1%xFhl zeG~X`Rjv#cZoH{yQ+4!bx{~vS4+lxN8&qln(+8AUr!oI>zUZim%XNI1(QuLN62Vn( zKuvz0Q0Xlr^~gCy3-ERS6CJI*&V1w!ZF5-B(XM81$VhM5w)3vYkd`M2A$8%Vd`*7# zFiH-~c2yWj+Qihl5IGk>*qZAUaBeVdBSH(;60aI9j#lxO-o>ZjMLb?SUXPsZM`ZBk z_}$A5aV>Yi#6E#x&oRcs7jds+;rA7Xo|L7LEQd?>%n__{X7-7PLWNHO+o#U=;fE zd_dQaxzzjPB&2s&$3%Nost}Pu`zz^1a8#~YhS(FHDXeLJmn>nN z&HOI|SWz}f=m!v2_$8`sfhzkyuR9~^CtB7M@K43fy1`++ zBK7bdJYBLZiwat?rg^yhe`_5YK4@}4GYWLdSSiFRY!ko4DWBh9VNEdj-k?4@GX^9u zO;_FrPq>%EmlUsGse9@Nu3E^4Sj{qL$~@e875M2;f6cSjbYgQS=^{(SRNV}N z<93&Ls`!*q{nV#xv~cjOpoUFWYYVtuqP`h+&KOkC3)h?aw2Uc6Qjid?77Tn4NQXC1 zgs2B&OE=ZcsN8hlO;1uX3L5-fCAZpQ!BJZi4X295WnCFBN;iUWsnV%$SM6Et`%z?m z5gH)nx09^GesiSDs`XPVrW-RkpC-g73>j0n-r27vSL{#J=yWgD)faNK^>saa*WQEOLGBYsG!*6OewsT-$I1)g5s9lLOzRE{x`i=@sApR#G{6N$A;+zgd({Eknau0N}5ZE!>A5 zZQoGrptpe&xCi;)+3yP??^u)O#06_c8nKoHH2FVQ^G}yFY?J}#uUyIPr9P{K^FLIL zUMz^*{;R+;A9mYkn3}Rii9skL132WoTsTn^leQf^>hPN%DuK5t00_O?jv^liok08N zPULTMbuH{7;wg|6NB*! zV%J$_NDu)^{i>o}>Gu!A{x`A4lJK&JyehksOFny^9~(iA87XTk%K=5o&3R;Pji`#8 zGtFP3i|o_!g+8U?MKyf`I1%_b;DLTVbJ_eSDp^7lB~MYXb&6^L3oTYxR`LtN znTP45owjwpq7An6?F8c7GI>nyI$iN^&@=5RZ%&7Uf9?ngNGXI}t+UUBMrgks&nXyH z3SCle^Lg84jD(lhbKcvRKvL0LDelC@*1lPw!f>rq89(g<`dzblC1U6kp0Pj1+v|O& zP#FO}X^IPFT(5d!xCRWllHV^!wk1LZq`3FFbn)?{m=8A`#XWvcMPljnDN;o|eZ7pG zL?K$vqIQ|Fv{e(o)-UIhRnhacCedo63&a6@7zgJjJgmr1X@9e#N_qW|ZL_M8Sy9~Su~bYx0Y9?KaAVHzR*oBM(5+ z8;>X0RvB;*kl%Tett= z!?SepteO`w&byXqOMg68HO!*p3(=}loW-BZhLGVC9WHPR#J&o9!74IIPoLp?tFmjZJ}@vPcqUA4k4wjbQaRW;=8q1lL73 ziLI*uCATNJ@ z-4$zuuN?O&z?LM|iNkBmt-oU6Btg+wVUxnYk1H9_J z*DNw9c#5M^zqez_P9+I_-7phRP>gQz1@gTu&tOrUEOp1MB(F_4!eqA`S@ry-G>|dK zu)K%uEWvhaYm{77>J^5;M3`>4Od6#;?yjCfsE)NquWeZ*5BiD%Ce_Cws!{FSVTSUC z72DnvgApT}`#4Oi>h;q4oPV*4O;SImQ>pZ`KihiJNsB`xVh&Imxr9%S5ml+rsvg+d zvC;ep4Hlio7)g`mXD+PH54=FdjtGD(Hl10WyA5e|Ojdl2A;TKf)}pp zRs!}qSYBG<^2_>B^!*up+``QJ0hiX^)*)movopZs(}Hlu8QX!HmqBqmIB{NIz|nP0 z?yu`|j#;@bKgZB9EBV4ch8JZ*PGmx&RN9Ccv*-dBh7B&$!|G>-AAW{cpglBrDG!;A zhA-{}2u>{@9q}u$sma#X6ulQ?a7ewx4nmqkyLi&;^f7CzHoYG8%2o!jgtNEJN&Xv( z{QE`$z9h2$p0mH1Cm<5= zG$~w)L2FPdL+tSs8-`Io9In$~NVX?-#jWA`tN=RQys0Tdav&S@Qq%ewkh+_`j06xk zE?EEIrZL6O`8PY#drBTELUV=x1f7WY++*@9su^7nneKhLl!}>CS7h+Kp*Ny(5I<_b zS#?U86~Y^I@)1^=mw5StR(^268Q>}asxtr>Vg09@&$gMppdYQ(AsLvVk?#BgY+z(v zor;iy5ry+~AHbhD#o5-|`{oCW6NTjkDQ3-@uI20@a{<1%1kE0Cp-8Y(iv0cch|2oQ z?&;$zT4RdtZGa_Q(a_S{)EWi1Hy?3ye-K5#0wIgTnnxrE-fvUyVrzkf8znP~s{{@6 zZ8Pq%SHok?YRsDMfvXsLP{}iw^sLl34gV5TV;U!=EJ`kx@l)h8&o#dunLR38%*Y~^ z)C;njptRtN6WnGipA))IB{&2BbLtQww6Y|+yYjWkUGOuE%5FnI7kjsD#o0E*y}ys8{j` zxc@$+QHGbZcd3wnvm|+@Sz?#hf?vQFXH$aS_Z%edW{kLme^L>CF_&JDYt4UHpk5lh zmiO=UJtj)5wXH!S$Zv(5(9pkat9@4_-9|

Uv9&$V?w-Y{ixktD{n)n-PcD5BAvY zt<_$qw_ID-Jq^=ta#Dys)MdP6Mb1eO^4e+A#LZYbyg@YX{lv}SyZrm)>uGd|8p4pZ zqum_x2N4XGKR|B5#;5j~YTu=zror!6->t}ctOqyQg!>k{*p^lbXc*g&?O%4ve$+sI zY^3DK{Ctg^IYHSrjW&`Z^Tl2usm+Zzu*$(f%5Isu>WC=3crdsaDf>yMySv2W^?vrg zk{v%$0@;G?|1I~Qf*^av@}pJdYvsd9$vVn@akRluU3S{D#ues#x5yt@%e}YX84zZmc0!d&C*3w!!@of11+#_B?R5Z3k6xVF zO=vW7wcwP+2}TzgZgb7iHBuxQQj14pNpN-{?)&^jl~(mEm-&*4M%xq2a>)tCK7;A| z>x20|4~P^7mCnNmpMLTaao3)4F6kOIWd|xUWH^Dyr#W7`dRw-4aH$+Jsq0r>CRPj4 zXJDc$NoX2z$QL^cpkpZ@;Mg9Oi0Db_zsARG6X@_d*kuqfqo$8Y#QHNded6KnTG(LXuP@JQ(V{#nGf@Lf~t!FMa{ zSdN=Pgk7|MW%k6zdf74_E}-S3!EugZla;|*wo@cc0uQl>5c`P=ZrcC7-T;&sK4nrb z=0DsI?5zarK3;(gv=ql%t29jyYQY9bEH%&`Fw`JSGHd6ml3$1$`dzlNQnKb8@B@mk~4Z4_Re9O>&imh34M?F)i{>kr0<9 zCS!5SFa@@ur;EybR_OjyRGR1LbQk#OpRv<|tk{vn0tSfK%T{KCb7#;4fx*Y}q0DIx zbmvA$8Sl&A?pEL6d;gK($(Q^BYjUKU2)N==iMcuF{7BI+z_5HkOUKCKoXo?h;hi8% z&DKKxnw;N_xBDE+NYf;BlmiKe^zz#f^se^dEaOzCTK_ zqx*T?opT3+Fdt!t7Z9O0hZmZdWMMtdp5^imP-ifL%vd_-JI*o7 zw}8o@w=@_sK~MSKh}A&N%0OKCTk|U!V)=Io_)^3Ynu5>sD+dj#k9y-Jnrze{AO%H> zKL3GB!7@P;6~R7WzHjhGbi*V$B~LDki4Ii&Ya4#XqvXY+FeT;~!4z-LCqz*(_k#U) z0RaRUY`=>YQ_-wwrT+uRU2%fT%F4E%86*as9mi%-C8E+j`KJ5CSxSlWMCYg7^TLEe zf=+)M;(dev3_POaC2Dyb{X+KpbKU23kJrY#bc_QYaC8il>AK8##^5;)xp$3o-MCj@ z&6SoZt%|817xMpLSNlgIDU^kJZ}F*b9a-vBZ%{|h$_4s#I)2Mkj+axp4*!d+v9jeW zPN7M6V{{+J>K?OxN-7eb+(0sVkw=)4}&%@owUHvsrp`y&^4N0rLPn9H8w_mmPCVsJz0KA z59>hYJ41{ZYQDDIyd3cbeWqgn0xoq_^3}sen??Z%fyj!PH=<6x^60}v36q}oe;UMp ziz1j0ki5b{_^C4ChJ0Y(A z|K3~D@YmBHFy*AR)Wg+gt;^Mu`IpQ zq59+m5|j%V(8S-k6k22WxFhzy29~>as)!1jNXM6YK8;*7^MZisP*)*C4tCq5$w3Gi zxMqYzo1!hA1 z^8{RlcA+OIlw(|Wbu1t`lXX<5^k!6LH}#W!qc*dsAD@^PQ@a5VZ1Mwg2pveF4Lu4r z;fsVX47EUew;NQ%mmY|)i2f2|Jz_pWZ%53Zo6c?+qo9V!U60)>dj7n|%!{a@BCsQ; zu9IrbN$g-F3w^uQ35mq+cpRbb;3R<7;Yr{>D(QhH0)osg9br*zFs(muj?cp>m} zXxWM#A?4m9p>UtnN@hfPU7hm~wNR4fJhaWv#}*}GC(lDClxv*!O7zxo!o?1!9Vl;Y%1``7~PylX)lP6rkB9f{Yh_PNeAl{9yP=&;nFr$Pac@z|sFke^k>jIiv09Gut zK4`G$S_Os#v(pVd*=NYGf1NT%!6DPbdrnS{5c=aG#35~`mdvN+CSL5Wwi#8iJm!W> z0&uf+QAVF}&1qmiZwgk{xv^lNO7YaK_j<0SA25IQdPZ2divev1jWIa~pSV`%MOBUB zaIp{~1a^LJ2Ul`CfH!KS^c#{!A{15@I`pCl@glg=Pvykjcvi4$0>+vjN%rpi68ac9 zPn~t)MsDk2eaturn7v6lRm~h;>~?PId?7pg1BNpTGsiyO%49lvBNNvXXv#t=zd0+o z7e;0&WC$H^w^!a=b9lz(p2pa`YzXT*|HhWi(+7p?7gc}Kdpb&~wj|}VIU0)|q-m31wcy9#*TJ*4bhi3#=}hdS2FFgBwbv!P17Ty|ayWrV)Mn!CN@+@|CNO;} zZ9e*HM~U^Nz7FG+gCnks612a5pjdYer=eEq@v+CcW|h)^VNw5mB!J=O7`m(%L2j#4 z9PyB(aybcTld9%ZG+D}sYm<=%)Js~aQ19HT)ZR&7VDj3BWO&=ujt09Yq(Ki!8Dr;n zo-f39divKtS7Fi?{yV{mg8>Iexkv{`7L%}|Pqd}xN3E}c2pJi#S!y9q_khcez3L0S zVoOgA0x9>2hNG1R%40x?(uj3)6xCl-hmaFqeO)TkbO{=6!WYFWe$QX8k+~u?`0)0blEibGa9Hs`tvgLdHQ6#Uxqb;OBdT? z$MuZyNFWnacJU)S3a5Hi@&)E&f{V0AE|%Kb3#oZ?wod@SRn2H#jKZV&b;KQs?=>~C z(+C=u{Xh-B^|q|u*l4*u^dRedmn-4=voG|kx6k{a1?E8C)*wDzQt#5UxpW!Nk=eS^ z<=0lx;^lsO&DF;K)@vd64Qs8xdQrvVHT0M4w=zU5ECe6@^QBiIAY;TR6^+Xz9uN@_ zdR-HpE)*SZe@niY_c*uIbua~rC;DdDE(sa*YzV79UIq0*pzjE%_=*HB;B!dS=cLkB z&Lb*-d6B;xL{^S|xH{5*`a^P7&am}XZ8Uh@_mZJjnu7TvWxJ(pEOt1p@@U5%0TnqfvNKHJ}YtxdLybac(gOBZ z9)SN8(w`;EV{O~3fGH$c0Q~59apuMUTRInlo^K^>_7oSsFZ(F7Xa!$T^m+wUPyo?d zfE4s;?3yEBS;?4Z*!BkCiowzH7v_;{7R^K;;`z~dqI{nt0#71%8O}vqx-}(R+C}-D zE?JvDDB-~mOZe}?R$z>7vLSTt+00#BC)yT7(zY&ZD{rFxhH)7{9?-LHi-g2;iG|k z*U!}L8!Y(A{)Y_4a*2nl4lcn4-}?SM)So=N8`o^<#f_SX-QM1^o93-^g$PI<+bdPo z4do+rEfDOb+mdx28tlAkXmkX<>#eRvZ6 z6~y@icKBVhEavv)nfiNjB~^{uZcSA0ug+>V3a7kU3k9BxKQ$9s@Lm17 zrPp}3D7{(DFtu4pnabKYrMiRL1~I)Dzl<~7M;Lgt>pZ=Mk{TDe*IdK)%lsFkn9l#d zV-(@)B$3}}62Z9brY0*E-qZgpL=pEmozd1C>Sn(Yp_oe35NIdT=w{~FLx=>`a(!f2 zK62VrgOGx==PW~S;4T#=@2zI|5BKEQn1l3vQt0ZfY2AcsH5`n+L4dmfd z1xTJ7g>15Sne+1sxkV7BtmVA2tNcEia%nAane@M?@(0tcb1yP1FLA9i_mU^(>M)eW zF8^6Bcw+KsBi1{=rpnpZEW7yJlJRtc&x>n&-VpooCj1^g zcV+)%%(err?s;Q3$^~*6r`_E|mavHy0vO#Qvn7X6)tdD#d{i!KvEt=I%6(F1!yd-o zuwUW>2HZKKL;bod)ml}^9X-D7toGJC~`VT2SlyBQr(NQpI5_^r8uGY;pK(G?;F zPxtlLC!pkpm*GggAzo?@D(hUGF=e}lud=Y+cx$Gva;uBzr>n27jQTaC6vQJMtVc`5d?t@CU z#fx^Rm5?_&*y?z+r(W=N?k$)vxkx@X4&j;U$v84bSBhzQKtGzmlRK7XM889??@t{n z<&Eg0dL(@kO8P;=Ok3tBHi9;qp61l4$Xux7TSxC*xXC1tk{J+9O_)wk7?t|Zlu*jWX*I;WY|6m z*`W+m?vQe@D+{XsCjA5p`DLkLecsx4qhKkmO*%t+^ej1DQv69uKR}6#)ybAB>=GMS z&3*NUe{mYljelR4O(LJC{+Ij%b|Qy?klAhy;8_W-GFbURSY~_j$GgOQri`unlYNdz z{RFwJ1`o=AFY5nGko-gionr6+Uf~P@tQ3w@BMxL?CwC>u(iEUR-NWoU@I-Z$4nfX$0@ zm<8^cjfke9;joobM?jv`#25$PON;2<)6VC=fDB7n!hFuFz0)EAT0aF!J1G{KaS${W zsf|j~>yU4!1%633Vp{F~Gil=rK4%QwjQ(>&5TsHQK#+)y?(PPMk<_MVss{o+&DNOB z{WPilK&q__wOQMn6az-79CdgnYiESN(CzpKJZ3{)05W;GJ6_oO*A#qfs@#MT;KF0E zEd9DHqbyF3kT>NBucr@eT>S6w*Xo-r#tj5dVGcdp@r(ahK zYYoY}2NcgKkSJFvj`?b$uE*XZhL^Q`t`Jvq*KbT?#*rOq(+rOWbqd2p+CgN-or>qT)TZGm z0;nR6w9^|NHaK4th=pq8!jJ#2P-GD0R<>N<_6JaWbb_>wxSW_t3q_a}IxO3O|cY^Hzc<3`teqaedgNei9AhP*=gii2#u ztF{US9wPZ{!8kD6+5^D7H(%pjrhp_hDeIqR34E}a@O>xY;o%MJo3gVUKpirW0vc|}lvsn}{;w|8Hc;T}>GMoV`*A&z z*i`U$uOhnSByUknHBRdE7(&vrXoqho%{#?) zs8aYer8j)pP+eRZ_TpSs8HX9gsD~xk(P;Xez$4>RN4wv`o^+#2lO+>5=z=6=qnHqA=nE7&SNC<^}2dH7@nT?|{lV&PWSqvF`yTqOeespnS!OK~u4J$?ZpUZ|J)|8=435#MbbBUMm2HfO!ZZ!-a?;^7*Mb!tal z0A)%cYa$=`er`OJ+aKPK4?c7|Ow%z%fKD5v!#%wRsf4<5+lnoj@IbY<*e=m;R1=+j zn-#-9B1VjY9PdSK)3W z5q*eR6P=j(L`lx%?Tb)Q^FP1^EGp!bp0|MI3YnAd_VU_N_!hWJq^r}~;ks$2jyc!# zG26(93V%4^;!M94zNPs>xzWFIwRPK-h_j_K*A zExTX4Jo@lz1xALWTdgKjdp(I%Aq*Rvu3&tzu4TcEdA+TP($BK=)f9{S>WH?QOCYv0 zQQYG=(%QkTPZIo3E^cmjHi+OL5FI)R@LfxzTD17Ww@43h?yLAns zS_f8vRC}Gq<9aB&@tJIUjTy7EbIgu)zgr4OBmuv+{;cabj^PTm`@MM&rvM4(jmNs+9$qNE% zP18sy+j5gT8klU6hl`zG%a}i!trqhkok^XuC+jveVUX`H?_;*K?ABwvMs@S#IOq%p06ho`~`s8n&;mr?w9Dif;M4X}s( z^P5>UM)P>+hl++oo8v^AwE(7iTt|eWaB29}&{LJ~MsOe6d9p5`55jDH0cUbFA<|gr z3+hJ|Alw(94za;7q-~%e5<~qXq^AJ)&v}MDqCbOUv{*`gd zMewD{*Oc`r&}o8hj7)>A@l3@a`+Of)DD||^ZDGvRmT{X-$UH3T?o+4VR+hyT7SO}u zo!0Ndtz2DQHQYbXk*RE3k$UVC!gZD>6cV}7^AvMlcW3CMwHw25Th!S9NCM+$8&Dhl zoN88_a?Hc9y&R^!+Py=BVt}1cM|V@v-rCy$ZZ~adRjRDvySp#*ar2wFf|z{n6K?#w zs6nW#Tk=^OYG`3o|MgLzbW$6-%B$1BKuhXj>7rWX4=>dSx^1P>SmMljVbhQc!T=!+hG2LO*C`X6kj-05i0@tF*haE% z+9&a(nlr(wh2=3X>CN8Ye22r))eg@ ndLr}T9}+eS1$W9NDmHY5tAjw%XI&|bt-)^CgjO+?WRV{MDa-~HNuqZ>Jd@l(e1jEM zdJ=!vV>b|=rf%DLPI~gOWHpHbi|pv@?Z2AgNGE`8ph5@Zq@MA>Xwu72H%SEfk!!Kh zsI>;t|Ce1z1p>xnz64*FEqRi+zcydzjhou>aw$`VYVt2YyF%3eRkT|>oWqrqQzmHX zS8$o^+CMdpT8d>F%DwMLB50X+lJc>@OT>f0S@Q_$V!DFHb|BW4c|JJv!bVM+Vn>V* zZHYwMU8|U~ay40wuisO&MGn;C#ZMp|3pPwB(@Gp!#}TENZ0m5!`1O-P`qllc#N0RY zLjo`5`}w#WUMe(=BFolI;QO{W?-?`IpLhn+<~Z-D@AD&t*+t)LU(dW8-_7rO&n`J; zdROby-PPJQ?1lEChvOyplf3h^1$#X!&of0hKlw&`pGJGjX=)4X}cZcL|raG)r2%HDbxq$_9D2{@scmI zamD(m`DCO9OMae_3aZ_X#qE_aeNL_AdFH-ot2ZWH8aL}-VlwdH&dyN4Cq6#KdGKJ!`!ail=*qS@zLdk7~%~gJK{^A~kb`WsNbUUxsj5HZBFGP718_C*u^KzW0{goRlKquOWbCoK)c%e2hGjbDK@B9E)5XJ=Oz zw-;&oIyKzwI~xag(AcQOP7Z?0Y)-!9JTgtXpy?@5ZU<0HfAcq!;4L=?o$Ul1%-ag%euO&EY5-a>;(>RNGfg)bwT{Uf3YZZ6a+>TM=;Nh z|J^^MP{+K$uWqco`av?ANcb>#(GS6`h;c?HTcE9v7weEmyq{ke^2HUd(6OzSsA>vBtkaEW4}=-wM4#)llo&{~f%}-`armqHZbiX&{}@oe=*} zl>Hw+iCAwZo4AyNu30ckPK&yKBT6ucjEVW-HyYvq5p9jntYA$@_%H~~elnd235{5v zl>H>Q^KOb`=3mk^gLX~O1nRaZ#ggbZF;jSMieR`RY+N!4Tk$?tGbkKHCMhcnA$YJ~ zAn=>Q*3rTdF2#^2Fbeyk;NrLore*>kn=A0t(GkV`#2Y^kG8K{xK`(+-WO+FaHBBw4 zkn)C!@oaH0$U&V5qFVz-s$Ol|$htHkA1Ov}_c0*b9n5^#RBTs6;O^#NH+91|qTvV^ zN&i&Xl-H)e_vmbi+4h{F9+^ulG{M&<2%t4Z-V)yfk_JA5f0GF6{nQie-k;#T^`;~XrYssk zA)y5CyJ+jkIJ22uxLc^vM6_Ex$S}Qmhw#)O+R=#&EW~B2tE(k%xz@CCRrj0yNcAR2 z_T~-7<3n1p8~flUmgpb`#S>l{y}eo186>xA#%%0->A;T}Q& z#bYeztXcKvUw-z-=NJ8fWdo9o18dZVce^hQy}cHriqSj*PH)ofI}G2OsNmO7%7>@! zJ6n04V{S2`WiS*GfA;7)8F7?ZTyERd3uSaUMEcnizOkYo_NHDv=}&mK-e6G-p>`S1_1vH30+ElmJ9PQ~i4*I+UL zgo89lb5mW5#swQgGx~&i#Q-ahG=dQa2j}amUY92mYD7-OK^mxzkklLSQ+*!Tgo`yS z3=uUA)(+D+U#XAmLA@~9{fQ#3FfK&?39Y{8QGe=AbUHW+lm;`VhJhSG`uVpD{?)Dn zp@yip$cW1cEg<-bREZL)k|PJGCafDk$xE0%drH_8Hi+_qCg9HT=13J!3o7>p!V1yW zqj~vf1D%O4CDo!TLp}Fv$#(E`j|FY*1D<`) z5(q7wgyI!7qReR?QuyEoqfJ1!jI;k`099gZKx5a04)cOIujm?D@Mkl&_VTWa@%wXW zDOtlO9gz^b2~1_ETI}t|K{BqZ-C$B`V?wO*PwdT0oNrZhT;*-)%VbVJh^nt~6>nk# z+f7=psq%p!4zC%?!^s1m0RfhZH51t|Ty zQ91@Vh~5^SLI=!bw+PVH-D!4d{r~nR@ZEKf9L{%FTa0GD_uh9NkStXBOToD2?}CfQ zO^@l7T67Cy5~x6*3Dm$m)9Mk@QjdkxHR8u>M}7{SB&y|8?MO z91#ky3^f8%1s;zlO4ywjXIIPel!mgu5rkHfX@zBx+~aY>7gJ z+kFL2qnJO?!qyU<3m%8YfFaw11nDL5mb4b26b`x&qYT|#M^qA7E04z-@`itmb}o7{153KH>YruI!xIKmIX@qIIF~**-ts>|jBe}_qV%#gh<^^Sr-mm_Ej7}I*veYWzkjzQA{ps(+AD#u zQukGQiBR_#0SH=~l8J5L|Gdn)Betai+Lu%i*9{*m6W zSE|-|AwNa&U?Ql?V#oYHA8B$>D0=_oc;BD%zCR~2BI9!wY*^JI5Y%3Re>YCf*Z`+Bj)^YGE^(?1_DUp`0l#TN8MyL}6IxXg$iL#^nUKKo?j1Q7G!4Dx{6o_?9k`ahZ;#8KTL$;rovovZq%$&@)i+u<#%RB1?Z6 zRoVOHQ}=A8BJq|&u-GYiH9)_Vobt+FQWU2;| z<)=FOj8JCM$3;B6&0CO4#=0sS)2I5ryuy#re-={&>(*wWPw-5swU6u8O%E#x7g5N6sWB858`ci zUF@(locZ=dQR9LZ0fvo0v*{?k?kh(KMOiehTbxx_Ni*OQBIF@bjz29c?+5sWBKG>* zDWYO!?U%}D;y@@2X{&iFZu`i|Me2``x+RBc*DtPJ`A^- zMW^*OYmMwMp79!Ofnhy6Nawu5J$f3XH{C4>uxd)dtKfKHE30pH>PaOQ+C2v;(&1yg4%Qb0X7yn)E0 zz*Lu`ks*L-ZHwf;*m`Kt!xgr*SU3`#--JX%hlGb{yGp8n3eNHGBY6j19<08Y%D3mz zF{V3X?Ki6h`8=rrR^~LAggvU6;qqeke60pYpFF&S#cSheDd8b!ed7;_`}o#9EF6Wl z*izb=goefAGR==1@=*TSIU7~~B&;$cFJ*-&ih8YwY*D}OM#(pu8s?vi21+3czb{wB zUgzK|8su2sT3YFN@W*H*_rL-P-(g^#0@MQ~*fj*I&G|!!YMdJv(!IhFhj9;4&@!v` zi5u}?V)SXy`u=7rq)y-fq=a1vKC#DiUt3U%QE#Er0HZ&`*|+v=m0x06v~E3yay-CI zQ!s(c*7Wkusl$*sCnP81kA&lRhaB~Lv3{z=DZ?&%8%4-tLd)PlikI85c>Y|;jDv8{ zVyL&~@?yGJziycN5QwNBcDGT6=yP^O56+Mt4rkn}9)qfklr(IHeL`~H)s5}Yy$@xw zCG#j`}D?5!>8$ema9phfEyNOy_YghyhC>v3F*C6|~Paqlrk zu7T`r1+?Xmq{2Mnv#!|R=@^^dO7pC5Vbdj$5pW_V zrH_k=knp6Z3xKNc-9IJR3bo!L5ocYa$oJ~*zLpW?jAOKvt1|I&45Dg>O-4buki>n6e6_C_{lf|?^o#UXdO^lrT zm=%L!y9LN(I^i}#5aJ_ZP06FL()niMCf2*ruej1Tb|fCq*J_Lrh{jcFI)Sl7b&W4? z)FItN>FMaiD6$Gzdb%hjmL*a5w-c-i()62jy%zO-Fj5I?z*&i`VvFmDg-F{gHu5wZ zOWRu}Qw)A}}6UDJTbW*Mf$I_>C(`&rfgneV!EiQ(O^f`G9K0}8aQ01c(g#` z;wpM0a^$?Qs69G5lF|9I!up3C5uD)8*2cr}cXJzEhRFJL7rm6iU4c*SZNvVGc+%V= zwSZbi55Phq<`oQCB3@+Ujw^u5vD1^88o;+<&$!?R7X4rEqF-S#C+tWQV*g+rFc62! z^cv@V;GC*Lzmrc!VX&9QnSizmGR9n6w^)EyLkt}jW+c;zjV)tg$AOT~gY`*i9@)m^ z0V*rX@pxr(BhK>#|5*J)adi#XlB0IguLaNV4WF{_k)g5(p71?2xI*~5GzlE%@EE8T z(nt{BMs%@TE#xpE;HfGpT>wK9_#KDBs{y6pS_JI_^y&ejROvz9Ip89Y;BI=bZw5U= z#M8SwgTTpoy&PqvuhNTfEe(41w%+(0MAo{qbZ(z%aW5^aX}5o%;)otrK2zS%JMt`k z=JLwYO0hRJ-nf;}(tIX!m719(Ic9`5he59nUL6bGyWv@~_)zNbY4sxY_5OaFT^?R6 z68d*#D>ym5p`a5cC%YNS=rXFV+V=99mXbJ14e*+NYTX-eu?QpYozqw!l{62(q*PWW zYw_rdGw3&3p_h<|XV83eFcg<3<$AmTE|CMMZp3Z2`i2GM+Mxc)T?I2>f9~$~y!Oqi zeQvi|<>oY!8OELB%q_40%=`-vIGWp5RK^G3`2Aq_Df~WL^n#nC^in8LmP2oIT~}w9;g+h)_3z_?z^`2=U}$tIhrL_Za3r$_+G?w7eg&|JB-ZMVlr~RgRF>4Dk7~R?0RLIv&3sxpoO|qev38c& zoEXXtox|1HSx94i7u)6Ont?$!i`Te2j`E9{aYM=lIj+MJw6E2>s8(|a|HSQVp2f8- z^b4jq-O7@{ZP+?8<9lr`LT3%SVmd+e+N@nTh|d8HlA`yq@MZl(t-|sa z<#nHt`J2t-!j+g>33pZxFN6y%5hVA!2F@rH|EY_W(D#pzgYwY~`}K5jRAL5xkp*$m zhDPj&(R=$$(N~%PdY;L|6Bp(HtNE9uVseo|4d_Rl_F8v7Id!LKsn3L%qzCuj$C1T2 zJ$?J-$kz$u)6^o@>#Th;eG@?bw?QTG{C&bV7dtxd9N^9n&3-f6-{$@DSzR4=-e`jA zfq?jl{3rS-+IY5rZ_3&PW8zzTVAjw185Ef*{->qHO9}VVbh_M~)eT;w)s(iEp-N6qW{X)kRAnMgb|Kz|6 z&>sH2ziW@}h$7WHE+pIcRBgmd2aZPP(L>B`0sN-a-ozx{{5=jBY5^Xi4;1^COxJEb zq@>)z&ATW;1*Z&F-$WpTbSK3B!7PBVReSa8)*SEy{R^N5oX2j47Rs}QS?0Cw`%C6m$e>uI;sW?#9wzK8BWD2Zh zlhAEi%XQsAIGv?h{2KUhDBNuK1m$b9gW_gTrT;*5g9~84Edw=mWjycgt`BE{Nssn{ zI2eTx^`fRA{-^Wa?)Yo2@HsLtE5vlv55)aF5pbVFAz_fM);kCxD1y+vgaDz;U@$T7 zCB_K}42P{gxc18wMH-~XR*^_X7Q7NZNW#!M=wds_%);#(#N`0XVpX~l&b!HBZ>{Lg zyZO!njblo~frNyl-JLzI+L~i?m+(d2e*KrFf;yzomS`tnkGfOis`!2d%Qty;+W5u^ zevLjocqh>)@iS9}g^ieob2V3sB0Vd;lr)L&QN#E*)rwyZh0$H)F!>_|H_$L~=l+l& zEAJLSaNd2@d46*qQ-MKUA@VatN>IpdtH@>l<};~42ztR!EgNE+t3UA^IPgCJlHLqj zSF3N+S+42E`QXd~sXrClK&*!llsFr>`XKqcNG2#0H%2Z4S_axs`7B;@T83jPKsqO4 zCwK>Cw>xsDKhp}}Y)D`vvRv#2GYzx;~#eWWYQ zD}^Gl@7)~th9F~?RkFv|(2BLU2~Bl%S3pH5jkvw87jOnpIZu{-AwdKIJJ;R|Xmq-5 zoE{4HjBq_+nsM?HG+L;j%LGaRFx=P5h4g%MU1AOlUp`71})3 zNn$@SSdf%SR_&5purZg|_5g<5`}f~QM?TmsV_5z>L8jAAFl3cR8&<|R00LbOCh;knC5LcX?aHz)6&wMJ{C#C z*S_uBP$HbD8o~b(+M5zuEhk`@+HdvikTl@JX|8rVeGvv!-ChY3_>5!M3n}-$6A0(# z`w>MHO&@d}-buRqZe-L@#bOCD?oI+Pb;9c5#A6K$WaZWL*r=$Y>!G0*1Pmw;~ z1PIPLlo7m~PG8{LMx?a$?xI4N|Jz($Gdjp@;$y_5D?F?uKv2;c6jHo*bdwsg+grGN zp>tfRP@`IJ^~9W{)b;%UTliO(uEZ5fXZUqp&86o;?3mSh>+#A(&Pf!wGmLY0p!PkP zN!X&aFK4?y1h$sHDIggDAoHF5Xtm;b>o`B~ZJIW}6=zL(5|Tadj-}^DRP3*!^g?gy zqejCUP<_N*_!Lwj;MGha_3QDuYDnlktmvSB6D%a*K6i{bj3zpS@lc8t^33P4viLjzDLKR60 zEWM-c9v$ZtbAY#)@FAu?(-6JFsPGMKS)2x;kY$!%w0pXS7N4bs@!kz%&|0^Jk0cMa z_xZjhk#fHEPI&&%dZRFoS~q>4onK)`KT}d{?C&x_A%&6j*GE=~b8QEJ)1(~`0z}b4 zOn@}T_{8~}FtK7_Ys+9rbDz^mLOV4jDzpVshR~~%#7Vyg70@K=UMsAqKt?0qIug9Y zZBxDzn4!!OW(o*lbxVyk1%`9b-4(Y{q2L1o3YqNRsc1n~&huCLZZCn0O{1!*%n&EfB)zMWgvVzR2P2|(&7RqDeouF-t!17we@#ro!NxyhT-mo`>$hASKU==?bJ7+rALo^)D*9 zEe&K5)B&$>Ia^yTb{FvxwRFaX`-UL8K3iLIs1kTAQa^R_(=KK!?mnSx1pWN$L8|$# zepCpVbWK^WHK+>7Rm>hgGDE>7(%m*3Ek>C1hugrgP-@ymUZ>3g+I-Lj862-YKH|rM z_bXzx{cfL@ZFOMF8UUvF^KSsf=Q*$_E_6qJ^7rxo#}nxJfoYuM_=#)$32ztV6H7}T zH^&-=*WGl9HjYW*Y~A0xQx>7%{d~Y6W5X#A9tGYl0`&3U09qNg2uY_A#C{bGAJj7Q ziQ?8E8lZ*_snwqP`O@;S|*-MQ_JUhbEP#-0;BH+ zW$F<~f%6lCmjw6jpv$ah#yrMAVwhU3c(1>BUBI>O;*bYhUPs$pjs@eJ4E}e*O5(e_ z>p&GORoYJXs4*fLs?epwXzve0TrY9bpakQ%v2h`t7%?fvi@4*vpU(Y8pAcVW20AkH z3;P~CBz%0xoUkk&ZaZ$NlCEq2svJ9LNf{|AOQ3M{+1SWS^zkzVP38?;47P9(xSNiA|YY*v3QovpD z7Z!tFNehItAKwxfcgvfNvk7St>o^M`{b0}=7Zw@AW1Rl3QHv5|(lwo?5Y$$ppC#{z zyb8Bzat?IixeJh2xk9-D>G#YPW%VTs8pWEjkZ4Tr;OFP(kIL#D%{?3obsy#i z#v2?nE3gzCbz0Z)aXOHt^FBJ3RJ6|OcsB8G5D$cfR3bbleNXu=JdHbjAHQ(GLT^GKJDd`O|F7OGV<&f6RJWGe8e;kF5Dkiz%+bZK(QC|%q;1@j@{ z|Bfa6ag!88;Pr2!PsKoA=QD?}u2c2GWd$Yv*dCmmUID~rmqHql$oiBaY9W*2h55Um zPoNl-!iG?bhd92p(}G5waOd_*p~IpQtBqZ2Xoz9yb>DK+Otq2PuGQthV@@OSUm?=K z9EQlIu^s@!3tLsv5N_NWoj%XMH9=NZw4GuL+2?T;6j zqyjcmqL6OXTSnuYFyy2p82n?b5|y`99gvE=w+CA1D+n*xJ1@t1HdVA2{KeH*iIkKOv_#^QNdJyN_)o=DwG$+JikVWvDso9D#pVGq8>La^ocD+!MW& zvKTmTxIx<&2x#Qu4wZ~UfUJN$9SJXjHnP%bwa$mxO79a(*xCY{Uda=Mbatc~;4Iz} zaOT|hgy>)p6q#2FXQF-9hg1rmAeWRrbTG>X+Vnr)O{~!aYMdU|hp1=$fb=MaP47wL zy3M%>A$@?$e|_MeuWgbf*`aeYX#06j#_=HVJ8F>-K^-0*>P;Gaf5*8oWvri5(SPD1 zfet&ElU=ulrAMT_il*~;ZfeYo`@G0dYQpivB7ztpvK-*TJIQm0h{>$~u60YZ;m(q_ zHmF-)X*)OPXTEo!Xpq}`!;pSUte$f-TS+5n+U}4oG=+G)nK$G_oM2vnWCKu~iB!(x zFP3r=_^+ZlC9_%1nY}f@fQ3SQU&O`574n8C8w8V1+RjRnrLTUyD}8fyN_5toA33(i z!TXCpp-*l2uD5j$8)l`hrVWgk8z^6@+G?h^piw#M7aMAea z5XL<64)Z65Dg2XPC$k&3faiB6y56{Bt8dTb@?B4T=%2mLg`QF(tYoV4&J)#i;TrX= zA=EK~@kM?&CI0JmkSxL#w7rR*OzkitR=4u;*&TdlR=hJBti2Ut- zdj%BaAgtde!wT^>;)7m_L6Zf${hQW<*3lL-=x7vMOudx2tya zjX-De_Y@TntrVH_s1-eRzjJ^+}}%b*Od{PqgPA7%y`fM!4@ zzqh7sB%l`N=T*p|_N(Z&$y-8z|MCJK21H~jScN0{MJXjtE3SEC>7coP+=nfzb`W2{ z!+5<~!Lr?kSS3J!BOsU|Q@!eEEz7%6JWIJFnoBpnj%eM*&ZA87OgTQ-gFt2_F^?ka zxgWza^W!c?LY*h3p05n8{9fh?;!*9{j>C46n|FGWq`G<;U2qVCvSE4#H9ed}0_Fg0 z(Y4tNNISiZueD#4+54r9XYqLgf{7DSoGSfidth(}jK2-{zwf054w&AvF-rK;;y~em zP}H~@IDu_M+T~%~g7clJ4+I)p8M^5DJ=gS*Oc|Oidnrz1YEX;ldRdas(bVkO()r`Z+R{l|6gX>JGc$S6Fx(hHW%eRJ z;!;NgaQLg&bBG1sutF1*bOK8*JOu4Ms>kW<>R-?Knw$VE!w}H@aesZj1*B4)e-5Gc za_{T0F+t-KdPQ##R^;S>dEl%h=WhQdpc1PXfg?RW2lTz6D`|ewTFlr`%uw+JO=V?e z@xz2ZaROz*pJI&CE#MUSHfsLX_vwLDp;1lAqjQsB8)Hs${(6^~pEgwytQLQ^k$WLM za9N@K`Z#LNkx+>8ix5grvKH#I@>uskQQB=qi{)m7G5Yir1GN!f=Be6?<4AS1B?LIC z1&V0&8Zlh|?B?Q4WBZ~zZmpUEduotOj)%gT*W-DE+U@-c_&6&wyG8O+{?Z7i2k-U} z)7ti_YW+tllAloK+Dn|ec0{~=_hVZN2O0e#g+|))UwhKIQW)omHt7|-sF!0r!wk%A zuJ`t!3(1%2Py|_9sa1SB^#eBxi8MbKs4tAECK2c!-arS6b(EKvi(3Te%9s@CCp#t2 za_Uo$l<95n)wG{gt7xff=}e?iP0&p1=~jbg$lkGwyfSJ9yRVLVeQHRN%jzqguV$)! zfM)^ckfAZ$W*qa>2oCg)F=97v3NYN;-PHxusjC(k8H;__y>G{|2p@+819O|62nxPw z*|s|aYInd4$x^i;%t|aFzgJUkE;zJQh)$#aIO-v)-v-1$eGt+VasmY`9RS5h!5btu z3LjCuPd2kNH~%(=*nW_w@iXcS74g6cw{DZ!uPeVScsPCG4DQ%xZ+~1alut)jL#ZY_ z13>-s4U}DL7FT!TMemO%2XdIfm3FsW99|d4KxbB5^8+NGQi9v67gO(XE0A5E53nYw zJUkM{UZ216^=h(p_E;gMMa$$}&1Jg~JT^6`t?Iw;v{KF`qcM$|bDg^t#DIHg2n{L> zSJi8aIw@U!dK%0kV(2SM9ii%Y9t__Te|Vj+`oZ2l=Gq(pp|FpQDBxW+h3Xx_!o<>o zc=AqERtdt#2{bu}@zF;ZF>qH#_WZ^!UOK{=(sQrkm%!7vn~v6H6OciGG$zX@?H_v< zR;*)L&z&8w&(|F58L-~Lor`uX=Y}|I#f1MeUJF9yQ3=v7<(W!S{_)Qt!C*c)O#NPC zZ1XrNzJsEhHO3=|90+r8z*=H?#p0%-j@czZSER45-ZR6SX5D3ubJ(AuG`iJo6saG6 zYMMy1a6tEcSHR)#=ENzkMp@4-Q9-c1>4i_Wf{r_TlBb6C|4sU_Lpos6t0 zueb)NI-_QUtywNER2z0auAUu|K@U0Zjn58|GUNX80i<%LUogd;&2(8OYsr>lsw5^B zRUScvOCD3RJ4_mKKvALXJRdOHhM2XtbjZ`#QFGgsgnn8vf?-pza`WS5?(#&@Rhs59 zsXQCqVF(MQP~~ylyUPBy*)TGTo;bo5e#8;sN5F+VvaiGL`OP%AZj9!s&$13akjm4b1`j}UIToLmvD}(Nymlk5x(zTN%7biwbxEYTp`E!{V*h)NbEN6 zqS#41&GQW+4%r*u{+yK|gSl_dMO*OIX1v8idHp{ibWLj1m+=;=719Ug66ws_$QN`+f^j1PgzQeVVFORXQ|OX<5!L&#!Umi*lveHIrhs_UmQs>+go9gV872(5oshN3t(y@#IhkL3wbL@J=uaA0pE_=h$9< z;Bhf}3&(3`{sl=*X>=7vh$f(2iPd{1odZhLwn8axZT5hx`Bq!UnuWn`I!U#n$=${U z{OJR9W5(7fXr-~wisi)PsHFH&*{w96#}ZGWaTvVWOncn4!>)g!G;ab_vq&MErIMAn z=XHH>YZ=9w!LP37*aWVL`du5@if8Bg4%OcDp3V`=*QJGR46T%moXUH%UPa{bj;7mQ zqta|?@z7)Epxio=Feb)PCe!j%^XJg9fUp9mZH6mF{P5K3_zIt01GK*c+Nf~X$Osw2 zUDMp9wi->gnt=u1SSkyKB_4|w-3gtK*EGfC&lBVVQ)=eA?`sRZ56hX?2NhSlQBiDf zjR8@il}caFE1vaV35ju)4L*Vmjp^BLJr7edov%9#=zT6!Gm=kFAzEcJhsb+*|I2wm zlle6_om2O+IBcS!*-`CN2D1P5FM@Gsezqm3=JzRUbT6vCY*-C<^E~R2<76Se3yjfl zrlNV--KEgC@+CGIT_{`xOr9ToGN5U~#*tXSQB@AMJM>pB2DDgami`;6sNN#{;j++#PE@i8kXMatom(9KaN z!R=>X-iNWu5tr7N8fAJIzurgOmy=b?^(vvY>g{k~y*$(V5smaTDviy@t(SO~A$v3M zF9!BsIh&t8M9L}b-pOSP9)H#h4*=_4pDS(u!F}OG{5Y5pUJ9vX14~FnQ2Z{y8c5M5 z4<;pYidR=xPj|<7XOiRK?qk6SG!D>BlG!qBj?5i}2G3ibPR3UAKGS{xQDD7=T^s1N@>1enxjq+q;o#WGqY%7#b5BV*U0QnT zzTvM5;!Cy*r||F;gNXL;E~Nb*xv8iiIruf%hwIwLf#a{)wD+H%(6X88WB#kE7#0M3 z+rjVJwTC2FO#OZgZwN`#M=FVJY$cM&esu4mwY5zhjm^X$c6G5+(@=(G4}*v9Q+(^- znf;^3l1>~&@9c8Q4+!)gklC^xeA87C7fZ5ymHOC{Lqyoeh6iz>-L`RUWQdgaE|HmO zpXv+l>gRcO$FMSShoc0a-z6P;=iZ?I@rnPbg&E19l-B}6=LDjLoW&CVGYBe5vI*N| znDTzOhC*vT)rMzM!oQfnt`04`}r~HabF97drgH&&Q%{FGXuw#n072o3sPa6RhZ^X2#x3ssY7THT|V_??^&kStpHz@lqT0J`3XkWZ~$@s>Q*6pP4i$FW;2C z`@+3;jMi-CHCN%C@xByo%4Q2qvC}ASQh6&1y*}c<>_e1<3$7%)8b-4}C4;YZ!lK-U zDXt}c#_ivPNFU{^0tKMP0zZ>50F(ym0vp`$WEM(LiX0 zPzc3~sCvgsVsn9@oB7IvoHw?0iP!h_q~UtKskDk$!mpGS4gc)g`0QzBZlX)(%aprN zxPQy`itob8<{>?U(y_cci;&tTNt_Urr6r=~doIVF2woF3BQ0o_^q^ht{-dPu4^U-D zNwRAZRwYI#ElpiagOmA%3>|qX!FGEXQBv#@m%k|UroRgJ4)*d$cJ*@MZ!?4-sbbV> z&}Zlgv|<-2OAd7F!3P`GiNk6(efwYT1^}@H05lJk z%u|~W0){_2Hg()XG z-@b;damJoPEI_|KetZxbXE#KuqtH=$&>& z=(ST#5#Gx=6)(ekc#3M>oq@ewh&41do^26faDYrGvKu);gpUUw9uT@ zhI1d8$n!t@QrEegaSv?H9oAI?xvz%tt*U5F@tHe3wsEcH zL3~-hAQi9wlTQDS7bl+Nw-g{@%_mS(9;SbyxmIGRPKYV?4|=8-1P9EcNUTC+gBk%G zil(wn**oUuYgY+nj800bPJ)tS$?53i19~fHH_=FInjW1F*~{tIsQ-hX8a-Zerdpb{ zt9tmrPUec%`t<42_048L&2d{CEgN4aRn_HK?@6SNqR5X@gcDr$+S!cw>`CouAG61U z;Y*q#IwFCB0mJH42!F61$jlXQpOFED$`(&0%@$Q1nUx#L`1*Q>PcjD-^8P63tZ|!A z-II$M({6JD?os=rn|kb!q1GtoMoiwX4ajBfJe4@~#~YH7=-%+QQfF2}mZe9j^$8l& z7e(DN;G7nQH9zm%RAh9`C0=l3CXUj?5+Dg|rOytAxNr0oaSuUWFzq*{+{T@1=bm@r zhse>w`fjtcs!xYnc`OPpM$`vyGLAvKUi7wur&FC)-#a%SLM$tja&RX9Gdl6_sY^OT z76CbGtZ6xaYv^@a5-NZ`;?;cd6!B zmWtc{iNnsO|x-TimdfG$- zyf&|Mk=aD|KV`efO@0>0WiYX?nxo716g2M?-_AescF2LO5}&pr3eoAXy}Q{mEbN=x z3KQ?yAK$>i7!0SB_FSlh0p-N}{Pz0GZtk)0-jih!@yIP2Hkgf9KNg7Xi+1dx9`;A%Y|XG&T}l+qb>qhs>_ko{}>QW7M; zTq-&isdQXRxudCQD+@AG?6N{+#;Y32c0ff;(hsHxJ6P*F{%K@je^sFU~ z7wtdf7!Uf9G)3Q;z^4p4Dp5|dwAxfBK@y`_FX*%cMVqECFWbg{nD(0l*h?Xk`p5W zo-F(0pyujH3>2O+Dz-{((q-|1l@Bxa4>JFF6_P~Qy1KgG`f|Q&O&7A(asU)BVOi#3 z?D63`;XxQGE*xwZ_fv%?lkLkftEKqQ3=xNGO8)Zxu&%bJqk58-v4jsjzPJ z?2AIpfBhLEx&M#^44PXADhFm}5EQc`VXc~M$Q#gPCL4cc(^(+;VVHx!^glF0??X>t_i1%_SvgFk`LyCDBCe||u4{AWKB`{*dlA}yA8sR4%*Pji z&Z?}e^>wT1evXnn7^t*DX?)T=Y-f1A35*6B;px%*?B83$J;tdheOah`EG&FAN?iY@ zz!S-WKTov1YtPct_|g59r1VDndQ=-3{PtL(HsE}WJ!Ug=DjBoNQa-u$CVz^^*0o^& ztPH!+Ydw2{fW6AN8fz%=YohVH8p@UxZc=}CddF^7aI;u)|L(U8c9T^#wIn%u=2TQY z=!8IFWOhy?n(sqdjE)h*23t^B90+%&%rND~-`PiIT^R^ebl#`jo!Z#G=Xw%8S@CExF@k5u#jpXV>y>}3gg3*+N`+qpouH$J_>ByABSp zJMQ|M5wP6gFNr;``?X$gJ7qt&m)5oXNX;(p_=-c5@w*L+!UukOW!}$F;p?U?WRCZZ z2hSY>!;bHKb=VyFraS4DZI?MIg4Qj@TV4vTnXjAqEfAR0$ARL*J-!$ zbn-0Wf=5k$HNuphv8Jm!A;LG>v}Dj4-Qmx zI2P}@w*1b35kID9SU$N<_7hOZH`qfo&oWvs98YRZ$dD)cFU4ONG)HePohd8ZPfNessYW2UdyXtNm$^+Z z&m|@KjVPguHc5@ z+Uf9t_gn!zS{U&;cJETCsRN|gHQeNtWZ9u=l_;Q1EeSw0F}hIk6e;fLP;|N>@!U~T z?;MTl+dzAqK4h<%w+@X~=u?iXG-rphAoJT6Zy`fM{`VJ9tJ~DAmStmsU~ z%O6*owQN6)J||#lypTw=n8=PSwFB@bz3Vda?E-ThL5uukkzCT;oZswHotMVQk3v4< z9qBZ-{lfn~u=n7V^Xcj7t;eT4^Sn=#`NoR;zfUXEud+8@O}H!#(IG^oHEFjNn#FRY zNOuOGi@cybB1nD~Wbq@&hJh{WpFGFGu6i{XY^!drg1nvSgGdxmc` zi#^-bTI4XPjzjyTEvs^qHh^Q(baeFW?EJ##xdQ5(xcNbRA~VOCtP3*UF{)#;ub?+x zaq2fo#!NNFq#lJrmEf!fCG0wBP5AB~X9pdYeA6 znCjC(Z+O;G*W$rn+a+MFy7zsygVj{R6=o*$2aD%44<2Xt_q+CYYORIkQO=D`Gy}x< z1M^GR<8R?5m2C6+;a8z`FTY1d5@eXN?5;ElJR2d!?~GH=UUO*g6&{JcHM-U@hdPtM zsMiAJC~11{s?&7JI%UQ~Q3uJv?fsMDLFvK>ZO` zMnf`!Jegf@Te~<*reR&v{}aSu)@dL=_~f3cQf90*iiK%H$U9`z|7TAW>tlow*JYQ! zP#i^ix%u^n&-$d_{5CJ7!wAEcGBJ8ZBMCC0>=cSiRy{)bvw#C?kh6VY;Cr&W2)I8w zZBn%rmsa>2*)1Rrxw7{^C(Tk)4e@K7Qw#|zviuFsB=Y}QR0f{VEQe7T@Q#=xmkldkc4(!}~Kz2$bP0cX6Mmi|yH2!bacvjfcma3vKS@Ru~=n ztJjfen76vTn7oiQMEw@=$K!rn ze@I#oI7$DnB}f?mPy{F|q)4-rV$RMo3j^7RAo@_j?2vGcd`o7 z3l-9`g8Qgi{qV4h%OV!ecbjF4ORp6Yn?#0?x1a?ve{y*R^)%sOx1!*VqdQd!p6nSaZ(j!h`+{+lkqn_=5wo{;P!0B{z75F$(Hz! zlkg*jh;4Mdwm=pUQ+`Jl^B$_`qXJj;UeM-T7t3<=*)xYZh+o+ik%fTFPziOpvvM0~ z60g1DQ+@S*5y}<%2y{Kk1>kF-6ue3#cbiWAD#y-(qd8)?2~X*@o+syMutrc_a56wV zqPUtjwiYcUd44{OST2Ihv=AS{f)5n`%kRVlM zp+d+6m0Cl|SQqZsN>)z-3ne!kzl1m3pEOoBZ8t#KkhA^aDT!f)nsMak%?_}TgmD1g4>SybARGbZKeXKJ!8Rya_vMfRf{DD!XZ3*&zX6A)qP+S9=Iv_+A$K2%@C3UT z6g)!Wm@V7O;I@IMN)uY@HfDE%y_c!@2`O9yX9fyC+eB2~B~kX^@*>$U$wQ>FAVu24z_;W9IU;Q*UzH0ySq}G`_jqQrB z6OlmrCo>fJ{Q@b@aUi5p(PFy*u_fda&Y9IU>l}OEhXe7~C4B1BC~&8i+K;N#9OahY z7xSb#5~u)-fAn+}S?Bs{wtnSwGd?f7O2@oS5f4{Pq3c&De@DC>s2pR&EOVOG_VzR0 z>zwTf=os5L_R^i8*tMQPV;|%cQi!dhcAL`qRQ^lAC$ZI90`U!zuZ>L;{Hk{`A}Z93 zYvJ8WG;oE)FVD8*OTsY!_zP0lwkrMJ1XUXJJn9+P-{vUz3)rJFl%T(tdwHoInY%3% z^esX7H)DeS<8%u@t{=mIWK6#b2A~fO54)dt!({tx9Y|#Rd{k3uU24(we84>T7+Im6 zu)yh0{9M)AmUq|UX{9V!!eh+U9$r}~#EOq$wpue;bH>+f6!(!;ev}@W^Ae*>XIJFy zN!Z413U|6(BwNo{1=+~@^KcW-v$tIC+oBRh6Att-_QBhCaBF*w4UL7_yz$oux)PSP zv{HcnLEMxT(){N=JT>UALvFPqTq0SEd33?Qq7lKvmiU-gzJ2=rb$&z;Y^IjU<&POz z1PMuhpBOYnPQwPNzp0?CQ^3EGxdupQs5HdGmdqo%3}5dSW^h$3JCsy(bt>9}R?m)i zvFME!7nf=kf0V$2UAHw)`M$(A*F|9W=V-A@h;#Zyl1XWwfOp*FNTG~oIF|C zGK_?Q$^i)nM8>GM3nCgAx%rcG+S?#ItoDmflUnIq8YVshR5G%R%`6!)h9m?#$4{)= zeZx_UELwYG2j4Ps3N(-!1x}8+6zox3S5PxHr`{W8z$wn0azYj#{<%(QSk^fX&dWAaYSntWCwmNv0uLEdhhXPEUB3Q2D z{@#%|60m$GKBgGG7FE&d{@n5WNR)BCG2NOyPCZp{k-_hf2bo7vuY>tyOZ}y zy??=UfD-2&iaU?m`Y{`}2;5geVf4G?v*++JlfZ$?cvs)Nel;yq9TpmDseSrF-EFY8 zaw$dDuVig`L%P{px1m-ZJ&aGh0u7t*K`VQmxDgVhnCE&uY(ekhUeXD;=Vp#t7uv#8+P1-TJ?8#pj1;Al`P+3as8~Kf`>o0SS zzR$#+K2H7l0`BKjt0$|0f#WmMVam9)_Vyb-ofS$&;)FxajvN-0zZKTh3KSHr8G`7!I@JlZMV9= z8-@S2Nc@!m(wslG`G#?VyA=aR^bq22SlB1|dt}ZSqEcT4d(izV8w%CvfgOEcr}?|q zBHD}^r{^iS+IK%O_I@V*yuK%UABTt@5ph?-MW%3>FaE2@c+Of>X+w&-y@#-KmL`t6kvcyiq~z%>2Cr-0uc)(x$)IV zSVB2VP+org0SGr);yw(}(o%!lqJn5+M4hx@&msF4mqpcbRGE9XS*=G6r(ffH)~2>* z8aO}Z17qaacJ?LKS09CxS|vd0DhmJF?;lK680x$MUw1w{bH?+Y5`m~?=I?LyoKU~C zC5N6}B+qP}n6JuiAwrwX9+qSKV zjfpw2ZQY)8&Uf!~e|GoN-S6JJu&Qe9s>!L{orGJUTqJk&@%;52CT1zEB zM*J5FxJu0{r(-K zDJTQr2L306!2ftWWE2JYXHhu_GKY~LvINHmipjyQ|66|%V!$X*$E$CyFkW#DV!$GW zzKv~tQ?m&C$+}9;lNpx}Bk*@*4!y2|I{#6UjabTuvn0Ajp)#J;%gfs&z;HMFQyM+= z#>O{SUK9SSRhI4!ZWB)tJfi=yNdJ2jeNo6FZyH~g_Ri2ux3*~*Z~T_M6O@Sb5&&nzn@G+&JuZOqJiu# zmr~t4!t3FZMLOKpV?Zf)d9o`S&da!qC0f0kcZuq@l$x_fZ;~m`e=j*x1QmT0md_QT zXzbP5<@0uGmMa^*q(Jlk9}nQ>hbqV^6v3~Z?@P9@YqrSl;6ox5|DF%w0i@{G+;13e zrJBw~BI5XQu)Ih~4Ud37H;F^T#t~2LMNP1}iLyEu&B}ig&b0<(-3uMJSLj)fa>fp_bii#c5?_Pn_XNy!@VnH(+Nx5?5YGnrOy z++pQB`AloQE?UYqHuB1it<^EZk}#Ih33L=Ait&1kuIKkOsjz3578VG_SX3aV5a**0 zJtV&?R+gkejutjBy{FaKVr-&_2O9H@ZeV{01`dNcPbV{FUqjfHY zX>)qMCmpjC;YHCVx8zXOY|=5hU0iUmY5P?hw^Ar#M6)*M2gs)1zQN%@!@$HCm-7Dm z(*V))HwEFdv02x3H#m~6{lQSB>$83f#j<>qvVVwvB;G@BLGOr4_KU{Y%E!t|=PZJc zV(a6slHu!G;%>>;cFd-g%g{0~n(1{Xr1#wW#N=o|ti5$pmEEbNx#*XYwv18Hfzlx6 z%h{QaiwodpN!Q1bfZNCQ$lcOb-Fb_aA&B*dA1}^}m?{2Cd=(`DE5C79KP-M>MORrkKhs9*8R9?Ts_UQ9y;DxyeV91?f=2}vOFY=j8 z?+?&rw^p~W*ZS@EKU*_?mva)OHQG#qviuZ6E%m4bXY`~mvbGF1uqpz54$@0!%q#v7!Ht+7 z-?bXq9kf{1zKaRNe?6ke!R(PLGydNKMS;M!d=At3{QNGC2zXU=b>svMg@K4yje0*0 zZMM4}?tbU?9_~V|+n>H2yv589FHsuL_1vH?DyzVJRHn-<~{_63Iy_AunoF~yghCqwW)#ziGVbWdj z{!MMmuXh*S_jNdvp!NP^)k!AaHn?j!JDk+ z;*f9N4Rk>oRWfo(-=cmV?g!~vqz@7lX6`sjpEyLiDlcxHpO>||ajbgIVyMjok9R!e z+GPr?=Vo@y1Ns2g&_CN<>Ar=+-Nxbp==%Lv=j##Qe; zhrnBR+J59{h#kA)K0#>~4XGe{OerK>gnyYg%PJlrAgu#Ii#E>y9B#J2Zs&Uw zNi%#xQ0NO&6ciyN*j>Qk$W57`Rnu()^#xMiP4n}=_$AyQ93;UFFlRzn1o&q#K-LUM z^(`!ADs3Rs@CsHVEeY$uhH%B|o7r!twTB{r8Y6!Rod%Gw$Ij3vWzKl`@MGN~Ak#Dq zd~-@O@%{l@hS)We+xwDTK(}5!CF^pwNPYNB?)X(yC#*f`X2zDphF z%VH$0>Z2wF+=7b4{W<}f`l}B7-(!#L3Hbqnbe!7Pf}yM4aS0vSxxkRMf*)MlV#1*d zwPV%uZlXixcRirzxU?zh;N~pEl7iW@=4ji6)K~j5vVr#&X@4`@2}UmwDM4!sFk zLB2TTwx};SV`*BMn>hX%j8j*}?a>(Zt=dAt0qmP_OrJC`{|-vzcyX2C^btBsmato; zVLc2m0|w;1F@~gSl!lUeU*XDmyl)lX#c(jz+et=O+>_*x@*RAZ5^Zg+j$cw*HsJoV zrYb$b(rvjtF`_&ODv~Fs`3O5DYmDNbxHG*0K8e=|TI#;JO8ilR?oa48O`FKz0%FgT_5`(jVj+Cl*FH0~#UyZ!vW(@z4YZU~(55D$RM^Bs@kxox-)A0y;bm9?G+G<+PfWjEXTy zjsinwAJ|GtjTMoBECFt@W0-Oro9ZnbQK9xYCJP=rfqbY-Dafvu$qoWd@_!X$elX&z zhcUlagZ?S|E5Ksnel7JL#sCcTF3SFs3>}D zU^Gb&7V`NrPt^IvN!IP(oF8xx4^?PZLJN z*Psno>t+cbUvYGf;BI(6p#CD*r?Gi zF#s4cEi*$li;$ZcO=H)DD?a@yyMfYd+4pVL>4xbKmFLG4*4LTdxb6yN8y{#>0z689 zDCyrP<)Yrro9&=3W||M78UHm5IBh8^9wf{}K{I7(+wYh0W0XR$P(bcx%9W2GUKttH zWMpd#%HzjSK%8|Fdhp2rdkna{mmm@Lzr7Rfg9Id4q7@2xMG0-e^!W!Jo+)5JKB;j( zp=A&Q=~p?0cE5pbD!TKx*)davHwWGGGVDQqf-4*9L9(;|6q9e9qUWGUBa4z10#&Fe za33$pFlltU+S+R4_Z(-AS4GVZgJ0KN&kl>b_N@&7p_EQUR@hEXbm$kNBxW5zrJ(vU zwq`fK0zkU>Rp*l}_8WuRGmmj?U)8Jpkh#nGZ49AjH zElO75f~UwkqI0Thc9$on{XiT-FXU|wS;hGYPp_n%D)q%QGVtQ-0VyOJKV#A5cF;OS z0Uu_{vp*E3Js4&pE+4`%RFS- z>8#LQDjCf|)%7BLPN0Fj(70c4gC!xy8jgHV3ijF_-=zkuWEL~U4@Rc}KQ2A^!`E8# zdog0b#9v=EVgKtOXzQ=9a==QNpq}FqPc9>C~pHKt0N1geEtcY5d#AL5k44h@M#G%K-3jZ2E28| z;C=);^1|Jmmd`k?xlMTQKJO7_*c{fA+5bc@R`1*jK`_jRUv#vdxdUh}&+Dy*b7t+R zKA*M<`#?t7+ZK2ViwslRDYfN}V?kdbfxA)l-Sl(O=jpjx)unh>w_iIv2~7ySI4Vq^ zniM@+mZt$w5d)q+YISyH_VX{+0Mu6ed~-V6)E$eSv=)azj=w|+AD{qhN5b0ES2&Kr zDXG>hFLuQyq5El=Bfg9|)a6S3|9HTVNPspMZ+*%p*JrX%&YC&!?v4Zd(oVFgyot7E(iYuf9^>*|L*s=1~ z{Tpf1JP!%N!Vy1F1kR~MQWKd&-yePKoq&stv$}es`_s;`d4G=u3Lz)eT!$8659;)Y$r@d z5t0pO^&Kx8Lg693

^T>`f7t8a`b-GBl)vcBp^WWh zmJm&#S0}Bj5;p66P!P1T&PWYmB@7n>V68_BY6}_=17&T~Mgw00ji5V!+Tw!|}YWJa_nPCTQ-fMF|>XZoA_8^HX zk2EIq4TMYMcke8~;ula-_RYBPiFv%qPb58t7u&rP*-oJ*(EUotFd7{#2f)J0(i-)( zB_2rzqx1{e*^h<9Bx5+~GI{HXt|M
OinLlm0)k41NJsL1WKB6Xu>H3E>#8&UY(994vvdCwW@&T3nEMXV+&*XY%^_6rs}VyJ-)mlg zX*ZHTU}z8Z?2O{7!{>QduOTQujEUZad1!vrO3b)iW`NEC&ahKut}(XrP5l`sXc8q9!klx~M%iY73YQ#_c(qf#?lDvy}Bo|(s!B+zYXwW zQ#xlxXhn+o65)Z$jSXq==tflOGqxIOiOj5S9D>G@Cf%;6W}vX#u;W;6$Nw0UsTp(N z@@t!BT{I_cG^!lbQvfrox69sJeq2cs*V%mTc3Tx5YS{Sjf6eWgy)7(#7FI+A(^cs< z7p7nSobt19KS5JfMyj~iiM2Cq`0$GR=IJ@(_oys>fvbrB)ptGI`(A19h5#Ch^~;r0 zH7Ctjc0*YLNgd?xwvYkc+$QxI9?rwLEpT^u`==5PoBW*{M_OU@#1d$5}Sq_6zcpl?hz4 z_0Hv@b8Z+H(e-rGnr2~*>wZ2&+4FR#>Qv;nvaICefFQJ>Y*SHy<l+9gHE?2VKc>&D^ImiF z8VQws_}Sn1$AwuydfG^8+;$lR{}@4aUoiCP z1xpaEBHjBWv#kV}ZmqQh=~Ff<2;yUu5DI&|t$oi%u9k7Zp@KIlet zDOrEZZ8@dV>HhXCe^L`*_PC>Kp$+iZOI}3^5!@CW;pio!NURzpui55%&zGpu2ePx; z_U^pgnW(DJ3}n2$eiPsR%0nU>;1*_IzdpaCixVI+b*f+9O6xfh{IitFo5_dM^JG5t zqB~OGY(I59!q4h4QEkfkOTlwBxfuaF>L2S~g#e}=IzwLlE3Klk%7%{NyYcQg?p<3I zWIrT9q*SbN18($xL1PFYSQ(dXsxVV^$)fQoM+dHH@WZpZZ^2?)}f|tU8#mvEz1*~c{ zw%15-Bg3?c%Q)ThGhB&;ml>Oyw>!{k;QZTPr-lLL|824rB<+T{68MZ}rHJO1r65p? z6C@3|64lDRL)O$SFKg2cuN?H?c7MN)J7+W;I>h8%-x7RaJum#dBd4qkJt;Bm6Cqq8 zDotJ27$GJ@n`P7}lRQTn4tT%%61igh4zVclQbW;6aBnMv_kNixzVF#e&JTKBw^awXbD?=Mf36 z^|4k~g!k84Y&JH?K?}{jg_-bgzYkaX^_at2Py-`96p z^JK)!Hn2U=a`^|LtxfKxxkOE#zFfmBBbi)Q-_;VL>-G3?+S-^i`6OJfKz45mheJxV+*^osJxjQWG)g?kO1iI4+Ibxv-|Ov z9?za6UnntsHX<~_SedU}55#!Ax}p3A_^PIehd4JAl&anhq8 zKgcF_oK?oUNTE^Omm|vFGfHgeCD67Y-|l| zS}Efi5z`V@E!YW-@(Ug>Ofqz_ zU*|R*#@SW&4*W1@{l^*CQ6s6h@eKJos-*lxk0e&$g9P(@JJt2kktyz;4y7^E7{4wX zeAB2_H{=yu7of24z~{xMq*Z=L<8cei#D9VbjF$#QwLa}X3-FKqIdg5PyPy9FOg&9# zvj@Gw$)1aiK%8wID3bVnnn#V+Qj96p#`_dYT9FYxZ*T8gWxAsx3iCHr$c%@~D6LND zo7lj?{bmrU6~|XLz_1+qf|#H|1&BX;k$7_(DLBktkLa2iQbW48mLR-UWY8bIk2L9c z>sg7M&9#0C^Rjha$Q;|ouinY+q0`yh_ZWIlanXN*8m8b15qh`Lzu7l}_E`!_xA*-_yg&B9C6GWh9kz8(Y5{i&QfK?WzWb?9GNM3yka**1Wk;#k5R-? z(b8Ic?Km(^@pogppCFLgfM4frv|=KNvh^4 z^n&$iM&-fhaJw=>o{2U}&UdPW3LaF69u=Zh;=+=-%2us-XRmc&f%w1{ zq4iO_;(pDgk|I?QO8Zz3L5s3z`U6!Lev79KfG9F4Sx_amFrU=}8CwpSPPT7`_2sR_ z$Kzek%^+Q1K0&lC?Yq+E8NxWF5tTxRV2BUm@bbb+W&vA?Hr z3&_hGbUoX(x{m%BuyBYk`>RuR!-(p}lX{X;x<=cgQPN22CXntvHU;B;72R*{hj!sl-K)S!BVHB_|FAEgs=uU*6uBd0c-)La!!@y2^!)ai^rV z+qqiKuO+Z_*a)wd^B@fp;^j{!IANqg%gW@?La>#!;ITdc zFPwxWDlPxS^9bO{+@qD;O?<(L!!4Pe59U`^mmE3?o)1@R9HMBYf0!d@VX;HOEB^(N z#CQcRq9GGjUA^mTRQVxjX62OaD=#i6r3JO=E%AU4X`_%hc? zE$n~3c^y_rZC=bf!^~+<(iZ8_f1Ig-q%wC@GbJzN*Q<|LJ;k8OQyglcSgl{_VOVmy z`N${zrE;7TI7;m+=b!NnfAE$XFP%nSfsPb{j*otsc*tZUTQ_Q^J(@T7MBeAS@ z@<7WyDjsOYHmNwiFT!bCR~+1mh>~@)CwXojg8)?hF+JdZIO9fdTU+f2r6TvB8J#*d zB>@0?+pzz`?7xd-_Wr6p3#8!9+CFJGyLgkScxt4Z4CVxF)^)>;8PYt$V13E35HmO{ z{+}+D6&dV1zd9p*TnEdvgdkJKKaRHO2yu&}rX);Y>s+RoC14E0h#6y;-Y65JXinb- zgbWFLPzbILB9+H7`L)Yo;lyqcVE+{P84QA(uAT$wi{6qYX8XO#oAo}w2z!0U4KrwZ z<<8t|7hV$Jj0PEv*e|L5^kee@h19F6HAoh`e&%UuKX2xFM|laIiW%&7$4Sb7GA)PlhheA<4nq zj^w4Qub9<^(DxYaCU^!_45f8bxbC@5oioJwsRI)eAwS~J`n-Ol2HlnShts|Fhv`J} zrf|bi_ctzv*JRwwjJ)i=oDjqvVYKiD9I(Fy``q+VurFU!-~FZ$HIZoU=(62>u%pRJ`2q$ zo)MoxkqA(ZLmqDH+<|c7)xo`YvJop@7Z+?r0Arr!)XRbU?P*?B)unv2U<<{Pra zfz;!(L_kIGzV2r>d*3t)=?_&@^Q)(Rf|1Sq?(uD#-B^5h`og?Cf$-3*jL7=hlrMF& zF5JjAvH3lb>V8ngS_?y~SHgPG4(X0M-0nQwPISD`EbhNFH^1m_B(zTXOuRJ)%{&x% z29nXifoLF{g`__i4X2j{0g)7HZ#i18&OFkQKX!idp#ht&y}A0c%Q5Z;Avl5oXI6@! zrEYhE`>$On3Tu}x(ky%ldm~L;&?O}A9k_bpiP@+Zkm#E{$vW^~og-9%{e+o z)8WZWMxm#zLS~Of7`rlzTCIs47`C(dH@?m}^gOOnIP5xVvh|R@b*1VJTj7H8;ykA3 zYNo$;E+H)~m61%d21P^+otC|f_x$32PkET~LlU@gcn^y zLxmm31o|7ViQVP#sZ$Xjv3MLB%vxDAQ}V<}-1?37-u|b~w~Q+b`0w$bwADYT1AqL! zAoC^c@LOv6yuV0@^}p4ZD5@aLBM<=uW>o@z?Vfr@l>p{3AQ6-T0X%EfuzDS#{7~D4 ze0q6<09WjKLG9cGb;4SU{{HgQn5DpmvgA-&c689JZOoTvY259S>1@49Fh-<`!mCpG z*dEwHGPQ{~3eMY8QxbN3aPwWXcN9w4S985?p~|WrR8jm#9z8h0pzvn2?tUYcL4)2U z-#9|<&@doXX-TJ5tzHSZ3e%HE_ktyKKdKoFYm&Z$FIL+Vl48tvibGY$%i^_w5y;U?eUqHk+26ZD3l~bi|cJNx8(5jwmCgpe}TH#r6Ke z2*>x@{cGea`vBW(#?1vC2d7Ae-0``(Z3RCj5cdhMe%84YWq!5_3X{FU6V^t07)pp2{R!L2>Zsx>jmzNxN<;{Ad%Q`v{AR!fxCgnS+y>dmZTHrb!>TN zzPZLRDsxZq=K}ZO=BK}wQ7l4X zamd6hOrz-@uGRX56mMLxe!2-@h*8(m8C+p6s@81ju)O%+y6>0y<-so>z3#h4dR=aN zTX69x9rwp)q@*2MOlzDyE_*6ViijX@sR-#wQ(D_TKV^-ZdY@@dqH>)cFC9jbC*;f4 zj1Xg^LXs?HWI2h2(i_iq+T;HyM*JGuRt>G{l4!8US03pM>;z@8C^=d9AvTq{+Qs5% z+u>2T1C#pblzdZr;zQOC!7{-{H{oc~BZtLHqVaK2b$RnSF*`6kiO2Ni*I_ZZ)#1;2 zRrZ}r!DX{;qEw7O;yDS}VET<}>h*LkA#ZYX1G7VPdY1URGQ&TSseLt)I%lC#v12ui z*c~8L^8iy{KvlNzaHe&>)Slvz_~-_WHu;S!Pz)PwF6%&)nN-63@JJ+q_)RCd88=m) zF$82W+ks@WXYwbOMO{!Z1gIkZv8z)%1GrV0HxU8}mN)&_zLGmB2w%(46W*O{Oxj!< zO5`fPc>oHe;@IM|cgBp2(->1pJn&ga6(G@^ykCYF&Xr$w0JGi$L=r$G`YS?^$?%?6 zZ_?p}G!@r0O5^z%T#ij1ZtxV;2)!>GFtuJ|X*R@KSE_V>&Y@fZNLDB(!}4`5?&AY= z^;))cC>P41X!xFA0LSna3B} zUc1|^PLZHNZ+;q?GsxN@%bAbMQsuYDKTg6F@-XSvmfN-xic8)`1q6R9IDawpL|os- z1^*ZVFSd4G&qeB^&Bmtc*DoBtBMRjjg8Tv!Ufs%;?m{N-<(3qiDMc~VSMpniENmX$ ztR`!S)1?h19*YTEgy+lEAAWh2?0-^a)=%(fc)cwD96Q6iUUI6PLlp4-+Cm6_D>lL) zOA-v;1dS=maYFfp1wMUrXD}4I%9Ab9? zO?#1D`Hn4PnAkjglZISs%d|324Thr$KGBrYV^nxx))T0URY>88ctQHO+H>jWjdqVQ zve;u!zU0p5rSiUeM+9HDZ+kce=9*Rrzf91%zKl7x4n{wWU zLxlBXb+8&xqRK6aM5?nm_OzBqLBt)+gGG*u>2768fy=*~rEpB3$$yNu7SiZ07Ejhu zR3cC16stxS&B<_3{W+?&>LJ3flGEKf*eSIa^`yEg-=z~1C2)7`i1#$QUVpMag&rfb zdzJp@u$C|Z=`-gBrTo~xP?Z~ubKwunH>y-nFY;CwE8BI|R3)(Wsg^0Ho*M@yj}7X5 zR;tl@ZJ}}#;ksCn`^V1dO*+gHn_?{XE1P8p$h1N6Z<{E?KZ!h)*SXDb3zEy?8-<{nGFZod@~I_x-`_>*fIAYtHM#S86W< z;pnf-7YDhn)7Qop2z+*Y#C=VE%w5$N?AePk4m^=~^BO?7+7_!Wj5Y zQH<&@29AJU$^lX-xVRfjbum+oMhcdjW8lQJOidl!+6N#w-*oXpH2)}7EnIBB(6V|J zWUH_mhIKb`z0ZbboK_u8Sg_AFz0IXu=Y;O~xtn1&+L)+8nA*u;^g6m9M=dv7$Iag5 zr-IVJ{Z)x7Y%uCDwps;1k)?u)oDNbG1_zywTf< zc}bj-y}^6}y5+$saMxfs=)DbwEpe9i!ot_QA#o5`j)TRT1F|>#S5~G67xIm(e3i!) z-h!iZHe2484?AQTR6vf%i>wX_La%(>5dF`enNa46=&N)In~TXjE2W|Ci1lc&;0P~} zw{Y+y@ILn%AY{!e*Iki`EmjJ0pIw8J`mwJ;3YpL7+kTc0S8fI81nH z)bYqr3<>EV3HRF3ra31FVtE-zbrRy)Mikk`4F$?x+b{#-WZ9s6nI@s3S@lq*9$^~m zU{2SjHY_kKb1fusPXR$*CX7fO63kPfAhcDnimjLciANVXz_}%uM3~)C z0J%zT^@NCnaO4$j3o}tsxv*YeTJW;m$gw@%F?a|lVzql=%B1fugq$Rd|9%+umNbOZ za@|rKLfj4-m*sg_JJYq)dK4?$Wv@cR4SkbH6fd*(5p)XoHsP}GcjyOIi)VV(7NR3> zmNw6IM^SsS0ENUPm0fAH3S$-Bi)*ul;*BBAIMv)!?C09_+x#mHE6yswr#L%HT6;;GY!gTxb&z zOYQAM;deyRG1u5KGY&1G2YTD8Pv`ol?PDA72=-YNq0B-W&xPGnQ@|ZA=?Dmbz+C}M zRR9F$$rn(Ye6@;6v40@hx$Z){)+x)_@1pa69LwZfaEY0g4@fY~Bwt~H zpF?3OtidK5fv3p9R$MI4?-VFw=ctaxm8o$OvAnW{(QM?JMizq~5ER7^Z${hh8}UQq z%~?w-)n5u%V{6DZTou-`dLsDP3cd0ccpRq*+`Zfy^-QszB7eF_@2D2`4g)QP1F!L* z5JWi8E;CU2Iu+J8PkTAJgO_`lSqXG=^CFZ}C(k!9+5xr*awchj3U<+-o+X3C$Z;W!?bG-6E@H(=|vf@5tMp1|8x4^5;k#aqPM};s(3ZI_j z9C*(XAt(C0oU<9d5%rlTdFuqrd8@>b!^is${ta^Z+6U_aXC`-RP5Z7dXDK%)b5Ztg z+4i&oX3X#Pd9JjB!k(R%7OCJRPl`3lup&s4E^Cr-0cB~M)~c-vpSOV1J>j6}kuUyZ$b>-yB%^QHK~ITZW>|9PDy_{as1A z#rny*>zS22-eyBD9b(T8x9jGBi!A>J7GBs+3RbMj;vC(hvcTb!-bk>>a*^aR>B274 z#Rx>qk5EZI5X9?YCfp=6cv0N`+JF>k;kgkIh(MbL7hjM)iWX4zVf6ug3R)~{8``+j zI5&2K@>IIycr2i?25+R-%r38|Q=Rx^FUHWKvVhV{__bgb`R9jV8U!DC0|6YB1^&%d z%nguB9(UdX$^jYd;ZyaZ1lh*DZl8w>ImU66lFw%H7Byt=8V>{Nc-`@Jm+_tN=|xVH z-xAND=QgikNfi>;i^6-``S|nGC6qKhAp5NDgQ2E88#q>vlX~h}h1$i7(%W4ma{nc? zm=ONUyI>$8!v6Ueix`7V>Pto~j~yju#T0gd7_D~MbhOx`BPOC3eUE1Js$?Jiq;>ad z#~0lTGjQxUNh@j1y;15zr_96nLqc)8&K<$e`gz*eQB;uqeoJuH%=9#g$KtWQ$+^bc z`~|)sAqneQ)pEp=%i8<)VQ8dXUHdf9r)Oh%{`qc3aoO4Vc%TOQ0a|5IE-SmxiYs}s zZbsAiPOXJH0>H$)_mBHJZo)mMiBckr6Z`l=cl6_nprSAr_V+gXWhJc8Ev z=S*A!3moteQ7VOqgL3S}_*6XRQ>JFyPv+OCeouzQSH0n+eu zP=mM-x17WF>T`G{F0D@Cmu)ecxwMWX%ITu|9ZrQY0od@c_$$DHl=9r!v7vi%YXLPa z!}RQ!At?b?lY9Bdyu}^W9eV1InP0M1C@M=6j7(qa7`$F2_qEoML&m$UdZ$MH9{3-k#WTCBnuyH_tDu4Cv!n8?fgDMY ziToBLctnbYCyx_!CBtx8@s)7T=j$`jnpbAM`$at?MxuRB=BysnJct>};zlLeLKnNq z>LIuPyC33m*@}2S{wtER8@?snRNM54{yjNow03RL{_#mb?S#dN0rF(C^BRH>3dwp7 zl!KP%aTfGhi;Er|Uj z9V8DT?c<$!{-Miuk!7;IiT$w$(c-%F4QX#DeXO?2*_O8QuEv{+Y0#*X^m`&|J{_@Q z-0GeuA~-_mhyo>Q)Q})OSdc`DiMl5c9MWgr5>_a%m`!(gcY^sZnc&hk>#m^G4@9&_ z9yY6#Hc{sZS^s0y^>6T(*L@UR@Qz=4h}6yeKUqksz|rI)x^$ z@}l<|!THQ_|JmXmKi@|1(vFg-z zAh9O|3TBWUefF0{utl^Q968x><+FmSGOzCHr*&O>w|EQki=4Hf*<6$6BV z;)4->A*u4=UgjeSh4ol13t2+;>R;>64zs0B{9a<^q%y54t<6t@`mJk(k1&kcy0{_wcF&3e-`UqaA!%h!X~1_uA&(vq&!{R3L)(p+=3U2Q-T&D>Ma zrJ$(mC>~Wx>(SG3#DQTV*Md*wI{IQ!h)?j;UiVPCJ?4c9=Mu#Y<9Z=hiveH+D;O!_ z9}R+uSktJldnr5Vm@Bu-&3V{(SGu!6OLvbX zD?7l5L|lq-jR(&!4KeTglw*nwSA>I-%WfK8g%Q}6s_?AVgnTTgWxwtDzR9;rE;iI} zS6Qj z97wn>P?!O9{}M#lmLl8Rb>{R6=C6BGGeLW|*z7+_9);a6)MukyPQxDW-zx_U$*}AQ zZQPFNtU0|kT=rq{18T&T%8O#PSCbzUC5p0xHGnOkZ@`6ARlurXzbPK@GDfhzxqU?)16z``nmp)fSDkMz30#`DxO#M; zR%9j`%Da>+oBvf4UURL;^aKLuSh@ zwR#k0P$3r!Oi*4!0D{7z8bE)<3$R|o#+@I`#1djo*0;Ut?l0{n>h@h}f0Ijrr%Hw2 zYX~iJX#&z6ICn+}o~a5lgA8&SKw8U@9swORW!90|U$mL3Z^8Z&CBJf1#0uZDEx&6X z3KUs*5mlVnZ;EhsuriW0`^Aw#{dm;OLNT%yG%o?)6hsJ9ERgpL#WryeaV{J2SLx5@|IR3tQSx2nukaERfDEo8I^f<_~Lwy&h9NNsYD>k*#2T8>}j* zLIvFg@&hJIoKpD%-S;h`>w-j*@`8$&52-4$wT@zoyU*CQVTER;;IR>_jE{|(tQ?`t zFl%kB@8-};-;^ZUE2~I&sMb-+JzPX5N6y#Bf`yA_80-;|>d`Anb{A)|uJ|mYW$Q=^ zTQqmxflG?hHye^S{rJ`~Yi$!&BOT^!+V_RUWfmOyJ@1xcV(R6;e#<)#u9LJgCyJ`! zIK3^Egw-B1T+E+*C#RsZgd#ov0i@R8_qr;!4Yb;wQBG^s! z-7CJExt9&t6)E!VgB>WxYISgL4%uD9F<{ajgusKt1A`xAzT8DM83vLH9liOL4$HH; zjd18&hpyyyS=E>~pSlieLSq(y{Erm}5W36RQxCL2K)xi^&aaJkDM-@6;MY{6X*fL` z*lWKI>g#8&EIRV(>u6UOM5+EEhb#S$!_3W*fe+vwFa`nrU_Nra2!G-upe;%a6n^Gj z34;WOmqXUdZ>+pbzrH5q@p@mCsaN3jI+Ss`Q%!-=!_QNn(ju2bb*k&!%bv`_A<%e^ z7CyUXq9Bjqz4dM8?>Sa=vMq9XXXVBYW+OaJK@IqadnOHi0hkFZBioo^u_Pe~393^u zJA1P2+4#_i(SBBIgt}&;T0=~qhytK@?C58Z$)GWRB=UXI_QN3jS}=$#NfsL#4< zi1u`3mX_P-9LyS77Gfvl(TvX*tRVXnC_#T;Lx`CO($;`V`7J#3PU`Z-eP(pO#87r>w@A>l%GQvy^~zNoKF^eE!X6B2SUk{4E`zW{-jV$1d_&@txxm*P7K^qS zgjTlyz?$bw`94o=FV_2KekQ?I>sNnvKVp5HdNRC%yxj)J>i^;D9ijwVx-P)9ZQHhO zR%WGb+qP}1(zb2ewr#WX{a^Q_*PY#QtcZQiR=x2r7r1vhLDcGsUP@d-c6oZ<&*i{` z`5->i=8R$wWWR{$Y9ZnEWqtg?zl-OC|FZjq!y%@81_7M-IY1zZqd~XE8GjBm(NR@x zVku2KXsTuVLaJ1)J`JmV$M0chD}d;GYjSf}moy1=b5`kj`2O6S_M6<+#hbA-`@psg z`t#Ve(ZUk)kCW~GU7o5YlcA`&hVjxthEb*iGj}JS7+HS?Gc-y}n|g;0s78`3@=pbEAk} zQleh_Gi~mInodN?Xj33Y;Ps_F7T}M7dP;}-B{FEAdQn8&iB!3ZP7W>n>K$+5OOq_q zPjy2VXP@g&9s0leoy>gGTk3_1i8q3x!dbyf}eHF|@c?^~0- z$Iiowrr_E<+R7Rtwq!Vu0`U+;W)=b2?!@CWE`JyCQ0V3y&5s7b4md&UsJ+YUW)2py z7diyHGhj;(kX%fl*;OvS<7{?uFtwftR?1^D;WO?qfEUlabtImI=KR%b9&RO3L#)ek=ytI*34@LJ ziAScaE`a!BYuT$0K9+4!qO;F5(aH+AbQJ;|ES8e0%p0C0gY8f7h4o*uDDccXObc0) z{O`@ZlkS)16!z=4CT!6%Daz7{%9yl})~StDW%0Zl-r?iO=Nj)MVd64A@nKUYC4<1j znCxrx{b^kCJgx=5H-Y>0Cve{^eNz9gLv8M6wiTZ@@ESg!LV8<2?m_&^92B+hmh1KV z_qMSf4!1wY>2f=bciUI6KQpm%Ur!g_kXGZHA~}hoh2W5_>1BW>t~`Hn8IpLGV^Whd zOHzy2Ia0-~Y!O0qWPN+#EPUL*uq$<>^XF=>Qk!Q&NsmF>DDQw{;bpLMSq4!}#Q ze1N;!If0VsMjfr%6`G(lnU)R$@~xM>hnit{vF9>gat|#ah?%;M#-C{`~=@ln`I;Iq=Uh>|L z&p9oFC*Jyki?=DShxCiuvE@w8=iA#@zCMz>RtRO<^v_cRbIr3%m#ynWb4o_nRTBro zyDEK__w{c(Lyfx%mOyb3LhKL|!aV5KJa-2m7(D0B9wOr}@01;?)F!?UHch~{k>Kq08e;+X){V<<_SuE|ITfw*E6b32fa5fOM+zE z{REdC&jammp?RMcR^5(?aB8s`s|>7+;D-T3`}*L7j)k_YI5j#_xE@klMxlzemQ;*x zlXAm3R5NM5myM~g?;mb$ySMftAOpDI^X48-M!em0@pg-gQXDx*%>Sb!EC2kzQn_;) zW?EFyTrLDV5WcpbklXNuyN8qhbVle6S1X}h1?^omGxuRZ?TsWW=25ODq6G|vMfq1) zN)EGCVoVGl*18Lw*3|Ss7&Bo#M15jFP44EQew&-Dl?M2n%Rd9JZ4$+{?=jfFYrc_Z z<*MW8wVCd7?jLxc+q<35sfs@#WDMZ{12c}fZM}hp7|-c8$5t(dr$@Y8)4xz}pPS2K zWWCA`<{JY5qM3jM1sVfW&3841tzAiwiJ^9ghMRNv76_OLI0zEgpeb{gpKtf^kwfll z6%J%vNu23%SLxON%c?U60)oa)0&xKIRpQ7RNuP<;X|2GPGxxW>!wmakjOf?E25N1z zG9dCv31px{#n*gU5@CQ$GeEs(DlOuRor$(Z`kl*7ZtAy;7G3m%t<3oIS62Va(aUe zj?Z3Z)j1V0>b|*P!vqVe&#Ghbol43QtBTxrdueZ{vnM)q6VaJE5P<5eIWvho2i;}M zJ=~eZiN*1w#xVGt$Ud=d-T9hFuc%2 z%nkvcxNG1Y#DzdTYr3qe&Tp`RmbKeW&HefH)J{;<%|>z1_vIN3DO-OUu0;1zR?qDz z0)UwZsP&|%zI7raeR6`Dv=ao{)X3=Y_3j{UOAyyXwZ;n~MTQK~kI{ncLlL7^n9&L# zegleA2H_}6zWLspbxTS0qNX@QJ$dI|z!OBeW9BEnqy?gX)C#dKg*#_~|7{pydgA-J zI%Ey~@jvSPmpJ4DLA!qMtU5TSrlds_&4ogo#oAwJ+1X^e52Rgp5*7wlSw;KXXbYJ| zEEJ~;2?pGqAy1%!9m&oT#e83OU75HfnD6YmbQC>J!z6;8NnJ18_^HWeip4=1l9-3S z&FI%S_}=8@wXfC6k<|JWQWuKHGN4@E;E7gJ)GAGOwW8DUqPe3H(P)N+;$I=q%aGl& zQ}@q%d+)dPh5Vg+?>hsmrlL-XifxmZ+_?#$if_5sJnTlGb0#{Vm&@s6G(m~mV;WXw zlpKSjRSOBLaI%brgAGGND}$Y5B+NL>n)Z+P7SRC0-nAXu>v&2G2i;Q^iMhyre_7Il zZESj36f?_ZFHz&ZVP<9)t>)Xs3n#5CLG~o2DJ$Bqw__cC6MdT;)H6B)jBn-VuwBg`4gmyAX4l_FD0c`3cj01tcFo&v8^>psd5Ah5hV8bW)+*gV zoY#p>Ptv6Tf<{z(u}oXT>ny{2-~Z?juZHt)o*@4il6*2Gm_S?0JcxTuAJGQ6y^^01*HhH=qJAUj>=%EbQT!JIG zT5JjPK`?AGw}ye8lbs7(cTk&7%-(iemOT=jxjT;Qs>T5Cz8I>$DsQXMC+`M!_m{>G z^OhnYPIljWNYv^fjQ}U4T_&yAKIq6eqZtU+J!vkWj zvx6fo#xG8^hv-CPYu0<*Dg9dCiJJdI5dw&6$`CYJbJQmv%F0R_7!d+bgaGZ*4m*5p zwz)^y2OG^+78j?z>-d)#61NfoyAK~g-G2M3D3Srbo3+m++1&zc!gSx~ z$@!nk*4Nivp(S>HRf4#Hq38{=*yQUkrxCHC@w5n7YRjsVD}P{Wo1)jj)FOs=1!%nV zBT_%)Fw4$F3^Pf`tK$Iw-V$2`pB^%xTh3~<1W5I%7XzE)4L{?%Q&xc_-m&3P+690b z3(=fRoa}em_7brmim6R(mQAuhbj7TM?9q=jNuu z;j`d>x4oq29p!Q2#}22_N})*^Sbr#RZsfXv$?i8IY!op0Gi_YRvD>JpLE063I#nO$ zjMC@`r$|=0u9pugJzS@zs5dJ-3JH~B#xL4I+6SSW75Bx{4#4Tq$3(TLRR=?J86T$; zYu{ASeqIZahrphrHw(kP7#S7!dO!$8ep;R}am1Sgn?R2iNdN~1B-H=;3;7)wKanZNXlo25n))T zMuOA_o5e2W>A)-^muYT*vBE-|Iw2@~ZKQz9_3K`*6PfB?4K=V{7ctYBSf`S!NOX^>znAl`*Mde%Sm_!`RbAiQ zo&rcY%n$WeZpT^Nl}LGKM|Ed)c6-gASV!dWmzb;dwPTs`>o$R-!O!Wa;=6qoc~-+3 z2b0=>q|gFVhE$?((#~2PT4RCOnn7Gscbbs{aRr&E9E)O&a&;th`HbNvx9!*p>uc}L z->X1)t9UO|EIIsbv6>K%(NZR2xi|P9DWd_CH}pAcNh>@iLn)ca3ahq8Sc#5+f*}?J zG2Kg9DgvCm4%Rlp*1E_d21Cg?eU4=|LSXpNZX(SG!{tRr3Y%kMl{iy!DFA|XP6#@R zrNO{RWBPd*{U_BL2Djz~xG)Cmr4QyF7Qj~zXP=tnW?1&t3=&ojOYuntTaM3kd2-wz`t+qA0< z4oMk(K*KB*)$~heQ4* zU+jdw=x=rph61Bd3fRP&?IRes7VN~b{Wj>R0m)%&_t_@5cjAik;!aE3U0XxJp;!0O zE9qX~N)x#8p6N&^ViCJChm()POvvenD+`~!g;ixKtTCYpGQj{pW)&_?d97!^lE3d! zflI5PmhIm|n_&~E!09TAajP+{twz0yIJ_IK+C9RxK!#ugMAwoJF&eejfbAJn`@>AB zYKJ}T`-+gAa=eoAya1I0c9yrHomCZhL!O__%&Zmkx}b|OGhass1J_A>HLE|`(L2Hu z|M}%Bng0)~gL5roVmvxh0UIsVY;d=WAc|zAC$~sCkt4Pz zjMrKFJg1bdrcN5OI9No_X4MvqERk*}Y=O^VKG4BM7~5m;%TJ0W2F6eN&Ea5oojf1H zepLBMJ9wwAKC%u(MKf}{d=3=E3RU0*2NKnmBM>~E+=qY&Ltu)H{y;H>ePqn?K&ZP< zLo@vnrX)~#%5fj_ZbD`xIkf_u<=75)(5jjf?1M#Rclmi-2|{928-GA8$X&NhCY#o) z9zKmarw&1lZyRg#iq>JkXu#mB3Wns{OnmaF ziE;PzqEDnf5HNC!d0FCT4scqn?_$aYR|Vq;odcU12hD4jB8#KXuvvK16(idNTM)(o zgXNA{mU^u!5zyidC$jR+!3qx`{P9e#+=W;A%97m$sVM$-b$xH9dGN7P1jN9f^HtQT zeBZh!BcuEq+^lV@<~s&2C~34H03E#F*4zsmRXbnN={s{iLXeOQ`F zv?3CZYI;b(3rS&)0*8yaePv$oVl<@CW~O!FM-+c&EeE9erH!TTu(Mt#zkM zO4ED5EHP{;c%8dXlPDN~fSCTrP4(Ijx@OAW7^rGC1}JiO!i0d#@u8yI$_8*v8sZ!V zcQ`*ZXjgwS+B+z*US_`N${V-kT5}9tQcr2;%Jz&4SvVXv<*JQ(FG^nQBR1*U+lPhn81nD1HWrWM9G3S0kTf!!l@@$4ytY5US zc^l?@Sjx;azC8G({i4+wlXi)+vo_>&2fgW!T_NvMvWc`w8pU1gzaC)?5AcvSzPz0>)jNevv5sX|Qhll-q zrzObVbEoj~kGJy${(3_}wIB;`BCYr-hs{aSmGB4(CVmLPoRkEgURnu*XG;{-J-8MR z9o)7{a1}ble+xG{9%Vm)8pn+X0$ABYK5io3(e;RPG_o;jMJgaxwbpp@5{p#!*Ix?p z+LT4>VP`7;P0wy4@BHyGKv<6(kpBxla@o;Mtl7EF1=`y}V$&W** zw}EacDu(F)-b?*#l!`I4t`r z^Zr0uWV4=LG58p+dQkm38l1T;l?{~?10pBs0-^DuBaksqaZGV~5*ylI>0(ZoUW-c+ zoSHlv%T^a-J)TXM!dNX7j1IHqJH^{Qs@B_9Ar|_XB-t~5fcPv=qjUd6?+XHG?vQj9 z@ohR;yT67r*In2qW)+m3r}-R`-lpL%%f{TDQ4T`_y2MQ@fbqMtoCnua-duNR3NDu7 zX~D1F*`B^%Xf?#UKT-aAQ9){XXSF?fTn~RuPG7}gQj#kKvnO|jHM-woX{$)5pFnVO zgrv)yl++Kwh>4%@zWnSP9bM2($7~;GX^EjSdj4 zh7(okQW0^Iq?n-3QGi*zj^73r3?F)32-}(P5FJ3zs}uSfm^iC@#iu3$uB}c0h4ASi z0iAlAb05pH@BBNTJ%p;V6y_qqojx*8iUWauLRVOajMj(gAQiB}{=;&N|7R;wt5myY zA^w#t^a=w=e{aTI1C|ZZ8yTnqOhz& z+0DGi<4yQ}8GjEg}eYT0P@aC&g7BtyB4qqs;Yn3c5`Tud=TYZ!T1uFUThUc~$ z`S1JjpU;^zYWJfHCE{;v%($#CC3CKM6$H>{ch_?yIIt;vdS4n|MA7fkebSvc*|f=X zp)AnS5^Q@$Oyciy*Y6owMRDs1ViOm)SifRB4RgonYptc+RCDHSFIg{RcDrK zkg?5boZ?Vq7EiXK^n9Q|QF}^IGAc>)AtD}Pf^$Ao6p{tDOv;CUUi5pPwI zX3DbYm=$?Jd&s|;o1QQ(F>Hp&UywEjJ*D!I4x#liqLT~=yx!n#AOy%eOjfL-s&%_#rbgaR!j4_gq_ke|2Jey z8S?sznxsk5lb{KS?6&3Mb$19)t2fwpHG6wxuVXAO*UNZ1V`&Y^R=BZK~ z&z;E)H*vTu>~98!Sw_En@hW!;-dO4Ep4hyLTB1E3H=BnQOG6o<(gFE{P_r2f8I<>e z`=gF4zi9nE6N!6*nTXfO!F4NqkmP~0V{Etrqj4NYnp00|A*3N!Oh7ILcmHeMZ>=z{2#%iRy9(n1W$=&>twf@GU095IH2bj+vOQ|}4vUs_f__{4O znwOuSI{1Pvqx2Z*vX(mQRF4Q_p`CKOR?2#(1u7mlE4 zvSwHmyWosot|eZaJDj?|1+UVcC(1hFeEew6+xo;17AxO(Y6)>?ecXDZ|2=Qhl9%(=kte2v zT2i<@VFb|mMR2!%7xVLhySHhOefz#6a;EWjGbU3xI+5cVJoLpBFT8miXSM2gC( zppVGh96uOCzSf*;+CGgtd$NBKb$*bTAf-3L)UDPR;t0eMSSCB(95AH85}eEf9ZtXg*Bh^*#FCi zgCYa~z0k%rk!^idIauT#0n_pIwv@B=)U*_KbY#Km^Ve||7Gy*2H~^H?6IOWNRYT)q z?cCSb+}+wG!Kl2H2q7y||2q*HL0J^HYb0hHDeHVSeZ3ifao$)Ck1^#*X@v~*AJKkLZ#CwOBlHZs`2QVxQm{^GE>Mq}O`j%`)X z>p0s_o*!xWdi*a0>|XWzbLYmGWuYwtK&#BnRX>lBXixnwkLFRMW$n#mNoTjKs(h`b5`zMXA3Z8#9QK)F*hxJ;(QqLH&9cQaz3(3Oyhvbj`=DFJ&}s8Z*5 zTeuRFY^BR0bXF#HuJIfUCxMUdx8HQ<<3yafTqn-7dC1T#6k4@$f5m1l-&ffLX~5GZ zB^I3Aukay>PBrCKWsX^I1_;2wdstP)H%IVH1|t~9zc8>pH$2Cy<<&)BGxJV0&{4}JBGJnP;W-O8QQtWw9)Tm#$xE=93{HU&+Z=!wM1){TKv zYP#aQU0xz0{4X2#n623X-w)<>FQCiS0xb0gR8@f#%knI*20KebI+r4Hj0CAWlN+5D zm8#F;|Ib!y*#0M885zWNL84?Nxg8DeSV61YJDi6j6XtBLIA5p~e4dwtF{w!A;?W_E zkaBuB&vhcy?*Jwn6P&)dtzlcyQtWh)G_>KT6vk}T5hI7*L~ZP*)yhqXq`MaQ0{8N5 zwoG-(EO34HECL!K;a=z@oHA${&%j96Wxxey1t(8PcTydvq#Tqrk-=LK5JxP>7MDl) zHwAxtL9q1=_4>5WF*ot1v&=HbS2T{*AySHCA(|OV%aoiN5loUa-n(iso}lj2(}nnDQ1@|z81H{)EkzpIaHdcuFCJ;^BNk5?_ z2-e7+#RfZl_6M6K9#aAasmdyF9p&;$WY34~DFdD*>qxoZvJf;xd1FD?ds zP#c1hk9RUtJ~MsF#L}|Bh;i61Tm;F`-p~-I(u+wTHRRM`-pv1Y5VgdHjt?uTfMzUO zvok%3%Cj$lq+1FpeL+}7GTn@E84YN*_gUWaDpb5)4$g!rMSacR{@>zd4?nN^1+tAb zL{-&Ie$jtvpXI3j9bl?>Afm+re8BTKJWu~i}Z_V`CFbzY(Sz_?Z zb+oTz&Bc@|4-y-}2RX)?^kSfjtr6eri6}oM>j@(03j_A6z2}|Y>vh0v(PUN~^?X(@ zxeGIEm7EO}{Y#wk*)Rs#H)5(xvoh=?A4&f7EAUg&2Kvoc8`F$CM_;KM#&&n!Mc5EG zyWIM|bv85rJ3-MJNNQk3VTZ=>crP0GWcwqesZ9d3vqL!GZG=+iTm%mpy#3$++9(Eo z!1vB12o#6&vjOiBU0&aZ57(PEXUE)O@;4EGN`qSzx4O{*d1QR~$ZyBcBgGnw%;%Gq zEu+`)Jw~+Rf1P(JDB$mf1i}IuAZcDg9z;%`xo)2X#MJw=_|wjpb;h7bNkOQb$ia=+ zgZoWhx37X{w~A)^{mw5(zin4wmf?a(M9TnWlMd9N1$auf1~XgZMHf2W3K-mpn~t`= z6Dlgq0bRCN9YxTj;(0t~4wz53a(_Bs0|zu~ZdTiSyr-Dm9NK>Vb-e&29*j| z*>QyAW5R94fj}Y=cnvrKpV~B5RF_xN=_Wsak6zL#VPhnig~kXFQ!_C!5#5UpM~=RX zc+T+L=iOa<9w){+xb957dq*wko10Y@@*zE6`A={?7Ho+R5}Z%)FZ}xg>~JJ^9hp|| zjm3xAh&vSHsxhs& zbce@51^{E`sAgr+Kq6JKh)R<4zQM9hq}#;xmLkB4r3nQ5PLN8Jl@3#^eW)EgL<3!7 zB+i4cSt_V!fe%R8<(G*{JhXrf6Ws?4wT&AR6N~f6?nS273G1(-b(knh(%X1vi(`$j zYVnL2;hK;K4VF7FX4tgBLEVZe1%3GeK@I5kR@lTJBEAdUp3gmI=;By?SyG^%Zu#PO zkTi_ZnE3%E&#IDul7OJa{o9y5uDp8ep5_xMK`B6OBmlS0XsWNMTpLa6xseR-^5$B3 z_^xH-3{<-dDZDUM$ii4%mqu;{TL_`C$WglJwk|4yfTTFIAcGrob)7wJ4r05o=i^Lr zQChbs03@r2QQBRGR<02_muWe7!96wRLrt^Muw{6gNoh;RkuF6nE6z5v4v7%cCt-ef zmY5vr<53+lf0wg$ruISzIrKzdsVRaJ6jV67Z5(7+A~tP*&|36yL%T-u2ge(%CnhVM zga6I(Y53qmfTntlQKb6$X^GmHupz&jN9oK0WnS_F^fjYHTRvU_>w8^A4sCAln@!=Aoy!He9$YwVlR&Er*ED_$MN0677|xNz{MQOXFErPo6!i>C8HL-zQLo;QnE+7KE0Vd8Do! z*@N3vNwPfzLs3lNU4H3Tq>Uh1P1k2m zLdL9LDMBa=LM5U)$8s-7Pamet%PyYs0y(n5jwLyP0=Z_3dQKrSvj9UiXcv?M4JQIM ze@Yuo0s|_*b?*>r-DMz7p;v;9aB*q8+MmH~yAPrP4rT75Lex-@inSj3nlx2p2 zWzZvFAAexvVoYnVhLK7B$U7hh!|QU}nuTaA6B+W8>F~w(1 z56^P+f%F=oNG`FLUEI^ctfxU<#|u=AG8}nAE=jxN z5abI01cH~5!KmpPv*6OD6CVyymBPo=`FD3uJxN{D?}G2;B6x(5L6(6I#cO2C>3LYX z>>9sO7u{w^-ccOsxUKRJIDuqw=+i$AJB)Z+0*5o@bW@LD7X^@!#DY|Es<7E=9Ztsr z9@hv!B4b)*xbd!jy!0*J$ed`oJN)Pl?!kkUH78b_o2|YxQSVVhpN4&NwmHTTBh!=}r{-NdFuh=c>-%+Q4Ax9Nc>_?9DpaxW{og$~oUK)GZ&^5}Q*X z+v5;0mZnQrX9E#b$G5&c5$#ps&97D3W-;pXMj!O2+Cx~ippqgio=AbiB?ZwE@$cm5 zxkTh4_y<&)-SR#SXH5DS^q!s6&(C+e2I+F0d_*b0-_L)jq<524^!Isry~<-%fRcyO zN(8dIg8UpllOI@_6rz`hs&s-{{Lk|zWNYBce+sZbxWH?f!2kv{2T(aaKmkz)AzO`L zZ?8PcnEj;DgTOU_gDoMeWF1EF?g1F`FD{UHIiYWzZ{!{}>e)I+Jt$2^v$7#WWRox8+UPb3?3Rk*vcq*`32~jV!eKBks zz5&1=79yyCB!mIkW%?LoT}m8GGPfU24&&Nh>@76(y$pZBKDjn|!h%?kTtU*ymF9!x zIlf#f3E;2=+r*0AH}HK7KL)HIjlZlg_KcM?Ltqn_B?i@Zm4Ont4w7!`)F@DX)&|b{ zQSp67Sq~x{;4VqRVd}|r2(Sc;HrCAl*5P9`Z2}pUg?o;0kJM~uyS~1ix#67-l*@~F zbYiJb#B_fPpT)DWrN1$@ftJtoMCMXBRR2MFO9F^@CaEB8vveluMXc;Ky9EP4?c;3% z8Cs@cOI%|4*V-1d@ozSpdleZJWIfB?YRx0`tm6FKM)rKkYip);R*9(a$>m|-_a@MY zlY_`>Vfqtr30{ADwgSH(qf|Y&<;K`wXFbQxz;bS=ir_B|$%XfPWy)w$zvB#sLfdsf zUq1PJ53)fcpW^PZtjw!uVScf-B)gG|B6h|z7%=cP^&2HaM!rI8W8c{|vMsYAH zQ8@&dOh_v^YQq$7mUtKx-2xdp&<6?EZOu&VsV{=4>Yn2Wvdb|g=;u1ITyh$Y@Q?xa z+`g+9HJo_l0DNh+!l&JI*Axzo+wuak+O&vdk$0gzc~9Bfkha+Ub+Nv&`{sLX{`2Ju zzMVm#y5j3*#X$Bd_=z72Um`W`yShLnR|RbA8!$^=h?0w<7J`!2;ia1(I?PlEV~teQ zOC%4lvkLRPHI8&w(2)zc?RdvaPwOQL{f^RTC*{8+4#=Er5c7}C)y8mu@cvQ2AuKFH zgpi*=9$pWNgp!n~4GlMrGR|@Zt`;XYayk#%i&E9*nk@@&-+LqhZK@rT6j3IPJ_=Qq`)A<1x)21!$Gs_X0k{hD!5=n%Nycg!4my8^n8HL35JwE-QNw6% zqB3^K_HmV=5)p|o#=&xoaCw9Pr|og`M9>!D#fWzK7K{3P(lJ#DZVbe6_CipZ{K06F z2IOm{5iGN-1OPv9AHUr3#4)4u0l~&He!gK~^0+N74J$~Ox)Ac`ErOAzEU-PgBu-a- zTlPRa^Z=j)8i}VBr|QbaU|bu)+30}7fDE_*FkH)if7(T3LM*31CiP;4;fcx7rUW^a zd8+Pk|0+-J^%jV}{7vl=}UY(s2M3}{?yI`DWrnJ~N2w6w5np^Zbh>|?anG$1Vm0R`qR z`#sg>MG*;f6z1aDgl{n0l3wO@RurwEsD^=&TX=Mvu+=bV(d{^$>Z7q05kP(;RHj_9 zO;nnP)42dMM95hxE&G&y{@$uxC(xD`+x$$Gdd8~E@rKG_+E~86oZh)hEj^SJlO}xB z-)m>U^1pXuB?{D}_L^Q_odTBJh9qTeK4otzb)V~wbJQ@o>yEmaEi3x`bVAFE!S3PN zW=70piYysh+`PblI0T78PX*;csOCG{cOR~2=DWY7wVDo0uU7b&s@KUaO^DjDQ&^El z_ypHIV@1xO?+KZm^FMytJ<$4?_bQ`rr3jw#GpV47Z?&F6JJY29>SipRU6g$_GitJRM7QTsZUeh zzkM_M@lqhMJ*C+6;+@A*^L8{(Mq&~A?=U5yq;VIlOutC4F=?u0JzDkI%Ric*@Ltn%O$WXev) ze9V25#_!`VZ8}Lj5|x8zIDuAO_Ul{c9Cuh41vS;5&z9yKrguC5mT4R9tzDSq>@uuC zx|k36EBw5wITQ{Tpm6AD&d56)`{GE9zBb1A3L2cG`F)FOzuUOnENZD!1B&HzNd>xC zpc-T)h0#Hy!`DaLke6p#^r8!q!)r9y{uuIycQLo0y7iJV?91bKBAVc_!NfkOs^AC;6iF-*GmwZV z)-R(32*q)Z_R;O^y(p)XvV#W_rUPq7uThECGTSx*A|v?;t(GN0Tj#Xf;II3i5Kao|*KtFj^ zOl^NF#B=UJ3Z5kJES>yVy-<8U%7o_;RM0}y(S_NvMji>MsqIh1p}yys?I$ z6&@Y0SAp)S9FC$F_}bZnHE(bpp$0m)ha8(#^5VmORDln(J+lr7QdQp7xc17G=&%ia z5p8x~{hN2zy4#hXPh-)h6DflV*yGjP>g|s1y)#^EXC2ex(lDNzl%Z|6>ke_cm#8DZ zyWGNheJdLXZy!%-*Dlq$rUV1GY!SkJP5z!&1z#jT__=p~7I<5}&q}p4gli%XoQ{^6>Kae{0HdPm|YsTB+LgLV_yS-AIkaYrIGI|P? ze*p&N?&i-JjUN`y`;9$B{U?G{|A5-5AT=m)l`LE`vSo)d?3inz(uX%W1pwv1tza7w zKnK9a^+u4*zSQfiw^x5D-2g00=5tFh$*yS9ZsoW#Kw0OG+Q>9XE=jQd2#H%*W)Y1D zskWUSDf*etNp^El!HjPHwDX446!Q0^^_R$A2_Ime+m%0l+`xY=H?O9|ZRp@Go8;R6 zZ!iKa>fOZba_QEcEjQmw0;jWOWj)6jFVD=r z$^a=w-^>%-`a`IU;0y$S=cP{AEnVEuw=B!Ss>JWJe)lF< z$K5Q5B_u-HGy?=ar(&`%DxydT8GJ@%q>c#)Tjbry)Nb*IdJOx|x2*a&(F47X^Sy#$ zidFRMif}Ao4T~h1FaV<{yQ}l^&ve1YCfwaxL8@G=?098u2l4&nbn|lU#Y+x1TLv-R z#fk_|&(Cd$GvkhQi5MKbEE)>!ZbI{jSE0~8>qvR8!zA=p_IX7Nm4tMqqfwtg?|7$& z$OR&Kb!XM179V;Aw7#1TmMp`cl+-R%0&W5KQ9x2{SSLo`WHf0kOATz;)<5D7cLCg1 zeukDArRNN_?69{}I~?#cI|t2EGF%22}p zvIj~6i|WL?pGzmmXLgd5YrfzvG#NyHjut1}Oi`7w<;>SA z!32!ARnaXW(^#lwGTu7vT3o`z`>!0A7-l9X=C zjd!#o5dly}89Ee;ClFTye4YwyAW5a%rh-M0BZXUO?%3Z8@uvvpa|i8dP#Y480xf}K zz*A>0ps}W=NGrnTWYCrTmGX}j0<4YnD0!sLJb!08N=1IdqzToJ2)TPnC9kvi$i|Jn z09NF7;*+{#q5t32IZ`5UwLd^-Y_5up|08*t{of||zmNJKP*cPFc~)$#Xe#VG=I0bT zOlqdJ6~TruCn2?`s6V^0y6$-P3Qd8Mj4Y#R>#+%_w)2Tt(KVgTh)Odf+fIR&Y(snw zYNKb?^W4(3rahxqx~2q?$fd}s$;1;v!>pv6xZi6w&q7isJf5qXq@J(w&+t=p-j3(< z7JkkteW^g`Yl{am(8SNi*9F`Q>Gy-CH`9#QS2Zg*yU5SsP=l|1t8YFKyn11rsuALm zAY-&&UZD&!2U*z;((ML8VzNyAM&Gf2T=yDct)R%RQU8$(VVwe{cmX2I^Sl{OhUo0A z;C^GTT|2~khZV&xY`eya7+6fPSh9ioE#ed{gAB!;)`lCjP8yhhs-i00lMV*FqAFQSp}pGZyWJcavW3~%+OmqGk{c)h<>C0%W)ng{Ohk1JTEaMKK`$JE~!RgaN^!BwngYqk!9HrM%ghyjAKiy$Ib6yl4O zL)bd01_=OT+tTIURTLdH8G0nqrNf2`J-;gj>~Yt|M`d^-BSjL*3+vL)*r*ce&lnMY zuGc2cg`!MmGLm%mti1YBeYhJ6AdW{#G>0Hq)T`A(pW9+saO&zEnANHa8&UG&+dAmh zQSpfWzqei94;7RR3C*}qdF(igPfgU&8Su{@zhFH6j|x$BDaGx&EK7(*z_>%5Qjd80 z;S0aMSpSNVFl)=^pTFzk6OHUw=yM**){Vgcv?}{Dod1@*jEGhBTc%DS-mbS+og5hM zHt&3v7`Awfg0=X|tc%k~E95V!hJjjTiW@7xP~snT=j z)EPjDC=xKY#a@`Nj!jl*AB|31Ru)HhU{4YX5fGJe_A|n%XgY)fi_*GKSb`Ah*=EV_ zLcxss?bR8VcuA7OulvNHkx10gWK^sL3K47Z53SXehDMi%+BB;u!K#7bgpjc7B-PnTIpOtfK)UNNL}7-u zjvVlA5e-UlXMeR>9@mRT!`}jZl+!0bZ9-ZX-HG%utc;_fLVOI9fpnvi!?RdL%zzO2 z(<4L~++$!!;;H_>EP#`sEJo*TRJEB73LFzVLM*UH1bF-;V<&wHB=%aHplWsz7E(JX zT`xk5da-sB1q8OnM(L($^x?Q?I`6V2{n=viw!mzI*&o0vN5y>v;Tcj*?v-R>tXGL$k+SM5UXNQ*mL6r$H!7Ht8=tn*vcMh#SFO0RfWadd-&-PIW^5fefw;634T zH9sP+0*TYA%6PJ@zdQVKEdh+D+0orw82$CoaPwNMcMs_`x44N_@G%og5Z=SxL`fL| z#+W>|JRIx*|u^G#KWy3+x4bsnRgv0>}|?}hnM&pY}`d&JbI ze>cc91PXU_=OLy#e1KCQhXYvYpvu4vqji)1_L1FS=y3aT= zl7ylx@cqm>kqSzkt_>{Z{?A-}qQ=G%>}JCi&{h=2qS1`RYShV*WJ%Cqn4jwi{JDr_ zyd&EuH7SeTK@k>E*M#;|OVJhaqxZjt%+(uZXv|f$z+-fxH?Oq+8PWAMj;Sy_7OG74 zVjxzT<{?W4$O!|xk9$>-fEgtSQ5+hfFN1`Y!VKkct1bN84w;(?`qps~O0(q2DHC+3 z0}e*O+9UekO223~fHNPZTZ-1E0VDV`vguTFeR}gMo9iH5G#0~b2tK0Rz2iOG?ybe5 zMmeR5}-;cGf$$`|ghgDV<3I?ODp=6^7a2Qtk4YUJRs48X z^fyy&U0BQnj4=OQ&g5}6^6sN;R(Wyj!4C29h+B9bT82u&u?iojBxd5PH#h%pkAJLe zl!6Axs{{;xwcBAQvZLhJF$|%B)Py4&Y)dX|s49o&4D$2s%cW6Ze&wpFQq=RjK6aqun_u`zx(~2@L zEvwL^Pt9k#Kx_Bxr|nY-p*9px2&A=q+{Gf31yvlVoaS6Kdh`ed7jnFM$$N%CK-KOKxSH0NrGO|2)JO$)>&m<(zW*)v z*FV|1G^@^6*T~w>bUWT_M8#yQKa{E|!H@)T3_e~OPw|(BA$FhGv}4n|sg)$jjJs4Z zT+{W{dIiNLsGHUeP~@I2WPjXQ)~YNp*czjx1sCfPLou~M%wh}6;V{}Fc+vL++6df` zF=qN|W8mHYm_YvY)r_MrK-67ZL})36!*A|!$KibZbM4H}PMkLF5982m*Zelvb<1?QKp3$2Zug(GJA6t)tZq#F(eH_nl@`46|5mi`=7?@$5G&KvQEhh8r&%{Chd17F+p?m&*^q6d) zMg_14tq48bizwdUl4vV2u-292@F8V07vHl+91#Ns9JpT zt@3b*E^Q=|?5+k@8zF9I?ZR(6VXEF71X zyg83cUpXI?L}Ll)kj=h5ClVB4pnmPt75wUuY-S1YVYxE(?Ik?LQL!Rby!4qU%|ZaE zOzLmHRAg|?P9q#wc1%4U9BjVm?hoM}DKvQLPqT0>gh4tO;Mmgka)-kJvBR8DI|>aQ z6L5F`KW)gVBOm>SF#ais-V1s!`E=Xa<1LXltm5ltzF+!n^BgZMw;kYHGnnhwE(0^w zF16k5((h0hNg;Ya5WH*E_x!{-tEN3}`o%z-B*3SoZq*Ex?{2$XdM#oR+3<`ogJTKq z_A=W&&CyT%YV`$p7rE|F-rj^tj`u^g2rx_1Li=r_V3P%96hGwGX-0lxBkbUH@e+g^dX3tNx)*9~# zRkhgvgYMzexX%g?Gx@$wi#6nJ9?Jg@p$G`c>)TQx0e8m;oN1a;5d&UyhKpYM`Ml~& z$5&;6xvC!Td``kK!qU5@5eh&x+XDSkBo-NCHYpF;r%9gE3M0PhW-w`#q~vY45VHN^ zX4^ZA6uD`C2w3EEbOg!Zv#-^LeqCl<$3%3=4L#L^=m51)aD6y=oWUzTflw7&7W#(u z;hlm6vMon`NdsXhW&!;5$lrUEk1mq+=AZRq2Ep(Lpr^O?mmz9->!9UozJsCR*_)Ep zbfiA!oc0GvFPB`ax8jGrUFzzK3jfSoeV&73yw`3QW{->&UB=qsaT+c|t3)e8t z_FtXSMsNxaZzLVe$=i+UnGX-JN}>7j_`NZzw?19h57(8mE(4EN8wkK&(PkCy9}6{Y z=6ASx%4x_;#cd2UG=bPf^H_R6`hm)`ae=Rm<8?ex24XJGp_6c1j%S0f;r{f{66z@5=wP%!aUhIwFDMS7Va++(doF7)Y|j9kI+sC3L=-Nx zbz(ooI00sW3j)(95)nuGrxDU4qFAw$GA242^tGJ+Od^}oen&wvqST8vLstNO(||BB|#!{?FU?OHCBTezOD2XXd81c)Caz zLg>ta%rY9r1hBhbO*$TUouC^wbD>P?juo_O&f9zciOc+^N(E17XHcB2FcE@kKw40mKMWEqgj~ZG||l=&qhze*w@_yg5B?d)bM;5D z7f&SF)nGt4AjHlB{@OBDI;n}^ImSuQT^e-`A)+}SZ%7_;1aE&Q1%?~}>-Szqvsj%8 zoH40sgGQ)A$EJ}P1AKwbQ;qbX8P zK^C4|Ih=YPDA3fL4_c_+LO^xC{qc2^8YBA7HA7K9RiO!%(Y=jI9dSgxf%{DT;JK-C zl6$BJkhJ-L;~H>aC;uPU(%L|-iBa5VSpYPLjV1Vy%L~t!bD0rSFNUAhW)*wwct%bi}WB{>}i$dEh7 zjtoQW&0X(eI;8lZ54ypBenjl+bu=AqiI?NcKGQXuV1_x8%52;e%YRS2F{gNEWh@My zg@Fo{JnmT4jqP#~r$l2T=(@p`w%dh?##LQB-|?Q@Tn)=AvpQ0eGwxi)NwH7aY$0pS z?|fT7-Afx~<3!_kZd=;`xgy@3(jG(@P*DuuFmoZfxQ6tpgj?VT8aKiN_X}shN`E z#EnzU!QbkmdNlgrENfBV7F+2eTT}mW5(6Fn##fg~CgXe##ULQ}OwjF)6iq<~|D&+m zQ`*w|j40;#Hjv}>%w;m_Y!{y@h^9XTjgPJ(vwC5&FIadg-GW=&w>7yOMp$74^kbiy z54sfA& z494N@)~!R;4m^Y(AssCBv!WQ>!gSa`h~H4WO7O1HD|=ZX34s(+31E;GxUz{HrgjhS^><5WoU6I5^^`-Oxg61LghYgGMDX2ZgL*A^Sn9Z zIz0L+5@W`EUqOs$Hqhx#IzjGR`MO6Oe&yNyJTW`oWN0;zqztzmNBf=e8&(u2*3-iy=&o-EnHhs6^fzQN0WUgYesI zF}^<@9*3T-%v2|r$isM}{jZR!gZ#LQAAra1dW|-|88a;f)?(g&DJpP5R-zX3UD`Dl zjjVZDGB?z$n2CWGdzbexHC+jQTz`!|FcK9b%@0R|7?1E;oufx9?KNvPl0`Y$fpi=) zoMyZWBn`@5FV5DHjT2W~=s z&9j~!pNev{+6)30?tzikCVy+UQ&%^oz2VTQ`^V?M-CnOZ$r8!|V7$TZ6`t$Yv-MeP ztMAK{5qB@B6pwS>1uL_|8$DTK&`^}ifqrsKu#hqgp>k>_G&s>#u+(MtKaZ+34xnma zEW#62&ZHZRv9h!j3DNt3`shtnlyW#MG67GWnk?g6odUj-jE~!1FXIT%B_)@Ub?s`@ zshkM%czyT=azhSdgT1PhhuofM-nfR=;B5%&-?c{D=2BUuwmI~EJ$eOghYB!sYO-J> zGj^M+_j_v35Ep#ScB|fQ)jO=dtFCP6-p8X1v?>G+AJ==m_a-I$f9CP-F%s0WZbhO3 zsRy$p2V&8!r*IW+1SMwsZt}x!8t0c z`=g7`%;zn#X9qTXi*LagqQ~QQ9Z`_7)7J)lJH5x~x|Yh68S5T`q3C*x>-{#UuFQ1x z03SC=3Dv$3^axr+WgZWWNKRTQddz~&be41nMHkW@=XllP0Py_?AU8pNB6vH?dGstIOPRah5ee<@&f-=>oiojFE&fQgAV zg;~<)qweyJCzSHd018%EZx15(5e#p^InBwK6$_}t^daE-mHI8^8*wXv*!wtjt%gmR z8QgH5zvBd_krb0QK0~FkSR@Expv58zs7UJ&DE9|smha{FZvKOjd@xC>y6Aj*M2Qfc z?EDk~UbnZmaRJTw%Va?bw9xv(as8Bd3mlGZk#*uS)f(9qlr zblQ5B#Oqkf*1Dv0guOq#1WwRc00D+Z-pdrx)3nW)HmF3z%``kmr@S ztd##`HjJ4d&zMFw=LxjTmN)8hm3_>NSMQhYo~Ue!51bE|%vy_EeWrJV{W zkvWqiiuX>P3!~W@F^)DSUjf+kHz#|iMWTYy3jvnmV@U3Zu7}LjiHe{8<^ESWpBvEL zZ=QU(V8j|lA)OX4x&ZJ0scUs@m^yxfUL&c!dosUSnp;pf6+z;J&}0H@mSM3F3vu2H zw@&QB`DV!HI(Qp_sqyqIDp`*!a<1A%T_|~V6{A}juo)nUk>;S{&O{sV4kObyy-yL> z2XfV}5}G_KOeZhhe4nPM_vF8)CAud5bYMF5MWN4}gabT+jd_5s49K-b7t(dcSe&#^)?ZGy!Dtt@^JvevKpL{){->)9VlR(Ou z&op3{+Bb$*Jtzfv(m)H^g`NK;nHlC4@;vsX;#*zB^rO$YTol;$dm>@Q>L zYBup3cCOm%y}Pt{;lXxJuqsmlt*Dw9{nhsQsvU_lH|vi-)My2z<`HGqt5+4q2u9Gg zN&^>RsLY^+l8t{6|0BY3*v`Bnn=tm1c}4g0z9YNzYc}coy1U;l7t6H(wNEMEWxO89 z-5Tm|*ilLxEx+Toa~^J}W>!5C%gQAYEYzS~5vv@D z(zyOlsfNw_ERGMTw$1xpraW$a`bY9rz7IilCdjdh9q}z^S1M>2+ZCo&Y}R zUm4~R8RCwW%axTcq}A^aic)|KL^Sl7UX7xjd`$di{LqJ-s&w4@Ub}|3so{Zd@40iX zDnxu$c)U32ucChh$H1y3 zOn@_g%PN9=vg*H+_jEfzgN_>OyWsgFArWK^R3Nq+LL;!M>l;bG{@!;spF8--iK5XK zWnVz51gkW~C=dj1L89v?fEj=`H4O{mwEA6;9YeBHsew~7?u1O5Vq9rT&M{Z_Nl7R^ z&@FfK_^MO;vlxHpht+x07x1)=1E%2rKI1HqC|byX=n6rh@|62Q-bh`~=rkx6Fp3`( z64Q*4ja!4$`%mI&brssp*Jbq?lEbqr24_c>1@NE?P!BB1o%#NF{XANK5`5o%gMMJv z#I|u0{46stBJ%fnnEiWK!)e4O3Fg}9SM&z+BkxM6lK}Il=Tsd96`e+-2)k$Y20!O^ zX7T|mNQViYY^V}C(x+~$Dnp<4hk&;x>L3ds(VVwTi+N72`FEbsgu3=5yAeRHYpcq z(_+Bm$*3PxGp^+LmxpDt-*ju{d==M6S%aDZU+zZmqBi&4aPh8>8eyPnF6TX;8Phr5|h*M>VkQbs7>XfY-8G5 zLbe2h{%@daYeq}^=ojpTv!e^Ku&7J~NvRqGQwOo2QAK$`=|39B*ZV**E`K%jpMsK| zHL$nX`Bdeij2J8Vt3Upo*Wa;Uf^Kf#NH3M#0X3yOTNMc9STB)=T#7MhNU!&^Fl)&0 zr95G((#rEa)(?OhR6)VEi8xS*z{>G~RIU;%Ow%8u0OhG~K-Kr=^V$7EdRSin`g(%5 zN#cUM*~SAw37HIfiGKRm7T*}(HegfdhMV`AGlPEPk6EXQKA4iDtr7P zaWKi$Y^VKiNL0i0k???#fI`Kg!JmU(gI4RJ4xV6B7SOB0^hfmumfB3KvPCG*Ae;4+ z$GL?0eL+(pK^E)))7uHuNQ<<;<758XNSlcr zG02d0>~L%bVL=uZb2aoB8n@i&OA^te)iW|sY2_i$+*=<7i;*&Lp-CB}5=mSPJT}eP zuXxw4V%_gNr`lu5xBNloW>68XPc<_b{9ba!dSAZiE>?H%`B4UdPgvw`ArTpo&mB%P zwt-9FeI?TV@zx5oPj&>VAY#!}T8xNyK;4>pS!L^82GOt~hnmWumTazw@*;%a;7#@` zZq6A)hg_ROhnf)8{<;}v5gTdf%{!B`wj<8t5+t zeRQ$b9yOiUAH*77*4Jos1=t@Q+qK6|f%ju$qK&dcrpCB*Mo>NrLY6h{)G&Sn=U8Ia z?u-ddfB$$BoEEbow%CeTRLIYgEBX zwXk>`|L`$A_g-3S64-AIVB6_zT1fnl_$UqJS+nRFIK{PDVG{+r6ivMyjj)2t-c_3( zHUuO-T^5w0S&p=1F{^n`PCRd#BLCu`|CClYT1@hyWB3;WtpupuvrR!>&v)CI?el_(6;k4M+52 zu$b3t_4X&Gz^rpqJx4TaZA(u9fQ$F!#YTCy8%IPqN>)t&3riPHwB!15h@HrQa{D~! zrb1r45{9FnIc>>)Ozd}kC3I?4*?I%EXW)oEK74CpA|$IPWb{w>-x8` zOVHI!NR`q-Q$)lzU7Sir^vF34BCrrrksG$swxi$-j-j0?8z-& z9EbIDYAG%@I$W97(Xj&kEuc$rU-2s{h%I+res|Bs`(o);P@*~2#+WDO>BO<*dPUid z>o617<#+_|t*Goov*WzQ+xDh|M+?*d=E{4)_H@OC%lCrg$!*8y=Ph;Sd%AT~YlGi| zpx!dzLUYjPI(3c5tq$-}?cUhFW&dlWKCcK3t@Zx%OM$ZPV->o}z*CUxlYOVt$J?R# z$h*$f5&q7`!&<}_YU+!F*UpR}y{%QY4Ei2l3{lQ>RFZTzuhE<9il6hG76G5r1Q07I z4E2}>h4kVUXg40PB*+>gBi=FuPoo|_G&VQyCd!_M8O)NC%XYW#E5Q>htHBG(K9*Eg){y^L{6q!)>nL5G zUB8Bgvu%KAG#P|(=*FQEzm+4AXfm+tgw*f`MY z@T4NO%>C@T168qN%sdm8AC1@_LANCFYzUrN^li86KfZ&WJ{z=sRS~8+4M28=wcW+< zUq*<0=53{)Z7GKEp_EX=1us)?ib_gHbDi)E-%CtP`P{gNFS{J~OGc8M){*F@DWbCg z|2QAGr5PDF!FF5^VQQOCr7AVvPruw_sxJTST+q`zC3~3W0^%PjYe)__6$Li+)Q#&? z0q43KX^i@_P5NI0u$1s9z!79!Z3cjxClW^P%O6&_Z2R(G&v!weO0kW5h^*R~AWg9~ zGMU4j2mimVxGoB2ev20vPK?g+G&cEqYYieE*AdBp;M++%zs!3rRW<)i>HqnP8G9_T zH6T!5|2_0VY(L=(tmC2;Pb zph!bPlLdiOSL;9Kr?)$KX1cuYjyAgmKKoy?5-RK4+mUJNJN_~t*>mY6P})8M?jD_a z`^wpBm;Nn}uAr2aTu@|utf)$B^6I!k#1uawxVe3`-N~)V&e}KnAnE>|JTJO(fsF6- zch^VkY#UIE{Lbf)v*4s>MvurO(AGvIUN1gdH9D7}wyr(VdNe`Ak#iH+$t<_b1+36H zrun?f6Y+$10~w+(y?k{dWD}Do>Ls5mA@lLh`&zZx-!Hj4Utxgv-be|6!L>2?37D2? zS3DN@E;=1DAEBxX*FNj*TK@>sJ^guJI-%fT?{Z_>PYC(wyimYw>FVReL1WtYgwNEy zTt=oICd0Yxci={n!5cEG*RIpc;}IWProJ_LFm?zI)eJq-+jXKq5Y-NuNNf>`jUhY|dyDVj&rHC*ZO5{F)ts~=VWO^_uu=rmE zD?vJl)7wk}11JAI%N!0RcnSgdo@XfvoQEgFo972JCOu_9A1xWa+K+#2sTgGvT5m0G zHROnrwfXI&S#}}GH!8(MG?O7>HThl;q5EZA_297(i*xFMhM3 z;IEa9bCm}fF=&wyH!jE{1}&!ObTM=g45iO{clJ_*Q&E%2K>L6C7V(uo___nKWAoL zAsS7dmW3swDl?&NU+wI!X}n)5G7`KpmsQiy2(+_NvBhhxBF zAq`qK;l-9~dJ!%=oqzLG&@IregjiMpE4VTO3#&MDB0(BOc)KRkFb1-=OazW;9L zXZa%EsO(dhU)nV46F0K^MixIYH9qca2QK6NNQ)h7C-*wrm9@(b%ai zWUwz|0BtV7>yaR7>m2#>(vu?KY4_SWs3G-YD|$G-*(+aJQki8fn#`$DhHbVgH2WX? zs)_@CGV`zb;mfiR!bev!KuZL;Y!7Xv>BDP=OH`D^uDA3J6rgQj92TA6_;ora{+PVV z`=zEXGwtG*ifss_d7s&lOhV8R8hXC#s|FvY*m<+`$&aHEFweewmNS-VH+dFy?I>-# zQTFCtLW$9_^2(s?{De5H>+WzlP62~wqol$Zm6c#S1j9)X~Ssbdp?*_^gZhAd2GEQNx8B_2nAJ<2>9^NLg0Kz8D32d@Cto zu9}#LF6_TfXiC-h$J%}sfAE;==Ziqv`)0lS40Hf8SA z%Sr0FvvvdUI*DST%sE!#_WF*y$)51ZF1ZtdFR+KX-}ERsoxyDe0hR246{Psev#YXz za_@$Fx-Z65#*tH}U69eAC?x9{Ck8nWE8%i11BArE#qV%F|Js#8tw|R8A0cI~h~Bx~ zu<%5zPY4rd=2BviOC=i)`=WJ4Beedr=gG!j$8+`!qA%5|ser0#8r4a^av8haptrC| zeVtM=Wj7;7nHnl$PgjT>E?-AWN!dnNyfWdlW%+j+_~~fBIH&xagb}!*&udifi$X5z1rMwIlCO>U$VSwKn}suZ6YYi2^nf^JA-2$=s~>>0O3 zr-s-2a|n?D8+VL&Ir&V(La4R=$FoY+bV)>7PpfB#IdVmRqj^yYIvo=@V^Jq)joyxO zu440QqKNx(2m5WqkWBHi62LzT;K2UCD60`IJ5tV8CIkXFW`1^X=T@FZkH(j+kiPN7 zfW%Tx74+*~vGCzKJzd$7KrsP00)FBnSV=HTR=0}ZSULud7ppS1I{?uyXJ{BsjvsXB zhU4~yzEpH&k`AbvKKr!S9rI_HX^ZVltzIb)-YWp<_7U6OM?88}Ke~ndBayz=SgEL2 zJIM@lm(53U1Y;D7OIvl8r8!bZJC|KwnmhC&wl~Ff6Vg*c(4E)L7t*4t&l~>r@;;+y zuZOQ6A>vI=#m}%oQo5KUmD*M6Rb*)Ksw8*9!*hZi^RY%+z@Z2;B^9ut5k4dKkYK(e zY>sog$|9Dn3U*%X$wt|>%| z3bXyY#*k8THK-qN2nI=l=rn&1V+jnM#F3%s@4hV>@HDQy2&5gx{oz^*us??tl?17^HmXM+q=C>K?P&}`T6HW-RgXSv6x1`glvx2nSg457sz$s;d-S_`( zQ5q{Bt^}^pk0=jOzl4kQ5H_k$Z4@79*zDkU` zH)s=#Y!JXO$zSZFt2_ZVlU|BWhi(H6Yx(Y-q3;jvyWkt?STB3|tSmTu0bo$uu`^u! zB)uhDah`ZdQ_NKZ0DZuCYj%j_xaVG_c$R4)Msmt)2&g5^RUvCCySA-`_&8? zC);duQ(dfI%ywgBb&vQSH39<^zzcI48Qqv@WIzwp4g;od!daQcYYsx~+|UWJSrm zPck?qlKwT7Z2pAy@TORP>5P&rX|9bCNv6+^AKO_WW#OcsiN47L;s2PYU*~41uz0_# zRPA<0(hM*7fYA#aDNmwPO%D>UCXJ3e>FSQx+d|eca>~h@;N?oI6-fhz@M=DQip};#(hbvo1msgpD+#)BFS=bGNfp1Z?gY!QukNA0B>$$p7=MAV8qm}< z+|9ZBdZEgfZt27P4yfBFVK&WDluNpq)KwnWQmfUjcT9qO|Stp9ygzqGo&Q) zlqzA4Kr+DO4=}=~!3(|Xq34(KY%#~qrPqVZYGp0kP=UDC~Dw$4lazUYD9L+B(0iq-pxXYxW^YrY_Hy-&7h!wU$#wBKtHS`TQfS#e{ za148GM5T({@MMT#0f{k8b#-ddiD;ZhiqTXr!;Q|TdG8F1Sx!uXPZ+{t*$z{&Iz)O8 zq^kK>hNyMqTF9oO5RdFp<`6CkS7U}Ww)EhAz95m7B2C8T4H~BVGT6^m$aulzh5EmU z6hs?46{#relxoV7-BEyx-%h38X)@qR#FZH|xi6%~+;*9USOg_KCat|%N{-w=*uV5Q zSA3ml5E;fwA0zNXUmCW7KiC3<-iV4ek(e>T zKHELZFf=VC`u+VBd_-*7ikw5e?Az;M_+}u)jMlbf21T1k9G7^g?!XF*m$BZC^6wl` zJF{G8CLOM9n%wGtp&*q}Sm&_JP(*>{Q~Jy_e>I!R{|DWR{m*zatY=WgTQQHJ1ha(; zo+QucvAZovh{WqK;d7P{N3k-?T3d-_HDSTBvcHY}QP*}A|80vOc)q;ol(QD32lH=!VjO#Unq0wd*Glp!= zb+>C;w|_{8^Ps%(x5N~2cW@HIELpIqPPa%Aj9E%21IVTy^evS^XP3q}1yzeGnfpWjxt0!+ z(FvosnxvbOSAHGqF`asaF!;@{b%yER!eT^W`b~7bGCQr}KR4zU@KBs2XrQ#WKwPJ1 zr|NlF|FD?pT<{?cKM|FtV^>;`)RkyD+-=3UlK zI-iap`b9L`$#-a73)76d)`*^#P-8a1C!&E?K;*E>T_!U-g2{;4r&!5mCC%c^%tsg9 zQp_i|R#bhhlSckzbI;QWYMVG(5PKI+Rmb#i%iT;$be_Rv@~lBsxo#~}^V#R1!3@*h zqTzFsV&tSgjE{$euk?TVej7MiUxlG66<`sol6p+~Z-FUP6oiJLc)p&g)j8~N6q!0U z*b;eZpR6BIY??{B3t-|&1IgWPC|}w-!oo{NMa@zSHI=D6_qk68}ck; zHE0E-%Kg!B(t^65GRq`uqXzGs-zb(pYrO#+2nCzBRbrBY3_sTI5NEf| z!0YzzrLyUeAs};klx=7IB#}czeTj zVL_D?TsYp!c7Omtnv-u?ktvE&o5-5dymgtB1oNU&NOi3a0&zWm?12W~d7iZ#h+J81 zh`~zRLd{KxbStvDG{3LR>XAJMS)rq^Gdk!imeqvSE6*HJCrhvU8(mxcD(v>m;qTNW zMBb6vrxj+Fa7`YR0^54IC9Z!3y~}C7ZL!O8;AwxbR&WLRg^fiThfX^hMGHKReW`&iDd*PAtcxpm7f?77q z8nu1b*u|jQ`!BOGmD7M0q|)Tcy9aEFHf;6`53MpSe8YE<9T*;!@CjyFMn^F^Ff{r9$1mC>d9* zX{L-LNRkB7`U_lNyc!TmE>Q5z1F4w(Vi4}qtWOwbRX6XAni3DMLEX~2f1~*p@UmCi zLa)ySP?oHeNDp(>DC))bQXPVtU^0^hxu>!AHZi+Rh;L3t>^mO zc7Gon&0Psj^}<}-+W(3Sn}9h5-R`^o30!h9jpE8+F4C&85B8gEjMn%Gj;WwIum7B+ zpsjumo$j9!t?@pi!^s`@om;jf>$0;1ZLN5e3YVtF{=WB>qFnucpXJP zHZZ0?l}Hx-6|DRsK4TQP@2L$6?R&iCEiL!BMTiq+-b&TxiFOh6|4x;yY5%qHjkj+M zHPOgIG<700|KcM+nD32$Y+=VkE=Qw^)>0+ic)~IH%Z`0NxRNvzpWpx<7}OO?g7(1a zy&V0MWBV3a>Y;yd91{JbCF8U{PE;q|3-xBha;GcVmF36GXK$}ZXQfI#T@aa+Y0mLF zhn;TbZstoSNAJzYmoF`pk1e`y|3Bp(wU6F*(WuZ z)AdD7r|r}EGG8qRJNxxk`-9i+$Ijj1DDuzyYVY?rMegI&iXWAlu1E3YaR#r4;}2W@ za+QK$p}%NAEjjotwc0=dz)2IQ5top(>@>d_LaU9GDz-qMbeL|z1QejXtNT|_2}wR! zEKfHaiXF+_+c#`-@xa<71#RAV*cUV{@M59K6D1kURd=5+a1wea(w2O~u=|fr91+m= zS&O5BTs-V5$)Ay#ism#?;AjwR@KAp6zTE>f`MfY_Q?V8HCs7^C?;>Xf|AnMrDdy492vDN2DQD=;D=s!8rE>N=-C_G$V<%~Ujh zk}!`ATInoahfJtDPy?+7z{$#VKQ{*EH7(7Zt-H5K{c8Ja&)uUtX9et3O6=4SF5$u zG&IsvQ#tg(_&X!>fGYo|9~_)-%Yxt5bIr{NC5A}!;)J? zjl9mk88-VAOIjE?&xhggg3a1d5|XA>ON2{kM}-VE3XGfedX{AVY33E=|10P&#R*5I z09H89(C4e#w|Z_1p*k3)Dza3FT3-Pb5Wqeu zPlaEtlbGwX#05>4`hL-j3hkYQL1rl)Q`YK+zGI@~2~a_T#||RHb=)Ed&R}VY3Lu}B z64^h@(4ZZj&38Z^aP$nWC-}{Bd{USv(qk8%b8kV7@3AFdP0aTJIyD*)} zL`T4`*!165^8fF#p$7NtnLpfp!H#Ez@n)lnOUb^fs9Xff0ml2<*l5bi0=8kQVd~>; z-Sz!ondJ>)Rky$q_ot}A)Ko4M)T7D+3!3;2$~XHSIGI|h?ua3ycxpa+J5^%0Ro(dz z#A4Cxy7Nqd*nHIdhYH@rbdH+-&V^d-D3Kt_k9a^7XIr^j_DXnP3G=9&kLOaWrAt2& zd8F)DP^`j504zx<|<7*&{00Y!Vlc2jbtal)}ld z)XZ0k%P^Y!6(1jGu1FBPgA-I-a$;%rb!O_8xoXW~dPHg#CC>K$(R5DXk#^Cxj&0jk zCmq|iZ5tg_Y}>ZeNyoO$j%`~V=hVOVKDTvIRZl%%&9&wj@0c^|vHL_rbt7vFbM!p> zBF9BadR%6TFgGzNmnK)+8y{J8=?)(1fFlXD;B=mQ+>~W|bTmNPxf7_m$jIk{O_p-w+KG|MHXZyiMeotw z86(nki*@SI>z9DeR-eKxiy6a|FLPMiS4=&|dR=uc{e$&OWop8tI9kgcizxDiPk&#? zGCp6)4l6BnOFmuS!CC`}NMmB6pJYgxeC@@Al>PP6w4|?J(YJf|so}#XEzRY5O230; zQ6XZjGSGw1y&Oj?cYJL3XH7DyN*YY|l~*?1!Ik+fZ*o!n>5m`8+^TC%KCyC%1H#?H z&tO~eU9N!9In|lW{{F$A1zuhtp5t1x2lVGRfDhXsN8i&MEa)CT;8O_5GOWY0#m_eM z{TD2XsopRz$69jnQM_W)XPK^LVG{*)tH)eqi)jrk2h5MObHAhmpGN0;8LyCl4KF^=5 ztiRTqtdaR{K~ns@j>Uj|;cAQhh6elM%F1jmuSb6nB!2`J5J%hy* zux1D>=?1kP=ZlrUHEN=2r?1?5ZwKJ(ur&<~uK&I7;ll8TKys^LPCkD;%%-QN_Ul6^ zYInLbR(F#M_F4TQk9ZhO(!8uB8M!M%LQn5V8TGO)4H)zo}K-wL&q=5Ud+aU{#F zU7@iBl;0o+a!B}=FJVu>I%lrb0(>awc?Z`qdCMTzi0%GC;pq|@ONW0h*{GiHYwBM~ zq;*y0AkK369bor=U@$d#kDV5`zx>p}6i`Z9M?PH_@`i43O)govkh+;RVy2*=*mX-t zOH0dksIYeT?z3guXvx`5 zhos?kS#|rkID6Y?;%%?>`CSZ1{k7VWR}Z6^qc7I6P|X8+xQ;bXv2yGC$XnbL!}jKw`zNCA>a(1#;+xC z>U~+sb_IR%F|I-sQ0ioTi@lf5ld-b`UdstLe9DZ{@977G@_kS5AHt@L{ z&f5+Dd)D>!cEM@CjxlMdqH;fkt|%mWD)@CE7?4Fmh_z<|f=oh;@jB}NF$xHP9^PZC zXl`cLfOX1xx>)&so%J_pv0L-L>9%CUbA~01_=gjM03#(C&fWJAhL`rk5Gmlq@cYTo z|LYaUgmqNc!6KVYAjax`7 z2AO`0LJF)h^r0YtL*hiZ`eAl`FS+&lpXPEHm>*ivPc+X1RP|ENb3Y%n65$jq3nps% zO8%Oal?9)E%09j}vO#6=kd2i5JiCbYJ>%{R(KuC?G|>fhx3Wss0QjKFHXeF2%umv0 zyh9ag$`Am`@cPukCd22;q>A)A^i)JKCW2f<#h$u5>4wJUZRFJKOm=7TF!H2zq}d8zpt+m1GakN^LD&$9~uG#BR)QaML~BZ@_SoapVc0okeOuU&WdvXYMuGKk0UNo z7hIrQ1dWL;ub;w&h9y}cDku*S`<3{##kH?5wTGdCIaSd#KixlkQ)9%cWf7iJ+)x+; z-e=~bYa^DL5BniKdGs8pEClV zihPHN14|oAkB$fla;hSa7V2_7na6vP5)&KTqKi4+QyY;&Pp0&XyVMt9Ag^uZ9XO7q&f=;3Do3t#p z47v7S4ar1OU^7xdd}PvQrby~O^W^NO0wTs%RjX~;ol7ybTzawgt88%3#+FW$?u@deJL;OehMuKbtNVgJM zq#PzyG=p4&SOj7nzX2Jp!=eR}Asjg+iA)t+;B^MIiWuh30iB6(=^cyO)Km{WUp9hC$($mpIieFIEb$PZhRS80iJXFf5tFLjhAPh<0xn=Y3!cs zpSM|R>gxfu3v0D$>%0^h`q9p?8bBHVp>IcSx#f@XwY{_ZtQ??=asg}TK|z$A8u zZpBLXZB*N@TDjA>g2SVUx+y57md6aWl@NV2C&*=A0g<_0OT(Acl??osd=jP(1j1lS z&X}N&`-_1xdzn7qUz3$}=r&&V?=B-N%RL8vTyN%5(eF_+&f?n+3-M#LCt6Say zS~ZrXwUB;_l$7#ONiW-bu=2Yy&r-a;UxAUo)jf=90^HQOC|>wQB=G}zv@mdigI}At z;9hnQl=P?cGs>ED*#74kcd}LrsI}O?p==q8P+Z){gX~ekkdl^ezxjIq1|FzT>g5cE zmEOHHoWH;KkcFKRK8u>R7Z)F=9sST-R_5M5|Lq{FkuM{zDuQ=__ir}N3QX4I+{p^Q zugyBNA7PPjdThC2gJfiQwi18rLKAF(+T2eXe*M%?6cq&z$OqVGua@Q|F8$U7-xB6A zDoLf+9Qpb2ST5LoKS6KIfE2d!^)f0579Avm@~6lJ$4M08dHL%#|C>8a5UaAL#`oiH z^zJi^{2GiafkIkcUH$uW=iB%3biUjB-++)F11lMda1nCEru&9HGd+gCboBoU1V#x; zeHVxxZRYtuAe`}5dWA>L){kM^2}t>kBaYQJp$$-_;GwdG)kQ5EXpBWUD{n%EQ6z0i zegxp8-{wciB|yMnm2tg8z0Bcm zWTAUQM_-In%^m_{Z$tawrx-84RM^$RIn=4oxzSVNv^n9GL9VhUU%3(RL`C~+H&^Mh z*lcgm_&ryaoagTNAy0m?%nq$}?z*@VtZ8r0U(@F4mT*>iQQ64ybgwFH zH$1=EJA^|+K}jl1YS-Cht;qyTM{Z%hbLQT@2NlIl_d48frlQ-AmsJ1VyH5pXLm$rh z4K#Php8pR2l2FJJL=*77CGdM&|MS*hF;-zxt`Rx>aJJ(v z)LKBvI5bk|jvga{AZbhIzy*iYt0#j8*F>1d6;{iQ8koSaQHn!*VvQD@6J$s#!OZOp zT4}d@4llA23dLI&Y1s%<1z`uW~qJICuuBj53yY9X(W1KrC&CR>`Ghfq{ zM48+Wcj-b)?5L#COWwJ8|D1cR1o`e8qNLWG4NsRBoZkTN4v{DItl-p?tH_p$$@5<5E z!De}FfdYNbqf5H&OVOuvQ7J@llv0(>0If|3Mp2ABd?kLulX{5(!Mkx0Y6;Dm9l;Dk zKZ<>2tyOD*5|w*})m3`)P;-~+RQd?c%_u*@NDutnew1Xr76i3PckzG3BcXhnxrbPy z@17e~OdYCNuyis=zsIc`0Zwa`+xpRv=yRN76$<-!ZyvA3m+MYJm1Z+9y-p+8)%|~; zH~xY>JU*`dPRc34j~9eGjolRv#??(hF?d9 zf4ZM_9toI{IR~)v_vD;2Jiv<@jR(WTqp^nXK_mk2-M|5J>Z_;&rQX-=)M@^G zWOt5~SH2Jp^j`sby7X#la5EbHc{F|94^C**;)iYgP*5|6z`3C=9eM$b*xhl6_#lWJl_ogfgC?dIo>mHm4l0sWBRJ?8 zQ_;(Jqu{QhZed~>$M3>|_kaC%9$uqcVTaj9at1K@d>}|?rf^8`hq-~EslD?hWL1)} z6O0{%(nUrKT(m*7gN)CQk;l=U_xk}%{-BmXq3VTjhA<1^4)_gtDKFgl?&!%uZ{bEb zV3Jliw8mkA7X-Pg1;0s~LsJt#iqI~Z;LDi{TfisV0NMjg!FPp0HyqDBnDagjqFA3Kc5R+Z;TI@K>40Zh~#WhO{c7?d}iYcycEn&UCfozimX)Q$Q{X); z$YCcts`5TLiJjh`4KtvQ%}mQrvGBjij0MVCio}0vxO9L!4;0S|Ys!5LaI{phwtgZe zAh;p`>{~D5@=U8_AU8jA}t>&(uX-!VOA>Zy`+33~`Z= z?l&x$wAd*vmJlQ6i?Ykc3vWj_pG7WCcgqfj8zpfl>i4a#0vV}qcQftEbcu8|xU9Jr z156l^B4KQ8k#7WPG}GXdhOlNZ9d(9(E_lvryQf_fU&Cc0Oi<^uXzE_NA13jGvAG42 znv3NDZnpMU=c{~4cI$Yo*MTS^li)+OOSKxe55^7eKvV%}n;Z0l7mH`9u0~{l*%Ydv z1anDQr;BH`1=6iiObV9dZsy5Q7|+5Cep+$ogY98e`1Xt6YjAwVD727H$CZWkQnOVG zHg;j$ufI=b!I!a1Dl{c_>kLzQf!RS8ihxP1l1Yxv>w3u{Y4kz>nyoX8c;Ds#>9O;; zIJudrppNK-*ZOtkXi9WPAyZq z(7MN7oLfA_$D~McF!i=p+m*)$35;J@N_-&iF3vA-`;YhM`Mg)lsuuB5fqFaVdwZ3vKluEJ7PQUZa* z1~hnHx1Sh&B7k*T=6OQxxKw}!DEaX7<>lrgmJ(-(U3gaN!rAqM(ty!J<$xoMn4LET zf&I1bigXa4@r2`bxjRL=kH_akLhN5duC50ijr#65cwUQ&jy{>n{56x+jPNN53R$#w zK%L5d6Z(%htt$apB#B{?s*%gHsO`}uiTEFki+(pk@K-hmy*#7YCZP&c#P7g625knW zj2~sLt=yFm;P>0$_t3VXrlnKZ0V!~lA{^OVj&Xo3nY69v$Rse_CZjsV@S zh5dOJSLz$Jy!T$K}qU_>U!jio{{ZUk@5B81V5$JpgwT3va&J z3L+DT&FGC=L&i#d$_>On@SV5W(IsPMaa~T49OOrM8p|XQXhr5>IBd&c-tC-0y)j%@ z^wFy-u#gB)f)0%%hQ6p324J!oq&rH};XT_4wjwdn9>QqpYc&WMt6LIfM4o@DY5oN^9+lo}6KP*0Qtk?E&ItFcF z(ut7m*XhE7jWD9NBxqP2>dfL#!90IUdPI3#4_vxI2(fGssGmCe_l0xx%~K>|Ml1pO}y#) zS6)-kd3dw^S8`I4iF&T%VgA=yz86*?8ZrnGj#IE;Ua4&QJpwQjZM$48{83bK^LTlv z$srziRHol~gl&KYT)DH09s>U;KS+_Uxu#7vIIC&_l{<<`n@wPFL=gWX#s$5bL zK!yU_I6-%Gv~m(3`qw&DppJpOm@=Jz>3TK& zmH*~sXscdqvlqjJqE!Wx?4nu=+(4X?hyx??DOq>=)NH?vjHGc>?F6DPjtN9;Rj)L$ zEoaB%St5csDY74n6fjmUYxVVYc2E>vYnbBrt>H4qlxx4O0-n&8M$p29yNlDF55)=r} zmXDr|e~F0X#%-^xk$S9KDlkJ>Cf=$ zkO?WLqO}VQraG8`f2@4bEb?CMs|bma=HFs?cq#_m*f3I)cA^uG^1 zc9K1A5=k@yWaKT%l9C}&F&DIL@vC1a^X&&(Ope}CnWy7FR`uWtq_i8;ig{&R8dTU+ z&PJd)-c%caeK#+(Ld+#`GD6x6a@O>hNPZ$r_Ixx zLq_Tg4-GcxAQ@-(g(dV~k|nw#`2-v4Yy0r@Ymvv$$C!E=DV`PU?S-53r_b539re~& ziV4q?zY*!2<^!_YuNB0=0;zLksVZ3e%xRorBx6+waFFA;+}ibN$K5gb)K8qCGYT4A zVyIonKf(R%o?d>#oc@4@BW8~s%s|t&bmUe1AP&x(|9+Dist{Y-D z&G=-$jOx0F!AsLF2OfviT$cW}jzB9j#U$?E!3pFWHAf7(h%EuX&sRTPr4fg*&3sNs zX|p>vfqT7j@ykU+`iBK8n_lL8MH_!Zxu_FQy{a%X?p~eb{KdNrRl>^%ehK95wCHsr zSz&!*|8id3S%(JXbY9<0eu~Y-?Y8E9*hSAfjn8*Ls?dhGl{ICgr7`H>q(@y`B11SH zfpIxG$Re}f8Ctjf0KXL@>KwPoS1$FlXfl*f%C7S!9!yl{_0&XRn!(rQbAl=Pp?m= zgn91bcbxqCaB{QL`+e2A1GRY%q^Jrk9_{n-jI5AsBoz7X#1yZ6%ifFvnCdwTq1#X}U`-<66rB_a+1!H$F&D>l;ADq8vGT03}d9-WmHkks0 z1r)!UM2au;fMfS$Tz-Hrs`{m{z7iji=LVAbENB6*j7At>*yZd#6rnxbZguzMw?cHQx2O5Vxz=T9dk!ZL%AW^St@O#BVJFCO*MQBIHa%-OoQD zgU*!RtAF_6>{j{)`F1N>mvmvmF|y@#+;Z!TKxJhxX3wj_Jxoyun_2Mqf`#`5(T>H3 zVD$b-f`Oya3yFaw2cjBd=?6gt7o@K?t4dDMO?Y)59`3Xf);j!~AhkBNF5bZRa~jXE zr9u?kK74NPTERN0&Pix2Y`UwIFP9;7IDHlRV-((qG4eqw#VF=7FZw1_B^ChB(srJ( z&ZVykZn-iR8UR6%7?d75ZZ}Z1aq>`BLM42ysqyy`&ZN8z(ZJsckG2|12-ga!f0%b1 zxtLT#*QAKwZ7Udj=#Fdj*w2Md5Fbp zC7kD`CMR!nqZhrpHx;(Q-xVcx8}i>x6!bqRU4HnWu?*X==bl}AJ^MSA?uHrjV*6X5 zGv_&sEc0~QG>e3XIbehCi>~jGM7ZN8dtsv{CkXTx2|dFQz|4YqVP61@P=FYO?GA@={r-aNJ_&R`J12#N@5CgLG&!-ke;zcgdUMO*g?5@S%cyiTEw zCOs*EDw2ewaT7o1GgNT4>c%lvgdt?FulkA`6%M*hYZ9R&JPk3K>?z;;12hy-HkBw} zkH!1tG^U!t$->SB3@W}Bq;tz}E$2+v{W7o`7Y8}Pm)Sw(`x1}L9<4?6BR2!gfQ`Sg z@`KnN=}tpL;(Qo{j7Psyj_g5eLMeBv9-m9J;t%3cGuM{pckr*N z$0nSI&-*1C_W#_t0%ao-TA*O*hnkKD?s=f2iwF6p02+V-&(wLT+55hg4-Y@Q*2vGf z!yuPQW@ZV`&7C-{=UaR6I}_cO3J^KE^Ysbq*hPA?>b&#$@B6eb@awtruZPKw@4u~` zujkhKc`Ix^CUNj41K_2D-ThL_;WuN}=Hw6TZA>hzff8^wKdeZBI*Av{k`jdw;Ul-DK#!3-N(c7AkVWO(R{Z zV%GMkbnM}=P=;b-qoai!$r>pMe!SUKwrj8U=x6dLgXA4h59qf3T+Iq@%nv6*DOWUC z_L!)5hxQc1FBWDKN?36o2f=<8k4M3~w@UxBfR0Sg=aRPd^EXZ7+R^8!le%x`N`R!c zNnSwLwbXR8>Vrj$lx#g-U_2A;OrUlP=kroZOIsqO#GFz#d>gG2j=GQn8ecPR!)uvD z0*Ep36U{nbX@v?OL%T6JIu(ClS@Kk#yXRmf*QY2Hg#`s3hGjT@c4wwIRkToUl8G9K zsIJnkFfCEXCU)*48NC{3@dLJ6{H4x7dKriW4mre(q$mu=9Reu~a{>u3A8NulI0l3a z_&-Z#NzX8_034%a#**DkF{H0jC@mU{fKUM!D3=kFd3XTM1^u2wd1V100|Po|a0`Ja z@Xv(Op)xDyN>lgUyPLc^BTD_o-t<=xWyx8K2X83Gh8^BpxGaNDR{fkBASbpa9T7deSDXb<)Ff|sY^2A+#gEr z&Q0JcFrowqEUT|7!78f6Z=c^X>ZuSK<>mOm%SHFHcb^naF)OU{EC3!HQ22**PU%Ny zmw*x+Lfo9QN+85j-5hBE^jvg@Gf^^lM0?1EQHfOgU=mCQI*npRs9GSuFjoTx)~*q= z3plEeIEp?F*HejVwrZ%88bi3wBTW%eb6{Uagi#fUpAs(mdiGlXec9erPFfTtwogi}oqashs_7h+G9Q|G}k(5?PxC{5<9obS;HM5ur@;`mx6NSHb0 z)4|KQr3Xs#J3^c>_g=Hs+23tYw>jADOimVj3zCW{-A%`XSsK`XB`+7;t%G5qG6pb) z0~8dcYI(vJJwrST_B)Aw*a9PN?({t930{{*^Pp$y<@{-OuN3y?k14tx9%D71QmHz;LT$abBH^5F2Ew;}R9Yr$CW!%phgR;8V1*E! zs9q4aoxzDu-lEg&HJP5YXgR{U^t(%Id$)eKm&@72x@npDV6mVEWGsX-L`6FX_J3@q zL$Zm(- znI#&=NwgVEf^g4Q0T=zMgy>1$7LvzKztCUL>UA-LBiHO2N|L+%>dO=Y9po3sHD-hA zF=7%=ox<`z1<&dw_b#D757V{0@OCMO4^*mbLN@h=_Ygk6kX1Y>?<*wknv$I#j|^jXV^ndLe*g9mZFt9%yd|EiHKVk#K& zLZzi_G3CT*##zdc!?)ycozD~gluFc<26UgBW8RI3M7kJ{$OJ8o5S)*PwsB8|BMZD= z#@BYQkv3j=KkGNzjZbUVNCoVdWBo5f!CXQ!_lqC(yN6h%wa87QSg2POqz+;;jpWf= zRtFqh0aWOxQcAx9zt8-gV(wa_WihrvEq2{#mHKb5=TqICw`Cx|1gpB`Bgv7F1S}L- zYkB_vQ4Sv`Tla%-1V24wFFm7{qobqiC*R|M)5sC!z}yi5&s;>{Byi9Dq==VZpV2v6 z{@h(_?YTi5bnOr)5kVlZ;SxHf~c(VhBj%h?Y35 zQGHVqB~NW@jsKPpnVHGh=bqjP3-p&M80#}?(Wx&@?0hj1@Q^_a?o}gfEzx7h27y6~ zIKe}xszzd2U)zE5NQ<&qnX&;S5Rs`+5T57_`Nf9J65Hr_4s*rfAl9mI5b>|AKjDF~ zcvSNArHm?d(K>TDs5;qUe$$lQqY+~2;8t@W~T^ z`KoDDN<>tM4);WGB^4zlCI*I)CYJ%rDh^UW_>Tc+b{qsSy-`ZJ2?Ro0EBkPbcm~je zz=~iS@m`}MMwp}MHhao?Om8=NarJ5G+MVF6M3`|hnKMKN$0qi_8AIlYi5rW4Of-|I?nefrL8p# z$XI_Mz9cM?60SM(a8o;98FH@VyX*$mQ)D^%0Gth+-4Wn;Jm5uXB#ZRxH?-9e>3M4w z7wZy3x8CABtX@`G`j`HkA~q3|(W1`SnO)KfddXG&QUAH%eJ!FyM;9_-0}6udb3los zt&x^v1V*3WakU_y&F~1i+y9PXN};Y1{(W<^KZq=2l^)*3w^oqMBZi}a>Czyr`%ewh z)gWbvVrvi?X4v%ox#!wFI`|qmTZN7S{it`|ka%&@Y^|E^Y2~AHfXiVVFVQ)_ix6#({tvb)q!o&3 zwQXzr9s9?3e&+wGhY&`O4lg)I)WVe@)07VjU? zAxZ4e*0+MjK{69F*7WsotQYOVP9-sIVJJ)o()|6E2y$VUa`I~3=uFt^9@Q*bMV~_2 z^x)-h2H)ZIz>!xs*He- zmw1@_k*7!@vUn0y`!3|96~#VoXXWC#)Nor)X14xF%VBC8zHys-n39~5V8XT- zwQZCBUCOUMXg?zTI_};mtRin-C~`?hxkb3$Kqn+dmJg#BGR_+i89BYXiV~p8Y~Qnb zOvXV_FAdZ=0bFywLv z5)|c#vI=#0jS?lPSGB)*d7rC!vl6p&(vmTDA_F+J?(dW>MR`M(+Yf%C+a3O(rrcUq z+v98ae9Xv%L_fyLLqA~2a0TR$7&2|>Hl9syOP1TlKtnpuMA5dbZj$ZKO z2mk~@IvURthar>0^|OEmDKD$)*o0CXL8`25X*cv?4nh4ziyfd29f=Ur2h>=MNiIH0 zBl;&<{*~ZV4wCs66r~4FajU(hQdiIID87~`I29H?C*C>ll$D7VnT8tuhU@^VWyCK_ zQ(_6rKm*gdzc>^x8CU!?UF|wO7w*dDNG8U@T=c5ej^PJ9dw|MQJ91Y}7Gt2Qhjz9g)59@#Wv6=JL3S2$HtUae#t_*O|h%2Bdscp13M~E;XEh>o}Wy z`I@|207OuZRCk!GhtnY+7V1n>f1kbsCvzU*J5u`=9BVlgMy4&C+dhRF;&~7(Pn{mf zRi8hKvNK_a7-*z}VgidOGBp3kTOZ+pAz61VKJ=d-pGzu-py##c&0g}m8-#VR z1)`htAiraMhX0Nle`q(8P#4aQL_Zn0Z*atciW1fI-R%`T4$A2-T2x37$pZmk{Ifu$n zd{SYMkcZyy_g;4J@K3Jb;WHre)#CMZ zJ}dB!z-_8IiyOE^0(cW{$^X2xw1QV!KAkU~)s~r@Q^Y9HWbUyf84djtvqhED55bms z*($+dp8S4_^1&C5ncUvld2IFL?oaqPkp)}Q2GKgNj9=k6guDnCYo07psRr0u1ebf$ zb(Hz)IzdgUpK*U!mmIz^_$Wp)|2QII;mOww5}|PdOIc}@;WwVoM}n;h9f&knW*!>j zvJ_k?QfWc#JP83K8miGCJ1MTd$R!^QeZLY8nAi(TB1C=RmUqQOVs1)Z1XK7&=lRp>O|!|KC<{x;f|;uw%`>T1+_d;{DFx=H7jWq8O-cMMLZwa8aVA)J^+F@D zfYA+=e8-Xs7Xg=1VP}XgrM?Q#De~RnKzmRwafm9u2$BHI19k{M2Sqso4u;8U_(34Z zo|94kSl(CI8Rt`;u$TV_P2A3toh}W7n^sn8N-gPUYKk2xi)c$JY#2nJs9t--TR|wi z0%*knMjUZkit)epk4^!Bj$c=W0mRJskTJg2(yu3u*Y>t{qib#4kkV%yCGnXoWZZv) zBu}-ZU*FO_IMkA2WEn{T{DHBKDL`k4=rPQWwoEzsp20nN%>;*^f5#{!6z{dT*I~Q& z4$y~f4OV)t$PNU_pA+XNr2dm z!QNyv;8Sl=dywDh@mXx*>mr+oUnE@jTZhg$SbzgqN*_dr#vtIZrwSVUsQuK6yl&~o z8K>+VbnGE#;|S8&&(Rj)sqBklEo+=&lVk08ySM$Bb0jt@^$a-V+(vUxWv*3}mNrYK zI*^}cu&{np>$fu7EBIJBNsZ0M1e``f9jUeZ+$yaeL{bH4Mh>h37pXZYR}1;1?vCce z^QIjiOJ=uKo5Lyx&AigC*d?(Bzwb`7m6C=Y`7mhAJeU zI&;tULL00S-Qap3q5A{ha?W|2?>ICn_iuyv%W(7Gov)b*b=IgZzlQ$jIn3c3KgU7aU%|E`)Tfk|lFWk+U;RtyX|XdPv&R!z!pebA$aA?`B=)k8 z!dFlQz)sQzf=HO7`l>&ZoA1V@(?S8y5|(%H4;A4De3iwjR5QsNbg#`hSU8t}1+K=g ztyOj+3NDZd85ZGpn&t3lw8$Hr>wiU$g}wMoz^q61fbSpTzxb-r{rOn(EmC|A+3gU9 z?gaD0%Hld_W5GHJv&weWdpjs}9~-`uiznNz;}PuLjWwUUU%`lJY{obO0fQns?pIye zXo5a(pYeiu-1aV{b|kK}g}MKy(cY?nOr|^Jl8=Y}Uw3qiSu7+cZ27lLev=Cyrq1OE z$lUqd15@t?`ubZyza|bNdm4(>A%|l(AGl^_b1IY7w4~#zWuwmiIS`sCfF+#x`(?D= zRSaaD?`cuhbRrnuabcu@_421%m6JP>O<1{54sk z+@keb4Ge>bisJ1s-t5lqOS^U;^2yiW+VTYjY|SBg-ho$wAk##Yl-eEdOdP&8*whO) z96lL{`jJ5#66ZKbap3%S=X{{<6@KrktsQa%P|n827Am}I(`En@l=B=y1;^cP{_=Ws zpQWZmsBSedP6}gZANPe=R*6L}9sw>bpuV+xdwkWcPQGCOyX+7LVc|~{Ryrpa$6%TT z3tMaAIe37Kx+ipvv4ruHoW61)RhHQzB$`=LNW52xd;D7&RE|m30h)WXsLU89i}N7@ zN(J&QaqGGvNj*(q4m0r)9KEp?@j2dN7U;?dh(?E;!ag}ERwR{GFLoSQ!LPl;y(RNl zO56Arju!uC8R@Ocw2%5s=0S)_J8Bs^UkBop%KH(Em~3t^D?}Xy;xs<?o3m?kT1+|M_QC8jFQC)c~b};^_ZtVXu;`k%Ct01Cv(iZ<*a$TC) z6OEpivQ~c<4?je68zHx<&kwS#J1IUDj`BUDGxwt|hLCb5hsnqg(&D;~pano0jHRy{ zTxSh`l+U6*Mo87R35Q{;kcR39e$Qw?fpvJ-n$6{*V6|mwqKyk=Qa{hMohLP%r?iGG z)F%|^Zv9r~N&PMa%=gUAawDK$N&>NYx{h|fweCF!YI{I$cYx39qbZGDEdi7uVYx&f;gaV%$!0OFS zMt%9_-EMT}bCavmmRv?o&Te`t1kI6Bc11iD1_)6@Id&#fin4uvuxKpCeP8IsyVK%$6H|6U z-_yeIg2nsX8=vx$5{Nkkge}lVu+((zMhmx2(+rbrHtU5Ozki*FMze?ie-;2R>J1Du z2aN&^BFt3shyGw_&*Pk&9EZ}5+=2AjF3*QKbexD_J>)NM02CKo$NP0B(3ZLxd5i_T z8GhZT-}ryO*Ws-!EP#zGw5|i`Raw`7h!)2khh`n)`m3DR|6x`6nBmlHjL6FVC|J4E zIB+(Kv%Br<3`3bM&q*~P<2W~HWl##wU6_Z9y0(lQoyUL;CvNCqro-RE{tUIR7ul0Q z3JIaY;d6pzesTX3&b6) z+&QIxz))%D_547tDL;{Kn87c=iaWECgcg%J{fGU%%0rV|XX11_^j# zCkMp8AMf!F^v}Rjh>*ncx{#tC=Y>f}RDkJ#)1bt-sYDN{lCUiCxZPzeOK4h4hwfnW z+`f+gKbo#7u+FvV#x@$;wrw^x8Z@?@#Itw@_50fJbn>dJ9XxV6-tphFay_ zdHBOxv%!q-PsaE_u}aLT#LR7m^qmdtwYRh`VM4^dr%WO;U8>*}lp5uZ9vu63q`zLP zuyO>F5e^sN=MR!G{-9>(1kF{>|Ga(qW;$oKp2)BqpfE?p33A%cuB*3iJ&i)s^oUmfxCCR`BLIGE_@g~ zgheuP0JS0*8ICBsfK_r{Hxnj!Yq|~>W+d{jCs@TRDfM*PcoXiGJ}c<-+n2X@*&NY0 zv#_V+>e}(=!f7`Xzt0O7Bi-wL2mwIy#9Q;{Twa9q=ye{*^U2J!1m zbiy`))WRz0Fy#h8itE6tPHXQATP%jKor9T-Ij2#5aj))bpeGxb-3q9#ZSYNgl-`qY zO&~YlgfVJIYC^%G&^ztVZ&53DtpQVa-)Ow=Rg%T7gFgiVe-7e?`*97SF$IJ}T_a42 zobxVK#0jWQ&N^ofo45P?NfRx&F4&`5?Kh?_0%0S7axpSMy+kJ1!#eTF>s-!DqoHxq#TKnP0oYr0aTTmu{= zcYoicSM&yQ4;~W}vyWTX2QID$NM9E~_x|q4nEqz#54V?Zk9 z6u02Mjrpfet7HLcepAy($uOozfFim3G;kj+-Z0e~C_O+06zEv?IQdarfumszg4O4Y zEY}o41L!Lj_aOt4<@V`c<#T^^%mdlz^k74tABa3RL_>k7J|DgQ8kFgVHL@#yN+VOM zf@{>t=cb@N#?6rio!_(B?&LN0I_3rT6`ELZ5yZZs!cM0N8WEh@?{^wtf(YkQoeWM4 zTLH&5qUo5DOblvzm15q_@RKL8&u`!q1nN zSsA>bCUSJ?<#z^=?HYc9oS;exOrx9zpOO3tGlQBBov+G%x2o-9Ml*O2L?B9O-+$DW zY%Flml;|zv)hd$;tbWGCFlT&OJzt;6;Ppjd2OX(3aC);T8ftUCNI-(7ci%DE6DDL^ zx%ZDcG!UYoiBXd=Hrfx~;o|pt2~Bfb$hIM8x>F9o-XF;zON^pKUILAWFw9T8E+bpI zyN3U%^=s*zQ%Y(f0jPP51tAzg6;mJGmAv-HZzxpLjd0aVC18?@{Hz?=7bPV@HS~L1 zwRBk}qH-dyX@09=gPw`A3Jb5#PB?d^blb-wvm4v_>Oj6<5gI-hvRt_FjaFC9^0D$V zGintPb(G7SElHGwcIOY5X)qd2$w&r~;LFJeI&It)=O{?aeu$kJdm5&0#$0}xQq{e` z(l`lp3c2Oj2-A(K`O+bwFVfoTO&ob zp%Ym8(J^ybW!zV~(qA+xQN*CXSij98<8^`3CpydEJ8}gdl^>mg$}{7)Zg2H%{ z3)f7*fGMLk6?0O;cxi*Y9Udc*)(e+z&yOa6HD)XqLs|o{5oOiqP@$tjH-D%Y&pxw| zjnP`+Zp0+=AgWfu{|)kSpsuqby$P(Rbb#E*ADa=TqJ4}5ryr0=di&t|^V6#CwRRl#o8uSdWU%}#bK$c{;SPZLO5{K*;Z2e`QJQ*V2Ic>hD0w`SSOt}v1 ztrR6WV}E*K=1COM_I`CsJ*r?2Io*$r1O=JvU<-DR*c@G7o3R%C&TkWW){4^4yS5e- z|KpJ>VI03m#`yuMvP?~G|9e4}X+Z2?^`Lj;32g1`LKl&0`z`Reo}s+$ZZZtKPAIa$ zDtX|!G4hCRcC3z5J^B>6?AF7twA6sQ6i+{{N2gl*pIc{v6^l469~u02*#*t5dpa6-}Hj0dV zsOi8nLarzal@w>z{;h0FcksbMq_(yrg;zxX7RGtc4qrF{4d$k&LHph}sHuhL>rsMQ z5cbLnL?V9rmjVO;{+gVfRWr|mk;uvM7xRG8e5i2Q!$NK~1IX;yB0B>;= zR5WZ9VK5pUOWV5{S}I%2GNJR0Q10)dzdStsxv$isI>&bc=XGK5z~JUs0tWXXE|uI8 z?dkey%F3!2z&o#>rIz8lScT&md(*pUgVaJ&l2IY!^S?#IxT{yt<3`uBPHy?%U&B1` zd!^`FiYbV5V~AU!t0hAqE!!Ywz%+zyhe>s-l3N*FD;X}2+*@B6a61}Z6ik@bAYOR5 zB(m8Sl^w=qv#uMT#kLi#Nj84o>Y7F8Xe)J8b-hB^)Okk3{>=`YpJ|jtU>ctmi`*c? z0FJ41^1OT$4bt_Nd59@?OG}N^$J)II&{L%7 zDkrDv$F7W<$!m_A8Dmw@bl7iD^}KeZr;u#t5E78F@X2}Hi#V@chaN6OwdrR6Y8 zWHBgY^vkoiCHRZz)XQ%+YwC1KL&g-I1U;s;>F6DCSCv4Z6tVf|+&XWUM11Sm>d>%P z+k+m@$#qvGUcW?4kNN!rkT?!?Jqy=cB5#2RXpNnGeyNlx$9pr{q^)?p2}Os6%BR{c zF&0)sVBH5Kw7@V(w6v9Fv1wH=%+1dF2ohnuf-R;UgKVd6wLR0edRqgVN`?+ougfGYIAmAZIf>>9~lS>Y>4HcCHKyMCSFS$~;Ge zA2XvvHH}GW`s=^+`fK)Bc&m0`j2)ZM1_I?018F^i+Ny&Tc`N|sc+l~Xc)AFsBYFX< z+Ce`!(??c8NOJw%i-{3rs>TFR|H3g(z>ug9`o}FQvUa7qd6k@rTBX-Q!N)3$F{bS* zFz@&Z3bByFMpLqeIqD~h?Ia`EPo{5BAuVvRFT=ydOP`ZQ@)ta`Sv8gar@{vX3F*~9 z>Ci4R!&Za-PuCZyYo~$%a@X~^YF)qh4-x*Kjd%AmzV*{;?MA4G>BJMalfu0IeEe5t z(@~7&kj~?r8D5Ye%iU#`>T+R0ZKPhx_x`cniK^|FP*q7gM_M8u2>-A@c9L-zQgNiP zHzT5^Y6(sq_;ygqt(ygL4>IU4W+vFqNhE|iZ?KQV=$EW!P24uvdv zt$^!whmB@Ze1vk5ib`d}DF*j8i`F}g*Hqf^6iSRRPJ5$=n1YF{hYFh8X3Yoy03|@? z?xxa8*q2|*|m=9cys9RZ1{2LXK3r z7)B~j(nt>wb;O8Bp?+d#A28BK99E>o^%02=E0F}V4kDk>G@P&C&?UP6R0s><5k7YSbLxz7vHdDG;2_< zxkiG#psk2t;i38df}pIaOB8PxN=Nqy>>iN|QXF28Jq3tHHlr_+*4Oq&^QIckd%pJx zOC!aPCdXSW7X7Ho&9QL&vb(5Pi~8KM%jnosgI}h8Hb*Ivd3>IyEZS!s5d}d+KcCM) z(m*KjeC`p4AmN!msrGWk&!|+lGHfkq-`4WRy55$AiD&l#w zmlEF_js5)oLV;%+GZ0iHPuOgZ+7KOy6OUQj9A%B5PuO9gv_@Th;dfa1j|=`0`C#Rt zF|+Rk30YM^M8E~2z-Wt#ux2t37501u8O>aw%^&W&gMfyvbNt^hX=4`yyo}fo=hmFq z^sul{LAh{qdW(Vo1u&x^Fo27e(e`)U70v?w`xc4>ncoaaPC0F?QSs5xpBSyU3q` zcc+CNjA+tpR_0>y@g^hU4U`NYtgsoiBL48)@AgiaO+qUsw`(zMneq{>&P2RR!;8* zOY}069Ybv`L7bIi!;>%>T~6Txj9^P+q`?BX!I{Na1sh<ACr(r0@2Ae7*fQE(iLOSE& z3l$ME&wWNvnUB=**~N;gQep7Q?a?XV`n`XS(L2uX9crfuo7=7wK<}>4`%_#V_dR)Q z8pjSkJ6(+&a;7&p(yVn1FP)EbfN(H#tb0*;iA62SZLNvVtb}+XEl6S&Ta;)f8WF zb;PANOZ<%WnWW)w%6cvU*ZCUOzm7!BI#l%g z2qA%!k~GJd3gA^UNg+BM;@>QPi=UPwQ{MuI-qd_AA}Old=B({GzVW+7am<)aDaa8B(o>N!V7Bm} zy>3skuL@mg{#U(!NCf2C-#8)yq{AX2$?c$dl(2qX43CqUraI`OWgn}$<)t9VMqAU`I>G?UF(uf zq|>UYk6rnV{DyC9P^l=Vtbur#%iHfVi(ZWzOYt+}XfQ3#zoR20%Q z{2K#>RtWykSniSSGLd2j{qgkKX&$8k`qNt#(45@iIVVy6&>N=U#mf6XN7Jz{6i zNngpwOm)%smdqpssHDiQa=anXYwGcXfB}r~hkSdH&I=N8r5Qr*)<7OnwPGvK z!yT7VV!HXL46i&~gE)9mW9OZu^d4ZUv)T5sm>v0T!77ajK&3-1(ExJ;D=99Q9rb|$<-pW4o8 zH3JH%d=|~O?%MU8U*1OmdwNH=MMCxch0-5dhEPfQ(lf6=->K1+u7j}hqT`z7H;UD<=M&S#RZsFIXr*j<9H@i46F@`_k-7rgkYAG?z5Bla#NCsLgWU( zf;}&xi37BW=Cj^mIhUo1O0ZkT;}sQk7M~v?w>C11xpdgzd0n(DW(uPF61;wJ8oh6ZLLhV z)vwFBjt*O#El`@4ZyT%_gE}mVm8~?IfW!Cw8Ap|>jC{47Vwf=nVH-9(RAdw+D-nFT zbTRp#v=$TvCV+*$tr1ierDOm9-7;#BI3C>=8_(ScDj?OOCnYWI%kOcMT{`xR7i+wE zvG&v#dV1L%Ot05Bj5t9@dK8U9wpiR2f9*EP(@kp@d5ZfSJ9IPHmZ$pn&mYn3WNR4s zzosBeXOm51Y~85k4_UX`wNrBR7v$7kUES3mWY{b1HoL~! zwK&$7KI9l_T8-MDWzd;EleAp6m*+py-@zf0&y@be}Pn1;6=539L7<^9)Z3*Q$% zez(+reo!gJ?-#c}8r>8Z#64()nOpQ z8wp7`*@X4(vK?^a@mHEE8X%_v5*SQ<={qK$Q7@N@J0sTQi-(w=1YU}E3rQ9)Dg47| zh|$ERhp<=O>BMZ9u|Fw`&`vdNFJ|rbGouzWCKgaq+0c;?tWObgOFHV*=a;N zR;RD&)Y?Dj$#*T2^P%N6|CFSC_x?63SM`0^oDq4dF%*V&nj_&fB5srz#nu)V7u)J; zg15I?zG!{43kJaeKej1(yPj=&HIoqgT61F6ax?nD-^`+RfJby=z@8T<|{VT7#_x zY@c`Uy3hOhl>E?boW;D4XRdArM1w!xV+?p}zwQ@>S$}RecP9;wce4!1k?h{gsTyWD5uXm$dqS24-=Y(M0S zRY?{VqA0_L`2>1lv)9xa>-V=Vsi%|Nf6Hy&u)6ia+=7Edh6);ng=?mDV{ zUnm=-DiPMYgEWjlUvtYzmJgp`>`yB&iWkT=ihczAbFl_?$9qNmU zibyA%$f7~Wl%ilT{>*e=z3Jei&W*z`gGKkkk%F|^_`R0-!NC(g3-kGX@^k;gAJ5E` zMKPyGzWGNwj`OuQz{9}6;OEc1TbI%9tyS?q|An23gJJ#t8TitE?mRO%JyY1njZ>V9Y-Ho zKvKc?4Q1()Atn%OkB^(7K*8eP#uS(0qOUFLFKn(NcE^e_2VTj`mr%`_&^fu;dDi;*#%s_7JqxxbA3Yd`r{fhgVQs6ec|JaJG9xUe+vbWsh-G1S ze=bgp0^4FI*=r~bLN!U@q7Cb2quK0VdX|*7)j~OI)Gc=-;!;5g#Z+1r+Q0EV@ZcBV=NVm1UyD8(iyM;ZMMd&0(%pUO@!Rk{@VWji+Y4w0XUqwYqZP7&y8J3h* zUt{(i2S4*V%u=KT5#ohyMHH5HpDMc{CUrV4w#7)uN0qR%0y>XA2MgOxTtnZQQzhc@ zEb7_eafV2_)Z>-*kzS5$rjX}TzZNT7_W1kc*4O!`^`f8QR0r~sVRAabN^-PaQIN%W|49*Gz(=osA@Q;K)J4vlng#>KaG7R5DQ_8dwNDR^8X-*~3W)x;zogv3yJW zWmjK?;gus6hV&gqab0eU!Al)8yx%T3GzTkB;AdloO&f_nAg+8x4VMDN7LctAeZ!dF3FLTrgC!OK? z5z5zEL@oC(url!;{9XdNhGl+g<~Er&HkjUlp%x&Yr(J6@K~K%dPLO+qgc5h$L6HR@ zjKMt*AZA$nY^rF7Rd?_=^jX&7#g>Oq$*gUWwr&Y2%&BBTxKeI0u)N$ri*}>^14Tk~ z&GBIKz|~|C0D~Vs5{YxMd5$d7hZCTKJ{uyv1eM2B97?);BID@1`u00d9bO7hl12MM zMt~u$wIKaBtXw#toTxqkGnp z$_ExQcM4iqmV?4eBR0h9>#IW7FXOd8wnG?)te7Go{@{LAc9I z$Uu;}+4ozgFSi|jUX$1Nj&icq%s6t>3U0JQRr-T>z9@icHdZgI?{9;3pspC-fxZ|G zi0Es~?Mz=OsfFyTL(l?!8zy>)qtNPGE^oEhb~er!_VS!z5JI#Ea5!;I2U*I)>?5Kp zty?Jzc`Y*q`C=umKi}6F2=Uk-pBZ)<2steZzsO?(SO5J<%tMqGkTc+7pzyp$2C2pTuQzNtE_u*|{`iL}% z&@f#&bUY zMuQj1b^a54c3Tz+#f;^v0PC{(Th9ga?8qE9HYW`L?36O5BD+Rct;Zw@ z8^%1Y%<1yt>dGN-;IOWk3SM*ZJ)U=UIXOmE!p?F{%nCzlj&!W1rt?HTTl9H)y(7sa zS?!xBqKwsQEKnF)2m>;?bRu3xIQfv;I63v<8069?KJA*$%$Pex9zFyyf+t%vH-#sY zB=Q(dJc@i@ZQCx~#>2^rKySCm2BKxdYj}t1>tP_yx9$8}n{frq`~9@@+U2|SO>=F> zLn@Kbi^vkrh#;~*`?DFQ)-DIz2$f6un&Ms^`%H&{pPx@~kJx}FaMqc$g)zt?Vwf!+J)RoX^>HD-MZ5n&2S zomly7UptDt{;?#)QwY@X2&;1Wuu&iQAIZykrA;*o{WhhbZsaafl1rR=@${YeqE;Kt zSV=0Rk67{nCUkZ3+^lZedztMrX4H84f_)Nwgw;XUGHk&c+36+^gkd!?a2u|ImIQ}D zZoKc`V`6!UG8nHQmxe^H63#`Gg&S~5cQ=TU_#p84PU&%Y{^pX=Wpx`HcyzY;{zS3s z$VIE(t|(j4>#yGiZS#7n)yG@P@wvb%Te_01-9r2Z?tP+0D-s&|0{x88&>LYxgK6(~ zh1FeVu-4e({AJtP#L1K*=H@n(Dg^5@2@FGLVx?zE_a#Q>`E46Xf1q7~sl~7Y<@!&G z)s62UYC+oRWP0mHVxZ_9{zTM)iaxz%EUh%ea<~+UC}I$RnAk5=)vmsx)zh<9-_=FDy`%qa7b2C3cE~L<}YPU_^r9Jz-1y z+@!T0ouD={A9|E{Rm_e*cd%)Hh9Vjro)jfg@=3s9tM$+0MV;?a75ukDuqY`IC2)t{ zA3mzav6n(0&2_dPlkSmqLlk3@AWB!>+n`OCbV|-z@yw@4MsaAG2^OUt9q=KM zZ9}AFE&Sv?ECTytx03_K=6-lBUs7X86HrKx{Un|AYSG=TH@O6-jJn`3xd^Hxq?GSU z!|vMdXGShe-4XvDFFZ8e%)lWvPUuN%8H@rbY-|MgBEV>8_rO`g>qpGjqIk^mb>t>M zN1n1hR>mMO3doNwlq+^FGrX}_Yyy;&suJc+8g9&`eD{DY(v%Y*tlrjs)ppfjR}flh z^4N6}@V4{2;teO%nQEsxVHch9Uwd z2CuOltzWMWB5oIms$~j;!N}2d^`;I5g`D=E<^55$Fu?Fpy_Bz0blZ#br}?E0`n$@# zC?;kKy)ru+8jmsI^?l;OhA?pwF%+ap@f`=7l#C^<*%};0j}$he=@fEo7SERw;3p6v zvQ0{p5Z4)CiaG*0t+XsWsliobEQ*MY*byp4Ao-od5`IX>@1P9jv7xpd6oh@g9SME` z>U;|ArMAZDd^C(8CjWRphL{_?cT;`-nRmPLI7YAgbs42<@FMFm9IJBX^pc2@?)9`W zWBzq8L-pBGtTTjRXM3~delfCAN|*Fg74r3754^>;xcPH0iI2-*uM)qv$+39$5D*}v zl0_Wa5qy2FGmV9Uu#Qj;s4vF7hwh{5!wZr%=BI%MBVh)jkKl!UkN(HEfnvJWi{F`< z;aMF*(Tt9`N^`0J_Wl~npQ{HBQfH~LONuf0-43`UKHCLzOjPt_pH)erwP!|ozDFDa zeg2m_o#HKGgw!IU*G4nUE=r=N#wKv?U#c}$-2N$Yp;0Yaszk6F*J?I|Y*2+-M&Ea} z&BKcwxZ$1owP)Wk_@G%lvcX@GC0f}L&0Jv$O-}`bryTK@`oA(NU*!deEqUiz_Y)EV zFQ0#Jgsuj}Nqirm7~CK!Ve%5!L7t|-&m__>P6&nKK$ZGtX-thCV}U`T%t=ic2WB4E zCflK>r0~RLF-72^=lEpKn&HChYiY56{{nB{<$mt*6@vcrrmy?$r>=N!s~Ub6lpI5x z7jz2!5W7!Bu-|SZJL_+wYZGg$=OFiWm+e|UciWXj>px&y=aDsF#7^kZy;c$+V+maa zM;cU~D?X$DRE+=SeR$$t)B-FJ3rof{42;Yp3rk9)01dg!BFI&pLDMN(EWTD5;~lv3 zU}~e(QWETZ*4;1qG-vJF_ulW^w3+J+>&?tABP9UB<8z8=trB!8!l0o$fC+7 zC4c`;zIT`p#QZS~)+)y#;(LgSb=8C>vn9`Jj;U9jYJIeU=XwaVt!!ZbbvSf5$x*O~ z*kTIYY^nT8OGxEF%vtMw4=T7G0#-sRMA}QnYnJK@s^C|Y=t(;dTr@Itm{GQ;xAq&$ zh@bGM1a23y%F}F4+49D#`>d&q{R5@jJK9E3s`;{7uCY3q@b1Ko-{ldH>I2JIlhWi% z;_?IqQsgV zE`7I{urSw={{^T2bUct@9K???Glu@ned=HM>e-^^?eHtvNxf7MnB+c<@~=EhNa2X_ zNhx6T;sL7Bf@P{9uEFbd%y=m%=eiIU`NVv=jLvAnGt7^TI;7hW$A!K%k+l^S5U`5E z0!&Nh1I^i8`tDmC9;fqZkNl6VQay9C>J6_2*^HXnISG9UOc%cAxm$0x!up%j!1GZ{ zC8}k9_2yH}QgN@H9i!I4))PU(UJlm_-QLLWn^$0`o{=fOu-?Z(BtoBumZ0w|(h!r~ z<~3)!zzcEPvxC_M^sq7Hh2RaTsdG*?^Qo(yUI@D2s_f{C@kXvLZg@`9p$Zgijw>V8 z7w1hM48OOO59SZ#ORI{+U|nTX3oI9>Bjnw!t2S3H%@Fb!6|-A#@v=l2^d%{a-sX31 zlnvToza?G=SFDXbFQeQG8?7mr%45G~pY$v*tyNnJrah60FrhhaF9aZhugvhU%$aa| z^rsuf<@X=qF?z0ex9pGl1~9iew`@BWY!A}IM90SZz7gf|axv%c6`(^y0t)?Rmj{Eb zaPSm*cw9|;byWw)L_;_Dy(&Dd_#H0=A{A+KQTQo}+lM54V)Skmqg@ogZ_A7_l&+iE zd<~+g(VZDY>qb(pStP>ov@$e_?<9=r*mONj9L+KB-Y^!C(-e)wpcmcl??3sy?-3Ds zCKNg!jD?&-Zw_QExgGd$Ey2fH`ydn~u|ckw=*y42$W<`<>92bMjZC&|BlK zwIiTIs#ia)jm-5M(IpKP5PSU#!2(RhdfzlEhn`TSf`7E=4qcSb|C5UuUB*(WU1Mn# z(g)=6S-#D#fH#1Vq&jz?&k#^&BtdwTkI??grdv7NG%n|3V8~LoPdyZuXp$y{g+V8= z7(-=eU<}3&a%)dC=fqIkxzbnV$gBO!pq4R%xex+IO=W)XG6^t7UH&RzvJ%|Fdg6av zn(|HZZN`k!aXi@0K!j3R+I;H0$a|4@i0c(wiXol`<{@ zlp!~I^yTUobuvw=5$cl~IN#+vLVA}aE_aCj#NwJD%{e6^4qmP;P@0-3L?8ha9)iI9 zaJA|RQdJ=fCognPekpd1rLxKBm7S-CmW-K9mU4l)tI5+-2r;{j=LaR}nr9qLC<3B@ zDeo!)lkmu6ZmskxU!<3>?{7#&Q@a{~$-COu%ATcZp?8#0Q?H*9xL$dmhNNW}r>e}D zN}>jRNSJM_Ewc$t2(;JV4v7Zw8xJvPd?rfZU{K{X96S#C#5=kL@fb`K@T4R1;L#># z&}j{4rQc@aylXJC8-rWFeOXh&WVCi<$rNL9cnhRE^gMrPSl40Lb>Lej1fl}4f;G7? zz4G>pO^dhDxJYR9(Ncknje{^a#A8kqhA>;FUdOhA0M z8ujOp67P4#%8s1G-Nt{-@evZ>R1iGPdE3Ofi8AzEJsG$NTQ)3 zg9c6qVPSXHZRms2%4t$h@d zc<5nCv6HAW1fy`VW`St^;u{idUVTwKIG&uuVl}7?i-n5-R)ALl-2KfDPgYR9IA}T# zTDGcEBhnu+vyDf$1bO6B4bSjWrf}XV0@>!hp7hS1D^~yA= zO@7&3(6PGk($c|&_s@o-ZK`&K03=n226!mm&3P*~nihQu!_3FcSi?~s)D#)@miK&8R`)`{(A zFTwm2?l81M`>R|mAQ@2gjA--MKp3fy$kC*!v_|r$)m7ucZdQN+O0Gjc~rht&&ZG}PIRn(@#&oMD%4n5!Ip3b$U zxdCT(EQ?K_ab+((US%t5V=>oe#;{EyZb2?%ZRGZ?`UkaJh+&Zgx>!c>N>ANtwCgdiannwkC*S z>Jd<1E%&|g{R70!(AB4A7j}>9Hp={R*X5LBrPbxXRVV$J4=d<8S^%8V%@oVrSob-m z6p$|?AO|Y44KfDwNUQcq3NpVpNUTW!=DHbl9}`JIopl>4OB*E>O)z78AZmoNo{*6| zAu`kEgEL!OU&ugw%6z~Byk5lt7sIjnC{VAn;*5SH#*>Sa5 z_enEOK_|#RJ`?umozGq1)^dC%OY!Tj-TqVkd{oKFBmJlxyPQQTU~w@t_A%D=3~AkU z_;1~3pUSv0OEQM)Uu_gPC>lNOe&nR>U$qOS1!_x{8qD)J(N3=>yb`7Y=A=AXYS<>* zxFgKb)lLLn1fwn;1xy871s%W=9}h&GWPqpurGbRdEp!`crE}VqEsdw|8~+&aw1o>7 zFOD2#lHs19zPtUL32#nAaTC%7(ux4tb_@-gv2;^533Np9amEdQ#9fFGyN7{TekD*V z0UldWC<7YPJ-t{hvFW+ErkutPq|;yoWa>_MYAd>^Nd!f5VN~lR*aX|;uVf}IM^YpQ z@PLx8^E2?o=Kb5r>)GPR)h@TwKFmBn@ZeNtdOCFj zBi6so|K2G7v5pOS^ZMOfr@;R&-3=jRMg7S{(;zfGL$GjwwN*L)vRq~J2oG&VT_T+M z15Ii)!%E}w^7wc|@P1%1R?|gNEY6Zl8&{NkKtpg8k+PUi8yHF=7;n9MaG*J702Fyp zggCAf^bDshR9QJjX1uFp4F<#45) z{g`C_#2gmlj!=bgdKslpJ*ba`nIa+H_xEo1O}Bwfffra1!geC+qjRf8>4f<5m{tm4G97P=hA|wOBB%9NE-bG{R$hLFOhC^JM_4!#g zW7!jv<6Xd}<}VC`&Wqhiz6poqT=yvcvI+GRy>dWiVx|i8)Z@otII0zY>jPzH^PI|W zPO!Z6*4nl5LSvZG!Facww&ItIL&op$#HU89vK%>nt#9WuYeKi5S#>LC-=qR_&E|Zr z_D_fepZbrNpEZBI-Cd+SqI*?wf4B2E*3{foYpTML= zTu*dx0C)Juna6@G4Fe)dp?um!BzyXER*z6THCBp98X+7mu6nwF8SSx^T`I6IUaays zKs-3We~Ji(+q9gejPzdUg?*2d=-(QBo)8Ra4&2@ajoF>X%9GHFqZ3;?-<(-nSirEHb`>zh>5~UL zWQ?TIjsztm#6h)=HaT|kTRP!7XtXb0vLRY=v@W!*xp>m0wedY}S1#ikSe`DOE|O^_ z$R(7Mp_+G5!RcNXbm8>cO*yVu3eucBtuXoeYh7H5v_`}_1A|naa4cD(Ly2=tJeU&G zp+dUpdU+A*9xKk+W!~$)emn%I;fm&PmqVOLoHC~stX~~15@2;v-ATU0b&p>hVW9HmV=eo?ftwx5C7%Buj|^F z9pCF{IU=ie<&49XG?31**L96-m)rev9Y{l6an+>|KH$U-SO^sU7qJ;IP+xQZFRfRC zgI?)YEN)2w9f9VuViYRmkf-vKrxGRviS4EVLke4}vx?v#q6}{-iw_oTFae!;du?Lu zN4{G_$za-ACe_mSL~azSXEKUJ22oz_TclQ<7KS`zl6}0Q8Oihf30c}wZ@@RVFz^GwP@l|hvghnGwGiW6%EZDm2tnWO_s8Pqs} zcQq!dF$8$RdKM5pY*}jV?O6T@UKc%c z*0gZH9R24|>28pin2`5LO4<)!%&{c4qsgz0r`*kl*VMD8%hlP|UmiXXA@lpioL;EN z0f{mY@CDk}?Hg7*n4q(Y4V4mo(L9;QYQM9O`G=~p z&ZAwy8sINav^CQ;hwX5Dpu`}SqwQGI{l|dG+5+Cp< z$xcXMjZm_pbvg+A^jTU!FIJn+J8T28 z(wW$Rdl&K$aDJ#7fQFxx89smizpn<2S)ha5IER`7694BSI{YDa@RPJ@CZW41hisEW z)!4CrN+AtgEOi9aj+_ve2LCwG?&B*ovgx-axTxs43d&3i^DOtvvZq5?dNwTN7>g2z zKky7W5lacZ?1GuYP@g3WgGne&ufdf4>zn2J5WVZdT>~XKO|Tmg!+~M^1+a(wNxF04 z5p!nkZjaWyLA7#`+km;HU%`KRbSb#?7MZo6cdd`|kzeDHVh~O#ZuqOcn%cC&svF1cV=+cU;2? z*(YH;NE&E}J1CVqi+*Q7WgSQ#!DGh6lW0y4s}JWd)xb)sq}!4*pVSJISq~UQgXw(p z86;$Vn3&I~iZ`;)->7yfu6fl1P&jnh?38O}qpT__v?L|FI=mI;i46&Vi!5klM->HR zGh+k4V=3rhO<{L1%}(3A4OkIZmkUsfg3#D98D>gF>64;5P&QBUx#byx;Op)RU-eo zv}UXRb_5vj`G=_6;nPvsqA+b~y#Le<4^uOwG%l4#42~q8LF2usJM9;96=d8P+ACF) zY5WXwTE(H>^0q$uH3onhbm=FRrQ zRqgi_wC!~fvZViW|t1a)I9zyCP@;sVI-&{EoSmp~Lmie$b;IJo$B zpg;vunt96;K?4yquE)8s+Q1VlTye+c~f_WxP{t@?iYo3Uu=II8vX zuOVqa2#InMeNlDS>2YKL$AV+XL$Jx2+A&aFWg>im>X`V`q$1XsY)RcWj)4bJxbgiY z?FEQl1x;XIw)_1PHJ8(olE2>|cTjUAucvjHUq$G&-l>04;4|Z{ve7P-&`$+eaic@z zcfbsYDH2>1*4qxyL?LKDjCJJB;LR~qii}b@hCdc`&W65mo&9z7c7JknEltAdfPzzo zEh^pM{9QJwUSQn<`hl$WCo@r=e2<3zVGbd%R`V4GD(JkUW;Wl+J6gD#M2VWNuAf4> zJ798;1|BqTDovqmta3C3Xk$YFYA?D@f?$_@iI*3Vqxf;;I-%Su%+yR`f9Gz@)IQaI zXbfj?5&~FQJotVwWNlQ7W#)-jE&iXJ#F2uJm0hq}zV9fE)^$xFlYUy~B7~`_j^F-$Z{B~%0vgP) zad+g#;6v_(XOwuO`RXUsIT9L^u%f#nk+P?rb?2 zfmj(c1a5iQ=#?g;*`f@?i^eVmrv;7sH9n-Pes770N{6HB?F?k4o~c&rvv2PV7GF_( zzl<%8V+?w{z2?3r_JA@@S}>qp%k}x2m#`Tas5>GmG7@f!Ckmci3yxe@9m7ZGu6VpW zZ?tZ`d*}Ib<62UunCWC3)a=RH79QUMf3ih+8!_4MQxs~SqCNM+O0fWFJ$M+Ohbn`Q zv7UFi#hdwlJ(id%Bv1?Y>)_fBqdR}#6PbUC>W>Dg6JPD3M6IkRLh|b}>3xC#%*5|9 zLV7x^QGX6Vm7sMAwu({6SGl{-3&u9`oZWBN*huyJ2G3ihV7u>^W+92r~ zw>3whIL>TLKHff2g3iGqr<_~0?;5VU)E>91>D@mk|84Q)hoU%e(_v7hE*D*u_+R)w4Iwp?bbA~OM5kdL6*mYK$pX8U1w5pF{BiI+<-(d$ zqzGA0xAEw$!|bWGJ-eCv{Ho;d0wXd?R!&i4jHL!9aR05qN`f*$yPepUMdtDUqv|cg z>gsx~QJ@mt^xN^HG)NbAc8F|0%P?^HL_zC1YG~lFwo2%)X+z4kf~fwT z(GJaV?hDNZBT0E@g@KWd1tF9OmQ*d z(WNyu!=FkNr54YSS{9>QE}pr6ELS>BmW(Lb_nMi167g0O+6&hk%58M>PHckB*;H1NCLW^R4j>%ZY+DDV}ZwL$FUgOP?BBBrj z+cwM3IxH!UjK4>HFxjs}i1^niC6_)JGeQYJ$Tez$TBrH2`j|WixJ@Ur&)ZlL@?kPcRd3&~Qv->-f zl?)M&L$1)E5&wVG3Kc+0ux1W%skaLH!Xl~b3{6`s?3Sn z;7)Ry3wxq8pMMTA!q^y58W*aQ`Hl>+Vxq8%mKuSX`%F}$ecb+{X=fK)nm$CUeY5)F zWUK2?UVroF)BprZO@ZHh8Rs#~amTY;*iek{A*k@wBl&IR14oRT5{g5MR@<{(iZF+Y z{E7qx_Id^OYZ3I(y(I9ojhs07p)As*h7Z5ntwwrlZ&s^6YYZ)~HVZ^FLB~&ClTC44GOqb8iV<3^nbMLJXat&U-FaX% zaok#P{9BJEsJqAB>>*hrFkn%p#*OW*7TD11+j%e^CqEF6VNEE&dDidqoOm3;Edzm(uKOPEqxnC%@SM~ud(|vm*457c!t(-`A zi4|nRQ{)h6XlSyZpK6#{$SiR}MGmcVfK>nLMn%G4-=M6wHcu}60ARm>j`+mWh>y+@ zI^o^!5#U7I{f~L=w2X9%*Zfpbk?)z-G7%jiwPDp75kP-l-~{M+e|bx6wX?6Q&uN|3 zUTgM4hSPgv9*>gMmE?lD3b7Wp_q6Va>gUp=nZF}EUu(7)R0#OqNW^II=ogNk*T2q{ zXd>xl2SID3paP%y{|Mr%V>01rJ}s6RE9kU0PC#p(m0Z1_(s)UpxjEKoZ+n@#>u{8< za;?5JEBv$}PLbn5Scw>n9bq8u_w{^#r)RS-EPelFB{RS5WlVI2#XMFCh;C$RuPnt# zuZZm!%1pl4pCHHG$%4aaQh2bvThr^S+g5b5y_w=PTwy{GdHfK_5rWFO zP5`w|MJK67VbHrK;Ch0ndgQ3{>3m{qSCxGP$;}*J)$Lw0r%s9g;nmLHcjbMzH^M*=g%GF}`Nj3qxV)4=$44kl3G zhLC$-9*7YPq|(Oox{hHmf=L^NnesyuGmp-htVGX34#p+|W_)=4hpi3fPGBMolv^@L zB!2NxzEd$De*yAw|Li(+s5N~q@kLp|IpN!jCBf;*&yq+QMf4fha)Wn5VCbUmOyX{V zKh@IaajAxsB{!_3#%VpC(Ukdmt9@`0$b5{er0$D?blrVB;Yfr|y_GDQDquT?R*V${ z+q~_Q`G<{NR3KA{izkY)&o~-ht7fDyH1pt+fQ>+oC5kae2?uGfMV}wm*zC_-f09kP z#RRC&Zq3zUslvjO#wGDhF>(0+#T5NS2~w{LgbmPoF+=R((1(;Aks?5Q<9zUqkgSzi zP!p5EigmdABijrhEAk;j%j13d688?@v?#@MI)YI$e@LXaDkPTsubaaHmi7| zuft|F*&dCkNa4xl%hFNX&ZkV{N^txlMFQv=3^XgsAcl;|d8TddRSy+Zd2h}ruUD-n zz3dwPQebMmz-S#!^z#yAL^2SOIjlCiBS}SD-gQzlv0JPhjcl>}m&}*50%%zgYltLm zxndq8_!rPN)Be<5=j#dZ>Ip0Pjva>FArv5p`ny#~rG)uMz}u-Dh=7?uRebw0i?=UC zKrHyzy|*9l&%P1?#~QQ< zV}h?J(`~L;{{(I=Ef!9-p^_LC8KWRP4CuycglNJ z|E3*?0dT8(D>tqP2tMK<(Cc|1N;K-yM~s#DIw`j}-7(M7=`WtC0Vddd1QC(=z#aid zJ?k`j9i&_CbFxXEQt|;XHrBuNA>Nsg_w!4c(Ni9y?x9|YpEpXOTmHR^l+YkPSy(k* zv3yyZMkfrlWowl3*K~fY=KP?SY<~vdHSF8d+E^F#IKx8aSY@9ktWvo1&8$;#q*^hz zX8pXy?Ns^c-8<3rTT#2Uyf=r1AR?IO?j#Da=Lw0dxTXhHJ zj@PJtFjWieGw?Okno0Q^Ngp=4G9~-!q*B@{%Wu+}Ki9qNBtN=!BW!rT8A_8thUK5l zlAr^V2q4z0fFJ;fguE^%ON{~o5rLA>hj-JEa0Gj~#yco<1(l5jkbOAI(QMN^>`Yvxww~8hdOU^p@cQI>ns~q_`%fu?;NrW!m ziHeqZSO`S4yZHm(Pd4l1-5}WB!6e4{(NcvowdKv%#@`B9$=!)5D@udHtN}Bj>_}Oc zvlUuZkuZYKm8PR7zwHivO)z?a^RZH5iW**Ra#VO`Ca{1faA|*5B(E)_I$N9ir(dyi zxF5cuZu}W`g=r1*rCb<4%=lu~Kfb1opu+74aXtp`JMpUszt6I-E@sYXw=uhvG@sY; zvJJ^~t;4R_VPy7c$>$v7O)-k_UP<0Ju3xvinVnv5Q9+pDgY=P1!g$bSDyZ0)ljFQ9 z@yM&8P|#>4qN7Z_c7aA_sq>DC?Ps(ca$gVxe@w(9i6msM^*%;!F!x(sLq{LklI8rz zs!m{ch0P!?(}FzmMi&!+a@6?;8lG)n0_T+oxbjGjkLYNV&bYQThUQs=0}f@7fX?N; z-Hq$o33EI{8i!k!&B(%^)@5D}8PW$%t2OFwwD%6di8;Ckc@adj!BsT0xVTpjGeWDl z(!G;&8xS^x`F@v2Ac78F@cU$xc7OZTj)7!^R>b4(OkdQc5_P!)x{_INj4`lC61BKV zMyCzy)7Z_$k@3T_`GM00&m(dMFgQ!=cII5J_zKyjRENrW_U+;cQ+{Q z#-qyJ-~%ykWLpOD7D&DQIvtcJ_Tgd)6*3b;)U>pyy_ys>`HV7M0Kpi$R*mjEuE_P0 zfpc-P^Ul|xwxdYYgCl$FaI9X8{~kWhJg@Zq7flKkSa^c;GKx)1DVVs@N%B;eQ!0>K zmR$H##rhi)C<*iD_@N9Q8sg=N@MWLyW%cx^@i>gzqN$7;>ys2|el@X88kq`nR0+-4 zDPN$Nb9YAG^3u-3)cu~rDz;ZuuM!?V8G*TTl$D}_3;xgyJ-7KHucXAL&#)6-E8^f5 zJcA)Fv2hg9dd+YRkhfxef_p?U-ePXf;e{uiA1sq1j&&Fcdl!VU_?BO(U_=EUTV)dvmrQc}pYJ`Vo;2>d+dZvS|3w$-07Yp;XxuDWB4xGT7Zh1I^MuJmxx ze(@MY`~RT9q>D2;4iU}Rx4oT`%_*kFf%|@(vq~7!n}WUp8G>|5CduoVj&@RLHz`j} z=q{1b(O|6ctiPxLpn0Y=t~W(Oos_q1`DbOQi`4)5ShSqy#YXMyXt}nvHCNa-2ucEH zqG!w-o*QT37fIlla#hNp-eZ+@hVo||l@%;6Ek3$lGOapg6jgYJIi>E|&Wivhj4Aek zn{QUZEC(p=XA2228)!V6ACg35PQ&hhw5``ATQoc$eBrhWwR5X=EN2{fwJ!*Wk~L~@ zI)+#9v`z9U+?_3799;)ewsCJpndej!wms$0#O%#qsSyLT3^vFrpi^Kx$HJb=k+VIc zi7U%7vXF!8@=?VR>2I$jR#jlCr{~Pg^tO6HQ=9MR-IuL4!)|zNcXwKm>aJ_*?CYJs zN)@BTZ}f&d*xzoPopqcA&dP#XT#(vcKCI5&bhy?7ujB+JC9M$9m4kK*s1kOlYJfH@ zqPygJ+$m#)9yJ^R>V`xQzU6tS8bxyQLr~3r^xy+Ukl9e;tRt9w?eNmlNZt6tx?Z_= zz6nYkyMhm=3vI@e)t~Q237>C2uW&hEb@=nP=zo#*{_5}z-XJu)XQOu?D{W??Y8hLJ zU}XiJgQDHh4dTbFWdPEiwGsB#c?X+oIM=k{3wo{6tzEP}Jz6{k^sE-wIERyyx%YEh zayh4EAAT~s`Pt$9ksaq7uif+hmEigNGxedVpA$<+#~`WYMt2~oXxOkUG+f5q&Pej^ z$A9SJHBZ5Wd^T8Jb&oHGYNY8HnJhT8XVXSg!#YhtN^k`8fnkA&9Ua`$+PWSaFglC+ zwfH@R#mkl9%k_xG%5?|*i|vlPP!egZu`O$@7hx=j3Uj+yXR+u{_;9whxAoe}@Nzgl zx_lE`e$@0(7DUu(QzM2IuN#Ibo~ej$>jf@AOoFdZV*_Usy~?|Y+fZid+Ce(iuXKg_~3B7VmDoe zz>}I}jti6P-Bp}NoNta=66e4MeO!9MM`HJbE|GsPwd$1b8=?+uTuP$C_pUGTvb8c_ z_OBKn%I4jM{>*loRPZ9ks@Lr_hhDk)pSNUwvs?gfX-sXT5B%|#FQ6Lr% z{xOov&} z|3lwb0vRdtQW9IB`l4xw1c#+JWL;+7q=bopY)E-B9$bF!z1Kz9a`KNB`_nq<I;J+`)lC& z$Xk?byzTi3LOB3T4^f9{z5|h#kI;&{z;?~5|Ih{oux`dx&}_2xr6heIc1*%=-^*cb_es!13o z|F4iDP6DwdY-n^I5S5Q>KX%<8hBk{R4yN+Z7y@92sLZ@q9*72R7B}JiwA}*|Uo%GY z!XocuTXWe;?Pe_d%Hw|UM|`)PL_WnKzc0=8X7h84leh?bmYE_`eDi{Jg9=Gz8(ZAY zq}-a`FK)UUh2OIS8xs9k!^R*-a&Q~vt|z?nb;tc;^r;5K=~;=Twf2)+rV^6gYwnvL z&K@t?Et|71Pfd@npzFdfB_dy7gS4px2hftmVmP*P^Hk*N3H|CEC4D>AyZlE}xmBeO zsr-Ju!nM3l@wxyLcI*$3&_Rn$VlB^H;Mtr7G5)dEd1`VJFNlf*|R zvC8Vo`s){z&4~+Nd`O8W%qFW7I6oFKw7Bv7aeqvg6R5Dyr)rV-I5ja1u|03`^`RLb zmo09$ICQtbc1p6i@|KgPwA~jJPkon}fEVV#kCQ4J&zq)&XO$m6taALBVZ)J1GC7)3 z%fG^%o8`)2{#_fM00ipjXNvRxtd#xE>)ksk)zo(09I2N{;k7zi{+h-`VD!T@$S=9s zdEtB-dEDF6qv)P>&jDRfK8yG6oheBx|I_Y=@1jtf`#IF}P{e-u!%j=|f=tZP`##3W z3MLBS;n2fU;n)^23MlCd^@Xm7Nv5+mw&qI}ZMC(zwS|Enq>FdmHeWk1`G_HW4^bUB zmDLh#-fI3-7@j0g)tn<)pJ*K=gcF8TIbX}kk!U{p0}9Y5Z;S&CGT)VhhU|>PPBw)S zPhz-^FuBudeaUQ7Wxs%Tp7=y2q{exw8zN)|$BuEKS<`bUX-|ZY(_eYcf1@E=miL~2 zoK46CJ`|@vZo4=BX}>;Ro*YxanF5LqF-ry~W(O(4KyfVReX8|QNMYicpM;Df)?bwC zjTYj12Z^+TKQy*IVZSb{)&d(}*NkDDO7Exwc5V5o!OE5JIWKC!VTJoSSoQ1AKL{@2 zI2(oIO|xt@HMRG#bzmwCV6oPK&6w3z7!mAzXd7+r&VTcB zF+TsW-xn7Qbdgvi77Z5+h5;C*H{#BDwYTqJ6x2WY0ey|Y*`o{tO=*kH!bP$KwKVJt zGm({wf*@Y524WCM3HBWHocuxr_jCiM-c)i2Cok8pG+I*N@|`A+)f zIZ~sW6^>U|o&caA^|)F&_{#U4r>Xt%T$|IT#q#BIB zTgLODA!~8nYnv3TJmll&hw|6kH)a(uO9$z>`qX>sV)E2g&vPMGu;&T{ispi8c~>ah3AvMT8YNA1s!$>Qtu7F1?+@gl<`5_Ks7s0Z&&W-b})>?B*E zgsP}3*R83q!;Z7id$n6Sn#j1**}-iRI%9R37JX zF6>=CoVV1nu|vOm79)d*&gg}zLD5?8w7hxKoGBOx=6V*?amVf1!{j!R9iIU`zyZKB z3Y25`E`p(QHZ^cnexeI%K_O3l2?G3O{bunD%}pF8ejA>HYKAv;$-21brm&mKYqu4|tqnn?n&^@B-O8hmZKF)EggLyc4mSRsFQ7CMb{^V&cl zV49-)n>6NvhUi>^@UCHFZKF0P%+k{UxBJED=!f#Pbxs=S?Vf3ByD-dI&B|KHc&KbT z?_|1rAge03-oK3ky%k(eLB){7Fckv#=P;0oAMp>qiMB~1U>>JGd* z*3K>vYDS@7B#Sl1njS5z9Zo;k*i#PewP0ESv#}0Au$Vi z&gIVzW&53(ys$8FTJ}a|kP6zY<7mF zH~Qa!`mNIjhwi#5c_ZV!6_r*StiJ2=?t+k7*sJC~!_u(+BhExA>=`+vFbH(Y5k*uU zsgWdh%$N~Zc@r`I3_wR`|0U5kU^C4zuQ1n)0FRf_@Ii>PDTzMP+%)cE_X|HyZL2)e zovUy0xw?s^iZDt77J=!Zaj{YK(i{X#jeWgLSO~pbzfIz9|GQqM*Eh{28rq)Fsra<) zX#j~=7ctHtJnL(o_0_g~u&PK5`yInVmG*)GGWC)f=1l1VUU^Nv_A2jDR;h`_XNr)& zrwQuwrgAC#`Snh6nznuj1gh*p!KRo&RAU;vZ^T`RwQ9P;ztzcM98V)OT(Cj3k?nT* zx26%$St?n<;TtIQD`b4%HwzXK&>)HI=Q1Jn4V7E;fC&#X$V2qkK$S>QH*=_J9iF$8ChCHNVL~!6K|i;L4mMgE7sLunWn<4uc@-qmg!DIulK4-g_#cROsr{WWZ`4>Vuy*J56Mr&ghvWaGCQT|8)0LhIWz z_E&I90??A#3rXb2q)QHPARVaqskQUYjF6`*sK4Blv1)f%^%#AJj@eh=b~m?C4pT|Q zTUky2&LWlu;3@v?uTiOe0Zmt$YAt`9_Js_O`_I(W01zZ3S9!uL9m+%eVEmc>_oJBV z(QiImvm}SJ$Y>UM1Qv;#jF_wzmN&IqtCa&%EQ1me^@zBMZ2+SuR*Rx-U*V)kY^w;< zg+KVP>n#nK$7arH2rlh`n z+>gpyl({Vx^_fq97RAVs zQ!(Swxk>-*w!-^Wq*~uHbEBB!fC2`(2g=HtZwoJoM)N+Ws8}s*HessbeLy4Jo9z++ z@YweBh(d(~=Wd>haVPdYWN(PU5Ul~TIV3Tbqc#xl?|(}$TML99k*9d&?+nV1+#z9UgD*-FonnR9smM)B^GTnCYPZZ+sIZ3<0|64-R?m=B3Sp(9s<5COzr55Z>)LR*ZKz$ev0KiVn80Rnv( zTzZr#aCqxUmP!=7@s`Dfr{@e6)8t>Y0C@JUR~_{Lgh|U+b@WCWnV7V=IE_`rgy_Pw z!ZVJR*ozw3m{EPhGnbKZ-K6ovqZ772(GWjdh|)EHaBI&pu|zY}Fc7aTuBcF*dY3#- zlT$%L2I}r5`Sdot>(u~iefZXYF3-~`N`BYT_z#zP((t`KJXys10;6@FSMr~HE>(9Z zaj}AA{^yCKYIDk0L=nQfJnJi zugFH6I+4CgjHO~mBTrC;!XK6u4LtZ!d0~T*NO=!RFY%(Hg_#BCt=eaY-yv8~bjD&@ z>oktm%bZ2&DVUV$4g$$1RMR#+hHj$+=7F?}XKB;}&R>~1tVce7)rJhe#7PPiN=*U1 zlnh0b|68&Ylr958_@3HT-cq-LgFyE`D_4R7=~v)ak=V`{O)Q^LA6wUF;7;eS4Tv!1ofqpst1jTgR&_XM}0&_rgKy4 z7bITwC*>^$AqEWrVSu!$iP_oRJ9v!p>%$e44<}cp_6Nav;TS++!JgL(^tCVV0Mhf9 zZ`e)g*N@p-R{vh)-4=zo%(X%_bBFS&hjG?%G$FXdE>2J2H3;oMEWh4=IvQ)iEWzS?siSWdt&Ih zacRv7ltSKOHf|CZMNd4q-dbZC>% zag~*X;v{G)!aX;_*@ZkuAJJFo<@9*!+nopWqq*x4FLG3yQw6P_GzaF_Gi0I7sqe~_hq%5)d>vGaVM|aV6$G>c|*OOP{f$8qse0r+9 z+|QG$!N~R-MM-wM)NJ#d?DRXA;;MGeC|HQRz*9b+`lS0$tAj;7X7G8L1%{AC!>ev7 z;)@_f69q+pFzuw#oL>;A9mUvVxewHSf#_39WS6hEKS<(V66x`&C+MHLPuNT_?47HR5IZ%kee|l*(?e)q<)*38`E!UHAFv!#0vp z`DNMj18gE%N0m{nOR@j_9yYk6F|BgvW#g!8SsKcEx1Lhh00t_RQtWcv+UTc)1u>0!HVjM@lP_a6>64K(CRY^IcfJL?k9o z)W8V8WPBouW8!)ro^&v$KR~goqK~KgJG|{_W8h1@xyi9f$z}qcKGCMiI;VB!noud} zT5b5KD$VTu9Pp|-m<}D-Z>ofJVB9;h`rD89>cTO- zJ#O;zZYcLaLpPfK6D;2}$tQcKM?H5JISW?Xs!Y#WN`;*{dmTr3Pj=e8(~}rabK%P_)>v?9eS| zhVR4380SN zj0?LUqUQ;q$%SY20sCO|GC- zL^WHB+Ak5i1$DdOg`=QKJ%XO2>T=_u-Ngu_@@GX48go9d=|Ajqu+)Fo+{;)MLQJG(JkiUNEW-&hIsl; zahB3yB@-cq>%RJi>p|lYe3c3H*Ow8{1wEcJcII1!tOK7`rL!sVpS<~$NzN+6T}vt} ztz2AO+z!@;&7UD$V*jFoSJNy)_ACr^bbr|z5O=3pg1NxcM`Yo4Hk+fQR*|ePE$q)r zSdJ`bn$cmS<)P?x+49POJEmNYnqvoxiM`ibzmr$p{y@(j5gyM^{T2ZZaelD*(*tT} zXG-}>0HFRB4BGEY+O2;my^K0!o3HO&2RK+u{-`9N zJ;R&cEV)R)@{B_9;3TgURj&h8=A|{gyrq;ClrP#TitDmS_`(ss#*0N8Li&hz`G2F1 zF_j;nOIA_xEsC+u$)5mS1FqmL#Cy`Zi-EPxy{nEV-Q#XpU#r}uKS-vLet?si;(}P2 zw!^2jnzwG5u~5o;3q?VRVl}{8uDKf|Qt~M}JbCLi2U$V-N zOL5s=Y4^Oj(cObAYRe;#!4l7QEGm%7Bzq^n=0}si6|GSu8VP_Iw z2Lc>|332975W4Wx4=zumkIk!^WEtiKLa$!M^_o5Y8^U4f&Tk)gZ!XsR;`HYl2b;e8 z)MaE=R16nKjH%L} zpWTB;%SVec<#igr&ll+@a60SI=-HOwK93ycuX3f!U=2h(Oo@CBqWsKX{*!hvI=UI@@+#tL|?JTCQeq zZ~xmlkoms=_w{~DfcKmevex5>+}o?MBV3SQ^|0S9x1&&m>jiH+_2^4oYAA(0&_O!I7NMCmsZa*GO_P3e(A*)Q6Z>KYiVY7*HWDtB+GmW*F8 zF4xuRPuZXENxS%9wjVdgMF~enH*EeBolYRi*$Mg838U7=kH$L*l24oh^7tWq-sH4% zxIfrfa_d&lPBdTY^4)VRL_gr3E1@1*FK%$seBdnWgQ0^1SC&T$HAgcyosxno!f)Bp zQ@-*4h9i(&Th1un`JKPrTL(DST!8i+_Nyh$LP;*F2**LNY*Nzbdw@XGs?lMr3;*J9 zL7@+ZiAQc2uQm0|6lS>XZJx-8oDh=&8#*wKUN((_d;(8Y&_=9MabYTFd}<#iOK$1{ zvM>1_s}iV^Ca}-;!UYP>o{KNj6K_^y!j}hVmmU( z&(u0hx5~BRG;62R>_Jj?vcqe}eV=4Yf@kZ1qzU9RQ$Yw1+6}V<#&#zzzPe;0Fx*rm z*FC}2m*<|typ9oe*xhxTBM09 z|JA#I5pTfLj3ry(P{c=?fK&6pFCJs#MD029hN#Ah6Ljv#8^TK6ZfAO@*Om3!9CtH! zD4%NT9gd>tE_M>d+F?GfNGwNL=R1GR z>+6VhxmM?;EzYFfuyx7W@K@z~#{13V1!e6^r`D+1TQ@3`Q*O8lWFQ(!;9yRqbAe>P zW&P4sPW!K+{1e34gi67h6KH4iqno?$&yAh`hd(nWU=re3t;pGUQL9nXh<9Q6B4cTv zx~P3FW?0Q&``j#}ZF%6x98^f4Tto3;7WiSF(c$*+Ln4xieqxhh|IjJT3>V1N7|kyr0LvCvFDuj#nb4;2q1Y5n5pdp~4F@ZkSsRMp)MFrdDY#j<&6*Tn@>Jlra!t<~gQsWe4N; zAmejHmjfGosJYR{@L=!&+e3i+2n%7X&3oB89Dn-MmhQBPkUq-J&aRqtjO2dXH6d6f zi%bNA41j?Ec#3-S?JHEMHkfLO!oOh>8J z)T{l_*4*4QKJ+Vmj6OLsvxvFLBt|s?97n#{wKmdIGW$u9ia2?A+onoqqr^VC>+lT3 zi%a$ZL#y9XqM#Hbj1?D+GC&3lB7(cnH6hg@UZqILXiKx5cs+h3-<|Z^j?m_6z=cX> z*+f84eV@MkN9&7zc72HG5oKw#I3bTQPu2LLAe|$dGg6uAi*vC~!TT5{Ad4$k3>UTp zK|u{#m?6&*!Cq}h5n-xKO8OAy6gymSroeMk-sfZ1pF+=`pvG9zH$g!`07pqtM+u+C z3F$k>XJ+c$U!HRXs|5d(Be+)wTChi6(fE9KC^h*(k%Hs+SRGp&3QOOXRNf~_4XWg7 zu`TohywoAX7&9hX$%-3wEsL=b5_PO3Z|U$6KeEO&?p3A8%=Be?4gnbh~C!Fnj`R7D)3czUe{70lb70r|^zDa9i0~XeZ z!6*MqMfZmGAH{v&s3+HqbLriV9)hi}dmxrJ58Ao?C=n$7t#**JE zk?Hy1!N7KP$CA)*M3YMuWhC9`;wPa_3;`%DD3RWDimc|;C#@9Ot!^?B>D6LNVoX&0 zO~a7P($Yk1le$B-9FO28j^wHy@5>XtOjZGxELmMtg!8)t!JR)zM}h-)$5wMfxiAnj z)%PQj?`WQMz=Es2iA&DSzUo5LHLH~XHrF(ftL+f8fDAH#jT8dL*dXAJ5&e6Fh%E~a zMFfqOwij2Q6R8olOos)96dfZ6=}Z`7Ts2HkZ%1n0`A|b8@;y}hI#hJFdJjx)ueD#W zj3uyFnT$4v-zP;tja4vm@wVMsQJEe^W;A3N;ZI1uKL5G1-WTbCM!dEU1+XW4sZ^=* z`kE9!Qt`|2nx{^FMlKtGlTyp_pJRRp0d9m=v_l6Rk>|n6aGNUSoh1 zV=o`JQ2W1G0EvHQeZt_9y9VG3=Syhh@dWO%<(rueG@TDGje|+yvq`W{`CQZj7U%~X zkFGM@)}Lm|e2R>;oEFZoM#x=j#I;2yq`df0Jhen5`k0e6bS&WVEky5G?|WbiB4he_Mu7&;|t&nIO~Z3VnRvn-mkxROq5d!^Mh&2wBZ6 zPCV^GHCsda8%Eq`F3=x<4xS|zdk%>br7Nm8Q7*igBrV(8(rl%P5G=V}70F8zpGJ1v z01pU&9R1=^yB?*YaW`9pM_MR`85aA&@bEU6b^7a4NE&=@bIVToY~w--g?|MnNg+=W zL-#H)QnkF-x;-J}3%uby^8}ndG@%Y|SBuIf0QRPUMtG1ICXtBmpPAy~feS7+*T53lRitzn&>aVTfSV0|7>Z(OHZLt$_7$q1z4gwoy;pge z%PrL7DhyIll9-+bj8N))2Y4AdP5tOGyD17Yt<QU-fpI;;ce6pM#Zt$T z=hGO<1nivX1>NX5gudD$`uiWYp`pHJ{gl|c5&GyVKrC}(-t<_-d9>5(z_S$Nii2Cc zy}OpQQIktp`p!o3ulWlrDd+44wlt}upNceWj%!$XMwswmh-vdVMv6&u8IJw0p_6K;$iVl17p=u@H$vpO&lDRDl)D+{{?LHw*ou#!{wXpNjYQN{067 zxM_cRa=)~AmS%`2eV!*k&LYEov#r-NFjD|ji|rYepzsuZ+$qa?M2mJCvlVKX>!6+J}=l;!rpm>T?PY zz#KoIxY0@qe>P}7DDe7JaBBfemsX1Bu3H`Zo=5XdIb|_uDS2@LPOoNo`ViC zyhruW>eft)Gi)>`9Fq|K!U2t#kIr$hr5Z_PF`rpMmV_{N+H2!O0_1mEH~gw9dy(HC z_MSJ4FCIKeqiNB4`J3CK{potp zbSvS1${G&mIYDhz7`C2GlO9v;)A7sMa{so>zT`*ow1B$b$DOAUaanT>4IYzv2RD*2 z2isuTWxIl9Ghzbx6S(hW6t1@Q0g+cAl;rOb6|EP{J>~ex)K1;0lOah$fS|!LTKpGqWbu#QyL7LB?A0 zJGkySps`etwgowO5vES3WFw>i9;Vj3+k=mf;E2x&7Z$l2_h2;y%7Qxr@0TL7%(qfQ zEicg@2#nUYLLMGrPW+im0sEmPYJMu50q?9nOyJa)MR5HBkIbXY-fs--A6W@(bP}(- zM^74{y2E40pnw#%5yQb&t~A)3X$%~!9MU}6-(6YSvABw<#)`ZehjJef`vVJ6FAT<{ z_??9wC5gI)Qk#p^HTlOMW{Iye{aspS>7 zdUe$DjMSCCS!15)A(wi_U(BrONIJyk4-yAJJyq8bN62EUokj3!9LQBwC!!Ex0=HKm zendW~E*2lqKdy%NId4vwx4~|Q_idVG@=eOvii-9u(i7n7IuIDDMpVHde|2kY zYH~(nOX0ElDnjn=VOj~&NLW7HkCp6cRQ(8#q5Peb=@fe zkDcdyYt-94N^=B>7lrF}&XZ;F_AD0V+jkf8wPda}<0~`@U%fiK`8@JAZD9RCV=iqX z%h+Jh&2Do|d$EdjEul<*k4dC~2mbi9(lg}SUi|z^M`r)=5uJPiJ-N7Jsf3|G@W)(3 zKUvRNP?tZ+EU^zJTwdo15uYs0_fKNz!w!#`C_n9+e!keywbS9@h}bDB(jEfonvWif zl-;sqqj;UOn*EhdP>AIu&gi5l2;~RDYZ5LJRaV}aC8Rhs(R&_e&C)zty)3Co4C}AqS%~LhOf61Ess;iyHUU?t7MF&C|e8`T-Fnkye zi*@EmB$QFmZ$EksYEDo*Cvpw=3GWFmsb?3fjE&cZ`}?PLiwdCB~LWPMY3rBRc0CmnTc+qP}nPCB;Dj-7Opj&1YAM#r{o+xkz> zH!~OiT%FtVynFAeRco!Pk~p!c2ZPIs2}{#mjUTh1h5V6AgdY4?X)-E}G%FG&&B$mU zGzcdKPC>^bTYdnxj&wgEbeCg7ii<6t_iF`itd$9P-f|Xt3HQ3{G2{Mf`aBrkI--P(l<(*Sa{5xBV8>>ydgSsd6%$j%#SYj@X`zNa zUk(q-TFWGl=HwP)(DJnqJ!137hG(}AX%RrP2mRH;WgU8nCboXpl+J-CGL@oB+=*ox zelOe+(d3+x)cWZ*KI<=B*W#oHiE3Un=}2SOYiK{sl}tiF`p;NdF$5c|-?x>}(_!>Eo?kvKKN(tlU7 z*}hxh_+qt=leIauV3rAzIM->wNCky=K;AX-{7HvKH1=b_Vc!Q9ph8z; zSuc*YMP@VB&3EvM%RB3am5$aRdB)ox4r_~%YR)&M?zVs41rn3L8--}h_q~9iLanM&5wUUYa7Om9 zGQaRn>+fcr>F(onV@X@qrZf+YgIUIi($70}o{c1VyVmSyBR8BkhLeJr4gbPNuz+^a zZ|bqN<+z>L9bQfrRt|P<%&Y5qDhtPQJx$ChjfKX6;y39w%D0Ehm7K0wX*hEojW%;` z0~$5jOt?Bdqo||vaH0B1!9M8ZNiMwOV&s3zgxuVs%{l4uCYNm0sI6ElGIr%0k|K0^ z&Zd=<r z$ZQT9>i4&KyrzH&P=>)5zUI~Pz3Z_K3TN1|bh@gj&4ew2Je=c5crld6pQ4FWohJ6! zKCs<|y@@H(N}Mps{;>;jDec=u?&i#T&KMFJOX;7se&yqXG+b zN_#NGG;+MLNx7ktP+^f23Ns1f^75Dm@S{QFY%6d2G?gQ@95?wR4@~~Z7u3X9#WdKW zGFgY|%v25CBKhB{-amDoo8sm3D0~bpd z`V;416ks&C9^S@tj_gQg`7EmBAd&bajCj~AVkK-98qMu$01*})Dv)4A)Nio;gwoQ0 z6;zZs<{7GG_uXhuA3{cfB9_cjQjx@XmMc9qQ?iV4ioLX97vh|FkL-wv68X=14pwx# zCT7yLOjqr|9BI`SKZK8c=>qsu7NdFr_fwa)SN-p;tsfT+V zDiCliBBCOcb(~-*rEN1k%$`Q8cn$%zY(TNwS4D|c*8&66zP2UC5*kN9LiyCDBV^b^ zsPOlCrXe_Uf~q`UrIK1lxIj*anir%*Gghe`?plhV>doSJ6j*?hn4n67E_sdu5heEA zwEo7l65J&C_rO>~NLmzxf&Q7?bg2VyoOKm5S^Ux1W6ogHr(YqJ zz~8cMsLks#*r9txUQ2bi_-CuV#eZpDud}$|you)wcOp9QDy&2qgvD&6QnWs%LJ{t6 z+9t~>IDMDg`0tPTFJb+Ma(jyENud3Kg+>|lHsMgPl`}THlUB;Gjf7$Bj1KCKlL;dR zEMSo+ho<)yQG>z)*y4ylblq9<-BfIhVw1=Qc8WNA5y`8Ymix6N=~k9u>s4&^J;11! zH!#6#+OTHD|GG{$$kHdN3&|+84B%zC78kcaxe6ie}GHg+Zk*{wnkfG$9c zX+@EPb|aZ?J4PjR8PV&Yf%r$f<=iy*U&CRxs|2exHY6rAOpya0$NoH_MGIh(w3z_Ta56eG_b3gg>zX`XrV#f zAqu{+5fShUv$^=vPuspYx`f0ME%_j79bb)ZM`nFs!x03QFVomeKB)nVvZI< zq@`AMTf=q+*SV@4f7Wx^#wt>{th|~D_02SEM{JG&rp<|k$6#F@$$_YHO{BbMZ#4OS zW=}Z#&)Z*OVC9s(Tup@be?F59ELf$wHFuVNdBuuNm*0gpeM(AEM_z=LM~x|X07n{K zK4mzN3X&yP!Kz55%M)^g0xg2t2i)EtuMbQ?R_74F!B2*g_rYC&3`3Yo?Exta%fqAu$BI6UP+z&)=i9;1J8qHE z>kw!_8Z&G9XizQ)<7N2Ih7MkZ%|wQ6(?~%1My!uIbQrReytFuTUR^0tV>`ofdF`u} z?!-o;TbyZ7=nI*z>g(SG-+!?S6(nln{9z4O1WP=TEwVs0CZwB@1YN#YwS@JA?#*G& zEcMPX=wgaap#&9n-CU`QQ)hZkL-B!3KuU z{$Wf+4>NdfNW|`-%}84#IXdrIp9nqjKOX%*)YpIB_$6|H4L1*7dUkqxIxbF4O$APi zn=RG%_Jzk9zbgajdXr!*Dn?#S3x|-HBTxBlTlMz26Ngs4q=f~xg#}kQIJN-6kSzu% zgVP-P197*}Zvo2WH7-B2A>`Wo$$LQH1G$4sPCBp@kX&FSEo zkfqb^R;cCB3Zs83IG=F#If^i3$PWa&1ZK451#B~|vGwKy5MhbAIr80((sVaIk3k}{F{kuBPulBPnj=(|mf>Z~N>=9wJ zvDvbQtddW4+AKCYLjy3?#PhGFeu!K78bBe!WWNQ*t;_y*B>vw+TjqrtuII%_2Tx&{ zUgGd@rsU*Gk|k$3ft8r69UqbL6a;}DqJ{1!;aCa=kF08s~OV8I7DbAL>V}r%F#O~qmt>-*)^M`( zCv$=@QdqrK#SCFUm|6pI9PMITHEXo6!KP{360TW@Gu(oYdm!o$Ai+tUPWurxAT+y0 zzTZ}XyV^XFb>O&mTM13ORvFkm>9H-TJ5MQr>X0;9xS*SC1Pl73M2Ju+0c8Mz=-qIzA0@Z37!O=EAtn|e$uu5UI!c||=^v-wN0aGZ09pM?}@(m}n(j);k(W>0-CWI;t^Q?93Ve;7?lq)=GB zXi9Hu!{c$g^nX36w%uUQK=KN<0SkzU{g!{OnQ0W$Sl`rns$MT5ErSsUuMZb1@ot-7 z_|sFKBcg3b<49&lk4QYWcaT<@EMy6q3O!CE%!2595D9bNLuE#nrYI8uo)F$Ga}}l7 zXNd>$4^S0fehs^+3OzR4q+BZvh6Y7-CM%Ayy&?ma742q$CrfQ zR7>;SyHjE#1RMr($cRahW;YITNB8K*+A8q*86@-Nl4kN77S0 z`<}&6PjE8VjgDkj(=}}{xI-1k{N(er7P&Z`Gd~a+I>bNc|`N=`O<{0xAn9mEeJUG(pbmuIFma+wR26b6DP(eLk9;3 z-@Lp8Jgzn-C#!?Pvb`^-^CBZ7ne{7B^XRX(iZ(Rb!04{jQ<~so=R^U7$=TE^N1B5= zrYv>XHBwTO;PS!wT4+0N_(SU9tG2&u+hR5D$G9y}FQlhhne(cqz{<6$Y2D`bN;gDk z6Ypn4z*0=)5GhH(L|L}h^;kH)=8_}CxcVW<$tqAO&Um{`hBa4pYk5-H>z-#yZOFa% zULOQ=vx;$xEbY1x((#kC>qiMMW8I<-1!mwENmBtBqj8J0*WfszEf*24u^aer|0MTOI=;O1!uY>ncZolr5vg)?H>NbAG1ILwfb{L=98&CzD{aEsYyD}U4WJ#cmHvRrSfA3>K z9~ix}2Uoox&oUHn%5J3~Tp5Kne$sQ8aOE3_5Y<(PT~MHX473hr4u9PsPcW`@KJ1CCnu!GG(Xbx5WpJ#B z%ef3JN~@Bs{mUn`|nAc3SwWpqO&(wO-pZK1TI5wd`4&U2d)} zs_I<;Rw~iKb^=b|L7fys(_(+t6sTaQ5-u2enaGWqJEy@8XaBvIHF06~uyrtTFrnsT z$&flC8r$AUg9b$g!xi$#vGKC=`&@0b9*V>Ln0)GLw*Q)g+61VA{Hrt(zz8Gj+i3%S z*;=p)zoqAv!ck=pT5`8Hp)?4O5s5S~y2PY4*NwaAXNiqpsD=2uL!itHU}+VTg8vMV zzGEUyYPgJwhHPlXn{s)!jydLWPd0riB=%M1_m$EXMr^SM<5in{+-OuGh>Kp*J1+}1n#8@1-6}{{whWhMBa1XQuOC&1|Jn_x$-&s1DYCryFehbYAjHLjd)I7q zxR@Gxu%=G#?;QQ5R4KV@T~~FW#*XT(7+Ko#I@8(mTRXZ#s}Sx!9pyxAFyJ!(Ivw^xjhku#kX)SpPtA_n&O5`QC0o-EO^ zdz2fJs9r%4s=aZT4@KmJRaMHf zHhwG!BnCAH6T?nhjOMheumfMiE8}L>MCpFH&wJHla8KnpL+k`4n3bUUMih1s^FMV_ z75Tbrnk%Ix{D0kjk9x%EB3wQeDIxORn&QgX2U8{m^#uZBTayUrSS4lo8q0h-((qT~ z^B>BrLf>M=?oZ0fNOJr?0l<7WJWN@Nhqm~=971{M>|t?!zHqi|Ta}UF2u^`0d?Js^ zF6FML%eJPorMtai;mS24@YiqdP+d%67?Cx9IH6$2*D;w;#NeGw(5r?uE9y=?+leXs zWj+p^69QZ==en!=m#(xGMoq$GiAfjotS){JlGhY}E?(B>t&hzsHg-0wdi#1Luu*W3 zTJivLg%H=j2{A)>P zyViYhypzjxI_md)Y@;Qys;E-p=60a7%KErB(J;lO3Cx8o|E;?T4YHkKK zb@e$~-nxKgrJn0+P;+K2JoTu4vnQ_&wa-zk)p-HTr9@<%RMI5Mai0h*+ElD-&f)Kt zCNavUav&RB^FHjP~DSzqG}NFAne_cL_Af1{ZG*DG18 z-Ux^5Pmu}H`9`vPDPmAfLaFEYT`{yk6?4W)u~l#ero7C9<2M*vD1GE3-Bl% zrl5{F=uiAVyT{Dq6uHuD?SA@Q`4{K#%cjvjW$qNWIPQi%hJ6+I+~8K3v<}j}c7M8$ zcWF??QkvLAPEy49$ zK0aT8*O@OK9+ryI@&&8SX73{$27>j@g8r+A@O;JE^MJZIq4UZnI@xOIdHI=2$N}sh z7@w$@0ZD;905m&^kg?5mP3P?syn=$>J#%_L$KBm&O~lJ`*mQ{kGjnr+_v^j1w6v{{ zkFBjYR>!3#?9Ki4LD6qM=A7ozB%5V`dardWt^whYd=53PuT7q~{W)@!V zgTLinU0$X)2=|y%A1)UBxN>H9o}p&E^zCy$GqU2i&y%hE!F8jyD|os26D5l)Dms$* zr@qr=n*1}t{6PctVlI9^@i$TM5kJT?FLGY0q53-5Vzv`l>%;(j)?1whSx0A2Nq_Ai z{B8gTBx`K+@;Be zVuU<4J#HvmY&KQ3cXi<(@l;ToIMaqThN|=zr@=K`6sw61-uZ=BH9S_LIpUPq0Dl57a>3{w1&fjw<+CsDJy>v+5&B3cqB7f>!g;hE%b_K!;ARY$7 zihps`uql$q;9OFaJ`R5C{)HOnrw%MY#(_7)QL~_|gOsK`FcGvLx3WPuQ|561t+Wj@ zs6}WCb7vCsJ*=q?B3VzztRC!hJK~5SlGwN5 zvR@3y_^{)FqJb=REpf+vbsXJ~7?)FXu(9>eBN$T$TZUT~Ph%SQUMX^=TQagh>h)=| zCJkd9atzW!pT{y;Gb`gXAi6EGuTT+3h4Tz=q{n#Yqa@YUHEG*D?NnW~m zV@v0Xu;&=SDyVJ(zWXbzm256;@2Fb0jd@wqpwrkoigrXTDO*=xytho%>?BE$?Hb_n z%xEooOa9OH!w7o%QXuQ%7Z1(x37uN~l)0GW8w2$IVdGU~Ju52CR}uw)8VaMC-CL)7mb~~K65)IWP*T*oV#00; zSUFp&$r141SziGJi3Fv>^R?TrHE;2DSbJlfXUAOkhEf`h5M|uZ2QAI8rYRSS8RI?2 zQhOPT@Stzq4|D^^%rc4XFkCL}_-5w|QMfj%FFosD*p$7Db$?k{n4}*Pa?Xg0OMIOx ze3E92yq_ZY{j#dbqH7gKT*ibFVPj=Umk6IV6&v@0(-qQ?9L>g3Asdv?Z7Svfh~T}whna_Br3HV@gL44jWfL^ zkFSrocJAuS_=BJh54$Fh=run;U_T*F}}9$1HwN-jtEXrskjW z%6;=aFyz$5y-I5V)OQo99=xfef{Ut{RYIq~QmA78)}CpZe{T$`Fyk_hQ)ur;aW(32 zt}%b>rEA1f<5l>xx+DY-mB_D+K;Trxi`cVjQ%2-*nDRJg69((%zLe*nn>;0h#_z6| zRP(I5eU>O~Zr)$l8$Xg^JZB#ztk-z8<0E#B%fAlEgf5eH)w<)zW#I3)6?;jSP$*f? zOAvcN+DgBT5(0j36Q&o_eukWo{!eLVMhc&_?8HbC&2Mn4Z7X3gFgz-@+;yl#O)2{D zQRlk}A!mTugZ)nVF{@?gm1%WK>6ND?Ef1uY!2%$|dt4ZwXTk`&zAn8*f`V>DaUs|j z78a)Pc|Eqbdk+uAH7^R4N4mOptd}6@4LV@0%=vMvZ#}y{24QOO{F++x(vOhrhPK_% zD}qVnf}APP9=SbRVmSJ)X?1m_0jZChUws&SP6d4QYiCne@Xs4xM_V1Rr%w=nYd#iMa>r0}gKhz) z(#Y$P#Hn&Wk9*r=*Q%1z9{KqsvOG|N1g;VPN3bCFTrV%9@o0x1X|!y}Y*OZTxMMvT zxUdQJP|3L7}W8_8(Ar< zXSM0nex3i*b~3;~K32u-*eO%|WQc+SjHjo;7W^Yly&{n##9v0ug$P=DqSwD!6k*i5 zFcrH?>IU7cLu;o;%5c$peGF{e!Q##|t5`h9G7;n6n-g}W_M}tdkD|X34GqfAq9Uz3cSlj z4RpNIJKYx&*0RP(ge>l%Xa9NSZ}R$9S}uOI_>abw_Or7N|%#YRr~+s$KATCY32>+;=BxvnouA0plS^K}weI@R?kU9So&iDogErKA$Ar4O6{eEOHm#2n>JaG@Wnm?)sfftNymA@6-3= z?-WDy4zM~ZduNvx(&m7Vr9ojYV6&65mnj}2~5n_BzpBh0atg5G2 zd^pL~V;b?v4^ZH6P`8~w_;4c1jrmNK;?@`nOa^ai!q=685MiV7Po7foF$3ew>khfT zpCE4w1d74it!Qjcp@a_pP=tRcZwe&7E-C!i=A5~Z;*6)u%c?vN7Bt}bo^)Lazr|#} zyg>QXZ{Xr*!G#!hu`uz+a@<#h<-GK|cdc%$7$L(ckQDK+u&OO4WmZepZc#-CjOib} zRh)ad7r>{6H#LHNZZ~fQF263P`DL-4dnjHId>s#k|GfiFauh@yuC(EU9ciQ>0I4k0 zcAqCZWKsE;`&3jrYO*y;{9#Y1&(aaWI4iVEvpHe$V7Ru~dMoG!)krfCJ z?s5WbJ4lVUxHi|^_OzRG2@;(}^ER3k;F-6PKlkaf8*X>TsHmx|c5B@rPaLji9dwk{ZhEl=9pe<= zl_)`xYRRX=d`2Z#hA0bS$_x>YN=I{|qR*)>F;2E=k^Ws_{rs9N*uQux7`0_WgyeqN z7Ub~#q`+G!W*s1eNV&jGzn6wX+brBX92gW}ki5y4YIOC3)RrcXV8~iDK@( zS0>vV_UlGzSF35%;D%-^Fi(N(+&YpiTy*!!$0NVlUviXYnxJ3Gr&l|}S|{Dgp3K9r zV9wolpJ0m%)H%PChbI!&ggr&T+jF~rYLfU5d@{Q@JgYGvZPVi=Ey%k)BrvGagZbmK zq(nOdbT5%;nk&yib#!baOHR&Jjljpv&W^rs`dpS0JA!P-%TB}HT0~FH!R)f1;q>2>9LISp^uwrNsZ?dk7&x1ceGd zjK*i?FPnk3eiC}v{{DB(91u0;ELzL*_Q2`A)&_5@lX^V3@L~DGOEn0OdKWoA zLT(dSnfTe+(1OjT)(iyRv+Qy0%4+Dwcco$7UDU*K@wgykAqPR8)QjF)VN{sMf9T+p ze?8dyvoO$@Gi*{i@#dwaeLZe{5=0HWF0O59yVAQ0gd}(a)h0}ku%T7oTCSAKs}nxR zFfhHfCT-eeqrbTQ3+cim(SaBs`Y5`vol^q{%Xe>LDm#62bx>#Q`aZVf46t6|ccK67>pTPqaOF#l+M+23QvwecO z2sn();7e2ynah(wG0N$zuOB_SB|wsY&qmD46UZAFsT5_7s|VfA?VzK(^@`R>M6Z^GNX!I>8leSfrP z_^;}50m)I<<1EXAYm#Kz*x}C3#cq-iB0B@VPx+J>oW`LfyhqP8$|=@yy{@& z0lyk)hURn(O-n0$l~+k}{Tp6#4>Dbo{WBC2skq?Y#F3y2YW$?o?GdKDF32bx^s@>o zo&_wxTj*O84}srxuffN$1P5doqS1^Evw*YVSH{+}xc3{n5F=Sa#3zX6xo+H`5n(al z1eO;Nf%{gm<#(}W-=(W_AY$^Q+-xJ(k0&>YCf*fTf6}RX>UX8J?SWnKwau~hZlL1t z{xArpR=bz>1LEsl;cH%H$M-V?V>{@{FN)|%KoLwjCA~;y8d6(Q%k;3y?b0jnql0_e z`$3pLzLP(Xwzk!bmfK6r*#v)i4VhRuR!3RAw%9HB2$~8Ri-qdKlAY$4Tl*y`4V7y(%{i zu*_I!NZ4Pa(8WWmn6nabXnxAiB0@mh5-+Oe7>`Q(f)&!hhzn&_{IoI`1D=34S<)!} zGH`pu=fwq`D`%Rl>8UkMTvn^oWuKt4o~xcMndfjB{heufy?yfE9=Hb_0J9lZ2sz9z8hM zB;F}}t`+CHCvj_Ix%fg5F;L6)9i0^2Qa(U44(g1;7T@VpYtH6|DKp(4 zomz73@;+*w-1zGIt@Xu~+ff*AV6g0SgUp6McY_Iphl_y-3#Dj6BFR=WBq@Oc1!@>w zIVn>XbuWST`AHMO^N9tR7snLi0X4{hC8&580C!JtkQEpb-;@t z=GJ-BtobMJtIY;)n>sb>YU;JDJ`p}6yQR|lDJJ0>F{hXj{XJ0!U-8WZX zXIpae6EhHbB5`qXUnBmndxCOu-rhWQyqX}`TCH z#s0b+yAgBYjnk<{pG;ERa7L`zBKdm!h#8Lx7BCU{ZTDv0uB_7Twg@FAUXhkJD}4N~ z%8U&^P81aqHKeZ~bb&*KioN3;ZX5yrBOiLn$VHtNpZYS?>N@%i=m?6Erl;l<;z>bR z2<^y|BcUKv_E5kp1VbXmPpAI!^w`pD3u*QYfhF(G*)zYpg&@D2JoQkx{fX*JI!*7) zc1{OlkK1eK_8suHuUAdlY>zJMp*(Xp)Wn@;Yo^GlS**(C60^gtmuJ%G;I5n1fD(6C zdf(ani2Y%U>5E!+HV7v!Doz%ImZ;6mh1(r-b30ghapN-R8Wf5=B?SK)d%pugXWt3s z$pJhyHg(qY@?MGG5w8iF+VM66J4t8#i7aJlxVkF& z^Yu*Ok0(=zBmfqSDx2_m-~GM>WQxDqfx>#268%Ub;q@@$H>r~fHctEu=V!C2%oVG2Gbr{*W8+rAWU=3`&>q-aG)bzwZyE2{m+2G2MO?=BPlsc(}EmSCt z6lTu#+O&bD4dOIw@O5 z{4JK$Q05({(+Vrlv6-T#6 zwd^PI;G2+JA+*uo@Y4~e%DiPX8ME24%A)GkGp2J(X^!T#SVF#d3t?qdDEQ(d5=25( z)fSn#%H<)Ia6$gB_szP9vR1r}iFWq}jQ&&9#G+ZPzuQ9LS>3Pg2x$y(s+MAEJHdE) zjz$qC@!P#^14oX}yfDFYk;<2+h-%8HWax4tptmD+cg1t>vohbYsc@r6iY;mFJSu=v z`Y*y-PsC~ONhm-bZ_61QyY7y#Q0Eh^o|;0$A)B&o^_=zPvCyTx8@LJoSVjak|P2*&?1?MS+e5-D&M_i&~;w5 zSUm>UmI}P3skkS7J^_AqZJjMlC@@|$G&4Y&>Cd|F&PHuBmVST*Ym{e(;)KLB-e^;} zL${rxf$Yf>Y=)y?pY$n;s*@Hrj2l2JwSdOfuJv(c^~!GZvYcqS2SOY)5R89|SqJiN zHc*BxXtK{cRB^6=M&E*ll=~U+6Ds8?wSRp%6@9|;)nVVBlVsV?Tr{)T_r|~=t*7yV zqTDL!)z?FzICzJK>jT4cf8*^thstnd>c-ow5Q!kbDMa#8P(cU2k|JYCY|A_pi&g`g}QQcZp-8io(y-B?Q-7KM~-N7l926{jg&Ew0F z!SJC#kzpNC)fj&?NA}@Me#)d-w9%WI>gzIC+$$T^t3-Y~DNtq|_VXlBEaz{`-BXE$ z(@#;N9D)$&U>}LZh7I{?qGF)7Pd_)b&!|-<7-9GYbnEz&@}3@%p6C5=GMic^qDRM)~(a zi{i_>J=ne9f?UG)s;VEZ@e{e7gG7LGC@tLYxRXfck{HTH7ObH)*R4uJhJ2}}tJl(i zb0k!xpoB~n$-jv>mDyx?$!p<8sL7WC#P$O+T9*u}GDwgYzuRM{Z+jYU ze89@u%E0p>sk*w_fs>h&)8+I1rNY;@&LcDA1!jsQ(FJHxGwZ(H{lv)Gpt)8w0g+H% zw^+CyZ-Q?(uBC$QpUIR2;JU#DfsB97>-GZYmarpSG8CIc8~)T}fgN#vIlo@MA6_=$ z%?c=zm$3HLOY--zT1<6Sc5$$R?hR|QSuv`^6e1WAM_fLJa^O9KYso}=44v#5zt(kd z;b-P!bGH>vfin*YTrokW6Q8!D?NIHcPZ*KdF%>jpi&)xD(yoI}9~xPq(TtnAta9=w z*LCtyHHrQX*-M6DB}=BuB^dw1~s zI%y;w-9W|Nkjk&OpQhTziQHF}T~7+ziQ=Epl1Iogv6u?`A%z{^XM0%l`n||cO}*KT z{mnNNz(yJ4$v3IV%)`&LHjAn$U1PQCLWvzi(cCakY=#gFp5%v6JsI3rUwLco^Q9y!x(*svQhsb>fLsNRIuhfId{(c~#brlif8 zB6AR?P7IoBRjzSFs8CH{_(u|gkqACzZ=07Ct_@%ex65TL?6|A;1MX;;eC;%_b|M@DWF6-nDbew4@D)^rhThdF&so{WMe znHn>KPjYZqUwxM;X9%FqoT!k|RkFOe^dvas;T^*|m;c(n7I;738`FU7pTX1xVSxpI zaih%+?r$X)2!Zgsy&2l_db=ojylg$X9gaaj3{ckVD+)UYIqa6I-vw)`pWLf(v|U+1 zOcNWv)HESSLMXXo)Ja5653Zr4b+m}fy(*iBGYAEx`Snuw1HNYB!g-kwJ!8nKDMB+V z>2a7`=xT-)Xr(LYP5UE^=+Oa1N}MK>vzgk4g-qKqbmidzhH%KDp|F0=$3I_Rg^i0q z#y7F60%TiVxRwoq@>nZQMGeC zGB(|K-pw?tsiAfsnVmu^roKKM-oo@hvjFIM#^vI1vL!1f_FY!nDi7CX-o9Q51U`5F zk~NRD8K0ro(#nT*mNlIUUp+YUExEfidaO?x^&9dAZ}U#;IMc7bqD~2vZduGE7q3<< z^bg7FH>C{#sl})j6O-z*5f}76q~?Bc%^b)O_K`5c>q3%T%i9-#Ea_uW?At1vnX5S8 z4b@v8$FUqis080F5lxkOZ)9+Be|a<@Wn;sq5eG~b8Q)_=O%aHXCFea zZ)D{mVj~WNh`JI?H2}sQ1l$eMo z>L&4H)bTK5kph*wySgKU^X1i*;6K~;>8Is6-ezwR0AGQZ5(PIK;8X&Ko%eSi)U8qu z{Y?Zjpehx77m!c6Ojw#^Yis}hm@)Nrzp;1avwyKt(rnJ$>Mp?TYHQKa zUKPRI(SI^x!i_CscVWTXVvlnhObU`ww{Bo8LL{l^g#;Or(Zy@N|Em%s(Y0xD>U{Gx zDo){Zk0Gw1d46F*cwpgL$jR9TS}CyWc2?g%v3!fjwS8$GXqS?tuU2 z$&`E+FB1kd5!Jk6LRn$U|<(Pan_d~BYLGWRJe?cOl zH)3BcBZ;yrR|VBIYwN0_gi0t@jD;B|(x=DNldk90oC<~)gxOL4i>|)dv~cLhbGV0- z#L(5}_CxwdDB!#bo%!c|J%EJ z;M_vMscSuH3KS>~28aV!N6*gCC2=)W&a@q=?|rZvY#F-HHa`;MzR4i1b))C3uUY5n z#eflXNVQv9rVbp)tVBhg@MIDVXc3r#(ozYnjx}d-0Aol%pT<`s(mU!E*qb?f z#yR?2^+nax`AKNXT@fylVs+P`|8wg`PEJd7$jA1CW2v7Ls5wi9PKK3%ucLupV@>L^ z4Z4B1V-9Iq=XVi_srz!M!2*?hDGU2DhU#~%*t#`YvpPfE?`wono?q2fK_ZDqUNr_- zm`_P;Jy<8a5)6npbGwp4@tAEr~JP);XZ+K=)y;t&Hl@xtm#O z$#eok{!0`)i2UXE0?WfwML*Mh)`a49>%%JkkX?wqN;!9k)@6|rcjxSk%+Kgv3n7e* z&%BF5prYf3Op@)I!{8K8SPL7VOaJszlJl`||JimWZNQyG9jhEFA5~29&D+`e?sB8O z+07mftQ-%XqL@#yIN(Hv7oW%faQ%4sc-X^%9@)M&{U+P0zHp?q@m-NQGXjaOb~{28 zMc`}u&cMgP#)LR^usPlbaj-V>mz~E0YUn+ch-wIzHQpOpm?^^ zdo6iTyF21&ZsmPh)xyxiiWM{Aj(W(rlB)8AiW)WjAI;+4^|WCs86PwhAb8U7KFOaK zSCFhAn5O7DW^PJLzkPC+rw(Zxl%RsyqCuL>=l+BBq@W!vniPmO{-*(Fdxsd9dw^{@ z2pDHE2|U?BB5%$1*7nM|Fs-$`p>eJd)ywDXQEQecVkEeek+xUW)sszw4y;&l6m{A? z31H{pOMAauWh|v}{f^G21e*PIShJO31H+n9#e%OLIipPoBmK_a?0W>(W&xB?>m1!~ z7EiJ4q%@Hbwfb9|E4nHNF#~fECaZ{~)2!x*m{8)#a=so5?Q&Wwu9JnJH6ESs zF>DLyiQlGdQtJ=vl_`FU5B!ZhyOHl*8xxbsnD?)szjVIOlFM5RnnJLHSn{ETUJq8pN4PJ z7#k%+<}d*&zBJ=Jf6lmZS*|D&iRLnd6hm!S6usan;l47JJY|HW!h`dB3{J_?;r)tS z@|utIEf!VcB*|eI5jrnTXo!=rac6u(=IV>!YG93|uN;}* zB{j$L4JRJ^yW+fUrr|FaEgJulji!QjSFu+75Z;pVy%0P}tB+_3Ub9_Jn6i9I{GC zk9TZ5nk8dFCIdf9Z4!S8T0O)4xnY=*2J^T;y!v!wYB~hU*4=|d3Rp}uK#fmKe4OX_ zt$G)kM{ER;Xf+RgtUFJ2-p|U=RZXZ@fD3sH4w6%lQ}kYvqKw$^L)NrJcz~jF{MI5> zSyAOKP@%SQcY#9!r~^BI>U4iOv-h&{@x2`he4d2mWXF=B5O#?L*xOgGY|hMUV_bXL zT$g1`jRAH%B&=g;& z;merb(Go)|%^%35kdu>R+Xwu%v(#>YfJE7!fIAw>wVn1un2ESyvr60e5%e_Xv|mnPBDG(2tFwl!_r zwr$(CZQHhO+qTW=o}TW0u5;G8*SpsH1L{Ln<=&YY85u!QKVO29p}tV0`mpYu)oqRV zGt;2aXBsW8gaT6dh{L?7AT(K7GqJUECXih3Xx&=A1SQW&b0^SoJ&Ds?>^r#NL7F@x zq`hUsstQoG?yk=60uJhknqqRo?8Uy}_FKL|&-b&4M#sF&!S)!<$4>3PV5nFBh$m7}>jjk@yQM{!%Pk32OUdAXP zkH-icHBD5EcD*9yIvpa{jTx_y z4x_g62w znC5dilcvt2%Ni%^YwLZFV>t8tKRx^R)6yT->k#(v@gZjBfMvOwibEqvds3UkM$762fPW0Z*j;)Rm#qD1iQZyr84eQ=gxu*cF;<}sY|D_jrDIJ+yPJYu|9Ne{ z7e~PgTT6Z%DSgI(9CVgAi1t+a=wWm4?l_x3uK3Z@Ej~C~dvvXWCgOB+&HiAk?<=YzYZx5%T_%1Q^LO*? zaGSbD?S8fo@IbOElY*VG%h2X}Z?89aT7JAaEpGRc zj}P9SB72INm7N`D$GgxK2#LQWE4r0;Sme3OY-@pJ|id?`| zMfyCIBE%W{`<&{EHK|@V!e_1bO4&&Viv1qx!TxRB>|WFYWfw7SC&E|iKV*+fZwmxA zN&{p6B^EEo^GA*wjg}es?XRr7%-Z{+oN=>dyPA=^xEdJP_y7lb4&0!Y!ig4wykW(d zk0q!z91ZG&+0f>^)b%aWB0QZXlbG|r|8rqU3phE?)Wzd}Ka1|}(z=;l4 z25VK05Q2UxTz&2yZ?-_;;ue0|3zVV@n+xUpU zrClM`F9#shmc{15e-<-LctY6E)l8u6%`Mk=vO;n+!TwWE1_;U7+JKrFZI$~@Em}`a z^dQeFa*C$I@*)}!^9+tG(KD~Aj`r=);a-{avU%`qZrt|T-ZJWAU+43E*%1IVC&~vL z#-Slu=&)vkV6eJa_?Y?9C22iI=+4lg$maZ3{7)|oxDvqx!YJ6Q3D2J@E!fzy?|8W= z%}dSuTWJe+;ZaQiK=%b^bvKwf9ChR z1I80|&Wl!elb>P1IfDrWeg`^E?vlB8azcR7XgsyjO{Wbf=hKdV_TKyXtOJ1mYp(DAd&(ha-IYI<}APS^Pf)`CQ1Ys!46pEhIw;Q=XhOrab5km z&IxXxiW9d+lp&&`+i7COzwYp9S>+|ubQ(5LDOEj7*Dl+}oQ+iUWlF>2)7>x;plBO!Nfn5C?x^Y@8{f#2&Q zL8|iN?#|ZAvQbuM#F#O1B(p3@kiG&_Kzn#LOKA?edAV_dCV5hv0+%3%gP+Ur-1B>2 zueSej)BO5+>M$8@#{KVDdMEb8u2v~mpE^M3Aonyv14JukkTK6)YgTF%DIz< z;(F7KwezPfq*_EVUiS7$@!`d^X5NSb^oI~6!EzsKMN8Yjxf1BV#U#8azU$&WaR%qD zMAChO28&0*;PPnTaBxwEp`27=azPwa8a)@o&T`J!nSVb z80S6Pv!AMKNLSTN)_wt=W+3Q3%zK(n)%iF(TFT;)P;Fw=ugu$|(ph^&Bx;FoYEZ6_ zsE!c$D-@&kP-7&%P(=m&-)~5`jf9v*KZ!TT?Pg_4-c-W*+A%uDEaQ3mWi~>>lB2|3 zHZ&t7$tcNL6A}II1EJIQlWH(|6Ct*&OCIT$>!Mq+tCJD%Ery6-`6?dAf5@dE)Tu?q z#h+WR*KXk;I6{rz`C@kx5r&Kb-5w_|TCrkKyrNV0o3I|lA1Lp-oHo@PfLPi|DNS7M7x}bw|5@HX6Aes6D#}i&lz9C3t@=cockr8>$Fq&^_u=<) z&4m|jlqLSYk|iA(_Vq65_$Y{G3Je)KYRuc{JbVX8#>Te`Ee z^HQ(r6n9wNZP`$6is?5GU0v&)zteMiXq74!=SX|P00!f9_;q&ZrcLYQ4Y94!U;L${ zKY%v`*?Anfbqm3MhMHvn7c5J$x-IP#J)<7i(nbDNoBo%k4a4$<8dJVq?+S;&0n)M0 zOZ;+B-92sPlm-RKoy~@!alIRHx}jkS`WeqMDUG8Dsqaxl{CU%^<8g$>;C?x~x-Feh z0s&StYu+XPmFtK{UsvDhAQi7P8(PH*jHxi8X=^B%-}1&5yDh$xoBjnYO}pU3b@MBI z&;Csg>;$^*TjBOjnAH6MXOA09=<3oE~E5~Iw$v4lx5^SF5 z*W&j74)<>a;KvMyub=}r+Kftx6dGhX9_Jj-JB-ZF8R=|?ClGRQ3K=Ov$rwI0En;Q$UM$3H_$s$KS*T^g6aiVHq}8Gp!X^M1;#!t=2P64u^}FzVrDao zXzJ&Kym2Rdj$24Cq2qsVf^UtLi|-ovhbB2dbtmXpN$lJNJ?CEF)-e%p*bsy`VqRrB z%X$=k1f9+4*wXDM>uYGzT%wurig%MY{NTl7xC!c;Pp*@<`2~UG;k&)E!^gXB@w*X7 z!qn>LXv9~+mcCy=sa(|!6CG??%$+)nfd?l=^(eOzx}P6q-ADa8Xr$YGOMO#pd4*0p zQGgV<&p=z1N03;BO=6tF9FON{-F*OI(PwAxU8NTX&LrA~+I!%CG~!7PpfBmR zY7y)={vj5vOZY&+x5}w@m*ZNe6tMGf9wgWlKju@y?_h=!u#EtCf|_m53uU+)N2CwH zU#ZF#J6|5U6Bt(i8iqL!FZ&tc&%)wiKx~g6C3MUc@BIz6Kh}m)XxLycJ@81YrN2hK zmo1&UZSo2jfU~o2!>!uzDip0rD|z7k&4u|l$esJX2QjXzucVbk_%KP7`|we_S=v!F zR`PPrUfB`033~|DY^eO+PrWcHO8bkIQ_Y^CZi{X1@Tj+|2n7HR+F)*4RDc93fK&ik z!w5Q3Xjj&qiV(KMCPrHT#VBZ?Rp-8$E zNIEeNwKR>|?tV@ImW}JE`|c?%J{+J#7Tk^RTnCTye%$Es6T5g#(T!AwV@N78iHbNG z`n}fJm)2N|V7u|*QLDx<`bQ2us9(X0E%)}WV}pN&+U=~*~5>Q zc|so>rgsz5vr_MSNY3q=SB-CgBsHgGPiqCzVh~K+AIJr z8o2Jm4|rQySzk*F8^h=wA-B-Ijz@iIyC>+`rd4l`lQv^WI6CGIveER^%1#}H^&#sW zz2$ukyLA5|-_%lEp;0Y9aLR!u?~Uc`|1Qf0H$=p`yyKfVO_KQebwrH@N?^aoC(E)T z_&WJB={~tT7z}yp-8t`v`7|4Tm$9b~OnzUDw&P!99&3~KL2SmAND-pNZsEIprsL1R zKQ`pel~%%J#HTKQ+G1nk29-B?qTLi@Mu`_4WXe@fHDjq&mvGI^sK5dzG|2ah>ggH0 z&JOB&bc}^2laacUReppL8*3omCZ$-`XSGU_bF-^S5u8MBu5NVJsw@vt17#2gNG@AE z8G84Zy!se$Bbu0sn5_{*xv^tFbq*(y%B`2Gmv(n3@@NtHcC_aI=Jdi|p;0#LC_O`u)RQTL5O1m}dv&(%I#K z<0O2W-QJ?c8kwTPWxOz}gn@7cKA&@)?@Hu$PLjdKeE_Y}rK*RiS%)#5rcg26*mWr2q%!qaaUv;V8v@j#Ch`zjShaC5gAX2D>BOjA7W zFrSAQm;jh*a0h<5-`wbN^yW_dYLvrc&ATYU>r&eR0q(4WKOthQnCPPOZU#O@Ngoeo7di+8b1lX!a-4p1+7+sHPsaB@P3vF&+pH{#-nNZn|*El=1y#Q z0TV3vNJL`vWS)uXQ=2eNGYrEuZUgr*3SF^YHL}Xi(y(FVqMnhe^{TJHg^39X4oo%6 zrIfAaG!c7R7~sqPu@|z7CY7|>&@i9FHRoPijK9T&9eY)kdwN<*$@PmgL3_IZVRWoqggs)`N0$UmIP5tD_@Z6+%Mr`Q*{)|@-G^UG*< z33}gy*0NI_dG%tERSoHw-)mG=bE=jpdgcB7mb3>jr>L0%?Inmq!6?So8`^yX6SsVS=`fJ3l2#uKZTS;aUlZ$V({39%b~G$x~wlWc`mMi4yJs`s{ zmr>kn2|bAzaupeR+)&tvo|O`Lzg=(Yev&M^i6Ti4Gzr)4XFUYl7i%`&H8VeR{&iD2R;(OSYF0P07q}0 z3GhnpTpWr;h&a;`V|!LIv>iw+Oi?EJ_iF8fT4%^K^IMoHcS_*t=wCazo8Wk}=3PR84U1L#&nIVqJ|W{dmUeu^*ynVu zD9xF?frQVSkSyGKX?>N0hxJ6I!ZhM|Bwa~z-s0l&P_;>MLTjK@^+PU^#SVt3L-+RnriAElDk2WhlGJ$^sgA| z8mmlm$B-@z3_rmmJYlv<>mv9B|YGrlZ6dvSLIsdiDYnn%}k8%SUZg zx*OGX)Fz9NXWmf~i+mwKb7*blRHHj54v-Gf;7wUpGspmaE2*ap`G&%>jcq8 zkfB9-lLjL7B?fT%oN8W=4si_iRSMJYJ*}Si&f7>gbF+seej*IAf4j=)X=?jh64cgy z7cJGn0(;Oap^=G~udKhiI1yYdt|_}Owkd#80;2}f3JY{NX`QtUuPXl4b}nR+(%Ta9 zd%U7*^TJ82^=GgZ8ip#JM;og;e3LOQ%#hSXLPJdxII0uJlk@@{D&udgKEux7X0cFf~8fnt?Jp1A_H}#H6`|NRe!=?F2OJ z_#~0j<9p*I>uG4JsTdX*Q5Kb9cK)&vTU4 zMTe6oh}%X7*eTJ9OL7RjHa>5FZ<5e(-8fsI8W2thtpWr#ZC96X>O%|4OV~~E-CQ&g*p3e{g$5t z$r0B<+C18YB{sZ=kXW_by|VTdqR9DLn+xVI!)va#5aC2g&@tXi439`md<(qBSMHDw z1y%i+jKhypbgIEDE(?O%$~&$3AtsSJ(hr9Ta*8zX0=FW^M=k!Jx73Tox{U(uh-c81 zIMh0EZ2q*zU%dg2U@Y?EDI1PlPjb>PnX*ypZxijQQz=+DQ3gKBlO_)vX~jzRQ;lyO zIks<9;vAqx-dLjYG}yof10xi;aiL#~4aXev-Tu$0dDMj5w3eNaN&NIp%MaG92?tf?4gC7Wz*XqF) zL2~;6-~E>H3>wrpUTm8tXes3Cu^V`~_x7WSp(>~GQHNak*ls;QZHY&R#hP^>Aauyr zcrann*n`-W&6WVPMgIv)7%_nu(U7vSj_uZWQ!pEk_vP`?0{=mwnoigv%%2C}jzEo` z0|(@}e16%WzEKQOX+Sr{4{Fqyv3)GqJuquhJ>KmX6)*uAi`yb@%XT19G9v*5JwuE=4ilqN1AdXH}_&KP!@94Z5Sj%Ah;c5npmG1 z5Rb!1>xE9rhMu~DQM-bi3fOwwo2#3@5uwFcxkNY7MP3dI)G)7d1S;?|qqZM4YFj5d z&NlE^V&4{N`%%0+PZ0P2hG;HgqdW{%vQ}|sa`Q2m zI(`@yzIma=_a5Loj1)VDYyr2H7LH3(ZX%Wk-ySz!j42qbW=4)<&A+kZTMa#AR|U)I zArd2=cqMxOZ~?=PZ`J2<8oq}e&~P3v-142b2O^6XD_yD&M%fa-E009yB1p&AfqyFbwzd4|=w9I+b{;MGpg<>j1;@F2(w*?)R2(R=tRE zH1t3+D3AoBF-S?Tpbm7GMJ6yzgI!$UG$M^WS zpVRhzjvrp@JJP5L5;;=yR0R`e)l)E{X`a^`9|nkz;R;ER`c&Y6nDQv-xV{f-)- z%u2kiTD+mtxZEdga2#IR<4NM;=1K&2@M;C&)Nt_UZ%Fd?&K48Yqhd1r3PS3}0xOZ& zkM>W}PXh#m=yMf8lw~=Fb8y`ZA!X(Hk2#O-3|O%p!s!X@ZadjRsC1Y#^)glMRR(vp z#fJ|<-L(Pu{IE3wq25+oFb;u_VK>?S?>{_NV_i)@#J9ya_axMZ&?Nr>eJ~xg@cuv0 zCxpgav)fm)Ek+QGhLgI>OAi)e9TOS`RlUbY>I{KCl>(Ae{yq6V*+Z;*8=aC$Rt`y{ z8d4@^TkrGEz5mPphmVgh&k98CJJA8uj28=ur=cY{xJ_fnWaT6?_92`c3z%*aXdACI zExB6YnUqC3NUC(N-&=Zq;QjRa_fcZYy7PK+$1H4|yl%%H)swU2DD^5Vg=PL>c3n@2 zg|O%2Xf<9eyB8ra{Sf_N#X5QKDmPxrK_-r?m%ArYKzXYK52DYGeW)J>DxP8u@CCAcP!v~^=&-F0QTYtlDLUCX?aY%lG zTMF|?lC{9fooD`|im0Zm&?jYe*YP(}7#THOB~vm)tO*29vm;E2+dra559ynDeT1*M zc;J2Om{?D+kc2^NRQo~Cz1C@jMpP7V9yIx|-Jy1zq5jve1A-q)k19I!7}^56njutN zbH!@+GmM}!&!V0xjq-H5atI0Iy1E8p4B=|ynO&DQ95FUb1ZCvyaUD4EZ5v>__#2GPB|&a>sz#(e=#v)U9z7mD2rjn0FNL2pbU zo|DdJfpysx}kt~r~UM*x*dfEZRfP& zd1_HTekV%z8o!3`VH(~kaNhPQNIRH+ew(y5igjD;x)|GnBpoThHP|dxZkRbBX#8Lj zge)ukzqGoH2ni8#tuE5e%{(r;+$7&kcss}CpX2inM#MV9&3RV`aa38nfK!j!l(DA; z>U?g(Z&*@g^ctLG3o&YdmxQoW2piYumV4j(=j+dn@|y~$fqMar*aG$>zqXlFF6P@d zOweY)25%N@b4gY97bxHINsUZmkq4t6lL$|P{WyZ3*URm3I>AO~CFMMix7#Ir$rS4~ zE4Hby@_NZ~u08842V}6o(ZC%cxFT`#UV*>KPn$+Q;S4APs5F%#ocBJ zxPRnNNl|Hsa>O4B(c3hM6;0QE~6^rl_OJ#fQd2pFfbBDZkhMVxaldV!{y+gfPc*HU-)1Bd`;{I1BA5|5{$Qi9Q^X)+=(^o*}r zf|iPqNAS_c`^J3Th~o1|jz3Rxl29HSREetTn3M|baI!<)y65?c~&+MH`&w4(DxBu^-_%!_$fWUmJEDV>?E?i|h*^ceBSsGw$8P z4~jtqhfL#N*x_E-V21d!8P~ZzgQRjbkCY9=I}NuPy*oR{bk664=FzX*uR#19G*wS9W z+yfTS8%>s}@Gw_g(QgCOuNVgaSUbUGCeM!@3Td`R$Oxc+Wn(b~m9j>th>e@pxwaEM z)c@sTCLw_fdvkEEP{@DHbhAJVwl>~0(MGbIB5vN*H%lpbc%>1-x<|3s(jCgM+H)O` zK!vS=yb19Scbkh+)L^0pu80B_FMWG^KmXp^0iq({7`E<^M97i5f(hB&b#s5ghP7~2 z5-2ZPt?*Q2q*-SQg)W@#Ru}|tuA-X833k55yeam$oW9rjUSp)lvV|!?bX!-AC}#k@ zb6hUR`pafIL2f0M(gxAcfP_R9ax^w*qMwcb7a+FZb&R)5t#jiED#SSrA{}rj2=Mv8 z4~78>$}`N7FE0q92BJs<_GpJ#s1By+7!A$b_@{y)GU>mr`U4n302LorrVe;8z61B6 z04Zw6_Mtoz12J#NKco&QA+Ls9=u+S_Zl4&0k0HRW=Pt3imu`Dq#KYK{PuaS`n&uMl zxkU@L+Su4XN0E!b{2~jLkf5f+;I>M|cA2@LT|u4j9^`2kxFKYC(sojv!v8dehzl)Q zpuo6`Ws2NSp-S9#{SH+m-)hD;D*`)ch*&_{G$k1JVJHErm!Jys90s$@mni!daXA-} z_%%uXM(sxexigUNE}3{9a;LcFCYy9?J5tG(1IL!F@^SAyjhtnunW;H>(5(P?(azH& z*yNk(=Bkq~2$HLzbFgIi7;-~g_pzdYwd#ssGX&9eG7-Iw_L`vDack$RwylzAUGz`v z&NX?gB3kSSt}$f7*}1Wipy^y41^ekqc2d|V89<8#q8bOjJ=7;LE-Y|TFnltL(BlSZ zJOrmcGDynWFB2;*OR|>^wmFLZjQ<501%71gv&^eL9<%nZaQ+q?_w*3;_1*aOT%YUi z@r$wkud$yr<=u%>S5UlUyCLRNZD{CeXC-ZzrI&%rTRrLXmiRR#dqquk&_EZMyvyZg zG$cbZ>cXjlRDsMj!T;W^YXrbYtuQ&$gT{-EBG@pfB=I66(i9VCyVq$4oc=0tUoILT zbctZgY)DZ_9ww1P z(b4(-9Ae6DdBs<^vydxb8aw~IWHm_Un+gP0s64&c>Mx_vp8(u<9Le#|PohG8DYa7K zGWzNH-5pHAto!}@?$n8XJT1ppq3PW8nG;qtUbtpkPicLH?f)432!FGwqnr$u9LzNe zSP5K2R@r7k^4kf?K#a4eZN=SbKBWj}==q+j#M7>Zu0^kv+M@C)<19j0bKB0ZGIZIB;@N zPPe$++X#a86n~q>-=t}_X2w-QYKadrZ8z7bhAk_shfw4B;0|SppDNbqXpZ&%YE8j~ zXS^3FNCAGHeb7|IDBpiu2**uOy(n2TKijFMsCw6T-rihO(~oE9K|IWt%_5!bC3ccx zqo_IuGH6O-|}XKDZL;@LtATNk&KK!M$NA zRhFY9s=6%=z=kK;>$`07=ElAO{mm-YDq-LiK_z&-^RZ97tpoK5)SuQEGOKW;iVS57 z7HzmBNAo9vK;+^GCVSmZPvh7x;ZKWLS4=ebOr(JrdR1X79VuN+b_#d-&~@2i3t<2t z^dDM`0TvP@SV&pe`_hNGbQE=Ha)(Ej&y)0x*UJ+5P~g`liV$s^%8eW{%Vhn}$8yp5 zV1-m#o%!zJMZ5tRNVndH{0h zsUZt`G&J!1kdpv?`XINr5PQI6|BL=nz%a@rDEM5?!I34@Hoc=PA8{3y8c}NALm{{! zn>q&n^TyqMuj9KmKhF_@DBtTp!CJAzHcZKMfshsmH@oix6JztA34}e)uh0JPC?RlS z3SPJ|u*Box-?{?cp6^u)Q!oa%v~Z!F9fBjn|Bl#n|J2 zNHqQhj0#FTTvKp2^80QNUYVrj5H}CjIBxd%&K@tY#kt7Kz;_u^v^c770z%2KodYmR z9jb|U2DG!q1kG(|OhF48rPNBqBn56JFs)mSouMe~NB26>c?5@Fhas!*dUliSI%{p7 z=z}GhKqeTGB$R!-_`I70-W@V;S9Yhz+8oh*q9QX%A8V#Eh6b|$|M=Fl9H(N(jq6ZW z1b}AtuKrE`7_ROYNvl?IS-q)C&pdp-Wr9qp`Wb2lfb~(gsan@=b(qAY3MV#tm0MbH zUaKw>QF)yCWQM~bOhHAG#`I*qb{d4Q^a5H-qi)u8(%|1cM`~yq?L4 zadF8dtmhF1qb^K3I9T@kCEihI#zex-o>wcd;E3EQv&a^0HMGKBIXrXnh*%L6&qTH zo<+`wyGUv8HWmJ_i#Zd?$rVc5rMtTuf?Ow$!#HZhGmyc7lI8VHX5b*Iv1fj-KQG1z zYu7FZnRaOip-U4F5A5G}_Yb*FJj?-oIq6B*6((?(-mvM-b2rr5wW_IhGt1eme)jKU zq4*gS0+(Tgdy-U*&t2`7?_1)z4|Qr?m58v=6Jb3}(XAUn8SEgo#bwK0A1F+qn{r*d z0vvabg5uMgdN(xbG#x`xxRG;<)dx%C8O4tBf00&zVBu5}w2Dy)fW-$;fYC}ADO0Pm zQ;qnHTC9+&yle&z8$THcY*(>YaNJXg`9SMPkl#-dmN^gK@?M7AaXpH&bkPucRrEDd zQQpLN1^f4k6f2cQ$!IjU|5M%Eeoq@+nu5+7W@e$1T+Hz*a2LycJD~jNDVG?gAc#M3 zrS6qmr)){ylgaS>E+5AoKtbNWj>^T4x$mFp?NWX`ar4Xr(pFenp7Y8p^0cDmGzCA- z*mr+r&QCB@#OlP{AHX>kNguC(%E_%AJG_2ZBa0X#Y zs>WI!_nEVRMv>@>>-R{BZ_1UXms37T0$;aWzH2?RMg_S=t}CT2{C{x>4Jgo&GVrz2 z0HNQl+rE-kp#Wc@$jEg0=!&BWF9<;Pn8*Y;&{(k=WLqEAi82SLq0Wyj3u2s62x;9-=s+iM($j*{W? z$H0>g&PDkO2@MQxR-h~Mnu3Z7!hVIqB;w@>)rP3U9KHN)0_s_X)Fhw~N=ug^e0cE} zu2{L^wJKGn4j@ATDn?V$P6V53AaekDrPsabG0|5W=9!im9V*SKW@sVG5n|x_(dbkYaI)U*GuS~m zX|5197S`9mZAZ_7 zdQVX9KEpm`P5qePup0$%q9&HQh2wj+*X}v~^_!fRG?jD8)+vh?HU*|L*H8A`8o{xL zh~Ty(fAx)*ZhWnrKTiYes)ZMz^tgP4Z552_+1Vn6OF1^|aIBgK!*RgzS~bC}AWmJ_ z;Nx9lNf1T9id$K^r8-StduB{dhQMuw#3G#)1@qbwx|OiuOsvDkKWkE`1=IjR$WblFf0c#xDksmJQ}Xh zqA8Q_Gxq+-k1^6r7i9AogY3y7_zbNxRIJCr9BPzYs-=O}tT~k!Ek^%oQe^NP;UP;3 zK8g4nj=Lm)JPO|>#wr(C*u@LYq_YNUW@93ueE#pXJWYd?cn!XHnfX4k`UO6%Vm$Ft zJuT1*u=%AUN$KMM&IV8eUxNe|2>RVJJ$R*Ol_lN*7P=1K9XHGRTIQe#u4G z^J)JiCHY|J^GtQ!$_hx4hIlR8aBbD|DGPQQC`Z+(mR{oWq;D|!E4h+w_oa{?qC)78o_iW?EV00^nA z`Pc$gB%2-0 zBYkJU)OjA>mYvGXORT3LgTxVcl>PpcV64F0*tg+V)NQypleCoz^qbn38n47waJKP5 z5Qm(fU=>2ggc-FFiPIu4XS;@`a$0pVN&X6koTHgvJhvDg+z`0laOkdm9Cqu~%+|J< zqG`SHC>gtlmyP8>I)&CF=&pzT!x|5^6R~w0D-P;hS10vE=wfZ$VS?V#hOi}W-Tl)f z?WMqN21liyBotSHqQ(v5EdWj)iclxMZyTkXeL$j!^ojbArLBRq2{ysARA5=)`8!L5 zu?p@uXn{Bz1&L#ZLPIIk;X^B4*Ffa6F5m$Usb0bd3 zStm2?4mT&_#$nZ^wi;x{L-7jzADVFnKr?#g;9f@?Co*VOdH0fgTHsug_tY3E(i4vo z>eG!2gm3IZ%}Y)#5HuJ^M>;&dV4qhQYJ}I_rh%~NoOKC*GcrC7-~0beKoGq9^07dY zD8YklV3!PNX7fJq94HKL>Fz^e$10oS#Z^mFV(bAdEkj*;*5f(W~nV@T;7 zR6w+wQcw{+1#i_VUB1-S8?Ttfj~;FK@~5J-WxGJox4p-~$`n~3kfTp-&Jm(YZ&WJD zG2p<45`W+bhNq`#z=qA@v=j5NwKE+!_^WeLHb#Rk4hm6(mM1<@?W4ZE zea6M+)coaG`yuQ9(AEhVV#!Bd00hqDPx_4x$2;k16tDT?u`%y~cUKGK1 z7Xdg(cBEnG{Fa?kdRod2=apSUQr6J)&Vi2Sg`M$rca)yr zKvuUAY(~N!OG*l~r`pl)phW7JW<2_NPa@;I%8}Vx-&dm><`%P8C3UI3kAlnbsRx>5 zL%s9k!u}y?_xl!@u6zMzXbTbAuAkJzhnNwii+z$E6xdfe31fu+;xUiaayjQyq890( zV2QC9d(EB%TS_SD?G0Pv(=hfwi(J$MX`!lfh181Nknhh!S$f9D`u$nI<)lkol_JJS z`GK#8iL}Wv4LGFy4sH&13Ko8&f8SR(s=H0(SF5L-H8o-J~8dkQ*(b`0`EW z6DzwJ=_6;EpZ&f2td4UG-!q!R1s9H+U)9UeKFeS(+bi~oZq=?HxX;eeRSuOmFUk! zKaC`rLb9&$9WF$hvGFGgbi*{?nvt)^{<9e2ZcjnAYTeot*8+_ZGsfi=w$J(GGBnf&P>=VA=_J2YT6~1j0U-mTJQDZecIv7{!Pc z$M4gbfv3&?^CSi40Q2=Ks3&JgCwj9&l?v5C5kjKiNO1Mxx9S{IJ>-qP^ARBfVH;~$ z<=os{|JT;~_YuQSQ%lc3Jsjz0MNssx0F#vFYMn||7ZzB#7w$uRM(kLc6Q*EQm>(Y2 z?&&FZG-3<4DussMkTqWi(M-~bh7Gf4c@JX3N)`gKiply$bvtUH1a{=re5v>pTD@I% z`lV|1SC-hJ^X&D^0T7bp>De%_rL3JeZYjimAVWMS>DtfJxt?U`p<+&}9ggOfl$Zcw zU2-b3RV%S&Z1Gy98=qRVgCMW$u+of!_^&%eBvRZVK#fwlshNe0)+1-Asz-{*uxTOM zlIP;Nw!?W_%(`x>N%Fe_uR1J+O7a^}EH}vBVxI1IWBcBV)l3tOAg{!Y%=Hp4m zK}Gg za*W}Q_2kSrV>vnJ9Lq^Pn8a-yDboWO0Qw|!hM`CWp!3ima@GHr-dh6%28P?&BJt;I zw8ll*fyH=^LJ>CS|G_UZ7<4ZeRqKuck6488`>LFjrqL<0GG18MvpUpCmKFT-)48Wd zscG4;5&hM-GMC5o6}+zBZhUMRw7ZD(#6N>sIe;_P)O>!rRwV zr%ZP6UKESug4K%OR}}d-*toX0m{~5K->1_M~X(dhql_kZsDf1i7O;ZZ-Cp#M!L zZoq{kw0jG)C36TcBXAA&MLObv-GxP$wTlD`WJZES|;F)DYUYsKqRVa8m6rX-Mn%H1ZXN+WJ8J?FtX6Ew^c~3-t#Eg zQ&!K0A*9NC#%SuB>Z(L=C_Oi7*<=^c8VJUr82UrjG7r zTEkmwZ@2|iV1cOOf7nq#Q3B+U5xKHE2KG8z-0eCLEXg+m0UPT>CunsjBzc2uOUqz% zOHP8|&v;>qa&i%|<{9SE0LtFep^Wq#IA~F##q#U*{zv@}HUUm34)HOz6;|g)XbX1j-YuRHNxKI0FC6 z#z#qSV`m8D_g=pId3z=-H0afl66=+-9Ua{+QrmIh;vULB%nZDL-vjaB!`Cn0KK~VF z3{0Cc70V?XuVG+lInQ#%C6_XYnpI<;$hwLy+WWtS@_(Ko*ucx*Ka8yRKhNcV->`P zYyTkVg}a>Vno+~$vs4X|uH4+s^Ck6!YZAX$Gz?*5nN*pZ2mXYrc6!A8wT9g*1VAp` zcA%qU54fa~Ai`}3JvfR+qD8%g&3#5Z9>)?Mndp?IS^e6uRd2~r#VAh$=>KpfSt3Fm&gmik>=q$y z`Y4Fgsbj|PA@Q3ykn7-sz>hd$Tv$31vn6{-?x(-LBA98ANaQtuCYv;BWz#QEz)+bwi1x{ISXWqNU z%DbBZabM<3GMKyL>slueVmmXAz^&UjNRX9t;waLhWI&0HUWa&Z$0-iyHVUR^q#r1x zEOk4(vYz4x)To}g%7_O_fk>~>FeUl@Pa%arAuiU`=& zrP~~_Pu*e?Zv}87KybaH|5xSTi45GM+v&u9rY-VoH}!F-L}SQB=KnAiGuL^B8!t9E zoDFX{W!;ksMjP3>-xkdnl8B~ScHRkI(s?i}Mp;V~xwvBT}npZ_l%{ z@P19)#q+y26!n;luf?98j>etEdGOfb5V&)c@z^2o3~l7@DR5LR@4>F-A)o2V3tR}e z)i7{`9>p;^coru5H5dsdB*gQ69lV!3zq-pkIUc$U!{rKoLG-^>jg~AqoW{0o8;z~Twr$%s8rzN4SdHzZvDGBsx%YnW&phWxo@dXV*=wy?Gc)Wg31w*@ zMBN`F%@P3KpA4WOVJcE4A~ernR(N#dW1x-^+jHUiI3@gS=rl9FD3iDBIHeZf;Q(hN zxwoZLzqSz;E(FPQhBF(rR>Hn6yzANm#yfKf=#|nHL7JLlR_5O+IOnK3Hn85bjL$7J zCXW*TM(*K5$1X-WmEY~g=K>4c0aZS~%0ia8&nJE%y(qJN+fI2#HXpp&j1W+&dVVI* z^6k*|JvK3pqX>dd1#0n31z+=l1yL_nFGO<76t{gmAosQ$r>3((iykQwF*g$J6Z*X& zv&@LAY_Wnr%}PG$VokMc12!lz`7{w0g8x6fRNy$oM9rK|#i9-yO&d)Z2~pfQniRATU98}=YZiMU6I zd4Jwa7W6xNUil^E*;80XXF#1!8Wy{Lk(HGbB|&m+d~^>PwEIgL#(5;-i`!8gX(6Mw z!>BM!*>VL+%yCDsci&g>m*=_o+P@QL;oUtb2tj7_al~6f5iqM_5x_r9Ltn_5hJ-Q< zFxzwSXn_`y#dU2@guOS&4%(d7O7|msC{o+95rji~a5|wIV-d4eiZ}0ihe1Ii3j*4y zgcCup7HVX=12V$LKmK)iCE@RYYO|JNd#;ce-|E-TVnQgt{f8ZQ2FVd!&P7K>jj4~X zIwCi$^fbSsxoM7?)-n7aeja^#O1z8I=+YbSO$dz6sPdI0Q?W}Nc(ClCe7q|^%WlJe z)NzWMuNxPJR{uz1Lk7P?Pb}h5))22mDTl$rB7%{Z`DOUFv)VQzcz^nh5goyV`)-Nb z#K1x#%dZfy;Y>*1Op@j23Y5y%E1&7q(5NJ@%SGjrMWUup_>Tg8gS^&9ugw{CyG$W) z)mn!&7v39AHsWfpZmAa9{hL_~7b2u*wRs(rBq~0U&qY`3t8O#0G#CBE4es!3X?^jo zc-fzen;DGC(6fki5v8f8FW;PK)lVYZpJVP0dF# zqwii{El6@P!U!Ut>p|=|?IX!9nA3 zYKftgK|;4RW!tZ%#{^qVnZOJ~y2t6cyU&~b;*SD!D+|m(oHovgK>WDldrNIxO)Lzw zTX6~fUn~W&j@RJ+ufHoew>^y0oK6nk@B>Z6PTufCG3-D8i{{NoB8{#hVNFzstTJ_s zCR@ooof7ZE_2dYU{YUPk>|^MVv3!G|7ytdbAc9XXGtX|~3hK|c#@IFC=)TKy>cr2f z@@D@BtqXgB($`J%#m{Tm&d-OS&Xpk<==RJu&leRLKZ7|$KAaoCwjJQ}=&WDxMd{ka znhsk_Xw!X;5BC=|Df2+`I(Srjd@_&iw883MTE@RySEqDop#uSQjD8PQy`Kk!Jd`=` zQ?d)@aeYRwRCn4I<2FJ&0u4Kb3~P_P2LEQFkpS{Av>u7=!d5(lIhe)9IcpF~n^^>*wc*9T0OS-Ddfu02v~}za zz>BzC`>@uNa1vOMt>s{SuGfP&n5~-VM z7$nt@6VZ46Pa67<|j=_#y-O6E>AMoIpiQB@d zm<&i(sS{=zZv9aT;gT}y4XK{j+fp(iMV{^{~eXF%^BD-&4 ztnE07Q$1SI?qNa3^CN1qP;RnONuP+ReKG}t!+B#&NTJEU+V5ETFBjV@w=0EONg#%Z zL;3pZXZwcrS$w*l76hdtc9)`HkJk7sP()Qr3^@;**HWdGh24GC87A^Y zv9Oi9dysD&4H7)?%ibm4$4;;V7Xl43g(jD)1kW03Bh>^Fhy$^P7kc=F?w&*z5nCjVxC52tUw4}>2hqd|o~ zJM%;3rPgkoz#HC%2-(Hs+$9Msl6pI3Z=^uJAW5%GeCaGnaaEZjHehnNhs4;A zbdOLIs@_Ssj&I8SXIL@zZr%6e{WJc4%<_JJmwSj*sCw$(qqn&p9&S5thtb9FwnF=q zALjIi=j*hhgC zW02_mySelEvfb~#W=zl<@Wch-yMk&c;?Cacuqww<1T& znMCW(@F`3v$M1%ces0^B_4@D%%6|#P`Q_5*eTJH8JRq6_N=Y-KH0(E#NPN!hyGa`{ zM9mwg!QamwpKIE0pVR5WGaEgQb>{JZ=a2J0$FB;7(4z;gTClUOIKXAMB-^6P%(Y^K zI-!usJsR7Ke^~r)@)#T-7n`(1tCVlfaJZ)&4MfBFxW(Dak^8}(-g%EfHFE`m(o_Tn zO32WFLYh`x3Q;{;P;&@m^)p3r=+HI)mTP5a1q~ZW49HV@Y}y$0$mC%sh=i;i0?!$` z6p}3=LI@h)!5bS)j98B*XNKsG22{fMLqW4=1R@mpKU*Nw>JoErW-t8)wbRP^Unv0)i z*R#BoU(PFxtyj)!?$U~5W{u$hE}XbgN3x=;(I?F~v1m1pgcP@4eH{DurySR&+|!B& zNQCH*ZTEHmuP`T}ZaM3*6mceatz0^YwGrjppxN!h92fQ8ZT4ySW4;#9ULr>TbV(bK zP8rL+F|9LM8%ZMQMhZ;YGmGGN*()?0Eo2`FQBa}z0WN*Ahz@BK)e@bR-`eabw)BswXg(?SQ` zhmG~$7IXVI6qNPa#$sbRpHiP#aziNQf!JV8D-`$%9FG@$>+sp7KBUJwI2bvqDMj!$VnwDcRb^Ua3KJYNF8340I=a_6JNX10_$K9> z-a}8g%@VvK8JVkNR4hP_%akgqm5Dpxxkp|NQgeiHJIDd*cW}RGd|dk-2QBXQMibp5 za7X4aXt~R=3qIR18XNj3>Bdt1{Rfz#0ERiE!^Rj_zRT~W?d{K__^s#hD}Q!1B>qCc zmq1mq6aTK)x-_4!^oBV8=wk)5`R~z!Iu<*2YmyTywCb~*-7Y3 zkiS2x>k7NX2wmYuv-j-{;&lo32ftfhg&wfc-TigPXwx=8OOt78;U0_I@Efj8OUPXG zPS!w!-GlC4Cg&bs-zIu{W_a0OKk$S;M;-ft?m{%_F6XzfONjqb)%#l2)6)~vSW_pK z!S>ZjZiLTTK}+kRFg@=*STMBXAc+%L{It;fpaUcD)X_t$A^7+7l(Xlf)qQ-|wY@>w za_mgoAF5W4utsoWNh|si4W0m`TQqIc|Ibh01t}I7Z_wS>nM>O{s(?hAcSq<1lb1fw zUfU8Z_#VgyUFvLhRswRzN@YBs7T?bJW_~wcGjDgnm`kDp?Xw285dL1TD79Mz3nO7R z@(c)!I3N!tF&vh@dWJn?D-|)K&Ib$z^k#q$RN|ll#q$MM76hpPZ{(@U14NQ_%S71m zx)kQ*viPTch#?1_H#fd`F^{F1ykirqcN!*!?W8J(>-5-IMT*SCTz)ySGS~!`?3IZw?TFBKHTBy-OxygSH+lgcb-J;q#C3uw2|G1ZwsKCz(y)Z=}@_pxtz z-n*e|@+}qfCu1Z)5GkU1S6Dt0CKag@>ZNC)V*6+6qZlLMITG9`qu;)znKXY7;P>}jg>}_e#4)?*o z6G>o0he-D^GZSUZ+o9l0*qtl|A{cK!=^a`9el+K-w14=Y`vM{%;y zL1w2zlH*pi7Ii25ZZ{-mOdvyU^ayYOPo(a|3%QE+Q34+44>5)+M3^L#iRwi3vPFv#~X- zWcvh*8@;?H!e`ph*3r@mqS!`i9&C`iN}xxLZo<=%5NrRPZ<_&5-JD7}Gts_m?8p$< zImL1FzO4V&`+4r~%-UI*X~33cv`G?>00ELAZXYg_AVE;L*#|!)DIIj|Q*FST0q>~J zs}V=ooyEl6%jR#Fh7lwZ1N9T63@i}0~&`%|iS%R+kuxZ?!NHf}vS3f#18j)mE8 zcwwDI|IcP=`p={|Z8T?1?EmWwjTm5N_dy~{61TaQZ@n6d>Lav%SB!gUUp?@(9gQ=* zZ7F7)0)eL2JY%s`mn|{khbmrDPhFrr9_V3g&X`0f2#|4v+*Y`_;dF+2av`2sYO1nZ zL#TK$QXXj`*aEo;fnnfL8MX3x}uP{`-L%#4z;Ptw6e&|Wj_*spa;ezm_R!JmkT zKVPq2s07-cm{Cv@0Z$3A<8OkLMlhnTO<|}mmo&-INRjEN4S*f2bb=&OgG_XRaX#1m z&w(D&(Bv9QxpCkbhv)tt55dh!o#c=;S`<@ccLFgEQXz?$?g`XhV#`UPi`5y)t zjeH^WX3_A@tnyKMQJiU|xoHhpRIeckrAwTll|04q^4Zw+6QIld&22mC`Feyh57 z0Lu>gbLF01-|kxyooEyaK|=&LyH|%ewI0glQHd~^@s($ zVsn=c2G*Brn(6`aW#E&FTy`@{u|0-T#p)&Bp3sKL^7Ff7bX)2S-G6TF`2X}gPi7SC zc3+21jE9@wTFDF*(=laXAOP(Qx527y^>wK*T``Lv-Gl1j<~^$viZB1!DGY%D2b9#CE>tc*yhE zR~<*CU>-)-WAW1ZUUt|;2vex-P68sYhYkwG0B1$@d6IYFd_k|FU8twTLW#qL z!^u?D+_y_rAFtlVA)fdJj##-KUV}Z4vLH={`Trz86Cn8SAj)W-;3$#*=a^MtLWUvj zb!gr3=%_4WfSj_LYi-aVK?~Uc`F;C5#am`-c2(Gn9ELGs$v~~CRL{TwMV08cL$tk4 z5x&hKWzOV)`>*1Lh@c#W7UXmDadJX!+$*=m%q-^;cccRzVQcK<#6|!)HsOxdi^YnF zO`?At49Mc{4kn(QMr*j+)UGsw>4u>jjT$Xw#yH%Kl-VA4rgCO~m}=GfyWGw{JjP;} z@^T(DH0qSey2HT|NR2oaIyBcoctTr+nYrqDubF^*e~`HI53YR3Tpe;@|NaH4qkYPxhFwBMq>v0gZ~#~|BT=dF z(psa-+XAl?V21#9oW;_R6_EG&_iF%pn~U0#7ljB7TBkOH5cGC2q_ zC0+B^DIRV{PcvEVyy|h%EL7(YG!Jd(&UbzP=e5}BZvtI2CaQOBVFniI1#*Fm90{xn z@pYxn*Y~AMRWj=J6lY1PETnO~al8TsrA_qJwJwDA<%Kjh`hco(K z>UmK;+EBKV?~7uGa|k>~(g7ZMmriOcaBK^8@euKwEk>yv9I8XmnjTHF2CZIaEUela*=I7|g&WA1q5&?@LTgs=}qMkla02qHm4kF3z3@STvd z6E9xZBs@O|Me^UXDu1dCOdoUe%KF);B18#&1>bde-l$+5gkZu=2_r22g|}xC>M$DS z+g~4tHcTXY={B74JTm@JrVW8w=R|(Ki>*fOVXy5Bl;dPjALaB#$R?^_S_U z4c?KE@F7>W-yPBTK(Lpp{!(?)GPa(%xm(WnHAz=5P603sIUR||&v0__xOR7cdRg1g z^sSUEoWX~1;pk|n{XHn=Mdq)sDIzl{O)nfJ7uDQY3Jg}vh<}li9{kypEl?q%oCSd)L$8T>?lP+wVXtT%OIBpE`Y)1L;t!xmYt(sg&iQeK&R|4vF`A&?af$M@z>8wHL}`Sq8($mp z323IN@3FMS`qvq$2>dXzloPT zXY&>UOLI4OH#3c@2kwl;N&4fT#~sdB{Kvl2Z$H&c@8IB_aI@Av?#ySdczr4xtECoH zuh}-cgYh@tFZjFpeFvE|K4Wl!Mg_8sg(`d!$n zHc=T4Q7B3|UHsW0rpaK516c)wExluK)T>DR9FOx0fs=L_~yw z0+=L*WK%9c71SrKc>O*5-S;l_*O2+KTL%0B%0jFV~22Fk?}Xe}9rdfuC|!uh597$Og|bDoH>#!w-;ul%6n z5MyYReG5Cw7i2EQC8{DRQw$WQ#N(@5H+e!U;dp#0n^tqlNiIu`WLCL+>wkkkx8~M4 zUDjI>$;Caf&ql-F3fv0T-Og1Pk{Nrtz5R11F8*%DZ)xdi$!LX8%J=v{If|JMzi{ez z|FN-_G?sNYckeB#P|Ob|rG}G{CX_UzrCJvofjJ7+Uak@;yjM%8s}%5rYAuLZYs+ zSGOp&CjAH}QR|l#Yezn|uQDir|yCW84i9 z8k%x;B9Xf!6~sF;4bg+4emYS;vE=$BH=RrA8X*Y!J*!t!82Ig2O(7qio0P6xJXo$7 zvayWrN-eoJ$b;UBj9fqjAaB)EFGTHs|5F8!!E^evU?y#aLNmlkB+Is=^}zUTsJQ7q zv#L3>8HQj)i{Bt1D)E5ROf{cgU}^|MIEic>;*DXCXe29-q^eg4eMmo&1f|^w1t3Yn z*n96}=z)znXsj$#dX`Z{VvJzNB;oRL-k%N&hct}3ZW^*hPrYsG717W9J=fCeZI01^ z$LaQO>lCd9eZ};Hx3C9Mk?H3g>r8j!=XJ$haByJg9c?aBd^Jxda8MeJE&0Zd3`Zmm z@1&fWYEUmrHA1#(^X-Fv(-)K;Y~n>q@Cs~dx9vGEeg@Lt?X;vn{zmQH?hSZ`b_leg z1=&pc!AY?ko)!RG?xx4Ew)s3tE_IpE-=k+X43slqDq1&T6cCVL5J%KiRwEsLz!v_T zmE=>hl5Z4Fb2}+B6SNf?z!&hi&Z^O2#3T90g#(YYhTSjOSGCIA;C^I8%=;y6@Gz!E z8;1%2Kpqi*qt2-&QN~hzQk-r_VxA2zNN`D@bO{{iQ)9a^C5yRujYadA)&F3gFsGER z-TciFxQ`Q~DY%BOrBhQ~Cr;2uws^pW_xue>20Dzc2mlF6uRyH;p}i9?r1(IzAX#&; zlL1IjzWtN!8R7thJn~@4twnR$=4$kb6@aCv-Q|tO9A#ees`)IoJ_$D+W?}=AuNj2N ze7$a%KRadYJ}W=lv>OJBGHX)yK@lABcXo%6djitoGsZQB9sr~nGq|Sl;_6(3`^!V83#iML5XGd_PG3kwtMXxD-hl~u<|EfvBS4#7FetT_Zj zWR!3}MvT-|bSUpP^tW|1+{-G=4Vh5m`0x8|K?$v`WbvuEa}ivw*d3h29`U6SwooZ} zm2<&eELesn!bqL}e=1*G}jIA*JPgQ|A8x0ln=RJNKrtnt5(Jt9EF=T{E| z)Pid3h2< zjS_v8niOJ3a%F?$Df`ZdNTEHClZ;W(x0tGF3_loz#T10|maiN;3R}y*GpFe)=}O7e z@6dgi>>>-TJ7_An{1`;RjKzr%1$S55{_S^u|59B3RcFl`CLFxe^eglJS}ZL=hdb}r zB(j&C8)x(BHl36-rOw|{q&}1%)?fr@h{~kY<|rOR{iOCkdz$LnID5bFHBo7N0T(A9 z9^~&1+UWpiNM%wYa_9gmO2x^aQMN(IA}P3MgiW89LXE{|%Ids01Sppe45p&O7ik_{ zdtDFZ#rc(CdS8^>^t-APWA~Yu=1tU$IRS7`#DSn(h6Hc=5-1jAjwIM14mm^jTmuX0 z^nt?GM<>=7k@%QPwvn7v^(9s*a@Z#1ekSE1L-WOzBdYV$aiEZnwC`z+?i7YPm8-%_C0&>|h%(2gPe-x81F;=@X6<57`1fGoV>J zato}tMp|ysL~O}BMJP_ts*vnbn2sF5CX^uO_1H699bUgC=g++0@}CUl%NO@={2zGn z%NLKHIr{Mn#uIhi=xUd2bwL5^?u7mXK5FLMhJpi$L4^)-qcp~N zrX!Orw8#D*$?ocDQ-i?^EH&xw_;6t)LQWlYSVL&ny606!!s#BK6X0*5M&T$u@v9FjZAB*5zH9!dnAe$yb@^UzQW})G!GiYcT%YTw07KWyf%`W>LEqgiG`e`Jgg%JqO3N@e2nvX-bIf z2gjlCHz9bpJSedviN{c|Nqe^aD6qp?EI0QApP%(;F_*4eO<#0XmUY4vv_lG)aWhL$ zD&xhVN)B2B>VCk5n9__B+={SUSbfhD10j5+Bs|sfD`6-)y=WmX$GBxJmF~@_C5W+Qz<@L`4G(!^g$~R zUk_bdB9}sPTgqF@RUT2zU{iKh%wd7cl)|VeGIl;Un9U_T@#yg3IH$CXYL zxFD}JC$iXa0Sk-@84a;*d?BuW-QD}L{qr?UD}AN`QWSsEwlrm;SHpFwyhj`#9;@>sz9q_c=Z$K7{q zZm$ZI6xaIVUsm0(4H_ZiNh8=?vVw`dAH-3SKXXJ7>n@3t^)zUl6PI{ z8bjB#ZL-WedBG^FmyKxBa-9WF#U@8wbexsmHT$J}H~M)*q3Si{ks#WQv-2sed&yMS z%uePb%vgkoP-F2tK*wcZe+TmN#m_GiQ6%`n41}jhP`5``VDb_o4{X6n%B-`%XQ1HnzwGpqZHz}C-k^Lw_oHQ-0_F5fRL#Y%K{k7& z&w~mLT%Aret)9EGDZu>Loo8FML63{@q3t@yrJ&PMZfDc64i5};uvbj@c)@-vo6ZIm z5<~{Z;O}6i&C>BSabR7r1ztgF_CAatND`cO2R8ygLx~FqNb+D!W8G#>*}Y?+w)-4 zFva(9P-vOnP!b~3QR#bz469u|fWO_d5-p$^4;xaaV>+KNxxs*b>vc#ZJXdY;q zxioIMm?)%_6T)D;opYM4XW>pDcZjORlc_abV$hQEl1kEOt68=1q7X?_ zMbhCdJs|s?#r4#_+mr#gob>o@6N}&!_lrtmg}!;0IOyxImCIj30{dCHbB2`u7w9Jx zY%b}rcpyIf9mR4ROVH0_Nbb4v?43TcF-em}Bj!v-$$b7{k|;Czg0wVxYeyv~TSpE$ z)SOAEpejC9$v;6Mji?KA83{cPdO^#bmVLkO5r zF?Z6=QI`Qcfam~QF_qx)BQD3s=r_N+=LOz$x*xio1$bR*Y78`!6B7@$y4t=PpAI_a zj9htTt$9ard6!#(ySolt(9q*w4uc^@5`09v02^ZyH+0q1U~MXVjPBoE$8JN_WHD2) z4v#x>z7FIm22iG!9B+FZ>l(wXu*4?Rh>57zQVWQ4?~iaOpTaU+jW9}cq4T*s*Asb^ z`KUN7U5Z2+)WL6;9d6tETv(rR*loBx!EBJ6SZRO31nmSo0tcrVOOzVt>fJ@*T@I$R z*YseT>$-vigP;W&jRv=ML(@7e#q)j^DAwjtrXR@+`|-?3PTZz?R_v>CGAj1G0qw1O zSJ{7Qd(|}Q9j)p&Ij{Tyux^4YN#L`9A>Us%eZU2hr)Fk}=|E3T3%9n%I2#(v_Q+qy zY%;Wh2oDpuneg0n`TEtH%M{+_{=MtP>1bB#ro%hiVm?==0-ozP73cISZD7B1M*`Lb z+<$~7M(CKK$el!KD1=}lQNYl|KpLQ4`XqmTF!^KMp>up_7e3THQwEps2a-VoeCoF+ zKCB00wONrQulZaALns>CUQEbuEEYqpS|I_JSv58EQ}R=f#o_&EjX}TS+68dIp~%UY zrjwi@Pkm!~ zDd>lH8=U)C%|;^`IKf4TJU=SOVeC7;p1OX#-y9RZ-!%DO|AN7qVHAhP7kd$+Vqlb< zdhn{8fa9#y#*P{q*tMWS#X$(^7YQvQ@=SOh5u&fGRz=71*sSe+FODbCh+i)C+^|bm zS+seeR{&oXS30<(*PA)_;0=4mm%x-A&ze-%BdvHu4W2A^f@!*55v_rF>b?6#DZm`A zRx{e$s?Umn|D$v=>}o&Y6Ge)RD7Z+xtDM zlY+#f3|3RA=3k@+Qm*VEi3oo;g^Rr|?9@0Xw8z;n0-gaVq^qtG1w4eLr77*Oh=tM;**d&g+B|6tm92<1n8IQ&O@$iIVF=}-+lS#{IWpIO=Yp<|YBmuT z&-vTfwu|leY5`_@U$uQ{XK{Jp<(#rXGZuPtCmApG1VIXwGMR2!rGQ>p6fUuSKDO zz7DDXi4{ExMJ)f(7VAju>q?-+*#?tQawMAEOlqV)UV6dz#TH^jMIW+sB`ny)l`J` z<)9%x)JU1Qj;F3?6&%&j*1bNb^_GEiZZl|tb7YOY3uiXpaSRRWax4kJa_qt*>^SRe zn!>NUCOgC`EX{@#p^+F|MZN0lc@ ziAX%m2%P7N{+H$8kUVZl*tcwQW1ldWg*gdkI<@p;7J{TH}WQfmxAK=7S3 zFv-*PsgO|c0gfwAZNiGEBP}hMfKBXevJ&9$+6UIveSEYjbvp&27h^Z0d=lAaqT?*(E9tI z5&Ebf?04{`3pdlCWB>vul3H+KJx1ot;YZue4LB(k>9c>BlfK>f4gU$I!n9S&tCGm( zcwxEb!Iq|@Pq@^s*=(I|_$lYa+co(ai@a3ipNIK~87{2r#E9FLmagEoG1fYj94f@i9KlhX_c9<-AwjwPBw z^8d-o8WKPNQvM5jDx5J&f0=>_Gl?X$8D*K#T+54SRluODZXokPb&^&&7|CP}vZt3- z4Ovf9mV}M6%QYD0`mV$CG0BRSq*DFMq9wN8-{0{s@Hy`K>}^r_6ySS32n5ik@?=RG zj(7NEu(hO1MMEBfAL=#VSIz#m6?6^=#3)HO1bZONQb$sb2#T`LE))N67l4!P@Lpxr z5~KH@(6-OBF^!Kte+_wYr7S$Ua`QC0>eu)1tLG8pULNOGdi2=V<_8drUpsa@^5EC6 zqa8iF^~$k>Q9k=6)d_(Ka~!Y@W>jX1105|!dN_P1;5LIufVF4EKvNK$_k5va-|dWw zOvT4om%d03P>Z6}oH&26QPzg&mLov~>VuX@C@kSq9`rnfr;}TAqJnggWd29Q>|-x_*<<8S1o%D6J|8;j z5>xJ|?~P?(OKcV-1?ri%4xE|77UUxvCXx@q9EY?&7NPVdU5 z6C{WG7e@6{K&{B2!b zo3^R=2lKz0JgE-(Jb5aDqebh!d#3xUA=NxpT1EZ(-y6UpQSb?g!y_zoC+9gGhDK}a zR&FiBER=i-ra!A+WRZrE%;=90FRmI6#d$0>+i!B7KfwHhu)=-h2WzPk8X(Vs|I{N| zbu(u~Yfmw-Pmb6STf$t&+y0xq&zHD}c5ON~sCeN>;lu_HZ8(WH1({_O%Y2i|K#FX! zbsq2drA7PIdYa`cd?m|dH7hYxO6cOUKKVU2E5dfmyN2ZB?G^u@Ci+{HSG}cFgjxd< z$eGU&RwHuzSuL_Lk^$SPyX>MHjhN0M2N_-GV?w?ZSN>dt2{E60by*M8MI9q4N(Z^%BwZ0L&_=nj4f5&U7AbGZk zBq%9SS#HXWZoYup&^PzrbhgQhULbafJmM$#iUZ13?~t8`%EnOIZ5kwMF~PUo(9%5| zD3O^aRpjTNONcxu;xoCND1cu$XScQpqIdiI?J2NI;jgV>l7%zk$QTz~!VAfNGy|Qw z$vBmlBs7`uxV?&c3L8uZ*GzA)g$ag8t0wotV(Dikf^*h<+w|T}{Q2L`(jO-$o;JPL zT_ZZAC*m=36e zY&UarLGp~Jr{W*<0(2=)!(_C?#uny`4w{2k2f{<9;I^yZlFF-!q3q#F4btO&ZIJsv#^h2I5)$TLMOw`PT_V?q&zBI#r~?e!Ms=kA{C4Bo~Tc;5-GY}Yy4 zbK-_NhE~uRSHX*yb7{j&$9HoS;2jAAZZjWW)vr1x1zX~6GM4HmK2t4m|wMwUm7V9*#ud5B!j z!dnTD7iFj|udAyomqg)<#M&Ew)z#E|j?k;6YYJ@v$tHbrBwy5bY|?QSJ;!t8Wco}v z>D6luDv#c_YAB}NHr!K>7EV4tb=wPJ#w^Qj$64g7XAQ3mO8ovv9H$Le62hLaG(GB} zl^?78+_$bgFhy?PZqt6>s!B}9MG)hvP?4X6aD!P5>-h_I^_UukEIV`#jzE@4SC@h^ zqq=PChNNi)YEh!{uiGLyu5#WBq5l^}S)i0|0~?3wDIGhHZxdvHvVhqh!hb($08i>j zcUgxbOv?QuYN^{=)&{h0DJim`2uri6YA0&uK^Qt6s=M8i+EI{BIS_rV!u^_k5FCry znoQ4IYOw}qiA(_tV1f@4|C?J|V^w9P84_=?(~zA7j9za&Z;sAS(nv(=if=7P;$%wf z4Nv-Z(U$G>TU;FJ2X9p!u#+T8umg2ywfb4im>8b69c|sWXPxWL-}@O2TsnVe*Xo_k z{%+y=3>0d=%IZ@%Y!4R+F8Ynie3X}FR zqPC1`lWcmJnf}CgH@{vG))$r?B4S(q<%1r{{1MLBL%^{dsV+4v2uzW0r&wGXD{!bK z#PZ#2-c}0lpJpEbl10pr$Rtim5{l?i~Nt4eByhu zG2x0XAd_Z?#_OgvK4JHgT3Y_0i$F#8)BwT#@6yC<^&hk}2l>7DZ(Fq%X9X3r&7z(l zbKqFw#~+0md0LDsC^z%p0zX?cG-z8ooVJOuQsm*+=VbN>@lu+RwQ3-2DunR{nmk~6 z{uO|ktWl)!^eE#x&0_;snZ@|Rnu3de{QqYj3X;b?Ns}^=m#G7n^__ualglJl8xU#A z5s1CcLKh;F&4@zKnL}gikY)v1+@;fsf_q`EU>)A9?2z>AOMI#1|&!zr3XmEZ)h_qH|i(bgFIq4#;Ao|NH^e=yb-xtbfS=RUZ z5pS^frTmtp0FPA?{cW46ekMwi)TZR1D-4@nRNhw|GtvKux)q&=gH~l~3VLQN?0?w;IOz zS2-oB0{oOD@i_N1P!B{){r4jastg#BUsAdnOF`C+PZp6rWw`|>a0|D)EN0#gzGS#s zg_Pu2m&3@2Sik9(VUXd(HY84zmmMi2WyVTYCv1)h!%HjADpqc%!`>#GKP5n@{k$eNgHu)3`26X4y=2Y7 zU~PByI=w%0wc<+kb+Po`*Dz;6I1gXpGGI0^A)rKxDc5IIsg_`3Ud;=<1+k4OZxOy-8Uk~B&DX37`1`}SAv++M) znV;yZe&GLcim`x7T`>~3J7u(Y_jM>H zT~2O%|3Eo|HCB@thoIJU#uRtWANdwhGY#xSe5S z67UU4g5nD*TQMEF)sZegrH}j_#$6*o;>|*!5Em*^c){^6cwJn9m>%IPDLM+%^S}rV zo=XD)90<%I@XtX1M8z;7#F=%NIJMOD&YZ75Fr}5q)LLA`NGbdi0<8jvqpfHK9%kPX zzny?Q5IuLv$D)NNwwyNCZ@Rx=68;Q1UVOEr^-RXfhb56#O6d$s^MBi|dm^lYc?!mzK*|$@!gIPH$8Nd6&cB+>l?pDps`1NV4Z8TPJ_m_oiw&>HnwdijcvQJZQFKZ+qRvvy7&Ii^}1O2 z-Zk&cqft=i|2hqjQF@-D3&DP)0UeF3HzTo%6D+tg4IHVmMS$k+M>pY}0IMr3rohJ!YLH&aKI!BQwXWL7Nr-L8fh|7gLpGlW-ov`b+G1Uv=lRbJZIWt(GEHUWXyJR+ zE#CJ%-A>Pv%LGD?2n3M| z$*`Gb1}{>dmr1A8k_F$3XR%n*I||KXq)*UTxwXUIudWTsCPaz=(ljBWG5vc{H zb||g65nmf(3HMe;A6DLP<}vnglQz3t`79tt)0^OWxnOB{jgaGtCD`dY@gnlUfcb&e z`Posz^^|L1!IipmUMj)nV?vS4+4-HC6F|0OAhDo+l7)2w82lkU+*UmA2&lTJG>w2utbuivu4@T z!G87K{W{bdP5|(~9cOlyrgY*zvffsW!ugguZ?PE7LNfS`_fOJubk}X?J_{7R}#BR|x+RsR0hT--%d`;k3oMX>e8$1Rlm&+{ier z$uoK{U&mmrb?Tw0m3iIn_hH8XkQsVi4~lip5gc@}THg(IRZ5=|+NmbvaB>mYQHxHq zHMErF^{Gk0u{1 zbS#Wn8$2pj-cRdyK>k2Pj0rx&QG)Vn3pjKQuQTi3$bibX;i3=!g*<`%DhPVAPtYaN z47ll(Sk^vKCo@-^MFM&LIUU<|1VGwEJeH}L@sB`}B47q$R;^9C|uH*@wbcfZPGQ4)@M!#XWa$f*})Q)q7aX0b%Sg#5P0xt?e%DqHr7 z&X|9SCK*cDGsy9{&$BYU;HU2qZoBfn)pD)N3`@1ccRmmG9xh%2$c59xO3Mi74`<-a z{iC8rWl;sZuqKzs{IDZkpW;ugak(E*bwMsbkf{-r9ph|fK?^DpzxVG!PaoBH03&w~ zhxD=3#4+Q_)##W-L@FltE9%g;Yy@CGFJeC8h<*6~cRk2{LpF=d1Fhd!r07h@joZ^A zhrU?(>P9_}v4-Dw_j9>J_gqGKAG$yhT_C-r68F));xG=6rKihDal_O`&I&Huw54Nn()3qTtW5`O+ z3_{xTV(`N&sq>mE=WTS))U+rCyl5-^w^J<9DSTYI->77RoU@i-@sg!rc(Zc!@|#pB zLpNHzQ%`>tRj}X?)8R$)yjtAzL(6Wnfn`ycFO-2^09jg;P9Xm`HFWVRZv62EyH+Qvt2g1WZwVOQ zWDIf72t>7<)JT59wg#uCG*+ z_Ul5aKWLp8;$zl~B+A&_JSLe(an!kW)Vr(f{l|C%c#~N<)$fM*dxAhS?Z)@tZ`?k2 zO2;n=4Ao>b2P?tH3()hjHjcCuzvod35T9#6Or!8SG2ISZ7%+@6bZ;0Y3lC*KxXL&e zY>1%Bma_*nifv-p75@RPwEO{5TTc6VuTg!~*CiA>KdMg6KJWi-Npa|JEzozW30b%U zfx)Zn-s26FxQ_I}AHP}K=NL?u64?fnSC#q@68lW4nD(*qM;ycF`Bto9QE3|j(2F96 z0?C01rR5@;p}d+PbCJL&=z$sf=@E$y(ci_DLDR*>qY^eeB8R_&QE&;4jt_8~rF8}l zKk+0xQwKHVK{mhdfoj#c&ldiVp_SK&Pl*cL-u`Eb<2u$5xO4}VPh4(}th zV*@P2ELB6TyQO$ffG6wt_=c-Cyx8&? z`TZd2SUePkwsWy#kGqBLx#=6AVo;A8g^Ko#L85d>B%s0kM94HuV3T-?v2x;oP~j%2 z1jTOMin?t|10T(|74vcAcAml-7OVb6P5G0U2R_M)nkxD8lveduR)=?2hjnWBGFgbI zwktBZbxLe7+#Pwlfow-U8$2P$miV!9xZ&{G2KyVr&xLi<2T-+25wb`}tYW-cseew4 zG%_%eziFB;>cT(T2Ec$;A)IpxQt-fmbUs8>Z%!gf-~DYEnZStsL70-pb-T+6Va52k zZEP^aq+t2g>bt#hHA6dk!6ICKUxj5FCsOuQt!Lf&Q@v?hi3C9`7%2O=zW)}3SSDPV z7dk=u^>GqRy5AiH^|gIir)WiI(Tb=b)>3}~?0FT`!XXQVL-=l=BC zYq1Nfrj@1f0_^{R2`1RUDdy2>Bt)*Fxd{0G4HtURM6vC$)L+C%7o`2eVJw+A@UC$$ z5QkGbjCdstFG52IySgSNfV1IOaZ<0};fJ{jEcpI7aJI1{rqUijfy>=n7gF&?nc7TN zt{gE23ari#q&0-m4bpH{FSiZNh*_FUFk*6Pvgj5OByiz-3&im)Tugr9ok&c*QbVcA z)q(U6V7NJa>@`Q>^rD}rI%kb6Tcq3%PI?@xt5d$MS+}Of{My3uc2hF~gvb-l)xCT8kaL4& z>4czQ#1~=kw?Dl$1<($38m9I1TJ3)rs^hELZJtLo^=e8$ZzK88Uuhmm7}`o6+T&FW zYwy{!BC=I8;`)_-p_G6*#^|nYPfzRoI^bq#66%Ho^{p?M&Gtx z8?XD!{19gxAbloD{^S0^#Ky%oFQZZmJ3kp{H8&%=$5kzy&0Q6e$J&^n{If}l zGbDG!*Cz9l*(Q_{=rq3b9|(>Bq2>IqC1>0lGVrhJ`EPBcuZK|Dl2F3ucvPNkju=2N zHf+D7ecUo0&e7tEKJrY8zv-zyB}Tw7;J)EFM4i*W5&4)b;>C^(8bu96S=}|C)A#T! zD=Zyg$gX&SQn%X@$wbust6=}e-qQMt!44i+y9`##O})p`_k58*L`ZoNTe(D?pH%}y z6y!P&hnwpR6_H^K-I(XFk|2c&TL70Pl3UiV-UZR>>|YF*`fdL#ciWEBFkCjf zqE@V~wlZUPT|-ev&JRBFsiV=GR)DO$jAGm5`pYl3XUeDktdI@9+2tw3P`-#*5CV5! zJAA;P169;%bbK;>4C$>_|e; zuVyOW3ir~OO)Boc;m3vxlzYbvSe+i0O4c0ygX#m2$Nc@lx1q$gr9(Inda%iMa`||8 z`lw*&CO)he@xO@f#p`cM)3uD<@~=`$nZrija6=5YJQ>=mF>Pn=OK=;G5&;&(bT|K^ zxBzkniVaY)wRV=V>s{~LK<|z}&-1WV;MI$_oyaUq@|`;vZWcc0Gpa=7Z9n}*sGYr1 zRXFAI86q;Qnsd@sx6?0+$QVV=(f`rB7$P8mPGN`-eE;V-08phnh(v4R&%ynH{DPZq z@s=n8NcQiibG@92D**aKO(Xk881axJbay`Lhl8QCPi^Swj70cB;0X&pTXSe^nYppc z5Bbc3eZ0C)ssY{fte{=9zvL>Aj&Pju#x^MYXZ7CYcEuP9be85bo#3*4>LtOeW)e$} zs})oUF6{`}sH8wxzCoOdLZRz>jx)Qqfg&+6jsM;ZPIUMnTA$e@PLrSzJG{GPkqs;? zSP00@+{01L)}FI}+>KkNPR!bc^Xql?Fj#z1yY2BGFLF+gH-=0g;(&GD5tyt?&t()G7)H!3W;>Em}I@C=nqMBg%B_6y< zSUWwUMcW(kblIEgSw<%AFW_D$KtLgU6Y@aF&dS7Krf_tds$)ZM`Dz7W;?ItKsHfh{ z%?lso8>A(w@?~3h|FkWWr361H0PHvJ=Uc(a-;N+8wnu=Apy&_({DtD(*pe;R&(Y^R za%j_UK?ZXU77m&O8b|#JY&s|?W?s2FUR*YG1$~WY*h(@)5;CsbnTTalvr~f#xhuXg zUETVchHcYwv;J~79F2Qnc6EG?ZF6>F`W-kR$KI9*4wA^SCat`V;W*FNK5iWevu1># z)}KD&jzTMuL1&uFBdymOsGF2{=m(PD?{|}+OYE4z+%!rLt;yN-jI`6;v^qBc+ZJ`i zmE~=mtqp?qrGQ?jY5angM{QgOU%+ihDA+}HHpP%BQO7$i^2R0PF8y%C(a{RiM;~8s zh~lnl{S75|?n*5esCLh$2MJey(MRq>O6iY?WR3)PqO@o}x`SlD83H%DJK5(?5P)69 zBtF{!izQ6xpZd|$K!6Zhlb}r|+ta@_!KBnPJc9PJ!!=`!I!4|s*_zcL1}YA7Cp#tT z?MWZrE7G>ZlAk#swU|E(sXPfRhDA0iAl87O zAYBXL!M!s!LZ)a*t@xs(&`M<($OHE4VI~3sZyrCeV{bHLjx<3)M;Z_UhY?DHgX1pb zw}nwSfgOFVG|ytKK&acZ`^j2ED(v8vY1|k+HYjP6Tfkv0!p@vU)xR^6_RG??`)qP&XU99y&u_=ikN*b>94JWc_V)DjjF8wQ>wLh*4dTl; zk=7L@2Wd{}O4czMs&fp6GwO2qI39E+9`>%~!VS*_i-o=O(_=W$K7x?-mo9&9AN#8A zinH&bj=> zX?|(PR4*PA=FG6L=R~`zV z1BPJ214XZBj~`z=DWXx+|NmYQ%_$wWz--_rlr~-w`Tj8!DQr^EIn}V`cB1Umw(&Zq zxKn&)fB__(dfiiwpcNq{0&@vkxkn$vS!ZOKXPBTAlyykyG`ynrC0p6uI^6kd)JX9* za&ZBjpnWQ>Xc7U5c@AK^$UB!NU#p8Xw^wdSZQ^er|A!_~Ac}4Aq_kI89z(90!1dSZ zW~@%H(fDUdnE(R*)Mz?6YtUmVeftiXMdD8Kh7)7Lgi~x~dx=6OluytjbVd>mu1=ZZ zG%(1~MwgDiZRhOTpOG?J)1!CIsprBy*MCW1CpJbr{uxN~#l67D#@4?XJ_iPjOyHZ_ zGzS(A&~pVN$OH&820V{rh7_Dj-LBd(t0i!z@=Dz|)0CrsW64;pUBNA>`JKp7o@^oV zyKW9Pj1>Vt%<_d54ts;Hdq5m9>>oDW`okpn4W-y*DoX-`*f_CYBaC%#CE>F(vo`=Z z{dG+nd<`5}ew;FQ79`v`q)>@zG|Gq_2^N6dV}eXznM_zqSTYxa2ypAQ2nr=8s*%;b z!l=1V{7$Fi@^V!m@|B<}|6}hK0@rEc)J$y2v}9puMW_GNR~1Hb3#1}V^bNs`)l)Cj z!2vPh$G_4EfB>2W^ptC25l9vz0MwRA3xWUc(A@Bi(QwTHRh}_{S}c8{+nuQ4F2k?3 zH?Ws&Zz*gVKV`RmKots);=a&0*xI?B{@1RP6j?h%*(ww&LHJAOgL5TW*j%%*us%0ZlV7*qFb zzk);HsD|Fm2@g)cEkVZ4AWy+%jG^Tns zfF&Y-T^&=EI=7?&{~OK$FKB^6sm9U7#0TRiwAseXfiOJ8<$7=;$J}70m_j$Q2wala zOZdfzW7Wo52eb#(8?TE5OU2?|o0=GT?kdWNQ=na#E=W6JZEB55S$Kz3yg2UCc8xgQ zh!D`XBLemAJ-F|&?l!MasV&0|V}HpUJ|Pr4FMFqU2}JJ6@$Jzf3SZ>({1ja9H*om_ zRcYY2Z)f1R`^@_d-qlx{2`fA(_*dZhfYVSwD6NHhJWG!xtGJ6UI zROK@2x}?!3g%_|D&)y=pyvSdV@b|EA_`Gw+cWqv@e<&RN!A+jv!8JU<1%n&4k}h*s zou#T8u6u+Y8Zt3`7^m&&u|$R9prz(B60I0u@s(fgBfrH!hvLz?hS(6s4`ZEzEbKj^ zOTxit2t@Bnipd-Ad*a65H9+ho%rrBI2{Ji&&rF@=EPm`f*tcy zruOfZgQO?wS$Vv*>i=9fMzDU+bHqBiG?JN(X{O2|05)GaSqOvp-}6cZoR&=e<`v(a zh4aCA;*~0#o1M5Imj#u+cQWF8-QzUYtGt>gXP&-0U;%u+!|E)U{fD7fANh)wpuhxc z8MnaMg@(MmTiW81DBiVO))RLCu^#+eMWF%;DIFi54buG(K@D0?JUPp-Z!B)2jpSzh zii|-P0}c0Y{wpGtU=P)6&snX>aInd=0}CC(VpBWrFCPOxbWlPXc_y6j-~ba8DAQnE zDaTR$8^s*;V37{I-Dw0!;qM-T9~-cyQ;p1Q7hWE+`ZFE3x3@2MM`C$#Nl9oCDXmKO zsq8}Tw2bc&(|Q4Te)pKG35bjVroTy!#+`C^wMscy&8lP>55B8=3o`1|64a2TqRk2I z^vYqF_Mm{LU{dZx1R@X%E=opo!4yt(TxTR43lP_FAz^Ksdu`IFnd8nHa-^-BB97Ei zc#7P#%Zh&QY~m*E){eKFcjz(8>ZJVpecD9q1EA2< zXFUiMLdN2l{3e%m8C6riaTu&I$b}TcGr41s{7cF>w{;wY9>l$3qgAT!`cjf3=4a`IsOzajA zxDiG}LLQWPVgV|&JqxS%p{(mo6BK*TNt|6ficetHd3@~S^OLi)i!&b0ts0jp9W{Ko zVEKp;JEG#1NFKCfFA5xpmG|%su1$UMmTiIE-~Jr&N|Y2XZ|~iEw9m+4o8pO?;r5B@ z){()~Dg+V)-YQI`n~{rBvjPK{3=Z?N`AT6cQwXqu@cIZ65%#A?v$OaSFP+y&sezV% zwA;|hv1UTeoh)ZUsk$Lnoo+5c`cd@~?YQOen#CiVc3FV@!NnMhIB~#&g7&fqoyo-h zYLSUXs9(Yi{JpU`Pr`&fMCVy*FmtEf%% zEbGr{71OJ2%o&jN*p8&RdX_7Fu+0a9b5TE(Xezqp#eXAi{dd2Cna0rVRF-5BjQ{bH z$>`rydDZVk6l(l%-Y8Fe(vN#vWbF*Pj{{AFP)pD_+T*_T4TcSF$F_IwiL6^XBYm)x zR)6dlI*PWXulwFK_vwcpXTLZT!Ck`RejSfqi14I1StuJFB)c||K%N}|-dZaT!gss0 z#MhaSu7x8Q;v}~qFqUK>$cvwfkdJgoW*oE;qa zIinXtF{jMI0bcrRvaX9#WFzoUQG;q+Gqpo$)`CS$oB7s;$=T5>DRtaW09?@-Q9x>@ z$Jh9HNDqFIJU^J5l1?9Dt5whb)JNW$N@N|OKi)V@l7m-Q;M*?pe@Pd>ejGvkULhO< zj^$dxz=wN}EYS&fG@;k$_oJ(gW9li{yZs1F%4^9dYlm}lrnR?o7!jX#)>H zAeA#NU>)-QgGaLMZRK(vYLWtEomI=+E?YXyT9KHcP-kvclboz6HNDwD-y+I#CSpME zjb2!Vq&Ba)PO4U;QmIH((P3)T~6 zldqBQK*jJ+>HPuQg)pLMR1IUoMEQ_-3MH(9X61v=TEMzLXns)t81bn_;0El$EWJiq z`xmv)66|RZ_bP@=mB-WOst4mRS3&vlWHlYmNCv|V_d1H*`1RiM^jJu#Q~gQ`O{DMy zJ;QDT?w}>J?zW9Q)tk^(7xB2w!lhU9i*&3R&*5tnw+dz@({+KZO3qe2rzi40xk}4u zZN6(HG3%oC)gL*Uph%^Y{Xd!v(Qg0(uM5dG8cpDOVg{P3jr5;ppa}b$Eau8HYTe~p z!L~)-O4;)IaI$@QYoHPl<$=E{>&9z~WV($egP#lIMtCWhR};pPX}X{vO--pk!U+1# zV0x>02`m=!C1c(B)o4?8?E6863BX85PYl7#%BrHIwEk;JyZils&M=m1R6(Fa%3Pv$ z7$glnNT9lwx<@2DGm}ra3Nia0ROB%aYbrg?6dy2Wu`?_Z)kpp|F3P-z2{a zHqz06pr7mZ%WQ^^#J~KOi(Z{==$7mTfbmIMfW81MAednO{q;zpPk3)Y())QTA@YT#@yKv z_Yo^8reU166%~{tNuNzH?)xNUi=pONXRc0hzoV9zh0K<8g;*HUCJV;nC%Ig?zP1ej zQ7(EQEc4U?C?IhQW4ZGS{8IUtLb74tz)~9J7Wx>0^mB9^O5Zh%o_!vgd@71xpn0r4 z`g?OhQCs9SYX$!RD+e&4fN@Jx`oOc%vkOZreOqBOPxyV>GZX#-o?1gY)|QGYYf2z> z+b0rnbe+}`!D;KW((UtQ%J-(e4%~uh$F*IA0q}4(lk6t74B5pONA&;GO!eFoPOlD6 z!fr>g3LeSo3MJ3uG%nY9zBDOI=Gva~Z^s+XJwrXzbaxEDY`}XgCQ~7f6v{9!^_YPo zStl5!tYhiCBF8WW!>Xhbdp+1p@$v3!sARFcQbzYB4MObBa6=K<&-s#4g$QOseGxY&el>&}{AOMrmO_=llCCjbvIF zN5*-QUjfGDpx{uyf}9=1Z;}Pwl0!6kNC9W!Bl_`hxCdeja>jYv!MqXoj0(o~@QM!s z@i<{z9f^3v6(jj^^3h5A-&ux9zZkK+Sn2Phx-P39ojQ2AF<^tyzyw_w)1)LiM}tv~ z^Vb}k`JT%B{gERLi>={uz{R`^D%Qc}#ulSa^S9Clu#3RpoH3-N0j8y}!| zT?ZKsM|YyT)4cxmWOle9@T>gpLU_(hy|kYoT1ze$&t{TvFur{hu)TL(2MM)KJZMztqSUm@ZlcTVee6^zgcKb`ug@7{|k^l&}xE$-o zmQ2%Ia?vZS<+cbonjH}K!2{7#4SU7vxt=QC%z+ziOqI0q{nk{RtyaSr&D5T@E-eAI zAW?xYZkLvP;9$UgEmqO=zUCR(*MFV?fsn~DnqsqRoV0Wm!H_dwNBuFfd^WhS_d1>M zvuHH569enh#OG=2{+DNdGm{ktWxf}{9T0clu@0y-p=HcO#<5`;Y!#_BgECXjx|5vA zbZpJpF6Kst48-La>*hx2vk*l zM2M$Tqm}BuTcx5#Y@`}e$=uZ?n_XVPZG|55Ixm4^Az?ztV zILD1egxTEx^nWI>e^o4tT2T7`;}WzGAyBG_H1SBsyQSA7!hi+)4@rt9EC|rd2nl2H zrCg$YrBPN?{{CB~98X7RLL9L=ffT`TW-msLMTT4LCdYUcJ}4aa5(-XDxblaNZEgez z=rs{WCZCESAYU-xx<9@v6Z~j}hC1AkQ*s#Y$n$`PzM7(Y*0cGfX>DC{^Gesc#O~O1 zsBT|iZe2{LvsIe1k#6Vy1|JcO$oN~V@E1(HjUgOn{kIhG9~^-8qa_2o5cT(gv2pAY zN)Uf7LR*lzL^G+V8)-Zb@XvpkTRS&I{vSgxI^)yGTUCJmc)&zF*ZS>_Cx5kEQQ(tcWY$vqKFWrm0K# zCK*m3AG48Nd%v+DlP+?^A@uW}cX^OmxGA+i!3K0@{v$MzI& zhR*-!ZH4>onv_7pN1*wc>MAKJhs(q72|%`aL5s*Xib}@bU!d*Be&&#|k3UKp zaJ#wMeTrUmlaYjI8Mkt*EOCv#$UHP(yjD3`F;oeskCnNso+!^#@^ zXWG>7%H{tT<}m>W3Q*#6b%E;#{g0#M01a=ByZ91F@(y=JhaE_hcGXbh7w1?H*&`9p z@x^7v$d3Hkdk%~wuYh2QK77Ywcf?3o7>FWE`v7-1SC)>_ov!r+RN9t#{Z0hvI1iD4 z_u_35j9Y6;a@;9YUx9}MQsU<(B9~G6t|f9GnEsaqz`(RiYJ?gGq9plZki!HZWiWYg zPsSrl3n&4^qfD=tUduOi-tk0t;5HIm0ux~a?|m>VYo4o-?@%ZQUb)x1a-US;o{_0w z&`^BOe(>J!)qJ1L+0WTNcQ@T#4~v?MwrVa5Ni;6qUzgplcQKV(tw@rIJZYQ88kqwY zCIzF6f8`3s#gJU1%h-Xq;9RiP83l}_A_I}<$P*ZbUK`J_7|n7kfI^OAvlJ@9DJ*d% z!j@7og3V@&Rr#wnMq0G(XV1SRA1xj5uSyds6ws+ijkp@MpSd00aT%G=eq?I_Zw+lS zwEa+m$95qvqq?G3K0nWz>!+f|cXwm&;cXqg?qsH3kwltZMcb6ZFswwn>Ix-q+8te) zcN+@?S9Qwd%n2|cq)~xg#y)PpS{L4eT-uW966+vZj2T_GFz@ENBY7Nf#z-|{M)ql-K&N`;Prr�V{ps_GIJ-`h%^72apdN)E%rzOL2DV1%vCK43CKl17Sc=Xr9JpA$3sftjji7EPgkX&P-1 zLd_JemUolnDTwydf5UH~Oc3Y5XmBfG6&``2xiDal#UbsC94Tzd`w?KjUzyt8%{Z(Q zzN^RbOLSTjg6Bbj3)#BMdh*-`!C!8%-z!0D=gES_X6=0MlVXRO5~m_HB~{Uz(f7ZL zq#u}9)^rM!e!>KWfznxzQq!#~6d0wGXw~m2M6%_+w)VL08W8>l6eJoIsY0kjPMjS*-sm{| zC=!8wvE{S9U#d79gJ*4H4(rNoEM=J)VHvh$NY*$BZQ6boZFy0NHe}F=R`^Yl&5mohOEV8CL-!5B)6&d0LDkJ|=XbfPSpJNx&nqC3+QZhv<88V( zOcv*xcyq4jE^;=17Nr)j$K$!c{u}|bt2}0gr-5N~S2GNpnAP8I?+)Y}$@a<5Wj6Bl zGt)AHM?=NJI)cyxq?I;q=U&g-0=zl3aF!dYd(Z5Lr>8+8hTBQ&Xl`l{OxfPMhbEfn zG^Cd=AL^U6eho4IYGfgViDFx)O6mkh7(J-B0@;oIVCHk7kcLSTM~eH4@rCZL-uhZe zXPvgKnEr~R=+o>s?Op!-Xe}3rXaNC}%p+Ng5IXkoy8>$KzU@k>-`V){dEy}ns6+N@ z362X))+#>0)yPbn_LFGg;eTZ`s6jobNWDzCM-!+I@cN7JXy=pY-Oq#Swng-EfOZyx z4S3Y(_T?zM`}6fI2A4}k&R0B%Lv5^@JojLKnQinXJKl_`wqpG~L`W=r8k3$BLFW{! zjc=U1f=V^4A3uh?BjNtSKsjd(ogD5tq5huj9>Jn$;DGEsaxidYhbYKnq-`qUUHMu8 z{z7lk(?4&1sr&PR!?23p=@n62qL~ovqfEJcd#~%Psy#dc@zA-{cUh;|hQ=&QjXJzj zyx%cV9ov<$JQoa|H9siPnq>Bj8xfvvWFY?+bGu<@9l04Y>PTMzGq((I@yMWBg3=%J zk3R{^E$j0K<`E*gB3O9p_KK^Sty1gcRX4BA!GaIr?tyAv-xG`sQbj$WXD{WF=Yo#0 z*133iWq;67Q%N$$O()nFuW4&3I0V>u9<^^-a(}(mm6e?~cW1xy&-pGijUk71MJLtM z@wn)J-3iZ49C9LjXRfw;{sa01kWq^F2#pv>=olaogWJN)Z)zAjEH})5+rqA_EENIH zit)bBe_D5xJ@gAF#`*jAv-Q5iejgw1_R#`sK?7y}r$5m{Bg#1`sOo3FPhB*nvJ=-V z+m9t*1A&xXC5JyNM(i?->#f0v#n)zdd;#yhQNv%opoTHB)K}^OWH#w~RUbtsL?i!H z8KJV)B~wL_90^GEP*w zMyt$-(2&KMS)_@tKe|=duBxDk*L262HvpH#pY6T(v>UtkOkUX>PAc+ZnVb7({BO+-1LH5^egbXPM5$tG`JgvftBDeXe>iu{+cMevJ`juyjnGy2Yzaa#3$V6=xm z5^NjWQ6Ac#wHxWt1l9zyDTbHG1eXN%x#er#j_+t{B8thSmCCFa=|mT!z|siaDS%i8 zC3ViIg1P5QiL45#{sLBa*DcYTwgb1qAd{sX;&YpjJwY1~$O3@XrDv_n%2epI!>>`e zm;(h0VL@fD%*y!@x&0%oVDm|p{DC36tJR7Ywklni{g!&kC&v`ZO5lWVuFY~%sl!P; z1e2({rKJZ6iR)#GCegWY@F5PWN8ztj=Rqc>N!xNt_Q%$qHMdx5kiaE3< z<=88*hkQPc#dP1BeYm_10Xahc@qqL?YHCpwhzzz({$)g3g$$dOTuQxBK(z2e9cuG> z>wR#uVsSbtvZx}ilpgLC0uG7iWl&Fg7*MG-z1#8*jA%gIs#qm!#PHX(;Am6kn*Q~a z&*$Ad+zl5xXs|?c;S{$chxYuxin+NcezELdMG1U_Ch>UF809chROpm=9Wti0KQ|T~ ziz9^%2s}+|@)XG!BD`*Ho|EyS*?z&h?ZYmPcI`OMoGq33G!DOXWm+uw-_&xICT+mg zl}ff+*Bh?YLDq4p={!cDCuDcgx##^6CYYgr*G3WG76s`_tB6qTv8|+0Zz@eJ#TTmG~#u|yeS$%IVB#+1anG_w$m)2%gS|Lf<0n4Ppp1BV_0yj)C zH;*C~dR75TF`ZcM%y|W~-`QF70}-Z!WM(cH4+gz*&GD`6$+N-99QL^~D=X{G=WcDT zSk}+kzFkYE_$+Ek_iF~;`+3jf(rvrB8w?q30y<#0wj~67)R7xvz1GKyKNXAvU6!E1 z=t)zO&4$e_7*`2cA=yTHnHl!a`Nfd-tn3Ndw7$7mNJ{$Sw$M~m5i1!`zk7fdR&_vl73t{jeFf}S%AZrUU=ev8$ef&oqc~&r$clufu3W@_M(t5d zm&v?o#`eoH5hXR$-fwX0-EjBrY4A+blo$Nx<4u&864Nq#w&Z7jz|D|N(XKi#Fhs0c zaa&~PFHyQAYa<$Lv~h3SQ(OwmP6{4~fVW@H6{8Kxt3?gwGyj^#QxJepsuxeH*O4u( z9VOb@+1E5TZ5fZc*zFrATaPrf35UWP64f8W0FkwLEB4 z+rTKSfNSrEU)W#k%87(p`|22S*|>+Ti{j6rSvPSUp@U;({=6!n=68&Tp?OEjJD!gmDYcCT(sHL$kf`}dY!u1<|(gU17b&1%qPPBXm>J=q!yZ|k@#98)DSHKIw;{4;D=7?M^%Dr)`)B(=mN ziu=Im1+(bpVl?I0b-^h{{#zO*ctCUN>W#;fs=Y2O9a+9jL5!5{l8kRk-7dv0eP#(~ zk>BvjEZ7yFhnA%}Cn6w`7la{5e`EUTK08_H9q|jX>}XS9b+kr`Y~D7yj@?e2=1r(9 zgeZNxm#e>pbAW$)orQW`YT->##hGhu4h0h%+7=*7%9vCKy z&~yHQKu2A;^mN>8v$oLrrfd|+P~$WhZ@KAtoU1#G6qc;(s^32pg3hpto{l~ zzg9l|7GnY7`0{R@1^c^>@ocNwBk#N%pjWv%5N4DhSk5RUSexD<+tVs;ERh>isAngk zp-9v*&#{*d3hv-HHur#nYb8scCVQ%$=q(_LYZ7^JWFHK;R`KClL`CY2S;gA zN&&B$exw7U!N1H&iJPnEc~=TG_O(yEn|P2!i@t z4Gub1KAw-n#SP0aE}-mmd|VO3GFKrHiJr&az}#$@x!lqM{bRK9vkWWnt>+CmieDMJmih6s=lw3*2n~DH)KDht%Z|Tx zd$m;N;e&g(aWaEODtZVsG;|!Ht(MSAi#@Sx=2OQJPY#DzRtoPLitR1&4Wh^2M}jGu z!!~xxd7Z&ubJVhxO>R|YQX9@)NJ3&tNHhI>278!_hg=PAu{ezOBi>gqyKIU0|JZuR z=sdfw3$(GV#%^reHrk-Etp;~&+fEv%v2EM7oyOL=+dl7i#u?-M%&+9i-g{xLx#koI zf4^Dfk zOmTl0gu^PvYZ{CW?sLL^j!sK4b5n~s|Bz~U;A11kx=zGw|BP#KzEcM8xQ zLuhXvv!$>W=55%rz6>v=&}~q(DDnGlfNf)^kp09E<*WgAlrG)EV$N}A7q z>z#8i1LduZeX@MLbz#A2T-JSPd1}ZA1)N_4v2;()kC_~SdJ96%P!xJ$e$WKLw5`Sf z#UOa(0HJj_1(z&-;=^Pg3mim(t&{K=BXsR=(#i;n!0(k6)!*|OlNZUh-YWgM>Y}Cn zO^5qFRn0C^U)d_JhDVK&*Wl<+cS$R@<&o?J)Y2%!i>sAehpxKKA@=gP-);ymB0m7S@@MiE;ELw%Zy6||2(Gu$cv?=? z0Y8iJs&QPO`|f>YPh{1LR?1R<^tjJ{z#$lrd3l1|)n^g2-o0+uQN!~|7HDtqP$^#b zPdNu1bXQn!+i7n_HzrJgbOJPPMtHotcgS53xSCCc6`Et^xkH6jWkr|6bRdw&3Tg=Y zg)+Aqj<>W5%Elxk&QL%RL2W(%F>|lwWq4;bG(m>L5SPRP*ucPmUK)J#>%R7vR(6D; zlV-o}QEcxLBQSYYV`3)5=*CL$j{KXi&n~`Rr z(y1P?V_*0l*?1`w?QzI){LF^kZN&&%NhmoE(%53e&&{hz9Tr1fHT1hlPfe3%egH(h zXib^1CA;dnV^M1^>_YcJ&I4jZ_x0CG^*VjrpOO|UD$xSI9%9w+Uaen_&(G)CB-hY- zS$D6RO1NVzgEf?#r6qWOD8)j6kSO*#|{w89-lPK9H2DQ zCcS-_lhs0Z6KdpWKerY(wNl;cGWMGW+RZ{;^?N>z>Jr#atrlB{W&UeU2atSb=sTay1DSe zcp2A65Nb0V_zjN(f?cOFZeofaebeokXQjF~8_18HBd=i@>99lQvOQj2(SI_C`=4vr zRt-r^8ef5Iig}B>2^2lAG)s8n(I2a?KYZXB6d%gv1%~*Y^}QuE$q$gML<$URmIKGG zXye17tx$B#Ut%xEk%iWI_{DY2vL=EAFthzktL=^zXLXcr!#Yz02gzTD%41nt!DzH> zcGQKxdBoAWV#N+RqnUCD;bB0J5r8WzTTSJ(>?eJ9fL|CkYTG;%aY&}f29RKiB*0jb z?s1uvH!!3SCo_3qMxgr?=v|l2$@Xush|IdLWCYPmF~?m9mchFMpvB%Rq|jLtLQs!( zaDz_lhI8y{8ds-k)_v?)3=tcZwalAeX*Z(#2j-4R+SfG|B)-nij5M~b*H?kUn&)jM z+x05X*I2ivM|DoyZ)<65uZJ_VDe5TwzI*w&+>7sg?VpqxrQv_SivEG$0LXA_V@0kT z5(|=u6MUKDyR$2+6`|7&qLk8$*?{}V?;r1w7v*@OT8G5u$)HB)b|O>$nX?fytOHb( z(#9$EJzjA-Ww+hntvSHxplC&3#XyYu+yR8N8P$+Er>)4*Sa3SPjI>b8X@3$>WGQ2h zE;~EpJX>KZ)y(5~wI~!uD&0pT^||d-w7PKC4y7Ro{74hH;IsDn{k{}_2+tcZ$VOs> zW%avp!a0RT2E*lGSX4-m8i5#>m@QJvFssn0kLXywKFt%zqeW#EBmEO*$dDx!y8NZK z$u!f>i(vgs8@xq&`2P)hCQ*E{tFBFyo1*aEsm4)!PMNCI(pv_{quD|$Qc%}$3|vO7Xd6~l5K(o6mc7x!(C@km0eNy` zaNx*4FW(U z7-J;n^jFKhxP-Z}xHHB4PkpYi(NcD8d~@qiik(T`o3W@& z9?Hp5HZGHJC92dvJZBac#agG|I&7-$3cwaMh_HVn}b3^aTBo z6b!g?Mx}!xQyX!Jl!~aX66y^SB_uLoSisulH<2J&_d4W>aiO?5sw^V~dT=gkyZEkW zs8`89hxObh2B~3&7}Whah%>=YBbufwa=c{7cKn7A)5649bgr_9EO^6s;ueO#Ia|jyvkdZb zebtPmATCl;CUn7y$F3I1r9-K$Ue(m3HT@8#m<{Jt`}^15F~Kt^XC^yA2IAw%pd-QHto{E?DT(2&$`tDNyGR;0p;&(@ch);(yeQQRK$ zuFuT((-uf)yOUl5m{2C=Q3J@FiogJ{>wsFHG55-*@l3okCZ8M8mX8rquhf+<@J%Kz7rHYQ`-RJN zNMHjwXb9*wo9GljaPB^x4Et8?o>J&A)PT-fvUYbeG}wTYvyP0grQCTbqIY zaSwoXg)sUsxNUA*8bv*>OmsEv^!@U4(g`=FDl#`O2hx~fp}v!g9;;qgWnoRS?-sVr zUDYHr>5K8bMEs2@wXxx$>9@gT#7lsLtvWWm&qy1}~SLSqkN*7%zz+Np)Shr}0fcengk%Zq?A$_2ln$ zFV=axaw;vF3pdZ}aK`G7rMmc|mfF_FL{V;LUr`}$6f;T^L-6hOQ~gL+aRRr6i)3%T-HOdzQ65XcBDf2= zC1Jx9vg`*C8Cpn~sBBN)p)Q+irp1-cfY;nN$)41@=E<^rQO6)Ns-~MxYQFI$HS5DC z>&7=)Ka`vnM`>dDqL`Ad^jQ5rX$aS)C}gwlR87z~hL8ViAX zvU_PdN5KZi2fQyZlVZxaniq?1-u-q7?8y;Qsq8`^?stapIyM*r^`!y#4RXHm$cDhK zTT|(gCBV+CiV23)3MeWWRTMO-yC;!N7C2;4#Wr-YNbPyhmG(h@TB@(``{Z4Bs2-yP z86XXR6IBPQQi3Hp<$;SObu&l2n@te4=%OwNz{AYH3zv)7_kcR8$Z?nOphlGTO1zgx zuSPlFZz0vMnH@Lr_5)vQ<Par>9IMLF8G;-obO1kj^}g5C)xP4yn%iPgnF$Sx2NL z)8HMP(H5NX`XzK8g^!8xVwg3$*WkUM_kp9E`B_$sBuWsNJ|nof#&fE8r?#U}_6@#V z#g6xUZi=T7T8-Gp5eqJn#)FfbmlzJ~B&(oVA>>AttTv@5sWO`;*^A7krfQ&t-oj|B zsyce+TO{OhQ{~t)Gms159bR&f<`bDmyV^l4Ebxo{irTIhyrI8cO>kTnk)O}mtwU4+ z0gjaXBSSQ2NEzM)tPgR??Dx>NOM$FQA@Q4J-2A2?@nK}Qh@Kn6ockzEK@aMhMMxlJ z8m+aSk@EIRsc_|c_1oHVJcT@-*IQ&a*?0d3J*8DiW-~@fERfO=+x4KTbb#- z&$WzIxNQLbDbfm0+Q!5M7bGj9Ts2WTK5ZF(Ygh1bA{^FzTJ#>@3u%9wXEb@!Za4DA zJ&JV?7w>8tXrHTDV>BGHj4~|^5|3&sgXiEB;oynmF!q}bkP$Uwwd+K&t6@TQK7RB%oh)J7pDlfdX7di7(k!9#!H#;oLxLKcc zA1A&q2~7=rb&V9d*LJCgcJ+zgR@gxLw1f;ZSQ_(KT^Z@5fag|c=i>cs?>99#F@IEB zx;$fNWQFBtx|&{;=)r9F-^&wG3ufNBC0;T=;Nm|0BHxFHPs+b_an5qP7UD~d3CBX| zXPOaj>MKrSB*n+W6m~{lwx;PeKEsDiNVrvcmKi;&g)9qjn|QhKBAk1w$zZ!clA~*7 zx!6VpLkHHu2gsl;!^j$OQxsykvQ}*&TS`x-ZUZy8hUZG@lSAs{MPOSzZaZZh=9M)= z$M6oDI&ZBE6x`3%D*-7BUC~i4V4?44de=n^WIWF`V0Nsm8nu;8F9AwED$V`Qu8#Q*3I(BR@E zReEsC?_f21vwhzYqNxS4j4{FMS@(Ldv(k1A@>1+vVeW)W+|K;O?B=XbYa>_ivxk;j z?4Aq@HoD>GHAFEzBhH?=sN7j{diiDy)Bm(%6#(!q30j@Hqr&gD_kgt~5+U(AlzYnB zG&1J1FuzZv!cIw=`RYm4by!Tuv;-XPJXA#rc5lPZI9W=SCwNLG;OHpO*z@G}6_m-V z!-Au{M7_sBRo_qOF(dEUGx?n_r(Y1?_gd+^0aIp{2aQnjLeT~UNMCauflsw(8&oO| z5sw$}dK4;2X7G6}bJlU2S3Xb__pmcX<6$jt{f=TEB3zJe6@FqUM~P)dy6cJ&vv3A8 zT+7H>d;xFC|1wj;$hm80*+q}g%evWOG<(esI7W#$3<3MwUI7!EDaz5A;oUhVQxoVC zJXU46U)_*0_lt(=VNPW{ZHzyh0~crC-of0=hRVTWw!EWQ3iXW>(y>6uCt=)=E3NA* z5jWOY6-)U<^e3{eQ7~5ur72rgLv1BE&^nE6&d{fCz#{6;gX%UQO8<;z?6`UuTxp6XNnM#Ey2CXXRh& zG@5YN;e&3y)9L)GOW`9#WNhA;M~1o`7JO?3B`HJzJ^2*gZXh-c=}e>~!6Pe~6eJ7F zu!~y&h<-iYqK6DufOxA^1h?pv+}v(_IE^u3li6=3FfO$7aMn!j_klblyw~GKT*77r z%Zg;|8GON4>%P%ueXejBnZ>eT(C@Yi7qydoSjQ(}4|n=}D}e*z^SfW+Fo^_z?O*!- zVBnYI1K}d%pt|JmJqLptPyunOXE6XtsISoqDo2^uBWNm|Lz28vL4B%+;lfU09zA#R zepC=Zg@>shR_Z(ny^#GFOyx((1cdL%{_9bQ!RZ}qYTJ|8zkCSf4Af@tITI;5FV%Vu zVTWu<8IiLD%}rc_C9hObP8Z@>e?C| zYAVZdqq|v`?>s06G{AeS>YlMCcc$ow!Y|-Ua4I#JAWNay*-_JcFVK3%HN|fWxy3mH zz=b;`vj1O{1qvx>5v++-{&(eoB!DQv}Q z1OwA99g;~?j?{@p8iNWfy|q>@3y|?TPMTJ*ZgRDhgiSf;H$T;lq!Wj^q&|7zHxprj zBR432(|KR`ZbMBO(8kvZAdTQNOtqZ?8PTsoxy0h%;}+kbc|kQzGVW`kLt6unB8rn3 zs2rwEHm>PIZq4&M@FiZhA@%=)U~{?0mjl{d_Kiw|teJN!S1(?u%*;FgX>3?TfiJ== zD+tOvi`P|j234^bgq#gf1S-^DtW{@I%8&r%S_dqa-B0F@-_2N2kGZ^mYZ1j19VFj2 zuw?>Xz32#3FllSV(MO}o6fgg0%CSiJ*9I663c9r zZ*Y_BqXf~S7yZW}UW7rj~6S-HwKS zWSp2Gx2I()h5*p_W6Z^uMjkjYEJEMZ7tw`mTiztVecc3n5_$eG#q8qhcqMNd=SBjW zBS(eH>$#1ZSQa=eIGJ1=)SP@ikpOYr8qgDlAy}Aq>nCG(s(luO*2*uFJVMaczfTIN zYZ4&%%tGoZG;hpoiqks2*PS*%FE;!C9y1|m7KG8PKzt9~-et1nhgVz(;$G5ccHOSI zsyO7xnBOSlh6lUi&WK~Vb3(xfA<~ind9p^K*f~@6(c&;ZegmFp>(0|5*c7rBEm3XZ zZ=PYat{kynk}nO>cl+ppRFg2D)pMOjul2jd&CBrYj|_V{7w05(Bvdq46IdcylOISv zkjV}Hv9Q#>W9LB)d=~e2OngymTw6?(lRrt(;%U<5%C=CU6yhED_1*XQh}>&4bR1qn zXUSiBm2HbK%o}%@Si$P+0A&h1SmzejSNPnDbxkFJrZP@-^A-KrV|9OG!rw9!6GSjtW!zf)f)Bva${yQ|b_TE>79q|DwqD7>(o@ z;YEq1{NEj@ME+Fyqi{XK;`NI0#&}E~4QJcj)`5OXP==Y{*8g(JK|ourSlR>AZ5^`GMhUL9)Th4B8R zWi;_WWfcneWWY!4AJOfCY)>SndL+vqAe zVhcu-TS#%draG`2LAKBP@;k8*$1Ek#AP@dhi$l99yj=eIM|4K#rcxW%N}@^P?3DL^ zvIG~YVx3+$&?M;TZc$eN=v-MLKnuC9P&W9>6>3-C^A`>3X>FBg{Srwa^Es1U4v8Cu z$=$K=cgQRAEH2!5Iy zQyH}gNsF}?a%QS64Lnl(p${VMg0ezdtw!bQn^Htu>mSPr8hF-v>8xxVvRMyZz6E~pwV+n{M zs?I0y$-A>L4M;_bR@_Yr)jN6iRGuWNi3G3TM(OB>{}$!0h=E%z1Hs9~bbq1rzwq~Ta7*qWxHyC0#ahZ1iwBMg#v_%#?lF+ROs2{LFg&awdo34>BpWb+!&p6#4FDozF zlTXu7oTJnqG{md(zkN4~*}YQKgnSR^niWpbPpJ;UKMw|H4I`tmbI9MM9uP&bl3jsm zo1H@dO@41$!Csl}`V|uiHV9@ev7AgHLHbQ09Xk>kOe9B=Nw4n(?aIvMd-%;Bimr|j zKuF5VNlsr!r?zB(?k0Nj1I57}QNx_`QQ6zQR#{!XeXAj8iH%5NIFmc!wk}<}d`i8S zdlOV@L!$hnVQdyNT+w87{nGb3um}DP7kZV7mUiu}@Dxa4!3~!QV`AWT`Mrh~&G-J8 z7yShlCZyKM(fM`D{iE7-G|8BaO(ggW|NopU@WelLJl?z|`Gz?2a+|b8U`O~Bz5LYv zB|&wBpdOyhOHn?ahdYu`S~4FggO1EH@k@OS?>?>|aOTD|Gz-t4JnVZJD=o%zEUoQM zS74jtkNSx04}DI}*77%m1)8ErPlV@T`UQeUvT$liK>&*1+$L?Rud*}Ib;==KTHp84 zfI)v(B>_VT&9V+~I8wRXe_2a!07a4lz`4%nJ*vukd%(bLVD!IiCWA7hE7jngp7972 zDOe@bl&I_L3b3`iubGP4GCu%XqV;@T*%7bgF0sU9Av9|OP7Ql`1o?^gNtf`?)V}Vh z3sMURe|yKU-`747v-&>VowN=n7cO)fzJ#Y<8MJCZM-R7G?L@xAYMudp`%1%O3CF5nQ zNa>BAS`h)FO|{+P6{miFg;ZvYXVW{hI6x7mt;`-O=A6tVSHjw7z%vOW|CWgwOATYq0^r~>ti6QmdfflW zu60N2{1Pts9B^~}xc!9()c)@jg%G*e($G^;s2vXi)BpxL={nd-94e#4Zk`z(ESipY zi<0VUIx&#QX^(Dxq{M-36DhOtR~-Rp!Ms?FNI`YqXJXzEK}{{9m&sEIW(RQ^X*vvAgyokq0aV) zUk>PhVS}@GjSxi@ks<^qNT+kRNU!?vbmc<37$^k;^4}5fVd0gm}kT z4U|mGIH5nF`~U!Oq}!K4t@c1zw;790pD&4aFjtS9EPYror&@D_?my_k^jR6{%2j^H zVN))Fj)L~)#9jns_8pr{S%-@O+_C_ja&v;Z<0fv;ZU9neo_p=E6X*sIYYgbx4gwk! za&;#JhNyMU==pA>A-Z)U%{ZZS)WfBRAx>f?$kxws_&$q5Y|)dgM#ph%K>pZToqc~8 z285W{w%rnl1;y7GgRy;D28Pbv37+cRaNS2Wc|4{2A=2oMhu^JOU%w%$qIv(95cKFt zW9BrpTM-Ltnf!$d%@aIfKv5cK`^&cyGpJ!T(j*oWO1_sSB77aGF<8!2`pGI|*h~N% zgzUE#YfC2YJW9=gDSn}zuC&EA|Jxv8u zw0|(2DOJ#LSH^xz3%QI48uwnDnM&jGis3wc!dE8!G~}QivOhVQ89HC$657v{Su*U$ z|5X5tnb3lttp|~Yt1seOF5(|un{-v=WjDw(;=`?@pj>-ZAXr#TArsldC8tb(6v?eC zV#l-AZPmN1@B?jLNYJ>VX`er7yus|~Fg6r%&euQwNinhzpajvTr}Ju?PZI+do)0|L zX7ENb6oyD>wSpw>$eBIRm~{tCsJZ#!GEXc)kBLSZO&&j4nncsdNa;$Kxjlds- zjlP=Xs|se8y~UVBvrB3!dkF^yY8xn0*dN`K*w#i?r02F9Dn!t_YK(b9E13^;o6ThwT=GUpT!_zdgX-n1X7encNj+O) z@G9A;UEq9M9Y4N67;SlqMSQBzAnz1^Z|M;^V*I3eWiXd&1Y}Q)&yzvpTm9GkkCAlQ zu4BnOyS0YD98Bs0RcX9CWSCI@!W4FaBwO%tPT9oUk$X$YQj?=()1&KYDaTt%QK6z3 z#1yfP^8=7PEX8%(%>ts?2!}f`%_%W@X`Hz7E^Z;F5E}31%AUBfeJr=IZ&(yTp4cc| z?yf>Y-3S~V1=ZD89Gx9l8qR{ctV|A-#t>xY!NZ{#gv9!(Sq&i(9B%83yKmA}F`X+v zuDrAcn$dxC69lYjqq(zaMmnJvcROYlHkOr9Lu0=uUQ+}LSz14e766mbZb=Xm9tw{JVuo zOA)c_bJF>_0r{;%h2MB!Yt$vo}F0jEuD;srYnXW^sit~Czz%f@E65sG94Ij$ZD!I#|Ym-@NjNCY?vO=j&vxNGpm@elwabdE|pv;Z3!7%HEdJ25O8vi3~k znoH-JlUXwL%-Zy&->R>F+Q&jfkV}MpjQtD1VkWSQp5%XnuTU$F%OqZyr0b=j!li}f zixG}ty00IN0LS9)?T>oIG2&0kFFRo{|F#3Kn}ruv1IQqDE|S@OKo1xu+Zn#0P%lFo zwvXdGH}P%^>=Kf{ToU4f?f{AGZN}wPQT{?i5*&2!CGPfkwG zt3y$)G0V!s!!x^u(NCy;N}UD6FDAsa0dcMWnDUt@V$P{ele&0G@Dc#}gn zw+Do;E{sa&ll91tE2(zzgNpx6yuJW&oWGOq0G#xBit*d)U=4&IM*06JF9)ok8>kIW zaCK%hlOt&E{(6ag4_BzlR&e#&g3{mI0aGRSM}-HH_d!)j1uju&<)3jBo^y1^<$e1Ixq%Dc~Kn)b>VWqFJxM39lk5KZ!%MwmBUVo zHR}elmCLvHRPb9QLu@}Jzaa$_BrA+p24b*ZCoH>`P4?=!1CSr@UL=y3^BjUV{5oLx z6JnN?l`czmV7J7*67g-{%NBNlJ-?SA@fCg*H8O0O zPW{U1jbGL@YN(p->f-!#Xa+9^1_t5F(npiU%l%GV;BkCLMMVYH$74^ayDpdW?dO)I z7Q>ec%L)*%d_z_a@OzuynZ@UyeGdgMhF!?cW$J6+(&2-5ZQjeXfgNC%T-i<{Nb8} zS*DZ#B51Q-?6B)dXwkE|ENlhfr58^w_2)$!BgF_7vqcD2Sj7~BP!3Ez43wbjhNx*=6#E9JhwCJ&xs8k_kNP(+0%*OI#TizQ$ zy*FCz0bXzj4FD@bn|w_r!^}Fqd2yw!wYj;uH3-P;`nrGAzFr5Z7MbfSxP?Q;U2L3b zw0UN^WR;=jH+!Wk3V|E*Tf_#)^0x!E&gb7Q9dOuO#T6I~v|@v8BCUFpyU~2ko3>sr z&!<2VP3!YXsjcVBs#=AXhNAircq3nK*OE4n1|OdN4&Q5v>i*Q991~Gg1zpzke_bBpYA5B_htc>uTk_ez!Ns!6$rG8LCO>bvuA zIl9#FQ_GX;Wt8!qcLb9nFLxK6F=r8)>WXpHSRe8b^ct&*Xu**CnmSPL> z{X&0J?f>7gGm{{feEHrXJz=xUs_+W#8;+^Ok?4$v^bo#6Z7n|#C%Ndmj1aFXoU<3? zhfQ%Ep7#a4+#XzJaUZqj%+TmKI24P(@K2Wa)d=|X7!_MB>t8#ETN^ll6winTT+O+dDqTI{a`aPoIhBpxrXamO>1_}yU28X3D zBE%Oy5{3%yA4fm%~?-?&a*lnvvt4K?j<$0TOZdH1VRK7 zWG|jOwyfD)a)B8|xlwW3T`M|-H@BsqIJo~bAav#&rJ?S^IL1`s6o?nSW^__w37 zCHQEwls^iV{hM5`(n*E)rn1jFfV;RA?8Is4TIDA-oh3Rc6jPakh~Z-F^Ew zCsG~165NN|n6~Nm5qO@MJPb(REgL>xeaov(ZWt&7P>U|V&yWvvf_@IV751-`QhCh8 zvt^GBf?`sL4^sSZ^LxM-IQkv{Nh$!zl zb068-2m2S1nXM>tRtMOqfV4Zp+i2;059K@JnjMA^C>WuBRih-Y_8wZ_PdfxZ1Dt!a z_!P-sx*s|>*XH0uSu$fLWXKTtDN=J(<^yFk{B}XT>7VF>1kF;@a+HEO_GtI+t{DEO z1wb1wh)QZVXyV>ng z1k4qclOphKjgy}M9D8`|P2`A>;SYnEK766I>IA~X@e!d?vfA2s8&@MgyS^V#%lP-n)Y&Nm70s4g{>jvlAB%RvY?7M->=PtM1GB){EHkn13 zj_f&uhhaniPscwp#a+-Z%wh^Nor}ew{uPACzPCe<*lA4&_k6Xuy%bg~tF8Y>N(O>m z5k4p6G#?r0RwTjik+-}OW>~&qvs}U+q7IBSp$YrWLx@Y6sKS`Sx#Kdtmcu2Tr2C6# zTn0D9usE0T4L97ms5YKj zJr0akb#OH*S`E4LO=YG%wG!!d4wtyeKPlI$#aXe8%^GU{aro1+=oq-&R$fsd9yl;C z@bMZ=c(&Sz#EWp>`?BNE{N_a{x>1r$zue|}b5d{FeVhNF)@lIZm&7?Jx*RxA$WkP4 z+h!r-22ssN4?aVYaf~)?&qhN14|xtHo+30gU6p&hrMU1|{7iK!5z%t>(f#)e!5k?IXju=UUs8>u*?o;wXlp>TE1z3`Tcu^7a;kEO8riT ze9v&*(CpwXAzwP+yex4}qazyqWj_p5vt95PZNlh{xs)V~YbaRvfO>?%4PX2RJ+nU6 z?etFtoha}E58qY6P;=2)4c1}gA*YskxBZ0oJ*iz;ECqPZCr%2A0yz}tob+;9(u(80 zrya9C-G70{=PxA@{Xbv_Sjt5st5*erdrL-n1Bs;G^*aO;qiZcV>&zQH+)} z$j}r*Xgw%^xCpBHnA+@L-=r}J@GU$BN?O3F97n1AAfLda(pxIauP#g*o{ar8eQd}S z{8f8$;!4DA#Ow%&G#fW>(davoyhw`QLQSxo1Y!jxvomZYqwDM1(`M+k5b2M@OB3h> zK6zo4+zsuj?AYw|n(X7YI zr`1)19q?6;t)7c&+riq@#;^C6^0-x_S9fn*r%~i#0=uvGNd1u!3zBC7$rXz&&8=g9 zl~JHV=?G9lXE}QGHTW@a)w z=wytX9&2qZ#zbh`6;eNL1z!c@K}x{dA4G?)=+`-A;0w8oG>`J1Y?LwMO2hX`NX(Ud zwRDK+phtv|e5Eq3eI3$RA#T4XT>_Sp?q8Ru&$UjZZvtrFyO@&Z9t9$X zk)1v}sG+u{ImsdXkkd~pqTNLa{&&ieg3mp_ZXv$jAZ5u!yuIS4e~$h8&3`Hjdg84S zQrvf6bZxOH&(I+Tv^=%nQ)TqYXN!^q#Pv4Hbs4lT-<6$sl!9ngJLmkC{xsJnG@t-o zB&-12rftaNQKv~HFTmMRHTqz0CV!H=BSI7m0T}5IZ$t{Q8ccpP!Ozh!k=S-Z8Q6Yk zA*HkaTS95s>TN}(gN*P;DaO_&BZ!C-vU*scmSlYxOfunvW_8JfxWp%tGHH+28nV6@ zArElNsc?srH=;aPa7V=$G-@NA?MP?xYHG<9H`4{u^KmD+I}ms{;GDkG!a_}P+Ewwr z>!wO*CB_WL^nh2nV3bf-hHq*%CU|1%cSiz2pb39J{7ZKbFCX(&u82zRl3@*2-} z;#kWAz#&OR_^5`?{qek0Ry)};H5?;#WR-5;5T1F#x|&zU)BOz3`^uH+e<&a zGz&K*hN4-Q5FB4ViGqfzU~E~jg>Oxh&%w)nOfV%gMR654L}yXG&%x8$D&JYh<3*I& zN`79eV|<1g^UDsI?GhGaxgm1opQxn93wk`dZenG>VJxJL%$p41JpJpx)7AO{KC&Tt z{VgU>mvqlGzf!~8ur4JokyzQ2MdJ+{wPYXd%iqeIE9BDu)aqUe`Mzcmpw%(1f~WNBn` zjwQ|)Ntt9i&#QhcCG2y@x`u9W4+z678tRBJL2k#_QS|?8?}e8{?#s;9HQ9`k0O~Hvt=u;ik zl4XT?eO%5QYWmNDgKqMBRGIJn`q(4+F#CaeKM_1N1jRVA0)kKZO-MgMMkr8;2f6F_ zs0ccClgx*tu%1!&^Hld_Lav)jG_}7HxBmEU>Ia#FG8mj3299Gc6vRiaf84%`50y6F z@(o-<4imS*YsL_ydFqk{MP#n3a~^12j!`39yoXKJU!jyt@0M&grYnn z9Sd}iZn6dS%F@!}>%}2>47tS2<*bvesm)WFWo1xy9=FpUg3ag7pf^An6>K5Gho3}7 zO8_yArlq1AA8+U53k(eOk58SscBovP-2{RX>(jHpYF1=|1I%=JCMuD?dU>4W;4XMk{@F|?2F@sS}w{-7_w zh<&3@y^AK$GbC|b>v#u+x21{`WSY%BlFeBIx@lX#ut3?D&=FEWI9{9XJP>U_fFHRu zT>S|4E|a;x7wHW$XvBM8Z^%dJ$bN?b3JWUnm_c-jcf#H*V>{f3>2s_|JGw3U?9#Mz zP3PBL+3?}_cSr@XoYC_k6SfYG;`toMTg}QWChliRB_Dd4B&;BE;8~21lS7)lkr=-T zKWP;e-~tLpSo^9UNJC0^Ozy~C*DyY!{?lD%@%tl|ijpEm@!yaUxEM|gt>;&BWKC8T zNBfTPt0N1&oU~@>(N&`Z2Kf5)PEZ8LMru}8*enBOUZK{T9KkMqOQ>Bk^nA-ipWWQkziU#LmO{BE_xORv zDF1BbZf$|b2cHx~M$OG6cwYxIoy?=3xQbl)Jp}^Md|Vx$*V9%v_Vnd70ChW~|(y^^e3hE8jJ= zT}!+#4|<1S5Y4qUs`srFn=?mMPiu~|Bu~}x++C@RjUbYl%tbYVmjSOcU%yHZu>^I8 zSU*poX_DD2G=8sSP4~pP8LeBANM?v>q*HQUC}t~w?Ypf~B@O?&<%?e2QJ6Fw!QYW_ zzw;HuD&9t(U6HnELX`aPRkK6!b@Vf}n_lhrzd5M{9+jGI(dUT&VqgTAZ(D)Upy{BJ zuIeAf>H}y{#IJWyuV53@l%>R*W2&6KpE1E`s1G2m&(DWPMJ|b^nR+ZUGJ3X;QG>!<(nq$pB{?1K16mFUJ9u}pnP}PGt)B3 zaVgJPqT&Ba^v|G}5XU`)xbZZSjI&eY_w%?im_yw%J3ql9{lY*hXYq|CCepn*%lO=9 zI2UZ3Pse-{VVE8EE`RHC+P>y&Fd{hT1OvgHw~7c9PDz`w{% zC18$f<$g)tP^+AooY?+UK#M-$g+@ML6+vg3mV5ojKRuU=scFI?f)Wa+p0d*u~A zC^d+}y2#{6hmO}nt^XcWj4;t#yPFq-1R8ksBgg-?B>sC;jFreSM$7vSKPX@o^LCv+Km&bitlqlw~oWBC^yxK_c_gm4p8@zZb$@!Z?4C=mx8jiTR zWtyTpFVS<(BfaRysHW@mH!l5}qLaJ9$s>%(FO{WA8y(Zl=AnL!U%aLkLcoY0G&7&- z2T|`bNih?QF5K(#aExszoT$``m{;_x75?HjvPhsQi}38U&{;y1eL??M2maR&=rxe~ zEV?A)hKG_@VSwjQsw0hkDZux(?~0(Fno(O_;m&<~b53XZUHo(ISaG`EJT-5RD#t z|9Z=i*Jx|^yZEb&FsfNl{P=;TX+-i#gvfc2eowyuugXLfz0mXAon~=oix&N*DXW=N zpyRpnzG^ZT_6T^3qpSS!I1R?$JGO~P81VlQMR!l!jMJzgzkAOnpQwN666Cs;q*k-+HZr`%z4Fi@K&-PK7oZ1GKc{)6EI? zTh3x+Eecv*1}_cRl(5Z7f9bMWP^E6?`&PHV1|t@*9RC3d(#EChzW!S-eX%vu%W4uP zouM)|MG+nS04O49vB=h11jF=h)~a6kD?EH$_n?!+?~vIWB-j1Fr1}pG1PbMdeabbW zFM{iogL?|RGC&G#@J)VEm7MHr*|jc4luCSgb_b#q2)&X81*z(m&Dn|kxUt4!N?*hD zFF7-1Sdx4fi~^}qYm@ay^{-RDgiPPL`-hD1Vo%bM%9M*&AQQTvcY;0yEtE@~QK1(8fs;*XsR^tBAxu3_ci#iCi*yLT zFLt6Pxu%yUwuXu|HF}GoWkC{5N1MYi3KUo&!cjsEx{9*(ev3kR=|=>9rSK<);f2~s zDF17=MtVw0`>T1C+=#$Ws$kq{Xq_+JKinUFzwJ@*_hTi=IZ1h^XGYx@t!#ugQb1>g z8d6OBS&aF={v23H1wwD#t#3O}m*gHxr3}Kyo0cg7mf*0$Qvv#_PRd^+2Y1TL%eQ1q zg0`d@v1-NOBYA=W>dIjm*qn$9!HTI;fcC~v!qA4s`o;!U+@VG^nUpV0^xxmJ;s7ME zA>vY;vY(?bw(G5T!c}-0awi1Cmpo%Xi_H4-Y+FYR?z-!4$N+YXTdMMnhT{tSCHe=z zktJizf1%AXW(&%oaqZ9jQptWDGX*XL4WEqTChNAkKUgG|5kf3g}P=+G!4yRL-qn0SC)+IpFPPs6zZn%Evw#6q?qiHdps5&f~md>gVt z@^?Vb!ZOQb%l>rval&l!s``wcXyK*6ht{NZ_x$z7-Sp*_<3$2a!lw>j_qGMaqhYFD zkEs8Dq3%EOIpBnSOND$;c+0{lQQEmk6Pa@7!8uTj$mN)LQ-NPaR1I!G5u#?vQ)>v- zc;+^VFcUyXQBaS2e#Yt>q8T9jnS*OFL+YJgo^mhGAYNPqrw_9_%NskLlOfY5 zkDoYj;mWEw5;8)%S$ulhKYNE*UrFUov!eN3siw2)e!rq1s@n5rv0clD2wR|r zxNZ*yBs0(d9vf}3QAgI5bFGB1>3#h5dAw-dR^{MonNy?ZHSK#mhCzCc3K;_wz5A=B zr|1`7OCsq7i9)3YWhHgo&it^FUKuneN(~GmlazPB;R@$SP*%ai3Q;E;^Eneo0gz}> zv+kj~N=05Z6lT?jz~dGOZ`4`G6Teb(p0f`nsgNHi51C>DCA-3;lf}5x9zKQ!y`4mD z=Eg=k@5`lUT>8dvS9GS6{Rf%s9_n$u#cGn_cvG8nDC*#rL=2%D1p11Oak0mjM$8WI zhBmE_wqC7=I4T#QO>n~YYG$ha=dq`*ev!S}uC%hao0q=EwIb+QV1eQr@%i=ORK>!`ujdatrNQ&ENgT`58zH(1SxvAw z9WG<)20A)o(__d5IIxW(G{5;c!hP9kku@#_VZb19P?6;kcmDa9_J{F{w9T)qvGHUE z!~7`2-gvmO1PUZVU=bfkL?B{8(2#FP>af7qhEC>v5Kh!g{dkak#>)8iKaKg*PcUV6 zkb)pmVd3EVV2$}(C-$+GS0bW(Xy6p~Dpj%CgHAjzA0thQ0{p?#a#0gX_OP*x%+qsS zgw~n-kq4v1NoGlcqg#fC@)Xc)9rU75A}o&ta$}Hv+fmWxqG-nl2gxF3BGcI|D@sdK zwSXWxi@UvdO24ikTA=M>C#~H7?27+@ZRw=2`x<$EekQr{prE078AJH&*ZG00effmB zcr8ongVpbB8|s^&5>VV3%vTcqwsAmCDrX z>@;9|ja&zBaR`Jv>($<^Ett*V3A?nEnC=bdhi*4F&WB@j7#>xVG`8FkaH#ukSlw}) zy2fv+2C~Y7h0s?!{U@{C0iRQZ$<+hOiu!2l*r)1v#S~T<^aCg?hV1B953x(m5NOjO zZ5kq(o4rUfN~IqYnX=99YbLLPph*2PWPpOaw)PTBK)>=~|5nGk+RsqjV9D$=ZXrOk*e=L}{9W=0wC{rt3{M7e?9yOnL_ z;j#>T!G`2jwMUjxLMrsxWLN8h;^5s#mtgIOo2CHAfPCiR=4R0DaM^tq^~KOHysJO< zddw3IIySdR!+%4l+Q||vDI~c92tLr5A`qQubF6^czPYtlKBp<;c z6WD%>5=sRg{8idc8>ZJ+HZ*84dPoq)u@ktH>1@Sj_8&Af!luN1*kd z8YKm|2rT_uP_JGb%2G&qgdj=T`oymi&BIKddhH8j_!i2=?zx^#q!cZT;qyxcKmL%H z)-ehlQuGly8`dU`8x{#Az##n){uXrKYV#zm*p4dz>wuYPm>>85dv`{-fufGd=8scO zj6nUOxS5dgE*GTC9X)ep7ZIgXQ4nd;08`EH-##>~M<;l z_c`Vyi#(?fJkY?Ug1tOTy!CRrtGkPvo%;^`eBtf(V7K-k`^G5HU5|$xYnDx|thQ=w zzXN_$R8$lca_3J1<_uyDMLv9M4xHMjv-lv4{`9@T$OnMRQdU!bE$l~YS~h6i7z@zF zd+(-NPJezh(wR$aYLN4ZEq3@r8M+P08(BpLg0jzV6;wrh-GAIJGIb8Kaa<6&d+wvQ zEUzUvJ4h^w#$61}1%ehPa=ic^1M1+j(ZiM~MR$Xq6cIrG9{Qjm6$zAxy<0#F*?!uP zICK%Ydi?P-RWrz2Ny1jZE|P ztYgmI>$L&2pBj&pjDvJ*X9RO{ZlHPawjAapU&ewrB7y5vd6t&IIn1pr+!VT>Zv0haOkY17;MH=lDd))4NGH~ zD1a8E?mxgt!pjWpItcg7HI`7#NN-DvO7+1)lRNz?=Da*_t;b-{7Uolt0gD+c8}suM z4&hKu$76#ishWyXVGdW6c3HB0*D`Wq50Boa`C1Kc8!z84 zb1;hvM265TEX#A4t1Aato1PN=E(AgpXtkZk{dO{Xo#ogE-%d#?iRc}wVhphq-u*iq ze3N^wNXcVBFMEStpiY8%U|ebpjsdK?54Id-);U5#z>X@OVMwBT?6C@!VB1{6*xg}} ztqJoXhZz7DElN-WnnS~y6UPJycC$&|o|1IL{{tsIcrxFJOrW7Nj1G+q@ zM#TuP%K>Q6SFNKjM6^qha6esSo6=PIYGbBb1QnDO5&&b3JeH&I%2lh%G!=B^l?n6a z@*pLDyIadmdw(!^v+}ZnW@}sHy6rEFD&>(1zX-t+i0t)(+d@tbXry=&8X`!$=3FKu zBzYXwXM3J|UIlMXXM9uK3jp1N$hZNS4G6_>sp;(e5SJ79;F0t6_I;T-1lS#|5-`zH zx;)^O*d@x7Cb-sXcZR$$(mcj0t()-Ux-Owja8+4wl~{Q=Ef`B=loeXEo7WD0tEc+g z<+`MLYUhzqre)dAr71ehMLmbY?22484MtESW!@F3q!rrAiT*{a((I63E;-!bZ}Iw; zMlsAMY#Z!!%5KezzAQNdglcqwcrC(EIldkCEdE^SXPHDAjHVd<3WnqE$B@QGdfS|Q zJ=V4ThXY}iQh2~>4`_huc1D-83Z0Pt3-RL|w=TGV|;sl(c zlb`{iB4JEC`Z8?E6?x(b4t!mL5Vr2G#_BEiZL1M&#t+|A_zI^zl!e8`k552RlK0o= z6$AvhR_$a$*f0Kp$q0jaKzV>LB2k8u$^O{T5*!W!zsvPyCQo=6N|0U1bk?JNRL}C~ zHNw@$Nbq@JAeS0T*)T%ejPvvYLFbA85lF-Mp5fs^FPRyK>K6!y z684!bo?!XVw?xh)>H<@o_;wDGv<3B%T$@&`%Bh^WxlPQLP=v;??-2a>PR3S3%D$ai z(K1ncJS~E+#gNETD@-h5V1-{rz4ugw1n*oP`^|5ClQ-?pc8rxpC8F0A()wE3D#&|J z^hW(KiM3O?3(pAH9iB@PP4l%b|L^riK)*F6$1d*6G}F+G5D+|)zt?0U{o4?;((({P zNg8(w2M+*MIJ70;i2mmW0dAO3PDRH%phc;in)q?3!EmhuO!XtjQvE5n5^)=M5k#k- zC+ZR}ZiVa4672jBTM!>qTt}f_wVWJ)R~5N>jG|+(S$kHps`?4Q5au=MP(i4kzd~&N z@ZoS~Ly%Nb%CN83i^`eCN$Xo<%q*PFx<202s&q0fe)vQ3Aaw7~4iy!}&c?>>?$*XK zEeF6{_&F7tVpqkHtsXfDGL~1CmI6Y*(n=aSy3&%87&<6&2$mJYF=`yp9sR4sWZh5y zY=QX52`ah2xDN$HGVzLQ;n=svo&eK*7h}0%l5q|Iw!7I>#*;i;k(l&m_k6i*M02Tt z(>>)Rs*)Eqg4!sEi2O}z{D-1N^E|k|@85x*dbnE~2qnT=%7w#j?r(myO9gJv8)6Z+ zPS{(bT$2~Gs@=gIQi-OkAGt4&3h?G*(=dWMC#aTZ<%{ruyh@=&>EP(|q`cewW6pR9rSqD zs`5Tf0;rhNmqP@G;hIl-v$uyBikH?{$zW;RXBG+g&&u#7zGVpnifm9W%X?^}#oP|L zs_BGCs*)O@ZPgD^6%}SjlR8&0FtGKjE*Sh3)aPH}BB&MXL8<{ah?Ko45VJ0zKz31! zem!FBEzVMz>xNnPPZ^K{B$2wNoW;^}67TPK)75S6Jlsr~0|`Nph`t_V-xw@MLGf8% zs#N;a;MG6Zrz7c;!HtMgfop%;=(U0*9%iZR^{10*j4O}n|{Qb2p>bRR;>Q` zedGVk%7b0t$kssyw|HQa{y4W-f^4*B@he7UN2jfTw@#vhGQqO`gfU7|ObETaz5 z@Xkk>@v)s;73&I(iV*B6(fUcw&ECuI>uVGV1{O`R7oDS)?`FH@?&&$heq@#nt=UiD_ zbUe7;4U0U=e zn*a3=pw#tbt!-^@Z|_~F(y6}uWV-><;py?*^L*2ktWe}H_^|+yjBvxKo7a&;^S)iY z9d(#GE))nDz;79ijkD3PV6 zQ^KIcUT+m%sa2#D{MMC#GmxuBFQR)U|J|&uEL3R+zOh`qN)!P0vgTtO`s9s@D+7V$<23H+Es=B2!vI>(jIZCtw%P^R+ zaLq*U8+Xbs5A`0P+_v}S%5#%5!;prejVr-u>oXcQuh%F#Pj~Sl@VV49RCH7ntXMTe zynpug3ZVu-<;rMsRBE(?=enV*(`$Ee{v%CQq|7?6*E{n1qOL>rg|MLjX~79p5r+47 z)cla=O^#zv<=1Uo*>9lq)Bt=Nmn=*55>ewb5*`%1O0`^D|xxR3~0P#+5y$FGvMJUdBsnAcQFn>>p!K6D&}Wy`$8cu|ng zRlQXd7_WKP&VB;O9yjuJ$6zqI3GsA|D9Y;)#)6ZzF_m(si?zVEoTJ}A0kd#u7`l+wZ_?=X5X*4eb#z^F@jXu?}O z%?CB;3}S9EOcMG#%kt9t6kpT7?`&PF>ULpPQl)Nw9(6U&wV})!csbto;w`K$N$z%c z%!9)eIP6{KsP=~3Y^srTw_!(4)MlMDQ+_1)48feww^v46#X`r?*a)$?wRAi8TuBL- zNJ{+O(i9F~7Z`kGm!rs7dM#MsXTKqDT?JD9qo-ykrW!>ozzzPl-ywoo$T6f)Uh6yU z77wQ?Pn*qe00FqV`*AJM{5HP|Gnp5_zrEVr6$$;G{Ty>4(!*C^a261UT}kXVn*Jpq|2+w!4_MeHqFgkOj$~iq}P?%u1EJNJd?l4l{(F5FOTJVx(FAa z08iOa7B}|CGD@tTixstU88_VyuIiYFUHE$_ppx(4CNeM0>6gvU>j0?EPjq=yZOOlz z{UcfM-G{P!+jrjGpkh}A?^UGXe zelP+7kCVr5Jw)tL*23YLoi3T>eCfzaT`^dE`|q^)Rt|=!Ij?8wWj%loDG zTi(z^O&o)?)Z?V^-CF!T0Y7WxWn0gT2(WA}p`PnH7BY+Eo;ddfX5G(K@vG*0nLl54 z%21S2xDgO5IXUmoS_LCRAD{V4LtzVI%;bl{JWmJxOAj;wB$lUq+1lsxS?!2$ zLB|$dEQ^ITO&=$vBcG0Qp727FrgRPZf!8lvLl06yP z?Z<>|$7a%L*+LAHcDcyx?+N#IK((6nZ9QN{Q&#o##%v|D z#5L!TJ5#@+EkkDJ=hj}Y^P}PKpLDPO{O)#ZEU&i*Jj3={K{OYSJmJfu+{-FDrM@I? z$!*ZJMtJ^+38`B8*c@gHAkmn|3o7pz{L$)bNGFT;S>^TI_0If4RVO;vSBcy=<0&%)@j;fGCAni~xDIfOOUa*#@g^mG!tsVsYzxxJ|z zUNoz!diq##)+W>($oW+l$S6vNq!VRX?f!BjV6-qKyBH?Uqc?Q zCsqnadI(1o${p|eY7<&atsfc=bpOt=3Uu#S?1(jxZP_o0mk)|pY70#Hp9Im9hunh< z&U?_OZ`5~&gNcoWiHn8h1lE_E67GLMf;iy3ld@@6NlT-xpsS>*t)Qi@*Emw-##n9- zI#lKS&;%Hx4F;?v6KG#hx0lD)-mHlvK|H#&y6L@p>j)~4&5dBPF57;w)1}??LXDnW z?WA+|2P&9Y2}J2Y<99ROig)#7S_8Qf+Q?02?`~~A#S|qZq`Q4TcN=i=2Oh%3>E zFCczAD%lmr{{=cd-dURj+laAXF@Z7SIU;avq2kRNA z*d;v0+zs6x{4O2q3#;U6E~L{uXC^Z+@S%KNYF;+^2hv>`6F8wAi`y`f$3wSm70cn! zfV&8w_=JM!y!R^qeg{O3Bt9lM=$R0Hu~}K4=9!(}N%9p~rUI~j_C>J!HmXQVEgDC` zfctIg7nV2cHmT#Bl`pe%EGj2`$LLM?T?T`6$%E-2)8gs%wEzsAm5ju|Yn&N>KPSk@ z$iI8@=GmMz)}ed=$uMvgvwUnH+0I16ok%e)MP#Bruh_(Aif4Z&LWzoxWnV67X!abi z1m@q2ovyi9xQ6J9OYXD&Q(pn;GbgIWR$mDf?5)L7aIvbajNkJW6%=2(>i+4GWJl&M z%~*ZWuv%zlZn9+N+8kNQl@tqq-DQ4AVczw8YXchxClj8+f>~eL;pO%)Y3YIVdc!b& z{%eRKF~yDd&>^za4zME!FhH52?-h|Is;P;#9BQzn78^z2yWtQ8`q@rKD-Nd_L&((u ztAE3eH_OoOF^-#<9P z1I8SXG0NEP&(kP6D59K!|9kfUBBeQS$RHkU&^nJheX}+}y3H9PE8$rIy1Fqdrmb5J z4&el+!LEEsXMS_N`8Ve2A~-sg#6or{$!1gV=pKF>LiPs_w84^-IUb_|pe(2&0#Y=C z5gb5oU3&dyGCaYF5}Uke%y**xXI>&%C_fT~wL@oE5=*_PG#xT4L=;eXtH&f=sgfZP z!emH2MjtwzQ7et^s@$&J_VRG53ARgsHr`@Bys3fa!WcVba8b*ZX&j|!xC+cOqDNVR z^}4XmYjuXE0+^&$5@HtiT0r*Ho-6AJ>!zy_4GjNq;!?4P5o-nv%e?DZxK zZeQwu9BhOchz$Dz57X`C0gxHZ#K*_`J_D(=`=jhqZvgR&Lg+B3rAH+;PlUVeO~>PK zGQOBzrls>{dJ1%ThUo9#7VJV49LNMHq|ngNvxMr8TCE#XU27NeLTjFSM!P z;G~v#ofvF6xo88K*>bDC>TLPz{r);ez-&!$UhMDM)IETe%mWo1ItJhOmPyyiWGi3B zorDqu{%hGlK)lm^yCa27QS-+P>G*Tf1}W>qm`9o~k*?qEwf(WG?3rV^M3^8bkDOR~ zE5~EeDnx>*jQ3~3oPus>w}{<{%v;F_$}I>+%lfv?r)7Ad{= zL>f!0ClO0+OVwbwJkwqlu2I7gbeOR~0`?RGS$*a}0*#8w%EMKOOsq{lJ_7>-GSnaw zXnFXn0dIR`z9RdUQ)M>5VciD!AQ|Ej+pnnS_b6g{uhVVpztMqMeNY@HLLOZxrN0{z zz%RgO*$T8GVm+McdatyVi&8fXoogSfH_+O4cWUZrPG(;yR7d^x)H#amgQNCug@y2T zz(m%Eb=arEO?Ev0F`Omaw=VTNC?$6uX_y~nk6|K9O|aZi59IQ2 z-(v|=@+jl8g8rNTPS(+%qa3Qv%4DOi>+G2pjN$Z8CNI(48b^68fc4J6EO}#znRh3uXb!<1}tiWnudnr zZ0GLfwas0vS|fhjKlL1_I0K+J5CS^`S`8?t!VVc2sZ+*KmuWdNeI0--@mJrdLWn`d ztL96ELW4vA$GX?ka6h7aNFk_iw~7Mv{iD;WN%H#4N#f6d-P);QU7Wcp_*3otpHD6P zdyAgXFqA=RM*54mOFI$~#s&wZ_7haQ2v-d?%n-V|6C~I(r_}nMc)DuTgwuf*n2rza z_ul-dFI%-CJikXUOXx`3VYOi#hpBC&uVM*SD7oS~9(<4-GD$u>`+-&hX8&XM%)qsp zkCAix7|L`Sfpw~kP`n!7rLa)`a;Bs-LC!o}9@o#Y#VNuKxD7=^@^T}1#GJ#CI*Z0J zBQnuNG{F1cLfPU?vshA%s7euov@%iRSsPO3-NzQK%Uy-!})< z74|A!C1v%Pib3>p3+I&`w5b+T!~uK1ezV=%H5fByYzF}0^yMX#S5T-eGP(%7JLbcm z9(W#DY-+-LI-e-(@h22K>eHS?-PZ5*CxYfR5+g<`wqNU9-nH3D;Ce8u`Rw(jx}ZQv zQ9=Qj40(dF&eD%>@rmeMi?==MaQ}&n1?d60 zhv|3{U3J%e-SC+&pmn;*bm#g$6`&oIj<$YpzkLRJd-$3qyAEPb?>_$ojMn+l*o_~g zuTIQP?_p3XS%%40(WObXU&7rx)W9^PV)A4*i8erb>G6XSNnIbVLam!breV&^Cfj~& zmejR5@qUTp|Fd+n()tjV{T^qLT??bdX!hP8%gCBFeN}&Moj9IByNgKik)~IpNPc6) zk)%k9O<$O;=0>I8vVV&;ohAQy5^0kzg_1*h2)vO3WSdg!Toub-Kdq>l(ID$9xKsUk zTK)NG{pIB1!r6(=u&k_JrY`?eOIbs8Yi7mnD@(%Wk_AF z*<|1l$C)I{iXD2ao)@D&`W9Aui z?fp~morG>lSTMP-Ft82@>)BCb|*e$fYf+cn%dHO z1y!fxx%gU@j4+Q4Xjpf-J2CFT?KhP=2 z%^Op^hjuA>W^+Hsb4ToxQB$Bt)mNFmCYU`6^KhIOLv+1H>WZ+-l=dbZi`?qb5hf*zM`q3q9{V=dhUgqPL`lh2}QDA8Fa+Yq;y27Rpg(f7Xt*`Phx;kt@YsM zS=*?3gJlP3`>mv1yI%$8sG<^D>9iM5vC2U0u+{1L{k!MG3(bX$BU@mgtlOZpUl=^V zBgQm?nwgm|Rau(0+p*=m0J6!|y5`$0rt$M96N>)3D&LUiQjmRRiZ*uWK_UTViv7Z zjm8~*6zIn|hEC|4l)D&95N)i`ozK6~exSU{O;JH=$6L$^U=5w)1-G+14?Iup#;E*a zi(WH}OHVhfhQAws`BnW`JPo>isX%mb5Te;M8U#?P8Bn|4a~`S^8PR1oI9q}pS)2vV z5Li-P559tmHC-*aJ{vYRZ>uP_4%4qaj}bn%yWa*Vwm%xHEjeohNG`gc*UqNB-`1bH zpSlPfK1_s0S(7`kMg^`|Gy?d?+=IqDeQ$u}yt-rKz^3iKsuVx|0VxXa z19V2pCN3q7zu-{8zJ5ff zxy9)~iFglam{D&seJ=1U_y~&s&Mpci6<=$r8oqdB^BfG)?ecj%C>fnJ-28wy(4izW09S&B2s1((2T4pLw;w+eJ6M>y-gdTU%jVoCQX5 z^2@&YHwjp8=&P>w$54tq6qHQwYY32E!e{@t7eF!p&9i~-TW2Iz8coK57z?11@H3~r zj<2@;T9Td?Vsd)T#kjH7EoGo!q^vB&`xX$Yi+ksTOcVBd03Qy4!{rSTZFDNq)*VY{ z&JikUPpELd0LKQmT49f3P=y-T0EfPv^N;t!#r4Ov$io@~)6R^$-8k=|zIdSmPY`%K zoe{jgLI|{e#z%YH*t-K}Re?1&#JhTk zlIn?u_X<0%hsD~#;jii^LWlvi1GKcs!N|#q4R4U>{ND48#eXlXEp4;CDLD~+aGH#+ z0PC0AR0ylRe+;57k&>)Wcy8$Ik8_LdwL3 zaS4@m{bu!Rw1c97?B$?x@->%ZLO%8u^kfvk8Wc6IXGyoWb-ukoSvMH;TDUlPfGUI0 ztsNm`k)fNGL`n40o?T34hv?~Wv9$4WgIr{hEstivOK{%no=-V}8xc`a(Iq5#AQ7M1 z#sdE9E8qY_3kVe!?14-b&cB}Ry$|9U%jsn2wpcZZvvmQ5eL#NL0G+? ztP2_g(^ZfYX8J0){#tPk=9MU2_c@M$s3njH?&_867C6c`SkzD!hJn5uimy14CIQQj z0L!c@MkJd!;#AyegkZyzDsaAwG`$bCgh%IyExB^DSF^t@hyda9nI%129__GaKYBWS zU=;(9H(53LI(c`NDzC%bw%uCPT*5Fei%wz-;3>YnLFZk?$v*^^Lw2#t#Q@CN{woVb zAj#*MtuNANPYaNU%kG4Q!!%i2(oxZC0krpqo#W>sARUJU{qcH{a5 z%p}L?L);+nECvXgMjT9dcntMX#&(8uJgJncs0JXBf2O&--DzNQ7PN%+@PO6oNLmVm zM-I#Znc=ZJ*qrsTs(m~?uA%};whxMXwzSE^`8;$qnL^`neIrN5=$*!-McA{De7s$d zJJ!L(gC9O1`Kuf<4tS+o=QU&eIh!E}Q2M4rA0PHjf@S0#O1sgu*a|CM*iX?mwsJ46Rr z^<2xUnK(fV_tsIcqWWSjoxrH6vMp6xYp7_e7&nsIhA`f}uJhYtnx0fCJOQHWrnAgm zJa-PrlsrdHd{w8X-~LDMzUO;tubLf{v?;mv8kH*D3UCpuID!F2OWZ)Vtv-SZ`ZsifbgEd&4+5%6Gs3z^ctGP9Bk+pu3(*VmiXsx5woRI{fsWlXWr zWZzl?7*P<&d0Kp9SVd0*noONgq{|ndUA5-hR63INjKe|_t7#l=ovnI+j6k+B6;{T~ zrjY8<0rW)%ZJj9Ny)Tb{$XH+4E({i$Xa`;rce7=l-Ux9AQcvog<#E!UPaY1BS>d+p z$eb43{^tOjPYKIJ&<({{UZ~QBt2ohw$64anN8x7oe2dYY)_w_**xGdWL*Z5N!1{jpM_M5OE_+3g`14M0{$0LD@SV*F_ ztsBejuT_oCuJ2w8t?JC?z#O9L7toSvZQ;_;s6Cba9eUyS4nnXA6eaTJ8`)wW1mfvVZl0yN&oX0 za9#Cd9v~+!^(J}i3lXg>nBOLJ#%BU-6$|T43k#^`HQZbxm5}JTKatMXLjkcM>?1xoa6K;Y zmDt{T+VlqSTzs$FQLaYp%&G)QsbL4jlS7_n8zt*;QR&Q8yW?xa7f^jje{i4~oh2QzLZ;(Kz7y_8>9-Y73+ z(^}9|V7t~Nx1abKy^rIwv&B_e4LA{g-*u5*k;6TFGY=jzFo46AYTScRrPAg6jdgw6 zZKb9%*|4wC2WP#A&>%Uf;>W3->a(#2b;_WtoZX$Hg*QJPRWhv|y2R5e)42CYZmR?m zEhUN9p&~3I2R;n(>@Eu@^yzXoY}QZ`G?HvFW8e_=``E2vesfmpyIZyzO#0CK#G@G} zs_|1b$~bh(^Ls7k)eiX03ytt_NDw(Kkua(*rM=vFJ%5c(1H}!APp^~@7@`}+h@aHT zGMDB(M%T*=0*B}QXSC;pj(UnQnZM3esEPNZf7O>X3m=&#p|9)xEU;9;I0m2~wBbI_SY<~avJi&D{5Vv>c4_9vTV`=m6-*Aj~u zDAxeuMT%^XS>t)faaJ=t0Tk%_r_WcbN!B2P#r)Bv%HpDGINM;SnHxKc!0vGPinjj5 zmta98`=ia$N|7@UY}jeY(%Eq zPX$%=x=SBU$A0oYCpno$lQkA%31U$&{<6=K!R`Yc9Ioo|8y+j;OfsvTZR6er$ZW5k zuZKdD9`%`zb*Y}098&VqcX+$n>4&#{scb$Qw;T*2Sv=Lx#NZW*2P-Z5jCu}Qs$Omy zRL$2k(5;DR!cvJfWntAchvs1~#5f3Vq|Y|};oR>DGHlR1Gj`YSqEfZ~YM(zAY=R?* znrAtR8ly3jU-L8RJBKUX4VWA)c!T$7nR-tW&1VfZ*ODJKvR4FHnkJ)+Z*q-qwt3ws zf6*o)MXwS^c^U@4r}G>jA4>;uF!dW#%=No@OaRP-%kfion~IW3-Fx~q3Wb6LkRy>n z2M83E>#g`0MX6bFbTK`=ufKiC<_^Q93B5ub&K^5Q64ev-6#3FI3W1d&|6?(R5j`V+ zFg>27-TR~X22>Z6xpA>HWwUBer}#)3z?ITN7M%#52<1qFBk3MfrcH02zO$L&x&lPEZXnN(##pNuZn?=msH@ns*c2 zjlvbsR#ul+(@d>EBx03Kv$frsM0U2Lte^cMqLLSqJBZardD}Mh;xCjv%)-hG%aa?PO#fXt3pLce z&3#XFgQtlh$-3VFArfaX+ZJVNRZJhb%;uO+Q9@#BiSdgXd@bfRZ? zf68E+!CnS2+`-DakMFp1fz5SR&Ls~;3V2hoa#npG|M5DOMI0WO$*kMO^@y`x2&i4a ziW=OZxNr|nAjpaWRW=T-tWTqr+{5(S&?hkFu%zFAPbt#*#XO9r*+L9M(|Csl=8@`; zwgy_9*xLfJDjOvlgTJ)}41EXoQ0v2g%CaG3K%Cd!2flSgG(Ex_91;F2b(mwE-iCkh z`Kh^YJ#8SEJtT>fVa+Mkzv5Ok+O5{wG{WT@;%C=(y{wzZwDb*C7gR{g{P62eHXOUu z&9JFJR*x^jrTgKqbpgY510KJ__4l_u`eznetL)wchbE4aVQd>EZa6_I?EaU+h@)YM=| zugq$^$dS*pKX#UBQcK5LFS2{n*dBF-S~84ge0)q2x{|$pkyxfxS&?n>3IRSc>fs(1 zHmZ%+Y8#dZA9LQ}@jkkh{M4*W@sFlNi5@C+I#GPqBryg$Ct@HT{Tn&eNf1740QrWF zOZ**Ra;Bh7&UHbJ0=voB3I~8S-KGo5SWX^IYF2A??PaAv1`2z6 zc7l9>V9;j-gGKkB&{9*=mX}miP;z{g``s(4s4DJJ(hl$eWo;O01QoMIa4^36l;sBq z@7i!8_sStft=6ExKl@uf4V0LBcsRHiT6kC(4qV*0cy`%PhP(|GpijU&hcbNP{`7jH zZnQ!NK1Lv*`j`nemJlgsk;%FDC*b|sFP;a`5ZY{GSe8u$)5B!%_#kMUyZR>KCJDou zaEaHM-q5G31c=SG?atB`5N=w-IC2@7gA9~F*8MHFlTOWpTfbDY?3SvJTRY|16t+;2 zKocrjiz4Fv02L5L`E0pJc%~4#A_v1yg7P`TEsiH!A`x6)+afiNp>y?PHqF5c;Iqe* zZq8y0jo#&Cx$>DNl|5_AM0bil6kjsNyyTm841y#K;OCa~mfe`qNa3e~5>PHN$Vgo(*g?Fd=1+J$s1k)q<9KuP zGS#K!r3Ga!yrqD#!rYTgK2J6kv>HXp z?#rW=h$+D5T?O^&{qODj`nJ+{%tW?*AE#beT13MNI1*ao2_}BZw5i&bJ>O?NSqDx zW+X(9GbWVlTJH>9yp%))Lz$z$}XY2Sx0C)n+R9<7PPFvQxaO9S7;8$ccFqt`QQ(1i#2{SP0F{-*U5px=d3e3@erQ5+Ty zZ8UH;k!(4)zEbyl^0a6ut4j!UT)n=mtEvfF-(s$m-Pntxu_*^9y2BG*f6;o3hlBp& zf?%=9{3KY3Ie4B|S5<4Fqs+@w%3i}%p^m#JrKB#eA+Mt$ui}1Wy?AJoKLEKc5fg0> z78LzweckA<$?WR+O7J;+By~(*I2?}4wx_L7+q-TYRknDcreCXN+yz^|)uGTJN$P z!?n@B$SF?_0u(#i*F7Z{;oDO!3NV|MR%VW)C9s2l_TU~Iry)&ZP9>^~%^jZ<(BN@QfS>CVDX@^uNYL$`Kq?guq!E>5zK9!`jH&q zY%MaY9so1Pfi3MWZM76xXz5p`;L}6ydh^#Qar2ZowzD=E8XKU?{|=KPibfmTFnQ8= zs>HaEXv2_FM|6OgS`{uL;aD>!l2=ta!f~FCqr7ZbKBTXvi{=q**Pvir<{xtlEb)mz zCWOg~*ejU$ZxRhKfW#LOxJz%0;Ya{b@c`%+O=&YTp!;}3K?yXz^|;v#ay`FnP-Lgk z-PBrM++9u+QNzo@C&Z||$cryWiR}9eLjZ@Swiz<2NN`9`rc6#w&dz+c``*@@vu@{{)5F@Lxux2! z{3OA3KQV-!G}F=1Mucy=!?_w93oc%iOp(_f2a2?BM#ju;Yw8-K)h&Fx{o-k2`_tvB z_m1Fro1+aVha=hrqVFIPqEo9OvI}~!qZXB&jEsv1@1@|+c2@)?lPHU1&3W7#BZATbp*pP)Sa5^3a)*^M8w-s*k=* zKmsZmh;3n=7fcOwSC7tq0Z`h{V_%|lq;6xYCUg?pw)~FOy1YA6WB@=HrqYp(BLUx_ zV849q6T@WTzKkib{W~ZavTxRcHWvn6Z^KxgdPQ4i-`_J=uO6>IK|)6&{xA;1h0hpC z_)W6{ZGrwym;JVtQbh3M31uLctvucBEtqQ&wwk+DKMNB-?Y&DccRu!L9Z-wt)ZWq5 zdSZ-k`mtVDRN8KRJ(qYE(q(@JJm2MG-n}IXJWWU5AP#DU?BnyZJFly)%xyRd5Al~L z|Mtkp$WBiWcR9;l+uFUr8dt1tw1)mCEZ{1bi;nqrhn3eX5%uOi!{FDwjZx>1z(O9 zP@IJB<0_1oTss-jH6N!>zi#x%Cjbb(U4~C?;V7_;I=HoCpzfAZopWV1m1t&I(`)ow zWl!U|#&7o#xM=KV=Ynv9?1xtHoCfjtTG1bw-D&3sL+6MLk&PCN7p|>VLi_vD-oA z$nD1{a5sQfT|AQt%QO)3L<&3>0};e)(@G&8NJ%@8h#FHzwlO2htH)Lt{Xl~0OD`1+ zm`!w`ijohlw6o${b0~_#{^&na6jLJ@I%(KIDjyBw;#cxQI)Uo#TvD+(etarEz%))n~$1P1sYqvkO{h!FpVV zUNV6CPOqu6fvzQNs!9xZG2;7p03?!BSEC)|NiCk&G212}#5FKq%JB*l{&&3XGDQEk zpVhqJ{r1{!tvCGAo=^S1{a<|wIFqc!NLeN`qX`!D9zW}dMWIz#a3V~+xz7OZH?&-e z!wIw4Q|oW~jFKGBAOXgh0Q~!oMzp9t#f_Sn`Yu5~a*sRyR&L{2X{t;!Gx&wX}>k zD~IE6lB{J=mcSJ^jPW-DX5pC*YU{&AH}rXv3l!meLrg9aKuaMHCz2Zg1I4m`FtQJ5jcyrbvxDDXVq2U>+` z3g8mw8+r&l51WDyODr?7?GQYu_)00Vh7`DifOmgQu zE^W|f>8RQA{()YPgQfid+{t*WY3~2yyr)0#e8T-be3CUpP}EI)=)%cYjMM%K()v#) zT(*~^yrwGetOTJLJka~u?X`sC%7YU=f#t?Xfn6&z^m|JHPULGX`6X=xMdauGLxTYL z-nh@{beb_+rR97@MCe@wx9}0MwBZ_xf65bPKFH(z0c#n`A{i3n?a^~LkYmr*i<0u6 z>@48|T#h`v)l?jW_W5pr+0q@|-a8bT#f#NX_dAKLs<9A_1M)>JqHJrxF3;-_tj>YVyg; z5I(Xf94Pl^8Ri0Kj-^UlrB1#=N%ciPDkx45OK0(iL;Uj{3>ZguyE;Qvh~APc-qlf& zdH7CCQGG1F_rD6SV7zqXOhYow#L@{6-dc(Q5Gc8`Zw)hWpWU{D9n|*nyIdN8R->yl z$Z-+W?8R}qH_G!OaY+S;QdwZZWYKvnlt`TYUxPA$ff*9)DAgprVn*9JH{?J$O|te9 zXi#K+&&#Y^g;|&{00*CwyfqDP!;NUD6;uC+x477s0=JvvItF*h`NvOK}?6oNWbi6VTxiX@F^+uAH$332Y%wjY~JD9o1OQ%FcN)Y zOEvOJ?YKi?%C-+o8zld>t!Ut#!H0K5CiZCRa(cB*O#?R-+Pcra*)>mYsP850Y&C|) zc*%o*Yp1xEQ~aGd<6mC}{4}=6;Sf)CIx8%4j)I{PI9AcMjCElQuI+-ycbByQG9JM? z5>{Vii=3-P9f9r+aHrWRd6ae%UVDR-5f;mkkRQ4c4M2+(!KP7_P$d~@ju(>)({3Oi zPCqV<%vJ-@J@10|?jkS1z&K|;kPlMc%oKrnZ~FwH&ZY*q3#Q#VO`?9B*RfFE5GO~n z?-vD~Ei5b>_B6kL34kRYT zMN~>d>8#uS6r)f+QeTEgb*dr@4kbPEddK*=@gMIEDbQ3bAUXt?|GTG{%N~MD+tc9i zW0srmj4P6J=M|bXkkY)vxo!0EiID-}>D6V&CpOn-bQl+=OoH>Wyy};-9B>x!kUGKV zr+U~R1y3VPgK*66J`UsSU6_)}8B5?INmoiwrZgJaK# zA(mddOZ^2oeueF2^?I3(BkrsVN1Z%#)#bE86I^(D>zOLq=BBx;JpQ_wKwsi2Hh((` zkfARjFV$3OGM+&}CG#K4ZOj$4k&YP4y{jV_$9M*mgXC@!BZ{V@J^kQ&Wo_;0L$;zs?RM(UJcUO&OTp{K$SLMuDzQ;}%^M zQ+{~IGuKaCX!X=fqt~}f4O9AqnF?w39Hv>#te;(f0ugo+W+><$kq@_eFjCsVEK_b;d;h*hVUk_;QT+7Sk&1>I|k24MSw7s`hlh zIFZe7CB|B8X?}om+{Xaa#f$sDg*4VEZ!Hl`IgoQ6t<=U^FbT_44o-E$C-mvUkalP_#Co2@14dh1eoW9tYx;r`IK~>d^T-9(2r4sj^NX`4> z1?(F@Ez?0+<)J3diR`;mHGb(h9V5IBEB~hjG4Wqxadija z)ekydBr-XOJy*z`}6`bfVO!azU31MbSO zWz4EUPw=sG7o;=2i#G#eN*cj((sIC?#rT|RKlT4x@07KfWaa^pzUhXen(hDkW}h4= zM<;~dh8bR{e7qo*S_}Nf(xE)hsiYJu72Uj~lr_GtFQtMmR{m`v-0cjmK5~~{z~Rvv zc$VVFCghHY!vvr8xvMO*b%YC)J?_CpY~8&$m-6q6*S$k;n@t?~*il?2zwJ`L=jq#2Lsj?;L)4~Q6xKX2aD4c+uTQBV zR5k&hcvIWhfHnhEzr)kAqHhl_zH3ev)1D%cgOl}9xAB{9S%y~1(s)9>1PEj=D1Y;= z61Yhw$56~PO5I#-&?U*tW9fGK2q5C7gC^9K``--%bnF1eGTU+5Di}B2+=1!WZ9*~H zW>&^yr-pKlV%~|H(A@h>Kt;_wUUUPw@vz{7(7|XwZB!-`O^_utHP%V`&Uz~k56M*F zUDV~pH6_&X_r3P^Z7aI=`>|y}^?mS~B3NCAe9pP7;E>1Aez3P1Dh1NxAxe@aN^zk? zlE4d<_w8F~wWBB$oG7o%&CRt<9@lSLYgP+Y*Lofar31fbB*_4H;RA!ChwGyl;%VFy z7s`I$wEUL4DKpb$orwg_=N2i&Yv!*@%>se+I6^~-H+76PINSNf?u`!Ee#7H1{tfbU zL}x}WhslVrX6V1nLn1mk{=CSZE%Y6wa?QWj_d?uEr+f%8kr?Pu+54ao8`uQrX#BkZ zEwC7l(i~b>5A_6$_Y7AmO=K-aga7MVfi5r!5Q-heRQSNd1O9LX{UaO5@D*FkX5ZN= zKpEnYqwApH>sM++lOWE-+KLED$lMo?Q=0t64jHaHs*rmRHIQW~;9A5xrHnROP^%qIniqg%OxKluL-7U=>DyrzE65IdDb znQp41!7z^s3tFMATRqj(sdUB5+yvaMDa`YgQRF6v&v)hm?woZAALt}cWo1=ur?;*3 zTbp>YLSuKZWOM$|9Q22tZehPV5nd1vyXaoO)a-u!iD{j3ky8SQ3vzFTpYwPapuoRL!Bk7kdpR)Op^#93>YdKw!YfCYIT!P>JZ#&%(pB4 zv_P1ro3g0Q&sXRm>tUA z6yNJB?%03B-3tvV%2$l*LD_CQ*p1kWmUqZ$xAQV|Dxj&Q zUo#R!epBf7%z~S)lQ$oHH(`@?1!~6sD}_FMp+Aj{Q)Y?20)iuw0KGKVB6kCi z+|5IWwSSvb?2=ws(et&sHv@%C)lAvv5M6Q+h0UIWS@Jd3_MXO*i8z!MBN^797aIq_ zA-%Wh-{D~h>DoVoVnoT{w^a?l=-69YY&AE$HtIS@bXlE7a}vp9f#zzlwAnIIvQAfD z8d_>0f;)0!VNZ5A+R6UZ8HpJ)bVL6S;W(K^=4VbV3tO?hTPXt@M8{B6E@z@QCmgX={_?G6h zLx{1>w0#DKnZt=3i@oRLd}wXB`twLLkrOey+Q?WSDZ=aApvu!szTmF95~b&>d>C_i zw5c0H)9JCwcORS*X%9)>o z$4^q+-IMavh8gIlBbAwSNun*J8Wd%i8d0d=HHbSdgMfHKxLBdXbd_9L;*mB#Yk{qa4ru%r6-6oV?O8YKeEjd2@N=)icr>*&b+oROG`)#rg7n9L}@RJGPwan+#=FEV7mCg0_FEvmv-)?3hYY}jQfZ*Jk zHL*X8J`-Y?YrMr&8ssus7{&LI<#H!QSFx_zCKUtTK7q~|Hl` zvVoVzttUTzD%59S6`&iAA0N9+ilz`n3*d-Uyl#klA54do&({S+`cpHMs9GZTWq`9x zyaOf(w}{g*$uPwJWP)q`iGSR`X=c&YB3VS@z+H@9y-+b*(l+Uu{A zFtp;WnRMXID{)@5nK_quEJfs}K|`T*&a>E&y5#eh?teJLfE(Fit)^uey@&xI zi1ciQm<<=KYkQ-otlB?;_WLRyPRd^ldY=XBE-bS zB2ZglFG zQ;M5J780DQYc77j&JUPwCOrSkq35%H`%?7%@8;lCV49O{iawf`n%yq6vdP0`kiy;* zKaE5@KrsF4%uH~tAuI8zu zCIUptA*0UUp6ja{uE53{F9ZJRbF;_IvZjQ5-B6&&TfIMvQd}E~Gg8#!Ucm%HVU?p88;)Aj~sc>!&zAK!wm8LP9T5qH(4c( z;Y&@P_V~eqh)3e=ew1R>FZWp?u3{m)7{MsL+M1d|kK#KZz;lG|gP=FT0^!bXGkXyv`p4{&pi1Y4uS=i5F8xfx%x}IozrULznQmy0N4>GK39WeH88av)kcYu- z3AihGd~;Ifx$k(0mA{Y7BW*V3tfCB;Z<&XS494;oZ?7EJCD6Qd70Ffzb3t>>5rYgSjk?fVnBO&QKv#ibXdoIw_i|7_>o#6 z-&nQTkSE-dHkTv(@#j?5LDr+S-*ep zQEi-qb^hPCsQwQDql&pPGb8b>^EP6CZ^$Er_PaV9db+`@oI*)v-=n%k_`I@7gpmJ0 zD}iH>gOZE)B%$OFgX3bdYDf@?Gw7V(T%Uz^I!&e7AGLjSXZTw0Xcao6OZ8bX+FLT< zf~eqXYmjSG`zX$$KPu=YFD%gAD&y1xeZ;10l6keqLhm65PG|IUT3RJKi(;*zg~(6-{!JftTD^fJAOlg$Y+q$BY_UpYpyKd zjOA7n{M6Idz9VceKQ;+XP&f4jAW?d1zY-;)J8@FbnhY3r7s=LaX6Uq#H|!hQImr3% z4-Z@w!l8Sa?vQp~7mx{ow-uLAlyoFdOEMI8j}<4aO&d3qQQ{KFrzW2d!2C4Qg>uT^ zam)Q1yG~P*wcOAth~dWG^R=}vc9iyx(;ipI&_sp8nq|aLA%~Ym2B$V{dnf>ye2}WL z;509pfCDc9)Z+XYHaOqfO228@+9t|6EvQ$QKw#+yK?Rw1UWQ@Z54dx1@?c7$wmxDL zb!$udO>lKC?S^-*jP44vYUPQZ+dBWxYL-_gdw_afg~iPIs$Mzhbc4BwnFn8n);(&Jg%E0EH(k1wW&C_XX*VwiGbE_hhGp``d>Nes>%w0NY#i{gIPDBBI9Ah zux!%8z6Ja$rY$LwjoCqP;Q+k@O=cL;*?F{lMDqYe*INcBMv{~e!((h*Xhhv+5ZFC2 zU{g=;mySbho(gyxJZM&Fe~&gzY7p}9K;+k>I!MLv;T;U#;i1&io8*%9?2ZY?IzHkH zVaHSXA~Q>-92XkU)gfBkXi>B78s0Hkx=TVOl<5GkZ$|;Iwzi%PO?+>K#DFS_nWHxB zP{xs6?=u05!kWWcGM@DcAHZRp_0~d~5M?d_MdGMQZo^-pA!#%SmWyeea6XZVPQ29P z)@NrH!^WCm_Qpmb|I+aqZzcGR>pit;Ww@T=!1vH53~mzasI@kG#dP(Xk>}J=6}pyv zK?G5oPQyWV_t?T+J-=GFLL+uRv?^tNj$ZcE&4}8?@Phapci_@>OMN zomIIT58ge^`Rva$X|_Aob-3c6HCfh$3@3_yBlw-$ptQEsCX(j~HBiWejr9bO{vtuH zdbA6yD(zQ-d-C`v0zxxWIzu?5fz0aaq19EVRn%&Hi9tV=+k$(t`Z#IfK@b%};s^v~ z337s3*ek(kscYn8G47KbNnM(l3eM;(!TTh1F1sB5&Ov>4e>P}=_i)&7WRG4|R@PeT zc9yo63?Dlk(`G1?yuZH(2P!*oSPsdRu%h&k-C}c89Utbuhye}O*l_rPmUeeVqmXa! zs%L1kPIh(fkLPLp-p?ybD+88F0xea1k;6uey)Q;2i8;~}Mc>#&+U_?yx*B?Z6;-sE zQh%tZ(&?nBFdNehqLK{-A0HNv(U04YW!&AL$L5$U=u)sDprDcy$^WIw+3vO<=52Qx zE?qAS*MdPI=I}aG;2{PkHLT+wADi;JT@P(TNRiUTGPtrBHZe=|Cmzl4A#pILwHs%hb%%vK zhaE>aWXtD`L)CArm)*kALSZ6~JSpiIl}l5P#KmSc%kCP}(|OJ~PWxJR?OM;jdTh(T za3Gkm=*M*qu&Pok_A^82qV1{=dmBet(x!57m6;*(+E27_Qgz8nhHAM2jq^6B{G75Y z#<-u;n5Q05<%|-r<(S=tQt6)$CK+y>qABd0%yFsd))^!P+SQ3>8rjFN&B>Ixs9xkr za^?4Nl;R~-I*6<&csUva4+Evx)_tDG7=P;W@HQK+(j0Im5~xDYab$@E`7s)T!=5fr z{tgaD;gfw~E)g#SHg76WXOe*Dm8_N0NF}**!hoX!CApvGd_;b6Ib5i*LmG9J&zPQ5 z6!&x$2RW7}jE&CnUU1~9uL+!f2lc{F7P1tJ2F3p>3SjxW!4O(fmzR&3ywZC8NGUl1 zAAf|V3ZdI+94&cTl#$~^!@&Cc3k+^2ZOc%@* z^A(0r&8kYu%gb#KM@9uc)<&kY|6l}@Kn3^bfBYSN-}T#^V;5H^u32_{(9Oa6BKh`x z-W9}v|9W?x&E@j4@$muWgYBRpLkuD!@VN2y>AAJ;ZFSw;UOh*HYMyQ|9-YbKLzg9U z8e~s27b-Nw8Xgw6)&K;ZypHI9nNp#JZ6t)LU_*W-Im?6Hbe#6^0}Zv?g)e~?yMAYw zy(xbRLkv3Ngs_;9Fm5D4W`v^sY;>Qz9raOi zlqp*9Dm@}SEzO7z2hdsLwxnqLG;D)!#dBJ7`nQPJ;ReziHubp99(Hsl^=X8X?~d`! zFGDh;w<^wuorE~E3D~Do1UdVuNRU=b!>b3W=ojlwdi9=m4EuEpO9})xN`>mv?GJ)< zS}Y3IsA3cTK?UF7#>NdzvcWIND(nWvO~-t?JY5#kaWOVeh04cHW{#%@@J!Q={Z z2wYAJVyLvY8OvB*<9iT8A|#n6VK!CS0uk77^K?^`ER)LQXs!S9 z2(S7&Sm4SF@;=yblPHz`)(fp+Uk6 zcd2vO1U(sA+wpz-7#4D^8rE4ljEJQ|x)R3;afCSx7q`xR(6S`&tp^kaZDQGt-h5yLt-$*6Z*AbvQo;uh$1tWDK=!Km~}OF7VJmP}oS z3=uSX;ZC_|DW1>a2nk{;{<$}18D~x-BRgfbm{XlZM{6k!3qCU2>i&p`xQpfnWzA9C zRDXaJ1(h3ccX!*6eB~QV+O&?J29e%ss2jb?T65uGDrW&5)t4(ugopV3@09W5Z@3^9 z9#+WLbSZtd;QbV(C{CQb+Ua<#wdHl^;NYm$1|U-T-O2X)oi9ad>Up}F zn6%9al~}M;)g!U96Wwfg0@m~&4x)ScTyBZha``@}zSmyX+k^*7*IJwy3_jIy({V_gd& z=kj|$Q3`&{-x*6P)8zL`B_3s=0aW&}~2;0CH%=7wxYlYbcxYLmH;YI9~l%nxP~W7sfR%mclkeII33f zs=DUW1ET2|o?qnyd@%lFQIYZcPduPUU_oW!lofk+AA~$JmI*J`pcef}m5;w@7~F4( zQ=X0xYd}p3Evj2q;J=Nn6!k%r0{YRjP7K<%~eX?7f@wk%hG=UoH6TOXajTLf;$Y#}-dvJ^@6c`IR|B|us+W2Vrz@?;Gs8-tfx~wL@vC7>y_Yh3 zUtT14s-~z{m%>@wmbZ-w1`uEpuk@LqIvkQFPF2HvbF??`-r4{Jt zRnRu}z3u;%9*xO(at@Nlh6nq-w|=StsQua&deVpaY+eb#-cjF z6L5KXX*dKL=+1yP*=vgZfMtu7mnX{TzcCVO1HQaQ1^y*I;&KGgig?bMDxdIs-)y?9 zo|LLreU0XSLo8YRvn*u9M0Baj%C54uT4|~I7W|I?-f;Z#$@?Sn`h2r{bDy)(;=THW z&%4cF&PMRv?>yVx`K?QqCyS7Rm6A;GdK>;B;4k1SuKcy1z03FVB`lumjMDnk|ER(H zc`KUbbYN*^86)pL6dPTjWOU{&0`djBO+kgY0Rw`?9HhDcX?}ut% zAd-ypC;nR`Di$g9D3hlU>@*93p!3O-$wkm&v{Ovru>On(dC-cK@1)Irlq-o01CCa}dMp4V{Bo^YD?F zsHGbi zk8)rQ#;%j*&X&Fi@naRG_6_8c)S}gIQ#xqQKM*cxWvzTQ-j^BZ%NBH4cAI0<_ z1u5i&2_9W`7!=ey?@lf-jbxpS@Kr4>X?eF6$%b}28tK?|tnGP_)Can~6mQx-d-EY? zr)oJLV8KN3R>*O|x4Yo!AV-Bf1#V%;fHj85duUh8Z)NK5(=14nsn+#;`8g)8UOaML zjxCTyh$v81)6i_OgcdQ^2aJoMSfVQg+@kh72;uq6wodPT*Ct?h%Oh)nLrW#OoDW5s ze_y>Txe*_tPquB>uCQD8UH@_M$gA4(E33sAf*FB*P|!|P%;9?7;k{wwQMd+uW8XkL%&d?0kxTwPY^uh+GK(XAcE(2t(Jl!`eKqXrqF zcnRq)?yf~-p}-tu{oOClUGUE2(I1qaZcq(-bI@yFD>q+=a?J>c2oss92Zh|x>(d{> z8Z(rVdB44;LO~J?jhU~Y1kJ!%)tI5_6TF>8*C@&9vox~fW+R|j1@4%f4RSE2_Y8zc zknRLf#>wP6tQ5ta1DWLv6ybJOZTN~yP^S8@RHkK+niw7A4WqZW_}z_FUt)&t6OnPk zHe9Au>KiF+%_At5572nxDt2FMnYo4_NB{6mdv4jt(>J*x0Q~4L9;XtQ{&j#)6kmva z2kPr8dEma#^X;yW7-1Wbn|~y#eicYV`Zn%lq7_a2@`qY--}_C93lV6lA!E#QihDKz zNjm;#xm}E_&nfUu`5-))jW7_uNe3>QrsA$G`d(xcI|BtX3G0d7v(j~3cUTogAm8(} zOfosq#exZb)EXmtTywv|s#dF|eMJ9Edx!Vg9K@y5&m@D`-S2Di zIaAVTqf=L@oaoKC39V%OGCSBY4nh(QEoF)Q_ORMrhZnvinnM=rZD6hARe)f)y@#b?z=Q z9y2zUY{lMO-fha(Lp%_HGo~ZX*k1l+l2ja>f_gfGjwZ}H4N!VG6K(H!zZk9&zpTX4 zm2#9(sH0mjH<(ewoaueDmvK(FVj~~6J!D10iNyhM=hf%-U)?$Bd}?`<)eU95$c|df zm5XVe&lSr*@x4H%%F$$;1S;JUfENhWUl{QKATZIggBFScxVZ*X!Jcj~#(<9)kih@q zHnGjAqZ;}SlUS0|!l<>|i|0j)GYrdx3(9;t;)a3LgZeHjI$3Nhq5LDaQl|~kD@;WE zhdw{`r<|FTU(lbZ5t0V7b!X{H@vp&*jM)pFZ597i@SoC%D*2$!605&R<Nk*G*`m9@W)-2M05oV=$=X zzJJXY6tSY@U?C`gl-ppaUPb8C=L7AvqVbxNhF$TLt~T0jMa8x6syiPj#SI)um;^YU z{>d^vKRqe9F$6d8;**n;zb*vPId<|B(d|cu$5=upOOf&L@^r(^z=awp=paQTY1(?5VIeTYuuBb1`LmFcd^2Mrp4`5^vN7=BL@qd}sm$L6ubBtT!Ypk+ilZd!`G?OhQ z=Kh>k$5m#k*3gxKxn(hI{nPZvA7$i+#E94XvVoW@I_`?m8sXsq_I0r8!0#I?_2aEW z!hOIlKR{+qCzEd28Y|!rOAd`aR{1vF?`l4H0^!=p#ub+Z%u91$Qx#WM$0WpNp#`k4 z_M0x0Oz6~Q3xh+Z3F7Mh2$cz5j-glCl1dD@3@-UlW_~uDEOsl$5Lv6a52i~Ac)5lg z_DqYvp6P5KG5z*9e97A})3WtvLq_M>sud*IzYiSGJ~cbp%o9(_nQSoDwl3O{1Y93N z?nfm7e&tu!lB+!YO_EDzVrT2F#2|5!#w-G6;e83+nD_V>)-Stlt0;0NGjV+lc*?bf zqXWw(Rft&WE8h5M{g$-C$6BaAj_b*FgrFWY#V*qhVM^jz3C5Ai%V= z4u|HT3E`L_IkxJDP>OQIr2bxO4HWs16ne_g&j*+_5@S4E$rXV?o|simQ4aEiX$-&` z<|~X~!)A$!MwMax5@5Qt$gmxA2psU1+15}-KR|VI)OF)A6=GK6_>F=~5S6wPda0+& zmTaiQNmqFpqlosL!PR`{SqW&%VHP81^ z`>)RiQc1$k+why6o^5@$_Dq+ai1h-mp6ta{MnaH&-D#BpFkLZd`iG~W0`lZM;_H16 z0;j7p4ko-(){GH6bJ z32JSFFf&iV?@PQ4+g{rm)!~$;i;W#IVieq&_^Q?XuPT$jcmO zax+U%rpR%ri#il>zpGRF`}#_Z7|LVBB+%`t^JB`Tu*0lRcVQ4Q5n^@+|DvQvH6hj| zx|#-NBI~$G0tA_rchFio_u=vIV}skJxX*yZZASbS{&z4@m9_Av9g+|X*+wiqwqX=p zK!`dOH*y-BGpo(Qs9ug(#+jfZ4qP$^w2y0Ow44EEFH9D9W61U=k*|mLg1knIpbwo9 z34K+exJxILL>8BNAle!-4!W^%$27zaB446iqha%}J%_g|j2BZyac%@0buyyjj!KKT zT7M!H9o7?u6}gpawe*JaBwif4YwoX;Sw$)7*RL|bFZ{*UJ&duSoenQ>G&7?Z&*{m_ zdL~o2p^Z)r=#eS^$JTjNaA~OxsCab#6EP53k_>9^V(pI>-4oc|t&ETXaPC#O}fcd>EHuT#l1+qKx4iVLzK`&99T!$U<1X8IQNP^p*Dp z+G|*8#?MV4LifX6z?qWx?04d-qqySp3zehHd{D zXLWJ%XduS$_*0SuSz}Z|u`t4gR8pAPjx4jHt;s5_8-6(b>NjVX|6ef!zG@m2?Pwio zILd7JS1@iHZRxw(D_lc0gzH3!5mBBP)gvd-s%if}-1SbNf+^dJZv0dQwvI%MZHA0@ z-ki2JHMKex`~3defAOP$G%KRy8J!LXW+t4epU4@Dq6o(q`tLL@`hC|1KQZMLpPnBL z-vuVplhO9=@p_vyPiWcRhK~WzvTXM}+9F5yJ$wU8=wsZDZzmV_J{N>v`@k$q6$ze? z{{ANOnahn9*)E>VBvtC|xo5f_SW9$FYLvnzaTY<{@6CirQ+m=FH_erH*C}OfAFLyQGf#IutO(6A8W? zev{~wW-b$p!ZGT}TgRTRhbVL%jvsUIi?_*PSs#jFv*SNldHa&uW!*}(Jc7Z;AhJX& zqthm)1@4tGL8gzJ(o=XKN5Ye3MwqlCb`okOFMAqVI#VW#JB-QM_n_&)@Wq!2Zsp;~ zWbJi1L3v2!nd~<+B`});vlqch(5^k6?rKQda_#X`IwXJ$t67yN$M$QRbaSz4)H%gg zwSn4POZDNdGzduv2@>=~7_);)3ff6jn#N|G&Z>%=@KfW&KS%VzRie`A#!*A&l9$@4 z7^;!~X56zh+i^M7Pbuu1KZ@U{SYFbJ= z3Omgm3><(9j1mJWD$;%fOyR97XRc0^GkdcppP4Xt|Iu z69i85vwv&tTgtvzR;=)W;Xi*`@M>k@TK&$(wQcpRcHafjVWSZ@_yK+!Fj_PpL$_I@ z@t%TQ4>2&?0rBMJVd9admqkL0dWg!@&av(dKX^*}&$DTQR`o7gHma_ipB3G0#Jxbr zZzLANe=ORM-gJHNz(0xW;F1Cj&{|^m$c~@Ui(zF)1kse=8*-2SK|AXw@AUZhxm@mP z`-nUbnim^sO+S4XnaatMGyrM`PFLSw``~z-kiADl9Ab ztlu3R;ojux6@Ptt3sh`&RCw$1-!^rFk2Cd?+EdKJwx-N*Rawmu_BX%Ia!+7?K5v69 z1PZWcu?d+8H6Thp&E_O(%njqUkUC9;(d_>uWZK<^)~|Hd6eOMmhLS#LKKr{~Z^~ZM zX#48ACu-VOIP?3HxSm`^UMW&JAd|^P)lL^8s;8l=w^#V~3T^@3WuLQa4qVLvDx?b> zzbfc^$Ij(p^2Aq&nz{Z%BmpE2tVjV5g!yUeEu=-pq(z7Qkw{zw(#mS$!z*rM*bNXE zBZEx(Gc#jjVwmSde3LK~$wUq$9L)c6*6H5(q()VLUEaU*0mUI^`wQH7D@i2z(?&Ck zsBbtz8O!%~PQ)%)t#;16Y4WltJ; zWn@29Xjq3mrn-E6d$lwI(9cAo6;;nt($NL?b|hTNyr58FxVV$hq(jlE>#f>r{v{wr zlk_3k1T2#U&|6p5g)PYWnfoU0YW~tImTTYNhuE>=n5I;%rDF85xZwu|R}>lVLl+-u z)s*894U~GveuU@$SjuJ_P(vLQKv?oKSoy`2Eyc}ehv`xAmPJ!%Zbvw*%3idK7ML9a zk~bOI5DDA>hKI-FC6>f&UTg#1N+ku~+Sa`g!W?g7%2ZfmVhU1#d*=o40n>Egj$(;cz6_dcAhGJZRR)EYCCtz zO5YbjZz+Ro0kSvi^8}x6;-W{5YIsd_qsNnEN>z%jPZ7IZ+2dNsiylvR(YYi2Uxc3~ zyk=Ds&z`46|2P^?R{(e^8GQcmd^9c+rn0a!r1Ng8jL8f?k;Yl`eAiapOx5moj_$W& z^f21#X#c`PN%etM0iMn(v$U&R%G%vDMyN#J!G>rh^ z@$1o;d=C})*`8kqj8Oe=eT^1q8Xhot)@QD6Y)g_ECwAz74z|lz;lh)gwoC^n1diJA zBQ6jkmlV2i9bBc~s>O2)F@5l4$p2+>dR?NO$HA(KMke}G8w@A;M)R^5q)2EhLw{2F z*TE2X)1K%H+_weOQgcS{a3ydoIocL0zauBn3PeT{+>qglp z*K!B#ygi-~BWj!%k?XY)-?y*}hHmLpj-L0L^vshg(3Zju?!?&A#?Bv1`W^z%L@kNS zma$_gA)i37J{o@iX+C`HK%PIn3z+N-n^^n<{P07BND^P-PxE}A)=vu)kzyJGfWh7k za2HpUg0D-U&Ue~?Ip6BENoe9UdPc@Ndz&t4^6q&PCF(%bypJK(GQ~V#IAJ;2-~}(Y ziL}?(=T`dznmI>DnQ;UETh6`NW*N8wr8!~%L zrN?WaUm}anrZ5r``3wBl-JR)qi(+ z-6nbYiNKX%zx;%lb{35vrmtD9k;Z!;st;p|l#5K&{Y|sAN&+;Xqr0znzHWn~NUnQ! zeHJ4Hl+v-P48JL;HfTCD3g21wVAN@w7OnU-G8xxKg-Q?kb8%lQ zryq-iSs>sshB%_bU>-skv$|tEg>}FZ+APPJ>EI8yYjUP?(-WIE_zA*%bz$BA$JRFn zR<>p9R&3k0ZB(5?mPmDR{7#}QEK^9?| z4!_lKgR5$_^~Xbi#`-DqWhH?#fb#%gD)ye7dzE7so}`u!A}owL)yN0Xqu5gVG7VC7 z`*XTqvl@P$?#C!N%|5L*-7bgG;|%cxfV}&M@SNc`D`+YBj7(4hxyV0~7)szM6d?1N z)=XD#%kuCeH&zz_F&$6vp^M_Gr~O&=1hh(n*=l0eIKtE3@;^o*6H5_><46~>gT&MXBUpGVxb8}2InUYD|~s{vTRGI3Vi&+VR1 zE2X>gm<)~`2S0X9%^kldz7v5eOIzRlF$%YBr0>-QP-(A6wS4HLFEO^oF|PLTab=+jZii4E)lYP zO71p!vJAG;w%^qOas`!?`d-(hh-g5a+OPw(6cfN*-PI)1y+c_9nr?rTv?^UKOCQ+< z@;n1i{i)*;HO$P?XP3-O5LB|^T>9Iw_sV`P^Ow5CSoW1>C*6Qn=7P}1iW=*vHV+Q5 zC*M|#gT;0APN@JPJ81Zt$J;hA%N}9BOmMZyN{B8BJDCJj!I{*iR~6eir5CeeoIEd}h)tnQ!B4_4ZB1 z=Z47tMYCHTGwJqt;*W7G)a-2qX%*5xYRkEh+DJgP#I^Gw`pY!nI;=p9d(+cxLuAV4 zipDUYiX1Ul?c>Ye;z0~<(P(%hNdb>=pjTP1z!WsY$7Hn%qiXiVNsow8k5MH2T; z?rSIX_21hqb5YX7t?1||J*=5!zjX9`o=3=2<~Z89J-Ky%$bCJjeW8}K+@Wz2Jf~!r zmcLH*eC!N~4t<^5eXgH6?kVXgH5iS+*O~!#IzD#{3=VO zZF5>^I~PLh^jlpd3N$B6099`JdY_7rrap7_)}tG(u@ZL zJGjt*jQWffi&rrT@B{qgcUmL7-kGY4JT#gKUB~0##QRC+GIv3ZQfVvtvYAu^?2{Am z@R>aamIy}rjkt0oSX4{)sKz3+39X4XJ`$nm`KQw=%Lk%#?sw03ZRdh}>uL>L7_X35 zf0^*DHUQ&~?!GbfGFOwQD-9*@e*012RLPY#OIr4}DM+32z50jZ(xS37G(woyMi_9A zf#^M@%SOWaQ!!sryfk$c%L9lR-X7A0Gn28#kB}ii6%|2M`OWoX+sCSzO>u?C?M!0{ zi)$QLjuB&1IwPcP*fdas3@BCYYVXEFKP#M~Lf)N&=J)M;ge;i6X#sOVHI@n=FY|rh zH6g9WOshkbUv5p;LqVdEf(ivL!%NP@>?C}u?Bwq>XZvPwKT|_7nOQrRslNKyxoj-5 zDR6cIKc8C-f4lk2Cnd&{v09^6E7jd3`uDSts09oKWcva1u%jJPJM*ceMGJTY*d*}< zs0$nCF)FtauR}xN$6hXg_@G0|vMO8SWvMw;$iUjENXi73O~K)OLLJ-2jjkR5LIg^FO!EU)V*|faI8O#(sRM_W%?<-zZ#z>Zs8K_*xB|@&S5+p5hc8htC?{pOoc6mPPuo3K=gWLg zmu*>|Eb`Xg7X<`uvvW5M$2oE(3vI-CybuV!*9(CAYn0xn%ZT(xp?8|y3{)78n3 zts{SLZqeKF#-yCyH$RX(J>6dsMQu;_8`~OE_m$5xY>pP(b=tTpba406h=jx6SY~F^;id*K%ig)3U1Z>q=1bP7&$7}@;CS_A5xrsGPJcTG%v4fG33Z{ z1%5|^c9!@Kons9c8(gY#0v$Z^_6O?QR_YD=x*^Xk2IysTznEfN(4w)UQ?}Oc1zzm_`+dU252WnxKR;^7E38Vd>**H=ZqoG&L(MJ+BC+IRH>`{d7t z&nT6ID_TtRmGnlJYNU=ymu(2rOwt&(cZ?GTRAj74I)U? z=?a+Rjft99XSn9~^`zzSe*%>L#ug}k|F%)~J~RZ-Mi&+q=4$SxO>g^frVIyJ*wWK#6T|6aOrPG3o3dYP=E0`NuO12~<83L7LO0fBs+eSW-I zou3Ij-i+|!%NaaNlRI61Jmq|RK2Z!0YeBm$qqOZ>2h1IMb(@A!O1JO6-|D{%DrC?2 z+`wdIWdS;5-?^OP6~_h<{GJDnmY2MquE3y@;3U!uGq7*AnAntlpJQ-VVl# zc_a8=2jZ4m6S>_Is9Aee+4xbgK*c)PM10pxPAhL1vM;A zKM(|-?Y?gL7I!S`c%Mf}^j@lJ7ME7`%os7fZ$dEI8Fa=S1b(i)cz?!uUl=`Yx$PWo zH($8EP3t`IuDvZfeLa3CGmGZePR{P`HhjHyb#G^HJ#E9pG0#Q^C^UC>cf0L)JY{{@ zQ6P_v@W0Mo0#x8-8v9taS+Rwe+dv{nkpX8oC_Fg&>;o<+*!di>3lLdrwNTP615oM< z7s~o=%u9#T7R)x-ZKwxH{tR5jI^3!p=Hx4ac;D@T1+1W~<)o?(NEAtj( zdpK0pdu%EVtGOIbHdLGGv$=$`a9qL%D$ahYOHHkr5_`Dbvg9Ul5N#?k-adE^SM`SQ z!Ysz3a_#R0Lrt#F5*!<}QNn3)u)hfVU9RnoD`lx1wNt`11O{s39#T_zEHk8Z?DfCL z4aNA64-QO1Fdan)Su`^}Y}#-*ced^skk3<@UuNRNY?=mHW5OOrN00dv-g424C`qJ} z2p3lOfnb%C52};-f(S{EkRNlZ`J+Tv2ZRsy;=HF#5&M{ph>DFOEE8bBb6FH&`w%=5 z9y*J}%i|{>>O??UX{N%rfa{@1RST{n=NifjkmnAC`ra^SS5i>}$(;WjPGr6R};yOg4w# z?`reBrV9xY6j~`rF4yiou4l9Uq1x+DS4(p@z9%CBKJWGWoj~etI&o@2U#7s$rtru1@A$qi03cPY&FOj1 zi3<~tTN3-U&2ugAwcjI5W%&mLVRs#;rpNd3{?zGs-9cCVGw0*tMTEJ|b&LDjj2*`U z7!=X{bFxfH?`vI+0DwV!y4+Ax(IJZ@rxeKIcGy>|=}wRdsN2XgkC)?pF(v{E)w*w$ z1ck<__&LxqelkIVNewT9SM85ETerbYlN$ypr;KK3OS2)oy9G4eQd7Sy@5Dk#Nt!g7 zTwt!DLE4kM66xI&*s92O-%310ND~}q-x?bpleS3NaRZ2irPFyHy+k{@obtetI3n<7zWsV-?{4viU}sJJZnv<_jpS7M>+tf9IaUsWqL4K3)mO$EwI zj%6b2e*`pQWTqMseJ-199RE9tsDTHWpO-tj$F1-O=GrTd7~4>(EqxvLl)k5XY%9dL zBOPt1U38nsmTdBndU#!j^BlVgm(8?o)Td;1gW~X@sa@E@#I6{{HxxW!S>YJ2HWyr}mlait=3!@%&%(~eI`PmS9#-;=Ye^ZV}+f!JV@ z+0HsuWoa^&7dUa@J$vJotF50C-Pi9kuby6f^etxS5~hps~qQh!I+uc1EeT2 zFaHc^#Qv@sT|aKqz7|LRW&x^^S8-Q+L?xljW_0k zR-E$D@mv!o%2b*!OsXP{C~&D}>LXM`Miiy|zYU9IB-UJ)rMK%Yar#Q%6s@WHn%P(cjRw5aD_ zY0Nka`X zYuo~iF8Bc}xCRpiW$o;61wb4m(6AMC>D1%}v9TciryNW@$ir`mGFq(Sb+!T1a1d7| zKMRJeg3Qq@s8R43xF5?APhuOdu&MP-_Bq9O%9@^*(<%e0QPwqA?ng!57eA{3sN-rq ztu{;HrY#$dc3cEN;Ff=w8~@ zh35_4=rMl^JBqMrso4bGyl!b_Wk5lfgB_j%LEvlt5eoMcR=fKf;ad-EL0z2?Ynpy# zrH2d6TtZqK5^ak(EOX#`x@GDJ*!mGk1Zgm4gW?24MNQ3=hQ5`xCc~8I;uuc+& z014)8qo#CcFGZsKRRDhfg^yVqV_C~Rv zjL1U4enmiYxx zppEzzRVdnj_yF03u(8`p#n^jYnN>SHVSKCRJTzh`g_k}~;g46vBiTZi>^S7hp__D{K<2{+G zWVn?Z#qwrByA+!cX#rMtJ}@7hV+b!{Wx9mBY^Tb)8@1{VyZT0Pr0GU)TMBy3gxL&2 z@|gS4GY;OHpJQcpQ)gXOBrV1tY`vsGdm<_3tVx451XVU_$b} z6T4YlkTCh@X-T~7sQjO=FEHY-3s5grU!0ytT~Q}I5kDkh#zYHrh6i&sBv*}4ylYco zAZ9saP=XAh-{EC)KGeq=MVzzwyGMoE#`Jt!f)BP)&3QFzRqt zE;d?rV^19}xaSX!m0=N~BLn;N_iE5{M(f5#?Z>Na;f2M_dsY=Hx+Gi3#@RNXUk8fW zo^@s(9S}Kq|2*fLibF6MWZ=J7YH3ky;0n2h97uM_!d5n6!;%FD@8>fUL*64*{-Dpk z@H_4JW9MTppX{0h_X7SMER))GWd?=*Ub&UXhAmi5g*;UAKLr*+;dyEp1=x+f;6RInV$|@_9om>ZMs{6f_T0Pmn zn9{!p_eZSdF@?3Z)+6DnQp2Xnn)I1R{TE@%Maf{!W`Ea%jM`rYWirfIl^2q-WeU?k z`t_ysHNewOH`%Oy!`yp@Nba{G`~{{sD)plzf59%Yw}~j%jQRBj<+VXzH{XFAd!_sF z_@dqAVqKLS5z+8wJYEVBRWrfBjI$Rm%Z%gz7L`-?y|}1brdp@OlW6)5%R>nN!p=a; z%{dz85r#y^Iu~Nu7c(9MyaN}&7q`09EVAu7s{uOC72<<*-AF2=VkL&vM)i`bJt&}^ zVcApIsQ9wJNj^`e)3k^=qf*K8J<0X&AZ(gfA;QBDCV8I~?S4Wfzyc94#baBPH#It4vCSU{z-vmGa#%g(*qg zn+z!c>ADn#JFkeM*_snvO+cavV^B{-^78ISzSs-&ktQF~@5Q^AomDrHDQ`)dH4QEe z9dH6sh9f|T+EX9~#_t{!6VqTN7)XojA%DzLv0AB6p(`)ZX3>IbtCIL`{T=fX$xHSI zvJ%Bad|@xO^BCd8L`Ksi=!iybQ}l-(g`z$#c#W~UNMGWtQv$(hfc7eK>+df6AFj{R z2OmdEa07e??kuFjRGH3Mn6MB=*M##9txseVZ#)DW^4q%*ZP_xBpBkNzn9&Kw?bk?{ zsmW!{%qqknQ|l%!f=z`K*r$*&ZlUDNV3yr(g$d<_;9=kS%)$U_IC*^s)UK^kqrXMf zVm@JjROTDy=ka!G>WEi0Ftvi$$M29j-5*g`nk3$l1X&B`%Cg96M$%XL3fC>yt zp^vytd5q{*84{xNP^rWMQ4~a|GXLD1_hUtV(dO{3TRs~MMXgRRN!M7274Cj`g&~(U7rV{jsc{K+g`&VW@e>ljnJv zCqyS&6NyiommS^7r{IVc+`9wCdX1rIRaTr8Mk?si5x5TUt&55`88{{)3oEQDk^6IM zq`Z_lmelpY_Y;Wrrf{khcc#5oPGBPBF$V*jTa>pbP(K_I(v+c`sx(MY6%zQHswZY* z4AVxvxv|2OXuz&qqo4|0pn$6EX*#gk?sfB2`>mtFz$oczHHr5Mlj^QaEtfGfOYIey zEbtZiZ`j{IMH+&Q(0F(Q40u}U4SCw~PVrMe2XgYwkhbu2OxW;1sSa)gdw(FgS~#QHU53wca05_XRY`4L+-5>AOu0C zk9`)*KwQe#Hxonu3yY#3Q<{1@(`#kZw>LVaUz9N7S9-rWbAd%+SGakKS*0xOXQF)J zf%Y2^QcAjegDrDVs@YOf_0N>=v7u--+*k=Bcx}CKZp`*OgE!#Al*T~cA>l+cscCqI z6G_ssP*}lX*^%nPKuj~-KZCMiMqY0!$0Exsuv>8}K!&sAwIAop`r@~504(B^j&o`p zDZ1vh4j^CT?~_$jQz&K{s4Q`{xHc6C{yab1UEZj9 zHZa7E!Evm;f%!RuM=D|VMzHT9bD87N#fRn(b;|s+1lmINw($#YX}2v#f1&ap)zAN_ zM*Q=UFznk*!-LCrB@)^a8JH)h7DCUjUHPN6;|@&%V^Ns+Dk19$qX%x`ac~({NdCP9 zp!iWZbe@5|53OFPEv%I_nqYRnR}=K)c`{9r=jU3u z{T^IvYYZS*={1CmX-(CYX{>2`f%0KWg_WU=ufc>p!=-#DopOup)x_U+Nv-l61Fj*1 zXh-3_6Ttc#Li2C2uv9m0H3ZXo1(?*)c-#IY+5mtCoB_{p)Rj`*gvLBQ*q#aNlT7woW~cKp5+2SNF-0R;v-+UA5Pa^Ly8Wo15IBVUyoH7p`zlyGD`dK4W1JZ#A<1i3uAyL zS`$fPg5}w6P$8iu_*obR$JAmdK{Tptp8g7wy4i4Zn2J$KJ^s*@wB!Lx_799&uo^1a zUB53IGNrG!zuTmICmWiSRoI1V(CA2foT@d`V^&^c#E7)j5K_9hiBDM)mB}doK%4E( zA8{1>w!A?~etp%EC6%IPs&>X%oEmmK=7)>mdYYck&`E<91X zS`&h4M}52q%;}E#y#8QhY#2JQ89p^=@1Zy28DDavAM=M7u?%>(ahW^y4ejl1(%P(R9d#`_LMOIcE(~1e7Ly+J@VgZK+uuJce2u zKrlng$33~7xH<~zEBO-X@wk`fwHtj`0F|AHw=D~tDPR4QKZmj3$lO!BU>08gN$dPS z!Kc?y^jo*(YI9TbNw`Evob_`Jlk#b~HM&r)d)&yQUb=QpnIo4z*~$3)&4cgJOZDAFwii&}-oT}+5bS`gTHF})$;!7G_M|!4id3OVFj=bjKMzEl$ zQQ6GUOQMjF(jIiwoU)Jtqc=llm(p%?n&Ru+%5q5=zDt$si8K~L$5&dFYS?N08_glGO+oVfd}Fo-C^XwaYRf$S(r#WH@BeN${3{5w z8bPAcbHZTcyHyHCks4jF+G@>|pU$fbnt6P$rHr&7pL6t(Ai}th>i4mh_XM~xsYb~u zx2gEROY9sXG4uf2AR>YT5aOxM^Ry-9z*+e_(*mDF%cQ#aKB78xT$M{Pz8$!AW(nM@ zoI1y7AvrKD-_>bxTekC69`G%2`QD^sY`QjJv1=H6RkE#SKjTT+lTB~jwNw139h3W{ zhHe6M2sCsQ&qF;TL8V^{eG96Ep-je@Hv^Df3oZM4fCR= zXXu(gfE}X|!5u;KzE`~Tx6aM~*)~zQKq&{&^@??E+bAb^x2k!KgAj6JrWea#lJGaa zbTg75IwNwKu@aGA0dX+Eer)W4riD;>h2;Q#H~XHc_RME*v$Je`xfu7Qcl6h)+1hR7 z#6V>uO<)8=+3m8jI=Et9;Sf|94sMdWdvaRN>vt&+_vu7fuyzJlP>?ByHI4QoOK+FS zVbz6M#rE>aF*qt(J#4H;_?&JpJDG)H%?6i4+l`zUoDZ45m~7X)JF}`zkE=f^(4{l7 zxETaL;*(Khj9w&UO~41BU-2yY4U#3=Bs&jP^7BG8oL6wyF(&GcjWHf$-O@+am zYZ%uW-v4fX!{OLofvrer$vu-Imsl5*DlRh@vI7SW!6|{}P~M7e#8AVgNo~^Az7&)a z{{Npa0A3wBOpj3p8|$u=m~rd?y0x%E@6F_Kgt(DL$Z-(B4z23OOFMrIl}W^XN}R1+2fG=S0mD11Olbp#N)V=uZN<53NHKc1hbem=7P5W7FA@-5^Scq z?Zc(-K7B(-eJ89}NX%2?srU#>@L`^UAIFmmPeLK@4%l z{^*bWyJgVR-ojtxW4M{EC$E_)^ny~o+FBOowZ`wj9X=>;qx_~|8!u^KMN<|w+sYa* zeRb-FyoVLw;r{XQsMGzHypIzNKZlXOX~37Wo61l{#6UV0cNF%w7}#xXJ!MHVW)Zzu zPL>GWPYf)@6(*$2E8{aCw85WTwX)odZyz-Ix;%}vywLDDh2*dN)#=QL_BTgge{8(h z;YGynLVQfZ$z1biV5Q_wm&xpPXz#ON&XLe!anK^Z_z}jb0;O8RLqNkvx8_XwW$+ww zmkox1=_ky6@>yAFfA;;Ae|GC;gu?O-fg`Db;B%h7#ES@Q7VJ)2mBRvuuHvXLSpzYk z-@|3k+pF(~)a_9w$_kkVil+I0-l_f-4b^2K?QJRU>f+SDK69)xMVZ2K!qgG!zf+N| zPU5PbwU1ZS|BTK+Avz!YNrjDw$+!48J8@7{!_;m`8JKscmV}0U+vgud0MFBRtjd`t zf=a$W{v9N_pd|QV>b*n&3asD|^Z{-;B*{+!ELuf0q{w%;IlRGeR~tN!;yb0^)gKJ3xmef3(~-5)V( zKKD{jo&Z{*R;SltC>5m8M(HWaZ|ZrTR};u)8vMurTf;<%7x7_C{boxEp$Z3T#BcvU=R)kI`f_95sSQ{1Eu1W zPtgFB;N^Y#)g?Lq2S@WX3~iSXQtK&X@7Po^#4_w-^eNIp@syUeY8ZGwo>4ue`V@N2 z$iZXlrEABKAsdu8V;UDBTukgRqM>XcMnTv(c7ABgGd9h6d#I20bK80WUovDJS|97j z``K`l4J67oBX@a13n=N84s^y9moZla_^G$DL-sF5KP4d$9B-)JeVsa0?)2CPCwpMX zIo|0!6oBM?Cg#D}WcY~sl93=~Nb%a$X?EJi%^W+mXR>EZ(MHBc&+p)KuCQeG;o<$Q31Ol z0?EZ29fz@Z-fZ67$eWMwRmrB}6wsDxTr=mM_Kl9086)5-u$H$rQH2Wov>OTAY7ArX*hhlB2t!GgjUfNT&lY!I3JK?+>Nu0x|v1O0L%mPWeEK|hR<-a!nQdhp5D?Qa1Qnn ziDa^&uf@c?*$B>!xO9h@ER)OU^L0>4@a;Ww4+tAcHXp8I5xxhd4&hN=+k+>$F~CE}U!|h$yqRmpt?Rnu_{w89YG`5{H24ggk5u4s&6F<_y7{~8 zRE6y{yU)6<)h+lv5d}VeHFv^q3=w>CVFmwT5Uo6W1whkC`9*IHkF?Kb!h*-t@sW|^ za<|8wp`gUajTlMBF0`u?2Hxaf+{V;gY51Bwx((#Em(SV@0nfuCiOe&w*3@E(n4qFS+`KG5DnHjmj z)Lzod0x7^h6AAfdU6z~?)g`+#_f!F(;%FYAC973JLJE1X zac;VQ7ZW7Hgq5s}gCf12oU&N|a70?1P;RjBTHabF{Qh_jyhR? zUUx$u(6W#J8n*yULPRPYk!sfHfMW5s2(h47xtmVK z@1V8R-fQi^(IeYxa-A<{RvWI`se0<)a4I|9=1+(a&|QUQ;#PKaex6)|TZVNJ71R)) zNn14%UiOM){%$UET^fR8H9Jmg-tMZZWtOTaV0$zAb*e{f-vm{M-J5jUIQ);i9H78b^DBIJqPaQr1#=SQ-Knj1?#e4DbGI&tDg7dhh?PMy#NBWNq z{MQ3IArL*k?#FW~-wo8C?i*P%v$xnR9v0??-Eojq#G;l+h1t@*E3Fjf3k2WULq*nX zsd|5ih;Tfj!0yVv=E80#PjQTloFTuxnO`yITxcJb-K0G?P_?eUYp>tCgr3Hv#Pz&x zhFk#L-YQKC<_s|FUXl!3afD1dIks(-TK8$#OsD6)1s2!i{4ZkfM1Pw!QYMQ+inbcE z>1D$h6t(TPN=q{h!vqV13_E)05L-adk@i=rJIshoU`BrIa@bGY`oocBRjIC!oR(<9 z`fHazFm%GsOI4BSLIw;SZlY_Ib@Aw)YMSq$olJaXFjUvEc1kecjdXk4$iX=+aG;Fp zwO(sH)lSn;mX-FxhJ6?)t*V4otU)wkJk1VtG{~$w2C@lYul?WS#sf5+tHkG=qpqzC z(0H8sdT9U{G*@f3u+OEBGG!t+$}GE+l8@zj=2HugJal?&LoErirP(7QyVwVbx)Vg1 z4H}+@6Id&>fxlePK5mP*TCvxYFBp#afPEZuqJOg9lobcr&(z}7@2k+hb%gdUTp#THo^l_?BT^8@r?PItKX#&sy(fQcISHN7+Hb8t zk5bibVI`Z$SC1SqDJ%i(okuvLS#Z2vrfk}R!Na5N_^|h(~vzmo3PiF<_|9AP-nnQ|w?nb{8wVQvN zY%PGyAFkFLKtpfcc{}j>T{zK( zk%FsHC3sm*;62@$einBW(o|MzM9 z?-hyxu%_!RCapu@>(y)VISG=3$LDt5z0O-4EUetDypNWaN?5XvU9fDzZt^q(w=-#E zSb1`5>uS%{L%YZJwvCy|dstPm#H&|t|C$Cg1NG!8DO^;6|EM+*Rnw-Eku1rsY2P&cjya@It>8rPOHF_J(=|9KM{m~F_#0^y}r;Uwa99IKfqz!;Fat72ZGd1}> zJgE*%5E*|to8Ul8svg5W#5QK~@2D^MzE0M~f7n9d^IA5gs|}P!xjY9_n!W_BmWP>u zVtO=r^3PdLmreel(ahHUX($o5aNZ>UL|*`u6aqL~~XF$$~}N;*y=g!3%E27Vs7h&E35 z$oYnUyBq%R3@()%@g<1mHteLxNimWMxN>b!RpqRldPq>P9*(KYXTFwXw^oY(>qK>H z)@#yp%zR&(d@OyGoRpDUhxDZ`;bfAa3L|c~Jfb zduV`)g57eT&26YysHXShU%zz%KAFNTArcJJA3AvyVD;q&O zPY%>ih#7BuJWCv;zcbALn5}>tP-F4TMbwu+`w$*}+L|GZQm2*rP&dGX-H64yS?okXU>Y0e~dVP%V#k@9zC?bFS8Z{cJ!R8Dd+$ef5-q(?%i}C7$45A5fW}J^VPnUZMZBZ4f&Z zCazylwp^SbBPQ+@!A<-7f|Tz|S#2`1bUS}d&Nf3TLcAk#KpW96_HF0W2FQ%>S5~`l z75~iUG`RRq&D%ha&-I$s62F_it!D$HR1uTruLH@h^k~7~Pr;*qHWG_sF0Dwf2Opmm z&i{Kcz$AS0O_4h80QUQedrs8cIBJYbn4sf->wMn&l6&7h;HomuL&t{%oyj_RAKK}9 zy}1+5?Vhv%h_mPTyyk5Cezk;M1XXiYg+o&SEOClKLm|XN+r>B8YpTpRcf~84+qi>s z?LW0^HjcRbstL|+wN)ArCjg%qULfg{p83t=T6#L9`X9dIf7{Le&q56c2YRQsJKyc_ z^4R#fI>%JP7HQk!+#VYOF_Q)!5r5QJ4W{AVi*sSs5_8`X>M5Q%;6!;1892MPYV^m+ zCHvBWhYtps+$9onkP}G4xR6r1Ks}BjA}`Y1stK#d&^+2SI+}TqIN(QsnSs}JsiA{= z>C}HuU5F6jErgjdWkkqa2hCN;#qfa0fP0vxx-wbaTTe zoT$+zOO$Lt0`3C|BcEX+f6Uy`*6f$0Sj6M!t%52OF=a-ol&SCvsG#`=*vI42r1HW? zYgZu*MZl6cP|H>+IBk;s_HJv+9}O?XuIdu$Iw;Sekh6$B8WWUfsYVa@fskEBQChLo zzgC_Kn!QFWaroc4&cEj*AVm6`!yIt8zftef_N>((Cqb6F!e0xw2`Gw_Cy#{+^h987 zVx1p&&D+gFEdv!C=am*WV{H*f?ge5c>6CZ9wC8u=*2kB4F+R=wJlW*xQ)wY%R9vb- z8azy0fg1<+&6#KOBNogsN^F`g?y#Majnz3Xpw0uNdSB9cOYd0Mr z!d;I6SMmNksqq;J>2?ux=ZrB5?)Rt0Oeo>^yKSqTb&StHdm9*s<2{A`_T~Ba_m0Q`DQfR(6%ORiw|?RIRsK1!B4QTNZj1gfX|SaFnRc!Q3Ffq-4g>0 zNae?aSy3NQ=<(`>wXFn_Rg-jp^sQ%Fu_-xOgp6#2G&vaX{L9^Z8FyJnDpOi(f}yJf zkrlgWF_s%WSfJvU|KiS!;gY8;I}SyLkkLKlbKr;0K<#{c=;kY@k7YI)o7Av7)3Z9* zA53SLru&UIF?Y7qObdx=Nk%g45_R4`4zKN65ZA|8H&_SvFB2@4ST~jrqEi@*pRO&i z*Vyz_Y(hnCglP*`mKA@vc2@s9K`1pK-;FcVTHJJ6mJdBo6S?`dY#ra^={ZOgW9qaK z6x7gLKE6}HzBEZjCJ4=e+z|ca{)3<@@Hx3ip#8p3k!+iIUIyBL|%|$aZ`1DbgX$tB5r*!KS2uusw$PYyeKkK zd8$?9;|R~f8G5KL1$twn1tSE(S4oHNSs#ziRaX8OldHgI%I>YqNOrifOr(~5zG_tp zQK$tMl&E>1<5|r3E>mHIz?p%@NK%g;*Yw10bs{mEKtcG*#tTmz=?PeM9$1#gxw%MQQ2*IetppReVh~7P(g8iO-#Ar%&Do>(9S2RS(FAL@^?IO;CERxfjoSc~H+{ z-3DoHDkVW&K!MviH;*?z_UDptqXv!=%ns{7x_Ky`*sna=JWEFcu!6+zS(Bj+R<8BT z26_cYd|S4Hpazq25qijMwql>Mw|BI+&>%rM(Q1~jDdCUpKe!%!eJo>L4qeE~M+vQw zQLZ84I`!P1Kid5{Fq1$!zz)l@4MN|<`Mn?QKl0Ei7iecng@NsR89E5?d@HJJd`^yP zv=h^_Fq_}=NwVO4>^{yKwo`uD?3Q+yN!p40J@9+6=~EbK?@sV`ux7AR@SBr3P5Pj;6IQ@@~xWl6yE2)OeLiZRHC%C z>A-^eIJQ%=z2bxRry)vnT#>0bsCEGw`Es>wL@5ki4 z@%VrfU*G=P>f6&*w*zn%#~Os}b|?3iA6ayu${dKmj_jn^D z>sj{z6`YCB;MF~Fv+pRR{km;4J-B<0^Yb+QvZ4rk6Vp$ED)Vu)zz@Sfj-}G=xhg*N zY9^3$VTBuHnZpr|-%Pjiqqrs_b%UBFk@fNX`o_bu~iTM140VEMSnFG&l?s z0|$7o2N`El3Hl%MWoiZBpVn$mF*$(`Q%Ogj9=!12}iXPkQ)8{MK5EN(8Lgm5(jhzZoAq#(fRS1bm{hYCl>QHNm zQ;!w!$1eF_ldz(ReT7nIzE+B-Zh7# ziP`X05Y>#xpD2MEm~6arKN@wRMm;J;+tKzU^!BXT*YA%dhia6+W+zomNRBn=Y4gtD zxRpOVg}{bheI=lsv>2w8nZ|Nnr+EiZf(`hNN={#QZ6DTO<}3BfX?yEBt|PdQh`12t zBG~2eY?&)ty)ir^2h-w{Hl~RX)dE5FD6Kh`<$ZE7E~x|5!NpB(wW+R2gA)Cl&Zpr| z%XrOy!LML!JF>tDX}lV)+r@yK$mpoQFgK|h++v+YgOFuBFml_0Tk$OKJA+#AaA(M; zX3EC~W+Fu9LF!lWjc8uzDA!tt^^n`P?$)(om2+y#*dDh6C#9zdTx5vF92f3dO%+e! z2?K*_QmH4j9RxG9Z{{&(anESDmY{B9z+S&dHzg5aW1C>1?7KW$JThH)Rk}A%**_S0 z^u;|WFQWC;lOrHJv^%wTO$5VQ#nSlEL!BL& zxlqIm_Tgg>_rT4rVKA^WBO+zxQoK&xWVKw_=JnoAkQvACwSp{FO|ou>+#dk@){4Dr z2t4oc?2@E_5cP^TLHL?QtF1W;r5-0|qDid2`tfkooJY0-5`9OU>t)`Wo-gI|jEfbA zHl6acqo+v^I0%*&`?q&6ACsCbQK#zpi0KJ)KXg1_V|I$$)L(0yt`Jn^C#h-p(-f+) z$D_`VQqMr;pVJ()MmO47XsQM)O=PLfo_3!zak4);|Us!1gg9CP16A z50JCkbW25!Sj3@Pjsx8Tn4gN_nR2){fSk$bC)!_L>XKF^bC>bo;@e6T!;3rvge#l zmkk!6z8sFfD)+=R_ZtlYtpvh6bSp17RznW-6JjP@Rg!FJ>B)|YH_FJq21lifftn=w z&K}sX&j-;o&uLT{NKO~6Qf)`%&-pfOEO@9~YkXX-I!E&dRx80BuZ2a&mQk9d6kRMe z037<%r@NH7ROOO^g6=yu&uqD-fo!`Tf^8%*+rOl>3exb{A^;49T0HhgaY!`QOR#Ms zS+`(0RTu*3tk{NzfPWP}=CHHP@%GYdndpx;=xSLfl0kjFe9RABI)E&>e*+l0R(IKB z5Mz^YqU0xWN}RtFy>6BS@pwI$IqokIu347@KDL#}AkZt(_V@DDI>}kLX6S)AenS++ zKH;b=M?kD)JPa)b{@dRDf~eLGs<9e0#Ur@IK*5v!IaYl$Him*itrC!e!YJ!a^K|(+ zg2PE6kzd|Z$jA5M2G47jz@}1GA?I^Dl~$`~mf6#-)1)mSgXsNaXm2FM;S{)9n>8|$ zY{jyj&~a16Lz+*GZ@rFPfl0qM zzZSCLn*;1e1}!g2`oChS*~d^LSk*zrS8q1x*#5?Ot%2L%nrP5wFFd~bYF2rAVReR# zQCdLhvD#6ddqJ*z1e2b$${W)X0-I`wL&U&v_Hp$ndiD;tWJ*(Na@Lj|PhndgP{~%t z%g)HPC2+@cmg_x%gL<7j{l0Ynbs{}$Yzz5eBx2r;h4i>wwza!P*aufwaej&>t_}l5eYJId`B=Njd8dU7 zTPP@?B#oq4%$QP;#VM60NS920$)F2r5#R{XtuT5Tclu8B z(y6}}F<_!!Ucf`}ttereekiJ~Lq5hwMJRH&N=ivVEUyc!)`Eg<8_^(ze(gA^;_N$# zju%c2O8&lo(W>41Kp~Uj^h6FyP}G9Tzlz?FURk2YVh!<)=NYn51M+KXvP)x==#Cpb zndm)POl=sXSme*8B9q$c9OjJX`SPcEzAPArcg-%!-M{fy+mc^oict6cMz1sYeSlL<^1w2FeOVa&5yc$Za- zXo`|y|7jI@*WTYd2~tU`$$Q_YVnIubjdPp4?D`K*Z1L{c-svz>s>WoYg)N|pbIyb? zPaI&J;J`e8^2oHke=xTgV@S||9ENykSo7=R=hK<*KkU@B=A!_A<|%@wXStAJma$?9 zh-e>ZIg57F=#AkG^9G@&6qjsh*32#yMQ2(w z#Hb###B+iKB48$?C@02f+mafO@1Uk(=ft($_>kdjoWwkGe(LzyA2!B<)|+Kx@G??0 ztm9SWcK-0aYEma26V0J!F=fkCNJgWziqBFmgpqKIxWB;Fr{!D5SNnFTyzPZ`x!|v$ zAU!;wrpOqEKW&y0WD(;3JX&w2t9VOW_k2gq6&GmR5S+;(5%{s((le1R!PzDp;=0g&(bMV*X>oKYgw#PUS8FYAPWn} z8Q8zBJOvys!&;J@Iqt%X9j2>2xjVVoKbcoA)@riYFZr5kfIHzP*AidB^LCqEO5}LvP>=u88mzd%jT0V~#p-M{TtL|kBr~Nj&PD4N_=khfo*Sgm8 ziR+^W7rZ?MX?9|u+>y)$jOsC=zH&)Vu<^K#+NIs@j+5I)S0Enp+peMrju{~%si|h2 z#EhOZ4lX*zg=6#*H&GuhxnSs0vNqR5jac1$uwp%ytu{~&R;A6HZ^VqUi-Y`q<%a!JepV@8TU_T=u1<4`XvVX4WP;vL0*Gq0%Np~}App?|`7s(e8 zVy+&#tZhT|kj|Gk-BW*#UES=bMy@t*y4#o4hRgEMwW+^`P@1MeN#HDR-2G|%g z;wE>%JJtzWwa2gMT$cGgSvqs|UJ{qt7$3X6JeB)5xcrCj9`igAdGMBVL^Y7Pb*_IS zZm03uMI*`=8!uL=cMLF-ji1>8fHVn> za^;J}#@(ypDm#03w6?#=G-h@yj>F*>o}@4m7xSF%CZm==ZY&(?+ z5CXfmA=!pP3mbBcH!Dbq$TD;cN4*bc6%rC-wHC1XZLV8Ym8%~Y)$qSEPlBo9NhX{Y z6Ba2v-nz>;4IGIYwJdZLwBqT}0vLi`Wypl|E+9hV16iSW)XEwxo^~Rk8_}v%4dp_- zbhmzz4Rs{eXr{qs=mpIRW#?j7icFNUKMV5rPjaN*kVt`PB)FslLl51gNh*x8qFUVN zv>oR{4NJPs9WenUtVS6p^9~xb&b{AF+wv}Iuyb@$JpmqKGNYk*=MfbW;#YkAaNO90 zT!UMH7Z?yABAu-XEXUo4D)ll?@5#C#Yv3>$?tC=o{kOEC+kn(JXDxmc9@t8NOQ1wL z8rb8!e_#xa^(}@7Ay5$x6qJQuMUIraMig-^wxtH<6Te!HXyz&#Xge?MZ^BY zNRW;PSMLtI*FI>VkxZlZkL~+qVT$kpb#L>Ys&R=L{t(z%%z7Fd)QB73{o6d@4|Xp5jNx zwllRI@y~(RrfS2Ij=Z7MBwoedke&IEpP*g7kQmWIhgh$hPtO~9Hf?7`3AQcgVb-PW zVaQ2k(^>t!U;~qoQSxVzs0~aKfNnpTL~beOKfNdf^Y;|#iHN;e;^>6)r}!x>1v}DH z7Bp`UE{9yaXP=6h1EFqUf-cv(xjah;v}&uW6lPQ*-vDd*2XcMSZ|EPi={V;-2U`=B zJI|BDOb+Ilu!(S_IS*?Kzd#w)V}mco32qZ9Qm z6M`BJ7(oDG2BGISmrRaV_rCKp=>ry?et@=%8P^h(w6)wZj77&8Y~Mo;o#4O0J~VMip8G^x;WyKV3wk0G%rf&xuftj_K-9 zF1MdknHI#r|Cq#WbmOQp=b3`+ zch8vq6Xf36sTvR#@q81?gC?z{h51fK>PD zbuZOC4jgSU$M)p>6x&}_&Ojqmrdmcq);F?0)ard3Z86l;05xto$t=+5;3oOoi$m~ZC+j{i>gx+Q#fvQ~C##flXaRdpCO^IhsNx#2 zIEt%?LqLeiiUl-=sS|^KEwj~g_aMnORZv7nzh-0v4TKP&Zs-#wE$SLZMU&xLD|O#q zKJ3puizmdRM9s(|x`tupBE;n(#V$|I{*io+qo>4G{|od4E+r|CSSvP0qmqMufchoq zV=^x~SiFFN9|b2sCQxigDUw150MVo)N4WTcBZf%akINqBCI9zWQk6D)G zr;tzm1)2SOW@pY*IdIR5C|jj z&=GWe96jA@>X!>ub#`kEBUQ$#)F#W6w7hw+zI}J^=-qfX(0*^GM$Y-k4a0J#&)##1 z`iPUb;c3bgr7=KDVYGux93tp;igRuHN#cAjLEs$Xai&kpYH1)3eH=bnAx$Oa?}MV0 zc_KU2iXEEa+5;4EaxIGqXiJZWne_88kQ~vNA2bsuY@uS7uWHJ=YR6&*>*-#d)jk{~ z9M;OmF*EW|Far6@IeZ=ODZ5zcX$`0Z;ojAS-<9ZAJlg~rqKrdbIdy;R(qWuqJpoTn zw-z52lQY+w=l&f1d_J1egsK^u>i>BJ1%PR13G`%se@ishV5G@L z+zgYQ#C1DN^!j~#`*eNy%fo_~({*biJ&vDj?J2o(Dn^zee|)EGtzeu@XMkzm#rvhV z_MsJ53ZKXJ3*v^A7mq_fgPmbV~{0kl74DndV3tR|kK@lT{9v&(l(fv*5c7 zc~432{#9ej0p7VuAg}tr38=(Nl0!^ z7wIm&$ycm;Wj`aHZ}PR~p@QfHYQ+!JYp+@A3m~p{_9B)zMe>nVmF=OB@%9Qc2Q|PfpI5voAWl`*6&4 zo3K_MSq_sxea8`$X~U9@vUlHy8l-IKR`|uW+ix1x$4%rGS&zxOp#hX9R<4?2OL=00 z`Rg8zGsK7n$_9wjrIKdzmyqRerv@09$0#GT;z2UZ#Gijrbj^{r+-QA%n5as2dFg>V zT=uZaJ~Y6apcQU)Y_4-HX{{bRJ}X%`>--QI8P)1{7OD*^wWn%I=@m82UG+W?v$PGV z%;kc%;rnfsF6#n-IruN;(UqtgQwsKoe|j2GXeX%F7w5G!uAX0cU9H6d5X!XYkyNVI zRjbN|!`j1zZy!dI01S~Q5&gLKQf45T#O{Lq>VTqnz2S2LvwmKw#Tk;2KDlbbmTkf1 zAb0chns9~eCx2t+g$>U-*dxCiHWW{s>aDV71lh$fvo!uK9Y)kN7dDE;j50C8agQMI zlrlC%S^7WKO@u%V=^jUZ-OD~ZTsCN$E1pe)Yrp&^#t?DfN?TPp?13d_pixdIZ0e~bhCEAX(72w`XzB8{J$^Yv zNA!NoG-|jZkwovvUT+tloS7yO2OcYzOR;Mgd2^~(QvIH7xy>-6B+E6KNCW+?K~DyM zse8oa{(j7ZljpYWZ61%+iRCV_Vk~4Zpko05CONa*>DNS>yw@>5kEGR6g5-foJ|SAc zUNBobuUEhC&Gw?Is}GdMllSBmKhOP;HOtRhtGgqia;$lUtJ^~)--C0z3sD-n;{7f<@B_brvStg)z@6nr*yBCcT1c}bW zJ*?euM+-+laTdGH40Ltce=sPa4!tO00#R+;>>6ZmH{E)pV|ldTU5_vaHQ$F)i&HJ# zng}>n6_0e!Tia6m47!QahtG6)o9zBzl|e}~Rhw19j5=BISZqGej(*{ZS+zW^yYzV{ zK1o&u(<^4_7(9fGOd?JRacm*VHy%?pxXz^yACS*ynL_lt((=3MOPIl)cXo~-fzK`< z%dr3up#IBV!vAvt=)|qX6V7|NLU(zgy`IqpfQ+BoMnJ5PV4WrUC>f|jel7_{y3Zj@ znKR8~HAz^XdrD{FL!G77k7KY!Z^c>%)v3$m?_9J*=ys_9jT(fxniQSny$0rdu)cY# zr9`Z|imkblD{)o$6P8cc^%+ig8*$=Adcz>&4PCa%qT?P|oECO^AH#na+x#FcpOtKB zDQmk#@xFy&JwxEettwJ+;f;q66OE_^iCwq%`-BS@sQ=xt#9L!4J#v!S8Ep){S%vCP zm>8ekxQ*bL#yN)*rK8!-QtW13RreK-dy-00x1|F#E>2I$d1;L~Co?D|hO6%8%Ag5{ z)K&c@SaWpt7U4~r2fr`TmKO7@aq+?$1z5pn=!&mcyY^)Cm5x4hb-PEws{p%UQLmlN zWnFqqG+28ckHuYkBfxg9U|+)_#iWF}tkPA9{i_yD@g(>=Q1jE7&pEM064elbBbNbp znEzVvw4nDiO>7+IDH32z<{+F;LuY28O_g8VnrmYl^9JH7Kxj`2wRq;{+4wlarh1%Q ztY_Bl-i=qs@s+Y;+4tNf*h932Ai3gFjJjGS4Y>hTGeY%T4bzrBXm*#lnC%?er**$x zRaOs+rkWE!%)esT?g{jF5t>Fy#c<$vII9MZ1Qo;IClee`5CEE|wQFU65#X5Q+?pT) z71`XnLTkMbvkt{^eUYoT?TF)s8G}R-U5pdF^^x zjT!^iYu&sz2$tfZXUr+FgXzBb!8W=SpgC7uPxSb^A^of zSo)ZWI&MLy7jZOiPrCFQ*O;ZvvD%$VRmk!yTk-`8dGt4-1qF67XGlJ8Ia;nwdQG<< zU-D-i-Lrn_)0ol$XfVB*mIeYE0a_>;Sf)D*RR^OfF)XOUHAkR*rE~E%bGyWK-)h}( zP3f&W>m~9Pah?r3hPD+EMgs=uUfkYKEG4U3@}W3Q1KWxeN>!|Alr86TI=p`tJBOkg z*@LgA_U|I7Q=GUEvAW4CNV%-TZoDQh8Sb_@r@QH; zjqH}7Y!^@?cRHB>kaWv!MU2gB02qZ}*+X+dr*BNBFp zd+tSX9IynILzjWoV+?;-eIPEU+La##0+MxmPz6V7PCSh!D=(m*RqR$W?e#TaGW77y9*d;7E2*--muQl!zg<>qxw?Xt!m8@3Y$Ni!af(-{7t-J~ zaFD!lFhYY<6m;*q;mr-sv+q*h_kOlkiA$cGvBbkWvA^+E009f115^$D)V9^HTnkW& zP%p635y3&~F#lLSxj$VlB)M7O{D%8fIjMn^M%LAnwML7>P1)#N=Ut7G(@jIAT&+7+ zSgf6@<^44(WV23PLxU2RgD{I@0tQx`11Bf0MsHkoPX+^*!qkD-* z4K3H2mipSoOGp6l5uv)DTrti_dPFB8QCM?Q)oNoNP+I9=S!+NF#~LgHNST|_fwK4` zq8hHiGmiToKl-)Yh6+u{cHo;p`T3HsJnRe%_QQKBs8;1dd7CeKI!p{+*5aAogiVIp zK8n8EhPVIB+;t>vV#4u3@ZIKSG7pgV-d{j^lOD&-S2c??aM~(at{46?BeN6$ZhHk* zap8*&$yI+}m*?-8Zc$Ad#`wDHaPSZLO!5;c>bO44@iM4qcy@n&w+$&N!t2X@ZZ=-moRDlo$EA3%Dn@d1BxQ)M3sk|)CdifQAzWzrG;HaxE3ZEx? zxmEomp|83}*QgyTs6Z<|>BrTF{f}?miWLo0v@!#DFAUOsRV2(d>ruu}if7ab=_*7c ze8v=N*M{V|!#&}YAp^cjQ2AqGR(^Op-ZA2qNyk2beHrsj`ila?-r99!b z+i^&66a4M?9_Iy)>tZMJiJ0t47AX`?2qRHc^%QqD)p|%P!b^GgFJAIAxn73q12wT= z$~?W>4wmLp&s13t!#)}{ApJstgGV2i!xoVy;`Blbk~0#5871#DNK%)a7A6nlJ!BiUk2vdzXD~oI$`wW*!w>fHU4mO`YJ$*`RE$bM4LKv%LoSl#lcnaIE@E_&@ zgzyg_k1AArR8_}ai557vE-x0;sTr-A)nQB}tW;(artMVd6bS0Y@M)JVP8%l-{FLTM zAYoi!2aDCdr%`A64w6R_T97f~Mf8|FA z*8g5skv1I0(y7GEz0-?qaX1N1Xi!gW?V`lJG&CW!zBHtaq&p*~Ow>rFVQM6?(FtzX zoi%DSF=6faFTNnJBnd&tztMx_h0cwN3{e__<%L!Mj0sj{5+w2cLu`Xi;FGYay6q1O zT5kHWMnwnyoW>GM9n46=LUlZ1#zuu+8mmq$S7$ypG+3;pd^Q@cP%S*?F@f4;q0 zZvtDVodTX0vPYGB7bj7%RvtB78Y{5xfWd zQ##IMjRrH{p6}r0*f?@pi#2RCNB9pLA%d=y{BLMAAefN*4GRawMZkVf*T-G+MI5z% zlG8IY!~JDk%LMcbCj@SCSfwPFAVU96m(YkQlE<;^Ivy0p-oD(n>)zDo{#LT60JL7< zO9>G6Qn4;()bySDgv96=dcJyT3RD=J-|+YOINip4MUO;F0{Colkm8g7L=gZO3-E$` zazXu3GJmo^mcsjX)mK?rSs!|EG@V~k`0JOG$|#rj+fz2T$C*o;!V2RQS3vJ&`|YGU zMJCi=A6`GUUo7HDV{+F^d8*})GZyq;s+-=ckyb$xrXvRE@BC!>`gduHQL_X!*PX8q zosHF%oz@&-6cb4ZUyO3UUZtT47&0juq-1^64IR~47DURE)|7C$Dai**&TF&yeSCPQ zCZjwf#_Y-vM!dgY`b^np#*=1R9ZqIpU|__J9A*A{mv$=?e~}l>pUwsH$LCvN_jNsyu#ryH1nzD$b2IF&cWT@+TvawdkxY2ufW1%(Pf|9?z znDa4V5X|S5lLIP+HukOad7{|mZT@@o=XktDT27SKm&&^PC0xyiC6LcgUf?IGHo?Ch zIpsrQ@aV>(Kx>)|f#OQf+2jrjVFFWpD~$Ae=%u_y@oQ;q&A2^J89BwbJc+jDd47q;Q67r1R8Tw=FGn+C^yKCiO{Qu4tg{M&La0L*IS=9& z=_|=A3a5?XTYj+an3&Klb$jcw^fnjIsC~2hkwjaj)u^+os>0>nxA=il^1XVONhD>+ z`lY4HGqOncTVZ=RC}EdUAL0&WBQccn_p1wGKfSp!x-_ZB4UfxxW$<$KHK`qM9UYp) zh<`m)Rjlvk<>#bpCrF4S`YYNXy)?c?C0aNn(&F6Q)^!x?hVSn2EWjD`dI+yttrj?% zN2mJR;^$Qe@7Yoz7q%0CBUDIo1D?1 zr4OKP%B&@Cp``G!^Q~*C(OT2f)6>P}p-Q_A6BE-8$C|Is@o;kcEo1YIbkdq-H3ZK) zG{NTy;B?8bV;|l~vHmE*uhV)@qM4 zda^pU`{sOgVcn^(HXMo^K~)XKxCmgB^j5T1PCSS|!|HB+AzW?;eUdh<)=x7jolf)~zvqOz?@doYol^d_$Oj;PpU=FX zhmT#K@6W&EkDBrTs?kZ5V~89a9J#q_fKW{o{wqN9x7Ek_<_GNNIQ}!A!H?s@lr50E zQL5@Sx811b*B)@ZHK$qW(1+ys`1p*B*V&v8lpjPMen%D{ARxFgWS-Yo+kHdPI9Qmg z1aDzv*-y_?K2@{FDwP^RZuX(BP%%FPz~MQtqXEXEloNb?2=vV8FfOjH`9W$Pd5J8} ztBDu6)}qnaJ=uJ}%kibR8M&F6QLW*cLjGi4sfxgEtCYxRU2kJu1g@$%KpQBfIOK?%9tYM!hvDAVCd5&+|H=t0Mg`nLxYU|?Xg(%i{#&}p?B zKh7`!W$s^L_co>5*5`Hgd9``j27}8=L;&0=<(rNh&~RR!*@$0n2#VdON~fz3w@&JcI`yS$ZH} z?vwV+XUG^F<^!ZInyI%Q>@H;qh53*AA=Sz;Du;w09 z$!o2^7eY{%aEF)Oiz$z*(k!3%hwqyTX4V15JP(*cvl zPKLMU--QkyL5Z?J89D{g6r4Fq05XOHl<%nC&!D-iJP%8TYMb2ogt*>M+hCVrEJXN; zseiUQoh_uE-*db5Y6UBd{^($mx|UWXy4iLFV2<|n8ovywO3xP8lmGKXwW}4NWJl7| z)b(!NZwF92|2)c%*+K)!LC>FtND@=e2)GBd$Dnw#?aUQ(@Y%4Cx)F;@N+8HeFoBVv z{1PVSFzg>wy59Cttf60ztd?rF7c^~!-QBMEnEb~OfBlq1lY#uRZMA`K9n+(-3O}gwrH3j=W#Sm}eo#i|mB?*6m{wqI5@B)^zamyC%kWu06df_2nw5pqS=g`1hz_$&#WsC<@77P# zv>KIg70o7*jseE3?6OmlB7B4{Bs7OsKHw>QtsOl}XB8`?|{!kjXto1bXH!}3+D=MQ$k106-Br4*iE-t4km z*ZsU7yS%~GT4zQljoQ?U164zJc8a@U-CV?WwT~<9t+*J>z%Yl(t0naRx+}haC)W4} z5Y&|LxJYgW7xyD;&WqY{h*m{MzAEJ%pC7NqVth^iNk;oY{ct#4aE(mQRT}_Dv;8fdob2E0?f1<8Wutrv9~_q}gVpw<2jFejek8iuWXt>G#>i;3y6u94>wV*~^D*nv^DyW0IVX;r zVQckILedO$zFdD8jAg^;;_3VG(9|XD|L1^Hf)baN9fCX&eufJns@ntX^|biOf1X73 zYavBAM#DR%5>!Dn$KSf121q3Ho~%-Yg#Lu0KadeAgdRx7ACvX( zA3NW(JuX@Rwv}BX8OTUTAMcx=@29$-tux+h0_k0^ZC$V1@IIkba$Y(f=M7n$uE8GQ z+K2sN$aXK!ACFz|`(H=?cqUKryh@3G2{YU5^a3Q?PXJS*$7xZ#Z3B7vX<45a#iy?n zke%=6wnVSfzk#C2w5Vlhmm7!s|6=wbri$U;u$u1KPKQw(|7p4YKN~28_)kTkww0@k znBplt-UkYBRf2_uRb--OW^SmjS6p_`3paB5cLsR??umO z-rg>5uTO?+wGqJWOu8W~zrMa6+S=&(>_E=3E1{B22#P)0q*2Mv&Ms#&SO_RA__yP# zc;g_8Y&4A|H~{qlz6B_$4A`xm&U&5k*hYs3DybbwVEGRAHBgk#N|$LgRPmqkJk=Sv z=aJX@t&OB!=WK`KD&^C*&$a7yUG40<^kOZW#oWPwlyA`~74~~eYE^9?t%@}zY=GC+ zT8pFV((sQA96Ss_g=i~!hy?oz$lL-4h82ld8Z5t+aw=Zs{d+ncAp(>MQ(9PfU_c|< zmW6~wahV^%TpF^`r4}htWVcMYoYv)A@63SB_RE}g1|iWp;s3jo|5{f-xUf^uSI>^y z>go~kfkpQAl8g^c#}LVEW^)k1NsH+J6HGv}TihUp^Tz?%FS?&uuLr+u1(N<6sZPP) zfR1r0;HMwhHnMjUlG7t0v-1nbOqo@rJ)6--;)0fi{rT|h zD1fQ^Vqm~<0)XVV4qiC-3h=-I5*wxluqF-t(A1n=q~~Jz2`nE9PKL)kkX*G)8gbSW z!^3~lqfM{J_2G!^+p@B}MUKD2KY$+q+yekp;@iA#1gdVJmYeK}F~wvgDVmK9$6~ zhgu676Rrs(Lkx+BVDKBuDcbG)y$_P2f?g-+_13hP6h2$c+x&Y`Bay`F0BE)-o|hAS zTyEz-j>qEz*Yor3SaI_V8g8R=XS)5g8&2;14dj76NpSTv2h`A09R+<76)6FtX{5A* z7*wSIprrv)F?B+G*;M}(9t17jMy31(@5EwX9fRb=)d0MoNz=r2L@mpNIb9%~#KM0=Qha`Y5#ZuubbO83P8j%T~KL zWZrl73jFW9a+#`B9HTjt8+pg6c6iwOI*1_;vQf3|jdO1{fbG&dL5O-bcjy_J4gN zoWn`=?|aWVxWkg+y)c?NOD(h~KY_!oJ?3P>&Aj58s?7`sP}ZPYF<0q*G!>L#c)!27 z{ob-$JlIC#-)@{3g#^H&{?{icKj$s*+_wC3-uB3L%etNqG7=IJhP3}u&-vGO_$%uD zx6PO8|ApE>?GF2n@oNVK0S`ZY`1Vc#X_p`xB#*EwGb^ae^9squlniJqpuHVJ>?TvUK-JRfef3 znJQ(uR!ahn{ZP%N0A@2uO^NUtJH{L{k1_7+OmsZhf_aC|=Hlh}Secb|nLY*qLC~(! z(_C{d>a>slRH9uW@l~Mw?-5U8Uvg_?Fx>Dd7QB$~mA?a4fPkszjT{T>UUSu>hpwR2 z8{AtXEoc6bx@~i(i~f!1XGxSK=X3#EGN+M9B3NOpzCtZ=FHA_oHr;qBGiK%Da5-$B z_WIAG7)in?Q|~UP5@*1pjpG8jjV)pSbQF4sk30MwCOT}(u(hvkQ1juLAd#$D5DUY@ z(i2=4$=%asSe)~%SgJKoR0**&CUP6oUh!EDBB@C1=6B*RuAE$E3nq?XE-^hO%&dtC z1G-9lM`O!dNT!W=m-~0O-W*-C*Qc05Iajn9kRJfG3$rSXhB83JZfs;(gDszTAERjs z%$b4w^Ns%Z_5Jr^D|7c++@>1;>wbw@0uX-zv&bD7Hm=WRjr5i>^;l4RP^4iVcK)(J z5udTMs}X;5WN}5mwjNbt1iEBl>aqZu;Il#rj+!6bB~G@XQ@cjx3}Ki^K0P`NH#fI9 zVD>g+_l)zq_{EcvOl8MQd9NVfk>6jXMF^-nc)(Qq=_wwBT`IY-u!p{|R0Z|7bRKp7 z&K{_rTNu+mzqN=QD$b1h>>@q!)Kl??!U*uN)?yEhq{b!a4J!v-OufSt<72e3vP$FR zsw8(nX%oTWk%&}>>43*Tg;|*6Je(lgxM+B%4g3Z{$Z*G30d}d&YzkwXnzFSO5WKj_ z-{kn4ugG&^!DSQu6rELtwrYoPJx)rCdxiBLaXgWscnklE#s7EXA20$9td#z`A|zdW zUi17*w1r-I)UTZ)i=D?6zz{n%jT zphIb6nZfqcUL#C}7!eZUD@7BVA11*JeeUckBvVtr6O8=8eZ(+~Fe;;{4#6vaa_|rU zIWoI^@uyiiz5n?Iwb*ZPc0-FQod4-sEN?2XoPjP*w~VOEN;hc`AOH_IoEHNR+us0o zj7O`RA1nrQl+5y;jqGGHsQ(yZLYaX<0zRgqUu9Xrh=C@#>THO{XzRv-^YhN8kzv$f z^idLs8_1X9!ngHT6vspBtOA60fX73JLq5TD8oxwWPKvl-H}4#}sezy_i?)#K;>|l4 zNt1WqIv5i$658|_gtqYq&s$>H`mG+_VmNU$tiVKpiwNdSfjfpQoAIN0$Z~qd?!+S^ zLX1uBpB(-5#ZZd@^T201stte($FDnrN+s6~aw zDi8$)1^A?^rvN>LnbFZvgkRPw;e`VKsR{nymGuJ??A}Gxl9@h*9nEkEd9(e^tafb# zqBDH^mEK1$Bw#xx$qgMzUkOF9jW-ng`|7J60Sa3MO%n3(3rrRNr+fjndQIFXIt#u7 zAr$;FmY}MUHyvEBQ0OTzT2ek7dn^K0Kc*w~WAFrn)6Tv9y5I%StsXwvw*vv;t)6&0 zYM7As86&aoT{~IMzNhiLcQ>)!xRJl8F#^Q+&h_!)+VyW0VVicmn0|eu?Ky_elw;YYDWv<$M?lWa0$vD zpUOnILkGUr)X1pcN~&McLpFrm9a%v=DfrFp=iO+Y1>nUuhlA4D8pOY8Rm2`Da2bNc zyWNHhSY~eTgGBKhaXC;v^SQ4DJJeH8Xr-mZBiVNMZ!TQ&tS~GYxSl@8ro?FF8hz^r zPVknFE+n4_P2<8BlYXCkr)ku&HV07N3x?4ccV6ShZx%-0&YP z{{P0mm^AIGWJk$G zwPyAJFn(twi@kh6&if=hovVUMcEX4RaC(=1c7R;%*7(tY2BJR5&AP@E@`GP}Q5H*%? zwFE>1?&FWYQP*0-MD;vDB+b)>go(UpJR}vDguANtwl!$r_LO;wu~0JCqp@M~-)NBq8P;CdNQ>Nnp zyWQ%MWiy&i78Dmx$%HBWtqT0t;a|fC8Ccw%c(i$05D4UJZvKi=6pd| z1qa8)AKPjtTy@_K&-|I_=JVx(H&VXxqRwNZ^qdReNW2Rxpjig!}*eoLsc0v$@938qxiZIDnRV`c%d zc-Bm;yQQN`5lLGOwQ43FFQ)f^0dE~n^mf`1_VoeoSHtq_S5d3Jcs9~DZIieNLad4v z6G;KjaApd!roHqDK!*@xXXl{ga1?$1g7f#{!ok_;$UY15k^K?oYXBBL zq%Wu~0+&#DL5XIam3T7EYHfRMy^GaY$WFy>#V&CMZ`e7T53T9B5o-u?PMG0yBF#vh zl{z>{IN`u6Fr9v-~ zB3@_d$Kyt8r4R0dyc|U^>)2aA4v)e`p9aa0l)>ww zooEumB=`{h37AI1+sDeoy=Vhx2lXX)d+oPAl2yOrb$H!}pX@BI1V6aOP?WV4Q9gCI z4mca@oIlTkyZ2(OS5P|m>?FTvVo|2Q4(gtSM-tcB6#pul%lQA;ddKKW|7BaaV_O~D z?ASItHafO#t`&8Rj%`~V+qP}nH~+KGdH25Oz2Da}#u}r3RW;`#ZcT-TSZW#%e--eP}1)iTBvP7QN+tk&7YtcTBUCI@` zNh29JXRwsh%=z^2f#u?qsyv~M;+DQ;>LW{@6wa1ma9(gIziWQ($S&P+;>%YJrcLB~ zZ0L3tm*nk!AK*#mCNS+5VHufmi4tT^oG@l-96=YUsMJ(Z)WnF2?Ch@gkWb2%xZj~a zMzE~y{WAXldJEq~3;@tf1NH!aT@C!cA?tX6brb`*58%~_NrDA$daNu6sd|ojsvh~# zY!)mb&!<=P=xoi=nmc%jFP-iN$)>!mCFf=kYPF;7QC6SM#(amif;6oDV`-h1&|1sc zlr=W2h)_%U{bArFSxe$$e?DqK64aI7oR>((ely5rhj-^(qkb#jM!xb<$}$d;DaE=q zCREr;m9vow#OTPet!vqY`B*R;JoxbGXR?LqusUwt?ywNYcqgS>$6G{|@BMnyudRRvyy8;I97bTIc&Ll)&ATnLyh^C6C$MTmcQorT245 z)rZkF58=klJUtecekl%ZCpzdK(xcuh#%d3z`@HDV-QA3=*R^dV-@WZg-@FrNt1KMngG11mFF@^@$F1TNCNXLiW4P#r6o zcGz#$N-ZuXA$N58Sa%y8r8oBKd&je+=Ib}h;`GFhNY1ZJP}4`ifENIR6w_AO?|Po^ z!AXRF-h+O**+J}n6}`U1$ir`K|s(_x?rk}b}J7g;?H!*bi5SsyAvpS>+KMe)BzeAb^dY)c- zzcp^YRA6WLI_zV0+-LYqev~>amTkcUa&lh_GDG=kn6~W@6Lx z6N8nxsA!1FvO)tE))>nG~wZ&C*VBp~bfScaJIorl@b0-i7&E==qw_ zgS~X_<=R0um3H7SKmcNvd4LP<@-D^^86$^ftr@2BF{|DL!`Go~qyGu?|2FUeEQmM8 zVOuTN_}j(X$d|vfADo`VeffIjCs*okluRoYM(Y^jM#RsTp8A;+ zWP4@f_K`hZkjq-G*c}&yEo)BX0E=8iAU~CpcjX9)x*g{-whS#~Q~gj&UQYec?i5$0 z8C^@06>nkJIMTf&{dcCr6rF@hOe%NS6xpWLic?Q7FPG?j%cHP6ufs)V<3sxq11y?65DqAkN6 z7bW-|`_BnIFBZMN5`rA1JSPoJWiyKFP6D*Mdc_Jq=aZ!}+QHFA^xH#%OJ#&G>a=U^ zn?+CK4CV|~9+T;z_MJw^0=6h}5QR9%PS*aae`r57D0mfBF}L#`*>$XFu@$^LW_HCQ zH{`Bw@?8y2a?zI3twD5r(0bH<*6&o_$5NVOlk&A;);AaoLUpVW9~7cNhRKQV z&Cro~K(!q$1q=zh-v^99R4x^xqFC_*n|b$ib@XlZuoCIy~} z4_1v*q`(b&A9oo)FcOeu+gM~CMT?oi_m>Abd}{yt9w$Cq2AMTx`CK#R1RJbmXr_{y znUE)#y|UYsmRGe^`ZVl|l7v~2cuXc+L)51FlPF3utD-h{U#rl3t-hm28u?-fLt?mt z)TxU47mSRKosN!7d8c4C&U6Lk`bs)G3AF$RPm=j(&&8YUQ@_6^1@L}0pcA=-u&iOG zHnag!KC$?bK=A}z!a<=|Jsll*Qi35%uLYulYZ-&j?}7?QVZRfk3livjN$x6V&C-O4 zn-|gxs2kJ7E}tK6!A6tms^OKd#tvcnapQ>!*w<}Z!6*XFs5O=2jvN+MdiPP39oBb~ zchQ!2l&INrJR-seUY=wv`SqD_dH!#7_Kzv;*NMBn_j3J-MCwEBm<~VejE@`@pO`oWU zs#(E(1jB_7Cfv%H>GNiPNzz$Y(#W~x8XDSb1#&R&1`q)(N6PW z+{Y|FmH0@!WbyhaqbwWX_afQU2a~aTte^GW-F|jQqG|CxYms|^CAEZ>7)nZE1~1kY z7%Y&o)a4%9Y$`pJJ*9RbqNBVKH^vE$|5r7Hol11X%I(atSc7Po1ra#Ak7ISN#0)sf z)-z?qpmCQ}ZOBIZvhD5hzDGgv*z95CKZ9 z{*0AtB1&oZp@V0Lx^0$!LGFM1Cnh>*$Y%NzL8jZ{z?X^BYH2ADmhp5yomkCC$J|86 z@KD^aM$+FpEdP`i_IluyhW_Qmli_$2ztC$tiUNe^gsG4v?mKsrhr5**>*$>MJkNEx zD~09PNqCVwa?JlVG4b%D<4YKUFG_INlIyXDsvw%+K=F(BEAKzZUjb2)>qb*R#z;?dGif#&-<8#8>9_y&+<5ZF-a5y@(lc~ea##= zIY%(r^w>YE(t^}Qt#UdLIJ2&R_vH7P497?e@WIauw>{I)f~A%H{!5tQG%1l>)G?<3 z`Q6k&;FISUwc`}y2WH3kiia0m5;uq5j52XmjwG2&v_(&wY4`o!scwPnwhY@#n{;B+G&cZs;c7X-hb_CquFRf>3ykq`F!5qkmhKjM>s++1i#mv`D< z7t;S>Q2#dPzX%mB>6g253>=-zH#5(~j8D&}1#s9Nj~pERUTqI&3stBy7mkRlTQtm? z#Jz!@YOg2xu8fZCg98N=7{0E*8QY#%ndxd{r*|6K>8=dTfwL+Lyxh=PB$VR!G9=1W zR2hF-j@@Ww{%m7c;lbMpxL~U1DI0m$;Nr&|{}W0CZ#x6~*T~ALy8kDgDq2kvD+LZh z#cLsP=Hz!nEO>mx0G!}n!Zg;tg#)OZV}~L6SwUF`Wljl7J+Zk1ALSB_vn^$TPoc%# z+Yjl3VF>S=(X1ysm&?w3Fv2Tv&{@1%HVFm0omo)sSW zs&iaTRP8N>CR0dq&)(h{<$1jvi7HO)3t@b^j;H$w`FwWxeydtxL2TAp{0OGZzgpC{ z@Ve3Hlu3n;4q`ya+c#zD$9M(LU;+J-W#Wg0%BCd(G-Vd*^4JaQV zUiasQb>w&Z@i46J6;XLB#|)uB0Grejh2oE-%_}@5PQs0u$q0fZkSTN%6=KKgc;oBPPIu|^q|jg_$Q0O(YsAQ|)5fNBNh^)D zp7XheDoQH1MZgub_gyG2@y?plxp%n+o+P!33Mxn|tUFGxfgJ&X6}MuJTx@{IPBen- z+bFHgq{2dvjgjG8rTl5fT8cDm~hS3h^x z-Q?;B$G!^86^;(qtfzIBcBmkN_)D!e;!bxIUg_P(F~7MiBvb8}eS(wHjTqC*56oE7ZeC*b@A|wA~)w(TB1GEta;D1UO|KosTDF2`x(rmca6zVVH zQvB^N*T{r()PcFik!jpn?VnaD*ons66dg$fam9q-^`cj=Ua2O!09Noa;1D7jHH0$} z$;Dq!&uSYw$lXk(tRRlP*x(g5obxP6D{d0-_)5rY7gra+DrpJ8(Jy5aMu1m?ou4!F zIEri+Rzv8>S+G9Vb&vw0n;;iFGAA<+t_DdB5eBS&QGZsFzIAFN5T{SkK4ggm=^PQ> zGiLC9I!@ivF3vuk$KX&zp`z-%_0Y#{3>eFh{$t~Q2%nMv`s5CSmr`#P&2#e=bK-sy zNDISvGZaSY%lyujI%-AI9z{Q0Zq5-OIG$wBDf|$#X25BYYM~KuJf& z6u_p<=v?@XMALBvr@Dbh^GDI2EZ*st`1r&8V`{q26A#mFr+H!%1m^#p&Hfes(4m(I zTOUBY(G?VTxfu*2x7kwCv19+6U9_PD_S8Lb+)2V`iw!SF5-0-m-H0)yPP;T@E;&7R zbNqNAq>x4Dl84R)&Y&q*NFB07-$VQLv|uUHLx06Q)!p$OF)_m1z`$fOFw)#goZoL1TkwF^dW?EEw1~K5 z-l1O{Yc_uH-5OWJ8UI1R#qoWbFDmr=LCVJ$u~?Pm>!Wr%PP(;Po%2|x?{Rys^NtBF zbhzz?&||MuXBaa_)@dCRUp+TX%n|z&&M~D(5O$if&D2SI={4}v!T^)E&V#n5@&B2F zdc?l8yhUSMGtvKsj2Gh27>L4Le||2+07Aa=rE%3KGKAI>1-A4eVp;}`1QiYhn;^P# z1#jmRtZ|!#UY&@LH&P6AN96Yi1&C4PMhVxGqVt2op8F$+;>4u_IfW4La3LJ4p_ruD z;tT=ehA6?HMJ27dBo9dCPFKPJ?vY}8P*dMrl;2UNObPzi%M`gx5KVZ_sjaY z0Y^g`QZ6teMhqz5wx(mR?!06pBe~1UiuAMQy>Fqv*>`;$thSuc2ZOGKlMieWq3`~@ z6|@y5BFT@0Hp06z=?cNwBP}`J##av3pcwwG%e+P$!WH>L>K<$tUj`6gxUST@tAj^r zgG_B_00q;|gCR#_9MeQIrSMI@O99eF7$_xRmzV=e^U7zM5);zfi zC}z?&S-Z%hzKb*ln|Xd z7c8V9M;C#?Q#RN*TR?+G-Z|^M8Oo}!g5~G%!3px;RVJ6YJ_l}&ft$p}*Z_(rGLFLk5y&JLbh&N-{7nvlenP0y&>=CSTI z`-%KzK^`WX$xly_ZEWVJJRqW$c$tx*Nj@~rBbW8in`Z|51+7$UZZMOVuR6VLFUZ?N zFi2b+R=?%UDfNL{DDpE7_R^j+CkRM0EY?;?*Cu%#q^l@riXi2hBO_@5EH+iiz*?7= zlcce2FMe{K6jgAB7QmnSe!QGtVm1Nqz%=VHd@!(Yf;brB2^EP3lh?Q4_?FaGtNCPA z1zC;T$R=AiHGDs_7b2+7xNErk|KP&kK{>wtgSJC9b8mNMu#Md13N6I{n-(8f2Y357 zyM*AN@04h$%?|0~Xtu{d0#Pyj6=*D+`LWr&FS2{LYlnqrb_BbPZmtvJx(UnSu;39M zI0KC>ykw=ORHg##mmo8c7d&rqKp!;P_KYMkoloTLvnlZs-NCFTw%5lWwgHJp+zVc} zaid#7*z&iK021nUFT7yV6H2rna&icA-*eBvRW)=^?psYFVb7OfNam7qS?74+xIJRt zxq#n_b#RNF#OOFi|8WogcljkeIl{6sRi{EH=Y|)bURQ7a!&KFPR9NaX#UgL6z9APm z^;yR{f|m2bw4wBARDa|irp&N`#6|7F?2*Np*t?=rPMnth1P&5GIlgg9ESJEU|SxILX4fd}jGYyi2YF_=_wi%%L z@9M+F5IN8)wHIGZBJnYfMiK4K{e;3C%sD_ zrek1)gOqWSM$3g~!EPwvAdo9bja%3eNvEaa0YjhEo;(R77P%WG5LizyEF6S#?%uV3 z0{;Qd!d7nHdPa*P+H`hRq?(s!S5LX!jq0d` zVKPuZ*1WAI3RGaNAEyl(wUjPb-TrPNKMXc5&ydqz!Vr=8Pfmu=3R ziBP`<%BblY8&eangc$C~cOHr8p3Gq7KMkKx#7Y}eGvz9goAU67G0#z2=X=-#gO-xr zRrOZDhKaL5xzBYjx5!1pc^gVXG&8pfwWhVGGY@!ybef7f`|q_-tQ>HlT0 z=JpwcG)QP-ne~hCqzjF6w{krpCZry3j0r-)is9V zzj;ufTgu&T%}F*LW(R!}*n|2X2>BKgL=!VY=8eBIjm{LFN>30=5CqJQD?T6e4@ZI( z_Vlm5`$ooH;=-u!91yK( zOw?^m_3x0Xiijz9A^UWE%wJ*!hxYRX%9UgW$7h{>83et))N6AEC0Ch?8`a#sIAbn; z$C*%Sy%r_=uEW$n1OzH^MjNHs**q*I&YS*4j?I8pvp%OWb6hbF1Jp@ zz+wuqo3r)xjoR%>aCXeC;Lyj3i4_@(1;uQ9+r5=RtcB^zG<}uhybGEyF0UHz`RfX5=Lq@WMX+WZ}5KF0pEo z`TruOUy?pQk>2HzUu%Ptxjk#_#8y!EI0jR zxwK!`bLPR7VG@&uYbmIH(PB2K=_t|g!0XkCXnybhnX-orZnLIUX@iadBm*ilQF1yy z8(~63h~gbkhz|Gi@K*MZqGueg^9wn5^$Z>}9t!>noD~$lgI4d7viXIjQ^=1QMuFN- zkX#lsZX9zH~tL_Kqv(PT?iqNqjOPz9Gma=tZR+l!bozYs|1Qpd%BPLH{g zOfl~)wZQVGT9K1qf1GD%sOq~2*+iEgNZPQ1OO`} z=FpwlYRbdnbyGY4xHS)#lvXW4s}+xOm_&?@HLM~u2{fFTqRH4gac^jTpLR~`i%*|0 z=CyrvRxr5#FZ}$kh}c{R&FfaWTQ+&|;bO*^JkjGw*r8PYmgK~p^u%XHBds>Cw%KTX z-T0DvLbAz*PjQ<5Z$}MCAvQQMd=0zCh_Frm3o#cn{yhoB%T36M2p3cJP}uwMdoSFH zY)cv`OxO?65UGtRCT30YBqcyK-8?p{aaEz>u5dL7pLWqtBjIAikZLk<%@yjSV|(_z znA*W`?kLORpa4bHq!OkIa~KIN+*G`cixBzUrZebZ?_5O{W4L$B4-_7`7aW+=%}! z#_+IHvm;_ukYe{+MW3GiTGI7OeY{#m&*V?&yf!ec5Vw*z?C6!aSn8;-RU9vO>A^Am zcWq?rglLKj^F5--^@49*ijmqr@6xfWpBbv^{n%N7dzJ<7c|PSLIjX@{SI=v5#l($$ zS)Nq>e|7=Tl|Xly{~`YmAXw=J-nDj&lNF9@0B@q%2!!LdX4hcKeWaT~7X3wRR9s4f z_XI93q2T81S(d~dA>+b{Tq_B;vR4fEBBMwiTuvK*6#Wwo64Wyul}f;xX5?J&a;oln z96VryrI)lFMR%HEigQQ4q8MSGwJRItp|bjjEJ9@1rB1}r0?DYasM^5Hi}+`xNI8brdYO})HfAcAR}<|(1jbX zJ?|a43DIvVo3wg&m?LGh>$YfU*ZK$mi(3O_m%LXxv9s9|^#@{-4z#^G?c$iw+~(9D z=0|&|tV5l?`+ZX5fG1(jT8ISW#$W{eJPX3%^}4*&MEmnku~V0$T>8N4{GBVHmxHpC z+6)iVQ+pP_%aMkYgu-U8otMn)iK=}Gk6C+N)PH!xA8JqmPH#&)E6Q?L9kCC$3L7eF z*vLUc-MrmbK#vx009*4@63+&UySIU}o7Ew$6%N5l_07b-%-WfZSD|3t3kN%B2;ox5 z<~avDO`X?$iKwX|YmUhTsdTYSz$5rtw^G*Gm!@+j-;LWiad3n{+zN|9WzGmSP+eUj z9HydSd&uaA?owE8fEKc~au#gaSPo`H<{Cq*55bLm1v+E!e(0WpAO{Y z?NsvRqgDfyI3*6~wS&ncQUVXW^Xf=Kp#)%lwGwkf#1abjUD))ra8pGYi;jTjk`kMK#a zdX7O2vbz-d3|0$^V8S*BlFfx}&O>-E;VxAhzG)EAK~g|#;hPh-C(aQ}a5OvHspaxy zQG87}xrKhL|6p_`mFJ3L%854H_Auu8?(?21+nkHEic$-O*=Y~sye%u{=k)AkN8@L9 zoT8qG&E=Em*>=;q*XqJ}Eo+O~krHLCyvR1i$1Xq6 z3z(kam|@1_THA25x7rHBL^-e->_xcvrE+Rfv}r6U=K|BveHQhnuiS!k|{vvc70SCWZ&WK)1os0>cbOX zc5A+2Iul#ik--dhLx0t?vnCg3XPpKg=-c!JLT63zoN`Q5c*X7EPs7hlcr(G*TwmhWK3L_{ zH=|2~H0AP`p+Yp+=)=jPhhia~SbiWpxm;GS<_QdUhV9WVJ|ff@U(PNyJ-u#c?MGh< z!}ylyC6~U=Kewk_(C9g<8H4X!qyN);W4 z*NkgPI+g8+kllc|F8gTGi2lY!qU{_{lg$q-*YgRQa1^Q^EMX#pQT@9P!KXhuF2l%w)SG`S`;4$%`SS?f19jhc?m7 zI3c;$JBmxw&8F;)gq+rO=+p8zSTI7k87t$(;j$S|+g)+iTIbfQY zv-oo*MT`G>k5RZbvW*3>7{E^#%3XQj8Mg2oT|ZScfp47Ne{G|ho}O-Vwf6Ct^yQ7J zG-xqRZ=y;6E>f;r@WglZ9Ap~W*qg->gv#UQOXRQV`uwb-^0O9@13e7WtuHLR8qm;V ztlKryAn`4Fhx}k@uaDYzuz~RUo*7h(4KZD%<5h+YXWpVc%vMSzTt5hY8QP$mA5oBd zH))Ygem)92|8UPmuMZQCIXIgF8qjc*%<&VW<#tUnb_tz4aSQ^sut#k_2An*P2<>AN zd?coRFp6B0jUYX~x&-jKha;gpUi(ujtYWvDYudcNwA50rcPHMaeQGB(lf->dw1@aq z<+Ms3mY_K?4_`)Yvb~Q%bBKJn)|6=5fFdG=ZFj|5IT|I2fi5kh$(v4L=ibiuW&BJTwt;*a=Wwf-* zJTtDx;_(-gtW6;|{m;ovA4DAdUP5VWpP?TMr`!#P)?TlVneIxe2~8U=JaxNVgiMWt z$j!(B`c2N>+h&X%8!P=`pznF}g}lkqkIoP8UfPqLo%Nk19S76rxEpm27!gjs#IDv^ z|2!%saM1upr#8=Paog|hXxBZa#)=xss)2rPRmJOdjh8KGHl6(Q?B5B1{?Y?GpRG=x z_U?uzKEF4qp^S%jV|3$=+=i-o%Hf}++Zhh>8jxj^%%IVIohIHFFA>58_#=4r8q6IU z^KtruvdIvNpm@XoNW&=Y@S^TPLL6~7TeC~p8KS-b<~v_)O0IoN^4*$uZc=X~ji zwb2|9v-F+Sx#AruoX}iB0N^xQW6l7(=K?AhyHtG+%Vp;1e z#7uDBz+w+9n7cn=LV^VsxxF#=rK|8=+-I98zz4(92uCp=Tt;{s2%k19=bJsTae zq+YF~Oas+AY}I}MCFZ74*b-GGwZR;uaB$YPVlv6}x68v+7?B&PLUvh{`y6B>7p z@fuEIa3HjZl&%CHMN%eo0Kq9<`oT>m<4 z%u7aplMRbSVfaZRJ}6jhKMZcDoKhPp%oU$N5@h*xbm4w`xCR&{Q28V(?wac(wRl}G z)SZW=-kcjNYk_kqg5Ci#{vn3-;4kB~JJmk2H11P_iM?k zdxcW?0A+M?J6~exeF^;s2^-LWa}Q7uYX4x((r_x<&Q@hy%!fiJAYJ`ebXP=M_7} z^)t?dS`Q!-9PMUfO`5?wpr>ayyrg-`rUXVVY#EDY3&#L7GNaa)zWe1`Iip-M&@)2m znp27?#&nlX;U`5==HXTv6cqlNyMr*ECKUA}2}(ROfRAlZ&WTDN7@PqW@9KCyV(418 zT1SU_GCVvksNqT-?%QUkC<;Ozi+_=R3#kP3RhV>%xN$|zw97xRIm5$Xb4Mvla%&xG zS>o#DUoxQ{sv&S2OIaWJlZ7XUpe}I-O$gV@@2F%hf*1dmL=(W?!sjM9wSm?ZoTE8* zrbrY|%dXF%s6o5Cj+F%OME#8*lREieET-qi1DS>*L61;367QY}XW1#ebM$Z~WHq4m?7j(}>lo3sTd4 z3OxR8*Bz01dK97C&bR5aQzF>aX|oI9BE&h3A!ZUl=A7cI?kq!EsHim9dT+9RIZ6_z ztKeB=Hq1S5>v#QC~QAm{P0x-?EA+8pw+%*ruQ>+)`k8b?1y} zoy@`mJc@x6t4NpQJj$ABoQfcf1FT@2@9!|KZzTrPK=uNEUQP}w-SQji?;Lc(Z6r;M z`QcAWxOb$9LBu}t$zLd$?XE~GZf|Bl*CzQoU-UEiqs1WKZ`J)=F})9%m24=Y^q3Js z4?XIuC5?K2KwvJ2tfPxFC(>|$p`^tuB}+BhX{#0QEWaM5u>StjfkSt6+|JKX`8>9q z*$qNhr+B(p(k1wkcHnUcWe**^NSF#aU(``@u1o&Ze4h(S5#1?$N^NybKP1n^7NY6X)Se4YEahy zY8*Zw1dT~fa^~f5EI!rJ)I_kP$W@2ou?)Y^)U5eC0wu&BtMcdq&^O1$9Vtbnb{k?$Pr98n z)%-OhX>c6`&ZF=afVR&VT-@U7abEJkFUOIy`vc+T!9uHp^;b98 zJDMhvZbToQdnLq4`D92$*Y@r>EGSSaO7@-GGpKF@uO=UwJ9BQpE7)1beA($tWi&d| z^B|t^MX>8oyIHcD)(Bb$2DBcpx62XUoA}zcsFyTLd92rK=0(ip6l?vtWR>H6%V4z0 zm5_gyLr|ZmHX)Mo<6~L(OAP#UxqN6)C1-xe)Fw1mX9jmg{E{f`cS*L~xIW(KSGC(- z+5qB6ZVx`!%etRW8?TmuckBWtT8bzyZ`zGnmexb&o~#)uL&n+bQZxh2abG1jk=U?> zkLfQ}ScgW85FzC(w(^4%j1)NOy4mV*|C**|t)b&!qxV_#_xo2|`9lj1D>dBH4{(|n zIQ^$fLSqOz2C)BTjD2?nt?^mY_BFE*Q1$Wl6Q{e)$yV$q;Y)n|uZv>Z7PIr6m6>n+ zj9M?boxIb2-mHjh3Y9`E<;SNHK1k)2Yd_WLBD~dDq8=j!=6UBb$2hmg25;binHaRW zVCcNAsmAcbpU~L*Q|hn@97&mw*yXSiY?A%%BsJ%S2(Qr@W8z%H>ZKM}DkQ&*pJ)OD zfpi|_$`rUJewYlPcW7as+XDc!ZIWi4fr@;Rb3Mqug=o6T-u?OIDBS=lV`ZC zESYKDB`m2ph>Qu85V#lS<$hfj@R<}@r&H{Ssb&L{Mq`*W_ZtxuXWd)$0`^YTP{iAJ zGq(#5pY;<}yY^f5p$KpMq+eeSpWUx{a!dEn?3p)R2S#cpP9h^50B0ZsA1TmkcS_1p z@$!C8(tXucEnbl_5kNWcd}N(}7OnXQ4xFU6PXRu-h&qDwl!0u919 zddJUOu6HEf4ae;@ZFX7=nw?Sy<{CfmI@y zek(96l0KS#t^yRAS8SGDT}UAVV>(_|Zpe!PXjLoQPRWCyLsZ|y`JL3Ufu^un;y?F%*Jo7=ReGwr zst%Qqqd&lH-NCQ4n@f4?OX=o#cw5{2o?2Mep-uTsG#vZ7o@RoYI&56tlQWm^d&1PT zCWv8mxHIGnJ>m19B;yFdNmw|6WDsC-5j2}Mxep!lGTTlTBan9PXV~BVscsQ}yX{1> zvEgM#eU3hSv=4u|PkykWgY(&0_&r+>+jru?sx%t=ibw=zkm#fa>U(ExvfGj|H>vrl17E^lLZe*WPb0t@PbB5z*Oor*;=l&eM%;=nE188R-g|^LGsq`@%q6!bkmI!f^(<6 zB6YaCk`%_V4`#Zy{^Ar;<}`;Bg>eJ0bvmoll+@?+dk=#jq?NmxS6Z|dxCiSidYNFY z`~LM?M1#6?pOkYswfxhUi7Ha}jQ+SY%8l2LcBYmnGZq^~$()GJ+M66H$0jka-1rFy z8Mrpx4a<_g^k{0arRQbs`1Kz0GvLCiC-uS$*Xgtrww4uFB_b2#JKK#_S;VETdRwQ_ z)23_2NU3;=vJ&e>NaHkMafRRE@}#M}jO(N4m~H=Kb4khVuiaZ!rj><4J{ThE=u|>e zAIp}wDbpOkBxj>hKY^_5)N zX;h}=;AMDM+b?Sv=I<~`sV$HPcZ}kHE*U?)Vi`3JuuZ$`x+v~i+tNEea*Sz0vTW#8 z#186DJYWgnlR4^Tb@Pn&*|&$PCS6@SZ(cec+i%aYibJZ}$_&8_p~vV!<)A%l{yscE z5HMu=9!rkexXfRgYw_T*hSkphVdmJS;v z@2-xXwn>XRwjAtq8lT*3m!Br?N53>kYDJTff1dvusQ1T4MeI=D49iPZ$hsnELxxVJb zqmCxl86BYOO*s21t?o+O*;lupjqqIKvu`?iG751p%z#Fc7qcUw;|DnXC?O8P0=bhQ>V4Y=)h_GM?DQVc}=P>E#eMuX?yH)SR z=Zg9nx`3CT5p&r`^d(`i>!2yT!$*k*WEPICp2}68{8%>3E&i${l;uBT3wL}2V%GGc zgp%K^sZm1ArE~XOTYlu1=bkNfJkRsJ=5WWY^+Sd@$CRAkW;t7?7I*~uhe`g9GI$J! z>^~22r#Qi?E5Qxez^w7E(bF3}1spirODDzSBZlrz99-pM0D9^D<-|piG{FUQXMFGQ z+<4;QCdua2{XC0{_^$63-P&(!c&yabf8bA|KLwOHJ@wKq{2k^BRQ>P}7*@Vx6^l)g z#TD?0v$|C8_>g<9iFW!@7me{1^a!+E%aWdR3v_Gz12lM%gkeL)dob+SH55QUt+zCD zwy@=Ce9di%xGE%$_YBZqGBH%e#2740JhGP#->zdn>d-8*z~YvquUx30nai!|q-@** zQsUqdJ5~g@Q6tjQxK0?bo}dD?ROd-0qS84XK!umsV}&4LwD5qlil{U~Dhrw0*MH&y zBow}mtH9OM9lZwXdU4j(Hvr8zI<`DaA2zsT+)%7sQ3GPHnfNCEWFw~M_ny34rk`t& zIyB&V^0MU-3|F1-1RE!evTcA85O-{@hC5-p6Na8F>u^lhj?vWy+R=4&yXO61RK(NS zQu4scir zTk#+^3(4i}1s$fZ;jQ|3yw4GJy69;Vs`|Y18_YOdTud*qHgX*FC4aX>8%O`E)uB+7 z2wUddr|_rp{lVfY(xo+R#l`0J!+nIhD$~G~Ve=B6Zsf3a;ugpsZ^)yHO~+27 zb8TEEz{_(!?XjT6HhVPk^5{r-sqn_2Th>KXK}0%wI!F-w$j)}h^>*0Z%L-}{Svl1` z2Lr>$b8N>uX;`*MeS(NHb>I+B7)~GcGm8Bdvg$9Twjf#l0Ie|=pR3#1m6es%Rk)-W z=?eK+;2=BaCk!TpdX>h?Gz$}fGWm4hE+D}U3cJns7tE;sBDCNtqEhz|S@07$7ESWNE9bIRDD&lwWQy#<6vUFY>24};NBB6<+eEA5!HBQDy=2Iy zGsR~4z0B^?_wsU6T?@Z&+*Oy9{V}Mm(8(B&$Mz19+}10rXrJwT^^^Mjqp0PiG(Tq+v9*k! zR0*;a-FL#aJQ?!VIV;`7p(bLA1A~J`H5!UU{y%LUP+r_`=EOzn^0^Tl5&xO)AztE; zRIkP_)K2kjiWxpU-tMhb)e8k*e{Atn!JY)o5W5tBILvf|Ox)G)fWxb&(Y~kF*YINK zH{?z7SFrj^jgv{#&LW>^(3c0WtT(|8?wxddc);g!IO*WA%_t@L;+Jd9(AFhop2n{) z3$aW)=>iVrZ0L7>9|5ZjA{8W%13|yMh8txQhB>xwfo$2%LvAifJu>=+eu1<5YSVq_i*SBH!{=XbiWbZMDAdO92ofA zFYw_LF(T}EFNDIyW!UT~&@$cs$Mq0GzICv|H&K!&lU9QA2bx3B=sCiJZRA!%t`%#E z6Uu0;1Xj`D3vJdo&c`m?89QKug4{|EIJXKK4W;B$?+u?|Z*eRT@;MpXPpNXO@GuB$ zqEg*sCM)i3W3xJdshBu9BT%%&{1AO~ho`t4xwVzAMt$JlxkUFs5dh+*OCEZI3*Cyy zWGROe$k5srogqJEuLi8P@x=95pru|9h2M0#tD~&|F6*-h&fe>iCp?i;fO92*B!3Wt zCh2hw&)io-FnJ9Grc@jp9KP0OQpDFB3;**jn7Y0jHgI~&H`UjxsHpk>kiEUy*{`)| zv+DV-7bK+26#@1nLDert09anjipnUSYSnYvbX^zVf`n(a#x#9kS29F zn1xtLx7J@{0$Xg{60&2+Kx)@TkZZyrwWf(SeW>&l3yN^1xtVqAi~1xVDb@}h>#!J8 ziZ1YKd>0WQiGLSSYg@l~d3ouNz}i*sPMYED$L#<4EU434Wl9q34=IUWf2Dzm-Ak1C z#^ymv^Y*)$0&HlP3o6BlqZ`lJ)yO6GwV|6Y_#UFQx1hzRbyOrekE{%lQCkvh~HuaO-+*Q zlC&QLKhAi(?PXN#HWsr(QmA zw6*gScW?XfwskDdc_D@@&k$e;0D*^_mLI-n+n1kykyEvAzP$gTHOqE1ESeG-tk-({ z(8*K75nhrXXV-L9fn_d(09XhxgHkGn7aMZ+gw?I!0o4rRwsYA8mr3!m zgEv(nljK}v2`x z@M6f=99@jbtdx*G({9*i46$kyEczr(2r?afUEu-^A`M9p5p#%j-#9bmJI8mUI&9pW z67O+JsjP|x<6sq0r^qd%U7yUK4JW?x-02dp{8#VYxUVA&(gTl`3XeVT){$cd?a??8 zDjXdbFor0{>hJ36>XMQYWRUVA5f3qFL6C-E458zRL?V@C#gRxyCM}{IoZ&+D+GOLk z(6`n-XWyxl_iOHo&+XC=q_ubMm$HEG&efmp-()o1*D;n}d%M63nUd=G`uk4qd3@E8 zp0(SS!d7g$t=vfU=6;U2Xw>ZJ%&xuPo+D}fz1}@PGLIZHw{3wJu!W61V%|VML-cLo z?Ot%r))5G7EMS56Gh<*p1?A0#C1pi&n`V;pN;pJ=>$c^kK9@s0I~0efw!Vop8r`_K z7}l2w&3Jt{dT=QAUQK^+!y;1dh^7Q(PUPH9rK{>)3i7 zTTnwvL*1TFkjyTZ%j@+*Tc1oM6h$2z8qrW_dEi7N_FJyK&9^yw84N7nICIgKhvG)h zoN4b2PfT3m5jzhY+^P6s7$2Xo9tcy`l<~zYGHzd_Y8mHn2m!>A6AZ~Ix_pMqlU}+$ z<8qtJH=>&L9Gg5|T^cqR(AoHkP1(WeDxNRX=~hLD<_qd8H>)0hym5yJZ=?p6*Km@O z9j;nx$Z{5Ds<2Z2ACLYAg6qVRPwhU`R2uxPEsfTajv0djTiq>n6`SiSe(%oB|F!?*f9yHZ z-Zx0C$YS?U_)mWPs~1~3{=fTo+}5y!#W|lbxaVebb@#-j|NHjEznF z+y0Y}Cdc>h+*#95Ubm(SZin9R^c$@L zo1;0NadN+AGlu9m_kDiB7(gRfkA&brJiM_)JgsbBQC)@@RfYkVYj`SP?zU}laaFPB z$^Bg^-RO!WBk}C^hO#QR9M|+y(e#ngo;Aguzx<62#26x^6%WsBye3cp9y{4aJlI7B zlv^QV2&InSN_oZ*PbG*#Y(dW<1A`uq$Ac^n zAKxsFfHXIjE7`mW0CfflZ2vQKGoflRRpa3VmKPCor@23JsC{_H^2+yYs{hu$vx>@l z1aaG<;{SF_!{B&oXUhQIdX5c5_nsZu(NO+dTWkO6wGIVA%y71~vOQclQs#Xhty9iq`$Q!|TGUNly7$*d+>GkwH9D z6(ik}lH@eabL3^i?SoxvVKw_jOdM!KMjY(G4zf;`WcYaWc|1O^*YELo zy;m3nb?R}jQij0wjR4|ReEQBUk%{=vkF>N6jQ-cI!^=v7JC;qoe11BYf)6gV`DgcU zfB)vy|Gww=f9^gsG!o`uA!ryc9zNC5H}u%O+dqEGdKTw=I$*Lo41pU2f$AdfAHVgM zoo(G_BKfPMr;pvc9mY2oUpu}#-oIrn!iGJ)ulcD%Cl#msgActU;5xS)_~KLSlE&)N zFWk4isj~FjFYSM=ts4uBCnjS5^5WiTEdD?5-M)#1-Bx0FJIGrx9 z!1_E>6qCzt8tp!HVs~3*ZHc#Nfx48rc4%K&&o}TKfxm!$2V3GN3emF}1aNLCMT;sL z9DSggSC+aWiJ$Fn``{gG{^z~x{`{9mqN)kkSam_yA3n4J;q4#a+wK;G&S>TbuXp^; zTUP(s{i}cf=gnCaYfqunA!QC&1aKxG3Sp_E^Cm2K@;V+nV{fT5TB|-uuUxiu{CS)B1bYVF>vWU{MYVSwyTBjP2#O6>m^Xi$wf$`+mUTXf+cisHOcinvNnwk^c zk=l}?yBcfa>Fi(s{M50*s9WMk-|%otTRT>d7?KR1X_geDN6)(R(TJUy+~U2%v|NJnhF;ov z60+%yyMfI!ya$-3yJas$_{{&1*`4Ih$74><92FXTCXK(r-qxaN2snBORVaA)q-r#G z_vQwC;mEDap{pSEt)`JJwgCfe^mB#D<#Lj3AL4VRW0!9SVnHjZmPH9(8(-PGudTCt zaCl@q8jr_g8JJ`%umUHgmn2J0hZA;10l&Yzq@=E^9ljP z!1(YTH}!<0$GZBSZ|$tzbF8kU=rX~0rg?FQf$>)l-S+PFE57u?>%Ta3l1r!Iu?;aW z{_Kfg@9XOS!u_|hIOo$ul2u~}+z1HV*0A`tH4VG=9p$2l#}A&IQo#cpmY0XV@b){~ zh9jps`hI!*Oiwu0SW`M3e2P8J-@bWmZP53%od=#c+5%r~Tvq+w-ea+3_KzO8_04Ft zr_(#D!w{Hz1PVWuZKiVW$H01gGYG^ZNyQK~cxV7eZkX`Don3WvOY<+9dLp%}E140` zo5A*KoqNO05x(FHKH@D4o*c$6Yh`96HcQs}a?}XzJi2IqNUE^ux%g-LEDe-^$ zts6EkEq|%GzsTi&=a#z4BKKeXUXZH zfQ|>=cwnA`Xet&nh|X-1gk}=9J9!HpYi1$vCR3ppwDW}u9;-4)3K}{*wvlZQas~P! zy8_$J7X)Y~!Fq|3AxYqzXetq;U2<6F3o`uEh)|ZpwWjFg93Ye=NLF~-)Fe^HB5K6} zr+!fG;f9c`s3J4EYF*Jd$*svUJs{{$*VCnoJfX#B4t6hVTFewYc`&g$oS4ZsJx7?N zb0Q$$`eidb{1r_xM>XcbqkK~nX8*Eh&%=Wx&g^6<aZxO#e|PuUXO8u@j3k_hUuCI^4i7{h zef-e78!K;HS#eKel>)QN{^-|UZrj;5=7Pf(%fu7ilJ*^z35ZlkvX&kZ%W}ucTk_VH z*R+pAmO!>M`Mh;xQV9gdQP9C7DvxE2d7fIxt~Li-3^?D*1%u3&FPIEd@Jx3q>GeW4 zG)Wp+xkYn%@HpXdE3}$9Zx2BEsO6u0=%qbJNN#0|GDy^5d#CUdA|372Y$A8*;jtO=R{XONgZ8(29A;jNB`)-+uyTs)#Y1QC;+qT zUJerL}SE}em8;m?npN@ufw`i?vASW#P0mt`3O3;~9~Bm%Td4vd2!;L`?Q z6NmspUqmPY)l7N%NA!f1p-+AICc$SOzAlKy673HzViI$-#Ways(~6F0#7;NSVd&R1 zGc~Iwns>K1_ti9%F$GT{jBJhy(U`?NzV0UzF^0KsY5L^16Yf&Px;fT0jOXF*rX^I! zbIFYUVoU!-Mh$1wPkj63cWkV?uW8Bs>zBa1=f&oMpB?Jj(;b0!$09O>xT=5T2Z!E! zeBf>CtM6Jvw!96X=r2zVd1QM2>;>Q;3h@u2cWlNG)u854t>J;Ipy$y_Gyt5F zd-RY&DqIq7SF=rd=-URzEKbHgk!`O}1eC1Se!MR_7Kf!keuNML-OU4fOi*iT4UZo; z0kn&t&Om~hks(?g?F$*6ASXMFDi^Kroft7QX{{n4LfHe>$m`<-8TSLPo$c?KIQ0BU z(8&248#kxZP`u`%en&Rr~i7n@szSX8a4e3NsecXUY4zX(lU;T-V@PriARP!u@VV?hprVaho& zds-kAsG|M&LtHG3Im)q}bZI1B!3NW7%p_v;O2Q9~r@r*lJ*Ez4X*`k9MEME`4q_=k zxt?h9%R8I@;gypzC#aeX(*dst2X2$%9-(C7oWg0}rF-ba9y^ zqK4w5Ldc}_K#4d$tQ41{{mi1}@=!N=akcE{Q;D&L4H`~>8H}conUUn+p(S$r<-|m? zq|!Iqmn^CFg$KqeYD)*Z`0?R4IsH|Y)sVGu-z;fY8XieEBUNH^e18uyL8VN@X zhe!98U>%(1@~Td^F2VON&LhuJvYLfEJ2{n5(y`d^a93a7iRP1sPn=xSuyotzO{*Fj ziWazE&MEO@g*Olah=K9nZ(7qn6#n`v`^Ul)&mC^rURyn7a0`6)eE-0yFFpU-*@?tE z*DnA3y<0CI=Nup+2F7o1U$?oo>Wk0s{mJXcxD4VFi^sbMKK1x7nn%We|Bg*8&iM^= ztFcRE2wWNgGNbW(xkS2WV03@Szypm-FID%EO{>24a!q-0U`c4|ueeK{i1Nnjvd=$o ztJ~}T*3Ny|cmmAq3(ak?^#AjR@4jng?PX46hZq73fdz~J7PiGk6VXTl3~&C!0_5Om zPdJ;%n6iY`bzl_8hyXkyKJL>Aa1JKop1M8YF37il+fRJA8KyPkFuMVZAvz2P7LU`{ zMq)hKe`a`M6z&Id32dM!mn|xCd7XJKBX^Ac;*ASRY?HT^grg1kK=2fq74`k?XVk3b zbUSbyjpJ)(%F_#*S{QjYN7sHHY4tsQKO@;19(ou+)Y&&8zR7tW^xs#XJp{#zZZn2d zh~=GN42e)goQ7{4X#3W|HaKK6G{=Nt21EgCy>o4H4uQ)U=D!{9`?q6#vgO!HwMcvq z)@;MQ5@#X+51uz$yMwd97$RXegB7IHW(=)IXHpM@)A0lmwM!2jJ$BThmFN*G*v%EQ zzt*L^QBG@=ex9Z20k^-l)_++QCDTY|lp-hg{bc}uz=M6*T=Y!b=V zErh#0R26Pdz5evE_;@NhIFV6Qm)lWX8LV7dR^M0?EcZ{@RN=_yaXynVUwFyr>N9vb zKSyL=m5d|ucUdO;Cg;eEI+W3ucoB9y=jI3+HfiNDGcYx8bq|ra#+^w*nD*%gweM_*P{?x9>7WoQnVb~G z8w|h{gn%VfO+!RvcHH`O^0`E!7_(a>!&e$Lh>c+d2!x&!OIs{!1QQ;dAfK({^r}S3 z(sW`A>1yjqEJ1rtuVe4p{Q!ULVw(@}LA#`k!}@$Gx2vJh&dklWP{ETQ!`yT|_q{L$ z&rC7LtuGXJw`LYKc)|neP=z}@gd5dxIHhbRRk_sJdnUeQO=;(e-VJx#O`D?NK64}k zwNU3tGgvNU63`NvKEE(FqSY*OH^0UOLar09rt8<*^)*qf?rzDJFLuPj|Bt=*0FNWf z?*qGX=Wd{p0V0wh31)IQeYL% z{_nki*ZcjG9*8h<0khByRrx0eGn^0}yfiQR8?T+2ZRl{yWBrD&pOItr!u+%+5TO_~ zbJKFicH#OtwfjKe@_Tx$G0^vc(z=;`>|oK%iM6AbwMb2%kTo>L@P@tV*=*gW&_w^V zKU_C>Y1S8Lynbdb)u6uV3mT07l~oqB!N>w#-j&o^3{!* zXuTe)Hhd9NlF(-u;Lx9_`Vy%kld_{a}KWyhdY`MHrC!4nE2Y8 zAF9dB7e4pQs-mD}L_rLU|Ma6TJl5Lur{6w4IurMBY$}!g!>_+Nnn-=&6OX%b&Xswa_vC-gu=(%HCT0cgG?kz3_?n4Sx?m5?a>w6zktXW>?o8$n?^3ez`7a0<%_+jSCaGq7?cAXg0ab(fwCWrA1#pR*y-FkY=m~=a)htJzF(DE%!X;Ay+pR z>tRW@4r zJF>pLSf|!p0nf)Fw4fAE>TCoH#D#R-nr^%9>NBx|f!_7h= za+h&?BsLPLVCU=h)bI(1cvz z4OLW%f?+~x0;jpWoF_T%0!#QfBr4w%3I)42cT7x8dopn;HLKRP617%K_^D=iDc*B5 z8*P+BF&GuWt*WiEL7V{W!MO>{Nb>~1WN>WOwuy$4%O<7xSpDEx&+PE!YyI%=_}Leq zU#$iAoJ)ZRiUN55dt8cw`ZIH~=Rqb!zK9$Mlzxc|NZk+R#q;}7^6nLm8 z08_%B*}MIJ?7fbS#p_oG6M3aNaMOM7PwejzN;qZcb;*IlIT^^ss-zaZ0`_ziQMI&#!XD7@N7I?&JFi|+O);-RgO!E$IZS{eex|84 z+_kT@rMn(V;;UA;(NDT#bk9c+eh8tCO%8HMl!6U4>1Y#DXU}1*8AR2#J4p@-Niu1p zekgBn9>I2@hJ}TFJ9NeoEST(WN^aiqx(c@d06+jqL_t(=7pj7greJ$B1Y(Fos(fL; z?jVK`!65j9$Fzk!WO0K!ai>!ZpFrAp9*AKsmk$PmbOpw9xg5goMx&8sh@pLzdXAgu z3q|p0fDCEXmRdz2H{F*Mz<7~fPBG8P1!xW0=&t}DG=}733u%w0KnRHGVZiP{g=ZAH^ja1A`Rkv~tv^y*4DHC;1$Dftg) z)tq|db9)5g0Yy;PeT<00_Cj8DjF`Gw$njQ=QD)oYtBz6iXm9WzzRczd*5_Vf5M60Q zC8(KnG+28!8*_0i1H5!N0#7hi@~}p{6YS3Be74mhL?*-liJ4;5Y`G40Mqtm#+sAm5 zailh&g1rX7pD7EUZ^SNfs*-8iP1vU(Jc1`%ELI4LNVY`n;2H9Sx$dZCcF z`Ln62+RSVm0oU;lj2J!=soIKqvHBcs7+Qv4)23c2XsOJv^;lI}@VEk=wWjvQte=(% z)(+-AebVn0EiF1}ACcYMk_RIN7_2tu6D1ciRK~swGk#flyJi-0*M*wx`e}PXrge5z z6*r(l>IQwH9+K9UO46H?UBj~aot{%S=qkMyp-$v+H}L|eo{APeein>q|E^tczV{*I zPvZDhrLGN)+$5QK$v@wHNQF9v$Kt^r=uMD&@)y5#{L3F+^dR`E<@x&Q%f+Jl#m_$N#yNjr=a&0mE(Pw4 z0-xN~^;^-1l1`6L&7B&Ve6gqFX7@>O2=IfYh+q8mWBX;({KId(0|W>=nK%3TzxZ|O z-~H$dTkETTu)5v5T?)7q*kBZ(eG9{CfpE5_1ws>ObJJD``gVrSN)jK4=(SD2qyaWS zPtJD5qpMON7lj@S^k0FPAaaAj1rf$u;)4;53Y&~lv?L!Y=HZ~n>&rzN_+pm4x)CaY z9fnm~oWy-Mqm(z~6*(QMavbn!+wBQaO|&Nsz82e;`XV~8{Y{H=r-LQq5a1AsgmF(fb#qDb56 zk9~dt4FDWSajGEtp%NtJA=`+BzT+JJIfx-v2k>~@TCqpRMI73vPz-C3E(5Kl1w_Tr zwB@M^8BI)H*8Q1GroO%o>jp)pXJ&Cd7zm=rw&pj%i{(Vi9>4{fM26$(j>^k+xs_+i zx4W5pNdZ#WH4xN*_LmeR)-f~yoQ-${8l>X{P9u@qG(v$OzGEy6rUVScIXZ%j&}$@P zXz(FLWek}<9k>W!wyCqa#zAdb*I?FQl116AQdNDZhJq>#=h}J*{Gy)Jt9m#3wfkv|*!nsyR?Wvpj5{79-}p7QDnfFG zjg~}#ZEIvdvp+pCI#+<3S>m8hs$hS_?U;PpDkS;V`ppZ13UER&bIX${XcKb0qR!%7$d=_Bf`!7c8Xy}T zNd7q(1nBTTQkes2`byr(EN&%7)OLHE!)5SN{OMP2OXQy3Mbkz@DCMwXv$1nOU*_EVuoL zDFsL`V?|PQJ*^uWq8Y*4Cb6H}C)=zCY>cmokNReYSdMFJZhqm(V=Fn!?dx(&xD?oc z6zHmrcGgwDdz~LmX3vk#%dp@)E=&v| zgbzS}i7A?T;pBi0crqYbPws2-d4**#pwm6nX2S&p5uz2GtVHo}(oOd#1#Znq%YDAX zV&(G4ly50#Dz8@Nd4b<{sD0v%UcToT6a_fbg5W3(x00|aL%D?sXHPAJBr!d8!{=LD z{ag06Qt*?9Al;UL4h`7?5l#R%*tY8x{8Nd;-QXV0xJJ_yp ze169hc=dDCcgm!IU$q4MQe%f!K6PgqXzhhEbB!0@60>uX-~Z_spKEPty0cr|QZ5BH zJOvuUfrf@!4}?0I!i~|X{+UFqWA*uu7#Pp=biiEirEML7j~)gCF);q^%SWSr=}X@| zL2=Fz1EX*73#Rpj&prF}R#(Qe;U|*&04@bS9tA{>{h9qeZ=AXa<*{#FzLCfus}6WS zUR8H^SG8aI!ZQbpMdh#Fc%ML6%lf+$7a+I(qnC~b;NZYbE(JC$1SexzZv zA&!bW2%95dj?g~)dja=z@;615fIlqUxNwtwb7^BzLMbZRS(b&TwbspbBj+Zva4RGQ zfVmi!kzs@kY>a?D!yEzvCp+ng&EpHj3_TQ;TDH~k;>JUxmOcAz58W}kEs55m0On+8 zcT;^+wW^W{{wfLPH3mRD!N<-$dE^Bi;I<7Hf*rvHP%@*amu6-79W2ab#x6E@9T6lS z)`_X4Qp2a3Iu41x5HTyGkQ#iyu63_3R0G72we-~ByRoLNfoL;=G(gWgJ$$07zQrHw zfMf{xcz*PQP<2bNek&G*&mWkYyreTysI?ke=Nu;}lUL+oHqd^Ul}#m=>uQj?I~yH% zZUtfp;Svx-oTg`UK3920_moS4)hMv*iLIyKxL)XcH?{9Y7_g-kSuvK!5zry;XW7nc zC&Htb9{*>L1tZ?dU+UFrbR3h+-HRo{3g)OS3!?}girQDC%>|7SnTM8X?%jNsxd z81muo78KMq%P%sVDz>(@IszU%ED~fVa5xxBivl!p_W5gOr^3{S>z+rWa4cA*>t}>_0v!O_}RE0zQ zRHU-Gw2u!MCN3IKm&s-loG$_gZHlU!#lqxNye<|18!vO?as~_<9yM?AXlr_Du^V_D zy#3qUn)YSQD|an;)||v{15m*E#8XeaclzvHJnp^trrOY{*0eA>`QUXDxJOpSZ|cX9 z?)VC+mdkJYClLdHe{|p8=bkviZ$P1ai4yL&OM#C@0mQ&K+*G@#rs_f_*E=(RaVmc3 zCITMR`?=j)KeM}AvNe-FS~GTO#K8FCvxlRC@bABUJe8(6=Wkyf`u$hm-`X?LWQo z;ojQV&p)>NL4RpqYtp*)yA*IK@bM`C)SzqcrmN>qCno#!Km-;)c}>m3ng?-6Dr%)* zK@3|RA4YuXJ&!gtb=GjUKevh^Zg#CG@UV>$w8DLU34gaBC}byk(K~j{1Y!sQ4@iWB z@(}R&6UFc(!>f$Sxbx+E&%F4YC!oMDs9G zeOI)h(}bZV4tX|Mvnf=w33ro+OpnJOZ3ymviTElcIE+WawY@LuR2KKcZ(Fow*K@XX zM~BHY>kBmW97EF({Xk8{(ydfFtEnpMg@eOIg#ss!&#ZS}ZsrQ{NwZ^_!7&-%2gqGF$=x1S11Xs^M zVzC@H?fnG6K-JKIwp*6gu>A@2P}L2QXHG+ixxk`Ic)9<$3s58Oan7J?z5Qz;V3E>TsxOu{_0a$2o}XI3D44AT@*@p^3bQH+d8gDPlBS zT*3<<&a(oG>A;DWAOMBN1OwDfHN(XpC$R8$2|ESyUIj1ym9q5(lE9j6dT8#-`vX@$ zxG}GWiw9oFd&5bkDBH$5w%sR9DoM@P#m6Ee=cAL?4?W%WM_e8g- z*%k2If;jG_yG8*?68G-d@zvMA3!$!f{iG%N(~rKwz%#FDK)RDvm1QjE77Y{6ATFXu zXNDX5Plkprvbxc|rSl*E$me!$U310ZTXf95`!p zt#R#of*2UT@x=bDX8!4S-u1v&8_Rz4(zS5F`-M+E<;FQ*JF>eqyA-%@DS&M>TkrnmPruSq65g0$!o{73U!>xe7s5xq+GC_)yP3Gfc ziGhn_@i|os)I$5aper&ipeN`i-D0tZ9w?>*@v-`m^IJmsCqBKauD$v}j_BZyQROVz z2ziWHnfYL_IbVqw;!&_4FGGILA!azrEK9U)FC{8(gNDw@^6fZW1u?`T zX4qWPuTBl>CC5JHzEMr~I>ZV?mF4r4Te&_<9dN(bhXUDTVQw;|%Q}rK5`sAulw2~) zvN1?y)PRU-2L7sZoIjJ~G`xO;6N@#F2?@c>98<_y zZB>?yAThz1$%$ahoO+qp^@8a2)NC?20VWIvWQ=kSpwm$A`22Jba~gdxbt9isl9S1i zzKNO1w8Dq;bq5MNj}^o9knR*cOxt*$KRwSto}_7fKE0y^Z^dYa(U0>q|tZ9oSZ^z z7qxDCVdoRK7$ZIt7aD50pZ!VZ@T1Jbq33NgxkW}ti^)!SL*cy=Ca@El%sEMeYyXuG zos9iTvVvb%mMmQ?eJ|cebuK+iESxi7k7*VSC(w>Pp42g9U_FBz`iXFZb_)v~c4}Oo zPuGHHedz4A@=#Df)wJpO+_f76AD%m}Yr6d!=3xpc-ZG%P1H+vRNq3eriPcF1zJOhs zX@ql$z{K_Xq4VCk5d=`}>Dl&EpLyjmm}KHb)J-l09x@6b2F8)*`h)e=#}ldE={fwZ zselL6MGTDJesZ6x$X|NnLyugkQ9&ivuqr~l!nUJ_T+n!M}8dVX8`uf6#Azxr>l!woc)J->7Q+P`mY_=8uT1XsN4 zZMhY?6mThU4=DiuAo&aebI1!=2IJ8bBcCo{7XW(!!z;qQ2nl1LBCe}I5pdG-hsa9+ z-chhS01TW@v&Ec85E1GfXW)y+kbJnEkr7rfVEMv`H1824Q}m%0k2Qjr+ZHEi!75Xf zI8H~+tZC?)kxAs{CQ?9BP%$6_Rng?Ul9)=*j3%Zgax-Z@RkKIw+?x)>vW5}N#3j@Z z8IPecsv_m)Ly3u6W4^o9yYHDjox59v2gu=%J4W|#f&73?1A`+J&<5k2s>-0DfW;+n zOr@a4F0taoLIuv3Q;GArxXEA4Eh|MAR)PWOgV!e&;Ts4JfJl-91b6`%bPjPBF4|WX zE};NDi<|B#1%@t99RJGs#9SV|z+l+sI+jvUee9LS$pKpozc5COZm@ub=f=otq?~Y3 z&PQ1`%)!?dWbHV{xLb^-v!)(MPLV91s-}~4C7+GN$3jCe2dgNo8$~t$+JE@~#zPBJ z0v9dj`t;L;95#vu~&I$a?_o|BMA z#`#=0wPs)FOu6m{9x-+;c*9OAv6R7NnJv3{UsfIMaFPn=v*9p*Qc$HX^j zD}q1FXPVP=l9pddGsU?J$_Sj=C zd2|HXQ~%12bN<^;A9UlKKU8zieOQ+Qpny;0fAiQQZ}eXGWDBpJ?tSx-9WVBDtOI3u zj{T>Hcb^=Y{_?vgNtf6#{^HG3Pj_y5dDrH3Xx(kXrGQI;b)!HruU`3ZXz21JZJw~_ zfw_%t%}pkmG!wG=4fuG$dD_n~tg1?d43keX#cW0QEG|9I%K5NdfZHGJj7Xy!8x2j7 z3RxzX77I>c)B>XglbfIN3j&i%VRJ&;5gU`>RoTd7Pk;Zq|I`QsJk%409LdN!&(06qxbsafg$|*&-z;C>9Dx4nYP@47lc{A+Q>RNrxn>TgsZmZi=^?=##xMNf? z-!^vE0Sz(hkw}yR1j1f-w94$X6y{ibSbKSv3fZR@4;FKDq2kES5(cq65jcpF5WOOr zgk5zRF{B!3xdX9qt}MUTJ-&w&kTY`H6CB=lRPl$s9PbrHm_FM5q48YWF~WLnX1e5b z2?PE?3+L?WovJaJEYWKO!n!wT35YgNsz5mfVV+c#m_^SpGT zR13b#r_dyjSfIaD7iw~r>SD=qLFb;;yJs_Sl(bEB=nhooq>7;dI8CB@4!#*}_!kB2 zl8bGvUCwFCslTF2YuDhttl1Ut+$r@JrvmIuBejM%poFUwUyuz|LoopAV7lm&B_9{5 z(Y!vTs@~o>*p)-GthU+o2F&VuBV5h$UIXe{{vgy4On*pgY}b4~p|KSr&eAQ6T2)uX zFQTbyzM#Q-!T2m+h^wkqB)=T3QzdVqw%PVeU~@!hvBm1mV3n?^W*`jR1Ww0|VZNqO z$8T6|t0{Pm`W6$Bu!#5Jza@vOHGc^55JU5uz7UH${Xw~^rXWa4b%Q4PlvpFl(w*+( z$?3-1Cs|R_nQSVPjnB_d&BjN^Cq^eGhsH(|i4;^V%utQiyhZ8Qr8jLge4(5oXVs!^ zOOo)K2)kKgRd9w-N^rsk-)p>~@R}kNvI4Aj((yoI%AcC%ido2hy0&y2+PCk?!w0u^ zbw2Qt*b-ITZ0j3g1(DxQ)9M_{bEfeA%wti+<#a9>g1z zqB4AS=1gBPThoHA0$?{tO{`#(74!%!Bb4AFdUVNShGMkeS^6$wMCv^eOCLb7nOI18L-|iUQS7QkTje#8-pg_bpAsFqZ z2?ep|=2#x+o3o*I_z&q>0x={J&mtP=AcmN}D`h-{4myo2Ack}T;vohqm`giQab#y# zlkT7+B80~>CPxW-0y z-)<&MZ}_mPUHiGbUM)ZcWlo=;mJJQ;+TnbwHx1|%x-0o4*W24$=7h+AO=k*vqM%d! zDmyt}E9*kxQVVp^sYH-`Vj#?U1BM0vwIJ6OE@?S?31 z8Wkx?cgj!doLN!p-8@m*!~?&&A9evxBAJ?w$8A1;t2u-Tn#pGIf`J9$E9Q#zZ5Hp9 z@hZ?5`Q6^YwXlT(UiQXkh5dQp}gMEq4Pqvr5oNwQ|-F~g&dg; zK^GyLRGK=`C1wL#h^uXWnhT>I2}5f`dY~9*mVxlI<9DaFsU3(1g zX^=>`A^(ADR+ElJ1ASb*a3jPrOY$s}Pp6JBHTG5f4BZg=TodFnFw%|dv_L&9g64o59 zLWV>-?WFlm$!VaRS<#+*;x1DF6vq(5OcBYcTGz;o0MbH$JeU=>yw4s?n@Kth_qLq? zW(DnJ3@Sv7w=Ha|2wR*D9N(gFiT}bTjVWqsLX}~-fOu(vVpf{GvHj&X_=Z^q{;H}b zie-4}7Ioj(moP@~oCVmvJVvGlN2VKZjJTJdVgkWsq8)fjWyOu{pA8`Lz@=n&%P~H1 zJ8&Lj8S^k5BnNUv)8Tc<*WbY$L}4d60uOxR5TAf^b4q$HKF5b@IBy7+HmGocnJ!jK zCrT`K=~+xBTp$S#F=SkG>PO|}DF?h=)ldTPFOT!(`a)-oFqy&d?ezpMVWeS!`N)l5m6Hq5Y4M$Z8-_u)hRkAzoQ4FfSe6wAE<&6mX|gAu!Ea>r`YA0Kq4M@y8tSKa zB70|3hbLs109{}LV^{g~JPv1Og|KF-B0e2N;S=IDnu~5hr3aVTzD4M3)o-zGK zxK6KWGOAlNguJV1R)baWPyh$k1x-(vq&w&_e*SAF`Ww%GA3`kpa}xT1zJLWlkgaR@ zIr4jbRduz^O^qFGn|5yP?(XbpsH<~`v-sF<&ZWS^N`YTFHvYw1rzKJN{pXKV`PXC9^&|VXeWkbmTPMzXunjlOzk2)Rr+0Q8?P|Hhp1Z|d z3b+)w-zb0$6OaH1MVyWy3J1J2siHsY4`Qq3_4-+!_OJ`vt~>5MLg~ZS<1{Zv+g5XAhtL@BFmbS?7k z7_9|W?^d0r!A=h5qr$e@P;Av*{9;4)g6M2;mJ`i)dr|`opnTh6xz%${VZeE1M9;}< z5O&IiL?Uf^!@Ms-TB)?ARwk!{CFd85RUn2a=cMxE07@Z`Ka}7rSsz)QiQ3s6?hH*aPz5ePxPRRZ*+D8)iHsYwA2Mg<(#IC;GQ@jwp0mdkBF z)XpuH!CyT!0{gISkF;-icbXW4!C}}XS*@hby; zzJ^XQ7{fY64=Pvb!XmWapenm$m$6@$Xu(@}1F}_Ih#^-#0ywlzCOnI`I?ssD#ZzmR z!y}jLUM9pXpMPL754(V;e|Yrdx%19@!SX>}GTTUELX*@oBcoSH&O&lq#nu zic$f~60V&oUBCIaQy2NT^@K;SP*Y3GmTPtj;v!zUC^iqn@YD58n;ROM8yliw*J{TmpZnobphN-0z<6a#`e)-4 zeBsYtKe06u{^w5}sC?O2BSJ*|;_p2F*s1G7@kGkQv2$~&KX~)hS6UmT2hk7G8qv_L z_MTIqawoh#JKlT#$hYW#4AX(>WNlNmqBdJb+|7M*)`JP9zRuDJ@koW&|OX3ia>`*z%bY?ubG6s(3V%}ouh4sQWh zsbn&DtMjPVNsgUOC1R*5N+^*S7#Yj^a=xlYK>ScStz1a7cI=J+vd%bV0B?|Zz(Q#5 z)e%G71(XtNu6#R{#!4q3W#OUAm)y)6Q9#bC*UpaII6Hdqg`TYk+X5kaqLC|)9RxKDtOf{x(seF77rb^#ynNi_^%oAl zQrNWJ6eVKygwmE~hs~HdwdlWgQablK?&EepQ|#KO`=b_ir({lG$FwdzHtGj%2CU7@ zXR6M<5u3V^?0I6MzJpPX$k6$o11(WlT%rSZ%ID-u@7#FjtLHj*Hf`RsX~S=6^9A$l zIqU3s_JwDeg9n&U$dW`5WMhpHe_$_DQ&S!zE0?q0c~Af9tL)Am&z{|jYT~64-$>sS za@_+dOJ$$SPg;4hoWJ!|6kD|bI3)c#loE>J3_fAI2_s9~vZrUumHuJHFokf;u;@&= zMU-<2m(S4cbht`d>MpmX=t4 zR9ZqGv3{(q#kIA(&dTFb{nV$*^Q-4e{a^iR#aS}HmHeg@(0u{RAGQKv!xuILAMAJ_ zcfvY%^xDv5QP(nxtQrf_9n^t)YiCEj@nQI9+qZl7k%Rjja*4bM{XSnb5)KCZ zaK(mvMYC?!rGQI;B@{T&T7O{2<~K$s&1B}pnZ6TkO;2^VE-A4#2Z(|3`A2t5PR)M# zOfNzar&F1)y?=W0ayN|Xv{NpF~{~urbKE2`a+-v8qo_hMAYs9k-ly{qODR9>*0LXS`WbWj*``UUM zwjFE_Rr#T#P1`b+L@&C>(im2>OfEAgP7QbjA7XqWZZ^7sy-Ve`%ihWm(hEp~(N!@q zCs~?dZ@}0BwMSQ{ID@fknm5eFheOr-rbiN(X}MHkZZz3{abj*FRmjMfPYeb^{sYf#uW628GOrym z*QjP#YFP2O_2%2wo9{C731ecCdG`dfqsKEj!KRZNYK%s#>I3G;C<9~ORLV0tZd|;~ zPD~(*hI!%xPbkbj_YBkASjk@2h@^MDTAEjodq8i0LB+qsxGGqo{W61=7ryCA3C<-k zo%64vXG;nY<6fwf)MBVTL@#5=s1oHMjb#wW!|-~&!C-K7bo8kskG}o>Dfpq{g4Gto z;|+>hBbQH(9o<4z<<=Iusgd_JdJZsToBx&zAcK2g!K0A|jP6`?csMQ*vo1YL7e@FP+4;Vew6oD=YK ztMeR~hP_E-(Eq6yo;vc#0Xhc_9|+r=x@?|O zS&y4{Dd19IRSF;m#*gjj>bo)WjdOi(4UBC*d2w@X)%v(fLkx@`KfL?uRQ&Sz6qGm0Fp=q0r0IQ*x zfoLL<-wwCqz&sgzM?~SuUVj>i4Ox15ze8J%Om4^LcrKrx&WHedLCJHkn z$-&-Y^H5V%(Q9bACo7P?>hd8F@! zD>l$(m!=(klV9;UefIX?sm58Bz2Za`=C$-hm%XQ!o9YAN7*IO>( zW_~aVsIoRcl^(n_F?e}$c5EI#p8O$*YOLvz`S@hY)J!d>#*(iANd+9w5$=mHP?q7d zqYX^-^E0Ce(F8L}OY{S8roQNK0-+)NGsNl;m{Twhfj7uq749>c(W}DbkRVEkX^k-0 z7*<9rn3}>VMZ`U`SdZ=x*bih`*?BgfWQQ&>B2)r+l=txDC>ZO(QVW)*f@341GhEb| z9D*KL-s$0hp?SURrT6-;oEk!98Z?nr&A_`W%LJodyu$XL=${&lckOQOg!l878p*do z;6Yh2r>4wHm(9z4p0P2gI9Q<|7OOQfYE4dgG@Z@mW68uu86(CkS?`_joH}Ld7)p&% z6-+G?fMGO(@wHBTU=9wkyLVc9cX?Xc*mY4Xv!~o`K1@@eNf^llW)+y&YK3iCwW0C} zT{(H<0xW50&JzQ***2+_EGC<+uEN!9hPP0D3D+y%jElIrbck;-NhH6>)%l`+e+WMr zC;)a{C=}xH_%lZjzjgdW*xLx9S_2ErZWy=QzW}9LsG|!B@*n1-kpL?zGX(@Y@^EP=_T#C(RIxNbtg;|rOhpW-(FkT5kyQ&_nG z0Sq8N6l0V;5f;0S`3VuK;^Ws4;KA}&nUa_AU|r>k88$P|C1*T&FfqUy95Xj5rQ^bA zwH1t-{wM*IqR-?+0!vVpR^WsZ)!kH&O*o}>6yfgdGb3~Ox=)kYBpU+ zOsD0X#<8AEBLBf_m*eBfwrx$dEzw|gKor-HPu^+_ZgsoLUk7B*lpOYkP=9_Qu~b48W2OYPO=+V z@T`g$l1ex(*l``;+>3-p(xu6($-?QhDF331R)d4mfQi2nP6dh(GMO+^Ncd3R35%%x6oaHz{%Mkabcu# zNxDQQ=n8eJtZPhXHH}r&N}VH}Ve99Io96mH^$U;JwN{bgfpa@S<%@OVBu#dnmAnF^ zjWBtzTe#mp=#hN^X4ejG+g4lJS$ZPS3{H#shqPC}8JF5N#28i2UFUw`XW7J1da0PbOo0dyAGe3FvVhI2D^ILRzOt1Ks#tU9blDgj z>dv3%Bb8J&B?lL6SWvu3l136KmS-duAC{P#?vw&$!Y#c4KPV<5?xa+TIa+ziex0)# z5q+nSf!l?rjvR&)w#ou--lc#`fsaZ7#K8F6?kyL`rvCixlb&4R&%XNsjN!ic+~IYI zbB-7oKe>0?^`VJxUA#&$FcRtSp6xr*-gv0J>7zb{Ti&IBOM#6`0aZ8t^6X^~lr=rp zCpNcv*D=lnwl6=vzvqv>b0U{ag9|0*60csm@f*(^+_=Pad+t);A)f(Ma~#ek=-0}9_u@jYsTgh_?GAs21? z143#bK=P<6#SD)$dm=3a+%TR%Ar;9??AX^_+ggp#+~~hA=&Py^w{3&>m<1JDz#xK` zFXrWZUPb^?OISomOWnIKzuhsq4~Atnq5t(al?&HYQ@1A^%}pnG33y;Z^#5hKb#)G8bRpTjly$HL64NDytp8zL&9LZw7m8}=qoLKbH_ zh$33FY^dEv46!z=Bka#@HDBy#X=`j2B?*Wj612ke`Fu}zC&_p+GhCo{LB>;2ztc_` zF{DF)g%m#>4^j!s(-jA;!-8_~gtCBvuDNOLDIiL`P#5r60cTb$ZO}px<=~q)JQkk` z#K0{Dk=FscuTY!Dz2#|FnO|DllB065a%Rcp6$j;F78o@Y5CzO=oO2Q*=$bD(w{>6h z)W?9ul1Y5RS(O-)ZCFad{JeIH_#)#jy zn>lpAA95JOWR!LBil=p6%W8KM{a=O3ID zFCmChM&d-AHw}&@UX2W++km~zt~mof3Q9y+;FcaXORpMq))tyB&;@y0mP0KTELK2m zaFPazD+NPC9CSL&%rTdS<1LJ@rmoK6)%a9lEiDMyY}S$S$k3_@*J)txOP4S$B>bl@ zB(#oKE|R!I5rmFyJp=oa{ejNF$u3%T-q!4MWueQ8`MG%C-O5$|h`0ityG7M|P!A>z znF;eZ6=K0DrMfhjRYflsW^2WnbOY2EITOM!=%0*Ha}%f}v>N#|cZciEH6 zf9u5ggAK8N^ytpg%fK2^20?&cJh1)L_%y{iXPMUq$M;+v+S^?3#yMX@qPtbQ6u8eQ zFqO&oj?ED8%1NKy)xFNW>!^u5+q3y=@0^5y2b+z*K6&XEj_eM5$w7ykTne}pxSuHi z?7$z9ngjJMTN_q_IM_>^edosI$K%8$Iz)7)Lp*lM%o9#&@17H|+K z@p6K_w*wNi#05GHMK=Ii<6%%u)qqOnin6M){1kNhgASpq5@Jdr)dw+ z5S0PGOQo3JvE(+vQx~oF`Tf9Q903nzxh%_$jAt%4%LQvqMqGe97F!@4an$A{EJh%O zct|_GkF9-P0S=-Jo)YCLcm-9oa2LRbrKJ|myI*&a0>rb*6kkzqs5%e|^1)Owj-iA2 z&-U@OTt;U@!sS~_qgOt$dZt{goLRkG=`4z2I#~)y;%#9dE-c)tvRvETwU{-3_){-P zJ^}HZN;f?q-!wq>AgxL(tGbE_`n@Tcz3<1!XJy$>z2x!vm|KpK^O&nX>^&S&{LF)u z(ooRU!?(jkZGL5+gtW$aML_{DE|SpDc@hCR&I|2fw4VTHNS(x%2Q8d%?lpOP&jZnZ zu@h)d8 ze-&=QyY6zU_o%v?3P0{qf9__pkOZ8))Vr;#D;f!}S}#V|dH-4UDilUmQpbJ~!OFt2 zSj))>2ByIcbjqk|#2UMck{I+K%ZpD=>%l1bBwZr`l1)&&a%BuH4+HWQdN)Kq%c>i? z2)_p&L3sM{!;N)ytB%65lkJwcum zY`%;FVBYXF2+|Zxt4MNt=XZgm-xA0AY&H(*J8Cje)Lu`H`OX3{?fQv%+X?I%-&lcu zzu#ftnM~%fgL~gUbzaGPSbq%hk4VO23v=vcYS(BlS!$YMr7+kZwF=~Etgu~^qix;b}cF($*Ki4F{1^dWYM>jR&W*>PRFAIXW-norK$0`qet#UZvt^ezV`a}=99_&J9n1a zalc&(xD;500!@)XV^x^8Wzg3UdAM#{rO+CjLJW)-cXhwtKi)qzi|S8|O&uSa*uSa% z7Myksdf`^-QoyCaJ)^+!!Er>&#b)8h_iwKb+zhQYu!K z{<-ZP67p_xDR6&N0K1ZM2c7+J`OwX90vE0dR7E&nG3yD|!4C*)0w#bSY6aD9LMH6i z7U)R;^zF;vq?(1(1W{-P3x6eu2X2ES&IR;DGUVEs{_Ylj>w!&yknh&@Ja|{#F}gQn zx**XgagWlsYM6)^FY>&N0WNIy%ZQ4wJ3knSu8lp%u@b-i)?VtbSL z$a6d4NAL$|WTW0n>t#!q_tK5~e2k=~8uq#880jADChq#sR=zMemD?CBbpW65002M$ zNkl~iQtc1xHla?L?^N{XP@G3h20!AW{ z6M@`hyMs^XWeJ%;Ab_jjvDsV>GM*35_vUgIlfN6r2mg3;33k(}olffkkdrDjy z86-$}6KUBi`PKk+c3Qa2XXQ(E<;(VMZrhlbSK6p-%;#9G*YWwp<*V0EoxK>Jn}7D$ zjVES(|9 zSbk%vf{*Zx_bVh?UWP6tni?QL0>DHcfeTOWC@PItg zltexFZtOoTrDr_?|NOC|thRfKKLLOtxBgeHL>cIO}l4Y3b+(- zDDdIv=|7Uh>y(e9%4tPKR*uH-;FhUy)o_nMB#@UhS$GV#DNoU+f zT?)7qxGNL@+z^WTo0@|C!wIIg1B;tNtMevB^uA>S*mU?=rr9S1Id)1hW>x(M*|z~o zW}ytO>QIlXtcQ>e#;Q$>Hkt7~k8P`Kq3Aoe?AHT$(H*1pAJh`Sx7f0S7{YzqC--*k zYOVd+$-#kmismXn2AZUll})p!A^g;~=50+?h?6%wn}74l^y!HteCODK?orT68)jR; z_sr(nL!H%#vI)iQ8#9^jT$#N%pTo)nPKc@?)TDSnHe%>3L-H?#6Sk5%mVlG!lr8M> z`Fw;n*ofh_EgjbeM`v^MmQ;nvSGWghDPoB3;vdvRJ-VZrP5{CZ)TJQn77sMBaxFTo zxu5r#0?k_*qJb;bxp*#6WpM&*Xm~_qMBFADXuetJ0v!98PYv~caQMKUhL+mD8A&I# z<&)xO_g0=KgD+L(6kyRyCX7=C5l}Ww_f;3hpZWZ;a1Fw`-1;}9vf>9S&urPk1m1|| z=I+}u@)u^<&;O#Qrs_c$Bk*2Zj=dlYcUUD@GWechKD(!VS6ls`y>_lh;z&$dJERWY zG8oVAf9k;JpWQi;%)fVS(l3g?@3MRknWIx$Ljv1G9DbF zH`!eF;O^}w&-P}sCLgHL96KJWW+B-RN53zS6lrkfMw$UjRxVVI5pA%sKExv?;=M?f z{lR12i!Q8Rn-8pjXFi$w^0!`}olk0-_WIlJ4o^%S+`DV%*6z->P1TWzGp>>-hJ!(H zbxu$>$RM+=MXEFOK#ynYfUgWz%uBd5aOPM<6*` z8uY?JZyk3afUzgPy;|sHhVC$7%*-kUE;-B1jY`u4!o)QupTSyVhc1OO38ia~(!9-# zZ8CgVW{8SNeY~}>2OYrzg?C)LeuI%SVti5>?~?{EF!PhRX3dQUPJUe*zs5Cp8?h!l zh$#g;94xJEEgTZ|+@^{0Bz4-N=eK_#6}3o}k6o?lKkZFS2IeP&ja`uOC{+!zKVtBb zDF~67q43Z-G|}1C`rH$b-)xz|`C~Gj86F$IaJldFrOVd_1~ko3Rqd;)Gk~UIfDNu-ofm+v{WHOY4wntq%8ev;;3+&tzO@N zrDh3+APgP&+P?33Mp z_4f7uc)UMdP+`%NHO)_LY5LbM^=z&U_f8~-=W?--_h*i5_KU*6dxMDr0(E;J+M)KT z(uL*$Pmos-`v~|0A%##LhX~e&4p9zyc>cj5Zp!7dTiTmFgQGLqc}}V}SkWNZqWmQE zM5qcyvBqupMdZSW&Q>I#G}S@gFNw=;tSI7U?+pbaHG$^#=+MR48FiEFFv_7eQ6J57*zVx4jCE5< zYMc@=UpqN6c<1DTjE6<+5LQu?#~#^(csz*L!}&?Z10L@bD}9pm9_xEq$TU=lgpSnJ zgg3R+dA(j7;uh!AxwkT3s_ia+S2lqMz00-VXVpD`0-j<~9-f%^{_*!;d*f|QBTt^H zrd_&nWpHHVWa}nKc)B|}8tdz;5t%0(5_kccA)G36x~%X5ttb-;VJ5GuLLnnr29up< zlG8$BLYN!(WD^YR<}k}=2K~8tVZ1|W+-lZtGD1~`KLk4%i-lwqG7|gWijk)Eg?ANt zJ27FY<1uwkDe}cEn@@8oNQ|cW_#~H}VdN}CD|nru3b0P*C%lMjXaO2BEiP zMag8diBuXs*{%%?^^c4U3=bD&h00hKuD<%6H&ji3=JCUAo0@&ngXZDdp{@JrQs84# zAdxR7=F=W58K2nQ+O&0j1w24e54Y4m+En-bbOwCv!t6X;Ha)ww&5d*Zu?N(>(WSsf zp}-{s;K}6ij`LJ|OF+D5UN4#=!Jj*_|9>8xqTLkBy?Una@>2(Qx2)o-ccT#2?dQWl zfyzB+nt68kVnWxN_8s2j zt(uhYq_w4G-~b9rshZBf|0HdqmLy1b{A#vt^2*i*X791CNG&P7E-CTQ9Jph&u0xGk zR*4wWq0>HnVM|?v>mPV8v(Moo{$$IsFdUhA2KL5)%&VDeOfJxitXz2GocK|Y; z*>v&m&W$Bx^{KX+-~03~Oyghq+f(1TF_VK81>@NqgtZS-lyx8-fHUCKhY&v7d}md} z5Y0k1QrDaJ>vZjDyK?bNB$p^k0Shw$%u{9!A0mU0 z9CBGQ$XTE#1NVTTYJVsIkT{R%Ijh;k>-+XbfB_aZ7$w8a&q5+ z`oo{x_0US<9l|hz!0y@Oy?E{Z93$z#PVSR0dVCLxvs0S$#E}SXE^{VZhV2Z;!l2l0 zDZ$F&KeIk4{lYUlCldMJ`O3+pqSc8oJmAjF=Kj~Op8K!A`t(0Jy7|rj`eBvC{ZB9L z%q!Zj{oQFucf#U=z_KqRW1aaJ4i*cv+!za}v17FZV3%xnYf!MET05Wd<-#zQVX7w@;ZGl_WY7(7%fRZx*cL6jtKnTA?~3K(3R?NTjx=W^8JjBqF*N>pvIv1z{rxuV{j*@Ji8#zhVy37q*!3=rXMh%m!yd zU>-s`GI(B?956#Qdb9>nc}xVG@dZt<-x9nYnDtc5cuXPcgRZcU>!>oT%1j~4Dp|Ie zk zw=|j66i*??6_b2wp3lUY^ei=DXtuPbVnW-`f(PO9DSmp;@&+tl*a}4SP?hBkl9)#d zT3)|L@FFabhldNt#a8SmGGW`~(Ns>(dz7pvpB6Inux+xSiKFC*UpY@%BB~Hr4LzZ2A5me8g#pfpOu+=!IQdp4hzUKBAG^6_)}o1=fH9=f|fJodVmJ{Vnxt zQ2m`&iMN%XI@t4<$IlFoO~HsKJ)i#b4=(=c4?l}{w>z!EE$33erNDhg0gmUY>Ox&z z(d#2{`%`Tpj3{v%jhsbcOR=)I&?E)^fLPt||1PWX$uYB`jj4t8|0-oy-o0=J+c&75 z={l<#ETY>l!7Hc;kZ;i_Hcnsf^k)t}yS=fq7D}%-t?fa-;EvI{jxfzen#|bSJAfUv zPHLvuEmL+e5yVH+x5_7e=8;bLsQ=QNeHSN_L7vdH;f(wrzkhXWQ{=_nE#K>ptGfPE zhdXOR((iut;x}*1La^e1hGVjd3PqCZumIj+UPCA#!)Ugs2Zw}|KwGl{X;|!ly?-Zh zlif2Q0A7joxD{32xue^>d~F)YfKVoJnY{1-@GsL|uk5^Z!60;`D zUukLB{q(01JL-eHkkalb1`j4b2Lnq@V5j|E1ntBovK2%dH7661N-z(9128y zOJ>@N@(;y{?c4eG=Gf5W{WeBoLzMlAANNEe8-9$yj-59Sd=lHll(&!V53o>(HY;Zu zu7h#Gnh{K|t2WYD9r%BbUmeL5BNBW#5nGQ6+}ZKuOscr2Ehh17cTH&LrmC->8$UUl zi;9R!xR6+mk%%cf%`G$`Ab=pd@MA${n56`5{W9|uMe*9TYhz<$pgf(m8$}*3HjuQ)Uq0`YfdeUa@m%@hUCy`y zp1VxYwW|&i!c-w{dhd-4jf?>sz(Afiu!P%aEH-AF9JARH zi8w|uS!!T$l89w>9>WR(dDvKlsOdn@CK2bgZ3yB5U75MF4@($+Bmw~n!1~!B^l2z5 zpdA?}Q;=Vn7ZjkN*_Y?&aEJz05sN_RNykEPfrGUcsfetOpeO@e=lWX$JSLL|iQQ9- zXG%VDmDD05&&Cp|f0&>>7LB6Gtt zhaFl(cq4S*$48GTFFJd&CpYcB@S+gbPsOfyFg=$>ycG5H|3e z9U|n$HbaT+Ng{4Ds*NSJ+14C~);b=>bNrhlQn|K;Bn5;UF&ueC`7sW^e7W;d0^N}$ zVwtwSyvWGLy(ML9@*nUc;h>M$NAm{B!ruym7r#WoPk^sUVos8bge*uv?J|6tsp=Ynwl1k2r(=J1(X_D{f>&Ep|cKK zpDQCb!2*Y)<2dN7kim^L;2v;SHWZKt-7?pd+&!rXd;A3rmo$dU3Izn*5%fA4X}1U% z{!rzl!%-I4;Egi?-O&gA`GcqP6yC>wZ%1a1;l4k`e+$C#k7J*#zYuSmLNP_ z2s|5~gU5rxq9~)$Xf~UnUJ3XdqH_ESXI3Dgh2N84Fxc7IY0a`yT+k75N)#EepDC=q zP%QaYK)@S^`e+JiN>&G}T0Fw5Yf)bARG8w^&CD2#MCQXyj-~4jvm-k~}& z15puQ7xp%Dk^+l@0XMmdqs@2^0uJ3_NWM;b6?i)W6wL{w4UCjxq;VEFla8l2s+*H$ z{td_JT%%HePZZryT?ylC@G3Sx&CLC+dN8?a1j7o3BV&^j;Yir)hp|d@Wv0nsoGC{9 zD^lRWqChAbYqd>xzyXg+S(tAv%+N2(`)@AJ-CS058TOihVfz9dR|h+4%9;w6^HBe? z&_S$Jq`-qi0W3)d+&+Aip)`}F*DmjiW9{>$jZoq_J2W3=y`vXLQ^mQ-R*bmT~-}bA@l+uC%#XAu7f|C8K-J2q%M=kHC>ql>$CT z7>6$$P72R4*oh1kw_fMBt94pL>SDQszLXoFMe0Orxu3 zD25l1>9OV_0Z=O#H-jTFs5Uu&Q%hMK_1BrGjBNhJ$d`d|1LT1ofiY7BR4Bb|i*YS# z@GJ&Z7o|)nYQU9_R;{WkD*+ONF?h>WdwM%)mL3`%>+IdAN9kG%3GVtrK)Mp~ zI5-hNW{G6k?Ec^ND~K6I3M>=_?7n~c~jPHOWI*lBi>VN&2cxJIKvGT|gRMu?CF;z9@wBasLshJb${^Yf3}BLR+* z8WR*8{g5g!2PG_I_&Dx8EAdYW4P%@?$c18D7-b@$G?Wdgq}d{uE7AN^)5MGAp#U@* zH#}N9HaL|ZRae*5Nf-$_9UUo^NEAw?QLod(#GY2GMo=oog3+_BIJ6xAuf9>&!5W){ z07I~d1KqI*M_j!VY59!<4Rj~dYs(otm#^~f|H77%#$4c+H%|Y+J)z+f>(NKzgF{u( zSS`SwW$h!SDlYxPm*~biSOmNOdIGGOXW9Il-C&4=>j8;}?prEyEOKcm zk$^}V4}!7y=6oaU@4wzM5QrxRrv202@Rl-5wo>Yi#XH==^Y)<{qv|i8T|<^HFsnpE zDUyysgL9@vL5YVqY)EiazSJbKAxg!3L;yQQ6gVvc0Z@V1!1z@>4uoE%QXzrg=c_I& zIB}szol_qr);h@-hPY&-S$se6l}YE=C+uQ@nZ zvONp<2d|JSO?bQ1?frMLx+jlDt-`;1=@peiX)+js{s0pVL=@@>!5!1egD_I91~!=% zC0-^e0Jn*;2)?UP9F&13&&>lCT(VeLKRqOg1)=Qs#3OswH`IZN+&q`~RHT4NfjgxD z9Pms`!);isK(8w>=W!IJb6BDC zjRr4|2jD)3NW8DNpCh6a#1qrpF!%0sF73)|em$}D}cK?&OM4(Zs+ETxA1*k4A|B z7?i?~I^gy<>g5=Z5#~(|WDT7!uyXsq(41eLEmzFyL!_#Bz%S^$T=vnuldr#ByR3aA zvudPY_##zZMJp8dV~>!5%JW7He*$u(0?&pC<`Ia*+z||0l1wd?3{3le)6($8Ip!ky~*8oI$>13~xVfx*nU!Lmw<|EbL-CzA*-n<-!&M`U2LgtUa zrNNw`sRLDKXz-*k9^pdxC6$rWFFeBqGrz_x9%6*2_!We=gwF#0SE=;sA#x-DT1C8D zE%1Km#$ibY>^}8&$qPX5Qc9+v$+4U{f!wBB+MM%n;coB+NfmbOJEk7p9~Y||;JNoc z&}uY!+1X=b<5Is%8S?w(T4qM4j1*2z2SjP9ByFK`?{%*363CefV(m1}!P+TKw-H0R zw|FC>tj$qbwl3{dKfdN`YHRlG+9{$tx1gH%Sfs#RQDDODoA7w>fi`PZX3cV13n?(_ zcGi>}zS`e60VN*#Nbks@p5dx2lj!ICt~yrCBvN3dQvgf<1UZ_C>5g^54Oa%9P~`g` zoY+&6_t|yTi~UN)7X4>8HvIh?AM>97s25M1|Js(iDywO+>lMovDR37Q5Cot?#?)r} zF3P#Tvw%ujMW)*3l31`e&ze=+P%oFsiNFv54=Yj16)Lq-rBo3IpArdl5!T0?=bh2P z0E|k_5{9rRUN}fBOhL{ZPvF!V!FBkPs{F&*#OYTM%s>>O4tg@}J-OkL^&70~AFeUy z>fwAiJ=p_(E%wn;4FDoBgv^kaknz_u72}3Pa^$$TH%ye#aF!eADkeR_58H;mvA5wX zTPlV<{;QLII!U6&g;`LzUowm{cy3g?Y+e(H!p$@wjhq2M8m^I|f zE?_ypRN+n+rSb|xs0y2X)F}K4QcTP&p*hz|(}WnzIb5U!0>EhH3WZ^8${vM@F5oq! zPH6DRq;S-dfW+`xt_aT!G9@QhqakGNfhO{_AocPR$Q!OhgocT~OF#ke#_EklZoBuw zu`^KIXxG)V9EG#RNFk8%u(_@mO_Atf{SSpf%J>xVn_%qCJ z5~pbVAV>}@VUS}?UMCR69d$I2J=U2q*;U_Y+VSk#>>`s?`k+fbuR|(K7EHp8>jHNB z(PO8UtB<6OmCUoBq>2h@Vr}(43OwITy$^UJL3e}^oHlZygba^6U)^07hRIpj%#3kw zT^Mi%Bme82^SLJNAMIPSz9jR&08e<1$WNDKl>kkF?%;O@^uWAI2AVGKI8QMH@^4>ngjOY^o@ z6!zI2pz@^Ced0>rBb7y>iKP2&%!%(KQb3>p=De^bAPzRKIlaZF!9lNoJ=yltk4}uQ zt^Lf}s>NRDW2f=e&GpY8JLhwIh!T%&>Zivq{>>+Mi!ie|WzgrTMYi8L>s=1+1$f{0_M zkX+*~>B9XMBq1OY7t?y|nYNx1rDx5~{KoC&rX2mX^6Og^^Z+~-`)Cn6S&&MkWD`V# z2Xb!$z7QeCOguzw8kOXUwWYQBCL$$|Q^&i<2OQp)Te`C|G@spEUYTP!+%Xo3QafvM zDzbFHJkkI9#o<6Srj{}Fap&CbkniMrs*O)cJH9v95bKJ)8B9KNBRSgIpSv*3^B{&jJV&*I= zNa@6f1Lh zxDJP;~?i__Y#}SJD{?86Q)l{^%pe!V*u?y62C$v1x`u;9?T{nI`{|I zJA&d4l3`{=`Q{1ejKEY$19)znx&U_-z<41>NG1UDfZZ`mVbPZNl#+yaV9O1B8;OOK zDdsF>U#{PDkm_X|;A1H|2D7hnvZPl6zVDIii{c_@G_ zicTu^VAm3jLza5oMts=W|99^kp#tGky+ec3o+6X}y2%zWvL?&)_?nuRKR!;$Al;>Z zapJ;rn;YtKL|c&yNMo@mkpfFi0bPc=Zfi-%-*NHM8NkUk^|i18MTEiREGgXQP*Dc5 zV&b0Fz|b2F(oP%K?Z!z~L8t;;2Pq*uhGY={4-vK8d{+a-9pqJpyhhh(rfaY)GqGkz zS@niujR7W7#2w=GdtdCM>(k}lV`m2|QI$y!YC&9>6duAu4&685JkvE^o@0b=4n}w| zlIXR2MqB|;B>vZLw*u_h-%zl&*a8Pu-4mYw^FhbqUVAV~tOKEeff)D0kFTEWn|i!H zw<=fPSeW7RNB-$x*N2^x)Bdo4%s?gxB_47NoE+a4WbwSMHokecBQZopf|L+a1z{fH zg7|OXXh;l`9vO(G5I{cgjYSp}D@tN`?>+h8l%UcowtZp^IC9IWV{W&nd3&?Ym;uy| zmnV@S$B!DA;iK1CwV(2af6?P~VKnj#8;?^s;_TSJ9_;|_V2D_J)27KQx~ajUV%4@M zx0clBqV^RcWxigkS4teM)kr?^gqL9toI15+eIzN)W}o~TU0+Mj_)9TxHWD$J z3r^HnMj=TqJiuwOxd|n>f4(&G&lg9eyfQIPxS&QA1l$Z0qc}L=0mdVsJBSw$Q{0At zc+RX}QWHdw@!}UTsYvkWRJ?G3M*?JmDTx>aqFsk-_-H6#BJc(9FS$^}yoYkq0iM)* z-9vdRI_olv|5gHR%8H9D8ODjpNws@i5)H>$Op~Nwf(Ls5XGB@Kd{LW94lj@?wsOUu zv7A{iYPAGtV zkv$xCMq<$5kx1Z0^7ba`om6p0Y4HW!Rg!T?U z1w;z0dMj@#U6_6q>{`6ZYxnPhR@ppW3r>*INrt0Cw9uKq}9vTOjEYLX;VRW@Yx+7ERhG&s*lqm54&>@zklAw-op!P8<7Um^klO;x6;7|&Ng|`$02nQj`Ly!=wg`{)@h)v+^ftyj2NHH{l zoADEq405lYXSDM6-e-H8Jb}69JDYrho zMy*$#eeb+|%C@C>Ye{J-E0w^S)6D0Nj14(U?=8u2N8@kwyM1vCG%_an(SE~<0OluX zmP*7do=M}>px!xX(Q{2tlr`+GG-m4H=I4ev#Ru~!pwqEWKM~aHPQLenZOY!<+`M>w z#Of8&Cmy1A@1!y1Pj0EHZWXAst&$ywaNT|*;LOz1rNdSz174p2DtU3e$2v=CL zu(S@yCEase&?wtjD|%&uDjqA z5DOG3uv#d9uNS`sf=GksDK};&V{V@-5YEQsA~IfUQZsPJJ0G{8(&o+P%3rXI{29+j>rP_EFF*#Pm;aFPyma z+~zeV^`bjd-B^_Ma6{EcC(eP%!8q{5z<3GcdD@{6QVpV)FDVOlF1a85(y)= z^ANbhRvHqZOb88eNy;YiBw$wH>^Kqd@wd1@z%W2SgmBA0A!9>$gf`*=kx<1WayOKW z`HR+)Sq#A;!E2H9Ms{C?(!}4DPXY9}TBq3XNR27iaQdxFZ@m6mRaNbpb?dCzSu&{< zNB+_Cr1vRKBV|5SVU|+#ZwB1XX#5(1HhEVb)WM*_D2(eOK?4Echve>YvzDpbQ&_*f z%#veNsDWS#19Rn*XGOFSi+H{j$z4I1=0?QDp(ryhsm%d0bLG$42ze1-$Q5$;2i9dXRWwqDJgG44Z>HjT<~b{ zLZXJl!}`K-K~dnn0it(FM|eI@3MY({1RhVq34Io65~CM3fzc&bSX3{;$`PV@X8f6#pJYFr2(0A%3v0-H~jz{f4a zk8pZGsF-|c?&}}|JUB=y-MVr8p_8YfzGHR_IrQ1WY)SPyCM6Y##uc*p%b!$=yS=~! zFiY9BR&B9LiupPkinmSq1_IF)g7JWjKvNp;&c)(z)HTa6ZsUDHLuAOd^_fQ7C5c#0 zPS)o?@l;uH;ca9UlZX@$DR7q*;J@kQ1O>mly5Yyv6O8=L;j^duM!&zWd2e;`T~>B` z*{}rIUYfhTH2)Qc6U&HEhx5qgp55j76{2+Y_6R7Z5-G4+D1e4T=v2gD{AqR;zIOj1> zWDfG252+Hwi`7B_nNnKTm}j!;I$HYL4!3uAT`edqEUT<4DlXFMw6G~fHnALKR572b zu~;>-|G4TH2|@;rfkg%$Kf)lnj;mrZRoJWc+LZn&rPm>k`SNl#RS(ycHs+YK4RV!C zPzzfvWLZ(|q?HP`v7WP9JuNM#4xhN%)m>DWfA4+7#sZSuEau_I>Bd?nJBL!KXkO7| zMZNoiHzvnKZK#S34#Sis(Lp2(lNhul$KX*HuVINR9{mBp4FhX@JUL$io=n&=i{TDF zOu%%YJn@)d!wom2AW{_J{9yRNijbA0nhB_YAn%AmGC&&SF^5kEDH&-1J7J<}<2kfH z6jOOT9;-EzaA1y^n6%-!My)}M;mZ(cPMY=$sYc+~{Dl;5fdF5TIti4I?+Q`riE}Y< zXcEZ>(z4|aE3G|?N)Hj>A>>_KUz?YcGdw)3aSkfOwPCp?Wu;A^J_mqD3Y$knwRxce zCXi0T{7$L4n)SMdx)cie(`(nYPHUgDOJAPe0Q=Oq!^QH zG<*0w&c5jk#%G?`-@Jb9N|@fa`3hn(kpioc0{DK?WGTP`mrcYW%HVf?bKeApSzkKW z@#>M*P%L`YKK)2l(PB@`^305;%Hnsf^!q$stSJt34>VsLgeM+Rx;nRK#5<7!tDXY& zxfZgMV7ODm6a1cH_Kikd-g5(^6w5C=g>`={@#AByPc)Wm<%@2dVQWd=V{2>PIdYaL z@le!D$Im}qTeiERV77hYl}G`R0;`Y$Siodt>6-Rb8MAd)&kkK`KRYxykXKNYlarHS zHY=4%h$ncl23@3AK2>JY$fSoxy(0lQ6iu#u_>CQrguqmgG#ZkH{Zhyc{cd$=I+G1$ zWl9Q4Err$Dx#gK~7j%p}PFmWmrAa-{+m|%m48iLCZnH|P8 zQBosHF9_tIweZ!$NpU<%N(}K31DOnc2s~osFE4)+N=Qiz`S>}K4>C`X6b~T9gYh6z z2oWwKf+Qw)v__GrlzFMR!squ#VgG|=A$)X9PlrO`Y$glHt1u;#Rr0woLQ)b#mX$)B z35g-(m3T>j6~YM-6O?thePTH`p|I^BAq?aeohO66Vj5hJA;i;1qQ zXX!R~eWXA~XrMBtY*-Nr74$|WN|}P_0wv-KHslJiO1ToHLExigg6A$O3I z%hdpM5Y%kg6NpRYYEn1HDcFcV6k%iPA}@I>mZ(Az z;30%I84cSvt$%TJMCNl@riX`fs^`r+sDIMoFwH2ja$2aK!Fp;}wtkmB_Q9{;?z+-d zT2{Jm$JPv`)Zy3vLx&yHV!>*e2283ts*O)Kw;MKwZIDBRpPH`mc z`TmjNy7Gby-5u9-CmFC@*i)Q)v>^ALwoW)yotSpR6VIOVg6b^OofIRc6)7N6V5L)_ zDL0E!D5!Ag?9iCYA2OzyUuzq+Qyw3YG+~#ZP$*=Qkjs1WYX8Tb1D{w^vDh!Blu5p` zd()YYKKry2T*yA<`pdTtjqg~uHZQBxVidfRE_NI4y8Q0$?(*{TyUs3VTrCv9J~`FY znGJI)LcF*c1pvKNG!uVj$E!{!GgXC<{v%MikqEO1H zb>(J@Nd@dRU1&H*Vu<<#dT<$;#W1%^Z869{GwGZvDnFkt%%!YWTC0Q6Bw7X+U=LORzbmQ8vF2n%;2?QEljcc*VI5BZwNx@w zJDc}Hkr5(b{71MRLpwd;NSiGbqoaXI&+w(((q^e#2}YO*yIccja*8*}lv=V}j0aqO zCo>DyD79wrPKJxQ`i`1%D%FPE1o%8nxd%=dt$8X_G1xxzg#1INv>ACCYbClvsZx)R zwZ&+;wjhJ!B7oS#W1XR(S6#H32_?clUul-Sq9j`YNQ9*++9vcOYDSC@Nl+``{rGqQ zjS`H(%?PuF1T#E+c?Q5tBT4P4_AgIU7Qa>z;2|^yj<>sc^9P5I*zHp$TaVpR5|PMo z;3SRyG)opSk{B`6wg7=BI@9#NLgVnoV=bpza6o0kW;=Q6bT}B?-n8kLU5=4x0&WZ! zpr+d{f~X7vfpGLMV%1;gHO*rLVBj@ba$JFm5NoWMDHn2(PqPWKF-8XR40fCtq# zk9o!MQ9v-OLv#IvC=>+5?q{@^zs5pRMl zbD+I<#Nlen%c>|V*jJJd9;H<*dK*&IxSb1!!|~EfFTL=>3*+PCV`F2tlSND;Qs9B5 zfJDlgatwO2wy-*9qR&1yFf~0sYQHkfMd21+D#L+rxeBsom}8*hT>f+*>Q8p2g zCSnRIrj{i#^b)O^&9G=qS-OlY1B}1vOj^hdA6Swu_=}5uv>+sv%OyDlCZkCc4ntW# zU05+wJTEbfg(O|)J!N^D%o$eVqk$;DzyuKO^+(V3IBf1%BO=645}5z0)~}BL=c#4$LVgRv5+PR5FNI6 zNMi|wsZ=7v5p#vdGdS&XDl-+pgIEYr90OsWO>U`V6-+D?G-+tPR;rK#`yl~sOP9Fu zQ9?A484?pguu2Cdr8N1d@OUQC%!`YER}}@&v#d%6qis5qx~L{Q6pH$t0r!;8?+FLI zfoLcOl`ZTOF+Pw+p^_@q3Z-7IF{lCAvoaR^1j<0HL}by6*y|A3Qz|8p9@><_)Kteq zAQ&^{+eS+a6s9Gp_I@nB*qmdLbb z$7m(#fT+h7i{=B0lV?DGho!@|Xf!J`lTlfrqbZRjrh1~;CX>`;#R-0Ni>E)K*BewB zS+QsoiyVjBDN`E&rs8n0lw;hkDd0eAjtfWQQ1lrX8ZXQ>N+g6`3vYo7p@p~wsskmV z3v8jE*IV0yyl8%e`wOl40dUfMqbKqb_PPFdQ@m%6qGcfQS|bV*_U`ZbO3 zetb~wnarFT7|f|;=dPJpc#VpLiBxnUm=q82P!@&coXv(Qpj0ZRY_?ZkekDV%|4*B@ zswGTf_DIgWmM+#Uv6Bl>2R7M=`9-3-Ef`e1{rZp|z^&G6&mB0Yc8=nkyJh{_&pq{o zHPd|ig@`Fd3WyZAYYJckr&Gv?m@*!Bczh0ju*A%}#F#1Z+~D}R-eC&X7qK8_>5fUq zv995VDvB4|&pCWxJXuk2u07|Y_HLrY<8&YI?0dMfu-sytX`1*~q<~0)RZjt(LcX`M z=x2i?R3!4t3s-hm6{b#!v`@gMGk=$eugkXl>F)LKpKl)-vmfi|dE-*omp0Wc_RB*} z=Nns_E)Gn*b)lW`K$__qwReq8v7LRV3bO0V3wD>}uP@Bb(h&;=%NU?l@b0_s0@r!# zt+yf(xJ{+9v$L16Jz~`&1y&ISa6HeNpJB;0ls6Sl+dR_~uIUMn+u;xUecp&C5y4Ic znmBZZj9YJoY63ZDECceRk?BktqeZ99P-!yMN{t+XiIn_e6}|qdXsXyp_t;GkPC*A2 z|87zak*-DLu5oE@UXIn0OAr$NL=}EKzG$CI8if9hB<{DfdZS6Ew?a>b3rl_DjKN^U zZ2~}0%d;-5k-*BVvUA`=uEhz}+x0nDr*F$4(XCIK2| zB6D8qo+fmM{1f1VTE!U+R8axs46qAEytwdv1dt6fK;i|AbQMX=_tjFjbG6q8qNQUG z;Bo{z;Swy9(3p#Ab``~=uq8{*2&P^Dz;RqWf6b%drIC14Cc$kqt9m;q3D2oA z2_0dXX{ZrFXG=}|8t6=Rkgjp~TNJ3Sye z{dOs@(;Lq>aw#ic>bgXLCrNP_5PM|%*5fT_ye_wKywhgNk4Tm2N<7#Tz$zgV3MZhN ze9ssU_#m85^?SmlCB;gG5{?J4^s-Od)-=@GLopjpxu(8pOE4rb<6)*C8pkgt9`yL` zE_Xkezttz(a7adv2HE6!~#P(dH_Dt76>(n$A2w~$C_xL`% z-1A}I@DZzdU0K17%EIks1^EW;(vz~Stqr)&&wu`Nn~gX)6~lu@f%H9R`d;)wqr<(w zFGM6NodW-IOD(YM6beMb!AK++i^Dx3e9-f&4xl?Y01lNNse*;dffUAo#FOCnPWsy} zLF}X3Ce=M9#ApYpQ@jP=VqP8x=R8rB-deihaX9UUfDSHwXtkka)8h$VM9AWhr$Sx4 zem@b?Apyj;jI?mgK5QIf#AOd+nB0b4fJ-64!?93IsqE^lF#7@NBE-H(kXg0?q6}Dy zhMbS)Vi|??P$D8A4aGr=&gb<91HrgXn=Z{1-V>~kZz0Kt7`Pfo;dlxD^eJ&QF^@MQ z=AH;7W?o4YAPzim#8_m-E2&fFwT0;P;2$8IxetWfd69FIhU%DO&&j-P5=U0)=A4z)h2>^mlLMl6MN#QuD&u1s75`j|*TIX}B zBIAAV-NENTg4qiAYy2d6GU4b;LX+}VBw@M&&ALBqCP9}l@5fM*xkJ{#IVmQGNs1ek=y8gzFO3B2r` zYV8OFAh~GP;GWDKJg0SzfE}!FsQ=1mo~bG;xkCzx$wdl?6u2h}Wa~6p1`W1e{Vq@M zw3~dQXUYTKW4Ae|V2B)G#RGhLY5L^onpK|Vw*$?tRxCFm*zAS=U?jT z9iMc(bf*1(?rTQtg3(yN2Y55jjTokq*X#Y=Ee$W7yf`#A+1fMsA1#;vzehGL!F#MP z%*r+HI6F{wY-Dm|WTI=*abe0C^986t@ND16*^!9@?LFJ8il1InwWqRBBfr@Z>4j=- z$`o5`YwJRlDV8HrK%@XD0BHgoGvJ@rpb`nPGASVT(Gv6vh%5-^2-ZU}mJF55pduTD zxe`M`=8@Vpkc|TWUn)juDj%P0H6D?tz=8OPTT2WnCg=}_CL%F7`kj##3K#h0FYp5q z1%ZuFWP;dEfjpUpv7mGbolrfJ@}jzQ8K&?={9O?g2>3$ra9p92OQnk|a)f*?*zIt?+WwwNISFG?OSr&d98b}{Q&m1Psw$MavF2a;DvVIfk5jFYm& z3f|msO+>N);ZJ&6fw7aA*{k%+t&bACz zU}1n#s!$|u1LNTrCuB-uv_qH*#W_thAPIPI>Cu``+Wi?y>BAd#=?W{jcuZ!_JQh;E zIpl=u>EhWFO+q_3V}r>Wg+aCiu}9B`yG3{CILsZm#&ZYHt6jr*)3x<=&wcJQH5FyE zR1>d63WyX~5(>cCWR@iZKFp`Q{yv-29S9p{48Kl#{1ek2Onl-WRW4KLbyP4Ia=K4< z^tDtLKU7t;*bQvTx9(e4b7pkX?tm5%d*Dj{$bY=?ne}y_ZmO=%Hm@i@=YxZT|Nig) z{%`;GZ@s;}i%k}>e31h869tSa`8T$&{YKv)6^;MwXzNq;<+V8$UpV6M`iSlz7WXn~ zk%_4Fl$s1*+rIYCUw#k%z^2kRn^Bda|MH;i{rxmlpufnaSzi!LjzC zi3^Tt${*|=p6s?c&i9QTSzq(jwbgYwmPKmPwWin>sUophkplM<1t4k&2cyuqfu*lS zeJ@BYVxbtEkHB`&qE>YO6)yJCO6b#cVOmP8CWz-)nN$wJ98tWP6Cit=Cw}8^XMa+u zXOqob3G(}yJ0W4J+EhMBEg=h0grL*|?H;0#gBB%Q!8e%v5OE-pC6dOF^RC456Aq&$ zkC3N|Mn)m+Z0X|FN}@o(8+`x8(__81ou6D^zBW&OQ=@|S*zgO7y3QZEQrT3vX@9MJ z@qIepV{2DXcCR<~;xD=0UiOopWY#v)3dQ2JEpg~5edrL?w3gfd2(6ssA$ajBU8N-p zT#;}?o=b8N(ek`CrJyqhKS_H@u*7*>e3mEAUXcuGcl;&KZ_`R5mxRY6CDuYv2H=Yp zShjGDUt}VDoa80yK48@>Z>En2gg6$~;rNg$HpT5zF-vwq8aoq~u^BT>Slqvh3#Tt` z+ZUnqy2|pW_C55oU%na-d2)x_!ez}-nF3=oy(+9jh{|y=oKVV%t+ttU3DV3iJ%mUy zC*#U^oJ~Zfk$@!VlLkCm-;~_#z%(6Pvv*o$d}qix>QP#?R#rwi?9q0I!x<;evlc60 zbn{khnP?1tugHwNw}6Ei@FBBnBxm@N+C2_;cUv~B{p^#E*H>5GTft&(kpdzG?uG*J zfdL_Du1+&LIn_Hl**E1}TaYCbWb^uMg3-1_B44BaLi3v8$*C8Qw;mf9d%Lr5ePPaG z`#A@M^Grj<<>86{cc_KnOz?p*F#ebQV?X@x8x)Baa|3Wog2EKrYp=Zq zm7RkJ52h~#ZkbU`AW}f2K#BsNT2t}!meS*wu6B>vzW>hAKY3&eF}neKXjm{{(V>** z@dnUv_ti~xZ?<-}b`A87P5$-4mY;rluY9Rwsn98`$TZ^L-kQYFwCD1u?c~U0@8EdL z(Acnj`h4HedHeL>(8TAO8~4|gB5~^N?)_qQnPThj?=LBtqm6Aa7@C@zHf`GUr+@mV zxW0FSh$RRVNZ)gcAE4`w3LGMJjJhwM=*=#+l+@=esvvK?-_h4TV#?GP*X1bHH}jBu z-KIXUkzyaM*sdqix?~`S>0wd?U*ua#42kHDe7R@C?9J>Id75^YyqqgRA!#5)L1~DG zHwefI8HSxNF(d}sh?NoY3wv%1&0+xo7eFLAk!lKWK+57nI@_Fy6aTIV3QXI)mxDi-y-up%hCaE(aV>~wc8Jupo54^XvLI3Bui zWzQ)o0fgYvWx97D!AhAuJ17z0nM0o08-bw_7Rhr<0YLv!3f)X zj8XkoGijEvh4U+Owsn*UTo`C=0bJ}|z%ZSAZsj5 zJ^gA+ThJde=+r!>OT(Ohp2?V@NuJ#+%FzFh{oB9ca5|=3FQ2^lxrWL|Yv*7IOF$?Y z%amCR_y<>HK*UcEj=yu^>aRO`DUbKnGi_ayQ|Bffzqh`&nAbYGm-ei7Q*3p0buyVO z91af*3;+U0c;%H>fQYYIv*upNBIXq-aQhU9h2z6rlgHny6VQW8U4llZi7Gl?ORrKO&K$=|0% zCJTif5rXdUAP+8XA~8e;L1IYknlRWg06HXQf=E1%RFV40QQB)ON!BI)SzZdDy8ykA zU&x@Yl}0aU<48qtH~@drVz?6uzq zQZfjP8eC4efD0RPE{-@hOXG6<<+byPM9=)6{yu3rNrE}ILOPzyUmK4YbQ8~GF-DRa zJOb~X1xt{KFme1);L><1l}LzqI1U?|vy3{uJ+UL3=6GXPIqDkEwuBoLUAfm_a4gy9z)Y%C0ay;M9(^kX7H zHX5emQ92ft(i}`ct5gbeVXj20Z;7gH`mBIT$G`;+%HT&C7BQF)e6b7qL*`{1cG$pD ziJwHG^2`xi15muVhkA&nAkC=4e#2z9b*xhn@~AYLJ=>Z;{lxx~g8X|ZSj;L?K%~IE zQUK=8s*CbI>=_uJcAx1Pe5|IVz^KF2cZZ{%NRG8K3D}QqOQfvPZMU;v-Vq`lSJKNTMwxhQ^Kl{0DYre3dZVCLH zi>BD_b+n23MG9Px0uNOeersRz_ue>2`2%lXByP9J@*guj&OKFKs!FrQ{!~NR@-KFU`v?O)d9G3NiaU3P6hIbp;~+XwVmqhNAPHF5oR@fye0!dPAORe-wvN#Bkdb z5c}x1NwpLS1x7c^)n_DzyeHmd1BfJ!e4gja{*v;rG`X0NVOCCtd;Vrl42dKW^CS5f zz9NVwh{?5dv<58{0;yu=5`Q7EfQ>|{l#Tj{&mO!EA4yV)VbxH8bmI~gz^y7C4_kAM%?9fzM_ks*TqVGpw4Q2z-)juV!Or%eW3tolOS$ z5wCoMV zsvnTV%exa^(M0Cucl?3T#cw7ruMGayft(b0{$4Q8VG(L(*31+^t=q?C3 zt{bkE7$R9R?M#-LzjF$hSMO{twFo4#1KM5)Gj!7kV_>#Ue-CM4qo6m zFGjo-DIiinq`>N>fUw>mTzUpai32;Ck>cXLh+ZSlgs+8p2rHh)ol_A!MR=>6GfDUu zp(xbF-_8CA_t+3YeFzPS;ot_$dh(mM^c4|4D3FrNVBU%*-xCEDCmK8~6N3rRv@*%d z%qU;Z*0Ho#>F=2S3QOCHC8|vXc&Sc2+z5(}M8lc~Ofdsltkj5qD{?^-&s?;S=59)#n+Imum}uDK9I;MZRl7!+ zc%rhR?D2>8?B2S`s3&fd#UN5Zq`8nUO+$RqA9jzADm(X9uNwcHHz;&ys6TpKhoCS zXLm|ix-8TDna%Z0`I+-5fRx{VWOFPUd*R@zpMG>=S6SZfio$tmmiSg9m%?OarPY+3 zo%w#t#rENe1DCr;+@5xa^FMEHgsI7!v}d)OV!O$!r9Z+#_p_hlq=?d{M*f>pPD`}PTi0N}ZJ@gfkQ z^fJY3kpdzGL<)!$Sa1rY3k^{?WOl@W&WwOCB{3v|gt>K-+u5k()ocXz6GZlqmV_5= z@VU~&dwiKfF0=&Zx*)D32G`GJDXPA)E(WlGtoDYuuVb;Nt;YeX$fU9(>LxT z;*|xyf7D7U(>gC1=bYxQc5{QnWQrFSFqP$$L4OOfyWDUoM1@0CZZ2&!+#>PZmt9eJ z=}-h3Mi^`wH${J;&t{nSoQBU$MIO!RTyWiaG5l?M;X-lgxrl){iYQ#BY7)|uk;8F$ zW_e+u8q#gT#Ti;IWrMI{H5J771rrbn5K zKN*k3)N0Mc_bW(#~>k{{RUvt5se+G6r)(BkSi2QpgIyjIT$;zxyAa~oErKAI%(&SReGl~kw?%G^ z-({HUvrKl&gD#m&Uf)pl?BkDaY-*G*sDX}qXr`D|q<}~P3O+DC)mRA)p5s^hUOnFt zq`1$oubXiD2#dg%S}Ij(RBGOBIDBAya&66}!Lin!;nzF+s!Q|#huN@Xe$Iuif)9-9 zEMs-nhHvg@Y9F(|bguKYvzLb_>{KxL>VeZ;eWU;Vv2D++sk#?G=l5=k?f3a0jxCV_ zA_Z=r0-0LnH?}tJDa*&M9E+lIi>cgdm|Z`m6$HrSd;6Mk;_GJz&-~zxBjvyQi9Dk= z?fLRtR$4Q@|L~^Gx#nLSzxbPu-mamEzkU0N-Rb$G9qa0I=fhOX)5K-1Rj}9wmpo|2 zrAwE7^rIhDR8$-~bSOJJ+wb>7g9ldKe*W{HKlj{ofP()1@Bbc3KLWrrG&BTsCo3y! z=gytW`UYacDxCu9 zk?}_9Q4}0Dc$|Jn?%>)7DGe5lO0S$J3QQ*uc5^;g5FUh88U@a%rawX>-3~wgC87z9 zL7@llQ!8$6c{>t}27IAtG_KJr6)GHoz5aWp*Rs;C#XeeTJ#v%wqzl@J%#ENjNzohm z!(S%v(@IG_N~TEt5z_Dy!;DD zC=j>ca4bx^N*qxLHUJ4B0m5uq;?;6fAQXu7w~e$OygJ%#i$-`8{w!_IH8kz2(wWrP zZQXePRO`o`-4{m!o)CBloU+SRlAKb@y1g}pRatVye2cH)P~`l9s}~P<1Ko*6;yAO+ zNNBBDRo7g$X>To9{dH?;d8hs4+iltm~qXtzg-BZZ^AX7j4S8ZX*-@1Bs6h#xJ(f9xto`QRK6ypi4`by${hvN$Xci>3;@G;tE<6sRQqy;EzHd4)7BzqsGb=P0d$KUyYd;2{uL(hKen^ab2@*{|k z;)zp<_dcLTC!jfkK1xU=i5x4vc@wjD52L!ypyB}Sd~_gj?tFY=QUUj(95}IJQp%n! zHFdkgQ_(m)2c{m~@*<_FBdp4>Ho{_uaA~+KgL+W{&<@dO2Cjn-EW(WL^(_m&MFR4< zh6-LhqIFMLC%Uz+5jGOGWSY0GU%z+Pw)(27J9&v))FnO^DIii{K`5}jJpY;IHCdJQVNx8Gr%{!g3<@6LLDasQ;x5PfL66U2n?881eMdp|?#jXiX~Ufs z0v{L~in7-iWB!mde|+l4Pw!Tyx$3!P_C-pdRY*VGR8?WgC@Rc3 zbgupU;OO%QPER{MfB5j0T@?k_t<+>P{UIOjV-sz{i$hrl7fz=WzyJ7;|A_6=cfb2x zpgXvQ3i}U#_`@w*wm`)P0MF+>_c;JOFwur&*jc^x)>}yP=}&(e2cd8whHEMCAPX_E zo$=Ux16RgJd+Z*YCmf8+R1$JzyDZaMm;svv*ODvX4UP3p4R(&%M_nOb7{(IRI)x?Q zSWuOX1KM)s{Da}xGEGgm2iwQTdhAZS2d7hIa;Z7jSXgT{Fe4>|? zGEIA`Gpt%+^kUfChMm2ahR3?6+*4j`&D2_DW+AH1%CF4Q@SxPZ28Tn@(H`6ImC31b z7mR5`ve~q!N~yl-$>@1SEcdP0N6Xy<_fS7X20Tgvp@k60jJn>D@MMBCi6NgbnTy{~ z%%6<25<|?>R0Vtv;SLf*E(SRuf4-YP;yqJb@-KENM7IYMgNQhQ14UujCwT6_)z%L?0lmbdI6(-35c-)L_J-|4&YW^H&=E4W(n443(2-YL$NKFN zI8jSl3FQ2~gx3)q@1NfNnRV5hiWI6@W(eN?;ftp`PxqrPE)KOWVv`2L45~lJ#?R}G z>iX?vbD(s-wzlIvN8h-B??^Bh4hO@Dgob75P#}ExwR0B^bo)F3sIU=N34Bo_-Ds^F z3O)P2(yY-d(uKc{G0(d%93Si$M_x=D@chZasc1Ao&Kj(ezGEQrBlG5Jy!Arj)z_*1ekK$mMiR-JatWV@a=7CCL+ochMQvJ7t5hih za9(fh#h2;RXDE*c6F1dI98`bCZJg0aZEuDLi&T1HL?n)%OuX?HZJ$a6{cJFR)9NrF zPXz+;*M5^YaFp`6fhrIcK~jB!>h^O(WAVWu>MLKQ^+fR~RT?)o8GGU1>5eO0z|Z20 zJ4VCcU;=t1sT8aBBDoHXL#{h#65)mBa%J*{N&JQgi<1ZY=j1^@@*0qHB6-R{M~CE2 z3&|UZnbB!H5Fd*XW(~!Kv{dAB)aJi3KH?AYoSEqGV0s z_gfDU;JMk*swgdf?z7Kg4xMT}fAFJ^eO`a_j;$uM8EcMf=an!uz>Ba?py?RPL~*&f(dIgiX2vgIn~6vT!xsIwVsLLf-0HOJTP{1i{=a;7@3zvsn-p1JU;np%`?r7n*MI%)yYGS> z0o8~^c~r4@2t z+@{jw!sYN@IX!UsSnv4YR3tz~9ri>xa14xRpnPo2&Wf_8f^_%<+qAJ>+l9kjJ*^{v zWa2R*l|-igw$Xw1G3rwk(o1A%(kQ+%+lUkA!LOVdkSipGHMvHM2BIgN2k*Z;df`xa z-{ldXGYFH|d{K%1_OZUUvAX8crahHv?Hq|A4y|82)_vtvKSZ1XUl@j{CP$oA8;e(w z5|7wN{D@eUKfJhrpfE&|k=h0buQG35-BMzRdBeNznKK*tqDU=C!Ne?;NHW+N!NekZ zIFN+^X1z8mHRTNUPda@eUK&Z}6ZAD86^4)&VXYlBzD550B{@OC@`tqjJLYt&?@0-k_<*cFRJ zQO?0v&d61=3c`Vjd>*U@kWBlDzJND~TT_l9yVMK-aHMz2Yx75gv4Af~bbz4jH4`vo z`r1bizH-(x<->=fYJH*Bq=py{gYDSs7cP9%?e+M95$VhtSY|^eG{$A~h616Ell`oO zdGu>rm8zr)5tKXM{@0^@ZKF`}#8*gTQ0g*N1dP|r*D;!@eC7QOg%c-EQSW_7!E72j z$!w-`a|psaFid&9wAagB>Y(CtOcN!pTm{5J_4Lq@7$uQ#xj9sZ5pWAPJW5BxoNY4x z%Bze@#WXe2GXCTQKB%v~PMtUfn1}F;oZLidAuI{Sd;3|Ng9(OU;f+{(O9yy>HMDoc zUwMt2nj#0&*RSJD8IYPnzM6RbP5PrFl*fgye!^^FD$58j>FH%`b}rzfPqx6SBm1>4 z)2bPW&E<~&>kCxdWiA{6_Qx3vT!w+(vy0JON6}~HbJ-$+Ufu?Zoa!}CR4%W#Qsd2UFSDu6^pkAR4s3EVz62u(6+C1plsN-T~F0rJ2N zPez7bB3)V2B>fdcfM;%JK45p4SM}AEUwY;f7~Q8YTsnQ?L@?mrylq>4VIg56*Nk<6 zJilPSfi$-I(#V|rZX%hV6KZb!xqQ?t#XB_^EjsOpL&Y`S;!Aem9S%cxo3IHD9 zI>ZPYkLBQ(%yoW0lEjB@gc8ju_YQA(n{1pa zK$%(=355O5kZoid5D4RM>y{-F~k#*mHg) zo9{R8p-;BayN`n0@mE$GU8|K&5MT zRVb86Ne9fboj-ULwNKl~NoQb13XLMOz$lg7%#_@U>PNAUR#bP~eq#i_np__962o-C zUQ%KR3?(ggR$_?VLn?k&Xh@1oMbEY*CGDGuN|zX7BSN%yAiEb#M$e?dU!jFM6;1|!97_ja`o`~9Kvra}oA`;3qp!2HyR>%dEA z9AhqMJrvehw|#OgAa)cr&^8KC-{T1a>{<74^@c~P^cEe47p!^$p+EP2ThJf5bfTwn zT~XzRLZm^jT<*~ZTjiop=0CY#X0+h=ML2nh%M_lJs>c|PW(&1z> zrm=x}=4md|0+}Hpx_1DU+$eayd2AouPz#N03h+sL z2ldYTbl(8H_;3dfarwD)4pG;RU%Jd4IS#mplSN2TRM#`RFeljtT74qHT$Jo5=$!eD5PkPm)J@<7GHF$2_y=&@g zu|Jo}q^HhZz;4##b~o?bSzcKIzdP48*On?VHIjLq9s;NGM>mP?Ab6`;#(c3z=XLuA ze9?IppW#N36e7$^U_1oQA#*l07*i)$76aEo%>>$^DMcixagCcCLu%K!EaI2Tq}4Tb z%^TKj+pxX>c)l1!3WyX~@f3g^_;2ss^mKjMJC}P0r`^?7%l`V3%1p9uo)_Q)<2&$k z9*3Xvmfsvd-;izIUPjJkFK>VkjDLRS;fG7|{_eeF?E_<7BNKo4%Xcr>roXd$#Q$Qn^{qB~ALZj~du+5-U zH0CdZx0^^T?(hel!O*lnyXBuKCqb%~N_*-|9iiX&TcQS$PAP87fQocI^X`I0Zc&wg=<6UTOJ$8zLEa*-uj zvLvfetYQTV00Knsi?&!`d*AoJJB!6)cR`{93(X}6%-p%<+;eBn+&TYy&S`4@`deqd z^YPmd^*{%Dmyu>k7abI(0VXPqRs_?|Ei)EWW+MI#H8JU(=xiJu z@3Rbajy0ZW*QP78i>V~Bn(gr3({_nMHHI8DjCHa~(%=!uI$`g=I^?jqut^+nOOy}z zWOLixt>=51KI}rs9JB+Y99@j0b`NAtUAY7m?jYOr^fAFO@n>S z!#Ftm`ZYtIwy@j?uW@Lz?Q-A6cdx@TNhoA;x8~%PWvF!WtkQIuGG#W*adQ;CH&W;z zcJngGhGGF4L#kYeNy#;a&_+OkF|9E~@rcqexyCR8(!{h%4^ag_tP%sdWmKj8QzGID z(o=;(uGJzGvYP5fqKc15!HC8XHIc$f!{j|CX_w1dRt~@tvFqBvK-Va|f5WNTo}<;7 z1$uUS2YgiOGSor|zwu<3J8XHw_S|uGV6bh>=k}-NXzqA?XF+AQSSG|I%0i*7dx~{w zs&{^Vwy$Hjqj8|JeN?HF;Zoz7PP^5Ox!!$b>#qB&fbL+EG+L7=#1Pa_e2XMBAObd3 zBNN4J6VA7Pe%9189`uDYM&)gf?a+sPX?nf>^KUoVCtOfa?|GnR@6l>NHEdHb?mYO3 zS`rMLeZAT52sE8;FR9G|zA)4=dinSbx5pz9X4_gF4En8(-8M%^;Sz(~l80_i_ z@fjYRsvW(bDaZ#dLc3Hd`5760iIAK;%h;?@B7z9I)7BY+XSSSd{^O4_m78hkov>P8 zuxD2=Ej{$}{|a@q2O1lh_KuKFi%Y>X=R#H+xX(Xw5AXi_h*m>Q?l2t+1rQ(vk025; za#(-|b3`U4eLw$i($mfJ1wzIQ!DEk*jC5MS>kYpB4z5Bj@gKOKcl3UuiugeV!o&wO z74!#Rf1B^{gifE~)m9MzJRP0>a%29El(z8dykGNwGJqMViHbif~cgK zGnXPqVIe&00($~5hL)d|;<6%E0w@wwGPj8&Btf5QKSLM}gb{$ksWcB|E99^1LK zrdp%bQdl+N4vu5Z?AspC2|Rz&y%5eZOFlutr z@N>&95Ch}u_wIvF$^STUso>OQY@c5ban2C~<8!;WZqLp5`a9=-d;Y4|V*9t>pBOPu zK6mfFTgEwuM(>ZqskgWH=Rg1X^Upuu+1YtZKCF`h*tQRJ=kNaR@74)D92p!3R*eJr z3P5ZygFu?IZXF*Kc*oD*5(y$BR;fEiNI=H*#m&L|o z2;7|w_nNPrY;!r>YJ;-wj>@Y2#o7!tkI(phfx<0Wji)t0M{MJc=-4fzwzfh!gZwqNc8Vgz(&`yG{A z_m#kb4?gtJ1w!U{9GeGK?6sFGz zx3lw}%_Y^jtkTfu_2+HMJSS~zKGy{hsO?;Lc7Z`E;7^$y7yi&ZVX_JY%+~#7wTDVG zi_;PN2N+UxMC<&3o1=9;L5bu=@8s!AR0}~hhWLrm?L{?u)5T_AgCmic47LodGOc06 zAw7*EMv>-d!RR4d9$_3n65bs&#?BnBP^j_)+KAcO6WUaQz6E-^6ncCpJ2py!QR7Je zgwGv7T(nK~g%GyT-m-&(uPq=kdmpHtFirI|nF9V09|&QH^tTQpBoD)9Y7Um>lxD!% zk}Z$92Bw}#{mu%7T1s!QaKa5_8L@UY4f%t9P|gSu z2ZmlktRT153#?(=;qtQWvz&a!_PCyp5p*Zi-;do$Onx3yzXRwFKJZzBaS+mK1P>hX zn}$hK3qoDN-w>X!U~3yf^1y&baBv?fEoHEoDV!*~U;|WM61?a3;ENNyQFG`@W2mB> zcs!)9hx)%OC?K~Vj-orLM1dCRv?DdPmGJN2;1Ey)f|xEUCI8cplF|}J5LQPI3l-??j*V(DiYyH=E}LcbiqGhZfeE-zKcWO|ME z-BYJcgM;tC^WLy&q^^EvZhkK8d$2}HSasa2W!&X>ZUiKBAWviVkYBftPk8-NjAwa+ z=-iBs>1(i^W1nN0$04(hX(k60Q^O*+mGSv9GL2Q6Dh}-4iQjrnl8yhiVIR{%21LJ3p-Tm>+#ozh^qVn{5qV1WHnDl04d&;R^Sw%IRz=}Ry}0X}j6 z{rBH}_uVZmEib(A0>1j5dFGjDzwpuqfamn-(=hj`t6LZ<7Hwn$AFMaWajVQ~kIOgM zHZn166H7&K-Lw6!3Y9iu8wdY9X*t^3+snOf&-pi7d^9>YwFnsPH4k@u9HFdK z>*W%qn3^C2naNQ*qC}udPxHWVw;8*Ti>kAC9oY)SC@qjuiJPO8nyy*IvO+te*n)Qu zDD1*YQmXt5r`dmKRwO6pa3%-OS@`uBu8K#Fx|GMiD8$27V$0(O#CbkZ?LzLNVxG_yR=enX3qwzk(dFPB%t5(3m@8r@-u z2NOa@BeQi&sCSUDSnzN%E*CH%;`PGvCbV@6PZf5X7Bf^JF$ldAJar~CG0F7ygdC30 zn3>p}m?LKIE=I4LhxJSdszp5HAvE>SgG@~gBaPe%!Ja;z%Y&x)dv*}5E_RMcZ*Sm7 zKOw#SydV?ENEhCDJAwlR{eIHfMWujJE;zWCNd_jx(4I1fxJ!{@$%TrU1zp0KkxMgf zonmv@a;#1_+5~FEv_}tVWF6WjfR|{?tmuxOMru>Elge1Evgo3VO7WPs2ckS)|6LN==27B^5fQ-ecRunag-MGc=r|S@RYvVK0fN1 z!FX;_a~Ur)JXmS-{eGEiN@^e1PMM@rqau%u9|{=shArjg+iSMeZmG=6%9-+1ATE!PS_B|p=e8Xc1Z?{?nL=1k%{2sZLob5hz>9w^WYRb2u%)8 zVPkksxv{t^5AM;Th37a3xorsDOd+F^s%(U9fL|aMl^GwJLTGMmPA{y;%->{;)c^wZ zx%@%fjSD?4r&p#BpDA5B?nre zs{_14|IrlDxY0i|dhh8AtWGRWV~C40y0UQnu*Q&WD5?sXLt{we|KPKN?g$TS+~UX^ z9UE?HY4CXAm@Cppti}+EIp|vOG@%8N7_qf9z<4OFTv|mAV9Fp0JTAXfDM`=Ps`N<& zDdbyvu1+cwIxO_Qf=f&|z=WY>c3C$21jesO%NG##24;8n-pxC70Ksyi zRE0)IcyJTXXZXb>OiYxc(BP2oM?Ymcx_N$oC@V|w#m|N)Sj9t>pwRdPoy)>}V$4WR zZM5}uf52*4V}&7|L3j$lFG$xISsj6PAoiw(@GP*0W=?CP**hK8fzpIq3(Hh$mct6? zSgs@$qo|0IxJ`W2;*$Q4t%Bo>Yd6tQW0LQ8!cAsY zR%Ur&;nvFXx~i?Y*;#_wnEx$?#9ih%z;PhOIe-`#e|}(Fxj}onYv7c+e}8H2`WaAZ zy7s$AZ_`R8|N8ES@3wRq-a7|BPPdG69!I}SCQDt{ZE<7aPB{*2Yz~yA>kM)kTurXr zFl`?1%Z+Ws<0Aqde1Fi7BblsFA^J8;dA^aj;rBqN z<`fEfx=b}3_#_SuIvtqc7>f+1PBVDrb$jiT4(!X$$kk>R8(@@^SVgIqW#sD-u)$$< z!#j{%Ep@;ahtD90H{w9VGCZ;FbPf&mkyB=@zk72=fgXlFY(6xGUUvY219CPQV|Sw? zPS2GSZI}MgY#FuzVXC>ke9Nv9*z~|&VRG0CA3?fo6-E~u=4ZO26!^={Q3_1bG>Po( z9Va^Hg=#c^NAZ`k@h44HSz(X|a<*rSWWR+*g)Kn+PCMX4QfLdYoBCSsQVxTc3&=|sd zP?)CJKDgtRR^hM!-NvHQptg=WSH8|u02UfeWdsaUKKMUj9})8y+v zINpC_41Fon6i1%gn^y@WiK<5NbWpQPhA33hxkfh+a@Mp61`HdQ0Nm9{5OF8s1}%Z2 zkXnUcHf%Pvt?|-u;XJh#4%rgQ(hLS(>=cTSNB|Q&ICi65Km>@yA&oN1*rXMsnV2K^ zb`u=An`qRtUcq5YAf8kfQplL;hd*dC1%B{ja-$P4QXnl&_|#{Z%1tbiMB!nK4o3AV z^<1NicA5$Y@H9syA(d{^zj1&-WK_z|-2P!?J`N>89$O%i&l*c2d>WO{xQOgq zE}EyW#{Lag!-I*35rkpBV#&j?Kpc{5gCg1Vn`62zTNw9Cgf2kxh(J%ju0YEyfj?bi z#s9{lQ*key74Li1JKzAG1+!cv6z$$#n~`QHFDW{C_QLhH_N$j0`g(gyE4s=mDsuC3 zl}Z(?A?c!L*26SsM<}Qi@a`$nNcq86yQXef>F&{}yS6#2E&KwcLuB<83W|IlncF6D zjLWR!3dgv_HO2S(gkoV5@q>JZ}?+fpotHE)A<3K8L086-gw-lwT zWLNt~RTA;)$2s@-1Ewi^$C#yS+|o5RX|md;yk4gl_V<453+tJ((=WbI=*7#|HCokk z_w1Qxob$LjU|qMtd)vvACt=X@t6%*J;U(hAaHkvxI1a2W2XrfDyTgvF(wMfRyy(KU z*2u?5AXrzD|L41_@-?bFy$Z5LHitf9CD;<{y6qRg_yyRBc>4eVKmbWZK~y5ym`tXX z=xBx71bmLbJS)^V*X#!705k}2g#u!|rZA(Q)DJ>t;BblL)r#E#7scN~NdU)qTDI40^C`K`e2JoQ4*U zHkCpL+#IFQWI?ZojqoXO@xvxhUPYs4jb zOm&IurPu=;-#mpvh%^kspu>m8E&3p^Iurcn3lTsis5C&`nYZ3+u=SATLIYUJM3k?0KEeaVDDUd!&pY_Z~X8i!n1|^0fb>Y`lbDl@GQDB zt3}*m^f5@u1j?cJL%=KEbO@KkVK81e!DF*goWVoIr%HGqz}vZo9rF1YOh}kBRK$Rr zem{O@q(NdgF*%r{(6~7?F~JxN@U%8_MF8C=40GYb5d_gW3XP2WfA~|TwTyt28Yha*gSR zl7jp*7aA@%HFtFPT>Q||)74c_SXfY0n4O)iGZ=&-d-(d1Q> zRuq?1mY0+j7aG&jIIEp`aF;8>abSaUfQ^CCI539Q_Ugnrha;vd{Ufh6wZGch-99`{ z><)r21jsLlG6+O=!GMGCl9GO7+MLHOw;n9Yd2IWprA!9rx@{@k(gj zx%O8=n}BVgppaF%p*-q9cMul=B6bK+K;a*4nkts4VuWfBp+}jgo`GRR5|JiN5w+c6 z8^@;|>$-%YrhIg1;&OPKPIsJtyUA>_iiG@4yNh?;yLoo#2n;WT{poX!E|%cUB`Ha`$Nb?(gT2}4{U4KXLXCJLIpM#o5znkFsvbsSHHJ9E9&U;Y#k8}yLlA>U z5;hviJhb7_gzXLu20s_T!*)8YF{BUl;>h4?Et&(ksYFsCfNkrP%R1rodVI0qceGhJ zvGjWUxU=|TJ}fMyQdrM0aF_`vnb9(_2R?6L(rov70#c<&AQZ~vVn}w6&u<>G!rR+T zgnYCS+*($%2~*{@_?&iwQ2{6gt^` zpW6#3pV;LbZD#`T8C_9I~j>CyD z=63WdDO<4m!_z|zA6?jrmr{jO_Gi1wVUa}$Fs>9T{9)_zU_AxUQhgUch zTbYr&Q~*=iJq_MO1Tq6VG-Ohvo#nxGW{3a`2Y?4l9(IWqb0MV;DHo>`3#3h-ZNR@V z!084Y-O3J0CXw!}sRq&mdzaG}8=7yl4Gay9kB)ZSxRH^SnO|6#Y0T888`K&NjDA20 z9?qL!J$Pt@JYIdKB1_6I7_gr)S$nJ==rtE=$}?@IgU^i+_P$_qg~;y|d0is6P3*A9 zT{f|6N{kRYenb!UtK@QHZb3m#PFYb=d2u1eqJgOgzB1~4A4KTwsE zzCv-%@txN=H1^9&*I#bxZ0k1>hnqlt;qwj)KkSXgTN%dk_@S?79uXnPxk#;1?a0kM zQc-ZAG`B2WKTYqF{&KF{mc%};GRJ|G<-omLiXY#<FWjaHqfEK0}8qiV9C>RsF zSzr#dnAFn=C4*`Wqgwa{Hp3NPOAg4?5;$iw4^0jCi~}_>7Q}yr`MmzVj$xPG%@^^s zI;C7AQySy|?|W~Ijt^SYhIq&vgBcr`7$35rV!BoXGe;g@C6ovlKhxIGTT-2!#Jna6 z0~i=q4z!KF|Ki2_p4z9@MuG#SW$XD89^b*caiQ0cr9JlIx!&t0zbAlE-223yvbwwt zhVHP8VbrP+Lgow&lb)WCX@n`ri(^>G>kW3a6T6LIOM=c2)ToI;M})$_jSj)!P{?42 zjfKP34Gj#0h6br8H*Gcp+j@R_P$K60L&1j2yy`8e8`nj$Q<&-acx@g27k|xr>IuG9 z8!3~W8-%w%2NSr^z|YDGzW5Szy#?SCoXGK?_+wsOZBl?LS^FuO2Vgrku`57qNP$!i zZ>)%3jJRK8&4Wf;XlrURyVu0sIOR*_#98p9^6afT?FOCh z;O_d8f`S{JotK-gH?_6*4-7S5YwGFjQfaj58R@xsc^SrZjZP<*E5s6rE|`h3%cSWQ z)%2o?K9)g}rQDn?d{zc1%%fBk81?np#>#Ykj#iPO zR_1EeS!x>dJc(KGEZP@&!n$tz+Sk4Y*KKf-1lMhgZ0I9X4GmE}0VabUq55J`lHb7krbPL9~d1}8EL)8d-IR}vhfjr5qY*A@H0Wg4kmB}NEt z%cyN^Xfm(D2*@X{ip%C1AD(jBJyNAuA{W8)8xG)B$UHn{8L=ugGZCKSDn?JyDfXLR zKih?13_I@GtT8CqOWF*jKqA0fXRv)Fr!;-2ee~R0&7%Vrs6saFDBO8Zr7>R*$ST^- z#yH^SD23)RQO$>&ncXKvcVa^$7#U2w#t`p6#I~SnLmaJ)#t_I8mTW#h)n29$OF2@b z9MGnzH0g@|R#R{5NLS;a)(8nZD{4zyLtjsm$>Z`HvQ_#V75rqTXKOoT18%G9%87PE zj#h6}M_a>z*X_GTR6R~h z3#A)KYiqEzo!{6TGNv;UN$gmH4Oc=Doc^TQB(1b(MB z$W$g%jK6OBhth=avvyuk2coH4TRg-CxX@+X$&#Ft}=3}U`y z#xI5ETaPO@n*-UI8Ck~k&E;hsJ-w}+oo!t`9o@ZS!z05(gPm<{Dz!$d(`ohirB|wy zO0`O%P>8Th#+ypO7YIe+-7ENDiL7$eV*cJ-Ri0MXIBY#_vUXT~V~G7rFR}%~5N$9Z z3s_cj2Q7 zLlB6CSFJ2GXfo9iW!%aV zv1+Xcyu|M!J`V*x+135Bbd4uYX+-SQc{hfy3NP`wXn<4d3t%B|8g&USAVM(_NEXr~ zAwDN@P9&o+M5H<*itN)&R$#t)C25*x?>~TVnAfkhAb2bU(XrOf)-m&y?!luq6$eZ6 zl(Y30^KF+a$8mt;z&tpBc-@9v?TrgP6T?$&m--FaTH);2S8k`b_xkW~?*yP4ol%Y5 z*$A|t)JQEOj-EzSenloQmDqk@3^dd++TS{i%_+u0y(Uem)S(J?jtfS5EZxmRhHMRD zddF5tJch*&P#l-ledU82D!qK`{t~u|K1-|8$}Gdyu4@R9#SFBMjP_2#PN;NSUj2Pr zGm8wV8k#$i1u;3?9K|prWe#H%TP)G6W-u}Us#L%s_L$P(7tj_`Z#=BVkkvZSYtu)| zGidC^K`#hbJmV*hQV>qd(*xJa3UfehkQZ#q>}?*lO*k&T+k${M+iovsB{uG&uEzdT zuUxU1rZB1Hbp@IwIA9guB@4j#r@_ciBA*?ppk&X3F0J+a7&`G2b&tvdilTl*GFG)ZD zUW?D+JN;IpDotKR3WN9SPJyWX?J^^7WlEkW`et)*i*T)P$&@&w2wM$o`TIr z@y7Df#xv+co63XDt<1zk@ZI-AcqiR{kO@cErjLAMW8jtFF{WWG;{nWs($ZiK%imt> zKYt;F4_*K^QW-gVgfSXfFKQu|EA-~up|j_R!$~wsrmBWfDX}9{P+Jwc-Wqf`f-k&? zr8Nui*m*#F)!#^Ln8^|mlgPYxZR);S6}|KQ%x z&~VUh=Q;6x5g=L(MR#`A0hbE4cLr^C-sXx8?4>QjB4(DOHV8$?Zxj|hJ!lS6I6xWk}(_JM;(wiK;!&+}>?uG^62RrI4&=<9Eu8XOt#8M6%8 zokK29Sl3S->lG`c<>|UD<%OTAt=ODtSRt0!Jl@|ocfNhCy|H(wbIN7`G>CO4!zhJ< zY`HX7qbkfWY%R#Tt2`ejXe&g+YBUQ3#X3ODon*3)IQodkLfmc&KVY!Swa$he++E)w zQq3)f<1~}W0ZNSBM4Y_@p0cqfp`RTK0eH{{HPI6jNhVR|5Je`@=23WwpTs%%EP71W z$hjZGoOxCzPOt@d=M$u0e|-L;&bQ@dJon%sgG%wj#YVr)j(9vK^W=|DT)fgXc%ZiY zlhsgAWF}CssGhh=90w9O5WA@`b-$t$hBe;|sa%p{fTvhaE7HDczQjecx<- zcvS1W+JE`j4R}|}t;hs|0NPi>>Ky5v z5Q_Mj1$qkSC?&>1gIX_}9JOEnunP!H*|vPSI-+}WOt}CB$yE4Ki*7|{pvP3FDTaE? zQ{(pYe`uDgq@~+(#ZnQ~Z*I;U?VFewnY4ajh3`1vMrGRy_B>FXS)_N{y^~J6QU_<= z(Wd6hhxJ^Jo1^tSJ2#;XvLz}~a6lDxvC(v(dkDKrgT|0*@9;mY#9Jv%UXUQTO3)7# zZkP_jkJ5INA#j)06$c>hifeKP+ecxJ1J`QDe|=$Wz>-sv1`C6suCdljz19&MrZ~IE zu&KTPXbp%@%hPS$TWTG5+03>pCvJ@NTgqy4wV5jT*zCGG*nQpPoO0sEtK3tZRitN) zbt-oiUOj$es1tfL>l;5gb>m`BPDwh9a!h^WomU3n|Lxu<_M!cVmp=qn<1st$+2ZpB zt{iLk+PxqA@}f$oEUuyM-KzGN-e~CUxi;+g`S3VElp%Uh!?7EtuJNwsA+OW(;1>^8 z))!94c3M}=UMw*}A+M$?1Zd>MDP~}R{Q5V6fk9qL5p@F8)g8EWnKv@ZU)SL4e;}Z32)(rY0#4yk{3vl+POLFuQgHj-Mc% z@B%j+`pNU;Vgs+Fn8@UTz8-R=kpQiE;xobeIySr|-NA`B^gVaM(kgW91kdgW{PH(E zot{^-6|M31@Abpg+qD*$%~Fwt06!BJ-m&AMuAX4?b)M51`r=a|07&?V_y{COMY@A7 zjC{iZ8xCmr4VYJYgFbp~rNna_Yhl3y2Yzm6(C-fi1B$zxI;XK5i(PD|&IlGjDF^T@ zfLZe_3gD)qxClS>wNoQwWBn#m&%j{+plNh`d}7M#u-kjO#sR$o(GiG*)O$40Z@8Tn z2w-F*;PbJ#!dp`$6yh0%1oruSJZ~`9>hk)7*zbZh7!UXjU6)~dO#nM>faTCrYVs%o zq9c>R7o1EF3`eOlrWSe?!DXd zjmu8MiZ>Yak>O@CaEA0>B4eGz+DF{OB!GQw5k4;u0a!lZ)BEsqa{@e=-_cewbcqa| zC*uHk28m;sctHu!oroG}0hHh-!4JO91X?1^Bnl(Z<&lh1lC_&;*AmIxAQK?2i$5WW zi&g}JE!?+U01pPT9T?C3`;8g8H_u$Y+CNOG4Tdgu_g@?tztJ`L$-T7?R+T6bx0XvB z2RIHaD+l;OUUpGh`HsR%@3vxh_UTuzO^#W!OEM7Y!EJXBb(x#bc9^=yu!Fm3Yi3s1 zftw*qQ(T=h(mQS$v0XdS=Crv=Yx7{E1J%dGh!u`~x~>f%01wP}3ahhKS~<3a=am_A zH)T$Zw-0xl&%AzZV$=qWp;RXJdVM{OLspA@>;AI*3L}WY;GhMNX=FR@tZX^o1CK)^ zz28k3(798K7kPOCqv?Djck8eX@UER59u4 z%;5vf*f?oy3QkNi?;Q&qKS@8AJbs?f4F;0TbmpP^NkIXNn&3C@@cy7_h&g#S=&+L; zZNZ)%!l$8d8LtOQNy?Y{oq{`WC;G_FIYz7FKm4)4kAB9RummlWp${(vE?j1?84DBY zb;JIH86BmcY4F|6KIwSeWis9;js~Wz%*BR~Wit5dmw9W&wk!~6j8*zlyp@{T1nHfNn5a~3@7e1cMjJe?NSGwO=6V!PcrK4BR%PmEh8CMG6L z<70NaV{*zmg+C6b+wJjrJZ^`bzHq`1PwYyEDnl3oYaWF_KoDJk;R(b5fW&;E0C0{} zECHSbk8d#f(ZKhB2EdL&sZ9E?UIY^xPjFJ;3AVkJ3Q~&khRcj#d{AV6qQ6Ga# z98r?$HWRu2GHHE>!YZ(sp_{kD!wQcqN%=)*-Mm5DFuC!2a^o#B)IhAAggW7ia5sYI zEwo>DeS>yjie$HtvFoIl5UGL~wvnP*vgt9Bvy;MgNzxpI&XJv@fq7@;>r`JmRJSuj z|BDZ={HD1JI}i!f6K>BNS6kagCN7Une0g_Oj#f4A_PBx^2RIJgngjS+QRw7b_myBL zc+d4AQ`gw!_*8n19tt~vG|RP;p6nQC z8$w zXNDgVrB;zysFw=Gti}+^l=OT(6pn7Eucvvazh(H+do55qre$l=b9FTb%IuRaY+#4a zACKGHezCvn+K^+?X&$vg%c)IM6m3So)2~RnZ7;PGH%F;BfAN~QuurzIe;)#kA$9cw z_3J7obcQ#om!Z1#Sa2uu#KmTh3*`Vd4d<0-9C~=WL@MmKI_Puzebm4JYeW2F z@+#8yeXORi3iggOzMADq>ETCqN+jaTC)%ABr`HK!5kFKgBCOixy~VqaY)#MA^QQxO z0&Xa;EeL)tbn@jZ!#!pM&Y`}b=`s>k7(=c)Z&QXq$YW4Jz~>3^Lz`0R6}LZDXR|nl z+sDU8ro1lP;{btpStV%?J$>8pmoK$l>IwOPad^W+ru61-$t>HJ7qJ9F)A)*ksR_du z$r5)Jg)61XFCLFyUdBB9P*5r%SDP5OhcIq-jsh@vswm-o>;Y0$#lzL<#9%x4BM*lp zGT8m_Et3S^(r605ykt^l?=IevdqTOn(JkYY`L)}~=YqkPf5+?TB>>*wN1DA202}9K zF`G&WT&x4h3-4?X3Bw<0dcEMWhy4}{zrD*hG8}Tbfzi=m6(vQ2r@t6@`IXS6OMG9D z_#JHTA&84DmHcfr5lbLCM?AbyF-O60n5A&Bjqr;VgX~CH(G7_Rl*I=lOX`$u)l)D; zp|ubL&z88iBEB{wV<)CQDM9ylnOb13iEPsZ+W~vhtuJ3xF&5@ ztIh6kI$a)*)8(`~T!;(jiipv8TA}QeOCjBkZI-_RWJ1%mM7N25^ic@o5STHV_K5+6opu0Q~*JF$>2!fgm-l1m&1 z7Qz9a8}{YnPiSvQ5ab#+fsd~t6L7w3E-yj+#CH-E%|{e(hqUTOK4 zeB(Ec9$H3$aJk>~?Ki{dPE>K0bhsc7T7+Qe^-I@J?WujRx-@C2^_V%?OHThIx%L{d zn!=y4z|7WyOc>3a{U&L8n@sl5?m@$Hk-^?M(fh|~C%q?#=>+L-CA+^wHa)Pg@H{KP z8@X8cP)%8By0)kw`~20m_>hIkiJYL;u9vjByHDyThQK6=csei+cSv>H&j;y9S4FXK=DZc?Ydy z;Z4E#TnoWEN2!&v`_1tiKhQ8jP~}x*?zqeAvU`TR$9ozFCl8k!GS#>Q??4{655Ur3 z*Qm$t?Yn{K)95=ycp>0S+H?(X+N&fbn8EgNx2WK^g zh^Ds|DnlN;j|dTVn4yGBrx*%Tj=>*Yg0bT`zcLfy&3al)h^%8XJ20o33;^3X#Z@_K zgHj}!rD?=0L%<$^U5-dsUOGYo>wTS(OwVo2l7xFn?`qfeq%LmqiZShXd1ttr&q zL#(!tM9eG5WwvhRSCkWt8sJJS6Bv<@x1)}in-jcxHQ0EaF`I*Ucj~lESs7EknW911 zqz=3N{d+?2O9V7XDnT!yoNWGQKNEcAH9lemX|*i#89+fr1^h2O$!p#ny3`PYXE}e6 z$;jZ>)i7IYNSb!0YrKjM>XWhb9>LXUBhe>4nzO+UHqP^q5)*^_31N8fW6+U&lFibL#|3`aVALOg zIbt|Sb&|0#1c(l*Bf%{IQr!D5Nfj>ZBftT-KX9SPbf&ZKbk7hR$e0}P-V?+wiDIql zzT%uu)Ku)u&s^y^=e-lw*PGg3Y3#T(JZ?d72q2n>vaZwY&NUt=&fQa#lcm`}an2E1 z+%rX77UFTz0ON3$6!XGZ!X3vDRVGnq&sVL9P6KY@wi1t(U<(qZVIgmQ=of@?B2^QS zV!l0MY+eVM8lYp3OYq<%(-L*wvcx$@42-`$dMGI1zkaEa+VKeZCbQ-5Uw-GUYwdq= zNBsj^OSw4baqM45r??N}R-g1V`25MsPu6c;E^RngnBO|yFfcrhmty?b6A3^GXy-hh z588WE2H?R%efl5Bxqqko8fSe&g@=o0o&AX){9@118qDtTJ1*Hr$%*Gk!wb}ECy@-c z@ZyNUO~L*|dt^^B-j274eT)cYr1Y-n#Wg(Gl9l$XL3_S9_x0B9t2cVygU2=m=9#kp zwB7d$bXlt2>%)^HRl{ActP_q<5D|2QTB9nrEUmaE7lCuID>){9iC$lKsgL+0T^LG#U+#q zDVm<_P5{P}mZQ^TsHzWEKwlOH-WegJvWSK`V4E$g&dV<|QYSazPlND#!Ep5HgJ@48 zp9%2@G)SeBBi2r7O|IVyzf%Z{z(*C4WX8-04T8w}d$(-6y#n^5Xlh9eS_hRGfgof| z=jk*|-8Ka64uQ|`a|c($c=VEu4w;Nx9;(^`>mTY4&+mt>fI=@Y^x^Z9&Ji%4P)-gn zBb_;Tka#_qBf7y(Bx3M0Jx9FiYEo84V{nS7mt)#Bu$-qqPk{@0#3>)m7YG=gHdw!d zS5r+8T@^YGp@2bXa>>j*vekL_-bHRd1cz@)N3jkVzhDwrXorq&`a$po0xs&|3EuP~ zeh3%b=+Y_xhX*a_8*hxIwY0{y@dk`ikW@59uWo>Y)Lvq4fC<15@aEvdDhKk8OB@Fl z$boY`Lx;ZcA2jARmgsm>hHZ**i~tzlYwq~o2NxcxuYUHl_H zGXy(-BETDfeka(10f=pvc8LY)1?>tApf3|5zl9>rRs zswBo5l2cFe>WMc07IDt8o&VjhljfJ>wY^})rlaKUXGvCVJi&`PM+}UAdGs!0y6$Jk z&-twmx<(|-#n#SG_YNM}zWLcZch?ngan2XT*_Ew?{nbK(&w#kZ=OO&ue_m+*-wz&) zyRx{`rlHZ-8k(tRwzvdz2YbW-Hzt~}GfeNKEw1?bJ?(s-T=-dXSOu$7tQ2HLM5Yg; zv&*rBgZ#n%L_u6onGm`5-{jH@RJbWalrzWW|4gTw|Q5Xvc? zQN(6qU6tY!B8e#jTo=43{Xj7Gjk!-0UO?AVg^bRHnL~=XIl#!3z$I{joyn*FN6nD~ z2F1p0>mO6Axz<#4TeoU3sGo_OAxByoqHD_~crAjDtp<%IQ0PB(nn5`Zb7L0IV<7#Q zKPZd~#*Y3+qlsf3K=}jvf0&<;qk)VL$O*Ysm`^>GXnrFz0=bCzOvC}1=jo|D(8J>@ z7e*YG!}$FXMsdk!-(tTKER&XmhbNH70B?r`UbPgsZSY45Br=8SSp(ZvTEP9Y??imF z4k6`MCIaMU-f39m9LOiBGG534PrOJ*N33Lu@j&dv*$Db@5k@`qUo-075$G@1rTFc5 zq?D7y0J-c;O>|a+oHXxC-M|Fi2~M9v_gi1lO{aT5qX{|4DY%jDtTG{nDiqQ0;ncSr z-xq(+h!bYWec90icD?5{h0MoA%QKf&1uEe;XNN z1swMs3R2J=0qW_UHs7BeY4smyXO}S!JEt< z;9qG-tI}a0?FC~+NW>NKrBGq&sKv&7NCbgXeuFlIN-~8#QBFVEt(|uc05Z!bp}9lA z&5u9g$I275x7!_`+5S0s9G-8K^;MG2t!H!fezhiA%KAM+eoY~sc-&DlQXl@Doq%dB)+>B^%dD;f?E2c@BR;8RVA*qK*>mcmpr+p*I~T*WaPMZ z1juud>z5Bf%pJYO%IGAGP1BSmFmk4J%;;;II8#GuNfUafI8|Vz z2vG`*7;mX?AjsgC4M<&E?J{IE&_+08_atUlFiXGx*#mwd)d*HIb{+XN)X9t9k!0Fl2CpM1}GM z5+ng6>+rK5eyj~2vDFVe7+AziQOE2vHRS`xn1x4f`a4$oQJ$%H9*o0|QD>Oa2obaZ zGH=mWZv2J}{5$#ZI-Pj>&9Gfl8+Wi5FpO>@BfIW+ll3#TX2mkic}+^#<8A&bMgRTX zKl+n0di>o$>bPdo@^QpF=C*I4 zR{=J}UO9OIcp_K)g$NwHm;ec+iFcyuP7<&8>5*!JcZudn!FR+P7YzG5x50% z+U~9p77PE;e}37>E%`8>ejlA<<@F}m0Jwb1BeqgX^Ss~M`y90>aDzS!Vm-Gh^ShSI zI-EAjG+M{Z`67vW=1H7mz41J{nNu1r8gw+`ghNw^M{n=h1tx9O`&j;l(5@WRibPn) z2mw?CW;SS|iHWx*RzL{wDyD`i0_eaBL+F24~q-aJ-2wL=?Jbm z2Uvf&bY=}CD7DB3;5Y=6Bk#NbtKNJcdw z7QwV63_K;Y$nihif6x{$wiP7z2PKwbL!F_<)8A}}f(b&k43cf?_|Nv_wDJ6Mj4CSn z>7F&j!iDy{%Gi=8sI93#ZLSAdjh*T4%{Dk$u#S&M)kJq^$g-CuL*$V6JG>zyb*=o} z4k`-6uK18O2u3F;fBDuUzI77XZ5($BmoAc+RcTp&Qc-wMF;u;e)Q-_6{?DgsA`l^z zmGf+7yZNC_Zu5I}nwDO>{d-f(Otfm+S%O9kWyD`)AE<&+yJ))AwwSY)pooxNY{6P< zsRuf}05#uQv?VCT4Z=>xC52-NyIMf7QXS7Fj?j{Q;|&Tn-O5E4u5|tpxj2lI-Vt5Z z5A>L@flG(~Cm7Y@cgz^e&%O3okt1LqY@xWG6txyWmW!fL7YToQdWVHOt50+r2eBac zexE8!Q+9cC*6I9}(y*(sve1w}xW{&Nd3Aw{F({BO+tSu)Er2Jeme_n`iSxof%)!GI zT%@Aw;-ag1?9epX)bNax8^8{%3t&cLWev1jFNx^MCNu-XV1vS59bGS7)qNx!K8hfJ zPzEA7GKrg=%WETYIKVjVUfCVq(8mV!^&>C2;lQ0brDV;)pk%v?0Mc z#rz8)Rq_W(Xc^SV2&k9wD4KzwA5yVK!*1loZ2qn^!yy^!zQ2$KR96!9TTybmI1fohtw!Vx`r5?IK`XCx{5zWtuG zWssFY56*8-|HDfPIX*kbuE#+{MUDN3N_&jd0U)W}NLd|4st_Y(vmvL+jRW0CYg)Ge znCHgQ;=u#KBjwQd+QkNffhdVqe$K`0BpMS`aF(hv2I>&k+#xE&A=kgz@ZD17n;$p|Ur^+Wo@?LaC) zO}q|8w4&rH%Ko8hRP!rrMhu78INv|g6+w+D6@05tVzCKbLUV{K%n>dd5Kg)>pb@V+ z^L!6?WcA**jxLlA5_))dLKS#Fb`9vJ-yXdil6~udmc)nTuBzT5ruo`WBcAZWjACv! zN~$-z+SHLkz9usc)vBAD;Cu7w_S_-BEG41ssApO{kc9OL?=pBX9NwZ`QuWv!bV4u_d7D7IfajY&p8#R8P~kV*|yLs!pLu;aEW%z{q% z!x}7)B4`HDHli4+1mla=MBN8hEh9Mog6H;Aw1ew@4lxp`a?o3+heyxv_Vd}X(e4UZ zBHEP*7#e-3=}*!o0w?o%o8X4K)-oiX(Bh>jCPb-&c59xtaqmd$VAbXL4Nrh4Y%63q zm$kZIxfwmBkuy@~w13sW7p6^D&f#YkoX~2u!~eB+_K;G$-V0LTI-8hXT~l4TD$WiZ zK;7;Axp)SRt})nD(bCdcdb{^=leGupn%Is50ali-McK2wA(z=O5c%N7dmIM8;d!&J z99yKT`1w2cxuiW9V9TsXYpe|w+>DMPgd`#4iht*3R_02{6hKZ>-QUgw`#LHP%E(UU zo)GZe*>wnrgRg^8F-8g!PW5*JET)R zGDs&l;+wswhVAt}KzPr(0x*%hA{yKeQEKqEKYIAuUQe4#DKo6&B>4b6+2|0b9(l~d zK-1BVH-A@&F9%-O-b1wBjrl3-73H)BSQ%Q$iC76evU=zUFb@JXPhZNj?ZW6FnqU2x zlEd7;O~wpn1gK#V8NX5_NHCAL{QQYYGN47+4IEwl4f|Bs$Y`)n=_D_gyNn!C{Gq7} znje)v>vgkR_pY&qv>=OKhAU_ViZ0H77tfx?gfFOV<-_e0h}lgzd;den`Ts}9xqPUI zteX-CjoW@QS;ZVbSC9*AE8HUOUJg)W2m!^qt2%Pa??4gA)mv~8=MCMDVRmb@48@Ag z+kZOnpkNhR(@mRSlMN5#za5RghS|N*BIGNwd}*r>^ErW3wPvztJ&t|p+JjKbF`85Q zy3a5n+XQh*ZP4_#5LRm221T#bBLit++HjW>l&pQSO4Rq+<8ApVH#h+pp=IE`xV-M0 z!iIYhQ^U*Z`ts?~3+#o*CUx)Jj^4%tV;}5?GlF_eTb`!l$*kM>%|5Qq?{VO#ClKUh zClc3xz`L-`D+L?wT+i>;UGU;HGdq&JO9aVt#VY5$y3ESMh>HIkRLZfU3C||Saso6Py%Syc* zMe|t_W2Aue%?Dh9S4f#ZXI7E*>^^<;d|!tEXHetSLHu@PhV^w^xo)jSvp%Hx%`FyHDmye?A;(+&BGT>O}_&(j@@UDI()g*qs zjy%*1C26yO_nx;Gz|)yWqiwAOPz4i|=Vj{3XPHn%%HCZ?|GZx1Tm^?L;r**rxAsYE z*NHPVJcH@cQ!Pw!o~Ho69h4Rx)aN7%*>1GD|1&gSn>iATuy?}DMDYpvnkRTX|9cN; z`C{XNxp?!Mx$EDeoaSYG+Ke_PW6tR%zqV>_G94N!!J+iaZrU?263n2L=D)tLHB7N_ zXgTNSgEk3ciu&bqEuy&8j3YOl!K*?8hJ7bz3_q%zc9V#nzYW7^zEd`YbW-={2pbu; zGcw1r35f~U5d=kHHW0%#;V4;m((!(56!8$Jxb`8b=Oi0d}dL!P(?fQy&=jRsc z)c5(i>x@dn>F;Krrv2_+`?xQ^kHI1j;>}RpUf1^Qyw3<7N3$k^4P4DljvTA}O~9=n zW=3hL!MCxR%peL`*sg&TdCx9xVpiv=WhwvSgGDF=fv zSOewDhkKjGj`waZ175$3EL9QFt_BvZ_EpkRZ?W{ixJz$S_Z%m}$5fhw9)9NV-zU47 zAS&24PR$?3MlKY^s68x;f}eMCxz9P~ zm_xhqABsp<*pbzZdC&W#1>fh-88pcOe=&TE*Mu3)YbsU@@tQLF%5b z$p?NN`6?579BO!rgV;a=dk74#WwJh#8@K4+b*O(r-7FwdhmE{^R(s{8a1k)2%r5(# zqfM7Zp2fxk%w(>Uof)UXg!Q}b6x0<4`9tb3{tzj(W+v_f=>k#JncLYvAH#q)dB(`3 zn#Z){RlPX$P_lUnjF43v7<<}{xCLyZP1wh|d1~}EYa3MpMcLrYtfaSsgccd-jxgAQ zUc^s;ypl;W5TvRVK+^|eVEjc(N&S<`@=xEpLK5^glQJ;5mh+{G1-r? z=L8|7^Dj=*AYQFHw#_a>w(#&@&i}ghmB?^^j38Li5;ZOYv!i5wSFnK1ptDHZMxzZ@ zlcMQv;?UMD44yYEGhcdy*jfz9ty(}vx~sl`qujs3#LNhR;axTupa$3B$K0id4-(pQ zN*((NmUR5k|JcQY|E)g2DE{yskJ9lRYxPLlIm(tphiI&Sr&!vC0{XWz09aWyuM zCenUp`rFIQWZrudA(T1$XN{snk(1 zWXTFSHSAU^NqQLMxDs3y{k8x>oNAZVEeNzVr*eIX0>#&EWYsM@bv#Yz`Iqr6A;N@N z=Wwp+;WB3}i{>g4xIdxs*KFcp08dnC!d%4AdasDXfen&NO?smMT>?O{1}LgnjRMOW zwOUT<0mh#IdP;t=30VW>v|q63AB3VQe5>yDm%LLYeYer*<}kl*-#|?a5PB$d8bV=^ zxpIbEnWXl5j|W1^Z+P#Gmy}}3r7S`!xY5CZvUxSW7Poxgf7NFFl;Aj^R^fVKQw0n8E#A@sG~am*!hs*|IH+nzM&NmMxOYnC-AtRTgSJgG-i@D z@V3t(w&EWC^Cf{p9@`XE)(E7BBi0=YVH3iSkpnY)X9h<$m2?11rUGsvC3rrSth|5G zMEZRF=ycvN*4cT$@GZHTZzVIG+rT}M=U1qTR{nqItZwuYgPiRz& znz7GQkwwAj#itzX6+<0}z~fl6shUwAqwm|o`^M=n{@7Mt((U=Vky+Dx)CH{H3l3Ul z1CO)y^Lij<@fJ8z&UwOLcdiemEFrMTLh+ zMnjgima|YHfNRVO#OMh-dO6YwZJc}&x`g&{iPIqRsVbyK!=3E(+eQ&>QGT1Sn<`l>)dCaPX zWA@=}n?q2UE>uq?w5`HQbMtfm9dq?H&)!qINR`s|FGkmqEdmyJezz7SdkxK|19_rZ$oJDi zNcNffL+S$>bbhjP)lp0C=7f3h%;Owr=0%bsoe8gmiZ>X%A z`bAIXuNlpm-tos6Y(@l#z?Z5@4iKznHaJ3(rEwV2d8$u>;2f6s+PN)`y~ccAx>9={ zap3ccxOnjSS8?je*k5ribkU=C-ymFNB!UPC%uzMOQ!XR^VhT_UOOkU zyzv`f_|#R>)v4)9ych4si;6V_BWZ#CQ~|s6c$fMy!ceFM_tzy*_^g2~U-8sfyuv37 z-^RwbY3MumET)s=pn56zKlQGQ=4lSqvq|6^X z0we}WLWQtbAATL6w}IkGc8$APRG4q0F2)m0vb>Bz7*(TCUV0u;VUSf&6qMV4K&j^8 zUO#gM;^{N{rtLH@u46br5OI1s&c{V5c+{`TL(?)XydJfM#lk6N%*L3CBAy+_cr<0e zo9zce2gihtCri^WdR*Hc-9^`@p#Sl?WX59XKcNBV<;e73=Yj&=mS$bob7H8htoxJF z3aV)AVh?FNlX0%*qs%PEJs!^cMJbghGk}cXmU-sK3G>99&v>&FCREsPq81#g?Q|L| zXoVUPQbfLdA89EnC}Qw43@5lIA8bRgG4Jddjc9OyC{@xXP9CU+0;B;xONbSnjDf0j zEm`Tnm8Q304?k_qGG#F}_m{kd<7$Z5NEKVh*4g8mmu!j@4U$Ccpkb6`lI^SoB8fgH zl5INU1DMYxT*%1DfE|wtQ<*Rc64jwYlc0q(yEy4Yfpnd;rc6OWa#@)~1#GN}Qy*!9 zWD}Nix^7NdHW>=dz+^k`Hg*<3Uk2_)H97?x%!04#kNR(bxWJU;4=1g1sWF*lXiP3aiQ58M z#P(tT3FI>z{*WR0xZgA@sc9YTY4NP#-MCI`w{_Hu)9mgv!rtqlXN(Nf-l_g8MTWrJ z*6jA~t|luZl$nnXaJ1Gd0wGXUYd?kuA7axcid<4D=qzvo! z5kQr~C?~4`@HRehFLT4xbg(F0Vx2&W`Y?NJ%4%{GF^S>i*ZWP*|52Yo|5X*aM`GV7 za}CQD0QSpp3aQVNRQ_+k)S!q&C0q5ULVUQGI8t$cRXjW(rnwJ8(Ulcu4L(i56Jt8- z!YDR=C#J7|?xjp7j!V#r$I@++kIIt=0=JRAsYPZz)8+G%b*vn;T_aP|Z{q-BebJHQ z@48vYxo%&*-JT%VeSfZ4W^7pEk-&YkKU2l=L#9sBEs$kDeYhWo=lwgQ$4jkLW~*Y+ zx56m)!uvHOl=Dq;TzYr3^XU7b^;~D)vNkz2=cOLT8Grf|{@D1~*~txuOYvSoUo(g^ z6%zlJ=+A}^t*NQIgo1_wA%VXLc1kO6cJ?wSe_YM#O$)E3;A%p>727*2rBL5u-qTjC9?_JF3{~V7q z7Z$gd`Ezlxu*gh6x2nuLJC!AidUEt~d+ZzpwM8W*Jsg-Ul8V8@tN&u5usZ!Hh%3HDY0RzYF~VTx3*5T7$(x4ZSC>%23(| zjM{@_!Ie|wS)V39#v1; zx92?FzgIL#ol&cLEE0Lf7DIj@SCTr3dt}9=6m6J+iw!?;)QPcO!F?RnYHbTxubrr? zqfz^YENz?Mx~5e}K)x(3YP@}~KZ#6|T|)SlNjTguZ;YPL)U6{m-#^gUkQZ?9uKoc6 z`=hT4`Nxn$C{yl_c_?T4k0Z+fUieTDH{xDRP0h@0t-p_ChBOHz%C+!m)HTxNN5SVA zJ7yT+l1ZBt<^=ij^=Gt(VJV(M4Am$w!cg*V#zRq@qby|XD7Wi{smGKBb*UE_7=svd zUW@!V6c&WQh!97=6K0gL#6XJl_lza;&oJ_e;vV{r+i$$T$xy3gT1*@~ zgZSL}Md_bgrAYtET3>7PiVpkT$tmkrH7MMKQJ_srx+*TFzd}}QwDE$i7go=rE=BOvR&;wDZ>Il&AE~hNcJVqDkZ7fs)6(6^ zhDH&Z%#RmI-8-WFdjA+Zc1m1b+H!1o++Nah#hc4rUl;c$*;wLgcS6Nmu}g_ZG8vj5 z!d9ySj8zvj6+8?qn8jrA?#wUC>g02M{C;nkSNsQp%NHZRv*=av=wm;2C6=Uem6g@_ zx*Ay&^2FBEgyb?VN+I6VY-o>{mzpb^ffX2|3`*&2%kSp)vHkCS{XIDvZoWGD;<|eCRVB1R~*zoNSa_(?7{3J#V4;bTneF3ug17~h5Z&kMJk1SSQ-K5GKDKAb7itvAIM0(OM&csFVahTG|aw1mv=jcvtL zyUamPk_YlJeNz&~LTtgH5UTMU$~W%}o27WwT4)dgAdqYE-paT=&=TAv%?2iVIzH*8 z9sWP2EM6+bKVEl-J)pD^EcA@cshR-W90c548Br(cvV^>%^wQjZAw-l@z`~-VUgN7t zz}uv5B0uT=A=e0;Qol>MwD~W5JOLd_O7dofjV-;P=N5S#QDIlmG9J|r` zOY?zrpP{D_&*5A2Jee6i;yOjmJMUYYma?p}%Hrbm^K(R(M9zMGootJIB9$MH z+h%{=`ExO>nC?d++@sOgXvc5dO{a8R_ zXds}x>SEcrm~6u=K?Za7+;)?ebRj?FD5(k;R}k+3CPt1q4OXYk{vu(9_u?1|FQ9>g z&Pzh|KJ4717`xHI?>LXb;*8^`)yL6eztDano77kO*8W}}eb0oRD*8{JC51`9M z0X?|yX{h*_&M_hwdnlA+md8s9b{DDdrNR3mGN6snoBqtNoGd3hg@N<$6(p(NzhUhZu4bI0M6(5*%7b;f%();2AP zND=Cms27J~)-lme3??|- zjw2DFKCnipYHMMf1GPP2yO1o2BT^rcuzke9yH_oLO}^4H2be zE#Qk*#m1;Q^NWk~ii(=M_ip{Yr5w>@E`Oq^F3&DLb`Le?uX(DgirO#Nk=75{^AEwC zL1g|F*;W6@GBUUn59>uo(ReA_90i7)Vg%c&EhZlp+epf#VhGEa-89{|M{o0OaGa26 zJwA+$Nsfm>X0!%}#YD%#Mm@5malAA}-`_BS_i(fxm3e2OfD->SOmg%?`Hp)W?Po@$ z+t;rh2mmgS({EilM_f<0c!jlH_2jyoBj4~LG=k|jBrOplo5dZ$wmBrG@GsAZ0%`7?%@85kFp2|sz_^3!bSIy zQBl!hVevFP{oZrxF5P%xR`aRqv3T7F*|x*+K`m(#FhK(9FA26$gTYxCB=!H{+^+&5 zcSyb!yHT&@=o+cW0tYw=i{9KbR%&oE^AavqiS`#cpXIdWx6t{U#8JyZTh4j5iFqgT z{eWlrR>sz_2PMj!Ggeee_0H!O_YHq#X>M~>ajY#ON@T#b{2#1zSuSC`E~nq1 zkYO|>95ehZJWdUd-6bI*jg^-S;*gKeA^6^I?_O%IEzYLx=$I;H%H7+#+1c47oJ3aG zRoR-zf+PY%mW1Rj$$!PWYKzo`(eqGF4Umw=|d>P*Iu(Hzz`WPC+ev9^d=;vSbJ8U_u&IMON(OXU67B z3Ft=@wU4ui-a9?_A3H|moSq-mTZWhyCKX3T>+v6-o@HglIU3E#@gR^0&`5;ipqu*F z$wW2f&S*OY?W?;@oK?Qg4_6`gGZ!S(l$YGz>~XhdH@GCRFt@PqyU1z#r0(^*pCl|< zV_2;AQO%rt1(>qfHrax=O*`*M|8TjOQ59lfSoKVP5$nA3RdCH5+{xP+M>%Jn<;9B8 zY%h({{a|-IXeGb!Xnph`o@}N;3S2DQ7 zSLUXmqI$n2T1#fs`C1imvn_cU(z)=K3WZzdV3;)ko&SBcwGgv0wgQDF#L~G6;vzc9 z1@kO?@`T=(rB?!XYMEQW%*0~O>G07@G?&o6658h5g%-A5av-pV6!0rmefqZqyOLm{ zD0sx$Iz(|IQZ*hEf?RyYccoK=j9K;TI9T*NdwF5dP)YLCg%@fbih+0*RS&NCXgm+D zwm46h?^zu+-+Ni%=a%CH+2bS&Pj`g&zJLJ)WnnjUaeV?xpi_!AEGt_pa!*{SBPI?G z5{NAKvN;7P?W0miLY)$PubaWL7rD%{E99+|sH9LBco_-l(CFM30(^RWCaK)rAe$@P zs4AO9c_XO^b>=1!f7qJs_N$YFhatHxX?wpxp?Tc<$>$xrE}^?X3I(oGAYcCQqXZ0m zse}86CxC_HWx5ls3~f~st267%+z@9lLF24G&c&Uk(Q>{-qvl#BUqO(@&Io$?CCh(QqoZ&ZCHd; zU};%PPi<#$d)LOYSBNaPos5s^yy8VYa;bv&NFOTgGIq`sBWB4~o`%DmDC|93DPdbw zHQpO&Vw$0Tp^aJHL~Fpw^hM<{&IS7jCwq*r?wZj~;8pD;=jmqQ$}SM~H>?$v(MIfy=Jps)zo8@QVVOF*c1^vy&^ohqm*P zDI}*BHpXmtIBfZq;N8<{Y3NM%Yv=d^*(lJ(>ABBW%+0osdC^uLUopwBokol4+kWTi z=oa${F!lIPF&I1^9fn)%CFa9N`SpS^UW++ zD!Qh}fUFbEq);R<2oVaZT?mUlYMqVYpUlkG)xr4TDr%Yp``Mj)!q|JThJ9Y~-RHq@ z=PHDl{dCPDi*8k^5m`aiTeO9fv-Dx~BbYCJh0=;YRuFNCm9+A4q+DNW1W_7vEExPJ zwh7xIymp=On+@}JV}d7udYm9k_yj+()&*jJ+E-MW`(oS-Wj*4VR@6f6&D0s{% z3eblS3K_-0g*;Q%Ri|dX0jrS}-PEHDa6`1B0Bi2?cM;w6&}|onz*i~{IYDBGeP4}d zr6ijraF%nJ{+oN06Csth)2HgtN>mssQ9T*?{T`84#2vkY53W`=MM_x}L8nSZ(xqP& z!}S&Rxx#(9LZf`+#Gk51qxX0CM!S$S$6Txs)(9W$&mR(kucvj!mv z1a$~7b6|yX6fFLQw|saf@8jIfhq4VeB%yQKF==yP%5ul5f7LZed}O zXICe|z6X{QAe(=^+DN%YcESdebP!U~@OX?<9Yu0(d>N`14Tvy53e@Ji5|l?rQa+kCw3doI!0T#he$#1`PE&HDDT%g~;#=!eH(zt%0u zKEHUrzPt|d3+fe>Mi&o8>k>#`ZXT4J>!VmC?7DvhN|Majez_BF)Y%VIPhpRii`po( zK(NMjfl*=$2K-Y*Mh2jvgbWp*+p)W-N(TW_DdP9Q>|Fyow1Y!3?N|bvY{~9gf4)`A zwZn>Oa`fC#PFCFWeZz+u%``DFo`ugkIjVyBUxY4m`70eAuB`0OauLzMxak41k|{Cx zqWE@U10M?gm@XHyo93yRt1XL~tut&4w~v8?rzu*b0f?ZSnwM%B?3QZ1gtYW?8}!-` zX?L&uPNRbSQ*%?vd$AL`Z5mqABiN^qVRJ0TEb7j7BHz2>r3z~g_3zgAEBNS)CT^pA zN=@?skgt|Xod)l(_Q*x!`yMkTF#IAgir>S=gZXG}EbagqX@ZBj_Arxu~lBgh(8A0xc!tigj<-1U;FBzdp^28WAnxwfEFt+~yO~9xP z(GbWN(EA7(JFBm3AUO#U8u`Qwn?R52F+KH6zWQjDV=u9d`EFwxhoR{GsLW9mSA!6b z;bD3Dg1=QaSpYZVHn>njKuK6vQAWweZS}5_O9J87`$^V&)MT09OYx*QTQIqE4Ryf; zSuu5-2W?t~MtUynqU()4JtR%<_Djy)d1~w-MVjp3wUH8~u=(v%DkSBY=3ZWNBEAwa z{@mmtUFGoLg58d?W{dWyg+H2IQ6!VH^Ru;hI{)vUMpx(JN3)H$&(3wKxI4R*pttdm zAQ#?!GTQFR%k!L36iVH&HuTnNqpQ>R-Kv7o6}3LudEk2ozB99)$7?bIj&k>Ew&yveZADogq8_* zTOIl~DYaG+!@}FAyX_Y5jXm#<0&-yiUPx%T^>!cHAj)<nzV^+w?}-Jc#dViEy2)Z29H&=_bT^Ew2S} zdzFEub25!zGf(Nm7bdKcAlw?0wZr1kk%uCp*3;q;P0;@cySS`t7s~okX`iu+<_hjz zTJ92iYjiuDKA^~C_G0kF#dt0JC79F=^KThR6fwJ4ymb!X=geT&3ffD zZH5GlaORFS-9pT+ETgX+{=A>t6fbE|MxOOAVt=3AU)iilSUzwG&E5IfztLgOj;G#4 zMt0Wnkc-G-(<0b+evel1CQgLl$19iTgdEaDN_V_?!iyf1k1U zOm8$ljj>0e0qPIgFu-Bs2FhiLA6ZYv?|V+vXqep2&cKh!-AaRT67mri)y)mHs63nV zNApK6VG^)o_67X|{7;{K0iD`rB<;reO)Rc)C5mO(IGuW`eATtJ8K=w9)679?(%O(D z*+fwDqMfb2^uPS-Cg>9Wn6iH~+mb2@>m>(L0|O`c>VqTSdt;;ohW%#i^!{G!<$Yx`$WvXf7gD&?K<*X#-h~B)j2f_lhzyEPG{HJY9YxrXec8z%W7f{|2d+(zg&Cd>+y$fwo(vqs$&>exbT% zYowQp%%9`8yyEe=@Rt@Zk3-DK?VYJ97Q?}*hWPKv7CGH__4?umhQ`C3dfvc!yi|2+=ZVPO78bG4P+g<7tDmuS+iE=N`hoY{MWX33e1f zbq?t5BqGp|LrWfvr4a03_Mrm!HQvy7-=iMf)|NEoq~77I0(vSyzzBb(|Gb9|SWnF3 zw!A+^(_GrgXW8BYGA>16*fL~So3&OZ-#e7R-;qZqW5C(TPP9eGdB1)9b9jZNN+QH8-FwxiIK{I|mfTPw?kD$tzBJI$M|^9>47YrSpup&4 zxp&InOY2_so=rMr;sy)Mw-ptxK}!cq`<}d6yp!K|qz9wdq#A8?$U$k&fGDHwfLu^K zVV_(rjtAKPNb9Ih_a;>TW|l6#^|zZ9{yb|_RQ@+sTDRD7sv3dTyG8Q_H;oChUI!)l z7pkt{CH}lkj!#e$zFGVRwkgK0|M;bbQutI0Y7m=KBj!_d^rQpVkUsZ^uZA(6);lKK z;?VoK`i+t4sqq;1S~$O168Y*<4&(h~7%bQ^J8(RP>Q<`=(!kr6%p=@?h5ya0KkDkM z%~u-{K0n1B|9W5lD}L&0f(c77B_;E`GgP&s2@=u?=(K7{iFqnXWl+#P$-9vZ4_$9G zi<}=!jLlV1jq1haoQTs3m;4$!BgBn#^Ko;d812$hU3XgrBF84;&Iavh4*B&<3aq`r z6`$vDu4D>(=(Vi03el>-`cU$^cN(`42MO8CK17rRE_~Ykd^?YqTX(aEo=AEdEyOw% zkk;mdN2Ms%%2cSu!?#O#ijZ-uNnL-QN43c6c9t$#4Plu_vb1k$u~}>Nhe=1=>u}++U}ClK-J}xo5S}fcJe%+= zt+YZ@fUe+CDI=qvf<$eae9h4wKmnh!G8TF_3@1*F_$-0KY1(JxYaLFB49G&U|Cxo; za*fjQrr;V2FhR7^`9GD_Sv4qOv}`qQCthY{z@N8&q&B|4on0cAFcvwPi z*8yN+l^S6ZHy7ky48G}`dOJ9aQB?l{q<081)CGk_*0pxGu0C1) zucK6UH;m9h_mH?pyTQ@@X=qsXtuu=#KK+f&#WEV&4#uG=X)`wIGApZA=AeT`XiUsZ zv$Iuko+^+mpg-xd?Q)O}q2fpVDC1(^2dU5^AI3xY5|$q_nUxgL|`36V)mxNPKYu zvYH(R{91YjvX8dY`qUk0fmz`Yw{B@#W^|7D>GF#PRlI z7i#u+LD=HBC{Bxtcrr-(6SuFpa;)JG;q!bq-=FtqI;|Re$(ZHm>->?Y3??$)zXqLL zZM;}TQN>0)=(wpx9Q+5)NAH(Y$VRB(=iqt8SKce}(vNJfsucQ6x9np2pE)!JmJ9)h= zhhD4rNNY}_{Xgx!V{~TC);1d3w%u{kaVP26HoKE_)Uj>bwr$(CZQI7VpJ#vX-o5wx zjWf>gGxj~!xJUh}RjY8-tU0f1)+?&Y9je27A??Gwx-o)s6Cv<1s?3h`4c$K5( zdQ`4h>tmiLo(}S-_YM!FTvwPC6@7g{G*`Gp#Ia1k(*{rh=})`o*7BMs?u z?4ew9p)#29v@!2%T|?vJshu7Sx>ic;`Q@R3v1>J+!vGX*2()Hg{Ef#ie8e9Y#L>6T zR1^`S0?G~2q$JGvh?CD*FeI1ki#J02k?g*q*(ai2?1 zQD#bT4dQd{nA%NC4N!Y#a`=EOdu0mMo)D;A{=we*e0es)SLHKVJFgobtumR!L6bxW z!d)ACtB7L3=Q74)o&hcQR#MgHE2H!(nL)3^F#f5gx|N%;=Q?%$LfRP=<#YHfT+CNo zR*#SG9F9BXaoHP62RaI~LqhuIE_Q2Be9mCj^1bm_d8MEW7OLH-`^G^;hi$xsgZmO=11^tb9HX?%TVXlQ>H- zRr$@gZ;v+A>rlkSu9F^_^*V-3xa$0Hs<(3L$Bt>Nowu)cUy7g&HZb%HGDO%o#uo8b zjN4i~9A3W~;{DK$RPVt#Kpl7^Q5V8!b@juy>ww<_^DYzwLg_8+#V_{(gmJp&px<_M zKke5C!YMsJEV(=Hj1(EdQT}qiZ*_qE=&p@A=xDU$q2)|=i~^Q0Q`)P8Rvf%W!*3=Wgk0$aU2T$r-w2#9`D zS*4VuxH5bJ*W49ZJx^G^M$wdT~`myHO&ZC+W+v>d_M#hBF z##nDGJYvnNWz|k5qc%GrT~_QlDmBJV>GMv@8-udaYDQNE0pjWSLrR9jjmPd#O<7-g zGIXrNU(?nzd3Kj5Q#<`}I=!amg~Aa3BV46(_HhK4s}&YSk3;}u4kx^4hUqm>%&cQ^ z;FWVEAdECuBmG>%bb{D(+KxxhpD!NzJF8Cdtm_`kWR^ZCe%ULss`pSuz8zvrbuhUm z1JoQy@+d7=U&Z^#Pkr=3_0w+{8KXYANsXQ5K7kC)hUe(g1^aA9Z8QqV_e!;;e2w$( zP@GhJ&$;978x1n~DRiGCr5uxkdZl;C5h(vauh+~TV4epuXQL73H%f(AqOYLbabVP_yJi;JRSiHhd z#yEc&i)oZuDy6b`+!#1G8qV(ni5~Bn6PQ9We?>T@`Ll@_Hqj<_#T!pi&d0&IihTw@ zRNEwf)Z+M!$403P9(+Q^d`Nn;uoREB z_8`?xM7*1IT~R@CRSPjYS6flbwT*JFx~l&BI4f5RHIs|0MA}HVn|!T`vWBDYNlfEb z^#sMc0X>Oy(m@9)ECC9?i_F}<2^+gSQ%lRf@VG;9UR_kUt&H5Ad0A-;`~%Y^(P{(0 zm6ziiwqum`%&lVU-AF!>&0o92ii`K-@1RK7JRfV99Rk412APL=;98PwmS$U$3z5?@6}~})N6W}W6_>PFF`4X|KG2+c0&@>AkwrvCK@z?1?>N`L5k@QYq>; zN6j-*X%lwiaZH$T{-TfhD5@L5ekdR!_xueIjPta0ZB<_e+lLA7Cjg>D zuwymEUcGA!3)44an<*%JhWXtwc@o;}AMS1~aa?NDey=Zj&LZa@XWUp7UN0C}WdEuV zCpM1mmPTGye~kQ@RFzWxNz)3rjPSg6{#uUOrgBxQCq-h9n^tk%Kpf3`gWXQ68x^S^ z*W7*RRfHGfy{D;1{Cv{)I;tO~JfQ_alh5TZB=YXXQM z#HTBk94L4j&S}!W&kABa*>E~43aMSE^XKh>W9Or%qsRT|RRyPX1Qi5`trS1%-ug5v z7BV~K^z32NrEKR*#6fq}mG~Ka!HVQDem!^JRy1arZ2XGiD8q=}m>$?nw-~ye;yJpw zj;Ih_Xmiz_Syl14#-_fl)lFm`DU*65d5dYaiJd{80UiZYFP_C+tj&@ES$jB(ITbnq z0m?J4xFw!MnW*SxeEjD@IUt1hjbZ*GDsFRCv)1IkgwOf;>W17oqK^9^op===hD<** z0__l54lXR+1p2prZc=w^?W9b0Cm0QnpZe;r&(^peWLaiSBN6Vt z>-kq$2ONj+>}tLCt7b|eHVQxJW>|t#RgrF?_j(jl!xYmvf|TU{)x(09O$#V4Tqg#q2b^K`4Bd>Pk$gTPn}u!3;o+cir< zBnGGN<0GyY1hpai8)J$dN=kx13QHyh9@_wV%)d{rA-h@T?&gN3#mU-QYHd>IrO#j0 zlY0>)r<3JE)6<9A5qu)vRvN;|`)Aqc#6{XZY39C|AVkAWse09-o zBNN;Tuy*BX88t*w1(L;zG#sPR--8xSFd5CtVwyl2UIWXvei2mm7`iE??dYUEv9-QU zo=se;Ps=(7A$*tn^57NXL|N@6bwGWcE>=y|Gn4FCWbO|ZokpSk&fsteFY>@|+Zg>? zU!ii4InYzSEF9XCup0`Zz^{TX45|bx)59_T0DS(K%dqu zAiJIjL>9o3{Mrma;ygYUYRQDe&-+o`gteya^_7oYBZQ9Uam&OE&ef-?vdVURftX}3 zVRaFv$gygsemYyt$|2RQ zq(1t}64)}k`#Y=nPTKv$BYS;ApRg~B_WSz{Zk~N zhmy*35=uWbi8`h*=Suyr+1pk}A*rPhHFO8VbfEsOr4$U$V%9eYZBWl3oi!>55y`>c z1R46em;yzJ6f*|t?}jroPgG2;X9$V@TuVnm?bVptonKILRLF!DHFaIu5d3%Ud_w{{ zJ|Zi1YPh*M>4$X7Z_r2>qXoVX5w*?g+P2!5E5gCe1}!ks?jha6a%;VeJPw@Ta!XVU zxR@q4xMPsj?hS};4!WGExg!JkM=u&+{xfQVJGgYc)ycFxic(wL-qy4fiSgV)wldfR z1V+-2PC-F+__Va)nssW;Q3_Z5XkR+U0^rZKkwsPvj;TymTuYaSE3bt#0`mg>WROOq z>}r#gZNl2v`YV3Qk2dME4hQ#2%F%~-UvE$@-{;dlW9vj^n5#aIalj2wJ$j+Vv0@x* zS%D3Lnmt`xAG@k;%Px)U)?_1KwD^_3b1jCL2rN~laQF8-fI*3!%NaG~beA?@v`vcJz*Lpz;8;Ja=$-i1;fTcU+1Kql1= zi(tk|y`4xg~?;tt&Vf>o) zH;d40^KmT!E`Yg<~Eh17&YtI5JuJ_fy z>;>deWrUvMU8KEWPso4gb`?Osd!aUWAJk^r8->`J#X8k8xNC9k^ zwkdv;UeHWgsL*QxJ4hoLkQl6(+`OJ#b6a92y8=+#L`pik^LEFLCg0|X(zpz^@v7Ywj z!RB(SUD29>%i8TVHB!xxU_Tn8g zWBKr$+&OL|SfwD&{|Nr*j&eeqEP7}#W1SG6X#x9H!p+?+W_QPwnymg8AhQaEs+s0O zIR;9Li;Ju-jmzcJ@Ja#2Ezn>-tvIwbAfO3|X6UO`X`dlUnmc)dfT@rVR*Z=?t@&EhU?Wz~*Agm_!Ow64|liv$vv zFFd=`ldv9A0GF0cB`}kj87>DVSKbSYuEfoymz}_wJh` zUEtU@I^lHIUcJUo=!iYcCE}NW$0gXC*cZDx*`I9Zd={!Xk)-{MrL{$r#k(*X@vvz0 zGrxc~8<9fKl)_=cQ4}bws;ZO;exB$qFJ>o|{Bl~-@_J(Lt55%dM%$%`O1PDdir-Ai zcWdQw(5K=%BmYtqr^y@>CTK^Zk?x#Ju{)#o8MSITwj|AVEAF;pCw2rGy*r8Jv6T@uRqRtLeRplC#o=DB z@R^sd+XvZt85>gP{X=j54oB0#=8&#$*1Cm4 zJojal0r0Gh=RmkPd)HlI5Ee6(ulYoGs!6TTuQIZbRw^^$aqOj?V?$dtHyo~RDXw98 z6e9Cuc}CQ!OFNS_L&7s>OOW2w$Wb2!=!D@Z3h*q$aKYPBN7=McJivghR0jGcO#-fP zE|zly4+G-?`<6}gKKE~rrxh^B47cfVdIKz+v;!2ZCL9JXI`-D-B0J*`=PHN1qm^)9 zZbuNCbRiQFs0Pj;^l!mXaX%<8v5`+e%&@d?A0sBxp#C4Rb&H z<4zs(a=FSgl{_tF8|Bc#P^JzjWn_RmJ7b1c!bV!4hHP(dvj`_9(hb93-Io3~<#l*D zD-IO~h#XL2$Uw2j$0yu8CNJ#2Bs@oh;+e_D1AXJ|L(zewwgJ*H9J|;S5YLI0^{-+&+2^DNxxUly??%V~R^D zZ(^Q4gkU=p-E2KzCfdi{vd$3Yw%(lECTzvu)UXr3z89k|g!8(Y~CPS!@ixvulZJHaGo}7>?j|^2=$GLu2^mB40hkV`_3N#W4h*4xcWS`;pJnW*# zH4)Mh@T-{{E&}4CJiS(+p6&5H5`GR0BYpWvS}r3}160FwU{%qg)4n}kV_F*@V+%eM zWBK@HVZmtvJ#Ji>=qP3W$tIWma!Wq`QOQbUQH;(14rje*NHM1)^wb3IJuFRDrY+hh zFph+V;Ugy_N>i9Q;ttW!(POmS+-y!Qy#Gv!9WhJ!iql9*&W+Xs`K@22I{K&K)D3+y zoSB&16$iSCRnA513(eGp=H!!<$mz%EHusL`Sdr7@^TgDdJP+5vjmqgpzG>3 z=KMIn0WnHtI{a;BlJ5dgJrsQs5eqxGxQI<98GGFkK8M;~CaOZsE3nP{z~^*eCMPCl zHbV)dQ9KEDH#C(jT*T*d)XBejTgraVPM{aJjajjE_Rz5t5>k{r4HDk6zm+-eFI{9& zh~50ydde?de5u#4wo_6mzM~z|zPssuA5GkIQ83msyY=cixbp*GUE~UP6E`WYNs;uU z6zLb#-G}`=|7YooC@Y_oBOb!rAVWpIW&{OlzFHava?iTENSfE` z4KEDpRyl6=dFtwqo$CA`U*0^+_P5W}PKFu{5+y)({P54LkxMy7FxrRWXe@jVP{(R? z^i(}ZE2H$R!c)ZA zYpK6^;DF=bAk&TVU1LCu9+%bYw6=O&DvzS(VKb zzd_ikq^xqLQdgq)jP?kX(a0}6lj8M`$b$YmCiKV50s>YCQE09M7$H(b3~~D`ohV;T z?+50;h@p`=greCEPGodi?XNIKTJ6pm;$Sx`x$Pb$+Y0vXdUF*G_@oIlpB(n8MvQ|=T^*6mDO!RPQck)G_H((xBHTFqyRca9 z?I2?0=L;aBtxH?Ca6#qP*3MR`{!Gb0s8eqT2rSj`WJ-`=U%NTn?`W$Z$M$iurwI6f zAeQU4aL#%+ZMHln6h4 zy#rb}{o7~d{C8R)HQ8?duV4b_!Y|yvz`5^eTt6CVf0nm746b<+fcoa}wq0$S4LUk! z;n&P%@7yJpq}+$Q#J7jK-$1QCBO=su3*-XW<^7mW6Zo5k$R=(acV#^A;Vxe>@na)c zt>H(xx<(?ZW;Sn@E)8Ua#wja)t#-A za4_;fSloLmk(-X^`yl7zYvp5ahmrn;tMl`BXOFMp@>=n9J7%*d?HhIL70hW}i(*P@ z#aNk&hEW`iv7;dba{eH+m@3%uGQ-WkL zco$k!p1XVRxWSSlg{x9dRQMD1F)ofv)gqzo;UW59LFIZ8?Oh8I4r-~^9+f((Z^!%N z1Hj(%P`h5jD-4umI`%y7743ZP<(}lLEhJ(HbW03Q{^hvB#kKiyl^Z#s%L)e{Z0=0O zW5dD%i*YCxnpa3xu94z-qnAxIRWK$fC@3nM{uM|1wZT$6r1ln@gwAZEkYt81#9?p{ zg=$R6#m2@|;B6Q-(d88q?3ni1lr%Nlk&jE*!bE4KlYwV!;S}LG$H8L5>BH1oTQD}L zJ*OZ+N9+1mbPuXXjn7K6b??S`>TC>3zB}U~j(s4J>Wuv4Ft0izmROU;h7`q(^`2Xr zE#PS87}((AR2-SOb|}JPz$7R;4hte$AwV>oM?y?-!{Au?8LQk(=4xN_nY-5(*&=j; zU3_NKXMj`iA>WQ_4iFCR@2f*5=BYZ|Y0*};%Bg$iH8=>YglbLL@e8-4n zSF{d@?;OMlQMgH$2Za$ih>r|71|Wo(0n2W>5B{%w`k|0<@wC25`lz{MDRjJL55P!(G>dYsq55^ zU`8y({;)S&(X3|CQGKsv%Um?mAXfDra1+m^c|{QYrIpSD;5zzGAA2RWXL~aINS)D& zPe_&{CxTb89~Bc2yZ<0o{d2Yf3Q5Gku}JEDTA|!SnUvGwV^IgFR0QbA9{^dsJ$yK* z#(GkX_m11*eCNmDqsA&1)sCcsTbc+<7*u#KsNp522(V^e;Ey#7Y9yoE-;O9b0^M*y~_x<2>(M494B{4R6(N(c$;NbNuuih3v<@P;N2Jz_5$2b9v@f{SDFpXDorVtbH z3>L9^{l}(er zHBV>Xz04Hk1Bld)Rz-^3OO@{yC*x+JrjJl!zzGj2BP^16&+4dezXdP?$7g0@ToEDQA*!ioE{s`I@NpN) zMgAHSySEKrCQIbwFv9`D z-V1E38=gw?D(dr#d`K8LH755VzOD`>$~3Pp{b&j%F*+w#EU#2SccOg*#X$!7sTZcDKNakHJN{m87@o+RdJnT_$FgYW@Kz z7~dA*Y7x(@!CX2Q+Gd5K51q2pV0J?f#wqPq6L5YxUUmk!8BQ#dJM^Uh8W90!db*FlQhb}UQ=tPO}q zg!Xp29%}Cz!=N1lVXBrDe$37^2nYGx%DJ#cB>U{n>PnE++?LF!#dc*($(C+vx8@kJ zldQn>VZef*z}{YSH#!dsI9Y$?)A-{_Bb!=UqB;O*VbTB3gfrHo3K|QAd}y$Dh~bPd z_7>&~AA-m~I-r|rC8O*U4hR4Te{puA5tYvU_Ws`9{$3xRM`c2uDL+FfPZl696ZSa= zJ6T9JFe<}IO~>nTw5ak`ac~1qM>k%$V!7U)h=huYgLiazC@F#cA}VjY!rfAO zdfMXZdYqWLgz+_Rp@^+q-T5iGeFlj`FPA3%&>MeM4B4jOiuqn%tZ=75UghD*T3S%h z%)o$E^~%Y)cz%O-Bd_M<B8K?%KUn}!~RWEkyy3HGJXjg3G3cI z=qqbR#$_h&tfV0xMm!1=kIM-hXU%J$?PNfp#`3kFzn|ON)W^ohhh{#avBARESvBPbJBGM9m9aA!4?QNml?3J?LP6YOHS4qHR=|8mj~`T@|Q@D>E^hofCn}x;olAGCb0lRBE$Dpc#;zkBV2*9Aps2 znKm2&!A|N?V8cqvL|%x9m4?HEt$BIP)F4f(^=GnJFj~MN|J#TgX7URP`l^{|X{9w? zmT*^OZDltT^OO;L~rV zB7MK=_*R-gAI&RbG~Bg0T=Ww_G;ViqxAe5ism>Q>bjY@=aA8V5NliK0-_D*ZQjQZy z{b~G`t@J{rBuw}G7?ODTNNo2d2)^FlJ|-ZbS{#i>V0LZ|_VrQWM;bt9dUm=P0{2?KqHi=A2m#*i zX2n;Wp7%Z`Bg>*s*nXQeX2wfi;>;25y1$)OC8N z;FLmDCfz~!`dvg!Vbb!z@R?Ad$PXNHRp>F>4)=RwT}f5cR`28)0X7pppKV8Bqby#c z44>1D-gbt0Dz&c~-+9zc0TDxDip%ShXye6NZWWhOd(!u!WF>BakUC-s;%aV_9vfyB zaQG-Si57iDKi-u*F~DGf89=k%Kd{$3RCQO6>ZS}%&a&|5+v&a{+K4nhcv^Lu9(t0Y z9;Vri01iHKqimf%9EUTr5U&+N{M2qn;} ze&J<%841^%zlS%t@0q8@i1%%`@Unx;i?~v9E+8W-Q%h^Le@_{!Q!?7)@#x?M__1LS zp}%=-@4VND(wU#5f%KFuBYi9#$vqA0Tl{9Fn#wHGVJp|Ew-Ej=CKttGq^zT2=mj|4 z>bJVRtJrI=Pa*m5VH2WNc02r7JDwXl5;Jgm0)U^n60-ziexitQ3!x+?ddCKusj=Hs z9x6$tHI3{fqD=bn$p!lHJ;EKL2CKo0g`#raZuXH7c5pDe#HFgrVibu4RvBxmU1kKz z)6o!=%|wIepbE@A$v2vH%0+Wlj_EtmGv*!Q%AnYk?K+v0IJ;d?U($oyfTqmNLh zzd#v|na=@2@X^u18r9dd6#pzal#1tUiOV;EkpJ?Umuc_Vd~ehA}BDeV17Nl9FDY|9ldm!!(HD>ZDeSN zKg|vPE_D#TkL$?(GDrlxq%Oy}B=hV`EA@`f&L4b$8JB z6_0bTY2&=0uOA*B20o}LctR`weoI))o54`9=?_`(y@K%4bjn}t;pxWV>B`;TLGNPC ze(^~blpzYD4gqxg|qAR>W)ZK@N&%=k;*TYMnn`^IsGwd|iGCU*P zBE-Q%!$eO`LrP4D6G|ro#TUJYI^I*hTuQ#htkeMn$|Ay`HtS7#7wmtpZP%n>g@)q4 z3s0W|^y>xnxvMIWEV0%$!F14S)MF7b(THrW?$cGpM4TV-hmGfxlBwd{RDdKsQ*IH_Gu2t(ko-J%NU#0M{8^%Xj zY%TB+O{2>lX64{0V#k8>4d^EJ9p(^dK?)Z+&!2WOX=@wSTsGi>s4fY16ieV?Wz8dG zJ%SvLU5@JR0`-WxWw#Z?F&&aD&o;h5Ro&T>$3X+{XM)4F(`ZLh3o<=`pY}v2Owlm^ zRj;pW-l-xPLIp!lh=#?lLG=nJ5Hid};mWx~V)@vUE0DgPw=g0L0Tm5BZFi7UWaj?lsWg0qPj(3S=X&xspIYKO8M6!WnP^4 z$`#bmu{!k}BB&-@@6M0YP7Fe}x4A;z`;*ej7F^&k%zrn-1iYlU7s}LacL(6JQD*Ma zxE(Dh(CGm;#a3>qBh7lr_;_)nK1Gf;>T=o?G}juj(K#7-+ssD<`&E`Q$8(8J?92eqpaRgerN?b5OBquCU&@t#K3(om zZ*PZunDyMOCyBZq%lLKNJ#Q^dp=^(w!AuH(;RX}?yt(Z`3F)5Q*!Vh%!GpG!rbB~% zdmVv@xSW8}$;WOlXzJGVdFT6sPN?UI<$pGgE&<}JhgC~LUN^NPx^F?E2!NY~R_f%q zB0c`%eE;(SFu|^J3Ez(_KMpKKj)*O9MG$48pJlA9u4YdrW6dz>S2D=V>%c*Xen@&L zF&d4Ydbqy-l>GEEO@MZt=+}m z+PZqG`0C2)MLQ)fj&<@S=;B{Z0JNpcjhK;xA~E{JH<=Uw!s>=V8Ek_w)$VtRQE88yg+XUE=xp_<%J8WMui{>wutmkU{Jw-(F5D zJx?_^OG-;oar1s8h(;2u*Z)pT+PUEl>fXIk3|}VTudA~I zOo{pVd4NRc-`dbM4jm+_yz(V9Ffx)vtw#JC(6$=DGQ55)ig%x;!6c?1`>K@QKRgt$083@O zXh|{ZzQ4bFczDorisIwrL-=-^FXlWmS}`zGR(jl*Zhky89^<7VzB~80d-)tfgNoYz z$Lon0p@C6QR0C-%4ltw>4ut)5sGqLQg=zgC+;N~Bw z;`9O*5*ph1P+zIt=5&4R2>$I)v`K;l^P^O1Qrm2kkECsV8%9rDqujlJdUD#nrg+hm zlS4v5VMZ*_)3NiOQr^o@{5TUO2eOiFQCwY=!?STKlp}(gl9!it@ z>x1h-d0t-MPYOIM77(N}7-o1Ypg?dtgr}Dmg)w4w%~D3KzhCCvOw!c}Tx>z*bO6J6 z@*n*V^eW($08X}s)*>OMRXiz?ALF3{+gWMksI~VigXD4S z1lw=q&&8$X2h@3+m(Ab3OL80h^WVVYpV;+Bp$Qmx1`t~&Vox&qj|2XnmHwY{S3MM< zUX9z}>HjbFQaP>th++C)qx`>yWRV02Jo5KAr3mOB7Q#PGg+B{h5MTgE^A#UJ`q%XO z=YH_-1+7pZG@gpDnON}uhHC$pc6;F>dHD>&Q(y~%{rgJzUkydFq63k;YFcp9WBt!{ z`e!TuE?e*e!4R+~8T_|X|J$kmmzDI#Wc}|<{qOSsFQ)$g3{w%n&|v(z;ZXJe-}jWK oPj(!TTw+EXkX}+SuUKv%GuMT$plPdgK!BgPu#8ZJzz?7Q2Q*eF82|tP literal 0 HcmV?d00001 diff --git a/images/creating-kubernetes-native-app.jpg b/images/creating-kubernetes-native-app.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2c766841e5fcd9a7f0a1a96bf5069bc209d62158 GIT binary patch literal 324356 zcmeFZ2|SeV*EoESv9H->8=;6I$&zh~5Gs+Skja)UWQ!~_C6PUQDHKYGDQkquUJ=K-X=6;QwE<`dxK@8)w;LRsM$pss6l zo_?=H@cIQG`~~hC5Q`s40?O$Z@31^tA?PA$GhEk<^z;xH%}ouA&guVZ2-ii2n>TOJ z?E?TW?*Kn@LtTl>S8ODhrU7<<4`2ltfg=u1{x?rwym)S}%s;<>mH*cVe&`qPK>yLb zvi?H<*8py37k?+P@fSgIM`wR0FA%N-0D4uYn|=WRz_0@1rGf%(f-obw{?tk-!!|h)` z{yRUsZUln;_3M%XKiS><&YFYYe0%-o>0`17vxD$QZ^sL!Aj|{81MXmp{s#Ya3p{HH z!h7@tCx1P#F987npksG-IAa3Bq982kedEGjKkngmuDcr;fiT$jbQbObMiw9p=B0D? z^D+Nh`EHv&^~>Hw$YJ5AsR(^_IKU zo)mC`@OOXDOM5)*)w>kprniUxMdKG}zQ;2-=1}0epTS-mK^myEmp=&osu!x^=5Dk{ z5Apy#7vN@b7KFjPP|E)`Z=EVkplMTd!eL@f147j-0KOk@Pd>#(QdvhA( zfnE+c2j~H300|%vOpbsL-~qUIuhgu7zy7#m05|}CfGgkv9Q_;S&m7jj?s$Wrmw*sp z74QaW{Qt_Q`|GX?a0|p!{=)u|_b71v*WFvca(Dti!5X{(6W|8;I}n6jz&wA}W(E8J zG46l;{sA4=|Hy0LUoHN7U;eG%S&p&Lu?Vu9Vv+q<4mjmFRsJevucst{ zzxTx-d^!HrTY7)>(x2n`XZ?S_;|@53R3`sQ34H*qfObHepr4@i&>BDjih;g^c0n8V z;J?*x@M~;b|EkUW*T{K-t#bd1=Fj>ZK+Im9d!y@rtmN3@G4kI^0Rmit1HkzR==t2d z?dR_579epN)YdK%M&3>eM9=Y0|1)>zjQi8;Pc;c_c{SU`xA{ud;fQw z3Ap-|w*Ua^KhVh(fcg`x>tmr)px>=O%A?==VgOeZE>KeT1LA-bAPctWIG_q>0;d3d zz!)$EEPyM3J=o7~;5_vMf`BmKE)WI8fvaOWkOkxd`M@in6etI3fO?<_Xa_ohufQNM z2F{f^U>U#ycz^_fK$svL5MIcB2n=!%A`dwZQHSV2^daXU=8!9pYY=W;ADtN8Av#4mb-L4ZCUh2bSLs~o zeCa~yBIy$89?|8|y{4E|}Tc+EF0#G)n0Q3O#C{!J)2Q`J-K(9mnph##e zGy|FkErZs9Bi#?3g04V!=^5#H=_Tlog7ft(y(PU9y)QkIKAt{{{uO;SeJg!G{WLw6 zp31<%Aj%-mpuu3oaGAlCA&4Q0A%o#1Llr|CC_8fuct%D>0Y(`{HBk1hFnTbCGR8Az zGol$kF!nOeGHx?5G3{rPXVPXeWpZE&V2WaT%=DTG!_>z#$3$ReV}>ywXEtEAVfJQ@ zV9sDJW^QEeW1eRwvFu|x$fChw%Hqs&nh1#LCSo!>Y|{!Roh9pTfp1KJIBY!C&OpN=fM}pSIXDThvnztKf!OsAI$%Rzkz>B04i`u zz)0YRK$1YYz>ok@P)zW&po?IPV3}Z_;Es@p&?zAop;#fb(4Y`;zr=q1{hs@i_t)$n z7p51M6E+hL63!8B6J8PF6VViL5{VTl7a0|$6O|J+7Y!B77wr<=784gU6uT+*Sgb{C zMO;Aql(>g@ns}r5B8(TN19OL^!7#8T34V#w5;r6sNwi8}4~QN(dm!+@^8?)nNRl#= zmXZ;YXvuLYRw;F<>r!b_%~DutnDlw+FzFKM(Ss}p)epKI%skj}aOaTBA?riYhiVQj z$Oy?8$%M$1$b6UOkcG?o%I3)q9A-SMe%SMH_TjH`baER84M^7KUb+qK@w1R-bIfV#?8ih4QX+;~wWW^3e>M@mLH;xq?8&l#@ zGFFOEs#DrHE_dAN_>YhZMtUtM{p{fy}QLeGBc~sL=vsiOMOGe8@t3Yd3 zTTa%j)cxNoo6~z@B?rMcpiM_l+-EbQ!h`=pFVur#9lt1Z(f8teGe)zE zW>3t1nk$)ynRi^`y@a?_daybH@XMW7 zgs-??d1u39V`cNwX4_WZHr;m4PQ@3v3@3g zFa06@SN-1w@PS5lSD!AqiP*YleXlN3Eb>m|?0v)g zB~hGFeo@2Gn$fur=pVR0=#Ej2c^pHHb&mZUcPuU=jvVh4|2aV^;ZXuD@p@u+l4?@U zL&k^R4~LTB$*)qlQ$kW^Q_rVXr%9y6rfsDo(m!RK$jHrP%?!+(dUWp5yT{UxA3i2$ zxn&JL(S3q`D*80$=}xwDc3;k^oU&Z8+_+rgGxulTo*O=|&O4Nsk0i-T-d` z-z>eo_I9Y;q`bL8y`rr0VCB;)-m2JYNcHV%Y>ivZ)VnM1dTWhqo9i^|s@^NSFRVXM z|D=JxA*qqIF|v_{3BhcCxcOnF$+c;=8PPn}a=B%o)vUF<&7|#PyFq(P$LS8tN9~XG zpEN$zepdfn)2Z57-KEl1^+n}NRkvz)b&p!lyRRp|*7a)jHuk~$n)=W5w+|Q%bPiq^ z>>aWk`ZjDgJoU}_+rr3=5!~plQPS9*?+o8#$M=nAOo&e8O&*>^PpM8dOrM$Vnz=MH zHtRII{KNkTc`ouN$Ipy;*nH8#@rC+DgT=2)wo7x%HF+D825{>{Rz6I;#O7q=(yUUvYUqSo9_K&^UFAB&X`$-Un{4MRF{{{YI|F(An zv>|{p@N3Tu*ZTkfZ@_fu6u6}T)1C=#aSEVQR{kr2n7zyYy7KW~0EFESG))LJ+NKx) zFx3G7#g9fKWz%RB(2jwQ0Kf;|KlX%s)O&l-XD9ZcK7rsnd*ArY|`9h9Ddk%^gwl?^1Q+6T}dM?&{|5;U5rqD=7GO$i2w>QPB@#VpCGn(lat2JaS6AvxwVbo zA?yCf_mio&SR*mt^@WX5I7%W`h`K47Qz%xv9VSn*Cv0q* zmPfZ#^9b#DZz?6;PKY+6c#eEL9^>@(VvGJL2_x0n?XNzC-9R4paY&_QjuqaLBqCo{=p zvfwT(5AjcgNd)gls0EPcPZ;VTBD)gIv2R(H2IRtbPhh&>!%sV@NF*CMsaNqI2@T0= z=ipn1X~2(38sLHM9tCqe2uwru z=3`V(PZKJcc<`SHU(tZ5#c1k%IQa<;Shd_8(}&?W%ae9D!A>$Ew@_|(Vdm$jFdr#A z|3ql<&tpPf%}XLgUoj-_g1z*D>OigkiP+r4{#ORikgWGQpBe`CQy&cghv)a`{T{u) z3+Zp!`z?FFHRpf6?ETi9dn@Gcb?>*;^4n_p!#MqIbN-)YbM~=ujomY?&31Z^ogDeP z?Ngc2F(rvEM#9LKupO61;z7z&uA6r!} zJFX_TO~38PbjouWKAlQEj`|7%caeHDphNz1o9iPDUVLDQfntTbc%q1Gn^lswoLAYR zYFNL1_)!|bo{LMR-k|}z!Z7MdL-KJ{yD1S0ZrR*WggJj2P`clTLJyWiXJ|-#p9pSH zkCPs6p6B1H`DaQP_7e@*4`1K{S$<9fX46oAWjswSo5jp`HU9iBR0vx(u%!W~_tAh4 z4^buW)^|4*YjaQZB1}*%GGu2O@C;5hh^7IFpk?sSd=q;!z)m{@2cI4-t!YxjI@esk zoUrbj{10sXOF7g^7t9tn7nz0Wck6zS-T#d!{GOS=rBGj@AdBYYW~|+ z`E7~)wlnwUz<<6XtdFj*4C5}f)ZM77E_|_Tg`M@k{Z`a@>ZWMU%X;RBf}$n)-lpp( zlvqAT4y3HC-E_pNyiyVV>g!YYCcX6eQ;%vIux@qU(+#!3Uqb^z>g2=N+cR5}zN4dC zbD0-^(11Y9mekc3z9=sEkBj{@VBCkoJnW9>P+y-l9n9?~z^7Vz8TicX3eTG9^8x=x zg3x+@Z-W2#ZGySR*r*0S$-nl%)c@KZcz%QiyhW1}!42vM)RtX3F%FT5_!q>NoI|LO zefAG1lXp-*>>Dnd>GSX6@Xz)6E<*kXC#H0=D-B2qFQV|$06iPBUM9tpavnJm_Mc&Q zrWHx02plqc7AZ{w>if4SXP0fk&Hlqu3~4acW)w!g??nS9(1V^~?BumSWG{C^)}02d z)lefnE6Wt8=gG0cC{ii-)LwX1p0F%Nb|+R4QA}i^5Q)54`G<@Dj`L|RFPzr*TlRj- z-v1FY`#(?i`hUybZ`u1TdwVwL|L5!8z;D_6EqlLZ@BbIs8~iPM|I=j84w0Z0K8*nu z!4O)cbS>!wY-04wcdlvrZOQk>1S$NkBZj@46HZyNd?D*^bjZ~z?5&shAQ(_V%1#^2T{mQP4F2Y1gPjmAxEna`vORWC2n=w=vDS15-Y~D&vWZGx3tSOzE!j0HjWC zstIZ*TR2!XHkTPSwT)!o%u1;PH(bx&{5K9fG!18(prt)EX8C@&?0%ivxd{J{Vt2Rs z3-dhdTpt`sz1a7XSMShGgHRq#iu`6MeuZF*6Kty?D!iTkCQNJ}>zOECgr(P?$uFTB zdjL1a`+V!1-bvKrn(|R>`+RK^GuV1X3|A633C8m49upDvZm&s9f{2AJvl zXa8FpeLHvzQZm@O`9kt?P#hF~I!Y(!4sqD~L%KwuBu1PDc%ts*@@?nt*Dtzkt4*ok zjph;!-jxzlb1n7K%}TAfrGz$XyDFbs7GhQEP-=4!pDUXovtT$Zwrt#JPD$0`nuXC5 z_)F{UShYUf_DxBs;5wjHbh2AN+}2EeWJK`7`GRkt$AaNLpV=wUiS_@O#I13;R3m)h zmMMk96UT|zMsKT-ZvQ8^$wC z74YD~%i_syIRfHP?9M`Pcsi()Gp)CM$UOoz!q=6-9~oOFyppXv?=FCnzsijTP%`atsttwpk|~ z^++fi26bpdYL`wAB?0xrKbH!I(&YU>57?0F{slq3fm5a$q-+_kkwF^C?WN{eS#(cQ z)n5)X0t`{lzn2bCO}ts!=!*IzJaoU|5#=Q?W3--?|- zH=d7`x43ti@(8ID^ZK;X6(dr`LukzN`7&yb3vz8ixYA}K*{i78uR;M;ScN=81DZ}( zCKnRM^agNA)h7yd8f|ptx18`{!})_c`xb@k6Cw{Qn3BrhQPam`cAoEls_t{k@9y&G zP{WJhkWbqjHoo*q@hN`cb?iCYIGK2q4;PexoH6kdXdzvBI2$7!r>*U%Q?;SE%8v`W zLTJH3v}C4Z&Q@5-J#!*G=vbG_DGnVTMlhnB(5!C40U?Dkj->M|`C6GGw;jmGxxdC5 z$4j7<>iVBla>{_bMm%s=OT8U}zMu64-spX$A(Yv31${qPxLy$FbU_U%UdQN$`>7wz z;CGqxiuK9S5Nk~|^$QK)R3WKW)V;l0;1u;RBt(4+-I*H)<02c9j)dvG9U`kKE9Clp z`H^#sE-z|2OK(VT;NtTOA%+G#*~THA7RcE?WjVU8n+!K$>||dj98w<8YVo`2-ZU zdFtC$V}*4?a=FNGd2T&-v;jJEjRs8R3-n%%tj7AAu+jj*(MPUH#94bK(n(~uAVO&U zNxQw$!-kZnEAWo&#lfp*o6JTrCPz@NnCM#q@M}>`Vf9Yy@G=zU>T!$b`GZg2re9gA zG<--_rBqSHgGW~`dxF!<@d8;gReG#0_2UCb|2DdOboT*z(Lm2+82F$);4n$^6qnb*|!ascIJI2 zakav*7sq+pxJ#7Q{F@$fMpT*SjlXst zbSdpAtyh|9;B6y1hg3)3R5ea=bM$FaB|sMsVr>3r9Aj+haYb;wI+>rw7@vEu7aJ& z23K{KQ}c1yNwA2$eo+F{P|<* z0TY?;daqBsd9c=NdhE~DceDMU(C9{9qpH<`Z!N~4A!tYc)!YF{k9$KDYb~nAo%foNXzv`T^QD^1XX?8FIXjmow8n)Yf7B8 zphC#(KJhh4IQR^=a$LO70G9J>FKpWzGY)G(IC`QgtXsT%E@*_e#YVO>3Y(%`#nS~| zh>5>`KbZ0Mt4|YX`Rjd{6g=YSRi$uF3#Efag$Qx)@VUDd$klfp5U)qPvw ziOwz4k)G+H7zaF?E&sPuOh5hmbJoH==MbYXyiTu(US{hAlWE0-t>0%hu@Pr`^dUyZ zP>Vu?3qx;0CW%*t9Q1`jds7#U7gvj!pv=dq+?iO)nuh z2=DO(?1gCGBf^|dC~u^uxTTX_3!S*ZH4(IvUs8{Fq68YHP|Xa}w@pTM$E>{>S3f2q zH;Sf;))Dq(QA#`+HXSZJvD{jO|6+7kVLoK%#?5kYC0MX{&T#wbTIIa_rG=X|Fft$f zZ2d#S62i3+#houLYG?3yPrf%`aL2P6> z4Mh(^E|rygw$!|n#M@kbT(-XDMcxx*;vku4E#!&UGvKS4su09^R5d@!56}#cD`es&#c!UaXv!G*^dJ$Ne(11l#r!xeoIvLgoI+% zQ}pa%ui!D?ww(|~G@yb8#BRd{!<75B#oUc2XHLtiuS(rt02UZr?bFPECboSoIrQDR z>M+@dF!!7+Sk+&qQex&;7yYDCKtAb6S%?%W*xCVGIHQn@Gil1OKh`(Badcg6^`cRG zhqJb&spS?r5;K*L*=A}+Vu z`ks#PfABwH?v_L>jK~JOa0vVCeF2JFY^M`q`xBbxye7CJ4fq- z1}kq~rQ3O{ke4CrEnM+PPD05`HgE=yGx_e!wj|UZVLDQhTjx2rj8&V~d5aF?CVbA~ zXyH$~&7&wvU5upxp`<}l$IA(}G3K%k$goO|WRg_BlbyDAVA+qF%<5`^^}ryBMCtuU zJi)MYSPkq>z0|9oLboBzm?9^m-X{S+$U65nJU!TYwFuX~X0lGOwuhxf%=0>-&lcF1 zY%h4C@ZOY%0Z+1dxP*ciwMZqe)8f}eR4a<+q?V){_^{iNgJFP^~eL^Zw#&k(zrLYH$`>ToX;)N5nE9tg%RC#=8m`^e=qnp=$Q zfpe*tKpeDVcdzX`iJaGdYW)-4(Q~h^=fl}GS8Qv+0*A=7SRfs6$AwnIT#eE2QZo_-GrXs)q8KnYiSw>h}PrI&|DBFeIeEYbr z<}ql}A4HF9^za+$S{XdDDf4{~8tt`-$tn)fl4~M#1LFKGIppi1G@g6e9@S)!XB06StW*$>+V@_IdUH6RAc)__8pbI0^|WRM z;dyFAx>C-6!(WbmMqXY>P@ZCHSygDeA@TNVtkO>Iu=Z%9C8dA{IHNgwrb99O!8((l zK)WR6*`D==iap29jxUMJ8!vSfex1JVza?Tg6tEi7u5DuS$ka#n^Ba2Cr$I!k7s#8` z7JC_-enZBKX)f;ePy9(465lQ6Z>Ntv>nXg1unTi1Yuy~jRk=KTs+$HZ%|JH3$#Kb*#(VSQ%Fv(=UM<<39!GGd7l&Yv$CJzBOk-OnjO4s@?48U9(cz&Wz{8rlQG1kDZQg-ScZbt{+x0 zl{DJfJfVtP5rdXl0{w1G$_gnXD>!T?iEaGC1 zSq%@VG97p?KKOyagvVFlGSC8XGEq7`C*D7RdmPO!zKomWY<{%{vu`0GR>84yhmAG)qhJ5{ z?oy|%(da|yC7mTWO^t8A%r4TyKVA!(3p&0!%9r?Hqt)W-{dc`e96 zO`v_qWl5I?1ohQFnT4$$nC^+RDG0g#_PT=jg~$rd&!JMNkcNsTVFpZ8y^3qWn8LvQ zkZ>cN!Xb<-#%Fz3>?jg+m7R);<(oY`_RP9SwwYo&Xc|4CT!Etj_e-@*9boAk6RRaD zzE!c?5+NbhUI-29=gEyTTd;yjqK2{SsuJ!N>^_XEY<()sL;3zb#WLwrlg%ypL9fR( zd(wrAMn$`FXNFg$rr?nqr5P{O-!-13e4J29C|s92wlIQDt%4C`s|p0guO>wb2Y2AY zi=y@fXP2Z!W*fk`aon11n>dpG29N+sU zqA7O6XYwxyn^^g^Te=qainC1Rj&*#XN;Smx+xLT^S7)*>f0F(-^@oyKZD48@HjSzL zR^=t&>+7j{3)B+yiP222(D!sSksUewm|lgd9=0~T=fWL{h>p6%ghLwF>uvE;gwoe; zjmj|gsvsT9k!#Op=QjGTaOrC!D~;n1io5X(SH!2e4p0K5A*mnGS=1xcekz1;#g~#B zC|lnM3pvo>+nl^f@6cN!f=3g2Va*umlhDJeOIeR3p^J0%p>-u(b>qv^sF4!9|2sm{ zN!)P4qfbr;J|CtkHb$ofN--FE#9CM@YT@NdUB7(d=(z@(U-H*8qh!M*`}8Q_@#ABw zi$8|ng3>cg2%jN+TM0aiDR)kd=ap`V0c~oueFY_lDt_e(o6u|PYda`AFhsHO3SpZh zz7V8gm$e>|PtsnIx&?8{u|SDFTFe=A?+dYMt+LUWm|u8N)g^edw5=jW%J-qZao4LC z-={bWhJV6WPUCVfzQfcKJZFqd&K0m-^KGXuhs$9uylgC>l#pG*rzY=>HsG_2?cO|) zXZIy7G%YbNc(Q!;2#fmq`svB;(3$AfyV&*Vm5s|b#|C{INxfXhwOrjv6SkbuTj*>V zaNm%S_vJyDz!w;Lw1)-=p;;)YIvfkRf-3JDT-y}!Nzz5EL0t9HVwRbtJj^98Bprcn z=KS?)uv;F>bXug*l_|~3}a%zP*dQh*` ze+tv=m?@QuOVJ6XbaY&=_r2FLaV%_OV{9F6Sno^HC5SY$N%KrFJ`I4_5kn4$El<+` zapVB?@bJ~ufWnO_SYYF&g%Ccbs<1~JpQHqm<9L_c(}@%3r}0o0cO(DK_u%o(O}pU) zrAOJ4yVlA5OH?HsXLVbZNQL4Q(}wt9SBmMSEMxiAMYBl$XRnh|vyF@3zcGx#dF9{S+_R0GPW2g=3a4v zeYqc^lTA7pX7LmnUbqXx_X$wVty3S&XfY_xSjFz<_vI(g%6YPyqYj0=)eo);A2Aw z_VtCFv1f zq7vXDW9P8V;V;#Hm~nEdAs5*wQ}5dQhcH%1iD3OwD#V`PF32twmqfZzx=iuxEh;!5 zaC<9z{-o3f*A)B#St&o5IDZ9a@x)A&Pg-fIjBA`}zC zpC<3XImstydju!lCL~cukJE@?XfiL^&F=G(~*s$ zG@!?S*91IK>`~RTqK0i8(X$(W(6smtd}`g{DWqXEuIgj1;{3W#j@o3iRfFNcR7i(U zI*BRuho$5i6Dsf(11>nl4z{?1;l0WdY+F&vtBZ&d1d|Rs!FAd)eLzR#>3ROIcC$g) zHR@)~P6E=H+KW$^o=m!Qq6Kp0Ht|C)!lo}W_0{O{R{i#rIJhmwHJowsG~r5tZHsuj zgTgv$qE@V@G&JG1VUs*z8y=OV|K0UKE2bb`esDRWu(g|rVm(%GG);o>xeW^+DtN!T zc54l66bris_jl_IvJQcN+B8}tjhGSnUOdd3M4sBGvzfbYjb5sIkrC&h@Z!~{Iop@`UgHS80`B0pT*KZyOWja<;k1d{Dz%*tzOgjE3L?i08WU zg_H=DcY$0?OBr$>8qlf9)XmSHXsVr?FAeCVn0^)3!0GdU2+ECz)fh%i^)?=OS>cv* zB|BP*lJ+2kG`!HUGCaez+7q4e%GUo>Asj1%a#%z_kdrp8YC6ZN`;yDCV38Zmo@#Sl8K~_p^R_T?Owwvw zp&nP8h`N&y=a*d4Ha~LMdKP4n1{}0+q5Px)y=E3ujmO`Xe1{vMuE$+2)!}eEEY!j4 zN~HnXpKpQe1 z*%QQLM?Th9koF@TMG^WX@C+=Vz`q9{3xyykjKUOV%2V)|dxkrF?Gsp{eVr$YRpzMT z>{$X7Pu!2AcOp~1c`kE(!!+YZ6_LTg9(vfL}}@`FgmCrBCThU1bJKrk}>s*ebIw-R70h z!&FOf0Potgy~X-;44U|i7k~7hd@Z%SjHsj@zzN%gX@^@r5y@Ky&)!4(FhV*9E@4!&$6(#nA&wCChAioU-PsaDtHGDwF zp}gR1W1es>;gycZ4q~_0rJV+>M=dw3SMR>odqF;tEvz|CO6&| zjUVT#m3U5$Y;1fP5!&{#q0nOuR};Y3YJASvn52KFecC9T1}r`jD;cSG#loixlODgs ziKo;Y$KY6`O)q?G`D*mEdZo;g3Hp(ijrt-dH(qe2Gx$XtC7dWrjat z6Mcsq%KA3oJ=a&4;ljDat7Vz_i=!&ZsXLh1?z_%HmF|xc*6^&^{BW2bcWi2el0J^mJ_>iv}tTvk?-;uo3K_w1eeA)oR zo3Xc39kV{L;USD0Sta07Q^v6LhLp@Ux*a|yNWEZv35<^#qX9-U=n)ia6F!&TMi(Bd zO#TuygJc#}_epe}QHz%X9R{8FH^g5H^(g{x!$<Q}rYZJEkO@_U???J66k1FZ;Gdpc>r zw^(Jtc9PIo1UrT(X+s%8H>1C$SUs7&A745c{|hx(_B5#7e+mh{qE!CDih9xmq=2bV zu+`=nm+$?rI;K7llT2;MIs;lqaWeQ2-1EUBN3%{?9XzY_l>AP6lC({lQ)yLQZggeV zKJPzT9X*W}9M#UrXSLTOpHdR1Slk<}leM1fE#(Z^I=O9nY05MjQDYtb)8bx-a(I?^ zB>(ko>JNl5XtlKD3L(XVH@>#Hhm;#C=$MPT(i2~8p{2oxaR_W?>)qUQmDH}sb&ZQJ zwpN_Q({Vm+ux}mh81~Wv9Q^L~p@EcP`iF4><89XYpGTHLt?DZaD=Lorm2eF95_|Fp z)-5)+RTIfKq}(;$TtC?!ZrXtKLQYa8k{PPwr9bSxqXC$si#C{Y!PUd)<>AsP_U&E1x+y?5$KD+DxjyC( zn*@x;zjG$$=M8_wV4D$~sQFs*3osC`%Fofx~IjT#I`V1JRbwv|ij2;bTqsx#g_&0^4(N!ARBOhy4KI8{x2)gS5xLiv3tK&5n>LYZ zbU>=``~4>aprXB1SkUyGAmF7iQe#MW4OZehFgy!i>7@ZOCLq5Ua3L?~GjGP+I?u(2 zmOX@=AJfgi<64A9UjcN5&jRw^ac*wf;+F|GYCoX?wWT2@Y zW!;!uL5bGlR*Y+uu_H8Zz?Du^ch7bZ7v>OVi;MQ4)?lC`Jdo<$%4PUw^v=WH^=W=} z3mY1IkL)De2^P&Zg^t8)qz@pkQeTR8X;Rm&(CtV)e*yDL4ewAymS0Ume*mrp@Do-^ zDAdP!&uuu7UWLSuNW|3akP~lVKt=LQ&n$xQMF|5TLP&Dp>#%sgB2!Tdi}KUC1lgS0 zgWXlrrNU~vj%tu$$}|iAj@n8IXepTfVJQ@#K?{Q*dm-~+W_*B{QR8awqa*@dv%E2^ z-PsydOG>JTJ7YL)38<-|BvJS3EA%hAL#RYu&0HIzM?EjzZD4kgy3Z{$`Dru{H3f`n zSr7bVN$Q?IuXvxoY}Kj0Ce%I3teqoRHHM@?kUxcU48~lDH@v-H8uoPCys%{U`4&1n z%WZYm{iUW^NY;i4DshrSq5h24QO07JnT+`5MANmkVg=p1k(X^1-n}6AZz4Rxa|Mm{_cXF;J$FK~{caPq0Dp;h!*sabNg9idiM9g(d7H=+>Rz07W@* zNFaYaTdf4#|4;#(PHq9U7rU%M|Ncf?b|jR1aUT-fWA2GJnYF&d)Vdxu8Xd(S)ESVR zwz_lGZU?bO9PZJOS@FTP7!5kqmpAJ9avwBYSMHy**wXWaG#+c8(4)*;dQ* zh2jw{z3wUsg;PmUDu@_VMNpgq|Ix=%Rw7YOQX&^lwt(wR;|dluD5ps8QecALZwoR- zsMdli^YRwDimX3{W(r>F`82fBhleh7HFN|@^)))e^%B=TxN|Q3f}8ZP zAW7G^w=ozuT>jB8gorS}R>wTIEiKEv@iuR33H4Z~YPa2erc5|{HtJsL+Lqz>IjWeW zOqrwy7y>3&FGG?KGr%#91PZ|`o*|R!PHKlTF;#^tPNTlnv=Du-s8BAE3J;w$q8Qws0Hrf1#w~2$8Mc?LZdbfxQTp4 zC{m`4pU}U`)|RccT9V@w0~){=;4I408Pgd_&QGPaBa5q>Fz# zX3=+PfDd~25PY5*(uN4L?m<;5HW@Xl5NxU33o42${oq3s`7hKH-cK?~ja8*A2TzQo z2iGtr<0eQd9<0|m`hg0>Q~t&W*;TS%?lkD+8t1~zPuqE1t*Qc0{fAM6EHWjD z%F^-rj1nw?HEAy}v7^|O5Okq-vi-FoC4`HrebrLR*xc{uN%y+9u}O#VI#ndQFt%R3 zBaIRqvQ(HP%6mq68NmRal&B6w3l-#u5N<;%Yfx$3%&Ev*xpg=?4ep@^K~T{}+|n0` z8mK?(9qo&E#C-N3`cK!aq+&a#x0{#tjj*iq7Thn`fdzbQd0vn$?!K_4<4uW0 zxrQ4W)q_v>9?!C&oMq;sjT5{elbZ7&HzRq!0jm6RC!8WzyD~-8an{Fo8UiA^g$1<-TBpJ(lUXvS@0Zp#%AxhDumI$(zg@ZnM8p!d>aH_ZY*6S)q-5 zF{F^H-{nU`rSNDTu!poUfkLE38g#<&}mBZ5x zol$5?JefRp-_&4m;pKrO5W3f;PvD9&}-EywH)*F`+ARvq!J(B3KBYN0xXxJfZ5*=jy`BSJJ+7 zk1uXHEiam2Tksscm0tRRrec%X&PAV}8^non4<6K+Oh-OgB!-t|x{IY}G0((bgP55Y zIEkO-zmWDq=XQ=QE_eE6Hcq)YWihC;HviHa|AF$HF3dQ(WBneAJ1}YF$xHyzyTQBX zx*f~IhtHCZ#%>f-)b6z*vDQ7g`)PnHnmG*7Utel)fPV&6^2l4$&svM=VzB+zD=QM5O`$cL4sO)wHd-qIQBvcdwi6s9c4)DM+ zrrj7huCC$_*Hk?>=WQqvT3!V@t4ysvehq25MM>R0A*0w~GVhMf`wY!oUP4SkeE34V zR1T)KB3!N^{pfB?phuPWRnw=HUYf0_N=#df%3}$q8)D}olkaHNEAGH+5w$x17klpk z*VLM?3rA5=P_ck?q5^^-B1msh6c7+1B1CCXQ4vs)CM_f?BAo~bC?F9KX+oq4NC~}& z5PImn*8~V8r0^~GnLV?eyZ1RWd**!i-gEi+L&(Z{v$EdxmVbHv&%;Gz6!@<4TOMg= zSblv4IRtg`wvPmc5Ej6w?ajNt8nV4xJ6&1%HdJI2bXbNdJXk85i#3nc%Kj*XmM9*; zyZ|!0OZ4bs%*G4{GawM>@WYxtst?(ilpN&<#$+c006%|ON}KdS?|?gZ93?9QE8az?utJf6oA< zGg|RhLl$CHP*&ViI(G`iR3Du$|H^UlVPCOV)J(+OoaKPE(V#)>`5Pk0^Qa0<)cB)T z$RJ>fpMQzdgxEl+qA>vc9yLwCN-&ZjcZ;#1W;LN{y9yX@PBp$OI~V|9p+PEY;RpS!JD@*|@RHGT^X#4K~Ftf<74;h~pE zDJFhHJ36zPJvG*P;u(A>8{!Him%n<1f52vcA003qD_SdiJNl00dM@`GcYuh9jdESV z1s}LHvK5j?c)J}V4LvXr~Ft?Dr;u*bFZ&esn-#O5Q3 z@k^ud;L~*HOl|AHBc=gp<^CLDm%EFlHOOyx8_qR@J424q4{uY9D?K4Tt8el`_zv!s z*$(1W*Hv#&pZoYpvU4i3Db;8^1YL;<5!%;$-U@msMYhDIT*28ub%}%9m zz-FT>rF$jIah!6J_JyZS+vv|l3sjpc*|)85UUA}9gMD;g8d>+#I5rK0@sEEwTp`Ze zaMS%L`6|a1Rl*c&AZi#B6H;1TfaD~xn^${Wzo@Q)I&46eZ(TTMB&yDfpPw=)*y)0A zgB+rtCq}=>6k0LwIKOBXU`_TmP?ZeIICE?TO1 z4$hxQzt);@ylx&#e2e6v>^l3nSUP{(+vxT?=z@3ORTS+gDQhdlsG>yo6euYkgPSLu zs=b=|w9KqBqZVqs0MANKe=iZd%>o@lKlmPTuEmD9E8^-(XuZmbxz8qty z&33kZ9O(FTFTF#rSyBD$S{J2rxXHP4-D*$k6t`l|1L{-9U^VE#xoLHKApZ5-n{s+_m$RKyGstuc$ za;Yv4PaNBq2lBn5JFBY(ByqLOvagK$UriCy@Bqi*h(<%y9qv%(;3jBQ+8b05C@xrh<;NVsji2j84g69rIJTw>tUpu- zCJji$21u?Y)$*~s$5s`19&a3veU*e%wBE~c8LdaD0=&O@XAffC=w76JL^b5$wNnUe z?u$8ovB4F7+TiQyYk@B!xICSlgLDRc-Y36*ETiT4&Su1IAqa?iNu~i&uZKX?iv(K$ zKl31djE8hXz7uMP_#oCnLxH%%i_D{1qu|fURTnQpYz|pBh8$k<2qHZ4g!OIgdgJo& z_>h6k%3k)6?phDL-W>Mo>4TRh5)Ad69MY&b+IWEp5ap!j9(YWt%PH7u*I!x#Mj3HM9TI%BE=?49nSa2G%2eK^FsokZZ>6N6!dypiyx+Ul)-SWD zKJqR$)fSM*9{;=q&u!C=5Y{=`|LKeta9^mkGekdh7Wi7r2j({%^72!)x3r{$Fr9Pu zS-#ho3-ztTOHVqVta6^WRIKJx$b3-ngxglcP5 zXp+~C>8v-**Z!2~j+@7f%;7lc_uvP55){@9YtZLu^&eQ$gyJJNoN1R4f#+Dulb7eyGW|=a zNiFG0UKgS)q18)0=3^cd$${iOYuSa{CQP{Q^86mx1FxL$kPN9DaBP~NAt`F zB*>xxFV8E(TR(fjnuhpMcR5(;Fk`Pr+L$zSvC@7DEopJdsMQh9Ylr%bt~NN064NKK z_t$#)hsjWQGjwSxF@xWm*^(ho2s(psAQX5Xw-X`V@<`uOsyVgMKeODiSoA!#QZ{%O zUY;6FQ*#H`f_ci0HjEZGh$wt%=81c{f7Nxqu?3s#Rp$e_$?jJcxLzAG1~}uT~|4Db&K-!kg(T*=W-8oGXf6-vn%Y8vj z)YaiYZyFECM!g0p$!r2IyY#|o>);IoQRIldnPaC*!zz(^cli($rWJ94uqsT_AL6KIL-;i)JK+xxR<;AWYLS#xQ z?SZ=ld>7(E9?3V*(fzwcsLaF%Jyi@`kCBMQpzd=#`!d8=a;8_jn6|Ty|6b4JLuweh zvB>MRWYRuZ6wF5BM6}!kblpP*nkC~ynap)F%jX}*>1h>x6Bi=tdkBr#p#j<2eF~NA z{C9c;^g!U|GHFIUxUol?|rXm|V5hxnp=F zBJh#7D(_XVM_{Y-!KNU7+sEDn!Iqf@hpR@~e;0f|SM`cy5XE4eCJ} zKeN@$;n{7JpOxIJKJ!|~NAr7nb;ET=uEc+EmES17HEPLj{DnKp&7X8IZVH?*S*8qN zX}KvjqYA06P1~nvLK{$W=)n&IPt%oT#d$!hWNg%RPREymu? zR(V5*m^J11v*(?9r+SCLC|iylK#iVT)RBVQbJuQeuVg`wTG0F0R+X`VsM~lyv!NiS z2&(dXXOi{9+tuNhs3qMK!7joHLXp#Ddl*k+R4Le!yPskyu&5U`1!~~Y8)YwX+6eyP z0rv730U91kVjx2=`r|jbD${V z^5ArA!1+y(l-g7PtLRf6mKH)dvJRdX6_IGGOI)fg0N)Mo4`a!RD0bErew19s>HgImTY=jEkYvm zO|6x0gV&H_HFgCOFPgTLp>r8R|9`gcErn?kSGX|q0OwWEox;fYphqj2BlS(P*Q zirPWF>3zB@qQOFApVzS|^phj%+!ZluDkpqF1z%nq0Ev)+fAJsIGigEYSqbi9S4f2w ztfd!8+!z@b;+mf%f1@^TNPhnA59isjXah{bj6t83>>Tkb=sew~`|ih{;3lDx9#(6=Ok3k4 zb}nK`S%jJc2{eVcc5Zc?_uh2XP-f?BJecvcLc@?{-3m~WOTgumc zam0vMU~PszHw%6_l*=8UrYm7NcaSI3&qUPFa{~yyFXCB{c;+6WlO%Qfq$AzKmAp_A zN?vrV5x8_no(+((5yI1rUSpx#KNaIz7^)|TWbO7^NVu9;`Oppo_wbu**L%=sTt(wH zHnPdsTY=&u=LPihb)rHWpGo@BORj%3`~UtpI0{q1;iG17 zH}wXkPP1QX_|(0lp`)T!DTEeS8nX$K&-zjxvirugkVMI{!-CJQ+n=0^`a`*7TK1FI z3If%nk|?-Kv07athx-gcx58BpcpdNzq*F=Q`NJvwLthd`9$u^zm}=aOv;xhxG3;}= zj)^aSAA6@JT;AGhHO28I&j>tqKqS=N%Q(85JXX? zkXEm9b3M}qz#KRO@TYsYjyL9^es(BbXDc$XU`>?h=i{KAf`u|t#K%v<&;n;>MKu!B z+2Yja`rp!SWSL*b(s&WV5tZ@$LDhU8R8J3y8lrl`pxVL-*b=)aOiHiqlYqO{;OM<5 zS0_VR=fNz3sBV#JZq_?_o*tmTt^iEI{}NsFA5l$jn?M0E*Kv#;3cw%&*sKFUb`J3B zjo}49VzIw-yp*EPoP*Kf36!A#$dViba#f|P1vpdEfl0p8^~Dbe%bBCrGstE`ED+UX zlJL6Z&vd*&6b}IA^%BFJLrrTJ>oHa8P<+TwOiaZlXqS*bkj)ua(fk!g{U?8OY|-=n zbGu8r+lKGQ*az!w;k(BSb>++HT9ky5AwLy!xLupiNm>|vNMSeYtIXhgI4PzS!jq&} zt(um3`%T5iFETv6&LOy(e5euJx=M{V;QO5qrQL3*6z7Q*rbNOs3+#=xP0-nCOb90b z>NACB!={Y_EY6IfugI|!<#AW+Q)%oXnLoO?g^r>J&(iS8kS^#mnX;k!wS-zUK8@4TPKSsP4@FJ_h@=uCq6eLV&OBD+%luD1R zSK_$nS~%JXKRe57A>Z|a?;NIUIkylxA*{GNGxnBE#@;{==&1CV@bV+BQ45;`@H2jjVhljy_N{A{E&#-_dVA_`*c!B@Sht&{|BG{KhJjFB0&BJ$ET)daCSK- zD1Rlo7KJvGt}52~ob$>c?aZ{Sf_MACbpaoXr8+cv4Y@1NTUhtF`Hc3i{W2jR$X8YP z=-5tj(10Cz342FcIBtasn?-PyL%ALpN|_0AqbRIX2w1H?2SS5)&>QYRIk6s$aK^e5 zjJz-0D3mP*Yq7isSAymxgA)T8%f7pf1AXQ1plfG4Uyn~IKB)|JO0&+_*2ShX_aO0n zexV;NjZJp(i&Kw2b$(aw^4K$Qtm&*7kBVWw#|_~OX}}&%m$+uqr`9+UT`_bBp;LA- z*T2RV6Jwk=<`c@!(s3JirHq%(LBu|Nc-;M=k7cTAT-5sf_0X547pjoF?FgBhQGAG7 zT)Sk?&i+VCCSs=^iI2q@Y8W0siq_QxmS;WOL<&Rbp-e zOBCpjez<(B_0D`L-Md4ruh2xCu zm*@WO05!K7Pvz5Z1s4-J=s|=GfWrW%q0#YM$l3qIzyAlq=(d=m|A9pc>*~7mHHWj) zi17Jv1G(5InvZ&O?%mZoH3f3R25V>_TfZRrUZVC<983Dv%57NrHRe!<4mj;uvhpV3 z*Wzt?^Q3n)mDhWm+XT6U3Th-}BK9~t%l9`qKAsq8)|ID;?Q-`V?~)YAn*pytq(-w8 zbk#KWZQzzq&3e9D#CnL5=`U}5P!=IA^yW?{;vNQHH8lrEd^0HioT2a~ckuFO^8~j? zPfgc-5u%yyb=T0ktw+x`9h-TjvfVQrko8<$8OF^IB~fH|%X@puXnc_MsP9|%BNiD~ zaIL86*0?|T@@`hO@kg!DUN0*5enOpLizyn`LV`YnW?3#e5#lxsrgCFVfdvzZ@J{_b!|s33 zvHzKF;}*mAf3ap|cz8J6dbAFc%8}z0Y1b{dbD|AqEx%8*{c^rnU(XIlZCPFHcT6wq zTEJJ5{P%bvV#%OYe4OgGcbgzbzr9a`4|sK;@@(oFafgAj@`ndYig`W*X{INdimp3u z`V5wxy?4sVL!MT!op}61QgE$gW8%Uj9bZSm1q*5u3vCw3KndM2+xO<0H{&r6Ibm}t zlh3R|h*Kk3I1)e1QdQTVYKQU=Vu&|8_U{om zRQhBT2S*AaR9iKKrn7C$uAP6fAF4ba(T_2s2+53@c_7pyrR;E3K+S8|nuT;_M4f)` zwYC^KA`Fj^ps0^5z5MD_clUsO3Run0P3?=)sj4JccB>XjdCW22kxS65t>i5&^za@- zUwOmuob$j^aH+xzR7gB{98G<9K@WCWt&G3zWBGo&)RSssochyo%|I zp~pbzc4*2YeU6PIsAYhifRs)jPefDv01am#&e(vV`faSdMd9%ksIdR)ayoyzS{1|N zJ8o)f7+25W2WmscKLa@s8=w>sy$Rw=`t7B0%Hm1PW6c13K?ApJ!@R{TnS`(WBGDcK z6sS2}N44Z`f+B#*k`UmO-`*=c?lo}az?3*}OK>WXR5`N=vT~NDQYq5Q5N)8?{I@*` zRM7tX9k9)MKtq4$*C;%e{&o}e28C6Qrqk)%RB#fC*dGO6Jp}~FT7i17Yp9>T1#C)` z6IHGJ4KhVwUfu+KhY;c#nMfo%9o>risTUzXeSsg35;7_<0bT+2))S~n@Lcm&{awjSZ{Px`G{_?xwCYZi_pw-!>+ z*w#Y&SF2)cFWfR`{%TZhnKN4!>A%e)U0mX4b{|2WB2IL@j0Z38D>uR|?96`2ni8ch zIijfgrXcM^td{pK=&O;(2Od|}Hq4S8R@Btmm8z{pQv4%(m-Uy~elIWG0Z7O^|6S48 zrT=!(*ewZ}|0KNd(#t)s&V5Xnbl0zv zZV$L4ST?qw7g@URKC=avXM4wCpSvbPxjy`y#!~;Ssubm^QY2DW_}fJPL~OVLvZoxe zukpbnSrta)M(*+>nNg{5Tf8*67YEq`N*xH2*rKwQ^GUv3){x~JhbH8~zu zA;!~$MAcf@L$|T&1(oVhJaQM5%zec|FnMXo@|2B(2t|r(GD58-p>6)D$2Gy_o59Oa zP3UC;fR{lj-*%Ljyyj|G+^~q8@T;nP&81@306;&k^Zz|w<_}*ZpCG%Lyd~U^=lqKk zqun K>cK0|$=ALW@|$C#UwIvzfnTVkek&ilwgkfvG-$p+ly2a9hgR*;3=0pw6M7v?I)Y-Nmgk`BZ~-*YoBp9jibB{u;2 ztfGgP($Yy*nPL=-ugSi!Y5SJplQvH2n=w1`-N**?Ut+xy+|< zstoHUNFZ|))P*Q!K0gMqQ%SD?jyH@nv@!{&u>VM~OGr)+QSL%QFss{v;DJAcG65!j zL-9kVcLM(F0Tk&QM^Y^o zr}})XMpBp!L)??^uCD7Y3#*$q`({)2NG?Epz_}k|hj;M3{Df>p3}~}0)QU_Ll$Ll-b(yw=A(OOYC1UkmpDa<@+mO*LGkwYhXJV=~y;U3~gebz) zF4r@iN3r7h%L)l^uB20KI}E6DO=jYWpH;X2$?E(c?Krkr)_=dg8XJZKoS~2|CPy2l zYU328!-hu*?6#BtYgoNT{tTQCDK&%JC$~q`W}mlswb4p5+tj;HO8ekwE1*<8Wi->u z+u9_eDgL&syC8_=JGOb_-0rKT5k#=Hevv) z`3eZoM(&By(RgqD+$*dbHqTLkdl3V#vOH*gsiN!+hYWSzT;LHPggRP?4>}fo#nGo! zCx>x^Y#~+UTxU`e`Mv;^(BsZt{#EgroRVC~iGzn~_Z#YthkHfLS=MmERy2;6%r@c0aG`1Te*fUd zSgmc2eX$*_!08Hs+s6tN(~^J>dC0l54meL6IK5Y#Jrp=Eq-oYR?+W+U$8L4PKR*&G z`URfbml5Hit`7U`j7p5d4ujv^6sqgqAYj^~mpC$QOoMH>d8PIdyVGBPOgDaVpK+VG zh!x6DzJTykt-u4J7NN)@D|w=s#|h#hL78nAWa%UJF1B9gwTVj3^J$dvgi;E89~<_` zb3hElx)mtUe)?yAAX^Mz|LYEbRiZ?YIP@sq@1f~K^(@-ESe3`0b%V`tV@m=vNS2n5 zUGlqjyyrK@#7?rLvJ%>GVB{$v$M6$wl|GFT~z-PQ}0O zTEc~ETnKcJ;gLYR>PqTs{9a2zlh|RD0BqwYyUUsp%VMnedpVau&5kox-dbD?$(Mh@ zo;^TvSe48gbi@|lvSfW(qMGWCG4a!DPm(_M;aU446=W#BYKuP%{L8j|#Mo8|B;tb?7kl0U17U82mq)+@}FbI4ey!VR#nqKiY`X)W9oE zsAD=CKr2HWn6eG1XG>uwLYCpoZhGLa=j-%L0WwJsseBCAUmNrQ{K@h5R5l0!yx~d# zVze9-UkLdiz`eW)TB{p}{n|nU523F()Dz~gzqZi$j~g~(0`$}E*g7T%Ref?3l!~SU z%h<-I#woIHYOZERzOBEXSi=WYP=fGokEI~EaT`d6eq?F?pJPLUh&b4Tn~bczIIe7*GIKZE4#kCTV@JQcSRN z>Rq~LCW5M`YaKBdj!RpfFN-=7eBa145E3s_`0Y(#XeR7Q<;Q~O=5M2R!<@6rwQ#rJ}01wdt}^=9QBTF?H>)@Q@dI$)>W;@ZkbCag00?hdS}$i`)T zJe_yH-JHq#1=%#TeUum#A1^{kXiiY7-@)kTw)%d)UO3-75X8dr%pg?GcZXxJ1aYSZ zE;2WG@QXr8Mwch9s-79}ZTZ`EJ&6#}M7k!Rd3kCT`2YR(Kl4+@4M?et-*%`1kX?GT z#GhVnoaO;QkW}!VQE7S*CSAh@7}cvu)aec5H9$WjEDHXkktZ#u>@Xu1H-0yo2uua5 zvUzNidlF-2LwMtG6^44`7b-Kv%=)u`QBn8LsPO*pXtP5RGApvdN6#VwjO_)uC28h! zNwW47<09q)JaQ0_M+l2rXZ*7Cct`I>I73d5j43YOHyVw9L{$}jtT^_I=vt!A5mWmJ^9SDFWw%@@+>VNgHO(6y!pXB*>`MC874i# zjes@^D)ayaq8CxcCp>kvi02e!BEBu8C5`oHJehh+t_><_Ao=K?vPc)!-*rK!`fJ`s zZQ9AGQ73kp^64{5Jc2uK6Ubuh#0j;6y=gi()Q8-dIxF}CHjy%&1G)&wy8v@LEFB?5 zvB^MOAnh!BXUn@B_pYh#u2_*!`=|}9K$a|55EzAQd z(9nXT^%U0B2aZKhlLa`I&no(E()F4YTatf*|IKX?i=H%s`Y>wejMdJTn=Pv!*uq#M z@XDgaKhzjQP-jRePN*!fM()gEK!+$cej`tlY)GX;(?HG)5wKeEugqyUJF^wDEc1)r zL9%!2&o8<6qDm3y7bI?`7QPdFwpEL3!-&4>eUhe#s&AAY0br@6-aYU=%%-7b*+xJN zSQ`EWHd$iCTxA?&>>0t+gJ3^6U&~x*e|8W4&5-zcFZz>v`hWN5ZRf|34*@LXF!y$* zw)xQd57kVrjU;DgA~XT}Z3z%cZuDX<1Dn(Fvc?x0%9sfEG`fn$r-`)2b-%7yLwJcs z{<_YX`2UYK&K9%lU&cgz7dcjn)iBzq>~YvdSzWwy=#W|J`G(XQyN0+9~s*IUPMbNQIS_T&Z&&Gm9mUsY8r>Y zc9T5!PV^o>$3tdk)~jnFNbuJFJ(ZFy`h8vSlIUS{C8a9BN;W0uBcq1m%W1+;G%W6P zu+Gj2Za9?!;$W5{{Z5Kx;RA!spXk7bIauuC)d&f#s}oAM-4wy6w>=tntwC1OB`EgE zbhVaa*97Cg<9`&#PnU3M$#!3KBT51EQ&dnz8SRCfIv zmP3oV2ndin4lnXvH?u1>S;CbA{Hyd~+WUWXj{Oj7yN{>x#Qv~Iu9i*z-Rq#2kijyY zO7|eZnxC3Qznup7wZ)m98n1sgV*s%Ey(KE|T{>iy8(S>7VgOuJ3O0WRFq2wA<5Nzty@xy;INRb`{o6(X;uaNGpr;X`W@J2VFeUxu6>%MzV{?$nxF!^|-6I_V3|i9mcXv6@5S$u|K&kadij#1cpI>TKftTRIBWhZ}0Cpzpd|w4f znqQ;T!Fqrv_cuGF5|!&l56)qNH`oaSkyI=vsuXe&rRnYrEH|CAm4LtYPJ{S2rs>3} z`|mARl+MrI-)<&>UW>;7`QAO$Rz^w+kRYx^y3FsKm3Yw%^^M0 z?Rq~UCQ=9QC(-7)=gu86y0rUNq%voZ$fZ{e$9j{}-m|t^Ova8jv6>drqC;y}+cDSyQv*!mrt{^bflT z&jo(HWEB2beB1rhu_$Hj>FzUYY5$02l$1KF;AxUH1g}5-QtdAKd<1 Czxchho)y zhJ^w%(d>}ukZwHh3_auC%cpRSa|G#l%+@g>I zVWlaVB;m`Hha+#vUVw5cYgAlFbxztiOS`KZ-(e z%I65M687K??Cy)xjB|{<(!XHbV6g(Xtw2bfsypa`9jHSsYZwPoUMle&LDxy)zOO)! zP*O}tp~cQvyXCi8r*}7$3y8BRf%ZRrl4S>|SvW((>{3R`48f0Q|&(+TiB*0mlOJd?YVv9 zb#6qC1UgghOMU^pLq}Kv5xA$>OwG?Wlok_Tm@+mz&9V4#1;q-cNjbMOTZO>E%*8%HNX^@i*%QRG?*y-0CfdnfC>BmO-LAnJNtkeR=>z~U?%MgKnoqYilZN%=AfK+fDD5-WbGIS%CpkDqj}}R8gz?7ot$W7<9Yc z5tX=OIqA;zwSv&w>Jq|%#d%>)l{c*Ry#??^NaYzYAYUvm>3U?tw-9{aNA$}h8}^ll zSDd)jU}c0+?7U#6<#mLqiolKK(>*?($VFv;Y)_N^)tt6dA3EgssHGmSovcu5D4$un zK5Faqt)8$C7Q=maRubKU@EU<1HW1+xB^J477w)_jJSBK;HLLQoUG0o7CyYIVu1CD* zo7Rvt`_=#XqI$xsq_{_*&~w5G36egxkK9gOx~g+Uk`%p`Oc^3MXi*|vKgkZG7<9m-)h z=PbiKED$j3ne(ihP>CX5-M$|)ysQkK!8G$tSJtElPV&4JGw9ynUpwuF^6_+5Jpw5i z2XrhNZSA5L0!S>iWxI%+1;_)h`sy9uK2KU;j9V{$R&KtOfy%a_hh+ld^r-0>`xy>i zRHZy4>=;RRJ3gewM`(;zLasTPm;R+?jK4j{W{P`Wmt->v3q)us_+k6yC*vH80<~i` zZkXphchbp3Cc|DM&VHdd?@&BQ*_|Qch4{`oCYrqt(6+CJK#j9o=Rb6~@~XwUXBT}o zlzpt?6%*KgHL@-H+U+rZMJEPVJq1-~b%b=GQOM8pwAo&V%A7IfxULH1Wn? z^os%G1EL9$5ZzU*vW>SxDYe#WUwM zW9+^ma;*dNc&2eg&!Nev`Sn>z7EO|d->5mH;C_L=vN)^Vigcf;`y-SM^q>I?t2Rvm2s~roWZ22r*-e_1Ik3=!W*S9fhBb| zbv-KKUQO^+S1MrK?ik(#p&?wTzLb?1oKboFVwtVj z6x?kb^{6K+qXac`F>(MhoD_&5y(LF-!j!R;f+{9=$Q{Dznw&DnSLPqpw~@0%Lc5wn z14maO{#J)bg7tZ=*s@khmu&(rJIF`(clKWCeDQ9LP4X>?BUnv&UU%^K>}itfS>l8VzIO)(pxDv-(Q}WX(8y>4 zt=to4t^2Wn&y7t|)ui*2oIbC-pdtLk;HgIs4hP5WkDts|Qy)X`bx-?f$+l_{i}*1= zo;a>QU6s3jIy5J6nd3112?ajF{xO>%ANhPlB{fugRo&#$v^^ZaQ64GTIo8lr;7gYA z5T8{mI-CuuPHO3NQnjj_g5~2cB?d~cvC)?p1sdWAP2y!fZ>+D4y`)D;Q}aQQA+&fX zuP%=WVcOSYsG5_g+FctjFL4yH&@Zqm3~|Qjbp>aN^cKClUvMNMnf@0KF ztHb3x-!|WoZ%|gPe4&rm#jf*}nUp-|&*Fa0xB);x~24&g$$Ok7v)|Rvk+Dg&h&y z;epFgPP!w-X1LGC7CXo8aE1v!t@EQ7mkfx6I_OIwsiIK)#V%9Z@n+7tdn0tX6}+41U8@aq-mCk1`jIpS36q;ZBnui@`7KBSxHmyFiwDtRvwHKVizy z1e>MS{4(?NjjZ^Sk*!0!<)0tITkf4Ih!AJqm+bkJ&#tb0(DAN#l)|*uAs_QVS?E0r z;7RGRN#YT@Bxrg3p~wV6zI=hptb!`J4p_mmgbf+VW&7>ij<@w$?lLYHy4=qcmFQGf zf1n>{I=>zxDI|ru6bn`P1lyMRw3*0}rQ$kwT}7uF&N@dJ({Ib)rrv1|DSVxf_ilSD*46&m zj*}g6@sw#9idfe|w0bjV+As`wh47kjHty4JmyRonI-^u^Y7hV{9 zOH~ZSR3*(eoBa({0Cz#kqFW)q34mFh{B7EAS*@KhicBGc_n`{gR)K4mtn~l(8~ihW zL;hcX9g&joX1^S(I1f7dWP-z0IDw3Xihr9OldSa> z%}_e4RvUjLDAQ;fy_1oGJd3g~pda7RzqRY)3-CIRtI^CHd~ZZVgsFqHSXq%*#>>G) zCj%4gY+TTZlxJ17z8bOATq;4aNy1w4Yv=alm)!9tfih9nPbDO1ifeZ$5=^mDA@Q!n znm92T7P`m&l%(xvbds%HV}D=;{L@$ALGy;NsE(Sh5I%cTuax;t;$4Kgo$J62H|WtF z+hrJ}9<=og+;%nf=5xvW035+iyqrrOC24b`)Z)=`li5=IbX>}*loyGq?LTsL9%3JU zu^9dnCB~oLH@0k(DQe#798(aKCv`+xWn=s}`Zns7M*icJld4)4-}2p!E!x8`&D?L6 zoOy9G+k_$9A6tZK}5qV9)m?|3teKBe9VK=QFl9VlDD%Zo>3 zNPS_9t7WE%r7>!%zJ5JL%kEif-_sezoZ>qKs;?iZ_wt7Xwk-wdwV3WnYY_-JIwcJjgFqJm;>v99pI zlM#>qlCof#f%X(LwQry}7(3?{zfqt+BPDS1qv-XNY4kL7KWf~p6j0;lT1grFYt_k5 z4j^r`jFnM_8-D|^cu?oK99~jdL&b(|&eFEcK2bojmgv9L1V>;QE>XDixPny_m1;&m zNRcMu#?Zm50S+4c+E%TO#T?to_gEO#EFFAnc!~|Scb)0o$7GrT7SU{`=4pc>fjr3u zEz!edDqS~GTT#938HG@f%Uh}|<*u?iSlA{Pu(O8TvQ2Z`<&Ic-SPP*hdJ~jeuwW~q zv!L)D$N*SAgd`C{8i!GWbP>`Hp`#?`^L0{!NW!~oLJH*Ig7s%JdQp6erHfpSviU_1 zy!~jb^phQ@$m~OccB$49*Q)nJ{PjXwOBs?2=w3}7zu`%oh>iWW{Tz%hdSf^n6uN9b zCEI9PBZ3I9n)HSmWphfQU#6&yI`DSmHMhyq0+c>4~iH z^qn9C%ah|Zhk0Z$>!R51ZEZM#P0&pURiNZN8OS@Fi@{m5>bc{C&!DQ#!+ml2%_-gv zG(-ExRWjE&Y|GaR0&y-GM6Wd*KoFfkfXpRhm=Rhg&nTD#rKM+{n4%43vCCh?fEt@+ zY#q2S+Nd<2f_d^v&&V0f!YtB)XZI@7pD&>10EQw7`W;3wWIWB1VZ8}zH8S#{RPoc# zt|qOSN9i)z@5_F3lR10`D9N$Gnv<~fAjP3Yl-V*zk2&*F)U#C}60Fl;h5M_^h&LXy zxu~&o8hI934PL$h7~)Q-=`{N_=8da7y*KuA0b;P#9Nj)leF%L&l))v1OkuXhpTtz6 z*z&#gn$|2*(YH3fJtgmOk#7$UnH6ywp0Im}VL^7l^B!R{?6Jd`z2m2BZ0`zFo-4Wt zG3H_@jx>G>wob*ujNt$1tuRL9Bzt#Q@A+v}Q`Y&2;leU8gyz>S`Z+cHTSF)<9TIN@dInxSlL0D}G23ET*K*O2|oa--%;;JTA! z^^7S@FUu%-sv64b4%6+jmYMa0^+w$^gf0oPt(pjO=Of8E8EK}xtf|=QrM{$*CX6q_ zGyZ5c{IuY8Vv>^3HOG^)Q0E!wKJysUsL|@y@gPTyHAWC*Jj8ZrtsUq>eHiY6@*nDwVAv+5Lw@6++Tj->H$ft(D;qk%XiEBonppon4cPf`;VCpl%w(Z+q9Cy8 zyYE`8h0UD2w=eB-qX=YY$$1eS`)#{^3{?uAxu zfyblYSE9rfTy=b7f9E(n+u_-%5_1yUiczV*-dFR}vX?eNX5pdLp0LGJG9ogc97EER zo|)gachYO^uVdGVQH-h^bEw927gWdZZAH{R zb~C|qS!Sv^nz;vSYr^na8*W&FeWT0QFX4mdi!5#i(zz%RwhqLpwYRw3&+rqfy2pBV z8LBn9iVVr(&og?>s! z-uvK#p8Np(Iwv!d5RUC|k&&+xM3I}?ItL49{j$~PCVmLQONTH| zRri)*ACu0}hkYS~g72V2!X9FC^U|1f-StJ+l_}v%%v);S-EX7^`WUV7k{Hu>50h_b z+1Mo*?NY*Oe_(_gg-*RQ6zVd(+D)(Qz}?~>P{1#kiOxxV+q=36;&rs*AH(=Fg+;>qYnK5EUY*) z_W&NF5K>|jRPg_?_nu)*t?RllRuCy7y+#EEDPp6ShzJNwL_~y@Xx@2_=Ly&zN(qb@rNTuYLAD-*vuoefwMf5V=Uk$jCe1@jlOc zKlgLrhRbq(>^R(f)D&Y2kAU#2bgiEZsY@^8W_?oqg4ag}O4*!7E=3BB^@F1cFA2IL=F&7OVfTWqSH8b3MsOBUjMVvH0mzhjI_oHlT zuBKU%+ds)_`^wfTR0IAES=qQ1aS4lOkI^RbTR2@vk)%O(PBE3ndC>J!5tM){f$t5Vb*|D zia`L}@KxQJ&^E?~+M{QBmAU%v11bkos_fYF!wuXn#b3uR$cxd%{QMzUg$~duXAmb` zR#;@UC*JJqVK!w|C&L@)sS?+r`?)C1rv}d+tPZU;2?m@a1yeMgYW#4+t*uY>5}0nX zX|ivfT}#?Mzr4YE?jh_`!?_i9Lqa+?Sr(v1v?zNA1<2UGUoGH~;K~G#1FWw}GREZ!8^q_<>VGDjXVd~1C2%9f?O6~k! z)}soAzwDw;)AHbl6*^?p+bdVTG*s*;&1WpCJ=D=E2J!}2%M4e5)Q>0qRWqsPd`eb5 zII8r_G}l`1wKxSMqkIX2vaoMx{9SZ?E)Wo*v}~v$ydAN9W0#!*0Q>_HQcIw}$6O3( zNI1YfnS*V#XRn*>eU0ko#PL1WRk(z1DW%PgS31^RJSQqvtXr{;jzFHi4hx+YL7EQ> zu}ibaDk(iuIYQggqk4|08qFidVV&#FkE{frx3RdKVL;xZFKW1>0yJo)O=nLsU9IL& z;8+4r1*NW2KDjS0kE6|-*r}ZOzLr2?nd_s$79S554r-NbepDd*lHVlP2S2MAv{9Hv zG*2Kx&A2Y>i)VNS*T(920UW_F?K0#%V_Q8aWSYI6aDMD)3$N2Mzw}yTFv5SJ(F_d$ zzA%I!6iP}rz`wfcQgailHUF;Uj*1^;gew~-UfB3h()Gu-i@a<&x(Xb#{5eP`=#ciP?obuZ3MLUy;ZyQ~KI=OHo zc#nug9iF|ztKTM(#xWUcbs6D#RyfWbf^a@Qy%ObrF{j#o6@@dFeof`VJc+3OcxpC| zj1z|6q;ipl3j$PuX&Ut?BpO3Tj@=TP7v$eZoMelcMq7fND(Yw9b<0 zGnavE14THi=ViKNrs>8j@$ZNx!p^>2t9oGWFF#riAx4O1;}ILB1SxF$dxFgO)-#{K zHJ=y-LVsoyZAOp6PPCz=H$d@W?K@7&lS{Wje0)azLSA0CDqt9QS}dcp4}MGgZHR(!?8ECR`Z*q8S8Uuk@S+;|+f@ z9Gnsd$VX3xx1d$t)Ocb4RYrPTY-J%8mI^d;2XBKxaOQvq1clHumw(kpqkIGU<$JBhbtIQLI;5k8YEq_4<1 zXd=KEZEf!YHoirP*&mw^zAxJkIQK-(SjSbM!cut8k*Avv6ygmlj3Unpabk z>*Ja(PhOTjWYj5!V9EaBh2kM~QxtsaXJd+dQVeVDs-l-)WD6AmMZh@IUDV`|i562f z0}JzZT&%$Pns9agbpe+FMqQq3ar;ki_U?4c*tEB|w^wU-&>P-_YG}9>zko;% zV(_W9P~x}g0gnC~D^o?RFV}yzB(VaAOR;Rr`DUq z21SP(jJ+nxj^8-r^~YC&jHZci-GI|}C=^LaTxh!sNkySRi7Ub!Ll$kuxq=SF>9 zTqA0t?QBj$=bW={1Ujx(aI8aBFuDiEDW4j;z0 z`R4HhdtGw}=^GtCaASsmjZbqo7R`1{a=&~w8iHlVT@gw75b!+Nh4p2K>)`9!%yxh+7sUKixj7 zFzpM)(buOE0cP{H`K@uryLA|txDW1A&celBAZ&?ypt51$M-ipCqjk1(=C9dR5FJk; zD}67h=kt7|T6xTlzj=AK#yHxZ6k#)zI7G(Sn5aHMZEi?Bnuqzj6K;_PJ3&Vtqw(PI zeS@c7A+K>D-0t0}QSTyX`E-SM0oAL@{!QhxCyIc`z-h=z815l{{Wk)vwYz)O=CJdl zd~b03IX}SOC=FVE`VDE~aTw{h?4Jy0x-BoKB`IA?yX#iSH>0^PPbadf59oJ?o)pp) zO9QfrW&*k$A=2jfm)0Y`v%7>j(Nhckx*80ce3we%L8P}@%zuN%}henm-ds}A ze|RZJ@xf3URfU%2u|RSr#7?w{2)Lc9L!`hICr)2nGpBS`!P{5&!T5C2A@P{%^sli@ zdzx`Dy(^c^3_N)RU=NzY=EK?yBdla;AK>DM7NxA)KeR--Z^$F-`NRu=eEx@@!Uo6r z?^)d?mWfyRSLvrQUo#I8j(@oyEWoab2{FdJpnUkr(4-96=nO(Q$c9cU z*63c3F|Z6@;EE4KGz1USNp;=TPqfK=Wg|bi4_1Jmhj(U5;EyMhe=>w^j+O|$KD>Vx z5vjT64=Dn&3gJCg9TUS)@`0>sD#sO7YSu%+ESd^tRi-V?s@DEIGVN1N_EE@px&SRH zKx)oVSZluY+;UfKc=>&%Wx>}U7QX7-^fv`RxEEzCPp*%C^PzmtgzLql$H|{#fBY2) z`ki_A-;b01gGT+k>;L^b{0sf-#8iC;x2D}EH|3v+oMt$A<>;QG5QCwy36qkpwj?f5 zsUZKN8pkq{4}JmQz_JF&UdmPp_C7=F380SJvi9<^9N&V*z@-T4AFz#&UCfPMHzXjR zePILYZ||x0q}+l4J%Z|fub$c2!xGzU>L!dCMR)X;8tl+{il7jil(~a7*=c=*tSDYx z1}h#@+ReDB0|Q^qik!kPhwZU*I^iA2PwjM<#SV2^Cl0;*0ht&o}Z2w(D5Ja}FLi7r%ZR&PJ6Tg!7ZkSmhZ+aLw2Bzsl@f0R-2+JcX2Db`%5&_bY^SWE3c$am!2agBAeA9azo z;gJz5`%b$C&lfGzk`@CS=sugx@38hX#;~lND(7-$+Q~oaz81Bj`7N%ce9k62;9AhTCHQ&l+O(oifh<5{PMP&*42R2; zFcvZbmVuea>8u6Q8N^^K6VE_z-?Pa}_cr{7^`uq`tQ`kU^SR<)wJ*sLPA`=zh%dE7 zN|g4V$%?IdEU2Fzc7nw=f|xTp0DIiXJ>!UE32-iE7%}tpRNH?~e=5~qi5{D8SEz>r zu#qhziit$c{9Ls3z$d_1r!?g2l`3?Yp5Y=Hsn!p*U7e|5t-K*eEL`^Y)k3^UnsJ<| zSbl-~2$O;TXl+{MJy7$l@<~4_oQ{y)QIBGso3=H^M#cf#E-57PkkIZu1hY?&7p1(M z$_jrs7H|n~g$#Ci#0$dJ(p@ZF@*?scN|^h5TEM}*4di!FgMiDhCI7dGlTv8xh<-nm z8rM}Ec+#ZqCxh{`?_|>r5PO69^FyIu047LfHW{;aBa~wR&>d3L;akbxE_>(}KsKft z>LEisFaSuCLgn0%lzo*6M$$~*-a)M9AgCtp0)SMz=M0M>(Zy+EW7`iX5`bF(y?m4l zR0GNa|F^y%Z(AkWZ(1;WiJUl4q#}s@MT;^%{TXp&F`$7WjiIP_~bE+5-rpkY1$E-%1_acrb1c0RB3u0MqtC@m3@WLBB zIIpdUe3=U9_M;9s{zj={NA=E-#24aGFJVV6gWL;&3Vb4D#XFsMAG4DmNgDk)(|RlE zB^J!?XeBz3KPac~nZ^$}Dx&N@^oUqmab2_+wIua|0Q8+APLddJAx$u{JYpvE{7<(T z@xg0SeEET_Im1=p)41vRS$o{Mku=p%?!g7VBgnlU_Dx$`nDtH^yObeb#Vob<(rXYk zI_ZFM4pbfBM$57pny)!8JDJQs%k=(?JJ9L#9PQX7<#5j##NsK#BRAP66o9B(3k)o- zKaG=^aSvuj@tfQ1fQU!tJFL$1m4L`chbaQekU52o(s_UI;Q*b+PeX#;6z2l|KD=1w z3o>Rirm9EbZq5%yPwUrO_Ayqp@0aiGOFuw0&x})pDR!<=7CK`B~13xl@<6Ivx<9nLioI8L7z#k`R(CX-0cW!pyD) zc;c5$@HjMi{iVp@YPFWlq@N$g>Vj!eYG-K7>UFq3{R^i#t$IlszAt)U04?4lnd3z} zJ_Rt6vOROxWj?8ODT$QVZU|ELD2SXjqQG#pSt#;yGJoj(SY=lqx#X*|X-l7#{Hw38 zZ_m(~m1GKQ@Ffo50QKQ|REZ*LkWyUlFJXmU(-3mFs*7rgo8ZNMN3}u2pO39x^&Pj% zyb*%Gp)w!meN2PWt9NTr^fBnR&K`RXf~>4u`X9am z25@oIf#kUv(8thDumfc^ zcY2)}VpY|XGT;Z9t-%@mWDwUWLtyj`Y2vY{LxihWW|;TD-Y%g}&mC#d2hNO{Q|RDJ z*QDhPV;(!+&tl$zOks~Yg>lE8JWj1 zel<$u`*UJ%(gO85+CNWufc4dd41_y_Jyt(ZqD$i_D;PKEyO~clIF-5e($kpcCUN8j z#%PoIwhi@8=VqR=O(Q(0vvl>|L}2Nq*Xey&X;?NrEKRVJ_8wbsB>dc&B!-jN4d)2b6V%pydFDi5>65o(xyR?`y74r@|i`d zXo?&q8tnm@HJfmPR&;iT(0@EWR~;YqjxFG;&ng<2w{f#ud=B)>1LuY+i(g*-xE!Ps zKgq7Q(a~ul=JqHfF=}$XctLi;Pv3-e1SQ>i6;y!}0+!sihp4i#;p5ASM@G<6U>Gcl z9JrWU(>lciQ$u=>e_>o~j=Z&iQX5zs60MbAK32t7sT3&KFj9R<#q=rgyI~x%8q=zR79_C=n|-k}5O)(qQJPC~u0tAzEz)9h!!m98g+de1T5@?x9y2rwMFN?Xg z6mIJH(cB^(@-@UR37fyxdX9ZGG}Rk@eK6=es*@e49zzJYcO2|9qQXBJA~hpv?gYm0 zN+ipw@fx}POVIeCWShjHS=Xz6Y1FHPybpzRfIlWOEhC0Jhr`%2glvD{U#~!=n9a^_ zxvRw|%Uf;9n!*dHqSZs9BpI%Ipc!A6Sme$yIv1|*W0GA%S{EmV_lofY7p*gZ_6(4H ze<@}zJ&-?4BE%4@yQf}7il^N2;eALlCa8lsl-eMu+B=LQW@pqUDAP%KJi4`)79M8} zJ*I%_e;^9tKvSTuT-14z8bLvalsZc>uN z-tBp1uk3H}doio^UM!TphbVTP(M+RaX`j`vmPN3Jy$VqCq^EgDw8YgXa-z=PRyw*X zw+b$=39F(Qp3p5MMQZylG+U0+slnyD?N3+Cgf`kCcLKuku;20BLGT}xx>!0}z+*%R`a<^94X zB~%wUXy$Z-K1C@&Bk*J!bN_>`(Ink_irP~B7JF_3AI3{y!#pT$Q;IZgTulmJOdPCe z0P~lrw!TVztuq*{$>U6{Z&8;XpxS0NmpVRYjC^FsKYuiK)(^4-!|mhO@5Zj*AuL9w zo_SIGE}FG=+sN$sQQ=XM`Uq0nH8M1b9}N4h@^Yyzw3y88J+y^!~7Q#AD| zV<(u*jHd0d|6~wGm4KhfMv)=0aN+TlL_J>P!6PT`FN^mi+(fr9YK*2)7|%ox54H8m zNR}QNYTEO@Q)+i!gstS)ZTi)^hm`-2hutLjope<|6a7#N#WO>Zn`DJ8mn`b zyJ>q^yHr=J=~Y*f;y0bbJqYezuKIE3NWYL8(1Cz5DG`Vur`0R4`_{>tj=_7JkcpWp z7&V^NU;Q;=Wn~=;C0+Q*Kn{ZZ>?+?&fB8#aG$WkD&bIthhdl%3@K(DADPSgV$ z8oY47AB{uPL%2Pxo32|&kobUH4meRbC(x{Q)oC9!UQfRr+qfGHRE;~WP6EN`Cj$_T z4kKEBGH4rE?NK0zZycB=;2KD4YY4$5E$}QgE)qPOKC{h1$RFtgqI*fzfgEt$^iPJ@ zaWU%Eh%P6ZZEiGORD_e{u+1L(lVSTeIpn?#kY7YWlfWPIsYxq{>Az6Ci)MgjW8tkc zcoRryvU$*wXlug*sUJTX8pY|6$yLBLFl&1Pa5yJ%;Im-_a$9T-u_;S`B~@R>iM6Lj zC?hFRfII{(#0&KsJ9+`lC29tH!rRX%&xx{NGofsf?#JLXp`dF?3>X{Z$BJG zy*n&VEvlJHz7lO6X7;TgZOa}9SFYxrmLEdiNqBa90+4~fq%q1oi?HstJU5RWw{VhH zABYi87F1oj=TVenh?5oKvsC&Qtrye5Rb_<;|^@m2v1RvWMeP z4w8JOm@v^MN;yeH)>`T;ZRf;!A;xR?R*`13RMNG+*1$7}3UPY06BSLXfpc4tY&tIz zq<+&B^X?INt~jIgjVy1=%#BSIt^{ptCDGj=L*;`0n zl*;adD@?UI;xQ24I6<3v`nU{S3*NpX20Y(5Ta6drWkJLiGcqT!fR-f_U}@t%i#iOd z49APTd18NF$aO`HX&tHzZ!6bl?4a;1@wqqa+GDS4sdQ^{K zU8%U`#sWvAZY76tZ>Kbaf$SS=*k;J2QA&c`b#Os?f8|lXywUh@XpgZ9oP;4z zF$DD#cEDG%mj69~bWb>A!0pfx(D+p%=vcn`u5C zXFLB0PQUd`xj^)&A7)^t64G9eIPBw6vSD}RZXI8Y8x1J&+*1~y#x>Muk{372EE*Do z5}b*$tUi!TD735+%8qj1-c-e2GTD=qg%A+&rjZgy#JMT)2WLq0juz)w5~2B^Po2%9 zmBOk%t)>=JhV&8<`A1*FvQGh%J76Bjo`So0WUSbTD9I7mGKy&U)uD4mp9iI!c*utxq z6OgzQ$20rvv$ZJA0|qmiKXL=gu~6Z4y7=d7GrhXRFT1-(20pjwu75utn0)f$2Mzyt zOznX@tevmNnYOXJQO*6fyjv(uPXhJZ!E9D+1M~EtszUMF02A(vO#LeDcHx^LlZos3 zdd6On^_LyHVwaguTxd$fqc$jIR-X(?+80}1Nq8O=&Mn>Nvfyp?V}lJK)Y{WyRpp`>QF0m)!3OX1t1ilNu1AV_ zTfP9Wv1%zCMFwOr2#{?~93Yp?9vvEJa_4S0vCWmU*?PXkg8zZx5SNvS-x&#^sQZ|; zo`#BNNbYyXL3QNT$&go-mgi(8M~)pKfP2zb3s8(4Z%vB{-zso;B_2*zs(@oacy_1+ zdbq$WIZ`U7VpwkipvL}NTlb$Og8pAu3_?YMU69P{O_C&Eq73usi~60?sJ4Wy*Hii3u8O;CYEt{`U&DUacfZ)8h)KGu9+cRj9{06@VeY$`ndvP{e@vUc)VTl;o<&CkbIq>iS*GG=e|wEir1Hz2+vcH<;Y8>C(<5CYU01kE9?JRxl5u(z8@?Vq(kv zbDcv>=I8NV#FP~yRULords9)80 zn{JYGMMg@_UGnYynX6v*^T!K;1xjij=ofy4Fq%%T;bZx7<4Ux`qp%_}MuQMcG#jQ0 z0;$){;9D=&lyjziYZK3&Xp4jdEqI8zK~c#9JV|->8>Z_b2+*=wbU5{DC^<4e%8^p6 z5omP%AY*KMYa@P+X5NZmmcEX&I|6e}-_8k+?moL=Xk$ztCbJb|K_MyQ@c2WS-eVz0 z*%CT_x9=H@EXrnP5ExFI5rj2(1(hFWEq5%ZMdJ?!(D;17U*gZ5? z`K^|cLyjcZ5F-Xqd+1lno9TI4%NBZP#%qr#rk+?J3^bM+ac)XjkXi1%=CcmfQW|bJ zXlK^dd`+#Vg#;YJ_MLE^KA+ykb75Fqe&U4fQ{3t0&BQgSJ)XDaS%K{*clq)MN+FpR zt+|E?ss*fYDf$5DWPp1ayuYraHg`48|Dx4U$;%)1PdS2{mDZ$)S=m=^b-JGsAByyl zQdG%uHZnGS;`O4mMAGk-5!n9~XLU z+(}tPtF&WNyKip0)6-@+#z0X??`u(oQOwvE>K$50qS5gAkpNHu!ai)-!GQK+ZGh(7 z3?Z733TjyMV$2MsUn@K{L>XR1-vpG003AAc1V@9c;2Q4{Ty{}>-!wwY=m%uzL0s7} zkA_?Gn-v+>g?-aHz;@y&_(|!tx#6}4)HIi#Sr!T8=j5i&py_-{* zsoYFb72S9H=7yv)xg+nXF)(-U_tM^NOmnU5*I)bj1j`n{kMkWJBo_bMy2 zuU{)bZF)+)l)7+Y{iDZpb{1x-I?LutjlgEH;kKF)K((9qk;qq zRv&eOR1Ib+2u|7Xr{6yyHzuxjjg8y(O^FXdRsv~n1Y&ZMy`}1ujx&KX9^Su2iR4ue z=2}}WHcqqX*e%@dd_!#_Q7E9A@l93h*;JNm_<$<@zDX;TIHth7*BFJyqer! z*UodJZ)~E>(9>ob{p>5tWo@k|GY-mkvVSswlDowLxf363M(KI?6V@hMyfy^TLtq>9 zf`oDBj;)J*ecRRiGDRe+hEheAqo~lo3y8N!$qnV7o4^0ocHwMF9$s;}HT!W>LDgW% z)5PxT5F>b!S6zbk2URX zhuK^I-~^~ft0^(6!nrXCj>B%w7Ca)dNhkp{`Vyg<<_3MmwDqelI5 zJnr%9#R!g%icqa5^cB?b%quWcB{)P`cJz9i$kP`vxNq}gWq_sH=p^VMGbAQ)ie1Ba z#KetFZ;I8?ddPFlM;s@cy{5d?Omj9E!0v3?p2FR|(R<;huc`<4Ht)vCSy^r$@O{Gl z%yZ@PQLZjtr-A3@!XMAgYV99`H)wj@@bd@vl`DF3rHpLzNnA_$qk&vT57PhWZwgSL zxR4C32#`>Zx7VeNpA!xyUf5dLW_8{JY=-ol&!MLoXj+y^+x7AHgt|Igk>%{KfZ8mx z&pBrUzo7{|w`N<|f#byu!vd5OEq6Ci$C3mtb*!_n3hc=eE@I+u`)Vj28`QXHbl3U@ zNF(yz5thqe_luIu``*1Bs8nL_L67o&flJ~8f$@U2nQH&ANH>&~F3rfYpbZ3S!P5qL z{(9|(Sygv@-Pw}VA8cnnBp?mzht-9hs!gvPRNOACdVbvP9H8h~4wMJv3CIxoMXdto zG~enM*O6B|o|Sj0aIQm9Vuyi8{ckxy5%;^t^OLFl_YzX9_u8p9t1|an1VXAlZN(3} zf@g2m>oG9~KghRtH;wB=j5*ICV*4{IucpY8AInFeI})RSw^< z@#^&!86Fs;527oh16C;3xexWPIR2g|F_kNZ5;`$lg!PbnkKTA@rN&+|LU-!;hq z8W7{MSB&ztXk9A>dyY~|@Nq5RNGNEJRE3OCkUf@TFUvEEpFE`$^BRBGtIV62K;kTW zK-!jqF*@4mOGHwezt91qWW(5c8tqLa--o;L5_i=ucKjt3yC`jbhbY*hJ)*iZCzsTl zF)Mtjm!Mjk4VkR%y}IFV3f!XrsccfKPP*2J&1sFx3_-V7G&VGFmX{e=*raw6iIpVu zd|wFs{9w+&cC|vtHvGyId`6FZvA9XW6XtT1@Mwm@&a`5mZ2yvtbT;% zmc^rJ6?P5UI4~KfP^!#R*t|-+lP4LA)K6B>(p6RP>2qbdg%+U7wX?<^IhXskm{uFZ z00ADbji9?P0W0{HKI;hHy_MCpNh?*Cb`%lB%Z$Vy`>LS8aQM9to`BrU>}ay3y=>qQ zyj10htv%Ov_qFB|#HrLkM6UA*u>Hl`U2awbS0c#m(vH_%xPm2Hk3ts1aIBQGldJnX zq=`9d=*6piE!l-FveUkbn|_eBjI$=zq5L9yoqEN=bt;kr#-{dIcswDWn1WZUTkY3g z6>wFoQKt=&N3?5-;Xjc4cKQa^!jIi!TEB{9q`>x_iFG@zb#fF~T{Q$8PB*a|)Ez8= zxRfxzfar`Q1Do`p3?}|c4goF{d3v`Oh~>P9*oYX#Dr>TsTdF3^xPMecaG+@<;d-)s zM2ATzQ=vqeS&~r3*zldS$WLH#w1e!J;lm#dI^bC9k&AcRuC^2@su;2NZ2Kn;G|c=) zqEUYHxik4d>{-D(mM8!N`~)(i<=aB|9x?5gNFh(IK z%eW5qQ&clrj{4aN;Y-^J4!N`Q8|y3nl4M8|ya(|qWC>Z;I*;P}`Xo&{DdxrOr#cZ+ z!+a;2colvr>WMp$EFd_|HB7X*9{C%onkGkn)I3?bu*q=jkd6i*Ecf+K#$cx=0cWbj z73b~x4&?S#0OfB$lbKM%|4e+9XmOU%Lrrftoa^7Sm-$^`wQ~G9ZEg=$2j2^U!tQ6~rqcsAI6 z=whsB*X-cfirwQihT#`QRLt&lEM3<_NuYQTM#p!yQK&h%mU0G57fpWMcMt8lUH;jQ<{3xK}ZqKQ?76Ki@Ws z-gpxL1aIRHM>bmLL7m)7tD?1a1>did)5wBAwp3A8UxfRNS)}(FZq7aFvs(MfptsO& zvuXq>loH7_Pkik5e6h6|-Muw>&>`rX|LEg~$AMDA)GK)B_UUsMB`5fB6Qq4{w96MF zs)j0qZ*9rsGVwoYv>aRZFxHPKt>d>V2V|ZM>`n*L_B?6@I!k>+SCK3{2_1$D=$?ZL zV7FFL7Ls4oUHXk!z4`V@r`6itmH;?(*YdcmsDAaS{Gpm1@sSz{)Z2DD?qcw?-tx+r zm)5Hmsk?b8NN589%eq@-SG(fQwVW_fhQK8~M&wyHs5 zm7Y}G(>J8Yg&afF3Kl9!9)srufJizYsG|4=5zs)nbU53AYpd78x2_NCo^_h-y)SBz zXqT}!txu@Y)Vt4RNe%(hTS4H*|GshL|L!q9eI0SZe(W9WEc)3_n%$p|>>CGk+V_CN)-&&wqV0L?W-GuRR!VPD~;37k#BOPD(GR-exVIWT3BC#0Pa$31VS5L=pSO zy7rUw4=OI|jTmmqxvtXv7htnTp{JgUTF{Pp20o*-3!(2<57^I#g@B)IMbB_n?Tx_2 zIbFEIWY1o%cFhSKU9br3A4lI;=cg+6H}U&C1F^3%aPX-hisV1EgxkpTdy?BC>usd3i<5FA-Z(`kin zUT-`MtQD&Jz|a4rcJR+%Xa869PROylf3L@t@j`Ob6IwQ0MspU*)|6@^KXr4%+Ev-* zFFFw)k`}A3c+Q9ARN3=b$K{7+OAd>KAE1euq#0Vdr}!%mesKSXFW#W`*RhI{OwEjq z0ha*Nm7(GE?po9{ktMImzi6rSL|w>z0cfe5uckkxyMN!&d)u{LYP3Zq+a&FoC|@a+r9bddRlbw^5@ibZ7$+p|Nh22~9F-)9W|$R`*xKK9xVrTd;L0;#uI4c~o?h z2tJ{)f$b8Cw2J!-z$%L@_&b_zhV;xVtR+JIS zWc@FoLuGr6`hBN#D2ct3ac;xVRSTbX73hu^G$sqNWOu-RV&F3;6 z#qLr_S2e@uYEr8pCU`Fda##g*s{5mN6a!E$I*lG^D*6#{HjCtUEX?r5h3Nxv%n;IS zO$u91HM$4K3G-s%=ma0w%>?82CP%N8=OO}la2Qnvw&3VB$5;y}4YZLlP(UPwTe1t5 zc4vbIVh_#NljOoCSONoI^8G&?{l7iXfAX{WWp4v8qyDaT|NrFo|9tCtP~Z9L zko-5eTl^Q&ApIxmh5BLw%=3L@Z`5Ku6;5!S@!J!VSYdZxg+?JxH=S`poTBQeSHMS2 zMIRHkUwrU;@wTk=!zI~jeYL6(mJMH2AEc5|CX;*rdcq4|v5x z_%<{M%r<{V)TgMaC2?Pr^f4>4#>Q6tgQswA-KH-a18U&BSkf5&oBT#>#TRv!p#V4R z$w1yaH~TIst$tysd9J4BOpKe`dbvAu$j58u)*xo{NNw#xvdm;=DoTWw(Zr@P%X-SG zE@*S;(icT=bLHoaHlbjs0cIx-L@@8bkQ;d^pXsu-6y$N&<>8C|PQ^1Rv%tE5>W%yS z&-KyD5N%fC1FEu*G33e?3Cw8btOrBpo@;R8Vr2cbQZp{>vci+P^iQ|rFZ!|#3$M2X z1QLpM@S_R6-(F$#V~^`th|~k>hs%B8NTEF_N5!ba%u02}iMbjNwIPnLDCAwsBI3$6B``+kPvqbDO@IGw65sl0{Hm5&|100Tjj+(x`=xSs z-aMPRk62tvfBd9&Fb9NMdeoW2Ifo7eFZXG!_iHx&*Rtn0bb{dtz#FJy>>%Vk)Sty2hVB-A6mz5UNz8{@2u1IstV#O>Gt3t& zk>u%V!`-p4F15?w=--4{;kTUAz7x@f`h{z!eG~Px4UfG4vKrfhhTyke1vZQ*oN65Y zv`Ha@-npSqQ}`Or^KPllJh8oYt=vvYJkbEKw%MgRJTLTO&IXiLWjDIdBH)Zrr&Fr<-)z9eo5;RR=xREhT$Zq^^%~ z1v0Krqt6mJqtz08gjDvCZ4_Tno~sUvzjnFjNR%x(6=? zLB_3CB!ANv-V{FFsGH}**& zHTG7+2FM>)DEd7$_D8lsW6Z#>k~S*0FTui0_$?dR@8yWF1}I<=yv}gxM(3%=vt7&i zhupq^S{QTP2m55&LLa8Pi!6OW{n(z&Oy>75+NvdiPr2O@zl?0cvBjGsnwceXO{BdF zzR9hFuAX;rtg|GdJgU7o_-*0A`Dnl0nWF)smq>RErtq{vB<`+Fbt zA(?j!zs>l5y5okt7wFv5v$OWVvs?H{U_Kk(R9t9iakI64eZ@z-wQp?UVw@4drw>s< z_B^trLlz*T?+i+N6hEo_7_N2NLl|KUSwZy?#*!>_a)x|YANwLj+N0B5gUldhlCTwPx3$(vxNWT#Gr$c?f zcTv>1P4x+i{Ruz?AE=eT6dfo)0fhRHdOU?56ilb-0&1TN zpn?7vGj`lNaVkBIel$)ms{upj*=pS)GqY3nfc6GXoj(tG0ferz&q(a$q$6~Io2rv* zWChUs|2(4qaDpKz!0&(({Oh*-x-Gw+mS0cHuSM)vZ21*ieueyBnb@DxxPO66O#h$R z40C|n`xvfYDlR&F8@$~Gr1fYOs)^^Z%|=i6rxWwfFZ+3M?=?t;6`nB6lmVsqq{d{M zW^n5{nKh=_tNk4@rV=dre~}pTFXVw&rrAN4uE2Y}bSts^C~}uir1p#-$ZRbyP4S## zu8PrdT~rrnTz~$W&xE}auB-Ur+X4sfn%_2aBC(bbV zh#TP5U&3-8oDq4Tq`EJ*FFsOVB!OCN(DMD`8v0huxNPn)1Fm}jouX~%&jMCyo8^Vx zibon5lpFdwXE=8Ijz8S;V!!4gMEFWnuL9asIhzz#r-dJ0&j8WR?YtNjbD`wdz4nnK zu%SxV%oZo6*Npj}!1G_?CPb3sw*SL$e_OvlpFO ztJpc@!2dM;zM+Tmx37D+&#|G6n)!#DIOVXqhLaz%HsOP<>Y$sECgpYA zHMi~#?oOy(m(=@`=&m9 zXwc#PB^7tSGzEpiy#5=T|Mshe*V5kgXE0-ftQVhg-`5-_r}NUa4!KRF5B1kJ8puZ; zp080T%m$8h;X2j{yiXf=`?$o(Sm1pz-`5-~@A3)*?-TsTPxy6OV3>am!aocGFcAlg zv<@@>&gRS;Ox$8GJ@OB6OPF-cSq2@ZaQm_l3!G(m!{mU+Zp#XA@)2j#Dq$6)0EH3@ z#_E4hxdTky{tFf@h<_a`AN*D&jx*V~b=UMSJb zthYR{Z0(wpUv%ihB_%sW;p2>kPTi=rSb;gRuNS{jC=KdOdx}u8AN&5B*V*CZDc&bs z@1M-xgVa;mN%{o=+5|4i{-lrLOFHv$N#XO)GmFO?E#S(T0@1}LUXiLBBBwpqB<~MT zs7(OfT^DJn=S_-)qu`ZtM1e6-M!hPgO(~5JVl;y0d;wU-45oH5IS*FMJe2J+_#&CK zxmYsu#moHDhER)evA!ZZ#X?9XZ~=&03rg27D|+gw>;;bgZXB~*`RB=EWjVhUDIgp4 z-eJM*{MgF7xI1$|ozjU*+*I!r#Qu_kus*xNaH*7-cU@UU{z~=I`rS9$!E1mXkV1(Ob51t23V_K&>~umU8lS^K+VnvL7;B?%M}>=zN?s;pQr z+a2p67iJc`hH0PS%|PcQit0~>o$qK-XjvTWD02Oj!q9!_qfdaIp40&ZV~i!HG~U_5 z&<`EBAegZWQ_4gur=JW(U@`SRFm(s^wV&wwlVLd>u8lUQHv>~%6|$YZ!2{iCM@XXg znh|~IPWqt~>dh1Y2?rg>QaQIZH+Fy)WO-@~$R0`etyys1L;v;t>VG?)<_g`}0;)nU zKm=xhhKE={;|;&DSM&jApV+=0n29)8nu$k^0PS=S5q;d$0?M%zDgoLR_t(#3K(t*% z>~a92a6t0%U%5~)<`~c-4fMv54goTLxqw`{4qYEe=cup$Ha0qAK=PI8av(s1{q4j4 z`q_ZG2rrdFMa(k(`jWryzVwu@ccOp{90ZA2(Q0F$gep0E93c<*8R#|>Aw>6 zf3{uvKOZM`WiCB=6v25qFvj1$6&(zSZ-qZg*rp$0&3d%$g^PcyeneUr({KFslnWyHy?{%+rt!rJ&#va60 z;%5E1&<%ZB`408RN-Ow3BOB5fegN4X`7I+2&lw**G&YzGeG?SE%rEW;a40B*F)W}h=`wk6OM2-Vf@Ik6c{r3=WW2C}DTg!$ao?&Gl-fLyxqLdoKjzpj z#7D)LO3nSyO0jGbB#@|Xi`7aSW(kpW+4v&a>kD5(*L8%DAz-@F-gi##TLY`hXJ@R{ zFs7@xLHu|v$s&ZDpN*S8_vAbvo7wyT08Mgpw($*zpx)oc;3+#GptuTCMuHy-FvXN* zS8q{7Qo^C@HpJckS{d{6@7wR0>95Dx!)Im$FGu#20%Z3oZ1@gH@WWbK=LSplyI_~5 zu?pXF>toekR_Ca%!`5nr&v+zv038c&L|A^h-K=B#$TclQ`gFO4y8LM&8m2>CtHEQPy=qe^waZ;g zo0~G}V=iuU%9nkk19SPl?^Zn~=q)-j|9R|;;Qf{JDvQic4Q)@@K}~jhfu#cbw`D7* zL&TI-X!S)bz3?9b5905qnVrsyw3bI;H?i^U-of%%gNUk1GEASK$(k&YO(}WkZUJ!o z_>*}_$aumPiQ=Bg17$ePG$1`x@$Dq`N!ATx|86ha6@GtpDJ|oVxxjMb*<1Us49b7_ zFV8H3tLC^J1}uM6!GXM8uqLBU{{}uEU@zSpDLO5gBAR}=?#vBevCh^M#th4XUCkrJ z7@{2~r0z!3rxZ0LJQ{LQbO~%?ukrPLzkxjq<3oxZ!7N_{Bzk^Q`w&ogL}r$$?4#}M zL}NWQozs{~%ik`5R+J%PXA#eTG807rW+EV^3Q#J4PuNeeF<8M262R58Wqvkh>gZDO zK*1%gWTug^NuOJ{W>T3R6aalW;PpR_{7F&#<8^*J5wd6jo3dTO|JMB?wJn0SxQ!>s z0Bplh$OboTtL|UelR)8v*1tZmE|p-o&5!!2fco-3_N`zwnpH|GTb@>+y_ZR#3%liCv z#{9CezxK<&teIb4!7tzEmxuSaeePd=?9aNaKlvX2nRXH*QKbNly58NDQ!@ z|NQL`epcRY@Q`$u6;`{7w=|ECHpY~{#zk)r)Dva!AFx)84>fYbbd65T)$VD?P}xOx zRC)0Sjhlh-3fmG{3rebaQwe3xHFpl_DNtua%#Sap6%53$V1vBYH(>IU(I?1u<8v^d z6`Bu^MEuOlv=`H_r+VFu5LMzx^?~Z#Z&w`w3Atn7 z^BT~xZ?4j>L)uP@`op+g(vPvbXVLo9JONEiv920%fBRng)wa9oO0o9Jl7W>UR^CTz zd+(Qs)@zZEgU`J-mr#|R`k0HZEjrd-b0#&%5`I^kXVe=?V3tgp)$&B%w?g0~kJ&1Z zzf6wL`<>M)hT#a#GPO@SIZ_ikRm@OaH%@X~nmn@%Xz1awMVkEn2g*L(E^+4iIPV*Z zBYx;Jo43pwhXqj~fg&RZwN}h>Vn!JeoTJq`HDIL)m7CS70%vBI{bjx!{$NjOyFU!T zW9O7pqPaO6DsTC+$Am+tI(lOMnPv2+bLn_Su>!5!L)jfDKCLKgXQS20%_Guzww-DB zQD29~LYPm5`~mDnFITL9;KKd3UJU~7(&pLr+5Q5sYC-?S{_WagTQbYH7@{Ox?_ohM zpk$(zhYAH4m$5A!1G7wBfZ8+;qwa~PHX}Zgs@~P*Y}w9LDV-p-_5reM2pvDdA1C^M z@H5+gc5zIs5(Tq*$uMYv7Sg!^%EgLrx6&Wo??L2&MlJrlEP|Qn3y=V zG644g`VxGhCPdROI;Y&0I&s6*dz|(35454_)W)*rtZ~JaD>p}19@7b5mU>+p3)oL@NgLbj1a}xaP70;?Y#B zds*}HgV1;H2ggtVtjWc2m~4%B13=yxj^9VF1q5HIuj$eLN;oaEQQ{s_PNiH)O5;M@ZUeRxX~9P&U5;HTM{6Oj z(@#+(-2zVe)_krMJr<-48x?vtZ~pp+hw9dos%{EbRpd8qn3ay z^oa&nxj-E8(D!|{JX^tm1cq9DDd#-Z-CDV4!S|;$sHw@T7k4gvGRwSV`Q+AHI$pjl z+xmWf1kg0u+bt;bK55J2-N+8e0>~R+#i{?M^yEJxu7Yhv4MUka{dCyIzQs*Z9MB}? z5q4Y944508V%X#f)E0cfTV0pA#4rq*0~95`(5I+wKu9VRYOz2%%>8_O!%C?%9B*%gXzjAR;Uo4 z7gm`ahk(z3)Moh{vc01c0SEL_@zp z{pB%F{`D%@kC~}7C*^0y@RzmteTVpE+5g9`_sd%RvKIeqpXryi z_+>5rQ+l`1Ntq!8K`BsE=n5Cbx?`!f2$MAYv4bdw^2RS}AYG9|u*Gpez6^CTr?F#= zdbnQy`ZL4MHM#E`{CB5MYDvx^eE@pVcJ1~yS!T%%h*zq1KqLrbw?4z>e*t_vE~8=F zFe#7mw6gC&!5a1A+@EM%f24H%9iRVGtCnX@#~+VMe?fp}WyFA5#^H^)Vnp-TDivE; zush!l=vpj`_wttDha0L#Mynsalgx|G&onZ{)q1J_z{{sUPg%T$A{m}yWN4}g>FbbC zA+z5d8m7QoLcZ97IR}EMo{EC_dW#G6nw*oTig@oW#ws5w8x?6uAQr?VhrCJm20WFF{g z8%-5CR6k|lQwv15LJs;j(^XE^S`-~LIFZD!8gCJ^;Jhf%kNGfIz4AdMT7I#2tN9TX z--tM(sW`IGedXhvb1BPH9NMNcQ(D~H8>Q}&e4 zyHQ$5N7sMOKucSjr3U$bBK850&#{$mSud3n@3eU~C1(hx0#0vyvVWwfC*B{Sf&q}h z9mzp5z)otMzcb$zlS)+LKX&8?TA$YQ1;*}fxOZ6hxR%0r`@D*9bs1+TKI-l4P13jF z)`HDMrMlaZb>^*?Jy26V*m}s*eyX>HVT0@nGW!isAL?O3h4-v#A3G@)>Ue8TS|7R| z>7<(OvwqW-IH+^wBCFwGXH?~Hm7tRvR{sc)@wd+T2S5&*4H7fG$W5uNuQ|~Dpr5YiUFzvM&gXi7GX_M^l8jM< zPd7IqDAM?PG}}mPy1!!gyT;hSCkyLlR`iV2;g>&{w^q+Y5x_%AOm@zR9nf;eVat|} z-U|HN{I~geq;GvTK8Qhb_QDs5Uv@wPnDOS?h?8cQyZW$k{%gfCrIYKZ17F7)q~A}P(2GB z*(aJF4CG!g!kXN;Qu%1UZu+w?M}>SZBIYtP;4f|skiPq)D}$Yr0Zv!MzQ?LB|DW(* zV1M?A)-i{@&6w;Z6A}Hr3Q6j1a+{h?an&H9T`VOieKUBc5%B9c=v2+uJXC{;z2gZ? zw^WaRpH`{hwheTP47v;a2wFgp0r+>gAq1rhaN6pb?12EZ3t$qcV%#crzWMjJ$;U2V zHKfx~en9-X=>|k0nqq)DOzMB|DycN%BY?0wWVD`P{sq+$3XnTaGCkBal;}M8I`Wmf zsir=Mv06z&Xz}FS{X`2kx&)4Ndk3@vp{JsPP+ES!zmJQ7lOfOyYV%=NkaDKhoim|- z_n2hK|DPC=|BX(n6fkk#gJ(yOXHd1b)#d#Gn$R7=P-7y;(B%#WDWeyzN7w;Tq;(sy zA*#UPKUBgd#3w}NZslD@F zLVad%#8Rt)^X%L=&8^M%;E`~soNm87!_Oe(W{<=Xv-Wej=j7T|j|uerLw)ytL%;sQ z7w8SAmoW-)Q;yC+NO3;8#xD$rj}ckps}NezrY7WP^&`uA>@xb#jDTs>k{N)$B{kE5 zRR8%H+)d~P6M%`_Epr)s!N3I)G69%wL$c=2N?U)QOxyo!=kRQ|&>rr9uz^fgdJ49y zf{t|yErkRRfS|1LsDh$f_rL=0XIDk(njWS33vE2gO74L`i|e?1c4-y|(Q-@E7R zSB33X+5_ZQ0k#Gr$EbHky{EN?FDN(nXraLzo|1NpmhA_h8&T-9&_DZCZ0{2o7i=7Z zE7@E5k&?VPj&Gm^a!lPf0TF!a+MsR%HXldx5M`iXP;@YEFKj#wwMUX-n>~*atUAn* zF(CAI3foqRddbe1fsJbL;>=JFFyZ~Uqe$=^Ql(5pVdC;pw?v;CRj3if3iUc$b=N;Q zT<~e!ay(!=9FP#|9uMkE#H$@pDV!daM?uYT;yYYmw46tV@}iHz^Yv4XiSgC}kKXmzEyOd9={%Fa-K-5l-lAv& z5_b!Q(OZ)WmfYUm%j}Ol!Di) z@IM?Y#zz;Mp0`mt*h9Drl_80gplQ5Y6`(Hvk=Qu< z8N=wKCWAQvd2E<5fePpupM|Mt4*XB69?opUt*QJ#{l1r8z!?%yTiOd9UEa=&L0E-S%Z#*9-3x`P@H}Uh+umNUv$m zzd!5?CoB_}eQ1{`h1)5lw#C{58|ap&l_;MS;b5#b?En)vEVH1`Sr26{8#i8={t~K4SBkR)1 z=O5X4r6cbyV#?6y5{IUcy8e$NZcXekW)NS0Y5E1@fu?-Xvt5`^aqt<*TEijZNk}B! z@9qch=VuEs$Q{tD<6AZlAf9J^pN`@T;-<9}6r_|Z&t={EyMEpO{+O~OHW$Sbr8nh4 zRA>Wuc^7F7b=1sT(4dB9kfCB_d@rKR-g&{84ed)Fca^(!vgC7I?0_j$#%e`e& z4ZZpJU~+Z4+5_C#Gdmz>M-v!tl8XEfQ|NN@Ygi>9n_1MKpLFz!jcR-L=> zwOTcK>X1KtW?)52#EpLH<|a!|cZGCI>dk-Ko4tJb`t1#FP%(Tqq!~SY1G2|qje@A- z9kWXw+cm?1uF9^Teu&Nza+vi)4U0|Yt=@6Gxcx|%ba8{o0QIyj_@imha3mY>gGv_LeSiI*z;YT@C>+md{T61WlqqE@k4TL96*E;gX7iI@W0ww zqj`;RP43oOao@e&Fln)VR-2X^z8KT)cuA#uwkFxnfCiOwxz1qoRfEV2jIsb3}u>XFT@tYDfmXnbRn?gGP zPDA4kDAlC8qDuFt#(Inf7O**d%qUY39e_%nIigxF*8urY)Z7vLN<5MeRjY`bN<*Nk zU>o}vw{)KTsfgv@sT=#Ra;$9C#T}4{8E%UM$;tczTl3Nc_~A(W4oL3-cBJDlkd);m zGU^e}QJ=u{q>d=qSji5k`%WgyR{y{0he_qmrJiy*e<&rZ=7TL{EMEI&f@SyJ8y&8J z!kcIGoLXEh+I*oMGE7NtV*_CRdEzKr8$6Ap%+4|JBe_}2XxLyZP|lSr8I=wRzg*}2 z&4`kQ_DZR_Q^m-u;T^hvZN;pt1)!u0=Id;czmcxqW4x>yt8HG7F2A2hFxlknP#->6 zGo;y@QnRaqQR3b!xkP@~Oq)V2tIn>#9EXECUZxWPNA^8xS$Z~s^Df`8$|x!b@sJ^$ z_El(c{m>%$2mXPevCfyxbwE%!3NGtKLr{`u)lMm*hP7cdgHH+EcTGA-vI$&wUwvm- z|ENicnr9HU@ssqi5~Yayv`>J`{PZ}`ZW1c$JY?K-V|Gt|K<5r%`>Upr`lGp z$xPd>umMUm&9jIESm{NmcR}e3U-uo5tLpY@jiUEv1>)1coMNsNs(ve%^485Qr~Luy z253X*$p%(ZX+pp|{7js6*UW?R4O#B$L>rq#AYh$BC;ao9HS1{|vXJICht}YMmr~+* z!s+eL6e8{X=eVQe<<^3qwne@$iCfT37IMYvnztlDgYKd^Tsi*i_+sJhEl;2*?*-0k zEM#qTfiGFH#xTbEqiDeMOrR>I4u1U4au zm9bIm5A-hY7%(rzq}ja8asF`zX6rTpl)}2$o)Hc2A1$FgcjFbs46w6Qe-;&en^cMty+{#mwbb`b`X; zEgglwI;1jE1*yR3uRC+oy(oFBf!Aa3wXd)a?8n&lDs~vOM=*LUp!=bNiDPC_`N&p@ z@8B(QW3=zCGgx;ZqfrBLph^ljt6~WxRO1`{UG7lzs1Yxb1wjj$E?ziSjZ-oMV6jh*7 z7#W;|2v6o`q$4H#rSYf7SA%jo*kkx#X)mEFq8J23F61Cn@*(7i`#z6quIq~;D#xbo zhHo8U^b8f{Xf;~~b?0Y%n0GhQwsPV&y8Z)}uqQ@XF$E!JtWElo=ykh<|rmJ%Q5C>bXvlNq>}ps{wpyw|GCb4T@@ z0TqAQSqbidS(f)2Idw3bjtFZ1(}@g{LCNE*2QLJPc;10z8vmf;7dIRByRp=YeETed zuLI)kg7lJ*=#WoLXK^zuDVqdfHMkfGv~9r7g5EKdg8{&%@%cs7XIfbk^|6Jd=$%_^J&5!rQIsWk8JQ*B^S$M5RmU`B9wSX{QfWWmjGc zx+;GaEzjyrog&?%PaC2srPk347z1tKN%7m9wT}?G)uW?qi@9uP5hdw zxiS4pl9mtd#7~&y_|Ps=y2dPeT8GVW$5D?*M@W8fr8IveC6Cm6E0PjsN#I@szpAdj zptKca|H*ObyQRYEE?VWaGf&1pu`OE4kaQ=Q39Ei}^SOg}PiWB^TGU*>r^e(k#`Gy2 z68z2AS`;t98zIrbM?#GRG6!f8O%&=zlASxeaILsA!9#mj%?scO8VdPF^!bAcOgH~UtU z+*YiHUdE=LWgT_O(QR|)tpyuv$gFurV9EXxnFFSXYgSBs zje(oQ9k(^e9WH>+dHtx_$_qfgZ>$&=E>i^|)f?mGiJK%JKVPqSzR#L9@vAaFH<%e2 zcJKOznlq?`D>TVhX_2u|Rm^9MLGOjYIG-087CPp+4BZ%S>!UC07}qo`AS-8`KeQqD zdE+L>8Zn`7i{A!EiWdfTuF*t2S{dj^e=wu9)!kUTtx|-Y>yNjo;EQ zU~~a0nat5XnG-%@$6iLDGOPm}hBbFH2NE9JcT!-@yfb^npF$Pg>dQthGXwL?KSkpW zrgV9+c2G$6t=Qaefmc6oKT!kwrf1nB9+EaHqMnYFyMCrlRwFNYPfp+)X#w`s?$Pdx zyT@5xuSSB&4yy9zQ4S_Xx*Q}|!ZaiMR5HAyLq?KW{TAoef0>Ma?Y@bfNvQv+Xbk~Ye?kC_8oj^98dX6+WwSz-U{9~dlT+e-hFa7 zHF;KJ-L%A{>Eh|cb4n)H9p0{%MY$t(%_#|s;7@Rx4L^->_5EbDrKNFs*}UH|NkH5q zJs!ErvS<_$dVrhjjBYhj&0mX=r;b=(NKk27)`{s*oxm^Rv zVMjN4sCnZUfrq1NodkTb%B9kEmS^WwxvIZtXxo@X1kd#5H2Denh`4%x3^t*qLDGF7 z&CZ8MC~$&jQT_mstSLRbJWs!oy~Wm#gK6$Y_9dJhro`S{eg4j-a83P_Pdj6r(a>RT z#D2zTMoV?!oRRIq;uWrdtpl6#y`P^lr@{fH6311gMhFlu^2tb4km4(cOv|?otesv% znzDVfuxD|r8O?~`p0LgGd-2?}{Gq71*k^E%}N@@2291dD@hfE9E{WX-l{Rv96Z&&vk5KS>Px1F~QB{vw!9Y zBj+1a(?Os;gfwZ?E6*+C14;OF(Duz?;r9T*FXaISW>H4zD=A0ueP05SVeU;55IMNgr?D1$@ZhN>^ihj28 zP^cQ(bo$jHJw*@QqrBY()U`2p29(q~8A4BFG}BghK=bF|D>eZ=8`yeSG-Sc-^R0lnbJpyB#Iqr)-0{%WKm?Etu0+zGaC znvariqS(u}LhPre9S4QXTcAaKSANqJwO0U!dgPwiB1m5KdD#o%^6fCs#HfM$Y%lT{3-lr!rE4{IQYTM*l;`Qvq|KMTl0ve5 z@xXKG-}aI~279=$pBXV6zE@rr8~4(1(s=j2uPWI3cf*jOi*=le=_qxyfYl3La0b9u z?}V_R9_562?4!#}LT#L-l;NinN#DJ1Gm{pkpLaGDCYl-Y$zbw{P(~=iGPKEqJF_Ye-DK9PryGH&QB%Y^1+_Ax+q?low?Yy z%e9SXRsh$C1wFjO?CCQ(h5EEikE7z_91tZ5bxF#ccTwbKp;CQJyXsU9`U5!H&c9v% zHT5jzW6k>f0(4x%fjrS;HNu|rTd*KCWrgv_rhXxCe{7nkQx1Nouo<;*M;hhrkoyp-1<179-U+}m_s-KNuxh;k{oJ9&OV@@lk{m^Tfx5_pL>ig%D97@hPIrz` zwr$|gAm^=VR`A#>AC1Wcl4g?#!`X?p(Vz56-xcrFgNgaFP6mM!J8{e#bM zoQ)W>-0RY7qs~MPF4{yHTHzR@$k?`nwh#6qUx#SkhmCe}vgp7=z7;;qSLsizk#u`| z>;CCMsbsf)9az}MH{u$%neN}9YHz~zH0=HcAo@?@L%IeJ^L3jwr3*;5!Ok9xOUuC{ zKI)s+-0p1UALAQFN?;N=Ck&)zf*ATLndOt?Vp_qTLrmXG%)pNt*N}0{4kidqn5_?x zyRbhb$~>025jmPc8!e&SC!j)2LMD1Ol~?_BsC#@4hQ(CkJdJ#`nv|CjX^&C0=c|OA z+3VgO9|~mI0x;?(hi`)H4C5%a9UTohya5)XjC)s} zFQN&4t4$MpQ&hqUCrVR=pG7K2iBC$T9=vwDG@UOSmJ<5oion-uewJ-~FtZ!8rsJ=* z5x4`=m4FmwG;mX*NR|5i)3E`AclM61Q!DT!wA9!VlWmrFnjX z2Ral`JV;&yfH2qq6Ua~4clr^U_T#W@ytP_jFPymbgbDHX`{FC_`ylF3K_o|WY_c9x zG{9jEOWymbv}!dWRCs*6!D^V2RzeejW(XRsyCx0@*y+B`aJjAVDTdp2+8IAm?D4RaBTMzO?pALcK^<<&+Wi# zKi_{_uS88?Vmwh$^Eo4lDGmu6<8UA2)K_d3{d^ZMc+XuZfmRd0okjDuW%3G*s?ssk zGtz+S2}{lA1j6v0&x~G(7jT^h{C8ZL7d_4ln~MStm`fEjd01DB)mHT7TQVX+4e-{$ zsHj&aTG->WnB=E~!vT3;@YlC|fruE~~ey z_`D5ceH$NN-uC}I<*c)aF=dp!e!$08ftf}G3X8C4Gt%mByV|%#`VB{JcH}t+`FF+L z36^sGCZ|hOKbO{(o9UV>tv{nKq@JL8tmv&#@O>bsUlZlN+0Zrg8ZI=+(B8oAx996( z%=IF3Sv?iPCNxwO9Jy9%x49(OW`aIzpzAxH2>3c6`175nry51XR->57@w77j278Sl ze9T>iN%t@}#PbRUxrk=u z?=28e#8X!8Dr&IL=$J^EMUsC7BZ`tVvF9D3np}{IdnfpQ$R(0aFgyrdt0#kto}nVZ zK2nUOGsCEHqwgG^gzABB76zx_)hFvyVP;jVq{Vs|*!`ru#_q1W!N)rQCACvN)<3C5 z_dV{~`T&!#E+1XH=T-$YtfktZH^u}eiFzb6NbRXNGF%1Mtp+4nOW)CwRFQ3_8w42I zQw7P=loBO0UC`|V!p$Xhk6;*W)a{$!$tKXI!Xlp4Zh3JBbRoEEoy<#-&2X?b z6j}_Mm9|;><_g*YnF($q81S03N4bz*>_}<6C`ZdYRZRzr| zBSGt!OXjUPS%ea@8?oOV5vsYr{Lb;)D=Fq_PlPqyC7*n8H>FyCQosJ8_(12_`gc6H zcu$2|26x|^Z64Yl@8rd6lt86mUh|9o4Lr>VsT0nIi->~dJ%_y1?VEd_IFRbe(rT!O z7P7jA=Kzx0r7MVv_?~E!H?2>Rgl$6rLgNWPIYsJt_0-kL_aLf6fC?qcKI5bKHcpx- zTfT>E+e!EXWH(m+9nc*DI?9yfCIlS3r@d>isUsurMrOmPs1G#}>Uer3Z6u}5&y;0o>&k^X|? z0CO0K3oexe(H!a5DoKVe)GAp?>O#xU@tqZ#*r6w8gyi4osaH`;I5zn^D`RAml#OhN)2?W$<=K5TsfYP<(o!kU@J^3WWfu z?Al%ck9cg-__)=r9P$Q1G8)_7L%G5ngKeC2=t5cHwk3}MlttLay^`6#_2myBT@=kj z_w;RE27p0vGgB&%3rtPLHiHL=1Z4UuqS=$luPbZUBPWZpR7QG&`ftg zExi3TF8~tc+~OpRBB_|f+LS&Ku?49rRU9^?)mhnfVosmvQVx~66>d8p5=h@gK<%-U zeMseq(b{v7Uj>;K+sfq|YoUL~^h?JnPHDAG5gX-slR2zLsaM*o<{1?53=tEEMj0Uv zySVD`G8XsqHrBvyd766@p-+n?pz5z3AYg+}7Ym(XHRe|vj?p{S_M>d8+98Q z5jb+6F@oCG7z2AZJ4Toyk*bOg1SoolHi>Rf+h(f|P{^HFs{G*6(|^19Lw<=sZ1|E( zxpBq`DkiV(;2Fh2DYjntWWw?oY5giGB%*kXXWnQVoFf%J7mou4l0)l^`NMLJeTP(SgXk_RsGuilaD&^ zxIuJ7nXnWEk%9Pud)=Vi(XA^S*juh5OBf8qtuaNz>R7l3{Mypgq`#J!#9?H z_$KGerkGGCGst~xo2`~Pj%^+r#!Msu?r4=e=HSMR6mQ~%@4MJgi7zi#-?R~K!dc$+ zymnN`H?$GSP3c01!tAfaW<}Mwzg+2ndH%5Zs3CpU#k}3{gZH53%>?63Fxi%JAjMyX z6j0JO@+rW2?uh*JTIyHg2MqV-W-0`d#t#sC#%)ET8LxFdK~vA_>lK}%~AX3uQ$0a*g_=Z zV!rxHwYx1M@`mURXI{DqZa^6hBz!%WVj#^(%r4L?Zq^AlRH_2(_Dl7xZ@D6r3l6+; ziGkgL*%7xzEss;pfSEsqZAK5G2=0(0Nb&Ui53zUc8@^W&%5*42X812p#)(_Mvy3kB=HdxV54sB+oIz2y~^>I=7BSP3C8WX|_Y^b}Kgl~edx zjuMl%H5OrwHAlMqZ>1W3TqtdocdAwqs$Gzsx3D-q=kRQ0OKqtX1}Y9cL6Mp-eMjbokY&zIXJdN5&R%I*zGYPE>TLc(W z(#uWuDedaQ(U*$=w#NP;g?Mwv$NatmGjtTG|fC1oa$#?Y8hqzy`4r8E*3Z z6SfoZrDvf4CMpWQUjIGencNQOR66!Yt%D8iOpaJ_|32Xuu62SZ4z^}ZR zw|GqqJLU~}nwSvjm=_XEyd0)E&g5KxuENO=LCD^>w~zU6L8#XxkOP!vrgkN9i*22U zf}zQyTDs>^#2=`6`8d+po}h5b1ablL85vKpU9D#tASW3VV1c}WZk=V+{}pAr0-#J# zK@!%O9HLzvVMxm5i%k@_iv`m4GRsFM7-o}_jjIF)^^&~UCr>y#-f+n-;YK#G{TMF8 z(`9v8vmpt$@sOA#<_3}yQxWJgv z*^&SVP00T$N`SKel+*AQIS_=ZT0<48u~Xm^Yz;C)#uJJ`Cjuu9oOnBW)#csYu0Rxf z)&iUF7IDC@-trWwdZh7H`2)vP$Lc#~#Tf;fDJ5Cp5mUpPHU70^>m_Dvd`xx>Cz-=hJ=*PNjXlbRce z;F>6GeR%<`tkoZPF18M+x_0-%wxT(a%_xByfd%JGf3?q!de$m%;j3bMc5U5@qrQEu zFZV3a^M3oe=RdkDV73Pp*H5rRp?>W2QD9(c#~qNTR?&2EowZmS;0@%_18b zIbYnzh7jaf%AIqA-aj;YIbEh7l?m^Fe3b#%_%Jq?$;bS` zyoL%EVbj{~ZHl_A)_fz~G|AA@6mpAxs;h(urWoLLSNHfymhG{Xc6((*7blBV`)c1T ztSRYrj9F=GczOBUz8sHcNiVkoD3q1k=rX`>y8at`_K!qfh9I`xa9Ee&4%`NY%A#dX`cL>m+o2O3{JXy>+7KM0UZ=H`q0C$4Ec3qP3#6odRzh&#YHGs&7LyL4 z9?xlOThvb$a0@Kb#k}o{JG&ky^oAOUMfSnCHxru_AZLKh=ySBH*8sFx0UatV)jxF( zVObQr#J0xL{ka0VRxz;9r>Rnkw1XctK^=Q{rZv;@8ne*kQQtYesl;w0^D>oT;5U8B z%57X)FvkH>46i(d@o`~(6dlaVL9kJtE!Qj3|1pa+7D~~<_rvS-^_%Xxa0#}-%njMLG0`Z`E`bkuWwj?Pd!e=-r z3O(*XhF@j8Q$PA~(uxCZy91h7>?-aE8s+IJDD5+Y5K&QHJp_S{i5OiEP2Oy#)NQ(I z32C&%4`54>a|fy5S&P%yx-lV3n0Ha){(s@1~ikBwU*Miwu| zr^(qu*9qF>k4T9j`W{M#S0HcXWexfOj*`9j2=@RIrlh>S^c_Yz*}?2U6o&1QWi;$g z11d@@KGezcuGwDoHVq%WN>3SzXJjR9()Njy_6BIFkdj^U}=e(e}-XX zAu_$WOL~O)>hgL}av#-|g3VIbJ%`DBm9!pjwC30YmK~6NKLfX;kPs_@>dOo?*E%o@wvdNlRQSip5BlmB!1ZQW^#5wK%TU6{n=Ft5+JX3fS>=95GEoCQ8v~P z+*Ji}7|&7BbhPoLd#+hZ3Zc!Z;SBKSH&#J2PT_KhKv*$6RHjDS-!b88AuA#H5+_pj zxYGwzrG>YuSNzz;grjN~Mmj$3fbO9O6KY4#lGrKOrdJSqzol)f%8w&4mqS*R+aIgJ z`%$E?7*^OEJB+yTQ?9qV=!-C4DG3$y$Q^3*{3)2JenmKqO2ywdGUk*#(Q_tQSXxTn zzaBt#!brx8-vR5qikmCX8W`D%cQ8>u>q>(rtaluleQ)$dS~i8 zwp=6Tkk^qdz&`C*#ksnB;y`Rr?@^rqMe*e3ew??XlW0Ym9>MYJqzHa}jHE_L96Rh> zl9dO!-~QHG>M5oM0&8F+;Rl3wG4e_bgP+osCg15mouA|ynor}NDQ*fa6x{T|RF)@m zl}Q(u!c3rRL?%1mZ4ZeS4ku(7vr+R% zIf+)*(5%qR;PGSK>`%0=tc8r~00-J}e^5DTZ}RAg{u+nMAIAl)?B@6tw5nIS=DlLu4X`%yDg{BN^g5Yi zPtG_ma=2BA6AL;E9+#{HjI3Un0CKxy^-2Ow87IMITXyYc;!;z4o0I^gpiYp(b`sKb za#Jm;TkAlQ-{AUM^NJpE|;x&vvC4#^#q7pERovgQ46Yo;pun3p$G+h^kJ z;j4zrTLf~kH(kU2>Xgemz*$M5_Nt2$??ptP;jk#`fjt@v6F`M9A)^lqlpR_;LFKUZ z?sI;Cei?uc)oQ8>`8hb|dxuj`he^iV6_56LWA=d=zm}Zzo-#{*x%%D9Ek}1xTe9cN zCzzYeX=I8JqH{YiwZ}M6 z!+2&8ho8(6fa8fnfUWO$u+V#q&VBC)VT{AdmZ^yYFtI_{-A>&vQHMFv0C^BV6S$!^c6W-C%2tlX2VjWiu!1-^QR{ZWquF)2(}!a`CkS|Bbsh z4~P2y_r?{HJ)yFWqU>9etn*PJ#AFvTm1GOqw=q+Q2othon@SR5vhT(&k$uZHX2`zG zSjI3*_q*?X&iy-|?|rUwpX>atbN#OS_eW8e8N>ZAe_gd0dML~whFJK{{RWySTVQ$>7%R&+_py3ISm-lTFe1}CdG@@Ql>g5y!8 z<1#%2Wj9yCg$U(nONB!MT{a-GS_b)KC&SvYs=FOZZFLfhJEklQyrUDl#l*<&)6|>f ztT23`jaz-%dF0VjQEnta9PLZk6?9^!TP_W*PRY-;N}M+p0LnVyq7$P2BbG5TnR*L z_P5tg_n;`hc-NuTV6HdF6RQB`_7uQ)^qY^-YT?}X+|$zLV59tLMknTy-?suD@|X0~pEe8JEr==lHZC_VmydXbxn$Fq zFNL=B7#w(2`_{H24v(GgHO7qOjl_>NV0F!7xG|jKWsDbc`ba*}lN$oJh}OxY5V=DS zRTjXl2`+7L{t+!_y!!3T=(44X<0lN#*7CN>L{4rhU9It!HKZ-(}yHF=feX@~95GQoc z_pnne?`|zD1GW>xTugw?7BU~s3cE8x)g|~Xl#DkqarAn9H>+Lt7Xx~b(B?_93&mYA z9%ft+{5Z_FeAb(35lAvf`^}JhV`RLZcqL!fNIOwW)k>)A-oTI$icCtUbXBfUw?a@^)+v)n8Ct&FQ*->=-U z!C^Kbd26B!wD$_sTS=+ab;0qHiGjL;n8LNNI9l%rVxbL7YlSMnN3q9aIeuV0VljRIDz5g)h>UXatwZ)VwO45YSx81if~o@I@+Z z*6a$=$74?CJGlVgz|9)0LosiDT2utK{>QL$B1Oqf?KcAt*wq+%s!?Jp&v>cY092b~ zGg5zK35>5+3LYRmAJH|v8tvwA_tEh=Egr%f7<}N4Iv|R#6%;2!`0v-j2AFHZvk9SdMDH#N;~0BSs>*w zQF)0cg^4WmL?f~|optVumLdLwuJ9D_E~r=z(>Y!hQ(aM|Z(`o(f7IrDno}?zv2<`N6)|_q zA3b|)xIr0v`WgYusE#@Ztm&F?`aJz12KYH2BM&w1DbgrN+NEU76D=ozHl017L>|Ub z@6a;MVm1x)lTQk5f%h_vf}s)DJ7GRx|A85>5mki@v}}@KA37CrbUT%j0mmKL;0n5# zohXnQVPS6oJ2o6nV-H%<`lAq>9N1VWB6HXtOObQhdoqE~AKfO@ke=0styJqLW(?e@ z7ezGD)@XeMpuLY}eg~eVR}pMBV<_Oc;F|Bm=m=#W&)TmHkCuDQ#=0M@8grUSUu(*E z=vKpjagZ`?gw!9qLRS1TBoSxDbPCL92l|-?+HFP1P{YV7lzZ+#9rT3dMI+Sf2WJhN z;l5n=R#$iJHuSG`Xw7ouizIq4unvV`1GI;ox4%}He*27nH#V&kNnR%AwS$=UCKbK* z7BOkzldnSJsrd$XfRn3?09ud$pS$u_HR~RSo7p90%C?HS0=O14i~jEMd;vKXgFiiU z?{AyZPIgb=DYAg3$RSty#B0O=)^?m)nbp*)u--(AI}vCfO2K$!P74D z`PexBSBP`2>-FJkJ~9kFyBvnIz=`N6LNFk;qHxub-Z;gx(-j6TwbGK5z8_TOn$Oz? zU(F+)&N$9?c|x4_Q_KndvBiIp4#tZwfE^|l6*uz{R-K1j;o^d(i9c)gi=XVEJPZF3`#4wr2Jb&w%Z7cCVV8QhG0Ms{2q+ z^K0=;HSb1GvHSD1p3`$Y!5lasVw&Ed4_5h+pEi!(#$TF926JL?uK@M$Cdz{%PW$XH z8jkrR+BaGbKr+F{5Nu+U@X0XS_gkEmIi{Lt==*T;sCg^Liuoh1g3lK!?pD%AEv|-e zsuB!8!LI`yQG=={|Hw^p3Xe&Fs86^041B{U>^Zb>j)OQ6L-qS~so>qrtCtxrVmvd6 z@y9ki7LijbBwYf~@3X36*jwbPY+L0cek_Hz=KH z;AwfA(e94295qa+GV(skx04 zn+ddp=6o0VgsJT~vuF^#hzj&$F~Q`Ejh0i3nbqs3r<;2gYd2}X88o#+&<2d=@s>?R zTJgha7ff|pQ9;3)9m>IZ?WFpSdy?je3kL7@NncDW(YlnXuXYCl$;+ZS>A~NJXwjD#W^kcJ)o)_j+LE6(^t8nyILHI6dDAwqw~jG z+pSIbXBMDT4bLiz}rQq!o2PyQydVjy$ z{S8~!Bz`jtuzDQCVVE!vP>0tNH2>VUJo0eTL?NJi<_{HB5fjMvg zg+IaQE>NN)J6Y{hcL@F!=KPjgh;i9;;P5-znD;hU)uz=UK=LxnO7;?f)(DNJ<=qRG zk;&I%6;@BXWcj(S+iwpt-`(MX0G2rSl>yi{*& z;01Vd8(>QE(WM@8xB1+SWb>}JS^DM+Xiw<^2ejm0|89uI|4NMbTfoJ!1U%^QIDkag zdK~H#(9~>590nT+=$=#&Hah14x>_DfiN<6gBM_Wg#sJ2ZO|;I!O(WFDZ>foVSI!C& z-@XjclM`>Vhu;uJ;`x6rKOX(1rM18x7^cU6fQ`XC6tUZoUC*qU47cgA*{5$*Z60xN z1P~~u-K0ADkT|{GAh$W$rez5C#o-{K2@;dnKKYiG6^cqJo0>79m#|xp#k^3R)z=j% z$y4^qVcA7bg^SM{v~BXTK!&0o1zrA*{gDvroS5**tt{kf;JP3ZsPQ!RT?sJEuSNR8 zNM2hq#gBsQ+!P`Ok+VuSDO?%0pu6IrAKm$3XI@Z*gU6Ra)5_uL6K@c-6GpTremdqW#wk_n$xhJx*#(?{8Q> zRJ#L6dq9s8f|0bJN2q^{4}E^vL+bwLG`k(lKWE7wA=rVbq>9$?SM7+c-cextAMs-U zmCO^*Jhz}_XD);TSdm4)ae)km&oUt+sXxbI@hej0~h&xyla!;kbI%4>z zhLqf_p|)+xXy&R^>0Qf|*&O8p{&F_zeOd!z@xyF5P{R-T*lf_?m+7;2Ad(%%dBRWH zmG_u{X8Bri9y2kUjD1F#AUO^D(_H4glAH>!Uy9aMuv|p+25C2^o@k?@8JtFDHQ z70uk7)x8+bZ5BCI&nEqe6C}5eDVJ=6 zjI=tNiD9Fowz!dLNL_kaj8ifJ^Bg8W-wQ1Z4(#-NAX1}pAX$7mEOXuULxmS5^P*?O zTM4tiXm!iWHB5rxe9} zme4FjR>Mo2vL`pUZw^g+?Q|X7_G$$`MQ+875Kxg+?fHZY^6c$bF;3+%n&LA934OLZ z9l0MTO!RbIJlInmad!roY?LZ_IlnD1S&mx9??Bn9Tx6&DAMW5&!F~p3(urqeC9`Dj zf0YS-EdE|~-O|jYr>WyZS~n_q8t46-em;ZyRV6N*Jf1p&t}M`I zhZa=uZr5E)f9tKW)%_0GO@c{!2Z+s)>`2{A+)(@x;ZqeEC_XZt?O5|LIj|CPiEKd8 z4K~0pMhv&^1J-FOG(luPH zz2yw;TG?=A3D;KMY*Wt=+0x1D^sG zB3d|js(3~-j^#r98g9)+f8cLYdI_#Ew2ue zdx`a-om`Vy)g3l9pNXsDx6b244gvF1riCEfKc|x-;tbY29|v^J@_w zvm--3TDxMfY*#TBe1bSzm5++MG8Dp}wtX?E-V+-EU*OiYgP$TO zzM!%`^xtN`ON*PWFD1dFa=#d^nFNel0t=$qDlEfMoG_k%|DkYC$gzfEl07+>_0})q z6T zn>mQ@&eC>Rg09`TI+~4qqn*w^JfcsSMgNo+?;o8Rh)@2@jJ#5c21HHb5tL3nv-FG=r(T?x|`e{05f{0l?$E?{vaZ_KD>JOa1Yn@vyj zp-VoO@ZXi^Rq_9@@45?Og^7~G=ii?8=63V&mht!8n)TIC0sMM_s&+1ZxlQAKS@ODf2CN4ZIJ2-_Mkk_g zeRv`1!6GvQT)Q{4A|1)xgo=*(jpmEvl@EMA_gOYJ`xhralbTl&Vc2x6FZ%rc?6W+^ z73on6UvD3&{I5Xm7tPPv*=Td8lDBlxCqlc(U5c@%Q5E2Tkx{BXrDTM`nVNUW61&b` zx5Nm22G>2tT|Z={gvIK~zf|yi-}rp9^sK*ktB7px6~*Tm4wxGLB-Jd>6^}K4<#QMI z(_ZLJAub~^iiEcN@ZwT63*P~UPuz~D{XC*w-rLce*WK<~mf1tr^dE2h?=&$U8w|Di zGITw~o-r=$_4IVR2itT@gOe%4IwW{lPg;zlpsZl%mP<6bf)qfuSzwPU zlnkigSV;bhT6n2M zdT!*qry6WWC*pjv?Agp9ehcf+$#$FruYg3&X~j>@AoN?uiFu&8!E$SH2Gb@?$uyQ9O*+*gw9LJtX16SOYe=U;!@$qlrxGS@N8%ooL2Ok)UDxN~r zf1~aIZO%^nuEhq>y8msOc?#;EvU6|aybqvR;F~z~fp^BA)Zv)H|5B&&FWn93_sD-Q zK%&D%YlKT-BAP{R;IdqP>R=j}0z_{nslDF_*9(r>U?;fGd59Bbtz-JsT$FgbWL`Yy73>RUrz#Bl*LEG3MN!raye-xp z#|$>U7RoQeHdQyv%6+FKYZ(|?=Gp9|(0oT?K=*j;+zDpL4d@$i_1TS~1#X_;6v#xg z$SUf|{YNSfZ$fbKibvg|(HNW6FGbkk$Ms5+Z_U%MnBIQOrHbnjZwCMeO%UHIf_2YR zvCFz^!m{RctYtlBF7^9PcKD2q1->wb&77Uoe0e1AWbLn2M9jCw0)Kg%~Y zwIGLz@^cI~+3NG{r`t?MFVNZ4@vDkj>)|K6y}*(E7DtEb>riicFQ&2*TgfN*eIO42 zt#y!#&W+nDeY-WJ$&9Xn(5Z4-3Y~9q6-ezGa$D#HG^9|6gXD&7019V@2@ulm>3g%; zhi9+Bp;^1?1E~$(*vE9@szYtuWQgu({l4wC!OpG^0g?fu`{T}fa?1<>mzwp+t&|5; z9kRm@H;)86n$DhB&l#On%A7q{0;;VRHEqA}W1`)gs}1s<`ahOj|)EiSKOFp$NPT zc%1^odow?1fKjr>bn-3qu-)7hMKDA2-R+C9BP`sLeEOK@H-nVy9-<9+E3}T`Y|Vw( zncDn&{3dk1guBg2k@@FA{QAVTGwi$ZPaQ%-mG+8{#)Z60P9Ei6o*>odA_#t@ujCM- zb6S_WKB`d7*l$S(PNXcUikwy5mlBIl)J3;Lri@7so5=zJXA@u|nw(l?p3e})U7c(F z9gAg~&^$yrv@4`pma0=7Zc^iJ{VWzKuw;0gz;B?pdG{gV^&L`WZuF#ZxPjXB`pR^C zxRZgA>@U<5YC*oc4SL)Gx>`cd56dh>)<&a6 zV;U~@4REF2wwA<3{)#=U`iN-NhuJ(l9M`&X;569Bd#DSAz+`FlFy@7IOqs_xW>Q6% znRyYVrS|d0AUt6IQUjr;^@!@p|10i6^tEq$d*Z{G%V0OmPxjk{{1~bPo%sf)%p^Ub~$3%e98D5&9Abty1B-<*pMj+2vustEEsfZB7k7?W@5?Se0 z2mTA#xDl^c+@B+Gd|KRxN8WT5=wH;6wBpL`gsIU??N4$3I~j+KODRPS{>lXh9!82k zK{yrtkE;t32sj`#bVSZ`N9eGU{a_Pxk*ZVQXHA#2UjlFEr^fn<_MK+ihTJ*PTxAoD zWNYo`(+>fx%b)kqrky~a3cZOrJ%Sq`ow*n&yBz+$!C)yID@$01dd(m{r_e^I#fNlY z|D_bi(CgLcDT1~fTJF+*{x^Oo`+dZHJ8;?F&u^)(F@B~b?848LA9`h70DU*;WO|^6 zrJWizQtUTF&=DZ8MSMGcK(EW3H?|n_zOkqniLLfZ0%#Ft^s`NWa9NM<@N>mnsHmL^ zZA1`w#CQWX0a~T@9L{~FEGPPXk)r+md*FIFSzkm*mFz^Qz?&?$4ZOV38OYuo7WCHg z5%>N};(;}&a~1jlRDs+m)k5%r`QYN;CN9sVKYH<8=xTcWPBO)RwKmB~{`*jo>{93RielE90N3{OLJonsdp>?Pff{~9C@zw4p0T6 z>*f4u6?U~2ErzyRO*-(b^&@UVe}gC9glodCQaujm3ucTQ<-&jMe!zb(|5A+q+&4a{&PEN@-Ft;~cZ0g;pyeE^3>%**? zZ{q8fNj-aLCUOl06wKSLp|xxhb$5+TML05~_51R#U4|9e-O}|g$4g|Jc?3W89MPk) ztcUc#YqQenbjXi`8!p>iB{!e$75y9@sb|Hn51Oi0ZBPe>2A-p;s?&)T9Tma8Bl;A^ zP9$pJvWv{-VAmfr&DV7zBb{|42xi_OUI(S|tHQApSDmh3S@Z37s*&#KFA)uCx*jK0 z$~f|;?gRs{?ffV34lluFd>1%Hz2M!@9dFSyNB0n`H&clqV9pNxI-KKZ4k0DGIWW0S zHvv=!#%P*Ua#OYeC0iTOM^M6S?`wm<9r?yx|4TrNjf^F|rWNdAURX=s9xn6yeEZ~l zQjHL||9xM<#`-JH>b@N$54^ak)^4b10x^a3iU4h07myfrxaU}^?8RQ{YeSAuTKS3A zAS+X>p|Ho}(Wt!O+wHEXbuXQ<*KYxFX{EcV(Lx+`nCwj`=)zQ{C)n5#IpWj&%H%qv z&sm8*72vkyr#v7`17UA;!QfC%mg`Y^!eYBUT4mmNImF9@QmTodWnOP1Qj#SW?Fd`H z1$TL8BALS8rynT&B>aR2o;w;`U7AFc#EaBC^W0M`2aW^9EK)Qq#C71{w(*2_NqAK_ zAU?MN?E`6TN$BO@3?viaM9LUOtmjYAU#8_T$He2*6kbkT?(EMoGK(B)LNFe&@tJc) zq%|%~un|B=gvgiX6Dy?prX>{ZU5!1V`2kW$faMQceXvFa^l=gK0PCOd5i#ve6W|QE zzgjqImezgdft}^bcb{319qbhOhZ_;qHW-{c+`S=vbvvc+vJeMP|N#Yf~DmAzJqZyS5FB5fRq?0`0o|yMh>DBIVqrGcir+I+hIJfrz@pR9lV{ESc25Xj!ljH^Mr;5`M8k zz&;SgDB(MFEAO55^l)_g$Np?bFd9PIaU9~=~bn*@{AYm}^4yeJg>%nz} z0&Rm21_RE)grqT$U~(nAx1e7hT}$hk!_I+&KP0|_-U>i3PXHK9 zD~b=^55E}WI4tQ>Ib-yd?J6O!L0w&O6b2eJbEo>&q@1 zCw&8}2jATB@AXEL;~~bpFB7EHd*4sP>}aLU!c?WpKR@P~k6!4yFZw7v5Y_bSE3h@< zs<31!Y^=9}uq#}|!xfKWch_V!WDiaC>J(S!H&aa_V}*`q(3jF#~s+ zEdjX!D5WT^Bw}lUt>V`1Xx%z$TX`RC?bAMVV7KHFYiiqQ(2SCU^oI)pCVPCA`rx{b za9HIQM;jP=OBRs5L?WtC?bui53c1OgnilZ1b zzU!Ktw)VMDo5k%-X^PS;j4Kx`M7|to&oEsr^_#)wfVWtaH@Um&lx^tTn!^{F-m@_d zo4Yq@oz@(T(%-J|K7-`LVki+*&)HyEqkD1yVs~WF3a|*MHUzu-rjY&DM-zH(6e+4)`S|AkdKg3c)s8pk^B0~1Uz?V(YHQkjX538UeN6W~!yh6w zUEZ(8F)UOb#rUm_=AXj4tjG{Z<66#jz@9CeL}sDrts;c|XKgHEAc78|wlAxlFT4Cu z(EQ*eB-@AazQRLg9W@J;ddUs=efZ(VFgY?3Zvt*R!n7keBbS3;wy+lT|&f!o-7$u2Lxa9e%n5X43X>LBTN9#h$&-m!b2jnCLf<(!Y* zOti76Fg6++0sU%^TVOS5jV@Y$)_U0gKHGtK9z$p(G#gu4vk`=AeGkDSStWm#$`fu0 zWvNSHIwdEnVA~XaDs;4wn|K_|?B1tWvUi5in~*YkG@h#TrpO{dYH!8G7-W}V zWz17@gsR;$m*#jMbK^(wayK{@P(rh7F^1!>J|MkpJM}~zbx>q^ z=hd5PJxClDq{T|e95$HnYHe`vjN#t({d1slof9N;cY+CK>l1asFg0AYZMPv4rDCMw z#y&cuSqP>{blJSIFr8yzTlc71E`PGi#=cDd&J=xCp+w^J&4u$Z zq4tHHB@Ht?uOqLadqI;H3%SJsS{M%M4QWKhDfxr@qOhN!4>XNTnt7UPri#$c7ukUh z+^qCY#aJtT%-!7wtl4-EsEm&dHOk`In@&JD-e!(|RwShfe2*OI(P^$CAU3TihLhCb zdlGk7B~9o9yA_m?OWP{F`!@mMYu!4i9NW(6ZI)%xRK|H_QZC4N(y3x%39~ssh-rgz zr0ePk)|?72-V{3Y%gMFY*;PA82qm))l45n;g9Y-n%=0D9jo(brPzcW)SsbV>2h?Jt z15zt@YYueY{izm`qJ{FmIoY)lfCh~9z{_MFYBVmjcP%5|@0gHCz`8lb9}f)?A1*^Y z+Wt9#TTY}i2ZwbW{mMuq0THUx{*juJNi1yh{-UXS^i+x*Okr3NKk6xWyqEQ&>0wxW zP-ri&mYNp&tRGB7&ZQzUKJ`oU!b*#nY_zwptPvC=UbDSdvky@IU6uip?qAxexh77&lXF5P%e!*f6E7SgBed&r@-M77g zcyM*iGaUhk=C}DbReQ;Fs_w@axAV6S<8O0w9@%u;#ZdVyx+YY-M=B+2kdzO4h|%`j zv~d*;VQOSQan=R+9+>W$OT)Rz%|3YX>>U3xsK7fwiUZjP;)nj%;S3Y+yI<7!)*%yogY)!KO*Yx>?mbT}GI@==aP8FrCtrvlOF~pWE+`n(ZCRiNZcND|b%3Pt z^%gLzym4^i9uQ~zdeg5#$nSgcv352xZwHiM&R$cS%*{={*I0J&=&ZniX&2iexHa?) zhH?k?C@En2gH!52##5!^9f5>=aIGu}+ybg_X&*Q~+3xB5=*x<_L|wn(zTdiI<;vi@ zKyR3>&kyxmN5+38z+AqbjgEnnhfOb!_HES;oYKwbp#srm6pJt(9ORdo>-dQ zz3`iXFXNoRrUhGuJ2JJa6X4!H)-C9G_=64n6P!#Z{K-|?G7q#XTT&k+WefXjkxDeE z@wDgkD{BDyjwCeLzV11lZ6AqDKYkbVVUd0b?bnAZyIsm^Y~R)=ZQAYq zi^|9e>ndoPC!I-}b@rPX@8mn?I6gf0oxi49SJ-2at?b{I?)--X)&7kH=$Pp?1PGw+ z3}X9UGRoulpb*AQ_AIIko4x3OA}xY?ipN)q%sn1lE2wLAMR8CCa64RY3nZLFAC>Lz zOO7{GS?N~k|4|R5cy6&?{0G1EADYC?{?8*x{wt6FzawjN^uL!+0W%`JVm~6LBrRSv zuNitGyH0u>fLfI$bk2Rfb#(GKLkV0Kj@7WNh6-Y>0@@t4oFV9RKCy`nwKh(7k+2rK z$L|@vY_6;9J=AWJo$U6wo~4734eS_))UA%1=>;$U3z=kT0pVb;cR8tT*vViFl>>$Z z?0C{;e>2oTXwr$z@1ybT&+N`}vE;*=C~k#=YXLLUQ>PDP^BX__e zfJOH=Lo}A~6@Jk^c41d;BjasIH#X6vsdcT2c51cy%#mkyVBmLt@pI?U-=O46%}MzPPd^6y^m741hE;-&PdQ{us{;!Ir_rP%|EMbX z@Rulf`@`A6qlr}hAgAe+@;b66KEaMcilBXP`daTNr)GYtxLedxT>C0u91Nj~<7UC_ zf)|_L7Ia)lR9#V7gZPGf95$1|L@C&1a?;rZdsB_<@nj<_wL?U>2)CwvMX1dRK*2;H z*Pa~%0*svy^EQBe0R5p)DOIvp(Bmd7%>tRv2T0?fL(B&sspSLdgircteiWTt6z4(u#r4 zIy4?rR`W`gn-S*Y&Ge1#ppjMUJWu2>uiXd__n2Dz$IY6o{|$Byt0Yd>X+mr!z`O~R z&RutKC38%ISeb{^V$k`^P_O&a6C(7~6`WG4xarf(p}d~%g+kL3^^wqmLfTpBZ%jex zY`Pj?i#*$ft`K=V-&H;RwZV2d5W59vG6S2I^|EJH19wq)=(>%U*Zs=af;t~TMTtzp zd-dx!iOVCF1%gSxGV8V0agcTs5Y8oPvr+k@6zhqN8m(VH8`9_pC85qRAL0$tAOMId zUhew72VFA*bWjY*1z>5)++D{Q)g4CwmuBWCI&azBCis0IJ@6F*KaZ$icwS;LSJtjI zz->I!Jd2#D{{?e%=6a9o?G3S1lE$vd?x@j)>=*R5w>_^&^Kk1ni&ODMHJGaIZjb4^ z=RMDl3E$5d{LMhDA6LJ3^GJB7f2GP3zVN-70RVPRx@>BCoX&a0-rG$X-`?LxW8Y{d z*^e?p`=PC~c0e#4dDFRY&F$Ar--d%v*ZKPz3>r3j&fG^Cz(^<1_sV|pZM{Q*r&KiS z#VR0#WMLR_AarQ6yGa1>mRCjqIG+wWGTo;!%^|u!0vbHE9rW__`~WrQOzIyiBzP}p z^%sH}@>f45)?k8)T)7I`H{2_fxrP{%9R7$FLH(uYL;nCdfB+9W>ZY5Pf{s?ON4GtW zj@ABVFq5Pd0N;w{1b9DexM}|0v`eb%zw9A{1eG7FN2un|{~Zyp%R_v-%zc0;u^<{y zpD7~N-=d{q2K98_!v?Y1R)KuR<>Z#8Q$e@_n5!5)8^# z-8;LX)CXsD{SIO5@1_C0aZ}t8PK-K`+pgdM8{VYrqc(wXbPNkjqS>BRtN7K|_ol}v z<=wgONU8*2M34}nex&0l=ipUUM$NN39fmr|#r_XgR~-RVk6CPVOOC3Do4(fkYho%0 z(c*YowgT8i@YO*BLh=+1vV9SYYIA#GrPdSW8SwCP(Dy$;UGji^vAlh#~J2fb5F%&;r#R%w~E)dkEm6v@mSHE>A)ol!})& zavBJhxz6{kN)IN7Bl{J}k_hUXX*8Le9+!#n@gVHdSr@NT?=*#U>5aABr+kcEY@%P{F#*U>vi*Ee3q%m+Katg$ ztEitm_*FCG^8iN^|KPsaR5O(5{X*wLQwgdA!9UfQR-dOL@#ULZ$n8_qB3~syWj9hgdPAUF!78m9 zk`Y&4R^|vEZE z<+mMJz?j_!&*rTlu3|W~&igDNU#8JP)k2vesQ87gXm7LCb?ET&i%ND7hpTm_LUzP8 z`H5#LR0GLE$wG7+WD;d~&zSUL#Lg^ZXbX6;@Qa;r`myJ6>-|MK0lfXV@6r};j{ZoA zh>$!T1T7CPY6ts^9Dpb=J{83F5tR3SA2f^ZMNLAwj?$yy$_Dd>+4;?%`bC0)t*30X zP_`BBLCWhGWKZDQ;-wk`32*i-$)FSTQ9=eCTJdcbJY5Zj+L6iuzW>eqK+cc!i=kFF zen-dOQfnWjP>LXnvWEx6xw>#7mh{X-=NEO-OX9rE*BzX|Y7OAY!Yqh5QO>s6WEN0& zUm_MJLF1TgNHms*r5zi~zQFQT<4xM|Ei1JR%G$S(sk!DiJ8@nk$=G)lYejqWq~?M> zpLqz!B-y4tFw-jPxn2tXlJX(92C>o+kzrZr*rICx=wf<8GUTwNb@>tgnUOt)TwOC> zAoh_VLFt^kUu9eI*B(8X1QJKK<_Lss;fN~g64|0lWgRRc*XYa~c~`t5=dEIVn?>mm zwHP4da#lyRR6PP`?+NWm*ywwk4#y`tL=@CX1-k5xP^@k~CR6Yy(~DJsE8!c@#0F*~ ziO@BH+iq*p{upKO*)c{%wBT2o38uq9e-nyCRmPuKCcPfNAo2@1Himd6_9Wj;b{4pj z(w@>f5`4(Jc8{$D^Gvn-eBGt-GC%eZg=oO{7!|HgJi7^C#1bkNz^aBG>Bx}7=&XnH zLJ*$dCGS&nHF9zELwzlRfHogEQUlM)7XaUDmEKZ~z=ioNT+!>-R;rx`npA&b|$KuYGm0Cw7;K zkH@rlDYxcmv9JvDTY<;uqwFZC9X;B93zZK)P9Iw457xNTcA?~9fgJT!K40)EAUX+zXDyn0$a}kZeF9iCYod3Wq0?jKTFhCkieve zWTyhEHrc1Z-r%Hty0ya0?pxu+4;>guX!;0HGvXEn z3uvqfz<4`gL4GKOI-2{s4o7=NNAX!!@3^dAyy3%SbElD40nqJn!teZM_*>x!BJ6Nl z(4zohVhfr+1Ym#8;~O5ds%W}*D@p_ARE>Cssr=PRiqd-J5d={bUhb>8GaD4-2d#ti z03wYls;Uru@}DmVio2C>F?VM;_aB#{Gt;;8;W+q!U}I)? z`@O46x0gM*!aw?~Xv?BiKGEJ>l<()r{1RSutKFqYdl+*GF`tj}YX^~lqqEXuKl7jd zBVCrRWxBCVn}RC#9o3M4=!FzlLIJG_h)=Mr+W!Gdpl|@$3zfkZbbue*E3!LUsw1L0 z(~C6-pq7{q)I@{s(COfywwkH7Y@qKls_LR+C4THAXJCTtW3RW*565#wWHgXkUK1OzbC3pH)6hWR0d+JK29}JC@j&IDR zuKlvaR=mlp!!4(#S4BzC<_h)@?_EiDT&dpqw)z*MnOn@ZdjRgeYJJE1Z-)m!c*MMR zN9|owqcbB&arhADpus(Bzi2yYI0x+!T9}NR4_E1$V;sFgAxBb+3>^Cl%0yESfpX^G z_67U~=D;Zu>u#6urW!Io&Mwsob_d!8Rm+_{sAi5Ee>0{RFZyQSi^FOL&p3A=--YBcdfDlM&9%$6!nu99%P z2!0$LbiRja;fb%)RzVQ{r2uY<5aVtH!B@je8)b5AKFw`MK*ipe*>E@lZ)L43t}r5WK}q{ zjV3kQd>X^@4n7s%n>GmXyjYnhE7pqcLa@TIs8cX?vB=JPS?u#xrwe+CO=-MqvKyEf zXbnr7_qkq?==?AmDqL|k{v&!og4!mg!8Tu*ZWUdvZ6wgdmg#3d>Jou)+U5!t!e0^9 z9+vni$FV5IN%eW7{YV^>AO1CujV0e~q)Jo7`lr{%Mc$h;5i&GC>k@SgY~wxPI&zgi z2S_YqYySj-mfmgV++JiqoDU76C+^PD9g^ zfBXe%uKX|JBK(Km6lcV@DQpb=xd{B@-zUoN&OqpEGziN}pf5oc5@^5#=nPOc>IwYs zW1@j8>ml|~cndraK4MBedJioiU??B^=o^2O1L*hvyUXsMU3dR~`1ilfm4qiu zEK6?Y(VMx+XLsTn6~qC(QWBoXLivImkO3xe%YjvOI%$cB#UK=&5eP>!gnkJ)RR8DP z$-VwFKvcROK4umZr2q=pdRwcm<`3qE%$gf*EoH$%@=0R+kX31n#f9BYefdS8YMmIn37H`rT@P zJxD#ln>*Hu-aCXE{K1b0vj74idwkamW_s4s55Jzzf_`+dbLCt5+mY6GBw5b($`8CN zFjsLH3IdFcQO_=$GV4CvTf#^qydm*HQ(?Bj7VYVJCOtmq9j9TRL~PdQn(Jxr!L9>b zRI#C{V@fIJ*{6!sV71a%C$0o&5|$GtfZv{S!n{370jz+o!Ue>C6#Q1)%1Uzj_H^FowNUOkj zRpp6c!81s=cWzgYOyZtI&pu@s@JpG{US9&rpuc^Yyhir^Xk-6TWt~!aj4S3F8B1q< zPsem({ya_9`hGJ;0z97yTBq6r>??}mh_PS4Jwo15{s!#ryj^Nr9Y|!=+{2M~)PJRG zDNP5p^*X#sw)B%x#H~4%q358S5Mp4E_FdUxQ2+Ku!HE>LZcR>>jx%`susTT&R&k*V zNin9Oc#MDg{6zcXJNk)@0G=la@WaqJ5U3GUOl2Ui1^l?k5%|8{{+4q816@g`GU2*) zXytDP7PzDqr@$lu`Wb-pQCq1@gQ7w`iF34f^Cz)Zit-ISq=WPjc+~ZeBJGc zBSth0s0682Ai^gd0Q4FZL2TCmKCj|YS2SpMm%1s#_hGg4!zH8vM1H>uEJ`srY{ZHvsOHM>%0h`Z^BL-wz~oL~$>|4^I^ zYXjQ6M7YFE@bsNz{hARW*)&|^%C8$U{^ElZ&N95IYUbwU202h(Q~o@Cr3S=v3Pjjv zxAcELTqZ{u+Yhi^ozm%xMpmyR-_RjVQ-x%-zLvFIwoUK-|j~eDG>| zHRjyUec?lInazPpx7^fH3!}@0L4)1eB~s>QNQ({RPY;sP;C{|&upkWba*l>xikqoW znEkQ_4_;i>64isJxk9T}yy2Bqm~=sNDzf6qp%2O; zw|RfU=8`18de5#(19L=y5vv~Td)xOu{|GQ;(ONS0LRrm_!lz_Dwqn}Q`d(?m+dF5R zo@r*16-h<3PnB4v)e;<%8$9K`rEATe6#v9)#o&6_=z~=wU$fX!nMP^d?*_|fBevzy z0D$RSUn$nHBkYv;ng=@Q4_*T>C4a!h?%B-H9RTW*SM}!m*tsaj{iLG{^DzT4+D%`f zTdlgt49D$gd4WJ6zn15DC&ZGZz7PH?s_CtuS`x?t(jQm?RCyPzyX7)}1`#JJ#_8%z;wgJEP$B zoDt&bjB!=4Et>5*eEv7&*w0d^yHuBc*5~@S{IW~0c!lmVO3|lYeJ6DI8Xuwp!vC}| zZ_44fcyCoc=Ds#dI0-jG^;bLj@X<`MXv4yt6iMqs)OO%_fgg+RfBN>n55y=&5+pfZ1x1k~XcFw5?ipj+DH zjW>8U*VxCz7+kG~(mdq$rOx9Tc2_bZKrX@##n*4S3`cjsf&kO~> zq-tJ|*i))7ie|U)P+|>gd<@wdfxPua!1d@V*a)?;Q$vc(H)^8N!EY}K9HL5(-mZ7A zQ^2oiAXJd8ux)wpFd;$dTQzF^_K??%F-HlwCMY(;^RV-MU$Z9Vwl>LZu-S1dd1><# zwwIrINZ?TsbI9mvP+>sNDt=#!)iT&47#uUNL|wx ztUM&ny@rZ*{kAshWV}F*kJKK=@RQ2M#uAm>qDGtUddFrR?7ti}2-%Q@| zZi)sm!zU+DjJ?6ee3N^J4!7Aq!1!_xSlD(?$@LwH?v=+;o>v&PD7CoxMg|`<#n7W>*o zY9GD5Omr>Pji0E}N;2L7QEi&|z7z)sqUzQ<&6+~k!#>!chc>po_kILNpXp7BG?A!z z#4|!cOP#q?h>((=vZj-Qep@5Q*x1hH>#76X*QIZ)?Dy>N}(xXIz84->k*_`WiBcXuYP$7O~36qgfkuF<>L;`9ZN`^Y=x* zR#wb-f z!T-BbbkCOk+nR+`mh&P4$^km_9O)w*N3D_dfp@-4?TK*sG3g$ctI6aK^A4eW)$MIf z)WTa$y6|~h=qVXYXTWYS>{7HdJap)KVfT~HK4VR@ubY52p;4Y#Q1evYaT$NQ=IbqO z>D$`S$KnttFILFMiy{tuQx)iD+!owJ+zwmZR@OO`bxtI;?U@7nYj9wki;+I;A z#8~fV%sz_OvbADuUhiCozhbS9xUHuE{Z2tZGk%UTnZyS8{w^)_A&Pr33s^LYmo$r z708%TM(Z#OB@P};e#c!P6PkL)MDp$Z9)DwjnRBotMBgEnzLhaZ0TmF|k4dic0A%zp zG7&5<)I<21Vz@c0xF0q30y|~a$Wj>whuSeRa^+ZpQt%~}xhPt68S3XI?O2%au0O(W zWDM;DhVj(LWYa@j=jxgx6c#dGHo;lFs3}@T9cnlZwY2582D2$clUz8n*%CMA!$hnd z2HZ6*ok}M^lv+(~cPa+54}0l10rcXHTAzlpf+KLq-w?HJNU+md5&$O$;4HJkBE(q2 zD=PbrvABHU17M%oJH(xhxcQx_hvYzZoD-|Ur}p_i#6Sjk4`|F4Dp`ummh62h`PM zP|z;)K+CRg1zv{{_piq#TvsXiKRBayNo06%9)@GeU5S!^zA(SwnrV>Ojv#}IP4=dH zpY^6JY_>(?;-$%OF*7|yOAM-;mxb-4Rg@A%hj-0oH%FXnP#!r<@{ zo&;(SS7MB4=IKE(&f$iuuO!K!&|RjWbrl*$Q4=EX{Tjy^5dTO7U9#tklJ^MCQw9D( zeFSujXkivC*dW6fSB0W!++<$uZj?1HqW(s%;U13RDu#gH?ttM%7RUjtw*xho!hz?w$-PS}La|@nDN3kT(G(+fKK-jkrN9_l>S)*i9M}41G4QiLEN4)k+ zId=(<$)g$EsuxsmP_P%U5{7Y4?5TJ36C!OSg>Hc>S^f1&FByC^`Xs5B|~YJr|t-0B~;Eo%9bXP=hl=Z`b`I^wTxT~Fmz6n#JU1jdQvHyb!r zVCHk_{&az{*wUco+ZF-cnA;($`jX^H3J1xtXavqT=rL;)slG4n;;p8aLyvU=zfGfc z=(&{QHn+Xz!*7aCM9jssa$ME$T2}EceyOy3s9dPaNI^28#{wbGZhC3xR?DZzn`3<8 z)o80Ytq}2@A{TmB`LGSp9?p@0LbHJMVZwo1QEGR2h=D^RCP=?9=T7|We#J54!(WtB z)QnwkXWN#^%XtQd05hsc{=h}%d0BXLT{$(Cdf00>&1zI(ZW}SaNomxWW+z2)_<4c5 zV9OXG@*QJIw{E8%GX92(lGY)Wgp&dt$)V}&3)y$2xjtjP#E4pRjpXV&+;-NnRG{k< zV8?zp=kiZXNtH78CwK6V+g^Z@H>GN@zy49euAKlX#^hPNDHms^$p?^H8U>cj?Il|V z3iSxypkWrpP~fk9-}^D})cyx0%QvN-q$;60=#Ku_21tff<==3GE(6#xz-D08b0(+c~e^BjdXl9ULaW4;sf$sM?L#rq8-m+^Fw79tC`|Z2+N1vP`3}JmqV)Kh5i6?zy>e~6P zdK~Uc#?xQr|~%qlgIdss+?S)}4N;=!*? zdYz)elIDh76~gXck*92jezmT?lJ#Qo%nECwS8H|x|App4F`&s4soEWy2V0<8%MXOg zN~>otVf_O>EAUraeYO!7K^oH_6lYpGIpg{*r&@mhB1PrWat~qJEjTHJ8A#J5g;rxY zUEOlqPropfUh6;Q#iQDbE3-_#lWJ=7WY*4Ve~8Ew`v#jO){Ax1R*k2XVf}ug42#h)wC~L14?hj`B?YrSuxvk8_ zATF#Eqn>*}uQ}meaq?0c0JC*onXRJD>^wBn}nOqOBZT0JF{#-GBlB4zZ;9i2Hhye1V@H*sme zaWhaGV-Ix0g(7mLf}fPrPCc7snr{VNk}h4d>wB@vWNcN-cE*LkA7r(pGuf?+uJ!PR z15=W1RaQm3B3&b%=NY+ux_yREUXe`vfs-U8uwOIjc-RjqW%N%~R0*X0BiWt71+CWX zzEUcGAYv3MzSfdTeP5sHRgR@|GER4%uyNorOP_*y^)SXWRku+@XnG%CI%WDV!1|(9 zeMK?{NEq(epXpXCp?k=C1<~Xf%4f2Of>*%wX>|eE!c(>|A@&&1XYh%pK+4sm84)=V zkm;8kankMyQIV$cB?>?SFq4FS<|8$12bm-6G3C@nEi737#!@6!=GN5|T1VGt$nrUNtRqH7Dcj>_x8DnNE%YfsK=DCphcuH8l7}X9^S|FXy2o*#{&hIoSa5ch zj_#i%CK#h^Zsx%48v$Twk|5Q}PS`X1Zf0o{){Q)Crv&wSRN z4p>6!hY zoD?^&#=&avN+BeUV1t7k9|%7VIBK_42!V7VE_Q5p*@rDxPb!NI zTC}Wats}n;VWfr`8=Y;_jf~7;=uO{`d|gC7Si*o*G)c6bhS|nE-`;;j=_xzY+hj2T zHe<*HdH(Mm!wOY3!v9lks@=guBBt!)mSah6 zN9dCvyFKns&d~45T8PjKjEZ2}V7qC7at1Ois(3(qr$$jf-4dvty+{eJU-DXkZSy%x zGO6}(Jk8)w4dqxUSo8boKA0=wOMwl^7l<9WWIz>u;BuSv?X8}-_Ml+XUl2UDQwBKV zXX5$u%K#_V&RE}eR9eDINV1-Hn78URnS*?N>YeTFG2OpROTP8;ta|G!S&`*`zCm*Fu-rKx#SbLBUAyu`rFpRoIJq-wH!!cncHVHAlC5g^ZLJs!G~e zk<=;2w4ixOkvcC`8nXy1RkqAktPm-+0^yI`g%>7^1>_mnq+PoCCCL^0d^nw$ut|1v zCu;yHZiJD8D@#`D_$%yE-rVmWCB0e$q34dH#tyufE5w+nO(liJxicKJ10T1ZWfg9s zfCxu;Rd<0iBdah?iGEP&3FS&tJUA%_oTiE!zYS9 z1?ZpL;arJ3bo!Eirap_TDWn76LVo6{n!m zi>V$6t#5dF8j3$SF3ye@+?K|#I&Mdo zhrnKRbNPPsrMW6)3pF50(z;Ve-=vf5M$uJ*I`BEJCjr8BdsORzQONLW1Ztj&R<^Eb ze#CEer&(pK_=Vcq#LR&Q%0boXtK2^U@%auSTzMXP8zJpJKp3?|Q$C;s4yj<~w0P;o z;lgBu29PNM6x?nj{T}{F{C73y)oc%ntd}@)$eBnuk!i1^Ljm?8uDkXLTCJ z8>Rer8^*go@yaCvS=Ic~z!8VRI<-q6Xx7j}6s)eD29PS1L>CT!lzPw_lbQh1=p6=e z4Xd%Aw0OSG5JBi2=e^}O7*UJCRc4wG+v%`<*|TGMhi@TW@HxyH!|7Mgy9o?>VIo)u zXxm0^yvd=pTs`7vy0C{A6SA<`ztcxY})Ale7!HX?YWpLHVJU*e)# zcKXa_BN2%(7fWt2p=pnv1=zu%D{a#q1-<*9I_KZJKW+fbKa?r} z4sqlj+qkG1YB?0*er)@gQOpq|5^PcPd}SQO6Co8==&MEl(jA5g*U-@D$(w_~e}S zc=i6OEbV$^YFzfIKNI|2vx|b<>cFVzGmLrU)4yF|9HznO2DHWD)dmmu8X1M2`5cv; z^g4G1qu;^*s?iYbA8 zU1OPab{|;^N|IhvS%A< zlvpBK=a;&@1hHlZY<+UF?7n~!7{8Aq300$e<%`d)JCvzo&9y_*;WdY-9IUBy(-+4? z-eq#rqECHFxzb>C-&3_jcyzI6?-km`+Z^@lEt{GK;bczMWu`6@kk`U0hqjWiPxCt4 zRV43IByTi#4lRa7!Z+W4SA2nSpMTyPcbjtYormv*v5YD$a}u@WyuR5h`*Z6pT7izs zOQbiLQAmi>`@6$Rieq}QQaO?F~A7f^eikrxr9RXoYn6(F*=e*YfzeKMO z5~0rsdwe)E_PSK9pU*AZ{*q;n?cBe~V5pk$h3o{j{n1d^SC6>wmg>_*QQuT=`1P2) zH(iMGelhR}(maMjm1Xvk(E1T>O-3oObO>aaa7OF#rRlMn_T5ku@!WS2V zLCuB}e6e~J`@s#{V@GsMmDGOlyj8(zsF2mg?NUw1hE|0O-8nt}t+d|00g%A98%c)_ z-#EY+ZPS-3p&Wb+xBUQEk< zfC%%#9)MR4dQ#=wO9iIv?TZK+lsY>PWk*$BT#~ zxB*@LpG!_WsLD$!QWO((bvDWNxz z#$C;W_MnO9Uf&>?a}IcK&M3Yk2||MStFT)!6N%mjtl2+JAjejLNW6{j<8V>>+Zn?~ zGIwJ3!H%5sEzb$NR#5(G2l899`gW_5_wB)wfrG)xMs++j*iNc;%mVAzD*n7@n2DJ;Sf+>|3}(vb)hA+`}m=YkJ#$2j7i^V z_3QM)8`#|2I2geB3<RjL`5=W0pT?wXMXX4InXie=-vz&>)X>ZO>I4-PG#!o5yw-l8r%@Hlklep@#oIb zq5}nf;K>2yr)d6U%t*_;SUN7iRkeNk`^vkMQ@F>wGsJ&1Xypc0*S+RC_SK{o)Sw1d znG-;*)4m6oD}l|p9JZP*+ZKNaEqr0uz~+f(g&u$H9IblsL%7K<@hrcZYeq+jXHth1 zDqH$EN5m@3#Z}HeIl_4|f?GN~?2T1(y{dV4c{@#>es2bKK*hz182mQ6n1Fc=DGnnn zfA{jj_jpm~n56icjB}h=qtHR;>1#A6z$UFh6f{2VRK30*co@pi=Gg3) zH%QHZPO&{R?7dunL~3|oBU!FbC$$)D$9@!Qwx&yY%fz@6Ld%cD&cB29y?lDtCrRc< z5+`U!ot+^9&COVyj2EvzcpFXCeY6nGg<;S;Jm@1WbM+&_R^H{DM<3k?>N)V-S76nP z9#xGYT|@7bFK>DvxNMbW;ROoQZD^Cji%->#OEHX&+%9%*D=fA;&l;6_hm8d;>g~k< z6GP9Ju$Sd;qj8i>uNLl0S@}k>M6GQh$tirdJd*^0tjg|fJLX@!VR0DlZM-dB={QO! zjefG5IK7QnUyJRNt=&I+WGUDDYVQNJYlk@_o&r0($_cD^Y9;e<#x0wX`Iz#xD15s* zLf6B)Y8~qTB}TtBQzLIz+hg-0K$Sd{I@_6^VGxmmdVv4381yFbRg=$In)gQ_R#sc? z_p$2u&C0>&`r3B~f2^)%5w3U4=j)2wpTYYW?m1wRhCN~EFqpJ*J969{8UQ;|&h9Zf zjIIgxxix&4@03BI=0)|wv!4qI55(T6YIq~6Tl%M_jSWm`V|-b{-yP3UVFRIpXA$Xr z<)NSK3Y_kIt`K_;FZ~EwTUP}r*jOKRx*+2yUNq8Uh3L5CPLSOMRV=?acTRF2X>?2t zJXMe>UUVnr-nwj~C`hzDXUJVC4E4cKQ-CKczwBO%?vFx{-7+2$Ai;t20{ht%>pHrx z1ir_n3p1ay_)oFy^omd=1qL9(wd+ukqzn5TCv~-3q8fjLGe1-eYjiXu+d>| z%etx#y#ih3lQjln7hp<77UxS1GtNBl@(|#9s2&aaMl4bv{_E8*+OcYDAbc%R=#t;+ z?$YTUK%S3{AvFp==pi#TeOBH`s8XxQ)!cx(3_d zUfp&hm8W-A(byXf(d}(GjVX)=mkFO{5)nZtQY(QMHgz7g_!Gv^fd1~A7iQKMGH#v9 z%5P(t(4c?gS2fyA0U{ikCgI{y7xwrgU2fyav*E;xE!_M4g-+U!*;B3dO7lz^D^Y*70u`E}x5Jr?4QW0|ehM5t!prsniQA z+!GU}S~_LyI%nOjF|m^?vu&=UJXy}Jpqt!z$+`#k2eN%Jm=$JCaB_nUWGAN%n#>9! zW2*=YxO2$hYQzEoh{#D!L*D(7<>>GE4%@*07ndmj+NHl|K!3}zGV=2$`YqZWT}x63 z0<+Im>xDllPA6rIioGZoduY5K5an8KhmPH+ zDSgxJ{-y#=@CTJZLs|&q7Fldgij1g?E3ss#mfMXRzOIOnKEI7+rsJ!3Lx`POl;_OE zI6WLPC&4_eniUuHoKEdkRId84mO%PBt2LORP591yMP?6?gYB_#u{%=RuI}Kt&w0mn>1y}yqjkIyG6W%N7 zCp9#cj{InPJ<^HrB)WE*P*`obTl4M#9P1}VDE_R8d z=y;c-f;XRVkC1l?8A<0{8z3+D9ZV05M5`}d56N}0cCnUJcjnVH^oz?xyvD@-k0{^z zU8&iPSL&yl4~m>>s$jdu-h;|6mi6c(23`cY>kE;%n|45)OOhz`o_JaUb#T^s&1HsP z=l#+7rJtdlXPYkyX3WQdqB-fMRDT`YrWs`P%WKgpGS@d#d`7-l5gQ4vm-pcYa3BTfT?^gH67(pBE0}g%&YTg_ zSYgdVbcpD=D3ZnUrmXg>Q~j*dkUBR3N*tx!-| z2nW{pBq)WsLX!f^7R3mMZLkSoKc%I!8mdt(`P4(q_`nS|jNO9T{XZU+%;FOs2Gz+M zY)f*WW$|%&+z0gB7X=0#@d)(*;E9ucf8%c@(CY$sJVJJgBb4Qf{k9hKhw-5Qc#(D65LnJv!t%Ut<*&Q?^{oEQ zYX7wY|7;3=y;uL+D}7Qd{;FW=&SdyBvP!2efI`j@r%JFG=9VGpu} zu!kY~q%cnHt?(X&IbWqQS)Gb4Khwdvzk_h8w`@}t^rN)SOL zLFjO+(-D026KG%mzFo^lq~ME}ec|JlR-mCppc6NB4RnN8Zm`waEgy~4hZ|}mvJHlR zmY>b_XW26~V|EWG^Pfjdv~-|p9t)uM=~5s|AHtH=VyL(S8oT%Epj#8ToqX7zUXqAa zbwKZfL4*QoS;VT?~i_Ddu^==NB=9SUHIr z12y%da=#lWqy7W>uQvqhi;SgV%Zp^|>H0t1dh+i_i=R;CLVf0QXRI?2Yk}+8&%l(R zLSg#9yYb%*6?b+jnR%@d_nTVPPgqwNgh6MRu7jS_-`)1Z-;Y*)4Xpv42>|WUYB*lj z9oCoIs7b=e-|(COWO74hacz}`TZa64O23}cuXkzm*SoZ7ntxfmU;EO(-9mruOMiBD zet8dn_40ms55Ob+<>UP&WccNYgKm)jz!L|a{|Y91 zaOTAJj6Ia|i++6tlFJa*E+(1tT1jD-3V)i;y#T1>-0+1npNPO|+s{A}a@51Eyi z17rx}8EEPrkf&+U*#HJGK_4cJ^iFFnS;Hb0=TaP%f!RBKb#vp^~qaIqV_)yk3jILB(3_mguCm+7p6Q|v^zl3kZYpH~d%{P?G{2b!+Q5=n`Z# z27ngFjbkYvVhh9 zt0(Kw0hqOB3ZaD&5q8Ec0IQ~BShAMH4~TlcdIe6wH@SL+y9f;MN23vuy-Fi)29n=b z-8b04-?qp2IbY*WnYmVN7kvXeXH92snOZJ`>(>l4jEhS1bM>e)U+iY|Pi}4E zo;1VMI&Gcao=N!>q?HRDgJfE^s~xAoEciO|xuTYZ%HMez(G|@ovXtzytS2is_FGgw zE_wH8v{RsSb>}mKhAY9}=k^|K;^Md$`~2h#T&F4Cq(dbZ)^50f84-Hj&fQYg5(+(} za?6ukGriZVs!jCTAzNxN+`>8&xiCF>PWsRwFAyQdktH6+Q=Ua?+%{>_N6m-eR|HdS z*R6X1!5+Ah6d9sJk-Z{%*gSjqUaq5?<9d8ovqzy`aaXCTXKF7jM(*>uLiWQ)shL4i zUlz>o9g@Uca50{(*KeLI@9IKNwA2zfRFLIr-#U;@!dIK6dl3tM%sE#qprY^1tyY@! zSL2!XvMhV`==oK@2wBOb8p`pq^enkoXSyGjw)i~AyYc>-A-`|2$@|dyhA%hFvd3@B zq_xJ!MK36Ye2|yi%DS1T;AI_GX6c4D1Dt=?wrB`T8x%#beRj4zC;R<)h^xCQn@Zd5 zq9?y$WYt|NS|6rJdF%e|m(%fL>yXSxTBGjO&amEyFD1-#pyA+<$&tt2!iq*AgGsp> zcWf+Aw63NZ`zpJ-}BcBpPc-?PCD(aWvdOBd9(E7vQEqw zWoUT1)z={c-{6n8I~Z*bEWpF)r=Oe?5^t41U-QO2{diKs=vtT3WAhKf#!{D?pNDAN zcP`Sun35*9`y?1Uk1Q;PtD$LXFArzT$xkZ5M;2Z^M;lr^$t81EDeXk%t!@vBNZ!8g zwQW{Uqb&G!Gd35rtg_-2b?oxN61Do{?+?DKAwWO!(cvz~> zU7GLbE5~!Y41fSf&gfcN#*MMAxd$_~%ICa%x%InQ&8zBWKXbNC*^Xa)_XnDqe@`3j z@AHDtlS1GQL%TV=P)3-x?L{PV8!^t zIbzdGD-lM5Tu-p~n1SWndPEbT`{kca44%;0$zJqEmfQ){3(jSk9=hy|SX(fA}SeqJQPqndkxJvGI>YfT<7 z#jcT?-#^lNaW6*tV!|1AvS-UQP-t6kuxqNjMsX8>Bcy7*ZFQ_e81}S5BKr9dzk{{; zBkI9zW9rL$$7IsiqRkHvK;&su=TTT6*p0g#UFVN$#p_&xJ=uOBsFDwQ_Q0DT= zN*&FG96AzHqid%pt10_%-32*eAg6x(MkuO5r*MZiei~Xy2vCAm5q9*@T=as({q-j% zMRH}`Pcuh{N{AA!;_0UWWWDWYhJKbruAiecB6uw(8OUXwS20bi%C#Lj<1NnrnVF!C zu7R5>Q5lEo>(}M<)B;DuI&XP$j-N7RTI@K_alX$`71)RLl zV8aXKFAgV?CLxY>Qg3&N_{?N^oPi*00gJf%Xr|Z*?8RKE>4&d&b`>>NDh_u8)7aBg zGsdC)Xkt>#2*#gkMhoI(%83Wpc`SQI0gJeJ%PjEoZ8rS(_Q^T)5pk6<6JVcosdrPT(BFcJ+m_^g7 zF++rzSI3McUY2H$Zq^A9+$E?wJLiGA2kn!&@~UmDuj0URWh;`o25RJARTrjs9Ld2@C##k4 z&TpaITo}-p@Esz|XdnkYM9rv`Z|o%R6-$e;w%d^~BQQpS)~3K!A19&&1A&cQ>i~ij zPD~DbYHY?O_VCD(&h!~16xk-@fg1*Ck6jCL^wfwSV__8@N73R6BzK|AA)K?%j*$$h zBQ!Xid?LwZZ@Ylo%!~9HcIRotR#c@r>qxkDWkHfc0UV=2zzH%Y{n=4Y>Sj0++$mTK zK|@JBn?zwQ)wO!BEnQ{#b#%4`TjC){LfyXIj!zCvzvv!IFuy!$gneQjnj&+H?G;eeKZ#{D3$1p)at;yb5 z-~}qH;#JG8BcJDu1lO>w{P8!Xjy*iHUrD-hCvG25fLsiWfwHNc>Gi-@!@lK}JdsYD zh(1;Rok8<7x?A$)x~M>hVnKJj=$ zlDD$dyA|wpERVBIUQ6b3Y)`Cis2yXb9sbvsmqj z4QmJaOyWrR7L~x*k1_|M5;G1)d|XfJT9=}qpk+>_=dZHjG5 zg1qvMc%;?1D9Oa~qpdE)XfS!Iv0Wo!4ApXuWki#u%X@Sf29BPOx>4qoQ!kQ|+p$&r zY=1Y-(+M})Fv=>zEy*(qeX&M?HTZMIhTLyPu1lYddhY9+arIN&>jz=7Gg${y&fuAl zHf$IwpSK1Uq9zhkf!Jk`9Q?Hee`jzh1GAc)r8euHWvI^dLrT(`@1)UiX1+2uYc4s8 z&JpYbZ}&YlAwYA*Dh44$>WeR{e8Z6RhQp;V5-_pkx)CLLUmF8M=p3^_V(39T;md9QP@5R;X5>ZfL?bNzQ zl%IE}ZhU2q2q$Gh*!ytqlB2MGvEjNzGK5&$oo(m1FZ0t5Y3c8<(nx7S4f~+w+UNIg zi_wue#rN*!L~DCnDBC1YW43^wqXiO65QiQf+Lh^e@#xd)g>1uWHZQ-v5?;W1EKeJ! z?_^jMwaNcL3d}4W;yeEgZolguG6ZFa$c}K~hgt!;O@*BF#@3Z3hFPXI9BT-e^C2Pf zC5S#3!#v&kD+jNrSynRz^?dc{SCQRn#|t_Gs}}vYC*JscDk`hH?z$p-qlmYcBh;&W zH!}@;#vb($vLF7i|NhQB&@UXy_$lg~I!V77U){#IMix*H_p zmR)Z8a>VPl6mw6Ef`7DWSGs0={t)*2T8ug~kad=EDZx%iwhN;dHyPuQt=)S2QRK7M zeO;}yz{7&^PCw$ybK)*zBqw65Z!(o*XDL_M&I9WEWm|kJ#aJUSKe2xmwq>P-*TUI|uY(;Dj=#?2fgNG)2y& z9i}uQsa}H?HJmXhY4G$)t!Bq}6`8`jxIt_^x;M4n3tfQ?N0Gwq?5{U#Nj?gMF8^q7 z-Q)Fb$}S~s#HL8pMmXY(hfW2 zCH08@7M7MF3pJi#MCV3C;0iH}h*ZV~TjTkRx%0xzH80tj^IA-@m+W#KsL~mL;32I} zJrI^wKZXWjg}%Yv%ODq00Mz!1GPx;C{$U*3E5T5Yz!3-^V8TEWgo>Yf4o-|vv-m5} z@Skfd`d7M+BpFKq@!_j=SUo@^YY{ooG(6o0^_>YF1#>lm%^r55u1>Y25h2vtsg{;v zp!Dtpvw&Sf-G;6SWdUh)==vEmCK_80>0xJSW#DM+n{#e8OWp_EHdM@iW{Ur-mEHfA z4yJ$i7zRJHP%+gZFel85OtYYY)v%K8%eIdti2o)dM`MGm*bTDn;;=Hja}4AJjQS_t{vaRkPG zGIxfnv(vhMK>0K59_B~XSorJ)+ZEW#W5enVwu5RP!P@l)*oYyk5-(*%ww&Oex&khC z9lvlBC@S@!SURiqba14GE}~d{A!Mf0v6taVh{VMcCJeOkl^7z{1$%i9gMyB7hj zWg;pz??Nitr`7{u>_o5>5R%hSlD?g1G{Okq$FrakNocU0Oa5(p)7Z&>3&cJ4V3J0= zkd3Oj;~UFGAB!fad=#!TKE~slzqF-{(3ZH%a9=mlEL9iw=f5>`%$~LEk)SQKTlot zq$|9i{L$@E?3j}abC^;5b5JCzrLZ2mDWqNQ3B8F@MjZs|)kG1XpaW{8J{aTPHz~7J zz+eXu)+9-OIDC+AnGG3X?el2vuSag}aHAU!L`m)e`iMSO7y8ziCk7#edl*1h zGd(kQn6mVq+RtoYJ(p`dnK#RE>OKE;QMN6JNCXKk6)^9fIb*4e3*CF~T2(zR?u6zP z>Ma)EFyHjqdxrC#XAw6~dU4LsRmNwqtfK8cNJ{OVKA=FLBi_OeUzxv!nY%R#W%Ga% z-M$>D`}EaPZNAznidAq4NEZi`O0sre0HtG9M(<-_C2B|leL?`gjTBE&Crj0!cAL`z zb;~5T_f~8HKD{SzM|;T~eF!JF!adI}iO!q74GE4d_!gQaY!7To4o6*pvreFY_YDVV zbcjzGPN2-KY3;X~Io4rF-3Hq-gAfU7_QuvfL0L$lK)f?*d5U=oyzrx>YL=faCF>3Bo1=DpKE{kAp`T^ITgZdVe~4R_c8k&3loMkG#nJ}t>M5j9qr-*aVUXL zzeKe3lxh{Yc(=a<*TkpP>;Q=ZiA)8(_JMU?$`113Z#2aH+Q1W4`(l~!!*#Yz05`P=d}N?uh$}DbaJW}m z?C`2iMsL@)d*jE{C-!qitX9}zxJ%iIKdv4r|Mc~Wvc1cc_qJI?EEI}zfbO?FvP{z@ z#wr>XHeXAPx@Od+bX2e%O7>Xh7-Jw zUZV~H>7l_T=vr{bdjw6)-LL+ZJ%(fghM5+oaUaS>{lxYMq@13|@p=Tt!YVwTPeR9#07Jfc2IjfRF zYQ{WjQ(qfkhXA9t&+iHzTgMjVH=p89UnLA>+WYqUf?qX(t~P;<7(2Ob_)TOnp*~R- z?lzF4>x`)qQyYqALJ6q)=z69LYRqnJh8eqXn#g>gHy*zzSICdzxc4O|_2G>|r)7oa zt{DSM)EMfLncKJgoPZsNIr5*x@->JHJ;zvzuW0Wu>^vGV(cp3aR>SvRsJ}&8>QVJ$ zn?^0}5G;wnmQHe?m%RsR`1<1vYWSK0z8m-Mv;+w;ZEy>Ftdo^sG7Jgg=)~FFmRsPw z;*7R)PRge@4u4Atta{_zM6Ei9lC^BMFJ~X~p#W0M>BWdR4NNEIrjZ5 zb}3mFRr0CSPgJ+B8;P1Q2bHh2YC(2sZW052h^Ij1f+va@NWU;Jvp+o4MMT7k8FUxg zjYFb!7^j#ejz=*g%+md25vOJ3S&fj@9Ctc?%ERI!*Hr2RvWq5VtmjnCb0ktA*bqwn zQD2zi#yyVMKA+{F#y8apT0}FAqF`ZRS~{y{%My~>AHC(`AL)TC?6rsgC{x=X&$IFf zWL+pkEgWG8GoPZy7JO`2Oe`z12~m=CY=iCPX#za}2XY|5*;2+FLE-E+*goc7V_(et zM^h9?CDssXNeQ#O!FFl{+w+;>#sr|qm>X=8Wm=24Qwd#cY{M8-!4^Z?FiCO-8pvVf zH>4l(4;hwl+*c?AT?hjLrM#z+KTJjh+*mB;fi);ENY~&lgPQ~lp?}H}Ze|Mq2YnFX zsKR;%x{U=}KO@0>jjhA=kFvC;HrT*Sar>gTp>WvEtQKlBspY$w)ba#LtwaCM^#6He z1#C$btiTG`W>O0Zl3FcvM%jO2irN3mMPC0=QUbeoM5()3B{1njej`3;LC?9X1yGkgxz8ujC_TmZizVfsClr{LEh1ij_$Q zKHg-+8XHIxp5i8A^3aS|WvFR(R)h{!iPX}k#kw+tTk!)@gQCXJQFQuwEK#=&1NQf? zfX*&@7EKLgg}lKz<;Mr)w>zHUeI6XgxX1*f8pB$kN?|KE^67!;=KN(!cK64=f5U1i zuq0Hzgwi0LEH59?l)`#=pOW(F+kC(~`{+=SVEz5+t5z&ka{XXiUg%zkp5N7uJNqkl z51*VISn{1+TknRNFcWYO;qxV->N(Aqt~s(vzY$P?!Sq(?@ucOEdOtN0n*+8WSF_mq&TeH;xodzc%6zhr{#3npCSKqzr*1ECyihc20qq zz4LvXrgh0km!PMw6xY1h?3IoziMPET{D6K^aGVP-HyF0quRgLP)vx@3t+bZ3`Ru#r zA`AWZ(e8K8?=0U1TqTdQs9;`wR8Tr&G_HuL1%iqF8*H|xjhE|K!h5aX+l-z~@2Pap zs-|AGf7}+))rP86>1d?IcphdQWFHskTC5964NQ>28jcSg>#D6v`BDrm=n^2z+}mIS zDK&}6*pq|&=xcV92UhEs^T|>y{e$rJIn27K-MT;rZnTQPtYsN7n8?<3d)6M>P+#K8 zePJ2#7Ry&+pM32ooW#X>VG^z;i7&TQskrv+b-_K41#Z7o>`|2*`1x*}<9t?N)4f%z z-O<8g)-F@)fAO_lko;s!po&EO{m`~ny!M~n`%>e$C)RE0C$iB0 zs3&f{>4~#8J@Hf<#6IL8VXszLIX5imogUSTcm1mpM(yTn>b7?)@>D@!|F!b)E8dwW z)>OV*TX0V%JXW&$hLwaERoPylWZ&S{dF(fXe1NJ{W1W=i6K#2lMLcYO+5=(A$zJhB znXMI4N@)RvL*>H#Q+-wwJ07QKsa}^IkaEMGDDPsZ)0KU0cHS=d9x1#m)gU8uMp2or z>ZLW8OrpR;VFyoi%ZsR8mnnd&UWZ>Q>n?xQ7IuOx2tnVB?MxMrGRI(1x%E$v(R@jG zz4o8pj;0n3z9^5mvhPz}3$MSoY8|OlI=O8r760=_{Og2R%Iowr4GmYA3~o5tk|W-j zG!Z!BeT8*Pp=CLqRV*h_26nSdEKLlLw$F!vDyJ%uc0adN<}hmdW7PTih@@32dktpW z^ak4jAHg&axZ$=RrA9v{KTA9RCjT|eY7t;H5Dtni-*poCc1q^0b8ZCE$Lc56iqp1J zP$xFTxqUZzVr3@yg%>v>-~DY-lzJuTQ9vQw#m`0PkPx5_bM(|KP}`8t%_<1 zSmU2H&z7pu7h_11lIJ6S@OtY=PC;cXF)}ZoM%awHrn)Mg7Q?O0OciFm;9_Bd&suG-{H0Z_ zuk=2PkEYBB>=OwYM%1{NVeLE=85(?hZ#I582=*1Nm1Sg{WIy=jOXi3444sqa!iNLH zX^(x!XbmVHuI9)OO0T(DR>?6_d&9#Co5n;<5@ltEEm)$Y4EfX7Jk?3mmggIj==I2o z5}9J3bo$F;yIue&6R;1ySUhkj=Hy7Tj=rg7CIbREMELn zZ5oLFq?js6$eUnoel<4cdMp!i^_oWMdo;%nBLfZwuD8}8V}+u)Y-KTKcqmJHbUgL6 z*A>mY3biOk{^SIY+Syc1q%=Oa=JxG-_eqBz!T`RGvW#QRpaQSdz?0N9#*rY{CXNtJ zEX(0#o}@Y)7d76v`~N^xA0e6@=@5P-VC^bhu=A!qne5&T)kaeDb<_H;Yn+b zVziujuII{-XS~ABCV}Z}`z*Shc3tu(Vnh<G*0jU%>mZNP1Wu) zeMPRQsZ6CkH%8T+8jGk{P)g9j>CY&xDFEHDuGiMKh+fw&97s0j+lYWSweT=ZYl__n!BVVcN(R zT8;0`2T<(l6uB@@bbC3bx=P_RHOrXf{sv9g{;t{?@I(X&Zn+ur3K)AE9nh+EiuChB zkR#*Fb)|Ice%j}~;aN$d3(bn>orxi7p}qrZ@Zo#AbUbZUa2A?CI9BuE_SB|DXyct9 zaRxUPsZ^5?x0N0;estXkL*60sz%3ZeslBi(O{UW)mp%(0 zi04F%@z|yUo%?Z6$I+D?=oC0c7ITKK)EB7pgx1tCwM}>ul~&JO&xQ*#`^rt0L8W$8 zE|U6ozY(*C&_C#$WKK&4vK9djuGM+=Ohe(@(I}+_XWJqnkvWT44E^&9SRX)C`0y~1 zqVW{VznVAjf@)nte+sYdwNZP+6sLk)P8Cfo`oA`cI~EmY3|&942!q@w`-FQ5MvY}J z-^yCXcWRYw7I0IFw!eRI3s{6Ls}(K{y_ht59vhVmnCGZ2_XR>3nFU#-T{LfWTwVeVEhehJBV1-unbbfriWc6`MM=8FJS<8SAT| z4QT7Eu{&Id^TIFl5n0S$yYn^ubC)`bwO@l*g8A_YQ}7ZyWxlem-&nToh~BIK8)i*H zFOzTtNCNV9$)s+g0U6m?V|I6_;~eEu<7C|^Pks>FG?psG7F|lpf)+9RTg@<6N``4g zkbxOrHX)wiT!bBBK8Ra9MA3vQ+dWb(vxx;#7t%d2$MSvQ@<H#)bUgLbC%rSbTBe`8FxPO{c(R!g9q>bW5xHc{?Gq^ z9V5<^mqlXhBugoz4Wq>P;bqK4zXL9zp?d;mG&Zed{ z22q5o%{HHhW_4XA&bdluelGma5f5+hL}E{7Q?0YLzeme1o9Ndd)&uJ*rP)a3p4F-L zK2I*{pMUAk;Gaeufvfu>2MpX?kEM#;BqaP#DMIpIysI~?J8ctA6Wm! zzuI2^mFvU*s?#;-n~cv0`+TRFou=GcH>V$rn2Q`Bg?0$7DAW70!cLIP*5!z?H-2V; zy;sRs?pq>dhr>KwO>EAodX|+$C_K>=>v0GatogpR_P#BfUBfl0Glp1i$=vT}zuu z3cWyH2;NrUwEi`#?=OxcPDID#_lx%r)g6iOh&Ni4JS;KO-Hm(&E~1o>kS%f4^!59+ z$G2~c&VKjOv&+2E&2U*OCAnN+v3>14)ui;<=p5xc`xAnD3rd0yoe=2p%S_UXl=Y~S)1?U&7Lau+=x z7gt+Am8?(^_n7T?QH{IpV!7YipJH#P?B3RtSzHf?7Fq^>y_+<5af21yTvNp?+d!@k)6Opg zJ&;;!j~Njnl-RE1;N>Vdb4W3zo@^Ji8wS@2!14ByNeDZ+At zm4bG6lY^J5L9)rft~ljB1LchlRvlYaZ~5-6^gRgWqWG4wt9y6Lr^{z}sGz1dY2buW zX4UebXOeP-`nP-0SERI`z+t{wl=H(X_;)B$s$TxC$FqsKT@J+DMbGayF zEGCXULDpp);)`;UK>1H(mgc`p?K53;(q=P?f5!P7zz|cH!CnuYDM z87Ntr4|7pJ?b}=RHtfU}+KQ2kb;_om&C+-A_M~)vQruh}?M>a{#$`bUwjkz338ytv z{7>KOT++=cGq?x5=?5amhC&3CP2(cj%;=Q=c46#$(fmZVIP8Q2a}jNu`JKpTaX1$nXUqN;)3I4s4`^NoH7?EzMR)xMALNTDOLRXJ0>%6Njf4m0>+V3!W zzl>oLmh@D=C#2`ZYD7f(K=fuo0FW_-N-t5lmTX`mO`{NGXL}F)1Sc z48Pi;DYIrA3ND5O!-Y`gyXR(hi>hxz++SUEg*ozH$PePg&6&H@Wc$nq zwgS8}jvDH(X$7jDS7+}-U))jPco$Z?xx&C(^)XW6Ty}H;QoOG}WrtYcvH6s}ccS`; z`?~*duel}Yev)L#q8Gzdtbc6WG=GQgXM-44W9nklyGN6lM&gD*#$Ey1yKsjYy8iD3 zAPW2^cEdy1o#72+^u3oM*$IG;06J$kUCWzprG=e^x6mCKX|-}>cVqXmmUA^{9^a@D zU!!o>C1S4y>d|-uO-OSq*syJ<^bVL}En*YLqpf5Fs( z1jx^!oayMCRd;fZUG+Z8yEH<5{s8J?5tUspI|xNmqMRkI<2tF^cwPfW_BZVgMa#QS zE@{aAP(Lw-0hoU~YuL0}{}~}@U-9nITQ{A>!AbcAcA<*C{jk7ML1}zrC!==~%L{fk z-n|-a&8!|rEQX>NtvZ<+z(Om4(euWBIubYxfuZK&^BWPaX8A&L3QQ+Lc*H)-BmCP* z(E%}c_T_a6*0nRYY@S~$%^4Wx1g!>>u@6uJJEpohn8)A_fEnA@;-ZSqesp>)zf3(l z*2LWa^%2F6cxWxkXBr71=Pyu#?QVX!7$fMdZ(Iid;Jl|36aGjnEvH31?CdMHVen9y zRDnA6(|}Kk<}p?7G=67b_7AYf|7rjJe-WtscQ;eSIU?Zf#5#l6nFiLw1#AHlz{T25 ztut`Ywi-lxJr%{EBDVC=pQqh-_Uv*haM<+^3}YAIiuPC-Uu=^74bGj;I_?VKCsv}K z7O&=IlAsxA1JZiSx~Q)^=n>`-@?OsHpmDzMHvbzNxvP0b7e5Z*8eRLMyUPQ?GE zhHih@Tn0@u6F!;FJ`fT4>)LA7a^SE#^mg7zsqq+Y`p9>d%E31!>67L_AF7yx&T=Rw zEIr;p2=WSvn*+)UTVOhUvJSaaOtseAO8OdTLq+^x@gkZVp!OUAy-LCIuzCOb zJ$<*!)hOKV@Fv&+7(whZLKB@viq-G4DB1>vw{*YV^Aj`rF|OzKZl9%TI1-A>*=Lbo z4qAw}4#v}KP?HF~%v+j*pcYi>@RV3A%_8tlA33E&9y}U_SWQ^2RwIs*54OEsJNG-S z%N}TJTFeejoqhs8^=W=^>Zd8y%897Szw3|%pL|L~lo@r~?Xz@9ZwG-W%hW#)D5(0= z<4|+_eioHxqCC4V@fiyi)3r~uaPaWS0Z4~Bf*4K;zqij~@ht{1lCsxoI}uLgBoRzn z=Phbe^Ar^3@6jDxJzi#UtK4Z6)w<8!`*F zMZi!}!PQ6K?0etr zw(VRAE2*}ubAO#g_K*FUB8z`05-`};V0==RR#qqiFOUV+0NiI7O^oM^BaBx7VjWcV zSInadM!AC(K|BQSf}SC!fWCwkBM#kkWre}m6W`DR{7-`5Zvn;~}JXUvp&~bDW8bbsf7UjEZTE z^|+vKmbiwao^QPi@p9Ic-sWb1EP2uD6BYpRj`#iCW!7s!89wezCA*Ai(0CTG>E0@c zmkkiHB^G-{eiE&0g>U^K7A~~Yfhy%?w5w5S*mXAM0Qv+qi^k>R{%{z>A+pxPdy@AW z+9yd==_Wd7j%Vh%#hkEP*E4) z!ghicR6_fnl%05X-e>(ifwjK_ouegr% zj1;iD=i0hO&RnL9t)+11I(nX*N>~{oU^HbcF&d2bS`s<{r|G~AdyVm&ibDsGkoSAs zjTl^k)%&@Z>bc#YE&xf&4M(D~o5baJO;u$Eya5|rN7*`!QlKgzeqBY}P()aY%A*@&_?}AO4&vKt*-7!^ufT^TG zH@etjQv}YTah76BBZx;%nY@>tcu0jZud71WI3^2@_#Xu}$@PC#z5o5={4;n_ro;_W zXT2Kpj3Z;#HAAmp!GgJh2m#ZNgd0bm_8m5-qR`RnPud>Y)9oI2ig<~CWix>mXvfGbfJR{rioOcd7ZRS? zZ>%};3I&m664nBQ_teubVu;B9&FO2F+n`y$ctCTxB1*&L z(X=Qx;lY5!iPocRlA3Lz!Q1F2WF%B1(36a6LXD)0y@cIswzzJba@N?}tieo2m>V8! zSN(cZYnzNB4o>lK{p@v7#gD!6!Asbs7lEuc=74t5TbD@>`_J*9oOFD5`;UE=P()$# z*NiVZxxR~n^3c>56C;R2l}izOmyJipr?W;l6=V;BBX`n)IF+8z!-(kjJWNcl5!7O5 zRKmr4)6FBzBWbc&nyJCc((x^HVX5))C9!j50?9h{>Egc+f;{mLlrv0Re@xyK zzmmoSg^-HV-#{zp_l861r>R7OduY12Azu^krQj47Q@*GpnhDt2iy9h^=(`3dq^_Ck zt)=ltPBUI~Hdz^B&<|wlcKlr zM|{r>*>bp_Gl<@Oc#<)U8~~X?j81p8Ma0iRv>J;CpQ;>4xq1tA9d;8V%!u%2K$^r3 zl+_J7D*dEw-2O9m@tS=fW-iC_ba9~>>hfJetp_2RZJPN8{8mF791A;g`FU&JVUQQ! zK+GV3TXVuSsWLM^ipLfK{>Yc+sMhfHnwoPlkv2X=D2V{2B1v7F`lO9d@|e44q?KqR zBO5=nGLv_$DN0m+0)ox=JL1Gjk4S(+%jGtoz&Hg*$=t2{M@I4~@%|sXz9i_JECrh` zW=71b-~8M6{b&DTbQu|d0K$uUK6fdR%YlNf_;kc)wWU6s7R^}Ro5!9-Ol2ZAlziOs_FDB%Fl6&dDqeX2oJ`v^3C}I6+golSJMD-Yoao7yqdmR>4Y$0|&*>ow=spxM7 z`8h5yu0CGvb-Bb-GHk{H?EE~iwb9m*F#TMaUMT*Uyp^9pzcD6FKXwn1;i4i0*&4JmAuC9d7% z)Y=iMub)y?=h)IYaNVTG!*-`TO(ZKkN5@;}$Vr zfE}=*Ry@tvF+_{MZu!g50Jr?j(~e`h8ZrtyusyYZX=zxVVdwSryV*x5C?_M=*gQ&04gfq)se=rIxdX1fhXva6BFv3>9m%hI0WK*92$OwbMh+~gOJ<{lSBwsUjUVK-l%YpT z2zIE-5f|uFi}m?>RM(vyDG&K0IXH{JEqb?A3?c-DGx_sttTX!25d0wByhn+Vm#KWs zf95_=YTp39ti4~BYa*}Uv-~u{0H<9@o&mqn*TOh>)K>1y#qOy%M+g41VHUJD+%ngt z*Xl#_jenE-Ppzu+UYv0W*B69e$jp2tn?3jY$>qCKNJ_NVDLZAdDhgR_S8P850@53P z-W-AI3-=-?#ByQeycvu4&E!0?UhNQOhfJ`43UP_?;08n!HX z!bl2CTXBK+Q$XFPwiQ}G4XHVhzm;VIImH`l!VwUTh9yqxF?-QncVr|R zs{-~}k_-UPMkO`eM@gV~9G-#4q3I{}>_h`qP&9uiQwe&J?na&VCMQ9dL)HDv`c`US zgCyzuHV2@mIg(k?Kw|RnfL(M?!GGX6IEu#$rU>GX7tcj@BN+Etr0N9^V!Rm%r5+jdlRaJ}#lpLY z_1rI7%(5+8*;~KAwuZls;sruup&IxK_Nn_Q4oaTS`$|RCmuq=XT^4yI3VC_R6%|xJ z#o7ZI1eY|2M{J?OtQ?aFGM?_a^%A@n9iA0mJ^fYhE~NT6j};4RHca|gv(6{o2-6p~ zx*G3zlim|e29kG;XS#{@Q{b?;NO7r}+NU#$wO(WXZ%4dT=CS3FR=--#oaSja4(QhC z4*RR+O2uNvocQ|y;oU6CBgV(>rkdSATf+QhE8QE%%MpH+>Wtavx<}cv-R``eTS#wF zfnlEAe&)|LOh2h~PXI;%b10V+@)$Xyb<@8B)NqJ#M1z>Z;MqQvZ$P#nf1&N3L3xI!dGw3&hOaq6tkPVKCU z)%zv!(KbsmM@7FczU;@{g!_4u3yX}h0fF~FC|#n0Pfk1#yD&r`t-bU7ai=En_d&gZ z;6;F%aE&5CUGln6N%fuyc-?*C;J7$?ZxpnKex{zx;GhV zCQDy(r40AoyXt)SGWeaP{24IyezS(c*@)fW-wQdOa~y~7h$=x#pa50D0eXO{bSr{j zTgvPm{3`P0Q?ky29P-0$NJg`_MYfS}NnEsRyrtexe;@iuU6Nr5gj7_16Hcz`0w3TC zY9}QT&>?WSr4`Q%n~+O1@^s79r!P9N6Dn`WJQN4&BSN%T0Hj~1tQS=CQ6t>t5UBbz zYwBaHxE9E`fX+poI2A8`yyzM`U(}^bao9=b_IabUTf1Hk>IoOyJzbGwpvZlelSD0k zXM=oO+(MpHRBUfTfDACTTVfb6q)^~l_IGHG@^?9+Ba^lc4AjY6zqrN2MK>We=TQ?o ziIy878}1H^li$s^P3*L1u%jtc`H63Qyxpcrap!n>P>Ks=|CadIeFkR!u_;3e+T;}q zoSN%I%A1kZQm75-zMZUHArZ9Yy_(Nhzrz4%ZQ;RGPiAm9lZIfbK>3;NTI{6ZXy&CG z8NSK=#FuwBLc}P>H7CEk_#*b%2R*Dnb-)JWx8!-~M^XDMkp;DPb8`dQW*b|wkk#EE z!P8UeBB0G<;2j{YlfQ2(go4WE!^x25LSMVC?v6OO-|P0hLNzVeiS2yjcc2odhJ49} z@q4GFG4(Y%&3(!U&NeC;1=tGp>!%X5t}Af!Td0m9;pZWRE&0Nl|~ zGn}bFe?s~A8324syxkWx75-f;XUS$jY7;A-S@#Uk%InV88>!IYY(>jal?7_qWj zfsNO-x4&zN0KqqeGo?AQ33J*_{rC3!F6e2oPM$u@Tfq1D0$rUX_U_>F%Pg~j`wj~i zZg0G33e3EDg$?w^-{&3XAYGZP6FG`slG)@I)Cf6^)D=s}%hh+WKhN>YZb31Krt5n> zP_>F0{-sTxNWA$B`iZAoGHF6-cr1&`O1rUyIE^R+c6L4;xr&tpD?Z(=VVEFl0Hr{w zPvbAdltt4XJorVQZ-99Vwfq%d$B3S6SF0>@a1Chk8)hYtWA|CmU`9Sf{e=1N*yCtP z6CG3ro*yx*JqZ2{WgLS{tv7Zb&>?d+S(U%5P^wW#47wzQ^caDx3p7J_k8d0A2ml_& z{M2n4=oMaI!K?=;P+(a%%aFeZkRF9x!c zT4ow2C{WkWM~yw^41}IFd1;bu3d5CducYppInB2=i#qNQDq5=Ch#QpgqTuo4$en;} zKTtGXku9{=2bOiSa>K@nY~gOU0`-RD5XZVQUe{Ns3;~I*&j4eYgr&cFiv7*sN!O)jOdE#@Ys18(YEF^4QQ+SQm+)i7_hmX=#$=AuK*IVl+4v$i@pDCd#n}Yx^6r*zy5vDOe zYc-uN5afC{yPLbmN_43Y2oO@xTuzGgNs)>|T3ddPE`D;mAa4dul$+`fQx7zS-j=9M zdr=roXY~0U1<=5GAL1_=9Jyf6oYwKUl|hZDgHfEQ?BEtQ=vx%8gIYO}S#VC(Nra%$ zvWy6QnO_%r^5kfEMM}zRRUpxt%BcQUb*{V7P!8)vGXiUK!-loQn7^4^WUXxOlo?LM zYVLW77u}LqBulk>AOkN^hcP1eM;Wij^(?}_{_WtzQ$8|cYt{Tdx6dG1a1_oJld8tV^v7Nqjmf%!^9gBA(Y{m7BlqIk02DVrv?DG5cph*ZqN96Aa`@ zpv7lI!_AE3!RG*(3#&}&INvr`Ud9Rqk9*>HS}CrbQx@V2?$L7;y-o#_uX<*ZFjE@ zT!`+OPMCoZ&w6Iv^gL#&1Arv>P)Fwa4va!*GWAne`I#QcfRUA^8v$bN4Vr1LxE2gW ze>sY6ta-+G_OGO zspPMc8W_0oFdA@gTZdL9Fi_I;Kz2bwR6 zP)okpQPbMmC&)}spb3x?+gz9|efig0nh}E|!x)5xfV@05|)e)Q4se?dxxU?AsWjv zOj`6mLB3|6LdF1%Mfb>I__@AkfqdU9x*bYyqKGbiUG}>H#@A zIRFb%w0SCZO&sBm1@v=|ZSMr|Al2citxr8`r^9K59XdhRfO!+A0D3Yg90&yjU{qgN zHK3sym0viojvlEP*?ZAXNg}z@<%_8CKW@vN!SFscPHW$o+vvX44Rzhsw)#`0*yz0M zU4#q5+78Uz@TdR>hhbb8De#jptk}BTRlNc#ZNJ+^1G?uMO{~xw5@;gUUzh z#FfA9H|iY}5WE_vjrxVmze_OU{$=z1MXrhMHE{x-lZoHsr;~5Y$jL6%?8S@g$6zNy z10Z;@6XdXPf~}aS`;AJz@w?o82SnM~ULRQIaox=#u&sU~ZMF#W4LeSuCP6$7=2N*IJ12as|pm2>mgw7z{As=BI_JWH#=N3e*Kzn z+r_o8;q8~58!K%ul!?hJm6Y^s@Qe;qGW#&RzLs3Pl%HL_YYU#-1;jB`=1zXlHVI0} z?t{xyqm?;jlvpu-zgsg)e1K11fhxmW@kg{y`!3WcRwE;nWBs1fHYT6GOVvNK_=D}b z%zF3Hz*AKBDZ?Ay#+)way5{`BDdO9Qyd1?dl*#e}{;o#CDza%?%5W*)wejiJ5QSQ6 zAAvOf^q&PDX=}NQ_{lA_B=rF8))Gi$VQ{KnP6tbY#@8G(3Y?%WM>56`M2~$I356yQ z1;}Kx2xx!|K9z#nI6=-C>fAMa&=}{HTAkdLWpUCU^uxQ=7uO0{Ga03ig;l*c8?Ebi zXos~>t1NfM&A$^Te153n8#d{1@$1(3ofx+30m*bZ@-Jx#}#y7V$;@}?Gxi0@&& zV#JHDZ@7JY3Ri|_$30iO`4A1)OGU2EhU2VE1{zZjN;q%!(|Q;%&Rc)hDcke(GbF-& z^h>3r?bSjCz#@7BE{5U|tV@5C9S@@QOvDuwUBuT$c5j7xT`AaU#+INO(E_9TZr_|& z>RMReofljhxw4gidXZ?a0wiY?bFVP0L;Tt1Cv$%s-AJOjswZlpoRfT|M zr=ltMsM%aC310mVlC;c-7Tm||eLC)bZ2c4iqa;nEV@kWLgAG?Y+pt{j~e2tr}e)vP* z$$^W->;&XD7FgnU9*e!k z$f4|fRPXCwB_Dg?PA`n~C4=vXb+{)Ei5kNW{|Z3mkG;%r`EGT7F8n~~6&1#P)weT{E8s8QpTB+f#)N4Q^{Pi0hu7OypCocb6qheXpoKk;@_BQ%3 zkiQ|{;>DQ2=6v#7e0-@-^$acQxU4j^nlCPGWs6V`DnZ~t2VL%sJy?RBuLsqBs6{EZ zgT1O)g$mN|JlF5E^zKNY~~YR4GzlN5MXiu0Fby^7 z$9-7JOGbuu)WTX3^u`i25k)cq?+Vo_GPk>#AzVgWRv86yOd<-w`~}+3D>eRt-BVea zTxkG1B6r2dB>SKY;q7(rZpe(|7Um?Nt8nnO(9k1g?mTAhT}m-#lp!@+vwa2VeErSQ za$!}|or%Dl(}yQ`S{$E+>41DM?|+ZfKqRJ;KVYLo(Il zg_pDQg~SuTCS-Px0z}qIqn^xvF$(7&C4}$}2mwk25+w001dIrFb1jsK1I#h(_&zxw%iz7hNc6Oe=YXaQycZ~!vVcSKKg{~O~7=(6(A z)4HRGhklX)mVqzTtiXw;r~5!@nD-90y*1k}ET48VkPCB+GA0>&c734>p)D5_pySim zrFf4j-5xkPZlK(RoBlTZ3zfFly~;M#nTa2TK<(Lb+SZX>RvT$hCi}#9-MiI3@6A%$h1hHNi9(5$aNCvh!+8tUpBi}rx14Nrf^9%sUTQwuT$sh-f)&dlNIV&^0i!Y zvvJuW#N{1f%PiM%$qH|VM0vI`DH0rK~m zm6R$*G@Qp~0iJqwpXF;jg50qh{#H=0_Q9Z-3eGzcd}I;*n-f3cG)0RZq6oV0EvaoM z0^exFp#B%CXxwd|CAb{`)RP0fgOj{BfZ*<>9d*ea`x}161e&h~Dul-%26T{304g6kZcnKR1Dig)%Gp-1t>9Z$wpj^tc{Kc#^p-JYj1nh*&?8o`sdlLNfb7+Bf>|X ztrtusx{JxygLi$SA-oUqd(~O&fP3{E zG(TbMY(L)JVYh>Kedu?x6u5%3OgevcB3%-2m@V*kT2xyF084RV_aBKB;D26hK^^}e zU475CLf6-OL9t{6ZDs2Q@XA(@{NEa5{8S*hI?Ue3jhex7FW*A6j-~=JtSr=jcvT1P z2MmilfFXlIN&yz^Xv-63^QKZ1qSNVJkX}wkL8?T)vgc!h|Ic~Mb*r7<#-{% zi=@pviA^S~m+k(dvC@sgO-TV%jH3t($7J)E&}YpjXf~ z08{_m`92FD;hr27)zi4$BLyEJ^J}evYb6j_#a%|5Z^STqz+)G<*6&uMZaT~T8-Yf z0<7jyh0r~7VZC6l@%#5K@RtCjt$1h*sO(7Uf`Xj*6nTd2sYHet{gpO ztLiJkeX|^~UJmN3_5dGx>iGzM)n=;$?alkkP1fkL(CyY~OX>VA^7w;a8o$a;=wGNj zF$q65Vy!o2Jf?e;*qZqMHk}(z2O)RzDDE@-tz#HLyfK5?{}*Vi^-GtACrxj>lV{!3 z1(c3@$@=FS@o(P;rA#MmfhVHDTZ(pJGQO#ThS~Py63UDI{B7eg`Np9+A90P^Ai3ZB zEO!`lCU00?@>4Bnmk{k8OZas|fZnhHSu0$WR>bnku#$?5H6YV^M0de{u#2eKDpr0d z1QcECO+!Q9pvv{=sfl^aZ7egsS(k0bzBQ0@fDuO3lrbc&>&JHKKLBs379d-&8*)J2 zQwjS>RRq#<{Ox^~m&Odi!G7v6Qzs9*adIhN}wGkJ27u@ zK^Ox-D$)$sVz!|+pr3(|ZOk|UsLiP_PDv8p(fG+~7H%R9R?&Ojw^%nHY65k8KSjLH zegLENHT%->)1eU@hzb>Ll8yn(HKrSinfq_{ z*$=9IEoa}yyxQx-A-FA1NmAce(XcuyqED_bzTi2RW+UlA#uLjjZ+r{n9g=^ER9BFM z`8)cSO#X2(e(m-ga~ZU~095gLBiLtXp^+tkYobuAG#~68G`DK!H!6EKz&@wWcb(1$ zTl4K?{>DY`>sh45nSzFg;(Db+dIW!=9$cUf^Fp${ULPO^6AEb16Rq`Kj)EycK z(f%-nu?hGjWoL&`B{cNN=ivE)z9g=SCZ6!&v0E#Z38r8+OMw2dn@tn&Wi9{=zor!Q z!4-vB+_y_S=^k@>GU)66vp(q2tB>`k6|Zy@HlUk~$B+%fcup-@Ybx5%?2ylT!lMvo z!Ign+IRNOt5BZs>?Q`=>(m9Cu1mJY1IRa8$4W`y;CwHb`N# zps@0lT;H!tvh7{{MXBkp81=?vj^W~gtiC`Iu?!iv`x#v@@(FK+Q-Fd{@W?0ct8pUX zkD{blBpwh;n0`ab&YP-}ErTmBz9g>F08Il{L*$Xng@i9@Hy~N;Pwq0G4|6HdPulT% z2!xCJC!a}KbsVf)+z{s4V<%)&& zujk^GbWzf6C@Z|5($2CmG^R&D?{#_t;dRpDR!uScnU0Q+;Al7(^(2ol@RV-0b>12M ztDsfu3P9%SBk%rW%T0-xwBdK}W|rr})F*_BIv;xaoJ>0^~%l~TejseQ-FG{>E1_g&chFbM*M+AssGrkOvj^EC2gq&BczQs9YF^n}QSuKg{~Rn*WENXwkiTgDDy`QSoex;PMn&arOn z8sVwiB|$;3om0yvJf4v%bor?cWWNhkADY3$GKcCdJGtS^NQ5b5lH0C`Y!o zdqVlDj@D<>!rx!~BXVv=mIZqj>EeUN7wBJ-9A2Ks$2D}2)kq}jD&hrAY6jv=5{ zj3nvU;y3rzu6xW7^`vrcL$();i}x2=`*Hhy>yf`zvr1Tx8puvyaF zEUv4BL);Wx+o#ApG2%ZQ^aSoD_4e_zhOlJ`1B3v3PIDX_bf6N9nU(7j{*2V|(-_=4y9xbPsTp0w47{ z64Azk+h?JWfq<-S+ZX3O7$ev+Alhxo2=$VBis8DyBK=BppXF2K-md2zOc{k3s7Yg6 z>9M1Wmh@U|o)2SmF3lUta1ZWqc|6uqeL2lMbaqylew|UO?7M}Gid~MCZ>jN?h@qW( z3n9{uACSxjtg=U|OZ{70Zf`A$1XxwSW}|uBf;|;zv5%nw;6+$<(GZ>gtLv@Vj_7@s zm?9v%B4de+eU@|b0CmI-vGv4)0>K5zE@C+Qy=GKD#sbnBzjw?sGWs8%B#4+1zk|m; z!4?80FEB6_#EeeFFwn2Ck9c-33?WEhhKO7c&47fn_&7&XGY!(VnhT-+0i;zJveW;%8`oNaS$t`;EC>8syU=^vorxu z>$kwIKbWjeMNl}%hwGBwu6fmbQav5RLn}tzxrjJCl8kMEfFDX_p_L&ACgDL$e7b1G zK;Vg-<<(d1_VW(`>m)Ld75HJaM*$qfbv^~`)XmhC$mz>qU~Bc29*bZT{Lt z{Ps=h%ZE2uiI34WYz_NeP+;n6Tm;blOcG871Eqj=Bi6lpmnUbJDF|EX9;q;_D5c6| zPu{8wK&km{k7~Svw4o~S^zeTAKFcu`iqon)5OQSz@oo7&%bwO9Z}98aN#rj;D zucCOUhgz_YR>)Y&LBP{&%pD+V^jts>gHJ5?E>>%{d?r?O*o5p^Qk&O50$)7L1EWq& zAni#Nl&Nyjmft87|24<5(nm}S&m(eX?fpJ*-SJ6uu;Ir>BFTT~?K?4fY9aHcd zyQ9IElxORA+{x%x{J0kWW+^q^9hT8jl&RQevlF8LZ7IVUaoa5`Hs5sB+CmN@M;rB$ zrsjhs!euSp8K<{*{}e5Jz}-l}f!1vRSm@I5qQWXr&P&(>`*y7x2t?^vEnv(49gnt~ zhwkY9mdW!w2w+kvBihA!wc>zR4x2IW4&OcYfSc?yyh>=o}EeF55!a~5L(Llz#BVgyf6z; z2>!pH|8Ngl{((=rNwe|oL9UPOvnLZ$?FJIG``H2~24usJ)sYoHJM_+MoANxG#2g?e(fRs1mZ#S>6bn+kpS^#} z-g;vv**)KfcHi8N88t;wxrp^N5BlX!4rx$ReZ6sQb|ffeSS(=*aS~|1fe-qsH&6F} z6@U~RH9s)iUlkP-hPi`eveO8jYKddGABf2jY-bl5O46hAE$H;230SgGwfB|bA4iTL zHBRg#+ON7=s2f;zc>dVB)7p;b(l{`)KIT3HTdBq@{;&wS-E{G&MmcA_l+m4+=y4*O zZds4=H|0E?@sq5qARF3{&l?S)>Fz^Io!P$6daSI^)ZhC#(Z1Z2{$<$g{&0TR*2vf! zZk>XKg#}^;$!MWP^%6Djw&Bk4wadZV(U3fDTp^l;kqCB?3S{$^3=On9cCX_)KJyBpm;YUp*jRAO1EW1)A`Bg+$&s^fZp#!>e_R({Cc&z&`EB#6|25{ z0Wtkm2g}e8Rnt{lZ^E05IhId>$Bs|tBpwOaBlKaU7&&0qUp)FfIV9v8C{bv^lid0f@F+7H9o;e?rt(xDe;c);2fuYeY%Zy1-eCteVrQjzNlA3l(`cN zKWXW|n_1PHd|>&Ru`yVB%hInTR)gnvZh>&XpANoJ> zx8#@JH2##*pWP{D4nGh73EFs3?L+HFENLXreyDR$-KUYO_B=S$yHkLb8>9*E=FEu_ zU#(HXs3O<>!fB&R&=(Q?{3iMZrk6^8+q)cwpMsA9?_q_e4FRcvF+Y517>5!>*k6Im z2I|(3Y19;&W0G&u2LO9g{CI;r;&05Je#rv`m0x8H9{EB z$(98$r*o*+a3PI$r%QVd?Xx(a$_W2d@i&KtoP3uMNWW8m=PvvOaUzO9nkkI0EpX@0 zMmC{G0+K$GVbW-=vC+w8;r3n|^i+hYVtkTIM zezowD!kvi{ru#7pw2rJya+7QEd>V>Ew9|9<54X0LiAQZeHqzGw{^alT5~nye15jeV zoL05&$0lsQ^cyt1&5N3Pwt`%FaPQm9TC_}`siji`mIM3@!Si5|v^1%v?c#B~LDF*} zkvVJk5q*~oCiatthT6m4sEw`N-n}QAm#+6RQdFjvthVY-@`1ig?>0PT3O--kj_NOZ zZ;5yVY=G`J>RdCEw&RSlB*53z{`IxnK`x} z^VRy?ffUd*x9PnT!COH)1#7{A_~sr|&>G-+EXj0|7*(qH^F0MB%Jrx`8@W$>)mP^O zpP<*Z%_SAP=e|p!E6dIRB~n3(#V|5nF3{NoGaIXQ-{GL34rO~{R?wD>G}qK~S%xDL zuXFb#^}KS`=!im|Br@kR`fJ?Y3Rq^lanwro=vSJtwchdB=2tTla_Ho|`^}}?fb)pG z=jmG+7oLU2*napvr0pRtq5EOC0b9hI(IRnjx=)c$=F@h#Kpvj znD*j;{ETJYTxu}Mf2Ia`q8%3|+jaRAVt;i~dTn`{$E_8i%$493EiiOAhA_!^)+^1M z1RSK>@(KmJ*)KaKN9NLfxjEepZFX&g- zb7o$k;-bpcu&1jSUH~rNg7Q~6H%%4M8}Gw8>UYrg=9wiDmsU15=j&YLG^&aD+RnR8 z!rt6!se*mtMtE+P$aY;<`|-ut8{(o}OZcwyL!0|(1ryae_oSh(40=bs=Z3En5+AMd z((aCAvr*8Mg*^vm9%P17c5ERF#}yPOTb~g;5@P)5?Joa;gMPa6PK21|EU}J<(NfJ? z5h3pI3KY9rranc8;tyiUHSO1x6u$P@a>&W9f42h8LGzh zBFh{(8v2N=HqsvIohI{XCpGCtRWY|Mt;0}1duPq^fx-K#ukSkOPR$?!JyIe^Fx}YN zOy|H#mFk=`=8w%#SSF42WX6Ytt#a7PIQPL0R^4%K{Iopv&81Q3kVuo^RZM+iSJ1Ko z*#v!xagv#*ylpu9qyRZxJ(eQi_I(8>5q9C0Gw<>V>|SMV&c|)S)C-nzp)=w0omFMB z0EzPGHodYM$JZHJ6!(1Vwq>Y$3N7YO{lO^8Y=ecJVKri)2Y8s-riyTbhDfb?4W21P zF!~eP*Gnm-G}WIl7o8)O?{M>BY0qgz9z-0HkBAG0S)ul3_vPj#e9}~6V@qF!&l~ls zZU!gwwko>w)pNMic5Wp7Y30`$hPKp->?+!qv0=zzWA_dy*NtGS@`>?#czwnNS`4yh zPka35(vWE-`~{_QGwysM+}}=S$JsNzzAtCK1bd(n#taZ{{uWF_q_|Mu{9TP;Ze?-X z#jZ%3`_$BxJCv7c3?8~LHmer(fpOvbHt9;{yP{>odt+~#jApMrA&FPU9n3YEQEKbD z=w3%Av$p`9jn`#@Znnb?vz>{PWd|Q$N`mKpEXUBabLq)Ws~R4Q3%il!!UqaVoQCkb zN8Jg(iFGT_EmyndAQuta53pvV0Wau)#XhS z7}5!X7L$tzslszI6q+rXy_D`Be8zF?gytxE_=IO1S-53a@Kslf6{mME;{xa6L(QKZ zGTX?;F@)HO91y&`O5i$J^6cTK;zCq;a-dPa%Xc?SWe+T5SqTlX{qSsd{xR1Vk@@FU zkq5}^rB0x3=~Y<9hw)&IvCdgXQeGu%SJN$GN8Ty_ORFxN{Ym!U01~`hUpb&kTn&B` zt?`*>PwfMU7wyRu;A9A9Z5w-UQ~2@bl<;B2`}PlanaI4J>T3}1FJoghvJX11@^E@h zvycNjx16v8P@fKzPDpk(avpO(mX2352u{&zA|A<4TPOQk;15!OO{{> z_m|Ghx}b_9zdstucAJJ}flgHqC7uL)G4*40d2OZ(C0mZ?Au#wN0O00mJQ#Cekp<1#q$!<((LPVfk4WM`g z3_)?w%Z1omiQSby-pWw7I;JPDa989x>S?%|?zzH~;kR@5fXNxB4Wv3=5*?~_j`GpZ zRkWTzO=IxCT4V063B$7tg@-CiW> zJ^T>36$c--E-QEfyt@h&*p+DIASF>W06` zOjSK@(lf<3Yv+mbdP5Jx$J z!|Pz{=1(uS#*nZbH-$!!J1oZT-8!{zztQJ5qbzi>P;k6@?PAm{BbZqRU>a3Wwr7sX zgIS=c2aWleg}_5!G?WZSd>lpW!*o!|3$rCCNH=4kPLT>~E?3~b1Wu4Cev;)S5KZXW zSdUsL!Ad?_wWJrnrFYF3DdsxA@(7AOE|vqgXtN#1_AIbKsu@&9Xge@P!^&P^u4S=vVj0SS9TUGo0B zaAN|up+djF&{P4}4d_$oM@-l<7C7j9XGx-q7BBkv{_Lb(^_b_Sd)o~+Zk+D6Pb|?1 zk?trvOX;TbQ*4K?#Gk~zA3)u)**)UEwZ);_Q0VT}QyUmF;}n!RG|;Bef>|4bFeLD>cR8Ni*3!g}uw=>kdT;6{EX!KCo^Pzxn#0O<= zDhL#6@!h9!!z-x{-djG)xXmpE)?nYZ9he!HTtyi>XSDi$q}LdSYGz_TFE4asSeI}C z!0`peGa4#LZ&bwv@ZRqcar90YZKo{Fmq@4x28oY~Pr|!R(4Thf8fom!d$cOK`p~hn zql9C{j~SIfA*D~ngv5ZH2Ykix`5YY zco)arLPqrPjQBKci>T!|0vCW5DX@m7Bk8v0celh^^2UCg4LkT)JaZTI*87J;oics( zul{Yh2U?kojMWE68uLF~k`qrC$lP<4W62?{QF6AC{;&dL#=cW-@stpb&(jlt`Efwb zq`MF2Zp+)C^X1Of@^BBvryo|A_5GzAp;8XIt3-OsoD;mOu6*${F`oXNe?EQA6wlxR zaw4s)lML1HO5;)-1D@{?b++7xbtmG(WKiPvvXsZLi&j5Dk?Zhm9$DuKUL!YJ7yDd& zr^pH1;(JI0$}T*FkFXWVW#iGR{qdUNRi)ULT6Dj|f>Aw!s2T28D9lYU%%m(1OQ4BS zA5TYkFk?KQvPk4*UQ*{AoOH~TnX7tw3M7L!)tcXeI zGG~gH9P#3tLFOTtSWnG98=osXa-G^4<1TzN!0uc}|!xq1b zF@rF_)&^YJ&&^`|kTVzglbrboC>fjQmP_RB|HTJ5`ycW*uTNPWLU#XaRH6?J4wCu|tLV{$;e6@9s5uBIANvFSyh@O$5$ zPw^%%`f5vsT`Wcs{)kV7mHnQV{FDvdnJ+vNx2Qza$H83!xgJAnOLv$6tck;#FI#Sf z5~a4PU-!dWr(sJ}V>hQ661dxYcK(J>Y2Lu+iu zZEbZu`d<%xY^kH}7+I*DL#+6bBntf*$L;IQKZmftIC@TwB*yI%P>M`Len)Q~q39E* z>o2}Rwf0HHVoT|`yCY6q?2j^0oMmiY%9MM>`JD3V$ER!hOHn0_%;{CZp!SzqQe7D` zg7y#lz+4PIQCfE!Z2(Axb#?G((>9 zc36WX_w0mfv)h=a!)bMTMuxnRZnK8XDpT%CVax2u#{$#b+EW~`jv`w_y{ET}_P1Hf zf7A2UUe;B25<2*Mj4qeTPl5;)pxI<#J{{3gH*{9g z=O2yhBqOJgHNEuqgqn5dXP*pqCsq-4ADjHGBhDi;l8>^p_?>#5KLKh>Ny&npnA79-(T|j6lC*Y6Bc`UjIRRk-gNXCpb7{HmpXn5-n3qRQhKVL=*QO?i_ZV zOQc)r4BOJtz6h>ejV=lX*1)!RDtIX0a^|39)|6P&FR)q&pK?A7a8q>Rzb0Dm-hYYZ zQu3shot-`!7$VS%^`O&N44IEhWQmRvC7L69Q3VAo?Yu4Lwhdsb#4-Kmy#;LQ^ewEf zQjGo0VoSpj>FP;_o(F{k1stcR&AUCf=&Q zXW}jYzrw`(zq>8Jc47Xl|NsBBYuNrL*N_dkha0kjrVYz)c?zq04UrHlS3u(HZ9_Cp zSlFaE0Jp$roq+3~+xMQujkwf&78;H`3B0v)UQ!*$T#48~#JRv_wLMD~sgJ%0ud&6t zUj`t&y5FM)DBqDl50s?>fR8hnJqeFt9N3|69|7!X3i?krgVMmS2B@mYyTJD?mSgPz zm{~;Ts(%n01N=yUSUpIZKXPF3!$3WQE+D2ijAjMR5-2Bu0yt(sP?VkM$=aVO$>inp?hD*7K`am-vtAe6%wvxgqsLp;;Jc~%kI^~PJtI|;M` zCK%(}mV3SR+UM+z#0adSMd^k!M{@*qE80fWo4a9UQ?Z}7=f@i=HjjrJVkUvUBXPDk zoF)niJqGsy>!}X?X&d>Kg0D1E-HR2ycnej8+7LI<@i~K~*PwiH__Bnx`Anyc?jzeM z7~>h8kKg}cQCqu~^E?`ue(A*kmQb`Kc-+DW20My%DF@hR0#Wm`vYxSm>IrRIv5_$r z{za)LGs(kfl4;tyFfTI``K8bfc^D-ak?-lgy^1p*c~N z^(Fk|F2H_rUkvG0>K+=$DJsA9PZEE74Tut%s~m%?zx7W_07DvM_pw@4j%I8DLw4kA zL3{Yos}d!l@6w{t8>hRMgVa`9BXO+nQZ%Le6NM-`xOp~rV!J1$mzsAbN|R>rJNPS zC<;ytfDq#xHMK%_z_V^9yR5Ojv?;*{qK4i$peAkJvVP>V_b0OUMy;~^)3FP^p!1*h zEeMRAjuVzktR0N-d_eo618T{ZIl))8fwKqtpwY7~@f56ysdzO|!EK5-{}h#5uj+Q& z?S{sC#?1VTBt>Muap*_WZiac(rKG)^tbh27hD+1MB3sV;D0Oxs*g@|rkC*b7!WP*Z zIvuaJ?c-|ijm4#1wD^j#6d!g)UWN}K-D_Dp^d78G&*hm^d|H?!AoL^tH~+H7VI$)L zv9p>}1~F-L=q4ogL6yBG`wv;nV$7$fZIQ#vuo#lO&8vG;?ZJ8`DM-jqZY<~v`mFo5S>n`y?`HcLFh7o zSMGBQ`8|#+MhU;m8fR7<9Qh*oOF*3Gwz_$k7*izSu>y7{MWd<=(t5i^S?Nwx)CV8# zi<)~6lkdkRt>#4haY~gR;l7sFI<{x_yef9_QJ7)j}*d!g!>&=K&ghB3K=4y)uzeA*CXwPq5JFA`0_I>2ti z&Y^xH)WJczI9M8)SPI9FXKpb=;sOH)75%R~;nTX09HgNRP&dY2J3bBUN{*z9SJcIu0b=XVVG# z4r(1a`P!pJh4;BT7expiWzEEa;w{(|oIC$cjRqA3ktPqLN|z}pqy%yFNm`PAnFM;l zan`X5xC92os?dQO&sKN|^QqXE3oJ_<$^LAvL20(^0*+)Fo-3MWHxV|@8zb_?BtAWp zwsKleGdsM-*t#2(iW(gF@n$eQ6b#X%3blLrO}2yi4(TWU)FPkMJix%;E1EUk95Os2 z>-=bp>UUql(D>=QJ4K1qB}y*Q)s$k9scGw+VsfAH@zzda>ic`J8Z$XWGd4^WN537@i^ahZ z?tYUFtR-?g<53e!%=%I{-)i~8r%`wbf(?*;d_rz_gVzdn`WA?iRfL`P1UfYdSG#g_ zls2R)#tB~Yz~xf0LKG+e@5jI3`ZGR7qmH63wEyoovKL@%}0p_YQ)Cwy67b|yiM zM4ktgvSD%!Bil~z_eCU`J6_G+tY|;UQo^4;aRm_=wU2k7CQ{YPWsj~YN7!G}uKzba z|NAs}|I9S{Pq*0pGwpakAB&V_5ul4O4bbd9`8jh%@3*kuiZUmQ^`ngA;GfDkRzVrZ z+yp4&fX@QUCDKk{fDb?#luHJ!3T&X6i0ld&mjF=quyfWMc@PI%SX|Z^^+K>{%w4+b zyd+%xFO}LqU&lY#hOPk(+<&u0^q(>eNe8%nn_7VHbOuOLN_O9|v({WyY~4E_!gLWLUTVbq1JE`>tBW)4s#7!&xO~BkfkH)jj_FqSsNJfg zWrvGYS>Z(3jllttmE7yb_yH5ui8n1iYIR8o*d4I*s(uv15#Pm0#o%Q?kq2jN3s++_ zdP6#}Dnn714&ShlzdLC=g9$*&0ClI~9Vs;p2v5>6AvU4SC~b;5F0KPKy44RL*yoGA zirhV*uSls}i*msU`JR}zqyY+nYfNJU-gnrbJwE`y)J!Dy1tVO8k$% zO2k~XrQM4X9Y4Q6W3Ne?EHN5~?S@_=85k+9vH7+c6ZS(-+%5EOZXacP0I{Kz5d1M7 zuTjN}@JS$A$FOLTSMH`B2eBYx9C=pP6cWgg%Ay!u^;Wu3cuFfkKwjm^^fsZHI84wh z<42auE@HXw3; zG0Epr?{jq%OcmM(oxd8Sy0e7=2mdOm0Wiqv#Fot?Na?BI_)Jhd&uN^Zkf zsG|83vrDW4UnV_`CaL!vwRh*Yn1+v@lp!wddx?*+jd|OnO`i1t5;kVJ#0|e!U-QrI z|E-6+B8C-0Il!nwa4I0NcQopHpg~_;wrnMa(}$NWCqh9#_-ANN+$!1*lWCT@nr=IhLsSRvWAo#Jhqr`G&<@ z_oI9z%;f8)w7lgJmatMfstJ@tzw`yv0L}m4>!WLI2B#OOWD0gk0oxt|VbB>jXdqDk z&3UXqbQ3m>SQbT%4g#r4S!%dCblTr^jZGEH&oXCa-9k+ehViZ86u0?g&?6GSx&mQI zYcP~u^?s0e0EtnKHMYH&1`pA{oeV;@04ty>n1aZ6s4AR`ByA6D@d1hy0~;DCie_$y zQsrTkMx;fv1}#1sg~!yavAsrR3P7o=ls|M|u=btBEN`SFql$UG90(6%YtX4sBB@`S z@_F@sQmvvcbcL_@~B-A)`x|W0G%{sx9|LuLt{N_R#xP)->d5{)CBZAl1 zPJvb76l`gMWI2WV>BbX(`xZLKRra7V_PR1?3>AbRlS<cKqd+DXPH-91AKDk73tuR z$X3-6BKDbG$@zOn0~v!4VLY=c;^~;+63DRovV{6+%CjPtMi;HK%h*0~OR81$qY2Pg zJQBpJBhG{-zBPBZQ8Emh!M>q5@Z6E)FeXuiYiIzhxu|;UmjsiDvLI;s;iME3?uW%4bkfjth3O8$c%&3Aw+ogr`CAKPVDGkz z&%ncBd1i&g8Qr&fEfV!pSJZBDFB0FAAL0&oe$C1XG4$Vx`7U++M)O-g^?NA9o$ii( z&~x%3CoudJcd3F&6ohreEqZ7mmn%E|rog}j(EsugVP~=8g?51YL9NBD}E77-8Nx(aR!Da~nJ;An!U562!7v&Nc&hX0dj^ zs{WqbcRny0nHEV`Ms{4aX~GSyvF+;voY>1oBFv>`5%v_N&VfSU1wHmtB z(r>)>ZU%8iIJ30~c40E3BRXoeDZ3_f;o_uCqqjLlCZRPCz++GTWqcQD!5Tn87INZ0 zkqjgDau-rx3#?nnR7!HHtTju4M0u+IDu+{;;RQ)gDHf5##V}fuVPs zmLFz$RvCqw;g-16I5^UhFzx6-!iR)u{%n}<%h_q%vDp5A)iGMK-+@76KiW4e<1WMgG*-&gh%4*G+};tLqcEm+_h+Y$+6>;bxg10ah(J0%=61wtva z2*xAONqU0_T_S@Qq-aP43rMgGWL zgp<)5z%5-xt>44nUhVpO_%U|Z$I|+o{_|3|J|EWS!@7M~w-3MdZPx9>x_wx;4}bIZ zf0_@9MDc}q`Ub{DyxgCkQmOV3VOB&oDtXVqX{eQ-=HUmW;j2f;HFr&dPwfhhNJGcR znjNmFl}PW)0|{mZQHWe*GjOsr1jj>FRfE-p%kIMp0P@5D&=%nDC7Vpq*9z8XIkc-{ z>wyg~+g|pejVHxPAq&!OC6_aMI(wTKnP+e%I_}rz0W{)%Easo7TSEPi-1|?5MtyA-ff@ft&+ns>3`rlWyy zrc1xeg?n%8luh}yQW53Mt(#8OjS3ns!u?gHP3^0#3qx6vkn_yhD3!b^F zc!E5^;mFz{@0YG>9PC!gGX2h#?C)+jdL!O0r*vyt8s;!P3iD+ZT@LT*)FU=l8%u^B zUAm7KwrA^I4e!77Rr$F%S-Y}Ky2;Bt6_++yxX~)GbYd#NEqPO4m}kxhbk1Sv@&6LIB>RN+JIe#Nw{pgCavIGJ~XK)>0x3;=h!5SOqOX3ceza%3H zh@^}-qpI#OHi8NfP(E=a2S(}FSc%)mLQ0kjL8q@c0MUX-7;ziw1zCc3v6FEIUM@)! zz%Ck(tg(fHLYAL?$$AJt9`q(ux&oHD(O`|O{0K{ zZVpsR+4A~LuD{9k;krI2*B2hJ^sle3>n7wUZ(!X{u6sP|Zr!@?y6&L=Dn$CDb`r}m z=SLJAR@-BCupUZt{Sj`tEFkxi?0tusDzXw=CU1gxe=}T9@f1gaod&;FV*X)tqVjB2 z0IBQJ(y-OH9p5F8mfE0cNp19BO}ayW=+^x=fIZ+I+2yVQ%FWKiCIL8WQe;i)%ecyd}ebulA7ug12?f4z~oGO43!df&!c zFuNnvTZo)!=L~DGDD$?RgKnopbXFJHn4NthxsBk2E$(k-aW*ET&WM%sZlZ)PC zH1@D+vMdeZ|IXn=Chu$pwhxawMqn^XA+Lr@p4^^Xs_-QB;_K+DLD`nBoO4At&$^1Y z$FQnW^-c%9GHr$J27N!~>U#_$ehnb-PC&U;`Q1#0ue~_=a(IWiSe@$1b#5OB@#7ks zvr%HRuHL&lU+(0Dg&aMZy*(&1MhTLEs=`%cPwLNd0nLelh$~8Q zk;HB+r~E())SfVkxNl8H8)#7{X~>TvFVRI8$){aR;J zqF7ZJW%7Zo9(Ic=7qY2Lf`UK^sz12fil@~Dnz`M7s2ezzNyZMj;s_jGm+}KmZfgSuWYz zTT*4^m1G?6$#1`&!4q#L7ZMo*OAdwV`(AIo)n@nB+3BQOw&Eh%=c5dkD?imK!VsWV zXE4_=T~qKFF%9a}>!Ef*^6`_Qm@lAT)ObIbnO8UbfBfTE1626`CF}N&=2P}6#5*xMy*Af^UqjQDhcvQL zXDG$xVVvV`*QZt@rpA%U6<+6^KSairLMPof3n?T7vAkhgpncS2MB{dm%7`V@8AzjO ze89v%*?C%F3l=6}pPUzKe^B0e3SsP}LVBgxA7T8x>z!q|ndY*PA7~%d#06#X-3X2@ zDdVH%;Kr0Kt4bo=GQ>5o)VO$Hp%+po4*vX#b)0YPNeuCA(@%TPqpC@1rJr(h!mWHX zvz7|>9IC4w$BW110C|_|WjqqZVpL zl{%V8@vjMrQw-_he%s3Euii=DZ#%0#UJ?1MTF{_-lk_&YNX4ejT<;{0(PgFvMVxN2 zuiK)z`qCM z&_057Ro3AQM}L+?h{q?}nU6U~9rMMfi_yN0IA58W)WC1dZj_|PR2OqFq2$<*29B_q zq4!ZgrSEA8`|1n8`FT>?BFd1%Z_aFbtH`0Uflw=ha2cUyM&s+TNMfYv{>%IC6iHbn z1hozQ($FiK@W<#n839b-sExX!sz5De4FbbUh0p=4UI%E70$9O8C0vKs*j$jQGpajL zfU#xBvV0k+58pxEjWK@oB5k_GUD4~PrI5e46KfZo7lrp_Sukv8p|w0U^8cOBR)vtbpLcajP3T71lBILm9^DuPg?*WcFmeES$9ntvJk9ID*u9!GbN71WwiL;*wOJ89NFF?B|luD(b zDj<~Cg4pG)SfV6_V+pXx|LycLsLE}_7%CCffC@!vpzeY~$v92GvmT`S*E<9KmE7RM z(E$wbzSgg0{kg1%RATOtK!U{fLPF zYvM)}%!i?4#~ovP)W9rNgkk%}*{Ulsgd5Sc##WPQ>o~}HHSpDb4TcEILjuX>mcLM?A)xdsYMbE+=nSaQb3-YT zxM}CRSWnBgg5o>57HmI?$(h+<`}L>om78+Jkx5jX!UBEHfdp#+$Z*%9&#&anHvW_U z|5g6%pCXj|cg8uh=Ja(wwP%gy6`9t7JYQUUqWWF45HYpZw4At5{%xPXYDpS+sSGdd z?4wv3cxt+bFl)(!Jhkxbhmk9oFy-SM&z3(03&g69cs!J@)i6g0nswvNAFLGaC}>T? zj7W0JpLD+{wJ6ySzm9Dkqe8zBYrS&w?IgpL*4W-$D_Ax)8@X5(w%GpxlwDr-q;;tC z65TwOl4bSY<{2kWPPYO+&?Ce>L+*|Tn9Z- zFKz?A-=^@Ge+0FO{EhRD1P7$HYiw0<0X*M>@v`Og$_FkxK|HIl3yC>IZ8*5xwcyw_=TN ziuq6DG%-&aC=ZjxzeSfqt66D-c>WLWJ_e5HVt|@OU^QBsnG;pchHs0N-tLVVn3mrE ztYS;1QDZ)`GR9`IrKP}IL%k#nJnYw1GG3WnjxWLjf|wi>DY~86l%{YfX*Vn%*E)A& zN8tkW%YF?NKcthc>FX1fIl@LM08FF~92&aJWv$(9ah#AhT?$VM3tlg3mU&kdTz9u*&bO~322nDO4!#0?LNm7z`2i03Wu{1kBTZKQww0MkbBVTv zaB@OlaIiaahE3y~#(5wVIy0rOV+sv1NEOwrwP^%7=gW`(=$ypVp?bqv8m2I2tdroHCAq{M$aAtM4W*BjtFJ)OCYgE zqmZBXOqEEpCQENHY)9aX@FOe%kWw+DQbm7N=KhEDUjGe`-vEjd`2ZAswq5lFpf|0- zOf1x37f+coH5Pf98@)^b2RFL`O=FEMkq?W;f+OoHT*=TnqiV-TL|6)goyChnEGiU5 zLCB;vwh~yZx?TmV47M7ugbCJI{(oZ>{w93%Ut<_BK$vVW|8wBXwe!0r3Bp(x4Sh4x zRs1GbDVfuu-}=?P7WO=t+uzEeEMZuN&zr<+A8(j=iV`a#_hYeZbJi|nWvKIWg3Em{n15;$>3yd~7wekwqV#y+GG(tk%#XzA|0 zd7j<{sY&q(TCS~=MADSwww|e`&csTFx`}x0d0x%AqWPl4X!&nL!SQ0!RCnoRo^0L< z3Z&)%C0zTQTP#Ndj3=zBrT80b=$n(4e4nzEjL^stC*sa~yPd_fBGvEtfR+Ek_&-|s zw+Fc$`W#Yodj8rpr?xu#m7c^gR4^2j9{75p?zi?fSv@-A`1-1|b-K-2J zcIrj-%uB0$6h2ZJLM-)_T0k`EIGG(GMm-xY=F9Dns7|XAs#V~~6^wXsq9wYZQb%X$OQT<^>mLpf^9gvPqFOwze-tg)RYO!Tp$vleE_Tc55xe8yeG zZ6r>t8iLO$3}XUT#iGBI6)qpqhk7#oRc4maL(y~EgJ_|wGaoM>D0vAx$mLzg@t|!`OpsZ1#XSl|Tgi$c{!&s?qb9Hl}4oI8_M8I6)Bx1EC>&Hg7?)3(nH` z3ZzS5w3Eh6B&H7f(~ed*OJ=i29;LGuO%>K=NP)pqy0m!pkJ_1ki2`&y4sA(a2K~nuB~UOVIP!(h!ZhqTCUAIe5T!%mV2H7> zqfRcrc=kcbzDzIF0%QWg7d{z~%iQJ#(rpnA$*7_=HrhU;tj?D|@0k7UFoXlqg|ZW2 zwTCc_dJUMEa>xFa(lY3OxTwJISt<5dl4QW#pTq{LCRjXEb*hOiWoJmjzK-vwA-1Om zonenvQ#;CLFfdZ*QweZfUESx;0#!(nPiN;4FDbF*jtyzbSA>u&;tQswo=}U*cNd6} zDv=uVuyV@xpgebL*~qHinT@&oBKHl4=ZOND&pXbKf zS?PhOBG4{XjzC?Rs3EyRPUyfKG(BfX8lBDtXO?Q5b~kS8%Wi6GXWH#Oez*fSA|W#i zmW+&FcNYG}<$AhB!2E^TUXs!ZiYQrK7iW60?YtLqwj2v}r5Lt3-!v-Hy>TsZ`-`!d zBBhUvdhVxbndT`YN7|u5XkOSI_&$n6&(g)fX}+7Wg9*31N<*HKWfzb1M>f7n*_YN} zU$V=VcIS@9zMES(I?9|W>2w)}DkXYnp>)p;iKoV+Lyz#264h0dKtnNa{58k zhnJF`1!*N;n!PAK;wWJJD!E?ojOFr4!&r02w#fTO)a9wj(lQIN>nTxNAUwS0lQv@F z@5q9xK@>Z>Hsb=tgJe?k$Ucil;b~x}3Vc8mm)?jBW=JzXGBymiYa@1jWkz~2PTbKv z=`k^Wy?k;rQc};yVC)K~r~Lj^?(-Msi=~%XM~JOr4GYIxN6||f8cU#F4Fzw%FMH{l z=iVp-do|v81RnMnDeI+1aXekGRlf7HU{>bDpw@Q)N$*Dz8R#k5dP0+cotOH8 z8Rb1j-}WOqf-wMBH6Z54v{TuC-a~WPxyDu`c?~5{CDRt-#Bd#8b)R zhSHVe-q~B_3|N zzup&i{q~#fTJf(|WS97B5Zw;tXzmNq!EE_$`;y9mZVS}aBg~hV%*pF7!nr+nEG;Ko zU+jDCb-|2Je2F0?;Z?9_*pIj{<;^0EcT)B85hMFVa93)+PTP)~WQu1VH~wK8O4Tq^ zkH%X-%)nc1-UiNtuEuxgRhq%VG{g=>=7wxhMbBZ9y`o@#d>`MlG!{yvM&7nWSwiUU z$TY^`gt7+8OFEQNUzwuac!aa@^W>JAWuehyrCu*Htvpf!9@x3a?>rsW%|Z3H9nF;& z%s0)rh+9yAao@@%PE;D|F<2u5xH4uTv%RkkuSR-IDpb ze{@OsCV4g&qJ}W0I1TeZ^)eEOKHtXYqMA`Z!yHr*r&ezY3*c>@Qa#ZrV|`6qJW8kZ zGwXe4^N@T_mUBi95*A0<`y|_aFlDN&7YKe;vmzqf%ceLct=lQFgXo^6+Bm))JJKUn zG}UGRfQ8TYzB>fz4X%=lztwrDTimtYbB*nh7)@D@+dRX5mVvKx3g4O+BGGqz2_v}( z!~zmZEUQpLEMZF65jU3&`wG>2COVlfrp&S}GP-PZ-8=NR@pPsxVQ$A->6o3In0Vcg z#~t8Rkp7vLcrHjfMqK~tuLsHhYg=(cu(mka>Xdh?;X5^G`X(AzH#z=*>|~bRzG}gX zSZ=iDyDPzMk2P-6Mp`~uS#1nD9&zb;e8l+zqxAP2%{NbbX%g?ebl;Y?adc!Klp$Vt z&FJd4hn#1gzMd9Mcn-X#_Zx%687<^?Fc&Ecs&F0{TN?DT0x z$oX?dJeCOctE~k}62{&3vxk>%+lOorA`M_#1a#H!+Rz5x#G5cP4PLA9r*B%t-SW0# zzZNu>j~1lZ-h5q=Vzw>T<8c=3^-504+p{2Q+4ER5;iHSGlXml=8wCp0qTLk-;?VGH zHe21BVQ)sTJ8b#2EmO^&-Lntk4N!|4=_#_!Ogk}X8ywXw=pxf|`^)X1vE}f+Zce73 z&lH&s#u-j;{6HoPx3^twY6WKP`QIWUz1?hI2d{2yzUjRh?sYt+?sf_Gde@^K9Cdqc z5(Qu7F&fGh?RM0q_UttwWMU6mod2|9(4SuX5`Hh^3K8?ha64`F*W+h}`c7i<3OV1V*4> zY72}hP{3eP3RxF-VAELV%0z+-pD^VHp~OtUNBN?$?JvfETI?5i(90N>AdFi+LeyYs zmPN9_(lBXfmS?8(+ik?hA#4EjJcEQs5o}Se#Qq3L} zGyMIim}%q^$PY^8?_ss{a~tUlIqd3lbJY)<7g=0m+eID>KRNA}zucm0 z_d^|T(|uWKLR<#Rm^u5+!!-9YPD5jZun046CKOKL80=Gvz;8CKU`SeyssiAsK>5J$ zJ)d9g-;fnJV*9viphOyn&g-ng3uzpVesoS8;3iSUN%B*SgXNRr{Z?5SO@8Y2vd*rz zq?Ir7AzLAG{G+qTON`6RK!#4@!WW_?<8s1&bg{4^3W>-DePNJ|Zf~pP;MN)4`B!4y zIht0EbzqaY3h=gNdzzHBZ8|&}WAHpKd&MU`av(O;63ZQWarv=Wkrgv22`3bwCD0`n zT!v?$2T^A(X9d{qG@<0JijvIZN3d8g?GbEO@o+%E zj>h*UC|i?bH{{g$Q}4Zs8!xlqucE{lFZR1b^Y566-|3ps=vmax&u2Vhz(dor8!fAh z2c9^07#Ac~oOXeF64L4xuHWIV!0gtw?nWo-XYNfFRZ7|8oDifQJzJnOr+D{s{_EPA zCNj%qh6QEDBagdz+8Ep5zCtU;)ED~V$r$d0PsxGl5FefW@(mcyz-t&C?e?SIy+iok z=vqPJzyfh7;EE|2Kjg8(ZB#~aTA3tfH^b-1bYbKw!OMj;Ye*GW_sDk(;Dr>R^oN!p z+dp1bC*+QK56`izn1p7S4f?6#pkaP{mBQTEW!O85mQtrn6u%)I8`$Yu>VEI7tXoD) zA$z0eBUB@Ai}!xMuB@;4g4K-Ci{>{#?B`Jx7aBae9WhTHgBqZz(3UKZPqC7;FYS4g zHq+YseMQSVm*P{ru8_*Y)LO}B!i6)k85NjwtypWoXA*OLhvo3bASLfOc1BaDmwZul zkFTl`?;xKPYD)l9WJz+O#8^*VyjAM~n$QI)f^dtd`$$(tHV@ zGucU1H+39-Cq6JU>K4g>d^ab4)&@t$Xw14G&$Ieai%M7!IwuTgb`T0XpAF&A3GUdH zehyX(CZD-+WUS#$2)9WDXcYL)K_mQ{vzEqx0>fSVi4P^pQTC4csD3BcbQzN|uh0Nktqx`v0)^-eFC(`Pyg{1q($4q$otC zDJWP#L`q&%Kv0Q*5UEiSX#&z)$dV#Rjev;KBE1QbCejk=A|jnoqy?mxgc1TNoaM}( zIrGlmbH3SUzU!Rp+w=aBi)4{SLb9G;x$pbAf8Tl~NZR_F>XY1J7q~UlxLm>3Wo+qz zk0RY$cA__Hie93d(JRg^2FqR0p`S;$y4fr3pX>Na1-m#b80;D5YI^hd@|~xh6tw&`Sb@3 zfwG&N2Y3(N-(_%^|A-ogV!LrcT=kXH26v3Fc3IxTilC3`^SvzsYFCz|g0wRqk5@g%k*i4CqI-gR z49%)-5EV%$71bQx{BCZ=9NZSBpE``@w9QOTz6sKLnYhe_Y5Bp&%R_DB(HIwxTxy==9` z?WRaCsgM!%Pg)DTpIu$3+r2-jho0+iOWp(z8#Zy~rPG3MwRZ|hA(1P`*TKn)<%`)rZ*&_s(6j<4<3+h_gIM_@YE>gUy3TU4;)>(u~)V7Xzm_Y{^(%i zm2g$J31{{>ZBvQ>xvUf-N;iq#&$gT^ohO=Kd9O02zYBTZgf#$^E^W*VI5KWmUHy%1 zk;o=ZiLPzAMz*t4dSd^+&aTwLEv?!J|no5yA)!-5LjPljm}c52xzww_8) zPqZzZyzqU3*GHoF)lE%r!7!NJSRkJ7@ok63-GY|;4&55Xj$B*N4kdtxQ%_mn7HmBE z*y&l_1+Xmh;mAUk#N!^Sz4xqKDKm1o+2$~7OzPHL59E zJx*Zf3NE!|V&!C9C$k{Erdos6PM&P6n9H`U`tYeTPQ|b- z#_bv0;GMg`ykaw{f5tS{Z?8a9k$0#z`Dfg`Nc-2}Ix(g5t@FRlUhpFZ%#W25KdT*? zGDXr}N`waPEw5|&DZG&*q*TJ6{d#fRID8dzfZ;^5o-~Q&;dYR9UG?v@in$nxoz$Oo z=nL5M=vcVkStmsa>XC5SB#LxIjoO5)bgv@4sXX9H>_M@2?swoWSFb&pJB53_0zBy$ zoNj)w2QR%<4X(Q%lpV;+g&H!j#6ga0skquvzdb+lZDd!-Jvv`spSG3poQ{20Vu4^* zhi!UdQBm5<5ROkd#*YKQ7K{V*iZ3VDFIhFRj*3Zle-&+NE6mDAja`r;la4*?k8iTQ|7!okTcXbnU?t) zU!>D;ip(%PTdQg)Nh$0^+xc(8ORSHV=!NU^vPHrC;!6@P#JJ?zWVFoyp%l~Lhuce2 zA9b{dOehJ|QTwnm+vu*i9(6MJj(`>!g& zg}O~~j%GL(&?59cIZG(qAD3h2KR(ZqlZ7s24FZP+!8w8lofL+BRI8cyc7bSdkwLRr zu_%x4mLc^rP1>16*t6ivsy;&Vh!GA>Z5E6-%itdO6c`REJ30ORjM0qb9^cKy+Y~4G zA(}7G0sAJIkoF#2mm@a8D%i99vQg(Q4t#wG=wr4P1PlEOkz6bg z17Fp}{F*jdZ0rq0i4bYRTcvb$#v>x?EQPz-LW`WF9I^ibhwT7&INWfEfL^bjEteI- zolrJm*90zkvUc}clb_=v6&O_~q|EO>X-q=o&A@b-FPSg#XDcuaaq8Kt@b!YH5M+VQ zk5i&Ys?V48n13OC?yacmD49(UUxVL}UByvnX>~qU`NOoE*DF^u%SWc~v!#%vMG@P? zIp;`}j`f5**XY7_(38MHEu9YI6WYJ|VRad7_#VbJN@?TJcYNcg(nq6c%gAc&O`zK1 z(-kjNM_N;4Zud_MxG^)Dp=dGzHt&>^f=u(3J0XLXpN-uv_U$4}S6 zEsFt#XeCxlKdEQZKmL>QfH^R=UzRFA31h?+8ZsL}`@a5yBLwd-@R?lLH#?9IGcU@c z3`v@cN<{X%oR^OSuBOZ=ToW*Gk)`XnNeB4_&30tDK&qnO^c(pJP!(hz3Qm^!v-c&U zg;2iIyg&%X^Mn(10C|HxB_FQ3mRFSJ@dECs(b^V^Um>djv6Agz`l+H4bM$44xYcp7 zv>USyt1VPAEp!4SI*%9Lfdo4-ESqedye5P%t?*i{?3#7R;nH10XkU`bzvJnf%n_gt zMR&rEoeh0aHtmq4v!6MAUoRMU^@}eq%fFU>8O2G<0ShSenp%o0R>~sDwF_ER#Ph>* zL!gj;fq|r%Oz;GYTnh8gFDz`_?Kt=I%fQIR4^yWX9cFoGwm>j!T?|QIliz{xni&^J z{V?ckjyh$cQwz5fJ?+p0Or)`RfleP_rt@EWRCx z@50tKq&NB-c2!3KlZo9{TqwyQKXf%ts9dFkMI1rwhDt;-$Q;zHK2pu7EY-u@ks-C8lPGF<5qsJP2zmDk0c?KPenGuFH zXLR;e8JLz`%@ge1u>a+na4g(3us(dcMu)Tm;il=fZiu2Oj`#I)_G0c_IVJrGhy9Nr z?((}JK0m1ke{?-x)2SoR7AN=*1G?q6_eDylzo<2x)-3#L0DsV{uX*Fo7~P-2mj7|U z%D;)h@85s~e@oc!+PCNaIxaGJ@*bqG@q+3CxWhQWd-u;Odkv90Koox1nA>t;o2Y zY_0g=a@d371Krvy-p3qOH6@Lcy2AH!h!^^wCs#KwcsVKfYZRXWo^H>$sBvuI-sX?Q zPkmH6?Y6%`()Y)SCV`UVM=lD~p_^5gOKZ}FzH+~+n5AD_XAfT^?4~vI92$C&OnS7m zMd#67*JTZSy=X7AsOoar$MOtAP&g>>YDSj62}r%k_9$QdXLZ?uyT;cv`FDMzh1Rje zD4J;{{ZG1WF;0k2YIlQ%@b@SQi-zO|OCG6E9;*2Y!!IkLr&O22o_;C(XXE`mBbJ1r z+K-&DVwxaI(oVZ;Zu1FRgH}yFLP5X0X)6*+eY3vWEe~JOb(qHTp$^iLCXRQ!pJSJR zHcCX*s>ni3S;6_4)JqB1_!@B=0uv7}8wcaryS77&CSTiVwebn{TD}HCumOzDHV3zZ z00IyoSIg^X)L>N&-FQce_Tsg1S!PGN^xz}4HF00S*OW{- zcG?uic8e)1OKCe>U>z3xjqbwMe@k>KnC5Pq%3WW!#vSh8fmri| zSDBqF?`iL}q?629u824@>Ba_9kop>1%a*!0&HD{wow0~&)%4XW0IS1oIiSqF^vEco z<{BgMlS7AO)$9_#RNdcMZzdC<# zj3Pf^8JIXwcmb@Nmv1wXFJXUoAe-DlPv5Yi;G$rwstL4*Agby&GnMw2KZ0aqjUiWW z0UeBua6Ancumf?%ZArmqSiVhW6L_y=^eVH zOgBc-hN#YL-jYi-;9tJ_4%@?BY zaav(XxQrTW`VsHY20SDPBn{tgoC>f@u#f!t+45HKQ=SJF%gJ}QGJU>g{^Y(_h)IW8 zV+H}8(w=lmPI$48o6DL^a^(H0RO72^Du~(Dlg(*|9%w@jFrG2r!0cqPkP+w}?{D#8 zH}q2nN5RkKS8=@2U(X^MmPJ$B8DFf0K`Y>eqB8~wN;O^&%zga3e`Ek&_Oo7m7|le zL||cJ@x09HX7q^>i~$x@Kg?z@9zQ_RPBoIHN+|~?Vb2wSU_9ShyX=0jV#@(&nCuG%z0D{+>5dnzM0Ttr#HZ{TF{hZPj4-!YXkut|{45NUY zF+oY17+d0Vwr1mxdKC9`;~S?oZZm&E-hn-<>ZA+RP*3IJX6iIa_MRBYs{GKSL`D!Z za|dDzKf<_5xsK(m;r{8UwAZy7o)oXcG5VWUHxJrj1sOG}l~{b*40-HxU|j^dIgMmV#~t@FaD$I z#xGa%8>c?3AV2gAez3pfGCF)|_NOpAK@fOSYf}%>Y-braCvMiMd>y;Qn{d=v66(!! zZHS<~19^-TXd@v*$(e&b8;tiBR9EXC_WX{PdZ4V~C(5I}!o zb)@btCUZ2X$!b?bo(V6yx2cl(ShQoaz&GCK#F2{gIX6=?;|6FRz<6@W@II##6z3JI z6PRsjOJJegk7vn0yETm=R^D^#`Osc*uUjlC%$MGbmtuZ`O(n^LRd>e2B72W4_DW@sW-@t_LtL%NzheSEz}L`i*Xl-Ex;|!k zGqy3Kh#S87xFyTSYjM|viUZSrrSU?pyj2BQZf6HtZ-5RmGo}vb^~?8OwB*pNi04LdxsL>FarZWDiUECv7*DzF#b-q@zihrCZ3 z^Q>*Yug3$St}9ZD_5W=s8IKkKdnj9A>@7e` zv>Ym2MtsO}fWF1b6n0a30G2A?T{8F?Q`aV9izxDv(0?e5p0B28fe%DOD=JW-6FR8oo_=pa8Nqgq7 zlTcmx3xvZv5C?fNxp+JAYdNn^{I>c4us_3 z+b4xRqoBoqTBFFW`Xky6Wyy^2a1QO~Xu|LAO|)`Y6T0fytcx4Eg;8XJ0^O$|lI3XOS+aU{+3H1?4KF8xsj1dOs9wdO4JGd{m#f^JV zQ2JHV5t{3af0~!dv*X7;`(1(?*wt^G?CsBN*Ghi%ZXM>tc!0|ZVZb~J&euKHfQGSw zm+!>-*-7M33kf&iByTz-Ed1=V!J#uKvTDst8*Y&Kuz|&gh17_e&Ru34C9mU}&K{m= zMwW~8_P@bE%60_FhwOy*b*Ye)Ec=KZdeHThOu=QbYCf%r6IqonH zrg2Gj1`2MYD$W{fun<8Cdn?_sb1eLr@Z8YpTgmaQL<2n?=`8=aabWn`L_$~wd?`#T zLEK}{O?lOAEj8KMH;OwD4%GE->ynZxvueA~Urr=kR$d9}z3XRRjHXyNze=b=>Ci<` zUVRMx!`{$FVZVqcm~-Pc>Q?LfRCbLLv}Bu6r^zaxS%>7EW{q!3$%F?>Z`mW%6!Udwa(2?vMEo#^y8hgxXs`5G2NEINfC=9 z$q!;C`Mo~wjU82z5}f1|IZ+)%pfeXXa_%zIRL|OwpMhLn!}ZqHme;N7&aZriQwfXx z97}3s-5wGqsOZthJ5`Yop`@NOZ&{L}g0*OIun^hrl%)#e?k$%2dEEqQF@q%*`s#=0 z$N6Qnzsk}q`)E^(k|1lp0PKK>iw=o%7lvd*DqoI6R6p{P4?xG+FwF$_b;pxE*Lz4k zi!UWyI5`nvtPUv0dps{qYWNmJaVp|1QmEtm;^iL=&-&K6TuJt8DraN*LNdMy#DJL0FruX7JYxvvdWVonm-)FtTC)xWi4tdYms_rRB4Fac6T&eI)D2pt^LV?` zCV7tMMRD`Wl~XBI0X<^n+eC6`o$=@{GWXkwqz9{J1!CASPdBh`@#`lX2Oh2Lw?M&Z zy)4<@7qMK*tw|D`Tg5l>Mu6%8#IlUSsQ`gjkaNmKKMYImSa7SJtakHpln8HDz+tm#%uH25$4O+U*N4M} z5+`)m7Ebd0@I_1$7HuYH$V!bSsu%MJ3C>Rk0NMfFhIeWmg^DZPYu5>`xHy=D!vtZN zz|zqN2WYPL%h024+!u0JB{pCy07h&E3;6los6CX{)rTCn31EE{LDV`ZJI+51ATQ1(J)pkcGa>F^yCBkqwB564^Xfg zm`bt7E=#jQ${rG4ga)&Y!en9hJGM?&jICgpf{N%~1c|T+K27oxoe3iO>C?EO@Q$OD zgg_Q3DH3S*03pHO_MQ;lE$BLIFiHjol8-oj)A6RAXsF;kDNbm}9z( z1!lH7q>ij~r)zPdHyXTL9#1t}4gU;Q8`)mL&F&_M1LGO6RWUNQA4JA1Xu7Me`p(ya z1h2}->Db$#B$#1mSSM+MaLS|9=sZ1K5cWd>0dppi{xNP%8L2WF0n+pBj-EjNf;v z&kiaC)Mr+F zDrO!08nkL}!^az|kj*i((8CNLEAnRj=9Q&M&PMjWndA@1YYWKe?Nl5^mRe4!>sgYp zbzgsSO3Y;6fk$if5?3o6m?oWJ!rzN4LNem)TOu*ISEJbea+noT0Ckxx9W+zfuP6|> zy5OYo*Cc7AtNL>0rh!abYL2+CsMK?CJ{FU#Rvw#O=|f!zRCX%qc?f1q=_Ul#+~4XwLr%Z;3<cR6(|vgVp# zNZHX8{qG`*Us?Bkp{5M=F2yKVJ*@05!cy$hm`&}A`};V~iM7Y|CNfj{%jzZxyZKdX zW?g7X6LVTpi3aeDBO)0t4*4*_#kl6Na+{{TJx$rD?6%{3{R@;=l(77B?*{v(4Jt zy~496;N>;Br*2~)s_a}hwud1&o;2sd#XX>Trt!#$(>%Cx zlnIGTWb#_CMCm7sOflGTf2PCAfi>t0JSY6n-1KxG(4987603GYbT3+)gk=LEf(cf6 zD6%$`b*tK!s@@R{c4ql~R3l`id>!_9m%&ZGh7-{mA(rVGxT%mtPn)IvV}@zu-jTX#|qpP6#)w^T^+?@TJ>!e z@LG`x2CG5HfBV+`qG7K_j{a}UZeF8@$*e5w$D#cfmm9W3NagZ@0QZLE$=m0RF#ao` z_XpyAmx*4mBt^pOPWyQULa07YtK;C*nC zk8qX>(zVMx?(2uu(S4ibA3n(j-KkPYpRMXVR)dIr95dMq4fLCi<6sX~cz2`QxL4qB zf%oOx1WnhDP%>^Ybs5@Jvy5fb67pyi4>~XEx-eZa8yI)lfl!9F3P@h{7+%!5-tTX5 zO+R&2tl`$M`gZ0#EA;Qw+ukPqI#;6v4FoH!5M4A`=PM=1561k0uPLwu~rdAB{!7gC202UX54|UG?M4L6BQ7F6J2Zl$OLe$@&@z+*rLaQ+T^4p)8DkBf8|2yRu<- zdL>vf*a1JBeb76vu8ke2ezJYjo5|8wAnzYtd4)DW{=X zu6o>Z-FI>McsWz0hK{Hmh+Wmyh=Vh% z?J)m~%z0)h*iJ7gHCYOj;!EF}$16k0Cqc)0k4Z&+8!@iIR`+cXrx;>s zkbz2MYSa?SdX`3~mo<*rUeLkSBZmb^BNdG+&_ev{(Z@+yhmWIwbG~$W^QjShuV26S zDJ7LetH-a-pcB!coq@)>JRJmAGp2CC1gDj)Qx(X2VB|J8+H_t7!@mUd>JH@NpDUhg zRFF!I4fVoMe_2eFLnkkSR*H4nv&h9HH_kD3g?p3AwvB^~$Fca=xHjAXF!?ye`7vsL z(e0nNX!Og+*Mej#PPokV66X)=9E+$F;$1NR^g{8H$Ajr)f#9SU<~xuQARf)>9UNsw zjoaN3yX=B-aA671!ToWhqEK=}y`bs$v@T2t4Eog zKg$&n+ZfRCmOoqtx9Km7k}0#6awDbu6v|je4r%MwV8=&Z@iJ+BYdtl&4r>lnAKEp4 z%FR3Y(hvS6^$50Tx0+3B))IP@p+&x{ewzGh=xR!q;lURAZ`&AMh9gasM8^(-&N3=? zcyYsH2c~Xfsp+l+ci3i&ZjI`2&`mehTZa=}2ewsMLIq907%cJE(uGZg*C3c{TR3pE zHLhpmqH%HGl6Y-#{G(|7mzr=de+R3BaF63rhaSb<^S8Hw_uWS$M$iq3%n4*UitcL- z)=$yIVX$z;3t2MGcgeZ86=M{nt^oVp29xN0WOy@SD5GUNq0yP9jXpA;di&k|UA$Ub zTm0CK;+dl5jcp{k(L0PGY3D}r!a8C53wA|qRqh% zN`LA8(IFu4J^7`D)(*%YmhV~E#Vt*(Ws0wUxSl_(W_uj+r#>x|G6Y*D% zcnwF!Ob6Z5N#IBK$-3i2<`A%X->ce9iW zi&pJ}(b)R$eis2>{C?0#;jZ@%#DE3{i+wBgBZjSIoRE{J#$YGoSRYk4Mn(xYSwa|I zSn!e@3hqHI7}Pl9DPzS6RJYtRLb&}jZ@hD78dYHRf|&(Oy>tL>5Z3JxGb;?fYL=EW zmla{8qLqbq1J}2@hl&bR$UCrq`Zvte&czP@g7b=Z zf#)ZTfZ8y+FfwM0eCO!t?-h<Z3;99 zc;o0ZYe<#~yc>Av=+s~mWEA5jaXoq2auT}MPta8LnXKjcj25jXmis*qm-P_K{P{xv z^PUa$_^mjy!(>M@HhfILs<^c8mZ{7|pRlR=r<_j^vEd`eFYe|^4CuvQE-TF2PMWZg zyDg|u-t(AF4&k6Y^k0agtkgr<8CRbtn3nOJo~0*6r^JV5*)Z-`Q7q%e4;J9KEVfL8 zbu)E}kJ5;$m$YvYo5o7uOJ_+~LGSFf`d_;MU}YW3`s}maYSl-yoIgEth0HtbGmO)- zOX17o?-jObL4F8{Z|?}^-oiM)@deQ4zQi`X1X;ReETDtBDfjvVRqWc+rz(Q5CS;}k z?7Z;4<3wJ{0VH?MbLSS&*!^Yh4{m?0D{e%Ar%g9)@=|U|ctwXckCnbt}J%nR3f-1UMi=ti!>zlM7HVm9>2Nv*lLpGt+mZ~8W0Z-IpZAamZm z*ZLhf{#Zty`|vYgM@5?QgtryAUQz?kpzu~;hwL;>@KHo=allz~ zxOyHi4Yzh_fZQN-2cpyNHBOCbh#Zw(F^A2^F}oP~y3x=L9o8(%Nb+gvC)qV;DwFn{ znKV%i=P+#x?X6g5W4$fYQvpnwg1z$MA4*>@TOK(dhYTJQDF@?b6jC&^$GzUVLR(`W z2#KW1Ad7k!Y6}JX4}M)qYmH3Fuo1kstMgU3gX`Mqo)2ov`Wbz2LvaC0 z#-c$r@krciC_|R&TsJDvgOZ)7K6~K1_zKVwGy}LiaJ73h=Mr+&a*GuWYRT z`nf3i7YxN3$}a_?6Sac(OBX@^1yGOoktG3pX!*lNec}#wQ>dvWHFlm`(IJE^o>K75 zc1Z3(@S123)+m&V?=HE_@@#3aPa@hE=+O>S(_Sq+MT$!}``!k5pnJNu;}iNQEq|;T z&DF!uctY==FvD&N;ohyUoF7+goKo`r5R~oDzE0O!P2TQ8pRD;9Fr`U-(sug(~_ZD8my=IhY!#T2`aq zkmT2m<*)wQS*ScPe$&Ak^2!=(7lOsZjXZoQg3i(+&S0{m#+(*P^2k%I<=G*o4SjppR4wM)H2QU&~N8sUC#Jb1zu6| zWl)#}s=Pkr1X2n%a{XJ=k@Rm`kS$a$*w0j0jK()I{Aukzlr{6tbqmRxl0_Q_R95av z=i@fHJgRhEBe##aht zCYOCe9cGs)LzxC-)9eTC|Mnh)X8Z@4@XcEL&8Chm=^-2){D0F5U3&A{4Q4!KP#=nE zKj~6Ht1tMaqv1?GN%Uuw4Okj^kp+ z*4|SH!rt=FDm3rX{Sz2e{I`~Y-d=EHUN}T{Rf%=r z;tu4*%6_I3h&pZmM8*)v`q(WVZ}@@-BdzE?l)~E{$vUb&%Zvx6!e(JQgmw4$btBXQ z9jxhE`8v^Y!0eptY<@GYu{{rUiV?vC-{+J%c82x+#^AoK5hR z)QOO@$bH|I})xYx~fnKrx^O-1y7OQE2Tw` zUc{xe{jg&B66cW5-76I=*on83R_cYIO<|k3Hl84un7Sg$nD+j)XJQ`ay0PQjRQ+#a zW&%^`6{Eiv&7>wVavjH>gga=trSvlI49oNq+In$N^rBu>oc)D|@5bX^=~qt@g2gy# zL2Oc`wqk}|n@XZ8vOK!9q+f~S?V365x0N^R$1w1fHjHTy^#(JGp-)=jAN59(#B!)_ zXO3%BDPWwMVv?4RAA5Esk$C~3edmh?`>U)gdPI>(nSud^0!2DNU7f)lx3p|=_Q9fR z$oROY91xBl{3R=yq6Ih&Nrzbv<~9TbUYYTk&*k_{b^-Symgb{+8m)RAv6#d@?nV?B zRbB{Z$kVv?f{{v4pzN{i?k~e1c1^aSJPJvD_b&D4*6Va9>ib^YKENS)&hU(BHyXf4!ek2&(+Qj|tT9e)vep$>Ct$7sby<=WHW7G7mj< zSQqiO`V^M-#z|{lb_)K&k}IxZC?IhJ=*prKo(V6*4Uyl}&yt~i^Qs;Dp{~foi}Y>S zH>On}En-OXfvlQwN7bTF%CE6R3<>5T(60>rM&=>d1unqPWm%*VL&Y(tX zhc(6|UszY%z%~HKFBRi>g z?5DFJXx4ztoA1@v1kFK`{zQiSB1Ha`+wV5L;fUeCAC+UtaTur~7;a$MVN+gEa_Km@ zml>qmOFC)@4`5qRX>bS)xJR_oVPhL4{r>*Q?FsCr=6uH_QUsXtkta8TX=Jmb^{?ti zyap4;5}~z`5CO0fRpAFzI&Rq~49X~r-0;-zy6x|;X)Z-t4ol{byG3gN_W@W;0wI=% zD|>1ja&{m#Kh%H7zdiQ0;frs^x!HPr2MFTbZXF3v=10gJ_(D`rQ)#^7UiCQV(ChtL z`;@n-Em;m-cg(aWl^THoI%zBjfBaG@Il6Ilu)rxSI7Yj4N|I5$>T3nxi# z1}{z0Fqwe$yvuIrW>;JF^b;U|@}0m!ukprtZXKxydVxWX;5IB!wpPjy6|%Ch()~o2 z%Q~cdRLSS<00-jExg!J>_l}UUa;3V3x7wj+N4qa*ci@L^O%A>--i$&o_BwMK_^)$q z<86Ff?cYI6hv6(;)txT#b0Q^5n=3IHFbv|}Fvc=R3SbdF-A7DIXM%L)( zc*E;WyEF5>E0NdXed0D= zZZp83U!A1{nWJ7!x&bX)Mn96~IK6`L#Vg%fRKBWxRfX6J?tcM3Bg2urD7Q-m6QW0# z`rjM!%v?`7e9qew+~KzEqG>I%r1PcU-4G)VJF`I;Xlqm&<68HJ-}s(H^4oPf7GkuS z@N)0gw>0)ck856ehf@2<&m09LZJ&K+bBuE#7V+Q;(LHz1Ifh-`1R@ z%e9dDv2-O?7gpG#+Jq(Xy1Y~}?32{UGa=TLQ%>wV#5|@R2o@8f!P7J|ZJ!35#H8M; z_~zP@aS`w)zL&|5JoP!}u2G}1U1!xj@20U_-k2cdQC26RBqV_H_()%w53ID@W9Hi~ ztj`w5wUfCME;QmGzwuLW?v+9^fBaioQVvat+CWng9ALnXnhiAH&>FFma&_y95G#Ze zItfoO5iwj!rd%#I@}tw=RjsfSGQNfAz8d{6=jMg{I$t(Hf7~JH&TWrb8o4FSrzodj zLI^f1gKRfxK2csxZh)=5!Ts^P+riGF##$W5!|BPVjD|8G?m7DAeKN`_rP(Qp0*}zQ z(B05dY|{?pC-@P8zs|b8h-USiIDW(Nz z8tH!5k?u@%e`V&7i)e$DBO5?_QPfrPgbo8q!vh3;r7hmp{QR}?6JrkX=J3qyS zzkB|+uFP$vPFv_H!#U)!|J7G=_o|cJY+jv?2{x<>2m+qKf{Ohy4V3cQ$Vb7B{joOE zYC^^Hy>#Y5>K9t&p&pNS#3QcEDw_Ec1m52eQ3xe!NR|}$fmZ`sl)gvzUH9Rt1g_>93GIFzKAorEI+CI= zms=nXwd^{>xRzyZC%8~`G$2rU(e(=N55y!c0VlzSfVVVlpzQKy9qJD(OrNN+Bz`4X4Zu9B~vXnSY1ENe?jG>)x|{E)vHQB124 z{7e|~n{+DJMGIcqo2myk7Fs8?V3kMnlx6m!6BvhR2PejYuD}M*2$_$3;dpyV*z0~V z^z>u4UB#BGxi8MWJK%BK{95eCFIC>)^72Gfmj_ZumYfhMS597J0zs5u1?SCq@tA_E zuMB%_AY__rG;}qex%Fs9736N?HEnbYf~w>SB^4i|&t4;q4cpY~bHTpbW>^Y%4qk#C zgPOMX)(FxqQ1{(?YHmfjI7p8^8In5Rab3nY7dNc+vBK?jB(y%Y_&V0*x9tP7+ha!j zZcM|BIQ3J^SP1)k@DXT8jmmQ2%HonIBH(^$en?hl6W!*fmi2!;8-EHIGV(fGw# z?p3>bTDs)*sSuL{2x2Nah<*w6sFb!P7%8V9ny6dNr!;gzRY=#rlU@bV$CkN;dB#s` ztKQyC&vIB77Lpaon}_gkpL%~AEFe3jwejuNGx zMt=hip-xZ^vOv|@i$vrA`rEx`&f+$L?~NQLXyqxMi|hR=YisNh5k7BqMaICupDhP` ziw~a&w0JMnf5gj^H@;&$pl>A2cdfhx+hb1!L%n+YnJL6)o@aoVIv>jPBT=$5Vs2lz zPZae9`*>Fs?&tv7tQ+6NEJTZW6G*Uzu|?;7_n1{=Oy%@RpzTHnLgoq^^4@)Yvzv#0 zyo!7fsw;PjmI3Bf#3VMvQn>O1B|C+QRavaxifK(&2+hCf`2gwArDt^hW9aojvgkvv^`Mh$s+@ORmR??*y8e~Fa5}c25y&K(Rmyn8s zlnFKuNIs{gx7Tys`1Ih9!HiWKswXo=U7jXOT345#NQF{wtC_UMbv)m0oWG0!_Z1hc zCy+d&au-+!2;J+sH4AU!uNGaM;p8#+ZOJkyK-0mXG|u4h-PzotyJ*fiSjpIu=EU%v zvjUta$wxHohnyoE_dc}0s)J?-{RlAGkgOdDh|7blnMZ)uUMEW7a!h!griTqliE$b; z3}Xd=*gLSe1EE#HCT^j@XRdPrU>^mpt0v;o_q{^={vdQI1MQt0!9vhq)!Aq(6OO)A zApBDHL7Tx3Z;zuRq<%$#7$nf8>2^2o+p@DBR;C1`t4kFQFn%)2S^F3_X*mTn&za9d zaB8Hd)?~Z7yZ+MX$zfJ;g7T0rVzy5pOD6j8w3f%|yq5S>S-LvS%$)38-!9(3I$1$K zyYV668^#r{q`RnW@GG^K>ucW1@^U4;e~)%y4M6v(OQCM|bLQCO6jBDe_U|(3Ms=fnN?dtINHStkgRF4YxotVz(jg?nMUrSx?T#Vw+oI<;@ zh6sDmk_^LsRTZ~!M`^m2ia0D0HJdS9qglbUl0(nUA3$;}lL(9kPRVYp#t;wpsR zh~g3I?n)ftIPj?i zK-X;d+^6x-^-xZI?_j*5g^OXzpMFI@Z?Lk+oXQoTYyC&Wl+nDd@f?ECCPJ?lJ=oCf zs}>U^!^hzvEUsC*2RS4XC3MB6L2mgT=2hvgX9J|#kIUT$7re?VEht~kX_};0T1$-# z_6TLZ$F%W)ka}@w*xB+{loh#Tyrx^Uen6(EdpB-&h4CcWEWd zER{U*S_oP|mX))anN_xF@l05pZ&ESUol*UB@J`6{mZw(H>gRI2-a>94G!Yd7QhLK}@AVP0#lZ24hy zX>0l)l5SV~=`{DlK~xn6}Ro zx(R$;yw+MD$#J!8uKjLMsTow;AlXJxST@_gWx|EsJZ>Jjl+~; zIXTRT5&`>CZT0)Vkjj7M_IKKcKU_}#`=XeCAf)-9<#PIOIP>7|!^>YDG7^TZX*ijD2^Z*!LIH@*r9Jb3BQKspu}V*UKJ;Gr^qKlRy_HaC$srul_R`7pH< zVWEs)-df(Ytnq6O;toTu=s80FuHog}c_7PM4LD4Z=0UWbcdaKKDX{4(ih|buG)Yml z5HQ6uE*{uGhO(5g*Q=tx&y_k2PLkE1rDoUUzdFpV>Gr6y9mRg~a~A@k#ONJOBk}iZh6Akzn6r+x$^xLb(WrHn;yO$_vK%a7asUM&i2obq{jN7>w#5u zFbsn%S(htuVFUde>+g>wHNci5`moJ-1yBdPikx@>il=EuYHzL+Jeq(+s>b16Nn8^oSKw(|36RMqLl&H)iAR+$1#jJe_1BT-gFMtKCVgn z{bPbPX#S-hx;2>!vabC*5I#nI_^Cq7sedIS{hONc|E`kxOQ->V?mimfU6f`kbu^|b zeCr^IBS@0Gw3Rpl>&m3#2iVj=N8ZINjl^{`e>ZsSSC`EysGbIO9Tzys;><~!IA$C# z{vJ6u+mSZ$ZPi=xo&6~8XavjT9Lq{CjJ8Yt@EK$a@d z6rfc1^0z+?WRXB~Mbzs(*Vauh)c~!~-?{z22KoR01Ni@08-xD?pZ7n`RfPCQ&-6D@ z=2!G#dj^jAI7tx=Qn)6#zh^9FnL`74Vm{IOd%P#LLbm)ho{l?DnF%N?r<=$*5*@!Z z^N`d)w^9m*uSVtGtf~_)dc%qE7I?dy8ngVj$`nqW9Z!-h;ohL+`I5=s69 za6+8#QfjrWoTHgEEmYNGgRT4H0%%ZoFg(XJu|O9QF_=NoEz8<}`Szb|NZdeK{O|k8 zf95gnf6v4?lJsH<GdkGCn#N3?7i$dz&grD^#KJm8K5M_ z380OjLmLK$&NZC+Zl~s(v8B@V{n)qj%8mIGgHTY)mQMrJB7YYVDa%&;saZN??(#Wm27LYt-g0|XhXU}Yb6AZrx= zS=b*h`VYREKR(_+E~Gz}%pVcyzl_^|1cyI@!ym!nkKph}aQJie`bP=+N3s3CzpYR+ z{I93b5G?U0(S3y79%K8EbH2zIjD@#>PmrbNa!@HXfmG!c`jRsR_6>wx2UY|B1$2q$-+7A7|3kaa){K|*-tXak z^lnd_;OSn#Z#jBKbT6tx!kO>Qj|ZXp$h1A|LaISXRTP{oWZ+S#>WbdR(DsSm2GqQ9 z6EhcH+XY4w9Oc{5IdkTsjXyrtBYfR~w6zg_P=;+MyY-oQW2r`T@*sh8l*6z}irz+p zS}h>Z`_+-4rn z)f?&dNow;JV;;1qSmUFs{ww6b|B2_y{MG2gq%>^0D*SS@$(#FVh0e@%*t?bAkEdMg zcg}$Qbwov07fhizm@LO-OhqY)jb{Z7UJ@HBV8$*&^|IPax zJOAFU%KwOMf#l`{s=P@Q4*BVno|T@hf7@hHVR0D~i$38E+JIC$w7K9%w#8DOZAjKezV5LMtNyBCfJ=B^YlLYI}M;Sg}!=0g3-VYQ%Wi|B+(>Z5HRnD|<% z+=R0o%2?E)51rH9d)Gxb?zs1NjXc@y#}{r}K;so1L3=6wM({fNr}Yk?XVEbAAR5Nr+55NG zn!mZ`nu{9aVoDj3RP0IDzi@0Yxx&o~cEnO3grh685(=tDkdCyZa8^Bio@WxTeDs^I z%JJ>~WK{iOJZ_7G`hf~ZY5ru>-Rv}5V9N518)kTl(^n;@&%X$lUlA-uk6x^W;H~$6 zn(V=N50i?ZNxD>3Wy4(%gmVi`PG=m|W8MO+a{RfQ;A=pPdbbM_f==!y!)R3K473@Sj{7o< zYCrwEe-!)}uxm&Gl;J)h&|V-1x3@Pi4xt*q_5w}b>VPX4>g)gLY|`oX&@>tiIuW#Y zD*w1gdspQ@TDf~yW$&u|vr%qut?aFpy|uEJnfLI@9*Wz;w%_*B;a)o2ONV>u@SjSD z?~q_>sqKEmgDnVwh9IH_yD9hkv2w$39*+Wd!?_Z?yE|T zb_cZcul|=Mt$>Uiyl2n*pR4oo=(=}Gd+jaz=H10m+dD=9ys>r{b65`NW>;R|cC^#; zN+)22seFqK&VqPh;`sVJU~XdR2uIlo`Icrr>+6@SFs;2w?th(4>K%;ZsvO(~-({&# zxq~|A5j|P`WlQ-9yJhq!N`}Tv;o!x`25SenuP{&~UFFREevx%NF)%{*NJ{SnG??4A zK+08PoNGcfCHpfri&$V)74Teo80f!%V;;P9lI&MLKxQSV$4WUsyxnM*dMNuhJ9tEuUkR zXfcvcHNm9nk(Qq;B%48X7}<0DiTs-_vzmXr!ZA2opbs@Ae5;d^h~AnTPqqRkLM}ts z5ATAa0T;6@9GL@71Gb^`ul=(te?O4#-;Ybv1kSueoB_0;3 zpt~SD@ZQ(kyL)?!^N(kJ@7Z(N3oZXuYS@dMK=#|qdwXdeKude5+6B#S0-dP)>C9~T z9ez4+VGnEkL2R12gzb@fuLu=7I^WkZ)9&IzDHptM^Ykw$a|m6@@9Nr-q-O1?=_L11 zI@n;U=$A4=6yQQoj!trpmz8Ch5u6(-ES#v-cK{_l3t={*SvzHudnk!NO#&7 z>+RHb=vyy9_r>Ls=@+D>!wg#G?kn&)kDnxSCZwYNoMG*#zm12vleTJcWTPa^tvTAdStIBF$J(o`O`c2QpE+Uzz z0g&MH1Mt#CYhYf1d(P6~1q5YjP`nHY16cYqPeBS2*P-+LDO{3J)sX!IPYfSz{& z-z!=F=~G1xVyLk*PyvY5ae=dacO+e-=(+_H@3JFcOYOXXnwGjwFPBJ*KNt#%(Vyt_ zyZiMFPx+Mxoez$Mm1$Kce#`}R!gb7LQ)6?^j|VaQd$(u(xDi*iTtgu; z5%?5V#?y^fz)-A!v#EV|&rR;XjB9_yHVbF9Fa7pHeMILuPhfzubyw-fyi3NWt~r&4 z)*dz8S2B(6MJ0(OKrWGkUdvP~Tzlr=zpm`Dl(eO6WlOAvb(3s4=q#RcZpq(B>cMmMDyc(+4sHB^|0%WpPQ{ZL^ zLYCH{2R&9d8=Ui;Fqa25j9+I$){&8%V=MD{`@s7`vPVZqf1w&0Ekf=SF(A;U6Xno!4&sckL&&7Ssk{ zh(#d4qSu#f@cuc<;RJ9~z$~0boWasnG5VI5Y}3=U5C#Mw5%$MiI}bK_7QSD+*TV8R#b*+T%sEa6CJU!eZYSN^w`z&%FqPv;9B7Qk)o z$j}ZafoYPu2Y@ODP>VPP{P1f2oA(yeJg83>d+~NJul&xZd--z@5B&{80+?|RC+y*bJ)E$&U-{G4bPp%&;e`JgPB1W= z%=X(CC8dOF*W3l+b8EA3BMu*O;#sHTc0q1I3}*1^fux5kG!x$?DzuTWW5}*Z^?8-v zt*?s%UVYLomgfNCV6>$Nb<&)OVw-W)YwXc8wP4t+bFi0bvr|;z#y%&3PjDl_+uX44 z8yi9O;4LNvfI9gghkQTLPK)5Dnygm4G6-v$1ptRlcR@;_cK2N1|66MCH$C{#=C$PC zO$4$469IiAglblICFC)A7t~-PmEo(SdAre5G9{TszjyLSrk2q}2%RNPVm7QvVkEx@ z_G|oT?eQz>CDLc1(bPaM7v>!7%=j_o3l#GKS|wz8HDrFK8r@TF7Fv-X)TH7CT~po{ z;uBeYTVM6{xmYUmou}d0kD4dLjM?Q}p zyk(XcRO}-JEZlAec>4&fCw6uGFPvY0Ej#MldsHwZ2%qH0pwW*4Yj?YXT=zfp$M~Q6 zY5ZUR`8%lcakaaDgqTd}z%mFwS(g^L3`yT&okgVm>316KCvNQ1E+`m?BoQpsi|hYo zAc_5#Kr-@g0!iPV`||HV@z4BZJZ5A4{OwBdy$;b!1a#org{S?M`otI%Cswtm#HaV; ze3|Lv2D8zVMwYlK*m&$WkNCPF_#5J&&7ChP;q7J5(AqRuM74Nk_)O!3v`{iklc7t3 ztOS*}dU)$MB1GwWWM1B34{V)KcAz-KDZUKqN}bJq7Cly;zmiCmfyJ)5E>v?duY(}FPm>^l1M7v6Wjfq>dLsw*nm&DRx^Y%y$2-^-z^qV z{4}>?P+xnsTUqPqp}Mon)#UzKmP~ivtxbDZN81w#qpcU`5Tn4u+8*F$hY8snu z*^1LQb0SJ% z&db|H&JE|136x%wM7kHy?U^k6w$eZ`BLDDzIJIKSoxn*86s7@$e}9kR>Pj@MgSUH9_@dUJa2Z>3fm{WAsa2VWIJHSB^^?Y5(8&8Tuf9V43tUFX9B z#wwdY9O9NCLpx{&5zui=8gN(~4084VM#rN6H_!PTc2juHwx@q5eEmx1=f2r`#-D)8 zUr`L^ec4Vg9Pl|~M1I``9Ta^dxyb1cUESvZ+>sl0Taen%RJC2uiL<~RIg9P876k5N zJz)0@F^j;BApc(AL^q*#_pIOlslfB6;IuE37+XSPw|9Bf2~45DHKSOwi`;5D4^q)e ziq@$MKR2*h^MY#o@-DOc#Phy*azsA!`)X8;7(kdCG8gG;5B!#am+l1q7ZmRndaE$qN^hwI=oCyiyqR;sR&Buizv)B;m;? zK6bTDZg8#*Hm(1A&$@ef*~vOy%7{$MV4&gYkH1I^7?8+btwK-=lhyKE=kQzIbLM-uGhB2I(CjBNQPbb zA(XARbFCKbP))0npJX!4(iguC^3^iRUYEBxi z4{Jjm#7zOCZ-N-FdpFLYEPqF)Q;!+q+u#t#mYubA;Bc^o*_MC4QS6`4tNb&XmjCU~ z80e%W$%1@;f)xL|vhx4v|7{<`tFLc&y%u5n55DPM@ezEN!q6WGd={|R< z!`<3aPU`ZF>C?iuupJMX1jOX(JmM!Q;dtH)u2)I9&k^N!>EozR{@^Z1 zau{elY+r$@eZM+iciyAKb)427_OxK9ah!7l%c-QKSvy`d<=kDWwPMaKd9Hy?h~fR? zOSJ|^ZhEj^SiA=EeL%B5mcvWG@>oig@-sn>Ls@#}hli~*({ZHDMs}E5q|S_s$QKqY zFb>e819I-^cB*l0}n%v$G2koSy2&kj5Ai9;pQi%mV%Rd3Vmw~piE>R94uEJHiRht zrsVIp3ku&cDx+P&5k+=E%Hxc!-~ZCD+mksmZ9IS}9^`@Ha~VCj;xoa=MsH60%Gc4w z3J7)MW!T^=qc@aLg0|9Am$ou&d93{gAGT+T`@QSCETY?gUNV5{O+M0qJ65g4@`fh& z;(~M*OVVWm8!R2`C4Lg0uc_H@?4EZ0e~$TJ)MpoFpn*7#PKWFZJK| z^R;7igr9f~2%%>wOzI|#zpj-;RHj5v9>J7kC0KLpSU~$Exf;@b54Ei@)RtDdq@k6Y znig_i<8Rl<92Y{fN~KEP_Jud{gP)Y%%`ZtT7vbUViRkGnmv`lXh5B5fPEK-$#P_=~ zc*=Kf8&x^&SRhHyf92AmFQ(3AO5WBzi{3~3=vQW`GYCg(T2d9MRMW`>^2L7Rsf}k!bh5{TyHQ742 zJ9q0P=W3VIPvtaaI^FMfQ{R?I%K)y?zuSrJ%ikC=;`?Ip>qTT(G(5JO_K=KfT0!Sd z>ayH^*sb$o;v85?7|r&{D`y3+!x)5e>%OWLHa9WRDY{n+QeSY>5?%hJsb&e=VfK-? zP-sv{QO1S3RDLjEPDyX`jNIm49ARC|2LX+<}mL5HBvy7=J4AS_Cbn=J>#=>_KgwvEyfNyI; zC64ekl)1g}5}L~nj48Rc@bH5O=P*W5itys6RwO)-oymH3*pD(TC zJcJy=@l~I63nko2GzbGHo|5oIRd8+arq}7D%Nh@CXcB#S*H!?z(vWZ+obOp0<7dTT zuffs7oGsELX@2$n9foZZ#Y?|FewT9rR+-j6V2zF4P<~vf)rIVj->1#$nV!^L{oe8{ zfehcz=%{6;2H=m>aMqWh5b^(S~SQs@IduI~OWxA9^Xp4Sv zZR?S9eRJvqi3P;E(5FlV{wL>Md15;NF6h10pOM$U`F+2Glq12LDZ%4p;QcAs42;-7 zhP7BSky(O>v~xMa63e$44^}(Ziq+B8***P37^aOH-9{)|aWV{+*k1WgtAfOTHS*;I zNRf5uzG+CjvqbORyDAal6pZM(Vv;PPRv@P3Owa|8LGtoT-<6gUPno3;WFqaMUxVlg z-^%zC%;TcgZubJdj8*G|Pez!qEs=Oa(hZWi{`vd~=bXt!N#5hI6j zF)~#bt9CX@_Gj~RWbqFB$Mm%<+w=aS8o-OmkXN?(Ch~IyJb=+=_#{n(i7j;{~)X zD2;WyVUm;fGLZym%MxmVq2iFE0E2fJ$4Nh&yN*4GntFk&Z;vBjH^B)iP9YN5#dNau zp7-p(i#Pu?%mHE*Q$+BoSjTd(=_)`{_(*rkz33Xm4$eZG))cZI4qy$^s>28;>#gT^ zq%7pl%vP>@raFsBD9HeF)p(rtTOh z$lJKcVCpEDmcF9;SVdU6E zPSwhppQ;7$(o@T}m*w*#C7zF$6TP#)Pg^wV1rcbuXORvNF=9=_biE+$Y-h!mI>*Ir&EKhDv3QB63um6PN%R-13 zqs_NID%q>#)9y-$A;fILOZ*by6o0PM8wE{KzUAa_A~%K^YN zftcTqj%jTMc5xio#NEa<0)w*1cdzb(kjOVcW}}-^C;lZtVPEZn9I#twp|cFHCb$m_ zU_r|IY0+T7e@5kDK51O&QX`EGAVXdw>)28bR$tc0L*7jVZ`sNr1L<-743BKUUU#jV zahNh!d!U-_tM9LDM#3HV2gV;h8_K%PO@R!dQ}S*>eym{XA#4=R76ky_dP^lzz)ZUR z)6aGtFRkD45oT`Fs$SDOL$;@W9$i@60{8$*IJM-12g}#~yzadCsE=IDvhBZKuJS~L z>uI&UH~!SBjD1Iyr0gf>GZG2j*n`E5b;#?7U!g}&|LkYk>R-Bpyukpz3TKIMJHG{W zW=58W?-`qCOZ!aTH}5Iyc{uiTzyY+XDM7mZLj)@=A@rU~Z?W}d?a++}l572xP&yV~ z?~a~yIM6a<;bQx0<50u2$G2ro(=85e6s%fXZ zpnrDNDmmoCP8M;$t7TaJkl&VRd|4BGZNfEM>Oy>Id3s1+ml5>d(=5$X+M%ay&NrOs z=O>Jd*jj=*p7hJzkk?>h)e>wwu*8|~lxeROD6}_e%ai2$hR`!5sTt7-U@-;VC z(riT9ZI=Svr(wL?HAtEJvc@d&m<|tSVbpBSSsykzWvNJA+_{G6`&qz`;1i#QXpMn3 zn7-9{(eG0w$+_xO+~iz|j@(i6w?5y#m^u9T>SX9e{AmN@z!#Kg#1yjL9hMyqX1(0P zyEaagapb3+_cePInxxr*-dczqq8W8>5H%a3R3~>q)U6W)VN^A44387_@g^-2D2b_~ zV;-TNpY@MS)pdMkUM}^*=W=Q(c!zQsHkq~*qUsbgZjw1eeuVK$+xHNtPq6(568Nve zj3AA$H4MqD0nA0y^!%s}JSXmnobj^{niT_w8K2Q=U9}84#)p~QrcMpA_9d{756-+4^%R}#cVyP@eueAPNGn2YQ(UjC$s|m`;{IikT+kdEh|cjZ~9862p|e#bBfi2Y2j6 z)2%N4aOWzc9Wwmd+Hl@UC6^2{qDL_sh=BE@R*gwF9$sn3KZ2AD()Q7FXbzK5 z{)P$8U?;xP{n#HzJ>im2TG0(NM=>0)?vc!iWeI#qaWA_6$z#>H6P#Yu6-rx*EUCBt z*5Q_i(Af;vITUyu$LA4L_YivE*z=@E)()E;yC6}P>xfkyy!*>qg}EUy3yjUV=|i3Z}X1HM<|`+S~tAWxaVJ%-&WPAlWloE}M><&7JsN}BZU zRG7CNF7Dh*GVaf zcwt?U;gTwFk*88J`0C}j9sli+PvwsV#^2p9yr5rpZu5>&)M}U1`?KmTxvaDe`X?xY zUGr-{aRR+he)PWM>2&s@Qwm_oX4P+}U9oZNGOysrYh_OMyDom0YHB%I$ZaDi!rKpa zam!?rPwH|zjDNQEumIWH^wIc-?D4q<`lL~VBwNlm3byHv4aJ<=Ht|$wN4@7!&AIRR zE9;E)puWoXSK}q|;mYUmyrk^LV^4xkzHqxRCRmyAC@D&*+ruokZbXOI##R#inWxX? zuElHm-Sd(Nz^18dhP&dl;K5o}dKJwipK8`H#4{vUSLTWAEPu)IqSN{#q$EJmr)ygA z{Qc?B$jy$w-H>)(OP$55sJ>sJ8&+_%kS0+V-zF1~Anvbxv6%p`SZlu%P7O$gS@xVRai4PJqA&#cJKXXaxembtv)iK#UT2DvQk99u!&m|Di@Cg1 zXRHS^q2(*}F4o5JbYI*QFhsIq`TCqmyAie;^@MW>2@O4(9Z0)TYgEL~^nDj(wmF=Z z_HpQw(wcz*Th~lz<@uB|3T!lka?J62vJv&E~bwl?a_y#Yv1VBkuLC3eEX zn95zp(%KUpoFVY+Iaei#9Qp0^3&afj5sDXekbIY^U@^YSL~!f)&A|mlNo*?7;$DNT zr_%AF#>F<^oWCe7UemBn7KVhK|8^qGFA^SXI=<9PPM=(2fpxBAR9|%Nl3RR5(^ts% zT3I6Qm{3z>xMchoVn1tSi!ZTOAz{xWTYh=0+%>k4;dhiWDvc`KE>f2hPz zN&*%TGM_g+@<*Y**T*?zk0h|#_IvC16EOAQAv9jR0nMX2vGV4<_pR?n8mrACxfiF! zPOmo9AgfI6;u31n^z7{KGip-L3_ok$j$mwrn_l2Fu#=1F3w)b!Y*;M^@?J1aEA7n_ z*!}IdlBC(MC}BCGFd;mzl9Hy_9=>;|A`%?|fa&tSNY<9l?lG$pUB%o>5KRl3wZ4U#jnrwJ&NyyKo%U zyd*Wkcu=*@V{b>R69LBYXP=@4#In{F;br|{mF++pI$5pjSWfN@J+IRnIwwq47JdD;djDSJJwtjYMYElN`*q9~mTf9<+Fk zb`GH_^f0b?L^q5JSgc@m2U4A5zqVX*?9^I7bX1#=QTWXkx^1<}g!ozA%KrVus`o>PnbvR&~F~)t)0r)CCp(M9OIGob<6MEE8%f%{?8j zCf+tIKv|q2WEznah|Fa8gaXhoGzlj6*sG+T&^jUzsL`eQdnJbP{(OZh=36b4!ne?7;}^q0@04~y-4crw=en|^ob1Ma zr|9i^3y7@&YS0IdZY{E+iAOecX#d2cY>Q7Q|Hy9MbxJ%5M~P1OC1`(SO9}!{7|+J{ z&Kad;3go8v3jNg9jiHyY0d%`&$^HOA@qN{(iDdMfD>8wmu4SPfJtvR7cY-mzu7Taf zvLx4v2=PP&0m?;-n96D-fqCE#7_1E>kYY=af_W7PeC>gyQp6~v_!I}?3 z$}t8RQYe1f-EL)B;T6U&_XC=AxrnptqA4ZY4W#V)?-T7zwn&k1`6G{=(VHW@*+JugCAaf>>HFkjB4n@xmgKYsjuIIY7?7m1sd`7i|1}d)jpzU*2Y&t z4w@iYoXCma1rfuayYU%Zx~zR&`M`s#Y(bwM;%sU~Xc8p%1`8o+a6_NuvDB%k_7p9~ z<*}@CnZfKjA$NbI@PP15*3I)xXlR4jcJw0rra_~TlbacLU}{>y=E)tE+Dp?HprgD~ zI9>=VRR&)Ww(VB+DC+^#{e)DN)_L7SZAnBC#DGMe*x3@z+i3|!P$nPy3G^|>Pv|lK zQvKpM%FnkG=U)q58i;4mk*zd#Y9fLRt>a7R`X)5TJS($_bTeE^PwPiB29WzdlVuv_ zSEk+wrh=Adt0U3C!au|KioBT8-`Fq8BlB+g4G2|Uq>#N)Fwa=UDBORc>aCbmBe+d6 zmS{W9>1C&N9OLL`-7!-wMv3ipaDn>^^`!&5*!H7A2sX#}Z{6GTkq>D+^p7Q2W-7c% zK{2Lz%>c7h-|1n{P>y`l@L|P#owTf1sd)Y&&zUF4J{?#4LuXA4FD+cpYVozX_`@LG zR{%yhPzIC{(Qpy%#qH_gp$6xuy#3;@R@aTxXksOXy*Qip2q=3xf_RLgQ*u|UQDU=BqOAI!=?a{LNYNbWHK2x0kiNoD5QI%I6(D6qB#Fx{*?_0$%{|#A2 z31Ur>BIE);jtCgwOc2ec2P5Lh?G5N5juy?@>b|mFkWG&8mY;wWXl`)P@&GBlVGPpm zYF4hyiz)s9mpz~q*fM^?O>Rywfwl{C2yO1uLM>-0M)Wu8sQ)5Z94ZtiMS?5VC z&-Js#<-=7{ZJQ>>PI=j%A~v>SCu9OK#fF$*=?iiX0obYU{;00ocKuoZD6W`-MSbw$ z>2no6%mzD}Un^k!`bn55&J-Ujsdf5SFDya2SdC;xaU_B3whCm5ulAV@QkG+6$z?B& z^ILTm!ZrJ6;%VHQ&9x`Ut}S7K6;96~Kb}z@F@=x0c`?nyGfU*~Jqghj#%@X6A)oI% zNCBOHd9LuBCLbbsoIpX7LbYOu+~Y4VCWb_|ZVoi0!)BB`+6iSu&ivck=Ul&>exV%s z*}BJH{LJ|rtC85THwopj{Eu^?&tdlp-5;6xOVwyhj^#_Hhov7BZIPpvDJ)^g(Hj1C z6rq@C^Q5SEVb)~^XB7=GfwPD|QX9YPaxUfz*O;j`Q^=!d+rQD_Nj#jz2SH_V7A-jY)~xaFVDOuWue1u(%0rwM z?lE@g6g<@QW}GR3>7!`?hsUg=)7ATWopyFs3FhBNd3CsdY?&58zNaPp_>}o>f|WK! z|A<3~!0Lu#Qdmf0_B|TRtrv%#xLR7tZ=DW{AcXEGjYgJee_~yt^ZWzgKZ8*WB%Wq# zuyh6>&(A?YwO4)2!b?_;T7G{hK6U)&aA9Ppg(L7K;w|a#8CQIiOYL=T;-Yq1>y$N8 zFk^4GQ+hi$o)UlNv;*L{?Zb}cA+Zxv`u)?oOXM9>56sT<@j8eieQ~N_t;CDvc`&%U z2Klk>iaVE7r${T|L?5s1RV7(3!~4CRTR^|^0RSCGVt^>H0_?cvF80syFAeX4P6yrG z*4YK6TjHRoi8T1CEEP^)#Ly24(f834QB!Z{Fj}Z}M8BcKL}PtCIm5dK`5ZaQIAB6k zQV7#muBjbQT<8?Xu4B6tw+NJkPJADde?h+c!lXYm<_#*-`Ce?+`x1{acG1T>$EX9; z_{rfp;@P!~07CcAD1$i_Nuo9lAje*N+pP{GD#xOPr^=tc0xus9N0u5sOt<~=;!E{! zDdjJh|4>8l2XCyHunTgm4U8!j8UdmsbXB-~8$BhNToC$?-?3|J^1C3fT!#1-rt2zr zqM`GEuN5hXas3V;WVizt|BFKb%n!PH6+UUx=Gec3{wZ1k>Vd4Aw41&l{5E!Y}5R5M|JIAaf8uGW4YLUCFhUWQu&=_&Q zX22VTv7dT3T{i<=EHt5rGER&WG_rtGpQ@EMcFcbK8%+>Lv7?n;A!%OBb`MNQ*ufC8 z@xDlIG-QTv8Fi{-nzJp^3RAKoB z`A=5U3(<%b%`Piy&;i)wr`qQ&OzQULRFq-K)wq37F*@Q}{smaxliKH5$MQt7`D!(D zSxHgVk3xFx2cBawJoQdh1i`{rcICFVMB3Z-r+*RnP2>Ko{=;E^tB@lv1BxIzXVX;z zeQWy!phwzpCcOR^B}y#|y%sn14(R#cTIBU5!cG^ZrS%c5h7H*R4n%bm8 z1mlE<&F!f1`o0w@>9AM{rLxUe8*&Tdr1DxhRC^322}Qf(>;2ol=(V6SyT_w&V_Ncc z=xUc8oqb0dOCtkg&tMkMZr)FDygE0T7~<11DGJnB2r515qFshyV}OuLSYwwpuD~>1 zw4`ZsJS#g^S41C!_J(aVp>D1kn`l=(dUXgW8pF^&npD9FUD0`2vjslZv%{<9HUigJ zQmy6iXGRu~K91_LxCLPS{pH&Wg8- zD=qZDFM!7Zr!;{v>RZf&o6BtX4_1+6Ve>zw9<3;Gvh4rhbj~WoNF-T=z2mXBV66t^6qMmYpg8sFg0Up_SJ)Fgff z*H@EGIapPR3OZ(1PVW($qCI+-K#ivOCs?QvMsN-SZ5Nz(LFU|Dl@i_BxvSR}?<2o^ zVtg3Q@WWC*N)zvV^xR%8%e*>&YwDRN58(E-rqYwF$3|@OU9Bt#5=HTR z`4Zp3)fZcf8o$VB@J>LBTd}nB_GTa#z&Ar9(`&^qD-OkB(x$-ib(XO$H-#x~r$yPr z=fR9{?U_t+`X`4`n>rln?n}-$KCnnkLO&p)O~9FJhaCs8BZ2;wx@)_jDJwF3rw_Ny z*1J7zW(|?I_G1V^kDw5XIyGRzqs{nMQ|WKc!}Qz8_t@_Cm*8WxGgyzyeSLj51C4nC zkFUi!+*6TErPh&z>v89&B^B>=%1G!Oy@t`ofrkXO%K+5?J|;axx|N`+(6Nc#)Bg;O+&I<2aRrgWurH$ zubcv&6$O`xrJ~AP{Vdt**jUuFUOiIhH<05ls6-5l zolH=A6Zh22N#R8QH)yFlEn0IMMZC4hDl@LJ!8qtM3xtFXP`I_2auaO^+cWV=5hS_V z&Nw{ZZM+RgD@$@W9?05p@rVrQy^k5`s6F&JqHQPv- zNF+Tp=$<*2GuKeQj`jL?A6BOaHLpDg3QsNX1@V*KJoq4KOG>_(tTlzL_=q!O&Y79FBZV^Wp_c zusPn${F3lWv+Ey@U3w{rZwJ;=NnbGO7m12$Xrrfp!g2~oB|32hos~Ce5Zzeyjc;-o za>t*Nn+>!dJCrfUIT)dY62g8cXn%YyW4f^UQ$72~Wn_a6#Zxc+ZO(pYcnGRAJ|M69 z4B_V~N(1ZjA$>dhEK ztxF*0v4b$iEm#mF9=xrxI0L6jplP8HQJ_|bp|c9jX5f5Wj8GL@qKF+9pYtWXacj*& zX)C(X-&q(#=`Icq+VrQly5tUpfOyG4OEUJ)G5fFN`U8xqcybjf<(?$*eAw;8M42P| zQ^Ud$nniu>a zLwTfTL+Ovg*Elm?;7jVD-tiT2a^9aT^mY_7Ewj?I>pa0)M}<1D(o{KDXK&J`PI;9; zWNZ;tiK(kbyL>9|*SD(NfXmlS1UNDb<}zeS;*`(`LIb?`NsjNQ4<3GBoxH#!;#s~3 zO5$WcnkIusbC_7zMxP4Omv?SHEb?`|fQ5isnw^A)xuuzGuh)jv5#IBuvzoSX>q`E5B;SuiDtFW=%(xjQFNun$k?X-FP((TSf z^32t*QG)3o;)##^agkpXUrl}Wl}9<@>==^pHkOxil5N~Ir58rLKZ`7`LMZ~JuTi1x zjxyT=$-vx@uW9je#`a?8j_~HCiv}$LI$alG8P53^^L>u~x8eoB<~x0ZhE{=sJPVSq zlD~r!#!ETBdXIB2$RkgZmb{<5sqjs8jueW>Q{lARMpInM`mU__^9gB`AF3?YYT!oW zBfrhJ+=g|8g^YVJZ}1*y{WSE0hiO$oH}m!Nc%6AZn`7UzDnDPd;@0d>AJk8Z|LErTvAtQPF|Vxyg= zC*Y>$i{abzuDhU&F4)Rxz&(X1z7wAXy9Nh1A>m@UM% z#D!eLJvFwDUw$tP-JHSsTq#cv(tUlkc1|zE1iN;dsp77>ej|;Uzi$~2-xsTCM|4?(6jGkysd*f`r_oPi4 z_b#ZVU!l@6dt#>{wp%$L5Hn_l%chL?;ch42&-um@LNlQU7Zlhd&)CIL2%Aod&uL=u zygPT1vy)~I3AS_Ae$g*rP!1^K1a#7=9y*`83+mMCC$jkIK_`U@w=_o>SvUbGLS{<} zsPFXyYKOFR>=5`M5WY*S8M7O5S>hAH0`LfRH22U?_tY`7;pBqvEt5Y{Kzk|2Mk4B< z&wW`$r{fH!2D5x|iSr|N&f}#j%Li;|tsLVUKrMv34E=gr-%47&&Ndvu<9b}OjI%z@Vw7@Bx0G7do z>;fN|_C~V@ni+e^FS;!*{c6b-DoR84VgwnY=F(-M&r){Ub{gNMZ!nf@dJL7%Z1`13 z^4p)y^Oc*>9TYYoDKvRI9G_pqkBue;fKbQPmA?7&*^K;?%--Y)k$Z#o)K@RIC1B@pURybckeTNWW!rF#{?Y>Q0DQR9(1$??ctl)bvy%eA45U5A z_JD`nEzug($H;;mwmOs+_rP}lEdLTXwcVAh?)teRVx&}ehx{YG#(i=4oFR1H3M(7~#yJcDSumrkh=ybg+nz z0stY zkS!}Kt$MARW<@WB2)9hY`D|;?ceTWoIv-HGjdvP?9`|0*{`}WrU z4?FZh3r(v*U&$E`R)_Hv&sSe1q}+yw>b6}*5f!MEiQYhn}u5S3v5jUjrboWTsOFmCC-ffC1eL7FY_+Z?ZJZRiC6 zDD|iMF-CVm>pre+It0x-@Bj=LfU9yi0-*(N1_s0@pWH9Mk)C>Ac?uAUVn}x!s8n)) zA8-c|{qr#N1@RJ6$_Kg6KSKP#p-iUu{uPr5F2+1K2*b5OR3ku}<3GEdx`q*$K*yb2 z!i{1F(SY`t!v#nYwMUD(9)x**6n3^QFnQ+=sV$)rNCyS55C#6qNq^uPsjDDYTNwGs z*Cox`eIb|Ynm40I+*mSw#CD$1&U_)WynI6Tjj_7mQfWc$ugQW zL$6kuY+`iu#0Os2wmvD~Wc|RydaPn@&EjJ+Z;!Mm=J~Lyy)CO*{k~IdOj?S=cevHh z>E~TUke6HV@kCWJJNruDgQusEFB7dJ#+!5J?5HFhvBpUuFPmkkg@(}mv>7F9oJ-{M z7$RCtjCElPxYUWO6wMWX|Ifne0P_DR%c0cu?Uv3`VN4=e+4z|SZVvUr=93TOdqj`|= zzgNiffDGWv3LV`!f3y(lSmn0|& zm7}mfxuLH5{+Wi%^pFMOL;!=&r%VJ@PU~EZ0eY@1z@gwh%jf?y)MMJ81os&yMAPCz zKvog5y_(Y|YjfN7zkKGHZ5AZ&hBOxlC9B&u$<2fYn5d6QO+O~V0<(bJ@Pw0g7}s3u zL`_bcuY_b-b$?im?mkU{8?3<1`v2i4fCG~q#Ow82lTGiZ3pz{dYi%s}mL0T~)D=iS zSDHUU{_UX}qdV|_n&vxz93yMVG)NbIaGpDen$9z<6((q0#k_ilBk5wdxE!Dp;O)zH zbWZG789#8bdl-^^{@A>tdQ8;m@^+BR6BUBP3ak{>;OzrEs;Mp*>g+B^upohIqJ1Yl zF5&`L$ZdLL9^K$xN%ynIRaI{_h0dK$pm{qCm*JD!mx<3#Yp9Tdwi?W3f@=W{A?kR? zQzcP^ZTV`U1NgSW*iaDF7a`{t53|+`M-TQ8N*3EtA-kaByfQB(&q(;Us$2U}IWL*} z9X1gg4U`Cy>v$;g&_aT`Khl080xa#sqNSIZ!<>p`r^oDq9_vzJ3P%V~fh_uC&j3zZ z;kQWPn8?^NNsjc8o$Z?@TU^`tXl-pt1+v?Ye8iU%F4tHRXesnG+@pN=+kJ`67>?DsyEDNr;@aa_G6s zT(7C~ma|qs{fWERe#!L6%3NSs(< zA6wkcd#&W{Dku(ZybnaYh#EYpN04D1 zn^P}+-(wxNnkU@j1v`Gm_N@Dzr(MuX(23;zoLn9sxxXVE=m+iSj&A3}zS@8iXdu$B z4`egSV{&G1t4Z|*v&|4ThjM!*&&GF?i|d+{XY?$ZF4-Gi*;SHwuSEo+t$vvwA6W}VbA>o8zD_ZBz(T3alqPC0DQm>TyU9y- z!LhDBYaQ)yIz|6%`VEA|upNiHp zBAmK5A&CJ~kFDr@nVDcJ^q8I1_dLzf+nQ06`71|tN68b+0SHr#If*ZHPydaygy2BR zC-J%=<3$d^S;J4rywLxrz3Yr>YTMQ*3Q85FOHolliV>wqOGE?&M5G8vF$YkpG(m`z zL_k3zML@b-P{pE%fJO(x$I?A~S}3tOf1d zjz1HdLm?WpfZ*L4zfG$J=FK%+9=r*~#faeXkwS#6v2U9mos;ngrl|h&!HG7|W=9jXZ6^>)*E=C>}mmiVnL2KSg!Xqnd-i?J-e_+b{E8Uau-q8YV5X zZPnlSYGz z&2<1wy}oM>y~CAe*{BYlL;8dO+?tb zHfVr#@+8W8;w6g3=!jl9?Y(Aa$E-q8Ai|1v8r>`&F1}CQ3@+e|;=_o1aQHg0(LTNx zV5kLjgLYa6&s5J>EZ1195GG?e+GwehD7LzhW*dIW1tS{EK{aFzQRK}q$Gn9x(iCDg zm!{@yb^mIgX1pQ%Aibvs;3dJ;JF{Uf;J!FL@n@Q?Ys0P9YkL>Ualt^nTstX}p!CIv z8)j_~ENe6V+gXFIu&)_Vt!7$hDZmzvi|%5M>$%^_^4rvR{SmJ}!<1(<{1+wY`)A_V3m5J{2zT?DXV=SH2yF zm+Q|+2MAvX5`7IHltB>8LE(d)5s0$wGMVASsT6Oar$HHDDp$pJK&gI6n{<;R*_2{U z?+`l%MN~hky-Z!ni?PMB+b{yE2 z%($p~5hA^kxc7MKIpp~W{;6v;Yn-}#Itlr4S<&rMih!WRDrHj;ptMVFm6PmVQzZ#$ zFO(kvi4TzQMKk^>`?i=HFDb2cKllrwk**d=oh-2vl4XhEdr;0B^1UVD7@uw`55Z|G zj_^=5Deq8ls$lBer^A{yzMl@IJ}>klX0hIPjw(@fP;_4G{E7nNku+5jYqrf|)47qB z^LM^8m7LqvD;A2CggA}3LLWlJR|=?CQI415p=y$9{ugJ!tE=xL^l8m_-A606hbNwZ zxU78mGK$6VCuIW?hIeoch>EJ^id@)d6VxOql-xJulEf2&b&gRar`8KHrLig|vb|k@ z@+}3IBco&B3x6^=tnI(s=AyquRH$?d92BHME2bY0I#p)GNy^s`Hg+%FoqU-PZ~eCU z8>Oi(fFIah?d9x8ue1w`KT7-LD9CkMz1?@Yt502IE!{wf-8B%qa+{*%hG2`wUBTDY z8!QzL)SE_QZqSbdrO1JxVk64=dK3nz`GK$j=};q^1e`{P z>L+<5a#%{=Bq3pddX@y#GD~W``2nw@Xp#Z6d82pWM3M#`+&_bGp`93%WS_NJH=Cks zjACJ?tK3=U%9EGI0nUhO*`_C%1;AR~jeRPYfdJ^$Q}<3vG&Mh3wBYXnx)T74#r-a- zLJYZf=q)+fCh-WOtvr$P9zqS&qfgMU55_O+GQ3=w(pvIWIjS4H;IYe|H5uG~? z06!%Z8CS%hs;ptIv%m%Z$AY+B766p1JQ`owY-*x+Po;QM#M-Zm=8sMW;25Z=B?&Si zmM0Pe#`#Gc$nhJJ4SdhjwT(^yLB=zx(7zq6#E0qn`fSkNyb{?F^oFV{qFdGm$66n6 zzn?1B+I}3hFT}`bB;U2Pi^*RHQ=3jkV1$49S>W)+b#7DbcRpL57P z^$9s!rM0vIEr{MW>Zxs-5NI+~n}CMNQsr`}pFIxCg`U0fjgmy)E)0mAwl7ATrWn4Q zSXkGVz>jT}O-;%EKwkcY{c^J$7!@j!h%4Jkp?1?MG3=y_x$7l_s_=1T<4ZOjq4{Nx z&=F&oI%Nh`s{>@{oF6N^m*pm3^dcSb8z~Sd<-QrNIrBr1To*9*+*G*ql*PzgF zRlMI8Z;iD6&PWEt4B0pC(Xt$C1X5 zbCKS?Wm=or#fgWy#7@7oLdn6#t^l(HM9~~xfb-g0T#W~TSj)TeAh4?KSS$Ji-;ct! z;8ESh5Bt$T@bLh;t2U4TodkINttHQwn>_9xw%Y*fXf~F{g=SX3`x*{5uwAS>V4p8^ zB4bg94P?=kgp?|hDF0D@ekxwd_m+ny?~W8R@cfdyJc9NTS|JC9=*!R#Y^a?%Ue~}M z=w}u_rrL87!u8FKKS%z{D@6U(-hluMfC*NW+~n9#75AJ{Q)u5&nq*p=Q`cM6(BSkN zE8%X|CU`dO!4o@cmM%?R3_z*gumocLyrMzOrsC-~K9FC2tj~%Sb%|C*=Y+yC8pm2{ zM8s}&t(|*%p-oyfY?tieRM8Z1fB!d0rC*W&H7`$uFoPN@Xyq168hr-N0gjkE~N5lfQpUrwqchW!MTzN!l;b9V!IXQxw~P0%;nQx zprM~suCG~5#BW`lg}tO$4$vsNHx8(ZBP1gFPxgP)sJnHLJE9tR_XA$9w<3&1(xaDZO&C>lDn-BziZ z_$6UNPVr&stg)1M6FnqjsN?C2;uEb#f6KRaE|i<9eCU3;o{6pY(lh(Kj5CD0 z3AvpTq@DdFEv~5^ z?EaXS1kJE=|nQ|n*O&Lb{Jyha20Ha9H!DEJ7l zY!E+^6}@=VEes(iHptdxqR!$ftjD(DRpN5gNCv>LCt_Fbph5MiLC2DgTjN=bc1TNt zJ>DrMO83mh2f<(=m%X;T19zJK5ze=x^?Esw07D@^fRp6*keC1JINT~_ys(Y zyXQcTPpRd$pG5i^R0>F7G+e5xKXBZdZe`U+!kh#;Gs($Z z%%@ZUk9L(lh*&o|G(Wa;R(eNy0@Xx2H&OL)MIDs&;8-PRr^byZVp*}~o9I-8yWpr% zM6QHM8)ugj-AdPV<^E!}McMS5?u{|(p_g<=R3UmJ1#7yr4iqJS<>(gYC6|=m8l;Wu z-mi1pJe_oI>h=UA9En2_<{&g5&lzZEY@-obP9A~#iRMGQ?#)r}cLI}qh-|Szz$f)D zz!fgF3NUxT{ZQoDuSM)Gf>X+o6u_w8^yG%Itdq~*jy?k~$n4s|P0cN*kLPnojtOMe z5-wA0jNd1l=&#a-U&0qm8;qe9U>W@KO-X#;;HA2g+TwAn0KzcIl{o>l69lTjvCU&E z?*(71tPDJ#*#-6@Okkj+4q}cfnH+0E+8(kucC6%~rx`?{NrMoPy%>}n7_roGCV1#hQ! zpw@MQAVd^kc{Jc^LH;BS7kpKVgLa8dYFC~U+i?8ju~E;p0Nk?0@^Ar)g}B$tECj+D zOTM$&jjSq*jyjXP6<_=i+w%x+)-MQEyHPU0JoN2s%OWB6TZtMkuG8c)wDe>lW-ILe;6hqRt>B>eBSl-kTXV+i z`rv?Ny3($1FIdL}i!`vb2oY^G5qhTHedYhv&bwFw;K*Px@sOL^kl#2aRfXy72 z`~M-S!05O37f$l8iGhE{@vnly!P>v->`~FJAo3vy0n|s~0@$zcn^8aW0kNG!-2YA5 z?_1k}L8vYueQ^S+Wj_C!14PpC6(_I9u4mGDQd#Linbdqgq`#fRKA77@QQ0Z#+iBse zsdt86hg<_R4it>Nih()#z@at%lpf2BMzr@1?m_LjqD%I5Wjd@&w0%WikS+s;HX9wR z6A=%CAOLki=`y$)jCr$z3+CXlZawzsqSX9e18vT#cgK?&0A(p5{k5y5SlcSEg3Muy zS7Slq`0KRAu@3t*dm?4@IejxAmA5Zt;OePW`vr_xk3yF5?K^Ra^A`%=N@e|2@)Al` zReAwyS#|HZ7#L^Z^2Vi->sDEn<@3!W=83$k{KE4^?`>`gPbs39U=B8dz*SOdb^C}# z{m4VbDb>^b&^Z_B62@C^q8ehG=z>-9A}rx$5o5@OFjFJp!DFpSie0|*>>QlH;- zE97T`yzj&q<>q`_P`wu>P!?h@S%d7(NNsG9=}`TAjjifwL3)5wrK^Fpft%Dr{*uR2 zqD%|s&Tn=BLOv|u!q*EaBk!#j56He9#m)`Zj)3>Ucuq$Tl-SOtp@Hsh7RAP9Z@N#P zT6I`R;Yxhde5|C(vYwc)oMQWk&N(e}TGbKVtv^-5Sy!u6BFjI2P;s91s@q1w-@sbf z%WfM@bhcfYi>g%-3VV7d&Yca;JQ7&0h!G$?ZsMaV`J;+^j-!};-vAmdYC!O3JA;0s z_?;;ne2{R6oV?}SMi5H_y4L}It3)X^Fgu99XT>u^ULWJ3xt(STwd*OOZQJLz;w{bx zvrnu&qCyTJTRV8UQjO4g;EEw>X(=;7<_cA0%lnk9-!hP=IDTwDVlHRC;JBN6TO;~D zy$X+QR9!kZdgG`Z9Zw`m^wb?hx|E+x7+!Z*8*d9CwsL#gSBtICEOBov7tDq4?kRud zz}#gX%{%;ZnAc?Bl#muDpyMgWCvkqosXk2@C$YYEr#oZ`H`Ilk^F@cz*?NwjwYGc> zF^61Hwk^nVu& zw3%*erY|V0Z|*p5VgX--1!P^<4`WM}6K9n>+NgvU_`FN~6G2V-e9^E2>8gk@ zDON$8C5Q{u`dZeaU1Zbq3L=nizF$=*HM?j?iXZg4P+BJPV_AhZz`Qxd5grKhxYnl1 zYC6!bhaU~UI^`q=PpsBfdlarKd@1yN>sQge+CDDserdMt_S!^6d*p@tF@^Yqh70J|_ZLQNm7WogSX%C>?UEH11Y|HHxiG20|&*xVO zCKZ$b>~3$16=#2Rz{{!B3vFGYYB7`8;}||spDlONHYmcgJ0|w-Sgl=@@~TjBNG-l| z30iuu&2@AaKD#=YmX$XUS)@x?wHx4S>oO<4UC&NyZ!J9Sm@*=I>udqx_~-6cZ39DX zvunsRtorwUyCWf8+{A&LQAF>B1=>Dr33iU+h9lC&68nt})%NG7oks0e-CqxO6 z>TiOmj%zU`bwl)OHOD=3f?J%swd+5GY4*t7I@sX)xal2Vy0`n@ zJ{|UlT+w}Ei5k2%ve!y`Wfg!}-u{u@SoDytjc+~?&_A_z1&eiWYTvG_)MLMQ+NVmY|yUp*l<3b@C(m-GvGinY}VDY|Y%S^UF=|dU{J`6<>FIspi(azUpug zl%B6(B6P9yN^$RYFp}?yzb)n~9LRpFn^p4i z-W!*y()(5BOA=z!*9VRu2aj^F@d&);ljBpC>a6W4r)eglkKE8F4y!sAHgT8}9@|}e zT*s6L3=u@2=utJ3>(iLtxt%eb8Q3k0C_Cmvx;!6LZ_uZrERs z>2Sxfwd;uH?@Z@*beXAw0Z0CzJ~qAct27x_AMZrhWv9xLJpiE9o1BRQ>?XYi!Vh6I zwfN_gI{y4<|1aXNKbK$qGv43YS`OZPP-VP-H?gzMI^)5S%8IJ;D6Kmo2mJUnc(45k zJN~18EKXZWFJ{RAKi`&Kk;{D|h;trYjfNCy|TR_h|Q$)fIm z+-Cg0I`02!i6^5y#J|**=%29^{}s=X^0Qz%^=I!DeQQpSAkZX=%sCvayUzCor~S(@ znfww_2QZxvQ9Bnzxbb%Lp?~Ogkg)V?BLtdE1BxsH!k5jwTcWki#@YaNsXOyjOGe)V zhNJseTt$C-PyfrSilNv4|I1f7l9WEx^sXKOYA66%)yxfhI*Gcpv8ZzE@vM?`@LNQc z?T?xUXo(jTkaO`e*f;OvIzleQenj!4;R9Q7_v*4R7BnMl?En@7QC0u>N@88XAvU_Y zx0bNvdAB9Rq=ptf1hMW(?;~I*MX+RmB_9tYUBL98gU$aw=>R@)bc$(PX1^ELGT zg11&UcU0E3oR9mOO8;ILpiR)9x&Xy7x&Zw-FZMtEj!ve_9~CZf00d+RGSF)l&!|TD z@|w}K(}g;>o~Vdb%Pi_qdSs?pm{2Rb%c=Drpo#k$%R7O4U z5&sWFi1DTW^iKf2K|OY{W|u=ZmXwWtPAnAOHm|l0bpCuW7@LPKua(J&X4MWeQZHpr z*gQ_z!L4BS&+ui%CV`T=kgjykbi7DlOm^{NG;Q`v4A5$8(P3SJA(r|-hxxyL08=o& zAkvyOzQ(oKfuVO@MvU4~qXFttBO0$cM*+xr$AI#t77K(ntb*iuI$u)tIZSEiUCxdu zapTGr0a@vfAagHqd_I*2FudQHsF9L@l&LON8eqpDx_KI-3^&`p_M%W zOmh$ATRj38MwiC?ooW3zihBKT&tpsDMZCQarR{D4>dIottXp3|bcDg0(=I^06Wbt8 zS1Kc2;(X#C4y;zQ>Fa1&JbiuG?U%J`2bxjxpdyOWa$0IsqSJiw-Wins$v zNX-#3TmO;U6JJaN=-d~I0i84nx+dfPexC=7=f{|zjP>`skz&}Fzvmtp_LJfN82<}rrh0}Kbbm7fyJUa!rWYL- z0C@WZ_?hc#i(a{EBg%vW*a04Z6<`ET+B^E+)Vg%(ciw;Af7Ac#vOV~_>_DH)KCQo! z|6>5Rle51g$owS`-NDJ<(Hn#-0DxY}@upt@05B|r>0-eFH$j*YEG59>2Nn>7#rI*? zKVX%8*x?U&znnaN7Unu2&JD0i*zN88TmgV{zg(h0j;_y7p^yMrA41D>cDB$9!x6dPp0~Po72ch5jLRH<|4fpZEGC(f` zxLFv0F!(Lh&e8sN*#tn?!`c7Jei?q3)5-a~E(n8sLGL;HSpK8@(4?CI+WYkP@pEr_ z{ciXBH0zzc^!L+`g77#0pv!-Liw^)B8<-BtgdVsVaA{vYVA<$-JnW73+ca1P`V+tf zKnFMvhysCNbO3w-55WD~a`iI!^v4@Lz#i}eTmff5<{v44eqsIljSqOe41@tIfDee{ z|JQfgzu!6oL0~%Nuhc)jl>uCSzYY5Rg(olz=HLw&176^BAP75y-~5@I6)+8^xc~M1 zXYP(*S-rrx|L%WE6NUYm#;D79p3#6&k5P(VkY1Bsjs607JqJd4dKLP!f0pZ?_zbfQ zJq(i!0}K-kKauYD{QkuFlOAvpm;lT8hg6(FiTw+A^v6M}iuBs_8X$d;Ha#D`ARt43 z2Fyu|9s%a92&VpPSp)y#_@A=;hum3Cv(T~dvz%j*{Pz-Yp5Rpci^{&FM1gmeq}8H z!1^zE61m{`3Fg(A?-=M8^anlq{Tl-~qrkwPatIIs!~jW8Y;4bh0cm&RlDZoo06UYJH0>wZXPz}@rO+Y*F8Tbwi03%>qnE{pn zJg^OrAP@)>gadLAatH#49D_(hm706>2r{kd$raMk2OQ%AoMQ2QBL3f?bmF^Z@7+oY? z0^M`EY`S-JRdh{spXmnZCg_&vwx9r%4ax^S0+oTPKy{#|P#dTV)DMb+#zIq}+0YVb zEm+fi&mI4xH1GYJYYy=$YZEvXaoDs48t}fBO@Q91fw$8_pUN}Fy3a2V@zknFn(n0 zVVq*zVq#)C#3aq6!DPy0&lJG)fawL(J0>hsFVhSYk(rGd&MeQY$85vw!+ei9mAR0) zk-3+7j+w;5#d3^AmBp0Bi6xXJktLUuuFgsWnEEQG`>xM0Jv2Y#d z(&4h_LUE;W)o=}QZE+vumg6?%_U4Y`F68dwozs%R)RtYC?`eu|j1+!-we(pEzuO`1awP!(R?>35y8p3*QueA>1OoEW#&p zPQ*jxsYs*90{kFc6YdUw3dh11MR`TFM7>0xi?)j5j~qT?a3t_Z){$>VNJk}(S{}W3 z6mxV`j8#lU%th>}ShE;j94>xQJY2j;eE1m4F_mL($DSSQIJR?K;<)wk=;PJL=OqLr z3?;%OiX?tXa!4X1Z%JlL_DeBJsYrQBrAvK3L3cv&gxiU<6W!8u(n`|q(&^GYCmBwj zJ?V2Y`{a-eyNs4hkW7&b?iAmt3#aa#sy?+UD=upz`$V=wmU>$8wAbm}(<5>RUluwiISKv_4Q$QnxdLFG+%2@AdVpH5!r~zb7JS5&gGq((~{Ei(0Zq}s(o79PrFikM@LmBLg(Xo z=z0C~vFE?(9?-SaeW^R9cT~?+uSgHCuc&`p|Dyqefw938gCRpyJapm(>!K?09 zKiDwYSlQ&+Y}xACrr6G0Q@r-z+TeAG>mk>>>;&z+>>7|Tq&>3y2Ez^O8-+K>_NMmP z_FE2y4zC>WjyjId9G9HVIX!h+aMo~6a-MV1a7l8RchzuBc3p5oxTU%+yX&~WbSHQi zcw~C)c$#|Vd(ynDy)fR)-Z#8!d=B_{`n39r_=foQ+?2lg;O4|F)mtgI@P5XAdHxXp z>;4}Cct9olOQ2+6R3I)$Gbk;H7;F_>9>N{s6Y?eWMCilNxiG!3H@E3;JKk;%KN5a7 z92cP#k%fYw98k?@F?1w)=8pcIw|80Zdfxqd@ASQ-dz<&I@7G2ON8X8?iqelNdcgU> z@4-;CT6AU%eT;j|w}%Q3Up%D7I>ml^box>1BXXQ$+^2ZC_~-Go1eb(wiAsqXj~O5P zJRW?4c=9%hJ1Hz_D*0k^)l<=@u}?QskSSfMXHqktu|5lYHu3zz^A9h?Up#(6e(Cmd zAWb_B^Xl-chp%?hoziZ9T&vuUH!5!m z@&xl9eCu{ z8xk8?8zUQO*f8wY$D1FQn_Qcwnvu;TEmvClTg_U(wHdc{w(GUGbZB*8J2g7%yHvYs zKB;`F{;c%5>Wku+%CCxFE59jytLj$n{_y?m_u3xyp2l88Z&Tm-zV?2@{?7v@13iP5 zgFl9@4Nd%T`Z530>nCA2XqYr|=NH4T*io+0)Um^3+2c~j@i^X_*}uf{Cxd_-op1q+r^osn@gnS2P@nwuT~{jE7s1feZgPB&k+0w zwDq`+LmT;OlEq9FepfUx_4stJ)s+Ykl-rdj}? z_|a&jbQ+BU>M_ut0PykFA2#7W_P!1J`piDm%NKHI|NgIm_7UJ_X1c)a0EHX{=(r(J zZV0UvfP-ac02LE3{D};qgVHlFGBLBTvVjPdTmT&e3ZR;`;k!(qGKM$CM7>jNqzSG#mnrR+&6jo zZwm^`$}1|Xsz21!Hn+64wRd!Oed_J&9~d0^@pBkAIW;{qJ2$_uNLb(4+}hqD?(XgL z1p%Ocu=N*b|HcB$iKlntMU;=erva(nlr@`8 z^qtHKk$6J$w{>G_blJ5l%Nt_9kZaF6j`5cbO~WF?W!|BBGWk|lOP()(IO&YyRMH+% zVNhKzY1zI`h+1og$FpQDc6_R;b+L_7Is20*N|jN=wlYkixg+IGt$L@!%(pb+RZ_XG z;$ef>rJg&8bk+L4K21*|4IpXTgu@a|JkwxRX*56q`b8>H)-_vsr)U;J0~la^FjLMs z>KKd$94dt4&3uWQ_-;xATE+7TB?;k*iOxJoKBQSZc^gq-l6Et=v%yY_d@3h9c2lY^ zub`kyB*Lgt*u+4gkp@IZyJI&`Ws;c@WS5GQh>5I>!@(mFS)s;DbEO?K9Jlq!a;{Cj zdki#S`YH_|&iIWTp+r)7J@}GJl3_ne1i~@3t{o=r*DN+JEnCZ==C_eihzi%1@mNAk z#qCnzxex=C<_tGeu$-2Ai16JR8(Gp#^fXqPR&Ct8B z5LfCLc2mG}0sX_8YA6m*q$Z>m24ojV1@cH64&I!PsL|!IKJ)&ccO5@(>I}yPejg zjyajv=GH3TX4!_AhI$(vh%%(BKgibzktp71Ba&F>6{ZbiO25m(3}z?7b68W826Rad z(ty!NG(fKyWrb=%tY1As143!Qrp3UJ-qY)64PT`cc3H$(U)~;$m4Zsm`(KN`QrQ`@ z)@DYOZXRkpmWnv*)tpq*9(>yFgJxsO(2(%d8XHV7#GKUPf|OXX*CYG14|4R0^$ouA zjyS*O4NGfeMh&DQ1nMtPh3^}agrwTn&S3F%0qaAg#>#EpaC^MiwmI^ChE#H^jwQ)u zBR#|FOL$2`csq6HO+=>6;LPlfl6+S8=e1QHGre1brAn;7i)DO{fyWGtk#`60XwCbd zW7Fli1o<6-`<=6(2=~SKUmR(@e-EhtT>~oSKGp{lQ%VfR5%IYPy=n$F?l6%FNfMhH ziElA=xwixrs&(n9mP9R*DY18Bvb+67K)UGecGb5>z~PAb9NB)U#r)zxd#SPDWxuHt zek~^XL+JOgmJP?(i>T9eiHtPh^ZLWJ(l>f*6zLhv2wPJhk!^~ncI)~2>qaNHALu@n z9FAXmd)CW{=1eAA;=zx#t`An)`F>uZ0pmW%{=L1S&+uy3mUe=E%Tt2PJKNQOYg5AH zg`<~$1&KVUcw@Ka`|DFNn_0EveD~c9wJKsXe(3^%9@da+%|QboL8e-iaieWkaxM+H zhYB}3eSHxQ+hUsam0kXH+j}Dyb$|CI#kM4DbJwY-FS?+c9&GPK$!c=g@6qIy;{5n~ zesXeJCJ{!BPpzM$vdYi^k&Ht$pc*sB5!$_6Ofn|IaLV^*d+PO1b&U&5>hw(e;`D?C ztZnWG9MBhf86WhM1k(*i&;X8cd^8n>ApFD<2mCd~D2-$$@BB}D9MvPIPAkRtO5CE% zmtMnKWjYp5x|ifXpaD(Vvwev%6)`zYwMaEIaLfweY?6&1J*Wc zFrO0nLY}o_E8wkq*wYm|%5ikM>-O5OnP!_gA0lg!P?RtYXf7p&<`J*&rUElGV8Tyf zuX?A!vxs}O@6zp8&t1DMzj~p9&`rAFq@aPzB!$s{@AjBHggtUs{4ol#r)smCNoAF# z0Y}!@X~4)3^*%?09ElYi)Wq)-Xh2VIZvmO!60!3jq8c1yGs5T&-nodfCEHNmH69@< zOo3$Yz4K9TP#VYvyyNirqKpepTbh-7kH-Tj(z{!fqU=y`@aNKnkGWJq3HMZ&o0^O+ zDkKV-wM*qiv=H*Q2O9deOo_UU0_S{6Qmg|%Wz3jSr=M_-M%(N@vZCam-4U=(#bTos zTzpM;icdt{k*!S(7VSL#bEzeR5r4g;vdHG)R?+thC2b#uSNY?Yu>sqdH{`1mnXr(2 z?V6q~W8EZ{gZX0lVp&XYg%?ORw}Q-SG~08ZDWoJmYy?olN}93>7iA||_@0@uKkSmh zu9faO#zF1?7)fFx$%8o3W>fzX(QF)aZ&kWb;?6vw9&R*sZ7KTX{&`_gjn3nSDqcCf;|Wvpt} z!qq%`rg7Btvn^)WaXS%zZQOmpXG{@gAZPn(>X%-G=NkHF(RN`URW9{=;B!>lJN@{a zTWDpqADi(O_EwFN#KZ|0jQF{Er`GC?M@gBex>H22P^m}?@fk_2>lO|<3^gi`D&w7F zsut0qM7W=_c@kvhmfvmSt{l)k-q&%w@fhVb+LHpqS^jDi{$5N2Dz=>3@xnRlUCb@X zLQ|&R@as< zqtm+9nf^3xj`y8sQ=Q-Q&m3nk-33g`TO?gKw9uJoOVhH&svjF!pEZ^FM>f1Jg0BgLFAW2gh5tFt{pPnX_|4s1F`MS&732mLoH~NoLgVRLz#KHFuUuC<^nB{hlTtf~Nu8SYYRCOX5u#(0>UlxE zhfeR?m_^xBr;|)i8ZMfKja_@1QXJ&p)L*!ETWno*XuFX}n2;ugkZo%h$0g)=?<(J- z%*0K-Rov6Z%Wg;EJKBp+cR4YXwU2Z%^mcV0LeK|+nhAUsz zJkc9uxJ%D9-faCn81o4kiDse7kj02t_X696ciTbVaEULt=+!Ue#DAqMX8PsYr8=Y5 zU!{!p>^&pwPz_D0smsVcMNh)gwgsW_A)W?IvS%C{a$YYke&mEp981uJ2OR8A7$7ka zrc;TA@5*l)>12|Kl|)5+{zrb~0o0MJL@Q{i&*+ZwHF`bN3*_qF0PaA*^b3LZ0fUN& zN0Z|t(v#f@fmfHQ!Zwvr*^dDci@FcO(=j5ihcd7$`XucTR;tpF&kr$?OKpro(0)*D z!6NHf@I_6HK+i;Pg*3xMF=0a&P>N&XTgTs1<;n8+?ga87w0*(PqQ>u*cSYmmtMN}3 z%mncx&vf*)-(bELv{Sng6?P8^bw(*>nx6}87&fxHd>fH)r8Z%Hb?c&V3&OFT$a=N{ z;{A&%A1awG@TiT#jMoX&6ygx+h=;YN0CB-NuV~rO?$YNWHTz_d#>%}J0)7ssR25+b zrwCT^7D)CRdR=GzW-)R*QyldW7g!+J`H@ zMd8v$&A&{#@$jk#iw0vlWq-(0jp92<<0KOgiW@PD1{_r?pia_&dF{9h2MU}746996 zGBd{M(O$@do*Ck`$QXWHz-Ncx@Q0&F_~I1PTizC8-}YDHn<;OX2n4RU{L0TPfP1D3 zdj~y}doGU46QuPZgQcDrHAO%BRrbJ`;gE(}p5O zd4+ZthEMstkg0gMeci6xH;vEOH8kA6)Oxif1Y<-j#rL&ns^L>sUliI%C%rCWzjwN+ zj(Y)_kXe2_D|iB(sAR=T;t$V?c2IvIx3pxo+JE2uYWAAyWy88W)Q0ulaCge|GO_ zTQMfIQc6D!|Ed}N9?9Jtp)79ymwo*x0(x z#GI79OWn|8nZIb=hxJ@bXr#x+B6tjjlqS-9c%l?Kc2dO_IZ(AkWgPZC8J63WJFg@~ ze>x7h`s$Ake$gphm#8~|i5Wh;XI1T#*nqNPct+2Kj(Yaps)*ZUd**u=ZqbH3{HnSM zHq6wVd1w=>Y;ubc?A?LbFC&IBiO&n~Kdzaq0Huu-UZeq>ntUEqneQXsGnvWn31q)) zc)dV?uMznszH%EwQV9>buDNPnV1Hcv8vE(ggVlM|w52WO?V8$zo$a6@pPaK-5BvZ% zgVsQW*EuR9E|w=&FQx40Kc#-70hPy#CNZ#(N3G}sr&@z*&)xU*M(%oiYY;mZcK657 zS0pzL@WIBY&gRQh-{V@SGCD`^m4=3r2u^++;W_vgJ%VXs;B%_5n>jaU;duQe{wG}W zoF=L_c6dqG@gv*TM8^r469R%C?d%}Wo>x5jFr8V~aHsv2?LfS$&&uJ=H|DwKCS$Ix zwcRTUgz{!2ySl{WlK`jd$Q6{M0Pb3-eB@y-&Bl9;lav>Y^4G#|;I$WRo_x#kregC6 zYEkorsB{1jda>qFx_q0i%Wp!~I!t3g+qQ@*;-M+|aQoLeF;F|3Fh3#k z1@Wl2Ar^ZIJMlV`%SitXk*{mrMV>8cvni_YEmg1P?d91deXFDSCxz6;f7(uUw9mpD9$Db2Yk)i?noHb5UwB8>=lnh5doW)rCml0tC~Cy&4AmoSELPzj-X{rz$eA zeN5m%O?p(c=@Wkq?O4fo7n@@xU(@{;L>R5_Z|m#7$@+q}>~~R-{Fd$qswK5TUY(_p zJEKPnmeCCh$h(bPgpw>=;-Qu)=A+6ABjr^Z!39WdwLRwlyE8%q2$nA~#=95oMDjN>?;@DgLuEm(VupPm{(3&hAJGEe z=sBbT+i8&AQ|{RkQ1n*f(EhcW;M}Q5Y#`kXKVF9CkMknyydDYVvng5UIiS5ObzsLK zRHCx2)6pHT3mPVcRLU!kBKBRl`2lNYHS>PW81Q zwztSQj1c7|!Z{O0cEIP>j=gys@zQ{>zUwu5edz5I&{<^H1v=aKXh5RrZ$oC|L4(jX z9SunG_Xmwd(2^+|(DONSSv#P=s8URpD!ehVw#Sa1=BEKum09`Jk8oJ=%mJhwH^jB1&#*nLT%6= zq)pIU2hg?Xlv~FBHSjwxCz!SqpL0K%&8NojlGa>+ib&WhYzBGPW}N&ga(rt5y?G8Z z?uj67MN`_SmLqY)5}a(IVNW~|1RR1Cy=y&@*^Jpe%YmTQn$m!5LG0~7IFXP^X(1Qk zjd9lGR}V62z;y|**c_%z7158zMDo|N%W{vr!fD_0ob-K^SrvVc3SYcDugd#j0E7GJ zrO;OKIM>fnbo$^}L(Xdf)5A!y~Nv-o4owueTLV`a<2-fveog!&u6Z1-^(t~GCFn2?2ybkWl=*et%qIpE+Lg{*N8&16J3sd^o$86xYW=5|tPfD70?>8{?z4B3zo zlo<6;*9q*yr1_W0yO%uTA2?<{`;=g=3hf6~utqEEOP~B;MjZl~XG<6`uJ--ir_|?e zboUmG+iB$w8wcOS9Z(ccMy6DObAER#mM7dA$r55WrQ~08rT^r0K8>%R^|qxch2$Ug+@`o;$Mnyu{liQ2*)jE2jZrzLbM9D4fMq6fSD;TjVNQM)G!J_kef_+0A>>D!H_lM6(HKB+gHG`9ZG@ZxdZq~f>lg(}xLW7Vt|Gds?fNKl?&rfbd1cp|wX zO1_&-`HqOSi0~0sI_?IHLKF>eIGKq)?MR=8XT@xGV?0(7o-g7%me(*QJ-fPtKUS9F ze(77Et-=L5NcF{Dt|$rD0Egn892~^Td5W8%|YU(KB z>A+skl__-0*PF&WZVx7TKHM|7=6u}%5s%^sH^XnueMBZAhWXfLwAk-7M+7%#)uvLx zs-!GozR8Yk;QBo5gh~kmaaSW0(698-irKo0=r;m}!&VVGhuAt^_#`%{t4-R<+-}~c z9G-dS5L*A=3gLf;Lin!O_ws%H-V#Nj zLAcS zDDi*2i`Z_)A0j+!_S`M*3mkZ?1IKiq3|&=YC9hLr!-JLYQpM6acZFrPE{c7X-nWbS)j>7||cV519os%W= zXrFbhz;`6Y?egp2F%9wTp#e9pOef+-dkx+gj5G;=OS5G|zrs=gnKY;i<%ft#RSN#|aC?YGM+-$J^PGgj};jL67y`#N&P z%KbnE_MF(s%W`~qM#k*h%%Ws=&9^7NwJIi5Z`BT{{0!z^dHyzO=y7i17N}XJ`0v>9 zewX#JS)ctX{-QMXZ1-U+V%V>QS%Fyf1?g0E(9fYP?xn9(RcO(-zQip$Z1_n~y&Qz^@b6kFgj_y%JF`mhtF0pck&AJXX48#zPq*Mid?uwGT#r%2mkdv%dTPV-jdmPWbOnDsT=-y>^R2J3Ly@4?^oRicXcHZEs4AjTZ0=B#p7oYU zMkUeKEAt@M1p|LHeRrkgwaZ5l+XoDCLvz9+*sQR*KrE}9$kD{lgC_@Z5$g!EL?cOu zvu8ZDJ#o9sedYQyNV2*@+t(Xa3%CcQONYC*qo(IbjV7wrB4@A(S^OyH?$fW7H&H!dl$}q+1yqs<2K)soX%@&+BvUO=6_0qc03q>$-N6r(iI2qUWx&L}LZ8g;Jp zM+Z{Rhgs_LysQ}L4AIL}RCZjKn3PztC74c$Lz`=5k{7;T$2%~dRbDyLNQgIRLfY)* zZ=a()uY@uDLN`ZDxhZ)_q=j@bxtx#rmGjt-c~N@xIzMhLQeD$xLDf&&I{)POmrR@G zkH8If)|bztK7VF#vQ$@?olUD#zdbb}fitNpWKnaNKo41BKEr>6jfJ4(&pF3G=1YDn zM!jl`tW@UvKHB{KnqDkfMu^w~h`IfVDu|Gdl-CO0W{&0>^eaSQXjgPz3;r`R)C%0;dT_mQF zZde=3m(nsHA9-eQwD22Rg150}LOq_(O(QAH)`s~+f8E}hjb#}tevNidyDLdfa-GW3 zfjmW4vzc-hlI!}ZbCZZ=GhcPY;dkU8KegOzL=$%%Wl&>QyonD2G?XT;U$uE>o)D)n z9y^-dU>AI|rDr-(>RB-_URjEPQbKtcCBZIR<4syELHEVbhIW$6GF?3 zEAyg)N=Azgdw%ZJV41L9yeM|!{cwxuwqp8;63+SBil+lFi18DqBk;-m7SA*I)f%_G zS3xg?3=`CyQWuZt3ADu;B+H$oaUY$U4 zbvh3ORKDmoKlF?SXo>YE8xXB0E;8TGbNA#HdH0pvwN-+7cX|BT<+Q~urMd6qpOh5P zWKdQ6Nrg)u8WMtlijLH7ss92 z_0%pR-&sD%n^!eS_%<9^`_ZX=+4YrG!$pt!_1P2grwrak-9mHKOS?~*GB1hOqomVEQq4b;qlD(`A~6lz-Up30!&{A<{!y3$IehjSC|#ZH>wJPPe$ z{k>mgf6vJOc|_0#{#!IG=tjqO4&@Em5ue!BGruR$DwDa2z9mF_g01(I*!HB9Y^^;M zQGL{mIx)5KVm(c|y>qmCG#*{6`Vx;ESz*uQF8&^P7PLlgS%0>A((ibn{s4pHg-;3qGW`a+_D|kR{F^VyT{VBxml7> zUzK>BA^bYYn%eVhoYA50q&i|_Y&^7Xk0VXJOmqG8S$5CmpKq=>gkGDFJdR~VxWxyN zCU@FNBSO&GR@Pln#|mcKx5(8Y6TX*ED}T(i4ctmdE1&m`L{yo!!i31@`qa<9MZc|; z_n`sSWV!O7IFYf+z4mB7X^{y*Y8cv;a<&4SV568U+^+N#Jf+N zWAL&6Y4+~(y%P#&L|0C|-gB}@X%C9xvG6`=5LA9Af`tci3>Ap#LsrbiHS!RbL;TOf zKZwI7=BDGZ4~0lzMv}~8qSMxsnJb`I<(MC0Cz=}jp!WS^^Yj;=_zz1YLyyIZYM z(s-+M43gb6;Bd7m&QA@puyAdsl?KGyu@Ns+6;FQ7EPvE6qG|25Jd*jLn1a*)IA}q% z%}y2NC<_El#&{$@+ebn~7(iQTRizxoBqCt?kZb4r0wF9SC9>?;zx* z+2=ICc`ZtgSWZ1KP>Eat{4?cE>Vh^m7SFcwz(8EbKWd6W%oI#ZM|3e z-D9L!SmisQV+DohYfLuEY%u8C3a(*R%i z9?$-k4p(xLBw_*y?rT^M7%DZd$`>q<^_q%+)kx-*XZtPlFW%CwRrz$?hTxN#!bD_KRdFz8j4c4UeKpa(M@yqE{3oGh`*|te_eEC7)yTa>y<1OaD7`99 zC;V8}kT0ncgaqv5(sqAfU5@&fP!(2HNY6Q822@Imh_H3Lu2pNJ0iuBhJi))x>fpIq z-21vFpD4B|(K=q`6XmkTc{iW(o---*-HU~vIeSCT3`y=GPkt@;)t7=U8Xg+3qDl;l zrP#9JsFsGdKam6F8?W>QXBCAhZyFA0j)FS>SAsHm7u58ZXh8e?zdbKao{+{+3_oG# zn#t%dvoyd~0lrJ70kHM|$fuGtz}yeBi^sw@U|Z#KG$6AZTzeN=87F#;A4GR&QZ=z; zpTvKCXgr#oxFvA{#AGATfYX2H_s^;UtKx5Y{@q^w?tlLns{ZZ||F_1^MXU(whn;#O zex|J|7VB8+UFY=0H}15=gG6EJsYJnW7h*X<{)-z`tlwt#2Mu8CO#a|gSQz%U_|pZL zmYtG4?i*{duBzRn?VF)Bjl;4+lQ<^9jE&W9B$0*Svop~c@9%E!+Bmu6`;zqWTf3ji ziP5ym+n`Th$@d|44qSABBdN+rZZo@j6)|Z;30(x1+^s=yvF|NH2JgZ_HM+%xm_`i; z*L%0XrV)VtA07h0)B6qn4;wt_-xH_lJ^8%i zW$;tSYqp>Y**!-@KLNLZJeuN*CzxIgEw<1KW>oJivlV*eID6vt^r;Yx3+9vWPrmDh z;X$P%2d5Ex-_moE;l-$TLlfbiN#-9N-{>Va_j-P!z3y}s$330;Egh-bfxtkRflXt=mHYN}w;$gP>$Qa0ySPSiv7l{n$$SUdBCK4Zl+lJU=AvV=rE?0~b zs5hd9s<$vDpzmQE0`66Y;Z>RQ1Zry%bR-1SjgQ#Pl@?pq7=L|%KvmX?tz##;$*)Ax zK&{JVe_;%X{W$rHn=-zDyn|#!Pvg$4?JXgw4?QRpI04+N08pPg2FEzl0Jbf}0qW`S zi-f9ll9q?6Zboj;h`u<)+TpNIl;NVrMbwb{aCZ?w_>&c9LvHk25p{7jh&?Ok0R|B^DWK88Ju>GBvw>-(|ofg;i49AwApLj?neze>DjY`jXTpdK9$ zf#V{pY#hNO5T|*GQ=e1S@mj(hDwlnX#QWdo@QcmJ4#T!T4JoeZnX-`O{W;#N*#tu3 z@~3_4#6tx{6Y7VO-Nxy#?y|hY@fJ3nI_t1BAB8UXxJb$Z{5_^=3C8EiKc+%p*9-{K zQvk%{UeLcC4~)L48cVrnFr~{H3y9pS^RTj@bc)?e;kWevni=U&%-N2nWK_mr6s>L> z{Cefu`B;+?c8=J-M=_zKH)^^g1l7;ZCeL_Oy^hE}WumOGGUu?r?XPYeJY~eb*&Qpw zUXLH0gorluj*^CT@r^++)d#(7nBMJkmYF*xHph$$KZl#_+na7x_OJSmpAvPsFd$2dUcf~KEKVmlnD(4V*Gud7KO*kn8JS&ie z(WBf)Oh3^AcN>6Bs&~Mc$ZSl(y#S9%AY-@I*g~ug(sXK-x@FTp+{sr7DeZ?}z7AWy zIH}>g7i$fdpNXcX@la(?5v#P#PlAWJ75KZs>vii)N%BaT!9${HC;1N3#T zZSprJyrcn8b=PeeA@gDQWu^Ms68Cmn*yalJ9K&+TBw z%porn8`=CjO36KlMWR*U>d7we$O;MKGsMlg5-D&X-h{>PWjyW+S9_RSCNTOS5)Im@ zIvF*X9#f(uBh%ew9)M|04PDI(j<@&Fh1D&$(?7!N_A9@n8o3ibc$V+bZ9C(PIg*?^ zRizIlG;B-UGFsF@O&Gy<-KHbHsXEk*4=Ze^^`O{(#w!gl%$DC|HmK$uiH|O6N7nwt zNG!8cQPFu!+jP#XldOD$YB5`MFrTrX5M)(||8mh|u;Uq5)&2KTj$_t1CJDfTM~q z#CI*sdS&*mJ8m52e$ze7K32f~8Xz>3+@IV(u(aKyD1G(hnuyfV$2acjB8NlJ!eA6u9aS%#2k>VZVM5znh#8T>S$!w{*pVD>OK1w3<^ z$5#{_S*=Ika{##TL&SNl7Hm8YYFHZXdfR=?#Sc`*OHA^htzVM zlBZi?9&`QlE;OV+PCjF6{Zx%R8a?ItNJi!0ncfM=-I4KOtbZdU9KcsMkF$+rmOWIw zI;8N$IDA}zch~r|07rs!VO^Ze@k*6>sjn*eSsUZa%CE^2R6PjO^U^{8up{?dvt>C`jB2rotAF1c(^T~=T^=i!yMLWI{PXPL|37!S1AzD@ zFTQ%^7)ks16P_^+eD3&y_uVtEBf39KJ<#UV61#7gPSPd2-Ny|^Q=a8jt(fDv(yP2B zO)bSzJjh9n4H?v#)e<59nl8n?@#t3qMU@F{vVq~!b!+n|@AhAp=JO+4u%WA@(6Gnd z>dy|GDJymHF;wK`-PCU~y|dTdy3`b`yr=X+MA_h+*WAs88dJ+Q)X&5(k~qoQftk-d zDARjQmaPL81=3v^inuja&kV_0zA%~6jLRcOZXfKQ2*QN*flHjR%7aU4?jy=U8%Gf& zbF67NKYp!6grP6l^!p0EmcGQvpVOI>68%GdfUpXtt=yD%sirdWp4rAg`IZ;9CH4f% zq|3;94?G!-2N!4#V&^`Cwgfl@C76NJP$EH%8jd9pvBVivS&to(tgGH#fdWU<92&s2 z0imp6sgPa8{}B=XZ*;7eWO&f6m%otoSHU3xPJo?;kCDLL)$nBGzhVB*IqLtJ;gP@d z`?FsDmgk?6_`ALQ-T#1uzx%`A{ek$unIA6tl8%rKQ!Ncw$WJwF1N7l{lCakEx#Y`V zR`#bd(Rz)K=-X>F;7*8glj*&%;O6wRUCArTLjoJ(!`7>*9YW6od9Jd*pj)0bTpZ z(kqW?blvfxP=SX3yz~|f7u}kUXhXAl)qAZoJe-Q?NRChF+2fq*txw+mMntLGnK@Rw>RleclpbN=Ss$pLlaj6e^{_}QG21XONT|H>V+d{AM|4NEBDVwjG{kys#~ln@_rv-V$*-I+TCP&AF+P(WiTZW zJjQUo5=BiUf*!m(Xd^NmQ32e+%>r&dzY2=N*m`X}?jsWk(OV=`Z}6GQzldTA2W`pZ zlmWz2F_s!*L|N#=PBoA3AEF>ClJr{xV8d{Wh7STr`Cp&|gt-G-jG7@Lu9Xdu)tn2i zbc)AXOeAibz!oLj!sXZ{+7OrXHKf61zwMt-;8Q|{uTlcKDSNS4kHBKzi{C4LZfTeV zK47*!*^yDt$)w@AJ59t*Z4wg+Ry3fs>X*$gL#{aVhnVQMmj@ywKelZs9g6IYN*qJp z$F7@fChj0)FLCY)d+Xo%c(lyR@Ng! zrvQ{CQ3^bm+SQJs1S0lCht@{1dokc|wZK2s-oiS~Hbr1A6Vh6-rHSntrZ%+=)v%e) zp8`p;rh7|Q(D+uQU<8~>1%Kk8!*hEvwtJ~udhrqhT$%@eecm0zBI`jZ*qS5&Tjeo!- zZHfv_^db~u?a*Mn;zf>dfaVV*Bw$G{*n2>ktL#Uof@<`;<@8=LU#pxN(Dtk(t$!P= zu{IB%aJr-ShASZEu`Bpu09E`mh?dM>3>5QwEbS;yac!jTS_MVrzo^&#az~+zEcR)onbI!_|4g z-+fm+8S7}mAc^Wy)=QNp>kl3h0LCTZ?1b0APKbZ@lg<&mtB|)vVFT)fnlTgx zV3oU#z_TAjTkD8SF$@5D__5g#UJqUlhVf6e^w3L82D(Ibic2260w08_HIqIb-4{4M zd=WqiAM3l2uA!FW>nzR^^rBUsS4zHNug#Ka|90G{XV}ummEndTW*3qFq&0W7cpgLKkth?pIC8@zF+)3wmS6+E5sH!`+U)8_jb51nI zRaaEtx_Mk?{p~97DPTib0h(l|9#q^*rHXXn+~QTyKk4|0C=5$b20sQWhOh0tLtwYi z*{F77Fw3`+SaAIm^TrDmrX~j)Rg>tiy`+cL3mx=H?E5ZpLod};473*)UoTt(i?MaH z_Y&ZYakQeTO$$j5*1GI_uWDR$b1|zAdBcvpKvKmQ>zW%A1B$-8#BO&V-U6g!cbOQ2 zgV{axfhs)Z*(6F@Oe&C2u>lM_3#1l^it-^Ja$=^~k|>fx{_Of@!rp&U?a2Q&5y<%R zyff!2jS;^wg3eVNEvVbWb+NlMRO}Xhi#;L?cm&Slm^mbI>01WNpip;B2DA?x=^c&9 zi+STIx0@yw+j*@;12Hy*mLw?)aGkX|TmS5{abf4sLC~CV^d1YyJ7o=m#%2qiighWy zMWi1b_@OQ4)>z5(lP+A&RyJKp=IUVDkyxeE8K0N~(Ro(}b8fwja*r~ejH7HR_LOQ0 zfBp26j?*`^9`XhQKp~V{;tyUHzYdape^kOKf>5E=Zsr0h65z&9r((?on0W6+X&$9} zt*ELc93MY)h+)G=VhhvpmhT0!}1BcsTffF`@bYc#;2A{(}B|iA|@)J$5?Wm8wBCdBiNVHSH!(%lYH>GtDlnDq{DTHb~^$mh}(t0be=X`)(?%OS}WGS9% z5SmVAit1Q!RakKt?S0oOUP62LF^it=wV2(Alc;Lc1$7N4ugTsK_fB)G`#%Pq*}5XE z>ji3B*C0U_^3)`l@M^78Rnfq1kZN9A6{I`t`L&~Xh{_AJe|gS+@U^g$?TmU@-C!@K zV2d~QCd^xi$4DtC*eHUGJcJ+jfci*gL#*!wOpoR@$z z!ggu+0x=`E(ylN14|If}9Apz%mv3m(5`@!R#|!83y!tkMS8rn0lFfr_>_k*hwd7aW z0OUlYov+l}ZU)}#6YViM`Nx6h(2kRyzR{(UqIa8bWvzBOKC9fjmyn;O^#0*uS9&t3 z++h|yZZ+S%ZC86Sf!271IzKn{ia{_~#P_o)X^ec8glJ63Nd8O`z9%0$!6_;EKKI8r zjg?K8z`LmN69dy1!Ws%yrF7K{`N{?ljQpL9`&se&4v$_zPRWwHCF?4!qbab51hEja zpr{Jq8RblDCTY!Y@j!LW^%}>YRg|7qbGhVzoN7{%p&#O*TAdDheIq;W6foFcv$b}9 zR7G@NzX3wLSB|8ZJ!1!A-c1J^M7IA3)2fhWmJvMLHvichyX+a6BKUa^chNezgaOBITjG7=UmHJiZWsrCIo>7JXKsoc(1sd}-G zdw7W1)Us`>KOd1t^kklxmK9$MFn4KIHql<@`xdbuSw!;Z0$dli`H*rpQii;D`_iqa zM>N2dUz5q1=*P9kDZKqOW|B5PYbWc}8R+V_oABiKubvb|kIrI>zXYpo4j)<~f%6kn zDxFNIfcnl- zi_gdgdw6qYc#G7x-cm-GP#zLk>fq7fV^E#TO97`|1{d=;ysv%w5PrmSwTch3bsd=5 zFP~zjpM~H0ruHL8qxxeGP!u(<>Wd85AqW$T3Ge7N0#(tpnqQY$t8b`d^sXq__xdCVib8s8=Hg@rZb zb)XcYvS(&mt{(+tfO;PHB$hj`B1AE37;D(bw_qA=8rj)TjscfyVp`5`dOy`T5!?fMO7tU{g3DD%hlYd*)%qLFlVj6r0uv= zGsSy5lyB+I3vyRh=02)JoCPbnA&5Vm^Yj0$dBdoviq*7i-3H_nqm)8jokmWxH+&QL zsJEz_c%j*KG-cGbdk12;j2KIRx}%v3G@FiH$)%73MBs~A^Hd1Glv()(h%9A`=Kczl zxnF&6;_!g}`+d%-^q}lLNaYXI*MMjZ_JDi8&Jb)1-0Jx`Tmr_qBrp)}j^CawolG3R z?`op)=IPQ&84~MHy6M6oLjnS*?7G-Xmhcgn6P_)cuq7{lIkmb5DK}7e6cM|aPrnxN z{L#p(s5@ck*M*3JJ(r=S!}JmZPp~}v1#*-~ZmewZ9-LEpfm5@59qSUi0<&EJRO;d& zpOU12K=vM=73l;$K6hXEOO`9#+gMfmZG4rIonaJ590`p_%$7H$3(tQJ4fEhX-)zGd z&k2{7dHu|q@3?U8E^7SfIdz6UL%KJ7uPZ25$iTkHD>at))Zr05`KeW3=hxgh>uoW% zwCwY%%LA6eVAHTK;_5HS)YpHuzUO~#yx7*WAQ=FE5b24^7t>-B zK6z*5!DCUq=?oeYuXpdYG}zH-4kziU?E90hPHrVLf(8fM{Gov8O7tbW{r4=n&7$AG zb;QEn(cwfsy)asgQ7fShjq}?g zpZ==m`hO>*`j3vy@F#0)(=V}%^Ug;hPrv*)SP}%d?7TG-rsPycf#An{A*kMb_)bG}&w&e|rjP*@mc~(zq`|z^Ty5=* z5@e@EE7r_BeU4V8WbY)ZbSWKS2eAGFNVz~M9&jX>B_NMy`)}THX#RrwMr*2v_X2py ztQZB6Ov5w-gvc@fElThI(2R?ZTfc&`9H@>Og#Ck z`7SBh8rPgFO>58`Ifc*@1@f2QjRW=dRYK`6+uZt#5x>x?*qr0>2;v9y8`xqw4jc1r zE}4@AiUF24@J6QgB}o83~|mH za(4)FR?!r#qaQlseLg0nR$dj4ITjX~$h+^lJ+TQ?n&DhyRGJ^y%TY-*un&1d9cqmi zBPNv+>=2DKaTfqQyA!*!PguCnFJ%I?0K{Z+iSk^AXC+X2x}m!nI!T%*LxNB2njG!{9%i-M;NSdAhi`Xm zRN2`M4D(xRRY-do!<^So7z2H&f|Vn6BZK;yzKm7z`K z4ao+tB{5!u%=2HO)B7(tS103`{Ev{CDXti(QLeq(S%UfUxGckJ z<-~l#O@@QPJ+a4IFNI4Jysk!Qh;>93SFE+0Kc=`)<7rCXE#$d9L?~4AOY#QF>4bMXRW2#TyAKtu6wL`SN8FSL2D^Yh&F@-LG2it(+#!o)1RId6-%ujIYTAIBC9Lb zQmB!H3^YTCikyFe8G*mj4IoZm(6m~E1kW(m@^=$3ttO@2{WFVcMSu^dro>KLJa_lA z&$X8J)pL8`TR-V6!4WCQ7^swAwnI6@zWV!EqA1r^Ne<~D^K0X&VNx{ttoLWiWx^Yh z$?Snvo{Sj%PrCb-387YNK;U_P;rS!)A11y$PnWxybRAlFcGYD`R@-&I z6v|_gGC_V0;1Nd>c}}Mk^ZF%Uo+AG^y!5j18*Kv?M;o=c@{{h3+iXu>p3CL7FiFaB z+Z_PfYJ%hS08#V8oH?pqQ+#vWQ!5c5RLTgK&HxhVr61nrT#>nY+v(F&#E9fsZ^V6k zV$k*SuU!|)%G4*D*xJuC(+TIITEXGK90N&dWEjt6B`Cn!!BkV(%INWo7hQ(MHO=#w z0U$|;Vw+j5W$mYc?H8X8QMMFLyDh3*ttmd>nFG%vk<2CczQ{q&;m(8?&Tm%-Y8^@r7x=-kp$e3!K*lS|9j5Yyi(ZXO1B62$Z@{IR4Q()9 zsD47VvQz)q{f_=)lcN)S))ki#kK_)g$V#N_t>&3ivgrmSkm{R&V$5Q~kSBleRUL&f zUDews9~rwZR;u;@+7mKA{&Vs3pWmP3e{}pmxo-!3C}E~?i7i z;(WLc1(*V-Re>zzZ9qO7NmJIOTmuv=Poob+;iL`!4Rv4i8VyK8g8{uznkJH#y@=jf zBfyvb(a%8t@2_tD>qmc`zJE=`@B8Rq6Y;N!08GojY{c(2_TQg~vwt+qW{_~W<%Va6 zj8M=QYE3Bq;GTf5mmSbP##r4!v+4dg>yn4_!hlSEy`p=TweEEEPr781>Kq{F?44B4 zg|x(MFjZxwG$v{>-XF-03&|TeTYvAL!|rv(d8li_v>6>8?qh-QVx_ z+~nAWsy7a~@hN8+!w|(PDhTELfUkLj6M3766}~e!1Y_+SmcV13^H;%VC~C4b+V|}f zZ!q){oNNR$zhEELrlA(U4vOc$L0wg~eu2(M_Km^RGK8WW41BNs(uP+4rIGl5@^zYp zZS4^a)Gy&-n)N0ap8Sikc#hO)hHQeJuNp4YU48K^p0+~g>ubEqwGl9nUe{y1jZJ+- zCFWFlfDaAMn&Z*_HGRIG*G`nSR6b_Z21AR`|LlbaK=*$``G;`-zPL(%9G z4%x^&-dwB}q_Rzwjj;u(4O$T|dWQqAmXI@PpYrDV_54~bQ#!a>DLEY9+rUXXi4Tqm zrdzsfsaLopj`R0-C~lfu9(|f9AUU0TAl_p^*Fh~4n`pOV>0bBR)mMirAD|OpoT4KEh?p&LEGPBO_XHdHq2DHc^K3R>cF~ zaEdose#Y);cqr99QK#nhin1|g;knfani*{g3A3r6VK;6pz89%u7E5szsUp$p-n5Lqappde9M@QmcXI*PJwpOvfHwj#aWCjN zB&n6^;gdReI8v&LyF$E{vx#x2Sc8<=MDjKw8A^F=Q(31E{4@*OvwE#YF~I;&top*D zEzeB&G6Lju_C#;tptL-E)eMjyWor@lDO{;#E_2>Ht`BJ4KyR&09Ac*7#mJzfdeF%d zM~)=6^Y1G5ro_H&`tj&ZW#LTr^C3kC+yw0&crd8Mzj^BWfz-s^U>4B!z_BUn)|T;R zNNDZ^?og-fqy0oy!f}?tAjmri{sHJ%ZTg(KWXsv8%Y@ik@ao%!AT$860Ax@-P6P*( zcR_M?teSL2G5;k`6`u&WB3dkfGXe@ z)Y$MZQa_qT=Q_O!ppiLMdkZ*f#C%I)1U$_t{TTaLtXC z_d%UgDLztXxCTiE%I49w;IowD!<&xJ-%LK9;$XX>oioW2<0*w3n?rx#H1ZH_Jio`W zozx5(W5eb&JrOwMb;+@3^+V&hd{Z{=J7kGgk(slFQaKNV^_uw-t1aJ!56HpJIqL@V2V=Yu|X z$7gSnh--@cCx6nd`HSvF2WO}MHa@0tG2B;VV;aq&#ts`tpQ+|@6G)z0$0a7$Gc7h4 zZ-KC3@$DRBwpk?i%CvHyIOX=ouFK^=js_m#KWVuCTJ`3J+1{59-gH@q7C$GCA$|3y zPJitv`Kznsf2|k%{|}8j3gJ!DJOdQtPkLFn{*zDNpN@*_3sgKRkTp`tD7rfI?otld zi|{n|%=ydiz0KfgFrx=ch}tDTd4mx{&}4hvRItIMljSwj8t)p|G{u7C_>3Ufuwxe= zT`nTSo|cZ5E}9oqoQR;9Vs`IsG2@C?m7-o;nLRI!_l=6t(7X*9+1T zVDRN77(?@<3SREuLruHm;)1g`eM<^50IKd-X!o7nJG@u-ra?{px(;mATevv|9z{`g zC|v#S9P!FuQ&c2o!A9Glv3FyiK=*ke!l_RH*@H2`3o9EL%oA@U;c~PdQj$RhG1CLY_ zMbu9p8}cZBdCrLLCtW}#>f>Hvl!Sm$UduaqN6G5evlgGWoY^{I$vv6LiMR>%XT=8ps?oX#K$b zfh7u{Cd0#SH-JxN4~Q?0dF9P&DARC{+t)R3Z$J{jjA$EV7?`WTsT6YBlq+J!PJ%~{ zyckMQ0>t_-`LMBCFdRgANNf1+b~*cfy(K*@rS+^+-JU?{A(AXd0VA+84K-q#*3CVa zq~+a`&66tld1@|$t#wR2ePYzc^VbobDAKd7X!H({_uX2DP(0&l=e*{fDR=-SbhHGR z$be6yZnZa->aW4D<2pZ7^WZQ5^6&CVk2BC7owM_>)F8nxA22 zu!f8)&MZ*FA9#Xec%IEG@NUDE_nzP$v>j^W?RJlY-EpI#2%0a&Uv;^GB`PIgAnAGH z9Y^5F_AP2}ii%4UuHn8>4XG08+C#8Uf}&MSv#(^DKjh|1Gf8K{K3#(Ko#c>ry3hUEf_{N)_Dy5;CB!$BYzW;Eg(K1h=Hp5 zeD09H+CI4A26U@jK{WAT>H(53F2%gob^T~i0P7h-@-U z5Z0HN6wh+~`ddRG@16@#FVGZfiZpallm$y?yvRjio3Ed79QHt+>-ll*+)R{mBUqd9y& zlA;+={{F{&Su3-)1YW*eBZsc=v+G;9tp34^|7Tmt|HSy9|7%?ozLz>W-|yRqp3YSW zyg<3bHp5?An*52>+3362I_6$AoB`Q)PDW>?V zG&(}`Qs~7wg^mQZssz(tJ%`^L@4mA<$j(zqi zDRj2eVF|w57QiN%(^Kzvq|IV@efU6=_&>IDp=W{G z%&PXHL5duZC@S1hsi&<*hdQr8eDBUFjP)M5DMvyR06h>L096T--zgLvjAe~5cbbus>U3?o z-oO-_Zm9Oa#`<~lTvohV^+*A`@wV*J!X(o0@R5|(6QFXz0H~w|6l?XX`!%H3=?;fA zhSs^Z?q?CQGuz$nc|yKApK1M~lQr4=8QwZr$d=>QYoTO>+;CK;`Jt9IiDl`4@G)j5 zJCn?|Avg(?#R#uE5;(zOiPM6FupM!usn5cl%Oba^?J^kt{5{wSgA)eq=N{4->HYl( z1@=8|Q``p83NO1E#lU1!T_6RvcF<2P$bt8IP}U?gM<>QxE0!VL6go63^vD;|6JzPg zFd|6jyV&8cZt=6=K=AjO<}IKc?9A8dyZPUHq-pj`p!e+77SI+I&Rx_H=-(Xjqk&6g zgoaGCqLx-oWM}#Zc{4Xi3aBJ!J$arb9DK&9S|96Oe$%5|P(-Rs7BUy-m%9P^1Rfvv zqU=L@5Wf#+sBI8bc{_i7>Fd}I*E-Oh%aA08^~*PR^qhgb++pB*>2reowK2;RmGAIx zE(>SVE$kWb<|D`R&UN`K5d8>BE@_p_Lz-!UpWV}A^>$Jv8iz;+NkutEKcuc6!Lis= zHEAP*=;cb9JKVu;W@ienRB4#TKOmHsBWEYm0DggdvyVIh^m+A$8=2XIz-ivg`!SvU zKqYlcZ79Ib{rf5S@;2Vf_D9djC(`%Gr*w1&?uo55ee1Z^;fGm<0Djb{vieAhxhZy| zmFQouxP84OwRVL(piQ^wuzUOOmk5{J^U5onLp5#qQ$H!vhkOOrC&1G&YNC8>HT)x} z=k}U9YQ2hl+dCPWeyit>@mlf++FKTFYK2l^if>m*(8EDZtyp8dk>$~)!ve`Aie4`~ z2>z`s3A243-Eo}Cr*r#!1{<|AWO>-0jY^aaw`a0Sz8sqU-z$P`AIIZmRD_W|75h9-e zFi?c%C3($Jopo;ueat*{n>9ve`EGsX zH-4@@On$J#7|utbWD4TtwvVWT$qZEcVN=1ZcC07QR~QD$3_s4r{o42I4g@q&hy*0Y zPdJAyko@OPc9=lUT$}_CzaM$RoB7sNuK?C6_7{dsoEgWet9I+hTjeTFHV!IwN#IS; zA|w!bVmXc0$9dwpW$nvJpl-UC6E$69^OwG1z{kMD=>$-5J^K~B_I2j)a()&AzHO-m z`Tz_EtoK9YCSuXx;j5o?d3yVr+nG7i4MJO_9B$*P7xWFFqqzkeCJZ=5rQwCDd0F+>c_2w*Ttw!W>#!51I^S4 z)`9jWOnnvQu&5&dhIvC(-3`pDCPFgBI=!oGjVM&>bvmdY@9xXnHL@BGz7<&$&G*(= zO^I5Fww2_EUM4Un<#<~xw?!;>&uq8McvuN;Wx>67QF*Z6NR5wx>@UA?Q{{wnRTb-& zg8sg>HOT!)pRe%(d?T3P)&Syws(*%%43;;}nsR%k7yg(m3fLcpS30bLUV}@G8<58+ z*9KJN3Lpo@Wwo{*+lye>qh}p0nzz@?OGRKCPwZbA7S?=4iL(rdiZxa16SnGLZ1~d9 zC^^%b6weQQtwQb0(6STWC-e*_AN4kZqXd{pCi*qW@;H~!hK95*0 zK)kt!Z*AHTB-=FXgKW$Bbw?}~eWi18*)@#uI?C5{=86#G)Qbq9*%+u z`8TUW5wpD$Vpz#Y=lMaMmC}s+v+T*0zxf_K1)d2TP8+CarjbvM#-Yzg7zV}62jA^{ z@gy^4{p`Jrh?A83WnR7Uc&y`cMChA~Lo#33qHE1+Zf&kD&gayK)Yrr=%9h&mcw5!4qZ5>byEi#$@9qaFP-q{eK@>G+aWRSI`>mT~I8m(eB;mM@^Fdxm! zkLPAf5)nhQjuMwjD`Mt^#@k$(U_%sU0)`@-LH~VN$bbL*wfG6<4rS&BG@HZv*4U}J zll|2XX&lw?N&ClRE43zfm&;xK9VQ68mkvjc3`gweu5GW;;DYV5FDG!MW?j9h45V5p zPL7?~huwJR3Hirg9Irp8yWD?7)H|{S=Ar-w`9$dJQgLg@%eXf{;pT51h5+|y?pRlW z?6&r@i>0o^ZWghF;5xRF2;+^~@j`n8U1PGI0g(SvF#Hb5Jdu2rzHp=%9ymzpZGzrv zCEvHqEDNMSc0*f2G*8s%!0G`VTxP(9H^p{;Ef$){ubF5*)4IjDkv3Xwt;9ZM zQrNF){&j8qMSwlVH?2VgGm7kczK=?PJ|k-?-yNAgrxT{YW}~=DkwEH0lu77T6HMeD zL-ml*B{(CYmCloX;71C1vOL6YP}^wFzY-KqUg4W?3H_cCRK>6-b&eX?m?88!FwN3 za6Dr71O;B`5F-%>xS4d2?GM1&EjSpG)ql>{mAb(N+8IVGsr9=6uC?jT7u(D>#bo($ z3Gi}0%-uh8p5|M={GeYw)=k>^ufwhUKKjSZ$n#6jXMP{8|M$^vt8?D$YOno$w5;Dp z4~e)MlpO#2XqtZ=+=Gez34_ePIaiVI%>Ub`Dx=YAR6DrIxitNVxdzC`@R0LoZJJC` z9HZUY_HUhklum@y@s@L{($PI-Byk(1mwjoMBO0&IVDELRS79V>-dz}5TqYIO=)u6E z2eeC{wXkV9=-Znr;HEjwXXgx^6Qy%e?(*_6(fqPx@rDvtI;Gcuvlb#5^ogbDVVE2 z4WyDJzj~vHPf7KQQ$c0J1X1V*1@D;Y8&IpROv&-agw3!NqLE@^7jp zKzIaAm$8vn36zy(nxI|Xk?gCrGI$eGs8+0&vj;DSAh-d@8wg~gnQ8%AM0@>XKurj} zd(WUj{L&*SavvC;dH&WLBB0{1okMJ=ew53tz5(O*mY&UQx;2?Vw>4SRhV)TrRb_%@ ztd(c!^+`zbesY0JL3)6cBGe;ig7PG?ML-;}!MS=VS*Z>flTwn$9;Lw?!MPPB-0Z1V)73CM3aLQ!4 z34Q{0iPT7RA*D?76VBRh1RbjgOHk~3X&4Kt3aUqqAR~Z=J|$oTjH9$TW2e3Re8Q`* zW1Ux2j_a;L5Rj7I=3YjMDM390jqQBqG+Ukyv^P$agH!|lSfBA4V!fzPs}RcFFCpYO zAd)Ifn@|9CTAe;`bL0A(6bV_fs~O#M7X&2?7kZnYA+9@d82pyCeo zX3N*jvG6c3O9_h09G7t)1ABI5QL&|++c#V@=krr|{cZ5Vt3FyWeB}&gdggl)C60-H z6--0C&HHTUcUcUlqjZM&J-S3uFB*MZ+=hzWgmUM=Kk4PFnyN ze(=7tTJs!r^RYxu4!mAf1(1&-sN$RJ8--)p;WJCB0OjL@c~=>6plkUEdm3gAUAI=D z#!4G(Y`_RxSI11-(CC#>`&H!yuxzL&!Cj4sWRR_iz2QuO3 zbchoz4Bm~|%(`j|KUO0a0Pcdt$r4j@!)7-Po)_C?@;NdWwy0uzjENVkA{S!9S(ri% z%mBUAZOFXl`ZcJ0R&5P_d$z03weOgykEq)2Tcm+Q5*2~A6}M`jT>((Dvh{Seg@h<+ z#MHYdeAMAf3vttED{J+)*P~l*|HcFd&y?hZVDX?<9c^(bl}PJ3@J~RuX>DF&tvR-)*cVN8JokvPxIRUjzYGRA-v|o4G>~~yeL45`YQEVoF zw-Z!1;#=mneCH<}X;o|kP}3KX9q1_{7|hl2`F#IEiRF2G`D?rwNc>{|W(q?CG6G#H zA7)MP$J)A&5$7ze1Q|c8Ev@uECJdl&0xTEA7z`X}Ebp6zL%vK|kudBSD}U;O@w3@7 z?nkmzOFhG~oE)_^ako@(?2A1Pk4jY4+I091|1op>@o-c5W z%19~}@Bkb0aDykrKSXg_#Yd(6QZ&QQ6W>;av=Re{H=URTqh~JxOMG;<0EIx7(25DS zS|m?<`xmw2k7@{r|LE@Pyz!{R3;Cf zN0*BQEI$RxD0##rMu0CLMiQS<--$yR_+VEFU)4OnO}4xNB55+tM~mB3R8zm%TXX#s zSK~zbqpb}>2!ZwG_C2o6%TukBNE8&WgGR{>;0b~2@sI==fdm94 z;;rN_%QMV#U`o`(e_uhf%n$Ql?}n$hVmja6P)3>@W}{OT_GJ!xtn~}BHc?3wUSt*C zlO9JpAzWPw2a}GdbM%r@o35*I54;$5o_OhUWNk8FRA8xA>5Qwr)a0Q_snv;!?lHj> zQt12!xSr=p%8Tx? zF23RZ>)u95ys!c;)%jMaj8_DqF)}!9rdD?o_r78|IcKJr@Dh1vlP?xi4g|lG5&A@ZeG^61SAtWH>e502V zDG7+~X;MU-LuPhQ1G;42fd`t$dWk+Y7f2){5nC*KUPt~`OZ^XzhS<2TN1?#I;rTyU z4hSI1F?ifO1pl#?7v3iW^dRi~N!JV{co=!HQyww+z*hpwzq^Y&yyRe=5>Fo z`Uc>LU?%(oEdY7I386_KNDpS1&?|s98)prajjxT_W|0j6SOW~22H`g8X;4jo?+RJ) z#{BU@-2d)6Ejraf=CNW!>1_M`vhbI{*Bn0g@bQ=Nn$4goHm2Pgx&(NrmbD4i1oP*RK}ZvRIiULgs%1=3gn z{ZUPv9F&`EM>(Y?k8h<2nVeQGRh7T|j?X8A^efDs2AR3z z6socc3*yt?$2-gw*99{{?AbEhb@t`;G?N^?ly%A?B z>oCv|6?EK#B^V~}e0C+wYG9B4{d=p$(YSn%7b?X9r3=X7uRCTGtA3&6az#5=_(miK z#h&73u0LOXS){?7n|f@`(+>W@%%n{k#kI|iT5iOm0(Nzb=jBjuRLm-h61AXp_gD^I z^q^k_Vffb$h16Mo`?&IuWdO|ucbIvTvfQ{8CJmrlH{KKkcZxBX@u#xNs&HOfh#&uf zcN=c4n2HXGFh>**5;;f`ZC|gtp6ERUVlL8#GVPqcV6YXYG1 zZ-n)kvGI53Lo5Y!fU03Ds@aXxbYfG`n`9#>Pv7oxtRdwY8kRK!!S9g=9YXKh9Y%ddux1nrYG8nu764-cu)wfED+?4TY z9|dIP_(-T`WGPqk`*}P?;+rz!GW~1z6D?cs!wwIg??B4cj^n%3SnxH7(1IR=VLf5h zCy@f5_SH%VjM`VUaOZq9bbb`DbRb9~Hk2415U?V@@sKBo2EZ=4@WEdg0Cu;H8-YGG zI@jlz;EgHa29~KCe1K@|w}{=|u|{G(34(PFQGx_hZu%2V$>3{B>j|}#!p4LsKywO+@9(<_<)kmRj_;`cBDS7jWurAF(=@!#m`VWOr*@G7`QTG);tb2o^ zXjdgZlFt_wv2A9odLDw_u1oIwi`^q0!RHp?b+hJtvRa8{-8r-ht-n-JqQ%iz)!}7) zG`g`d-hqu{gJR&k+Y9lf;z=j5cknh1Q@l5g#-Bbp3DEM|x_|N9bIuCfm77ZY;j`0y z?>kr;{*s>?dXiAs+gt{(^4ljj(vq85{SFU}zrn^Ktn))OPH!*mZ}DPt$KC-{X_U`J z!7qT72dwn$2MZ2)7{>rb|EQ0WU2NO$v~5&Rf#j1BHA!zo<23s_PE(Df4nFvq>8T{Z z=ly(c2gtTU`T@HjZ<5JGLEZtmu2A{S|9YAY|9c<&v(DTk?p>yzYXYo?b(XcihC2 zLs%;p7ZamFhDD5@G8)iF!4BpwDz^zUhNWfB(jpdgpS-3{`)q7LyeUr9G8L|mwp^S7 zN0DSrgLiZ}ECM~o;I)hA7uX=2w9c91dx%K*nJj`vb+l0G%jRtHfFG3#7Ew7N`h7;& z+#kY{-M_geBPhXv;m6BY~J}0e}hG?e#q$|lkJoZ*D zq@T9B3CyzN+!e7{($? z%j?-m$k}8ManY-5ob*yw`Oe;r9@c!K`>0;YF|m2~nUfUzXpHicBjf=sqD_R!+n+-&&@i>4=7M#rnfV*> z20h@7_?GF6y}d=4!LJW^W@0cJ;rr^JmqEl4g;63ARfZ>vAGX-&uH7Z~E|$+hbGYYg zLVoj{Ht`3*K$f`hOA8}q!8tm4xH953re48UYH1_t{$gYak~-x!L8ynQKILrvD~AJ< zr~k1jb|`dU5U3}EAT6BJv4eUDjF3Jc}Y1&vUgvzCaC2L2*LNX;i1Qd zcAaMTId;b~p7F^E%cF1a)C#oWtO%k}*KuUQK^P}R;X7XU?bo&1LziSwTwj?&J3ov4 zUW5p$!{uDKQEIMyJ~;cD0*|-7ewMn42mm28t^HwB{};&`{I8r7_|F%B{RbqN*x2Jb zuY7x(W{$0}`6C|=9X0Sh>IpQUySZ0zML%uoL*cWqW`oZH_TTj_oLc?VeZUlnnGnF} z0h_JO08WsFH*AoAJl9PZ+225wq+e(^o=K$$Q+qd2o+-U}Qr z6!bqjobjHDo}{E;d6o=qtj^8N&vu4#kJ9@%vl>a`a&AN%GQtvq< zx8eaYvoEw>6cF6bOm>6IzkzpXU=ph&#gJt65&8&U&>OthEMP_kxFYhMrNii80Q)(y z1-U{xMda>>GR;-#vDKpvmij8rQqz-^-=QF~fDzHIL5k8MprRnXw?sui5Qu;jK|n!7YUoW!q)QX&z4x9_LP+zsdC&ajdwu6S=gjrZ znK^UKHGdF7%ER9KSKH#p{+Rz< zvm^on&cI>ZH0^oHE`#o#YMD8f3AS(n6*#pM*NK}= z$b7b7cg0E{psEd(*p<@i$l3 zoFPI~#06G9@11NiN-c)r1ZWR!YLQRJ4ieRDW{EC8j%75xm6=y8_`=dym6>062Jl=kcmb)W27^{z1 z7B^4DNW{^OPEVp?Z6$a})cYCn!4cXgTQb+*I=Q)L0~^%tUZu8u_r}15yp4is2eEBS zO7xTG{fSzCbPV`6g^0GOQO;ODn(9RfPY`Ro~mi^nSF8j3Jf`VFm zn63PEXlxAF4G0!yxgY~L z{K-VjaPe>bZXumdZN9_6jz2<9+@Q2WL&urie$s^|GhlEIY2uMQYaL_vN=GBmKUit& zxqjm~6hQ*FQ`^B|LacFjTs>+pyP_+QMqc>ly~&PbO9w>_fAj4A<`n;?!holmJ^kH- z)lG#TSRydWg6>d}=AU%GUN33Ctak@l1-oZ?(zLOEwY+S+@dWBQLUb|w!Ia3YhqR+E z6KG*whSroykx5azJ-aj3riP$TGI{Z;!?rGDuOojo=SROTduc{cgN7RYjugY&Qn;x3 z@5|8<<_B<~gOne+;(S$jhit;akD0nzNgFhtAFB#~50E)b$fT&GDBso6(|w(c!q0W} zb&dG9rR#w2^2-*b@yn`6OJDkZZhVISS>N5QP9Tur=6mH2wnc%(qLdZYN+L|&7F%X1K2 zEwZXkk-18X$YsSBs&xGcfBX0@HB*>Yl<%p7zey}zF%`+vw?LK_((xv9s31FrAV0u&RQq9>dwMHdOb)WGa?J3 zyKKj$vsAcHh3obKgaVVA@!1{ zorJL6vCD3L7LOda1%;v{<25CX*ozMta*x^eC1)ETG$O6XxY%^ce%sz!LagcCB^hp+ z#HJit5v|}i3m_QYm{x*s+KX@KKGD3^QCodp+|19>YV_pSKb;e8%67G4e}M%e9$A9W$? zXNTH(MD5xFVVVsXi*s3$lEPv(^~my8B0(HiJN!5y7SoC<_BMr(KdC7%Z+ftb3!j;H zKCp=%)_tzR#(RqB{b}+0mxudW$#d$KBc%SKsUYhWmu)piS;we;3QD|g1ybBtiGC#B zu;4SR*Hd!LU#!J+`O_e`D*yI!lWjFzg7$+ZV=j9`_xKQY-Ju9eO$uc9Sl>o}y~M6z zk@Gk&W7&J`F(dD3uMjsAZ%lA2lv&B#8E{LPN0u(`K6`njd@+!Yz#CI@`|T^1_GW46 zVu-6`1LxXB=TNFcj9mBgFb|E5IU#0+?FU9aEjsg=6h7SLG0Fa3b7ZIJ_XjXT`|6q{ zMdJl(b(Kf{#}nb71L%z3QiO1}UK(|`8iaq+rE491O`434m~oR{GTcq}jjvL(aSVKd z{LRVYO1_+0TlL=Bwax%k)hl@5kL2AN)tW z4*drU=>A<$x0xnEOkJ*}K`JT>6dkx6x|^U10X@t=`sa`Lj_kO+eP~rr>R&|HV~;}F zi6-+}VN*uOim4oTc6eiUYjTKoMnBftlPHG-c*<3%SEY9-Li%xqZ;TsUalpy`Jf%`} zw3wPyyo$g|9S{slC?bfUcW1_9pjMTDh^D=3+&4n0ZFrI(Y1=W7hn=nwJEf9q0cCjE4lP;vQ>VR{s%p@Sc)^LVzPtv(b$C|wgiJI zHD&c9P5hn*x+=9OnKrf#ioA8xb$y$*aGJeA+ zI>O`GkB1m~*>6I@JCiv`J_7CZwC$Z`z3jCf3N&MvHy0o*9Z(O}Su8t601{Sp=<~vc z*wo|^AkkM@_IV4f|9LL##>vLIv(H$&1L|AC@AL-YA5$CI6PUAg3}%arU@5+zImf21 z-?{hU+y~W?0H((3#A)n{2RLN?aaCPD<_8Q2iA*5X;8l&hchA`wtX3ZI%i$xi3?&+yeYvqIcE}fd$F@f!6q5f0L&GaQ|qZhA0 zQ^rd(LZ1fT#El#Wd{Dp1Wm;}r**5z75J`v-kLR+(o|I;#m#CXvrC8krxKUCzsVG2l zJE-+hUA8e&S$AGXxRLDVe|Jk~+#@)RavnED%x+R}2tU*E{X25~&0b~M4(E`-^D#N& z)bp&no@A((t+qyEcup1aOMa6)Sd5T!r2ZvygvQh-%TjM*9f|XmZxq?;n2dXohKmb&U;LnL z&v8%Z)3AXbvk@CV@a4z&Nn;WQ2|H)Aw-->q(Vb^d02C5*TgQUaAB%i)R!#NFy4)7G zk7W#=64r#Z6x&UshhbAkaH!hWYOx{hGN0HV0ob+Ku;h9-X}9UNu5%xEtp<1DQ|BpH z$-ntoR{P5etV8Gn%36ZQZOmhqu?4!Q3>;g-VO;X(F+ zu;J+6RW^l*pLD^-%tWux^Lf%@Hg#8pCR?$AAlH-U2v}F1C#&p_vad2YtRbr{vZJEi zP-oQ|gfTYjm(D)5{L$5a0N2_;z{ueqrc7}ekCic~BS);Nziw^iZN1y?FY?WHh_}Vl z_*y4C1%46~!*3DYr~Nf*k55i}5}@Kc*hd$>_|PxTa$CzSo6%F0jKoQ-y0k-3nv`kI z#j#TPF|h=XC)D3kv0^C7DU&mjeit>U)!LY4j07wB=JU?o&Z92c$2l*^z@ z_q>nK4_uMSz-h6>uK9%$aOtzr_{3|+gNK46c>7na>VNag9hREAp4orA+cVvper1(> zQw=|~X;6&PqJ39zjESDxO|1GlPYwxtk`p7-FRN;7By5IuUFJNW2X922<2{8NovX6i z#I&4R`93oHbU9#o@m6G-jh>pH9!`-^j+%yqX#^Upkq>KfQ9y{Ds_3@S4>Bv;p5aRB zN}V~QxwKY&4IE1t)#x~641(YzmUh0)fqfqltL0-A zWAcJNGNRs;UV$?r9Z^hP zq1zzR=LBh`u*|Nm)U`zhbMo{;}N1Xz~;U5meN7YlzeEo<9kI^3rMj!@D!IhC-wU1sx`Q_~iY zM>QjFA5yoAy`}BLH-lxTefMbN-K=w2;rUHYsGInf_6eiWTWqiTTcxkEI^U_xp8L=q zf4l9%s{SeZI)9-?!XbV=)-GzZyxfe<(<}1(F)1QbcWXWy_j)cN-0PEh^|INF4_4R= zgA0IfQ?S%c+kt!)`|4Ky>!yo$Yo^O#_z)RRbbB4zegdTZF%}8HDxDcR&JZ^y^h8 z8*cF280bS*p{Bk70mI%-HP+3hG3Y+p5brT5(5F+R#LQd8=AFMMd#6td`0h6OF}}6# zKm(kqL{#3`?S;BAE^;qqB4)TVKlgkY#wv$i7-w{a*~gJNl<%!>_@H|ziuH{2g^gr` z;sk7mbsnf=)$|jXA|LlS8C3QVw6E;}Dn95Dsu+n!tCp+PVbg z{ig-VmR^asUupbU+ddl68{$LNf#o{%w~v>wg&ux}Ya#S94Wd1sz}B?ht}fVhwp`L> z4$Zjg|0%NW?Nl$LvQ|gx6G}XuH`h6LnN1WtKR)z@zA4{sl)h^Wsc2i4a0OS>RdvvPknMODa`gf~9FEi?jXgGEPcW8akTNr_ z^FJ(p+)Gxs>n!^nLlkwZp;|ZB70cmggki^p%}LQBo#;@YlFv$klcv*-XK8vUEr{tU z-dZF=JLPN`v0^&D*JKe`+IXS$Z`>H~m)@`It8nDAr*x|p zFq~;E5*J3)zaVI4HhLeR+xs3ixhQZFI4P~ljpQe)b`MfvspTR7H@ zL2s)u&9B41!s!}t%afpV9#@_q?XtRDsqWmUyHjlkI2ps_$*!d_=3;!wN3UqD@wrHf zgXy{-xmfcLaf|S z4R-$Z)^@s1xSEO`Q@X7VcbuXwQD~=ryzEgwh7OBNXtf!vS+7oyBthovw!Vgiand|t zHo7mNxMzNb-R>1!Ev3f=@%`)SONu9Zt3^m$#T5~2Ly7>LdcFU+x@1y1Vs^rI^V%qi z6AxKg$3-_NRFT#yQjaALCt>a46wpp?9k$;H$n@Nn$z)e^)r3N5effT@IBuI7l_Q<> zg!*`vmZCPl9B+pzm0A0i>l1j3NHQC(9eMu_M$oI2TWe4zGI|Hur%uVHD&%^tuV8UP zcZtnYrGQ3!v-y+yaKs9NEOHJ~*`nU3H=ICCEGc>3GAf$e!|xnMC-SLU$8e>)KeP2( z+~w*tqZPkqvh`D#%{t4zMSFnFG50r)<8`da)Pr9`x(E2ymC(s+&=(sDyi6s~O zq+75vyEV#FG?)?%%|}yW&(c5?yO3$OjhZ&5M4MYv2S6&;s()sD1#~D7c!{hUCkdNS z?-N0dBY*#$pL8BDDtpObC{0){A6X0Ar^Gtd!SX1sj3gekk8G5+RB_04g9Z7L`A3Qt z^(@hpeE5-ijWq0~tAE~{V|=Zu;Fu{fEsma#b;Ecia5B;Hgr8$RS%3YGi8U%m(if#g zow~((E#j$FNu!aoM@THg^QgJVjV6?M6$e?AB3S9jXzKq~?@FlX+HjfHzS**^&7*Tz zS(pv#d1c%x>oivfvsd9mf0@ykPeR0<>_BqPi+k@EL>92b^BASLswYfn*1t$a!jOd+wDTuF$4y&%tML0N=i?irIxh`+vQvjQ zoTH9TpW%KshLtO6KfQCcN1z+s-5SgH?VGZ?#u=bj^a%Y9A@3(;j;JZ2iYbj<-FbuL z)k5oEVL#A?$q07!)Us$y+ksCY?3>;R7hE0_d%iP`O#m`qeq0uVILE0|V=VnFec&eW z!-0vPH~x4;{kHT}UOIcWlNv()&ul9$q_1F@06AqOflnu4Ozku)_V>&@KH3Ev320 z#A`&!`4j;eu=LX5tm#W$HV&pE=z3JR#cAXbKh>TsmoNH~fyfEHY~J}ktfSp#8sR(6 zS9CQv#6Ehzu2ehhDrZE-E(3X;Q3_^_B8VGlE&x-AVFe zVn};qOsYDEBz{|U5&npnayR%2(3ZYSn>mA=u7@B^D209QB1xut353Nm z#boT&P@3nY8C2Q$?iQuvN0oNFe=;xo_38)NTX5XX%B}T43y?N{5!c`m>mo?JQ{ebS zgn7irIKF>)ya@H8uT?9tUMkOI*)i5%Chj(1_V(J`>97udQ(4Wtn{o63+^p&Lr@mLl z3#bgskb5CkSD&V_ik>o_rQaI5s`kfLiLCmhB@r%7hfnsIoKf0z2wP;$6`CORl~z9P z_^X+*-IT`%O6wRydXbTNs)SqA*edAs`T^)^VqZrG*^z{L+t0-x$+af8#iQnj#*3Tu z(B%B=l5*L<^+%n*+C(V|xw^zgjv<##Kp2O;RBJsr52Zed7}TLBx82zR(6JAzz{=w1 z+zy#2Lt%O4$W1gwLbpbobP(`C@9r_Zp|}Rv!6t2Rk30t;yC%Ek7Ln-Tj*8(G%K-{A zHAm(6;yAsZ`%xT4@VMkwG+isV_ha6*yYbXHWxH)vf-dn5mRW^$K-*c6OujN@xmbC3 zKjAhC?cAoATiBENl(ANpA)1r0)8k(7TXD8a_fF|c z(Ng8jwt#?zIcnz(x|dszI%?-;vjlB)e5C-bg&i633NU=_{T?E$;xbd1IU{1t1)`<)~m7| zJTLsLg5l##o;UgvjRnz-G0NnBeqqFsh~`|soLOt%ZW;pBKiJs z%^}@UjivQKio{1RRkgAo#dm%?T{;nXOLIw6DN6UY`Edlmrt0e4V5EIuQGrV{Ep63J zV>p&-Zr0>T4Jm88SFsrn?!2Ya-*CN(JlU+e6ja6iyT+arJ0;3 zE}f)=@3Fj^Vz%@ehJfAh*gEV2))IJxY=OuzB$ZVZ(bqqfvLXqgh_U{p1KEr% zgA(%H8$5W_K8n;P&2y zc7y+B(}%DS^)_VpP3yL-4`Q}{AxmoR&3f9okU>YH9+nZdz=+-1(4`4y+QR4&Cm$QR zjE^mfa7!>2aqt6*11~z(PnL`%3a<6)e0e~SFZedjx7pzyT17uSz8d%0PrXY3=lRT@ zb8KjNzoubXY#kpYX8NGZk!3wjR?~uy1*f7aRj@?+Y2Cr1284Kk2F}1r+gR-6{+WA$ zR^IN%e(-J+?yA@Ygh#3W}>)wY5c9}3;8YKEd}&vB2uxc%NNxua^B|g zuE|E_@6{9iSBbIwEuoe_zPIn6CUp7NbkE9>FEL`ghjB6MAvZ0MgZq;zhLhUZ@<1`P zpU@;^c>ZQDMbq)4or3YeYWhQ?#P8x|bdeh`3tgN5@!()G%1I6&3n?XilV@ zW$y4u6ytMYpKm?qKY<%*^rM5;GhoQ1+ox{_o? z!4a!v6c3bwNi-@gA#-Oq; zR|!ptbTBR(k#bW=-$K4lm!H|D@5Aa=$>t3aQ5pe;#qtsNkg5SaKmq9;IuB+8=iYTK z3q(VY2`atzwf2?0N@=&OAM`9vUrnn*@)gzPkO!;~*?Q!n;eOgtf|?ios=#Cl@j8ZM zY2V=oPX)&t4LY0`Eq>D7CFO>3#-=F8)UHRU$wNP*Ls}(Jke3-2um_(Iqo~@fEgQ9M zZD_~VtCAKHw?SW2ycpjbZu>L++S6!A^Caq&#X?uu<%^)+VjqBUTv(|A>N&PobrhwM zRW$2t?e4@^qf$Fulk_3hc;lyAt42~ulVFE?`el#nqk|3tdlXO!_WyFVM@d$3({gs# zy1(r)JAqcRBY+gbtDA3|`I{)8?CxAbEzqD%{?_z5Xrq#C6TtDX*thbJV0!%d80@^~ zd#8RBi&X14Rxp6*dB5D2N(u35o-NB#LTs|a|&PoH`g(V&Tgz+-l%B#tO zkABMLhW$?}BRrAA^85#|)d+({T-8qGZvg`FjK*u8b)&rE5HE$s8n$~O?9)pP)l|E* zh9{naOsCj}kk3JW+H&$cB#~92V+Gx|AwDEh-#M`0ykJQvnR5lb3GO>;5$}L5AwK<$ zx+k$OtE@dbOYPG{X32i9lD}@V&ctkZ^PU`-Oif-l@0H_-bcjlQK{sT zLYPec#+n!OJ64|}^-=A^Lg|p3NQpK5HvdmL6C#q}kE>j^I8W<(>#1tJ6HH#3@Z4tI zA%eYlll$(gxlB%x4OHk3 zC)4iyj@%2H0v>xmK*6om;tVwx!}ge|^X@3MP;E&@D>Yyhbsru zy3mrJJOcxYKBE*qIS0KeN>EwMQix5o$sW|q7|(cnFpaT-PfmX6lHKGu(cw57?J0bh z2TSaQ;zl^dsdU&4MI=6}m1+qrvZJG|gY0^YGXI6>NU$MFO^VuzpI9(~46aXJ)Ynz9 zskwJ`@?FFW0@{mi-?Jgi2@uvybRriHfm-9?pLF4RKk1Iwq=Fa@NNw=}wKUGGIOa*- zZJ30HLapd0RMmNO0HDd1Uxt(RA%FLW43KqBkFZ<&fs?Shr|^R8V}w#Dlor}Z?N2&+ z!dyw}BIa!O!!BtrKk{81kEFVji#)rsn%Bc{vIqt8Lt1V-T&RwwHey;@UjA;{OV~@2 z`9}iY@vqMEGoF{({`RK8yqs)R=TbKUfJa4J8t35h70JeWcq526FPbrCt! z3*P2Ni~ik7g8d_Zd~;`0h-`X@(z>R6gT}s0@CCS(HGs47V8mH%NAej8UCCRa;5t)t zrmJM_COCO_kM{lRYNgS*0@MKGo{gsANR!qewS$cz1sjho$%nULrq>3huczHfV>e(w zFkl=>Q77`?mZzRVrgATteSbSNIdjv7gB*q(W!?0o$R2(LDm}$15#tnF)E^h)+SJ8! zy#|z4?>^>i`7}fmfs3eU?G_ipDF6{(Y`*=x0#YeT1SKkqMR)(_18Z!L;G5p`sARJScX$?|mQBa&1N zj1F_!PYFPGWlBP6)X*LN_04p|`^?Fah%jH{LStalf3$MgFUEJ{Wj@OI_U4iKJ8G#3 zD!4q22HGK9Sw!(%SC!s?ta_jAnBF_JNPfz-@mDoF16!AFrOze=<}fGkUxFGa2PUdS7*$2jAt#H@CaqRXs((k*E45FF()VE4}58KJ^@3Oq1zv%`P11xEH-LsE#FHG6%PF){Qu8<4*Mr5$NsgH z`+<7cqm*5KQb(6Kl{1F2o*uK>P~UB$ZrbuB3UYG`y%0bM0maC=IaFXFotf}A*aA#o z1~G2x;*enGiG-nLrw8&J!>2yFT=EIs%8!~2uz4FEe0loh3Va}OIwx7;hv`?^netI& z2HiP+YXl*tLyhjJol;!=^{1p8?f=#3l*H*u#f z1#(%_eVt|GmEEHz^PIsMaiA^ASK>Uh_HC^?W^k}u^~1y^$}+xVN_VvxPhVeIKLoFH zmpNUrkLTJA>{B1j6MP3bH=8k+X+1s{RqrP6b_vdVMnCbde4=jKbD-Z}Sb0Z^!J^j>sNWG~@S(37sJ6aLEl&Uq|9Tm6>D z!>w#}Tm0qod5#ShBArScX}V~;o&7^yBm3K*k6zcSeeM%E1lqufAV$V6_q)Hnj1H_Q zEWI5Ve=xdjQg<}=V%P~a1p@))K%*+l3KES&mo9Vya~|C0!n2U&Skeh}(=&WDA+z(A zUDRl0y>~~<6Bh-y0e`zOB`&f}-8A&rlFDgf0CO?cU5~pydhWyx&=vM8I~`L=S?~Q4 zllYwbyeQQgq?O~YXl%;5HV}kMM~{S3WT+wRvW;IP6;8*~oNvG@E?Z+H?!kgi1-@~Y zL&wt;Y2CamL`6G+SJ%Z#-uadOq*DNGNT{n}Sob1$;!a9&-0E38%aW#_qGE$xN-Rul zWCB_B4Wf6qyG~1sdR%xajm;rrb>hs{TrBoPF=}Xhx~XndU(U~1#C-UI;)e_SZBXku zHa(nF4I&?Y7zia7TYIMGhi=d*EvQ%*Ip0&U%3``UA1XYV&g$G%-{{clI{G;^bY5Zq zADho%E>k~L9nK#?hpa@~Hm6)h-3zVO-C!k{yN+G>VNcQgFtW(THxmz`xCQYvrTfTR z$WJZ{Cn=D4AuRXtO_oAe*DJRd5VwrJKg%*(82Ttn2!2wK4C%SI-C^tb((P3)#JIfM z@$jG(AK%ixJsl$fM8Vp3m76pxTeplMO=>$d8*{4Po}!j&y5zB1ejT!KeTz4yGuaCr z0kaY3Je91=XaQrfZsv_2YaiZCAoF5{+ArxsObkeFTd)i-N7*|o+#k00< zWsV0jV+A|#R>!Xt$1t2l)1{=?sQz`}>-x)*MTqeWaF(1^o&wf6>U95o-H4_>ti`^) zCe#Nh&^+zQ^%Ewxul*MQyWrYEA3!PS!O>C^@*?g1IYtV_;Tf#EO|LkRf`Z8cqWrm= za2&^u1-G^LXPp4D7;smf05-wy1=5vYV@n{F{~uOtJ$q)qt}Pvrp3sF?tYLclzTQP( z(-U>m`-vieXP3}vKk00tEJ>bIK_3okS000l+fW$zJ1j?xzq3+}YrLS@vu`YWv$4q; zn`qW)WsFKrrI6zn)gA=XqvKQXmz)oy(wyE@z;LG?Zu7>FkKzxQ!Wq3}=C+yo!f%OB z)%a8GHs=YKaqEQOusrEL`0n9V;2Pv*0+weNvN$7z)?xqneeK|;=WamvA$AXR8U9Gwt)WYr>7TRdVV^a8;R75Je@U`i9P{uXb1Ac(>yYfCYdYGHD zIK=ZNAt>Wg(2=0K5xTb{0uttx+^0tmeC1S1z*NYEobpR!)SQ7#9^7HwWyPbWGhSi@ zm&CcQTXw&ptuYr*Sh{Z7>$?~$^}kBAF~Aj+Zmf87d~(!f%&qzYy-6{n=JW9T3w?#1 z=jKWQipMjNR5iZ~Q!q|qbb}}CNxwMfl!NzIxK0b*s#zxgyJT$vglk0GvXy{ORCk%` zgv8r$+qG?I>-&#UF5Z`85FIzGv2X8{gd~lBI^<)RQKS3PvmnHNtF{F>9EOz1({i(c z1&O}PPm}k%P}KYxeSc5q2QN;0ra_}=02w6UlQVEl!KwRER0;}!y16Rprn}IQc9m>n z@XKqIbtMJJ2?>C>1MINl-#~bo1p^{QH-IY*zKV?mGED~lT4fusir`)V3iFXsT&dO3 zK)0wKOF9H4?}CI8>z=Gr2y%-7wzC1F38ONNXc{}0NV)JCr1P|!ap9gg5>Sb{0<09t z2Qt*a*ZOsc+rs1DXme1*Y*6W_H)0(@a#W|5=WEc8CJ zPW$rWH?PlfB}U4Sfqmz+9p@Tdaw_PTM0QvAhFJ~9yxj-BqNV}=ge@3vdO-}C6vQ=_ zsjQmnnB9-KKh`Dt;MS!q#RN(VH5L2E&zo4@o0b=R5cWA;M^a_9Q-@iU*E#fc`y6n91ZIILnV3Eb{c`sd1a1X(S)&NF;8FBE9^FwRnxML|^iP8K&` zG%f15Zp@<+!!Ac|CrCGGZtz#8e1a8Xf&BFNelL^$`h?NXDQ1-iW33Q3r9Mx{-;S{y z!hB2A>Yfk&7JKkQnDwcIo6M$|XaeUa|3>$EEi_cRyKa&i1eo)-bP>=fxoLR9V{} zLe{@Cv;;dOO*Sn&_~guY|4*v$#FTvH)dqC^a^m^JR^@z1%Ki&4^9C?JVpJv$F|C$S z-5XR4P1wBjgClOLrnHH7y?7txF7x#yW7OCFbXvoB5h3y;(P(^%2}_p%07&PTG8%ES zRguXNSHJOB}LWeQW+A_%wphS&Yok3D?cPxcyzwVP4bEw1dmFBYqj* zM*Lw&UXg58uHNf2wOT!iuuCc&M7vwm3>E1c{!{eB_9AnB6B8G2o?LsgJR1>1_MRL+ zLUgPh@-ulDZu|P`k*^%0{n4!~J_$50467Gv$~YLkHcxat8|pDQ_qw3$3ZPEifmrz% zS0QVAsoB{14PElKJrHl`8wh2IF7#oaEMN}(qO(GEA*SPA)k05J$vdan$t+g%&p6$& zKNCbdZAR@J!*bF%>?%7}@c5p`Z2GZ5gM{zzECXsAv@$`pGuVP*Z!%|&4FphA*q_>L-i_WOU*70Zt8xL z0g^#cBHOncLAPZ*EI6+?vx#M2z+!RG3P@;J*jBnX#`y{)b*;E5i)=ZtH~DI*w68;klr1T{7naT;_CK_-S)|=z4!WxKTC7DzIg-v zF+GG8SYGWI>}UXOG`jEOM!RQo@w2{bV(r7EKC|1;=09d@xbN~` z!WD|%$)5DBsM;O7Idd|BasUF$dC30GbXhQx@a{2u;9h?D5yBj@f6NG-fwX}g7oDpR zE0A=35C+}p?eD|fzgqD;d7C7HldT&c7-I@rx*C$V`GvNBHG;Oytr1G_dKkN|+C0Hs zw#@P3e&6eU3ijmCBhVP2o)*e5O>vyohJLMNB&`*kk2rs2Tb@5~<>-=z?Si|gR<~wh z6peHhWU8a=@WX_s#C~tQthr2|3v>Dq%Q?7tjjX$&vT?Q7?sq$GQAKJ-!R*7qcTPR) zgybe6e8ALL5gu000w0x#=!LAMemjdWq<@|LZnZ;t+RZ2;F5P08tzh7e`h$eqfqE#k z52Y~oZg1k7q+ytJwvnOWtf;Ao$xw&4aE(LYXgQtn2>b~4zPW})Y4BZqA~lM>dE*k7 z+(ng*>yUeP<0|_A9m4VX>bZW3_K6fj%b44dv(!n7n8EDz)dv6PS*H%^Jr!9kL~s}x zN{GzTZ2D)-6mqhxRMt1;#2Emj?HG`2)c;%l9RqiPw?+MAP)O{SU!TVC6q2r~<{zWr z99nD`xt*WEo~?GrCJwrzF{|3&k+#)Ydg%ZIPITY&8yT8 zy@`|Rg;_`CUOfW{s6JK#W|{5#5+knuAmsIfU11GMVeFR$B!k9bK4L!1)-Dbo7w9(5 zLG)R#Bjs3K7c;vV648cbzxm~e%uNsK4{l91*!Ti( zIV@B~v=HCaAdhCDEUidSQv)c@d6AmePMEg2oaOCAnh9hZ^IhnCW9Tc`uFL zVPf?SfWP!ect9;hGfJ8^R|wof++Mb5X9%@R%$%~iP_AMTyfvK`KBQbos;6Is2f|8V z!6lG`e$LpQPuC*Nl(adflU_G{`JhrlE=UDv1pi9%Y|epg#Ez=K1UjcMgy-F-zOT z<7vxSz^i)*VJ<0bB($cyrH?t|bZYvYCJJlO@fk6m!Yo#dw||f>UGn|XgR$@w;?ZE% zk2n_x{J?kif%L}<9hpc6(1Ok58Lkg`sCII_B)pvN9Xk4=?t;b{5@?=96pDMk-nV0d ze=uTd=_u6fAU}2AVV5cbI6#Lfg2XcNbz&kJX(ZOE^R?NfR-*Wju!?pwwarO+nSh=| z*0qaIqF!92h@?+O4amR`I$Ug{9(4$;e%x!Dr1IOgU{s7&a%V2{E{yC$ytb)HOoYDj zOPg5=K8sxrb3&tmq_yLq?v!>g3d1&V$VvQ3T}br1@Gi`l(_XjF0g4+@)e;|2Q?zh0 zKXm#)OZ352#a*s#sJ7vw2CCDz|;{iK?wT5!s^{q$+JdYZ?aZ~J_pQs%Xa{Zx#9$BSbVDE zseQC+--N%WsEbS_x4(n5&Y{PF9sotgZh*e$ljB%|($LfcV zCY0Zh`yVB9Y5x{vsQF@8*6&$Ii`8m}&n{mhwiMRj7-AG$r@~X{FT48g1fz??kgEHm zI39=(VSSK`VBS~k?~FL-ccpA2?}K9Q^hK`qoR^f3FWk;-2(unKII61g(lyGUqx&0| z-+8i!1Ko7DwCEW+My8&UTF*a+>-#u>J`m8GnzT;$dPO!lk7r$1O-^Lpw+{C6`xR1XNqX!U5)-rQqL{6+$}y+deEvwo2lEA zDXe#wn8;SwYJ&4-?cf_r3=E`K^*9I*sx-qIMALF^=F&Kj+X-ilLi8oi*OqXD_#@N_F^j3HSv`hfSsQy zTIa8bbYKB^#)x$AlP$-F$aYwl*-)77)T}RvOqau0aPn1rV(1vk7Qoj?T*@c7njP1IL+-F$8Y3qy zvdt+J!<)`w&)-Q5qGws?3pDKfguL={IP#7sU3YdAPUZ_+h?2N<&J~yF%qGgg7{XRw zDHdg?@09qWOs8pUTR7BN-xPGR6gdYx{rA5l7ycuEV|b{=ei{TQWU-Q^jDp;)fNr;^ zIckw#Wk~NTlg+(e*p8n_)Cu3wucfKuT?30B|qt6@l{_3?%0giG}}tyZhBP? z79F|qz+;@LSOyObyG%W+ezUEKwyQrj7E|BfJ6>oU)R>89615C6))|dnoKrgkas~@G z$z0ohtR|zWG+Igs@soBSftcm?r394HlgbF8{nkZYVBwiexept$h<95vjrU>X&=$n@ zJRa*txbhtY_OK9AS_37KCJ_CYR88<nWpUxQb&ZA${6 zdT&D*3)gL4B%jTF_>=As_0oHwk;poUGxtva}vEO!$3FX8#L12b2AgYP8D4NRk^%Sw+marB^!8$~0E#2)j7>U1BYrylW!BPlN*Odf#wYJ%i2X#=oMF(}iuj3r@GNMN_}4pr z7c9YEt#*1<-7+4?YPBk{-Ts{rIbs}4Ye62Qf9V+Ah&dA=bIFwI4pV(z98*Z)BtCHg*lvYrQPZ7&ajZA7z)0>=k9&H-hoR+tikS{M>k{t-sh`+|GdoJ=kI)Dyi|9l?*eY_rKxjKSV}s= z6j^6QQ3K*50BT-=Qyfr#IvTCi(vLO)*hWrcNM}+bo!+yrr%HBZ|2(M40tDa?bRa1n z;xt(eQh(qkIe_}G&YRk_WRX3&%4PbpE^{bnVbry0NhPN7Dh@e0TRi?KSkusT+&5*; zZZqZA^>X{us1l||B-eM2;ssgPk>%98MfpYfFI8<=>I3pw9k7Cc-}TGCr^txc!*{_w zBmy8Djf=?NS01_i`wH9@5i^R)!54~W-L?KZlNtO!lUeA29zpO9n^E(4 z1dQP3Ye?E{|i+zVH42|+Rk1*Nk z|2rlU4@~G^CUR#`^>~itac9n^I@Dpijyc-|+neL-k^q6`_c5MJ`E?BdHWJi81m>g+ z@=^m>;;3{(ykQA#QbnM4m)f-W2{i#dx*$hu?e5~KmKUWoZ|7B=*9!#XZu!5tHmXx? zU^@G!%z%X<3jf|k;D+tnGey{T;_>+%XwDnsneksoxwM_Knlv4M|L~l6u~47%Pr4zu z>4rNR<7=E#W_#C6wjTTt(S|VCna6N-ibratt_ou(i^k;`g0d#sMIy9`S!99pX?&Sg zUNUyWB^tVQumH;Oj~G;y>xML+86NE%?~i3F|@Y zm*=@&WI<-G5~o$FS4xAb{Odma8T93qWA7m@DeX#A>PfsykC(!-r^L%^Rt-LIcz za#R>sUq^aGh1NZ3HmdEZ(Op>uCv_s_g+lwAQqC|!H>0FPP6Pi5HDAxwLzA(fPj8Y| zJ)=f^3r*s~o@P@Z^2URK5{nli8ms)Y%}s6(j?P>K-LywkTWYl5O;S_cc31k7(xl<; zwkXxbRk)$?9C`&lCFPiBt8lT)*?o4!4C#E|dI+nPy_*<;gBrIBH;D3Lk(84=-%m99 zKEWjV=Bi8p|7QmB>HijT|G&N_GTN4JdPC&hA$>bs$CN|f(0rKVwxIo(oj6V5y|0h) zN=gHdvy$>exW;W)l&Ar6T!#(lNga2Xl`?BSD?d`-zlva+}K8b*wqd3LQm>h zJjn%J6kA_|ZWi=j!3~>`xD$s|DjuI4@9)AKx?1sdTbgv7+ip_#S*1)EmVfAdVmoxZ z#lICaw6$#N#SZcX!yPpJV$mBa;=x(I90m=rg3~ldwcTV$Dlyr|-@yDkpihN{yD;)4>?9}`dy9*vL&p~tIxkPMP|f1!>k^`Cp1w~! z4`4r_y>fC{YA&~n9XsFH8|FsZG$|p{WKS&h#%T1W^m?lwvdk}D5C=W{SaMfqPM}8uRBRPh z2-6PJHv)xgx7q*W9FTtmkK`Y6jD7#b(uRK(ZfZkZqL$HEX_*AZ-7mM1^G`flm$dFiMycDQS=c3lSJ)k27-pN zKtI(WAU!+p1B5w2zybxl*oaFYTpJvSr+_esCl*-A^(fK6@`Z2&|8fKGZ6SE{{=Kt& zgj)lV*UALrW~wfkWZU);G7A68A^*)O|1qp4k`ObXBQ%CzVj+P3^~gWM|Ml-8+>xLL z{rvxb#s41n!V+kN|8i-$pdfKa4BP`yAmbP*dHOfAt|U&~5lXZ&Ec&8iRCw;f(?jiw znB@UO^;pG6%2MX^?*q@y$DZxx#okPr!Z_yp`#cSD_`n+y(on>%rsr+nJ@?Uu)iz6I zl&DHFae#5fiwOZy=h}$*$y7m1jX(d3)Obbgn(n(A(@fu{$FTag$yEB7ZJj6f3$fB9rx0#MS39smj1wF0Wzg zPjy27K`Q=bL<%*;D;rHl(E%w2?{O^P>^^dH@#l~r>Py&zcB;YgsNXEM9$L!yPAzUe zY1q=d7z${fy%u4F`X6s!X>}h;WX0<5&M^Nf@@Hykao`344|T{9LM8ZKJF*IHt9v<{ z`uC* z>EFt^v+oZ^kT1JTK$--_)HU%Z&mq9Qd%C1d0q%|22uc7Y&y%?@@OD-8I}d@1`#FPw z#VZxacH5^>&kk6Q8AE|iu~U+ogzp{nvjDPB2*Kz`!-*fU8o*c&zf>NoB4}rlaOb){ zujR(4F?H(=D5}4VINXM(aX6w3it~&dvoDZ3>j9tordn6u*oI>mGWDpliF#yJH=T%| z7F!~uaXmGime_;l9q@2kyD(Jv3_|VTa;y9|Fy_zCk$7?jDOUIUmc2VM0S=q+ucg1xuE+fsQuQ0Dacprot1 zbqGJ?Dd1Q}-F7Wo+|VxS*?k;9VftLKo-}^vi%Pb4!}WLW8i%uK96T_mS%<82G7so~%$cSR_wgM@QE?NP{FAD8VCz zcOlf0{YnNOiGypG`CofJ_q~ob>o>S^J?a!2%K}gWgYlS>6>+NHlS^11vvtl)ca=!E z%Uu|+jX*zZ2a^2!KD~fi<6}eZH-sxs18R)`j^ic|K!5#z0*PlX(f~}WzzdppqQ-gw z(L#}$Jq%gFI#UDU=@USFZqq;f80xLP{UM85z8JzAN){?lErD;Uft?i69>r@(nP-+C zGr#h4TXSoS>!6rHe54xe4lq(YW2HJtPo4LB?D&hdx}7;!?;)GNpvNGDcV#Hq`Lpcx zy-K4hh`oVIFtFnCsN5to;)7pJUyqAo`pIlko=DbwbktZ$<@G%_L_gCG_*EH;6~ni- z?Z`+jMSu@tSl8W*RIcJLARO?DUev*%)>bxsFb2QR<5Da zUqN%x^?KWrAcyosaXa@k8T;I}Pn3I?ubDTzftuls4>@|1?I(mPRfVR}mlr6xO7Ngr zZK1iH;1%1U7kdwP-#sqhQEHhRq>I{@k%pG&ZgCckroG%ABd@IRIMOLCl!l;O4zvPw z+Q3)|TTfayQXtOTk`Ypu(GPnLx@-a++#7) z<{E_UCWy%9sl1vU<`Laz(44EGil{zhC3yd^@Qb54xSmxuEKO>tnx5kdQibu_ff&(t zj4`KbG_H5Z)T{^#t}@GYB9h%bCt?zNd=YawTeP@vZtrDwv5eQjHZ$ix^|EeV`=M&yM6ZbM#=yR^9cnH#ie)>&=lHKwswDTXDK0$N?CrUrO zS@O=XcQ!`Fm`~-KtHVky{Vd}a^TLM9*88~xW_Ql+*S7HrE@qyEbMFRry>F#USQzCa zDxS<3_tHkZa6G=1Z$K2DhttFQpmg8N{bw~CE>zB&_jO>;9lBE{xJCED1YJt@OJd23 zX&$`jib)yW3Q(E)5*6LA_ssH{PqzN~f#t|5?we0PR+^NkOacqq$MyecX8B*eANapp zSM%?Hw*1RueBn<$55JOxjQYE_JzGcPh*Y|i2 zXkza1WgvSz9MivFHQ8V-0W0_1&YR0FDh`m+ueVShA=4H%6|4@jO-r@UE;6aR2I<;U zl89P(Q#E4NSM}V{pIRS*Kq>y&_H%iW%HBylG?Zl>&H%vFfjXCx`o+TMEt!}dzFB8Z zBQM?s&Jxo_zwV!dOUd_+iwP zXQXZ8S{;9{{n%X;NgwX{qcwB2mkWEMSkC^V03ktrF{G$fB)}k_eSCU}U%#AQz7TLJ z8Cd1m>ivAHsabuU_VA#{AN~C8$IO&5M5mWOtmhLY2(fli*iFn)h<2FeXzFJNO zS3@KHd%ld?>yLLF#zn(&s6BN8B<<*xup&;1ksdH6-)dQVq9P#_hb8mDnGZ}~B{NAz zVPIIl$%X@**-PVra}*l^L3$A~F;H{8OlC_b$l9O+T`3>0CQp{* zNl%Q0j`}&XJZT!ttn1Xmfj>!|S1&^nBf0LSOg?>V&&1}o?wHRoXjLwK%ZazkskoSl-nRT`EBr$n4!}~x6U9T5go()TX z7%VqTuSx@1F*6S*3QOqkW6rs97&N2CBzmatwDq5q4HO8XtDWK=RtguBzlarEaJjF*L5_bZ-V&=1 z4Rvz{r`W&sHYr(`%i{DCIrqK5Dd;_kk1XrLOspZsuSq#%Tg$Ho4#lavP5N^jakLI- zP2_IPo452zLd&6aHFWf^l`hzRKh@H7`nGJ5`6nOcl$veME(!`}=fn+fpPeH!qLAH$ z81!NQ6YX8Pui2Rq6eSAVUwd9hbcC-IM9l+y`x=v76=Kv0Xh5!sLK`)F=bK2#6y zR||-FxrWo5rH~&2EnEFL7v{jHdo{AWpEE6&E6&TllzTn|+{gp+w4phgfEaXyxVVBo(v@7w{JeqjFJCslHFf(>*7~S(l{|JgQ!`J8M8IHgN znE*ASMD{s&ozherbwVTJxV5;sqq4+N`f_mb{0tZMsvY!7&AEWg$$9`9D+%T;;x?{g zS!W$VUkBTO*zLiLDpL0{R-9@#s=DP?x+3n^eyQAoR?9v-XQ#1Nj823r69krsgdDZm z&pMNEthhf_rvI%^RA#TERcV$$CFueQa%bm_1+R-iIrxZ2fgC|es=L~N2hTg95|6ik zmc5jMNw!0>gx72H51I%$81fBm4v|lY|(rL=M`nFi}R{h93mX*nU-Fj$j-5GcY zoZVLY+#!=_MhMq}?qgD)JZ;T>}X4=uY9gq&=)ch4OO135jCr{H6Tm0^z zoL>YOU;3<3e=Un`(Y6>lzVi-U%Gv}CSJ18;kVxYEP~ofDekU&O@beX}cRdC~3`x#c zXyK%UTkd_a1At}D*{iZj${6BdWjPw02L#F`=3fT4LeH4rDWK=22XG=hyT&-k7&MStE5%h{~kvZ8$_56R~R0h@=~D zb_CPu-chsouC`V^fR=Ygp~ppM3xdye@DC114KisXnom;?f=Wm-u+c*m*^x|~PYjl| z1b^iSbZ6MdC6SNDm1c2BK`II=S|uVgCoFFB5*E|{=nU^AKvVzh^nZnv+~09b`2R!8)N=v0g7T#SAB3%F?7R7WqzQZ6p1U??f~)hrTmX@M0O1Y_^;;yvS@K zb%eu~DgyA1bq_m*)R&^+XU=~MiF&Qh$CpuHPkBViqn?hvG*%u(RzB@wx$H$xd8v8* z;Q479F0)*d$y3r znd&o>FFb(Ily0Oh_T8QX7_izxEC!HeQ;aDVE`#LQGGL(~m2ByTqF-Syp5yiwGi1X_ zcqCBgzaAhzi|*?#l!sj`pIvj#G#@mvb|GKDDz{le)4s9r17dCM%PH;`5q6QZ4~|>yTD01dW?o(}f`dm1o2%L$(k#By{;gBwXWiytZ2&`s1yr$$vJ%2}T2#33wx z%HEmTH6ZI4mZ*9oPdr2QilOS?Qo0KKO+(W3jrIh+JC4tHN^#@CG(N)s%G`Y{IG?Q? zB)^GwV%F(psG{pJ+b3^i_AHroL`p<@8jGIFn{&G(F*+;K<6SvD#{C5FMf>puQ99kC^wqd`9O^tC|!Au#Ug4(6?@3I;x&ia0uNAol=VQ+R4o zn#fnwuzeQ>%OpmTkYs1uy&-ZRC4$v-jpeVT41UOIy`-C)?qZafCx)P}%wWR8T$Xyh7cMAuKj- zmJ)p`Xu1G+3J93ze+`<*wP0SB|+A6VGVKgb06hvW-yaNuc&ckzK zWa&lI`9*|e$zkP2%F%BcLMi35GgagaS#Rz|Z~ny{Njb`PAd0K{N0dnVFit4!CfW}3 zR+NIs$Xe^H0mwP($(7)iYSQrP*t75P0X$aDX=>yBi0b6==%h5qL!bBpx7OPlGDHaJ zdEny#m1}`IE}QZ-B*m+=pp(v1Tz+#Oc^FbQEOE(OruZ0}pCiW0rO7+x$0E*?t7j zwrXvFuisTbhc^7C$%$RM@bFrfQB_Ip3HYe~f<<)J1Fu;z*zDaGV+6BckC2AT*CMu7(G50cV0Zwa6(m!l{)j$UmxPg{xr>Q@Q6Xls{2BXlg zpRKud627{XI~Al;T!Vt_hBMeZDt^7fe!TuMDQq3GbqWtxC7RC$RCg2wL%YA;V>_92 zG;I2z;53~NvJ*TeRPUb?OO4_nffHRgY^`hpB0m(><+AT<_Xsj7$S$dy3^y zLWWA!i0Ud(xHRID-B0@}gGP*tT<6aPIXNXRp3VYhA!%6vV<7dJrnU)3_=wy0Y+ z2dd~hp=dMja?#43KzlrZTjs2j!ic~55}xD96!UrFaww?*UgeM7W_+V0?>5IsYrOg) zt;AGqk?Uf?AKCoCIFb~sy$sDLSdI^ig7l0e2+Co23?F1H@!QBhRdj;}&R!}n*ECAW z8ng|Pq`4OeV^*ew>c^!)K8UBJZ z_Qd+e0*E^ri}yIM4icPQoso8DazC>!GMjj#`j?2Tp=Q^EeLeuU*ut20NQfHlyi3Ez zeM-zP=lkCglKod41O7il0rDF^ZM>_ftkLj3<(%%9XG2R&6ZN;d-QCg^fjh$wlc!Et z@L^GBYrwd^FIuXKfjE}#RAJJuhi4j2dKKE~eVh)J$!}%Yq}W$sOCtq^y*$I4ie*ed zS5{f+9c9DYHlU#?x(%Rl&;pG@dujZfz+$^iuuAGhS-=iSRR0Mj9{vi<@Nm2k6%|$d zJZW!kpo9V&G5Lp#4%XPtIX!5g@>cd8@C(aSvaj{MZpu)Oc27Lgt<^D40ZScbi1&bM z|9I!xlL%oPqWcq0H_(cJD&3i2;BJ zdVrVG)+=)WZ4``cft?u%WOW^TQv4C5X2+P5Qy7?M(Sb8<7Rw@*P17qrNBDZ|tF<{E z$Pm~IWktmGrWKF|c}G4B)&^Xj@>KQ!WdCZQaPEYLZIDc9J}NG}Mc+`QAF{$0vatRK2~MBRNwuBM8jnLEj51fS4vUfqbg)9LW3SW286j!g2_kM zgEh|98wl+qn_I#6DkCTd4+&+J>r40S?;X42`xSA2iuy7lg1oadK6R6g*Ysse( zlyd}sw70%55k}Z_v8gp|8^gCMY@dXrW06F`;p>AZX9IUBJN2oM&B+JQgmVHj5AYl% zOVPcffS{z^D}qGs`mJ%b+r98)na>u(IN7@j$Y?QVqn9^iok7nWeF(@-w?s;s@UtMM zUp1V~R(5>=0X2y9b*+E(ShY4P@TL|5lR~bh{8ZunL$h5tiJMCOTsBbBPL+`1bDs6& z8^W&n#3-pd&4JHbILL|>^F9F}ps(Kjk>XUHSGHI%9p3ih%&O8b1+1Xycah6fi8j9*{lYgAKINQVi2R%!sP3$~Ux zaRIDajW_QOFxj(B+$#R2wnGtJbT?xyZ&qV;8r$T)|wM% zZBB^ij3gwxEk(^Aey-(!Wo3R(^9uOASOCcQ15U;6v%|a0$$(ncX~5HXzu<#Hs(xVm zZ<-pg0vU`2J2FO&SE^X#EDCQeHM$L=lSG4E#wF$#Qag&S0vHlWMoN5*G2ZFqex&Ux z5l-tNFFKTe25kr&>hQ2TmejPD^^bp%^^C+<@CZObnjL(&E`>!{%D9%~rSeycU zo(5O)VdwsuStS zce&t{lMnC!FXj*r>qU6jQ*H@&_x)^Q;hSM00fF78}Ql8!1Za4 z=lAHhlo?B9_CRE5a0~qSJc4upAm1-pST0~Vi0HP#i5}$CmnNrb+s?@7QOp!wAH>_d z<&&SGrd^AVZDg~`>6agG#pkKHkr5foev} zkb%aH7F1C0W_7(|YxWr(FYBuVRX$a_{vdQD!ISg_vK9gV3_1*3fxZUe0DcAwaLsSS zcZQ&3)xU0A{_~=8Bk(jp)VqI)v5g#yl^Cy0tUFB%bD(QM!;f9K1Ni_<0InVn2{%fm zdA>p29Nxu7s7M8><8`7{B7fMwba05SPUwoVbvVuSjy|_Ys=n?P-u89Ta#f#rbQUJM z&A!Y#n0}gm2%8r+Um(?_qpLoVuVTwN8{>4oHSvte69?n`-L{}(D!1<1Jt{_MW_-qs z!%E!_bd36QQ)tnDxdh<9Z0`cRgJUQ#tLfIz|8|D){9~TMXy8o`$Phrm2oMHB>%s9? z{>^sb5Il}b`_yZ3b@2aY`=39B<~L0^9e@q|8V&<50#wDLIH2+Y^Yr`2 ze_ljjnE&21*Zb?k@|)%#4{Oc|M&4yK0uT6%8qxC^1FuT#4Pag!V+k&l{2J|`c|{in z-|>Yl*Qpo$-$We8i`w4Crlwv*^}m~sU=jQ}2vJJiX;A7q8=7>TIqJw@98z2Nit;!! zrGWcE%-fe~P)*qFzbg#-yN(0@@7Jyb^gT|-Qig?guh1UZlKcW+rpxQk#l#8HV#a|t zmt(e0J~>*QbeW1Vfo0qNQFUoCq`rbVU5Qsuu%4P58a#5 zht2i?T8l#G%i2UxVa1_BBXfk3PoCY4frlq4Yiv987f{k??c%?$_DN z@92KA<$O5Pz<#9^lIpMC1?Y^s)L&qnZwnPlIrGq@wxav#`NJ1etxM3Vise5n4O@5w zM|nf+`?x0W$KFpXwLZP)n&T2uG>S1*g6yB{WY^$4= zvQy7w9SDt7;}Vu@*jKDAJBJ@3st6F*8Iv|2I%f?UUMy=7V4kcVCL=f?llI7T>O1ZI z7#QxPJ1zGGS9FPU9rZd*Q-otGm5>mC7$|Tljw(`_)fnSpy`w8TemiBIGX^~Q z8)c%Z_s=c(f2LS{L%sC`YX%B0AkX=Z4=;C)I; zonTqYp0vJ6Rha(IP(~j0!!~u~G@4j`99) zpBrqb?5y1M_FL(fC4Cb5W==eNOvxmYj2iMjO?2aFw zsx1AW!MT4O%~QtCDJ=Ke`?1oE*VQ^jH%i5wG4jbE$0> z!OAS2v{f29e%?&c>)A{Bm)z?%p^=d9+t1vdK1Hco6WRgxqu&}aXVzf2_pR48!mO+H zDD_L#0=HVPyuJqP>R1=P&UXo zH@dUx{{x{16@ZP*loB#<1Q+6cB`&wTuDF|7^uUvAxUpMxOa8n%eZ-M9aJN3=K?@gB zh)vDoAx7#x@obqI=u;nYzr-a=y|Z;M7WT-#Q23PG({;)l_QeiV8ryim-ln_+OZSLE zpKt>+|Ac|CVOkw9Y8YU9S#9bJvK`JvZBV({FFoePcwv1jx8n?b=cSvbm7wNPFBn#%6S+r1}Dg%f8T3V0&4_ zT0`z+E-7BZhC9C(rU>gu7bGfY$wu5nogTVr)%L03s`BR-dVmEITWZpNA>aM9BR|yb zr0dz0m7ZWX%gSJ1)n*-zJM0(IMLU)B@Dp=kZ-pGRJ&cTOp>vnor_+WX?LcFvFHv$} z{K(gUo_{ch=^EAOd$*t2qZeRN!Ha6@(?|~-W|Ms$&14b{X;D}(F`Eo*IWH=x5R-q4 zTT=hP1{57_l`S!skH}ErCLZBRT3GeQUL6ngD1C1H^v!HRDE|g`{;0w5kOI@aRaF=h z*@N;~$rz_H0%mG~Vaw#fHD`)98w}QJ-**6<>S_Wnu{xi)(TGjw#tQ6x9|){1ay`3e zVz_$6o32)*!ZQm)?`eMXVQQsd0j4H5Gh$Jp@h{wN#2*9i6u{!dkL+s5doHhK&NM1$jhzyf|a{^ts@* z35vqxC52;~Q3*M{{s73l0U+FZJ5U3Dk~qz&_4&@|%w?eP@S1OHQ&-=@hw-IKW7Sil zO_mZl47BOMPw)p7qh@D7WF;^78PvXAm8nEE5G{Jrqfr5+?l+4IBHYR#Kt4xgmdcD=eSyx^YwXHG2+6c%{yl>q09@g)rTLioV}veP$Z$ z%0@g=byuZHt|h?nVfNq)Paf7j7bUI23A?N3s*Jz9X}fz?`O&86VL*cg7Yc_6EhrVe z7rr&Lu-;s}uIsS@{S5!b_-4Wsk0A3itAFzB&7oF(i%vGVuUE0F|=ysDEnRPYCA2K6=fF>M$iL8vn-b*&au3XBbMY%SYce4^7fabY>h&OFB!nlha@`7YVF0SM=DRLYcwdw=dtV51O;U_sgf>|uPj}~qqYsr!% z24dH<2*qftJrY_7E2hY8h1|PHirr9+{6So|wh1TF?ygG=irW8Vj zgUP3WF457we{VO?`iRsK{0a!qw##@#2Fw!xFN&p8EXQbopGBJi3?;k_CE3l9|%U2Z$BWG|031VsylM z1K_+oE~IBHA=Nw&V&DBVC!&64a(Ph0I2#EG;90X~eqeV+w@c);{qPmYYB*KxW6?H{ zf1mhU%FH+Z9-TDsf47Y85C7nj(cT}fnzFn3A3InB9TC$fVV=C;)HK0Y0BtNXt}F2N zdnsp36_+P+wW=Yp;H%B*6SxQ!hPDH+X=D{*Pu|1#OA|ZqGgZ)4_oq769SB{sVa=iv z??H^EGfXBwr`H>>h-8Of7n_P^H1{P_T5qeT1$%$Y#8?1DevC2`vK2hBy?&N@ZM=c? zxc$%>Y&72j>ohl0cW@DC&auB9uMJm0fC62FS|lQqlTBQ9c>VZ{!Um)-3<1c&jdD5= zY~`Yg#Zv4DdUTiFk|>&Wlq0`a3}|kDLw3(hIGsd*4#F{bnF{jOJaRNPgy$;!G+BPY zlO6KngJ?W>#@hvOiiF0(%gV{Ag85W{SQ{RJWY{tIC8>wH5<4O>hHWV0IY%`8wpJGR zF6bAlvtM^iQ=STHJF$hV1iT-?I=^XL0qIKfY}cOkZJ!2X{@7OJ#Bt|2c(}jQqQiSZ ziHhT`7$(tu2ZCd503swspnJ-@TX#+ryAn&j0?hqrz6~Mjn+vcv)NU>GY*8p;+PG~+ z3hU%D_i#RoY{!*eNwOyM;;4Bl3Ay&1La$#pb$vQu-(u$mcH<10MNm$nce#KUrhNtI z3bm1g{=HXMn!Pr(%OUtIR3PRS&Bk)9<_b__o&M9)`HG?5bD=6f$x_c3?yTK^bgW*F zK!KYW!%fIMTzwLO6V=fZkQZ5s&x)6_^!ogo4f=(NKmtYkOJD;iFGbH za!bj;DUXt=Wlq!;PTt`Y(&vK*4Ajmn5ZfA^9N{I5)M%5Q+`S@wqsm>CAWSQA?7`eY zc??x-tOX`(VpCxFrCk11h_F~#|)H$ zKcXr(()QJ#;g(W>%ITUsVV+s*G-bCqFW^p48cB@!3(i7jAJ}%`-*e{E2q{PIyrBEa z{SGD@=v*A6wIH5j3lh}04mvvPaG*ut9~3Gk%AVX!E!Ys8A)f!WjJ~7f#(*I`m~5q- z&jV~UR6XbRQ#U=VP&sr2>sfeHkkDND>rH9XGxe?o<~a9>v`=1QgFc1+>4n~998N70 zTxa@HieF}kT{Bp{1W&XfrgV`W+^;F^mCc~YBW0?qO9u3Q{Q@Hj6`RK)u_|)nwF?0V z!OnTLz0|ztN)-{vGqmej*|7Fe*dX+TH0e>n7Ha9y!V?-M-9JU3dG*(IJP3|J{c@b@ z7G_ih%U_w`OsUpOb!GEvD}=p{B?_6`I_aAWEVep1ziDD2IpNW=HiCXec0&+1h)M$( zC0jp#EAajq@O~qldL1ncWJJ*}b2H&eK`!$yH0F^BzRkV&AAQ;#kg?!md;qlgK)CI1 zz|x!xs96ACUKnNu>P(7GA3}s8>iNDq0S_W7U%{gD;#$gx!S{msAaz_;a|tn@;8i-b z{Uxx6DCVq5_dLk70|6AO%g<+F$Iq!4?4OYgX~E`RR!|RZ2dU6~yYROTEl3BNg<0FLviBmgC?iO+C-S z!LUJXX~gk}mT~CfhvyMXt#j5UAM>>GdLo!1^PIWE#l#tv`WdQq^a=7f{52a zs5>s34b1TRw8niJPmAIGOYA3k4v2>&ak8e$d3=qD*M#G*p|h6}+TlyS47&QzpRXGF z6n5EOqMx)&QfDrHnR*3i*gy6u|N7?>Mj}>)heI~%du5cBXn_-uerDRDYkj?f21)9^ z{)$r!7hc3TmfKQKx7VR?DA>%0`Jf^WvF%dui3PLj^|9zjh1oxj{!}(u>jxhMuwYP&s(bJY=7Wssz9DZZd_z)2!e`b=@udn*Wq$_6^RM! zbxiM!cUaJ@srKP*bI@!nm`T+i<*sr8b@f=YP9DCh?KQk>yv#Rl^_^sQ2;wuCe!N!0 zm&F|bsKJGuzOTD@7Es5jq~7su1c%L>FWSy@DpzEBSXb$FVKgSR&Xt1Lw;2kwJT=FD zwYnUn-98JE0er)|F}X&*R%5$K)3zz#=}Wh^C=PhtOC?~N)5)sDl7rsUIR}AdWugUX z$z4=#1TBV7TdK27AjYmZBW^@;I7T>e9Zp}vA84~ssHpm4Izftg3&mcmo()$X_`V`= zs+vW_* z{A8w(ts}kA4=S|G6?wLx?a6SXD4>N$aG-`SbJ8KDIg``D_1xvN>vy?mboEU;HAG8_ z5`~||zh#-|=R*)<@tk*w4s%DSDfcH4AxqFNH?>Ytncy{wkhzsscodbFh(gw7)Q>Hue&TBX_d=mFcq@vFjd?)dikdHpi>8vYQse0lgX`rc&}g#j>`PflpfZm z$!<$DQ=4&^rU*X+mWbwjY&`HoAT}#g?~GBU6}6w8ww`qG;b`*xRbt^qGcDiOirI5M z&OL(#Z4k~8yeLW1v9#k2^SW0aipvC_ed4Z!u0uC&=7=mNZ50TG(RXbYN@IE%!6Piwp$Mx=W4>JUtcMw4=_;_Z84K$2Ret!_9=0 z68ABTV}+%cVoq!Ed^@nxed>^FU_krqIfDsBh&HjmskP4pm~7m5UCxC)|01K$q}uCc zP^>pM@Z#Io{qVn1NdBK0%l{0iV!C;W3C>w1#O0dXb$oq=(y-h!Z4`uk5@h zroir|bfI1y`L)s)>MBS$?%8uO?QJOBfPm@b4ke6lI*)qGDi-|ag%i_Ljuc*!os?D{!hH5N6U+a5!EI$l|9yMtG5Xu7ZD z9dd|dT)B>;sYf=VCrqh2%m%+{U=x7#f-1CX_y~hPB&g%TEfNu|(=vlQfz<&?Z+|%I z$BIio=!nxb*e}2(Ua|3Zhzj{A#=NZ?hS(Kao7n~}JCqg$G*&&U5)85Xt}`)xiZ|7| zE@&g8$j{lc%^E)eIKKlRkk+_^1i(lr6nzh1g&*;8Fb=p-xDEi0P$_n^V;&SW&G!pn zu%WLjIvtru%=kU=5=Tz~mh=~iB8Nbq=(Xq0IcUdn&rANU=jV6gzI+wJj#T{^()svw zaxq&+v*6oEST3M%(CU547YmeX0K>r*QZamW7-@s#0D7~(Y4{Ol0DSaM{O?KSe@wj! zqX|^OCuIT6xmNEw(ZPGS#}xP5)D0gD@w5$~160^xeQ?G){#Rw#GS&_5?$@E=`Q&ex zRrW+e(khXih*465KPXnoJkK`(`c~V?_1UF(g_{NPPZ^&mVt*0`3Fq*JiMiTdv(ef` zV!I5ys*hczx_Z{N{f*K!GzYwGliv~IVxm3;*l^}U!b)w}Q^!>5u>a-w=Ig2szS3MF4M9=Dz`D$U- z#A?D>d{RPgkJ~IowWzRSG~=?8tJL8#=BWcphvoXmsT4VG`N0HN$YY8gR>^G+esQ!c zDqNk`q#+vdLx^3hQ_j6cu(#xyQ zk~Gt|3Sx~>eLG}X+}l^jsoxzwSqV(Oma07T76LenXlvag2@W699#CNX>_7mLp{f@t zt?#|QH%{=_&~VvuVOI|85o63P;PWs6V`Aemsk^+~PD>~$L~+53>pfd+?V7$?=3?UX z6876Ex)T(jZe@eGMCQe35hL|$_N=Vas;|K5ka~Yo<~m9t-s5RnKjv*n z)J;qZ+5viW4FE?BFj7F=fdF>zB+xdA%W0l5{gXaWxQR2fiaYN?7g6A|c7uM*(|*2z zr&~K%jxT2sgB(`SV?G2&Y#qY}M^V;~DZj3urRRS&~4}l$&o(m$#ZaYbJf~CdI+*Fk!MF3 z&yFgSwW||WJF^C0D0iZ!Pgdd>qjyRn(8-La(HoUODnVH%@`%-&I9y_>iz3F{XfV;N z@n6M=qtdpDE_*6!axJ4BRcMI#K%d!jqGM36p?hz$23MZfzFrl@^VGFb;$Zv$%28(F zG;Vlre!`$lh#2@Su9|KLp=aT9VWBqWnrf{;&JhosaUm$ti?y^AtVIsWg1h*d?CG4V zOce&u07cspGk`5I-IWGFM$fwmU63M8! z$Av0VLOX3rLBB70`ANDa2dV7sMzLtg@@(u5;l__02qyS@{VIH8qe&Co%Ud(JGm|mX z2YX_kb(cnL`T`%={oJ6O*=@q-?=djy100ikovZcABs)E**G@3b?JN#L0Z`SG;Fc(=jHx0mp z)LO@HVOJ%M;;zN3M!9Qbr|aG^Odz$%l+Q)#qSpI`*$sN(d{FH0i`;&fZvf_7!XS6T z-L}-oG}83ZET|ETka7RunKyfnzdbBN4%wb_XCgrI*!h6k)VT709(S!7fPMb)o{BPUp^W6tXXNG^f5$Z0Rs^d+-EEqFtqOPp|qn zjV4f0D1AF=rhbNf`YsY!>)+H}3~PWgdTt4y=Ast^Ozz`8`KX{~adph&R`S#R>J`YE?dl7CZLUkw18KP-{MZHbmL2J?_CmBJnQv|l|ILI;ik4dL8hw2G z#I5>6Bv3(d*og(>-mPxU_MC9EhlE_d1(1LO9y=@K(Yq(iM34jH)f2npxk|Vapv=u< zXmXQ247~`~BudrWcgeQ2kRDRM`R)xeNZ|`smB9u@}qkk@pGTh##pi+bt#27U|njT*U_EW8ypIN9lPa z30UtP1Y>C~S!@0gor5ghH4Q*RBY1->-sxVT2pRus{pJzP6B@P?Noa#c%WL;ja!zG} zL-)v}OP{T>BbqyR;qKQl*gcNNfYjoRT7>h zOs^-k7?x2xku z&R_P>;g4oW@06@IMU>(oA;TVzueU<-Hy6Ybi)PYSsnOuM+qfBE)(dT@lLaj}f$Hxf z#o?%)n&rpx0;YF8$lfYxiK%n^^mIc64U4UjD zptlmc`kRKz47$Kmo_EBdOMDzC^_vC^e5AU<{_mZG%66y$Xv#it8Z!VKp#o&2e{iG! zWuuX}0K{y&uu^!V%^dpwVDHW2p?v@Saix+9Dat+-QG|qKol5o)VnR%k#DubB8#6`r zWkQm5%93o2UDmM+$!_e7eK&&{W48D2?sLxPe3tV#kMHN4&+q$tJkIa&`9mJ}xaaP^ zuKT*K`+B`z&t;!!Jd91{s+a^m16orb{a5V+MBw7J@~V4Y6waETOyx^gP$M3%Wk;Z!T^0WHcnk_tN zU!mF}hTHM@J)_otFTwe*l81Ai;5Rmfb@zI9KEDiFGyXlz?=n$?8=JF#T8$4e6E0L& zq5V#M z4Yzu$H=gqCZQ7S)4TvH^N;h5lW8&pWy= zQvpDRmhj;KkWg&r=eb9n{xPuS<;PA#C`BbtgjUi#rGQsQhu5&l_E`XriM7bH--3#e z7G#A)^AD06NBivBvpqw*G|JYZ!ZTHxA7nh?&203Mo|ZA*;YdUc{<}$RkGX2-ZPkd` zpI0O5_)Pt- z3L#UqwE_M9BE&gI(!AbyTcbuLheXeKH3b9|JI=YQ+~Lm`Zr42nV5e2dKWRP!y7KhD~2Pe1G9S5XlH~AgNp7KQjR231kBYE@t=Sc&`wUpKU0U`C30X;VX|bl}pcyPA>X^ z+f*)nJZzs;zEy=1KDgO4Sb9RIyzp{*H$k*ws>TEpMuf2DW z6#9Ki&=>l3Lhw=;>1=!4S?8UOr^iQ>_Ufei(uNge+7D>Tzd{|wF8m4RDFF)OX`wSbX(y{+)e;r;thvr*NS0?h} z!W5lO#7}U?zrJyg$CJgAwfyBS#Aya}#92d)sF)L~&?>Fiabu5*$%OQ(=+*?(n!(y|u64C&~6+ zGTU%jne3_M`iqrHR?>+vA$#YY~^eo&Nrp zu%xzfP9j_p%uWXnmUo}0*QC5iF*-o|l~2DJ6WX6@KKXVsp~vKhLis+1CJrP?Pu5FN zX4kZd>U(1@8KSIDU{P72c^A_20KI7~8!IEupAqKwao2#jsZ`)5 zpOmv4c_T(Ygb;AbK4bbn zo&5KALo(o|cM-WLC`)5HHFpo#QOALgUjIjLKJ(x1U6Uy~3@mjTc>%yKdLCeW6s~9w z?_z*A$loWM9EeeHuEWK?f!@#pj}eN(*J?J)zQ3_51t;FafMy}|FbhU*e$aO(`SQ2<_4L$QCb z{eUsJZD61CDsc_UL)T-Z*Po*+`HBy&8;r`xU3#+giI%tSm%RPbSK~rN?Ocw>raSng zdsLe~pI=uJ;B(N9jK%jpG<;5*aSt7Y;J`<~FfbonJx*HQ(&OujJ?hH3VdkRx9S4-e zaQ&?(o~jQY(FS1g9Fe)&4|gY%wjB9IPG8Mfw!fUKwKwUf-pAmkD^W0bQ23a>v#oH= zDIxeuh-qysUB8O%GV4{A?8jO}}XAT_;r zGyGQIk21ERD`88Mn>GzBaj>?R5%htGtF}Gd8)y3IRCM}FajsF^lC{B(Dk_*M6W@rC zE5>mboqGz}>rVix3A9)z57;aRhwFv?+3!YRE@mB4Bd_(r)m14HmaI#h^GRvIjwrcQ zFX_`-_I94chCWg`wt*M&me$cC4f&-#l#3|KzHKeP&~p}$d0*SJf<-AjdC4TOQE~E6 zU1pyOkDqr~{JrRx2?wU+Q~hOI{H;#rPJ%)mzq5t3nSUN-3Udh&voY~`zRg+K z-eYt5gLGe5^{^^te+fPK#?})0ZQh*m z&OBgLHm5pjobiN?B7>xX&}-k4`py(nZ)!S$XWGx+b zillpw{RZ#m*|9vI(=);3?BgyqLxG1y>hDuDGCyPZi=(ruqVGFOWLq+F%7=o>P>08a zWZsMI$W8y^{sI04BZ5xJgYy?QFRw{+70gsWESH=Gjat~23~X0ztM&Vql!a1aoUL`O z{KIBV(%M037MSN0wwXm@VvC_l)167T#A~3t+VkGRtw=??25kqc2E-IXHdBp5is5w{gW`JP-Q%JV{Vk$;(t@4TArx(eJ=uBA^;PaX#*ZLCZ&&+?D7Gvo6mm0K@o4rdIWkdC zvA$qGXvRLmK2|{cXpk5|iZn-7popnSXCRwAzTJvkoA3q7^j+%F7nHQm&2tYz8mCK< zO}+rkVOG%c$Q5ZLN7qjdLKwmK+mLd^=7Z5&Ib%AG&uc`(9vvg!fL>Wl92PIL2C>e$ z@vA6&Fb_?vlXhI`gH!jSm{Xrz zTRataaYis=CV7+ce(C~o68ok=?5CCGg`3KgOJYVUZ+cN?Hhj+T0UZs=oGXG#;78cd zAZwhW+gWFBH*1%4Gaxuuw|L_E7AZ*pM^Fu?8_}r5(+_guKJ=eQv8ky^)qLC9f~*;@ z*AP|^B6R72f_tIJ#Ul~FKWSBJE7(cG=+*hfiLO@Qc8t1&UQA8k_Kw#G%uEqs^D=XIBhN^T$ zM*fO}>VZ9LBT$_5zVc5G2Dv5BblYm^|DPvVx5>%cZu4BEUj zh6Erx<@%S+OyzPct7Bw8Z}HB7c$qgSO8A3T*oZIN!BZQl%D=j+Xc5y>yz8WWU%(QV zi}eSAxJ%R72`Tw}__A2H9Iw!$c@$QSji}m;Jqf>3FWsBHbLHbH!4n#5ylU5ihK!vl z19PDpFhZ#nAiR$kG$_RNEPNHbxGHR2`^ZzId>L9CoHq^UEU9=mqkEB!2@EP*S>SnI zC%A>uW%GU71SfIa&mYwY1uU7X?Z5TjEnI0k=Azq{wKxNm20Oqd=+>TN^vlG;#mZS< zc)oAMhU&vzJvKtVNP!e-3l9dO3|9K1^wsI6N)eB-=Tp~ze4pY&)kiW0pfS)=*~Y@t zb8L~DWx|i$BySvh{h~cjce<&IF>S?gr(CAppm@v_4lN8vzFbn=?4FjAojAcZ^@>R> ze3_c)J<)z}$zg;lmt4c@E=t+1@z2y$Z*DcFI3b!*+@0xSHsoMv));fjozZt=?SNd{wZ1U7vSN=r%TGq4wxRUDD)A!15J^V(PQ54 zW?48d^C(sF&30bMzJc;8WwFpKTU|6$KEbMJ*m;?;_0?78e0g|ITHpnDJxmJtDOXhJ zEbcYFdeDI@Z!A#eh}<3kzcvSX`9q=*K~N6J^E0#!kC?nu4(nXdD>0;{Gsw!G<|+p%FSi8=`)W;k z1||VOte{5ybtpC2db2&eHm((>$C@;SN1b@+T^C|Ig8(ieJNBjXe1E&Vu}F^2lbk-O!$`XgKd| z&uGtV-k6gTmz33w?S=a+Cy9a6mt|BmlrfXY!?4g6TfX99!Gcu}pd={f$EDv{$18Ya zy}O7s@hz)d$54$herLd_ky~3&aa$)EsM7QVi*OHK95`Q1dro|G^346!18WPhQ}vG- z6-Mq#o2ju{bMnLpJ%q>z{LSD`HMhXim-iXAwZbiloN zOQzjaWo`@5M)t&Y12~@b%LKGZ(KCgwh(?VY@$z~-!cpHb-F5oy?ckEV);Dg4e)e14 z7#q2J!>7$hwV}Z;TR#<7Rdo$xif$_de^QyLeMyvR{DXz-O6yjHTX4qpm)oWBfoyjU z?ePGGEXprj%M^PB(FMzGEO>Qoh}~?EbFLqIverBwU|1wl29xj07oGjq#WHp^@J0x; zBXxc2LEq<}lVoku+IF_fD3v>xIfXsp(jnl^D0l^~GwyA*C9jX`Ym{sL3)Q3cen`SX;WeGjcNqd-K5I|G?Q5nEoA{9vX&nLu0DEi`m zBF-^8r?;WI)6^9(GdvFK!{72ENigG^F&-KOxW!e{6B$a(YGVNJ@dPuOkEIY3nN{co z>p7-^{Tm82cSk0PW&yu41b>vyQU0Cn-n#_WuK2Gu?dU6<>7?D5{Culr>htRC(eV>E z<>DU1Tz5+!){*yQXe=4=HpCmEv>A`=3#fk(k`YTa{$eW zkXUPPW{Ol#yUV|`&{{5-Zi!I2yafa5jXD_DKj zqTlh*5!<&hQ3qDJ0Pm35t#!GX0en}qPQ%hVO$O&ZTp1K$A0N;pZnfsEN}+5mCY#%q#2LoujPnFDdrpq(3G^Bk{bC8w~gq+Zi9XUID!DOo1TX@;c&B``DJyDdNVHb_tMz``nky*wffYSk8{@hpQR~s%X6P zk9ZF{OJ)7xaXWZbADm`rukofRzTNFkAf_hQ<05*!$|_c51Gg)FtIH zHpT7zO3vGP9+_F#BZ$!kgzV$1*B(4?t?J*Z&1Ay5Ho<89k>2tRvu3|kf%;gvz{QEU z0v7dKY(cTTxD2>8MROMPcIteh&d|GBWyZvJmaAh~1Rw+&na9QA8t&()cr3(~>X2;D z@@k&FWu7(o`QmD_;|6LSAU=)IO$N*~aZk8mFHQ&nTeQ20>al=<76eD44^g~ZR4~D$ zXNrXa9&TQ)dKN4qpMn|>c`DL8Xv%P(5=tq)C4|ugJCk|({MLtnI$p>o#+kMrq(<%< z8A5CWcpY>nQ_{zY#IDwS5bHjbzYpC<=nQz8Xa(p|e1`9J-rp7Fr7X})e59D%&@gYa zrW%)TogV?vh}@UWr7Fw_b^kcImcT~@1s=h=h^(c_Zqoh@9?CmI6gfU_mtx&+*3v^r z$8nBLIwSNUgUFH|6d2D#oWFirxMFWUoQoE=;4iXRZ{r}TUwO!UQf1r!_{g`ylqazj zw=msBKB4*r&N5MS+Ht^6Njqa1(a!7>;2(rvO8zPJp@hGO;R))o4CWDOLpL@u$i(L0e}mb(HStOZ4zzo6%T$?r=Ch2KypRv_LRgAt(w zPP3MZY3F`%+bsM22tq1%;~G?_N1*%oZ<(NUxAV_ro`ipI#EgE|ufwYZlxrM6v3J|5 zD#r80%b--;Moy!;2MviIPK0;Yjp(vdb(?$rAhUcxc4=g*1@@I9>2g}xMb$N+wmr5^ zs)=ZJc;hN}LH$oG+h?jys89XfC%ND^vBY6X|8R)SaxL(HcH3TAd`S6f7vC?rHH7{|V94ARVD7kW_8~H* z052m6!n+3~@TtygTnP&x8nqCpmoBMYdXgZ@su~7LL*U^j2!1|F_{Fp@*Idw?52=gV zb`xYpB3wFnsl}r%BcPqbu8Ju18<7b;ju=CpF>7M-7_&VvsMs<8weAJIJcrzj{h^Uj zKR8@&?Q;6=_rcBaC-=VTY0$ZBNpuGN2_-my2x{9KV~lU=@!ysO$qbsSvoJRTbWS}L zX+w#5C(a}9QB<4I(4UA>bHwI{dJF&{mIREJQgU+(UFNI58~B7Qdc2mb3y7wjgNs5% ze8y=oGZX@yebUBX>0{i1ACG)Qr?47LcG*^0PVKLdkfqV?D_Ax&e(PuX)cVN9o<8ts zDE=Xo6Moor&*b?a=9WBY&xQ;mSkla3yfK7l2kPY8Z#3CljZ4R4Vn(@{LbGGDGyQx+j|bY*kW;;9d?`VB07qh>YEAc2>lXYrl&sS_U4lflzp0mbP!&)CF+7*p8ys=7E_@ZX zw|e=^gJCV@69+I$=s9eY0p^W}hJa07$M;H$Bm5QXlW(f^-n!-g%Vp^8Bv&Lj*eqB9z&E2o%T!HTD4MkmaRpYl@sCikRi!iU5+WjhcV3 z4Ei9MQ{YrZYB2OYmSDnRC0pU+ZzzPNY9lWh+P}4>DzxIe)#md$fx0t)3f|D$PmOqMDJdlPv#ss=eY24|IM?6 zV0i#0Q!R3+cZ05shjEAx%bmKN%8B*)0KnOs(5UNMcr1#VG+r(mcM!~HrU5X&n(^uu z(y-=XtEemd7mbdRZ!i;l$~W8Ed=$63>wd-5KLYfcFFMS+1qn2FankZlxPwb>*$n8^ zd>48%RgcD`Vx6%k$0`0Z-)hui&OlDSg+iy#siO+rO(_X6ox=;>V0GXfSG~3Z$ehaV zs?SUB8>z#;FFMoY0T)_{Jh%lRp>8t>g#Nm6>*vdPjwx7s%?Age!{YUq7_WD=2Q@C* zz#hytuh{m6vy>9WUOgSbFq}m3xN=?!G@;zfVY{^RP7vg#3UTIxs#?PQ&A<yQ>zaBqr88&L$m@a!`Bzan(C!U zIY2bVE1|%SpwpH=_oPd1w@uAFKS0jg+*i_&!wTWna@O%g@vYgVrW;<;&$sHg=TCcX zh~_bRGc#fm(wxuen|N!N$#ELC>HnmM^t+s%g|uf?4ubtEOHyh3CZ|;FY2}Q=Bi5RZ zjvQsji@9$e#G+9LY-9rw8dg1sSI3>6oG@l6guNNiC}$iQ*DV=uzaSKH&x2wDg zt9;ea)_Tq@2+YYmZ>xixt-E4teCAbU(giOTBr9hFBo`74H)njJvl2qK``cR~h`W{k z+Nw~jM3TZ=S3%>h59?D2FbfJU6WmBcM^sB}6$GMWRTIU!7Lcdr z9|4mY@DV%RPxq(YhZ~%!PM{sW;{Aj7YEYLLJkLydn`qsaH93(M+hbJ{9R~^E8bF?8 zj)M>FIzDqzG_JUdZMu-AuDX8*(g_v-yR;M0A>ZI~c;QP8GDRntcCNe$^=mEaN7tZ6 z6o0Y^MKfEIf3Q$vQiZWJea(?@D%M6PrkmP2WxxRGS^GY|-<%}o`i)K7ehqR~ zPo1~pJxnoQmp;P?{fKR)Il5+5f%g`z7@vtKdL#t9iL|2IGs0gWt~JM`0Az?hQ{4Je z6ms6G?hh6ZYF`wkHIT_q38G0MJCFm&HQ)@YxISYHfFYy!k!wjzbBe)#{kI2O(Oj$q ztQkS70le*g9Z4F7GENTBf{CY@cYu|jx`_S>!Ee1r&d)ywq(%#{)bsg!$7dzzVF3Do z)D8Za*Ge}z!-bEfWYaD@dqwSOH@s?8^VSW_j7Xdw!}`5gjbA3S7WSy>K@1+pH@`TR*JirU*E`9YmcvB8gRW^W@A?+@mJyJ7u# zw((WZ^ig|uk-rsg$?`La^mug;V-&n=51E5*gwN(fPtjitBAS>EwH!yyf_?!f-rSTT zoBj2P2notdHM(?N*18|yX#SAx)BhAF^N)V-|2N)hKxE1KOXDJ4fw0#y&1Lv`?Oo`U zhR*&Q)1z*9DMsHE_Bd09E>2`?LUMOxu5p7dqc*elUo<<%H#}`qI?#bxwypJ^12@@- z&b9}C>D#)0u@g0E5U8Kl+5WmKb)(&i;8B}=OH;e;~?o5sB(HNX|CrGOymHOqbHQ~G0MaGopR&M zSG(TW-WXCVZn2)EOn$~}WBif~)w>Ow9{_&_ygYb3jUy0egRJ}KMV^9vQWuOCO@<(V z^b$#*&c{frzcvNGM1(_5DXX}2w6-9pn!ekX0`N&%v|+b5@>G*%HH+ZWzUwD$6N=F4Ox8Ed@3ckCbLVbi%P-mfY4zjlgBp zefU<;(>-QriGCxZN?2!}AxbZzUews0;Ro1LEX`G!V!t5iyF51Zqtc`s<%`X<~tie--$N*n7W|Th#u$h8$ zB2iT1Xj%Iek$V^hp=xUjdYLXjjG8)1+)OJBVGL?DbJ_gpR0UZvjxsV8!Z7Cs)bt#8 z#2UNMG`tR!fWgsayJZ+rb>RCRFF9lj7C1;>wzf6&)umeKHZXF4tv2Y@U3Q2{W*uR8;H~)=B7{SuhX1i z)=7lIy>pzOjc}I&xyXasP|Y={Wy?oU!Hks2CVHw|r+@xNk6lzYq!c=jm^CrEb?yVt zJ9j=-Q{1+u&MrX7=6U`1b-91!_%eHgME{4c&WU=CIX&)6&$QHrw3S}ncI{l6*V$0T zq(Ns9*(a4Bz-tk55|!of9VhnhRWaOjoqfoyIdvDh=B#SBg;7ZnXz$1IjTN1@g&G6- z=k2z?cmT|s?~VNr)DE4e7eiy|%_)!TzFo+6-6$O`oIBYNi^B7&1hyQQihF<6?E5b7 ze$+y3($31m`Fbgs4()pT9ae?~1K4+3@XmM*11g3B>6L25*O%*UuzNqc-OcF%ExxGX z;vv|s%lN=&YJ~{;;*I7KT{8GhI**1l0D@hKNx{1>)Kz3&1(c=t4j%o$XL|9suFo&y zk(|a5Sl1S>VgGd3cHJKPpLlx=uAy?l014xid(|$nx2wc(xEB?IqxiZ^niU<@?*lzL zB$!>0L}=eAC)&Zm`1b_ur0X)}Li^uA-d)d9z06XBR;9Mj#Tglk;V^VpO0aATQe;(N z|9PwErCTfIyvBBnF~DMXeNLDZv5nTs0QYWjeRpuMbWq{&-EC?<{pN335rH=SvR#+K6 zqM6b33AI1}rs_P9Mkxoe)t+^k0fkbbA%*eJDj^cgBkRl@+;F`s0laRAhaRDnjJHC^ zkTCzJuZ}+U`bCn$b!*fokhi~@A5%FHvDRX4c&Yu#OZ?XkrOTV`0L0s4>KYZKnHy)23nI5<<E zS4@)UV@We~ZKZP2?yBabKS%*F0Xncn`V49@KGEQitd{cWC}^isf!4k5cWk?7G=*Fj z)S&VNae_FTP#SbWIR^2tv})h+XgjrkP-OqjB&*9oYr;(4*p-kn$tfAc; z0mGKsibHYpGo0HjDt@NfRx)bw!d9nDmvX8mZ4tx2Y5Ja9l^oBNMbE>N+TmW?YddGj z(TquluRJ-+|gTBbF*ApXNXaABuK4^(-V-Ce`=1$GfqwR8+ zJ5TLWV~gj<`#uH?Z`jJy;PmUmiru0H#F(fk$lV?Ic45?hh?uPsmr(<5nF6S?zLxxu znPa&+?EV1Sb}`T-#B3VtycGQ^=a*+^;EQ>k1;~0i&H9-L2cZ@Fn*M5W^2*-PFRBx1 zo14>;+Q$5>@le+enkoIXfd`mOxv?>C_*;1{hO$RKL-C!3Ee)!HmB^~#i?2Ga8?o(W zi-hiAE=4lgm(zWB)U@Fk>K>TxSD9EGzgQRJ&`9|@)g0yYzJ}F zgQj%yY(5RiX3*l$*VLX|1xE8*~jf<{{Ua$Nx^X4)nVE-xj z)s3R7S5aHRpSmXs)Q&TG4;5OLop(Sqo` zVMz4iczfq~BBP(4bxya<_q=>8$d4ukKp6x5oW@b)oG=57P(bR=z9IA(U4|k$;}(48 zxEsFfY9gpzX6a(p^B3zQw6uh$Wh0+Gf4!;b~A%no^gj)4f;#UhT-RgB#V`=s6e)YGJa$WIamiu&0->; z#l~;c{PQ{rolK*}XKkU!#cE=BI7?XMP{Ut>eMZ{?3|j8RJKviWNss^>4h^paPm>8^ zQ5)M5OoR>Gkaim}H%7J}Z8@FOZKXdzCqeNwd0>lKt>tcw_k#B!r0 zm~}n7$@S$FbDAPuodR*~{$dr5d3bUlUP5_7V@>ry^W(G?P4ir7i{#QiP%HpYiSdMn z0){{?vUC%r(ZX%~^obFT`#~9jezDNX>JNYcWlBTon$?x*XQkhx*5SOrhR1x)i2zn4 zM94aOrZYk`-n_3IRXnagx?TIhrIrO@FXO97t_)%dkNFrhoMEyCBCj9!@;c=rsyb0M z6X#FuZy9`BC`J-FlRBIFqzudlc5vI5rMfXZ(SS5i2eif;Nt}~qcz`?WRVgoNLOy_) zVzhLvgJbUg_P9b+?!vV{SPHbwI^F?N1^xA1adKjY+{>LhOJ)&y* zfcBM-gQ;t>%rIw0GbQ{FmJ8Bh&XVN_i;;6*Yu(P|S3gKf>DvAnFn<~M z6L=ewrfM>^zC9aQ-mVx-x*Z?2dyb0IFVK3{@g=<3^u5_#x-laHJYPKm7zP3L^AA|} zwK;%U3@qeIp-f)NFD|}OXQ)&1!Fh@IU3-0rxv|)~73ib;kMnXniBjxF_oA~Lou{`s zzXKAexS8oU$oV-J@LCp{dJebc_4D8&+nX5HC`g+Xc-;|S2WmI*nNm0N?32HW>a^Q? zmjk^@rJ6MlnHbH_I_Vopdup_$_Y1DEI=!fgyTgKN!m=|};nGCo2DfkwX8GmE=Sm^v zA6*_HUwx)rX>p(TfYx5=IAooQzvmA7Sw6}Xr0Y~7kFS>Kz)Oh>^i`wjl&_5&Yi76e zgSKV4wG4gHvT$ieCH+9b^7THS&bf7gxPxMXPvX^51Ae?-K5P$%GQO*c+Gjjm!E$BM z5ZRZG6|jDpNa#7G74*&!3{>m`kv=liO@lPhecB;cchT+iy9jaTHTq%NNxCv2`p`cKce1R?N|j%Or-BbzdL&QpFV^t(57iQbqd z?5F8K==R-Rx4kDo<{}#U4q2g5WmgDy2oTghKU3tBa66E#hvyDSWQe9YBP{2MWOt(a zd4_W_)%5mAd99s?z-#G05$ZUJO4;@@Xws{f8SZOn|pMCKrQRQuAFIX@(vIq>8waCvj?<%iT$vNAI4Kd_}K`S00U# z+Kp^N7r8Y}%Cw$txy;CMXOm5OOj<9Qt+A1O(QmYlH_sa!KK5h23`M!*{nkzV{*b3} z(fQRW{|;1xCKO6nql^;vW<9Ox(z(*jF?l8f{5^X^$ZFV<+bL!fn7v$8`ShJo+VzFk zElJ4_?S+%?sX!L@T0ucbfQ{m)ju2eV_Hg_DQRe1N!2=%Ku5fqCVp#ugI2gNmKF5=9 zsO3`;lr6w|CF3)t>NE1mnnc&|^VeMQX!)1!Di^*s6$XSq^azS9Rdwpt^Ej6gA$k12 zq>E`g+_a2g*)o8#GDiWi1PlW7tYZd{`_bsuymvz67HaajToL|#(7P6|*>7n|5>K@B zBczhj2Z&PS3<{Jmew`ZgNxWA00=uj_INSt~ic_=&%$3y?Ptc^{s!Q!afLB(+3=!D( z{;e-af=0(yM!+>+4UgHnXMOvFMd{)oW0W`XaDXxf`-u_CWN&qw@A%?oKU51qI=HUa z2wm3HbN+SXSJ}{TSR(bx@Awis>4X({VeilAm_T)O-FbZQG#jh#`E9b`|c1)8O`*@$s;{ zGUhVilEjC4pHXFdyO`EJ(w}ft&o9J-!Ivj2d3M!+e;rc%uyWXj>ny7Bi`iX@3dSK#!@ESp4 zCWoGC6Uy*R0bV@n2&xXV5D{4YQsKRT?qbuhhhgz9V8JB(@7{~|4*Z)3+PM*_`2?N#qY3i6vUesq(%lJ5hXw*3qGdi;$?uwQNi`OJ5)Z4CJr z<&0-d1ts*84ilwBhhG2&(xI0aizNy5!*I9L4r*O%Ji~R!pUhB#Bg#g(AO}z)p>yma zyLMI<5$>LtzUh&!xq|9|bttFTKDIBrXEdBC#4d)2jE|4`g1wQqnTGWEVyoltOcRTm z=Z0QPoO1}x%e%waiSTicfd!^>mP|X2?mF(a96iSkuqt;#*C&Y~)3b>|7hs7%sv+%T z%i8p!_i?(U#wr<;xkgG=_ZbSzR-2R#-b{t;Dg%no$D8KPH}X=D#Gb|yABovBnOVUt$L@V5O$rom z`X^I<$u_PDto5POh!&?G4BJL+UwX2g4PmZyBXTIRra($KmcU{6ZmYH&B)2KA-2Q2~ zd>;aJX0|hBtZn>gjeD4n<&SC(Y%GYy_KF^~TJqWxC!No?1BNbrnZ3q&;8#o`n2AqN zBNdTz+Vzwa7i}Se<_GbBN-%$hFn!*+9A`kh957-ey{RUyN?^lM3(G zboZArz$drz^3MX25x#7{|7H2_{_xJBMbvHqWah1A8`JIU(-O#$IB;{UBO#h^}S}2RT3U!ODL3zPcphR-!s`(d`mE6tR zWz4%D-sCcY9(dbaT`p>*ZoEd{Q_6YWIMjB3j2#Q@S@SNlF>emVvWlLh%?{tUVsPrEi+t*ut2uKKZY1FEt* zB{@vWxdxpmE#7Ki-h_0`ey@J-nJvg<>z?)bl4_IN21G|RvpnE649kqGto~|Rhx9(r z)Q7!GA%CUF9cW3O0&f6&WV$O>Kw#FJ_RH{&o;s`xeHMp20h#qJna!KLfEqSsC{%Pm@tGrj9pgxgXJ9nADHt* zHe1lo0Jwr_=1m}=h{nw+@*oRG(z}R%umI{{Rxd%FSkMmmDH5NLIF9Uf`-9~ifOc(L z2FSRD6d**ng{8j%|G^@NX>-2x=l{-*SvCt!K!6t!KLJ0Ozqx!LunbxgteHW`{ewuR z5;SVigjxC6U$&jz%y4Q!O45-(830D(Z?B!n6sVl3WE=r(WDfnQTll{McDOBMpgp0~ zfiR$o_y03=@a-=GhKyonBVGJG9q4Hqego1PCeP@e4gcb7)} z=UzmSYEF*=jRCD0iS9$@n2zf`e|Krn-*%}NrCSTM_DZ4zXbgBbRc9O1PVX5-|J{|r zf7_vJ#N-I@@Vc=B(3X$Le?6^#J+1%sb4z=ToK2DNa=BQWJY}IJ%D4k;-)DF!}jJ@kc8p{{Q?hE(5p1zV&Q!Er=6{Gt&=~ZVpWSvc!0N?ewTsW^~jodK;azdu6oB( zDLCRX?>HYPByvq^=a(CYlUcV8FMxzsuL)%wxZa^{bm_wjLGvRiIuwy~2d%xFSUP}Mh zu!El(cvcB=4`E@ss%@8LQc`&+F=aBDybSn1H1`+(ym&rE|9PF+1P3Ej;R?j)-jZID zi9%s7-@H)^$w4Gxoh5{y?Hq2W!@>WhgG#sS0>6uq&9E+YFlV}C@<2)L2(4&0ar`2Gm2rP{LBkO=h^(02vh7sCS9sKKMG27}* zOoMFTD|rfPPCV^UcU{(p#~v{+1<`K$Ta8k-I%t&lLa4_c+YGqb02Bh|Dq|cu8fcW) zra7_s>g6Q1Zq0VUF<2sjKX&4gsHa>%c5EmzXcs-4krIQXM_8i}v;~MA!U~xG_H}W{4GSHap z7V?QXJjio!nSXXn*HrmgnA+)qe(ZImGwV6Z9L87Y#9bET^m|MN7GP+p>SFr~YXgV} z<-ISq-MuI0FkV*cxTvSx!xn1}7#8q~-+|qKKXJhqNEimb@!5Uoo1uxGtuI*UU$R!yii@wvV`vkz4I4Xybl7O?_c|7Zeuq9ujG0R0j9_bC+ z*oQT7K7(s>Fx}`KvGj&*sjYErAezJNIk;3t!3zercl$V3?-$bTscagwQ2xs&Z{Q)W zcxS5-IWFHg&HW`v2WQXh7+m;q|H}4e+uYt;=gc=$+*oJlD!px1{)&|A#Fg^#1CrTa~X?Cl~V?E<@FAF{$}@C zc&+FprVF0ET~<{kQ3b%9m;;I6Ly7@PcZ2k4$nmJxXa$|(U~ z+#vs)Kzf$gWv4{&>cr*!;cF^FyO?*#nG8!23BtZGvLC$52Ll5MrF6E|ACNW?<@Y_= zjm$5nj_zEu5nR*(N-iQ=&R?5j^|eh4&Bkursmen(-lW6S8v|7DJUh>plUzvs2pd71 zVov!Phk4!USMzXWBrOWDN0uesu-`L1zsxoVbsVMhQQYhvd#<^hw>WrJdryO5@DzIq z{2cQp8Pvq^>uS;9L9LUM-?-Z!j_J9gU1#|*^@IA?plqcs*kNPX;ckC)xdra(WJBWh zjd~@(^N9!RcJ$i?O1aoT(0r8*`mM=RtmYpqPr!V(KBY6vMDb5dhn9mkyW5qMZ>q4U z97B^)jP$aeXqBVStkSU|BzJIdtky;!vC_kA*N`F9=M!>_ z`^)|x_TD@k%D4X?CP`YTklj>Bkt9U6se}-dEE8fXYh=sLm`V0^LdZHrw!~y7%w!KS zcCruIw;9VAX6bkJ`5wRHx$pb?d%nl_cRbJiJjZi9pFirjPGg+cb)DyRUg!J$el4jc zSDg=My+B?Qg_rTnzv=+xmR+I>b0DTixmd>7@}_aZ^b(j?bN zhNzWqe)EWk$>&W$$xwqE*@e`8mO+uBabGaw7|RS@f3AvjHItqJF;i_%87XU0D_Tdg z(zJ=;Ehp6!WR6c4PssGH+o;`JngP(c>;{Roy_3qHMnnVZFsu~8bw#~JeVbrWr5+36 z@>N&@QV{&Y0v~C`)j*7P2phU6c_Pbu?S+?1micccpJAN+=4WRwDYGjWt{-6r{2nn3 z6Qabr5z&Z%3B~d{nh%G6Jnq!G^&NRbtIhV7ubt=zd1626Mv>vOoU1S|G*Tc+kn2`CwL2aD1t32vS`Hm_QNx7Vx3Wp3@9$ zG=R_^4EHNo?ei^y@D{oL^7Hf2uM4TAe)K;^w;>{%F>I7$H*<)^sL^D&>;B|+wGy@E zAZ>*Il4X&j_3Mze%$@U;aFSp%_{0u-y~eClJZt@am`1-B&0``WchBs4>8kH5MGADy)x70DUK4pD@u zGQMn$2)@dip(~nGMVBt>=qdkZ;Klfjkjl3lH*5X>KOK+G;&51&pdX?~XPz((baaSWu@b-D8y_KYspRDEW5G zc@osSbd#TF0mh|%#aEkP6*pELxzTcy8d>?S`W)zMF{)mwAljn3jM(p_h%Xb1II9*I z4)LdnQ(|Y)rx<GE-PQ2w+Y6`$iG}Pzc|KPx5}?wc@GLXs*IQNe<-;ODZ5~l0 zWBucAyM7J>`rhCWoUken$-g6?xc*(PjV<-8kaNe2mz}lL{9OMk;8vfjL+^2JVH!7& zXw2eXA^H2ymHBx?!kb?6>6~ugJ8;EJ28;)Jz1MLA$6Y}Fp$(rcE-eu#PxF}sg-o-8 zf6cA|b*kxu^Q(LP;9aL39ttnfh+%NGnekO0;EIai%o(b1BoU}3U^x5+uGI`(giSOi z3r1WLTsZda$qm0^QoPG^Q8b$ZpAL3mu?-!k&PP%Cp#I_0J@Ip1_Kso_OXa-lvmHqc|#+rV;E%Dio8=i&y*ZSZjNaKs)?`bj{|IW3~6=MfAB}L zcu?}Q!VtluEDo}Z7q2N{euBcLPt+?An{P+oA+Fa%o`QELi^u69Pg@V6{Chs0H@M+n zuIrJh;~{-|DLOO9hfS!zPt!f~!#^Z=|4Q~g8e>FS&q`mMyEkZkrctTUD63sGPbKPS zcA`$v_2t30V!6C_sKv#(l%`^B0xZTwK0V@lrN=v1jAG|!-xH&aF1xdp)Jm#8{9O2K z_-k9IN86ts#e55(^@IWwS$ooImtu{yRY!Y?XLhau(3q15i_Ce{Nz*Pu0BC$l&^@Y~ zc+3dd5SBm(;kVWDXF#p#)1g)hRfEmwFP--?!3|s203_l}wIh9@Pa#?uc|X`>;3IIJ zonO8$H?GF*()wTD{=N?T1S?f+b1yIh)Qf9}rq4A0W)f)um@0EuP<_7bwHRpVwz>}8 z%2m6%kR+l&v>(2NbbNrqofKab0i3+P(}gj3{5Z6+Ex;Mv(kpX-lj{Z>ox%9iSul#g zK9NFmo{81ac)H#B$@sw8t-WRkQ{3&AHO(m*h&0+4lxw0op2!G?b83sw&NQ~q3wKirp;IsLaE9#vAH zJ2#mnv_82!Wce8mKHF>9*6BXxoidG^^JuDvIFp^N3}cS8fzLmt`}3lt58-)6K>^@A zR=;(`?prANor7N2mg*-LlJ*3$6csUrZ-IX0LMH>%!f(HsN}NRfq06y|D-8U>wgc=; zMP0(U^^yAw*)PJVVnknCv&swHdjY{yS~xeET{I(^^1jo7T}t9@w)%j@*qISrIBOlR$2~b|P`kX<^}dBtes* z?fAA!@wqHkb7Sl=SIq*%c{_bg(3l`Y#P3?^vrORY*Qjk#W#UU)i6nZ^0MA~ zi)TaRT*F!A#{>3kF$iBwMrw4a_29@hIJ!2U_{AMLg-oDBatSf&WuNGWF#uHl8Rkd> z85cMk9kTyI6?f%eKUz`|xePm|!Jljr<7FL1&!nf(F2Kc!Qv-OuG`D#`XH+O|u@v*e z*PeY{RubZ{f=F+!fjy6SxGa2sdJH{#h5{nw&RTSok%<6qhUJ3G8bAvVO{h)KdXtfS z6H|(NfsSkj9N9p($^wVYAxE2Y{7T!}U;okjgga@UnhUHtpIhMx?!&X!xKhEFL-HC9 zvbJ{uWa_rX!!D+oMEaMuA{GGN{7EnY23H9xu^tbEv%lL@`O5@tXdepjNR`rO6DTOhNvUXhrPdu*J}V|Eh`-ROl6Xvzsw*7Buq)}%TF2zp!U;w zSNFL3G4=$(!XplPAwM zfBWgC(B=MN*lTqU1^k@W&a#tCqsThRiJSzuSJ($gvFj%Tl`z$?;qUv9sZmbDwEZ+4 zo3{&ER{~d*Uy9yrdW}fhCrj`=tf*a4DJQt|r6t*R8&1aNGG!%g!Z?ujI3e0Co2;?H zW}E4)(YH#gH1WY*K1_N2c`Bc;J>*^0r-HJ!v}lJ4s<^PVnm5dXvp) z)L5IckNtdam{A}5!7<;;TD1!vKovlNnvz4~rZ|6KlAjdQMwyAZF0>EVMX1Vtu!t=+ zKgUACyhO95DWKB(JKIB4~+|_aY1&0mlz)3%wVS-(34jgzBu}p166v2MnY15jqLEygFhni zpAg2#C(EGi6?78le}41SZzkhu8jS{?LE|z1y>1N|dANNl1=+~?cc*{f-oGc(f8$C0 z`;qy3Mfxw--ViJ&ku9>`wjehw+HT+;_f-zXXQobub2shW`1puN zImcq&%8pdvji$kcqdc5niiXzb`MIwZ_w2nOU;mt}WH%r1N76l@{m=jZ+okyb&UcXi zX2W`x6jE*u%-8PVr`GK9z>)_HP^;8SM<0hj`;s=d=X|1Uqep2{kQwlB?|WPz9-HH$ zaY%}&K`tO(%ps1?&JV3QJnWf|FCMkdjMuKKRkLXDnaQmW$U-Q~=#+$SI@htiRur1Q zO_y}sA0RtXD)ZqFD1r7b#X(~?R+@bGJLe~4RE&i}*I}9?+}Q)=Dmh|9y+&fr1>H;R z`6JPAPs)dna2cZ70w%yswAS_zccXaON^!5^t;yC)Z5!WTCg`rwOaQS%4CNkH*-5d$ zCsJhMM1ZCM0IAr3G{s)j=ORp{^7SM`*1W%jpVW-d^!kG1gIr=v69ifSRX6hYEKpA60+)AVPX$;awB#;7=A zgJRVrt$4e8n`MdicY0g}YnXkr2!lMqgq5=X?ja2=fpP^y7?>=P6t^RlO|9@P@_V6P#d8ZNE|9d6iT zuoDU<%~zEf?+8g()$xsd)FRmpsZGv^MFNoUJOnxIn>QrAp3@Gle=+}h$FjYMV~Dd6 zA`F+IyiQ3&H7|vYTc(e)vrSE~9kjwU?ygE4$q_!pZC;R(>kVi7BL)bF3=}m_TcGGi z1gODe&4Jlc%J~l!fi_%=#4KG(Fyk1IJrae6W2-?2ClvO=*8r5ami`nq0eyhX zfUwNm_k*iACGgA#CS;RLF6dwtaBgpgMY%w7WH43^l>lkIS;M<-Ix2&qT zzx5F@iPGI-VG;iX=yj9*%dMmI?uLOE3YP??$UMo4D@QR$N!n`w&ad^@x?c{n)b7(O z*vVj&?hZa2ajl=`O0baj(0!-J{YgtpTT(acy+}@fUE0D!cYk3QNj;-&2*4!fqTa5n zd`>%wBwyw}b7AY}_5kH$`YH^ayazjtvv(EodszNx+WgwY^|+$|#3BfzjakCtgmE+P z_YRD)GW~1CZv#s}iru2%mN!Y{Wy?|Xt;G4{d~|ekn^cyv+|KySUW*eXN`&nT3Gk{T zzZwr$<}1Ik4G*ot-hm&I=g(J+4=$;g7>J0$4%n`}@+1xj4(Z%Xc z|4SV)x5q)l#r+{?Jcd%WDpqcM$kuK&N3Tbcidtv&V8@nGHev*WCG_gvtVkGErUBJ@ zoOBY_#uTgnQRAxj;zmqhkMc8Ij-Dl(R{sarGV_nR%u$U|bbufRM5Jo^GUKed*m>fZ&0NtE(<8pB`Ip>qL-=CO%q`o|m3e z5?hRo*IVK&*5o8~wvEP0Ha4IpbuWKv9K628DKB^UVR!F|ku3is@GOb zqFIt^Pfg_0Lk(X6v5+K1bXH=$796H7I_WD2=(sTk?+GZFg?>YM|AK<0?z+dsk5mQ@ zsr?v12nD1^&}zjZ5^8&tc8hHd49o4@3;8WY_X%G}xg)RCC2)~z+1qxNK4Wqy9KXTL z0bpMnsV+u2;Kzzl5+t(v$P3n`+K08&P1@bqbeHPc36|hCq+s#RNn@9)nxrlczwy)m zL3rChMljC$+QTSF#aGFbPrt--U*cCf-;()~b4sLgdAQ_*_=jJ8&9(W3Tj6UeB-rf{ zxKI|GjYXg~q(?bX;@uW*?-n1+F!nbSW-*Tp(6!1Tt-bTQ`uR8^%#Si{{9Zfr>K=rU z(xS#Y3k!8t)ydHX%0nA`qO6ZFZViLvk}D=s>bB)7Uy@vEMV5AyFJ;$qe9H#Jx&GxJ zGrvZmDL-2S;{~24+ODN7a~`8Ca`= zNBvv(!CRv=7kYj@2V8=5WM!7;r?amyAmI|L=p#{xNpsrOCU+67SF+AjlQb**oMpTL zGMFQ&V3RJjRhNtKLO)fDyQuSGS9^QGCzSjsQoMZdK!zMRqFQV>*L)_UxriC?+2QdK zZU|g0#6}XndM0kNz~m?%X0^Qvz6w%CRrfj;SB|RPUUOOjtnn%oecFx6REk{=t+}L1 zz89(A#SGI#yJe2m=%`h~3W+o$?X(a{q*Qgs3QjN!HYWuG>-fngo|UDahg+($r)VLfM}t0Gs5@vO?qYt*7g&q3Wiafj@EV3` zP;yNq{s%4lBjzGrjuKkHT^-j7Iry6DP=}i?#_}F2r1j#Zk`q<|4Q4)Hi@DHxE41EUJ#}S(el|9*y zw5s~v8+K4PJ_Ud_p^dQ8{9s(A>T`)m2}p`w+cecN0fX5o0$0OYrYok$e7W=!qn&~D z8w-}e9XW^;wY+&3TNIRaepC81oq1;Zm}^^471kH`!JSkE`OTsEp92I{{)GATf$YoHgK^EnScB6Z33}ehY?IZ254t82 zAD>~C62L~_XRt7muTLXz>F>NP9cxy{JN=9X0hb)?;%AwYOWp{fK4;H>!#__P|F&4$ zv6{6|?X+AZ)9yfD;+YJ5--8q5{W z$(=)a0F=o3-v%Z6M_AGSQ^%aYN9eSFdE@-+8|eR18^c1O6ZLc{-2}S<`xfZ%%>J2~ z%#UTsb1#G6)GRb?MHO`9d*07ZD+i;co&qUn7%3w2S%6~4jGv1TE3nlJY95(TKk=s| zyi|QwR~4E|PjesH`#g!?d)l6h2=+~BoDw1hCO&7JE4L^RED$SP;WTMnlbUZpl0_-E ziyo^EnGYBS!E;E7g+BuR3LDs=d8T)qSK_Nb?)H@}9cI#{{qo z*{#xvBXcq>_@BX0*W;A9?h^TtF?d2>0n2kFw($JwIW?y!$~vH}EBpyLqKBlPN}}o) z(UgcjEg(+wpOkWl#b_0W;&qYsPp1No?R*+Cj4@8?oA1w_zpPXzkB=85(*ZP-S?UXL zKdA*_lj*TKZ=DMloWlvj@6iir&L@jAi|)OOXT@4GGvBUTJTI4-x_~o4lO-rOM(k#u zQ#5EIHq-cSK%ify$@LIi=~W{^O~IMP+I?!nK#op!HG1@Ekns6z-qr_GfBhwQSb24m z2}Nh-xc-@CZF2AtxoQt;yY12sK+R#8b6taI$&m3qtEVr^Jmjne2wHf2Kzz>G!1y1y zM9q^?>FI^T%U7+e!hx_59}7MMwzGV>*sMICu9ju2R%*laM2<58xN+_TLcUDLHB)qH^-g*d70Lk`Y& zfo<$3BEi?cW;(tc)-y_$cWl1bz}W+ihVjw-sQSJM1I-5t>DqQNO_x|iS!zvCJtmyS z_=`9wKuCjyI1$6Yo&74b7`Xk~9&Qa*DI?W}GuH{M(@qo7-_RU0*L;ocTPUc#+*riz zb@`|5ICfTgj$AQ`FBUkd?x#||#mhJ*@Oes-NbbvW`GYmu*!+2KamFW{0nFK@eW?ag zdm}y^d3xhX9Vsq}yzgA?XWyNm>cRx7X{Kh^$(*D(X%UlAM}ps5(9_z+eNsdgUbu*( zuIQ}jbA8U>);=a}?6Ki8vdBr}7UCkrig7?TaXXiscKmC=f`E>XeT#+fIvkcy|9Z)z z!Y}UK#BHz}>b6;5TFl#2ulMu3+;B7%0IGLf0WUJlyB;t`kmZ)`Ep?hCSm@K~qVZo} zSx}1V=5c02T09UJr!~+K-{Va<&YjIQH2crMs1a)VBsFoULrtfo9do2`*bz<=e_((X zLwy5>!341qjGnQKgDdeW7KgI+tmlKiJyQ=vO>P5&Stl^;Mj(p-s&-@vB`TX2)y;!)vioBYDyKcl%Q3?uVXvn`*VE{k$(-`=Bw;a-D)*?d|^I7HDM z#?6sSYrsBo0|(VrBC2SIDS+Lfl<@s_Te6&8!O45uYXx(_pS#bYj*WADyw};;3EjIZd3mPRI$5F;fuYGzz%$b! zRnW5t?n>&-y77goMd$J3Jk+((wz>X>i{!SE2h{-<-&;z-hXvXc1}{!{D@)w&Vb}B6 zsv|(x=hahl8$VJuzT>Yj1>X#;t9_>aXm*a}P=&e#VyblNIV7JV@-1?aZ(Ad=I-VxO z3!s{IwIgvhGQn9FYMpRv9}Z)>KsT(jGV2?cP3sk@Vzjfwok%x%LwV9xiQPMEjO(?y zT}(wzLrjwL@OLC))KdeH+toY#Tc zDgY1)6#dO)io0zk_mP2zvCFG@sp{o9JQLQsuG_V!>-LC8l~?q`4#t6rIcTitw3&Wh zM6;orzT7N&oOi|(d0JIu>U`gN^~!-m=aZT0ccj9VMqd-PB0xB9ba5^Ocd{NT5vrPy`tjDq^B!6YtXXU5WJ78wH4~ld*6;dEJd0JX-s0JRpalR@q#!T<%4hnDe|ry@2%5pG^($C(^F9bS{F2z2#`R{DO&IlbwU z%p$j6vamzZDIA?Qn|6=G@U4NBQ~~%Sk&~ogitB^KvuRb#c+IT@t+o6@^~#Rs+VIli z4(X2pjH|?aa-8|<0IWXu)3zma1yMPhbN;On)BZiG6T>Z1sFNLoM{08Hz z=G0m5<5adG1=RkfJ$FUIO*~Cs5hHg3SJNJI1@vderU}SQ+W;dU~V2-}4fP4$EBNEQkzu zq-fu-QuOJwK+Q_*@MGNRA8GegHAxyq$v1w;zH8!yOrBEiiN39#J)n5+H&cNE zU9))yk(+3QrB&}Y0HCbqC=zyD*fd{ne4}@FbnpVJ@TcTL&noZhJq9;|4$^utY9xTy zlK3P!3h}%g@CF8m%Nt@phPhavr}lKlhn8?$3vTji#SYDNsxQq1%mlPn+TywEScR7> zdwpS==oW$bk`$M^kEu1FTRIFATF~d*1^i%+;7`$AL9*+>6C${YmvrMj*09F(%JfF( ziW2`W<~u%uAyrFCiFU0V&!5`<0yt?JUs_X>+Q9=DKd-=7Aj(SPCko@5@yySsK(R=_ zGo{|UQ)_dzjJ;y$i6;%YNCz@%A&@gxReSe-`N|i>F4UFk<;u`50_3_5ZqodDwxH|c zq+w2`W$E`031LGeTGzqzS;s_HjOTSWtx`#t*~*k=&J$gMZ}!S!G(_%6oOr)o8O~hZ zjgaw8YyrTv3Vyq>rq&8Azs|HU_cAFl0c;nGh;ocI$XcI@C#pNCchh46lN2|>-+o{l zY6h=btt@waUldC#)C|CX+YI(kY2`d6c9(C5Vqll@npbk%cjDu(5+w|N=|QGh!EMWO zLO~yJpl#X>3xXZ0ghd`?Pd@$-t{bwT!@==VP{_q-$Z2=6S)I3!CZ4PL(c#Nc=~+G& zv0bq&9f?ADU+ZRNXh?;xyyb=SC+nGBF-sJRZRw~%iaiw$;vjD|j&p1Ul<&k8Nrnjs z|Jnn1s?aa~jsJOq{}Yw|FQn!Fhkl3cckjO^k}mOSQI{BiD>{>-_uBk9{FZ@Cyr;<) zv^Ai1k;F~x+Pg4b;csi4MnRtP&GR#`BFX3Y-AKT8U_b9sTrb5bo+;G-DQz~@LBjrg z6f#iF3A+X`xw!EZt5bl%tsz5Ac3Lo9uM4+0J&XT{B=~ClX1X;^i$GGM z!2j6T{~86Jy=sr!-bc>J0{G+KOb@p<_xC8}%b81aRJWft$^X7pw#oUWBLATY?0RCKV=-kVP#mZVH-Qe00K{D2UbGnga}(M< z2n_`?_8w{g*bGQpGNblIFb63qNIb3;4J^=2#BfFsZXXyV5;yW68a&%x+{l*4c;q4g z{r_goxXs9&pWfai+4qBq^}ry#cV*lK(=&_ z#rF4Q{XJTm|NoqX{J^vIXALouXdAylB?}?};v%5&)2FFKsmsFb!~f%fM6&`5Xa}Wy zg$j#N7fE5LBRzgI1uAZtuT^XcP<)K|tG==^4mIxw!8nbxpL08(xP|dxoEyRC5o?UhYrHD%ym7QisUSyf6{(*4f6iZ>Wc7f7wBuer`Ttr3k|@fcm>`N2I3OuaDKx^V6? zJq44tf}PvX{tsiDqT?N(ug=v~JEQo&E3Ci-7$Q?NMN&68_+`4a7OP;K&xO(re;3RJ z%=eXnq?pw`Bql}!3UD6iJ>1F#mq8x`3 zJJhprPrc~Zg?;YUd`WqmDYySpYUlAu+t@}Wy#TWVbTd|p=>N@ng)ocrQL|n;fTz+9 zt|y5@3{9o4idEL=%sPM0FZNWI>s=&nM!4)0#YdiZLC)MdJ#jE11xs^c_siK&b}M(c zA0Yc&o3hY$XYN#*BhJ>M1k59Ms@|Na2n3&zlwpAh7N39W-KB{hiuF8rduX=vQAE!z zh8I1vDYd#PC%m z0~o3^QEy+6?H`K`=z|&sXkhzKcV3puFkZVasY&HvGLW`_4+QcBP4-K*N2+x7 zFJyz%+?O3pPSHM#Qe(ZkWHINwGeTl#!7PeQ@dt!!nxZaYBTKUDmeWDSsztmt;H?2Z8avPYL)o-vlcOUIOT#$pf$U)F zdq%zjU8QBnL{=9lv#=c6U*0sf44IIFY<+H#w1H z_7}gI2JZc00XQ&;-#S_Y@x*N-5W6~yx8X@3uaBWr-3jQrr9T*HqogGKQU zg)`^ln`BItcse3KuNWt35207$43nGLLi0oOE6%{&ckye1L>i@fRv(+I8nEs;dp`?Z zA9qV7co4dt{YPw=L+TgiCkvsL2x*S+lKM`Dt-MOr&QB2w@;{?2GPGNXNO|+5Ax;?P zQ%G~3tv(IJOLn_ALA8)cm^5Iu+C^z^^>+=HKb{$2?Dl3nC@Vkty!I`Kk)_yx_F;Yb ze8++gg?&XTgyxjkHr4?MEy~dJq&a{%C~qcq@;2X-fZ=_tt&oC*dIRV8?fXqi{sxx* zB9}?7vv$QKnU(>^k&V0O7#r&9hLyK}NiM2*Sea-#E`3!a*A=wpE%))~zni&q&Xa#L zik6+LGnoFi@&&CeSI79z!74+5rZ$YY;2Slq**_~?pV{5cblK&XzyHWOw0V@!y_)g*#e|;G;Z~ue~*v`RxaO$4=@`v8FO$9+CLQW|fn} z4-?1`(OxnWd03#4h2)|ERXa?!+RVS!)83OhvTGrNzG+>J3_dI{$H$*k87AUQfx8oc z)_u-iJ_n1Nt}?+j;dwg<&T(W`(!9Gx&W|w3#~rv(biKThsJoccg`FpyixAAUUEpfo zABu9B@3}5#036)8^(*5cn?`RirI)4fvv;BE?KH3JkX`2;dFK7?Imo#3X45p6OZ}0r zifM1;9mTi5^-!93w(yu@M9{e2;&}TF*@oM+RifgY3@sxo_cxQ}6F{l)IIO|nhvo}Z z;%ac6UrV_Mjm>Xf8xoh0N^!{Zi(eEBUxbtxeU<{OZ>B3_ryDoy2tFl005Cv%#rqMPZS(j8mY2Yuc-zlHF#5qwO@Uj{w-92A7En{EPG$-UHs=xbq zi?Vj)3+Tf>7uN2&MGj{EM$sDKR%d8CtP})d^_)mR4)5LsDC}D!qm`NWepXoC^w7MP z{nvdqJ}b@LyX1R;2)S}XX*iD*xb_kp1f-ayaEzp*)lT#SNlXopaUyKf{Y zQrVR}+-Q%iSo_O5-pYR%UI%e?yz)ci+D!pvH((f@fG}&+?MBxlr}EsU* zw#)$i3QQoeanYhl>Q{h(IbasGzYE^$2Q;TZL<}5=jnwlQ2QU(*f{c0p=ZoN3R^ZjT zeSi$#XQSP{0@;#jp_DUtHlx6cg8*Aek0Fn2Y9Ik5B{X#z*@hh$!@iC!TKW@u^8mB; zuMR<^Q>EbVIIdYd1&ReeK?BimpbVw&U(Xci?FP^(l>kvSxU-yhP!1`QT3K5;o9vIMN^6&q1!~z3){g-a&eunfp#w-SfNobo1hQ35~}A4 zR|u+i@+OOns`K(&2ch+=(`>HKPF_uw8I$dE;~l7VB&E|q4+bvY*o)_W3T)ub+MS?$_=NBdlSs33;}q&RLsb)vhDLGH@}n*e6}j2;f%GO#}i2Im5)}Z3lG?wy;vvo6zW{H zmKI^r3iAD~;)k4c2fl^k8H2DA@Gw&^4^0Voh4g;zvGYG9Ks zorX30)fY-*g*{9>>tD#k z)u;L~{wu54FSOxfp_?eO!oJN;GWu7bc67Zg{X^2%!I90&0bK*1=m>2Osx?KF%GG8_ ziJO!?Ta-cd(-AnJ9za`nslrr4iY8o-IJLlFom6{wC!D^|rPlrw7MCEAJg+8oRjeuC zik{PRM`WhZ?NG9B@-W`0fyhb2jYQSS|;`%BZ1x!%{OU~X0OWP_8iJZom(-V!;u2}QYwVVX%riIP-eGDg1 z`_D5NOvh%54$7r4>uyAJ{K7!eZ0f;D>RHN%G-thAnKks0Wbo#k$n-EyqllDFb|Zml zqA;$Vj$EbJ8t0JIof&VnZbVm+%Y){cZUpnJ3(M$&LAJTR#P&~gy&H$z*o?6V$(ybX z6sPw4$b|o$DhrI2F^0W(>iHNnoB&>>OY;>hJ|finAjy_j#tZlJ0G!1=%&C&ab{`-L z9YXZryk%CHO9XJS@H^e=m1)N1{UR&-t`u}&Jb|7iqBfOR5}gm=Z6BCflq|-7MdV>$ z(~ICb&bea_xDz{7^x>GY{V4D8s}I7)^%#5;?{CZ2HraZ&@7gzGsFmm4Ie>V=Jba zD9<)q1QMT~V`0hCyJq!mPd(|NWsy&_V=11-WnQwKzHQ$Fs#I(Xnpp=9039ZVPJy z`3ptbx%64ys54uPE6d*Y*E4n}Fx5FMdnuKOrg4kb=Z7r5)fpaP;@Ko(jeao74IQcg z-O>p{Jtmd@%*g<~00GK|Q|K4}D=H<=J^=Lhh2PU>`$nQDPg$l2U_DR=)?yI)!05rq zbi(&it^U$)4DQGvalejQ0nmvvYp4F${K@-%RirA0$Y<}p7DeO9a zHGd5DMW%T!&)4cVQ|q1~p;AA~3{d!Rl@wtxUSo~vB@ zZY`;;>Y4zi6^I2dx8F>)WlI6bHK6Kqh%;jPq-VFE;C1C1_vI?B+oMRI{)Q@?I3Owj zJY#9iP#lqUf8T{1h?dw`8GDmS>j+G*uBH%&2V;F%N<_Z7EKlS6szo)-Nv6B3{j26ivG24*=&V)g*M&ZzeXHx#X3_+YkIqURb-WT82f=K-ej!r#zilU)}u}oNPS}XjP5O+OLD>=#bgP*7m+NGNiIsRitp$62u2h)P9tZoNI91NKIUeGEH!gn3LXS2jcQnl_) z%aLj292;nS04yAnQ>xc?B|0k+l(qa}+(YbP8YgkH33<%o=j}rs$O{Y44G^xWslHbL zsUjuJ%i4j@nDb5n5hn3%M_bRdG^zl8VuZY*c?pm0YCy@b6i6L-m%9?c!j0sB)Uva` zs(em;ZvS=fl`zvPEa!eB;cEW5LQy~@7dIwB)+c3Fw{U%?UGI-=4edH|O#JSs;VBWD zL9M!bnlIs z$mOd8r7GvQxQ>-;MKjI_Qfoz{OE#~L0?w|zmv?tCc{sc2am7cZvvsaB_y}}E*4+84 zalC>i6SVj?wkOIX%Yjqr)&Vcb)u*J|<_vEo$z~2$qzrk8L{igTugmfAs~|l? zcEz?H4SOyMR7YziSu2|+B;luRpLFths*WsZubWYU7d7QM>9i;3aYwLih1f~#V*pBa zWzMA}#=mqE_Ce&(5XhXEg)aE+BVXWma7L{`iu9X_yz!eU0gK+`r1PA#4tecSuGrdL zUca|DYo&*jdB*p2)?o=aUaTYyLJ7Olj9JT!C8<+ndywD>0{j@VrB;s4DhoNM@N>@J*;d=fTlPC)Zf? z8Lvob1nBI*y3XviFJ19Lw1--p5A=b;M}<8~0TKu18)*9ayfw3Y{gsXNLL~n|s2a2u zGyI2Z!tZ0Fo2D*>n60?JiF;I?`fAPh(U8`mxLtBd zakj(P$+u~UoWn1hFEt8-4Z9H{i2mv0=`)HgMb6OXGy6B7CFavY>!XplO}WH`#bRS2 zM{2QIqM<=beP$oEE#&Vi*Y;%DEjhf^NFR_PYg@3CT^;f$4OTVC zU~42v&bZHivBkSzjzDq(rh)A!W(v}jo`bmY*+2E1Bgd2n(K{mBx&Bb5xA$j<2m000 z@g*1PbS}B{tTK3~*JVl9w{%q1ycalR`JyI(=^b?aDq^{)gL(%sfIhL2o!ByB;0#sC z*;>cCQqJU@?m9+KBBEZy!86?gCj~VtA3X_q!tn_B6QuyCCQFuR8a$_eY*Ay_+4Pab zgQN3)hp)~50m&MZWE*NdC05ES_jlEO>6u{1lcIC?A3jfcbNr;~3A^j9dy9T6r?oY^ z{gCTVN#V2j=FL-PKYh>GMNPTNA*PzEbB{ve>1YCQNScIbL~VF!0mVdx+kCM^k~vDOa7@MMaE-Wmo4H>KDnVC?za- zgr*dmbg`hJTF(?3)S~8)U=YgDiV{C`}~F9NIOQr344^HP~=;# zkcZR6!|U*Pz`*2j|S=u{&FXiSTb&2%Vv8 zm<){D5G9k8*ov5^FVvZDBcst;J4UBZ1s(y4!T9Wso57#CKQb|2L|b#V~U zXZK?!Kjq9$oR-ihmLW|e8Mj38;|^zd zQVid`X0p0-Xp%Ipwo2Y4@UGPXzLfs=k&czTW5KVi9$LP9*}k#)w6D}+*q9(y;bG_{ zshtE)MN!ZBwzrDNP`v0xv#ZH9wZD>W019Gq0!I|n602YMIt$hQr|(q;@)(6j{bzyp ztARt%r|0?L7(l%_ez?Jh{DG3l*|-@d7klIOJ{VO;eN)2a2ZFz}Z*E zp_t&9EN4QSUVuki_+P9G{9q~F{Wg>@kP}m&UMYgfo^}{H=^6k^UL8Mo;1BZuj#+%y zi9K66L{P$Mr5;!qul^7A-aH)2|L+@Ck`_xsvW~K4i6ms5uY3_g_OeYSgphq3Gqxz( zSVE~xS(3seTh_5p5@R1rjC~(wFk_gd=j?Yq*Y(`j{rsN$x$on+?mzD1`2EK)=Ww2% z?frgj)~oZ(h0X2bVe`G6lsV!3Q2G5_$;16tQMS-g;<>jUr#(YuLm$y3%@gD38yiYT zsfW1EkTsux5a+bQ{0Q2S0edniQCCL)iYa?qNcrGiE*+#;vIw^jpyZdZ=`P$89$Eh!<0h3OWSfa8XUG^1Og{Fs;b`6I3t9U=|d!|6Le+jFjh+%}dQP~Vtd-^FLi{KZ9H_Ckj5IT=-( znbO}?C+W6CsqGG|AS5he6ZGqgIH`ZhZqot)+xj+kG}>=xB@Y(TRyd!PJS!CX5}J~Y z%k$V{CsuV$`gLIw)b@Ag^#W49Dq7ftSWW`P zLAgLJ2;%1Loo!rG_WR_svpWba&c+((7?b&@nG?*@G|)2zx`nqu??=B~VP<`qmlhCg zeOT5HcWCWyn9)?BQ%EK8VvebAE;di?0_-(G^^!LUs0ggiDu-|Z$b5lll>aVAPTJxVK$_z#ns*d?n+_T70S5~ zmFtIQo@bbLKs_!;yZ}ww(-1{K5FMS5@$7#kWv;N)`SRNooZ=H6V^jzCv%FOjC|Rk- z_k%kCIYB={^wuCKxG2C2S0>_YNAw>U@I0kIyd4R0pxbshP5X%CKE@H4XJ>uPYh@3V z8to53nbho*rqbKas_=}Be2!|GZqCTT9q`4DpNkSkQ=+J3x~JR>80prgkjz-fq_ zgN_{-8`ztu@v)j|2XcDKqrujY0ZjWg`&1aHEe{s=`1LV4019h;SyxIXqk6F8!*gi5 z6#C2p&UXRtQM5C=r~5eI5)k17_(!F)y!(?ST|K8Q`d$uO@r+)GfnA`yOu8E+mJWPY zWK{1i&nrvQ%2rlsz^U5{n~WyrhQ%>?n&obdkDomW!Zf@3uAgWVtD$%hWJp#@EMdB) z@*2ve5RSif)HHpWX8}8GdO*I%w^tw1m)R?s0#y$Xi?`Ru8)n3c_vpxz*(t_U?I6)BRGdSyfc>2vVAnoA zd{Ny{e356wzSm77vUBXVWAL7?pF*p)fh}7dM1Ku{ECD_QU>@VbUo%j>cO+eufR(KD zU25Le*qKX4YBoSDYhR(Yj$J_B>K!Lk%R_$ZESF}sAuJT{vaEInU2K*!vo85Lw(`nBvE0Rm zq6^?M2|uX^(Yj-F`0#l+b<}MGUQ^OdfSV$3j#P*_Mn<#BCh)uwNy}J9d;VsbV<(~* ztiynOPYC?qeBHZog=uZ^X@nVgFiEvtFhwV#&r}28XNzlOC^S}?S$XOZv>jo;7y5oZ z=Ws0BuhgM-b=fIyZx*+JBg_$7z^B6+uw4ze#GG47n_4uLUSx=`M`o=-Aj}S+hhvf4 zA|#Vofe)TZ-Ny#5n96gt28ofwim_~#Jn>D^YgLQzskKn3ekBB;F&`=lF*q4Hd*7>&$9=TV$n9)y;QBvLyu&-j%S7E)#pB20@~(Z8i&WKYW?hu{YB))VL-Xg20m;MVl4L&#&?j59T~i4r*G zDVml{|6C|Q>|AQ2Tay*VvZ*mNUwF?0zx}M=ECvRUx|L#>IBW{JbLTY!pw@ZJBF;N{z$}Ui81Jd3Y{LZawkpDZu z^aN0-{EZ%^1~O;hlvHyN^}|cloXRSA9KpO<@fvXG0`CRYkQ1uio#j@iQd&Ag>#jKjL8TP1&rUPlsuL9_cBx@U{0uiGvS7;`tZq!Irk>&Czp4JNq_BqK4zzSjhaVxUzz*8mtz#USox~3dw@W3!cO>VUhre;4-}Z8@JFS zI0$w@wVveGh!=LAae>B9H44&R0ye5_jC?#ZTxn5$L1~heQZ~rs^d6fLs~qCBYY1q+ zExFT#&K86(y=G0HY5pu1mMilJW9{|rVQ#a!Yio5Q{0!BRqB6SI=kROF;zTk_x_`r7 z{=}wt{2xx9LkBoDII=b(eYcz>{bP+)m-Ame{1kdL@sF7Q;AAOt6nTz*mlQ)+opp9}f2H*_ zL77A5i0dzU(>*eKDGI0MFJ5J#QK#FMHT=eI^wBqmWt-2kGt=B{_(=<P_!;w7Q zi8l3PTuN&Js+HJa;0itS7Bt{QIpB_J`px1MDX0pHX_2UY=Jv*Mf%uja28Fljs(NdI zguOMLw$6DEbLViwFH1`&f!oqQ!^LDa-NNg-hS)o%?;y8im7 ziqoxgTU7C<8oiIR)q{?4u+G%We(sSBvB|XbLFWBtspO{TyoAsEnQEpG6jA`F#PhxI zf38yl_!iN)&foPcTXeS|5^q{>uy?`=?4hz5YVQdf}E-Xgg#%Y={)9_u}` zWmFu_zIl=)Y?O7Yi758YQn6rS@wgy9OCh!Y7YBF5WVWeQgT3-+b8>m4(*f!~kNtw& zE_@A?RBeU5V0T`^4MnxzLQD{_7q8Ku1_CXVIe)|6on7-Boaa zy9sMcY^cE`=trrcB;LQMb%h(DhmJ-<*Shc#{teQiM|5jwZDPue&y)t{7z zQKq_JJLJ9na?U!P9HrbB&0X3Vc|*WB&^}dl8P`Yt=ze+Az-T%YKz!@X-~Dea?pS{h z|F;F{ErbJ$gX~qEN8hipRR51WG_4*+3=RDM$iF-QS9Ig_KdiKWFxCEX?SIB7{@?hT zh7upuq@%MoK{oes;TBpuKJKSZx+6YvPGyuIg7n=HO&@7b5NDT@9D~r`&6Z_F{Vr<2h22{ zwt=h1To-)`x2C$yU;OD=CU%*5s0Chztl+o77v}o~!|p~#$p3;T(phT@lH@78^W6P@ zQ^>jJO+!9Sq4&kiK5SR>FsBTJ-~xcO*%C4YPSsZNmjzv|_n$M+z~LNyJL(67d}Ptmmxp6W;xM9x1o@8wku5e@IgH zp#kj-^zRg4Q|llY@8p^k`~uxqKujIvEq?c|@Vvi5>~mc9UO=2Qy7D7mS{`5Wo296k zNx&vaI%i%L`{eJ;{nRft4aK_>_l9ZM0C--yaQm@Qw=D;6%wK3*0Z_F0>xxwu~s{mM8PiGB4E7q|2|}yXTMiZeP6?dU2~; z7NZrz;F8B`hkJhwtN3fK3fX!DGVOFBE#*^*KoO44yR&ZG?W?f6P8+7klZ@*|RV}#s zCukPgOWbA!b~>MD8z{)=8JR2qN8`CRMQT4fhdw{fg&Au>CIe@) zofCeOxCAO)^pARo+6QVb@LTGXFu{M+VZw7?EPJxaW1fwge{h+PbS{t@?LF?EasPQy zV01x?&uB#-nO=99{x>X+pQ@B}Q1Z3Sh*O{pKZ~9n-}&ap-FJ5+ietToZPD&R$qNDo z_Kjfc{b~#q&^;zS8?`-<#EdLy#;$@-P%UV#x*c;xs1q%o@B~e!V(;<;r~ZAFYeS-4 zjUzW3e<_uAsWn4pF)@m|7s6;}0}|(r?GO_+4X+C{(=Sk7OGBjp*u5wR6arKd8HW?gz}~$eK0xW80@0*;M38 z;yHg-luNB{#f}fKH|k!Z4w5eJ&KspD273yhYmGSz4?XoMyj43JC0^{VV5)9yS5_|F zQg8Z5AhN!d|G>=aYpJ@1*95iP;m;SoSU(1mDS#!(bn`fVS9X0Is1+h;FHbj#KzT8W zzE;Jyk7;SSO^CR%*gu0|ft@sK4jVfgy96KJI-(-B!qpX-G1zF|gj+k{awJFS?so!@ zQey9MmJ)@phMEu1O6=3VfNFTy9e%qAa4W(Qo zrM;iEc-~Xw?D+k1&DFnC^jdcc)6>-Qm(foqY`bxS?3JH9Ka3l;4NdbWK6xj!TH-SB z0HFnE_5oSs%x=DDEnd3iD$<-E3b$;ZWCaQ zyJOAro$xyLE;x@Tq2$pR!c`$oxHZjup%O@Y-!w0udwbf+Ldm+?ICqC(s3Wg#N)(9S z^T3yQuJIh^FdrV`tUKG*7fQc>sdA8g1T$k|@n7-DOmS)fw=i8l1yk~{>bM*fpFI+V z;b^?WHA>^>thUREon)KLs9>SXwO=$;D|jjv`}U4)+45F@(ltK*45};%VTOiJf~~T} z1M(k>MEf(}%kRT2&R5p417+ApAI&a&U)Y)YEgSa=>vG3eQsXv{0@Sv@0f)I)G;hy+ z9O%60T+DQ+j5sd<@D`w^G)ph1zS&+jJEawoezN!mO5)qb;7V%66VQq5thuesiYqt3 z?2KK=3uK+aQv#-|GfX_xxg%W#c=LPIq^hl6jUmgXT=ky^(h_DnH4^PPSl*-PeX3k6 zZ0FZ5I99m{!cqM&o&Vm~hoHmN(Om?gkyzLXIs=t(Ev5Q$S37ySxUaU^PA@mc89T28y!jLj#TwdUGZZszYbpSuz(d(3ZJqBvDQR1dgeun)HFT(&C(z@ zVrg{h*z+5gHswu5UmSY5-H1z^tltqJ=+o4UNO|p?7RFiMM<2&!AB6hN+!m{Nci#@J zrmaFUT1ebnic(6DzI(AQCHh{cxmF#165x|#9T=-+&ek0biZz1CRyD>i$XL+FLtCrP zy8%7Urz(WD^Avy5;iGV0d=da!J+5wKulaT#p`whoW(;%5CAJ|Z>M;28beH_A)bx~w z$l8?X8*|146a4GiXk5ZQIjDtkQ`7A`(_qOtuh=au#c4FOUm`JOg4tZTQuzi{@O#0 zPTqK*aQ1I6ofogYdHw0cm>Q z#dqZQ>{-tf{$kos z2ELW1*|PF4O-~)XDU=r!>PGm@qd0quz2sZ8zID$HXMA5vBU7zOymN`2hzT4f(Qe4L zFil!|X6c67>~~h^E2t|>1BGuWgnzi><$qr3{oQoR#i~N*j)4YhFxZ+j$vuUczcXxy zp=r#la)j4g0M=*M*KDh~6G8l7PtF?)-DsYrheKt!pnGL-wWUj8DTPnIYuRr|)paNx zb8uo*;BOh^;sQYfZ|r?Uo6caC3}dVZPL0lAY&kLBbN?Yz!x{7(S!tLX>PE`rBx)L_ ziuWAjy7(d#g##r&!xCwBe+x=k$1tfC6nWw6pq~kRhGZ62ySug{pHt$-$85a=O9a6-RKkN$BpM-tRai5fRPiSnN`#U7^VrF7d;| zYtpTvs-*A3%m=Fb`FrrgQECqfMR`ENw)H`&=+;}qp0hc2x09EidwC#5K#Z*}YREd}Af^+lgpjzoZ7yhDmQV6l|?~VF@#=6K~rchM6zv{>?IJ zzNW2u-|s?=Yn&ggc+aopj}e>$<9##LZ?@SGUJXRJ??fymu5t!Bggh-;Q+~F2Ia6Hc zxW%$JysXWuuoO$a+bpHSor`uuP0Vx0!9d=G+kRuHRu!?=Au5E`O>}Uh{n(zRQ`YOO zMy(5M{0{*toUz_N+h6~SUSt0~@*f2p>k53*4+B*29q0t1#AkF3rL1uF+?p>X!o>#K zSMCqwj~+OB(_En;&FL8LPbwMX`s_d}6$L0zrW~0!bIJi$5_%6BOx)i*S#oDx_u8mL z^D~peAKma@InAd|$-U09*WNVE`c&{X&6FwBK>=W4bl3J-*`pC|6*l7W7%#z8KzgTfdYwmApl zxm-6@&w~BN`9__H;W9U8gB6Aac4|VBf}W{1i&?6cf0sIcR3%n#Haa4S{)z$gcRk=- zRH4&;!c;@qxBL2(O{gEEz zYXC;g?Rn((=j+&spA4yj!tjD*hU+Jfkr;E1vc?y&Iz}dQS{Tc__7Th34>m+ce@Lk7 z0c0Mz(PKOr2GN&3?M^#7%7!o9X(fx>Ns1{D$vpXbSjDG}+MFxD_rrnx5f zAz*ZTD3051($*6MPlJ9mDyN%_T_*w;wceJUB?+({`9qRPA}Ob#~z9#dH0Gy zzt6!+_@WOw0B~P@7V^|#bs%_uO5^&POhFy7kb7Qq<5$z=d3mLCk3uFXZvy>-&Z8i6 z4@y3=0~8%#;!Ra%w_NJBjN|Nh&(6HtwFW6g9`(J9n+r+$V8i~n$rrRc`1w;g-@BqT z2iw0DLu^;W^np>A?A-QMJJ~C@E<8KAO1Cje9%cFE-4>g0I*uW6!22{7= zjZd@>_vSDkCIpS{Gw>#OL1U!|8QoJ=p(%(bf24uO@?*bo|8)>a`ABVxBED`#L>8D_ z9NF?;Byvw@Ucrao3GyW$WPe}+sF;v%@CorOyI)_g9}B;|hh+p-lu(99HP}vzq>L~v zBk8c}f*Y}35>sysPws{}{^hZ*sd@ z#1a-Zhg{xe4b>Io3FZbVibTZag&cLbRFKZb?-1GR=yqBAQ*=bS{8=RdYCt&jCF(Z1 zSc?r1Go(r->An1Qg`!a^3mCAu&uj>}zj?i_cvgAq5vxhkk7073jExmVUG|b#BB;OG zE%=2jjwSxve2kVV`1EUL z@@}_VeYUG5PWo$EljXbhs|RYi&$idzz%@fyVM%}KbCiDn(3H`wBEfd}#)@^tF9}_Z zieaAJP+rZy5LNa4$@Q#3i$Bwke{<#x$sjSMNymF$=gRWrrm_pVz02Y@H{h6^4j69# z%;+JBb#}Ma`zC^S`h^(sjJI2thv&HB%P+>B^Ht|QSSOEKW!j9SIRw8hz0{Ie4@gQS z#22k_vLnD2_Wugg37C;3Hs?Ms6Pld8kD2xjTNLYzu*bl}iHxlnC?(W8wwbzhWYeNN zYRbtYrAgLr-dP)yVvh80))3qRx@xGiP~8MigZU@xqXlT9)WiYBWBdvFZ0||1n(N3Tu2RU{A;vs_Co(<41p_DxUh!(%OI z)iwo#lhzxTwG(^azpq-j-T`i4(9%iBCvfA+so=Hi&8|}WA7ELq;mP3K zq#$R|d`QMmq0FA;jC>CF>xoXG>@`q*h%=py#D?w!$2VZxUypSXpQYA@gEw3}Fsxtq zy$;lNCJ2fT>-=LA`B&Y?-W0lkJz+W)y8>P2{5h6yY&%r9W;XhQbP`+DXJBu}%EEyi z%3Zn|>h?%6=tGvEGTchOzv*-o#pliq%u?F8Nd|HLS_+jGB`lKz_aADnyKqBuLvYO)hm?EoLjeCgTJk9wGk=;FQ19e2x+P@`>` zX!5rvy*WowE9n_2D;4GjKV`3?y~Tu)-gG+{p7^=o`&7};A@XoTGIfmralq|JM@5G{ zc5~n^goj$RGGFzOl6jt%I5t3tHvkQl_+IS)OegIV_G2v_Geb+5G0&LI?scXgGiJT9y*2ofNw?`r_;^m$gE97Y?2cfrqxW#>h? zs=VMGI-R?MBkA^vckJ9Bj%>c^EA1HOV8;Y)IB|T{E_2Y&^e)8tUMck|+4cyXi$jFfu#*IO}JQd#h8-a28JV^Wj;l#q5+H3zgTLxOckR@iPD&v4+5{ftV`GcB*a! zbC3Y~Ny7?XPh79}YHOp7LYO9XwCRlUg>8X5RGAKeJO*tujYTJ-icR#$R+Kyu+bt%8 zCiyl8mV0qFcz*9$c*g?JRr$#)gm98=O}QWQ%(R-*)LnQPhUCZdZN2clSPgO1qjMs+ z=JRG@HC%033!l}jsy2gw)3QZ#wiAeX|x z9qLR)93e~}ElvSnzTXDTVAC4A|8P&(u^D)lg;-p(i)V?()P;_KG~i$at>KXnyzv_8tO zJ718vE-;G9Iir&t0T0#v1H_TIDL;X1>tU(dWI(F$n&>L#)a3uG8?YOZTW^sX3x3ry z_|d64GlB@#gD5A>|L`A5@uPyO7h$aP>&4npw%k&mfso*z3?vCmNF^fw@<^E6CuEfn zAHUMkI(Qh0I*J3bgKozTKAI6-sA80$%)5o??e6Xxu32WNQ+2%a#l-4np2HBxY`3w2 zlIKaY1N7h>8lVTFY*J&i>c-2N{oEFLYD|ASUgi|?+#Pyuq!A6+iAbNQElG1Zfl2Tk z64K1H<7ein-qWCll?fzq3L*erp#c#(!19s`5TZD1NP$c^n)DqgD4-q&N(87qoemF2 z9XHlWw{s{0%CaLv+Y#vWt1k=4*R0Cf&cr!T+h_00|=0kvyGL zX8bHi9>;qCqI>&pK1&8F3FzJ+ws~hoa+xs~_CdhHU4Mez|hlR#sAn9{|n)M@bA(8r}?#L@t?X_BnX`s&oqj%XHLLZ zdF>~WwRQB2%Z`7(qi*c*&SJYEJao{gZ!n14@L}eH@2nc2l)9(8Q@5WQVI#g|TonHD z)yJ`9k3Am)*cG6q7ETBFeMmz((D`1ly+Na_QS1E$9iNN={IeG{X;pXAn;jA8>zgYkp6pJYD9%>Q+^a$oSOWHxKQ;%K;LR$~Eu|Go_m4shD5f-tl zxonPC-XS76g-dk3+IEvcDFkar13cFYVJ4v}Jy?_#uA?B~nDG7mC8ZOjB37fV<+E z#F;R1?j9F(1m4gw5X^6BY4f7!)|6{3ydUY}7-cikm zfZk&x7o(KsV9^FT)dr6`|E=ZQEq$)}#-taVyKMKA8?|xGQq9dxjt^38Z~MorL7swY zN|I~9qEPpSf#UR<<}AZX8IaN|KDYW-?8T~-=ITuEzdcnq?@dNjDeiVdgn_>c%-ic% ztny1DqxQn$JH0wur*z=gMjum(9(C!1=YnO8ect1TC9lq`M%`-3K738&or5J4dt2=s zY~v-c!P0j%dZC*gv5w~f!@0Wj6K#AY$yL>MaS5l^cO(A9Yo3C8qMpOqR4GJU@?gd0 z!abqm${&6u__ciO+-W7FCb^GO>u!w@gcGj zyD{nFj|-y9uMpQ#=uGdfvw$<}arSo=VCKmU602o%%_vlj4Q4H(>o1=83ELlqUNRz#+K~COL|9t9fUILC)VRd?!=7hw8#7!>f^y}}W=bj@4~ZV80) z!V`p`nw(gKs_>J20WD~PB)!q$&Q!GjcQT))$r(y2o2SMI#dB@sV5WnjO`VP`PBni( zauS0!lfN@WyhEK;|O9_`P|-69mY&j(4awr3nzzT=*tH*vo|+UNm74 zMA-~ABD-dRXfy-=uBJpr7X34^9V|%AWeV{>hE2{Gr!PGLucSrE@9VBGb(Vg!1OeS( z`UwQnTo1lc3UEvL?u})hQ>E*cYF|^n>+Cc%WNH3+yN|=DLdK~>qftd%3cJyzS zPv4D1u01ZVzq3Ez@&iY<*!1JX0%un2`4~D;p6J+`XN$*x)2?93-Tk8^4&vYV^O>c+ zhSyJ)oHkQix||W#P$6rmRaHRzqfSMNE&H8O=}|Z+-jCBOG5uj8d57|pYNACa zT-ejWfv)qs_)%wq8A^z;AP6DHM+lLAvpfe^DppoN^&3eZPgRw^O#K}EYIxCps)pY& z=Y@x)*yy3k`$^}!Yig9b?PUw0xz{WnJH>YXR@N~#@6`oPBKtv|z`&p%z@UD%->~cbGVi}=^OOfX$>}^1ld{IHt z&Xi6N{{cX2WD{DOgVQ1a&`9<0JtX@#3(>M7UjXQ(p%{hE*DX>-#<6o%FNqd0)IKf? z_U$x-bRK{vT#UjJe6`RM8vOW`wBVbf zs<*2jl6!p|gF)0?wmsdNVNZMFdIVHpj8HFiTMvarOzqBtDfUkl=$x+7Zlj?Gtsy)4 zGA@NJkvvX(rMt!wEj8wM>Os=XpJIabR3xdI8h2ZGuEyhP7IWmP+@1ujW(g?@?C+8GVKqLP zPgmc3wjYat(BsO{8vuYMM|E9T+03qhJ|Ojmy$!~Hb&M^3u)|M^H_$#v#`zV8SKEX) z6A`BbRfbAJZyL#U8w zs7srd$WA{B7Gg%UsD@?TAc4CZ8QvX6!9_#*m96*RR$k0M&pIN+*dw|M5J!{MWjx4* zf^0?Q-@aH!bnjPE=V*>pe5WP|!*G}xBNx;@FYg~X4jwVI@LJdQCsoa(2(0BwaZLWs zj$5_19D0cT*BOpwpW^RlG~%+Dlg}g>?}9tQoj!>xe*CacuwKBW0>uKo`o%Hq#*A}< z3RtpsLIiaz{fcmnkA~rul=9LS60$Zg{N5ilM$yEeeh4$o60B6~v4hBYQ3e!mvcfzG z8*tI0Dlf1Rx09sZhvV4tj-TwdO``uuiTI)|#29JqR_=ny!ln;JbFxV0Q@5-2gz?Hs zX-ualG1i84)zGi{DE*0wFck{ibi{}3cp~_;-Ly!DIIa&}jm1!PPlx&FcIJnChH|IZ zPo<{3ZYbnRooPf)?BC9^p;{C;_wK)PV@Ta+$akP0ZBN8GP^afl>zL#3(m2jNtu4v? z8nyISxFUMG!o)x6ZE@v#7## z!Xj@zUmWo}tEz{EFXheLseM~eedUg-ci%eT^Z;TMGA}>{Ne-*JmqPM;?kLm}zPQSD zLjT+&9q)aKTO;D5k1vc|PWvN2fdKnKjZCt($^v&vLjr0p;p|jn2W>!5|G_2P)|c~b z$I*aq)Vefg&LEXW@4j9HrXKK-p4eH{F3v7*Uz%A1X~9N-E67r9ES^HIYWF?ROMK$L z6xA(q^=j>sD?W+7M?`mQBOqcWnmY%%S4;xjr~gcfP?=zi2_}998cN+{n@|Wo$puUA zB2)p)E&SXxyBTpg!>?#9{n@>kC!aJndKMSSs#ckWe3>IJ4aTy3R$;OLzIGG<(ZF$N zl9i{+{iG?j;^`w(Zf&Y*V?u|&a(>+Qp56bfjx65_A<=N9cUU<1jlB{|s;@8f8PSJx$ z4gO6oi5mi&f4ca!j7Lzn)EC=E6v`wB5AJW5x4C`D(nynP$;EWBM~;Ra6aQk%=U9h|#9=-*G7NKR^;Ims~9r+h8o zq)&yOTr1B&dITT_J;2g}!nE2N@CgDKO{eFKVz+(Dic<4@N+`=cZj*KJJ&>sLhAu}7 zv3#>J*La8i4!&0PgD?;oDxLkPRDV+H{y3(4aGGO5S)y(5s z3NAC`b~BnD-HiHtZd!g0*}mr*5UY1Bu2 zjPj(NgQ^WRsS?|VJ7G6zA*GE!1GAW)?(+F(YKpbm* zd6_GjOZ}2EsOcRd&mgsBM~vJP7Du^B#TR&%f9_bexX6)>KeNRSxq@4RZ#K_5%e3K7 zRo7HUOLcuu|M4qxJo^V(=AG)E%}10lpe5maK+>)lbPzT=`Lq558O#fJM^Az=Vkb#& zQSo$(0N-n+g4u=llB`b=_OnlmRElm;x&`gRG9NBpKW0BGK;ZtMKZig>WVgI*xtVtJ z3j6ixQ@pqhNTa6eO4i&X#YJ#{0@H@{#KAb2LnOa4iu*0t6Fli^Mpb%{=M#Tp_f7+6 zr8?h4^?U6XZh!+&U$cBW9t3|x!;u8GrD_5@Up-YGjd0_(*{g809z!mmFl~%V@S*Bv zeBi)|_mnA|8w^kN+HZ3?KUiMvBNsMTSLY;`l3yZ|l{GTjXl2rCeSN0-wcp<`^{_gM zKXrxJ&x)EnDVa8vIX8b0dM|34kl&KMXS7*Eb6T%EEYoG9rYdLN4X8A4Ra8q1+`*iu zh~zOOvbuByUvUIDcmL=(#sf*h*I?tDz_WFtu01?YYA@rp#OQ=msz`Iv153Im<0i+r>p#*rl7RC^bY# zC*=I9LlnA$`ZLRDIXwEB_h=yoFEM~y2DMj1PBI78>?jbqF3E3#%stKxqyyss!FiUq z|F6O`FQkik?xU_>??Lv}b%EKTe-YCv`4I>S<>}gx*zeOX$Ty5`$WKBAi4aP{*fnn* z)4ys1OMV#Nkc@tEKCa`sPM}K$Eh<`9^3JSl4zON6!L5aEZXTiPzsm$C_%VdY!&Z$p zBV~cs9WqvyEf_!IdWZF}HmGELLqZ_qv)`ha=~+$xp~i?yDA~rw{I3Ax$?E(+$yz(I z1m|fwn@2`X5soW9SOLxSvwsw2gTA_URqsNYUab6{NJ@&(!PGo5rt|USaz0WXUM7YN zHnL{O6O{z6vrl#{zr0TCT`>K$z20yDDkMU9c2wib_9DRp2fM|QiS2k{PF=`moKs6T(A`va|Qmx)^$VKxv0pSLvU zDCih%4-GS?*O(BAdaC7uBjQNd7^d?KGUx1EE-vC~1Q3D!mHv3hr(D}vt*`_L;>w=2 zKBtn#qaM6cih41UeId~jVm#%2azQD9h>>^S?Kqf;L{CaB3_M|P4(D6xGqFLM*>6x` zMr~12-@qGp9|xLP?w4wr0-}B7aya>EuY)t@Ni!gI)0JOy^vs!O%MRy|U*<|^o@!GE zxMM@-t_BQ{>BguLZOJ|jW&SIvYf{1M=vkQ2!C>cjWV@=p8MiLUwr(Q$dDtB3@Q4#3 z!ULhOQ|bJ%X9fI@e>s>@nDs!XLg?&46sB(|dmd2*cE$46zFrW#W&}0*=!!U8&Qvdk z71_@r+EHh@Pf_s}V}iA1w^{*lLOSX_8THH2cx-u6V>o@C69$_90NErBR^((sl9(B^ z#KbVXXMDCm6J)mcnxz5E1s46!Xbu1UYy3nHLh-yo-SA%d2@sj-JI7p0P4uFDe{
`G-zlcq#q|6MMF_TFi-J}_n3AY`@-D?OD zqQOg~rFZQ0iCfsmLU|t04CH?#^l0RbSf_maZIANU#L&l)y z4#~xUTHiEedmmZ(@UG4+ceylnuI9BG{@sBkKN*@?N8d!{p}*~;k^(`Ii)7cB+wiix zTox8eZI-QrRL?F0XzqH{bD9txtSd)FoU77~EU#|b)JCBX*T0BD z&%*JcvWL_6OGFGK)n7hyjD1Ju-@j5y=3ry~LPK_pNTAdvSMr5J zRa7Y)3Uo}s`rHQk|12Frj0uVFZQnN_NE~iIBDKS5=H|nU14-O>_cxa3XU<1KU`UBW z_Sh!AQpa4L*SdH1Vy6!&bOv{Ut~_iQrbEq@ZykOus)JsN9v`QnyLl{Z`=EHquO^{4 zyAaJrRz`f8Z!1e(2@qZgkLK6qk6YL^jw;!Xd!=&M9lUs)pRN5RIUC@;ibz0K8FfOQ z4M|dR3{SJ}f4D4&fy^lu?I=TF)YBAkqG7U#HT9DP!FX6>{$|4p?Q6y_`@*;$WqxwX zF@4FVD zPikY_T$$T4@HWR?u;4Z1c;UR;VM0Ul<|5^%8NImT19)7k`|;KNMvFggO&D)7Fk&Ry z$88(n;XA=lzy7HvCwoQ%+7X!n^mW?ufet@l*ho)6(nAo`d)V`EKuW{Zw_s{PD)?*x z_loo8*zK!;oBKpQ0DQoSd@!K4=r|tM!G!;2*%5-1Cc$%Y+V=tA8IYKBf~f@9H%p_# z%J|Jc+Vg+C>0dYEUnAmQQ{!K2#J?|=0kAfhqWh5=(L!ABpEb68#FANdUklu1(fxD( z06+KMl%k&{ol;{ zlb4=%Ve|Yx0=snmGTVQlW81%93-o*J{}XB^f;M$#(HvB4*W}xh=L%03{SO@V+@0K{ zWVSL^<+ri_BV4-Bcrv+i#m*kZyH~;#q|GTYSaO+60IcX5`2uC5+WLs+i1OHC;A;Bx z$>xNlIm*|q(8%kQD|>#Nt9i%VFH4W!v)f3?M>-vjE%Di?IU{yXo?PR6nfsPP`O1I<4MbAux6wimBYlC6L@Y_2rJzt zt1Cm6Cidyaz&8xNGEK_&s)hBhQ+EU-`$ENZm9lj)go>AgI!_z*_fIK;<(eUvnSEdp zI+&!=RvUtN!fRi5%!~&s=rPqlYQ0kthWaS<$i~;>wr)-D^=q5`-^`4iK|0Klok;_t z1tfCjL8stv7IaTqKKY=FcC;%9&%5PTcRl9ju4ZV`<49Zu>=#@t=ru!}u1s8NeG3#r zh^C`zgh$1QZAETa;)b*R=4oQpZQ|C`k72{s&V{x+^mKpBi=o0G*NtDN2(e2A+avR@ zS1QT}cKoQ9itOd*!Cbu;ewtC<=HnGKP?dlqqO1ab7M4=Gz||7MMDNbrvVY5*hH&~q zEoUD!Sw4KdIHdlN3*8cyI|4xA39t^Za5aKE>T|2E1`Ry3JI zyqp3YZ<=3oz)9n;`9@uFqofsoz%UMS&|c(b1#exk*Jj$`gGUYmAu)NXGbKd!g$KQ# zWK;JJgu@mx6pTbt@%ZBSv8&HW5$0FH?=|HSM32;_|BJo%j%w=N)<#iLY!Fe9PE-Vx zB7z`25fKm&BT^zY3erJDT7Zx!3MefCB7!t20wP2}dI=DUiu4kC?=_)@G{5EE`*+UX zzq8Lb_PBT4@7#OG{)53-te2IQx6F6G&z#RQq0@mJpJH;5pY}vwR<5wrSUXic!Jb3z3mZHOr{y;xa`!#j) z_Dk}QOz!ydUo0v4R@?v5a)}2!s0YRx-LY9 z0_RR%o}1S--8`~Br+-$6Md%>fiw=qNTZXv^ zlt163|6^P;>c~}sSp z)CClIZOPQLurNNutps7%xuR}b6#yIXoPvd2I*s9>#I3S<%m851!bz+KB1f24<@$op z^P3N8SL@5%mx(DQ;8%2uG^D=XdsE`2{;{#Mtxra8vmHIMq~k0s2H!P#@g=%sW(@-! zKMfH!L&;IHWP+u#HoLo~_ek2l+WDPtfogVWVA@GCwcdkwq4^jn^fW5Kg$$bNn-7x) zQKcKBS(oXYpM=wdGrI9<&;ymalL04Ra2A*Kmoy{ZD$3PcB$`4`LS5JQl-wK`Gz5c_ z8$CqfyiPHK9ld-NV}y8xDD|Z$L5i6S9P?!u-i^)>?>Gt( z2*Do!LO8*hfp$Q??^)cS9^D0y2?D0t_v_t|BA^3K^xm(PD;y_4wE^fhbx*{fr`w$& zxIHueb$nIVo%GDM^DA#tI}YNx>9$c=7&@kH6z9B+87zj9KFX_Md@um8Nm-=njTm?U z;|gc?i{8<7k0u`Y4%7Q2|K^iwSPh)t<;@O;V`;p^w?)iNQ!hF3hjnCWpZ`PO!~Xfc zm(h(FZZ)_BJ{+aJ^Spe+7Q@(YxFETeI+;xu`93kKj8a+X58!B!gT9={se4)pS52vd z35i}qAM>Alv1pKMBj(mkJlvK~5z`%#AVi(KTAkcc2Oc^wX@G5j3gSXs*C2hfMRNvO zo^P}l%+4jP40>+W0tb1QwEdlCD4&CzuYc1wSc>!e3Tq zJppjdP8jvUoFiP_o$|x#s;oht-I4%(msuGEd&wVIDYGsjQ$F9XbQ9N)=cW*@k$fT{ zWd&AJr#o59DjsF-YY6g>9>S+$v51qyQ`pA$>nE@2eoct|G5xGmUbD}5+dPR2tnj0y zS5Gi|C3_{hqh6g0q>JD3irdqn*$h}&&;y7z?`A5>js{{pxX}enAmM*Q%L|b?PKD@E zT6d@vr=0o%bNcmSY+!X(#ER_3DyMdhg}8X9NoS0SnNrSJ*q*Ri7PAE-;H`hyJztU6 zjt`LnW~!QlNlh4`KroXQ3S^iQM=(W*7U@$hwTHiw7_)WQdxgi(Yo_?BMO3ev(9&z0 zHf=>zn;~C6*<|zBhr*8B)>|wlYftZ-r%ZbCg?Tw>4pc+wKDFhH7$ysHcVZ_mr)|oy zy8IHo+LMnK1Y4^T{w$f!_zZPoqZnnjWK#4fVpSeVCq9(A$)-*LGuPG%x{^GNdkHAX zR13(2q5Qla-CYLT|ImInv;kOyykm^6Uc`8%`TNtW@LsVc=`*|#3MCq{+RaDcvE ziU0G!o-LdVMHcwPaln`G1~!tbxI6BZ=(skcDF@@LIur4$X1`d(kK|#9fc6=u5W0_% z;(`kuyT!)Q=|O={W&i2+H1f$=XfRNlx#cueWwi-)LWzVI?3f4FGmjHd4$}+OlIgSF z#}Y&bzp>4l+U~v-V#^P7gxYjOWjcU5XpaHv!O-UVt0C=N5_o#px$6Vxu#&7I@7c*a z7c)LIcGE_n& zDa#ALSawH%AHb%$fj-0N-o;ytwO&pWzR`W{X zqnPEoh8;R4ayYtd-B^U->cA^n$^YZ+xlDi#W2vB&_q}+@Jp_G_%tpXJ3nd7PSe2Q6 z+na@rCAeVU3pua!hCJD3SM9E@zg47{wv@n|Oy~ z%eu833o?SAw8hq-^Tj>+bl%|KQ_uge^kef0)&G{bl~!IBwsbnkZX~kTc+2?5#XR+p z_qP`)^HXdu&XeZS%w^yEp(m;PQsygg%D76pLE11SRR3+mgH~I%9#$~Gjy&z+uhkxo z5smC#^M z5>t;+24~;l9#}Ew0`?$SdP`FUsDFW`=Q56+ssh`}h1|{lFooyExZ;Az6x9~QFclTx zrr;WPAbn4lKSf_SlX{eq5QJPiXO^xSSINJ(x$~Ct5YNdJ9<>;5sipQF0O91$x=$TQ z-^DE@tsg4nb!g2q&zMDBnL?~o_n*pHJrYo~6^On?=Peknqb@HDrHt5Ltfg$%DXj#$ zV9SG7=aCNu%#N_~N8ZajB&3D5;!~>1puw)R=GtEFyIc}CDc`f@kTcd?XnEPFTzGqg z4yipcrt8@9O7__nD=}ek2|RRy&HiWjZitF!lbaKdeaoG&8O#xAPE&xOYC<3ktk2n{ z9w-d>j7&W;j@_hg%0t=BJ&YY!<*;de9J-$XshbxQE2%iWQCe(D9id{TczLW|6fp@7rh|EEjP4>Uf`8NEq@fo8~ z>Ypt7xc>2a)$kUd({OMtse6J z%J;cVB~`aad8lM-bYo$ww*r0n6wP9xwE=AYbIHDhr%KBV6NZ?<%ZvKs>#?k`>2+bJ zq2o>!cJ@+%S2GlIn5%wT=yWt;jODpb7m&+4R5U4i*ll2Ez%G(1%$%yZm3?G2uNmY9 zXmH__HeZ^W#q~FikCkLB0isNp=q=yXgVW2H5GV(}SiY2VY~;+nYpW$am^;7g547aY z`OP|6{w{e{Mh~RKd60RUiHa}x3ST@sx&uBJvXUKBv)r(JYxgp>$+wl1&&j+$Ujfwd zcv2fqE#J}U3$aPbMI-Rnm~)Hr#07R=7lJM))kZeryW6kX}oo z-ARv%+lU8BEa)wOk8bUGwaICC=9f1#E3#MTE*ZwJmAV|iEyi_!-f&R(qe0A+2j%7D zmoHCZnFTbFUWg>!mV|FeK4}2E_fDR|7(raUH>)vgQW#mciHI2Ig+M6ysU}+TbzAnJ z)X}=D$r;B|-$}&ARLoTmH=~uvY*8hC-jV2;4zZHWvaNPLl19UCdZ5^enUMql!#-1Y`DBJCN^^ zZM*Wrh8^bLyb(a}+S`f8;d=4Gh$84yI@@u$|5sb1UU=-qO_BLICg)ip4#_2TksCeo zlJZ}=Cr)Vc)*VPiXN#4%tZ%FdXI$@&uMydUhtth|UJ+aDME{7ltJ=nNpD8O1>Oad~ z$*Q$*5T?bwo*H*`>@%=RS^pfi{r~bc<`?ld&}ggA|9FjO$>SObknGUH)3*^n4R^vZ zdrSCt+4ibnC3K%P2N@g%;aXDR30{Q^1}erEsV^lAOUTZ?6xz3rkVZFy4?@nPl*slU z0;ky}S1udOuDV}UZ0~&yE^iwRMkfZQk8DRnWp9d?3Z2)WgbgV?Oq;Yla_6>l;H!lk zbeH=t7IpY0S77=$(;qh{DR3p=l;Iu5U#~9w@kaKip~`r-Uo6kZPt=MzrwFFjCnAQs zJ60i1o3_|Tp$3Z)N@?Gs#%GFv3w(5i-*^2jzxvBORt%LQ^%!TxXl1K6?+oteHZ+ z;|sdRYD886KXQq75W0rik#0Vu*aQuxfNL+A9` zj~h;$U>orR@`O0Kqdfv?W-k4|SO#LNV;ZYnn?7yN=d@(!qv!bYE!I`Qxmh=C%BlaiQ7rH;u~W>gkvMwndO=>+po{6r#mWe@TQ%W$ zW^|J05MCUj%xp|K)3rX1pxg_Y<58_`Kl-tw5<)N9jJdak>mUDW0Wb$l=uKx+21pMz zDPb2)-UvT44N5GA9{0){A8IAAEb_bPvMiRJ#;y0x6Z>mKCIS8bJhJUnRBh#Y)00l~ zr%gpRZu2ht8t-1@m8i+vIb8B;<{ePI;Wf-v$?OFBTp2RE`>O0c(Y{i&97SZumXbY* zkGNV{t#SmbTV_w#BMTqb@^_c+C9cdk(c-k33v3%7_jd4`T~OISzU5~ZlQ*kgZrVY? zE9eKzUU1Q&iYfLdn=L8(V32^w5Pgpl^inz~QI}-}w!hY)D;p^IcoCR+hj(JSEq)J2 zF|};Wl}aV9eJC$e<2%7Tmp*E9TNeurb~$bwQw*1V%dYq{8T)NL6jOw6f`ujiG@ox8 z96ENd$R}00Z1yHl|K{$c%O(M)0q8cw$*u7y-TXxrj<%Cx0*H?A?AWg)gzp$`z)=0m z;yF;G;ruYB;|9G+CLi`CQ12@Eu;g*)(d$n!5_{mCa6r@@7U5A^^59HSe}>>qu3Gwv z(6LVu62$&-N5D-^$=eHq{B6x}zYx7pmGRN5H!c&~0L(p_9{o)1Tj`^tt5d9Qk9XFq zuxrIvA=?N%$$iy~Y_9ON-zK2SXCk@3YZnlw%w_U3TAZYY*X13Sw2p};0%b1)4;SF_ zYxcaAA?c-6g=Co(B2CPkE?FTgPL{&E^kFY#`RrM$q?l@(etR$+MvMBh!6-0~d6PWY zcRMERt05_M)suGBW&HHRVlpWBec)&TgEXnl7&FHvBwwq$>kn^K5eG28p=^S-s z?QhE@^UAPr(OJ5d)y49vOf7TEu33jxL=jL|*bs=UPsY{0nyP{pT8piG+(s9D8@&_V z8Ai?RRe7V%Ym>oqdO8zWaQ|Ty-VDb+v3;R@_a@zokq-1U5SdR*ZOmJBs8@jx%P%*) z_mFZbb9Drb$BHKr7eFg`Qtj>}hM=!50}ub+n>jvTpwWE(Np*-w2E0owz_$Me^kvrH z454W;?xL1ff)PPm?uq6MYjd6V8RAq+Q^z2FS{P5p7?MxNrtR%q(cK5%oU{y%G$vX8 zhR&S+m@!-_fjIgW?5A;8-UZc6m0qsC1nw6GLJ=pvUH7$UjpYsI~snxJiVUe1E~^waD#r!HIoJI$BgtL#Z2C!lvON z+}f7k=;4eUjKg}kuo9&&ml8`b7a!VM92?23_0ms{*3J%jj6Ac0>q|XiHgXT?^P_w9 zsy_dN5RDc{Uj;Bpi^S(H{;_d!b4&ct)yj`NtD${cKUV?$06mCdtvyfICVx!MMe69% z+^P@mD!xDIeKS<&zJD&R2lIFKbUPX}CE;XHKYF%5qeIx#Po@qFS{mb}ATk{jSHSkw zWhd9Mk6jNI`kczSc?P!N#A3l#vD2=U9xA5J~tLS2~p}R4BYg+ioxd1Em zZ$HNu=A-aFibj3caDxr~I=K-{<4J{245G#sV5`fzRg#ujOWg+O-i$P;Z8EIB(mqa+ z0ALRyo9Lc1_mg7YBV#m|{NbJ8Lc6ji+9|(mWS5zIcwnl7(X2<+Pw5}1@c<#0t&1N^ zJVL!5tZ7|tb$vNzl%6Db|5@9O!jm4ej^&eFc0aMnOm%cSnwQd%y%#%0GUy3G8psVCOHp6M+fu$Lmyc1MxtnO1Jd3f$+AdS?KknNkJO4I)DByJ+CR;Cs~hd zQ&@)1OhGbDo!JfLUcU?CUoZR0#(1>sGmZZd1x#E^fQ+k5B5G1}IG#d_GQwm`~sG5t)R;K(9tv72Wm_y0cjRU)dqe{sHwMybAJ0Z+KNRlu&+7&;SC%q z`#Ge3fiS9@8i6^9w;g{q&Y?i~Ojok+m3}5_hAJ<8%%*$({64Ow6V~m zskjb1Z7*S(@sE%1w1oVgwd(z6W_UL#aJ{D1Rmr3d*5Ce$KgET{3(WlXeBmmxoOU5r zHCdoF9>dCE`^GkH6;V(t)G0rH-03t`EcQgm(8!5TN7Qe_?!t_g&4U6Sbl(8~rX1$211H%fS38Fj6wBW8+dgS#<7j2{AnZ~5NIQj-h+q}5 zV4MXbkvd>7r&E4)w7ZJK7_idYJ^$hG{l&q&5C`C}10HE_Mrg6{8DI#VNFq~MEK{tu zc+w5L`qu9CE478#OpmCDy6r=F)7ro7wSeUW{6@4*oJ17IB1eg^_m6ggNuM}OI#S?P zkAhP5`*$XUj~dOEDQVGGv?qsi{^S4t@M{%))*Lh3KrQP-;2jjDVro5NDE^fi^uznS zBPsi1{741Y8^e$eY(XGrhAV0HYk#72eoez_^eI;Fc~U+MGVt4l{=?bHcQ$zwW8i9c zJ1pGI8-k%M(;!HfY3;WscTZCz&g>aYzJH`QKg*Fe11ytPE7*kiDmy8_G#Mo$P*W~) z>d^DKXNy>~wv7AGv#{b{}Hv z9&J8nC;;A@T3ohbD;~U!thOZ|q%Y^}MU7x9Z+b)JZfY!n^cei0hj{M{@O25;AEwTj zEa&|F-n;wVzWzzO;h!JLpcMRUfzY+rKKL02R)0WNZ-qcQFoh3-$P54#3KPW#y(nWg z-FsbE4``i>;{HAZp!Sx3SX^Kz05=M4Q=DyT7#?Pk*qfHw6L_S_q((cY-&iBY_}SgR z_4yyC2h*$e{ZJu!@Zj{+Hm>Ecnjs{(?6w(if*J*R=v+qR-tsDl>2avaq(|E zByc|X>v_byL~zy1RuAwiQMxn>X1fZmX*odMoq&7447##=4_v*eR=Atcd;FjR$uyEA zzLQOAAC?vY{F%NGD8^O$ul~%!fA}*r#NI;cyZnn&cHAy}y$#xT=axKvv2f&lI#`oaPesgssqqRDQnKj=n9ad-j3oP2uNziMt7$c99tmUK7~J zxTl}p9ln%Xg!>FEkJbJp>4cpli@hm|ebuv$%RbLR=*Fj`&Q81M8#3D(TgQWT5y+iw zMD$)f{|uxVW1d+<#K!`mIAVk zVVS}meh$J6o!cxxbi(aC*4OkmbV(SrbZ6eU>>W!6?OlyDq6o6zp(!GTdKQEmnS zB|K92CZXExJ+Kz?I&HO5?M7DY}QlEJF0#gnnx0Q3t~6wU#G4myjqo@HO26 zLi1JaAT6a~aJbi1&O`!V{> zFDFZfU4o`y4T>!`{)u}`oKxA|L4>B~5kkg(cJz3{{ppv4-Vf55i7JnIv#(c6$`tiL z)3ku<$-aEqa1d#}(PawHwn!aN-s}rykeUM7vV6Jen3ow6-lDfHvTdnPB2Dfd^%z_N za1I;C|2^1oXGHId#(AI)+$jUe=MI1)nK9Mlo4?59FLMnH|GBACu>C5&Y#W!TcFb;e z_`A>O!)EC5Fhqo~NNUp@Izmy8-NQmPq zetZxkOo@CS=yD~=hA|M3w@rt6o@pq*W9}l2`8KaG9Q~#g5jg&VyKOsnP0Zz0q7Q%* zcBkP&p?#PmN?H#qxQ|Q4Ws9~!$@++TGgM>;hZ7I~TCrQ!(4|FtR~FTYhg3MQFM>}F z(|IZSAL&=SX7=waB0Tp+zKgb7(B#le+1`Y%JfEo2Rnzh$vsqY`n)9-r>Z+cJfwZUz z`Y6=X&B>!3So#CM{2zll5?uanCI}72(9S|HwU)aLWz^;G?J$ASwZ6ibk^3Yw*+<21 zAQI-DsNS3%ju}Y|td{YRYj`dibx;!lXSPC*g+`*x6p?3x==-Nq%p5 zU}v7`)Z&9VH~F|b&(J7pg;urWPI0W=PcoJp(eh;U3}CD$%bM!%T_6btZY6#{MV5M^e2OD^8(`T680`rYasXa3-^T(8}-B0kaA zGqJ{bo5wrkb6=&(mo8vet=T84FBydH7@N#7FM7LgS!-AkpVkUf!l#5wPDf7CY~`c; zl<%;yZriZgj#ZAo03W35lO3$F20bFO!v|=CwRcar*(ga>w2ycoO2J2KK@|Ap?Wjth zP?pVmAh}zGQW`c&_O7-bi&hB`9X0*Ax6#Z}HP8t)eZmaM%PF95tTlZTGvgOphaG%R zKpc~r?)7Hnscb!-aC>&>31KbNj+P%rO=3C3w*PGy1I!*0%zMGeRENy;NVfS)KB2)6 z7`12x2t4I?Nd4QKxZnR`&f>Yyh!CR@qn!V3Io0&PRLHsTKi%fQ zuTDl2C4AZi6k2<-mTa2E{P^>dj#@mRH=^E=h2}$PF|upVQ>HE6QV;?buJq`Gk);v) zc1sVmUqR$!NX6OEjT&Uk&|$IUaurKAJ# z59al}Gh^dcahhA!$LqS{lucr)aqgNu&`A#@ka!;iK#C{;H9Vz5za#?&az704>u-Ls zOu4i!;JP5$I(s}L*m7#V%nZU~2}~-l2J%WZt^y^@i~!k*3?>f`76?;|jqm=zkQHXP z47XLF-)d*TCRz0GH6YIU_qXLS4O2ko2ZME(xd&s4Ro907_36)-zT%?;TXO#KHkM*( z4J7!dy~m7S1PUM$P+Ur2m_zjuzZO}h90HWo`!drm2_bGSUEEJIErCy*W zgJ%#=K%csqTD!s@{&fG(-`*Ob1vbh6-dbbZ9PzDoy7$iw{%iMSEO}v~1HVD?^npI| z5B;2xlN9EUt6sOvgFy5cb z7}Rhe_JA`0?4+A}#E^AwAgn#Y8zI4&+VTU3%_%UixOx&Jbe^frE%#ZTDdDmP)l*Xh zuy%Ih>R|(nCe58@?VpXa3#rADaI1YJy!qhiFXMx;P|}o{dce!4tB6-zX|JCsDNNfS zCHEOQKEaFm-HSR^-#c`53z~&sr{4um^V7QWR=W0^D*Hdz;>QkjS8V~J|JujZqg?h- z#eLfXM_5aXi{!TIaUbaWXr}agB&p~Yj?70=J)cF-@h#J}<$z66h0P;S-^EbL&nQeUzB{CE+Z$G?c8c>_N^YnkGuqmsZWy1BI&Zzs+>6#pS?$E zk0=-Xqr0A}N20W$;*FD5c^WuKs^7-F5gcczPu?NCNFE>^`BRPbH}Cj>s!0zZ9duEu z0RBXYsiQ!42%|%xq1(EyvojZY)vvq;|Go7B5Z52f$3`DsdZ|kg6BOi&<9e| z+;L#|n@wy^&NHB5{f0dPoBzB~hf{0F!&hgeJV{=%+iR{MVRuTsH2*;b8u`%@J_%b@ zAgwaNbluvL;H!h}#)JhAen2ta2|UE7Mvb>H~MfFP7hlS%Ey~vSt$uyIMAI@18%}7wW$|es~G$-fiU3D&|Rv zMHC`6-*1@YzHrC*`R8|RSiB+?bmh5yb%?TTaGOV31bs|reI9M)9^qQsnwwz!z;2PU z&oUC@JW+Tw|3a^o;U+n*ucNnPfY#)(ga*Z5bNHF)1Kkk$HkLA0vZIB*!Td2ME6h&` zcYpAhCXL%Y0-fOereixoT+TWB^g$O9cH5+=?etq`^Ta9XpYGG`#nIO|1k^Q zww8qXOVCPN?^TA&f|7+abqo?G9AC9%UfeV#i0S*{EwNpig43LEI|^Er_O{%RYh<1w z7Ft`1W)-*7B|yxQ$72TzhgsvJAk3>wzC)Shova;~gI_f_^^=&a{0vy5+K!U5x64*g z5s-Ns3Qcq}^v-Kem{BCk_EP! znWbMrqmlso{ss&;J=QtDZkuH#f$BDdc72{#$a>b$*9`ZGIORSQC_YtVg%+fX_S6cR z3B{K^Xj>FQo82g+Mnw)VTxJVOx+!Iiz9|=pmlOhe5=4hoy-bd(>>Ge9z#|$1bQg=d z{K2y@E{DD=YJsk9@Xy2cA@SLOI84{QA@PLs*(~tYj(S+A&p1alk5g7~ThD$w%(~rR zd_;7eGsR)&I>k(^tN_zsm#g*O0=(_FKx=&Ie_4dpAHH;kg@sK6;na}eE+u?UA71uh zh18B{WwW-ZsX3T=eqU@|77nnaPPidG!sJ-8Ob8u%As&FUr$t)pwVCGuaUAvKC0S&- z;qzR>1@x=eGfGLVrvxvZR*a?d_|7~)_u{{1;h($OlR|^DM?dMYM7Pk3>$L^txRvh) zkP)wU%gXqJ38PlZeu6Wxcb)y?#`66{x0e(Cf^z=#;$G+nyf>^$b$LP+)_{Unm)rqcLA2Oz3u1o_*&0BJXM&>Pbz`iZsC^#@9~SN$ws|Q$6B`2~pY)F%tJ1rG>-unzD}(z3 zJ@jRxalWGIXPkJRD*$L@ocO1(LsU}YChjyHPU3m?PUcmKR*7vVtDicw9F+4hHsHwD zGWF9yNlzw!&qV5wTIy;&7I6;4RJ1OelzC-+jQkVr9-_cqh1fL%q1)L$e2P<+?vlu~ z{<*f0UXI|V>yaFS>_b4OocFiQx7QgCt9qnXf^u?oYr@>5+n2(=b#9N!H2Qm)mtqx% zr0v^@^?dbbNRR!)h?$quE1+DIX@c;vXO~dANNEw&M4Wy5W#|jA;a!SnGen4N7w&9+ zaGu6AnU0Y#Quln!$IWaHT3etF}*`s#DC#jUXg+v+~kS!f4z z^VlWQ?u*(BrP))*3SHdG=}l(%*`y4L#+sIQcLq|oB)7%9CYviIbgcvpV~(N4`?sK{ z7owfYq>om-KeKP}+%}cC$P)q)M~XMYr5z^QZdm<%_v3~4(h=qvb#^oYQ-2*l^SBdT z{F~&`-Jsof(#}?tR@y~nqsfhpE?21#Y(B3!jy!jfLT-*P0@J>{fczzdar`+ zixb-7BQ2#49-W@1-r%9)iI2(9XtBy;2cFYj0oTszY{|u^wb#{aId%OLm47SodsBX!5zH7sn9m0PKftN|r>6SDaBS=ws%=lNdTbVnlCN0M{pdX@+vljC z_~vs*`#Qy!7z|wioUGRX+!O^kW*k&fHscV$BRTM!H^AUbuW{cllLb^Fa3po`cB}v( z^}x3``sa4g$#H8g03jM@Uc&&q=5%>OX6YPsbDO!3S!9cupN6X9|DlBrU3Ix3bqjdf zH_&gDiY@IhX#YqH6OFHf_TO@+(;eo0|FH>x;e-YvjhN3dd;5N|08E3IT{_fZAZx3G zNMWOXj|AQ@f({ske`o|It>cbjI)TKhGXQ=igzMtJ@QcO56TC+Sgrmyax`ZmC1zvt+3w&|J1bA6Yk%_*tS@n-?zG4Wq)c`c6fu}zO z)HHnXHykqe%k%?`ht0rCHluz2)CT*HT~Azy_oE z3=ZhuWApE^`S-f{_qzGFP5bwL^Y8uUf6f;uIR%hU)#;ZBM{sV??X_{bKVB>6zjy=p zf%qLv(lfOX`5d4Oc5EkB51RmaBY~9W?yG^SMzjYjwRgpr>~RDBlIijMbDT8a?VAhP zO~?U86&+g#I)(n8%2%you9?A?W?dlm!ruoF6Rn) z^$So#%Xa)k2RkxlWN6v+q6goSGD5TOaR9vP1^MKs?izW9ygEuz(AeB02pzh4?%TJ= zghTHUl_?60a5NY4&G)PEPK;z_T7IpWY)|29jy_U+BWA#13LI=3vq^w4N**P}zMQO^ zrgg`TFio}xF}xFbi)}yXfZJoc_{GxbMSixT>k{)cLU zl2I?b?(2A(`FDhrLYK|tLI3PW=-$B1`$ zzjnMlo#JZy+`AgI4z3VHXdp7Jk5jOdAn-l0jLG-gm9Ij9gV=jlPwul)$D>f7%J3tH*s060 zv-tvkTbL`G)db1>ka+2P!y5`YfV_T(+Qq})>a)_HLYyi26HUblC9R^3PXbBMPyu6a zwYlo^5#Bg_iNLWyy>@-oxKY99(OWa{CZ;T@uOrmkOgfV&j#hezHJbQGp5^PP?t8Ju#B zU-ohbMXBI6ph_4=a7Y=H9dgEc@gSz_Uhb=3EH!VmAIY@@^-U)E;s$`1TAu7n1(KVx zBeuTLTa)EaW;%%fjFAV5%K-;GWtBHJp*gs*Vpl0AlWK*C66**?E1_1D+0%OVZZ`Ss zT5BIBQI(y4y6r2c;y4yJ<G7$`-NW%B{PUUxCu-uTA2jfdF1($pk>ajPVv{bKqYsr3STYU+#)~Dn8Q+| ztu+zXZk{skIXXZ48YA>MiPg@R~3WrvvsQk6yUc2rR zmw*K%a(IsntPD{96qzT~b^Bh_jZYYIw@%s?uKvtTWp3&R8kq9a1s*S*_3WqFNxwCH899{v04FU_b_t25m|{ zYfr4c4#=Nle(HnkrhB^a79{+k537ZnLlUh--1ZzAF=cAodYjjq2;UgkVn}qkx{AE; z_swGyW(w`7yJhWIH?+vTT$3_EKu(L--`{dbgQaK|aR& zn9p|ihO#gL-B3F;{_+8}quY^JF)ORAX1RN7Xv)^mpx|cWrYbMG>q)df5T*D1*%V~} zaghdvQ*V*S3qKqW2g&&RyrE}wp~mnsvH)JEPlyNbI_a-2+@n%669Fu`OF9oM2M$KD z86h(9h2sq`%1Ye9&e3arvO7S+G{^C1y!QHR9-h%w3 z1WYW04`zMZlWMAp1s0&27HXHgI~@{73Mb@Wcp#SjKmR24Yv}5!kMxV(V7=}EgZz=>5)qN_F2F?j zZkyzhVP^d$?3g?8M!mCFEN?ye>9HlG9_&P^YN~9B*v^aev`Fgke5_%AhU4jl2^JNQ z-t6+G;IKl9L>BYMAb7Z%t!_)sQs%sOu1pvreM7?#KP%YS;AYj_B;O7WXlMGQ1ikFDHuTRRJh&o~}ai1vpb7L#!&NpR>F= zfohS`J^z+@t|d=+n;1QV5+e&s+Ky=dp(#v(*9R5`+x+Tu1%*;CxH!SCO`<_q6%aA{$VWiJ;;bqEhH^ z{Vgwm_XMv6h%0z-DI)dG*FJ_O3<|n5x zGNi@>TT`;IrrQNSyx%-Z)F%rn&RYg(&;ppwG{9uu8zV|(}oIn@y?#_D(A&ZV6 zFD(s81-O22AFYM8y@`Y!|AdsZ;`Qr|-7Qu2FkWo37Dz20XIBG6B94?jmj~#2oFxS& zMJgMg;hlh{)*O`1bzKK`3h83Y8SB}@7;sq!4ZfJ4*a_c?|HYzRg91>MgC_xb3n2N+ zB9(E7pKv_iwB@$wSY$^T!2XGW)iKz_q00aSLiCed^xw;h29M!)$Ird~&ask#LTpp?;0FYkCbk6lrjy!=5MkQ`HaU4t55M4Ul! zQB(uT0Zk4BQ^k*Mz0uVd+$Vh&D=<~dh-$C^W+q`7yapr^Wt_%egg%hjlVWFXVit1q zoE2skzwgNAhlb^HRi-*H+rdTsIFfCnFMl8Nle?)K{$hl0LUPf-kb_IsyTJLf1kH3C2(IeB57s%+Sq50_1#G=0$X;KRTJTP8iPL zauff_L1;aTntBKhuoWy4o~unN0UR~|Q#R+*q>?z4cALrai! zUZHsWfv6p)S;^<=AUT0A>c2@PGt1Pw|#6izzB>@KndOSxek_jJre&u-Y zPJQ}2n7)QNf`(9r0Y!=@3%gP~PwAxO)XN0SDM`-VNMy}+@YwsuJ7tBuS+@@>+^)rQ zhGv^{aC7%S5=#!7gi7`GKryw)nLT5ND1}@;PM3@OZqALYKSJB?{*c(RB=(#M;WK*e zvWM*P)?Yohrn(9-;_7oqMc8DTi_C`KA~SMStzOo+aY`?GA~rp4mmjK zoPWBba-m~<1J%u2$(avIja84I?07j5b5ZTeC;CpA2n%?!d2U4=_$C?Cy95DTh_zaE zH@(_S+346WmVgTpBadwBA8qe`y<2_hyk?N3;!Y5$oRUvk^{&#|RERt*MEHPFxnVd8 z4@PJra*>$EUo7UpOtZHEGGAplq3KW3qZ-EW*C4V=HE2KI(_-t3dD$yf?V@ldV-|Ykv_6ob#UD`f|$=AY=}d>Yla$l{hub z9=c_PIVnjUu1`6)GvNIAeLrNH0ooK%Bm_w5yEr-fu?JqENmKAlPW;Q#(iq5a!06ch z?LmR*!ZnvU=l(7Q{YbD;5Oc6p%`J(6RG{^{J-zo?;g(u$w_|lNFNyNf)QXQr&AfbE zh2MM%B7{sFZmQ+{LX<}M-~Rl&T>2VYmd`f*Ilvw8j8lG;j9vVsU_6m6l=49fkTQR! z#dT-jfBhlef7$N8FOh}ve5aL+EbNr3zHjVL7@7(u-BvkkG zmROa{5V)xuIXq?6mS5%fwv(GPCE3pH}RS0b3FVv21y7qwLEk-DSuu)|R20ea- zE9N%L9G1_W@o^}y;4x?fQNxgIhW2or@Ay)2$-;=w;-*nlI9`i;xv(n3Y>!hQl-Ezy z-=;L@w=n*hr50}XkzXv*$Z$xbJ#wg=u3tR|gjQ};$8-rNlVT?sKoT&F$K3I4?z^Oz z^_JFPd%=?`tH8zuT6l{V>XzOLXB|Aw@>F^4iGS<}5m>E7Ee*XS-xoxy-m`-_&IhDp zcYRc!bpZ+_&bv4;WmkjZH@UaX@b@?_yD#d<`53OiPS{;l_@X8BX%v5GNp6Ne{(NBk zv6-H)wui?&7m42mJDNe^{yYsBcKy%JkN79BqgvlC`AVc*FO{uQpqP+PO`WbF3p8F? zruJsdyo=yC#Ig_bVqYgmUpL*JQdWS1lhuX_GS1TEQb(GaPq^LjI|?M*Z(1uOq?6}C zry%#}lH^Tj73wO9nF0&lDT!PD?o7MXx_xxZJtURg8%V@RITL!L>+Gd_^tLzw?e6hL zIPGI0if@A9&`e#eeV%^Jp*-pJp=%4rgmn+3sW0!UPPo^*@l48u-t8Z|yt~O_AD@(T zg|puNz`nvJZG$)Zr&RI5kGzntT6&9Ip;8L@GQmEZXsfDZa04c|64cOfyguEDlVt+E zar;S&=jElgh)3$ny$k91+u64-heepgK*MFrhswi>aOUtg$-@t{xJ;6OdH&C)_+S3U zHs7%tO4D7!fc8^F8ep@|Ml<>s>Nm)Xh%0Ti4DZgwm?Pr2Oa`8s_>Sn0&D8B8e}?2NhH}9- zq+mW=gUo6e%Z^X2Dem!&NiWLCax-vQ28yf~Fk&z>p$pKR@n0-D-x2!6d7Tv_GNEC8 zBBz@DvsJjuvecflN!04PNXqqz`vJsZ$XQDAMCvi~lDXpC^RhV|N81(|oq;a=ov2lT zuAHP(6(k&;OXf~lD4|k4w)Fh__ByJ(SQTP`hp8I?ggp0Lq)}|NyY&0-r(t1W!>-=a z&`k0HFK~@9d!I?|#%^2S9&JCMTPB0q1lF@%!Kc?A>9NXAu*70Vs_(QrJyzY80YEle zHpE%+6`(UjpW(;c8Qco!Le#^ux66~wI98EMKi<)W4A9ExmN9vtRM9s^JiaJ>2ZV1MuZzE6^gwnB?lU=*_zym12s-AV;|u_llmp$2e?{{9d5OehI8 z;Ai0X2+|$Q9?uqr284J&Xj<*A0B4``IhfQvH|fXF|NfqLE9LMEVja{;O9{JBaG?5M zXSsy2{GV?7tLJ~vH?BxMi6jtePlhR%&1M~($}FEyU(eb#wkr1SMv zw=PRu2;FTml}8F7RMmT(YhIDObR{WpLY>{)c<755@QgG6ihkKYZKHjM;Bx%Az7`9r ziJZ6dfP`pz@98}oE_5ewu+C?GmdUbI!^2G>lQ9!Kkq}PjyBc|Qc10a)^8%1ZJWz>Z z&Y*!ge7ar(85Hu>oolgZQDyc^Md>@1AG$(^MWUbg`V;$SEG-a+d&So^%iJ{=pP)ta zL%xav7kl8JXL%>OgEYQP5EntKM`+w*j-0S5OsMYskc^D$<*N=hjJ@;cfvhO_yuxU5 zTSS|?H(w+%h#gIMYu_vD#mM4AZ0aoWq3v`|arpl)_TD?Hsc%~wMo|$_v4DUGQBeU= zktR|jA|RkfKnOh|z4smx1?f#dKxt8sBE<;OTj*7KlNv&3QWIJdQv9~(ch0@%oLkQQ z?pw}y$GCqu)ag)9TC5XJs4Sv0ixhozqx;S~mZC;Kry( zw|B)mC$$w3Jd)qXfF_+^`cQ(I-n`n-=e`QFfeOw6g@%mC37Y=4T%-s^)t&jXQ-06| zhJcHOy94i@J?ZPH=r#=eeUEG|?<30~H2`>o1a5zPpiEtIGWy3%)KQVuH)^~2)#1^ZnqUC&}_=foa zc2HL>vH^6Aana|4q^#AuK|x+t51281PC;Y^bqLXm+-!`!wfJ3W(7GI*I` zz4=k5OFSkDqVp4$t(l(XZR`#UBJRpdk4-z=7|v4guw5Odql4_A`zhIBL6I>jIi)`< z0sI24wEOSs?56fa`_}GRH1o>&9}eoLo@R}?=V!E!)?KgHr)LA$A6vVhZh0_{54`Tx zSJSI#VH+hbPMGKaup_nR!Z@F3Y$sbmEv20Z{2dExQ8%XW4kyw`LiWkLb5O@O0AT@zZ|N3yOYd-TO^ zeo@K(riuOf<@kJB#kKmn5sU>mhNOMH`J+=b_-9>ZFLFKGEhTm>l~|)0*6sPxQ47F^ zX2D$=3@b)=OEyiK1X_j+tlsJtFRa(EHStlcv;G!e`=VU0KDrLpL8(%@a(F&v!-J}1 z?If_${<$FDhoz;Sx5`#(^+<&neMZ@Hm88C#{pEa|tCtO}{U`gJ?DyhG;UTy;U`g00ZOEORW8=QhYzO)Wt}yEEvDHx0X{Bip^Z)L4epj65B18yu1 z8I@}ydkHx2AW3jg#(@VlW@j%aW$0wMtwEzI9{-rDs|R%l)S3Fy#HMH>p(X8MUrMWC zk;}2_M6NIodsklSNr^G+k zr02I^b?@~SFm*f=MX?NArQPcC7CAWuO-@`$Xc=BG^0c0Et>7`+ zau6ueF~Ec_PhRQ3d7VVYUeTb2_PbenFAnydz*9Fy={QbFzg3{O%fqezEhfL1XFveN#9 zZBHRoW7p{-4FA}U0s2Q}212xseWeVXNN*2=`_fTx0L#*hk!w3wu{XYz+k8iwv#*O! z3Fq2_@LU0XMYc>|eo?KkYAk4V@GijLmaaoI{#CQ{oMheJQ1dXd+-xK|6vs`MSV<2F8;;u%xfeO7HsI6Vm5UqpPLz!FW&d?Lq>a7s>C^JtLKRbSr+ZmLhHdx28L_8qz+AtWtih+-#BkCF7u@}tA zt+VqfK+f+uXB#mW@QgR>fG%QX4=qElE;Rw-@^MBh(ia0V`}nn%!1Be+LlgCYxDE6% zF%b;kowVQyrbRP;sE5(*ZOgQiB@DANQao0+1hc>}q0HKfISA0>L$sSf-P5^9+yUpx zRFlZ(FA;O)iUhR6wduH@cajQsuZwK2Z12$tlHLa#Z@_;bG4)FO+y$cyKExRE7%jlB zTHBOObBtP?nE0&`?9;}gWZ2Cv`DNg(>nW#Blr9u_L&e3u#)6H9K$?txJj19JW-TI; z=OM%L5gmnbf=~pY*ge1rsMe~&9(@wi&<9YP0(~VEa>Go16oai++-L0Qa_H{>p8-kp zA%`-iF=-?FzPb_Q@eFpdjS%`UIhXo?CP#jX=tG+ZdHFp!r%cDrJu@44g-*lZFtpQq z>6me#YB7WK9%-~;C4{CS`(6GCS6ekTi#oB{NvR}F6^)y zF*~@&-4?>bz^NMGsDki~odnPajO~O6cG`~;`-`cqU`COLF^oE1e|6?sa_&r+^QYb) z5uLpU1Q#u|`gOZr57skNos_5K=qPUNGlJxK3jg*J=I zp=Xi6hbJU|ehSinRIIh}innK^N9n3*((4uZjK#0zI`7@{EC0z0 zdKpo^UrOZ;u*w)yBxhDSGs9H@ij|2pRi~*UbAx}${?diGfd&PoFN?)?ld0D-@`MUO zr+rc%5OPv)p}$N;+S}Z_0ocYb@A4=>vf&1E?33NNUX)zpCzZX~T;6Zj{k+7^S8Bg4 zfAFNxoG7t*UIe#!95D@M+x4$3-1TDgid#6fl=x)?UV>TKOpLxq|4{k-QzA0b_SN-f zfxCB?eOH8^yv?e!2CS{Wu5!e(V^x@!t@aITL!DlEBz`Ke@+JMKUn=u?WEk-4o zZb-ddayibh@;Vu!q7@9&po&MVH&}7HjBpx?f0O;{Gx7FXzP6Y3wmKdlhVcuNoeS>7 z&DkqE3f;X>z}%ql>?@EfceY?(nVd}sYJzd@s13qZ9QcUTD)ENTT?r%u_w=kfgb>_~ zo=SreLKvFiX410i;F?NaRaGvqke3tN3ycMB<7Rf0MmoeRpc)L>0_$|%P0A22c^%T6DSAL^{NMiJ0%N7D7KWpr}TbF*Gn=WiIc7;oQq?GQ|rl9%g-D78=uCMstV_0;Zc7cvU z+KHZ}U8{}Nz`U8UiSi9wH_2K8`v^$SIE<~cH)Fyx&XM0|sulUzEkk(O-?{>g|=ev9En{&cti> z^pGzfdp1DMX$I0Bw&Hq(k9RT>1#cl<C2xQB+_ zojfh`B;<_K?eTG7x{i(Rd6gFz>JhJB-(k{C5kika`X-(Afb;R-zl+ZZWxo8I-|v6e zY+{{j_M`lu4IlszwD@?)jU0;GU#giriS-DeT-gV3YN?0! z1P)o8-zHKlZ#?j|6GiES@+E7xs16yDRuHa;E`WOx61@a}KzuW5WlaK311sDNm-#$ zy$r+D1+Sqvv>}&UPfdMyIV^QLJ@8_ug2;~d9)$k>N$wSX1!6j8|N8u73uNXOlYVrU zqUuTZ%eBTwj`&1Gq_NMCQQ%akLsf&4`S;jDT+0~e9Yqz`vm*1iT8=f;6Q6|UT9_CI z3;aOMi_G(BwlOY`=+pC;OEM;(Auo!g_)zn){((PTv=hBa7Xn@#85}1%v4LFrnDwWN!=^_2TEJ_$*e=7$07qb-vf7`n=G9j0 zWC88pI`;r$Uj^vLA+dj4)&iPV1KwbKBOLKZuR(uY|MPbV{r~Q~#;v8}tZuF_W!g2O zNzY_?qSs_7FPJ#zcZ|{v^nip8+CSVrk*b^M)%I>8z#S;Zu`d{0R_xj<)XwfNz3Ek9 z1O$4aBw9%4*Zj@vHTkOCo0|kjAMX|ygU;GBxL}! zW+D3^dNAV>S-I)_9iWZz0LQVSke)8izDPrH+jO=ugW}$C_cx0Pz5O z25wbSTRVV`&Yy6Y4%Yj#hVJJP@8KwD;p=I6)szUeURR@u#s;H)y9{=Jy9^BclGi~5 zq{gcaQK|lX)J~fY?akPdAWe=s$3j2C%b@4}V&c!2ja_3QUS_jO8l+8Y$X@#~1`bat z*^|V)8y_vAWrHUQ_L9Dm8!i%(gR;bbJkXa6^4^r@{sAwHP(7BhySurjMX=gz;O9tf zNoi4hyoWAL6ui5Yp~A-<6WR}5L`1vJVaA)=Z-T-+>c+%M7I=nO=64UYtD&c`uQ8{0 zP!ascONgrRnUZODo?5I=tffDMZS3pL1^f_dH z>`;<=g<@x;^4)rtW6LEmKm2i=F{yiUkuXJV@Tl{9|E1jbFiyk(vP25o6v;&l4ca)q z?{J0dUdJu+Xm&%Nq9fE?1~-g7Tg1&P9LAR0*IV2PFLw6BfR*Q= zDbWcU7;5+Ex}nof zyW4`avlxavT=^H16SmimfSznNsjMLvPFNHgvK_o5O81DkY+G%oTi6*5J;s`eX#kD) znb1F!vBmO`n~0f1X+f2F&eEH?2TYF50$G8*9#c=jFC{-dr)gXM6d1^Nq&~=T0bdKHI7+|+ucR009tFfk2NEBllBhDkkHlDbAB9{keO95 zOyR;~0o;EMJdQOgC7uiR8PeC#Zx?!6bLX{AqqpEyVh4H{G9H9yh8qwL*DX+yExkWZ z-`%eXUe1>lwIf6DsP#}BEE3FK#?Bhe(pg!#MXQ3798)cz^1gO5Y^yo|hs3ic(wW2F zPsw~?+Ls$f(lb}l*Lb!mm47V?f4)3uAcIczgzJm|q2TtStjk;63!}9?EywYlHt#%j zR{SF<_&sxa+?wK5A2%?cPnVmcQ~2!`-I7U@fqKg~r#k@cAP*r)M$SSweZ3i?B}TE| zED9uJQoWyps?|QAFOMK?kgrgzYdT7g)g3>U_jV-7V0mNk9jY56QP=CA=-+(2qV)v6 z%O7%rv<4-(t+W=I4b{FsilcSL$n0#3Y*pZ6o7pfgu>fQBn@2$tp+GqBUJddUn~Om# zk%aC~NA3YsX3KYGA8f646vE;k(!q7CKx{@dh*RnpQzYUR|FzMgp8SW+U14v$aJrC7 zP?-risOots(M<>U7YpiWcXv|iEGgVc-he>T9NnNKS;8}9Cf4d+?w~udlkRh-WpnJ| z-^d&fwfi)MIF7q7sh!Pe*d+mp(Z>r=LHif?QUL>40WL?iVf66}48xVBTad>m1M=mf zcwZj-(i3^XmV`>1-Qf)3Yt{9Zb=7#16q*U=phq%5Q)KNvW8yk#K4TnUh{e5>a@;>pjER@17iI@C`0A?o0({`2H-IO%mHkp-uRVhJtt*BdUSIyRbL zy*wHJ-S>2{dJyi>A;(`jJ+po>X0q*Z>sx34ds%Z zyqT8$W1?e~jn&qT&=A~apyB#Rf&=Q*Apu5>Dm%Vn1pij3;+2t|&5G|2#Sp{<=dP7$&gzXFbTt5<%$f24Zhe2CL^B8|f;whm zGijwXY+cj4tS2VcJ*U4KVFZF6pZsedgwm$9mX?Iin_~faeIIgXO51R&3#$x z0C2-%?%#0M&zp{&_O%t*(|$6ih7g%SF^*lrh2GC?o;n>%CIfpGmd21I_MjpZMIPh% zj!mweHdg1+Z|0f|8)3h4vQi;0m$mcmu#%n!{;I44BPKCO$gDqVa7+RBSf$LFeH>PC z?E%2kQZ_YGc%B&{{BLp{|25lI?@8ImpBa~i7KWs34()9p7f*|!tjMey4(mjvcq;>c z)bN|@XQ@0eve_+po!QfSk*5Rq@XX!4t<=GX>YZ11<}h+OA^I2E#owAW9>IL(Qi;>J zdptK!ir~F|g351GFBjr`%IBJxb-_u4BSt;cpn?ugmjmlPi1g+(M|-U{usT4HKq2E? z9g8$M=pX>UdQUDhO(k^Cp(E^}f^Q4P=?oYoAdlcuSNH_QV`rNw7V? zUR+W2@@3aUg?j>T9xmIbqX(Ccd49CLGS^q1c1iD?Uk+lQ#!AnX7bN-(-1uZ>St!(e zPxFrY6k@LxR$Yb8S*i$GQHYQ3vZuTrC@WICH+*?`rz%SNRJ?8oyH-r7KpmnV(T?Jm zS0+L=08->b`N{OVs`8s}on|b+o?7=Bo{DVU5}94Ne+I3l?+MAJWNn*X#Gwmv!#>bw zreiDHF94O$32LqXowN-X*juoyN_ga9J4L=U3*+l3DlTY07GlWo_LO58+7({oh7S~T zqklTNzO&%ZMyi67$QP5ljg>rSfe8PeY`5(8TJC5VqKJvutY+K>xqta4YSz#-<2 zMU3&yo6k!sH0>=}FPJkO3wku>s-&|&RE}SQ|uHS0+Fnke^-40J|hv=f& zWab=B!uTCbbV0W;SF4t1-O^OAPDJr4gr`5XjA8mBVMN>v-BBv{C25Y|bMo_M+8_Hq z(buAYgkDis6i|S&(Fh{1ZTT3)zEKQEAI;6M`DQlY{L$Re+_j#utOu>%c*fAA^?8pf zikm#HR!|hZ$R8?)xnDN#mm4k>V7yVE)`55l7<5M>Xx6g$3*EPo>$<5X|+bdfq69_zyVF8T}>@MbhzhyQ%PfM`& zgzH|{O!*XVY7K(uPcxQ#{1@X0rn7S?ZKO!6wu{Hb9WclGv~k>KiA}$=K}791P%$W zR;g`E?I~o$)W6T(M;Q&%bEBpq4Y$RqLJ}THE8%vfP+*LF# zAXpfq~&Os@cg zbL=&EsU0vQz~lj-Q1ogOXd4ee^B|giMP}@-^(`PZhz)X9gcd|K;soqsOVT$$;|uo0 zu~UdZ3XoU^kdOASU%S4XMK|ciET0FFj{$%_W3(07Hy*Gy_cz@E5x&#{jx$9t1yD}g zYW#$LG3mK&?jH$d9K>{h%Zg#!docgIv1F*}DS#;lq5*HjtI`Tz9q?7v3;i6%Mn?@V$#4DD<;Tz0EL5s8 z^c>c9o1NOm$+yb3r$J^XB+P|2zjLOk_IhRb=}xa&svn z|7I6(@;w;S;Pm}eYW;JGCYIaKw|@{wd=8`sT7Xv5W9tu&-y=!ODs~LT!+!!2t>Z2B zBw`$U#4{D5NY$a9eMOT0j2f$qov62^12VvY4l?R@s=Y+m;hC?RI6v!TQfXnfrOw^! z%V*E6THrN2;g*@Wqd>y7H-i8N7pGCrtOt93i;>m&C2B4>pm<0UF?CP0F?Ck%=%1`v zJoal}W3;)3H25_9VPUju=|rH?&&4)6@02I!8mvN)7FV!IwFZLQ=JNe?HArW?Gp7rE zlK;k*C4wirRqVkLaVu_u8|XgR{u0Oix^E0tV4C?&0ohmKyd)LCN;QUaI!DwWu`jxH zbH6@?^`-S8N-Y0IVF3wR2P@So^0N3`^|VADKRdS8O#tfZ0-p6#-Px=Qm1mrtWnH^! zL>~iT!F8Gp?x9(0`T_8kJdnc34I_`VcAhZTC3Cq7zLqzz8~phapES7;um|?!PD|oj z-BE+kixJ06egxz@ns+9E+&j}Tuve&?}TR(5*9m;DS@>C(q{!q+&bRg1pb0PyAc}IOV z_Ha{En{v2?o=n19Pl}hV@{)%RAQGz8uMT+Xm6H~l=XpT2E-FQKy>E46}#icp1`5uA1||u z{)#xl7>Av7k{7nm_Gwz8if;{{YIlgYzR?oUK^)>NSbUcJO)OTVK-4DSG!D=*&^S=& zozxTD2kTgqbXHfEMlljdrve2*O(Fx8W2+UGbcUjoA#ve>C;A)Bo$5l9=y%M9odoZd z^(v~g6k8CdGX*d|VF_5SS{*z)Klv5nl26k3hfuvM&GETm_uCN5Fr|s88-hJOm*!TK z6OsfsKx?D-$l%PgKOO_iETcvQ9f4)f98OiP9)4@$A@*)+muIztPh`n?Tolh91Xr1L z6m0Y&>5t&g1f3k)*aFW?aBW2GqC&qBYFBWu z_?d6M)XkIOV+%Rie2@_k7yJm>9PgvCYN*kTNejOMfoy!;LC4ny0RC)1as|^16Uf&$ zsdsc2%qO;%a=nvRjgSCX?j*`=x35x%ivGmtsS7ikJfogrbm7W9kArn;Ey!}UMo0eE z4F^wTc?S#Y!lU={-eM_Qd&OQ#f~s~K?ei*t&KZlb77br)Q?(H-wuq#Lu6w_zbZTPas`^8eVcDBicsqvHH-9gSiKVUdq2o zkH>h)%gs_9B^vb@9F_bQ>O3qfEft?P0_+`hDn$I>Js+WDmj5OPSO66R4w6IhK;RZT zdGv?zx3{N+ea~)uio&|YZp7qZG{LxN#Usdmz;iADsCexn+ahz&Ah<475^j8+5zKZq z?S|ouQfbKaDfVWrHnzBqeJB>mj&Xv8OzgfDp7UmORx_UzyHQ`c@PhdRC|zDnjx@U? zc%mcu%4jao`Qx^P+wSymK-F3Tw41~wX+Y3}E!J5KWmR@A^O7HSIafQJIRJ!od@K?L z^q=z0W)l|8w?5j)b)$A}mB||>ZN5HX%a8>aJE9N4LuP_$7c+%YS z$pPgNH6?>4iEr+X0*3LS;!&4V!S*|jvfb)KW?zSr0egR*M=;PGn=+K68hCe9Y~UU! zQL)ls=D*5r@d39%UYedFW3p?{r(Nb<0M(>9$vuIBua^Y9sa(WxAa28t5M!He>aLgh zu#G2Bt3t5v%+x8m@1oANOVQ6lrRGpamleea7Y3!MX70{Hjfdj;17)IG2Hy3pqL?(i3n+yHNh6(*9%Ehb2nl1wE*CHjjzz)-ct0 z6{zo>-0P7(ih93arosW8v!2KOcRUJO%7>m_>v$`)_t`e~hD}ZSOXkr8Te~)N2}sp- zw06~mobb_RzLl4<&5oml=E6^#3eK@d?9Y5UuBWz^=MOLXxl%^Bpgd=)n>-`_xld|P zmLVswyfwqw_QGM<7wQ$jQ%L@%(m$L!-t=`n5cx*249|a=+%9wN#Cfpj;G?CXZ#R7& zkt(>3n?kQ-l<8qu=Rn-sH%xHB^kH~eGM1sX>}$=8vzgGXBP~@-Ij+buvhyEW)V%!$ z%H83N0DPA5h0V4KErCLBk`zR?Qp)njP?D2xQ9UTLqxaJsIW6>ZuS=u}b)pUW6**RA z;|r66*1A`l{q|0n4t#`*_j^RCUk}LCq8V83u2+sN%Y$*IjxnXb5vNf@1TRYZy(@y9 z-6Q)uRdid?dVt1Pw0iY??^N|2H@72TaSYm=aOQd&mBY#p=ez4D~l(;MxS1QU(Si%mjI zI?L#h+r9zE>JtBd?Y=)(V*bBel_fL9rU-9q z^*pam9nDH)Sr;HwG$$9C`{%zffa0rU6gh>4ZM1ig@IJt zRC`P;n01OKW)8LuK`W_uOSbQ%7mm+7%ZS2E=g>IGT$wbk^1AZDTYy4TzBc7mLT_pb zSajZm0#f#bIU$r0h@u>EW=N~iq+**u`^S*;JQ(scY`^S1z!+pTqTT4Dx{EF@?hRn_ z6;m?nErE<0#bm&AR)WxifF^je80Xt@?DWInt>ajlQx;wCJ7$?5pwvqO-h3DE_CDB8 z0XYYL>46S6<)|TKpj){7#eMPSp=-}}VTP0;9BM5H&?-U{ZF&hUqg@qJa|b-WQ#}>2@sH9`qW#130fbB#^}FFhadW z`Uop>_Qq#x@{z(-V3d0*kOau%i&Kk3Exea^sB)ksDL>Rp@C#6=>6zgK?GY^106%~5 zN+io6uebY*L_oLw+o=Npg})wcBSrom!-}sc#=-CAsSBT{bz^5@KRMpKx#l=HI&#QW z^)}PGsV&hRXPQiOZW<)Rbdx(nj4i)H)SzqK(*gFq$bU2}h2~9^c<(=k#Tqqx{#y~K zju&-U@xU_ja1zBMGPaa=Q`B{$J66<7al3+4fPHxAc~lTg;_p`6-*x+mvDkZQ@B6^? z9$g|IDlAajybm{_C(#ZNOl7hP%|FVsy0hF_Exm7BIO=@jWt)JX=f*u;$Q5WR?S}mdv}-vuDnbD*#8F$vmo@uV%)@Q zpg-qS9z|u$upVC&|K?ifo?h3Z{r4;XZ+Bj3^ZP7s4?z-m>u|bdr_n_FrWzkXZclv( zampu2^lUU_J)yD2j-sNtRf3|}G-vrihxu<2pdmE(9L2afBPDMK{9#RT5=;IYzgNGl zvcF!fe=8dvk?kd%HeNFppGkbL7fIBMuG5<1zf6A#=S6zn9F6=O5PTsT2;V3Foa)v$ z^ajT6ff~!xFYF+I&1@O?staX{qy0CiwJLkvh* zChz^JGs`Dt*m$zzwyFW-so!UoJib-jO?L+CE-p2E!h{P-y)4!NxrzlvYQpaQgB@+C z9J2jz{XUY+b~(p8S#$39vf{yQ{$Or8CoN`dZ+P&p_l2O7@R$S`oHM!Y#`mCVyk|?C*QzcuCWe~E^74yrXpMbBpc$+1buel z7t@dolAJOWiib8=b+CSpLHZ1BPgIrTf9!+M8Ih5CW7g?i;mWu415lisfy zW=KWl^*BW*KCFB17ZKU_awiTc_T-GJ-)(TTwUtTxz{B4aL2duqLw4Xov6?vd!qb(h zQ}(@ZG2r0mKSeyThxsrWX)ARiME7v297ohV~;cbId#y)|~i9v9nm*|Cg%k9Vd&_|4J=sy3`S>lo$CCgYgy$HE2RkY!aE zHYYRsSNFL^z9bzf9KoLrskgkJx{y`|ewv5wa|VpkMr8z#y|>oHYI*KMY@2$@BhQ`J z{?I7|H!(C6ZvLV=HuTHr#u1*5!~TD&=V5JhLL1dqXs^tpfo>E@m*Rl-V7Y!L>OV;1 z9d+1&oK+m(WMyp${h&h=Y9AE|;9_ctVq9jvzNt!{L8~vjuuw;VXR72e zkr-lZEL?O}L8c6R*0iWlWGcjb_+$j@h8(on^g3X`M9}UII;F6m)W1_2*)}_*0`Q9k zv0Gmv?iwa49(GSY9i6R-PmCR~6WIo70=LJL?R>rL`FR2qkhRuvznI=~?uIt?O*rF1 zTWwwrNO2lv5INqMZ+d%Z4cM*B+BKSLNo2i$nzzmadZ6|2s9B=Sk)*1Vvzh-|wC?~r zQwoKxy*PYryifb)7uyIyO)+vgXmvCwu8c+I>krz`-m_PYA6&(>hxAs0kjoR+dZ=Z6U+^ zVr`m;&1&yS{>b;Uek_fh(+{dkXY`#uS13qKN%^bB;!JoS+`IAN*mSRY-y)>Q!vnfe z#lpPs-VSm?-QC*{Q^yEhs)|8Iv?9;>c^O`0I{8*hCBSy>A9^G?JAJa+6}bs^4;Ytg z&t<~by1rRENsyFwIsKdK1qa}|&#N;}aa|L;dB`??7K4N80dO0&f`Iu17 zw?o=J4Hr_FJUJK3F;4x9~>~XWogAS8Lwoi1nHdcYCmuXY?e{uI#?%k>YhN|os{1#l>SR{eC=~u z!*egs{Rf2b|I<+Yzt=uHP|uO}SIhK&IjPfsp)KYG&;+J+BIMO9YK<BzIR!y`>aA*4bP5V=wE)7{MQ@HVqNy!)~p8%LE}KJL;?~B`b*A3 z_OZ~S`9~F+$GeMB|6a3_{`w?#>sxIBoFD&k`@cNDZa(!tulwf=<{;lKn%i?=4*!?~ z4q-RD%{@nw4QLCDQyOw^*9UD$`ak?VbFB(Js~T5a2civR1D03WVjG4|r1!6$Wp5bL zF=>u}@WNJ&2`YD+cCWBVp}y2_P6J94*xNsqu^5d#m$EPD#eY zN+?AZZ7$GPK6J-a=8( z;H{suX96UFa?~k&R+(+U8r86Gub*gfB_O{YKw`_oFkP}Mz+vM^&1}3?tMH1xS;{_t z8U0cSt%yEosXY@&xr5OuM{zG|O%za*DWDuLP&wk2*ul*wTXjgkc8ZEgK*hQ*VMArw zjRB$4z|kj2C}9}n}U@bwcx-1R~#^j}Pii1`Vi zAQ|VPVqd<9H9Zvc9cq$eUu{}k`Ew{d5TD~uwe<3u+?gjC7b8iZC)qUqlgb=zjcJNV}DD_m-U>7lket@pXaT^^7fHU;LvK5 z>WOh%p}4#TUclg$FcqZo_$1#St!K5_n=p#KGpxBND#M$Tz;_QYR_z~oU3SWPyk?$c zCfpWfkvIvEylEQrL`4p`%8=>A>Z5a{i+u5+Z`Wl&O?+y0wCaH}_6v+w7&n*Z@D=;$ zlf>enudD}K0o_MI-_aI%o<5C^0o%onhu15%XLnlq@3BQPU%I)lIbGWx^Cc)#@7)bc z)VF;4sO7r_lIdtf>6rNn_T*~0;?;qovLYE;w|`<+hW`cr58j;QbpU%m4Wt!H#$oq= z3Nl_DMtq?O5*+(p%CqEjmu@Ro3JYnoNWhmjEPd*yu;3jIE zS&3)olneDFu~Qh9sVxy`dvb;+TLp?6ieCv-9;ACt293U&n=FyFvU&Ik01auPn> znw4?Ck}IDukX~nBOjaT=nsMk?l32DMc?YNU<0X1Y)*BE~o&C^CvCmKSt>%YC;SCGL zflDd18S5EFg>Wvi^Opw4T|T}mI*dv57Lc(x9aSp5c2AN!mxyiRF`Tf6OOJC2R*Pbj zV2|5#%Mc|)SE4PWv&`N!1d&H2#S(>UavUSNf9ghms<&Rktl3f@_YjXvGB`J*Ku&em zTUq=i^BqgU_pU%Hr*+uIx$u#Y=r0{iEuKJq@ z-Y-Y&ANl2e&xU-?&|lfQt7}-KpauZDO@2y$hV2hZUsr*zdJ`(NM|a0JmYlyeEVxYF zajE=S=?5L~czIy3-ad@1!q9D3dGmz7=-O>fICrttm4ro486e<)Yb*-cr~7lHUAUV`-({4v=Dc5ejzI{2GlhU0J7(q(4 zV#ix0i+I}(FWAi3@eQ;$^>BQ}W zo>(38GryL8A^r@*!8+(Pb2Mdj9abV1DDc$pr3jvWcCUe;Sb?{xJ?rdnc(|~OSCn>L z{})sDZCQ3hD(jPl6DtxX+~MkF-(%qNmlGDe&qZvb!;l>CON4F>M`$Zo;57KPqm|$r z@JlhqB^bvFkV-X{tzG!BhzeB$__B@PUv?@>*dg+>;c4>weQ6cAC1Gf9knE4I(^lZ8 zBTK$m_KsXrcYmoqulm77!QQ$JTUHSy&xse7=ksPymA5?L(nqto;36RKaX0J7lfAFk z31u4rwkoxg?an9=s1nOKlk30m}ZZXW3tqGx`H^qpM904P$HJyB(<#~;$#-;Kxv)2&=*K6Lz>X{%A*>NXYdVM)GbQ{RxCdXZn#E~r;%;yfE2ao7SGBI~c zv1B2&kPMZ_kOx~FyUq%epqC-_{bXCBc@vCdCobQ|YWZ<&{X0vM_-Oo5-tGAz1 z0eY9rTORe5nyp05$Ift)C#hOAn|_3tPfELf|J)Y7LbN7HW~%rVDG^z1pT(+PWm0rp zEHOShVVIsu=A7fxX$7;DI=UMq_tsRM*@PcRFDZ#(0+RZepu15)*z(v{aL(Bif=v~B zl*-a=X1+n>g{bCF7T&w)LC7(mh{;yJ6JP4A?lgOF=Pi-dO=tb|q(0xRDRopWv?zQ( zpk=B>6u?WrQam?7nGwRA(x8m$o5^Hp&Oz`^+KekUzE-lp+$LNJ3xb$iNsJ0yq0 z^5&Z*)6>(atY^@r_E3!T=HV4CTl>PFnCGfsWRG_{%h!l{JOLdk3FcWQtY&~Jx$Yzc z>b?}|f8+C<{i(lpy0F@uaR7k-^`?P&JX^tz}#Hl z{qwl?F}j8@OVi-%VatA+o@J1(GMk+I?n*bc;@2}g{n(^Zjn9HX@=}7>5*;#%_$4!< zEF!Nwf!g75krZ(^T=4=~7J|wB3{Y>S$)j)#EBpX>fBGhbb^_q0k0Y z6MU6P3Z%x(yV2OlW{Quh?*eCbPDw01<(nWqz5eVYcAC0Z0+8wnz_8|4Czp%x8Rq9k zvVmqLc!1PZbwczGO4rh!DH-ZI^kf{hZz~M2j8=~z$9ZvSxo$>l;7twWUUmRf#dBz` zunl~|H(+fE#s1u~j+@@zPJ$Ja0xdzy+rBUY5Bqj}mE7hJjD{Pc^YhvceE{66K`n_` z&uy0}L@rQe1e**zFbiE^8Gy?B!sK4DeSmW;Ri*hNWN~S4*HgX8IbQSSI z%QRZ_jJy7(BwhQioy#?on;*t{Q1Q~(5@-uqV$TQ+X0QNIwTwty9X*9{TYtVM({*(T zu;5z9))p@5DyZ5IWY2whUN6*2Ls8On1S0RaJR^)olI<{+ zG?rwwVYANoBiwuW_KtEFjshN}8Gkej*^LCYEz#?%0(DhQqF$reH(9068ne0Et*pNz$riP@toZpc9lH2cbY6S=iw{a4PNUGA{omM zn&+Ae@fB?M0e*~=j9ZMh`pXrgG?~u|y@_(5nYFMZ$C#*vgK^(qd*pl_zW!5ve9xkr zC;VfPPcA>;H$k`dk%eMuZ!srlYA}TjINCohMCAR3X!!Jtq+5&p|N5!iu@i zj%%%HG|AM7W%@w4S6k#VQ|_=!Ia#aj8EEj66h4EvXbAY~MMDwX!=DDF>I@ve%(cyn zb^Pc!g^WIGGr8n;8R%(XJ51%XSU91tB6&^GhoKM=8 z{35J8_P7xsITr(roJ;>cr1YQtee5tCfayR?=A%_GW@G8F33|boh2%K96**Ri2uw@K zPIv)zdHD#@d5$V8EGM$o;s@kT{P}(8-@b49i*Efa-cI8#zLXTIN~WrFC)0kY--x26 z4y`oih+XOtcm1(e)^o29^J*(l-aSclv#h5~J<4L|wV6w5z~e*n{XlA%y@Fi2><1e6 z{8=-LO3t>4pBYOr+N+w6sju>kA`4_IZjGCW@4CscV$VGpsOxUSzZfb2s_8bvdCGm@ z7my*m0@YH+6PJyTGi>gQmOv&=Qc~QckD6dE=5pW(NIM4k^F8bs>7Zo9%g#pP@)Feo zia5wW7;@)?{g+w$wSCTRs$It-X2> zSt4}dBCPg&;-($J1z$-?i9}G{05Z)(X;>y~)O-C%uzo|U5NgpcdKb=^-m`4?M8_-V zl2uTG=(U~>F~{db)P@|<@(r8&r-_==xYNf^Ads8%(LAs?lEgkQNjKd9OZ{O-O>>Fa?k}k7+<2S#>;toM@`As+S5n+Ygtckl!nhE?&F1W#l37t z595eU`vScMuEOO&kh9$$B}QnyU|30MR{|95Us&K43xiSlYrGN_u|}Ud z+>c|jl0##xuSMOD1GG7Nuq|JBnLu+xv@f>Fd{K|W5Cu0hSM8n%Sczm?MWkogJf!Xr z4k-SG@Cnv(10$A0Z&rYT@2)`m10&;WMfS2w9YZuh(qczrf{}2UIzT~$q;PX< zA#>t^0?hgGY&NDDLbEHC7Cq?@8^^FhLM6UyGsp6Jm&&vB^HT8PtChMh`)Y9Wmyi1IBiNZjbdqRhzi*r50c8~H z3nc5#18YTzAFk0)!IO-4Z53K~SyVaYei&qaxr3kOOV=32B%h_(NIAYHebOfBJ#v#=A5-n~TYi|8R}O@8j`}V&3ntmF(Ny?{J4c^T}QPU2V>Bhr;D| zv^dJ4@9q8sWzwN#j?mcEcr&H%#Rq$+W%(=TPJ{l)IMgF;^z|eZa*_hc&q)K#(-&#t zU3j4^^Ox4`-@nq+cZSJ0yv^LM>+-839PD2uS!UYYRG+njZk$~*sd0!`%A5=O6!l)e zh7{X?Wbl>roDw3Q%rl;FlVXoU&`C_5IEortP~xMOqt7opOBWU%j4dJ}(Iy=$)9upY znOUq{)>)4wW8+j_QIx44Xl!=!q8s>e$olwCoDg{-_Huxy4Zs1|iQTs%iUaM>xg)c% z!jCe&^sGsI6^&#+C!B)*Fs~dHT~vNn&UNeLvS>;BX_gNzQHh6UmtB9Nf@tSV&8B>< z`^DYplRxl2pEx#reYYq~3(2o7m;d;@(>%lteFN}lU|V^vPHPYqU=~`(bu2@sZeTa1 z4z!?=)8!rwJ!dFaiMmm-`1hks4Sr4H-l7)2d23wMqWs0GYnF*2Ny*)a=?{bo0UX=`_5KJp-=9hA zb=#-_`l8KT^kFcCb~3H&=_fu9t;tjfS&fFJo$2*u_5(zOPqgfFhx5G0CisyjS@PTb z{bT#9fc0l9(k--B?HlQ_SI3yfLeO)Dj`G(o-M88f3M3+t70^&60g}T~Bbuwv_-T37 z_@eqEqpZE<%ch5!egjkAf6={orP;#CxatJ>QnCww2&^Y1B5w0o%)oHKX1Kun?SFHh z`p@>S|2Loik1DhCUq@sC_i@VC7Y6QM#R0d+zlpv8Q=~FAXBK>xV%S?A?h_x}Y888@ zRmlL)yOfI>bd{M@XRj$qbDQz&x@2DSt(6 zw&&?0{~Gj|=NuNoiUGo0>_(nGf0Xzr@?L*)eR+gVM{?o)!yRy;O<;YO##8TWmY-ki ze35;cycj$1U>!jcgzSe&1h;^sTvDr`s1r8t1Eq}w?;~ddDjSBhTjV5eCErY6nb8rbE+#YJ zB@Wp|A#0ye?V2PZP?Lt{1ER_oqN8d=$)>p^$Kl)W&2?&P>mqx0jnHjz12|4nhvG6* z$Bmuht|4^ST!uR@r^zYzvvy04NjsNBy5@+r}h1Wf$^hs z@Y7ey4tY6wxUHP`@Ocx!E#-2An@iXbA>4gLlE(LhR9k-B5zp{xauQ;S7rXXG>g>%q zr_1{5j_)CV|F)U=o$6=vamn3d{~7|Bgghp>w9KPIH$&Ir z06MX^_Y@)8U&8ih+0Pd^ zi{uX|z-Ki?5v7wa6=ia1uc|6kw1YXaaXax0(#`m+u8?oDlw{RaX@aY5zD}537xpqF-74w$)LcH6gUs$3>$=+=OQ{&NcIR#Nj*ICTN_SzTSH}`d~7-ol;l=@(Z-Xx0T4Hg9WJA&hk(4G-!9xJZ~I9AUns)qG4nHhUy$;t+pQ~M2c#j z^+8B8+%H#MWy3sfuXSmB+SjV(o8nI~N9;}}csW0n^JiV4o)`Nv z8*cglpHK3HZ0$}sK#$f$3+{UR*#QCV8Gu;3eU5;m*1_-C%|Ze&%nIg&2hU51S>+iT z*-*Cj;AhvEgP+zIM6m5H*4@!wOpj6cF*IK^K0*u1OHFfO&H996sTfq2g9Q10z&noH z&Fojc5g0;YN-eOHqqkFbCLSt^ENPCh@8brf?XNqYp~xsXRn_}M$F-ihd3Y^6Wm_J1 z`R$Iw@Lhoq0s23Phr)9h#Pvs@q{`>SR>)8b$y z1U0r2d5WX@dl%0|lbfz~aJ#-xIWHeZb;QsabTe2$+ACVRT!e~(4LGFP1qUf-NcO+z z1o;66uIX!_dH_v+E{vzb z{jZ~>g!{=VXj8((l@l)GCtUff3MHQ}q!^~MDfGoagfKIyQ-b9KzFGyI&hFzf8T0GB z=;xi-$~_cr6<}&ZgB3i;w|q5C^`@>z@;Sf!_=c}Zh2ANlRdz0kdYWLhfhVkK4_qL9 zgj-ErIm8{rSj-<76!)G<(G?Wz;c-w9uw-phQ~-$bTw&-Y}GZVC`kbmx{otW(VNhx1{wIqs3@LZgEGP! zh4V=2t;1vP-(@BpPbLILwsW{%bCoV)3lFnTe6=QQsA_(6n?PbpHx^#jW?kS)SYII^ z_3C1uq5Iy)b_8@Sw%_uP@?RTPPy|v0?4SoH?e)KO4)pF_)}8>Ic{EPM%o3FjOQivr zFT5zQ3343J<5kr`81|N>IeM_UV-oqzub@<3$+L{r1%&aac?gbTix2Pq4ukh*(RvBA z;){5OFLi>A9=+-R(-i*h9s*{gu$$6UbE%>w&cAt zFO`tBDh>r1=?Bt>AcD*6bU6HV569%lYQ>Y6(VtB28Q$a5t)MhO$%nHwFY2 zL9Fx>*L$YJT8ZcLt*tv=$+-OM4>K zCls6In@0YFHXKXQ_#TVjY7%h~)km&fX9^fU^GL~%|GCo2Wv-=QE{)NQ^z7@5W=kDN zrak3h7{qW@$XOu94EFImEsngCYkM|pc^is(WPS>h>gd1kp$FzDOEw-FmAaiA!ZT|= zhWUzQu_??ay;{<|A0N+$XgXEdx*7F`-nRt8NYR`{a?tvK?HN+izw*iKhe5`%w*hw= z9Ns2&ckV#5rWmQAw2R9!(P~A80F80VKUUZ)V}6PBFw^x8svUlk6rx805#1H_wJWaA zX+$5SU2zh9!*J#1^4`%>7Hkx-akp+~)zu8qx|HO?+rNBwLDcdWN-C_jb93UoxE->W zP%D6HCz^+3)NXjYPWa_WWSfTrmfU=VoAn4-44B&wFO3-qj_9~@;+*R{+{iTG;O687CENf4CWOxZU9oi_h~)8hnaV%%O_3X1rJ36LNLLqm!?fF6e7+4tGlY1ybh$!V9y|WSf$32@qXyPxrOAyDx zyP98id*}?P<^I^z%h{4;)KR{pu*3eP{TUMecB`FNDo2RUcABCeC2uwA{G0iBnIw0J z!!qip<|%l?Jzw3I->{&F-2t~_0!+yfV{?E%T)o)D}F25vfwtV+Xp%Rz#wEc-17ne zl&kg6E_!uLDN+?MYew3c3qG8K`5c?D$i8ReKVJt4@N?HoOTV`MTMWVg%{Mr%+PCR; z&2;0;{JA*=+U;ZI$}k^5^Q>6f@dFT!7pvuV$@ELQ8~o96Cvfi7T~}t*fuG&(HRoGK zI@z=9ohw7y`k??CPW%cKcJUxTf%VIW)RPOPraCmqR^yZj&8-{tlTtrtTp3;8W za7>Eh&L8FrD$C}MTi!z|pHhB2%hJ1{&||)tC-RMmkKeO-}JecGr52mU8a87Ga*IM)eoA;S16VzwL zuq~it2BeC`>Ggh7$94y%$Py{HWAK_4v&6jBVR^BLq^JzRxDy5wBZ8cFHlEK86hBtW zPr9v7k4Jn!pCbkh7<$4eCY@h&_>1I`v-G_7Yc+RbZJ3UFiD@G$vm&q;)iZAENy86Q51 zpbm1(<-E=4ucMTPtl|!`GBrbIdF2{sP)~U&WU+*Nr5NJY##uWDUTh=Q$93li#|lLh zsij5hL0=uJRn1-X@Ge{Q(oE>4Tc7(%-;cie>Z#fq)R5K*Chp|I$0753yNSX2s-VE6 z?DDNa3Ser?Dwy4sNnh(wuVd>_@k<6sPfH=2_@82M?XFa*@ z$KHcAzYXjsw)*`rZUQ2tR=}bf=Ks3lyWl-W^TE^twN1Os^su-ki4PNi7piF{?7o@# z93&8Vo^}QL;2Wv*F*}ZJ{*$BB!-gyLjW@xKA_L#PMNF6uNT=`%*ob@_A8)cbV_aWj ze*c}~NeyLI0aiy*d+F0$W7~Y9jtwXDyJ6R5pJ%NSsZdpo{gyBj!U1rKcczZPf*!c z&kkRq5kBmC3F^6W$h!{;a~gEdg_~-ImP1Z>hChB|I(`Xi+76E{vfL0SBCpXiL@a-RjQL4=6mXtN zzz>+}SNe)R0YZwMnx(QkwmTZCGn+Nm>PFylpVp3_s8)s!#_)ptxU5})U90vhKVc1Kp1Z4u?XfEa_&23b1w_5 z0{m1mOFVuJevy%a_ZbLL+ zNJ$C50W#ugdM%d5h6%zfVQBJzV2uWt zP)-p8XuZGah&Z4Pn_qM&P6+kn1E4_qeX4&XkPi8r90i{P)nm?3ETdW!9^Kjp-Rc0K z{PutYY)#@o*=l*ZG@+?WwmyG`J73Z*NBphqqfE=rFCLp~Tc#ltG5A^d-#Y}s+mVcE zfH67;Q0AJ4iWJ)I9SfU)mQ>)YQ{o5v5}Qh@IQR?5@rf>sat-<7MVM&}B*#6e8gUBN z1V2Yn^kSkFLCUg1V;Hu~_jt0D%6I=+1H*{s%Xz~`6TtPqt}=DQy8RPsLeJ2_r+v?gRSWCd#a9>p~{ z^yb>&*L|Bw-jl{E4jiG|ZLU^m62ot{453}SV9ely1mB5*&h2+n9N@SwArDrdacJ4W zuSDDipOf^5otLSP8ZUtEUZFdJyVoJ-R5!ZvJh6{rn_!FsU16F{PcN&bu)f@`eZX^i z%@AI^^lsxehPiES>oQ2<=aq5cw#hmRRr32eP99~;sIKoL96oQ{bk8mqtzUW@+oJKj z>3;k8D7b`8QdH(`bL~lGnyzhav#%cEI|tw{{&P?Pv-m8uae4wx!*vKQi?jLbe-t;N4QY|y`*W5z->-uEHFiPoZ1S4zRQPc}Z? zim-CDbmO_Cul3hv5hNoT8$fphX5ktzv81);3exTtm;eLuNtFl3&=>XtFqb-jGAV@O zY9_Ya5{b>jr*Noz-3~rsJ(tjfFJIa(jQd)Jp$qy{7RY%nXFr#{^iDeATDZ)5#Ivd*yv187fzmUm@2%|4OkcWJo<)7MONO}>2e10 zz^T~OOQ1MYRpl06z-l;8C48@$lGdC61j*XrON+Qs$X^#3z;ymglltHbQkezCzgtN- zOza4j)R^pryW9ZF)sU$NhH@#Whrm#DFy5DG9uk*5n5|16De}`E`p^A zGE!7jaflQ_yWap-<@hiLz@B>l4~_iwy3`RRpSJd@;4_bMn(k!(@C&a@V`zi~C0R8( zo>RuxNoh*6$h~1sDV6=>ccHK$QV3z75y|2jm_f?!HECAb>g#U9`SI^8`p>8HF|En7igFt|;c3+j8tK4?L~Laa7GB}f z%y~$yWslJj<}a)2KW;v%z*E`$a~ynJ03J)hGNx+XSN7|;r=4;|-BKVTwn-s7P1h&*mXnz%Dt zQ9&Td>sJIJE}ZlyVGqWnX)DsK(-FYE(>w+YPV4EX`vi$=>*X-gI^&ip{JEczoKhWc zH+IvF^{|PoNE@@8?b4HAB<({=C)JxhPAoYc8=>TK(0}Ez+jTNDeUST;|LTvt=T7cc z8aIlQYE`w>)2WN}L{t>u9|iL4o{;Y3r`s_5)fmRg-D^>(eZwEiE!BbR+jLdY1YbVy8`%eTHb5x}hnd8%x!Bztmap;lL)X^)91M{93fMOg|}$cCh+A%y8{>JZ=82fOQ;$4s9j_Tgk_d!Rs?r>z2jTE zZ6VE!C}#8NVF6Fb>()+e4LNtWe`B9W5TuNXE0r$kIE6*px3z z=dlY>H&4;g7rJONUMnh>e;E<(i+ZrDhjt|4$@124Zoy)tZY^2FU~7E)mKAJb_jm2V z%LPSggY(SMP{1;(@Mwke`rfSTdqD^~7#ZH`-Z}qs`tAu&^6K=I^p}%@BC0M%+hO6( zM&MVz;v|okgr-J!^reBUowSO{;0Df#bt)s^qQWhJHUNu8;Q0U*usqW7sDAKhMyt_Q z6ph_xijT;O9di^{=-MjRzia7sSqPX7`8z)Q`vBi2*M>#}!eDQ4Dx47OO_$lReIsxV zv~-CQQ^-vnh3wktivdJA=U%k*q`{;u4)fYhj0{t8L|@{K!b}$B?E3=0p_>7(5u`=h z9@_19lltbSgSq0v?ePEtE6$q!vIyIh?QUQO`x%p4OWs?o8eJc{tb4c3+Ao2C!EYd% zBpAm?+JO5f80XG?(&fkK(Z%Vhy&}kdq#Q}D@Wotc4WK=uQvi8~mL&>C(3r+NlMRwL zxFYg}#me;sfQzAHxW58TfLv6%MeF6tLT6;Iv5ULnEYG8L=-<^x{b5z8D=+nFa;jOk zkfXuo$1?Q$!xA;3;4>-7JHms%Wli$;HLo0?7IAp12-+3W1H&Y8{qxUTVD6s={cj|JB!$Ap3zFS6lT{TPhKIh8BBT+Xq&#sndHYY|39 z6k8yer$+Q?g0qV4f_ZEq)@5O;hUSuuwNe*T(pe^4Q6pq8Jp9o@vfe3 zV4d}D#pJD3+WX{oTE3$a}0TC>z?99yI8-+fpSM*y_C@axZ|osXJ`wJltH-2?IDcTH5aWV|$|p4Su(j>mY>Y{s$Prs~c)sDuf%GZh^I{Z&V?2s9MMbjQ3^-rHDGyTljBqM2{J|3 zB3tjX&b>aAe!~j2()Ct~*xpenDq_nKP+lg-11`U9&D*~ne_@aK65W6L`xm)r5gbGI z?pJJ%$Wj1iyuOSwpNL+obfxw9!}gXTf!9Ow8+Aa5e0J1 z#fk>a$e(`E87(-7iWed+ia*+IZRdH1Who_FI%mBOy$K1~MulKW%pvhrZyuQCLfWyt zFKhq!h6&ezsG84cAXYT^dF(cPrbB3KXol|64qoyEoZlp_r(xsbA4X%E*QlJBWc}8; z+KXMnYbsjE?&&1OvCDpe39U&=I_c=JT9SAtq;Cs`1eT@s{PS}&!?_n1Wp^}tLDVF) zISW46OWpS_ED|!fC+y^K9J;-;qo5MDJB|uU6%3sfu?KQo1i)lzb2S3F+8m_uZ=+b=LTPffE<;2xJNLWBl7qZ;S^ z`CP}%MtF@*iXt~Ox+$z4ZwC|+U{1(;vLhylhX<5d=@$<)O1T~aJ(jY==;#31(DYwf zeg1kOi>>-CuLOEb+cmqlc4V4*Gt`V?)VvHYE$1U&{s6=%bo_#s$dQNog1>Ktyke%$ zr+@g5)}A-Ay=hGK$J4MMf9W4sKbKPQKmLSj@cGqBqt_X%ouGgaL^HD?^v~O-yG%Px z;LXW5%RFFQ7^jcHG8Av5?4hhewreeYE$_U1akyErvNZ2?m?|D5FXfh0jwrAf*xBfZ zZk;K95+(hCmHq3bBic@Y?F}-rd%JuWSTMhR{I6-4|DVc;(*Jjr%0r2g3JD2-LxHVU z=}0j2<+*#*iNrT)69t}nCQH^^W|^I=>b&n>*FtGr3_yAYrub>m$zi9;_JpEaxYrP) zDq%C?RR@_eL^~pMx|qEgAXz*;Fk{!wHxlmooPFoaE|y7HvCGRX1fithD|x)1JMm4t zcz%hn;!&KV5iPL6@RIDaCh3H4)<#eB*<=0Q#0YOhz;CP>;L75<)n4ck+##101jrt~ zk9#x2BrZ0)38^@P5zkgX%!tw+FUHAawvo#6Ziimh_6W?4{EnmPq??ld=?q(;~ zt?lK{V49PALg6P174&ax)y4}4MlnoA_qx&nsYw)XYU+?Qc#K=scrM7x!Q_0%;vA>uID+WXplZ$^4I=wpED0aSu>MKZ0(9O->=_}+AU z=6r3crl4Rh>Z6hzY}Aa4Wc5uG-*sx5tgjPtZ19%q_wRa_KaX4v-rIlzd4r96)I8WV z<3rk|3TuG%f?2LZ}zvxa`Rt&%tR&tbMU4^fWpFh2J)+!MC2c{lAhM2wWfEJF%m%mV#v9H1^G7s)) zLFIh;`Z9qoW5p!Y;wV`E`IE&Fi)>2Sm1!C(Qu;rFuV#Eo-!MPoFi)0(uON3(^}w!q zD8~q217snV!>BtD;yirq_van|{&OtDek2iNHAtPVWw-w-HIkDk7kbjfKL%E-zhDK- zsZo|$-RYKa=tWsCXet0>@Tw{pr=k8~+`{@@48SRx{YAHd z%Od$`#oi|{CG4lPJ1vDs&fqmCJQ2C9kqnAD%td?;RAq~gxE(DmEzjIPXTx=+o`ZD; zTsD;OehbF3x`yIt^yeL3;y=ixPWg$<d)z~%cSbJESq<`$Ab^g7*XSh)08j(LT`c=g@MZn4LnKV zKkOywzZds)o9lox84(Vg+YW_9eWIF%F=-%xr*sz3LF{PyXr08G*Ls|??%eW0025lF zMBIW#B1)CMX#Vwm0 z_3X<42vi{B7u~oc6)fP(O!k4d>6HDV^8|)imE9J6WSsUW9sv-~FuT5S1l7Nwq^%>5r1NJWB%u&Wu&>;2UVJ+jdFL-6mjwD-SCBwgdbWczKospK zngmT0#=INGXbos?4*c#15O^%JtgEJ6AO8IC>#Vpt;J_+;JP$X=99}h&rB(oldt2Xz z41{{y1cuSk@a4gREQ)4gCOl{xA_IBhITt&tlFsvhscq1TX4nNS$v+`<`k`)*37rVB za{mNy!Lfk5y%u>SZc9xll*Djyfbkhb$o>S$m&Qa&N);;&>8KbokoxFC%f9pJKrnX~ z1=J(NVtXlnkf2#>XU4OdLMYeVctQ^BqNqhME_h=j$=K7Le-i%cQln2~f2(mWefk0F zBW4DPxscRkI7p#~9m=XmI)FGvu16n%7H#JbSG+RrkmCybkl$cpxD5qK{92d8VMUWe zN~(A9m^}RiZWi1CIzR2}OxjPXC`z?2j*eK0l4-=tvnoDwMfD=beX#2I>7bqEXX+J8 zI<0+x2vODtg!u@LkxWlfjD18X+=|$`IyRQwBZNG%*v{}c*uC_PTsBb?RkDH_HgA(thM+h8BdMAGV|CN@U^;1gJp&Zw52ZHyo{lDm zRr&mRcT{n$BHj<&Bx7flRnM=rsp$+v3!sohCxIJE8xr17pmvQSp2p8=`x)sB4nK5N z5UhvLxps;1*AcIfAI^PDDVnMq`l*<|a@Nk+=Z@Z4i__t~XNr}5c_n!CYJcS2MV4C{ zigTX?S%m~)WLF!jG)V?#U`I9t!!HB%l^ivnoYy(_DZY6A(lz%Ag%SQ>+s?(yj~Wr!k6(zwHoe zFqxsnk#J;9Si5|JKk79R8aR7vc{`>097u3eXklJ-{B!Dr{_NhF3b5nx*OPN}O~HdtlXHP6fz!0XBY|M8 zxR?6*M3#LEYOWhN*n{Zl(cGU{CihgziCnwM%#NJs6^8hxb#E+#iE7TL&yX;j`z=_y7J}l z?}v%((Vzo{1=T-~&-+xbu3zp9{~7Z8)IX28))~?6=U*$&KcF@oYW;cCDbxB{!*J{B zHXm(<^6k%Ko{`YJX<%Zs7vOY%3)A`Ai>}xuXYq#4pS7XzD##6mKacTYa_s+6KU}0F zu7Wo~?T9~*S?N*ABXa*!x~Ai=3tOgNQCF5`?bY}WR2U8YJ9+EB^Yy=}D%x?m*v=0F z#}cU^rcUDrs-v~bR`nR(}p zc)Vg5ckK0@0FIxXcRpa<$>Ttl#u|{47zTd>jzCKe7(ecR zI7cpq(I*XHb!r!@C>? zqblQHf0%~cjeFo5^m*@<97>}fFjlw1^^BYiOpFv;8Y8@rHDn`0J{x8?A?@U>zO!dhmsmLq zx^G6bN~gf6x#gEChAn*TSzQvYf@33gd5tfZx$PW8)iTXW_#;_TxgXkaYu+&bcx=Zp zl%*9)9wh93a^zkWRn`}IJjQ>-`5%Kess}X{&2NWcQR2a(?0P>t@Nm4CBbg4Ape%Xv zw_ykOhq?Dd&euxvR=dXBu z&(DX8(oYTbW%=saG%MXgLkF0@(n*(mS<@M1vdf=+5?1GV&;SJ0FDTG^;6*A%dDflg z=M*v+cusFB+7SQ|RI3)`Xa^STP!hSZ;Qt{?ml-{y&9Knn-!zwWvzzcqd(fb4a89NO zv3!Qx*v_;0S!-T??=drGvYKa)8X=^y&S>`NS;?JRr+4A5d~0937#btyI70P z6g%DFkRYSspz9gEN*938d*+@t%5xuISXOpUtG&@>JNiKDBXyo9k3Y!Cci?qQi?FzX zNa-hPND8F&^k+DW9Qa_>i2CW@aGeN&rJ*_}Qt2c`MZLz@O1(zr3aZ_i^YrS2^O4Ys zDITIs{v14jND9$;-{R$UIeFkpbj9Gc^INb}&zDusz}7;v3FCOd$bowBr8mB6 zlFfSC?Hq7)1UdxQk4cvR*^N1d?k4*gtAEyTlwU9abVtvFv;Yx{CUBs9Wiy-`asEl9 zwF8$%k^U8XJ%{9kq6s&Y(EvpQ@K@|xqWzBFzd91QeVxx2QR~(G-kDZjbUh|n&-?Ss z{5yhfY2>KRyt9(^tfWb@7v)JA5gL9Qa3`!dGwQ2Ko#OwI_{wPzq^WJ7bc0%gmhjBi zu;+i`zn&QzP_J@0&r1Y7?k6213y|_>*$TnoweA{)Q5z#KiY&<;l#9X~SKJG7we}rT zUzg{Te=lJ;GB*kEb~FHPG1uLptJ@9^9tkFa{WhoPFUoTrs{aKlVsr}%W(GFQODjx^ zn{S?knE>bA+bvYp8l)(^`D*hAHABG1iAjaaBj4TbL(sSFj8(`#LyRa4#++;%4hcAV z$gcHy9(wo6-lB~N9y*?{JS0IAGU~8#=)YqAlM^r;opgI?BK*vhOR!j24|4l)nd5;4 zfcpxe$AB(8gv{T<_W+Wb&v472dLe;Il5s|h)cT7O3-HnM5CKlP1X3Gx<1uf_Za?)& zVod5@w-n#ECd^b409XA?$vgmXyx51?sH7ocYBa}gvk(%Du$j;!X^|Lew3ve^?d&qa zBrOt~{1LuZ7PkSEki0!5PH<2AD#+xVk06~rwa1%U9)aAFShPP$E~`bfO^>jC@_jsb z?9SZ;0IPICOXV1{;EM6ykXDGll0-aNqtb`_+sHiPhDkssXe$Z<%>e5N14n3lT8l? zEj`%aML9oV8;)T~u>i|}?5SsQn_V!=-_Hllg>}mqA<;VD*ogt052czh4qQH*&+w(- zmNPng1u>hM9!47) zPl;)Dye>DxNYN!;Aypini?%W7;HABB{<^8CD6(I6j*h|BGso_} ztKjM6I@LiVq4cWm-oLu3fhaAVP(j@T!vWW-2;9KwUbn5IGk_QD60FyZt3UjH2BNZU zZAW!wHMp8E!Zq|Ix23eR^Hz#Do5Iklz#$de`=mWX^xhFJg_JX!5TqblP?>PAv+jb3 z%RtrhySNsb7-@7Cc~K(nT=v*0a^Jhc6-2VNVm{WgD(Us>xNx?$q@xrNV`ZV2Z^9um z*@{vtZr@y+0`kmPEmDLdu(wxQ`!)f3 zL4a@F3^IuKD$%Fld!u=UqJ@Cx+Q@gs{0#^_+ge`Xu;w|V0im_>RdXSlOVf+nZaDtV2wiA zP?1i8V88huZHnZvF?1GU77;Li54dj5C93gn%I?O{tO@!{U6yAnbv|Lqe7;l()JbsJ z_YAwJ3D_yo%eVDYCV`YP))=n@`P3Fb1@<|3zC6f}xP#aGVtY`T)EmV?{85_-;LlzG zAHz4pQpoZ~hz<7`aR^%8w_R1Mp`fT6f)e1Yi|t#hK$vQpr=wjes0A>3({{c};`Jp( zbk|$R{ugh62SIVcBQ6_pZ=IW@7M)xj=eZkAx#w}J?#!s!^%TXd$Ie=t1CRR4;^fhd z8D$$jiaa&rE`{k~BJaQcb^88GL@6UlCqO1%J4h?WZYjBEtY$_zsA+7{GuGnUTh-AV zfJDQGdkSXIE2Mt9TsUen6~&gNPQFvQT;#{9?27|ZOx+0dwBVOO@iQEl;!q8g2JVG2 z)*4zw_=>kW9cPSlIdX@o<63jvZ}^1)OpP43m-Ir*`iHo@;)>8OhMXsN zs04DxESk+1-F$Y!_tkuNHP12j8~ZmPULbW?79M*<|GcI4>HWuLJNAo}ZJ`ex-Q611 z9DJ6sU3-C&dxdlfkxT-MKf6gOvXmg8hlpz# zA>a21C&TknwWj{<4S|dc_-Lo+I6RZCkXQ@)Y$j3S_ zeB(dF^@w~p{QrT^LBDYS+1$U=)WErF5TgU~SryMh9L>x&>gNUi8>-S~2HB#q_g_`p ztBJASi&8Ys-(as+?hu;>5N@O3)`4;95sScNfxhRNHiRoi*qINAi^yJp06_z&I)q{} zEX2Yp^lQkbzKMy#PC8WJ($07XT|!f3K+TfCE(ZjT+yO&-4%u{rdJ||f0x*stVBp@- zrx`VqZUGr&J)k(20P_H(4R+7Y_K&Wx_?)>K?mz$Jo1VrD^aW7|zBVB)=xW&oiD4&H z{IgN@+{&*!!%Yg$T3L`H>lWzt<5l`WaM_HvUY+Mc_>?MkdW*d38Lk~WitzvQgtY%W zwiBlNc<1TY3$PPWnF9oCT{>(N=Cv-I)G5!=0#F9C5eV? z6HL@w}2S+vB%H=4;Ki(csi{`(k}Et{MAJ76>6i*(ZL}gM}s~BL6y| zj0yfmhIQ0^ka~E_5LNucS{9#4pS{?RRA9XrAUFEQe_Spn^K1a72Kl$;9r{vI zHc&J?+#_{Pr%xLzD?CzE;kq5?Q{0zKJpe+~mt>#-2mvaptPuJo6!|GQ6t?sG>AeN6 ziPY_kq;C1f|% zDOrh3BB^E0fLI+XbOQoyW4mg3V#fGNELq%I&OH_x6lD!6Lj(Y-dw)DF|88^sb(x&h zTV2(E*=Z2<*vUz`@fcC`{SVxjfKLw{uYKkGd^LB%15yNM{3jE9@GmQdkpFeF2iDq5 z>oYqH&_4F+-<~Iahjb;cv7b3Mct7(83v>zd=cDrHvjou4Um?PL6aRHMM6l5Vbr$TJ z4YV!y7*_31JN(~m3E1i<;VSU|I7N_~F=wU9WtmkOTD11?w|#WDD~T7lsep0$55Gv- zNn>LkBL*gGgYQSK?_6~x6B%$zZU1aJ+6M){U*I2((Su7`av$IDzj&PX&#eOZ_j>-# zmJE`utxm1E(>9@=9ew;U@bI~FejI|K|GaPhYFq#@-hVMJ;MHQ`DoPXmk_(Lb1rXgq zXPlC`(xeo$>9n1>w@1dXW~ic?(m1SzPDYd&q}Lh6Y(w|C@lz4l({<8LYU1Apy~bLf z+K=@;<~yV9Ce}?oIjw(ooSXnT>x&Yw{yuJ8!IzDEO&BY)(-0`)PAzgG^G6FuUULO) z_^qSw)g^2q%INJ_7VL~ARlL;c+nyumOe8(W_RBN+B_(;sgM6%V2AMQB=9|FfTKH%c3W`7Jax7gk--)cULCzaCBn&hS5eoubI0yZTC!jj@({QQ{f zqCw6_m8h~~`q=?K!HA>26_$2R?-89KpALD9^IU&jA|6%m%0h;>xg+UmZs?XBVT(Oh zZU1elBsza|DlRaF=%;_ghVUGoTCyb$In^5pm&TKt)2}4i42VaD`^Or);|5W84k{Xu zr9E+<@-2Md&~YMk-s?6PZ)A4hMJIb_yVJp=*TQS1Oq}KjR2?f!HOLzQ@jFP@o!JAO zau_BpELBX+r*ywCYGZ2kn&`t5(~V(lHeWoV=|2hNeU3s#6br zmsuqb>}8OrIA43U;y;C`$)1b};{p+VMVj*8dYss@_%?1UV(2aWMX|&;*WlBtzU}XH z3ZcXDIo3lLZkuNI! zQf8j?_PUFN_o`ByaCv#crrj&{@l7`dOrPr!Uzjc}jegz|ev=@|SlCmr(CQgan~B`c z9$c{CkB^{;QJEdG^})UgD(W11Rbb%Ce2D6(;F`}tRH`74|ZKGjuSG6vFPp2@DNvrIKOfE3O=-Z%!RLa3XS;~yH z?Y8}rDicPQ%qmw4pOcCdDX1*t!a!{iCt7NBK$CL?)i+OIZ$N zg<}Z?&03Q^;%AxmKlS`#c*br$T01-I6)ykn{fv=qGTE>txp@*ONRXcSxLyF&AKF-f ztqO;zT-Ft>H>iJLrY=^u9e_=NgeV+NC5MDL*(+dQr7abA?n66~F}2ftS2iVH&*(h9 zVSu&>f?=^ta6ZmgHK(kzHk$LfDv9IkO_fA|1ZpIrerhaTzQ1r@@5fEs>!5r^zBYGQ z*_jVf>`a;rQ2p@pA(ygWqV)af=)5&M-<2n((PFac{qU0QaQfQ=k&jrgTGrnyB07EW zvUQ0(PF=%8_P&vQCy$abxw2o*mDGq>FQF1dNIn3lZAz57PphFa@?)kuksK}jb_4Q2 ziWCP8Qw6BKV6X<$kk75Twb?oQV70lNCz5FRA=0pp%8L8=e_SY{vgjB^C^ zi-AE~prRK3i=qFkR|IK5Rykl^K}vpS7OORv&KvCd?y<)T?c3K{fG0s2{JGJ@A)Q3P zG-))^B@!MUG7m^Sw?xUU+zd!w^y(~+JzjinCfI33GBO#C5XH0Sglz=xUobqV!c^pB z!_z+8<$%7#qPVEi$$*!oV18cdz2{5Mi7zxIzVt#a>Qcn7;(7m>WM>QKZL)Fmr;`qE#Cyq0Ya*~T0 z*V-ZXKxTA@xpZzng!nhfiB97MAm4sbJ^!;K;+OMP-(Z8RDH9?{g;D#?V zZci$hl#@6*tQQ{7*O<77CG8w^ed!3G0FtKzNS1^^EGh7LtobC$e72zmMF zWFD&>5b&ejwIybt3~cp_wfER(>AKb>`u9*3aS+Fu9^r*D7TtkNAT}Zl*M*bWDe7eO zG`$#W|E}}X9qp78qt_Xp!CSZ@IeOWrlhcy|vaEEhk>|soMxWPrA>PE^ZJx=)eOeiW zMyS#2_|u*$=m9ajIoXfE*@QhSb6j={G=9S7oyWEMS0`Yz%>F^6)uVJTwNr2>f@CA; zjAs6d$9Zub+Ac(dRqiuyJOC>7WvArfQ|3T#5aP=^1y9+6E}KQUY2hd~%uA%t!^o{EoMjQ$)L;GbRr5ZQEq75*V;#p_db;Zw^Vb}b!Dx9h0( zUY|beME49R1hhDTIco`;Zr1_}#sCJMm%Be+Sl-tDFa1Hc$|O%eJ-8V%C=q+X-xIy? z(15Cm_v-gESiNn&JT(IgL7XRB6UOF>Wko*5ZeBTYNspkt3gU^Qlr4KAn!!BIv!)-S z=HJ>c>|Erc!bevR3OS;+MO*Y%GcqW~4!I++PZXVm8-b@RG4)c#@aMF&vTwx_BkQ+? zgXTMUBE9$PDa3bfAU#^oyo2+`_o-aGir3JrS8Ijnw+Caf=?8$=#LpKg)XV3GEHWc&DQue z5AEc(jW0^=89(OvcUQMvUy`IJsM39#5-@n=7u$4lOu?|z6)|P*mI(`yk3s!!JHAYr z-CVdFZyfU2u~DfiSmSK~ZU(af(DS1lkeO_w#}V5h$N zZSwpc4?r4f6^DOSe4HKFbp^K=7geqb=pPn)y8YcWDjTm>3v5iXhkzP3ivh3|HA?b=tCH(Qr3lCdv);Q`s^;OY} z*X4PY&d=Jtu{}u5`&jq2nwGSs*A;b&zT6{4DQTlajDavYi72^G1DsJCfKL+B#}W9N zj_jgxTd*e!yFR89>GmC7&>=Lw>mZRz2T}!6fFKNDh*)~_kiA=U4KURbJoZz6iP}QX zUPSNA(7}uF9L+$}KmNoYf16hO`%gdyYX<0WNE5a9-7=;C@`q_;($AxT;1E4U4Tg}b zAQhj{z55{tiOI;`5Hdgjo5oJGr%J130~ZB$p9M+qLN-aS!geS?EDkvToZ5SPY(S89 z?H9v?9sPYEd;;yL8GfuQ1RQ<)LF^I<$f2O<5-9SMKwUZ#xec%<&Mwm@8%u#9>;Lm6{(tx;avv)pMQ?h-fD`@4 z!|=yFR-(vsmzFh&Wf6uGX({QDgG6AwonG~PDoDGSS zz+vj7;P%m?eK4~zdY8|KMQ20cxPs^sRmzZ&;-TXdt!y;q2{LoobN1>b3i9QE{)j4p5wc2Lt zu~VD3+TKLsQS4LzLvq!Wvhz*PqBszy&!sb)!jpOI0;0* zw$W(k+Jsf@oCY2mOjSJ{a_8$A5z-TYIxCD=TpW&fH%;@OYK~JIlbBiOU?ZRF4?LjF z0`1q~W+KImqkn6FE%s?qyF5@M{BhjLEcKPbBSyicN(QeZM78+qvEkJ}AH?2S-`N5H z@#l5{tDwf2OxHfPrygpvw7CCHUil%*v0hnn>8$G+M}CoE3E8N7hF5pq>Yir0!7$CB z#Jok9f&(PHLgaMI2P@%Y8MHw-A6hi1-MHuZVofksqSwIS7vE~!rzHht=43o^GJunrSS16sMHPtf}FJzIS2*L6nyI+O7g+1NtnawNt z<0*#I;G+%@R=D`Fh(`?36Aw#0dmdJJ1O_^}fWOmAAPxor5d^vRFNs$09jjHU=^_}c zg0#V-X0m`cnWf>?QQ#7kN+&Q9S(N1csh4Z3682Vg^Tq<&J@f)~DNw2zGPF)9w&VNG zr;=a0fh~TH($^RSLq+P zw`9c{Y@U#q(!SB|iM3fq7F(i(>UNfK4Op`-;glh(1wMf#A*?3(D-uMz9mQ_W8{T&}Q(@IzC^4XL{ZwN5dfOJK84lQhSWC-;S3K5r2z}*c1}&(eIKa4>wwD0> zGza1npzL+(w1YfCtP*u>$vzf2ab3WN9#69K+iIKwXpBHd97cHs4j=XW#SmEOYPe?g z2G(1npRDpksR$9ZvRtm@!BT7O_HLP|3xj?GfX>tD^?wiO3{8bU#E1LN)tlvn@Y-*P z=FrJ*^h^MN$)CK+cV6F3(?P4?_2699V1K;!i7VoL)LO@3>6NO;J*Zgu?f0)B@%GRm zNEz8BN_$-Ga5K-cLtB8532Gt8CPi;c9W}XhT^c5xG zhS?wEW?VbB6|2R{E7Fy7dyfBNAVT}iyk}viZC+`-NC~m|4eq&~vTf941ZLK0q7J-@ za4uch^h0rxz6;`3>NxDxn$BMpoT`@10lK_okV%u5+!pS#3Ketblk#+04v=Q)FLR%Q z@i7l|Y>s!gs_FL%f8#p}C#qaG@A%PCZK{8%F1|c%xr`arIt*=D(ZeSX^&4?wRGs7XcypXV9HgE3S?rl zz29{$Hbb1GU#bcZQb;(-#@#peq&kz-|MrCz=mIXCdYv$dCNMRYClezW7Y}oJHkgQw zg8&-Y4BU2tFi1)E@|};R_qym&8CJ3v^3^9)4e)@L6-D$RRkI?ch5u}*W|2teh3`8E zAjNeqyx@)DSTfaa*vnb_#^B|C;={X~r&F(|@@J}QY2Tq|h|NPrrT1Vz>coGRW3pb_ z7d4+?J>7%0X3vp!mb?x2Zf3bGz@e663(MSVbk zG)^GIJ)3aHuIDUSTw#&>@{8dt&M@`-E;mb;B>oQhvw;;f7 zLU;IPAe>I=EJHKHU&Vr4Jr7;J%|JZDe_S9#mI9$DBZkd5TaI7)QFBu>`fDb`S@SN+ zQhsaxwR~mVyozqaPLsSjD~5FfN@d*P1ulon>Q!H5L*@gsH7`7B67@JG4!=nFw?wc1 zw$F#@JF<+_0wufSM}1}XpF5g`9_IDl7ex6O3uO9Wx&V_5ozAD)5Sn04ge0#5p;Kx) zJdt5Slg9Di?D)h60V+Vc0NQ}inl`hI(39`32cRq^jg#o%x;9z}T-r60$}v#pjk0Ua zUi1r5>`hep;Ik{coU^_>;nFLMaWLz(4Rp_39(^8U&5@Gbc%kC8wUKA!wx7yy_Bm*( zzjhM!0omaLTn}G1{IcjK*X>G^`P0Th$P=*5Xd+p6CXcEQ&;d8Qo_lbjQ1HTwW5;8T z^S$?Sq=yxVTC&|q*N`dC?FoqK_A+Sp5K8KXfqDk+b)R+cDOpwW;*3cx%mv+r0s$;g z!n`~Hgd_f78XG%eUG%C$$uWiv7*Jv?hfx#tbZN`T-Mhx+jB`&O0^t-Ig9b`{^9jq? zO`yV@$z6-637UpbV<$LiBgkDFC~KP>Gn`Tl)>1ND{r24pY=h`hcxI$^iWfZn>$ zf<2pQBsGrjFElYW&^gEkQ12E#QT&Uc)(N1Xxr7-(y#xv~cfJ0KdyKLQ7ACtZT$^f1 zS_5(%eoMZ8&&K{Y``)4d9fHIEb$NuFH_po(;vLX-Wz&;(zI9#@H$Mev?o}Ff9@j4pZJiVhdbYje@_NaIC3MO=RPl&6`;O@o*i}8_d&=H{D0JuNv&}~LqbKxT z$c%XZ&Y7Ix%!Ju%-ntn{==82J-|?o+3dbf|UY%%86Y@|eU*Y>f$<~WjOE#ajW_}xi zPcFjI!%XRhWv6YEsp7O;giu8@EYwHRS%{|A^^##-Uh5@=aVP*c_2g&hG`_AUAqv1f z(NplKARAQIMNaU9AqkDP+d<~-1U7GZN_SuERQ#UHB+OXLz5+W52TMj!bx)N_IDZsu z4^%S|_@=6fFpX+yd~?MlclXZHUY+n6ZkdG#0fK=#c*%ya3|}ut(v^sUDXX#bA;kbw zD#8Ws>TGwG7@#}R5KhDMVe!y{IKUoiPw#(vv^to6sh|R;37GpC zxOrfeW~m0rvwHVZO?7isp}$0gXKuUu0&ucXya5L|T^~0R6r{XuR_8;WDfTlNK9JE) zoWM>UeF#19p3y(nCeqHM3d7on`86H^q=E*$pztw+xJYY?xUzjktXn!tEG?v9ZYy|J%&hC$KW78D;nr} zpR#8o-u+@QU+@ziICjeg$j)P}d)={y9R((Z&zLDRB0ZG&`~ zgod}ik}={t_ZY3oPVMo}_`mJzJa?aI5m1R@YtZ%1bg;Vm=DIU{N(NBk{_&ASkKY%D z^&vM7O%D-z8z*ekPgLocT(27wgsQgX6d{#xXJpvEpL8<_r55Z(GqwfX<;fyNP|qJ! z9Gl(87pF{0;-8RS%Rc{6qivs;*Y?gFV0{!gz>Q(~zQ*H#CLoM?WQw;UWb3+QMK%Iy zf8@+4RujQ(r8$Ru6P9Da*MHRYTA<+Lx^zH_aQEZ)iSv&)swt+AI!oVA7*#L7odP6@hy`DR z4P$+6iK_~_M2`{QKdNR-`@;Ad6_Z}=5 zmAyuo@D(sW^kImh)kE%mg~Cd77AtWl=QUAs)S1^lPU%Lek!N;(Jc9~+j2pzQ z3A%i_5W}j$5>AKP#C(jg)p(V22@8N#pTNS`<}fEu2^+*IC8RTWAR_`4t-_fOIqkTf zQ*e6L7`~kdp#C&=L7>tNyfG{lX%wglMcUM<82dwsX6g1aAp#lHM%1iHpN^cPSC#@ zr%7P{949aj|19bO!hcx;F8Dna`4;bMt|NoP;6$L7Wf0@ei^s%%`b@0_E%q1xbma86 z>$ft!{WwATt!5yI(7WeGux~td(=;n`d_|DHIqiI&3z*1A@JXBG{bKBs#@W2bSMuLY z%=QjlBPV-*$kCgS-rd=IPAiiTPY0ib--+uA#V6jfah;g|vJz_u76cvn2tToS>9J~0Z6U{WUv@ycgLP%J(r=Uu~uce+~ zb+NW7KPBl$ef@opor{Q8i)$B~1LWh7uUvqJA5DUDVZg_@-kmYkYZo?4%ww_57$m`2!`BwhDYHhmXzAS6mF#XnC`i<>DB<4IB(Ideih(%xMd5W=zlqg&?n z2q9;UqMtw5$AjrQE)n!~?15APZmnI75pb@=`;ZfW7pEobPxCM$CyjvK9A(Nr#~*(b z47gI|5NfJWHJu2sKZawUr~wJ7?^1o(3E*PTl0`_ht?@*P? zE~W*=?>lBfm#dULEtXQf5W~GJ4R=BM6*$K_|cBc=31UtYffXu&NSEwUv`J3_Gc@V z-j6OqGu}zf*V9re4`5nv!QXV{aqOJ6e4<)qnbFQfH*;aFeSanx$S*Y=Ib(k7^UZ35 zV+`zJGpJK5lk9!deP6#dyVW0Ou&)#(3KC^YJ>B^VLGD*ZLr(}{O*%Y&&Nz4wS9=G( z6qZ@9u6-L=`uVvJxu#QZ9KqK)v66p zzV5$YZ#Y-7d`K3*gz|1GCWp^1^R^y8`HKPmA^>#fpl$Zp7a0i4_;^sc*2z-(0mGhJ z8G*%mQ<#yQiaT(4cfS@W!SfwJY-2)(rTxb=J#xt3X~R@~C|&|+QhCnXbAuYjPN~If zHWW^b69Mv-0_xYi3w=I@vL-xP>!zgHn!{+ANhgpu? zm$G6C|F(I@%gr6(x(&-)aBK0iDP^&;hm;As>P~H!JO)S*TTpd>%K`s;5@)fA`U<)J z8*>*2!|A}be+mW!=y(ValfFJVZ5}P?l$T>0fJa&Wf(1jSKk&1}-6eR3gvYog=Om#i zwlNpBXh#Ea40D0$;{5?S2$2IiON+;2OZmOTd4xCgC*;d;b-A;~&F8cFl8noOj^u+{ z!|BDkQKqQ%dGT^ZT^H(^7eLx^h=2}Bw{KGDQZb&-?5efdLg=824C1q*cjizb(P4Av zUCvocl%UAWl9j|_HjNbTl8W8~%&HD17yEPqv{5DdsRGD{-RG(V=w&`e1MX~{R^QUL zE4%V<|Bb5Js4(UFA}F8;+(|#Q;#(G{GWFWdM06)}3puR>p(es&Hq{OfM%7YGD;xDt zKiI4WU{Vh!WaGzut_*>8X6xRSelsudO1r1QOwY!n<@dgs&Il5Ns&5LD5cc3hsneY1QKJGgkysGbT43yx45jJ4J|#D$szlYi z)*Hb$WZ&oQE6W2Tmaw6kPb1R;TL_$u+{jjB@Y+>|ldrNG$5X9aH`j2k{xmVh4#BAkHJ?a5-=hau5=~)AGEkrS!?^+c66JUkWp0 zL3%?=2MK7;?8#0$uN1-1>F}%*OqZ}O&DJZptd(x7>#&!EU9o#nnhMG~(xkPfgNek@ zuMvS(4WuRoL4JUO0%#g3N~Z47Vi6MBG1KD4d%)eszrRKQq6GwExMe-02!UTm1Bi-!F85p48jgLyOn{@YUm&mP!fg_m%tN16Y4Cb(y`=@P zS9`vKg%_qB6b3mCx5s0Ro#mSov@W7!NCL;E*+`}(q&JA}Sb_wYhevbN61_4`GrZE9 z`*`T%hD6($4Bv={_=#QrgeW)XrO<_k4E~W`ihOIk=xOZ6m^)X5CK&~K&jfawKy87E z?wY%Nfot1)@6IZ+k0j0igj@JIY0(1>aPRGmqzo1cz-Yq*S3EYdMsbc$RTcu86dz>0 z&dQvg<@E@3KLBa~>xRBm6Y*`EnIMh9rQyR}?lEN+yoVu?XU1{M*e1!b?O_vLFxb@Z z5#GfoZjE}TWtV;bxpV!uaPdyiRvQ7L()CTPR%k(c5*Bt4sOi@mgb*Whwk0?P%oORk z*ti;dZE9;X?8z?YcQlEr^A8bZpZ0(f9DBR=mEL7DbzXiVe$K!&>0f;jeq}=2Okhq6 z>JZG=3YlPf^Zi&cb5-S>2XSl-7M#pcHisS=n3Yc#aB`|XraUlzF(`vERbiP(mUsD_k75c?=wZW zCR_a(!Z?(W{;oxCZ|N-%Ug5-!ZxvpVq#bKWPKue*x%7N3$6DjeFd(?y5T@ApB5>L@ z_{0}>6yw1L^=~|JPnhK=VpVt}qsw-d)?M*xOS4Q7JZ;>gF*Z4A^=Igx+rJ1ud2mfT z0|O%iIQJver!K>!)&oom_(;Qo4UXIo%fR%bPm-ax@Xupd!#aITpqKAcJeZ+RE=P>K;MN}wZ zgB0ORM15)TUYLbkhVj@Ktu0C*wzBPC=VpqPTG+{pm;qrqdEk< zTGb)VnLlugAynn2X6&vMn#u&G`q#zLQm}Tlu<43Wm$HbBhf`N62$+bFTmc3qg{?Im z%+?WW)q0y}Tzy9goDTU%-ZLBt&}vB!(?#HS2Wwni3u(zb11BC*-#FJ;LOAc>j+b zSjPW+p+houFN2m-NIx?o_i|?cNHR6^L;H3~2{!EfJ!2F?bY;Nf`7NW|kHx*kWqh7@ z`tPgKI+{UwzZjSonVPp4li)FVp8%^|L}w?beIU!v;Wp0rdiy5metQC0lh@$lI^xxV z^kkpXqp(nIqo=x10a%iJGXy4ARad}HxIKR`I9U=Mvc8BeTuILNOW-^yO5q;FEKS`< z0YoEfg>5n$La811hK0^+xJlCpRPTcdiZr6n4B2XV#B7!2((uNOow&GaPc_Bk1te&h zJ(Oj(GOkJ{GyDO($P{hTr^(*s{g(jh;zcceh6gyvY@+QCNlVt_8bNv&WM&7SJ&EfG zw|lb7lw1u1+Ht-lhQrl!;qZQOi7oEi$+~GBHyWC+pMh1*eJFD3n}05qVV-8y2#gx` z`OhhEW5u$QVty)lpNx<<{R-mAVU)+@Ov?c^Wk-`?A3a{tJ+eWxG>a+SyNualvsHG@ zLu(J>5+B5p*Dh1m3RjC?#0RGdhIpP+gy01%(`<*L3;{RoFI3>nI}>B(3@(Caf>zz{ z1<&%lnPLoq8NgBjN!3&KM*To84OExrW<9YN^>eFB0b@Z6lNF=s&#-5=W=N$QVz1r+ zZ;!(6O!R$K+_~(&H+kY>%|wJQ_j0(3fKL7$>Ort%8Z>Ik)@K$oZ()YcVSqj-Lmt;Ua~n8vi4` zi|1up9&Xo}Zs^RvRbr-PlQr;IU9!zgs8@mUDUT-77tbAdSJSKR zNSC%_HB)<@S&cQfoRU61z%TGJX{6YR|J+JvQ%McF7 zJ(a;f6^m{3nL^Tqiu<5I#cIcX0DDU1bUGtGtu`wng7V_-5{~7Lk3hzTWy{mm#KVgl zXs%YKd4|GO`m^30$)ed+=<;NGgbh9nnr5n0hPdKLf1+0Dx2MeHhOmi5Jg$jMO-&~b z`xcXWsyI4)BMx3Ty^R?b8P!%WK8CHLM^i5oS>o2bIkA>I$gzUkbj%`>iM-VUhV@u+ zr__j03aBtAYMnMfsK7&P+)@A7JT9~`w^kP7xD8?+p%pwZfp|-!o4NHl*FZ&UB>7Fx z8qD&C0Dq|Qbik$ZEoD~Q)`49iY|*xbcq-Icf5_7e_LG;BSE)a%Uw_!YbJGm`5mJJF zO`n(zz@F(rZ-Yh_&B*EN zeN^wXYMV987Aia!#jc`06R>dw=a$3--h3NaD~1(osp1MXw&Yg6PSwq{V&~LCUXNss z&B-m`@Buk_^EL7nMk*Fbo^gm<;cn@UYMYAcLGnAX>|&Njv}7;Z=R55{d?JgBQ3C~K zs5UGMi8kXE1XO{rGb^z4Sb?#wjwaUNx&c}WacGDtAdx85J92{oSqDgVEF!~3rG8ND z0-Y-y7mkDR41qK!YFC3@e@AJ}bA8DYIb8NGl^gP4y5nv0gQB2PEAG=f?*`47qISNcL_5A@|s8+baANAAQI4YXM3j0F#0C#9{%CKvt zoHi1GS*n<`p?1yDdDB5>!s@#N2D5PE0O5W>yC82k1-j=+2(2wiY4Q<1#^U#gG;Wy! zNV@w;keH1U+_+S(Hu7>%7x&hE*RJj@ev2-HDk zKVlet9Cxn{=qR1L9dWvsjs-WpxLs2$4w%g@{^}yAf zTvel>K~^HJ4|*@_PJZ0Y>O}1f<9B zj}-a(K=HJJk=62zBPFS)-H--t{&e*1Bae2hj0tShQ!mwo9_rZ7TB<&gM7@{(bWNRp z47M3gp{Afj`|qS*@syw;r_m1ags3~ugD?K|tL_E%19^hs9t(jZMtX|OjNIVy?2Rk4 z86#;=##*FqJI%PV`!cqqpRD3QT>9+faL=~`d)McIwmU4_Zoh};97fAw-uSBE~J^W~eTbB9d|&h-ufl?7yjhRnXPGgU9cPj3>Yg0|`e_1C@p z0UTuTtI^iMK*RQTM3|^&bS7xU_PE>kW9aOGD_B72`uyZt=y zi>N!8P3=@Ncj6t?CM0W$Y-V*7=nVUaoD!#>LpJu~C22?5==-h!V`k?D3A;1#i$O_k z3W!-$r*;50F4R+(j=)yak7EbTkUIog+inWLORQUu-94vIS->^|;(&h}kDWRaSVy7< zV3)MVwvbEJbT4GH;&0qE3>ftY5I_erjj->gAvy zRAQ$N7xIKk2@#_#GtGG?GT$KYd0DA8t93(qPh8#|AE#{8*yxqqhQVu_hX4)$rpEuy zg%0q4vZmc(MV&K>d@nIerE_HyyN4)?Em640Q(JyWVm8*&!u-lDi>$`_j@6bx> zn8+cGR&s!zYveIK32MPHu6=+E-D~{e0xJ9YwC%fOdsa2^78yhxT;Yez6sxa>{(x6| zv%A*oNKexHIGk#?k2iS)58Io?!FlZXgXYXVbImmx741+@FNe)s5+t&dd|o2~Y)}xt zs8OWj?BqgK>*M!xfF)!dinmZ1_&J;)de%(3Urh$C+*fyr%umrAENU(^zQg-gowevE zY7;kCvA$$8C&@s4JdmR+AAZ?SV_OP!rJhK{j5`araI;o(#oeve>8pB9V^tM9xQd}Sp431(rYd(fzvU)c=1VKktNk)~n5kBeTf(KiiujIwuAdKQm}+s z&z9LKckrY-%edF(>F8bGfV~{D)a;emew@YO*?XbZDN~gBw|FHA`_ZSAf>Mu7ay^>t#Wq+(Zs4e!S zm<=y3K#ZMT#W5M?5vE6CDlP;)oV*G+;{GIhWir*17LQOS=gg3zAK`^Vvn8{~rNs)I zX(L84Z5MTUOZe9I#9Ac4O+L}J+$zKDBHwfb{AJ$_^^qMY0=$SHRoXPKDuAHwEOZ?k zg^;Bxk^?D#N5o)2u@iVc8t}S?kffvZjJ_N1q2`vJpQa zM`guI97o$DR2b)bkbCbjSzn`^?&1pL?)T@gH3*9~MU$8PHed-bAkqD3h-MI5xfuDhrS027Sfr|{kSJnFv96$@R@N{w5y*lucR z=}1u$2u+TS(yXKy>-pRstbUuXoGgE;y7%rAtQgV>ISPfjp4P?V>g6>Lzb&oGPopF# za&F~tEoBt@8>LRf^qB!#TWUWE$PF=kVnd?1b^h8Rm$>}n*)H`&BY}*XFc?+LA#%jc=s=B%q4*RE^jsDAw=j6RXtHT z-wS*%Q738y)Nb5(Q~7&QcMX7o zKfzwD6NOs>?6kMY-{Fec9I+$s3pv(4+c0O>Ip-pSRFV&x8en*RGS6)txN%;L@7!g0 z>GpO&);E>MRce^ym`Q2$nsiE;Yz9gDRh=AFoXg=%9nA?|a&o`>LUJkOBRT}OEwqB&1k;WWYlj031S0FQ&6*Bd~NLk`Yh+EHUkKjDuroISPxoCyOcpXeRW zQe~$hw~*@)`>);iIv;*P84P6YI4^!TX*l1|rV#{P>!@S4rONc}^mYo4M1JQ5@IU}=V0p!<4LSW0w#|Yk0TK(#_i-hN#;I*>z;H|> z50uO%k@Z>zbY-=dgtlf_Fyd^ruIqWL%Qd}e8JRxokgA+PsMAcEy$0^<`!Y$^4xulJ z%3N3OyWLi~#*{0&FGfL@N}^=Cvrxn|5KTI`ZlCr}2XfTw`aNGXK(SS8qokvLdpPaM zs={+s{Z<%T4cjzR1**kA^ovu`@Hr^#QKu=2Eg&d}OlAQ@>^-Ib2 z#++=?xi&FUxa#}^ZgQ_K@H7LLS8~_~%YrWVfdjr<2caEGU?8d$S!WJ3(%F*WnLQxm zA#9eyQ$^vddK3GMZps|-tXpZ~t zw!=c-gKHgmlrgrwc+InveJ=`2Wt+Rptsd=cFQ3H%iSnqYSC*Eft-2#*^+ z^MkT;gB&&ufbv$vi9$1*=p{>P0WGb^I+srXFO>wPi>) z>zmwHVuVFoN#h|cpb?m*!^Lk-Q4-5%7H4J8THl~lzAa{XPbv=C<4o5@Md%8|?Ifro zSKfV95nb@OD-Z$i5O+k31DKq~zmG^Usj03gsR$^@Il*fqnCwg>7NyQwcisf)X z`D}wpI7$N6#LpX;Y^(kIH)yTP?ab3bQ_1A^-%KIrBhGet=!ps_9|f2Jw`*Lk1Kbwz z{5gy7Yb@6E22%zf&iPqfVkzZi5Zfi~Jt70Bo#V8x2x)?fX?5az8%RUc@_4~pkK?Oi zuW9eG_N6|82swfUq|)k9B#8oiL<3#I-ZnNqcPaAF4=4`5sO|JvOj7dzZT!}DxXizx z50UwT;gNwRw7N39XH+*S#Zl{oP~A91Fy*;+`&2no3m27bTK>m|;21JLQad`cBlh}4 z=yJd;_T0RuKxulRr`0}1ld4xqj-R7c)AmbWzhpjuhw_k(wxPAYM-k9;8+wUl*_S@> zID#9G-Lrx+d;SF;0sXRn$;sDj*7l8}gqs*KWd<8GL$<-EcXZf}<8 zDH^Sl2fPdpP$&WtJWE$Sc#^Jz{=7dI=c)ztE;&+2vmAM}%;NUbi(HMb`GL+XYQdau z6CY9KlQ(CJ;kXsrfi?>7O(vXFG$y>;u5!Ii*&$NIh?HoS*n4`=$|UC*2iS`;RS&hR zwbw?A^3lXvj&q7JCP4k!s%)mOX)eh`1~87 zHCki^5M`NJayH4IU0QrA0XfDvHC&y%kDd!IU9>b`CX}lbjkOJmG5=zKezB-Z6B%gg zH;Fu}<99iO%2Z!@cPDc1o#9Dtvv6tk?x-5LOAS0t99J2m#Yg4j1c zzeSsyZ>puyl4Kj~W|gCaG+KryGHERy0Z?;__VQ!#o4pp1$hICDee9+PCZ&#NP49zm zYtsY6s5fo2-E1Ipx1}F>aABCrrrcT?d*O{frKfUh1J;}lSwN4B!?gM9xCv0A)&Irb zdq*|(ZTq7jC;~!6M2bojP!vR(bVzK7fQW(+rA0+4A|SnmL_vzQP(&1gCd_Q2f4 zz!(|6)quc-MhmG4>#6XR3w0=V>wKkBrTmB&_Xr z5qq}Ox3mY+uMqclsEW@0*qFx)>cd};=LrrFewHz16X!2ce0?9E*O9_~LF&GrvJWkg zQBl)d#;%l6{1BF(K)lO+_IHY=6j?eetG_Oba17Psu#M2L36%i^_l%`=fH72CYSyj& zsq7dSa2zWp+eefF!jm*4!Mgv+rbH!Inx2e7+9!ZFNCSn}s%&c;51p5HM+GnAOsOsh z@vO3I!vjbL>ZsSfySm61f$e`igIPLWHvg-m1j%{l~9AOoz8kK+l%G=Dwa2 z_N`IMP!smVht~Sgd4Ut)3B1Sj5UbUD^^M&YYb&FN<+Y&}iF$|!J5Xvz05I;zL(k*3 zk^S^O!pgSLR^>(cw!6g()tdws_xdAsjHu!50IZJjf@Cw49eLgt?nCX4=n2`1SKVTf zz!n=zTLm}9Bf4Cc_2_4DU)6%R4@gB@F7Nk9&JJQI=TpAk|AGOIP#0i*(~KJ!MqTEP z^5o|>rQd8xZ&~vwrL^mlV~=QJ#xn4=WLD%lgbg8$lNxt;HoMJS$=FqST(QzX*6+4# z`i|M70VSAwbPy;&>gp;o&L>s7067!zX9|M ztD)0Lkb|h6S}0ip#k!csk~k~O3hV@`X+BNA*(4dDg_4dWdKUU`mj}2z|MY4!?I}J8 zyd@7}90ZP6kr{uz8MHkKVxWObhkL%I|->eX<0e zuTNJ%9ElE=4-3^}Y6I|#7Q#1QRFkH3$V$@fo>kpa?%pfXRXVqg{WnW)LqZIcvaj-~ zW&Ub-Pf3wBxRb%n`c6M`&>u=RbxQ-sBMZV@R*K$aNNHP*w*r zB$V!!u9}S71>Fthib;vN!H3ep9A})i1(xef3pF~u{#4M-2=+9!8&Q5zX<5}v-72Og z7Aani9g!xHJ-HdZeaiDukXOFP3%a+&p@@7e;Cz>}5m>_>1uWP^V}K~i(xN6~Q5iyMTgW)j_ezTj4 zdj2c5dTxA^A!}==oO^4Z1oxG>CsZh=*a{&{$%0d$P&tMr1#`Q%FacrGsUgW69VJo_FPZ(W;uu#)fS`{Ss!~7QPAO;Ysj42?OTZ$^D*0Ol@4}D;pB>X+R6eNvYy$ z_AYGdc1+$EWS;^v{JB8KaGP__iS0eYdKaD%(8*c?3{OE|MoJLitpzY7&rj9&6#enJ zY@y5-?}36`y&Ss5L=w1|GhDa=qyXa>MK7cPB0oy|IK$UEdOz*zirHwsOUBxv?Y3vk zPTJ?}RMY99j3*jOy5PkQCvwu1Pa{)nCaF2R_Dr5NGvuQ972G3KWek%;ekbtRw|8@i zKU(2ws!(e|YCG*jgn(DgO}n1#{KY9kqa!5{^`s!bP39RO%WvR79-n%2sbGLto1p)A#h7Y^R+dbn<~K(fhtQd$ zm9M5^pIRN(MQP(oQ9-M<+>{Kno(nhRKF9A%Ma5k57)Th*+XG9nCyd|@pq>}lOq$Sk za&@0K^l??}h9R1S9A6?_>`D0zX&-boXpBlwOQlY{i-t zYOHm)KSy_<{uomteSJu7fW4&Cvixup>t%80;FW)bw?Z2OM@B1HCKxK$!>GzqPXq#grYB{l4UEgTqF9{~V2H(bZf zicYW7frO^#Gxjkv1UDP=064?^X4~>CU0`sk14|nNun8J2LuV2F(A8{KWJEok`4&1G zu?W`(UKK$2jv?Ox?Se1i>5jnjK7$2X-%mli64kIB z;6>=4qF3e-)31Q@TM+Ngde2@lP75e*eWJcFoGQ3DN7#w;t1%v&K7hL}ySc#P;rh)c zAr!_0j=jCHLn81QNd%1i%ZbGe_@H0F@VUqh&q<7y`rAFw#|is&)h!IZ+ylD`0sk$6taCLXYEN--uEcYzH@T1K2JP zqHEi@&y3wCs`r}n))%PMHLkBFe!~TW9yY2@DgBLY)=!5wJhVsB#5744+ngzMY8;Nz z+($z-#fw}@RDZ01n5>)xS}P*r3_@Gf&y@!;`d_5dfcneLhQm+N26x}di<;Dq8^o+I zNJ5dRj++DiL3w{G%>Z%-zi?Dflwg-|1j9@~%sr(Ih)(Y?fyOdZQHQlWOj+HNCC+W( z(&!d(`QWEPnH^|G>{4$kuO%W0}o^CQl4JF*1zLWF!;xBWQxKM9qxIi z)p>0j%IyoI>nWSb$o6NiL)HF(}Tl-~1$4|N*jYy{zPw4#(7 z9K{}C3CV+_W2XL^}};l$jrdh5fm1g9O06V5&RTpjeZO?2KkGon@hRns9` ziaHi7RV1~89YRJ~onLh`EAfxKdGY>U-Q*9)O(>>5{)F+5t1)IqBzb?E$m{iVTx^?X zw3t4C&#DJekOgN;uwxS%-8Y#>L9PaS^tO@8-dYtecmCtNXw-WU3C^#0p`|7PKE7)t zcLFcCNQ!d`|7llq1?jq_HOV>R5^q?({JxDC>$DA{%8@5}b4i8Jg?GRbK^@(}cFFUO z1Y&N7+}y=zFuIVxkX>ma;wNcOvp)zw-`QsbKy$D$vz{u*)3y-oTe?u~2s{tsejmRI zIg5>)P;{Ay9EZ;bvI7*l&73`vzn08@@|}I7Z51{BXvR1=IU{%8@R5KiYpZvmatqra zD%cqZ?5*4_C?l3a&ol)bChzIjYT~)T+ULVh?8AKo9dlH2NIO;Af7Tkyy8+2i*_#Y| zU;j+yTs;t+{r2D-Gp+`nHCoJhPNh;osm?(o`f`X7{>0Z`?~tFWyAK^k24EED*A+}>G!XTxj6I-9Oira{0B z5yc8)tV-{apx@9WHxY7t=NvA6QNlRl(`}I`E=(#fMKHo&ro?mAyw~ zY>VRzj>pov#%`MZ2*~)u%M~pi#t&>3&K@m%HnreKy48;ss%n(yaA8{+GO_u+yi?e` zWgC1`OB>q7Du}r^%lx&pxx~`z15EBA;=tAZxh7fh*IoeQAi|#& z>G$bbf@4ze`?=Rs!CUQ1D^5G4uX}_BR9R935yi!j);s~NL$e^4Sr>L4QZ1tbcoxZo zH!Q5>vKgjIrxwEy+=6Jg>(t2?nAmLjYy!$O-8^(t>RnnT0$Kn*=p;DPwW%Gv_BxG?si|EX!;{}bav{Wlwq?kiZ$_bD&7=uj%&Crg$m?lje*rm|bk zyL!$kS|;&1pKHLc^`kh;HJ*FGaf401>(hh~h@_*vKE>#LO<2Xx4!u)Af$WYJ;f8pH zB*7LVcrDejYQwbwG^*8ryx8wtiV?vxQ*%F^n$`Ze`xr9&=mwk6_8bb5_7$BKQyR`p@p9F;Xq2k2PT%uA*dn?S7sTSF>1CCn zL*+P)&uafr(>m7hSl#=1Ah9nwhsSYa`G;q&nN#Y6lVHYaZzY;6OF*=B@lb%VP1X4D z9-S}^!Djo0QJ>TB)VC&G*&64@*c$Iw0g9fNS=-EmgeZgs?f4Wna>TKW!uL&$2kN70 z<^03M*W7BsfpmYGv{9b+b5h{i2kDc&(PTU|n&v|hzENIRjF034szc9qN#k|})-slxq5wjp;$rqogl4Z7e&jgu#lTv+ zUC`>z?O!5yPe|Ke#f+`Ot01Qqu84>ZF5$zMs!dbFHJY=qdcEQ_*^VV=~k6o zQ1gqQ!}QiVwfjZRhDK>EmX06X(kk=whD_qi+X;NSjTh-9WHeRxH(NqTAC#0yV1>uy z0>S`og~119kf_YGVW zpWTdDVCeuIZqMIrG7K&>>-H5WDF-+g3(>b_{Nl82qraoOahgd~nZcwP2g)RMrycB* zLlCNx`1h}Msk`wTaIH@`7*EYCZ6LoYm5_~?6bd;|K|XakDtl;5SVuL+y%^S6DX zR>#qasV}FCS_1b!7|Q=>RU>I|RLltPjerXpPmLxSVK}!qp@6eIqZ#Z6TjJp05 z+H=JV(T%1sF>sK!$0NS^gq-t!>>Wa0t=wxJIERYks&X!wxXQVm=0dPvVk+P9}JamrT5IIopaBYzfu0KD^JfK4!B` z^K;r^_n!BAL}+t@eIcJWR0#K|m69FxlyK?feaCVQ^pkxcWtRansN!3}_^R9tla!MQ z$_2Lzj4XD+9tcg~P^ktSx?TVvg&R0|K!Xb@IB(%Mtc`PTcCL7 z_g>DPkNV_1CQecfluh?sQtSHZ(U|$9%Fj6QPY4ymwX7e5Mu>A9dk?sW^cTwRpwEu*q0~AP{nU zi|acs)CbC=&1F9=n};j^;e=Cq&Yjw)yme@DVms5K)cR8Fos4q>?r{`*#!*Aol{9ex z)Qe{Fh*9N-Em=3`Fbog?%rLYzUX`aq;0X7M{!=9bKyju8{Shza3^0BgY_QsX0 zH(KI?>$$@kE0QK%eJhZgodOF=T7SFD*TkFjGvZls+49fkdKe@iOMZ!J-_1ge8Z(83 z0(a==P(Orv>Ve9{;tp1YnE`#8YDk2Kud3u(+pciC)|{Vxqg-{@|AR2SYzx+~&Tb6A zu98u&3tf;yRHWUL5kak4`$b;=ykc zqBgE91 z)sy5Kn?(Dt)lw$_A@WT`?P+b%Ssx{}tDVe|j@R+Y&j|v@ruwwiSlqjS^v2pIiye?v zJ5z|E4$^eg&;K|UBxhMHTt^(3>ltAcMX zoD%B2D2l5i!LNxFKEyw3^2fvVQwXRUCs7jfQ3_{Y=#;& zlbK;lt5IXf;@Fy9p2zqYaKn|4OVJA>Zt%nrl8LBC*L_)l8;e2{P^rP05D3M{?)n2X zF7hOdq9x#wo8AiP@hMcILzaAO&@qXdYAbx1Ox%p>=bSizCmNN_Vo5cYnV?JJV_L1^ zsYM=R%p}0Wu~X;HztLNgyTg@Mjs~Q_(7rpwq?ZhKm(`q!&`PkkLZL^0b@W$DUq((9Y&h_z4w8IlA)F_DT0|wF-}5j4!wt;XJp`b@-}KP%YpQ zY)id-yp!C)RLzEO`=aAr=-GKU4$g}0Dml)jcK&P`-q6QJ|D@$SXyV%e;?~9kw4mIh zFrbvyzOEfek#5EuE;An}>r6O}?@-7%!Ol4455wHG9N`KrLkYrLvX%~uF1($-{C$yI zJD%$bT=@L$FNNFBcexGB&iLfn>D1K{Ou`?Utc?M<5?LyOnKWzbI)gu+H8|Vm62RPs zDnWa;u;zdUpQ57;&rL79;HRoVE}3k}&h9BBDN%#dd#3)o*}Xft3i|`=4Vw_1K~fzKWfQTeCncl3ZE5@98qQi=&@mD4@g!A4W#8HL=uuw?w|Vjv)Mb99TBoPtuxg)oaZ!F-MIsRz zC%I~FqtbU>cfgl?6JVM{1v(U>ANUhTkH7hT*%R@B&(vOo-zV=!oqFshh_x`Y&;K&I zTauQj-j|hLs*lv;of#dgwm7|%I{CR!qHtN*$j?t|br9Dibp6|%2h{+W#7J9q`sv#z zz6e?Kk5)ihd9)1W9XX3iv5}-mc6lVzN|~c0&}j>K9O}a+rcKUizM?mFDr=QU(7F%9 zdl!io9#K$Eu=Z&_d2g9@uDXg#MwQ6g<=8WWa*w>(J`)#euDw7xfb;GE&a0fD<&>cK zc3<6l{)TZsezQebA|1tdrY1o@3P|1-DmXbwH`=}lK$aHgv!!)#Yg}fc z1&9=3#{hYcAE3T{{~J_H&PP=SX+$B~X*!&yO@kB97F;kaA6aR(1te*k!~@diaR}Ls z8WxYxwC%zo^7I=w9DXA3#p%0~d$3>Ri}JsSZ#DH*_AsW6B5AiQ(3`U0B%!T^l^+F6 zbTdIrLmTeg#w>w7+l{HghbDJA*eDj&n?@Z2Ckw2T;FdF4*O=K&!($RK7e4HLdM8DX zw3X?27Vyiy=@4ppT<)pfUNjv|Z#n9Nja!-DxM-VHys(Y!1xIR{kRbbO&5lQ0c#0A> zaoUUx;x-)s&%qZa!)SQNAgb$14I(Xj`Ik43XqrA;KBx3;#!dlb-v5S2`+R*U&CR{K zB6Oh?EbI8uZ4#9{OScu-!G2t^ryG6oKm`CGlmO|5amP!TW0!#)@$1JIsD|U<9PA*? zRMq!Fpys%|{o)q%=OMf-Vo<@$EZGGT4c*kyEc3KBSkNDMR9C7VuiQ4H;7`AoOvRl2 zkYng5KWTLY@=1U}I-ZHLbX4C}-hg%84opTxKw-7#3!7$icnIdfYd=9OO@lpW^z+wLvzGY^9#Ff8$c%yvRmT+d6$|~D-MjE}Jwh^!WRPDS69ykI0 z&~`jH#DxU6$(|v!K|?CweA+O%+#FTqltxyinOy6<_OW%F({SM2BE}N6VB939P-TKw zi(WoL8GkTzOFR*RZPkGr)bwKK`q%HYUC`r1*5&OfzpDITCt_oP-ihnorP>oy;IZ1r zB$Ii^-sLf3NYk)pIlbx02l&7-agGed8ClR#hoQahl*((K|Ljd>2dO z%KScdZQCDbp3D5<41}2}%-5r)W{CruDZ#g~3klT!H}Lk^{%s%puQ`iw{CC->e$7gg z_5E3*plkDGplmmjZiRR-1hYo%cZ`4UT=1K1hzN$vMYox?H*Bdie6SN}GVzisd@(Pw zIK5OI4}J;R|6M1Ar~+k^0iFjZ?=%QDg53+EOTRWTJN9VeS#jiz}R-}M*rconKOVcpw%-J zzHBFK?-z6y?nehYBD|6fm@NVL-}z_{#18@dwus+s0KrX&rtyz!f?aC?T3NwvFRKzb zCSw4^-*z#oy?%2C%{rzR3|*AK{h^pV4>4be3e;2KyLw2L4!>#jIKi$@{pc^^_wjJiUqwr zqA4p7TMHcmBEAG2KLMrv`@sTiymqoZlN0tgHeLh%AM^cd`rED93_9EwRgl2a-NDqN z{%NS^{^=8h2V?%4?qAakN;4wYp}=AL-w*ZfKYfNcu78{FKjsf5{OyKK zv&xNkg(r|eqwe1g72x0fADi^%YY9U~; ztiHY>AS2vxcs+ z69+k;l2`>5Typ(|rq+~_wUvBbXE?)FUcXF8NlmE%zf=FAA;n*|2U1&ABZr=Z(yxGvqcAW!7A6P_5OTwN=qqrCy&~WXGzVY%T4h`Z`$q_@78r zj~Ub;9>>&9{+D8xAL{SHYXjJa{t-CkX%8U(F3OvuBI-HIP5PFf0U*XlKybowv;GmW zzo_ll=NIhuA#DE^dW|ENQ4&zQGTeymv|R-EM^irZp1piUI@p4|E`%J_)JZS35$}f# z8S85g-SS-Nh0OWVE_r$8Ve2Kg?A!@`t}- z3Ho-nCCDs#%U!U~J7(5q^Rsga{t3f_dScDnlywr;%15P_{t%c>u=C;1In8w{V(t5r zK?-Qh+|bWK@R2dj~jOn+Fxmb06nuOT8d z$E?i44+jIH5F4qJYEoMlhUtS#<=(2*2wrDiDETc?s6TVmD|f-|B&@@Cv^wn*;0|UnUu6Hu(MPO+ zzWd^GmF*AG_oa3C%gN&`DQm_p$~d(S*b_5K&PA*5dZ*_s@Df}z5$gi;;;m>Yn_>o< zpbefj#?H$MnjzQTVY{~A*mer_@z49)Dmn$VNv2-}fFu*tnnxgDiq#q)Xa^GC4zAmd zXDx*(EqvZQa=A|Ak1R>R>n4M;oBc5>tP;F(#Qj5OZauYMYk?DrcT2W1sE#zaa1%Ui zsOJ`LY3K5X(W>Sc80|ffo^~M~5QOg#jN(s3+F;0MOqbw>PB4l?wRVMc$eYXHH~Z?3 z4_tgwT@&wj8}Ll2MTsfs++jR3xF^4fietQK+{i<>4gJhp9z_KJH6xQkRI^Xe zNW&n@C)!ybNANOF1Bq;qcyPKIvpQQQ|D-UfM;GWw(DJa?sA&nBUclFLolA5@03UTWN*wU< ztuMURmdt5$mYnuE{ijmGCp?2QbXz=e{N3Z!6S|sp$$->}UNM#@^#kLl`bbNljG<%l z#EeG&y5u3V)n|>)%J-^@Q@v|l+@Tho*hT7FqJ6IY8pQ{u2jHy2E1>`bH^!1+ut7%F`JWjf>09spIcphKztL#R>)pHN0>pT&X?gI`ag7DQ}~@WhLD z`bEjj17l^oF2u5$qY1yWr;Qa_pOLCJHm(DIeIc$in5oaqL{)Cq9k}y2exKhhTS}ML zF&N0%<3zjn>M6qhjIOx}87g9dEcA;oUWcD?JwXecW9WzL2bAe-|7LS_rH#=yTjKO) z;Czk{_5ObMXZx<4#v7l1ZCePy1Ug7{%rq5F3hI>?mi?BG)z#Cg8O!UE#7~P_$ulkz z4tBo7&N6Nhtn8^#{Q3(G zHUfp!4)`2$o7!xi_KC@s+-*x1IBs2t3|_S-2je4If(6IqKcI-SAiBMnmn-%P9BHOS zKkj^^&rqurbTpV#YP;4vA|Ag3o{}uD#ambj}T&U&=xd$|w zW@Ts|)6f+?B1(L?Dc`qzyEMRW1Lpo^cP@9}JXOc?r0fb%)c{dHy9!GoIVc6)t46*RvFL3a>15?~wukHk z1A|Sn9#_VN43F+s+2!MV0oU&L%Wx40cuBLwA!+Y@4CUZ{RgKmgdZ%wL*}P-4wha@e z!rJyVQ6mSBDmLjK@YDHZfCuRsm3Ao#r5tf!u0`BSR(YqK=yD}TRuTZs#_D3}L1m-^ zF)!02&i`8|`V*2&T;O-_m1)=<=bnq2PSh9L7Dop;}_=#-+cWm!2R|8&g}z6LML#`m_-MU z&TaWx9db46If0mA+JOk)7>&(i3 zbZtc~#lzOoLAcrb-+r@s>=BIodAL;u=096^{eXm-0x?&y@1zD+X@I4Gfc9?N^{VjQ z()E3Pz$nHoF0S1~Q*5h3^oDpVPMoEH#7`dGUXM-=3->Hg%6M_M7<8P zoOPNc&PxjrPtCafFyY=9FB>;=EKYoCZvYTusJiPV-6?!FK<^F8!8Ii9Sjy-#i|YKy zfrGQ^OXK(U2mw>rlQfUEk%xn|Y97>&9t_I|0dc)Ab*^ArCVwqZ9}Pk@EXSPGs!Q%$ z=muA|%`dpQ_5U;-blMJ^vTu0L;$2i)JiH7Zk)bgclIFJTTtYt52h05LL=^kKzmuo2 zJQP+1==62`UH$Jrkx)N)lP#eK_8!^_4eMQ=97gB7^5s{3j=Hry8k4rYb&&bcg8e_U+H7setEzh3GsZim@~1UDp3=Xk7Q)Xe)XCaf4iIq)KA4<{AR0`8T3d! z{T~GM0X*8B|LPalf0q+1wsU=CXZl@<(+rG~m~z8X>!+(E_-sow7a|4ojCmzDsa&-> zy2fJS#+f&J!8YDNtQFMn2XyXiU&crzWR0U2vRC>sQ{LU0{Tlpsh#3QnOqKZU4XyYz$|CN7l3r>S2k-pV+Cm3y&$YE zFwTKq17KuM@Xj9mkD2NnFgMVaJP2Li!U2}Be|u**DmM|83JyovVd$ag5@Ti)yaGf1 z1Q7%(iic5?OQ=7bt$)6#Cn{$fmx?^i$~9woB2P2_j26-b-Qhz{pc-G&08-9nAS|FV z|4y6=VJBl)FM|OP3(z=h>1AaPvAk$udpWS(URoH{@i$wg1O3u(wlNhtP%Pr%n1(Xh zxtJ^fJ`NDPj&e1k_V$73O!+^Kv3nTHUBV3bUlaY;9sSp%`>!?n_XzkuCm4NE-9%5R z(1BbQh+=yG7RB-u7`!yN%V>JSXpj(K~fsDDTaI${eL{sE(u8J6)uOb13o>%Xl|T>9ZL?Mqx5HlHp5WMh9fuYkz*qI# zD9DYAek)Rvb%BpwYTmV(7Yrkpmffw{)Jaj*4b8;LO@6>+J6a$U_l@4aDiQ6^&QkfT zE3|sX8Eq~596+Zk{#~{lrZIyT_~mPQxrU%}fSa#6Dj_ZZYRtGl63p`Yi&-Jlb7b)k(tGBgG3w~4jgj-hxl^;i2}+9lg1(NCKe+if+lgi9VT5(S z%t_rHeJ#-Kh|hYLwLk`&)eMYkTdl4=Bvl8r#^Rjzozw7Cm~hIZpjC!ZlmPBuRx~6wy6`t~&tq9YoOF=~)ZxQ=qnZK2P z7^JspqPy~)3~=|_Qj7E!kD_GukjFZAA5)HyifGv}+7DJ`^v?fBz9*erWhyqa?f-XlceWsQ{$#x3fOwkJcsleQazu&2Q8Yi7Th zkG^diECLihWEOT|1hI)v)#izAU!mjSAcpTuJ!f5IKwi$p6SGxVQf=0|LOZZt3VMW? zXFFRORJ@(U*iUEb8QLJbu|&hv%O%oja>Q+`14=nk__aBN3qvjzEw%ypTe`kY~$Npd@TY_n@D(EEan z7GXgMb&brD9etQ{Z-Y?^b^W0zPT;+bgvV`Htke|jYKE{P+E@@oT?VsfB>?73*LOT% z$jkTOhJoJ)h5^rR+Z6JfCno%uQ3|&jE3Se@7@Is+`gBNJDBws`$15J>#V-W19{TYJcIkm~?l#WHTB>nv4kB;u$y5C(j$~}JwdVYdQ4YUXfO^3ge_Itc zt6tUjE&@8^UH#g*qP8IlRUKKA0ssym7YfCPmNkg#=dNG}3+`vPfP?*gBIQ^CHBa4Q z7#)JlxZXh7Z7A<1+ScZt)RsNv9h76cwmws;ZZV*9n*oX?5$gG91+|JZ@2zpfiO1J5 zLFE&`^&}pZMWhYN)pohcbl@5Ri1&o-)+Kgchfeq-esVn;hGh|Eg%+zlUc6gmuoU{EdMu~k9@<<)*1pW`l7GZggTJ4E=en# zoJH!u3K?Kpb}mDaBKA!6OI>O9PVWnjNB8*+ibTfKWPg_FC~e-Xjc?skKGBwx&}gKH?{P^b@hiSHXn#`{Cpk;vD(VPq*iX5DL_x6pFmvy z4v*0cYjU=};5zo>OZN-X_tBTl?m{2eRgjfS76mb7w#PEE+s}mFO_baKICi^Xk-j!r zmf>>~plMWgF7%VSYq%qEano6)ni}bu56V3Gd3z~Vi@kOHL zc#xK#()G5Yl|WrY8=VhvJhCNC;$xfU7{s|Sa=H8+v9s+vviqr*F`u! z9bG<$Ro{v-7I3gYy}1;S3N{Tl;v@xS$An47#Q+LDSg4(G^aK|zd?e|0si{g`{>zqg zY5Uu~nqG7pyUXrKd@M7*Bm2y3Ox~7f;OMG{vf;aVXBkm>$&IY@p;il=ZUeAsP#btx zy~~hQ4I~%oM{o%6npzKG3iVIFt$qO&RA@uU4lsv%t@J)OoL-Du*bv&t4uw94bI~xZ zWST@@zjO1-G8ovV_r22=`Myemn>u1PE=n`qpdGk_NIrlz)&^uH+W<}T@L;N#ffeCW z^Wa-6lVv0;XY6yGgKe3i|Dk>$B{j!gx&EgD;S{?J(V=jyzL{DF(I()WV-_xzZxRh% zQ8SmB+QnLX1j0~%!B1{Y@(B+3D~u3ymMD)ReSz|br!S>dPdxO*X;f1Nz#B4pzkvmm^wYZN; z3))+opaO>_)2t-Oykdb#1|nIqTl+j6;mdJ?HO zLZtd;hAug_wB&zJd~vXwn=6QYS$5Nh8p({&mj2>NJP^RW>GHLZ7UM!a=oHoMxhV2%R+*oH^s?=^gi&O9vMNWfoSUlthRCjmL52%V_TS zOSv|B;j6toQ1v(|!Y%5{TmE5#yOCaeu8{!$`Xy4}YcQ-z5Uw`P@_Drq+A+iVZGrNSTSm{rF7erVfIC#_wCWBg$TEg=&``1 zeB>nZa2Sl$)mw@$zkdH*QkQ49+plzNkNTO-PchGJ{X-d~O+dHtn{-HN`&1?68m1Wx&Bz=ou>em=e1(KqLg7T8wWXGcEjP5>$MT5Rqpfgzu9t8pR@bV079F9o96wD zPb@#AHV`p9263c(@ZNdZo&E;+#)bQ+Dy9=aio!FJbSzob4M4z(#|3Hw_!xT2Kozw{aWzZ>?A@cahfmC%@eGF(FC-;DZ)3l~ z{}%4T+8Qt;@}__kLo90XbzzE}%GrL6A~P$XE}KUoG&`$(uRnxo^`WenzdS7)525cr z@zU6Q3;P2+@{yWACbU{h-~QHHQ2XWrW4tXu)u;4{mo94+SxOS;>C;9{>5_*k)1GE_ zoSlw`kPd)7g9rQxbEs7}y0)+2v^Y9*($@xj(GsVI>N-s?5CQv*(p$?K-94PD-l-TT z8HKWb!pnq1>OnLEBM_{7(=Oq}np^L9&5b5nozu4Ev`=%O@(en}xz;RGI%$9R%AtfE z>x3L0e?SV=7TEzUv2R+B6hD+?K~cedeDTq594KOrwJ852&hw z7x%n^iokwFbwAebZ$G850Rn`?4pm+Qw6Up$^Ie^-E zC#DHr#%e!u!tdkk$0D6$r&r-GmNe@zh4*!H96h*K=Q;EEm4jo#(#m$LDE-t^J*RsX z2=ynhzY0IGy=S+ZjL%rtzj)w)vB4&E7)x}?nZ^oYo1g;S$d(rdjUu4&w-27UMCkvJ z%*oVLcZNNBbLY4GIJiGo&p)i*hFHdo&VQRP}G`mB@oo(9t z1;S^GElbvnOaavDC#jZ8b-ef}FmB-+@X-~PG-4_$nwYkCBigJ>SsnW7n`%GgANyap zHc}&hfYpNIF6C2IS8?e0O(9 zUqM0HB3pOK_H15eK*pBvbqasNjLt{ zVxvVJI?3&Md%jN@blwTB7qm!&w*tl1A*b#mrqS*G^$^C&|R*8}wBbrP2r^7`q3UxC+#JaqVGJ>A814u? z%CY{2a@QLAneeCJtqJuMZ}GG&+zp()d{`_(h4v_WsZVciT&d=k2Vp0sb3E+Es%3Y4 z^reZFMj@+Q^@6z7>hhv&2X8>bAf-{{w5XjnQTim83K<)Tb477DRBQ}8s6t25mC?#S z<1c>SazNK1a_A2c4g`c7F{gI_G(Fw3)6Q)86JoAUZTX%+C|6w|FW{qcn*38!O02Ef zq5fBZ0zNUxDCHZyw=*~{7$&9=<(DhkQM({57RY@GSY*awQctZ-uhBIVI{ns$8JsHA zq$W%xk{1wd;ROJ5Hjz=F58l;@vuL6jeq3cb?IJUhb(mBg(3E{YFm#He+*rxJfmjsF zKe+3bZA5hal{Yi?PPJd%N%+pJOl;%L4Rx*W-~BTUQgzoHnwJN%2#~L=m)XDB?n=}D zNYJPPb#iG*HKrGwd02MGscB>t_XeSHpzN8)$>!&JciY=_F0TP1#&-ewVBAMmHQ*QUwH0^(isK63@t!_nV}5Hr|DO|(sog$@^i-{ zQYou*HEkF9uLzC%?ocIYme{3uG-woMH}Q&IrDk}M|Aw#Q<=I0py*2YN==w-~SbrKj zEnyn^M$=%Kb*xtSP8v@3{E4{UptMhf0023|MNx~wJq3#EpiFqp1;y%oN^0ec-uT*S z3R+nkwl*J`XW^9mv_9uC>xpvVm4ynfLaiZ{#SI-*)h5D%#agt%x%8@Tpwah#%X^FC zckllMD1OGeAa0VA2})DxgLrKWvUEbM(dXY_ELrNn2 z5}&S_(hvZEony18+q^Y61MW1EFJ^-gedqX*{f5?|w8|A(ahx9NG5EFpo%*D<>RQpz zNwA$Xt%BPY8O>-x*uaAm{fn~Di`^Gy#QbdbWhrIrrms!3(}`2KrbzkFH16V_Pw+{Z z1j*lQ0T+L2RTxc_5qt^vt%a{t)<|=*qg!6?Y!{XuI>fWvNmQ+TQW>TnPC15X*dEah zgdblJ=u}EBdz4XeAR7<`CX}aJDbBC z2{ahxHi*|$z~ZQ5UWd_{W&CK7dib48nG)|PQ`cT`vzeT0@Jj(D73Df1?p_(W6Lrk3 zBz^^b)-|D2_%e=XHn!7!Swti-JaZH9M*7XBNc*V%?$LW#tKsSoT;di3BtdJ-;|23E zqxr>&74-b9`s-LbT5o<}3EW!4xYG$B+T_2?;!MRu~nQPAZzT+M5 z81RK4Ojto}fo{9uZqJe_@-4DON@5F33Gs7mi~F_XM0k>DibO06#dta%F4Pib-EgCd zQ}2?tobiVX3QwDs*m_KCl*8J*S8vZsDX;99e6C0^p#an;N+K)}eZ-&+IzPebZgu zvJ7v>dmTldiDcP5Pq64SM|T!~WXW$TNE`_{fE~bx1=SDha^>7p^LTTRGZ?YPy@_1} zXAc`8AUvv(bepi9eC!WQa9Cn8N%NJ@LVsCmDz}KrIhKKI$o_jq3J$Eb`>>A)uYSJd zxzrp^9*wDfb?liU%o@$BZ9x>Rz=OS*B%-kj0x6XHfjM%k4Gxm$SftHdh{{a6fK{ z^{Mux^7%@Hxoh&dP)BLpY$RZ?l8vx{4<|#^VNI8a@S?LqX7$wti>{ zA#%)^%uPMtk6d9?d~Wexa$6_&sP9N;t-!=An6UD$vR(>Hhp-|WhPdK6>j@fjEnj@S zXCJEiPp`_(4f%$rgBI_4P`9^xc(>3lmo1Iz?7wWIe*NZ9`;u880Q-)~0}CPtP@K|a z!56r5q-`x+#Qv+`w)j_Ck2vmgj^b5L#mvNIzIpI=I+7ihY%!S68wNCh=lp4QJr}6! zG$W1#skMRt0s0b_*WA z&@kk_LFr!KAG|5o+eD$8J$rf_6`nTxV-qGsMe?JMUi;&7Yuoko%Z1TnA8l=+Otcv8jucgpEtYcqy7f-<$Yv5MlkcmI6`O8fdevC!Z{}O z7~O|T#u^|g0)#7FVe^F>gDeZ6N3&b+c%kwg`pcQaLX>Ty$xTZS>DGC`zQ%Uoh#P5e(05Z;2S2(2~#5$75$eHOneHhe49d)c6 zebZL|LoZOzh=OhGiG^-@r6jTbSdU#j<#Qi-+A)R>=(Xj1Mt!fuP=IQS9Pl|z}BhBK#sbCBJcy`jEEWGT__QJ|MZTR%n0+ot0lZtq-$mc$!cR4AOXq! zA!u(Oc(-i#kYm%nP(!`F3A?bpI_UbINX({J5}wkp%sNuWXmZ+X>s&O3H8jEHsg7>NmmLemhz&e8;3DWTEY3ooqe00vm%+@f3`R@@dMKL3pYvKj#ZCu6;-4gM_Cp^B6Qi<(=L#HcUf9@9Cy6Dp85Z!%P60NQfN(5W}h6 z6Zy0#3-{4G#8pvfHj?ij{wuc;p(-y-zB0(#LbI${Tu5zH!$r#L95OCFM+jn@libEw zK8XEU`I}=1+Y9>zW(tTKo(rfk9Hb(WP6LgFgD-*x?*qnFx=&eS@=P$R^gJqDit8IG z%T97@lz?33YEmM0_YAZ{Cn;DUc#ep4nc7wWGY@*Q)okHwL=xw8PBZ9b_1(HW=h+pi zxpE5IXEn`udJzo1E1GI-rk1#qP;>luln z`jHlrm9Sr(Lnr%rJ-ziBT4ytrmUDMxi2|SdHZe?r2)4s-4n_z7RjDswNC1C@+tQLQw%CHS^H5jBYw8BxGS0wd z3@xHax4=6I02&4KKsyuqQQvmun0tED_-u%xfAo!=6Uz1-fiQwKPFcs-yk@FReQFTtsR$mQkmk|xwq@rU^EZVZ(Jcy4!Z zz|k0xYA7c7GakDoSwkD#O%+A55_ayPUFN8?LjHdm^%*&$y)T8s7)7s%w0*UHlx0!LF}3q-cyNupsOA=`D-Y8z~^+} zNq`f+9_WdiQ=f(x`8V}MfRB{9+v}6dUe%=|Fvmq@u6v#JtTg2hGf})x>e2Ka;O&L)ZJWD5!AC{S)@g%^^v_3H z>`|*de+W#3^?>n{f(C)z@Yw+3{ovO(j-4}t8c*wSKKn*DpH@`6ZEDwMVp-T5FQG$@3u|5T zDU_wg(*1>-Jzwr-M88h`ko(YcefH|3b?y^xp`xI`UpI+cVC6&P z$F7i0YYZ$jkIpJ5#Ms^wpGSTkg#- zx-ia`vX}9Iq)?gN*(y`zCSh+U*Cr`wHbR}j$<$4_MtZpzgMHYvDz5t}IM7T^^U9Q7 zJ+?~Mo_>w8uCBD-ju}Ur8J^5{j zYdw2NxVhC(0!}nQwwaJ)K3qCfS?N(wNBF>5B>B?6Yzwe>9!;&E;?*^h4Oh3vt(^QA zv9VuM?%T3BU(D*gH_@rI8}0Ftu()l7ySKi==RW4?)mXvX&E^wOo2hvuV0S< z2sj_WmAR11ba?adq%x@)a-zic&4hHK1VA8`9|xHkGZp?}q~CpZUrWrW?0QB`kk7^; zvjFIPnlxn=p8?uO-&N*g-Uw?An#WT^8)ULG6-)fg-#p#395;&p6l2_{jL=jpiH?ZK zM(#nJBjF>a=`&6fU!eLJFzt4!B#^qyHA zJ&!gMcDHj>Af{H;qy2-Vg;z}3E{$<^rQa3LRjPGLC;93KlsdB7rE3G&`LMg2T#-bwyUTs z)z2F5Kdq5c4a$?q_b)`)oj&hUoiHMW(@(hDsp3O=D_ke*X$FlWO%X%L?*q14+-8V! z*Oy1Fjvwt~w#1;hE{RAF5?aa>`fE2!izQ0ZQXWd{>@$xKg8h2mc_;sl@2MJS-BaH9 zSlh6iu#cs%$z6P^4uG!g$J4`R$CUXjIjd*Tx)NNsCcbY2i;}w@cXXf@G!q5ob8$Y# zV{u`1=zYztUep(*7{}GQ?8(`=xJMsgTohQki8a$Q+15va^20M|c8Dg~ z&_>(t8I>$y;zdJED^wBT$JK;Rp>M;(>2nzG;q0VK%izG>Iy|UziwW(ha_Qk$sy-pN zo+rONDN>z;#Z)*kxpuygKE~@`n*2yFk#i}sq$5wW<}+Aq_M>5N&u|qEJ+4MsqY5ry z51FJc_L7+%v!L8x7AJ&V?$BU8sCrWWQ}3JiZoDhsyk|ta=}jLT0wSM_fPf9SzPsiCg5nK{<)u@X%;!lK(Crtf%#$9fw@e^Ea*tj{12b z+Tq<^yf2d{`>=51u=W#iM42pjLj^jF7l-gVgJ$8cQtrlQp1gemN)j2V0~lCSS8GmQ zWiRc;>ykai1N)kC%Ep*+%$}J-{)v2BN-@h2h=1rmcHjJuCV+q8?>&aM%56GCiI~hh zLFP2VmP8^|ped$DcmsJOz#E?`@+jElbJ3xCZX_&qh%~x{5>e-QKk(Q;!){=;Sfpfp zm#}0f^{~o|KWi8Jo#Any*}8U_29G1phjcKMSHUiQ%B5|(P4h&i)emfV{<7Ifrlk-> zE~TZX!>fXmHRyd0eC=`SYTDVz;+mWe^-Wj6|IumXy}yf8D#OTmB(lxw(mPMJz}?5@ zaA{u=!Cf#p!BDIsRM`UNQ}kklUJX#V7SUI>v9lR_T7$2Hq$ic8zCa(($bG?a^=IVS zLJa}iwji#rlOe?>8D*J{YGV_I=bdrXY*IUaL)4 zS`_*kd1fT_BMZ%~v`^=I7P!1|)o4>` zzrf@JSr21>oDxb;X6fH>HNO?xDW~-?HemED>}kb5vG~Zh9Z)U2GiC%A94Jdhh0RVJ zZFqOIb*eHf2^pNhh2UEv7a59M`tHYnSe7=r*>*Dg?sOQO8-ZjcpTJuo%%|U%U)PG~ zK+-16w{*zkBzbD#0W#-I!yG0)2@qE3WadW&C=q9$xD7+U00g})WL?FKax4A_Q)Q6T zMkV9N_pK8;22aGnDtD#d^$5;Ylxu1ui0(cwB?FER8M9ELVVSS9mhVQswc2?nEC>Uc z-^n7%Ag6$0Pt6SKwe{0`%ijAcITv58?*v|$PrTGUN^xcG?`2-e4|uw(XLv2GLDxy{ z3Fx7=L@b3C(YP_n;(wGpH?$KJlB1`Msggilf`@x4hx;y7z&JMyD=gTkYChL8fbf&qTcc0=4j$>Nt<$vBMrH{0^`}U)(XujIEoUgQc zm#_~&0R5XC)1KA$}a2`k!zwf{Z4#ND9GYCg|p7*X7hz0iJ#Hh z8Fgko$}jL0pJS;@0!iGiXadycH)i0WO1Ic1mqL8fA{6Dkzb`Kw?IjE!X%MV+6OcP9{ESQP z{j1!~0Z9ejw5k=AZWZx8OKv3S9GaQ-2|+jlalMHn>#i>!2}nkrv5w1|@AV6JvR zl#pXVcgu1*X3*hpQae(=DrfL0P2ORRoNW01ebx5V29uxBP;8s5s1BBs`Y~TcTwNfP zG0(FSAfgDZI@`OGgSK#GBa>^DGZ7Q2=a$^=E)|%^ExpVal#Ye8p}S?}4Zg_#Q1H}0?95fbrI01{J_ z2nhZeBFL)59a$AJQmP1KB|#t#Bj+o1X_f%+a}iLygid?Deup!RRN!htS2HHAb*ugD zOG6}mVb|MnR=SV6FOEcf{vi%dE6dEkfs$G<&%BjNQ#VLz#*QV-lqY5OfWmUvL$z^F zG9MTDN<)zfKBsZQtVEnxK*TJx#SO)8a?pmy&@bak1zJ^WFS*Lh{3mPz31)@i#9g+s zc1u{5vZUENl=V~CBHdHd7+yN~hOK;)#>4Ir)C)Dur$I%sdFo1~Cxku@VjpPz&`v)* z_EWbDV#&?ilHL?#?pOsxum$~FDTFc+o`M&#RtLv0swcURYtr&}#aCMp%VaS10o|4? zZ|)Aa9%vf_Z&U^vJmHX#&s{3ZkSdNDEPwPqMaj=ADDw^7iD5_QpEZ1?z*Ncp+)?wP z!N39PWg=CSyM!z2MGiu_&02amOgwWw6Yq9D8e{KUg6E%M-??u|oZ)>%x-r#CH_SQNVDat&=`cQkDp}i^$5hSZ8nC8TP zoKwi3pVr~CkVs3CT-}D6vdSR`C#_PW3SeuS&)L}abiBD_kZcQdgfKJJ638`K_Yg7+ zozHdQrWxx0;gfWk0%Bq;l1t4orOa2sIq%pRICoFEw>e)+opfD>!{dg)y2t@&SrEcv zpe?la=pG@}G$Xkp$RklnBrN7y!>tSIKv(ga3n_>OAX;HlcEfWju41`dZ_|jIlNYsk z4NpY45+Rc8AMiu9d;EMw9_tTHiM*-J6xrwV=Ph3m(APf@#^JN~Q-Sm5<`sM6JEyvO zH1l?Bft=~YZ;oUrdkRI32~lPoLH)olf&atdfsn;~KtBl39-=p0iUU6B5kCjwU`i?r z#0i=ts#=@iLL$s8#E{w=hjIM924g0+V}@N!k8UK{eh3|O>r476zRZ_GU#hls0CcMA z#)n*>l=z3kbnDhR-fXigEgUEvOsT-C8@a+Nz12c$FX})%{$eS3Y^r|uQG(~gT~8Gy z&;|gL%~=p3N~~CS_4*Q)`^<&R1<#+HY@@OsCO06j*#Un-kI*kA^Tl~(*%i`r?j79* zP>0D%_)GXfgcd2uD8Z04>mGjH>75>rCRxqj-Y(C$bVE}{;QpY175x-;38%$+hf^pl za^X2}+)w4)+1H$H0D`Eef>x_ThSI^*Rr-UkBP{?Vpo<^T6xnG#l`0vWW!JQY{mS&5 z2Js>;^~?0tnmQ|eK#n--U0TVK`rwa^rr>{q>VMAy9rrpX%Ne8ci!C z#@Lec@eJ*SW>E*^t-#v`fMDwXH=N|ON&JE*v z9MuRF13cGN50AlxPPn1_Z8pJ>=b~mxU)dF z`jm8aNMzH_Bo59eno8q>(5Kg`{v3J3c?^yKso7sYa^N8;(o7v0&%rL+sTcT%(3etD zxNtUmiU?#@s{J`P4PCBn*vRU;It=+0eB1)BT_TxT;0O^dhuwMalH%|0G{QM(1=0#w2 zpTCnNzzULCH@v8K82Ml%Agl!}1aE?Z9s!cZBLx>;F>7Cg-j6>()r=F?&4>qQ-NB7h zJiZ}@R(oc6L`J>49W2%kuIM8lvYWtsJ(0+h;{nCaC2pDDwDh3J_}q0J^cCAm*LR=C zmb@Oi*nKg{+ufK}=C!6n8{bKROKgj5&eqfX6oL}Q2{tDca{DU3KM1?Q$;nx!7}zy+ zyWhBN7Dn_U!SB&S=0*@sA1c>%5#TUEJwa0f^fR0e37uDL4QJ>ZJ%c}bFRIr2WbIP0| zzGeoe%sN+%whe0BSAT09Bmi#5gMml!4Aq``sS9k*#}Misrs+TK`rU&17uiQYu}p|J zb7e(DstC9p+ctl0QFd6n=T2F${8IC`@8|9e`{Z9RUg|FsRDXnYW1eKylv?!9_9Krs zQQck{=rT(@=Ft%@>4Mq z$e}5r7sI5|C*P;%d6jRUO@-W3-qTCD?q%!qWVtYK&5O{iu5?Y& zEUyerKhyH{m|#CbnOerCjYhJ=#U@FMOaaX%*ybmB2vUN@)Dih*z~ z-_7S81&^L$pVi$VKqHdvJjvqJV|4eg%bP~imH49dLC3&%kJHc_?HtqK?gHbe{^m-8 z^#u4X0zxmC#=rK1KPHuts@f)P3$q;FKz6Y-(;$769z=vt>&4JunjbN~t|=6|uSoL6 zxqxo1o%)$AI1YJ?ESM8n%46#HqeA~_R&3oRAkC}%>LsA1ZX-5RBtbLt@$oU&|Hs~b z?(QiOymQYXggCxrfU-StYMqr<$#`zW$eo%LT!DggeLo}gNiEpO^Knsb1!F+~vrtg) zazriYkn8?0+#%uT+5LD)cgW!Y{hV?gid(&Y zbr8|vaQCo;yi-c9{*MuUt=#;Qa(V)}ZHCr3=7-{6*(+*ztx<=6pjsXk0Bf0)0f>OF zoy#?qrNV$nl6atP!n5>z;IyB!+YFv7fE(B-&(4V!R;V{!tMAd;tiM!B+=$_l+k5?mz;mW{%(QkgHO6;Gtee!)?oUy|JpW7eoa%c zAFkowW;Ubx9N?GWG!gsA=ry4BVEDzVbHHV!Q}ehS-`=pz4eudx*U^_}mnB7RWV;4+ z+(GV!9{B!C%$nXr)n|Q3n5_t(x;*UAUGkYj20FDRcDZ(+g@^?a?nIE7o?Yu6^%Y3$ z1+37y{?=Fu>|`Fxr%@##*@&t2MT~Ww-?g&0=yl9I{Dt2r75kgx@KH(ZsLg{nKpgnk z+omtG2PJ)oQl^2^t67Wsi6AXS>U*!@oW5CA$x>_X!Z*SWH}Aw;8KaI<56c z-J}4u-%H9<0jXW61b1OeLOFM(%^5eqv6>mR?cm3C-Rjbj0Q`u*#5R*|`Rjsgm`sKl< z9MSJU_5$!l6723o+BU{!5vpC@@VK4%C@Boh!&bGsFuDy_anOay9xlWE86}BRS*U4z zJ;Rf#K_U(6IzdW1Fl$7B=28Coi zRG!5KOP&H0epn?;)yd;@t=x{i9vOs=5uoiZMVBV|GMK3xNxiUgYYXUwB|f1_kIbFz z=$;CsUjaUP(_t7xymd5WRpMk4H~BI*w&00Ewq@RmT(Hq~VU_DST$V6mAlpj9V?XX# z@C{Y;WD3+Rs-$XnG0&zjGrPmX;A$@PhTA5vD2f2#vdSW)*y%;GD0aZ?(285Wvqn&S z+UvxKU&e$6yy*Fa?0^NT?tMrnYKn)g&bA~8%wG|nm%HOvQVc(PHlqoo%&kzU91l>htZ^>4R}dT zG&+nevrO}bjKP#-r0_LQ>d@Xmu%t#}N}VN*?;`SRx-5>f4tSa`!$U{iJ4WqVv?a4G zB?#--#tr^Uvr}9@UHc#axz&=Ny(+jZl}M`Qojr0-wF!MwrNW`ziL-kOfRZmZi1GN2-C`YL2HtNeajU>sUHCQ$ zF{k*BpI=wE8rt-c23jM|3E&1~NNZhNWA+_VHBm)GmfP|!vz(5awL-|xnz&nt?^V4d zRr3}L^dk+didSL_?5m$wNJodkxtc3Mj_^@K7o-z$Qfd)y)xZ$N2Z@|`=*9A)xzp4h^SDWTV7oW-w6+92Fb8lTMz?wECh@;WQVg)^$(c0!1SIJsX6vGH zwX?%qhImdGjE)X#sb|F=;QK(G`JktYBcQbp5-vV42-T>gv zobL`e!NAnS6W}K>s+@2i0Va02=r)Q_KiuK!F1LotdKQ z46wm$-uWj@;=e;@Ps}k4*HH_|%6}S}$rc-!@no_A1V;-+pFkS^>Ps+WULG&0#ZlOo zfFKM69)$wDAvH?VKaK>GhA}*-MSw|lq60XaS2v-jqf?->Li6eKH0&mb{wjdo$H44h z3jb*!kX3}2(rtcoK=C_#aJD8qZs)v1-(Mf{4XXJ{8x;*??VYSA|1{K)sh-m~wZj6G zI-dd@v}z{$N9JiX{biL?V8|*njD1oHT8dm`JFQ0i(?IA<0R)KCdp6itk^u4&;TO_W z^gOCru|ub01O5w4K0;YrtsRWaGo_23C2s_;Wp2&Ys8`(q1N~oa>tU3X-0|znJ0IsRuq#RznpzM&1n|Fb zuDId<+5)99|K4=}zdt?u|7QpNQ~>{+Y$)V7KH(lHR7#zNFW?Oxwj4ltup@4^2zJjV&-1nfha4}ukwUV0u83RNy! zhfe%>aR{}>N#<-Et_DJ&7weyeb`t`FFCAv|q`-9NI5{VA6#7R}$3=S1OfpjoYgi@2 z>E?*LC89}6T+ggSDj==`g=9IV^pN=N0ph6Jd9BzPYxVE-VDDflo`-dZRHK&}8f$?= zb7$^QU)zG~zO2m2UD~~!Pl7k0%HVC_eLB|n;JUBgi}T(@+p(=E*enEqUvM4HP<%J3 z*r(fw+{PhQ2=rUvY)1u&;n9>fNp4(38L}5%tS}{9`*2czfCBwyq{JK6K9!yusr?;4 zv5n2a0d{|YBQ)qD6Rvr*}pwq`nv!GCk_FUqU0 z0BN)M44dC>?SQm&w&N}D3Wb=wM&U*Dt=OEr5ohWxSFz|%A}2*3>DBmQyOpEB6)~R^ zzL;hO?CqRVxC$!We>VOyrjiWyvV>QlTg7vKRnT2BuUR-KDE)zGve18MF_c0E)tdE= zDgR)`1S(8#>>*mGPtNXfvQ4!m(Ba(DlKHMab0S-vjoDVl_uepiQa`b(APUyT9)$(A zGteeR5s}c2Ekts$M=~Vni6~G6;)=)lI^%mUI!ZGhajGzizC?3LuCADx%%GlAf}!8w zF0W-Sv}&HCtsGF_0WmBPdjPi&b?jI`enb1i0T0plXP_(P2MGB!{O{~X3;EK`{DvJ& zEpj&OmRRL?c{!Dj6Y8O*AfjU-^A!EL>`R67Q>Bg3VgB#)X>sBoW6ph>3(uk+>@lmz@KB(e?@@LrwGVR5_2A9v>^urD)bFsTohnR;0 z`LaK`oYeWQR&1tmJ317^3mAOgx-*>;RD5!!D^J!TV3DZCFj)n2`RXY(nSNuy$C#ZM zmD}R`oT>pvz!dT@oxL>+Z5jg$jW_%<^(mH24ZELNOsQvYWFx&%4VUR;<%oI;1~7jK z$Bnb%@xLOSQAYS}pG3hEf=1_~a&FzoM_-x81115Gvm_XLk(5!B8R%tGonHi0{IBo{ zoeYkp>{?{Go?9995qrUXR`nO`R}PUunlHuk|DJ7FSWsi9U zC8Z)x7nN6r(5eHx&TedVoSv-r5m>#AKAEO(kX4uYBAJqnfHJNlhcw~)DPv&J(mSX} zRRStaw?6erZU-&%{!{T`+-PO8`*dL1^zqYpZuaT%34{~t<6>F4r3i^2H&+ip zg)@KoIcq~PE}*(lrPSYP&8Yz{>V|`mB+-$&mt3gwsdYQ(r#Ox89YYr-IrQZ4Ej=-I zN9dO_H8N)pwq-QYZq5|bwRr zaLi$seP4xk$j6>sJ}@heuOc{VpMU#n?5D-*1^rVskkDa3!W=HOLU?L8GofDaQFgv5 z3$rT#wKmnHZqi?7n~1Fj+21=I*zpN<#%)`A#($0cr&NNAaCdZhSYmq|;?ROwkV&AK zd=r_7oFo$(z#IXK#wksA2P~44=I?h`YH`yogYIUGCK5Y7O0^bXyYjja{H#w1X>!I; zc#3`j^P*>%!q4uF*)tzVRoIViB&Q~}Ai1J`D*E}F`$Xn9goAs-v1Y^T6{eKk88UXZ z4&MX~PlCuMup!C(?PeQ7z6OhKe40UbeREHp>KLpR37_5Bm2he^8%=5eoNn#iwNI)a zz6Xf;aagI)%qD+!&}`I6F^j}Lvf0fyJvk$~N6<5@F?+HK>=Z+jDOJPz!1NEC^%Skb zm{mBsl(e#zR`6bLq6H+mkPR+S9$lVKRy>DwOs>tes#K*?WKxFhx-r#5BX2{p%>s3# zw{q&?q?*V z0c9dw1lf--PI`ebnsMtn8esh0dR3#r_*wVZ=6BHKlgvXP*PR-ml#(BuNiT;-Sf{mo z#9ep}BJG$c#B>h~?xrLUyqKNmdwDzWg6ZX;aaJhMHu!2?j96wORiDC?Vihud{pU8K z5|&jER4Q+jur%9RU(}2Ohup9QAsDAg;#Gbyw=kh(VE6lNc|TKHI{zGDp6q zq+G?wiYw5p)ib;tBZ%P64KSE_;qJrlHmBqF9XkMI$o|81){*9Z{#RUAI+;W|H$#SA zG6O;r5x0Ed8p9{q4tBScn5|2-JJA7(%kWR9Lh~L?^Ocq@Kt;-IVFvWTFH=ij>?nf! z?%St#f1vF1QO{afRJ=ZOpivvF!F<9h(bXgoZlyK_9j>4jW(#r7jOryHYWn&(=L8x# zv}rm;Y$3yMv#9BhNNqnsiug2(IL3K)4*{n~X6{Bb^B4r1_;C%TiqkUZ-rhGjAh=7; zM(B|IqFl?Z{nFx^GdYfR?Z$_=eep=wQ}la!Wpt2+bwJt#pr~wecsk+@J;6ltT*4Or z$+lSNNI3omM3iYtT5M51+%l^Sx-JvbFkbnff)_?l49+9MCD|i*E+#KT$~K_5amZa= zUa#h!TKd?p7&ES-fzjGb_QA!klczHaUATTcXU0N8?Txc97)y}vN@^#>+%VD^-v-(n zfKKL~*GS=j?$yRozj%*Nug*-$SadH9Rf~)|P)luHQ}yH3 zie1?Dsae!Z_I|QJiEB{zdnKhq@ixKOOR>#Cu$eksL^c@Rx3mD-Ry!fCidu{fqO96H zaN^OCmT-{u_0R4p$AOTe$TQ@PSf8GQlS`i`#p|Z=DFrccB|b+%X4b5;wfORZwaHy_ zvzI~D!(bR@34Ra)UP=<$%}CBx&{Qe3pjEXRfVhf!38Bal0$bKBBI*d z=C9~jJUVDDN#WO9i`buAmlc1ElcyrV73G+0*JI0e+}{|UZ=X2v!2nq?>`|KFEI)D# zC7hi^JU3cxxXj1|5UzXxBC*GhbeIfg;>gNqpE;SLokv_w4+g&%(7?-suw~T%@x~wS zRQ+$4gjAw2iz$=dIous?X{47{g--I|R4@oS{XzJaYur!z+zpknqK32|1AkU>&@A4} z$JFx%StxlwHmhB>ta4xGGlM5j0xsz3Bd@S00B()+bo!NR1g6@7(uA8^(ceW=A(pRU z8!BzLdtn6px5=WwT&7e>Y1PwJH@pXGEajG~+dx!=uF1z+t{!O<-OB7k2LaO(Y6lx2 z@nk*#-T(krI(c}J$qCeoObq{P<*4cJm7|`&RgO~Gd!SoKHzB(?eA$4z9hG`X0Pq3@ zBIPkMic0K9K=X_;1NaCGqfZxH!rt$`g!;}xvMfMjZF7>X=+8~0RimdebraN3K8-b% zYw0jtuow`IJAg*Q49JgC@JDog-Knn}7DiP&)xu;v+b?W$#~CM{$=&Mobu+$A_31v2 z9ES>ZV%!;OJCOJ2WgwyoqMI@Tms2LQQVh`wfGR@mz;i9X1PLIXYSw!h(hCs{IBto0 z6r%#^iFO~KtXfSYEP?<yH_IM6 z9r|{!9R|H!N;K%PD2d%|yriHSozF*k@@Bu;CHQ;HH~&_X{GSeVm#)Ma`eN;B2|QD18RrgO2pkFNZFA3jXwv- z57WW?4Os_-8YziNG_|)t8<6MYN{=lHb9FowL{D6c_#l|$$#8V@hYAhUKXIv9`*ndo zFn^$ZOFJ-dwat1J>Vg?@YZL>r3?;;s86D+fH=+B2qupaPZBAO+nZ~eDydv^IX%{)z zka0=zc;F@3(bIxH%L6G-ucaiqHsqG049Ml%Flh50<}yV9x&qKge1JM$lbc~xMoR(M$8-ic|2o}pCZb9Ke5-2T zB`{P&^on=cv--zd@R#f}WTm{>F4wtxLknu-<43Gsaem-J8nlbrNnr^yIrkfh!Bn+6RDNVxC^f+1UCJm)Yjq@zxmnLh`$6Ap zXsTXZXAWa5v2okc6DT*%ockGgH#e`z4`sHf)xx`9D4_OA=|g;0?Fhwe&53koS;Kx- z_|=2v0aV);%uGYP=2Ba(bS}2fI^WUB^MSi$_LsFtm1kVm*x_>`XJ(e=yr3@v@?Xn8 zsvMbJ^&RoEJpv17+7G`Fdb1qqwXi-x7nCjwM#M^ACf4u8$&n|^cPH^)PsCLmxWM`x zRH?+Bil`mhkZR6l<;gF^%~dvB4Ilb+!#&DEs*kYPJ9qF$m$S>=c)Ll%efMglxSU|$ zn9{v1>t9FTcDyZJWbfXUDr43)Y3UZxcP+26a;1}<=l z0)fB1MS8*qS+D6lAAfV)XP)mbu<|-Bqe%bWjMAoBf6N|fd63iK#xeRTlXJBTj{Q~J zbZixt4x^@`f+1Vn5qEMMGLx*Qd@H!{F>wF zNr&U8I&)?f&cRM_cTri#mf-#p}fmMRk1u;nku9o#(fWy+L2iM7ad0y7YO=h>QV6P~R?8 zQmy;*K$zI9iynXORC^G%li?xM!qBzV@*mjfrAyx6bE_)}p?)8kAf!yo*Y#DHF) zLgT#CQW&j;=+xaCzrdD1^5#C3Wh$bY6v1;3Cy*)Z7CQ~$YV|)jhWP%Dj?IWdYT{VM zmikVj9R*NE`N0H}k2Nl(U7UXbG=kSW(UDrHQ%DeA^HNC|SzMKvk()MUB$a@vumv$CintsrCNP>O(S8Z>sHjg_v#~?tg$bh%@C_E6^GpQ~%96l;J zuwGf*dXTNSMqK_0vZgE5@y}o)F^w6g%Lg@yuZ?(B2y*p)kh1OMVeYYBFPj4M9oj)B zWv-YQ0~R3@P)9*#m|J{Z|5RHZyTD&1;UHE5PgHjYEx#RrUD6l0CGw|B%(qas1qDK! zozt;*$69_}9b0f{Gh2^ak}n6oJUgFXIX6rpOWg%$x^+unc212rC02B4es6z(=S{}x zAlY|tNu5%R^-fPSz8KRyaq`yr8s&;sXcWSH*H0NlT-$mmkR@ox6<~LV0@>jXtj`4Y zjzGw}F?OIVYf~Ohj{?BF4=5%OWq`@Nb3m2sH^(NNE<5edluqek?$&`xv2M_<6$)dH zYU8={3q$*!xmG^4^#i&QXUNoFM)gA305zTiIxE`7Mgc`Vpc%b>gKnsf*`WkMXC#JE zi$kllw!bbJamO#lqoz)x|MSKisAe@Z^EU^eQG%MrJfHfH8*ZZ-wY~y?eLi6W{ntgs zW2hb(fSI8WCJSc+tnl9)JDN=3!dFhi>4c+?lJ5B-gCnWd(xH}MT9#|;(%>lu}anS#~6!OVgWJSuPG8I`$QUQ5daKLfL?)raR$x+4_~Le3&ey!@YCGPD;I!|{F{TB#6C!u z{Oi_+f1e8^o%Q#LJHt@?kLfoDpw%GwHb6K^K-Gij;lPm%U@JSyOn#Ei-zOgZ_t{{P z|GMa}sbrtr2KohlbDRdwkAKhqzvuto>;K>D|KI!nzxV(DKcD}{*?i1>WN3_UrINCH z7cO12+Zv@nnF>+03$01*1^_by*gGy@(5-46)WLIWa&DjzX)R5WOu z_d%sf8*kO-xX4||Vm6S$h$}%9=_l_DX$_W^?X_QB&`k}Ji;WtsMX?WozAnF;JUnnd z@1k1`RjcSIs*06 zpvOFY3u|lCk!kY=^2X4vBnSBuFL8eL%-4b|U(I>X9Bl#;*?diEC#e%BNhf$=jii2- z2}kFxA#xwy9^c)Ct%4iFNl7uaJfByoenr*YUv&U=RKa!bYzyPKn^CU6Zs*9}GBR$P zUu0?|wV{DN8$7@WBY)X~MuJO0EvU#*o&Z#{SM57utcC&fwGG|=_K;x^3#hfcFl)kZ z{lX*$Y_x(UfW`;CPoJ>`mTe~zbodq2!Yy>aRM7IzJd{THiKX6Y>Ym)*h5&I}D8eMr z-4Sa!-ufLhLXL~w65AnA=grjS^xMI+DrjbyAo3IQFg+PaZp8FjOJ-`0IvsLfvRwb+ z$UbHF(i90tj!kXaH~}s?P!_+S?Lkee@@Vo=<4~h<))iJ70_r}q@nTmmy*3zPWyEy! z-dMF;weu#Wq9b;5ZEFCVod00sVVrZtTBS=E$;40cExU z9K`y+*n9JEDC53wm^O)w$`WEKN`(|z!z3ZZgphSAWSNjH+n6c)K9*3#gk(3#zD{BswZw$l5AbWJTdunvi+w0nd=e6kRYMuy{SwSc+~x!=a@T|vPvek*M=aLK$7K!)@i`HyrB+uwyh*csP>DxGJ3yWY`c9A z$j_FopOsK+$CGI(e%FK2y?hF!(GMp>HK`9_6(31nRyF9&YoXw%4zX>? zy)y$6n+?}={cZ$|w2NW(W|L(Ov&gk-$t@`M^0dp`liT7ppnV%-AT*h>Nxe)VfRECi zRhACnytoHD1kt3mh)1r0t45(C@8d+=&&!_`ydM?5hGX_s-n0)3-PFqWD@Lwsc2?vE9tmu zG*Qi+8WJ|kr@UHDub?U90v6Aw8*1k;a!`w^j=PAl7eZMBHA{o9KUvg#O3c6Bq^%6< z!3Dv}KrNUM{yQlyL24(jw9W8Hf0914ABQ%US;dhEFd=0agZ_`3i~ztjY3|L6716mA zbNz7s4UnY$tL^G+>+uU9Upi1w{O29vVpaAe#Z_TRKQ+KLjzW}MPbCw)(|IS28d>&V z16wd5!ErxIAO$k+SmYet|(B?RvJU4$8-YU+rYq+0%$3(&AnM z#bI_+fI!i?H}J%wXW)3OG`WNieVRWe!fznh)2{1G7so-eQg8VSg>Xz0}~p~`!k zrue)gKP6*9G67?=xHPXC3{lyt=}-5Fqv@W%?Z_pU6%tQi{=-ezF#?a=B~f z0MK%14skM3VVE32--GmE%;GR~z( zzgY%Es+35oJx!XVJ5=)~q#)n2z0v5bTHssX4`CbdMX{;z(dm!GGa4`{jqq!z#_;5AHTeG?5iO%Whc#ZHr zH&mJ>)J}Jcuoi(F$Xuw8qhyfb2J|eX@X|TB(4A;p$$Cj#s%!MG&;H$o$Y)|BxassC z*l^F+em;;x_TfG7dSuZufwys8n@>BdM|kwJ+na77Uyv_lpTSD<^2V6Bz{CK~K}z}F zry=w5D1VHEZBg$CwU@G!`j1>$*>ON!DS5bp+EGUySfSd#O#M|4y{nkoa=VSf8v?#* zVi&*!*)}@HaGx+me}LLR&p(|M#+N})FT1TA(z=d+IWPVzyCj}z`iO6wx$GaN72$t? z?X(hLqz|I>8pJpWG`%V1u1N0Lj`eDfUj@=^`bN0&w!7XhV@=|3YUX&4?`@Ivg2l`z zYEZ_g=j^=0rJc1t4C-gy3zu=xS8kA%X&!g1JYzkciyhb;m1byZGa!eGW&pAJ8WNYIX_sLi_f5KzAobb* z_j&U{K`}fq0`wkS5u*V4u*-1A@AX{2?36BBV!& z*%t+Hto!2uxT#SUGM7Gv`T-cozPbb0XKo&FxhwR#So|dVz8iCxk;&!>4jm{QREYqZ8ET~va+AW=q>jK+&r;khy=G506befm z&w6NCuhfx!bO5l7uA@o%6e46opO&M2mrW-I(Xvn9d+lMh3$)9iQt`+p<7>ztzMgep zP%$-8O z0&dBEo9}5CY`9$JVZHcwMVG$QKd(y^_2}CsfTKk48!oL7{xNTsMn!bE z>jU1fPywfTVbYsJUCmjQk3OQ4Nm^*eeaEFn?3+i6ZROouDD`iDnDWr{epX{@0t*S)Zfj>hD#Le44;gX+5Jl_`}&9~)pFIj8$jH92bM*# zCo2Rk?wLvh4b>@-7Ajm!jSD~7tFKTGB!_pTfY@^#e1oWWyuY7mEap~pc>B*I`Fm{i zenK4~Vi_SybTVw?pJ)jE^5A7vl?c-<#UoIWcYVJ2h41h4^)`)r*L?5T6Ya05UTb3O z7l{FUnVJQyXObg|c23`Zs40s0^!%Q&a}d_Dp`6-D+WuiC*Ni=FMse4O5_xm<||yDb}`dl)0qUZylKsXUhiS@S#CTTM)Do zboSuoDEfIQsybw9(tm0Rs5kr;=S5k`Y)=`Tq@A_Uvz^hn{l4C9Y*hB^ji(+mYg6ZC zwte!yemsgus_tzdD{}{Wo6{m7V`nWfE>q<<2D0 zAHE{UeGcfr%?&xLm}=A7SLQ%4GZN~AJ+>I0wDh^CL?1H4h=p3Kni_-$S^UJGoKjaLVspmy>1({9dV(uq9o8 zKdL0fbr%z!#OpmFSWKTHUBd{{3l>hxf=b8KPtNkub&+9=Rx@UJn)y>C0w?NH@gUz1 zZ^tv!P?TsCzwUxlqvVHNS}@(lN|wt>#(QZly*OhTaL8LAzfmHBkjS$A)61M-qss9Dyi3Oog9n0p?>jKy zEw$u3&~j(4sER+a5`_1e?ikn;NPT%X8?fBlU4hD&&uGMYKzXO0&I+6YcQfzr=3B80 zC2Nr-vR9)U<)Tn^Ur=W4l34z4!FusJWC(1QJyW4VE363suRHGg5?0A2Rn5mDDxZ5^ zN&D?}reYw918&z^bBV;Lb^XdSF~6?Rrm_{Xt5V)=^z-J;UJq2F(M6&tZw3_k$?1e`qw%RoO9!hwb>^}%{*FfitTpOjA&>& z8AWQJgfJ<&)igu{_&j|swOB%BlWIOTJ_w8<{ksQF(_rpiqU0H~l`yxsgv8z7UpyND z@mrDf9B6%x+4dl27HP8I4w#uEy zh8Q3^$1oo!F91un5X(mAa4f~EC)U%Y%&FI^>$L8SB<0$x@8j})2!-#4ee^8(W%?4b zd^f2Cu%+c5jy#tcwvQTgKg&t4+ zx@oo}2x?A|3l&d*_7^2hp4={DkC)O~aq}CQur}wM&+O;#Yu>tB`3A^m(x;tpiAv{H z%LYptm9RgT?^$0jw@#?ZO`TedKUjs; zmM^ADQjDfyf`XG*2aT6Q)#?WasX@M{JPN*4+Y|?zgS@(VUn00r(-|XE&j?CV$iWz# zgR=Yb+)RT3`9wV#vR6W`FRwAb^ZJBS+=7(wab~3%rNu7nuz<_h-RbhO1=q}@>60E) zG`ISKCdMgJZ=jP(GjaS%dFbF3v;p8oR$avH>pH?_C>NS1Au7QEn+62Fa8Pji$r4i0 z$8G-Kvipq!k#`~Y&4E)w`$VtkpQB8?7B(0+6DO~-8&Tm_E;F#8bSCnfFP<)(4lh=g zzv&#r)H59oex}H5`WV}F^CjFW^p3*g&7i5Riw=+t2qOgEpPv>ST>Q~NFfrceb$e5> zR-4bJi5TnM+B8;Ea2$^GIMI%*8QRY*k{T08{{UadjEMx2plU#!aHyia+sgwmd`zMj zB4%3%-gWU8*6vQYezN=9FaFY3c3QOA%){3>l4WfDMbA*@E1P&9k@T;XX*ScD?oEM4 z7pSQ+2LK*({1}P3dT^(gpMaHZaQN^JQ}E>MR_9tVJzi=#dYIiLrm)`!Qm9l>2TOO6)$=iXBRT57G zkh^U&-iiC4@4ilFOIg}0tDIM$N5kE=?UBC7PE=`S7)@)YigB6Jc8)u(TKruEM$We* zbfZcvfJMI>S?V2l?;DvXQYAF^V*I4%DNW@V&HHyF@F6FQX3yMz$L-DZq`cd>G@-g* zBOvG%y2&LB!iK!U_|*W^c0mcv^9^lqv6a+sG;jKYN)NW>+vbIv7uSmQ7%F4tKdITN0 zrJ-7^uD|Ur7?UzTzXUj)%`1LrHbkhA%L`i&{uBlNq5Pru`#w}SjmRMHUQvyzZm~uV6BXtP56=b>Vz!lzB|6i z8-e@2-&kFaS`WW-&$egHXV9|^f{TE2^L(-?ey4f&809<3WWMXsE+UkMRH6sQBX}?c7~+5xvkos6g6>7tytL+ zI^IsysGjkMJR&DW@+ICOUrN(}%x8r27J$_HOvVl+8-eUH@KEM(S=vQaPmVW`)JuOU z%VA6Kq3nR`Gq-+b#71#C)btw|1c;7mKXy&6rbLri_R2;YKn_O%$TP*0?R>-ZP2sVm zH_E~<+{{Pn+gkv60DQcPAAFkf>PvT1oNKM%-8(XSfb`@gHb_M{tr%PO_$8{Y$yCGl z4F`)O5N}wDZi2Bts#J82;IUb1Gu2bvvlFg|-xx}*jrppdVS3X@0@Mx*Z#YiLo7v>B zpjnd}Hg7m(O9?A5CkcgV+u#^2&fszf7dQ^83#x&{tG}S)&S&0+|Fu?4ZF&ju8{9evdfrIiG8cS2=vgp@7s! z1@JAg<>k8=b5=k8z3H|WjqBJQ45;wP5gnFU20=}S$|NTGPBQKtyR02#cfgEl*J{N| z>Wy|f+uoA&VuPz^_;*)SEGh`CszZbw_l7qb7ugaEZac8e-r!4>2beo&RKAw%b&YKa z`0=j$^a|0l7^l?9361d%6~8VIedY^zWXIN zZ1|!=o8nbosocgXnQRuh3Q17hr#COOe4}8X=H(W%3pPr#HE~PV`&-pKIRA~oNKh;W+{$;n9?%J+>Al%lV0<6NIluQZ5$+0mvL;<4hh^?v%h80tt z4&M!A@QF@iF^hkgUI2L&gDCRPJrFBkLF)2u*1k?UMQ%Z08X3^Og3eB)UqHtw-= z_;LQ7b0&*dn)bDOaY25P$TGq1i_xBQb4xr-i@0&SQjKgluAjBPh_g5Vj@EPR#`3a( z;wSx#D@E7sk!FyUGL?yL1PrpI*cHt6j`d?HPSG~f;2dpo$p zX0n|$@!@Ec%?CE2Mf?cPNv3}q8)(Ib8-BGO-!n47BrM*(;FOh61$a`wqbILP*11#F z9S(Qmk}Ja(q%gXD4|xX%2g+BbD~ zv==|0W$KjY-iEgI5_Fm{r<{ZaK((8J;@7nj~lom06=iJb8YyMmYLj7)-Detr-3UXm2dSX#GoQTLeSAthDL3K|Q7 z?^nw4o;%BRgyo*re3&3JHuQatfr)PgN_!w%jZO7F>-{rU;UWfo9)i%8luwF!ajfYJ zZufJ4EU_xZv#7$#05(1#<$6vkcxY*2ta`KjwbU9a@c56VwYma*suH8S2GUe;a@Wb| zcu(cEE`^SY^>Ub9kG5}z!C-}=8c&NY_EL#Sm6b3|Wcd&JmR1dE&@@j!Jgy;r23psS`B}=;tF1UZq0$3i{ZjeX_5+@j>Rx0( z1@vF97{S_ih-1P_u)q|_OkKOt2yS~VrT#ko=V!PNqZ4!B)K|nSCR-R4 zRI%;55kNj9rgsc*X8L^OU;F_jpl6f!x2$sN5qlRnj6Yur27v>*V0390Ov9+ z{{LXe#o6bA0SezMR& zpl9s7^CWuja`X8ifc0HZKm!=N*l?_W8ac=@9EPt40swuY%y^&{XR#iDaXriWm{hCY*b%Xy%)4r{-B9E}HbAT-Z%O?3c3*vsV9VP#|*-bs2{uRHewZ)lPA}s9CV@tN zerxe}W4`US1Brz=e-^M#9w$aW%(k0q|XlEq_GmVrB&WnEemtbWAO$@kAT!* zr)WqLn?2$CY7GLlwD0qqg|vRMzH>zv+0`_L7Ec#~YL;xPT(MEw8Bt6bQPdgCIsnW3 zjc+VR*`Y276ljoIS84s$oxO_?&>-iY;rJ^@fG)!hF^FbC8BVeeu7ga4ONuA>m6VQ~+0R!<2HRqi@r@s@K@WC9L? zfKz?#2`ZoWCi(jTQQ##*v~-VsEl2m6&6HiqaZQ7vLY1?Y+2J?h*CsXY>I#XG@5Spu z4PoFGg<(e-?i*&F)pQo3T&O&IGpqO;1$Vej@xWj~MLl!?M)7``0i>lRGxR_VX-%4v z^kaY*0lokpJs{0~wbz0fKU$979)f8={3H5>rCyr(g&2fP*)f zMCw*!1WVFBfoU&svi4c@7i}oW^j`<5e{b&JC-dKz%D-=fe~&8vzZx^$Rc_I%lG_@D zYDtq=BGX9ab2U%}$}^B7?&y#&ev!_FDty|zNaHikoNGaa zGz5a$w&u0#Wkjc}x1+|FwnNfPlule3jHF5ce!&w+(XX^8^bF)>gcTbcYsFLkXy#%` zOfA$%UHJ(=fJSf6?m&)8xUgFZ@kdW=bPxDvb+t87X02nQVe&Wf>HQI&uGWkmZL3=Q`vyB&jGSe;|7r3Cf-e4fdc`kvb1(eofrCu|ygVxuAl?W7 z${iq){maxp&!XSU0nQFfn6tH*^=JP?G^z|BKFPzkYa)_c&bRLYs3}9exv4JT<|=!` z0qWzfWF!ubi9}wDUQafV9q9!mzb&sT6j8iMfIUUm6MVVlOKaC$i#x>=fM$w$r_lE? ztNaCg8jRPtpk7PX-`O;7n&dj)q#vbx_P)NQS1|4L4C?%3F@{LJTK%!s)}%!F>AO+< zl3f#Y)CZbaKvPYP!P$&6NK|syj#FiV{_9$|r#F>tr)cPT6-Dx0-qcDB zp?UDUIgYH-kP0;Xt(Fho0*g&%u7SeacM4x~nE(pO5rC<Nqh|zx}723ai(b;In%#)}HQT7p5-AN2X*eRiutjCS%nP>Fp~P%!vf4R-NL= zT>-0pQi3XPO&07YPND)L*ce*mJ!Eh5IuOUb=h8gEWN&ut;q3l^4Pf&4?(AKR8D)yq z+i$hH(6QU^J$-R?kc^-6tF!y)$1WV<+XdZLkc>fIVziKqTI=l2SvrjY-;rmp^(cN( z=WI*D5Pld{mQ>{|BW5Wj+fhDMg{tF>IMBoOa|ks1DM|TT<&ZSP6hbS1*&f>o(Al*y z)Q)Z0>Uf>6h;KxS4uhZKZ@T=R>(E@fO&41QD!;?bWh<^41aWS>lH`h6gmK0{O;XnN zUoeDfO7Bf&6>DzHNb}w7gQe1s)SEQ!O&Vhs(!Znb!+#Fz+|r+{@71oP-o&sUdsd7x zgjv3Co^(@*Y^M#5OTNJ-NI z04v!pXdm$KFBifm7c9w&D4&P4R+J*(I{^s=1vf~nrl?F5uszVEof2^|1URF%$eWDG zNq)38CHzELAdBGyQfUt$U#LpvSlv=#1>A^i>&-oU)88)& zAc_L z@RUX41nYuhZV&CNG~sL>?|slXY-E5KthB|QOd{1d<3p=+Ap=HQqRim5ATKcI#Vcyi zAy<_&=MU6YY+nNQ!Q0OD1=Dg-Q;jIjoplY(ai`?7Ad9x2ch3zCbh;wj&#~55prjAX zueJJ#Kd{*wr8g8Cwjuc|$;N@M*?N!>=D_6uQ0`7~y6D-Kcu5T=hUXPzyYy> zE>=V4*>blf5MH56iPWg=lX2MT8XgBe$MyFSML+&#Yz59+e=jN;M0 zu?rb!xOy?KaUE7@>%AM6nJ;m8i1}rq6|ch_k;q%IystR`V@6&AaAL>M`$Jz^%SU8C z`FCfGe_Iqe#812pVGg~BpHnVx9)DCSYNUYehX&#}Uyx%$Bo^iC%Fny5R#n_(?lb8k zYna;YhDpMPc`{;A1^z^ zi<3j=S*}ukg$ila40XQY%2S#c37`)}Y@XP5=5?HlO7_%ita0AKHb;v=$y4jn_UYzicf*ZA$x&k?=2=_b>AvU2b_|6ieUviwM5Vw4%lsC%*XR^eZM^G{e<|{grxSL zmT!qAR;zMBN(xFEe6Jqot(z(Cj6_^@3^plGtd-?6uo8*dLLFnWc)m9PhNzOy_+T6n3r(5|4 zbFT2GdO_YiC2H+(AJmfB_BzPmpoZ&~6rm}d?TeN6Ly(lLyR#>03lQdPV|5bEeD$x5EfCAOavu`hj8%y5;e)_l9a&{v#|)4nxcrzJIw5B--AH zN+GXH&MpEZL`x{gQ^zKq16DS0;oKPUnqh%M45K+%Mf>_`pPOI*FtOK3y3T*O zV429Y@%wU|%9d-4FSy7S*){e3bY;}esFEl7xYt;3f?!F=Dub42t-@9D_lj79bfvD}EuBv8~ESPx%To-Qi*aj*)~q`Gn(QfSC{ zd{_@%RPm!oQ*9{1Gf(1{eVd|%4HAOGO1r9}TWC!(`hYLTTC1%mN9e^J8HyM65u=+I zb=*0NEE!@e`-Qg$LEuH=O;q)wxoWhfMeg5?sLnyYjiTC9NNE?z>#f5?pTsq!hA& z>FavPQ~ahpIGRZ%;#s`x(htU+!~aHl#Q$Tu|F0PI|MPqQ^!@8NRt$MDpbP`1IBqtZ z{dB3UVN0(!3gig-n-tbDC4g+jao;J@9c-3`5t3R+)S1XAj_#{K9DGdF%euOD(X8;& zyqXns6%ouS&Hhna{w07ICgs!jHGyn5DY#S{>ptPfmP(d0!&a4+pc<=1WmDnKKz6C8vibxaANmMnl@B2f$l&HwswCL1Fx z+2@7cqw=4k7EqypYqsVmJi}jY14WCY0I1l1+17hhvoOL3fXe`M_o~F-By!~XGhh98 z%bJJTp$t0;3NRM`_on}S!vB4l{l~rf@0;u2L(0Dg(*Jv^`uFtlzj@Mp1hH{JLv{cU z{JB=#AEpaMcOj#zvmR^5dm`ts?u++EXb*P0G#TVu-(4>X9a#eI4YI}R7A-&3fL5gW}U z22Xmv-0sjp6uuR)$y06VY(yV+p+LgB03wtXA1V6t_df9?!kO{^U^&-&DqqX24{N<* zgYJc!q-iM(B{=N@egj5){}7#Al0TL{RNn4AYr$yl`&y+I{*W29`59!rYGE3qlFk!E z=C;h@8b1&qzTPS@xRX`;M#Y1InK@Um>T=BPw(34;=%$}Z^3Wh)@z&F>ppZMP^G$yU zPjdE28ESvA@?ynB6tx8tK|PqK$P<^^qTW2*Y!F<+j4STk{x)%T5motfbaNf{qS_q( zM)+-{l?e&c-27+&Qg&6h)Pj+!12P++6wELaH(4ev#hl&;vjHO7TxE0-%ABIswAEzW zw_rs#@mku0c`YD?+pE`X&9Zf)LsBuR(0+ee{+>Xjz#Mb-x`#YbwP?m|_++>%64cIS1r!$DiPXXRmwyRH+`sFa(3%l(K2&Gev(xY+4FkAe*4P?OHjuHvHS-b+N;?Wj4cQ~ZVZVO+qkQMQ z01vX`s9R*D+$vw9W16H(t=fvb9UCC*&X`Bx9s@Mn1+>RMFSY->&&9T_0ZJ`#X`ye& z^7YL}LdeJ0r!oYv|*22f@q*fu>pW=B6fzVW zJENR_^HUv=`V0C&SV3^-s(hUB6Z;}h=*F}@EPmdG7h7`Z>*)IxpL6^ncEbG;yX#Z4 zDhDU|D*8(Mdgs`0dT@+Ux=133J?p@v5hqlJTHEC7^x!?R38u_k#Fip>bHC;hBL&J7 zehcip5ChiU1QYEEmzgL*bFE?F_{eK3`R|y ztzLo@p&r<{=@n#2d`$r|f+$n5h0f~wgblz6O#-usiX$WX4-Cps6Y@EPpM5>?d4wm6 zGLr}cQr#qoT%7UwSM4MA90v8?WX?$4R-IYdZ;_dO&4W?{4k=f^tbNe5W+VV>+@^Tt zI2tq@kcfayKA-0EQIqfZMbS0W*-EPJW{``>5C8Q0bftFZK)}{g06QY}3Rcm4-~SCt zc!!q5z5b1yY)~}+g?FSE?p8Y+`z(SsLCKreQsTvGvBM)B^sH5?j6|!Ib&*LXDF`S@ zrZU=tPl6no8}b76=zBujy*ZBE-2)!GUEhZNCV*)s-8icv2!K3uu)3InU6}Otjan2l zrD3!2a85C?q{G0i&OKH$W49P)Fj!nx+bO;=*cC4wa3dI*YH%ixh0pOLZBwIMoBPT1 z`ZZiDxR7w7+HOYhrl3LVe(N=Pr9Ba$stqy623E=?vII>IMUT$yOuY{9;%=SaGtkq* z*3HIURgafWbXLXwEVZb3>R2r49J+&L2Qo}={mi5!#am)%XLj_}eM0*I?%2J5jA_6l zI1x;GyeVi=SR)r$Br?vo&FHIIWn|-X8Rbj3Z*t9kf^v=V-dpzj7AxLc{Sv2xvnkMlOTJ472jJfyncYoSvkHd2iO|VgG_V++ zn^IfWzLr_H-Os-y$R`ln%tFRS)oD_>5VC+sMYd+D+UKLW{kGnzw~7Zq254_)G53Wx zmpx`B`Z3lvHOTZN$MGQTdomP-{{rz~r+BH+K=Cj5UtTT##j!Tmju0E74-b6E37zp= z`4$ghhQ-0yA@;C{Lwp-m^7M*gV=^>eOHF;QMT+O@^X+53`sw9|i(gS4RhV6Xym*6l zi6RFuF9dV`FyVn#%Z{2sbE^s*&3pCJt3wPfOiNv~iYxxC@C(Ja_Gr2p{rK5GO!=sc zVy-us7z`&0*bV{OM#Y{UgP%}Vk&8M>`+z=i#k%tmpt3vs!}MIiN#**2()(j6G;T^B zuo*r-Ky7dt5M;B#YU0Jw#DH(~Z4fx|&CKWc=ooR4aW|a%cI6(L{@oJ-@(zHpAQVxP zKkmi9W}Z(xkNB?P|8N@LZS0O5X^?s6Ddx0)CS47cnS=0<-@&qx*6EcY3~@F8S^21b zRk4H5@hV%KI4_k<+~=uN-la`Qk-_KHea_ZNXR|Ii?JfNfYOBZ{7Kt~x^(D=CJdRR^ z1LRlsn?Zy^wH6a7=c5H_!P=wk`-uY7CY%Az4vpsbYD2S^_X!#2II>`bv>IQ&u3Ojd zupnF7wV{G*V6864CcS+h=(}Eo-FPfz!toEq!j##;)SfLpg=R!DY>}(P?@!s6*Nr8P zM(%S1>`32oYsJ`4l5lBXk409(zUEs-AJAa~_Cio>=4!!LiZO9^z_9D&vDddQwggw( z_8P#qSMLqe%!OYre06ONhoWe-l4STNW-+X>3Ei|5a9Tqsnau0|V ziy727q10K`4qoA}Y&O!9)Qv$H@q8(tz-Tjiac3a*mUc5Qlb_E}laqB4er>D99KZ0O z)yIe>m-`S?;JC*xlzKBS@8_k;w@21g<$DFISB;4Q$JA}2Re#JeiES6`?NIV^?R~E# zzw^jb!mTSad;ld%lG8b?(f%4dfs&vlCWvVYbo#x8ib7SOBWTb-MkePB;1k=ObLsSTR97BPL%!}-yfRYlr1xl!DA>wD-+o2cWC3-S^Lf1tsu7#X=Hx(N=fb$cN zl=%cK9K16ZS0fg7Ku5h{&75EOWg_l{55`X&o>3_u!C&#sH~wCNzaJUAvBBLz*=)H@ zF>gvfoXcTz27qLtOa7oD&*5oE*$x)eD9PCgo;(7g3nhNyo{KA+NU)o6Xq4;}{%^mjf z!A>;8$HwiP*Ccinbo$|-x51?5%5%sFE)65(fak}y6m+o}O^6I!V?xzP8wLg-@M40p z@6A-ckumF9DU-F`#m@?6PUl*Navkf7=cGCo+Mq4(H)_m5+-*@!RDy2Wbca5 z_R{p}MX+3&NY3Y@=Z}sg`ntOA#jz`EX;}4G;~b_4DWITfihHwN_ekodP`b(Q=lfZ< z3dYK-(~*(_j3FGi&I-Rc*3+4;Vs6v!O!O;q#g)s~RZupIbn>oy@0-V^OC7ttn2oGw z^iHafKw${xujUkA``OzCfoIrHrL9l==y?q{zwugE?BmsyC$7$&;KajZH1JIYze-N| z-Rp1`Rq7;KNeW@?`U8fej3Hg}GhAq$-ngeXxdgZP)vJW4|K#}-6a3_5(nk19x6q>c z@Fkr{zr<$t*;qG)V?tRYI3OJy*{*Vv;zCyGZ8J7YS~WM0UEAujQF@hzXhKB}lzLv0 zuPu;P*M1%E>ZQ5t$rW+(G04II&pP?PTLX>y)Bi7{;Y^tk*Zx2KqvQ^|fiM9Q=wpiL zY0gu%6?CuTDv6s7cks#g`mffq?0*o@9D3{T51%Q>BP_g39CuY-B|!w_F=PE9rdq2K zw#)s9sX1iy4^ybpZNj|)$hWHyRymvFt-8{X(UPalcrd-_Tmj!AruAsh?2RQ|*R zn}Bn$^m|q^@OTcn?G4xfpInET-*qx*|JIFkpjna6c?Y(6m1>N=(+&du$Rz|sq3o~) ze~Wk;G%z5>IL}ZcgQfPKvgjRlkMG(g4TqV|M40xwJx1M9Hz`E(Pss)YolBDuuJw<{ z?q8y>b#T}F@zbvt5X6c@XMbNws6*PSoopCNq$@I5ntUe+Q;rD}l4Q)>ZPO=<0PSN} z^v+w`x4YP~s8{ithf43Cvq{7&LmMj?#_0obzdke`p0kR~1l+9h8l6CsFTt}GJqgZE z9+YRwPryS9^2;b>M9r)8C%*M}dt}-+#gw9-jKd^&+zdSb3W! z{jb~?+@0Qw>X9u%+_-_Sd|I(Ks2Vk2wDzjw^;j*ojHMqDHSVtQ@k>ucW<~mG4J#vt z4u^@YUuNN1pCWp&uJp{EGsNCV-^rB7a$);0fXwnBdc_~!TySW@X8I(XKQjLjee<+Q z`lpvQUwZq|W821zeiCr**;zpEC$led0={U6UrQqhHFb2Ge}C2k5W-Z<>bhd0rp3 zW z&&iuz*&%J(QpQu8X43u0vy@_&6O>sf{<41n&ok8Y!>%QN_`Y`2hS;inamL*wbru=g zw%Vl#qgYSF6#6H6O5JNEYtq)xqp_MYb`hb}FaGs!m&68%?6%oAK*hOBL<^(r8Z+%T{rp;L6M)NU~)uDb(B`2;dx5{HC zKc@IqcrComcFX-0yhe3;SV-3P8!ev_VD8O6tpm+NzB_>CiHza&1{g?4L3_p(pxXGj zz4afS?KTK@%D3-r4d?M+5864a1(Dlw>W7{(y^R+aZUC*{J=~3lP?^qbmow# z*u})x+{2DAnF%nKDTFCTH5>bp0YW?rH4RZ_vGA`HYkG)6Z=U<{_ym*S>hfziqz z5X^dz`P_AWWBZKx#)p{OpFFVz4R;V`eGOa`l}+=ZtA_HAg(E+ei0At-mqd>(!y*7k z0&xY4@SxNsyKC5J*-R*5t`T@-$0ktKgxXs}hI$60Gp!~Lxy|nTm}dT>g-YV8zX4-v zt+c1mOu4aQ>)xDoIr2Qh6=3v74dGV5fDbQ;?SDPVL+HM!4+Pb9m`qo;gAe4pbr?0O zKy`4o><8`9)6HOCOpUJ{Ei(KMlLbINY61tMwpJ&-|eE?cVzMfo@kZ;CDuO)=9o@o+!GL#t*1JBTg(`!g&1#AvSvL^b@Abt)N*&lQW1O$Y zXO54~p!)&1G+8neS+g~(*hB6tn&v5c2-0Q*VO@%MdQSMfL7vZ&Q{~?!+{G`2Ze&}IvV&#~ z*YgecZA`x97*2*$%rhB7Fd+uOm`p30P>OBc&$OTBn}r}WrY5m~rebV=FNd_*cF&`r zstr?6e4Aq-!D|ZQ_X0a%fydJHDAv;`0gCU+cuPf?;Y+Xim zIns$k;ffEw>}~mm-_b`)i_6t`zk^0+lBZG|6wx5sluYBDz_|A zP0i&JjG{^(f5_q|1?*VZ$4Br0&=K4%z!B~9Buhj3Y2h~G4l6e3-EJjBZPc}TD*eW- z`11SYl6qN091+BV97t{uDz74y*EW0@ugetrjXKlI?v7W60}7L%b9;ca5u*j!oX1k| zleiL-mUYoq@advD$vY911mo$SnvnV=4R8~S>I!@vdsDh!bKY#IrKHj_%*koi7g5!hD|2_SM!CSAhR?XhP6tv3BTN4Wd+#09M7#fsqKGJp zsDL0XDoBwgAkvA5G!Y}c69thDA}Bpk5Rfh)ptJ}G2ndnh6M6)sm(U^fDhVY7Qatm% zd+&4h-uIk!?yue7T6gU~AYn1{%uJq{C*SfZHD5};!BuC=h2G8K;TUGY1w**txXsJ> zSolix8~@|}skGo6d*6}a#hWKg=g?aad@30SX^W6mVi3u<`R?+HH&P(rs2tQVy|Fh- zgpv-k0pGKjsoTo(nXhnle6srZ*iIZ3@+X^r-`B_!$ak{U`+}lbr#XhF?lnZcPuQ*>-b# zKh|XViqtw+du6##j#a`x@)XshtkP1PT|6cG9vvErKGi zK$M-JHKCY>Z$VBt^9eP^**a&PRlrb6{XQJfZ0i#$dHvIwcTG)J13WOu@w687<>HB> z2No`;47+?rzuU~#c}hvy-A`*{KZ5jv%;({g5OX}D zZH;oF&~fr{^;44%KhYRY&!*^1TIJEzCqYhy=Iws2E!s<$8H?{07^Iqx)f(g**So{m zEdrgW*|hUk*BkeTnnJ)=V;QNh0Lh3g^lL;_N6JFRT@bVHsO0yV?w0#zs2zGK+%>w# z$7WD-1fjRea`OSFU0+V#%VY%I?1ZY1XsZ9)A|)3=xy4EMaI z`yYxN6X)37Q_H2k=4k6T=3_PyJ9}tN2f2Od4DS zeCu*cgfK7x=(*Lh2jK_XF(CvxqDsv3W%E!mK|bHq-hs$^#hV|UObWOjSycH-wnJDm zERD&`^~>DundD%EnvIs>)Wd^|pLkZ3hp#-HGsUNzcaZfuYyZI7dXcs?5ExHoPM)(1 z%i_AZ>J7Uza2;Fe==$jS;v)5r)DbmZS{~Uls$_O@SZ%NYj^hYx&xmqi&EpGnBTks{ z?QgxLs^d7$P<17bMl`lZu!AZb>oA>;Pv}u8)h@~D@Ri|(z@Xwd7Lv`bK6B-JVGv%B zzW)fjG`)!w?`{b96jC_CPQokT?*=iNRRe0vyRWuM^!(>W@K_64CeBJuoKb(@VFwZ6 z7gy2I5$vx9lcX_(cZ6x3QcT@NOh66Xlv))T!?n6agK~Ng4&zfO%Z>2XRM*rhB*R;E z$*ThJCN-f$)9%?y!3tTHNRFdGy^msTYT!A2O^AoZ(NCt8_Pe}^@N7z7!hH7S!R zt*b4rV!mtdxMl*_{-Pr#KCc`;D=#X;yH%$JK6}yzALb&b^Qt<>Ywck`&(H>)1ldta zv7{+5gg!kT%YCvgQ5t`UCWOvZb$6tdGt%k=qBf|a^XpViY79j=neTk<=<8)=9`Kb#32%m8@@%_b~F-{fV_pwgF7)_ zQ7^W1UYD<(?jsilqn;sI!w9fR`ui{was@c>2Z0ZJ`XLOO9KM;qsXO01&k`nyE12|J zQ$GPZauX}0;=PX`-(k4_}mibHtd zSm7lP$OiO2Y=ZAJ**(sFzTR=iP3|zCy2~XAUxVhFhL1{*^?hT(A^1o>>xUUm8jjTc z-C08IVI0%}hOvdL#w_c zzXP>q?~7~$%8w{}I&qX!)Dg(%4JeOc(kD*o^b0m9utfg5@ar18s+IzMo}5gm(U5)X z%SQwcPv&9WTW5K>v)E-}p1%AGLhl-`Y@D|9asr?Su>%_vOPzI>O$bxqV|;Z)j`412 zBgXq`tm450qcdgq<`0*CpGvGZY&+hwgoW#aRjtD~Ev{hWmXVZ|nsj(XEB#Pv+Lb*d zd(OsC4FK8?)4HYK$E_|r6?;9HP$W<$lI8;1cV^4M+E!-xJ%ROhv(yfV?45_%T@RO57e zZ^dDwkx!&2IM77+LVmFKNdMa%?f9p{QY})QDhmKnrGLT_)(@>%&cBrMO?ySBbC>(P z;6U^qpSgkv^!tJ88GO&wZ#t)%Qq4qc{1MY^Wm#oew7Sd1j-vi!mJJ3oQp3w;S{1oj zB$WlwTRGYp(%{kU@MnGYjstF<{mG%`b)_mg+t51Vd9ga9NB89&o30^c8_rHn$cnGu z>CrPcE?IHelgm-NOI(Mk!Uo`FK2a2G`+t+qL`};v(F>Fq{=GM1S9V1AWFUB2TyC`4#20uNJj0ca@pztZT8UZFIZj>O)Q5|P_pZOe zZFXrbi{k4SK+TeoAZAs5$UHC8c(hWr+Viu4%d*Tn$##@`FGLS4a(uHBskSGNUb`sD zreDtWXL7QlAgGzNn}Oa~n1GX_P-MPe#aiHipj{|z^P$%CBmXzhyi9rgvv?riIcx1} zXZ23k40e zP+4s*-=VxyYDzdR?gRafOzryO3zZFB2o;G^F;gR6zV0W`acl<%B0ogCK8;k+95<3b zxLop-B-$#tu%+3q+-CzQxY$SP1rTaOT5eH0%_xu%0OJC8LY~(WEa3}qgh<>IK7VtM zK_(-K%!r5MY~t`~Ew^X52Rkfo24WW7uYan$`=bfwh<-`Y>Zr&(lRjm8tiZWSZ$6igV|{8zIOAN z@Gt#=T6FI(kgL0y^Iw#bzt#2*%%FKuS71HROE4yvRtx+OmngFcsi}ep-y6pcFOU5p zt`|wg<|%vrkh&Jew;j%RxVd?_4W!Qf0uGLJf<#RD zGpi$j!`uR&5b7W-Ga04UO2_@cuXtOtcrnFr^%O9`ts%z%)ES``=#D2Ij@1GX5hv@O zd+6|xmheXzqsY0qI8gcRN>P%tskQWi(?y2R5S$ji^uS2a zb*|9Uu{I&!Q-laeVlYYlG@jfL-|E)VUUq3R!HhD=tuUQgNv<3%-QHsc8Cq9Y6eoJMpkExFl|3rffT9h}X7%RVf>hyL4gw_}$9=X-hjL?{AWVY-## zHYa+3uYQxVm(|NM4nGWo}}Cutt4NXRHJ*Jtq?GE$juN~VP_w^|j9t)YVS_Zr>Y#_dPb(wEE>0FNZeSh&t?%g}?-(VW%Gll( zhDhNKI(mxpR8`e(R;1a;hy4?VM=Oqr`w}}oU_ROz+7#?Yzr@xa_+8Vi#YLl^uR83= zm|vJn4|R^Ctt+e@9uiKboh#1m60AzM2Ou6HADv#SoBnlcy9!k zHY+MUe$MPasZfTzSEr!a5N6_AE!3V>`1yqed& zII}Loee@Wy$n4Gasx;5J^stKKj5L|cP0DI0yL{4kXb^feUtcKCub4@lN#r!$Q@m*M zh_VYVhHYyYnZQp%EV2~edQ)DQUAraO4g-yrQqz1BR`NYotXV9$fTD}Y|D*iS$!n|5 zpU!}cL-CV7Zq&7&O1r6X06t&4sT4_!i?k7={rIB*CT__(<+C?obOf-PyTsbZO4nTu z(H*gp1_@nXRVonxogN*36nFe-738wM2jFNURXOUb-nu={Y{RWmt0^QjM2FB{vnmvM z#nh-5jhA|Hb7Uhr4M`eKd|7lB!gWg&jQG2b-bb3PRh$RK1d3wk1rs$1%2Z3{Zhj* z>|XZ*RZzvN#Ukv}wO{6sJZWo;lY{8gtrsnBa>ljg>2BU$@%zhYEBr3IKf8!3&=Q4H zq>dcu=Os10$x75^?~At#1tOmthr+&-lnkt)t*=Zse#Ap)irTcKLz$seDi7%x07i52 zWttdHugu#7rfSijBDdz57+_U&CRW$N%M+FWU15WThWb~ik?%MGj=Rmj={)q4%hmt{ zl{Y&+36|;G_aM#`KfRP5aiZy~9N}o)hC8MC;Kj1;S-(56C_C*H zz;^tA872TUK{1r8;PYaVD?!O^HHovrO|OIskdwmp=Gi)r7H^(J-@~PQRJML?T*yBJ z`Lx!cox+y_Fn2A4vr+F3TFPHq3$~h<2k_p#Dh3&Tp+dH;_?mzD)~|?kw#4Y9A=R^_ zM7-Q1Jf+N;>!Ke~EA!@UZ%|uXdV{Y_`K%L4ExBq^se zfpWw@Z1+#Ym_`j~{dG6+sB$X}-VJ=6I#j@0D5NC7Obbj2Y_G>R%W2ubKMS?e~w5vv1@7=9v;h@WBKc+`|oVZ$RtW+koR5GQ=Cn zOj^dwi^0V8D@Q&aBIf-={s4YnY zx~2M>Wgqy|*#Q1K6mb!~D)gJ~JrI=5i?**8!^^(U?rsHPgpM#D;eChjxd+6!f7}Yd zdxRbmL}|hBLncYuwFa9=@L%?6rj;&(c4<%&t4a=yW2FswbQ<9C38P7oU>q#){%^X^ z2-@8!K#~Z_ksY`BO^!l_K8Jtv3kM>k6xxYC$A296pNt>QHeIg1TRn5T6IV$B8|o?e@So zrrxBykCdBXd^A23_PWqY2o60Zv4H|PFzj`#ts z3#S|9z^W(}TF-AfE@l0NX=vL7&w~lq8S}dW@1BR*ef92*lVZbO8J$Y6!mZq|M&|f#UY%|Dbj|fw#PIVHh=G#nEczntsuSpJ!*!qeV*A&T z!;;;AJ>Hd7&iyA7^?T$fKx;=7$SUXws2tA{ry1rjgR1_+H`rJoF$pC=09(w%#CgSP zOYRH6tmhP6xQoGekwlx zlAY@=bz6y7bQ3<7f-}j$LBiMO{c>2~h1BIMge9G8ar69Hd+Z0iE{5z~hfhjl#3ZGc zzwSHC$xXm+!@5xd_v#(W@A$K_z3G^R&e(Bp>`xK{NVt5USpRkS?mLey%=SX3tGmOQ z-VvVAKK25_0WLFzAcG@jeX;&*-ck_xfTx7l41cQk?mSumY=_{S^;KGsEqN~>;uJV2 z*4e?`6=~2I!IL>WUD(_;k!|NMqmPQFzMy6cCePx|)r*PgH-(>oV~vnq(D9T70T45H zTivh9=YwT_d=Iq1QmVF@cdMjo_%vXJuz`FqqwMIu8FeFY!Y=krAxEI& zKFu=%Bq%;lRFZ2lP4}aEvcZNBr7}KT*s`4INT}83(txsfZfou;vHOIS1l=Mz6A&8K zoiDx!J-;evq0CDCR6=)LufA$7Wa!%WqKIXy0(0)c))%N}zxpuX7;pu=%_AyPHnD~8 z*t2lRvq>J&C4|hP30NjOKKRRikp;)5D8bUa(z53nVe@)oLU?x~wk8$TP7){cnNj!& z=`g8UtY#j8<3`wQSI3^c@?i{h`#xFlz-rKYc-H4dJQ54;s!YDFb^qC;mfC<32#{dC z10q&caSHuY0>2DkCe6W}v!G=9%foAgFo?bbz;b`n#Z4gFdaiA^stTp_4BzT`^R$W2 z;p0>H$jFn&WB1|2a?%K%2o9~w4_m0S?B1A_?0zR2S1aZUbE{d~hF$}*IDlwQGOKR( zH5rcwgua)H1*X=juNR~ngc@@;UM7pb^kU9vNRg=;TevKMUEl!Vg~q<%_bqMv48@n4{PNVxlv(zW#NB0%DZ0ra=?g!T$X^QC81!F@ejp&jZUG)dzAtt) zhTkh23c-omPx@o$Lu1?~bk!Qz3=VAaamwI}`z*W^mvCt0vq+^XXrcoBQzgigDSy-L zo-}yYy-oOQR)Fb~k8@+MV2L6g8SaNYquL1obZBxGA+f`NbOtAwJMYWwnq1n#L7Py_ z&)VNBmWm%0xe%P3;7Y0|=1{-w$AGMANIrP{R!y99ry{PrRv*(vKp4o&?)HTNpLh)T z;uAi8tUv&`Vu!OrFa4>vIY~e#~$pW+?ALvDTU#<3C_mS+Th#&MnfwW`gtp+_4;8kf^NZP-| zchy`ef``KRHdxIm!X(nJ@53PCsDWg1>lM)L3-#OXzdF*Pm3lhPBABsG`9_vi)yTd< zB>(U&lEl1%IUCPts9%2G-_Xl{yt2m^uM>|1GsLN#Egg*Op4%H1B^z8RUav)$BqK6q z+@6ufEaZ-ES^uVMe4k&#dkPu7kSU)3;c>1x#Y4Mfo04KUjoVB$n^*_gyHjEr+bpI> z>%VsTy?{Mak>3L-_$CbOaJ|wr{d!$9j)2-Ze*(0fy>O&$)GP5tI-;ivwtMPsjH>Y$ z0~JPGei*mq$lJjZHyjywyj*?u+4l#pb-dwt4WI}R;d6LmB73^sa`RNs3mt%aI)o*{ z!yANn@{e)7pb1na?r!aXndYsQ7NjI}aTb!hOo#QvXaWe>$aWydBbBT`9Q!eowD$&3 zKNvY&BFwg(DI#?&PoJ}^nqZ}!h{&87Il`R71?63rD;uWd$CXPKEe-1WkASR`j$h3jL9~nW{Muc$FN#++QZTA z0+7;X#i@jQLT_LPp3ht!Y&JBQE<$JKA4#_QUJq|J_>%N3hW7@&R}hcc4+Wa>FSWa` zi~3EIeQ(eH9EJ@EXJAj|fanU`#l-D)YvKF# zpQCzaNo^#m2CQ3nA}fdi<^gN3JDq4eydPosF0eR!{kq`{TDu4jW+kAo(nJ?W0W@@Y z4cPz8CZ1<|3xJ*80WdtfiNF3W?U8@p_LMunIUqZB`Ro_d-Fosp$9y#s^4l8jwt(9u z0LOF{m}x4T={#g>{PXQk1Pg0@Ls$PW;K-Bm*(XuEt(#gB5=H(EQrP(pzRVo>1bP?NtB5ATV^n`UR2>2SpOcCNNyn8u&yzmw!j^~L z8cm}}z>sR>AJo$mB}_MeTp^H5m!WaD{PrMhHs%k~ZYMPzeO1jLML*a8Fos z6i7vMje%#J%!tUY)T45uI^h$(uifR0h^2ia0S;6(YT4etQt5#mCfKTE-e6fe$lQ)F zhhU|xG;USueu7Wd(X`h3wj}3{EqH`ii@T>GU0Xgv^43^=o-3Z)C-sOu69L+)4@Hl- zYl4-!O~ce`pF&rl?+`qJ7v$aq`UQO{+j`3`9P}a*Q@fA!iu!KiS<(nvg{LbUhm^y2&%)vjL8$QfLhPTbqCxOw>yv%Vreew6fd+l%(}* z_WXKdpW2c@%oyLwXhSRyCV=NI#L0=KmC{P;v$qzir~sRHJi4h`xoolB!ouJZn&~2*AZK!pHXXupfB%N4vwY--X`Cp}#|8#2!csBA_2rVl z#zI)n#M47E!%Bh@b0|6uD==F|UEYkb6d8tDDg$^2s4=-B%N>isqYt3Fuu~Y?RiY`7yv@>p>ef$E zH|=upHZ2%&cm|N@1?paBn|*-N6oe8A+;jrAfRZ~R}Cx_;9XT| zAUuCu4IuLn3SoHChY?GR#y@cRLyxKY)d6^Et*IMIqHzCgk_Z8#PU^Gs^ZBB+ z#I2H+A_hBrtO$sVzXB*66Bt}fA7KBPBDL0yI{HC>k2pj6SMW^$^rIxB;4gwp zLPOB>9>sgJ6QGBw`){=c5CTMM_oy#4OtS_Z$(O&~64ZXgM0r51eUb`_PS8{M`GRed zq<+bTm^^7vTEq1IXGxnk;{;kB?E!y_O)W|CAh75$w1;h1&@oaf9CK`nDWMJ_N%;l7 zV{&?9WDX*%zM_F-nxo^;;s&(UHDv=dU>2};XoY6{lU#U99l;GVKEo$&7VP8hBml2V zw~Ng~KgwyWNWOaOMS=RPY=*zPv{P#T*Ic#%Et>@Sy{P=eUvCwVO)OzICeWBP{v1mE zQi{7uw(I_p={m<26bOycEEIk(+!oiWPKnLy3r3(6mn`@D(&lI|SoqBDKUiu*ImbtW@1q@GYc&VCf__;70mm@12QhZL zLn$sI#ZT&9k!F{;hlGfVFdsOayeI>WWWnwxI>LC zfp(~h4K}cUwjza_C3aU;(Ip1M2uzB{0cXi|(S_|vPh`s6t8Z?{CdnC#cC>)@$c23F zPX4klwYKAUOcdBDF8VheccAtRC&@3(Ou+Wig+GZh2y9?EcotkZmq1+)H$$MO!WuU;kDjW(^~LS!g$ThQCO7J7tr zb8d)BK84HD0D7U*eO;2Aw6;2}H>x0K8^O_+Az!B^Pjz+JKJ8dLo_X(!&t3^D@z^88 zMad7|4B$Phk$sk}PlWAOf5~8X3jLR$6vk>XTYejkL(1EhtE&JPoS=JJIr(beC6YM z$yVp^75^5_IHL2(U|H_Afv07vNnquNOw9Im_2+l0KV-hKJT!TjG_S{DQ3hZPAay!d z-#_I8gyw)01@0)xOp(G@jDGJG8gcKq_RgVPaQmaElt-8 zS*>xA{Lq{L?I$fSf?F+`thWbvyiJ= z9ef;79X1TL_)U!S^u6>2<*c(xaS)5`c2)CxbIm1=-9IU3NRog)5}0$JdkcIgz4dxt z&)8%y9#fAz(EEe!dG-Q9od$;jG4gLZ!ul@sEldhUO{3jvM7(2*a-8L$kBwn>! zVw1=*xa~`;e8NLn!+S23nWw{yb{gfCZ?otdMhg%0jbf&0w_^?ktIiy#84^axK z>{sLX@&S%RB;5uRqr&4Hp_yEyO8I~`&@;L7XYGYg;g)G42|&H#SD#mgM!DCze}S{8 zPhI#SyXf#~86?rF%NmiCjpi8l%czAxO<9c%zLboI6O=G6PTYNzW;E+6#b#zKZoM=x~5Ux_)% z?r*xG%7T|kPj-onjhcg!eGWmFJEqj1;=OIqE|l;A4R)#k%)*N%GBV#^AzJ*T13(s=y&fh;-%DrQf-oL${rKCK`9_K_ zA>@E2CB=*&f%GidNb%1Y@ZB{+9S5uv;naUBP2X&jJ=qo+6ko3HX`ePOQQc>udru8ZCRZnMa&SrLpUK4S4PeG zsahX>LkYW<6qrM{B{?;yem)M7QO$5o@@&kN2r*Y|!GK@h=28zmJccm>r*|E`|I}lt|f?jFn`tJfo6- zS%Wg9wN?7c2mCm`8Xy&ex*9wdQo>VA@K)GT%(9imf`fd=%A|Ea(N3+e9kaoVyrK4N5vABoeIe~(8#@xV9WIKwoo=fW^-?|Cqo;>&2?d1wbCBRLy z4m38~9&ST2cgg!Xp4$3Ki@ftANTq0PmetV&B`_f29nf4SGW;~?+#iwc;9oTdgE*Ug+(w2KM>-S(X`z zaq%(uLKmmEQ0aaNA6oc>{pnkPcKu+~+iwS2zgAix-0BN`3}!B0Co}IfH&)e4e%{+d zK0y9Nj?K=0rBHk=6Su0q? zQwWy)sq%}Ad3*;d5a1|`Pu zKqb0J&NrRi7hkla-<7Uo8hx_0F~`D}g?IUBc*+J>O^V8RrP!?}Wv9>UaBKSCj^iMgJ6Z1buU7Jhixg4Xs-l!wtdijeBR9Tve%oe!8?}xE;)`Z++&+^$2O9}42a-SI$gR71@SGK}1e|F`emN=QQ}lPK#?kNvpH}!7VqPoy zW!nf!;8{z5des6?J&B)Y-gOF~mY7nG;mNItbGJ?fs^IE9J>1SK+ovhm&bH~SL;SZ+ z#2)Y&{aN3EJw9vVAJ?>c=-fTShg2 zg9+;dXc29*p))rdMjD>vQe!6!>MeUBcHh4D*0LK3g~j;+?LuFe_g`Of-c?wc8sw@e zh^U5aBoPpAy@CACf1|bTC(Ny~|5%td`5UMAm3{g~Z`xD%3@C%|&)-k;>Qm*XC=dc> z(}m#Cl)6I959U{VdrHS#UE|W!Yt@Y|oplWjmpSyrY}*@e;k5BmTL>Y6(R=oSH<4UEwb-9s6?=ntN)zFA)VLh314VzIWbQ4Nn&TMzj?t} zs)x5IvJNL}0R^5X+KKruJP&-=puDhtXz6G;fy-Q34R>5j_SsSs1EqCGm^^r+a>oa=@`{j>aPHC2$qz96Pn5ToV8?(vfuR0z$-+p4Y z7pa_{u76kBC!N#wI)kGgpO+h1fi?`g2s8yv;-AgaH>tiX?St)rEcQLZ5Tz_pm2PG( zvzG*=G(VR&y+Un@9^2KB#Lq+Sc3lV1{d(7vY z%DQK{sXnUxvF)K%aqSLxi-CNDpiiV^^l`Y;Y^0Zl z$GotW#cdxQ)a26kfVIw&2J9WI4`4rq;!Os&pS_1V551{#%ay9AA=i(2yP6JXW$`k_ zb^YbR$j%e*w0xZVR^MkjF8kBpJ=0jQw4y6!;K84Z&U$f2mVL1HEQFxc3Fv{1gxbURlWx`V44- z%|ltGf7UUJ6!ab1ZXYzp?|m|5vL%QIG?f18gAlN7tFCfR6EeTjNC-qq)~YF^?5^?j z6gKhJZq>5$n@{P#>=N~mOR=I0XKg&76Z5>;RgvTH=3TH&=gxik8-X828=)rblfE)7 zDK-MbCIXxhFtR6xk>{y{{MSHH9IsYeimj+`UotwV=FnGufnT9L=*?%qw;NQdtEADy z3sK&I1Rh~WeZ6qg4|)sR=8eAslPtJnV?91k&|<}$X1t%+s71Q8S1gUyE{p9NxP1+7 zn7N+=gyvqF(N4{GnL1tL+E?TRmjL}4&8tSGTe?;baSeA+T=@3oJWFQIN8q^!9Dsg- z#truu`j+JgR-0P|U0;azS6ac`fsZB*qDro+*PXik7Hnii*!tsZB6HXCCYI zDEl(!wS8Ig$xFex4{u@b!9|{#joXU$HdH6pZo|ue(}kZbogYSf|9l1ZU7`8&znPtA zD55d3&R3Id!f`*iML~#Vman1V9ve`1%1Qhzk>zb{j-qSo=EYXzQB|OiN({h09pwZ1 zV{hvh_)3*o>PA`5$4%bs;Qp?n;b(sBsMmGfI~WjehnxC}#|Lmd`jA}p0V@i6 z`)N(XN{0w04y%ltR~KaReOq>0hB`a)%ez7Qo!7l4kFXn|a3wS&1p)}e!1G*6tA^-- zsYDH=0QdQxXqiJu#B-}8W3w&EpGCMB_l?$x zk*b1clh;FQlzmR{)o!{RKyU|9ViYHwi}oDWH6s3Ey(~NL0{g%VQ=BM2@-tvruDs7u zD&bP30*mYz3{=b6i1(YY?&?KN!NS)pWv~C=6H^oGnaVt~&c*2|F^^R}I%j?Tb}_Gx z>+gPlWOuhue){rZ*v zh*Zh{XvmmgQl#k-V|5_?C<@4nW2QfR zrw?vHv5nl?C|2CDbonr`>c``=jAkNK-|l#nrI~bc;WYG9NmXUae(e#*vB#@M1q#Ea z36n`w>akNs_9te0X+&7jzAyZSEV{}AzLG^*rHwcKrppZ9 z{gx_xwB@k^4`3;3J=8CzmIwUdAhDuF?AfO5fis>CXklI8O60L18&Fq_$5z_x;VN7a zP*MD2kPX2|)q2ar*_qaLJk7}EZwDRt+aTt*^#5{fK+tO^$CCW)faVYCjr;2$o!N0E zqcop`??(;bB(B*>IrMJ--=8^K(5vcQbYu7kj`DT%Z>O8{*Fnl0b2ci`8PrWxB+HU! z^`tz%r*9z93pX;ySzO69>?W>R*E! z=;lmO`ZmdiZ8e+B_Rurt9lCv#UNIOXtZj&B{5e(i641X3AVob3EINbPi6bLMHaxVb zACY-AIxoo!--DJ!5hG6lIm=T8F{^~L|)gc<$i zLr?k7`{07DHJ9CaJ1Neg!YB3$k|l};dZ?T1rzGBaX>Z=dT>E5j2*mN;Si>wJGRIzr zmNi+(jFg>JBntSOd=#rg2|8!;ozvU9zi8Svl*{Lwpu>||B-Ij)&G(hV0ma?>)vU-Q zCo1inKkGNF{Q8ZHt1mGS&2P!OiFZQ@@CfXv05BdU+MjP2o_e?-6OplD#iZoo~4@2)K zq^E!R1=zN-mmY%((gtWEX2o;;*A!;!fAW1CYpo12^zmZPs7PVGJ*<}CvdcIx^lp)$ z73scj0v5}Cu5$Y;RBa8a+sX-cf=;O993l|DF=+v3Ilf070p`KFb56UiP6Fa`N_dH@l0lLIabL~g= zBggXZC{@PThtqcgp$NKY1De_NbRKODw+Eg3P4_fVs$!VOR(Fex-byb!Hamj~X#3_- z)_wh4nG+11MFGHhW>BY(n$WH^D%fysT{s(^8aMOLkhXu%-z=@OlZaLa9TNIwQQ=mc z!DKaUT>ZSt#(XLj=+gy<`kMc8^9G;qxGCO1ogo9oT0plEFq)brnX+y@XG7R~)KGK6 zrTB&z^@t|qxNN{RsK;{CMn=(3p6o#T5$m#d5-m#Oa-JPS$qa3m$K+y`-vR){_1CK` zUm5q1bLd=my-oSxOcC)L%|Qe=AmAB&lKX3{dcKi^20R&|uq*xgK0mYR&jVPfthAx; z$WHV37|{;R+X(&=RA7HDcKjv>b#}4}pYgSd5Jy(D`Q$gRuXTJphFkLP7;*-LUq(n8S_+&v#R!ivm zYzo!Sz4hU%Jh+_4U@P2}Hk)5SvHUnU;xC;?v&rA3TGB3Kk;<~)EZ2#WOFgH5DD!;= zcn=*_uq!0h-DXvx`hKJ;<%HK}7bgHoZpALXh|UsK;R1^#e7NqAqqQyB%EB{7xr0y2 zYqE^0YB{W?7pC(i_~|_j@*Yl_(+-MuI=vYoHLZ)uGQZ?i_xC`6Beek}9L>D+{d-E+u&mlUpz4zMLaV+oexPU< zNz@{VaI_=1bc#q}thZAhH|L&+Z}?Hhuw<^?D!gW06&~!ZkWGwpb~QS zF(}{g$;g7*wzw5MuGq@rjd)K*>yQi2Dg3|8k4u1H98n3F-o= zv3__uO@z$%N!sU7kF zUTbu*+ET?(H{ygfas;+NqiS8e-#6O>_dNxfX7i-=Df7G^Khdzbg7o@>EUyP9?q~EX zmCFdN8)hHMug2Uizt^RU-4Bw{5{szA^nG5$R4IstCfYjG$ zdzs9SZ8KJl!JWY#@v$Mv^aVy?tup-rIj4cP!qFH|S5*`AR3cHz`?Z|^{eAD%7w508 zTRLj{uQx9SfKI9s+7%a^Fu%F$-H~rJj`;Ff^RBnE4jHoBF3ZkzFGjcql*&ChZ8=YV zoJHaGn)XLk#ch;#c8!`YLbpuo5u9{U^08qn+#v#^Qu1Q;vUdyd3T+}lsM-*!H{_@z zgjKjTp5vkIf2^E5GU*!@s3j5V6Le2Ax5-Sw_K(8g?*f7K*8$miC7CrU5H=a4usA$3 zSXRm++QeOd!kZ}5mu?dvy- zeQv&Koh}{iBKbo#8N)g-qao%5^*fXRUF|~D_U}Z|{9IICW5_}cP`zP=yL1jmA{K8b zZqmCIq#>=PkCGd(PPlXEPhRg*>KYAHq_^q!)?z_9#$^fo zF4qOiK?Su71q<$_eUvP+Z%&gVtv%a(tJuYD6BPUXyjxHKwe85|n?@#3eh>igRCAC8 z9Z;-`Fn7L@z^<`M`^O&JeJJnVna5wfpo5!2k_D+=*1c-V+3oo*-2Qx@?4L}O$z2g) zOUuqWncUYVeDC2DRTGxzj}Ae~2P%^p8&2aq{n{%HJ_xC6%SLeHdFPxMq*9;)NqRY1 zDN@nH!S_o{ni-aUV(XT8IwgdM z(b^q-drDW`LX+N%QmZ{lHV^e7_aL}5y{e3ycp@%+nz+iJEsr0@IV(Vh59D5wS0hM zsAvPS%a>Hs2&3j(yw`6u?&59P`mXz$*fN6W<3wjbPAKORA!!;4dkYm2v zk0ETYt@PMj7hy{Ux3)zCPtw5ctxX280rc_i7J?!JA-N_LJLV1Km>N4-ml$P@J^~#J zN^<+flFBb!Zol^V_nQ`iI%(w3Uo0EMUo6|Z5T>#%U=rBB0;?snz&1gZbSV&`+t*VC zC8nGv#x;jEwA?gtm~^1av#xM}l5JVQs*w&-#0gZ~1KDBTGdeE(b>d7bz)sN$-PYRI z6V0LDzSOhqx+NgO+OQ&F+O5Wsy^vhjJbH3_I;6i%R89r{9)6LC74SpVTiQD_#Vcsi z89qVL)V6&Y!2xmBx}iu_x-_{cFIo1*R=SVh&TyJpZpz*Xm69c? zSSj)#t)oht=lN5)8!Wk!me8i^C$y`9(p0aDU&v>RPEz(}QZH2`b^%u%E~Rh~HC2Wj9(I2o$8ZyduTw%!#t z{%OM>oGV-2WmwcOb1w%UVs3~HU7lA?Z_at3c)GtkM>{YFiyVUN-?S=mtZYr2e?+KA zvCm=1D-QWdP!gzmQ*F(!!wMo~&?JE@glT`AuBO?18+0(#TzYt}#`t!#Q zO9|S7j43u!36x~2?u2S~?^fzPY0&lCevzkeP%JN6MF{5r9d7z_IiF?J!by0roK{Wn z;=!{;CzDOChVZmr-oLe8w}LPlwzWcxWc7atzMgWAQ~##N*fDvd2NoHTKl&A2VcMVL zZ}!a|*W^>Wxbpive-$GHopD7`nyF_pY!u%N1TL6d-y;w?2oi8mU*a=z4WrfN@i(S=KM$LqQbS47Aq-w!k&k6m*lTx0LB!=jqdXR z1~N_2+>i1{yLtTdLdY!<%3$^Cd@BX`+Z$uYgtz6RAHf#h=Eo?j$8n|DS7{VvRCClL zzF-?*qN7Kpxh*-oQ8RBNZ*8et@ObmC>3?4oJf^vNPG8!wRZ!v+xP*Op|1TJQ->;wl z?O+;9k*+PtJc4*|L-u3`X1txl=&CxjDB~-6D2VNrk^$!d$=Voe0!KuyzV+se!?eEz z)o7C0_>1K#U^Dm#u2zPVpj?junoJ_GW*MMdPqPg-&j;~XKotlOc6ON?-+?@+sUNEM zfAJsaE!MN3_WE{EJ@b=2YHo?v1^vsLW-@t|!WghkS*BEtNhGXKKOU$&qbvXIo`IZB@KF^;;*|BqL9aHC}tn&TjiC<4y|zpYhu% zF=5Z>$~lwoOB;{T-tETQSZa$n^YXBq+l2p|SYOr<_57KgQT<9c6~SwgS)-@4*bpgM z)cCI?!__PM(}JI>IZTB`pgsjc6gV%*$as7TSn=tyAH>i7fv#zv62$sbcZmIXx*_GL zzmnBZ5Fu#s`S0wgC^EfqP}V*SKDLOuOdfFEDhNzKRI&3fQC~Hr-OGFPV=9s%O*NeC zqUx%hbaFpMbYQb$((bBaEU96WkmrOZ=K0b{?at49GC>;V;$PLZk*|Pi_5<-}LYSZM zsOC3aFM_um@Ki(s-LFk{728Ko0yt3!wPS2)SU1!av$y@nMiM@tVF{$bW=o%Rt(+w8 zs5y|~@cwo;T}P0AtGlbcDV8|Vzl-LWV7xe%s!>MYf|Y#!L$Fm3TI^#E!ObeVFl*qcAwGn{pl#O$1VXdf& z6lfuK9E5HoCYdo}&n87-`r&qV<|k0q%{2@ca@RyltYTNV%7e7_ihD9f6VX96PzrJU zBHKA363-sQD{4i|_Qy&5I|l!vho8&yZD%~vVCyXCG$n^@{lChW{^ubvEjwax#npad zHQh^~eR#)D<6SnIFpN9=_HnW4Fzy33TIeuYsj<^a=hmo$8Auj-^w;QSp_K653jFcc z7`MW)gKFwc8+JJcCG@Ud>ZzJ}eT>6SguMsFYS4{USa=7&1P{Z)z9^^~jLvnKcKsyc zPBDN;?H?M4T_7W3`%01?>x1&-?r6Zcw?kX-jjx!v;5S`+SYMVo4oAept~AtGsyFK5 zf>fBQ374f%7~Tm?Isfy_FXUsX#CO-3HrF+_yXYI&6MSGjHCe|5xUF!IYIz{iM5sfp zT-Sa3B#Y7glezBA`c?D*N||aHIrjp3a5wABVmV!~?5M~m|1xf#O~bD+Jy$9Qow%vH z+)TuiFwL$~^C?$nuYmA-p>Fv}^3LQlbE$yHs_bz?yC=M(?8TXPybzg_ODaC%irpC8 z>`w+tW1{Wv=9JtY2fU0 zjBiL8r?TtT`_bq1?SK5&EFKA+$5wSQnSq&j-rOPE#it-OT)S8`Hx$v84zBL4EvxUR? z+TsJ5gg*Lqgy*Wo6TjtcHQRw%E4>~fcj?b*PDC-FgA8jS4uzay;w%23d3B^1a1Y%*3rucQHs=y|16|L z%A@Kv1GO|Vr$*Qhyt}wi`=Gm(UK-1{$trX=jR^6;t4N4#+fxDf^1m?Hh1WQaEj$>52MY~0}$Dg${fcHzLu$T|K%lEln z@PYnw8N3z+E28CJByVp^$(uX#eSLNeA7{49b%4d70euC!-b!g?WNk5It&%3^f>*1b zi5+MRH6N?6kMr}^`0&VOv$d(b-7ei`{9xU+r7a?EuZEyg4B2!lG^;l2ZRqG{KCxQs zKXk9+#)>-5D2&9`0`+Zo1W`YF=hPT+VKn2MyCobWF#><50H zwpTYxRKEi1m^V<&Bb*bi8Uo%Jf#K#dqLJzkyR=7k+EbSo=Ar$+Sngmr%W_D=hGKe3 zqA-yF|5x9Gpq2|aj%3I+mLG|IH%LWL`YK(3OqWHyH@= zya9ibp)iyf?{Z#EO)$gG=bXPr*i~Hmj&uFRvLmh%#wY>KtlM1S!IZ|W(eWC^%U&9- zvt7%Rw96c^hjVe1913bs4ZCyi-q@@==r{;)#SXxR)(~@%_=d$xG&9gOWzgw!tonTy24iR z_|ne-a6{sX3(%ZlaHP5YZ~qupkzKYQA%yL91P5(%`cexWL>E_OTh zqDUOG6LQRZEete@Y%V$4LtIjn6KUmZB3Q5JgDW7(5>L9HN zc2vzbhdO7`uT+Y8`$jEGkRaT;{aRy&6%t#Huv1ja>Nus4#1|99SN&#zgRMbaMHh(%>FPXUP^2HxB=Ra6nEoE zrcAx;80Io&_hSC~2=mDPDyF#8qUsK2S4t>amTUl;yAP4kVN6l#TiVsp+f*gP#P~mY zec9`D!<*hBpQkIITSEAN{MP^-Q&3_R6srC3kAy|bYwL*^YG6OSGsXLhPuV{f>$#~U8^R{KaOW`|6++FVh@1r;sqcd z9B>XR_?=E;jtRLrx;HWt9**pWt?T(Z4rteci{ z06by7BA@aAwZ(<1L?!wV&rX@lPc@jDWH~`P%koJ}G*zN0m3zgi>;gi`y})kQ?Kjpq z0U0z7lpq?ynp&IAqDt!EuuE@%ZeH7vW$zhnM2 zuSd_~t(~NHxlf7O1`&-e&=8=W-Dtvh#yNiE@szVjR6V%0iEVmP11EN^7)`}Jc%|4j z`dD)PrOpvBF>3YSVrzlT1&32JTtsEBkmx3MvdHcl)Y9AEg?Bbd%SR{j)s^Z@sks>k zr%XpSQ229aj1Wf7KN8c8Fu7SSeYwlKB3&yxufjo`sw=*T8)kS_;tq@eU>MTJYwKQMq`| z*MjuuWZwrR_4WCuktk#<Y}3EP7RKX^#WTufSaO}`-j=ubru$j|Jco*!m}`LzJHRW+K?`=S7g@VB%rETGTz4~C zwa@b7Z!|iBYH%gXbTv9W^HHc!_b9fg(akAD>5L2i1i2{o`GH}bx5pz7@rMeodIn5y zDfUMeep@Y9zVIa|<4c+D4+uawLF4Mjz@xpTjy~+VD?QF>Mq;wdi??oC=WDIvs7|;L z!lp1}d)A$K>WepKSa{O5{G)c(Lgm-j#K0AVlyM4Mn*N@iTF~+p=Ij5= zx_D=LPDf8ndmd81f^CqEh8&|C(3I(aM1rh$T;@|{7mH7a#P6={7{Ip4ok&_s6{jif zJ;(QI-KnnnG3XR)h>U{EhogQW7xaF>=gi&qNh%+WHFO9o{Ju<)<%2Jgd7YHfJ+x;*-sXFGthkNTTPu)U8K z*@Ri{s5XwPQbTDfOS{;xiL>KQkI#3{Uix-Ke_bmUoyHy{N)rh@@|7AwJU~q%zM+q4 z9F!K2U^v;Gc3Ax??y{%P%f|V+Q@K*PRt!GzT5jPk#t7MpC|RSQs60+iiZ}%u#y3g< zX1x629*J9?$IN2;mkFir^p}L=j1#~kkn!CYTjEZ&3Jm96V8NyW9WTPp$=Z{?sNKD)H<2-EYN|ta>!YQKlJoHjK#%#&M%lTu7C^b5v$Nm)=AVd-*8%@ zSpL-5m8|Bv^LNgD&5@=%$sE&R_rokJO}vxkrW?DO@2Gv-%-p`=zi~I=F(~hU#B*Se zE(`pDOjUFtcNA-w30eW&W)ungq?4YxP~swXsL!zPz) zAbI9f{;b;gqKdpcw51-YRXXa)`DW&gFLuQI=6ADocBC+KSc9j8RNf+}2tB<%u=6t7 zfoaiv9B=^fNYRWu@83Nq^*DuN^!fumqF$yloCV!ddx-G~=h(g$R@O~bZ7|&l$og?D zmPNx>txQDjxM#r=d=G3NU6+9)Wg7O|Y{QadjA%u5kHr$HaYQr-5l>O^uy8e-b2Ra9 zY7Lr~hmIRJumn8&Y3}EIu8ojhBvW$Jd@+cdoI11hWJ5tZYT@9mzzknc`S#9dEiOaLFjv^K*L)rpBa5nBXIS>| z{9>Uy7oi&6>7cISH`^%aYmT|4pXD-kN?bqK+%;%!j=upN!Bl3XXq*JZ9~=v9=DEPo+4==hHzKO2h)k{@D;s}=l9Q^NJAtbXhl0e$cR03 zQ))h$n-WjnA;CjM-x!n$>&cdBLR_ zX9qbt?1JEdm{&&YF_Qhd7z&ZbODk)LvXIyV=29iVA*hLo>MO-jr^p)=Y0Tty*aNcf zb9ck0OK)V48|KX3)7((_ur2Y-7EL=CXfdF8^c`XElDyzu^qq26i0m#N86rVe++`(O z9x4hvkk@!&G&(A7^NtI7i6WC`fQkXa+znN2n1*wX&r=p^-9q(`q~2cNg1N7`O%|8F zT75hi7p#YE>_Wpt)jY^mEnI%Gjjt~D3YInRUEcYW*E%azaN)a=B}1K&M>kLkWgb}y zNUhrO{P^_TFP4CNrh~iqIruW~BouX=u3Jey^KE=s(_{~1V^|rEANkp5myyYDoD>%I-Wh`&@ zx=O#wTvN-*Ig7KpQdjqJCgN;N#>@M+??krUoqICcJ+Ux606PJ*M-1 z?iW{{FMeIO9(n$iSVVk$Y)&M@qhHZV9daYdtoEhueuJ-|-6$10i>BUl4{am5Q}y3} zSzm2*+pktFT7{a+R(WDByOIOKat^w1O<740`jI#bnsPR3kfN~Y`NJjYW9~WAklFdb z17s^AuE}Uvw{@=Gmg}*r;>g1bGWFg6b5C_6IsSBS5!vU{&h-Z_P>P&iegoVU<#6~wLtg95bRN1!ocooL^Dc~XfswHV;|n3tTDbkoe@<9^ zy8`0;f$O7EEEiZrk~dQgDBV>4+}oY^y*{$W-Z~Ki9Bdf91lztA$~^shk61Vlv%UGk zrO_BHUwM{wjcrSTm~~lE;oRlkWbg5?O{lh(`-#_wrXfx5`}$|aJ_50;-vT3hes%x% zXOjK}Q2sxtaF_hw{@Ws0tL>?%Mls>a{?nJ&>-I7C*8@F=$Gqg66;UE3Z5GBp0(r%@ zQ%eh$qk#7F`nhVqp{~dC03k^ApQ94G^!8v}c74uB^l7bUThsU^AZ;BMb1;aW5$c|9@}MFuwB7eHDNGo8vY*XUbmxW>-We(#Heu zlE|d;CL6>xaWO3rT-0`3^?@#!S2{XS(ghkSL|{ z4eVa571bb5x~TfK1JhO>Lp@TA-Dqw6quy>KnSDj#JJXr)wy|l`n&jm%u7Z7?17s{9 zY(|qDe294^R#~@37)g;$6@#!OF0FKpxYSF(r)vsr>V^B*hA}I2xpseFuVZn(t&@yI zC)4Gw`6B+#*Q&DVQAQTp2t1aEEMhijsPbF#u7$WE&h|RIf)Si-s!)~;ZfpZ;$f_l* z0l}-#EACqURqilH&$)|(Y1ufRxR8ub^H&siIoR*kWD9DqiE_-y3VLck|8d3~I`*-m z_6e+YL)2Cf^v^tmaM8sT3#1OFyU8Ejy`NAnoZxM{wd#HRP(9X#^Qz*Vg-*w=Z!X<2 z*Do79K&e~SCi!F%+H=NF#@>!Sg2{ej?x4>&#V9}?D*cW_2f_~CzBBq!9o65GYPIwe zK1<6K=dlZhuK&@#=e#2P20~VAeYtVpd3Erk_%dP$J@sMyXkXY1#R?^-q|*l^m=%myRKKkVAJhs~5dwPYZ&8lTrwDc9vPtoA`&*d@R~mAD zf^fvoUDdcutfV)S!SQ~QK9#i1Ynuw0M}`|4n!|6-dva*qz*cJv{%27tJqw&4VSaoQ zqmH!~O8;o3f!>aq%c1s4;UqE>6@EIwDQcLxpVqB8+d6%BO_|z`_l%8ri`A~lV8@`4>Hz)A{!K?Pp zTU@d>MM0H&l8K4uBp$?4`=j80tQDQPe$qc>#ccmHegfSB3um4om$%ehpNAKeZF5}} zzqN1QLw(_oqV?E9oRjQ`{MCvc89IbSX7-<+)Uu9gD?cE-+6Gac_Q9TFL;(ThDR(Xi zWT`mr{$1Y_u=StcKA)Vgi$1OjzkkS6`-kqqQ#oH=*BeCN`tT9j9lp3s1&PW@y=UDxnNy_iyWdM{ni zQ-NX0S^6?8XtYfEb^|)$2xX0*9zb=O=)7d8W!7)zDtLrTr|Sgj2*z>-Op=SJMevexhdX9rdA~p%U}(T85s(1OsmFUiDu*86*MGnw!gRKZzURMs z6*WrC*jZ98jN1{XIRWsl(fJq4+6ZbsJfBhykT@yu+A3t$g=|3S1?2KC0GLlG1Gi;P z_=@{8*t8jVaTM0kM?{6~CE=Rh;^9*$IuAEfcST}Jg87Uc`R#8n++dxO6{TN7Mlw4H z>$(cE9|*h3OC6}cVBYW?kIboTA9hdCdBe(qAk^F7!MyitTy>-AnOrcr|hu zPlT_4IXm+boXG%omP+AhQA3fs1|Y$hbwdUB?R#KuV-QJ@a@s>`T#iyYiows0%%p?p z>y=X`s!Ds&e0mDihw=nvxQ z(1ifH0V4tDbcU)e@8{ASt`p}I^O+0fQ{G|UkSTxvjV}>rv5Ag`aVWOK%AJNkw;dNe zor&GJ`NZ5WQd+OXTJgSQ(VP?U=+^>om;ARQcTb5V)Av(Vxv5oCt0^ZZI?s|5= zf|eCV0nE8(ACr@Y+3R--RXS!fVIawtW;+nmdhCs|poq{0VNh#gta1xSyW5+Ghd z%qa#vqZ#?>FSS4pD+bAR`d+_w?U<^5%qUtP6y~FA6&yjiaJ8Lp7WEb4RIBlQj+1PU zcMQ&6zGdHLXR4>9_d%uW^%pCcNB-Q}u&+s&0adxk`p2X9TDu*=Rk^j&etVB{er(Un ztJCo+?plb(&!T>l#5IylU{pAagJ*0gWoJFQi0>efNhI zWv?rNTXxVF+7VJA#kglc1 zLF%e=2TR=c4WNcLroULO9bs&PU50kVjUr|cU2DxpztW+!s{~tT#OcTDI=5*GM;CzT z<2zF8${tdF6YBn#p3-!MUdbc1W|?Pn)o+6Ks;^}k-D$_1l+W&B4uVah<88q{_rmA) z4q~&AIeL{fU{{tRfIroz!9L^^?m3{k0Z_}c3A=C?nkgNOJ`_WmC90ko`)J8>2kf}( zp_QLwqsDO;ewTy9;vsv~4E?kHUT}FYwP;F82&;XX_STU&)2i=by;)&CZ;Jd;EU!K@?pmrqMhD0K0QfQ@;wge-TykFg zRHcWf^>=7`XK>VvXOK~Y`;gE$6g#Y{6=|F(2l9&Z5zy_4M=aR7wLu#5hQU>^uai)_ zM$*WgV+kP9EpSJJ^*5i7hiLu3N0zMsvTXkwS>^@17Bnwl$*hbj)QU610^0%^d`B#Q zh7>UtPeA zho`9$TqzOTt1v+MgC0t2CQl6}r_YsIFkzflV$ zQbnl&&1yqMn|V}TQ2V;G6D5~E4fP&dut`md+*H@Pg5b6o4eK4{vp6Jv`SSb~b-+g{ z!R&_dmm<(W+T=Y6|DvxgXGH$|k&7E{*JQU7ISz1G_JoQ#xfH)Xg*)w5uD00KFn*bl zNw+AVCV%HzO|#M`oTwYT6_C(%co1HLY()rtal%{v!9;+E>&d^3eZ@_k&_k>{x$ zx=85e6CmB8dY|;J?)9!(xX`W-%n`K~xZ2(_D{^oJkqRsGtQ2-OJn13r1(@4Uyyj!9fK>-ku|h-_;;Wgyfh`2Ev2#SPqVe|=i133eT} z8R#>ODDGNe!5pWnshueTI5fBGWfm}r?KUgo-beB-@ebH5`*#QKd?wH4XI;PYWP^`Da}a5wV26a5PDmlY-rLkC#BXYd54jJD z?W+Xl7u!Z52z^Ey)rDq6mm@P1-Bp}V4R6(R9M(J5a~R_Hp7q3oN{=rHB{}6!ib=9i zjW}#GVFYgyOC$Q!rV8#K28lTl4j1)Y7u zC5Tu)4?i^J74@5pITytEk%H1Q9^DkFS3nkksO&CXwGh-Lx*cta-V^Dvae9izPFJu9R*`eWF#??W%L?+4X%G zqO!WzD1KULC5r^|2K5>$i16Y4_bd^g99lIH32{0q#S*8}`becgvsEu-FclE-`OCNo zjK+~yiZ4NE;kS>O1tQfEXB1s^q#m`&2ih}bL2@KNkIG6zP?ef1c~sz0tEiGKE02(u z>+Wq2gT6IHOY&~c=8~UGQqVbNIYD`OS2C3c=SJR}1oY__-ma3r4tk8d(b3sphoStA zu=(%yWCyrKv@!><*-XK}Gd)O7Kd!Z_-9^o#2jyQJF$1COH{*mH!pGx%xWpWn${Lgb^^t0AywwOqs0#|u%N?mSr$OB+?N+^wy=!mjTX zZxxh9FsDXx0rQcpm97s>&+*ZBxoiU*gYzVour+ywf~`Xj!I?J@|LHrn+1Vi+=Q1-m0IT$YfpC-3DTwh)Lfj_9qm3YPye0{_yLNNBo?xSvY6P zeq6X<-(9QCI2%wzS!U!A+=jiVyeR5TZl2^YedA_HQ2(TUO zyq9n8*775fr^xR|DD9XZx7NUkKG)JFxjMb0Gi5h)RKL570qc)S_KI_Rx11-$n>9rcI>CiR$oA*dEC6d;1Ba@YeU`z*HA*87rEc ztBQ<@${xLiehTuXX_)oPE|}Y>&BOT=E}YoEURnCjS2Bg^I_9iY#|b2e9KNA*f3bYU z?H}Vxg=5bl_9U z*f4u}SMqA*Tq5i2V0#_lAdO7Gx`@UWnACJ#t7W`rayt!A`FXBc9sYQdrFHtdzqt6z z=ib}c)^hU4#f_%Ll)X85 z6kZw!S*h+d=C~SX(4;|TsT(KDLvFNmV%Sw!)shW&v{NS#S(!2C&?5IzEE#EPO5`GT z1kmHK?7Tm{Y!f|1n03g+Zo4e4QGLsU!3szN{Qv|AcHm7IHZ@ccvjfIcROsVkLUP+n zRIk?*vwAoU-6W(zDf{X%UsxMug>P&7W;y5C)PH|7ZGHUu;&*juptgR*xT*;Cwn>8P zncQ{!Z{LFWpS~q;Ae6>V<+#>A(I@?CTGPxYTdj6Y{&kC3Tb)<28e-Flz zEZsHa$}JdFFxYmqTT}gdk#9XjKy4rS7|{xGWbx|yZI7@1o@bTz>JmW1SA|WH%>5_U zo4-2%?6m#5A~iGuFOZOhG7(tkhVfMqcOxmCmI(pRJY6l@&`AWJqpf_;DZHAx2PI#* zq?Gb`Zh!9NQ+3U}M@}{wX7P_JdW^L-y}Z~zaCkf#5!ZYbB&9ErQLSq5yA#Xhw@!0v z!C|+V(^CE>(TV%h^zGkc^~i{5Z;k^C97TRF=9ULMCQ4Qwv|Qg8x_|3bZhfK989jHe zj~Uf#bp#$dKlwNanGY3ik7#t2qrNT047H9MGkR?!^s>L$8>$EYl?&)MIn%uI`vV!( z)^sHGj^Y|s2Ui+4vJnDFl(lwSDiMVa^}W~=rqyX0Wfh^P9tsKzb^f!k@PGHay~}P* zZPOQ#j@jNg5J(H#O}7YiD2B(iw5sQx=db-SS=GpW+n#Vk=yVglIex zYZiQh5=RKY=dER*PJdW3@JH%SYMgek9pU=)61l9jg_`Ew!YM88|%_0;BDUh0L4yd|ifgp9*0XzR5o!aI4(?oAB@s(6NE~~vSN#n0t;bURI5CfHzQ|rXZ(>&vw}kU(pJG$ zPvh+Jcmq(Rv2XtCw!&_jaX2=}-_jPV_44c8jViI(cU*#}%|w5ki}utj#Bw;T|hQE4vu5=3%$}79w)g#oPjIS^;)uBx}7HS$wpqg z^t16ZAt3mXJc6C-=;c`=N|~da%Bh&%K{l1Ft2{~9UIMtd!RP-AwDd$hBGnE%lBRuNk?msDp={g;{-Gk%hUu1H2^xZL2> z1x3;;%Nug&@no2DR7nNxz{@M*>HBuf=fd1a4Ex_#+aBTG4TeUM7(~K+S-HNasL{NK z6+jdUZ9O_sF{BjI;B(3&RHQ?eWpav$Sf?nc{X$P9l6@Nn23AZvDtqcKQhu>mowuuc z_7hHQJZEz2>nU7@q0AvQs|k+c(ZD5@OQ;DMkA(%qGYu6IeO!R7xQc#MWy>cBSz~Dg zA~{Q=D=noem%q7Q&Y_BR1uN%Mc-M!~mY$WHPnb}$@`@ZKNMl#H>cPMI^-HP}ziC+KoN?wTa6=W5o{h`(yWu;F|GpkZTY;t#21+L%BF^|4 z0{*8ZGVbJVi-m|ayhe}7p-UZP@Po4VP1Bv@g}zP)R;q?ai>T8AC-BqTE9Xe35PO{w zq?e92U-uO0NkG=S_V8EnDc^kpU%GYQ#Zm@CWLI&8nuQy>U)Yl{?X4^l8-Pzro_W-4 zI`d@H;oaMr;-=o;KWz*d#p*^h#ihutvj?e?*C?od4dF$T#ehF_Y=pED@94HSgw#K_ zf9<@3#6g=T9dz7MwI856FADhgynEiC47({N43;nVou z`yz%?dZyUCwZ7f+*zhbvDW6iCT}-(GyjAQt;}e}TUsH(s3{~RbKFOG_k!K7t6PL>f zVzqo@C;YbJx)^ex2_gbpl9_C`2yJoP9kxh3oZsR|k%G42CjICaLB&NK*k2}qWgKWe zld0gss`(5UIY6k1Q1_y4rq6aJ;?HksCALar(+S^aNa@?c?pzjODM4+%mq%VzObU8! zYl9Fpbc)o<4B5wI82n2XyLA%HILOM_1tSR05eYiInq9kj?(V#z!>K_N2>u~e;U=s_J zHe{=*<&#)CxLxb^xS4|hT_}n2a%pTS^6VUj%v4PgzECbQ!w{qv5Mvj)tqaeTE&%Dh{f zfIa(9Z|3xX=Ur7Yt-G7;)}j1!vei4jQ>f+p{$zKe`dn-V88XPXkfC8T^!&~G)HT}p z-K~+d#BIK3S*gO;%X=;8jeDmaUrY;9#=`#LbMn&e8-7R`NOOEclsr593g5AnOvO?n z83{$Tm;GLJS%ugzJNTMEV;s2VJl`+P0_Jf7^r6nsSlvJBIC4F$Nv2xVbCw1`V(kuDWkYi>c zeaE(%{>ePds)L{eu&P;6nJ!+~YN|sMigcshYGlVvL6(aV%%^O|*G%rMGTg%Ep&VqN zmc#aWM=>r zhX}YB`iFu|U($GB64{+{x`vuamIEf=O-))h@ca2GY&o5>j8oq-pDOVp~OwoRE5zXobDQYu*6 zX_w>3p)0nr2;95Ej|To;b*U(P2V!*>4(_{ub60=gsK#vH6KwkM zz=@L=m7eWoVlfgDCg zq}Q}SbWk{LLG~wpo)VGalqS||*W(?2jqm4PUY#3qp8m?}hW>U4Rj@~8Xo^od(Pan!%1(fGiZJw?dECmgD@JP&%;2lJ`2+x?x3IK(U|LfDhfvw?BbB&Ss4<|J@>pDm0rnOQ^1PT;i zr^&7z>V|142Dmp8s1h?Zsk;a&!JPK+%5>db=pec}K8QUCV4;6G0RA>;tLjm34s+G&NXyJl!?KTnlNrwm)=&RKZkGo7k>s$sM4)V zpTYW%$4!3jo1XY;V3OKu+K2LW4A#^iM6SF4<@AL7)u_>b1}PcLs^5bN==9@QVI|#l zkFn(u6A3eBc8W7C{cYXJnckzgv8|${fA>Hs9-Hx%;G~mlHg^8uvIbLUrE@PJlLnA^5Q#cpdkuR! zy#B|?nE5Dow8J(T@AJ2hz_K#3kmsrUY(PqIA^E|_vlR)N#cCA?J>lT_^cT}LFrZY-#ryH zhyRUEqIpWs&YSK;v1@sXDANsJEdX{0QY$Glu*nm6q4DGX2YCZygU;^4=W`m2o{h6p zwX)8<4sl~|1A>~D5VGx@jq>)w;rajI#d+II6`v=jYHw#c`&=OR>jljf46sJ1fh}Hc zkfe(*Z0;8uI<-cjfmwV*tz9(JO-&lX%>hjTxN z#$MZdit&`zrBr0--vWy6_cO?2qqVkKMWqGES`*y-S!_?&zuU137EN5`UqF0{=zoWk zqjjAJ;^UH3$mIEM=Xz~Qp%@D3n`6_s>K|S6KiGTEu%@=IT^L2hf+7l16r!SnAXS>Q zST-OkAkw5pl#YP(780dPi->@L5NRScA|OQ|0s>M4QbG^CLlPj36u;%(`#I0v?>WzY z&-u>#J=Zzcw|{UY>spJIHRqah%sIxm$9IZX22}6eBR##%G8n(6`0lo?_GQM|Ar)XgX7cLQgxrOX;UR48neh;}`B}@q> zg^m4i!LP|?wUKo`UHyp*9{pDxu@nH*h5-X2BN({xF?u%+uwF;%WC>%xAg&X--&Z}m zoN4yFU2^cbx=tbT1oR+q5p-_>RZ$gH27n<_=o9H=nx_PFyws~o$VKU|Y$lK{uefLp z3>R(fq4(zFt9UhOvL`cH3bxP1bhLP9k^H)DWRtf0(&Hm4Gz4>nu2`4;SGRk!C3w%z z3lhky_S+lWP&3X}imEtSB4oR%`-R>2a_rGS4+t-VhZ!%e!Q;On&EGjAT@^KgNFv(^ z`-z|?w@%UlY3s#5EmVItYDQk-PV&G4bMc{=&bzsaEp^)?vetWv2yzg?w1qCp8hm== zJ$&yMDc?=tZNB>XHg*2`Z5sUZZ3>pZq+`0{Aglw#S=E!tf>O0uS%;7__kXax865Nn zc9_e7*xm{M@~``D9#_qh@tFcTzI$7Ng8z%>^;TO@{B^cq2i=L%bz{^Z0uJzE4$B+$ zsmVG^P8(plx-)P4Jk&Q)Uv2nL*Vj^q_*@du!73(F(I$qFfG#MYn4-o@jOhJTus`xH zq(*ttchL=I?86b>ra;&SWX<=`h#5)Ehxx2rUO%7Z7JmS{Y`U!U7tiaeuMp;K%r_b9 zY}BkX(H^|`1$Gn|-9gsr)KZ2OO5nkrjj(2HDC4$b;1qVLtr;IF%tv*~YRmF}IAHPR z>EeQY&;2TsbEHR;KxFGbxIX_~q~fAc_r=b@7`&!o$?qQt+|#h_^6SUe zkRwVWau(8THpi`56w=*K`AO!z?3b)4xnQ=^5Ytguc}B$nP1~RHM`Lfx1uhwi6Bc6uMjA$F1i{H8pdNci@D%Z z0v-T(f8#`vgAfa;o|_|amK%a8=ENO>UExUbJt?5i(gfgl%oTi-YO8@7!+idYJ^aS$ zIrdd!g+saJAJZ0YglnD~FxOmh7Idwj2_E4QB%()ETSXed2YgEGP98rUb(%CGei0z1 zor!f@^SfoDi;MvAF>ZLx!lQN?K0BlwuXq5AY7gFEjyzZjyitSU!Yr5;(r-#ccjpCI z05R%-H)Phk1{dZ_7LWF99AC@mAj2YLzu<0t%hVg*Hp`Rc=tp4g_GwU@#M=p^lq%)P zEUTtP;>gn-Y?1l*uQxdY5aek=s!?1tYHTcxM>>ZFf<*4T_ z5~>G76_bvRiwyO~9d4qkq@>ruv7+IT&RdRJTQ^A(nB>Dq5m<{gw}uoIy$|KdnLO%p z#p4avWQ&;4et=-)4Qn8C54PbnRH~jpi9EBlk+(@Mowr<1R$Tdd(jD7h!8Ynyht4s3&zbm%;s^tYJ_RHXvQsofB{;!0y^o33H`2`&EA7z+_^WjrX?t zOf(JG45@&JDj*_=j9Y{JPD97`ed%nc*vk4|4Mc2UJJTo(9trK}`K;b;cCUk(=jk`3 z0^x@Mam#pNciSruPgg!>N@O?erUuXh{4akUtuWVq{uo`xXuA69b9d%}r82H-z&61| zX{a`?(Auo}rNBPp$iTE$d7!95pGW8YKThzN_5j!$xRdm5=$Y7S1wB$e`+CMa>Ksmb zZHddAQgnXAA!=*%F1r+yEq)U5Tjem!7O=T{k3K6^x-p$N2cY$U9nq}9HdhmRlrjwj z8J8_?zKN5*mtgVuT>FJ5{K)jqyULbBADeZZ`(An6b!$hy_EZH@gjy>l`nZs!LS7nqvsiBvqJ2R=0>%c zD7Wy+imS(NN)9~xnSxt~?4=@E2MV;l&1K? zq#Um2NI*Mu;T^E*?mgURTYb}~n`5o7Y(ZZ=T`1_10>$sC#yNk}8NK*4OoUWR*QF~o zvzv4Au{

N|jf3LcNBl1JFx&$6}@lb@|hA>#!}gtc;pgms!vO*qHD!qEgIh!kbaQ z;l+&98v;UQmAAb&pRTVO3VJ5LFBpjReoPOHcK%=o$S$O9A)}GUypiEB?pmUCaf7Yz zn8uxi&t8Q-yK3FG3vO+Ed$(4KxtZMnrfSn6{)RK?7*~dfg9F(jJ!6dP^?Jtpp1{_e z>>cbduVH9R`2}8{q!}L57mSD0&$$4&cba&lh%d7~(ZiC_QGdM(aM}+|MYidJXRR{k zH&P#wZ>}ra;-xd@BfTLTeQP*rrK{wGh53g3`abLPM2lJ6QyBNy;#a>cxAcwcygh5K z81Z5bPKj9;ApRg0oOlylUZ+$mi(3~(_6SzreN%Ze`gN4H&gJdVXa1f{S4lfvT8Y(y28o|JZvdNCfce{7Q?^5e^_!M{f&(g=8 zNOI)hOUjHoDXGYQsP`9;=p~Nd;yT#_k_P>jL`BC0;uxN>2UoSc*3;t zIzVr-7zv*ip{C}I$aL6WhwmTXLEIu9q$9|e2$WHVW`WB~DUFu-#~Po9G}Tu}XS&=6 zH&+H#MsZ46jFx79eb@Lreaz+nQ(AL0y5Z?TA0E?9C+6w@+2rtlb{+2mgdYl6n**lJ z^%gf;@#=SHnQbN{;iTnClO1jk2YJapK!2t(%ecuSZE3?vD<^CKP@wUzxxxFjEFrE~ zd-bj#$M<^hEGRNEyy47Bt=LIu#0>=H*LsxKrrVW*(NAKr#$J>Tg`WyE$akPUmG0~A zkvq_0ud@N^X#DIWnrW#wDhY1KgsclHuu{(~fTV%+TELsMokDcG{s~X__a4Hkhh5Gq zVsq=~sh5ZIBw8-7=!cHJ)X(e8N6>5#4F57z4`?hBFs%#`mQrZTiL#l)04wcS`_04w zibhDo7*)OCkX6vyxmf^Md$ueeh85-zmZVw*_#Jes{jmQL7~d%z6yJpQn0l;_rFtG( zH>{<9VsO2jX0*>k^$%1z8uje}yI-mYoO)SNX_$W~Y1+rNLAYw_>i{{Qm`E1bkPRJk~fzk=l~ zcU$~6<$`~mVDEKXZy5w;I4v=3bwDxcF_v-QAXZ!!)NgR9k z{65)1zng}_Z<#yFm^B`$Yxf1DCQ3!#;8c;bZ*S z95ov*WHMR}J@wX_%gV6vgofy$p2P*9)C%i#76MsvDLSC4d#d`Ib-KoDW;)}3gB(>m z+eNcVE#xkaAYi3!1E81yXam42Xh9C<(k2uidoW3xe7Ljbc;A;*2?t{QGc9!lx18x3 zWQJBPeG3_}F2?n|3&HF3 zlTtA8PFGi*9HB?q-<=7d~J1y`J#Weqn1Xm;u7Tpc7%GY~W}l{7W42Vw)XLN)R*}Q*tyPmER7d$3+MSYj z+oA%@xsRD&0*%pQNk)vbHRo*O&^s#PL=;strmWxRD~A0W%geatiorlsi&lM#;hOa5 z@tHTEgDlXbi$$4t<1aU`i<;o3<{Up%8!iNQ5UNerrtSoFKb%e*y!5Tx~=N~wxN4HH20_*j7Je2liT?-1jx|R8@uy9!Xnvd?Ie7Fw=!B zO^zNq@|>W5zh0}k*mnvh_ZHk!TNaaMpnc)MW~(Xqyv7x%2~nvH^_(l)@nc2uUh>#4 zim}jW$jkSES!&ox9AFx3@VguZB`e$F&anCiULRYO(~M1S8-EoaX4(c&k5olG7=no_ zmZ$}>zUb`9%(T51pUO(2yY@%lPb)nUrX7}>ANX>BnlmH4rb}r4ng;A~2?z_g9qy@G zjnZu?87-K}t0PxMh^Q|7K`OEpUar)l)Nh~g!lL&1jOvr=obv(AvnRJ&;MaI*+`xQ2 z*n6gucp;{wk>9jttFBdud`bMGjms4coyTfcyu278L9h`+wRCw>hOpjHmJd#M^^fG( z;edeBo z^<@bAuU5~H3^}qaeDXOt9Vf3$8t(nDZe2y^@yD``!NU}Nk2lauEL~48JGu*U9550nr#-oM4V^c1e0aFGMR)mT);OeLKL$*M|?8 z6c23y0nVDe67E%^C=+Y5q<(P*DR0!?B&BnUV-dsP;cG=`pmP1M59x`>Jx^gyACh*T zb6Ew7_8T8fV;fa1B6O?D+azuS1}s^({@aqvmP>08OBvSxaun6Qx4#tUE3 z%xTOBXsR_KU)yc;YR0Ec*5l?Yi}|mEUA7>z0L#P3EFvQFBJwLs=#!k2W~jE@M?07? z$GlV*%wD~y01%D$Wn3ea6r01NA6uB-q;ht>c!Z+ZXyj~`Mwh+%?mg12We}EDSgDR7 zBbW#^p?vCfN{cl+p#L;q+VqeUkM_zY1<1(A1)4al36xOtb?XKg=Zhmxgm!MS@7g3v zRGoH6TyYtMO3;0vve+S2$WgKWOzXF_k2^oue9tUHEU7oF2=M3Y`%0$G?)j%@bmZyZ zQ<32)Tes{lP&i_iQ_3^7h}q}+3aEh0c&5p7ch|3fygdEd?#Fd6juy)49ysvX*;81R zXom`uKR{BAXcjxEU3#DE#wsLLa3~)DtarmbS+*~QE#5tmkr-mhQ<)62(vvRQ-APs4 z^o>J5X@2W8(08U8<-_+DQ(8o5i}BSltJO#cUHXf$?qA{X@HNU5x0nCq1@xU<<;`;3 z86dt@-vT90gAPwpbId~Te6*|!+iH8D?+N5)@y&s}s7QH#`E=Q+X$}zKET9(eWy#P@ z4N`%%N}Iv7-6K;uhGaqJ@9V$0E&x{ZWCiv?R1ZAtIL)xAhb2z^2;M{GGbHjv?w&A$ z;~pOelx`Qdovw6WS1qTk7>&0BJI`ftqvemhjo(G{wR>a1bgH7%v=3u`V zi?B2~NVMz{*0LENBp~o*{BKNezemyi;4R2+nMw;RoJ<_gcC-jvVhgN<^(}C>6Kl6% zffXtU%i47=bOIa1{_nys+^}5#m;?ciiY!` z_dxR_?^{D{3(+6Zmj?jVG*%$k1&E70{V(~uF1MeQm;Drc5bg>-av_EoYEW}6ympfa z%-0AoJdjF88^EWsU?;fsv#9a1NL9>auJX2zG1^w0^@>9YuZ^UAwmx+oj`>Eka zv)Ir;Fw=az9A-rxv~iCEfsLm*X>>AL8i;MWzVlzzd-bFk0;V*Zt_V*}0p+hOV+Qwj zaXT~eXn)K3JaZ?;%qTn@k>AvLTZk(hZ=pYQy3V9A7CFETab8h8Q9G0(c$ zB-Mv}rS$l)X_7_;5Z!`tV3O^U`dT29mB_-rA8e^R+VuSlO@9Sd3_DI|bNn$~Wykf{ zHzvoH!cID*2LWV%PmEI~S_5I|$8_O+Hb2;c3m`XEmm$49&kM#@Rg4?*9*^audUExL zLY$h0djh2rC&3Af7OhAHjQ_v#VYL55;ROlR(HeWn6X3h`5)qbyVSP30>?xs3g{XQCYd zs9yX}4Ea?LK!3;%z>qHiNG2fw7yfjX#p?XQ)`BhWGa!kBzctKh4_<_e(-FXy>rWQ| z4L_&~0LOQ=F_{2xJO*W!jVLf+X;Mz_|s#{B$Hh#y1Hx z2l3=8yvH)82m=U{KeaCB&ktu^t_ixA!-(0qM!E&*XqW{s`hRYn_kZl))gi~66(e{; z@HdTu+ceX0AnpCf_96cCaDm-C!M6~b(ls1^_dQ$8-+ll0xc|*4{cUspwmJWLz53hU z|Gn1!x|jLy)>II~Ier$X4CeD&j<{A!jl5%T=>0OMk*v&z9MF;btg zY>U)LR0*sO)|km144U5FVG!4Y_1(Cbo-d6_lx>tD_eM+Qz~%?+tC`QBAk#MVCoChu z^)_n+AVQ!Pt%*~&$fE$)mt+Ba49o=a1`KG(2l}w~LAQV|*jS#CruzZ<$AkUw#$SH0 z&6yMM%sSv7?_w|i>GXFXZ1d;yrVO<5ENdf>Zw&;H;JlOuf?e9TAs(j9=nrtl%bREr zQ%Ewn#j+X^m*Ee8208&7M{eFMg-?Nhh9%)tQ3NBDgrnOq7Eq4N8t_~K5cv86-}W}1 z0Rmw9AR5Oinv7>GltuxMv6w={X7Mq2ESf)B-S_vhcn-60ufT)xuZ{dzNz3pL+A)$o z;0Wzzz}0}~1w7`;pbk`58{&;Gl!lXah@fvgS&Jd1sF4(ULnSo>fB~NH zr;{I2pPi}xH18C34b9rKMP#3V$u3q%k*Z03s9t}m)P{wv zM^?KA!?aaci21pJ+gY)}zRH7{F2jEDlT*NAZA&!m^DLn97)(4iK$V!bwQctKn#fKe zuX}8E@{91>41I_fbRBKG1MsFSBc+xCyHtQ-9YwmYH=kyRp_lQJw^1XgQH~q#N1^^@iCg|tuorE@Q3 z_A?}M_}9hCQ+A^SF2F9Znpu(+1WL=ZJgfHk7_!Yb@Mh0N{rq~^RLK5sct|+9!o3lE z49ZVj19-B>C54kj_Eq^bj)iH$in}{yS0q*-;h?Jg$oPbP8AJgX#*hC@J}uqvl6L2* z`~3Vo4=-LBeZnbeeY-z1a@hx_{xi0xrtebVUw*UK^#a#LLG4AItz;7^CrKb+ZB6-11&lz?~UCEd&mLi^? zg{K5dG_j;dS0Q^FWSq$=PO|*bu4?d8;`&xN9I666m6>@0^au+zR_+|>)bJ>b^;#VCEZ-o=o2}TXiiFatYST0)n?;O4H}#TDpRy-6^T{WZessbKdGjD?%=M z2$Sy{@-!_IyB4n*hi*k6X}sAe>^F^zL37px!wKcKV1^-)?C&%KfK5*hoW(%^iwQhC4FP%GDOs%tHY3y9aS1YkY$FPfP2b zKROxS(E4~*fHs%Xf1$yW`Lf}Hf9MxD?;1l%ZO(~Oa<(UuL&NeO@0z~hFm@Dv9PEZ; zGha$MF}XL9ev53}bV>&pszBY#36byW|A21+wP-Sl@j=w*G?mpSg{E*$j(1)gSx=)`xo4pvo{%e1ZkqEB5vB?@a!-Xq*l;*YXC-ML@3A?r zmoBSRzYAW@^~U)(Hha`(25(94jeu=Yk#=JP@4LEMUe9G#6z07aTXCw~7HtGY%K4`l z3+6E(ONz_@`dH|$-Vhnh;CIrz2jwSp))3!5)_r;>Rp)rm9NpYS0Y3wC=|^stmqOY; zt*r8cCPwcECVW%a)|FHKevbQX6Y82is=e%Wjp_Rp{Q{=?tNB+1fB)&_$7qlYDg^8X z@F!^JQ1^ikwOnZTxkCp+OD{#YDI`?=^Mi+kL(qr8u1H}7FT=RHZ+#BulDoc%ai47A zR)4mlzPjYE#J*nE=kB=5gkmbEEkTtnE%iAC#H@2~(v}$Kb97PuA(-1$0s%AZ!i@gv#m^5oIqJgk@pffzf zl#HAMfPF9{Sk!-p@B3dl4|M^!L2Oqc9MYA_Snp9-cMhvo_o#ZC4J6{+gBf%#5Y`Ih zu@)g5DagAecitFG>5Kq?f{Y2sFTE4p$2oZ)N|3p@Jk*Arz6XOAV8}-0c;{pJBP`om zyScRBZKb1nC+K8*yItl%>xzRHv%PtwFxdpd@4qxF!=148+b}H4q9XV znGj>3n6@)^qG_B2L|ThXH$}r~`Ga1ClIzkw3^LiIy9TbA7d}be>NFwcxb}-5T)UXjks%?6b~B zI|Awu&)Hi5o{gs3M4yEO3dU=p{fyAlB?+QiLk{zm#_9Gc)N|h_9rVZe+XvuNvh7u6 z>bg*}W*k{Ls;ayXSxiMX7s+;rYf0cD>p{t@3yAU2iwU*ug5UDbL^zRlQW+}$EARXN zmTTcZhW=gN#}^{Ub8iIa=*bCd4^H^?pl8F!Cku~yLd(ok2aEVMd)(SjZkvVx;QVX; zGFzuog{$E|*)beJ9KhCc8~9!iT6=8DpgV+7%K;YxNu*-}v+MX+f*b&30mlS(!cjSY zI-sBA{lTU&j2;@VV!_++G8(5@9mqosBK~QMErFzz)L_4=F4xW2wY#C=3EG-MMMEqh z=oP|Zvou%BG={Tr&b;LfXSvVAiTjCzh(KV=I{g+3Vgkf<00AE>jL zIf>+6@x4-0D{hH@9XxnYkbdw`IY+R4+p#wqAijIACV{Q5Zcz@fE+^p6eOhKsKTuOJ=`~?&uW;| zF;%5U_;R@aATAoW4|;|O8|vwAc@ofWhwh|hc+?euJyPh|AK-uNwm`OvFdqpys1kkJDkUdZVLbz07Q!?x7=MnUR zZ%_9yeGNDHGn;Xk9eD3}7UDPdnm-&Fn(0Xm&H3`}3?@yoCY4C)D|Lx8kVfz>s8Qvn znsNT~3D|0cw+NtRCAYW{{RLoleMW#&05=WDe~A6*=%Th*ONp`K6%XtCgBkwJ#U2Xy z14lEZ=9t9<&4X)d}`W*)?9zlXrs=H zE%j5~A}wa-qjj*YVILgVvhex-yziAo9v-d$gVM{(A^YbZka&r0&*3)u0MM*+e7S(U zd+{8MtPh9>1Sv7(xk7=Y%nm^L!ma(s13dw2IWQ-Gev5!*gLoeubEp^n02LwVy(;w{ z=Vr`0bt`ps{5IqvRlhM)h@h7$oG{&a&BYa@xi`H={eI#=CTad(PqyP7#FvwtcV}Wt z!H0iNPXp^if6+)2`0y}_M(zgxJ~hRmy3DkOQ&6KJcop=v!@>lzZfs6~W(*G>@HQB& zXP-tjgE*lQy@t(`C#vY`A06VlCd|@{%7p=4Ex`W#Vayo(>BzV)pU=fp^=Ngzgx*S> z`$|yrdmpht8Idhiq7P;mH+`bMxU*#PzI}k%eb%O1A=~H>_t605@vpJFeF#U9xjkc)3V!b2wr*ZvPGaU45vw?h^g(r^ogAG)-Ru@Fb z0UT6G7-lAYm$&wW>qf2vXxwg_cQd|-&JLBYeK_ikv=e@vgnmVZde55198_2BagBuqF1G>zNmw3@~NL*5vFS z&n*7czM=nxd3!$Z-|oKe=>N<2s=g9Ufc)XkA1V=}N?M=L9=c;A zAH_W)%y|8*!NHh3q3;Tp)_AzgX=5|+)-j>oqW$uGTkc$YX(u(7mpZO}yU_527}<*E zq2|lvMvM*s1jMoxJAR{F2G%MY-%piw2TPeXF-r8gDezPph1)*?!>7Hwgt@pQJ@C0; z#V@UXErqfvaN$*KCdak-L8PXi@P2|KYF1bG@naFy%UfWO^d5@H_lDZMn}(6{6*n25 z;sh&K#I9Y@VXjpXw!TnU@F>8SU~dyG zz(6zKl`*)d>Mu8%s|C7-Fup*$F=~n9Nb|MTtvjnq{;8wjgJTTLlH1W~$NBA={1P2E zdcbL*B&u2C+YIZBf131uf9<>21!W665S6}nTcZIGP4O?fUyZF?tv6ZPcxgWeP<-9- z<)sVJeW>eds)T8%5>L960-^@TyaNc*1c3}7(dS-z2BtAXo7Xrl2pf+M=@Yb<8N8wM zIpUaga3kBEHwT%?%y7nWuE+)Adn^}Oude6$>+&h=lbE2xR(@I0V!buAB3xXkedh$c<=P97Z z;poyl9o>xyOdoo<{OZw>mCqqVjC3NchT74XM9|un)5doNF zuMeZFRT*DwZ(Jyea{qi|MDX7_6LI&x%YT8)mY`<3x3R6o5k#7u4 zTfq$obkL-;wKcFivP{-en3&bjWac#A?@}>;za@Qt1(sl)f7WoVk1Xj8JAt-RGL2Y| zSNus?i1dgQ`5ra)C~_OMnQ|a1NwBO)Xr`ZzaIQap?+BMYqE|KGm)_i%-e@WS7yU+c zhT0HFq_J}oLAlJvaoR_%`~vp3+X~zHPf6DbscxflZ^|Z+jZv(m@6Nt5D<+rkk4r7h zTXI7tUy0uUg~Sf^N5ccb2Rgh2YfS785+>iEtxUT?UG$4u;^W*c_dkA4>sU~If6}GZ z^Kc-hO`1)Iur;$UCo3T6Y}~7Sb-%M^8}NGTE07f9d~vTQ$xAG1b2ed}6L9~P`loZ!&XhEUH#53HG(Wt~Mmn5yz;|qA%y%)H zrX%5NIDj;OlwFNrizn?Q-;O6!{o6Mx9EhUN)#Nt2hVsNvp4QvJApOK%bFu9DM^o2- zr=Z>a-##Uz%?Ho)dAF0UwBFm5NSg`Um&d!Q-7Gyi)&xGTX2lU!F$R|k3Ypeh1bdhv z?hnAZN`l=@m6bosol!XN{{i4dJ@j8W0{>n20{_7NTU?i22flQ}V9)>}??%?la&bmf z1okAyySeK@rc39c1w{0;Vou2fh}-ev!Y$mY;c4%KImtFlfd?s6FK)fevm~ylS;mR8 z4ZGTopzPCc7ivhf=Uk}eBYY}5$@6j_r%F5*>D^C^JAmxPoM51LNIUwK2rn9Ox)3fw zPkolCrR)C3)7$WWUb@XJ4vHLH`OmKlx=-OM8eo}HF0)^VO~V0^L>FVAZcCB|Cjq|6 z9}dW68z9pH25Q58pRRxc8sOp`Dgw&IEC3&46`0w7JX(J=u7NV7LB_i{)8t&>EbbCh zmVXrxJ;mny@%w07b+C2*Cc5UefV}Mx;0TGBqZmYYV4s1~{^R#b#FL$R{du^TjGSY8 z3E?jRsHYY`)8N^U)y`Pmj3WcE8VBi+*ThfEbh(GpE0^` zm+UQad_OH-Wye9eoRlwK9^98k;o<_+95ahMD9JzA?84D8o=T_(00KD-w#VP=`sRj$ zEO`Wr?s~ntH0Ots;6qLqNJrzp?L00Dxy)&3y?808Q@`13K{^u7K|=F&NI4j~+N_TXhF!zAMYUFV8$`v4E69Q0J z)O!DfrrPy~zItV{#;|8rYMe&!GxSpKR3-Y9zPgHFjkX8?prP{ zxWV?`-o?`-UI21v?H}d@Is$M~++7d;g$+&axQWRNs9HNY#hyhKUu^M{bxkxG`LM|H zuX`Vagj}oE6r9Z&hmZSca4@ysnIF zHY#RyJZ3%Y{LoYBf6vNXd$lUkd$=M_aHxqpEY6=1P*Ug|SIF@;-;}tnQ=`*P0_SJ( z@??iZi@g{7hvM3Q#izt3g>?`#4N>IcqQ>uSOClrWnb0I_(oQ(!JZnUQ-<=zXy$bVo z4NY42FV{M5jR%_6H}j(&#pj7m;$V^G%~XV4Y`(N=@D3={XALN4XZ^If*cPjkM)plw z_8d5Gy__-nT~k(xK8JB1}_CjFf>^3 zlJ{R>!)`Bz(yE3${reM#(miNy=*yc@3XSwjE-y!KEiO{l(S-LdFyY8H6;Id?HhoM6 zVs=YV3|r*O{sGl~cE7T^Cj3_B**;0N8#$KCd_@fxv3tA*PQ}08kf3c*OQwQwku3QJ z!upJ>jNA+VIk_TVQTk-%iiZjN%WRZTCMzEgkOOO1HQkmhWUd6ZF{ByDg(o{Ed0x3I zM^{cD<~tABa_?aA*e{spCg|hPLk`6`@Z9jT#;gsYq=2UQ>)5Sjh#RaG$;S~|_Yf*w z(D|{=)X1`ucIgPnca6}NNgh^jr&XXc>7sKT)_4c`@s*)* z=zE_ldu$Dhw#%vC{xqJ70jbyiZB+puNj-zZbt0GALA`fC8@ z_B>V}AaI@qq^LR}pV{+Z$62kAO)KvoY?MVHpecvl__AyqPyO}<$~7Q{;-)qe#~N`Q z6*+z-UBc^a8*gc>$GjX1#t&qxg``htWg?g;4^yKw~3vh0}dRRR42s2mh4DsbR&BrrXqigg| zPph(*!?b8IH7Wmm!^2`LO_}1yeYL*>_RxxWV+I%2zym0v3ZJjxegJ(QalVF~YFtK% zbZ)#bk?`3`qI|#+)r#b1O@NO3tn1DYULCe`)9>I&k@6QBy%ZnuaUJ~_A1`>oWY7;I z9LN*yMU6s~_yphby`oWnhgq22aEE$;ZbW@Fdw|tmea)sw3hZr$^z}p5&lhf6Vu$W@ z6q?1eE!5<5@Y#DeIo~3lh{b+E3!`0yM87C)Q6Dvu0sMwwtpDND57Xs%BM+Zsf1w{p z6W1T(n4QeloG30W<0Q^s5B}U@(R|>F#A|(l{hvXV+op4TpP;)#Qm2kxjlv2AU5a z|DT#-fdABW{_VT)uP2ON8Tg*~`=bB)F1UL(%3Blkv_LF-+ExfKh&TS5FUTM<>se={5~e!c z_nD*d3a*n6wl}Rsh3c7_u+ZCQTL2-nUDp(-eDwQN3hOvLx;E@Z1Kq+iM1&MAGZE=b&gH-6nsKYxKP!ZOEnu1n?C&9ZfUOVI}2bcyaRUUuLUB3xvWSn2v$U zKu=G81;oJ4U>#z78j>>Fs_+k!v^~ZH1(}G;t(P=}eUQJv@?JkoI(WHoM}#c>1$MH6 z;1F8uEphdPgwaC)AH;nrFcDD4@>}0ks#rsmFseCRtJb<}N9!$+`(9m)CHYryaayn3 z%P2Vb3`vPHSY3sLjh3v9Mr5fEtk+B+6gFWzHbS=siC){ZdNYR4jGKCOe_pVKYw93n zan1N`MM~|b53^CwH?XtDsAp{Frc75;){xUV*^JAT19ny)aU0{{)i)n^;U*iJzsCBN zb)FY4cY>j_HlYVX$u(hi-!G#3Ng7!|Y>t-8ocq%PlpW3Xzk4hZNk!xy_G?UwPc!vP z3q?@w10a_T@snZ48qe7}cqY)IzrPx{&8_9rbCbJg#Whbi`Sh~iAoBohv!+mN?37Yo zAg&E^1RM?%brF<^Yq+81!D2X1<`30Hpz}TqCf)LQUz>};i5VM)+1KWf!BgEdqV^go zcT%QvInxckUee6ve#`C&=)M{Pyla4^_#E$Wt-w?Nz zCdJYNV4c}rFnSGD!s2l*33_`-8>~>p>go!XXeTHsll?oYuc%)rtv6+GP=!UK_@$hW zmNs-&wP)r-kbqVvuE`tFWd_B211|J;c?PYTB`uq+xzhX=koMkX7C4k(SGruj92f?B zd^}d}rM8~-9Y56a9T>N1p16AffcHWUbjl78U$Hs|{j5yK;qtM0I)($r=@K=?rH^W# zeNuBc9Zd$u0|b*O7N7ql{$?_GTkCWa1ByFa?okuf=ti0`rk!`bm#=id4S|R5Q%M#N zljX6RKJT-HnFNg>XL7;E>e$^V$JpPGgop~U9G;+?a`-Si6Tm*_FyroUPVZ+@#~8&x%~_TB#?qsFQ=k{7wRP6gU|Jc#TD?qr`kN{|$Fnpx#y4gI>r!X_G8uhjd+788p00O9_Cw2A zAIO}`ak3QijdzBHPUwt7boLfw9p?}`_sB)(xF>2`tprM=JK;KmiMYEV*TTZz9#mJl zal(ZRn%c-Xl0v}ie5_(SD={m)(YAxlGT6MQ=Dya1+aK9Uch4?VL4dHUZqePSYg^SIz5HZ9d)AS*-c8#N`Ef z4l4VYM8)TaGJ%xI_MOU|bh@Fl+%kF(iIzpDhH~^@K3!UDuGi0cTr`#iIc^gvl&5B>XRn0xmdZ?o=kp6Ck(aU`1 zT@-|gqNK%EZ%H0p2e$TYG75C{+X9Nnv%~)4PWeCzHopzdp>}w7LU{JpkUYso1|(x| zdGs2{C-N=oEBF+kIRGh7X<1K}GBT@{`6k285#=_OvIIJX-98U_3DnsizQT8(z}z8p zlVj<`{duA9X*GeevreVfcC}!gx?QuML4y7#Fs(As7d^2a?TuPufzJSEW(MAd=39XX zvgC9hJ)B>UDg2W2migU%g|C)^LWAv5M=^kP?p|z1ZreTqPnjO$xbz0T!Usg;2)+OzK} zYY9X~v4s8gTNIDocP1aW=o8+*&uXxw_4>GUequQQjMS%;YAZdq-)+@j=g0 zE3Xieit0;pqilTa1h@n@=v}x;br-S(1gmfe5uUiZ|D8n6(b*8~9v%P#YV3H%t7g_j zBbVL4YNRp9<@}q~*K_RSeM;7;Ix*4C z-Jr3-bytFd)dhq#*IiFcHVuJA+1b1pTaZ_vQ-HuJ^2E|;#pG$KXT;7(4_=H1(K)R> z`O@;uOZ{8Kdg)oEbuNOnX1|-1Z3yowcl?3NIQ(Nc-qh-*_@^fhyaKhRhKMfa5oQ6} zenPY@^H_>8A@8Z+`X{vy{BiV*UG$zpnYW~3@NoGTs@vIiTkjmor?s_OjT%uxD#l!fw$JIiAfsm zFg^d2zLLxAzi(tLa>PvXNE)oiYcXbEdsHmDy64M3pmhI-eX{@P`}gcW|4r>?{|a`L zj`DyiQC$jFpAkKI_qeRqY8np5$3c3S^Ia)^_bfZc(ts*tK=!_YB@)B5Z);S%H~+W} zqpnSNIgT=5#KD|R*WYj@(@%+3xR|YOF2Ty1l)3C8$cO-!&7~_wpZ8ej&Y-=)5AhL; zjSxL*wz0pCmSqPgRB30%^Cel9!valycYR(Np^|D}Y2Gq|X=e zt})l?gA=*)ld>sWZ~#jB>#5GYsktr<1{V9Sq7w@Fc`XetmuqZ}p(w_S=&8*)YW z!L*5Zp;`4(8?o)^+a#v0x^Szjwn50yu9}QUUV6z?oUl4a^9gg?8_r+z5tQc|tV-Ge zI?#J4=65IErkjOOLsnV z0b43!Gl}j0gE!0$Xs>S)K75CX#z8p`0}*V;mSa0~Y*W(C4>o{Af{pOTZ<4F*|6?Vp>WMi zCZ{Bu2zFP08@-LRTxYL`D=;rAdyP%y#6%CBfJ-8$u;U7*!}c%SouhH`P`dR5!O!Hx zPHwW)`1mY4v$wTQtCow?Su%IkgtGGFktk^okYf~`<{X*w66ep1zQsalQ$aT}rW(q^ z3vSmk^goNecb(Un*10<;vubEcf!e0kf-}v_%Qn&LIs`kIsQ)jMPOHGSU0v-sAKhX^ z$;S@l6NH^+P85Ig<8oD5_f|fQ-R|iSy7l3knjj;c`4P5PIuZ{A`PB#v&{E13B{opa zvr{gE&##=@<|Q)plf0yhZ}^~T$ll3A9v%78Yq3W4Sb_$okhZ-AR=;+7&WCqLmwt-X zd$z~1;go-}9=|6^7Kk!0yt#?Y>Lzd2EdQ z78-DSxB)=#PAQkb!R^6Y?l-89qWs~y#}Q!I8AeI0G5A7vt}LN(5p*VDi{pStyC-b*5UjpMX0;1-Q? zP0gcPrT1X63%GHog@1Y7gUb-89?yAEJ%^objVXniGgF`r=RazZ=WpVKMQlzFd-R;* zbVprREYMFWR9mjPm6M_czp5~Wtnfo@1LUSq*#VNoyM0>w{~z++JFdxYTNg!9QA7{~ zK@p-Npdg?qy+lPoK)Mj5MMOYAM3E+hL_t7l5fxEd6a=J3dMDCBdY4WJy(W|p(meB9 zYkzC6?_1~YbMC%-t-H^?fA|YOe3|e2&N(y37|(ddGq`0vo}4~6&$XR3r2?Uxwfz$J zG-Rnx%PKnnXujJgQ1?IXseVf0%mdAbQU~spmBT^}iuSxXt*XV&5RrNd!5bSMJhLVh zDRFnI|8tfQjsJ!yWhU)M%C3l}jSx#lAY}qyM!!vLvglfQ+xFfo;`w*l8%g0Wez0k_ zWjj~7^hSn+68E!xkKb_JzlR-jG@3Uq%wCw6VYdEFS{u4Z2zQvm&SZ8*0G>bmwlOfDb>e~Fy<(&%>S zZX7S_AsXVGBt?1UieI@%ulsV8L+j~A%BxT#H8p*A#C1l6s{2%cx{X3@xvR7KBUWFp z_!Cj>hnE7O$6FW)C`TwCLWgu&U6EAisCLdGZ>;cR)S&UU&28`esL}3gu?5?+d`td@ z9&Dh&RRK^=&%KNrT5Y{i7Vc2r-Cl-#tMZ%F)LC`D#dXA4%7m->mx{gF`D}YYX7R=c z&UjoEyi2pH(!5KwP_O9lon7!)Xwmj^Rjy0o?HX%kYhp(vnu=>eEyCfR(Z@dmfbX^<>z$8%8>STPuP`t%Q^K@i`Gl$Y743X>oBg zyDkvn5swx;(jv5kBQ>p%rrJGfa#YWy$j6=u6224d*d$VLO>I$_)2FK_Dmb^(7RXlF zE!nbC!k+-L9p^A3O~pInA!mw8w;&}2I&WPlEJ%DI1O-c)8)^h)Vm ziAd>Y%Hhv~nut2Cb3fKmB6GMG&?5d9Oc_@!enoxMYV_Rqy&nte6g-XT>P4HjDW9B= zXS?w+@^y0tBTy3bvn2HM>NCtXFk^674{P?Lix)bQ+l)Yn|-HQ*4p0e zo6**KeRdL7gUfS&B}0dj-goK-x0F;kAJrtVwT4-Ht1T5#QfN+;xheTXnGm{MT5;^5 z@m;$H9;`@AG#nQByf$9tLzZDA%C$pop{HQsBI`o3ZUhwrp9A(tGe*s{`EX;r8HZ=$ zt%Z~=juX!UobM-B7WDrOqxr8A`uHyz2l_AXvHN$5&wsW2bOPh^FBVrre)>5I8ikh< z``+&EIy$BR;x9~v}wqJEQHeTf0$?)x&_2&cEE`VYHC!l9DR=hIcT=aOZ%H3L3cZvX*qW#@Js8<0XY zviNHd&dXw~X}>f)n}~0hcC$Bo`=dBWaAXhR%&zZywl43=(|pPn2;{YtDZv@V(f9zp z6`ynch_WMjYPYYNng)mgK*BcVfT6*o2tHZiu2JVgpXkUNdMy^qs@En zYp`C(UezZ7q$7B8;#7_nRKg#3$z%G&gXp(kS-dy1V#Jcav=J<&4J4lo+)oZ``;F`L zA7WCHv832n*@#p4iiyK6QuzlRj;U_l<@MaNCD#X=b6>%e41@&hK=1b-@XbhV_4|;z znMxDROuwK-$=;Ur?XddVu*I#-MZaUV2c{$&-zk8LlvO0&k? zGfMd+6q@?j^9vxL4HO5ZS}$1^Rma~?y7yLoHOcb> zx&pRwhl>P=`#SUAPB0$0cl~0y2+znJG5$c_GZ>FJL)0&Mrg|vd*Kb@cWba+nk zbEPgqlO$;pr!cWwpcXYz$o6ejvQARtR>d2dQPQGmSt45Q>eag}i>}mk=~@W^U*o85 zoq(qGhftm+^F&yGaX>eX-`6g8YfQUdmh(KKDwMGmZHQWcr+TfgEvEor5uq4Jy)`69 z_5oFHXI^v85U^w%DR}B!35014ZCZ@PBJvXjt8%)rOtLMytR3UoG8# z=KpuH$lr)RN&GcAj&g!uUTPEuI7TE9dT^K>7M=@2QIb?ok}aCE3Qud7c>L}o$bUcE zU`Xzyk>q3LBXeyQ1`ifZJ&E3$2I7WJb(Lldw^D{CfD;%3)tN+SE_B2?p>Rxrw!E0n zl_8ocy5E$r^OAM2XkWr)f8wTio0A5b!chvs{kg9vJS53WWFgXV*NOeTnlU9U>H7df z#?A3v)Odev*hkgJA6;J*`xy4 zle=}hWuazl)WM7JnWHKPRMnijlA@%rrL;CRGm``-p*Zz!icbdUXXooO_JO`o0p0G- zU9EmXxs(G@50G`F(Y(3Rru<>uF+kd#`6#bUlO5Dl!4^TB6o}-i{{YD%xM-yKz%OJ6 z_-pjiwNnkhxI`atn`4npAS(%7@C*ByT;1}e|Kn6|##tR*>2}BqNTJO4F+Du=)xE>l z4PqWW*^}@jSby~4-3#sG)eo`rgH!xthc&DW!U?Bl#~GEk6mL+#2MJQQ=R zdOsx|uUvZQ9++u&*Bf` z^X|>3b733XlQQTD9(7?w{evdw+{b03i~Y78?8mEsR4c+(ajX_pqecaVxo7}Jw zJ4zOpD6KVp8`Hj1^1LkP94@o851oT)hi*V`f5rYlQ+$sCuF$nL=1uG`mg6Z*;l03% z?yC63@>UX@&sph=d$lNq$xaCKMAsbM$rC1VwkgCGe*B8nwZ$`gvi5fm! z)J{d-N7a2V!cqBA>kcSNKZHn>a%Tol&_BYNa@VG;#W#=~kf7V!^c^^b7u7}!fehjB z=M|tXHUc2Mo(n2NNM4J>?!kpZ!u1QoUdeS2eI;c5=Q5Pm;2-}mmNs4WAP6EqgD=#RzhY$yM9%*E>q>g)j5!8;`(!xwq?bMt@V+p+n@!nGS5sq^56mKf0U{BKOe~{ApFvtgdPhmPOR>{TK=$1CCMIz1qaP1%FPw%gb3M%LiyF)=$5m@Qn=pD=_ z zh*{1|j@Ze@zpV(BACxyy0`Rf7JtNQAMf=BkZ!tF@mYb=Xn(WWc3I+d1GpaH5Kx7;q zbe!n)w+#qO#!37~vy{rrw}ZwL?Li5b0tQ4YV11u!(>we9B-b~z8Mxfp>L@kCE9;-2 zPpi8)6*svZaaTaC)(zrZefvH&pSprq_77GC>`FkMJaF4|xZ=Xo9QL*M?~2sjU;6T2 z`Z1Ylf6h`S(-~VO3<7FcGUDaG`$)g}N&m0^+x0K*8}$#c78W`Pa!inVYS3cbc%1aZ zdyR9_O9?D&ET9|wCu4p2;v+h%PsQnHJA)@S_rhs3@n0-_@DI=u7&!OvQuaFGb>zF< z_cluGJ{CMbs!mc*!#TRHt1#IGfudczAaew?w;nV<>egf+FI$hwFX#DKGs#!GxPJl_L}H3MKdOrlwYCiOQ+&CW&jFDOzd`0+P?(mAEq0OY zL#&zJRE3@KRDy6&_-n_xHc)us{G&(9|Eed z9+6~L$$Gy2NJIkX$2B+_dKCJ=@TIyk=>(OnPAM`c{&1e#tpEzUA+@!utfBg`y{v(@w>w6CnIjw(3U&zwUd$$OkA zR!qOgy>iqP+9}tGV{&d*W`?h13ty+l%sAKyU6050KVPvC_+hzWJzE3ozSdNVfHUGN zPmxe&LAM5H4xB$W@|sbL#(Od6u{IRcWR40z;Xbx!Q~vLhBpQCWfX<@FZw4 zx?uuBInqEr?i0Lq$zf0GHkY(Q2al}!deZdZ5T*eAvIxb8IM*YWv1?DfeluZM*!YvYldiq-e%!!@EZ}c<!XZkQILPS`B)iS`a;1YEa&U2qJnp_59^~D*G216du^-_!P zl<$eEG7C%k=|OKycnE-$ZOZApzDlGu>$o|VM_Ai zDkWAvDt=B>m}O0ce@ByGW7hDwg6*5K(XV~KaK4zl0F6W{(N!2Zpsq+)hb~GwZSLK8 z)Z}aR?K*`U1f`BEOE$1&G%00DJPIi;q)=~}S5e@tXec2^Xnh@Jc6D*@&2RZS)eYMG zM|*4x2XxOSml=R^hO@sPaIBy(!mZFoHb9qDtjRu&=|UFKm*{Ska37#6O+~+w7I+u) z?ljd+Q)38Ku>enplt3B{DVp?4kig4NNl!E8+P~A1Os^9%Uwd%nAmt1A@3yzdQ72et zggfFJ^j)R+)9FVxt2tVgPLgCTtN1?+)9T2=6axf|@OFx!r8=}A@Ldhkp<`OG`+{~V z;*x{2+}5c1`=x3fP8L4Ker5;!kSRXBE&!_gUM)6Cpv^y_c^Ow(w>xf9$Pn1VCn%lxd9sRkARur z9)=v8Itb;uDo%g8w!zQXiVOw4o05pq@U4iBceX8KboAV~bR&&pua1{rz+7Fn-bMSq zfuwLJDRS!GO4Wi3*_qqWp*cVaAiksP^`4)$>B@MGD_I?lB9vkmB3h4rwOcfQ->uJR zn)Kzklw-*oYaUJAm~0pIG(0rq{6qfbI}(#5P8~GX*ZldHwn`{r^RIIX3Kh`wj9%lg zuzA2-bNoA?;=j%S585>U@DG#($#_Oqr3CXBr6K!t?oY9c0t^zzy=}F2)D^RxQy|O2 zM(nj(8dk61!9PT-%niEKrc?!#?3yaOScz$iV;E*nE^CX)yxobVAK9?P>7XW)#--%d zPY)n2iJ+P_T7E*8;@ajwPk{DJsuNkCBtr2a2&2y`YRiUz;#S#Rpv~gq3C&>gPl<2q zxUmfC(Wjy0Qr+lvdJ)tbDSQ8mb;H+s=z!Bx>R#3=-$Mb3q=S0OB_U$_FP&tEk?!at z947&Qk6mADJQ2TGR*=P)s@c9V6kVgZ<{D(=-=vXAbW9sIsBrg7-!1zDxT_{=4A+(B z1fApM>-({t&G6P}YcjS-^*~mnecizJu~xf$wF~wBey|~5{zO-+UE++H9cUMke)anT z$2D8SO@=`S2=eFaHdt~s8k6>SA)cn_C%clT?u)W`os2j<_FHSAs_Bdb4A55V53NGWBBLlbBw@BJy*$g{FvcN=9tKO%*EacJzM=JGy%k;SIt3M z8@GHmK0!Kkcj7+EOGtObL+{m4Lf^p9moFOI=1wG{-^^n(arnq;6;v17FBXv+@CqQP z36ZA&PQ?`n9SweYBKsFhejapVh0)|hU3v=wKpQCi0P5p<4&N4xezBy&dF3t?Vl|l!tG0Z>!S5E5DJ7ezuuV*ubCV2$J$`hlI zrCjiCS3SX@BFc37Q3&gY|9SfU^aPBUS0EsBU**~w1PeZ%0xm0r>Mvl7SaQ~+Ziz!`-(F)l@YZY z>8$z_4tW+UM}1CIm@JMa`qZ_j4=W@-^49t6E(YEulI51;D@ijTAet9nzw_T93i|sn(1Cvf#C`Yp~ zqjb&q7UgEm#9efDAa5}T5Gp>j>j-z{&{s2Kd!lYOeA{>Q=z(or)FV&qXrfr%_3j6! zdkgWrMzt3*_p8snZAU4iI1qBABlU58WZ+f5Zt?u`uuiKg9 zlWp<{5bDky)f9Gm=z2oYax|mbY81$t2?yaPc%7d(TUI_c+|wGUU_a{dTfSY{OJ|D2yRso=)^*GMJMGJqzD`*PzIwI2wP;m>BBj~(Zd7Da#AquOC7iM zF2=F=eqUw=#<066R2B;PMR^z6bPk|~_pYhx5f56ty)K$w8a0}xsCqNLobXL<2XuS) z-LA!OF~#fr2`nPdhn~d*$=A&^t1o7B=7 zAQ8}SQSnoU4D?;U!{9@+ww9ykkXMYizr6E1V$qewn{dYcWvW;op^Ns5MT4XeiqHnu zNv-WkZQV(Ndx=KSa5aMQ1?n_{inP?s_U6e+oylFv-+L%==2GL9&{3SMoZ)kLp-kPu zka+DtLovwCePPx?#+#KTOLZPbp893VZrZZ?*_a{$J3`C7kn{H-Qx+|&=bAWrtK@Iy z>Ev2Q=TYG_IRyVeHOG8xoP7FIa_Q_)!St<>5O1|(<*T`^U-C*i*@Ir1Ts7LuSzHvMI2A?n8D&EeC0NsQLa}@99PVk;IEL zWaMhoEM!>#SODHNzP<~i$`L?;Kq-qHd^b{kYAZ{EHNlZ z`gjKwf&}n-%@yc0q8GZ7&I}K$rZL_@r^DvqI>09R4}QlGjktCompf_B;P0le)u3{Y zYv*)1e-DZwyk&+8;sI7vSQ`k77_WLk4*X&nnOF}g@BFEPm;OMNYeLKUUhmZ}S>EYv zElQ0nm*)&<@W2p_J9iFH^q3%({Z4MHWZ9nAnJB5@B_K05M@-VqXvt)+H8Q9u+p-8d zdExM`O>7&6a@H{Da!RVRuy`Q1(2u7hKb>Y{Z> z7_MJ_1KUw8DwQDJfSPY{oin7IWi>$fr@hi%eb-XJq=d>eAgaac7;^(PpW&MMEL{)L_?P1biI8~Jxm_M?28Pz&d<<6Igmb3GQLjYE z<(>oscFF7LeB52!3-LxK2&P}e8;5&EMF(D;A9uNtlEPXe9D;K*yLMJ{a#t7${%nGM zxtiCPv-9e!CtiV$Z=)-u+bSR4xgdDrIQa)%@KpgS_im~jG3U4Vj8uKHjal^41J2%k zcvSb>l?|ouM*7H82z!#GZ^0C2diML#0$x_;?y%~h9-I~P5+#($jR_r2N)vQFd4-Dx zMC$@vNy4k?CB!wJ#}jpRGO!()J}mqbD3Npu7INsL zYd)%kNdACsZ&_6#0~#yFV10wj=;a|ij-t&JSO9+Kz|Avb6kW<{fa{{FUYlB!i%uM zr?HJPhs48~LW{!(Bf^6gD~uuyFxj?;;vySyvhSsr>y>fwT4Yq_5mz7wq}~(bziqOE%eQ)k9aeGlQW(aWV>p z6WenZg%(s%&4iJj>MJD2*Pd>U{?g6bP$RC4E$nqhw5p}58PVy6bG%pEZewx6<#_*H zUFIc?ZjjcvLs}P~ORvl=6mo9Sj}e*Ia_<8sgQgXpq|_RpPs4QAXOh9F*K6bI)~pN0b=VwzGS_ zv-{FA*TAEkK3GHUH&d>(-exLT@O zbrD&{RzeaQQ0ROWxg!3PE=1WiwS#0sxE%L+m9lOpUHc$mrOg1g8q(nt6+_Fg^LNUy zZ1TCkbjE`tP!caQj3Vp7Uk=JrT5BfC2RJ#RqY$HeuC_L3mYS)t$q_q4k!O(IP&REc zuStvhp|VrEvnh2@`%ReY3Ro(X@)@qBxg(<=F_ zz(Kf;Z+u;xPgtQ!BI_-5TTvN{>@>Rks?dHl?HuGq#7#yV@+|#0=~j{`@w%I6gxd!v zQ-Q`?l8C5fzq}J+Vcaj?hL84gD>Y=i{|REtS%T1FX-V#qcbq_ zU8uwrD-lFq#?Gy0zGQ6hd6Wx!SW@fx1Bb++SoD@r0BY`Iu9ZzC11LgTed)>VUb7Q_ zcY6P`c|SU4H0K=Nf#16=^DaU3NAwu;`(P|cD01iKNIjR|aN7Xhvq$s*&H#6x#jrNV z6*3kM&XJs_nB_W}(g>Zu+jhgmUBRN7wOz4sEcpBUGR~_4)~zm2B$5n6+&#qA}%t+cRop=bhD^saAt!H`IH@2;H_deNZKa2}4oDA<>E5rqrOC^2g z4)Hwl(Bx{PqldEpEi+X4t6Q+OF9>kx?>zw`9n1^d3gA`T9~^|{Vd$>|*qGzc&1Bu3 zcqTh5-p0S)GB!9W+U4O0L~Fo5^%Fx^20twhZwEe*+RS>Q!gWdUr2Sm@ zpm0pcG1~md5xP96WUhfirxV9f-=UlLit11`S1VZq{GGl~I{NkM9AP^ zw;(@ZSNaRNcH3IHd=oga+;~HYF`ZsAP+dntzwktz@I@}w`7NJz)%(OMeS?-Q03w|$ zYcPp~BI}6HdU0W|i0R8xBp>a*O0jB|>aqhU2WHLSS=WYz5Xe7oAD%o7WUhg3hh;V15ix)DRqZY3*F3ng`=V+LL z;*3yJM3KXppz zut+ol081IJDa4ZH5rV#B4+y$q8}f}$o1GH)yy4?0od<0#nPCdMs*sNqxT3>)t-nLP zSP!EFQg;^`M>zjD_o3vRS;R-08HoDd`+@EQpwSy2Vd9wj1&TkOKX`6n>o%()dQ5X; z_nb$@w4}XYL37SG%G3V1bbXrw%gYWXx`$y8HZa3Jga%A4lwI}cb)Bs7z*E))!P^R6 z=J4{dC8GbWzSS-N1+Rk(9B*%WwD~j5M(Bo=k65}9Nvi3k*N|T27dPv(!@Gl-hf^rp zE*oKtiI8=H$@@cecONPw@1D}-NAI$k;-#Q$jy>yhtHj`<1$g^>FP-&a(Foa)fY`EYyH# zh4+mfde4AKr5-Jctz@Ez=TWL8q{@;2^^KJje=GAqpYNsXvGulR3VKcoCOJYW^HQKS zV>wkUe&5^Fk)KwPdG(cCSTWrBUo7{~2c$~~GXwXI*(hKcg;7_RR^ZvEaTiEdvU3d@ z9K~T3TPPw|m#WhDb-Q$%HF3&GP`0ohw0Y-$W>EhtewVrZD5s(DTZxv5L<4W{9+VPX z7$L7JNm7Az6RgkZAA=Cj_E#C^_|623(R?Tt3s9brk`umY6L5Z5C%+1C#UE!zv^=&Q zIbe|Pi5`$c z{BVSxMY1AaAYJJ~UT_|WcvO<)xEAobS=)XIUcr%U(JXQ*7%4eruFqbS34~r;v^UJ5 zDpS{qLxDM5V#omH6_4r!6G^u_<1wT3E8QK@*wHKd`UFNOU&VI@55Q-uou4J~A2B^- z>RVh*AghXr$z^C|d%Mj21m>r61;y2LBJ|SL#EEcu6@SUAQIy})@F zq8_tlG#AHgA`YrknWnlPKJNksfo#%qdm>Y!wHg}{r>C8{1KdS80x|jD3~l>+=E&eltxWOresYfuGMFBF)EM z=WL&eAe@LQTF+yA1ICQ!3GfMJn0i)#^Hfe#g|OPYnwnjXq*b9r7Qa=V3d`E7#s{FH zItWflxbU9e;G_&)o_~cE1bn~$5fuLocOODMhFZ6r0JN(?7+qu3X`^WkHSPI}B|`u^ zizYC&*9Nq6v`BU0)CP0vGgqQGQ}7`9IUsAr0uUw#Xk3j!h(F4dZOyp8d0jqZRWLqA7{M6O~IhR=6%;`u1{Gz#@|olcA@?_5#9f@ zfkR6>=gSiz&5l=Aoa4nBr*`!Dj1>2p|*qlj{5Vx>+FzS&`nv8Xeq+|Vi8PX zo?5-ML)gZxL#WYV%==8}G?LW%`F?Nd8>{;*!u@`NR(f6sIPO35+Xa6<@eh;w=PA=z z{_9f7Y>DvEyXmNA2DN!B1%S8z_udAzs3p0R0)z>pY0KSL=;e|DS1zA<)YRsZwaEgT zH2jB^`G=W0ynT#LqD}0u{q3b_s*i4X%x4FDzr3|S=-@N8ETZ6lYSRQHDc%E|-pmQ3 zLrT~0%Y!;mkFBbhO^UGaa92C6+!QB=vY_j+w@OqyxMUKM#pBkfjK2PhS z0K3{Z$yw;IJwY8ubkfKlDD;OEHvpc=<0M~{HVS1u!!evP_iOCOKF?5?l`uJFf*f=DK68d-$wz!WA$lj3St zij0c=zW6d)g?3s{Q$Mq|VRBnn!VTdGXEfA^k~;ytTT+JFWmZ+n_Pklmc1*rS4`vsX z-zlHj8Ea$z5Ec3(d{E7Mwh{s3xcgb%R(l5{2|M3pjEFf~Ob5q}$d(?tmD2y&7yiAZ zn!Q)rtjKjN=x)&ZvhPH0PbB~`gaM+6poo~1C^v<##<~@?yr?%&8krOE>4dw`OL2X2 zLY50y_ZiunlvZI1feI`4g^^<^4U;)Xfu@wZ6-_jLI zPRbB7(RiV(PN&kinYOoO-fHa{aMYVe1CZU=Lk?riM0CLq+2e|uzDv@08^fT=Q`wXW z8_)n=@lZp1&B?PTJ@=x|Jns_T>}2=Vwk9Rz?F;{GB^ds(g?$(6r!(vIoX@tgqlQBj zQ8g7S6ddF zHk@lF?^Y~gv3&&Uz!sv2X5m{%LE%Ts?@1LCOd9KeZ!{f|)RdRo7Ry4iDEH(ac4Z!z zas1E_TGy9N6+@a+-rkC}7?74pqxC32L-|6KyL9gAEV|5MoaI)g<-4Tkyjj?pl{(kt ztweU->vzBwJ9}|I*J|6-d=2Jo;?&arT578hZRkgnEk%dgo(Qi^I{dBsH0_%tgtijy zhp}8)+7W+Q0LdY^46cq;Cu1am0U{$O1rPm3UIkiv|Ce@(>xKyE| zR4F~Z;;P`1GcWhc(Q#TMXUsQ2S?NKVB&7;rMGV*ZIO0`*+E~2U;gJewZK>+@n(;*C zr8LabrX~TaTE19)&QQY9XVTes`=@N(HY%yB5^R1u;i&-|+0LXzKSZ%dL@2;0DNe8g z|APx|UucU@v6B!%MbuAwyQsiz7b#Rdu3flf*Q{+%P+hC<%6#*A>9m4!ok>Lpxot?O zoNrN%*g;B^Mi7-pjV%>%3Vxk_VlZ^g^P|D8ruWdbWPg}KkgDd^gPO@%#xRa*GP3q_ zFe<;(HNnpYK%$&u1m4$*GU!w(?Z*|Z~Ej(2)-n6}a)`x<|E zxLhPP(tC+}lg{^9&bPVlc{@+}8$SeJi9J)6^BFKrxbZ{v{w#1_f|BZ&4IJ z(NUB&qLcN};mwY2acc9iQ+0@Rh@w{0cqbnqqtO@^|bN<%7m4e@NZ@5mx%_(0?d3qPku+L-|ND?5!5OPTu^T@}`LhKj6 zx(fAOG?ZDrF#VU;gD1eCJh=PlD8KNgw2tljT&)5YA_t=xQyG{Fo6E%fHx=C{iY`75 z&+o^8RE}#AbbScA_SsZ2h0%CuQwdwsPl3=3oWH}H(wnh;81+#?$@XGXFsSE{iGr|s zrGb(R$xhyT?j;Vv{ztZJ6}F$}sHdYsU?NM!zmY4;G!YDt)tdvFcK6d}yAHZi9?#ai zcPl&3)@F9K*_vcayC`|&+aiE3C8pn5dOKO)`;r^HaoEdlr+ZeLf!_f@eZ$%?D`DST z62ECrVdSHxJrABAX?a07o1S5M>&TSCivDRfy@mY`G3CGIVqJ6YBx0YuUFxpN?i@Fi zg|2aRBnSwH%26)L)T4*v_Q{TnQU%2o2AGw7giAhjDrHyBJ1aEmsMhcN)~ z{>KaJN{hItC!p&$P^3lZ@3#wL-2V>fI-NDZdDw~6L?(c^>+i4o!y_+6f%*?9ij8@d zS$GFEH#Z5~jQ_=(p|_fVkBPy7rved?hcH z?{8bu^o;wd+H>xy0SAe(WH-`wgVj%F!-Ik0r%<(;v)&Q%>L!Z@x>Gm{Iv))JicTZ4 zJgB~vrI1EjhfdpZJE5h;gF?!k`Xzc+XI~hnUy+>MFg$GkgYY7%YGOy2)4nmUMEbb$ zHfq`zRs$Vct3|P0UMZH4_Bz0_A-&&WEb!-!-Zz0mQIhi#XR&Q0i_Dm`MKhlVloMKF zUXF9dV}MsJnebh5eUSW}a9&Mv%aA?4(>WqZOz(zF17IGO?aaOW@QcMMIC5_#W->~; zYMcConC1I+N-I%Ndfr|d&SaDcH9fZ73r|7ig$9|cNUrgXip%aGb*RUm+?y2RO||VlJps&r;ofmB8*KyjR+*e zO#!eQfG1yidqhHJMW&gi&fefLC91 zns7n2_1kEdZDr;n-kN;u{qFbDLgv+mnb4$-h;j?->+`K?THxPB#B)rhQ zL55-m=X5Wsc*6Wpb6sOl_H+sH+k8J|h6u%{rO>V*bk~b+x0E=x*S~OrRNSrHr$HB( zss`bEV5o1kf19a#@v_SR84F#{S998|g>8qxhc9@Abr!PL!DZ zoUO-~EfdOyKqc1z^8&F7{n8E9KX$IkHdFU6g&wo;LhXhdOkRXq)!>oYiaW`C3kl_t zsDqrWelRS>|4I*C=F8RO`&~Ony+?}0^z0v>e|q<~ikTQF(P)j|@Nv%lYJo_SP&%d( z%#sC|CEjxx24n14zF_lxyOU$(K)pw|CDm51 z=Unu%iYXUS72*@v@Ur(l#pK?By`no&&f(p|)dfp?ixiKlye^BftC*=VSCy20Qo^^= z#~mGHo%hyYe(9a5xzk1+kyfvMlEO!_JjS#mpJQ+SAYgaq9%nm)$_zX&L+HQkSsfnp zUKXXh(QZG9s+00Jc?;|Jd@R@Z#pvX&ehu6xHGm!iro# z*(<`%L#_hHLK901EN)mi!f|wcOC}(!i)d#^n}&ta9Vw!Ct}tPNviv?bKIxbzymgO; zLQH(kjXd~I43-U^BN*~5Ao++hvw4|qrvmw=laM5K#bauQsgE@aH&xQ!T4J6<4ixFb z@IH-_@`0v;!_AAMoElo+QU0`k2B#J7;~o#x)2s9blyVZJ^lcMP;Pd-2kHI%@ST=8T z_V(bCuwJdX`pe~C|7-)pHaj(yh=AMfF{_+^igE2V(Y3! z^ZO^=(5yt%_#WY=fp6GJf7cef$AwwkVL5`!UMP(?=F5az$y0#6#uN6=(*F|Gn^;}s zK6PkcuX&MD)IH$tSJZ5-O8Q1bz*?dCT;YHc$>sd&R9kWFl?mdxFHLp}aMQsJZ@XYm z*N(UI!_fUrc4HE~=|hR3x~fDax@lbz``mSm$wR4FE@9QEfL6pZZf`cus#Of4 zg6P)Yd`q5ZGk(NB)njr_Jnn29k|{~q3UeJt&j0~1s(IM1=H_Hv7P@KozjYtts_Pud zO9gEOTRqPRMDUIPnfbfz$PdUCt?uf*zP>{z8p0z`6FZj%sv*GIhESxZMr^w2w;#r* zRBlxEGRDH$Wl#&Y1lk_@m2dq>2&(qzTJfCDeC5V$izL-Py$pLTx6o8OQBQZ>FoU`m zo(L~0Y3vL~ajNP&*iWjY1rFQLNW6yK_9AO3!0*v?RbSKXUHhOmdhncZa_NG;=2%u@ z(Vgk+L^O!dD6a5fI5Q9pxV?YLcx8+se@+ydg`ce$uv8nno6w!E>w#4ls>;!FSm)6p-MYDwE9~xx+i(Bl2X`OF*J-Z|3S1`j{q6VM_5i)G?N0o(^!k6YJX#sRi<)cb1d z-y9K&x>W$|I_aj)yYdZs-9q8DARH=HQ&3_+hTBJL?k6s9np{9X)HU8^2SqIbXKZ>4 zo8h|AN#0Fbd7nC|Q+afyn4sI(-PsOiF6rl_e;k7&*M(F#to&`g@(BCcw2D0@O`ncs z>jyV)eVzNoVk}Ne-+>V(n$W}PQ)w|ISK}^K{wzlP%~yDfF3ls1`K}U~V`@5U@XYap z%t3XoLN{+v3|si{Ucl!kZ*Jgy7p%Y+;%&f9=Rp%U>ClZWQs%lpSGaQOSYCZ}L$Rlj}E*dl8L z3vh`JL(7)U8Ks`nwa(9@bjqyc(XJ@-p^ZH4CzgpAMv85~6!a*d*^70eiRe2sFQt_Et2jG1Q05v|_y9d0Y#`^niyB_!`BafCwVpY#Jo?`qT9too>fKJb=N{Ww)e!hE6jW{5^? zwlv6`WVkNjS=__sP-*sLI)2_@X=ZQ@l67ychbZ*>#{IYwiTeNK;%19n&iGbLlW`3n;H;r}v5W0$gY~$;jR zNK*8X@l#1jrkIj?YoL6EH@wt@SIoU3l`lhg3@kWr!<2^oVfljbj}(sG;`)lMcjrbn zV~!(Fz!@cVr?7U{ynh@QpzYeW)L3*C@^l}Lnq!KlDal7&j7MRu`zz1e=0zXmllzjA zL48ZYkkLf#+Qie^ansg|()+_$PCQgHLf(nk$LbVE7bDGud96Wwqt+Uz-A?a`xfaj% z@$L~~5n@?^KcKLZA(U5BE4FG@Qzur^M=rLq&a~nJp)(1eA`VL~bb?~K{4dI}b@5yr zh|Ayh5ez4ad?t%`xY4pWA=<#ZRD<|@A4fAg^W+M1n^_@d~@wXd@pA%>Xr7X09Op-qq-2~E$Gl&;m+lo3^~0jHDvQ+?4CDscU=nS z)~46vzQlA-pKJVq;m}0)ka0DV%rb*?WF_aoZ7GXacA{F7#?R)sy3}2_1@Z2U6MN|< zUHJaX{zW|BRHjaa^c>{O&PrI3o5g@crGxu^1Uz!D=Ty&Ldt^uFfNP%K9?h$q|uwv3%5jtKEmm z$_T9%rPLXrjYyTDH;bRVX`gAY2vk1vDJn2c(N}ZaMB3tG>y7Ih6GQQMB$gCD<>9LQ zQpWzlOs+M7LI_A?Ps$p``KZPOw$&zb?}x-Z1wcGkaMDqV=_j#ZeAZ*0=NrWq?{`2C z?s$mrHdZ@E@A{(hYt1K9l25ASl`r%uiR8|0w2&m3dI56`ufK%Bs4G-1dY}|)ly66d zU^oOJ>?8hCc;(4-`P5Uu z^^`9Bi>mu~{r#_Js{C7E-0%KgU-WIPC{)HjbAvm*^*d$Hn@osOgKs`Gao+QK!B}yD&;-ivYL>sH)jh;D z%NxM@l^>8erlxYX?PCf>X6H-b8Fw^eV$+0^91g|k=-U$>LYK0$LkNKuI@OO_C83#| zq7^UwGq>wgGFubme&i$f{AW2s)t>#z$-!F)b+A4}UDya3{!Ik#74yz~=UQ#zC*DqP zLI*62dm`j>`KHp&C!EENn{{wrmVnz~IPF{2H(Txb+qUCGM_Q^(X_mvA_iecJv1IA% zv7t(lILwkfsRPO<-$RCILpzo$F$En@jz_Fbi9ec$nfH}n*H@@8^yA#%#9()pOVA|z z3ftyNIz2vdd+YZ~wmP7_>PqUmj!!KR{p_b#sZ6+ZMG3mwz3t za1FiT%4()r$Ru?oKEZVKz1TxxbG2>*K2^?MK)0(Rr9e}N&;87*Zm2@czeAs#nnE|b z;hMRhxf){Ry9!u&`LQLp4=anWV>o>2nRyXAm28>kyV2LU+xx?Ay-mBi%nr6iyQ)TvsvyJn)AEhwC0kTlMS%0|_QLQNE2AuXmhhp=W=NuSAmDEwj@Q4>)} zKuo0Kyyk$*>KDXWPh@Wcdy@t;l_+lzo8YF{-00jh^lL1wqh+}ivh#@>ybg#|@Q`|ANX>0hq~ zV+H|eC!QR{7NkZVk-j^#8m#@Ni(+^DdKH%OBG*rS!jNqUGTgw*`b=bjMC)S@#Z?Xy zs?c3syuw}MdiuVD0w7JM_+O4@#7wZ4Ff9?x2ypOA)Q>uQ21x4e97iv*!0g-)SRAk6 zEVHM!KUE*RhOm$yg68B~S0~9y@W0&du#O)09{LB_SYHrZFCJ>8R&lqi5Y}Wo7I4q# zNA_OQjHpI^2H*ZpoZ#{Ne--9$9}lmlNF_`d{%I-*_lSQMz9-~t549Swf8c$WVQU*9 zrto(j7dV|C7&QnYj44aj{gd<~{xlfP`#1A)lNA4f zQvghrUuTZR9|k%}y-G(_b0z4pgp@()s>4@X<=k8OO8C0V>)MMpFKdZV_fL+jN}#5+ z1?zflOe`tHzYRh9Ww0%DDqc=S^%W}r;YzcKk$DwnP19Z`LF&1 z3I%+}J4WqeanB?dEdjiOVI6t{?lZJt4iFMUh&2Tp7MW2FtZUiafPJ8Tl1{3CzlF1( znGl&UZb)p|R(M5Z#YH9^aUB?9i!BQ$h@JJ^DegJ;G#0=zqFOT`(s>93b(T=-Yw?J0 zhz4R3GK9Er0j%*EZ~`0ZXR7oAcs}A)7$5g127b92iot<_6=$H4kxDpY^#?=tvfz)@ zh)z1k13iY0gJ>3TCX*J#DTn67T6eHdyM}TFbrKot!R73~^GtAU8qt2F3iMlHcGL=Z zc^0~DSltEH*c%-%+Fp>c=DqZ{;&^>vUG!HA(hKZ4dM@Ffws=6^fSQ=nE5p3;ZIb~3 z@>m5auPlRSLj89PG=k}ATdWe>BWoap`P(762k8e6uybY@0_e3@^)qho@^j_C2(CdUv#x*en<(Q!Y4M_=@qKlM*C z4mRvn+Pgxi!t^_QZeyZ{L)QxRZuq?cq0at8seC2Ngn=y3{*S1=r!W2fZ0R|S6qH}Q z8pQordPWN*@z+|m)v_c!E&FVsxZwcVtyeg)#}CgPh1H3%3>gRLuV~~NlK{C3Y89FD zgF4s`Y0~fJU5BSBDOMP8!rOn^95e_7G0X0KOD5QP_C3?`nJu-uf4e(crL>% zaXMluN8+xbpUpU1I=5n25q4pk(OQ0j^ng260XpGE5nssi?HOK!KWsUw_8e5)A9#`e zM0;NvGwNuu(G%hMVr|bQ>GtyLbj7MLSG}iSGAB;FrkskdG@^@4O=%*X|$?Qr%ZI#MpHi)N_4kAZtsmw)t!M2ZxSU8o&>DLd)v@p$ya93 zTG>A)*}c$@Fc+_3Uu6q3)g6yNMMd?8n@Y10z11K8TJoP0RA_7Od$Rp2&MjZEBejf ze}7lzZ0BTouByWBuMO88Z`e1&?LUaSimSHYA7a2P+Jl#OV_mr1$`UXR(3$!AihWor zz}uzGab9xw-BHKx*wV$X(uth2_<8&ox&>cDUOmzm#PI@~zkV)snw0-~kB7;=zO2yO zxES^bGkhE^f<2!w+=FqAj~4mPbD4*+l*fkkX|pRPBRN8ohy^~Nrt(}ta*WE|st5RC zq;`JiDQ@MYqBHwEPi@=g6GQ{Xo*erfFt(opuxQ;w#M~mNxa;;o4@=*0q{YvYf45hc zZ{^%2j>Zwe-XEJ!8Z1hd9=ZM;x(6(}^WuMN)@`mnbi`*p@hAXi!=I38k<7RMD+l~C2Wu=u1Jt2evF!@6l(_^ zAkEqf(?*^RUOwkk7(qY*Y$u&h+x&iorC-^#sh&)`xVvmw-x0*FJ|}CvmFRa%T%jX= zkv)DSiA`o3aZkBl(A&fszAym~u?i?5q$c;DjU?)^YjCe>P7K@X79QC|y?8OY?6~B)7a(S4 z@BjbZr~dE7&E#8jkNTLw#eMj!XrnVFH_yR1Gz!NMbXe6Io?()F!#7xmfnt$<*(^vYBI;d;P)v&2wlAkSE*%0-+Z>-@9XOI~=2PdQ!g) ztf($DLuW3z8yAV3v$JrVXKJ6yHC9EP^I&}EVMx(ud=e>G9Q=b@$7JufWMX-&wPeuc zY1g>~_RpL^-M#u0;$t^TV_c=H2ynXcjx`JhkvdUxw@&bcFEb;Ahjaop@7S1}154~a z&ZxQv;Sdt@-Y*0o#G2r`|5?!I|K|~zfoF$>t68kRrl?{KPlA8-LvXlb^C*K z^2V@vTMKm#YPua$}_hU+wUeHA4$DJOqpIJJce(>O48-cG@;-l zj~YhCwBpZ^&WX=5P--|Gb#v@f z&SgU$RNErB-gYE`3-H+i@+iMmB& zQQQh>cXjRV^6Mc5R}fhBm7+3{|F929sQuIoxcpUxlL*H99hW{B==|4|)oysu(x*av9$le{q)ztejN4y*OG`npSJ&nTLWm^Bj1QoX-1F zHZ$8VIsMw}3-u7RH&`UlNSoy;w0eQ1$M*j~UtayXB)Zl4=UvAUV{cBuhkrH@ky>7s zI9erCAU7>weV>WPFlAfm{E*2y{RW%*Sz=yL9QGO$57CvY7hH~)?0F}l2Zt-1)AM*F zd(fX0(T4J5iH8}_YiXS$9|+<>E$rLODOYkoL!{7K?KRmr-ed9Kb!Y!*7oBbEMR_}o5 zGO7Z8<)5uqa2{MT5oVA{aa9!Ss|)H-U#<3Dv_Am+gRi6!4rBrsWFN+iAUC>bZmh%z zX1#zuRFnag4rz#S!!>|e)=@VoTS-|A_eJjH7KPOLOV#)?F!DNwtmRF%^50#0=ZzWEuAgE15wke!O)D?ohIzi-kh#Abb zTpZ`pQ|AOBAJQa(zU3Brk%10cg@Z7tfGh|t>y9w-Mfl$hK~Qj5IS|wR`h#K{*jz%) zbjA~6sT1ooc~AwaK3D;@_Tu29aQZHH^x{0?3F`vrXue26aW(~UR^veYF_kMkQAt?a z0b?yH{^;Hp3^^sTY9k6wIU|PbQ?|lvP8@{`Vo!hEz`v0Mpx>>5QEelZIrezlm7~`J z7`-t4UOC6~+n}6xFMAb97jrl>w-WAZth@V?^ntMxs{P0X-d;3=HPtcDb5HjCjnDVfdi!bP%*09-CmP)JF z5nXxu>x=vC9qulEqGql}GG~2-){bpIRv}ikYffK;2Q`X?>IB?=NAZD% z{+o9`XsB|@kq_v=yWX~^m+FD)O&1(#|O zXzrPL;5^cjEW}Dj3v>HGD6^x6xZ?n6Z$!XgbQqq|eZFFz? zWP2=}A(BCMcEX91IuZHa)p#a~ygNs1is8I+?FVm`$m?xi1-m|>n)uU1UQNsSYsDW` zaQN`#eg4jqgnsuA=Sit@$9uEkIm0x1uCdivf#I^Kcur(a@NJRx2(bVkzS&;-M%{gE zr4BDtz>1uN(eyFtj9ET?0z{aQu$HoNF@9ETR!2FXuq(~NYV6}35Sl1yI`zv|!>?64 zs70Ho;_O={3fgjFyN<>7HFrLme3|?jr9K1?9pq(3Yrp62?$jk6&=}M%5fsliB%5kvGz(_H^{5-=11 z+ty!y*6-q}*LeNsTnGUVy3p)&0sRV0vt9ZQ09?nXpoTt{p2P8%+Dg)|2+`1G@)Mj2 zz5G+&$ziwCgZk?pTzRJ7UvjPW8}C=^wv++e&}$bCXLut_ol=6Ud;Xrc_n&=tIIRu0 zxNccxn*D|Ip$v=-k@r+;wR-ho%+g)5q%2di^b*ZCvLQ)-sPE;pEQm)XsIzk8G#&)#m_(NNz3h# z5AJ`4=bk8^T0;RFV~&rWIBXciRVGjNndMz745MsKwsPdDez;(jY?PbP?s3OJHT#f! z0VD98w*K%A<;u0wYvb<}qn+Kt=I*>~y(c`ho63fe-j2Z-HX*&)7aWT4cs(k|?Ut8PQj@;b`ZT6Ude^OFaK)(lZL$1Dmj>~=hC zuwV^VRKDszBI1SkS5+(j9VhRHF~J+e4546{XG^66Uh)>@tsh!u<%oCs`H_Z;LrR77 zn3fmfjJu8|oOrfG0YHjfnqnME&QVmAjZD#lOq-m3KQBq#^Cd#b5LYD`)ok|j_`{<2 zEfZhph^=)>V+nCbBAJI7+gevg$uaOtKa$U;tEl-PwmhB>`loX&`z+e=}n1LYa^H;&kz$;tgleD z3W&8M;s$WS0rwZVzUpI!J&K5lXxuqg>UW;sjrgq)fDL5(3Rq6?!3W%ZX_!@Z`3UXP zFHf%fizEo&;J*``3RGD_^XBl9xN8`CDtG@FY?a1cg41`%L#<;g@JHld#BctKAaEdI z{ksuDtq+KkQpA-GC=C9~6ubncmh(j}r1*iY@XK>Y{80dnrAWzZNk zk?jPAmFRJ?h%!F5o-uiGHK2~-{paVl`b7-rkBGk=^1JEOZ-6gSIFpF~xB}Ak-VCbTAr_btU-h)USS#qKQK=mP29f$A z)lR~ba>mS1R`rpaY6>p`zUYQPiMqwILpd@6nT_N)@3nndS;0aEJWJ{3`dCrMZ#Rg4 z>gNlZdJFALOHy!wr#s-6=1wc6=0b(@I?}Vb%|s;Dd0wou#U`KJ1x& zgv?t)Eh7e(2}cRti2eymh+R*w&wM3SXmHlF>20BIKUVZ0t=G;ud zvf*5bjLsFMwoDQsb8Q=IE7lsr@ADqB(ZynzbgqAFf{A?{)2e@96mP?c+8_`)NPLdq zpF!EfyXKZJs`nn*@zn4J4bnrm1yv&8=h=odUHtGx;XzqzK7zPJ`Yi(qYdgj+Se<{IdytqpRS0N$?qD_kQa4IR%#lFV{v9fJ-LG3XenH6vL?!D$ zqmqR3WDX!j7!Cgh5@Y~qdTtTM7VO{}>_BrtpaAUfsYvFL#c4k2 z+eq|Y>|xO86qN6&gjDWyMmroAPL!>kumPYI{dg5$t^i=J9iJyI|t_tbw!Zy!JsYGeYRwVX{yi74wlRKkxlys zB}=wK!TVvBH?^zVn2NAsG0;Jl-?U7x!1)k3Yh91OQ)kznUS8M6Xtjk=A4oD=Gifc zueSsE;lp)v@w636+2}{Jmp?n5j5>1lFkC&h>Q$?K*Wv>ZQq%L2b{|&^ghg}6 zg%Db%9K(55zK7{Oyb>W3Q9~WC+OnbAWCtrsZzbkgsF2XwB*GE-aJX(3(bHG}vKbc@ z^B|&4i%KRd*S7Y>bf;~7uaqRMkuk!AMlSYv=H7wNUM$HMFfmda<`w+%c-?1sQb0_G z>mG*9*@TZ*&PRH0mr5b-CAz?St$YM63rIMri_wm^Z3~T&^Y1tCj{9yoaUg+(XPp1g z?bZ6wx=%600pWC4vwZltp26F$3u8D3~0CLU(BrzDGhnGBF0YwhW$2;b)(a@FiA4mqmZ89QiJQAQGf z7|#kNeiQ22V#&DKs2Am)Id^U_eJbKJzEAutb1<&?aysW>udlDvV4dvgVi{J0+!ss9C`86#7*~= zUSg^1!Zt?Nhdo58HCeai}(H z4wX{UAvMWTg(ONugUQQk1&^>CR32Y?8cvftKwdac=e9$W8DPZ^;n~qT;+$T@l7rQ` z*UzL6*oozP9Tfh$Y_pm^LYxzS!-hrG%Fn6K)v}uUSshYpHz%K=BxF=!gFLYOSt z3a1nRUP85Zzlm0y>Jb~IG^WzSzWl6banb1XkBOTPG4k% zw{tg|r0HiklW%IKCl;>+>XchIEg80w8F>&x2>@me19H87GzWeVTL4rBytS`XI-EEK zu?+HdM`t%Mlrzs|+0J;(Iz!G>UT4t%X;xR-HSdU}`7V){jb9*d#aP_6vujc@3)cbS zV&nF>Ia5v~rHXOO2UF<9rVzGr;&HswWXabi_&v6GrlBXa>^>&~Y2kbgtwpX%6|H%| zW-Dm-C1qI$RW6T>h%R1RZe3ZlF;;mwyL{a}=0bK&&qBll;`{Um<6`==2KOJ^=aQdmE9eu``yfJe!kYI`-JHh|CPY1Qvi(r|@&%9g z>sw`RiQw^#=O#;L?2g&HZ_)x8hSOv{KqmLMP z#7mU_>|+n+F8M`1MmLkJi41H-qoI0{~ir#|Kb`Qda>NK7#9@e0tjfE+k zcz$us$WRCsKANq+1)G_uRRW*k%o0NN)vOv$Rb3S*7fXzktzV$ zn5FrEam!Bg5vE-Fa$5m98c`(F)r_CpGAy2G`QFMkAbkcSLPcM}wW9NDh}Bw~xd*Z5 zJp6U20}scJjm3pf_4i5&A&m#E_K9RFC#z1MFcwh03YJHZi-`y508*)0J z@t|NAlZSWznPRJGTfD1T&fRm4b5k<@{aD| z_d#F6mriU+>glQRd#a-&Pl5M#Fs?E!M=*~*8a`3|raUKHEOm2b@w)A7Oy{jUzLQhv zm%0)(Mm*&x|Bh{jA@zktJKo(kQ{Pn=Wi;(7Ve*L6mHA%OY;f^wHJw_G-pbZ!BiqnT zhl=h&z1L-*Y95l!l*}F?3qlpAmV6Y(ifAqOF)6lgiLF<^eZ|MT0&Odw6I}=b%m>5j zwf^|+l~?^uu7_P-=OaEGmiR)^eSkEnSV;ieMDasU`8(Y=^ck;b;;$Xu?-*s+aA5Bx z33uzvQvzoG1&Xc~w_U^zU?<){6?*e8lb|BD%2~rqbrJ5*q<%m>$kp*t9uZ1xeA%>T z{0Jrp7Bg^BB0mo|HDwD`!+VhwbV%Waxs;BY7nIvkd@=8b9(4HdY1H71*}F=+a1vP0 zxA}or-ta}9-+t-AHV4*Vk#Uk{B=)03;rcLyQ z0Q(^F;0{iyw$|RaUkx4vA{3}P>kU~z2-x{Tu1Oyadf(Hn>^h>h5*4wreva2dzAp(ezY2=bxSl=1Rp!bE!QplBB$ zm7<2`UQ25@`7RQhskG^IN_cJc3#b!TCL3Q9fuTj%H)ozSlt*YP^q%M zU3%@K=8x#G*~|YT{`DUcYJYP)@Av;*j3ocn^Zuj0=MAS*ALeLdC>N>xF7hA6(?n;w z&MQqCy3e3nP!Xm3GMM5u>g5t$J6|)Jj%!Rdhg1BaP81=;l>MBZwC8C50ECVxlDennSYRZGRt{EB>nE#d51~Ai6UHP;o;%kP9v=6B;yo%ur@TE_pd`YF7XQ1wKgv16SS-(9<>Tl zi`dlQDzh*ni;SE$X5jvZnWh~H1t;})TW{J2vITtA09}V>NA?q$ph0AC`DI{v_(5MP zvVkQJ=WKFW^JjqYW0aORVCMjU{Weq+(2K}b5mkug0OF&~_F;&jg4Rs7CoapUfU7@4 zUK@9%c`z=uQ@Xf1&eqYP*+#eALCrcxekGy?zn};J1SDm2AR-^ZexVD9GVBzHOQ&Nr z(3CHw7NxJIaVvZKf>YlL3AM8Q8-F1Gl_urn@ywMtP6s>tk!1Ri4sjk3qVDdkJ&0J_ zPju6j@!{_nHoT`$D(|~;GfG>^FI%1={SnmAT}3b425Qj`$$d;1zz8u{D#@@VpCXLV zU_da*X_cY1#{YQBw7&N+)u(5X7F(Ey+v5{KA=glQNj1d2?ns@~&($@Ws}?sm{)A9K z7mOsVAu90202F;=k33q)`8ri}$KgnaWv)=r6#7m1ZhGBaoLo<{dXPe#<$Vc^jQZJ| z2VmvvxPtc3Fcl~qSV*jqqUWVzkQqL@p}@W4?JX?}pT}i#r}v1NcRwrCo&k#Wy<;U@ z1SuJqt&txOBFRw~lzeqgU%ErcQ%OuxV>BzOyg}W!w<7Qikwyq}qAV%(N|0i(^Xab= zeq*QG;T)xtc!zrRA@}J^VU7S47t8oT>23o>;(NS!{ zII0=!IKi5i9`4 zTa%G{0jt%9KI3vgdEKsjT2N(}G_ok9@xYmjQ6BNOE3(~X;~$7YS;7@-x9m4n0oqHk zIr-R|b4j>fC?L+1)njyUuRr1w`SRF)5@vQk~kMn}SFKPTNj%r;7yGT+&Z znZPgx-bVHnR1gNG&tho34A?AqEe=>3fhi)c=^C%Pa<}RLWU9RYcz1{ZnQFnUg)iu< z;90I9&^JnmB@U&M>B|&2%m2p((hwt}-Q0D+Lo~gH!O#$AK0LX#=-VHW*E?1d5P+7! z)esj>sUTRGKMl^ANg&p?WWhn|5rJ7oGtVcu*BvN_r~|5NM_I=%>VvmSFIcGWr71W% znWzacTiLO3`VbkVe#Ck}=HsII={J!Tc02G4`|4+vn@-#rEHxdk+7K;%_Libx`c3kMIE z2hD{>?P@Ul))V10%FE7PAyX1COz%rU%#^3lNG);4vJU(Vt*)-#lOU+~vEeY^^Vg9{ zw9$Zpl3|W2y`<;&b+GP6`fT zL~mOvNo(Kn@S}HPYy9}lMK$Pck{W3WGpnB696+;askA;=6I!D8hLMTX>%m2q z_l=x(JJnl_M|T1U>1ztWfjDt)%S>K<4-tX^#>S#~_)YUZ}!6cMG3VtFx%)g}rdLuEUL65JmpHto2pglwSQ`zp{~ zeuR0+OPTI(=7#3#CKxKd@D3eGmy};y-@)ESQRKJz;`wRAT{(4&i|@E}05Q6d_O0KZ z`;|V>x3f(peS>@ZF8LeU!=2C0g&tEGePNr_I>`@bGLS2L3j|PE>cKS`2e~l2ePBk9 zk*HP(`Ye7PHio`WESzA+H5+n2>VZDB4mr2bf||!bMa(k70}GDbKNrV}83RRTaqP%k zXzEADT+%aEh4g`G;xT4oif;Ts+B$MshIUQGty9t8M2;%6gX|xr%mz#vPd;#amX{>_S2Tbx4X23vdh2NGoN!L%$iCM<-kL4wA|juP9)%p3VD{92eDDBJ6TZEm z(-ZC}7UWo?qkwjsNh{_)HZGN|lNStSN_(uJDm6WAP(e3xE{UA!h^U@eBc1WHz7BNR z-fZ;BM(5<1&nsGO%qJ}OyF}r73CVgOxfDV-@xg)Os<6HIc_KHkYgL|g8>+WQP-VHn z!9*L}cBo#x9q|p~?8yEkz}9$i9vVKoEOsd61OL$97~ub1gOvAw6PMEer7Zrtvisff zdHyx_5*_RdFTXiyrufFL$no++4+B9HNPSP0bEnLX8r`%48AZekWEivq1E^4^Syjbn zPMhaHC*Ku0E-$t2W@4G^m+Td927xIcTmx-|oTl7ib79(o0+JW~4L@beak{AJN5$Qi zQ!5rWwwXrK8hc4pta4JFc#s8V&<3%%iq9Vv&9C{{lx+wKs^l6)11f|Al!E+T=c=k*nZ> zh3Yh5+vajRoR|&sUXv9cQx8w4mfc2g2Y_xZriquk?+^=?-{?{7yLbRQx1EoChfB}< zU~QZjVHxWwo|XP}iLeLyQU(kj+h&{y_se&l1$Qw7$MYk2yLXYOs&g;Y8TmH0JK7qQ z42vmfZ68GjifJ^YD0p(Xm46Cy>^vIs3{tyZ;U(weym=DolA!v`;ZS*Mu2F-z>#L=^ zlXBv}cmY{{t{I+(zKL8S9c#y}tNG3&yPvq+33jev&OOdct~ip8pj=xCx6saHh=gjqx~pY|N4}1} z3Uarwn8s~kZ)biR#`M0S!=EB7I^xx)GjhH7fw#xhv5 z?U5_mCH>0+sP}ArCNGUVeB*|;;hQH{F(a`3Yj+Gl2yAtDgt3uncd#yg+p0U*3|G>Z z7hT4zpDo9Rw(QQTI`b)Y9N&jxl;tz=luY)vp-)$C%rl-iZb<64lf6342wth}qgASdM) zABPzqvJMUjjeS5kWfAp4Dx;ez&HBj+E!Iw73c(a49!(4!#}Gfk zeT7*?s}3w@sQ$AHg8IV*d7Qo6NC~P?T#r28ypNg8E83J zu1QDF>d^4jC@SR0*JH02yZQ!85KO5>~DawLtO7SS(`B#&ViWXQ@lsUI3&%t^Y6SR1FG zC^J6VK056iH+vBnKhnTz0^2=sq1IYH5O|n;Rm2_a6Br7#{UpOx+^t{GL+wWwsy#aA zSMs{wWy;qVx|yj3pJlsnh`?rjHAO@LZE}k=NuCAvzyU~a?N3fn8Z(-uja6l4X6fhn z6eC&Znojf$Ud;}D#P`rkayKf2iLCr+_&`Mqf;qYSssZL{1rrC4J=*VNGnaF-tj)F_ zYDkjFI-#3xga$Q>?^oEdp9p_(k~U&lbY6xt(~d4%4I4%aPLZ9;o@OeJ8xy@wg$s0g zn)k(uWj)NR5;AI0GlXc_+|xRWoNiDf5r!3J-Qc$N0deDZVP!Gpp7o!2?tS7|xpT%P zd35-?iD#DF_P3tuxZGa2)VqP_k*KWbiFu=3+0&86XI|Z#b=h~W*7nQCXU|d_TE8If z;omo}+%N2+)m4joXsfXek7=E$zm7QKE;ztnwA7N#^2JIQb@YvxCeC?l3?D6!J;9}i z2u-4QBCbX};7trmXoWJ~hcp`l{6ML5@k`Kqr@*3O&F z;qQ_Q9oa9Ei-3)N6a;|0ST!#}#9%*(Q(?bl7~hMcz(d&Q+(&k}nk*KHM?E4p$KN(T zMcB4ik}oYl*bO6@1Ihp^4V3tbG?%TOSxoMB)@U$YmL*gZ`w3j*N~n~$rXa^DB&=fw zvDDbdm`OR`d2TYZ%~+;AC%%mvm#q(fJgs>sEimhq{JUZ26SZHaboroqr8ekWG-ws9 ziUp4SMgiK~q6wwhWl;G5dP?gH7rwj~=e35R%<&;j^q{+s)-rTHTuTg(s!AQpx zMOT01i4dM5yq@W<9PcyBGLo#A24;z=W@*8=q>b&@F~niW5k%D#Pctz?@~}-a366;e zlW7MR&)+0V6ThAB7a7nZO|{%3^s-H8yZyRy{P%X6GxCgBp{UHQe!;YCd{6q2X;bZ; zHl^OyHC6;y3vJ?KStW*g;33=n3}nTB3zX|>sio;-S!)^(g;0A-KP%p~whd2w*8<-Jt8 zL9#6H8b!P<*@%H_+(|+r9&BZ{B!i!sA*|6&8W*AKY)+To~Mla)ko zt^}@?xC1A01-WDD{^2@$Tx6NQR`E)hJ0t{1@DziHLaJ}*n^NJC0W#|4SGC?|rQX{g z9%7cfUtq=J>$zJ+2dV0r)97uwhkU$;(}OH8&_$jg1$RF_`VY9HIMx&_jF#ZDk{9_^O$gJ^}7Kr*99io5*WNbD53VK88rv(dwRo!K=Anl_uw_QpHwm&Ie~% zjO$gky9dc5)^Ydx!0Nv96gXD9=QwA4tr3jIDD_m_`$>H4AU%vZUOvq9X9Y0T7EOZT z5_%l`od?@jebsNfSL%m9CzO_&QjYS~ta(iNaWYDZqcaYK4=r23%xBR1W{7X&D(%dM zTGRCzJL@!>+MyHrZgUh>R!wH>vTp@Y zhzboO^6NfB0&GKiweE4c&O6PsN1Enq3>ND5hYQ67itu->I6gvLK|a)#EDc{9Z=H+Q zZ%A5{(TrG~4se|*(=yF(&$Vc%+_(K$=*Hb?_QxWQs%TIq4trQ#o!WX5Gg_;Z@j-|w zPM_eKnSc%EM~G&h{L0W>1DN0Hr)oeYQiJU?c_nOQ6E!fC{s5Ow-#S(0}UbX{E{Aope8Nw=Z-u9(gm=w}@dsU|G|TM} zKZ{i88`v2V4*M*xb`|xy~Ea`e8Fr2=N)wYc$hm~w&E+hRQx%L+gn>&z(KAjTDg>U1y7@!f?Zv8i`Bcop+z!)=*t@V3~32Nbd!a zBZ@({e7onG-haMVFn?(Eb+oLk>^*cs+@6j;1^xc~tg#E7R;D-HK_D<|e^m6i0Au*lWs5!THC-$3lE;v`mGHANC>u<(Y; z8b>t9vG1pRSmYP2xN@!40;mm%4}G?$VAL|*Mw+dh7SE`#f+*`2;ocjP zv2Xd-mzNl#exAGy+e=y)anp`neEsg2wdGTN=pYeyEo&b+XQODC7V!#{Q8JZB?*9+@2w?x_6p;TfS!qA+ zbzM%@w0Z5x=^fa3x3jeSV)T6geF#Ypz$iBM%_+~8RG_I9&G&0ET z3lxo7inUUz+TLlk-sQJ2YKj8fuMMNrKOSQ20b2t987N zE94ypO#-(h;X6+loR(b*zzILCut{v!H8a-HV)g>)1!n&1skr-fkgE&jz%#Z_2St9Q zK}Zepn=znz?vdRHdX0}bP8>nEd-u8#C!!$o{)qg3sPzCSu+xYCatlFg%ZOD*`Qqxk z-wrYS{qu06%NqfFL_35x*byUx>kw zecBJdI2J(RmLa(Tgi`Je=&y1EA=a&*YKcoQs2QpyzNY^5#;<{pWU|qB9&9jTSQ>~A zFHeOnPeY4y0TYW3za169JU0ktm8;5;{q@Fwl_sdrluBo>qJ1EOB{+{O6_E)4 z%@|M(nMWsZc*%=sI$=udFSBR(t1y|5mk|>Uz}of0vsq|gb`MCM{yGAXG5%&kAtUl% z?_2j*VM5i0Kz6#Q!9Nzl9}D4+h46<)*!_n___0I&u@L?t3ju-NLA*|QJgD70To73E z@;%>5RBY&+i0pjah6xeEH)N3gmUm2liadOqGrPt~JlR*_&zYjRJ%^&7{X{$&998b1 lJ8aX#G62l3Fno*0+K0x&@&~^H5|Pz{%Ks_j1>$?}{{fE*Xc_ var gitbook = gitbook || []; gitbook.push(function() { - gitbook.page.hasChanged({"page":{"title":"序言","level":"1.1","depth":1,"next":{"title":"Kubernetes与云原生应用概览","level":"2.1","depth":1,"path":"cloud-native/kubernetes-and-cloud-native-app-overview.md","ref":"cloud-native/kubernetes-and-cloud-native-app-overview.md","articles":[]},"dir":"ltr"},"config":{"plugins":["github","codesnippet","splitter","page-toc-button","image-captions","editlink","back-to-top-button","-lunr","-search","search-plus","github-buttons@2.1.0","favicon@^0.0.2","tbfed-pagefooter@^0.0.1","3-ba","theme-default"],"styles":{"website":"styles/website.css","pdf":"styles/pdf.css","epub":"styles/epub.css","mobi":"styles/mobi.css","ebook":"styles/ebook.css","print":"styles/print.css"},"pluginsConfig":{"tbfed-pagefooter":{"copyright":"Copyright © jimmysong.io 2017","modify_label":"Updated:","modify_format":"YYYY-MM-DD HH:mm:ss"},"github":{"url":"https://github.com/rootsongjc/kubernetes-handbook"},"editlink":{"label":"编辑本页","multilingual":false,"base":"https://github.com/rootsongjc/kubernetes-handbook/blob/master/"},"splitter":{},"codesnippet":{},"fontsettings":{"theme":"white","family":"sans","size":2},"highlight":{},"favicon":{"shortcut":"favicon.ico","bookmark":"favicon.ico"},"page-toc-button":{},"back-to-top-button":{},"github-buttons":{"repo":"rootsongjc/kubernetes-handbook","types":["star"],"size":"small"},"3-ba":{"configuration":"auto","token":"11f7d254cfa4e0ca44b175c66d379ecc"},"sharing":{"facebook":true,"twitter":true,"google":false,"weibo":false,"instapaper":false,"vk":false,"all":["facebook","google","twitter","weibo","instapaper"]},"theme-default":{"styles":{"website":"styles/website.css","pdf":"styles/pdf.css","epub":"styles/epub.css","mobi":"styles/mobi.css","ebook":"styles/ebook.css","print":"styles/print.css"},"showLevel":true},"search-plus":{},"image-captions":{"caption":"图片 - _CAPTION_","variable_name":"_pictures"}},"theme":"default","author":"Jimmy Song","pdf":{"pageNumbers":true,"fontSize":12,"fontFamily":"Arial","paperSize":"a4","chapterMark":"pagebreak","pageBreaksBefore":"/","margin":{"right":62,"left":62,"top":56,"bottom":56}},"structure":{"langs":"LANGS.md","readme":"README.md","glossary":"GLOSSARY.md","summary":"SUMMARY.md"},"variables":{"_pictures":[{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.1","level":"2.1","list_caption":"Figure: 云计算演进历程","alt":"云计算演进历程","nro":1,"url":"../images/cloud-computing-evolution-road.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"云计算演进历程","attributes":{},"skip":false,"key":"2.1.1"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.2","level":"2.1","list_caption":"Figure: 来自Twitter @MarcWilczek","alt":"来自Twitter @MarcWilczek","nro":2,"url":"../images/cloud-native-comes-of-age.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"来自Twitter @MarcWilczek","attributes":{},"skip":false,"key":"2.1.2"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.3","level":"2.1","list_caption":"Figure: Cloud native思维导图","alt":"Cloud native思维导图","nro":3,"url":"../images/cloud-native-architecutre-mindnode.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"Cloud native思维导图","attributes":{},"skip":false,"key":"2.1.3"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.4","level":"2.1","list_caption":"Figure: 十二因素应用","alt":"十二因素应用","nro":4,"url":"../images/12-factor-app.png","index":4,"caption_template":"图片 - _CAPTION_","label":"十二因素应用","attributes":{},"skip":false,"key":"2.1.4"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.5","level":"2.1","list_caption":"Figure: 容器生态","alt":"容器生态","nro":5,"url":"../images/container-ecosystem.png","index":5,"caption_template":"图片 - _CAPTION_","label":"容器生态","attributes":{},"skip":false,"key":"2.1.5"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.6","level":"2.1","list_caption":"Figure: 使用Jenkins进行持续集成与发布流程图","alt":"使用Jenkins进行持续集成与发布流程图","nro":6,"url":"../images/kubernetes-jenkins-ci-cd.png","index":6,"caption_template":"图片 - _CAPTION_","label":"使用Jenkins进行持续集成与发布流程图","attributes":{},"skip":false,"key":"2.1.6"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.7","level":"2.1","list_caption":"Figure: filebeat日志收集架构图","alt":"filebeat日志收集架构图","nro":7,"url":"../images/filebeat-log-collector-arch.png","index":7,"caption_template":"图片 - _CAPTION_","label":"filebeat日志收集架构图","attributes":{},"skip":false,"key":"2.1.7"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.8","level":"2.1","list_caption":"Figure: API文档","alt":"API文档","nro":8,"url":"../images/k8s-app-monitor-test-api-doc.jpg","index":8,"caption_template":"图片 - _CAPTION_","label":"API文档","attributes":{},"skip":false,"key":"2.1.8"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.9","level":"2.1","list_caption":"Figure: 迁移步骤示意图","alt":"迁移步骤示意图","nro":9,"url":"../images/migrating-hadoop-yarn-to-kubernetes.png","index":9,"caption_template":"图片 - _CAPTION_","label":"迁移步骤示意图","attributes":{},"skip":false,"key":"2.1.9"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.10","level":"2.1","list_caption":"Figure: service mesh架构图","alt":"service mesh架构图","nro":10,"url":"../images/serivce-mesh-control-plane.png","index":10,"caption_template":"图片 - _CAPTION_","label":"service mesh架构图","attributes":{},"skip":false,"key":"2.1.10"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.11","level":"2.1","list_caption":"Figure: kibana界面","alt":"kibana界面","nro":11,"url":"../images/filebeat-docker-test.jpg","index":11,"caption_template":"图片 - _CAPTION_","label":"kibana界面","attributes":{},"skip":false,"key":"2.1.11"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.12","level":"2.1","list_caption":"Figure: Grafana界面示意图1","alt":"Grafana界面示意图1","nro":12,"url":"../images/kubernetes-devops-example-grafana-1.png","index":12,"caption_template":"图片 - _CAPTION_","label":"Grafana界面示意图1","attributes":{},"skip":false,"key":"2.1.12"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.13","level":"2.1","list_caption":"Figure: Grafana界面示意图2","alt":"Grafana界面示意图2","nro":13,"url":"../images/kubernetes-devops-example-grafana-2.png","index":13,"caption_template":"图片 - _CAPTION_","label":"Grafana界面示意图2","attributes":{},"skip":false,"key":"2.1.13"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.14","level":"2.1","list_caption":"Figure: Grafana界面示意图3","alt":"Grafana界面示意图3","nro":14,"url":"../images/kubernetes-devops-example-grafana-3.png","index":14,"caption_template":"图片 - _CAPTION_","label":"Grafana界面示意图3","attributes":{},"skip":false,"key":"2.1.14"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.15","level":"2.1","list_caption":"Figure: dashboard","alt":"dashboard","nro":15,"url":"../images/spark-job-on-kubernetes-example-1.jpg","index":15,"caption_template":"图片 - _CAPTION_","label":"dashboard","attributes":{},"skip":false,"key":"2.1.15"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.16","level":"2.1","list_caption":"Figure: Grafana","alt":"Grafana","nro":16,"url":"../images/spark-job-on-kubernetes-example-2.jpg","index":16,"caption_template":"图片 - _CAPTION_","label":"Grafana","attributes":{},"skip":false,"key":"2.1.16"},{"backlink":"concepts/index.html#fig3.1.1","level":"3.1","list_caption":"Figure: Borg架构","alt":"Borg架构","nro":17,"url":"../images/borg.png","index":1,"caption_template":"图片 - _CAPTION_","label":"Borg架构","attributes":{},"skip":false,"key":"3.1.1"},{"backlink":"concepts/index.html#fig3.1.2","level":"3.1","list_caption":"Figure: Kubernetes架构","alt":"Kubernetes架构","nro":18,"url":"../images/architecture.png","index":2,"caption_template":"图片 - _CAPTION_","label":"Kubernetes架构","attributes":{},"skip":false,"key":"3.1.2"},{"backlink":"concepts/index.html#fig3.1.3","level":"3.1","list_caption":"Figure: Kuberentes架构（图片来自于网络）","alt":"Kuberentes架构（图片来自于网络）","nro":19,"url":"../images/kubernetes-high-level-component-archtecture.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"Kuberentes架构（图片来自于网络）","attributes":{},"skip":false,"key":"3.1.3"},{"backlink":"concepts/index.html#fig3.1.4","level":"3.1","list_caption":"Figure: kubernetes整体架构示意图","alt":"kubernetes整体架构示意图","nro":20,"url":"../images/kubernetes-whole-arch.png","index":4,"caption_template":"图片 - _CAPTION_","label":"kubernetes整体架构示意图","attributes":{},"skip":false,"key":"3.1.4"},{"backlink":"concepts/index.html#fig3.1.5","level":"3.1","list_caption":"Figure: Kubernetes master架构示意图","alt":"Kubernetes master架构示意图","nro":21,"url":"../images/kubernetes-master-arch.png","index":5,"caption_template":"图片 - _CAPTION_","label":"Kubernetes master架构示意图","attributes":{},"skip":false,"key":"3.1.5"},{"backlink":"concepts/index.html#fig3.1.6","level":"3.1","list_caption":"Figure: kubernetes node架构示意图","alt":"kubernetes node架构示意图","nro":22,"url":"../images/kubernetes-node-arch.png","index":6,"caption_template":"图片 - _CAPTION_","label":"kubernetes node架构示意图","attributes":{},"skip":false,"key":"3.1.6"},{"backlink":"concepts/index.html#fig3.1.7","level":"3.1","list_caption":"Figure: Kubernetes分层架构示意图","alt":"Kubernetes分层架构示意图","nro":23,"url":"../images/kubernetes-layers-arch.jpg","index":7,"caption_template":"图片 - _CAPTION_","label":"Kubernetes分层架构示意图","attributes":{},"skip":false,"key":"3.1.7"},{"backlink":"concepts/concepts.html#fig3.2.1","level":"3.2","list_caption":"Figure: 分层架构示意图","alt":"分层架构示意图","nro":24,"url":"../images/kubernetes-layers-arch.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"分层架构示意图","attributes":{},"skip":false,"key":"3.2.1"},{"backlink":"concepts/pod-overview.html#fig3.3.1.1","level":"3.3.1","list_caption":"Figure: pod diagram","alt":"pod diagram","nro":25,"url":"../images/pod-overview.png","index":1,"caption_template":"图片 - _CAPTION_","label":"pod diagram","attributes":{},"skip":false,"key":"3.3.1.1"},{"backlink":"concepts/pod.html#fig3.3.1.1.1","level":"3.3.1.1","list_caption":"Figure: Pod示意图","alt":"Pod示意图","nro":26,"url":"../images/pod-overview.png","index":1,"caption_template":"图片 - _CAPTION_","label":"Pod示意图","attributes":{},"skip":false,"key":"3.3.1.1.1"},{"backlink":"concepts/pod.html#fig3.3.1.1.2","level":"3.3.1.1","list_caption":"Figure: Pod Cheatsheet","alt":"Pod Cheatsheet","nro":27,"url":"../images/kubernetes-pod-cheatsheet.png","index":2,"caption_template":"图片 - _CAPTION_","label":"Pod Cheatsheet","attributes":{},"skip":false,"key":"3.3.1.1.2"},{"backlink":"concepts/service.html#fig3.3.4.1","level":"3.3.4","list_caption":"Figure: userspace代理模式下Service概览图","alt":"userspace代理模式下Service概览图","nro":28,"url":"../images/services-userspace-overview.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"userspace代理模式下Service概览图","attributes":{},"skip":false,"key":"3.3.4.1"},{"backlink":"concepts/service.html#fig3.3.4.2","level":"3.3.4","list_caption":"Figure: iptables代理模式下Service概览图","alt":"iptables代理模式下Service概览图","nro":29,"url":"../images/services-iptables-overview.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"iptables代理模式下Service概览图","attributes":{},"skip":false,"key":"3.3.4.2"},{"backlink":"concepts/deployment.html#fig3.3.6.1","level":"3.3.6","list_caption":"Figure: kubernetes deployment cheatsheet","alt":"kubernetes deployment cheatsheet","nro":30,"url":"../images/deployment-cheatsheet.png","index":1,"caption_template":"图片 - _CAPTION_","label":"kubernetes deployment cheatsheet","attributes":{},"skip":false,"key":"3.3.6.1"},{"backlink":"concepts/horizontal-pod-autoscaling.html#fig3.3.16.1","level":"3.3.16","list_caption":"Figure: horizontal-pod-autoscaler","alt":"horizontal-pod-autoscaler","nro":31,"url":"../images/horizontal-pod-autoscaler.png","index":1,"caption_template":"图片 - _CAPTION_","label":"horizontal-pod-autoscaler","attributes":{},"skip":false,"key":"3.3.16.1"},{"backlink":"concepts/label.html#fig3.3.17.1","level":"3.3.17","list_caption":"Figure: label示意图","alt":"label示意图","nro":32,"url":"../images/labels.png","index":1,"caption_template":"图片 - _CAPTION_","label":"label示意图","attributes":{},"skip":false,"key":"3.3.17.1"},{"backlink":"guide/using-kubectl.html#fig4.3.2.1","level":"4.3.2","list_caption":"Figure: kubectl cheatsheet","alt":"kubectl cheatsheet","nro":33,"url":"../images/kubernetes-kubectl-cheatsheet.png","index":1,"caption_template":"图片 - _CAPTION_","label":"kubectl cheatsheet","attributes":{},"skip":false,"key":"4.3.2.1"},{"backlink":"guide/using-kubectl.html#fig4.3.2.2","level":"4.3.2","list_caption":"Figure: kube-shell页面","alt":"kube-shell页面","nro":34,"url":"../images/kube-shell.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"kube-shell页面","attributes":{},"skip":false,"key":"4.3.2.2"},{"backlink":"guide/ip-masq-agent.html#fig4.4.6.1","level":"4.4.6","list_caption":"Figure: IP伪装代理示意图","alt":"IP伪装代理示意图","nro":35,"url":"../images/ip-masq.png","index":1,"caption_template":"图片 - _CAPTION_","label":"IP伪装代理示意图","attributes":{},"skip":false,"key":"4.4.6.1"},{"backlink":"guide/auth-with-kubeconfig-or-token.html#fig4.4.7.1","level":"4.4.7","list_caption":"Figure: kubeconfig文件","alt":"kubeconfig文件","nro":36,"url":"../images/brand-kubeconfig-yaml.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"kubeconfig文件","attributes":{},"skip":false,"key":"4.4.7.1"},{"backlink":"guide/authentication.html#fig4.4.8.1","level":"4.4.8","list_caption":"Figure: Kubernetes OpenID Connect Flow","alt":"Kubernetes OpenID Connect Flow","nro":37,"url":"../images/kubernetes-oidc-login.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"Kubernetes OpenID Connect Flow","attributes":{},"skip":false,"key":"4.4.8.1"},{"backlink":"guide/carbin-mobile-dashboard-for-kubernetes.html#fig4.5.6.1","level":"4.5.6","list_caption":"Figure: App Store","alt":"App Store","nro":38,"url":"../images/carbin-kubernetes-mobile-dashboard-1.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"App Store","attributes":{},"skip":false,"key":"4.5.6.1"},{"backlink":"guide/carbin-mobile-dashboard-for-kubernetes.html#fig4.5.6.2","level":"4.5.6","list_caption":"Figure: 在手机上操作Kubernetes集群","alt":"在手机上操作Kubernetes集群","nro":39,"url":"../images/carbin-kubernetes-mobile-dashboard-4.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"在手机上操作Kubernetes集群","attributes":{},"skip":false,"key":"4.5.6.2"},{"backlink":"guide/kubernetes-desktop-client.html#fig4.5.7.1","level":"4.5.7","list_caption":"Figure: Kubernetic客户端","alt":"Kubernetic客户端","nro":40,"url":"../images/kubernetic-desktop-ui.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"Kubernetic客户端","attributes":{},"skip":false,"key":"4.5.7.1"},{"backlink":"guide/deploy-applications-in-kubernetes.html#fig4.6.1.1","level":"4.6.1","list_caption":"Figure: API","alt":"API","nro":41,"url":"../images/k8s-app-monitor-test-api-doc.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"API","attributes":{},"skip":false,"key":"4.6.1.1"},{"backlink":"guide/deploy-applications-in-kubernetes.html#fig4.6.1.2","level":"4.6.1","list_caption":"Figure: wercker","alt":"wercker","nro":42,"url":"../images/k8s-app-monitor-agent-wercker.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"wercker","attributes":{},"skip":false,"key":"4.6.1.2"},{"backlink":"guide/deploy-applications-in-kubernetes.html#fig4.6.1.3","level":"4.6.1","list_caption":"Figure: 图表","alt":"图表","nro":43,"url":"../images/k8s-app-monitor-agent.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"图表","attributes":{},"skip":false,"key":"4.6.1.3"},{"backlink":"guide/migrating-hadoop-yarn-to-kubernetes.html#fig4.6.2.1","level":"4.6.2","list_caption":"Figure: spark on yarn with kubernetes","alt":"spark on yarn with kubernetes","nro":44,"url":"../images/spark-on-yarn-with-kubernetes.png","index":1,"caption_template":"图片 - _CAPTION_","label":"spark on yarn with kubernetes","attributes":{},"skip":false,"key":"4.6.2.1"},{"backlink":"guide/migrating-hadoop-yarn-to-kubernetes.html#fig4.6.2.2","level":"4.6.2","list_caption":"Figure: Terms","alt":"Terms","nro":45,"url":"../images/terms-in-kubernetes-app-deployment.png","index":2,"caption_template":"图片 - _CAPTION_","label":"Terms","attributes":{},"skip":false,"key":"4.6.2.2"},{"backlink":"guide/migrating-hadoop-yarn-to-kubernetes.html#fig4.6.2.3","level":"4.6.2","list_caption":"Figure: 分解步骤解析","alt":"分解步骤解析","nro":46,"url":"../images/migrating-hadoop-yarn-to-kubernetes.png","index":3,"caption_template":"图片 - _CAPTION_","label":"分解步骤解析","attributes":{},"skip":false,"key":"4.6.2.3"},{"backlink":"practice/node-installation.html#fig5.2.7.1","level":"5.2.7","list_caption":"Figure: welcome nginx","alt":"welcome nginx","nro":47,"url":"../images/kubernetes-installation-test-nginx.png","index":1,"caption_template":"图片 - _CAPTION_","label":"welcome nginx","attributes":{},"skip":false,"key":"5.2.7.1"},{"backlink":"practice/dashboard-addon-installation.html#fig5.2.9.1","level":"5.2.9","list_caption":"Figure: kubernetes dashboard","alt":"kubernetes dashboard","nro":48,"url":"../images/kubernetes-dashboard-raw.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"kubernetes dashboard","attributes":{},"skip":false,"key":"5.2.9.1"},{"backlink":"practice/dashboard-addon-installation.html#fig5.2.9.2","level":"5.2.9","list_caption":"Figure: V1.6.3版本的dashboard界面","alt":"V1.6.3版本的dashboard界面","nro":49,"url":"../images/dashboard-v163.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"V1.6.3版本的dashboard界面","attributes":{},"skip":false,"key":"5.2.9.2"},{"backlink":"practice/heapster-addon-installation.html#fig5.2.10.1","level":"5.2.10","list_caption":"Figure: dashboard-heapster","alt":"dashboard-heapster","nro":50,"url":"../images/kubernetes-dashboard-with-heapster.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"dashboard-heapster","attributes":{},"skip":false,"key":"5.2.10.1"},{"backlink":"practice/heapster-addon-installation.html#fig5.2.10.2","level":"5.2.10","list_caption":"Figure: grafana","alt":"grafana","nro":51,"url":"../images/kubernetes-heapster-grafana.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"grafana","attributes":{},"skip":false,"key":"5.2.10.2"},{"backlink":"practice/heapster-addon-installation.html#fig5.2.10.3","level":"5.2.10","list_caption":"Figure: kubernetes-influxdb-heapster","alt":"kubernetes-influxdb-heapster","nro":52,"url":"../images/kubernetes-influxdb-heapster.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"kubernetes-influxdb-heapster","attributes":{},"skip":false,"key":"5.2.10.3"},{"backlink":"practice/heapster-addon-installation.html#fig5.2.10.4","level":"5.2.10","list_caption":"Figure: 修改grafana模板","alt":"修改grafana模板","nro":53,"url":"../images/grafana-dashboard-setting.jpg","index":4,"caption_template":"图片 - _CAPTION_","label":"修改grafana模板","attributes":{},"skip":false,"key":"5.2.10.4"},{"backlink":"practice/efk-addon-installation.html#fig5.2.11.1","level":"5.2.11","list_caption":"Figure: es-setting","alt":"es-setting","nro":54,"url":"../images/es-setting.png","index":1,"caption_template":"图片 - _CAPTION_","label":"es-setting","attributes":{},"skip":false,"key":"5.2.11.1"},{"backlink":"practice/efk-addon-installation.html#fig5.2.11.2","level":"5.2.11","list_caption":"Figure: es-home","alt":"es-home","nro":55,"url":"../images/kubernetes-efk-kibana.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"es-home","attributes":{},"skip":false,"key":"5.2.11.2"},{"backlink":"practice/traefik-ingress-installation.html#fig5.3.1.1","level":"5.3.1","list_caption":"Figure: kubernetes-dashboard","alt":"kubernetes-dashboard","nro":56,"url":"../images/traefik-dashboard.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"kubernetes-dashboard","attributes":{},"skip":false,"key":"5.3.1.1"},{"backlink":"practice/traefik-ingress-installation.html#fig5.3.1.2","level":"5.3.1","list_caption":"Figure: traefik-nginx","alt":"traefik-nginx","nro":57,"url":"../images/traefik-nginx.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"traefik-nginx","attributes":{},"skip":false,"key":"5.3.1.2"},{"backlink":"practice/traefik-ingress-installation.html#fig5.3.1.3","level":"5.3.1","list_caption":"Figure: traefik-guestbook","alt":"traefik-guestbook","nro":58,"url":"../images/traefik-guestbook.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"traefik-guestbook","attributes":{},"skip":false,"key":"5.3.1.3"},{"backlink":"practice/distributed-load-test.html#fig5.3.2.1","level":"5.3.2","list_caption":"Figure: 使用dashboard来扩容","alt":"使用dashboard来扩容","nro":59,"url":"../images/dashbaord-scale.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"使用dashboard来扩容","attributes":{},"skip":false,"key":"5.3.2.1"},{"backlink":"practice/distributed-load-test.html#fig5.3.2.2","level":"5.3.2","list_caption":"Figure: Traefik的UI","alt":"Traefik的UI","nro":60,"url":"../images/traefik-dashboard-locust.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"Traefik的UI","attributes":{},"skip":false,"key":"5.3.2.2"},{"backlink":"practice/distributed-load-test.html#fig5.3.2.3","level":"5.3.2","list_caption":"Figure: Locust启动界面","alt":"Locust启动界面","nro":61,"url":"../images/locust-start-swarming.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"Locust启动界面","attributes":{},"skip":false,"key":"5.3.2.3"},{"backlink":"practice/distributed-load-test.html#fig5.3.2.4","level":"5.3.2","list_caption":"Figure: Dashboard查看页面","alt":"Dashboard查看页面","nro":62,"url":"../images/sample-webapp-rc.jpg","index":4,"caption_template":"图片 - _CAPTION_","label":"Dashboard查看页面","attributes":{},"skip":false,"key":"5.3.2.4"},{"backlink":"practice/distributed-load-test.html#fig5.3.2.5","level":"5.3.2","list_caption":"Figure: Locust测试结果页面","alt":"Locust测试结果页面","nro":63,"url":"../images/locust-dashboard.jpg","index":5,"caption_template":"图片 - _CAPTION_","label":"Locust测试结果页面","attributes":{},"skip":false,"key":"5.3.2.5"},{"backlink":"practice/network-and-cluster-perfermance-test.html#fig5.3.3.1","level":"5.3.3","list_caption":"Figure: kubernetes-dashboard","alt":"kubernetes-dashboard","nro":64,"url":"../images/kubenetes-e2e-test.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"kubernetes-dashboard","attributes":{},"skip":false,"key":"5.3.3.1"},{"backlink":"practice/network-and-cluster-perfermance-test.html#fig5.3.3.2","level":"5.3.3","list_caption":"Figure: locust测试页面","alt":"locust测试页面","nro":65,"url":"../images/kubernetes-locust-test.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"locust测试页面","attributes":{},"skip":false,"key":"5.3.3.2"},{"backlink":"practice/edge-node-configuration.html#fig5.3.4.1","level":"5.3.4","list_caption":"Figure: 边缘节点架构","alt":"边缘节点架构","nro":66,"url":"../images/kubernetes-edge-node-architecture.png","index":1,"caption_template":"图片 - _CAPTION_","label":"边缘节点架构","attributes":{},"skip":false,"key":"5.3.4.1"},{"backlink":"practice/app-log-collection.html#fig5.4.2.1","level":"5.4.2","list_caption":"Figure: filebeat日志收集架构图","alt":"filebeat日志收集架构图","nro":67,"url":"../images/filebeat-log-collector.png","index":1,"caption_template":"图片 - _CAPTION_","label":"filebeat日志收集架构图","attributes":{},"skip":false,"key":"5.4.2.1"},{"backlink":"practice/app-log-collection.html#fig5.4.2.2","level":"5.4.2","list_caption":"Figure: Kibana页面","alt":"Kibana页面","nro":68,"url":"../images/filebeat-docker-test.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"Kibana页面","attributes":{},"skip":false,"key":"5.4.2.2"},{"backlink":"practice/app-log-collection.html#fig5.4.2.3","level":"5.4.2","list_caption":"Figure: filebeat收集的日志详细信息","alt":"filebeat收集的日志详细信息","nro":69,"url":"../images/kubernetes-filebeat-detail.png","index":3,"caption_template":"图片 - _CAPTION_","label":"filebeat收集的日志详细信息","attributes":{},"skip":false,"key":"5.4.2.3"},{"backlink":"practice/monitor.html#fig5.4.4.1","level":"5.4.4","list_caption":"Figure: Kubernetes集群中的监控","alt":"Kubernetes集群中的监控","nro":70,"url":"../images/monitoring-in-kubernetes.png","index":1,"caption_template":"图片 - _CAPTION_","label":"Kubernetes集群中的监控","attributes":{},"skip":false,"key":"5.4.4.1"},{"backlink":"practice/monitor.html#fig5.4.4.2","level":"5.4.4","list_caption":"Figure: kubernetes的容器命名规则示意图","alt":"kubernetes的容器命名规则示意图","nro":71,"url":"../images/kubernetes-container-naming-rule.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"kubernetes的容器命名规则示意图","attributes":{},"skip":false,"key":"5.4.4.2"},{"backlink":"practice/monitor.html#fig5.4.4.3","level":"5.4.4","list_caption":"Figure: Heapster架构图（改进版）","alt":"Heapster架构图（改进版）","nro":72,"url":"../images/kubernetes-heapster-monitoring.png","index":3,"caption_template":"图片 - _CAPTION_","label":"Heapster架构图（改进版）","attributes":{},"skip":false,"key":"5.4.4.3"},{"backlink":"practice/monitor.html#fig5.4.4.4","level":"5.4.4","list_caption":"Figure: 应用监控架构图","alt":"应用监控架构图","nro":73,"url":"../images/kubernetes-app-monitoring.png","index":4,"caption_template":"图片 - _CAPTION_","label":"应用监控架构图","attributes":{},"skip":false,"key":"5.4.4.4"},{"backlink":"practice/monitor.html#fig5.4.4.5","level":"5.4.4","list_caption":"Figure: 应用拓扑图","alt":"应用拓扑图","nro":74,"url":"../images/weave-scope-service-topology.jpg","index":5,"caption_template":"图片 - _CAPTION_","label":"应用拓扑图","attributes":{},"skip":false,"key":"5.4.4.5"},{"backlink":"practice/data-persistence-problem.html#fig5.4.5.1","level":"5.4.5","list_caption":"Figure: 日志持久化收集解决方案示意图","alt":"日志持久化收集解决方案示意图","nro":75,"url":"../images/log-persistence-logstash.png","index":1,"caption_template":"图片 - _CAPTION_","label":"日志持久化收集解决方案示意图","attributes":{},"skip":false,"key":"5.4.5.1"},{"backlink":"practice/storage-for-containers-using-glusterfs-with-openshift.html#fig5.5.1.2.1","level":"5.5.1.2","list_caption":"Figure: 创建存储","alt":"创建存储","nro":76,"url":"../images/create-gluster-storage.png","index":1,"caption_template":"图片 - _CAPTION_","label":"创建存储","attributes":{},"skip":false,"key":"5.5.1.2.1"},{"backlink":"practice/storage-for-containers-using-glusterfs-with-openshift.html#fig5.5.1.2.2","level":"5.5.1.2","list_caption":"Figure: Screen Shot 2017-03-24 at 11.09.34.png","alt":"Screen Shot 2017-03-24 at 11.09.34.png","nro":77,"url":"https://keithtenzer.files.wordpress.com/2017/03/screen-shot-2017-03-24-at-11-09-341.png?w=440","index":2,"caption_template":"图片 - _CAPTION_","label":"Screen Shot 2017-03-24 at 11.09.34.png","attributes":{},"skip":false,"key":"5.5.1.2.2"},{"backlink":"practice/using-heapster-to-get-object-metrics.html#fig5.6.1.1.1","level":"5.6.1.1","list_caption":"Figure: Heapster架构图","alt":"Heapster架构图","nro":78,"url":"../images/heapster-architecture.png","index":1,"caption_template":"图片 - _CAPTION_","label":"Heapster架构图","attributes":{},"skip":false,"key":"5.6.1.1.1"},{"backlink":"practice/prometheus.html#fig5.6.2.1","level":"5.6.2","list_caption":"Figure: Prometheus 架构图","alt":"Prometheus 架构图","nro":79,"url":"../images/prometheus-architecture.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"Prometheus 架构图","attributes":{},"skip":false,"key":"5.6.2.1"},{"backlink":"practice/using-prometheus-to-monitor-kuberentes-cluster.html#fig5.6.2.1.1","level":"5.6.2.1","list_caption":"Figure: Grafana页面","alt":"Grafana页面","nro":80,"url":"../images/kubernetes-prometheus-monitoring.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"Grafana页面","attributes":{},"skip":false,"key":"5.6.2.1.1"},{"backlink":"practice/helm.html#fig5.7.1.1","level":"5.7.1","list_caption":"Figure: Helm chart源","alt":"Helm chart源","nro":81,"url":"../images/helm-charts-repository.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"Helm chart源","attributes":{},"skip":false,"key":"5.7.1.1"},{"backlink":"practice/helm.html#fig5.7.1.2","level":"5.7.1","list_caption":"Figure: TODO应用的Web页面","alt":"TODO应用的Web页面","nro":82,"url":"../images/helm-mean-todo-aholic.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"TODO应用的Web页面","attributes":{},"skip":false,"key":"5.7.1.2"},{"backlink":"practice/create-private-charts-repo.html#fig5.7.2.1","level":"5.7.2","list_caption":"Figure: Helm monocular界面","alt":"Helm monocular界面","nro":83,"url":"../images/helm-monocular-jimmysong.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"Helm monocular界面","attributes":{},"skip":false,"key":"5.7.2.1"},{"backlink":"practice/jenkins-ci-cd.html#fig5.8.1.1","level":"5.8.1","list_caption":"Figure: 基于Jenkins的持续集成与发布","alt":"基于Jenkins的持续集成与发布","nro":84,"url":"../images/kubernetes-jenkins-ci-cd.png","index":1,"caption_template":"图片 - _CAPTION_","label":"基于Jenkins的持续集成与发布","attributes":{},"skip":false,"key":"5.8.1.1"},{"backlink":"practice/drone-ci-cd.html#fig5.8.2.1","level":"5.8.2","list_caption":"Figure: OAuth注册","alt":"OAuth注册","nro":85,"url":"../images/github-oauth-register.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"OAuth注册","attributes":{},"skip":false,"key":"5.8.2.1"},{"backlink":"practice/drone-ci-cd.html#fig5.8.2.2","level":"5.8.2","list_caption":"Figure: OAuth key","alt":"OAuth key","nro":86,"url":"../images/github-oauth-drone-key.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"OAuth key","attributes":{},"skip":false,"key":"5.8.2.2"},{"backlink":"practice/drone-ci-cd.html#fig5.8.2.3","level":"5.8.2","list_caption":"Figure: Drone登陆界面","alt":"Drone登陆界面","nro":87,"url":"../images/drone-login-github.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"Drone登陆界面","attributes":{},"skip":false,"key":"5.8.2.3"},{"backlink":"practice/drone-ci-cd.html#fig5.8.2.4","level":"5.8.2","list_caption":"Figure: Github启用repo设置","alt":"Github启用repo设置","nro":88,"url":"../images/drone-github-active.jpg","index":4,"caption_template":"图片 - _CAPTION_","label":"Github启用repo设置","attributes":{},"skip":false,"key":"5.8.2.4"},{"backlink":"practice/drone-ci-cd.html#fig5.8.2.5","level":"5.8.2","list_caption":"Figure: Github单个repo设置","alt":"Github单个repo设置","nro":89,"url":"../images/drone-github-repo-setting.jpg","index":5,"caption_template":"图片 - _CAPTION_","label":"Github单个repo设置","attributes":{},"skip":false,"key":"5.8.2.5"},{"backlink":"practice/dashboard-upgrade.html#fig5.9.2.1","level":"5.9.2","list_caption":"Figure: 登陆界面","alt":"登陆界面","nro":90,"url":"../images/kubernetes-dashboard-1.7.1-login.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"登陆界面","attributes":{},"skip":false,"key":"5.9.2.1"},{"backlink":"practice/dashboard-upgrade.html#fig5.9.2.2","level":"5.9.2","list_caption":"Figure: 首页","alt":"首页","nro":91,"url":"../images/kubernetes-dashboard-1.7.1-default-page.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"首页","attributes":{},"skip":false,"key":"5.9.2.2"},{"backlink":"practice/dashboard-upgrade.html#fig5.9.2.3","level":"5.9.2","list_caption":"Figure: 用户空间","alt":"用户空间","nro":92,"url":"../images/kubernetes-dashboard-1.7.1-brand.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"用户空间","attributes":{},"skip":false,"key":"5.9.2.3"},{"backlink":"practice/dashboard-upgrade.html#fig5.9.2.4","level":"5.9.2","list_caption":"Figure: kubeconfig文件","alt":"kubeconfig文件","nro":93,"url":"../images/brand-kubeconfig-yaml.jpg","index":4,"caption_template":"图片 - _CAPTION_","label":"kubeconfig文件","attributes":{},"skip":false,"key":"5.9.2.4"},{"backlink":"usecases/microservices.html#fig6.2.1","level":"6.2","list_caption":"Figure: 微服务关注的部分","alt":"微服务关注的部分","nro":94,"url":"../images/microservices-concerns.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"微服务关注的部分","attributes":{},"skip":false,"key":"6.2.1"},{"backlink":"usecases/service-discovery-in-microservices.html#fig6.2.1.1","level":"6.2.1","list_caption":"Figure: 微服务中的服务发现","alt":"微服务中的服务发现","nro":95,"url":"../images/service-discovery-in-microservices.png","index":1,"caption_template":"图片 - _CAPTION_","label":"微服务中的服务发现","attributes":{},"skip":false,"key":"6.2.1.1"},{"backlink":"usecases/service-mesh.html#fig6.3.1","level":"6.3","list_caption":"Figure: 下一代异构微服务架构","alt":"下一代异构微服务架构","nro":96,"url":"../images/polyglot-microservices-serivce-mesh.png","index":1,"caption_template":"图片 - _CAPTION_","label":"下一代异构微服务架构","attributes":{},"skip":false,"key":"6.3.1"},{"backlink":"usecases/service-mesh.html#fig6.3.2","level":"6.3","list_caption":"Figure: Service Mesh 架构图","alt":"Service Mesh 架构图","nro":97,"url":"../images/serivce-mesh-control-plane.png","index":2,"caption_template":"图片 - _CAPTION_","label":"Service Mesh 架构图","attributes":{},"skip":false,"key":"6.3.2"},{"backlink":"usecases/service-mesh.html#fig6.3.3","level":"6.3","list_caption":"Figure: Istio vs linkerd","alt":"Istio vs linkerd","nro":98,"url":"../images/istio-vs-linkerd.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"Istio vs linkerd","attributes":{},"skip":false,"key":"6.3.3"},{"backlink":"usecases/istio.html#fig6.3.1.1","level":"6.3.1","list_caption":"Figure: Istio架构图","alt":"Istio架构图","nro":99,"url":"../images/istio-arch.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"Istio架构图","attributes":{},"skip":false,"key":"6.3.1.1"},{"backlink":"usecases/istio-installation.html#fig6.3.1.1.1","level":"6.3.1.1","list_caption":"Figure: BookInfo Sample应用架构图","alt":"BookInfo Sample应用架构图","nro":100,"url":"../images/bookinfo-sample-arch.png","index":1,"caption_template":"图片 - _CAPTION_","label":"BookInfo Sample应用架构图","attributes":{},"skip":false,"key":"6.3.1.1.1"},{"backlink":"usecases/istio-installation.html#fig6.3.1.1.2","level":"6.3.1.1","list_caption":"Figure: BookInfo Sample页面","alt":"BookInfo Sample页面","nro":101,"url":"../images/bookinfo-sample.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"BookInfo Sample页面","attributes":{},"skip":false,"key":"6.3.1.1.2"},{"backlink":"usecases/istio-installation.html#fig6.3.1.1.3","level":"6.3.1.1","list_caption":"Figure: Istio Grafana界面","alt":"Istio Grafana界面","nro":102,"url":"../images/istio-grafana.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"Istio Grafana界面","attributes":{},"skip":false,"key":"6.3.1.1.3"},{"backlink":"usecases/istio-installation.html#fig6.3.1.1.4","level":"6.3.1.1","list_caption":"Figure: Prometheus页面","alt":"Prometheus页面","nro":103,"url":"../images/istio-prometheus.jpg","index":4,"caption_template":"图片 - _CAPTION_","label":"Prometheus页面","attributes":{},"skip":false,"key":"6.3.1.1.4"},{"backlink":"usecases/istio-installation.html#fig6.3.1.1.5","level":"6.3.1.1","list_caption":"Figure: Zipkin页面","alt":"Zipkin页面","nro":104,"url":"../images/istio-zipkin.jpg","index":5,"caption_template":"图片 - _CAPTION_","label":"Zipkin页面","attributes":{},"skip":false,"key":"6.3.1.1.5"},{"backlink":"usecases/istio-installation.html#fig6.3.1.1.6","level":"6.3.1.1","list_caption":"Figure: ServiceGraph页面","alt":"ServiceGraph页面","nro":105,"url":"../images/istio-servicegraph.jpg","index":6,"caption_template":"图片 - _CAPTION_","label":"ServiceGraph页面","attributes":{},"skip":false,"key":"6.3.1.1.6"},{"backlink":"usecases/install-and-expand-istio-mesh.html#fig6.3.1.3.1","level":"6.3.1.3","list_caption":"Figure: BookInfo","alt":"BookInfo","nro":106,"url":"../images/noistio.png","index":1,"caption_template":"图片 - _CAPTION_","label":"BookInfo","attributes":{},"skip":false,"key":"6.3.1.3.1"},{"backlink":"usecases/install-and-expand-istio-mesh.html#fig6.3.1.3.2","level":"6.3.1.3","list_caption":"Figure: BookInfo","alt":"BookInfo","nro":107,"url":"../images/noistio.png","index":2,"caption_template":"图片 - _CAPTION_","label":"BookInfo","attributes":{},"skip":false,"key":"6.3.1.3.2"},{"backlink":"usecases/integrating-vms.html#fig6.3.1.4.1","level":"6.3.1.4","list_caption":"Figure: Bookinfo应用的拓展Mesh","alt":"Bookinfo应用的拓展Mesh","nro":108,"url":"../images/istio-mesh-expansion.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"Bookinfo应用的拓展Mesh","attributes":{},"skip":false,"key":"6.3.1.4.1"},{"backlink":"usecases/linkerd.html#fig6.3.2.1","level":"6.3.2","list_caption":"Figure: source https://linkerd.io","alt":"source https://linkerd.io","nro":109,"url":"../images/diagram-individual-instance.png","index":1,"caption_template":"图片 - _CAPTION_","label":"source https://linkerd.io","attributes":{},"skip":false,"key":"6.3.2.1"},{"backlink":"usecases/linkerd-user-guide.html#fig6.3.2.1.1","level":"6.3.2.1","list_caption":"Figure: Jenkins pipeline","alt":"Jenkins pipeline","nro":110,"url":"../images/linkerd-jenkins-pipeline.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"Jenkins pipeline","attributes":{},"skip":false,"key":"6.3.2.1.1"},{"backlink":"usecases/linkerd-user-guide.html#fig6.3.2.1.2","level":"6.3.2.1","list_caption":"Figure: Jenkins config","alt":"Jenkins config","nro":111,"url":"../images/linkerd-jenkins.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"Jenkins config","attributes":{},"skip":false,"key":"6.3.2.1.2"},{"backlink":"usecases/linkerd-user-guide.html#fig6.3.2.1.3","level":"6.3.2.1","list_caption":"Figure: namerd","alt":"namerd","nro":112,"url":"../images/namerd-internal.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"namerd","attributes":{},"skip":false,"key":"6.3.2.1.3"},{"backlink":"usecases/linkerd-user-guide.html#fig6.3.2.1.4","level":"6.3.2.1","list_caption":"Figure: linkerd监控","alt":"linkerd监控","nro":113,"url":"../images/linkerd-helloworld-outgoing.jpg","index":4,"caption_template":"图片 - _CAPTION_","label":"linkerd监控","attributes":{},"skip":false,"key":"6.3.2.1.4"},{"backlink":"usecases/linkerd-user-guide.html#fig6.3.2.1.5","level":"6.3.2.1","list_caption":"Figure: linkerd监控","alt":"linkerd监控","nro":114,"url":"../images/linkerd-helloworld-incoming.jpg","index":5,"caption_template":"图片 - _CAPTION_","label":"linkerd监控","attributes":{},"skip":false,"key":"6.3.2.1.5"},{"backlink":"usecases/linkerd-user-guide.html#fig6.3.2.1.6","level":"6.3.2.1","list_caption":"Figure: linkerd性能监控","alt":"linkerd性能监控","nro":115,"url":"../images/linkerd-grafana.png","index":6,"caption_template":"图片 - _CAPTION_","label":"linkerd性能监控","attributes":{},"skip":false,"key":"6.3.2.1.6"},{"backlink":"usecases/linkerd-user-guide.html#fig6.3.2.1.7","level":"6.3.2.1","list_caption":"Figure: Linkerd ingress controller","alt":"Linkerd ingress controller","nro":116,"url":"../images/linkerd-ingress-controller.jpg","index":7,"caption_template":"图片 - _CAPTION_","label":"Linkerd ingress controller","attributes":{},"skip":false,"key":"6.3.2.1.7"},{"backlink":"usecases/conduit-installation.html#fig6.3.3.2.1","level":"6.3.3.2","list_caption":"Figure: Conduit dashboard","alt":"Conduit dashboard","nro":117,"url":"../images/conduit-dashboard.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"Conduit dashboard","attributes":{},"skip":false,"key":"6.3.3.2.1"},{"backlink":"usecases/big-data.html#fig6.4.1","level":"6.4","list_caption":"Figure: Spark on yarn with kubernetes","alt":"Spark on yarn with kubernetes","nro":118,"url":"../images/spark-on-yarn-with-kubernetes.png","index":1,"caption_template":"图片 - _CAPTION_","label":"Spark on yarn with kubernetes","attributes":{},"skip":false,"key":"6.4.1"},{"backlink":"usecases/big-data.html#fig6.4.2","level":"6.4","list_caption":"Figure: 在kubernetes上使用多种调度方式","alt":"在kubernetes上使用多种调度方式","nro":119,"url":"../images/spark-on-kubernetes-with-different-schedulers.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"在kubernetes上使用多种调度方式","attributes":{},"skip":false,"key":"6.4.2"},{"backlink":"usecases/spark-standalone-on-kubernetes.html#fig6.4.1.1","level":"6.4.1","list_caption":"Figure: spark master ui","alt":"spark master ui","nro":120,"url":"../images/spark-ui.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"spark master ui","attributes":{},"skip":false,"key":"6.4.1.1"},{"backlink":"usecases/spark-standalone-on-kubernetes.html#fig6.4.1.2","level":"6.4.1","list_caption":"Figure: zeppelin ui","alt":"zeppelin ui","nro":121,"url":"../images/zeppelin-ui.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"zeppelin ui","attributes":{},"skip":false,"key":"6.4.1.2"},{"backlink":"usecases/serverless.html#fig6.5.1","level":"6.5","list_caption":"Figure: 从物理机到函数计算","alt":"从物理机到函数计算","nro":122,"url":"../images/from-bare-metal-to-functions.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"从物理机到函数计算","attributes":{},"skip":false,"key":"6.5.1"},{"backlink":"usecases/serverless.html#fig6.5.2","level":"6.5","list_caption":"Figure: FaaS Landscape","alt":"FaaS Landscape","nro":123,"url":"../images/redpoint-faas-landscape.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"FaaS Landscape","attributes":{},"skip":false,"key":"6.5.2"},{"backlink":"usecases/understanding-serverless.html#fig6.5.1.1","level":"6.5.1","list_caption":"Figure: 服务端软件的运行环境","alt":"服务端软件的运行环境","nro":124,"url":"../images/serverless-server-side-software.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"服务端软件的运行环境","attributes":{},"skip":false,"key":"6.5.1.1"},{"backlink":"usecases/understanding-serverless.html#fig6.5.1.2","level":"6.5.1","list_caption":"Figure: FaaS应用架构","alt":"FaaS应用架构","nro":125,"url":"../images/serverless-faas-platform.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"FaaS应用架构","attributes":{},"skip":false,"key":"6.5.1.2"},{"backlink":"usecases/understanding-serverless.html#fig6.5.1.3","level":"6.5.1","list_caption":"Figure: 传统应用程序架构","alt":"传统应用程序架构","nro":126,"url":"../images/non-serverless-game-arch.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"传统应用程序架构","attributes":{},"skip":false,"key":"6.5.1.3"},{"backlink":"usecases/understanding-serverless.html#fig6.5.1.4","level":"6.5.1","list_caption":"Figure: Serverless架构","alt":"Serverless架构","nro":127,"url":"../images/serverless-game-arch.jpg","index":4,"caption_template":"图片 - _CAPTION_","label":"Serverless架构","attributes":{},"skip":false,"key":"6.5.1.4"},{"backlink":"usecases/openfaas.html#fig6.5.2.1.1","level":"6.5.2.1","list_caption":"Figure: OpenFaaS架构","alt":"OpenFaaS架构","nro":128,"url":"../images/openfaas-arch.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"OpenFaaS架构","attributes":{},"skip":false,"key":"6.5.2.1.1"},{"backlink":"usecases/openfaas.html#fig6.5.2.1.2","level":"6.5.2.1","list_caption":"Figure: OpenFaaS Prometheus","alt":"OpenFaaS Prometheus","nro":129,"url":"../images/openfaas-prometheus.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"OpenFaaS Prometheus","attributes":{},"skip":false,"key":"6.5.2.1.2"},{"backlink":"usecases/openfaas.html#fig6.5.2.1.3","level":"6.5.2.1","list_caption":"Figure: OpenFaas Grafana监控","alt":"OpenFaas Grafana监控","nro":130,"url":"../images/openfaas-grafana.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"OpenFaas Grafana监控","attributes":{},"skip":false,"key":"6.5.2.1.3"},{"backlink":"usecases/openfaas.html#fig6.5.2.1.4","level":"6.5.2.1","list_caption":"Figure: OpenFaas Dashboard","alt":"OpenFaas Dashboard","nro":131,"url":"../images/openfaas-deploy-a-function.jpg","index":4,"caption_template":"图片 - _CAPTION_","label":"OpenFaas Dashboard","attributes":{},"skip":false,"key":"6.5.2.1.4"},{"backlink":"usecases/openfaas.html#fig6.5.2.1.5","level":"6.5.2.1","list_caption":"Figure: NodeInfo执行结果","alt":"NodeInfo执行结果","nro":132,"url":"../images/openfaas-nodeinfo.jpg","index":5,"caption_template":"图片 - _CAPTION_","label":"NodeInfo执行结果","attributes":{},"skip":false,"key":"6.5.2.1.5"},{"backlink":"develop/client-go-sample.html#fig7.5.1","level":"7.5","list_caption":"Figure: 使用kubernetes dashboard进行故障排查","alt":"使用kubernetes dashboard进行故障排查","nro":133,"url":"../images/kubernetes-client-go-sample-update.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"使用kubernetes dashboard进行故障排查","attributes":{},"skip":false,"key":"7.5.1"},{"backlink":"appendix/issues.html#fig8.6.1","level":"8.6","list_caption":"Figure: pvc-storage-limit","alt":"pvc-storage-limit","nro":134,"url":"../images/pvc-storage-limit.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"pvc-storage-limit","attributes":{},"skip":false,"key":"8.6.1"}]},"title":"Kubernetes Handbook - jimmysong.io","language":"zh-hans","links":{"sidebar":{"Jimmy Song":"https://jimmysong.io"}},"gitbook":"*","description":"Kubernetes中文指南/实践手册|kubernetes handbook|jimmysong.io|宋净超"},"file":{"path":"README.md","mtime":"2017-12-15T11:49:07.697Z","type":"markdown"},"gitbook":{"version":"3.2.2","time":"2017-12-20T09:26:03.762Z"},"basePath":".","book":{"language":""}}); + gitbook.page.hasChanged({"page":{"title":"序言","level":"1.1","depth":1,"next":{"title":"Kubernetes与云原生应用概览","level":"2.1","depth":1,"path":"cloud-native/kubernetes-and-cloud-native-app-overview.md","ref":"cloud-native/kubernetes-and-cloud-native-app-overview.md","articles":[]},"dir":"ltr"},"config":{"plugins":["github","codesnippet","splitter","page-toc-button","image-captions","editlink","back-to-top-button","-lunr","-search","search-plus","github-buttons@2.1.0","favicon@^0.0.2","tbfed-pagefooter@^0.0.1","3-ba","theme-default"],"styles":{"website":"styles/website.css","pdf":"styles/pdf.css","epub":"styles/epub.css","mobi":"styles/mobi.css","ebook":"styles/ebook.css","print":"styles/print.css"},"pluginsConfig":{"tbfed-pagefooter":{"copyright":"Copyright © jimmysong.io 2017","modify_label":"Updated:","modify_format":"YYYY-MM-DD HH:mm:ss"},"github":{"url":"https://github.com/rootsongjc/kubernetes-handbook"},"editlink":{"label":"编辑本页","multilingual":false,"base":"https://github.com/rootsongjc/kubernetes-handbook/blob/master/"},"splitter":{},"codesnippet":{},"fontsettings":{"theme":"white","family":"sans","size":2},"highlight":{},"favicon":{"shortcut":"favicon.ico","bookmark":"favicon.ico"},"page-toc-button":{},"back-to-top-button":{},"github-buttons":{"repo":"rootsongjc/kubernetes-handbook","types":["star"],"size":"small"},"3-ba":{"configuration":"auto","token":"11f7d254cfa4e0ca44b175c66d379ecc"},"sharing":{"facebook":true,"twitter":true,"google":false,"weibo":false,"instapaper":false,"vk":false,"all":["facebook","google","twitter","weibo","instapaper"]},"theme-default":{"styles":{"website":"styles/website.css","pdf":"styles/pdf.css","epub":"styles/epub.css","mobi":"styles/mobi.css","ebook":"styles/ebook.css","print":"styles/print.css"},"showLevel":true},"search-plus":{},"image-captions":{"caption":"图片 - _CAPTION_","variable_name":"_pictures"}},"theme":"default","author":"Jimmy Song","pdf":{"pageNumbers":true,"fontSize":12,"fontFamily":"Arial","paperSize":"a4","chapterMark":"pagebreak","pageBreaksBefore":"/","margin":{"right":62,"left":62,"top":56,"bottom":56}},"structure":{"langs":"LANGS.md","readme":"README.md","glossary":"GLOSSARY.md","summary":"SUMMARY.md"},"variables":{"_pictures":[{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.1","level":"2.1","list_caption":"Figure: 云计算演进历程","alt":"云计算演进历程","nro":1,"url":"../images/cloud-computing-evolution-road.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"云计算演进历程","attributes":{},"skip":false,"key":"2.1.1"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.2","level":"2.1","list_caption":"Figure: 来自Twitter @MarcWilczek","alt":"来自Twitter @MarcWilczek","nro":2,"url":"../images/cloud-native-comes-of-age.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"来自Twitter @MarcWilczek","attributes":{},"skip":false,"key":"2.1.2"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.3","level":"2.1","list_caption":"Figure: Cloud native思维导图","alt":"Cloud native思维导图","nro":3,"url":"../images/cloud-native-architecutre-mindnode.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"Cloud native思维导图","attributes":{},"skip":false,"key":"2.1.3"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.4","level":"2.1","list_caption":"Figure: 十二因素应用","alt":"十二因素应用","nro":4,"url":"../images/12-factor-app.png","index":4,"caption_template":"图片 - _CAPTION_","label":"十二因素应用","attributes":{},"skip":false,"key":"2.1.4"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.5","level":"2.1","list_caption":"Figure: 容器生态","alt":"容器生态","nro":5,"url":"../images/container-ecosystem.png","index":5,"caption_template":"图片 - _CAPTION_","label":"容器生态","attributes":{},"skip":false,"key":"2.1.5"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.6","level":"2.1","list_caption":"Figure: 使用Jenkins进行持续集成与发布流程图","alt":"使用Jenkins进行持续集成与发布流程图","nro":6,"url":"../images/kubernetes-jenkins-ci-cd.png","index":6,"caption_template":"图片 - _CAPTION_","label":"使用Jenkins进行持续集成与发布流程图","attributes":{},"skip":false,"key":"2.1.6"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.7","level":"2.1","list_caption":"Figure: filebeat日志收集架构图","alt":"filebeat日志收集架构图","nro":7,"url":"../images/filebeat-log-collector-arch.png","index":7,"caption_template":"图片 - _CAPTION_","label":"filebeat日志收集架构图","attributes":{},"skip":false,"key":"2.1.7"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.8","level":"2.1","list_caption":"Figure: API文档","alt":"API文档","nro":8,"url":"../images/k8s-app-monitor-test-api-doc.jpg","index":8,"caption_template":"图片 - _CAPTION_","label":"API文档","attributes":{},"skip":false,"key":"2.1.8"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.9","level":"2.1","list_caption":"Figure: 迁移步骤示意图","alt":"迁移步骤示意图","nro":9,"url":"../images/migrating-hadoop-yarn-to-kubernetes.png","index":9,"caption_template":"图片 - _CAPTION_","label":"迁移步骤示意图","attributes":{},"skip":false,"key":"2.1.9"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.10","level":"2.1","list_caption":"Figure: service mesh架构图","alt":"service mesh架构图","nro":10,"url":"../images/serivce-mesh-control-plane.png","index":10,"caption_template":"图片 - _CAPTION_","label":"service mesh架构图","attributes":{},"skip":false,"key":"2.1.10"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.11","level":"2.1","list_caption":"Figure: kibana界面","alt":"kibana界面","nro":11,"url":"../images/filebeat-docker-test.jpg","index":11,"caption_template":"图片 - _CAPTION_","label":"kibana界面","attributes":{},"skip":false,"key":"2.1.11"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.12","level":"2.1","list_caption":"Figure: Grafana界面示意图1","alt":"Grafana界面示意图1","nro":12,"url":"../images/kubernetes-devops-example-grafana-1.png","index":12,"caption_template":"图片 - _CAPTION_","label":"Grafana界面示意图1","attributes":{},"skip":false,"key":"2.1.12"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.13","level":"2.1","list_caption":"Figure: Grafana界面示意图2","alt":"Grafana界面示意图2","nro":13,"url":"../images/kubernetes-devops-example-grafana-2.png","index":13,"caption_template":"图片 - _CAPTION_","label":"Grafana界面示意图2","attributes":{},"skip":false,"key":"2.1.13"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.14","level":"2.1","list_caption":"Figure: Grafana界面示意图3","alt":"Grafana界面示意图3","nro":14,"url":"../images/kubernetes-devops-example-grafana-3.png","index":14,"caption_template":"图片 - _CAPTION_","label":"Grafana界面示意图3","attributes":{},"skip":false,"key":"2.1.14"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.15","level":"2.1","list_caption":"Figure: dashboard","alt":"dashboard","nro":15,"url":"../images/spark-job-on-kubernetes-example-1.jpg","index":15,"caption_template":"图片 - _CAPTION_","label":"dashboard","attributes":{},"skip":false,"key":"2.1.15"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.16","level":"2.1","list_caption":"Figure: Grafana","alt":"Grafana","nro":16,"url":"../images/spark-job-on-kubernetes-example-2.jpg","index":16,"caption_template":"图片 - _CAPTION_","label":"Grafana","attributes":{},"skip":false,"key":"2.1.16"},{"backlink":"concepts/index.html#fig3.1.1","level":"3.1","list_caption":"Figure: Borg架构","alt":"Borg架构","nro":17,"url":"../images/borg.png","index":1,"caption_template":"图片 - _CAPTION_","label":"Borg架构","attributes":{},"skip":false,"key":"3.1.1"},{"backlink":"concepts/index.html#fig3.1.2","level":"3.1","list_caption":"Figure: Kubernetes架构","alt":"Kubernetes架构","nro":18,"url":"../images/architecture.png","index":2,"caption_template":"图片 - _CAPTION_","label":"Kubernetes架构","attributes":{},"skip":false,"key":"3.1.2"},{"backlink":"concepts/index.html#fig3.1.3","level":"3.1","list_caption":"Figure: Kuberentes架构（图片来自于网络）","alt":"Kuberentes架构（图片来自于网络）","nro":19,"url":"../images/kubernetes-high-level-component-archtecture.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"Kuberentes架构（图片来自于网络）","attributes":{},"skip":false,"key":"3.1.3"},{"backlink":"concepts/index.html#fig3.1.4","level":"3.1","list_caption":"Figure: kubernetes整体架构示意图","alt":"kubernetes整体架构示意图","nro":20,"url":"../images/kubernetes-whole-arch.png","index":4,"caption_template":"图片 - _CAPTION_","label":"kubernetes整体架构示意图","attributes":{},"skip":false,"key":"3.1.4"},{"backlink":"concepts/index.html#fig3.1.5","level":"3.1","list_caption":"Figure: Kubernetes master架构示意图","alt":"Kubernetes master架构示意图","nro":21,"url":"../images/kubernetes-master-arch.png","index":5,"caption_template":"图片 - _CAPTION_","label":"Kubernetes master架构示意图","attributes":{},"skip":false,"key":"3.1.5"},{"backlink":"concepts/index.html#fig3.1.6","level":"3.1","list_caption":"Figure: kubernetes node架构示意图","alt":"kubernetes node架构示意图","nro":22,"url":"../images/kubernetes-node-arch.png","index":6,"caption_template":"图片 - _CAPTION_","label":"kubernetes node架构示意图","attributes":{},"skip":false,"key":"3.1.6"},{"backlink":"concepts/index.html#fig3.1.7","level":"3.1","list_caption":"Figure: Kubernetes分层架构示意图","alt":"Kubernetes分层架构示意图","nro":23,"url":"../images/kubernetes-layers-arch.jpg","index":7,"caption_template":"图片 - _CAPTION_","label":"Kubernetes分层架构示意图","attributes":{},"skip":false,"key":"3.1.7"},{"backlink":"concepts/concepts.html#fig3.2.1","level":"3.2","list_caption":"Figure: 分层架构示意图","alt":"分层架构示意图","nro":24,"url":"../images/kubernetes-layers-arch.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"分层架构示意图","attributes":{},"skip":false,"key":"3.2.1"},{"backlink":"concepts/pod-overview.html#fig3.3.1.1","level":"3.3.1","list_caption":"Figure: pod diagram","alt":"pod diagram","nro":25,"url":"../images/pod-overview.png","index":1,"caption_template":"图片 - _CAPTION_","label":"pod diagram","attributes":{},"skip":false,"key":"3.3.1.1"},{"backlink":"concepts/pod.html#fig3.3.1.1.1","level":"3.3.1.1","list_caption":"Figure: Pod示意图","alt":"Pod示意图","nro":26,"url":"../images/pod-overview.png","index":1,"caption_template":"图片 - _CAPTION_","label":"Pod示意图","attributes":{},"skip":false,"key":"3.3.1.1.1"},{"backlink":"concepts/pod.html#fig3.3.1.1.2","level":"3.3.1.1","list_caption":"Figure: Pod Cheatsheet","alt":"Pod Cheatsheet","nro":27,"url":"../images/kubernetes-pod-cheatsheet.png","index":2,"caption_template":"图片 - _CAPTION_","label":"Pod Cheatsheet","attributes":{},"skip":false,"key":"3.3.1.1.2"},{"backlink":"concepts/service.html#fig3.3.4.1","level":"3.3.4","list_caption":"Figure: userspace代理模式下Service概览图","alt":"userspace代理模式下Service概览图","nro":28,"url":"../images/services-userspace-overview.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"userspace代理模式下Service概览图","attributes":{},"skip":false,"key":"3.3.4.1"},{"backlink":"concepts/service.html#fig3.3.4.2","level":"3.3.4","list_caption":"Figure: iptables代理模式下Service概览图","alt":"iptables代理模式下Service概览图","nro":29,"url":"../images/services-iptables-overview.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"iptables代理模式下Service概览图","attributes":{},"skip":false,"key":"3.3.4.2"},{"backlink":"concepts/deployment.html#fig3.3.6.1","level":"3.3.6","list_caption":"Figure: kubernetes deployment cheatsheet","alt":"kubernetes deployment cheatsheet","nro":30,"url":"../images/deployment-cheatsheet.png","index":1,"caption_template":"图片 - _CAPTION_","label":"kubernetes deployment cheatsheet","attributes":{},"skip":false,"key":"3.3.6.1"},{"backlink":"concepts/horizontal-pod-autoscaling.html#fig3.3.16.1","level":"3.3.16","list_caption":"Figure: horizontal-pod-autoscaler","alt":"horizontal-pod-autoscaler","nro":31,"url":"../images/horizontal-pod-autoscaler.png","index":1,"caption_template":"图片 - _CAPTION_","label":"horizontal-pod-autoscaler","attributes":{},"skip":false,"key":"3.3.16.1"},{"backlink":"concepts/label.html#fig3.3.17.1","level":"3.3.17","list_caption":"Figure: label示意图","alt":"label示意图","nro":32,"url":"../images/labels.png","index":1,"caption_template":"图片 - _CAPTION_","label":"label示意图","attributes":{},"skip":false,"key":"3.3.17.1"},{"backlink":"guide/using-kubectl.html#fig4.3.2.1","level":"4.3.2","list_caption":"Figure: kubectl cheatsheet","alt":"kubectl cheatsheet","nro":33,"url":"../images/kubernetes-kubectl-cheatsheet.png","index":1,"caption_template":"图片 - _CAPTION_","label":"kubectl cheatsheet","attributes":{},"skip":false,"key":"4.3.2.1"},{"backlink":"guide/using-kubectl.html#fig4.3.2.2","level":"4.3.2","list_caption":"Figure: kube-shell页面","alt":"kube-shell页面","nro":34,"url":"../images/kube-shell.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"kube-shell页面","attributes":{},"skip":false,"key":"4.3.2.2"},{"backlink":"guide/ip-masq-agent.html#fig4.4.6.1","level":"4.4.6","list_caption":"Figure: IP伪装代理示意图","alt":"IP伪装代理示意图","nro":35,"url":"../images/ip-masq.png","index":1,"caption_template":"图片 - _CAPTION_","label":"IP伪装代理示意图","attributes":{},"skip":false,"key":"4.4.6.1"},{"backlink":"guide/auth-with-kubeconfig-or-token.html#fig4.4.7.1","level":"4.4.7","list_caption":"Figure: kubeconfig文件","alt":"kubeconfig文件","nro":36,"url":"../images/brand-kubeconfig-yaml.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"kubeconfig文件","attributes":{},"skip":false,"key":"4.4.7.1"},{"backlink":"guide/authentication.html#fig4.4.8.1","level":"4.4.8","list_caption":"Figure: Kubernetes OpenID Connect Flow","alt":"Kubernetes OpenID Connect Flow","nro":37,"url":"../images/kubernetes-oidc-login.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"Kubernetes OpenID Connect Flow","attributes":{},"skip":false,"key":"4.4.8.1"},{"backlink":"guide/carbin-mobile-dashboard-for-kubernetes.html#fig4.5.6.1","level":"4.5.6","list_caption":"Figure: App Store","alt":"App Store","nro":38,"url":"../images/carbin-kubernetes-mobile-dashboard-1.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"App Store","attributes":{},"skip":false,"key":"4.5.6.1"},{"backlink":"guide/carbin-mobile-dashboard-for-kubernetes.html#fig4.5.6.2","level":"4.5.6","list_caption":"Figure: 在手机上操作Kubernetes集群","alt":"在手机上操作Kubernetes集群","nro":39,"url":"../images/carbin-kubernetes-mobile-dashboard-4.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"在手机上操作Kubernetes集群","attributes":{},"skip":false,"key":"4.5.6.2"},{"backlink":"guide/kubernetes-desktop-client.html#fig4.5.7.1","level":"4.5.7","list_caption":"Figure: Kubernetic客户端","alt":"Kubernetic客户端","nro":40,"url":"../images/kubernetic-desktop-ui.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"Kubernetic客户端","attributes":{},"skip":false,"key":"4.5.7.1"},{"backlink":"guide/deploy-applications-in-kubernetes.html#fig4.6.1.1","level":"4.6.1","list_caption":"Figure: API","alt":"API","nro":41,"url":"../images/k8s-app-monitor-test-api-doc.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"API","attributes":{},"skip":false,"key":"4.6.1.1"},{"backlink":"guide/deploy-applications-in-kubernetes.html#fig4.6.1.2","level":"4.6.1","list_caption":"Figure: wercker","alt":"wercker","nro":42,"url":"../images/k8s-app-monitor-agent-wercker.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"wercker","attributes":{},"skip":false,"key":"4.6.1.2"},{"backlink":"guide/deploy-applications-in-kubernetes.html#fig4.6.1.3","level":"4.6.1","list_caption":"Figure: 图表","alt":"图表","nro":43,"url":"../images/k8s-app-monitor-agent.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"图表","attributes":{},"skip":false,"key":"4.6.1.3"},{"backlink":"guide/migrating-hadoop-yarn-to-kubernetes.html#fig4.6.2.1","level":"4.6.2","list_caption":"Figure: spark on yarn with kubernetes","alt":"spark on yarn with kubernetes","nro":44,"url":"../images/spark-on-yarn-with-kubernetes.png","index":1,"caption_template":"图片 - _CAPTION_","label":"spark on yarn with kubernetes","attributes":{},"skip":false,"key":"4.6.2.1"},{"backlink":"guide/migrating-hadoop-yarn-to-kubernetes.html#fig4.6.2.2","level":"4.6.2","list_caption":"Figure: Terms","alt":"Terms","nro":45,"url":"../images/terms-in-kubernetes-app-deployment.png","index":2,"caption_template":"图片 - _CAPTION_","label":"Terms","attributes":{},"skip":false,"key":"4.6.2.2"},{"backlink":"guide/migrating-hadoop-yarn-to-kubernetes.html#fig4.6.2.3","level":"4.6.2","list_caption":"Figure: 分解步骤解析","alt":"分解步骤解析","nro":46,"url":"../images/migrating-hadoop-yarn-to-kubernetes.png","index":3,"caption_template":"图片 - _CAPTION_","label":"分解步骤解析","attributes":{},"skip":false,"key":"4.6.2.3"},{"backlink":"practice/node-installation.html#fig5.2.7.1","level":"5.2.7","list_caption":"Figure: welcome nginx","alt":"welcome nginx","nro":47,"url":"../images/kubernetes-installation-test-nginx.png","index":1,"caption_template":"图片 - _CAPTION_","label":"welcome nginx","attributes":{},"skip":false,"key":"5.2.7.1"},{"backlink":"practice/dashboard-addon-installation.html#fig5.2.9.1","level":"5.2.9","list_caption":"Figure: kubernetes dashboard","alt":"kubernetes dashboard","nro":48,"url":"../images/kubernetes-dashboard-raw.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"kubernetes dashboard","attributes":{},"skip":false,"key":"5.2.9.1"},{"backlink":"practice/dashboard-addon-installation.html#fig5.2.9.2","level":"5.2.9","list_caption":"Figure: V1.6.3版本的dashboard界面","alt":"V1.6.3版本的dashboard界面","nro":49,"url":"../images/dashboard-v163.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"V1.6.3版本的dashboard界面","attributes":{},"skip":false,"key":"5.2.9.2"},{"backlink":"practice/heapster-addon-installation.html#fig5.2.10.1","level":"5.2.10","list_caption":"Figure: dashboard-heapster","alt":"dashboard-heapster","nro":50,"url":"../images/kubernetes-dashboard-with-heapster.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"dashboard-heapster","attributes":{},"skip":false,"key":"5.2.10.1"},{"backlink":"practice/heapster-addon-installation.html#fig5.2.10.2","level":"5.2.10","list_caption":"Figure: grafana","alt":"grafana","nro":51,"url":"../images/kubernetes-heapster-grafana.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"grafana","attributes":{},"skip":false,"key":"5.2.10.2"},{"backlink":"practice/heapster-addon-installation.html#fig5.2.10.3","level":"5.2.10","list_caption":"Figure: kubernetes-influxdb-heapster","alt":"kubernetes-influxdb-heapster","nro":52,"url":"../images/kubernetes-influxdb-heapster.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"kubernetes-influxdb-heapster","attributes":{},"skip":false,"key":"5.2.10.3"},{"backlink":"practice/heapster-addon-installation.html#fig5.2.10.4","level":"5.2.10","list_caption":"Figure: 修改grafana模板","alt":"修改grafana模板","nro":53,"url":"../images/grafana-dashboard-setting.jpg","index":4,"caption_template":"图片 - _CAPTION_","label":"修改grafana模板","attributes":{},"skip":false,"key":"5.2.10.4"},{"backlink":"practice/efk-addon-installation.html#fig5.2.11.1","level":"5.2.11","list_caption":"Figure: es-setting","alt":"es-setting","nro":54,"url":"../images/es-setting.png","index":1,"caption_template":"图片 - _CAPTION_","label":"es-setting","attributes":{},"skip":false,"key":"5.2.11.1"},{"backlink":"practice/efk-addon-installation.html#fig5.2.11.2","level":"5.2.11","list_caption":"Figure: es-home","alt":"es-home","nro":55,"url":"../images/kubernetes-efk-kibana.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"es-home","attributes":{},"skip":false,"key":"5.2.11.2"},{"backlink":"practice/traefik-ingress-installation.html#fig5.3.1.1","level":"5.3.1","list_caption":"Figure: kubernetes-dashboard","alt":"kubernetes-dashboard","nro":56,"url":"../images/traefik-dashboard.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"kubernetes-dashboard","attributes":{},"skip":false,"key":"5.3.1.1"},{"backlink":"practice/traefik-ingress-installation.html#fig5.3.1.2","level":"5.3.1","list_caption":"Figure: traefik-nginx","alt":"traefik-nginx","nro":57,"url":"../images/traefik-nginx.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"traefik-nginx","attributes":{},"skip":false,"key":"5.3.1.2"},{"backlink":"practice/traefik-ingress-installation.html#fig5.3.1.3","level":"5.3.1","list_caption":"Figure: traefik-guestbook","alt":"traefik-guestbook","nro":58,"url":"../images/traefik-guestbook.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"traefik-guestbook","attributes":{},"skip":false,"key":"5.3.1.3"},{"backlink":"practice/distributed-load-test.html#fig5.3.2.1","level":"5.3.2","list_caption":"Figure: 使用dashboard来扩容","alt":"使用dashboard来扩容","nro":59,"url":"../images/dashbaord-scale.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"使用dashboard来扩容","attributes":{},"skip":false,"key":"5.3.2.1"},{"backlink":"practice/distributed-load-test.html#fig5.3.2.2","level":"5.3.2","list_caption":"Figure: Traefik的UI","alt":"Traefik的UI","nro":60,"url":"../images/traefik-dashboard-locust.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"Traefik的UI","attributes":{},"skip":false,"key":"5.3.2.2"},{"backlink":"practice/distributed-load-test.html#fig5.3.2.3","level":"5.3.2","list_caption":"Figure: Locust启动界面","alt":"Locust启动界面","nro":61,"url":"../images/locust-start-swarming.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"Locust启动界面","attributes":{},"skip":false,"key":"5.3.2.3"},{"backlink":"practice/distributed-load-test.html#fig5.3.2.4","level":"5.3.2","list_caption":"Figure: Dashboard查看页面","alt":"Dashboard查看页面","nro":62,"url":"../images/sample-webapp-rc.jpg","index":4,"caption_template":"图片 - _CAPTION_","label":"Dashboard查看页面","attributes":{},"skip":false,"key":"5.3.2.4"},{"backlink":"practice/distributed-load-test.html#fig5.3.2.5","level":"5.3.2","list_caption":"Figure: Locust测试结果页面","alt":"Locust测试结果页面","nro":63,"url":"../images/locust-dashboard.jpg","index":5,"caption_template":"图片 - _CAPTION_","label":"Locust测试结果页面","attributes":{},"skip":false,"key":"5.3.2.5"},{"backlink":"practice/network-and-cluster-perfermance-test.html#fig5.3.3.1","level":"5.3.3","list_caption":"Figure: kubernetes-dashboard","alt":"kubernetes-dashboard","nro":64,"url":"../images/kubenetes-e2e-test.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"kubernetes-dashboard","attributes":{},"skip":false,"key":"5.3.3.1"},{"backlink":"practice/network-and-cluster-perfermance-test.html#fig5.3.3.2","level":"5.3.3","list_caption":"Figure: locust测试页面","alt":"locust测试页面","nro":65,"url":"../images/kubernetes-locust-test.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"locust测试页面","attributes":{},"skip":false,"key":"5.3.3.2"},{"backlink":"practice/edge-node-configuration.html#fig5.3.4.1","level":"5.3.4","list_caption":"Figure: 边缘节点架构","alt":"边缘节点架构","nro":66,"url":"../images/kubernetes-edge-node-architecture.png","index":1,"caption_template":"图片 - _CAPTION_","label":"边缘节点架构","attributes":{},"skip":false,"key":"5.3.4.1"},{"backlink":"practice/app-log-collection.html#fig5.4.2.1","level":"5.4.2","list_caption":"Figure: filebeat日志收集架构图","alt":"filebeat日志收集架构图","nro":67,"url":"../images/filebeat-log-collector.png","index":1,"caption_template":"图片 - _CAPTION_","label":"filebeat日志收集架构图","attributes":{},"skip":false,"key":"5.4.2.1"},{"backlink":"practice/app-log-collection.html#fig5.4.2.2","level":"5.4.2","list_caption":"Figure: Kibana页面","alt":"Kibana页面","nro":68,"url":"../images/filebeat-docker-test.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"Kibana页面","attributes":{},"skip":false,"key":"5.4.2.2"},{"backlink":"practice/app-log-collection.html#fig5.4.2.3","level":"5.4.2","list_caption":"Figure: filebeat收集的日志详细信息","alt":"filebeat收集的日志详细信息","nro":69,"url":"../images/kubernetes-filebeat-detail.png","index":3,"caption_template":"图片 - _CAPTION_","label":"filebeat收集的日志详细信息","attributes":{},"skip":false,"key":"5.4.2.3"},{"backlink":"practice/monitor.html#fig5.4.4.1","level":"5.4.4","list_caption":"Figure: Kubernetes集群中的监控","alt":"Kubernetes集群中的监控","nro":70,"url":"../images/monitoring-in-kubernetes.png","index":1,"caption_template":"图片 - _CAPTION_","label":"Kubernetes集群中的监控","attributes":{},"skip":false,"key":"5.4.4.1"},{"backlink":"practice/monitor.html#fig5.4.4.2","level":"5.4.4","list_caption":"Figure: kubernetes的容器命名规则示意图","alt":"kubernetes的容器命名规则示意图","nro":71,"url":"../images/kubernetes-container-naming-rule.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"kubernetes的容器命名规则示意图","attributes":{},"skip":false,"key":"5.4.4.2"},{"backlink":"practice/monitor.html#fig5.4.4.3","level":"5.4.4","list_caption":"Figure: Heapster架构图（改进版）","alt":"Heapster架构图（改进版）","nro":72,"url":"../images/kubernetes-heapster-monitoring.png","index":3,"caption_template":"图片 - _CAPTION_","label":"Heapster架构图（改进版）","attributes":{},"skip":false,"key":"5.4.4.3"},{"backlink":"practice/monitor.html#fig5.4.4.4","level":"5.4.4","list_caption":"Figure: 应用监控架构图","alt":"应用监控架构图","nro":73,"url":"../images/kubernetes-app-monitoring.png","index":4,"caption_template":"图片 - _CAPTION_","label":"应用监控架构图","attributes":{},"skip":false,"key":"5.4.4.4"},{"backlink":"practice/monitor.html#fig5.4.4.5","level":"5.4.4","list_caption":"Figure: 应用拓扑图","alt":"应用拓扑图","nro":74,"url":"../images/weave-scope-service-topology.jpg","index":5,"caption_template":"图片 - _CAPTION_","label":"应用拓扑图","attributes":{},"skip":false,"key":"5.4.4.5"},{"backlink":"practice/data-persistence-problem.html#fig5.4.5.1","level":"5.4.5","list_caption":"Figure: 日志持久化收集解决方案示意图","alt":"日志持久化收集解决方案示意图","nro":75,"url":"../images/log-persistence-logstash.png","index":1,"caption_template":"图片 - _CAPTION_","label":"日志持久化收集解决方案示意图","attributes":{},"skip":false,"key":"5.4.5.1"},{"backlink":"practice/storage-for-containers-using-glusterfs-with-openshift.html#fig5.5.1.2.1","level":"5.5.1.2","list_caption":"Figure: 创建存储","alt":"创建存储","nro":76,"url":"../images/create-gluster-storage.png","index":1,"caption_template":"图片 - _CAPTION_","label":"创建存储","attributes":{},"skip":false,"key":"5.5.1.2.1"},{"backlink":"practice/storage-for-containers-using-glusterfs-with-openshift.html#fig5.5.1.2.2","level":"5.5.1.2","list_caption":"Figure: Screen Shot 2017-03-24 at 11.09.34.png","alt":"Screen Shot 2017-03-24 at 11.09.34.png","nro":77,"url":"https://keithtenzer.files.wordpress.com/2017/03/screen-shot-2017-03-24-at-11-09-341.png?w=440","index":2,"caption_template":"图片 - _CAPTION_","label":"Screen Shot 2017-03-24 at 11.09.34.png","attributes":{},"skip":false,"key":"5.5.1.2.2"},{"backlink":"practice/using-heapster-to-get-object-metrics.html#fig5.6.1.1.1","level":"5.6.1.1","list_caption":"Figure: Heapster架构图","alt":"Heapster架构图","nro":78,"url":"../images/heapster-architecture.png","index":1,"caption_template":"图片 - _CAPTION_","label":"Heapster架构图","attributes":{},"skip":false,"key":"5.6.1.1.1"},{"backlink":"practice/prometheus.html#fig5.6.2.1","level":"5.6.2","list_caption":"Figure: Prometheus 架构图","alt":"Prometheus 架构图","nro":79,"url":"../images/prometheus-architecture.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"Prometheus 架构图","attributes":{},"skip":false,"key":"5.6.2.1"},{"backlink":"practice/using-prometheus-to-monitor-kuberentes-cluster.html#fig5.6.2.1.1","level":"5.6.2.1","list_caption":"Figure: Grafana页面","alt":"Grafana页面","nro":80,"url":"../images/kubernetes-prometheus-monitoring.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"Grafana页面","attributes":{},"skip":false,"key":"5.6.2.1.1"},{"backlink":"practice/helm.html#fig5.7.1.1","level":"5.7.1","list_caption":"Figure: Helm chart源","alt":"Helm chart源","nro":81,"url":"../images/helm-charts-repository.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"Helm chart源","attributes":{},"skip":false,"key":"5.7.1.1"},{"backlink":"practice/helm.html#fig5.7.1.2","level":"5.7.1","list_caption":"Figure: TODO应用的Web页面","alt":"TODO应用的Web页面","nro":82,"url":"../images/helm-mean-todo-aholic.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"TODO应用的Web页面","attributes":{},"skip":false,"key":"5.7.1.2"},{"backlink":"practice/create-private-charts-repo.html#fig5.7.2.1","level":"5.7.2","list_caption":"Figure: Helm monocular界面","alt":"Helm monocular界面","nro":83,"url":"../images/helm-monocular-jimmysong.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"Helm monocular界面","attributes":{},"skip":false,"key":"5.7.2.1"},{"backlink":"practice/jenkins-ci-cd.html#fig5.8.1.1","level":"5.8.1","list_caption":"Figure: 基于Jenkins的持续集成与发布","alt":"基于Jenkins的持续集成与发布","nro":84,"url":"../images/kubernetes-jenkins-ci-cd.png","index":1,"caption_template":"图片 - _CAPTION_","label":"基于Jenkins的持续集成与发布","attributes":{},"skip":false,"key":"5.8.1.1"},{"backlink":"practice/drone-ci-cd.html#fig5.8.2.1","level":"5.8.2","list_caption":"Figure: OAuth注册","alt":"OAuth注册","nro":85,"url":"../images/github-oauth-register.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"OAuth注册","attributes":{},"skip":false,"key":"5.8.2.1"},{"backlink":"practice/drone-ci-cd.html#fig5.8.2.2","level":"5.8.2","list_caption":"Figure: OAuth key","alt":"OAuth key","nro":86,"url":"../images/github-oauth-drone-key.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"OAuth key","attributes":{},"skip":false,"key":"5.8.2.2"},{"backlink":"practice/drone-ci-cd.html#fig5.8.2.3","level":"5.8.2","list_caption":"Figure: Drone登陆界面","alt":"Drone登陆界面","nro":87,"url":"../images/drone-login-github.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"Drone登陆界面","attributes":{},"skip":false,"key":"5.8.2.3"},{"backlink":"practice/drone-ci-cd.html#fig5.8.2.4","level":"5.8.2","list_caption":"Figure: Github启用repo设置","alt":"Github启用repo设置","nro":88,"url":"../images/drone-github-active.jpg","index":4,"caption_template":"图片 - _CAPTION_","label":"Github启用repo设置","attributes":{},"skip":false,"key":"5.8.2.4"},{"backlink":"practice/drone-ci-cd.html#fig5.8.2.5","level":"5.8.2","list_caption":"Figure: Github单个repo设置","alt":"Github单个repo设置","nro":89,"url":"../images/drone-github-repo-setting.jpg","index":5,"caption_template":"图片 - _CAPTION_","label":"Github单个repo设置","attributes":{},"skip":false,"key":"5.8.2.5"},{"backlink":"practice/dashboard-upgrade.html#fig5.9.2.1","level":"5.9.2","list_caption":"Figure: 登陆界面","alt":"登陆界面","nro":90,"url":"../images/kubernetes-dashboard-1.7.1-login.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"登陆界面","attributes":{},"skip":false,"key":"5.9.2.1"},{"backlink":"practice/dashboard-upgrade.html#fig5.9.2.2","level":"5.9.2","list_caption":"Figure: 首页","alt":"首页","nro":91,"url":"../images/kubernetes-dashboard-1.7.1-default-page.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"首页","attributes":{},"skip":false,"key":"5.9.2.2"},{"backlink":"practice/dashboard-upgrade.html#fig5.9.2.3","level":"5.9.2","list_caption":"Figure: 用户空间","alt":"用户空间","nro":92,"url":"../images/kubernetes-dashboard-1.7.1-brand.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"用户空间","attributes":{},"skip":false,"key":"5.9.2.3"},{"backlink":"practice/dashboard-upgrade.html#fig5.9.2.4","level":"5.9.2","list_caption":"Figure: kubeconfig文件","alt":"kubeconfig文件","nro":93,"url":"../images/brand-kubeconfig-yaml.jpg","index":4,"caption_template":"图片 - _CAPTION_","label":"kubeconfig文件","attributes":{},"skip":false,"key":"5.9.2.4"},{"backlink":"usecases/microservices.html#fig6.2.1","level":"6.2","list_caption":"Figure: 微服务关注的部分","alt":"微服务关注的部分","nro":94,"url":"../images/microservices-concerns.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"微服务关注的部分","attributes":{},"skip":false,"key":"6.2.1"},{"backlink":"usecases/service-discovery-in-microservices.html#fig6.2.1.1","level":"6.2.1","list_caption":"Figure: 微服务中的服务发现","alt":"微服务中的服务发现","nro":95,"url":"../images/service-discovery-in-microservices.png","index":1,"caption_template":"图片 - _CAPTION_","label":"微服务中的服务发现","attributes":{},"skip":false,"key":"6.2.1.1"},{"backlink":"usecases/service-mesh.html#fig6.3.1","level":"6.3","list_caption":"Figure: 下一代异构微服务架构","alt":"下一代异构微服务架构","nro":96,"url":"../images/polyglot-microservices-serivce-mesh.png","index":1,"caption_template":"图片 - _CAPTION_","label":"下一代异构微服务架构","attributes":{},"skip":false,"key":"6.3.1"},{"backlink":"usecases/service-mesh.html#fig6.3.2","level":"6.3","list_caption":"Figure: Service Mesh 架构图","alt":"Service Mesh 架构图","nro":97,"url":"../images/serivce-mesh-control-plane.png","index":2,"caption_template":"图片 - _CAPTION_","label":"Service Mesh 架构图","attributes":{},"skip":false,"key":"6.3.2"},{"backlink":"usecases/service-mesh.html#fig6.3.3","level":"6.3","list_caption":"Figure: Istio vs linkerd","alt":"Istio vs linkerd","nro":98,"url":"../images/istio-vs-linkerd.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"Istio vs linkerd","attributes":{},"skip":false,"key":"6.3.3"},{"backlink":"usecases/istio.html#fig6.3.1.1","level":"6.3.1","list_caption":"Figure: Istio架构图","alt":"Istio架构图","nro":99,"url":"../images/istio-arch.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"Istio架构图","attributes":{},"skip":false,"key":"6.3.1.1"},{"backlink":"usecases/istio-installation.html#fig6.3.1.1.1","level":"6.3.1.1","list_caption":"Figure: BookInfo Sample应用架构图","alt":"BookInfo Sample应用架构图","nro":100,"url":"../images/bookinfo-sample-arch.png","index":1,"caption_template":"图片 - _CAPTION_","label":"BookInfo Sample应用架构图","attributes":{},"skip":false,"key":"6.3.1.1.1"},{"backlink":"usecases/istio-installation.html#fig6.3.1.1.2","level":"6.3.1.1","list_caption":"Figure: BookInfo Sample页面","alt":"BookInfo Sample页面","nro":101,"url":"../images/bookinfo-sample.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"BookInfo Sample页面","attributes":{},"skip":false,"key":"6.3.1.1.2"},{"backlink":"usecases/istio-installation.html#fig6.3.1.1.3","level":"6.3.1.1","list_caption":"Figure: Istio Grafana界面","alt":"Istio Grafana界面","nro":102,"url":"../images/istio-grafana.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"Istio Grafana界面","attributes":{},"skip":false,"key":"6.3.1.1.3"},{"backlink":"usecases/istio-installation.html#fig6.3.1.1.4","level":"6.3.1.1","list_caption":"Figure: Prometheus页面","alt":"Prometheus页面","nro":103,"url":"../images/istio-prometheus.jpg","index":4,"caption_template":"图片 - _CAPTION_","label":"Prometheus页面","attributes":{},"skip":false,"key":"6.3.1.1.4"},{"backlink":"usecases/istio-installation.html#fig6.3.1.1.5","level":"6.3.1.1","list_caption":"Figure: Zipkin页面","alt":"Zipkin页面","nro":104,"url":"../images/istio-zipkin.jpg","index":5,"caption_template":"图片 - _CAPTION_","label":"Zipkin页面","attributes":{},"skip":false,"key":"6.3.1.1.5"},{"backlink":"usecases/istio-installation.html#fig6.3.1.1.6","level":"6.3.1.1","list_caption":"Figure: ServiceGraph页面","alt":"ServiceGraph页面","nro":105,"url":"../images/istio-servicegraph.jpg","index":6,"caption_template":"图片 - _CAPTION_","label":"ServiceGraph页面","attributes":{},"skip":false,"key":"6.3.1.1.6"},{"backlink":"usecases/install-and-expand-istio-mesh.html#fig6.3.1.3.1","level":"6.3.1.3","list_caption":"Figure: BookInfo","alt":"BookInfo","nro":106,"url":"../images/noistio.png","index":1,"caption_template":"图片 - _CAPTION_","label":"BookInfo","attributes":{},"skip":false,"key":"6.3.1.3.1"},{"backlink":"usecases/install-and-expand-istio-mesh.html#fig6.3.1.3.2","level":"6.3.1.3","list_caption":"Figure: BookInfo","alt":"BookInfo","nro":107,"url":"../images/noistio.png","index":2,"caption_template":"图片 - _CAPTION_","label":"BookInfo","attributes":{},"skip":false,"key":"6.3.1.3.2"},{"backlink":"usecases/integrating-vms.html#fig6.3.1.4.1","level":"6.3.1.4","list_caption":"Figure: Bookinfo应用的拓展Mesh","alt":"Bookinfo应用的拓展Mesh","nro":108,"url":"../images/istio-mesh-expansion.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"Bookinfo应用的拓展Mesh","attributes":{},"skip":false,"key":"6.3.1.4.1"},{"backlink":"usecases/linkerd.html#fig6.3.2.1","level":"6.3.2","list_caption":"Figure: source https://linkerd.io","alt":"source https://linkerd.io","nro":109,"url":"../images/diagram-individual-instance.png","index":1,"caption_template":"图片 - _CAPTION_","label":"source https://linkerd.io","attributes":{},"skip":false,"key":"6.3.2.1"},{"backlink":"usecases/linkerd-user-guide.html#fig6.3.2.1.1","level":"6.3.2.1","list_caption":"Figure: Jenkins pipeline","alt":"Jenkins pipeline","nro":110,"url":"../images/linkerd-jenkins-pipeline.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"Jenkins pipeline","attributes":{},"skip":false,"key":"6.3.2.1.1"},{"backlink":"usecases/linkerd-user-guide.html#fig6.3.2.1.2","level":"6.3.2.1","list_caption":"Figure: Jenkins config","alt":"Jenkins config","nro":111,"url":"../images/linkerd-jenkins.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"Jenkins config","attributes":{},"skip":false,"key":"6.3.2.1.2"},{"backlink":"usecases/linkerd-user-guide.html#fig6.3.2.1.3","level":"6.3.2.1","list_caption":"Figure: namerd","alt":"namerd","nro":112,"url":"../images/namerd-internal.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"namerd","attributes":{},"skip":false,"key":"6.3.2.1.3"},{"backlink":"usecases/linkerd-user-guide.html#fig6.3.2.1.4","level":"6.3.2.1","list_caption":"Figure: linkerd监控","alt":"linkerd监控","nro":113,"url":"../images/linkerd-helloworld-outgoing.jpg","index":4,"caption_template":"图片 - _CAPTION_","label":"linkerd监控","attributes":{},"skip":false,"key":"6.3.2.1.4"},{"backlink":"usecases/linkerd-user-guide.html#fig6.3.2.1.5","level":"6.3.2.1","list_caption":"Figure: linkerd监控","alt":"linkerd监控","nro":114,"url":"../images/linkerd-helloworld-incoming.jpg","index":5,"caption_template":"图片 - _CAPTION_","label":"linkerd监控","attributes":{},"skip":false,"key":"6.3.2.1.5"},{"backlink":"usecases/linkerd-user-guide.html#fig6.3.2.1.6","level":"6.3.2.1","list_caption":"Figure: linkerd性能监控","alt":"linkerd性能监控","nro":115,"url":"../images/linkerd-grafana.png","index":6,"caption_template":"图片 - _CAPTION_","label":"linkerd性能监控","attributes":{},"skip":false,"key":"6.3.2.1.6"},{"backlink":"usecases/linkerd-user-guide.html#fig6.3.2.1.7","level":"6.3.2.1","list_caption":"Figure: Linkerd ingress controller","alt":"Linkerd ingress controller","nro":116,"url":"../images/linkerd-ingress-controller.jpg","index":7,"caption_template":"图片 - _CAPTION_","label":"Linkerd ingress controller","attributes":{},"skip":false,"key":"6.3.2.1.7"},{"backlink":"usecases/conduit-installation.html#fig6.3.3.2.1","level":"6.3.3.2","list_caption":"Figure: Conduit dashboard","alt":"Conduit dashboard","nro":117,"url":"../images/conduit-dashboard.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"Conduit dashboard","attributes":{},"skip":false,"key":"6.3.3.2.1"},{"backlink":"usecases/big-data.html#fig6.4.1","level":"6.4","list_caption":"Figure: Spark on yarn with kubernetes","alt":"Spark on yarn with kubernetes","nro":118,"url":"../images/spark-on-yarn-with-kubernetes.png","index":1,"caption_template":"图片 - _CAPTION_","label":"Spark on yarn with kubernetes","attributes":{},"skip":false,"key":"6.4.1"},{"backlink":"usecases/big-data.html#fig6.4.2","level":"6.4","list_caption":"Figure: 在kubernetes上使用多种调度方式","alt":"在kubernetes上使用多种调度方式","nro":119,"url":"../images/spark-on-kubernetes-with-different-schedulers.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"在kubernetes上使用多种调度方式","attributes":{},"skip":false,"key":"6.4.2"},{"backlink":"usecases/spark-standalone-on-kubernetes.html#fig6.4.1.1","level":"6.4.1","list_caption":"Figure: spark master ui","alt":"spark master ui","nro":120,"url":"../images/spark-ui.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"spark master ui","attributes":{},"skip":false,"key":"6.4.1.1"},{"backlink":"usecases/spark-standalone-on-kubernetes.html#fig6.4.1.2","level":"6.4.1","list_caption":"Figure: zeppelin ui","alt":"zeppelin ui","nro":121,"url":"../images/zeppelin-ui.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"zeppelin ui","attributes":{},"skip":false,"key":"6.4.1.2"},{"backlink":"usecases/serverless.html#fig6.5.1","level":"6.5","list_caption":"Figure: 从物理机到函数计算","alt":"从物理机到函数计算","nro":122,"url":"../images/from-bare-metal-to-functions.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"从物理机到函数计算","attributes":{},"skip":false,"key":"6.5.1"},{"backlink":"usecases/serverless.html#fig6.5.2","level":"6.5","list_caption":"Figure: FaaS Landscape","alt":"FaaS Landscape","nro":123,"url":"../images/redpoint-faas-landscape.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"FaaS Landscape","attributes":{},"skip":false,"key":"6.5.2"},{"backlink":"usecases/understanding-serverless.html#fig6.5.1.1","level":"6.5.1","list_caption":"Figure: 服务端软件的运行环境","alt":"服务端软件的运行环境","nro":124,"url":"../images/serverless-server-side-software.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"服务端软件的运行环境","attributes":{},"skip":false,"key":"6.5.1.1"},{"backlink":"usecases/understanding-serverless.html#fig6.5.1.2","level":"6.5.1","list_caption":"Figure: FaaS应用架构","alt":"FaaS应用架构","nro":125,"url":"../images/serverless-faas-platform.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"FaaS应用架构","attributes":{},"skip":false,"key":"6.5.1.2"},{"backlink":"usecases/understanding-serverless.html#fig6.5.1.3","level":"6.5.1","list_caption":"Figure: 传统应用程序架构","alt":"传统应用程序架构","nro":126,"url":"../images/non-serverless-game-arch.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"传统应用程序架构","attributes":{},"skip":false,"key":"6.5.1.3"},{"backlink":"usecases/understanding-serverless.html#fig6.5.1.4","level":"6.5.1","list_caption":"Figure: Serverless架构","alt":"Serverless架构","nro":127,"url":"../images/serverless-game-arch.jpg","index":4,"caption_template":"图片 - _CAPTION_","label":"Serverless架构","attributes":{},"skip":false,"key":"6.5.1.4"},{"backlink":"usecases/openfaas.html#fig6.5.2.1.1","level":"6.5.2.1","list_caption":"Figure: OpenFaaS架构","alt":"OpenFaaS架构","nro":128,"url":"../images/openfaas-arch.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"OpenFaaS架构","attributes":{},"skip":false,"key":"6.5.2.1.1"},{"backlink":"usecases/openfaas.html#fig6.5.2.1.2","level":"6.5.2.1","list_caption":"Figure: OpenFaaS Prometheus","alt":"OpenFaaS Prometheus","nro":129,"url":"../images/openfaas-prometheus.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"OpenFaaS Prometheus","attributes":{},"skip":false,"key":"6.5.2.1.2"},{"backlink":"usecases/openfaas.html#fig6.5.2.1.3","level":"6.5.2.1","list_caption":"Figure: OpenFaas Grafana监控","alt":"OpenFaas Grafana监控","nro":130,"url":"../images/openfaas-grafana.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"OpenFaas Grafana监控","attributes":{},"skip":false,"key":"6.5.2.1.3"},{"backlink":"usecases/openfaas.html#fig6.5.2.1.4","level":"6.5.2.1","list_caption":"Figure: OpenFaas Dashboard","alt":"OpenFaas Dashboard","nro":131,"url":"../images/openfaas-deploy-a-function.jpg","index":4,"caption_template":"图片 - _CAPTION_","label":"OpenFaas Dashboard","attributes":{},"skip":false,"key":"6.5.2.1.4"},{"backlink":"usecases/openfaas.html#fig6.5.2.1.5","level":"6.5.2.1","list_caption":"Figure: NodeInfo执行结果","alt":"NodeInfo执行结果","nro":132,"url":"../images/openfaas-nodeinfo.jpg","index":5,"caption_template":"图片 - _CAPTION_","label":"NodeInfo执行结果","attributes":{},"skip":false,"key":"6.5.2.1.5"},{"backlink":"develop/sigs-and-working-group.html#fig7.2.1","level":"7.2","list_caption":"Figure: Kubernetes SIG","alt":"Kubernetes SIG","nro":133,"url":"../images/kubernetes-sigs.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"Kubernetes SIG","attributes":{},"skip":false,"key":"7.2.1"},{"backlink":"develop/client-go-sample.html#fig7.5.1","level":"7.5","list_caption":"Figure: 使用kubernetes dashboard进行故障排查","alt":"使用kubernetes dashboard进行故障排查","nro":134,"url":"../images/kubernetes-client-go-sample-update.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"使用kubernetes dashboard进行故障排查","attributes":{},"skip":false,"key":"7.5.1"},{"backlink":"appendix/issues.html#fig8.6.1","level":"8.6","list_caption":"Figure: pvc-storage-limit","alt":"pvc-storage-limit","nro":135,"url":"../images/pvc-storage-limit.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"pvc-storage-limit","attributes":{},"skip":false,"key":"8.6.1"}]},"title":"Kubernetes Handbook - jimmysong.io","language":"zh-hans","links":{"sidebar":{"Jimmy Song":"https://jimmysong.io"}},"gitbook":"*","description":"Kubernetes中文指南/实践手册|kubernetes handbook|jimmysong.io|宋净超"},"file":{"path":"README.md","mtime":"2017-12-15T11:49:07.697Z","type":"markdown"},"gitbook":{"version":"3.2.2","time":"2017-12-20T10:50:52.717Z"},"basePath":".","book":{"language":""}}); }); diff --git a/practice/app-log-collection.html b/practice/app-log-collection.html index e1956d7fb..a0af4908a 100644 --- a/practice/app-log-collection.html +++ b/practice/app-log-collection.html @@ -2937,7 +2937,7 @@ diff --git a/practice/cephfs.html b/practice/cephfs.html index b3bcc9526..6bcf48843 100644 --- a/practice/cephfs.html +++ b/practice/cephfs.html @@ -2779,7 +2779,7 @@ diff --git a/practice/ci-cd.html b/practice/ci-cd.html index 63a561e4c..0f9896ee8 100644 --- a/practice/ci-cd.html +++ b/practice/ci-cd.html @@ -2779,7 +2779,7 @@ diff --git a/practice/configuration-best-practice.html b/practice/configuration-best-practice.html index 9b4e5cb23..e10a6f81d 100644 --- a/practice/configuration-best-practice.html +++ b/practice/configuration-best-practice.html @@ -2826,7 +2826,7 @@ diff --git a/practice/create-kubeconfig.html b/practice/create-kubeconfig.html index 6ac6c9130..9f6d9622a 100644 --- a/practice/create-kubeconfig.html +++ b/practice/create-kubeconfig.html @@ -2866,7 +2866,7 @@ kubectl config use-context default --kubeconfig=kube-proxy.kubeconfig diff --git a/practice/create-private-charts-repo.html b/practice/create-private-charts-repo.html index 63f9841b1..d545e2d8f 100644 --- a/practice/create-private-charts-repo.html +++ b/practice/create-private-charts-repo.html @@ -2877,7 +2877,7 @@ $ helm install monocular/monocular diff --git a/practice/create-tls-and-secret-key.html b/practice/create-tls-and-secret-key.html index 2e45ed6aa..16ff48ce5 100644 --- a/practice/create-tls-and-secret-key.html +++ b/practice/create-tls-and-secret-key.html @@ -3101,7 +3101,7 @@ cp *.pem /etc/kubernetes/ssl diff --git a/practice/dashboard-addon-installation.html b/practice/dashboard-addon-installation.html index 47d776441..e327cdd37 100644 --- a/practice/dashboard-addon-installation.html +++ b/practice/dashboard-addon-installation.html @@ -2890,7 +2890,7 @@ kubernetes-dashboard-3966630548-0jj1j 0/1 Terminating 0 1d diff --git a/practice/dashboard-upgrade.html b/practice/dashboard-upgrade.html index af0264e2f..a896b5afd 100644 --- a/practice/dashboard-upgrade.html +++ b/practice/dashboard-upgrade.html @@ -2904,7 +2904,7 @@ ca.crt: 1310 bytes diff --git a/practice/data-persistence-problem.html b/practice/data-persistence-problem.html index 2a7bd6711..29a7bea6e 100644 --- a/practice/data-persistence-problem.html +++ b/practice/data-persistence-problem.html @@ -2796,7 +2796,7 @@ diff --git a/practice/distributed-load-test.html b/practice/distributed-load-test.html index fa49bf511..7012b3e7c 100644 --- a/practice/distributed-load-test.html +++ b/practice/distributed-load-test.html @@ -2836,7 +2836,7 @@ $ kubectl create -f locust-master-service.yaml diff --git a/practice/drone-ci-cd.html b/practice/drone-ci-cd.html index 55d5f9ded..78e05e3a4 100644 --- a/practice/drone-ci-cd.html +++ b/practice/drone-ci-cd.html @@ -2842,7 +2842,7 @@ diff --git a/practice/edge-node-configuration.html b/practice/edge-node-configuration.html index b41627c63..f2d0972f6 100644 --- a/practice/edge-node-configuration.html +++ b/practice/edge-node-configuration.html @@ -2951,7 +2951,7 @@ traefik-ingress-lb 3 3 3 3 3 edge diff --git a/practice/efk-addon-installation.html b/practice/efk-addon-installation.html index 31e2c1aed..d9561ebad 100644 --- a/practice/efk-addon-installation.html +++ b/practice/efk-addon-installation.html @@ -2892,7 +2892,7 @@ server.basePath: /api/v1/proxy/namespaces/kube-system/services/kibana-logging diff --git a/practice/etcd-cluster-installation.html b/practice/etcd-cluster-installation.html index 9e79edf97..8f95ced17 100644 --- a/practice/etcd-cluster-installation.html +++ b/practice/etcd-cluster-installation.html @@ -2882,7 +2882,7 @@ cluster is healthy diff --git a/practice/flannel-installation.html b/practice/flannel-installation.html index b633683c5..0b2d27717 100644 --- a/practice/flannel-installation.html +++ b/practice/flannel-installation.html @@ -2863,7 +2863,7 @@ systemctl status flanneld diff --git a/practice/glusterfs.html b/practice/glusterfs.html index 79bb3fc14..6aa7fdba7 100644 --- a/practice/glusterfs.html +++ b/practice/glusterfs.html @@ -2779,7 +2779,7 @@ diff --git a/practice/heapster-addon-installation.html b/practice/heapster-addon-installation.html index e1aef5377..e331d15d4 100644 --- a/practice/heapster-addon-installation.html +++ b/practice/heapster-addon-installation.html @@ -2934,7 +2934,7 @@ monitoring-influxdb 10.254.22.46 <nodes> 8086:32299/TCP,8083:3 diff --git a/practice/heapster.html b/practice/heapster.html index bb71db00f..be7dafaab 100644 --- a/practice/heapster.html +++ b/practice/heapster.html @@ -2780,7 +2780,7 @@ diff --git a/practice/helm.html b/practice/helm.html index e579cac04..1820942d3 100644 --- a/practice/helm.html +++ b/practice/helm.html @@ -3296,7 +3296,7 @@ Get the URL of your Node app by running: diff --git a/practice/index.html b/practice/index.html index 3a3412d65..d0820f613 100644 --- a/practice/index.html +++ b/practice/index.html @@ -2789,7 +2789,7 @@ diff --git a/practice/install-kubernetes-on-centos.html b/practice/install-kubernetes-on-centos.html index f4c0a5ceb..4bb8cc208 100644 --- a/practice/install-kubernetes-on-centos.html +++ b/practice/install-kubernetes-on-centos.html @@ -2852,7 +2852,7 @@ setenforce 0 设置SELinux 成为permissive模式< diff --git a/practice/jenkins-ci-cd.html b/practice/jenkins-ci-cd.html index a458f0a67..5753f2cb7 100644 --- a/practice/jenkins-ci-cd.html +++ b/practice/jenkins-ci-cd.html @@ -2793,7 +2793,7 @@ diff --git a/practice/kubectl-installation.html b/practice/kubectl-installation.html index 47e63e7b8..f6b56da06 100644 --- a/practice/kubectl-installation.html +++ b/practice/kubectl-installation.html @@ -2814,7 +2814,7 @@ kubectl config use-context kubernetes diff --git a/practice/kubedns-addon-installation.html b/practice/kubedns-addon-installation.html index 3461ea52e..372f4de46 100644 --- a/practice/kubedns-addon-installation.html +++ b/practice/kubedns-addon-installation.html @@ -2928,7 +2928,7 @@ PING kube-dns.kube-system.svc.cluster.local (10.254.0.2): 56 data bytes diff --git a/practice/manage-compute-resources-container.html b/practice/manage-compute-resources-container.html index e1eae9c94..b73a2d844 100644 --- a/practice/manage-compute-resources-container.html +++ b/practice/manage-compute-resources-container.html @@ -2978,7 +2978,7 @@ http://k8s-master:8080/api/v1/nodes/k8s-node-1/status diff --git a/practice/manually-upgrade.html b/practice/manually-upgrade.html index be22c5964..e1723c0cd 100644 --- a/practice/manually-upgrade.html +++ b/practice/manually-upgrade.html @@ -2857,7 +2857,7 @@ systemctl start kube-proxy diff --git a/practice/master-installation.html b/practice/master-installation.html index 375a83912..254f0c797 100644 --- a/practice/master-installation.html +++ b/practice/master-installation.html @@ -3025,7 +3025,7 @@ etcd-2 Healthy {"health" var gitbook = gitbook || []; gitbook.push(function() { - gitbook.page.hasChanged({"page":{"title":"部署master节点","level":"5.2.5","depth":2,"next":{"title":"安装flannel网络插件","level":"5.2.6","depth":2,"path":"practice/flannel-installation.md","ref":"practice/flannel-installation.md","articles":[]},"previous":{"title":"安装kubectl命令行工具","level":"5.2.4","depth":2,"path":"practice/kubectl-installation.md","ref":"practice/kubectl-installation.md","articles":[]},"dir":"ltr"},"config":{"plugins":["github","codesnippet","splitter","page-toc-button","image-captions","editlink","back-to-top-button","-lunr","-search","search-plus","github-buttons@2.1.0","favicon@^0.0.2","tbfed-pagefooter@^0.0.1","3-ba","theme-default"],"styles":{"website":"styles/website.css","pdf":"styles/pdf.css","epub":"styles/epub.css","mobi":"styles/mobi.css","ebook":"styles/ebook.css","print":"styles/print.css"},"pluginsConfig":{"tbfed-pagefooter":{"copyright":"Copyright © jimmysong.io 2017","modify_label":"Updated:","modify_format":"YYYY-MM-DD HH:mm:ss"},"github":{"url":"https://github.com/rootsongjc/kubernetes-handbook"},"editlink":{"label":"编辑本页","multilingual":false,"base":"https://github.com/rootsongjc/kubernetes-handbook/blob/master/"},"splitter":{},"codesnippet":{},"fontsettings":{"theme":"white","family":"sans","size":2},"highlight":{},"favicon":{"shortcut":"favicon.ico","bookmark":"favicon.ico"},"page-toc-button":{},"back-to-top-button":{},"github-buttons":{"repo":"rootsongjc/kubernetes-handbook","types":["star"],"size":"small"},"3-ba":{"configuration":"auto","token":"11f7d254cfa4e0ca44b175c66d379ecc"},"sharing":{"facebook":true,"twitter":true,"google":false,"weibo":false,"instapaper":false,"vk":false,"all":["facebook","google","twitter","weibo","instapaper"]},"theme-default":{"styles":{"website":"styles/website.css","pdf":"styles/pdf.css","epub":"styles/epub.css","mobi":"styles/mobi.css","ebook":"styles/ebook.css","print":"styles/print.css"},"showLevel":true},"search-plus":{},"image-captions":{"caption":"图片 - _CAPTION_","variable_name":"_pictures"}},"theme":"default","author":"Jimmy Song","pdf":{"pageNumbers":true,"fontSize":12,"fontFamily":"Arial","paperSize":"a4","chapterMark":"pagebreak","pageBreaksBefore":"/","margin":{"right":62,"left":62,"top":56,"bottom":56}},"structure":{"langs":"LANGS.md","readme":"README.md","glossary":"GLOSSARY.md","summary":"SUMMARY.md"},"variables":{"_pictures":[{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.1","level":"2.1","list_caption":"Figure: 云计算演进历程","alt":"云计算演进历程","nro":1,"url":"../images/cloud-computing-evolution-road.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"云计算演进历程","attributes":{},"skip":false,"key":"2.1.1"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.2","level":"2.1","list_caption":"Figure: 来自Twitter @MarcWilczek","alt":"来自Twitter @MarcWilczek","nro":2,"url":"../images/cloud-native-comes-of-age.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"来自Twitter @MarcWilczek","attributes":{},"skip":false,"key":"2.1.2"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.3","level":"2.1","list_caption":"Figure: Cloud native思维导图","alt":"Cloud native思维导图","nro":3,"url":"../images/cloud-native-architecutre-mindnode.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"Cloud native思维导图","attributes":{},"skip":false,"key":"2.1.3"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.4","level":"2.1","list_caption":"Figure: 十二因素应用","alt":"十二因素应用","nro":4,"url":"../images/12-factor-app.png","index":4,"caption_template":"图片 - _CAPTION_","label":"十二因素应用","attributes":{},"skip":false,"key":"2.1.4"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.5","level":"2.1","list_caption":"Figure: 容器生态","alt":"容器生态","nro":5,"url":"../images/container-ecosystem.png","index":5,"caption_template":"图片 - _CAPTION_","label":"容器生态","attributes":{},"skip":false,"key":"2.1.5"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.6","level":"2.1","list_caption":"Figure: 使用Jenkins进行持续集成与发布流程图","alt":"使用Jenkins进行持续集成与发布流程图","nro":6,"url":"../images/kubernetes-jenkins-ci-cd.png","index":6,"caption_template":"图片 - _CAPTION_","label":"使用Jenkins进行持续集成与发布流程图","attributes":{},"skip":false,"key":"2.1.6"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.7","level":"2.1","list_caption":"Figure: filebeat日志收集架构图","alt":"filebeat日志收集架构图","nro":7,"url":"../images/filebeat-log-collector-arch.png","index":7,"caption_template":"图片 - _CAPTION_","label":"filebeat日志收集架构图","attributes":{},"skip":false,"key":"2.1.7"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.8","level":"2.1","list_caption":"Figure: API文档","alt":"API文档","nro":8,"url":"../images/k8s-app-monitor-test-api-doc.jpg","index":8,"caption_template":"图片 - _CAPTION_","label":"API文档","attributes":{},"skip":false,"key":"2.1.8"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.9","level":"2.1","list_caption":"Figure: 迁移步骤示意图","alt":"迁移步骤示意图","nro":9,"url":"../images/migrating-hadoop-yarn-to-kubernetes.png","index":9,"caption_template":"图片 - _CAPTION_","label":"迁移步骤示意图","attributes":{},"skip":false,"key":"2.1.9"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.10","level":"2.1","list_caption":"Figure: service mesh架构图","alt":"service mesh架构图","nro":10,"url":"../images/serivce-mesh-control-plane.png","index":10,"caption_template":"图片 - _CAPTION_","label":"service mesh架构图","attributes":{},"skip":false,"key":"2.1.10"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.11","level":"2.1","list_caption":"Figure: kibana界面","alt":"kibana界面","nro":11,"url":"../images/filebeat-docker-test.jpg","index":11,"caption_template":"图片 - _CAPTION_","label":"kibana界面","attributes":{},"skip":false,"key":"2.1.11"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.12","level":"2.1","list_caption":"Figure: Grafana界面示意图1","alt":"Grafana界面示意图1","nro":12,"url":"../images/kubernetes-devops-example-grafana-1.png","index":12,"caption_template":"图片 - _CAPTION_","label":"Grafana界面示意图1","attributes":{},"skip":false,"key":"2.1.12"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.13","level":"2.1","list_caption":"Figure: Grafana界面示意图2","alt":"Grafana界面示意图2","nro":13,"url":"../images/kubernetes-devops-example-grafana-2.png","index":13,"caption_template":"图片 - _CAPTION_","label":"Grafana界面示意图2","attributes":{},"skip":false,"key":"2.1.13"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.14","level":"2.1","list_caption":"Figure: Grafana界面示意图3","alt":"Grafana界面示意图3","nro":14,"url":"../images/kubernetes-devops-example-grafana-3.png","index":14,"caption_template":"图片 - _CAPTION_","label":"Grafana界面示意图3","attributes":{},"skip":false,"key":"2.1.14"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.15","level":"2.1","list_caption":"Figure: dashboard","alt":"dashboard","nro":15,"url":"../images/spark-job-on-kubernetes-example-1.jpg","index":15,"caption_template":"图片 - _CAPTION_","label":"dashboard","attributes":{},"skip":false,"key":"2.1.15"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.16","level":"2.1","list_caption":"Figure: Grafana","alt":"Grafana","nro":16,"url":"../images/spark-job-on-kubernetes-example-2.jpg","index":16,"caption_template":"图片 - _CAPTION_","label":"Grafana","attributes":{},"skip":false,"key":"2.1.16"},{"backlink":"concepts/index.html#fig3.1.1","level":"3.1","list_caption":"Figure: Borg架构","alt":"Borg架构","nro":17,"url":"../images/borg.png","index":1,"caption_template":"图片 - _CAPTION_","label":"Borg架构","attributes":{},"skip":false,"key":"3.1.1"},{"backlink":"concepts/index.html#fig3.1.2","level":"3.1","list_caption":"Figure: Kubernetes架构","alt":"Kubernetes架构","nro":18,"url":"../images/architecture.png","index":2,"caption_template":"图片 - _CAPTION_","label":"Kubernetes架构","attributes":{},"skip":false,"key":"3.1.2"},{"backlink":"concepts/index.html#fig3.1.3","level":"3.1","list_caption":"Figure: Kuberentes架构（图片来自于网络）","alt":"Kuberentes架构（图片来自于网络）","nro":19,"url":"../images/kubernetes-high-level-component-archtecture.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"Kuberentes架构（图片来自于网络）","attributes":{},"skip":false,"key":"3.1.3"},{"backlink":"concepts/index.html#fig3.1.4","level":"3.1","list_caption":"Figure: kubernetes整体架构示意图","alt":"kubernetes整体架构示意图","nro":20,"url":"../images/kubernetes-whole-arch.png","index":4,"caption_template":"图片 - _CAPTION_","label":"kubernetes整体架构示意图","attributes":{},"skip":false,"key":"3.1.4"},{"backlink":"concepts/index.html#fig3.1.5","level":"3.1","list_caption":"Figure: Kubernetes master架构示意图","alt":"Kubernetes master架构示意图","nro":21,"url":"../images/kubernetes-master-arch.png","index":5,"caption_template":"图片 - _CAPTION_","label":"Kubernetes master架构示意图","attributes":{},"skip":false,"key":"3.1.5"},{"backlink":"concepts/index.html#fig3.1.6","level":"3.1","list_caption":"Figure: kubernetes node架构示意图","alt":"kubernetes node架构示意图","nro":22,"url":"../images/kubernetes-node-arch.png","index":6,"caption_template":"图片 - _CAPTION_","label":"kubernetes node架构示意图","attributes":{},"skip":false,"key":"3.1.6"},{"backlink":"concepts/index.html#fig3.1.7","level":"3.1","list_caption":"Figure: Kubernetes分层架构示意图","alt":"Kubernetes分层架构示意图","nro":23,"url":"../images/kubernetes-layers-arch.jpg","index":7,"caption_template":"图片 - _CAPTION_","label":"Kubernetes分层架构示意图","attributes":{},"skip":false,"key":"3.1.7"},{"backlink":"concepts/concepts.html#fig3.2.1","level":"3.2","list_caption":"Figure: 分层架构示意图","alt":"分层架构示意图","nro":24,"url":"../images/kubernetes-layers-arch.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"分层架构示意图","attributes":{},"skip":false,"key":"3.2.1"},{"backlink":"concepts/pod-overview.html#fig3.3.1.1","level":"3.3.1","list_caption":"Figure: pod diagram","alt":"pod diagram","nro":25,"url":"../images/pod-overview.png","index":1,"caption_template":"图片 - _CAPTION_","label":"pod diagram","attributes":{},"skip":false,"key":"3.3.1.1"},{"backlink":"concepts/pod.html#fig3.3.1.1.1","level":"3.3.1.1","list_caption":"Figure: Pod示意图","alt":"Pod示意图","nro":26,"url":"../images/pod-overview.png","index":1,"caption_template":"图片 - _CAPTION_","label":"Pod示意图","attributes":{},"skip":false,"key":"3.3.1.1.1"},{"backlink":"concepts/pod.html#fig3.3.1.1.2","level":"3.3.1.1","list_caption":"Figure: Pod Cheatsheet","alt":"Pod Cheatsheet","nro":27,"url":"../images/kubernetes-pod-cheatsheet.png","index":2,"caption_template":"图片 - _CAPTION_","label":"Pod Cheatsheet","attributes":{},"skip":false,"key":"3.3.1.1.2"},{"backlink":"concepts/service.html#fig3.3.4.1","level":"3.3.4","list_caption":"Figure: userspace代理模式下Service概览图","alt":"userspace代理模式下Service概览图","nro":28,"url":"../images/services-userspace-overview.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"userspace代理模式下Service概览图","attributes":{},"skip":false,"key":"3.3.4.1"},{"backlink":"concepts/service.html#fig3.3.4.2","level":"3.3.4","list_caption":"Figure: iptables代理模式下Service概览图","alt":"iptables代理模式下Service概览图","nro":29,"url":"../images/services-iptables-overview.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"iptables代理模式下Service概览图","attributes":{},"skip":false,"key":"3.3.4.2"},{"backlink":"concepts/deployment.html#fig3.3.6.1","level":"3.3.6","list_caption":"Figure: kubernetes deployment cheatsheet","alt":"kubernetes deployment cheatsheet","nro":30,"url":"../images/deployment-cheatsheet.png","index":1,"caption_template":"图片 - _CAPTION_","label":"kubernetes deployment cheatsheet","attributes":{},"skip":false,"key":"3.3.6.1"},{"backlink":"concepts/horizontal-pod-autoscaling.html#fig3.3.16.1","level":"3.3.16","list_caption":"Figure: horizontal-pod-autoscaler","alt":"horizontal-pod-autoscaler","nro":31,"url":"../images/horizontal-pod-autoscaler.png","index":1,"caption_template":"图片 - _CAPTION_","label":"horizontal-pod-autoscaler","attributes":{},"skip":false,"key":"3.3.16.1"},{"backlink":"concepts/label.html#fig3.3.17.1","level":"3.3.17","list_caption":"Figure: label示意图","alt":"label示意图","nro":32,"url":"../images/labels.png","index":1,"caption_template":"图片 - _CAPTION_","label":"label示意图","attributes":{},"skip":false,"key":"3.3.17.1"},{"backlink":"guide/using-kubectl.html#fig4.3.2.1","level":"4.3.2","list_caption":"Figure: kubectl cheatsheet","alt":"kubectl cheatsheet","nro":33,"url":"../images/kubernetes-kubectl-cheatsheet.png","index":1,"caption_template":"图片 - _CAPTION_","label":"kubectl cheatsheet","attributes":{},"skip":false,"key":"4.3.2.1"},{"backlink":"guide/using-kubectl.html#fig4.3.2.2","level":"4.3.2","list_caption":"Figure: kube-shell页面","alt":"kube-shell页面","nro":34,"url":"../images/kube-shell.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"kube-shell页面","attributes":{},"skip":false,"key":"4.3.2.2"},{"backlink":"guide/ip-masq-agent.html#fig4.4.6.1","level":"4.4.6","list_caption":"Figure: IP伪装代理示意图","alt":"IP伪装代理示意图","nro":35,"url":"../images/ip-masq.png","index":1,"caption_template":"图片 - _CAPTION_","label":"IP伪装代理示意图","attributes":{},"skip":false,"key":"4.4.6.1"},{"backlink":"guide/auth-with-kubeconfig-or-token.html#fig4.4.7.1","level":"4.4.7","list_caption":"Figure: kubeconfig文件","alt":"kubeconfig文件","nro":36,"url":"../images/brand-kubeconfig-yaml.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"kubeconfig文件","attributes":{},"skip":false,"key":"4.4.7.1"},{"backlink":"guide/authentication.html#fig4.4.8.1","level":"4.4.8","list_caption":"Figure: Kubernetes OpenID Connect Flow","alt":"Kubernetes OpenID Connect Flow","nro":37,"url":"../images/kubernetes-oidc-login.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"Kubernetes OpenID Connect Flow","attributes":{},"skip":false,"key":"4.4.8.1"},{"backlink":"guide/carbin-mobile-dashboard-for-kubernetes.html#fig4.5.6.1","level":"4.5.6","list_caption":"Figure: App Store","alt":"App Store","nro":38,"url":"../images/carbin-kubernetes-mobile-dashboard-1.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"App Store","attributes":{},"skip":false,"key":"4.5.6.1"},{"backlink":"guide/carbin-mobile-dashboard-for-kubernetes.html#fig4.5.6.2","level":"4.5.6","list_caption":"Figure: 在手机上操作Kubernetes集群","alt":"在手机上操作Kubernetes集群","nro":39,"url":"../images/carbin-kubernetes-mobile-dashboard-4.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"在手机上操作Kubernetes集群","attributes":{},"skip":false,"key":"4.5.6.2"},{"backlink":"guide/kubernetes-desktop-client.html#fig4.5.7.1","level":"4.5.7","list_caption":"Figure: Kubernetic客户端","alt":"Kubernetic客户端","nro":40,"url":"../images/kubernetic-desktop-ui.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"Kubernetic客户端","attributes":{},"skip":false,"key":"4.5.7.1"},{"backlink":"guide/deploy-applications-in-kubernetes.html#fig4.6.1.1","level":"4.6.1","list_caption":"Figure: API","alt":"API","nro":41,"url":"../images/k8s-app-monitor-test-api-doc.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"API","attributes":{},"skip":false,"key":"4.6.1.1"},{"backlink":"guide/deploy-applications-in-kubernetes.html#fig4.6.1.2","level":"4.6.1","list_caption":"Figure: wercker","alt":"wercker","nro":42,"url":"../images/k8s-app-monitor-agent-wercker.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"wercker","attributes":{},"skip":false,"key":"4.6.1.2"},{"backlink":"guide/deploy-applications-in-kubernetes.html#fig4.6.1.3","level":"4.6.1","list_caption":"Figure: 图表","alt":"图表","nro":43,"url":"../images/k8s-app-monitor-agent.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"图表","attributes":{},"skip":false,"key":"4.6.1.3"},{"backlink":"guide/migrating-hadoop-yarn-to-kubernetes.html#fig4.6.2.1","level":"4.6.2","list_caption":"Figure: spark on yarn with kubernetes","alt":"spark on yarn with kubernetes","nro":44,"url":"../images/spark-on-yarn-with-kubernetes.png","index":1,"caption_template":"图片 - _CAPTION_","label":"spark on yarn with kubernetes","attributes":{},"skip":false,"key":"4.6.2.1"},{"backlink":"guide/migrating-hadoop-yarn-to-kubernetes.html#fig4.6.2.2","level":"4.6.2","list_caption":"Figure: Terms","alt":"Terms","nro":45,"url":"../images/terms-in-kubernetes-app-deployment.png","index":2,"caption_template":"图片 - _CAPTION_","label":"Terms","attributes":{},"skip":false,"key":"4.6.2.2"},{"backlink":"guide/migrating-hadoop-yarn-to-kubernetes.html#fig4.6.2.3","level":"4.6.2","list_caption":"Figure: 分解步骤解析","alt":"分解步骤解析","nro":46,"url":"../images/migrating-hadoop-yarn-to-kubernetes.png","index":3,"caption_template":"图片 - _CAPTION_","label":"分解步骤解析","attributes":{},"skip":false,"key":"4.6.2.3"},{"backlink":"practice/node-installation.html#fig5.2.7.1","level":"5.2.7","list_caption":"Figure: welcome nginx","alt":"welcome nginx","nro":47,"url":"../images/kubernetes-installation-test-nginx.png","index":1,"caption_template":"图片 - _CAPTION_","label":"welcome nginx","attributes":{},"skip":false,"key":"5.2.7.1"},{"backlink":"practice/dashboard-addon-installation.html#fig5.2.9.1","level":"5.2.9","list_caption":"Figure: kubernetes dashboard","alt":"kubernetes dashboard","nro":48,"url":"../images/kubernetes-dashboard-raw.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"kubernetes dashboard","attributes":{},"skip":false,"key":"5.2.9.1"},{"backlink":"practice/dashboard-addon-installation.html#fig5.2.9.2","level":"5.2.9","list_caption":"Figure: V1.6.3版本的dashboard界面","alt":"V1.6.3版本的dashboard界面","nro":49,"url":"../images/dashboard-v163.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"V1.6.3版本的dashboard界面","attributes":{},"skip":false,"key":"5.2.9.2"},{"backlink":"practice/heapster-addon-installation.html#fig5.2.10.1","level":"5.2.10","list_caption":"Figure: dashboard-heapster","alt":"dashboard-heapster","nro":50,"url":"../images/kubernetes-dashboard-with-heapster.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"dashboard-heapster","attributes":{},"skip":false,"key":"5.2.10.1"},{"backlink":"practice/heapster-addon-installation.html#fig5.2.10.2","level":"5.2.10","list_caption":"Figure: grafana","alt":"grafana","nro":51,"url":"../images/kubernetes-heapster-grafana.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"grafana","attributes":{},"skip":false,"key":"5.2.10.2"},{"backlink":"practice/heapster-addon-installation.html#fig5.2.10.3","level":"5.2.10","list_caption":"Figure: kubernetes-influxdb-heapster","alt":"kubernetes-influxdb-heapster","nro":52,"url":"../images/kubernetes-influxdb-heapster.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"kubernetes-influxdb-heapster","attributes":{},"skip":false,"key":"5.2.10.3"},{"backlink":"practice/heapster-addon-installation.html#fig5.2.10.4","level":"5.2.10","list_caption":"Figure: 修改grafana模板","alt":"修改grafana模板","nro":53,"url":"../images/grafana-dashboard-setting.jpg","index":4,"caption_template":"图片 - _CAPTION_","label":"修改grafana模板","attributes":{},"skip":false,"key":"5.2.10.4"},{"backlink":"practice/efk-addon-installation.html#fig5.2.11.1","level":"5.2.11","list_caption":"Figure: es-setting","alt":"es-setting","nro":54,"url":"../images/es-setting.png","index":1,"caption_template":"图片 - _CAPTION_","label":"es-setting","attributes":{},"skip":false,"key":"5.2.11.1"},{"backlink":"practice/efk-addon-installation.html#fig5.2.11.2","level":"5.2.11","list_caption":"Figure: es-home","alt":"es-home","nro":55,"url":"../images/kubernetes-efk-kibana.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"es-home","attributes":{},"skip":false,"key":"5.2.11.2"},{"backlink":"practice/traefik-ingress-installation.html#fig5.3.1.1","level":"5.3.1","list_caption":"Figure: kubernetes-dashboard","alt":"kubernetes-dashboard","nro":56,"url":"../images/traefik-dashboard.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"kubernetes-dashboard","attributes":{},"skip":false,"key":"5.3.1.1"},{"backlink":"practice/traefik-ingress-installation.html#fig5.3.1.2","level":"5.3.1","list_caption":"Figure: traefik-nginx","alt":"traefik-nginx","nro":57,"url":"../images/traefik-nginx.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"traefik-nginx","attributes":{},"skip":false,"key":"5.3.1.2"},{"backlink":"practice/traefik-ingress-installation.html#fig5.3.1.3","level":"5.3.1","list_caption":"Figure: traefik-guestbook","alt":"traefik-guestbook","nro":58,"url":"../images/traefik-guestbook.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"traefik-guestbook","attributes":{},"skip":false,"key":"5.3.1.3"},{"backlink":"practice/distributed-load-test.html#fig5.3.2.1","level":"5.3.2","list_caption":"Figure: 使用dashboard来扩容","alt":"使用dashboard来扩容","nro":59,"url":"../images/dashbaord-scale.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"使用dashboard来扩容","attributes":{},"skip":false,"key":"5.3.2.1"},{"backlink":"practice/distributed-load-test.html#fig5.3.2.2","level":"5.3.2","list_caption":"Figure: Traefik的UI","alt":"Traefik的UI","nro":60,"url":"../images/traefik-dashboard-locust.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"Traefik的UI","attributes":{},"skip":false,"key":"5.3.2.2"},{"backlink":"practice/distributed-load-test.html#fig5.3.2.3","level":"5.3.2","list_caption":"Figure: Locust启动界面","alt":"Locust启动界面","nro":61,"url":"../images/locust-start-swarming.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"Locust启动界面","attributes":{},"skip":false,"key":"5.3.2.3"},{"backlink":"practice/distributed-load-test.html#fig5.3.2.4","level":"5.3.2","list_caption":"Figure: Dashboard查看页面","alt":"Dashboard查看页面","nro":62,"url":"../images/sample-webapp-rc.jpg","index":4,"caption_template":"图片 - _CAPTION_","label":"Dashboard查看页面","attributes":{},"skip":false,"key":"5.3.2.4"},{"backlink":"practice/distributed-load-test.html#fig5.3.2.5","level":"5.3.2","list_caption":"Figure: Locust测试结果页面","alt":"Locust测试结果页面","nro":63,"url":"../images/locust-dashboard.jpg","index":5,"caption_template":"图片 - _CAPTION_","label":"Locust测试结果页面","attributes":{},"skip":false,"key":"5.3.2.5"},{"backlink":"practice/network-and-cluster-perfermance-test.html#fig5.3.3.1","level":"5.3.3","list_caption":"Figure: kubernetes-dashboard","alt":"kubernetes-dashboard","nro":64,"url":"../images/kubenetes-e2e-test.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"kubernetes-dashboard","attributes":{},"skip":false,"key":"5.3.3.1"},{"backlink":"practice/network-and-cluster-perfermance-test.html#fig5.3.3.2","level":"5.3.3","list_caption":"Figure: locust测试页面","alt":"locust测试页面","nro":65,"url":"../images/kubernetes-locust-test.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"locust测试页面","attributes":{},"skip":false,"key":"5.3.3.2"},{"backlink":"practice/edge-node-configuration.html#fig5.3.4.1","level":"5.3.4","list_caption":"Figure: 边缘节点架构","alt":"边缘节点架构","nro":66,"url":"../images/kubernetes-edge-node-architecture.png","index":1,"caption_template":"图片 - _CAPTION_","label":"边缘节点架构","attributes":{},"skip":false,"key":"5.3.4.1"},{"backlink":"practice/app-log-collection.html#fig5.4.2.1","level":"5.4.2","list_caption":"Figure: filebeat日志收集架构图","alt":"filebeat日志收集架构图","nro":67,"url":"../images/filebeat-log-collector.png","index":1,"caption_template":"图片 - _CAPTION_","label":"filebeat日志收集架构图","attributes":{},"skip":false,"key":"5.4.2.1"},{"backlink":"practice/app-log-collection.html#fig5.4.2.2","level":"5.4.2","list_caption":"Figure: Kibana页面","alt":"Kibana页面","nro":68,"url":"../images/filebeat-docker-test.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"Kibana页面","attributes":{},"skip":false,"key":"5.4.2.2"},{"backlink":"practice/app-log-collection.html#fig5.4.2.3","level":"5.4.2","list_caption":"Figure: filebeat收集的日志详细信息","alt":"filebeat收集的日志详细信息","nro":69,"url":"../images/kubernetes-filebeat-detail.png","index":3,"caption_template":"图片 - _CAPTION_","label":"filebeat收集的日志详细信息","attributes":{},"skip":false,"key":"5.4.2.3"},{"backlink":"practice/monitor.html#fig5.4.4.1","level":"5.4.4","list_caption":"Figure: Kubernetes集群中的监控","alt":"Kubernetes集群中的监控","nro":70,"url":"../images/monitoring-in-kubernetes.png","index":1,"caption_template":"图片 - _CAPTION_","label":"Kubernetes集群中的监控","attributes":{},"skip":false,"key":"5.4.4.1"},{"backlink":"practice/monitor.html#fig5.4.4.2","level":"5.4.4","list_caption":"Figure: kubernetes的容器命名规则示意图","alt":"kubernetes的容器命名规则示意图","nro":71,"url":"../images/kubernetes-container-naming-rule.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"kubernetes的容器命名规则示意图","attributes":{},"skip":false,"key":"5.4.4.2"},{"backlink":"practice/monitor.html#fig5.4.4.3","level":"5.4.4","list_caption":"Figure: Heapster架构图（改进版）","alt":"Heapster架构图（改进版）","nro":72,"url":"../images/kubernetes-heapster-monitoring.png","index":3,"caption_template":"图片 - _CAPTION_","label":"Heapster架构图（改进版）","attributes":{},"skip":false,"key":"5.4.4.3"},{"backlink":"practice/monitor.html#fig5.4.4.4","level":"5.4.4","list_caption":"Figure: 应用监控架构图","alt":"应用监控架构图","nro":73,"url":"../images/kubernetes-app-monitoring.png","index":4,"caption_template":"图片 - _CAPTION_","label":"应用监控架构图","attributes":{},"skip":false,"key":"5.4.4.4"},{"backlink":"practice/monitor.html#fig5.4.4.5","level":"5.4.4","list_caption":"Figure: 应用拓扑图","alt":"应用拓扑图","nro":74,"url":"../images/weave-scope-service-topology.jpg","index":5,"caption_template":"图片 - _CAPTION_","label":"应用拓扑图","attributes":{},"skip":false,"key":"5.4.4.5"},{"backlink":"practice/data-persistence-problem.html#fig5.4.5.1","level":"5.4.5","list_caption":"Figure: 日志持久化收集解决方案示意图","alt":"日志持久化收集解决方案示意图","nro":75,"url":"../images/log-persistence-logstash.png","index":1,"caption_template":"图片 - _CAPTION_","label":"日志持久化收集解决方案示意图","attributes":{},"skip":false,"key":"5.4.5.1"},{"backlink":"practice/storage-for-containers-using-glusterfs-with-openshift.html#fig5.5.1.2.1","level":"5.5.1.2","list_caption":"Figure: 创建存储","alt":"创建存储","nro":76,"url":"../images/create-gluster-storage.png","index":1,"caption_template":"图片 - _CAPTION_","label":"创建存储","attributes":{},"skip":false,"key":"5.5.1.2.1"},{"backlink":"practice/storage-for-containers-using-glusterfs-with-openshift.html#fig5.5.1.2.2","level":"5.5.1.2","list_caption":"Figure: Screen Shot 2017-03-24 at 11.09.34.png","alt":"Screen Shot 2017-03-24 at 11.09.34.png","nro":77,"url":"https://keithtenzer.files.wordpress.com/2017/03/screen-shot-2017-03-24-at-11-09-341.png?w=440","index":2,"caption_template":"图片 - _CAPTION_","label":"Screen Shot 2017-03-24 at 11.09.34.png","attributes":{},"skip":false,"key":"5.5.1.2.2"},{"backlink":"practice/using-heapster-to-get-object-metrics.html#fig5.6.1.1.1","level":"5.6.1.1","list_caption":"Figure: Heapster架构图","alt":"Heapster架构图","nro":78,"url":"../images/heapster-architecture.png","index":1,"caption_template":"图片 - _CAPTION_","label":"Heapster架构图","attributes":{},"skip":false,"key":"5.6.1.1.1"},{"backlink":"practice/prometheus.html#fig5.6.2.1","level":"5.6.2","list_caption":"Figure: Prometheus 架构图","alt":"Prometheus 架构图","nro":79,"url":"../images/prometheus-architecture.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"Prometheus 架构图","attributes":{},"skip":false,"key":"5.6.2.1"},{"backlink":"practice/using-prometheus-to-monitor-kuberentes-cluster.html#fig5.6.2.1.1","level":"5.6.2.1","list_caption":"Figure: Grafana页面","alt":"Grafana页面","nro":80,"url":"../images/kubernetes-prometheus-monitoring.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"Grafana页面","attributes":{},"skip":false,"key":"5.6.2.1.1"},{"backlink":"practice/helm.html#fig5.7.1.1","level":"5.7.1","list_caption":"Figure: Helm chart源","alt":"Helm chart源","nro":81,"url":"../images/helm-charts-repository.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"Helm chart源","attributes":{},"skip":false,"key":"5.7.1.1"},{"backlink":"practice/helm.html#fig5.7.1.2","level":"5.7.1","list_caption":"Figure: TODO应用的Web页面","alt":"TODO应用的Web页面","nro":82,"url":"../images/helm-mean-todo-aholic.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"TODO应用的Web页面","attributes":{},"skip":false,"key":"5.7.1.2"},{"backlink":"practice/create-private-charts-repo.html#fig5.7.2.1","level":"5.7.2","list_caption":"Figure: Helm monocular界面","alt":"Helm monocular界面","nro":83,"url":"../images/helm-monocular-jimmysong.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"Helm monocular界面","attributes":{},"skip":false,"key":"5.7.2.1"},{"backlink":"practice/jenkins-ci-cd.html#fig5.8.1.1","level":"5.8.1","list_caption":"Figure: 基于Jenkins的持续集成与发布","alt":"基于Jenkins的持续集成与发布","nro":84,"url":"../images/kubernetes-jenkins-ci-cd.png","index":1,"caption_template":"图片 - _CAPTION_","label":"基于Jenkins的持续集成与发布","attributes":{},"skip":false,"key":"5.8.1.1"},{"backlink":"practice/drone-ci-cd.html#fig5.8.2.1","level":"5.8.2","list_caption":"Figure: OAuth注册","alt":"OAuth注册","nro":85,"url":"../images/github-oauth-register.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"OAuth注册","attributes":{},"skip":false,"key":"5.8.2.1"},{"backlink":"practice/drone-ci-cd.html#fig5.8.2.2","level":"5.8.2","list_caption":"Figure: OAuth key","alt":"OAuth key","nro":86,"url":"../images/github-oauth-drone-key.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"OAuth key","attributes":{},"skip":false,"key":"5.8.2.2"},{"backlink":"practice/drone-ci-cd.html#fig5.8.2.3","level":"5.8.2","list_caption":"Figure: Drone登陆界面","alt":"Drone登陆界面","nro":87,"url":"../images/drone-login-github.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"Drone登陆界面","attributes":{},"skip":false,"key":"5.8.2.3"},{"backlink":"practice/drone-ci-cd.html#fig5.8.2.4","level":"5.8.2","list_caption":"Figure: Github启用repo设置","alt":"Github启用repo设置","nro":88,"url":"../images/drone-github-active.jpg","index":4,"caption_template":"图片 - _CAPTION_","label":"Github启用repo设置","attributes":{},"skip":false,"key":"5.8.2.4"},{"backlink":"practice/drone-ci-cd.html#fig5.8.2.5","level":"5.8.2","list_caption":"Figure: Github单个repo设置","alt":"Github单个repo设置","nro":89,"url":"../images/drone-github-repo-setting.jpg","index":5,"caption_template":"图片 - _CAPTION_","label":"Github单个repo设置","attributes":{},"skip":false,"key":"5.8.2.5"},{"backlink":"practice/dashboard-upgrade.html#fig5.9.2.1","level":"5.9.2","list_caption":"Figure: 登陆界面","alt":"登陆界面","nro":90,"url":"../images/kubernetes-dashboard-1.7.1-login.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"登陆界面","attributes":{},"skip":false,"key":"5.9.2.1"},{"backlink":"practice/dashboard-upgrade.html#fig5.9.2.2","level":"5.9.2","list_caption":"Figure: 首页","alt":"首页","nro":91,"url":"../images/kubernetes-dashboard-1.7.1-default-page.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"首页","attributes":{},"skip":false,"key":"5.9.2.2"},{"backlink":"practice/dashboard-upgrade.html#fig5.9.2.3","level":"5.9.2","list_caption":"Figure: 用户空间","alt":"用户空间","nro":92,"url":"../images/kubernetes-dashboard-1.7.1-brand.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"用户空间","attributes":{},"skip":false,"key":"5.9.2.3"},{"backlink":"practice/dashboard-upgrade.html#fig5.9.2.4","level":"5.9.2","list_caption":"Figure: kubeconfig文件","alt":"kubeconfig文件","nro":93,"url":"../images/brand-kubeconfig-yaml.jpg","index":4,"caption_template":"图片 - _CAPTION_","label":"kubeconfig文件","attributes":{},"skip":false,"key":"5.9.2.4"},{"backlink":"usecases/microservices.html#fig6.2.1","level":"6.2","list_caption":"Figure: 微服务关注的部分","alt":"微服务关注的部分","nro":94,"url":"../images/microservices-concerns.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"微服务关注的部分","attributes":{},"skip":false,"key":"6.2.1"},{"backlink":"usecases/service-discovery-in-microservices.html#fig6.2.1.1","level":"6.2.1","list_caption":"Figure: 微服务中的服务发现","alt":"微服务中的服务发现","nro":95,"url":"../images/service-discovery-in-microservices.png","index":1,"caption_template":"图片 - _CAPTION_","label":"微服务中的服务发现","attributes":{},"skip":false,"key":"6.2.1.1"},{"backlink":"usecases/service-mesh.html#fig6.3.1","level":"6.3","list_caption":"Figure: 下一代异构微服务架构","alt":"下一代异构微服务架构","nro":96,"url":"../images/polyglot-microservices-serivce-mesh.png","index":1,"caption_template":"图片 - _CAPTION_","label":"下一代异构微服务架构","attributes":{},"skip":false,"key":"6.3.1"},{"backlink":"usecases/service-mesh.html#fig6.3.2","level":"6.3","list_caption":"Figure: Service Mesh 架构图","alt":"Service Mesh 架构图","nro":97,"url":"../images/serivce-mesh-control-plane.png","index":2,"caption_template":"图片 - _CAPTION_","label":"Service Mesh 架构图","attributes":{},"skip":false,"key":"6.3.2"},{"backlink":"usecases/service-mesh.html#fig6.3.3","level":"6.3","list_caption":"Figure: Istio vs linkerd","alt":"Istio vs linkerd","nro":98,"url":"../images/istio-vs-linkerd.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"Istio vs linkerd","attributes":{},"skip":false,"key":"6.3.3"},{"backlink":"usecases/istio.html#fig6.3.1.1","level":"6.3.1","list_caption":"Figure: Istio架构图","alt":"Istio架构图","nro":99,"url":"../images/istio-arch.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"Istio架构图","attributes":{},"skip":false,"key":"6.3.1.1"},{"backlink":"usecases/istio-installation.html#fig6.3.1.1.1","level":"6.3.1.1","list_caption":"Figure: BookInfo Sample应用架构图","alt":"BookInfo Sample应用架构图","nro":100,"url":"../images/bookinfo-sample-arch.png","index":1,"caption_template":"图片 - _CAPTION_","label":"BookInfo Sample应用架构图","attributes":{},"skip":false,"key":"6.3.1.1.1"},{"backlink":"usecases/istio-installation.html#fig6.3.1.1.2","level":"6.3.1.1","list_caption":"Figure: BookInfo Sample页面","alt":"BookInfo Sample页面","nro":101,"url":"../images/bookinfo-sample.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"BookInfo Sample页面","attributes":{},"skip":false,"key":"6.3.1.1.2"},{"backlink":"usecases/istio-installation.html#fig6.3.1.1.3","level":"6.3.1.1","list_caption":"Figure: Istio Grafana界面","alt":"Istio Grafana界面","nro":102,"url":"../images/istio-grafana.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"Istio Grafana界面","attributes":{},"skip":false,"key":"6.3.1.1.3"},{"backlink":"usecases/istio-installation.html#fig6.3.1.1.4","level":"6.3.1.1","list_caption":"Figure: Prometheus页面","alt":"Prometheus页面","nro":103,"url":"../images/istio-prometheus.jpg","index":4,"caption_template":"图片 - _CAPTION_","label":"Prometheus页面","attributes":{},"skip":false,"key":"6.3.1.1.4"},{"backlink":"usecases/istio-installation.html#fig6.3.1.1.5","level":"6.3.1.1","list_caption":"Figure: Zipkin页面","alt":"Zipkin页面","nro":104,"url":"../images/istio-zipkin.jpg","index":5,"caption_template":"图片 - _CAPTION_","label":"Zipkin页面","attributes":{},"skip":false,"key":"6.3.1.1.5"},{"backlink":"usecases/istio-installation.html#fig6.3.1.1.6","level":"6.3.1.1","list_caption":"Figure: ServiceGraph页面","alt":"ServiceGraph页面","nro":105,"url":"../images/istio-servicegraph.jpg","index":6,"caption_template":"图片 - _CAPTION_","label":"ServiceGraph页面","attributes":{},"skip":false,"key":"6.3.1.1.6"},{"backlink":"usecases/install-and-expand-istio-mesh.html#fig6.3.1.3.1","level":"6.3.1.3","list_caption":"Figure: BookInfo","alt":"BookInfo","nro":106,"url":"../images/noistio.png","index":1,"caption_template":"图片 - _CAPTION_","label":"BookInfo","attributes":{},"skip":false,"key":"6.3.1.3.1"},{"backlink":"usecases/install-and-expand-istio-mesh.html#fig6.3.1.3.2","level":"6.3.1.3","list_caption":"Figure: BookInfo","alt":"BookInfo","nro":107,"url":"../images/noistio.png","index":2,"caption_template":"图片 - _CAPTION_","label":"BookInfo","attributes":{},"skip":false,"key":"6.3.1.3.2"},{"backlink":"usecases/integrating-vms.html#fig6.3.1.4.1","level":"6.3.1.4","list_caption":"Figure: Bookinfo应用的拓展Mesh","alt":"Bookinfo应用的拓展Mesh","nro":108,"url":"../images/istio-mesh-expansion.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"Bookinfo应用的拓展Mesh","attributes":{},"skip":false,"key":"6.3.1.4.1"},{"backlink":"usecases/linkerd.html#fig6.3.2.1","level":"6.3.2","list_caption":"Figure: source https://linkerd.io","alt":"source https://linkerd.io","nro":109,"url":"../images/diagram-individual-instance.png","index":1,"caption_template":"图片 - _CAPTION_","label":"source https://linkerd.io","attributes":{},"skip":false,"key":"6.3.2.1"},{"backlink":"usecases/linkerd-user-guide.html#fig6.3.2.1.1","level":"6.3.2.1","list_caption":"Figure: Jenkins pipeline","alt":"Jenkins pipeline","nro":110,"url":"../images/linkerd-jenkins-pipeline.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"Jenkins pipeline","attributes":{},"skip":false,"key":"6.3.2.1.1"},{"backlink":"usecases/linkerd-user-guide.html#fig6.3.2.1.2","level":"6.3.2.1","list_caption":"Figure: Jenkins config","alt":"Jenkins config","nro":111,"url":"../images/linkerd-jenkins.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"Jenkins config","attributes":{},"skip":false,"key":"6.3.2.1.2"},{"backlink":"usecases/linkerd-user-guide.html#fig6.3.2.1.3","level":"6.3.2.1","list_caption":"Figure: namerd","alt":"namerd","nro":112,"url":"../images/namerd-internal.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"namerd","attributes":{},"skip":false,"key":"6.3.2.1.3"},{"backlink":"usecases/linkerd-user-guide.html#fig6.3.2.1.4","level":"6.3.2.1","list_caption":"Figure: linkerd监控","alt":"linkerd监控","nro":113,"url":"../images/linkerd-helloworld-outgoing.jpg","index":4,"caption_template":"图片 - _CAPTION_","label":"linkerd监控","attributes":{},"skip":false,"key":"6.3.2.1.4"},{"backlink":"usecases/linkerd-user-guide.html#fig6.3.2.1.5","level":"6.3.2.1","list_caption":"Figure: linkerd监控","alt":"linkerd监控","nro":114,"url":"../images/linkerd-helloworld-incoming.jpg","index":5,"caption_template":"图片 - _CAPTION_","label":"linkerd监控","attributes":{},"skip":false,"key":"6.3.2.1.5"},{"backlink":"usecases/linkerd-user-guide.html#fig6.3.2.1.6","level":"6.3.2.1","list_caption":"Figure: linkerd性能监控","alt":"linkerd性能监控","nro":115,"url":"../images/linkerd-grafana.png","index":6,"caption_template":"图片 - _CAPTION_","label":"linkerd性能监控","attributes":{},"skip":false,"key":"6.3.2.1.6"},{"backlink":"usecases/linkerd-user-guide.html#fig6.3.2.1.7","level":"6.3.2.1","list_caption":"Figure: Linkerd ingress controller","alt":"Linkerd ingress controller","nro":116,"url":"../images/linkerd-ingress-controller.jpg","index":7,"caption_template":"图片 - _CAPTION_","label":"Linkerd ingress controller","attributes":{},"skip":false,"key":"6.3.2.1.7"},{"backlink":"usecases/conduit-installation.html#fig6.3.3.2.1","level":"6.3.3.2","list_caption":"Figure: Conduit dashboard","alt":"Conduit dashboard","nro":117,"url":"../images/conduit-dashboard.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"Conduit dashboard","attributes":{},"skip":false,"key":"6.3.3.2.1"},{"backlink":"usecases/big-data.html#fig6.4.1","level":"6.4","list_caption":"Figure: Spark on yarn with kubernetes","alt":"Spark on yarn with kubernetes","nro":118,"url":"../images/spark-on-yarn-with-kubernetes.png","index":1,"caption_template":"图片 - _CAPTION_","label":"Spark on yarn with kubernetes","attributes":{},"skip":false,"key":"6.4.1"},{"backlink":"usecases/big-data.html#fig6.4.2","level":"6.4","list_caption":"Figure: 在kubernetes上使用多种调度方式","alt":"在kubernetes上使用多种调度方式","nro":119,"url":"../images/spark-on-kubernetes-with-different-schedulers.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"在kubernetes上使用多种调度方式","attributes":{},"skip":false,"key":"6.4.2"},{"backlink":"usecases/spark-standalone-on-kubernetes.html#fig6.4.1.1","level":"6.4.1","list_caption":"Figure: spark master ui","alt":"spark master ui","nro":120,"url":"../images/spark-ui.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"spark master ui","attributes":{},"skip":false,"key":"6.4.1.1"},{"backlink":"usecases/spark-standalone-on-kubernetes.html#fig6.4.1.2","level":"6.4.1","list_caption":"Figure: zeppelin ui","alt":"zeppelin ui","nro":121,"url":"../images/zeppelin-ui.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"zeppelin ui","attributes":{},"skip":false,"key":"6.4.1.2"},{"backlink":"usecases/serverless.html#fig6.5.1","level":"6.5","list_caption":"Figure: 从物理机到函数计算","alt":"从物理机到函数计算","nro":122,"url":"../images/from-bare-metal-to-functions.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"从物理机到函数计算","attributes":{},"skip":false,"key":"6.5.1"},{"backlink":"usecases/serverless.html#fig6.5.2","level":"6.5","list_caption":"Figure: FaaS Landscape","alt":"FaaS Landscape","nro":123,"url":"../images/redpoint-faas-landscape.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"FaaS Landscape","attributes":{},"skip":false,"key":"6.5.2"},{"backlink":"usecases/understanding-serverless.html#fig6.5.1.1","level":"6.5.1","list_caption":"Figure: 服务端软件的运行环境","alt":"服务端软件的运行环境","nro":124,"url":"../images/serverless-server-side-software.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"服务端软件的运行环境","attributes":{},"skip":false,"key":"6.5.1.1"},{"backlink":"usecases/understanding-serverless.html#fig6.5.1.2","level":"6.5.1","list_caption":"Figure: FaaS应用架构","alt":"FaaS应用架构","nro":125,"url":"../images/serverless-faas-platform.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"FaaS应用架构","attributes":{},"skip":false,"key":"6.5.1.2"},{"backlink":"usecases/understanding-serverless.html#fig6.5.1.3","level":"6.5.1","list_caption":"Figure: 传统应用程序架构","alt":"传统应用程序架构","nro":126,"url":"../images/non-serverless-game-arch.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"传统应用程序架构","attributes":{},"skip":false,"key":"6.5.1.3"},{"backlink":"usecases/understanding-serverless.html#fig6.5.1.4","level":"6.5.1","list_caption":"Figure: Serverless架构","alt":"Serverless架构","nro":127,"url":"../images/serverless-game-arch.jpg","index":4,"caption_template":"图片 - _CAPTION_","label":"Serverless架构","attributes":{},"skip":false,"key":"6.5.1.4"},{"backlink":"usecases/openfaas.html#fig6.5.2.1.1","level":"6.5.2.1","list_caption":"Figure: OpenFaaS架构","alt":"OpenFaaS架构","nro":128,"url":"../images/openfaas-arch.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"OpenFaaS架构","attributes":{},"skip":false,"key":"6.5.2.1.1"},{"backlink":"usecases/openfaas.html#fig6.5.2.1.2","level":"6.5.2.1","list_caption":"Figure: OpenFaaS Prometheus","alt":"OpenFaaS Prometheus","nro":129,"url":"../images/openfaas-prometheus.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"OpenFaaS Prometheus","attributes":{},"skip":false,"key":"6.5.2.1.2"},{"backlink":"usecases/openfaas.html#fig6.5.2.1.3","level":"6.5.2.1","list_caption":"Figure: OpenFaas Grafana监控","alt":"OpenFaas Grafana监控","nro":130,"url":"../images/openfaas-grafana.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"OpenFaas Grafana监控","attributes":{},"skip":false,"key":"6.5.2.1.3"},{"backlink":"usecases/openfaas.html#fig6.5.2.1.4","level":"6.5.2.1","list_caption":"Figure: OpenFaas Dashboard","alt":"OpenFaas Dashboard","nro":131,"url":"../images/openfaas-deploy-a-function.jpg","index":4,"caption_template":"图片 - _CAPTION_","label":"OpenFaas Dashboard","attributes":{},"skip":false,"key":"6.5.2.1.4"},{"backlink":"usecases/openfaas.html#fig6.5.2.1.5","level":"6.5.2.1","list_caption":"Figure: NodeInfo执行结果","alt":"NodeInfo执行结果","nro":132,"url":"../images/openfaas-nodeinfo.jpg","index":5,"caption_template":"图片 - _CAPTION_","label":"NodeInfo执行结果","attributes":{},"skip":false,"key":"6.5.2.1.5"},{"backlink":"develop/client-go-sample.html#fig7.5.1","level":"7.5","list_caption":"Figure: 使用kubernetes dashboard进行故障排查","alt":"使用kubernetes dashboard进行故障排查","nro":133,"url":"../images/kubernetes-client-go-sample-update.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"使用kubernetes dashboard进行故障排查","attributes":{},"skip":false,"key":"7.5.1"},{"backlink":"appendix/issues.html#fig8.6.1","level":"8.6","list_caption":"Figure: pvc-storage-limit","alt":"pvc-storage-limit","nro":134,"url":"../images/pvc-storage-limit.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"pvc-storage-limit","attributes":{},"skip":false,"key":"8.6.1"}]},"title":"Kubernetes Handbook - jimmysong.io","language":"zh-hans","links":{"sidebar":{"Jimmy Song":"https://jimmysong.io"}},"gitbook":"*","description":"Kubernetes中文指南/实践手册|kubernetes handbook|jimmysong.io|宋净超"},"file":{"path":"practice/master-installation.md","mtime":"2017-12-15T07:17:07.278Z","type":"markdown"},"gitbook":{"version":"3.2.2","time":"2017-12-20T09:26:03.762Z"},"basePath":"..","book":{"language":""}}); + gitbook.page.hasChanged({"page":{"title":"部署master节点","level":"5.2.5","depth":2,"next":{"title":"安装flannel网络插件","level":"5.2.6","depth":2,"path":"practice/flannel-installation.md","ref":"practice/flannel-installation.md","articles":[]},"previous":{"title":"安装kubectl命令行工具","level":"5.2.4","depth":2,"path":"practice/kubectl-installation.md","ref":"practice/kubectl-installation.md","articles":[]},"dir":"ltr"},"config":{"plugins":["github","codesnippet","splitter","page-toc-button","image-captions","editlink","back-to-top-button","-lunr","-search","search-plus","github-buttons@2.1.0","favicon@^0.0.2","tbfed-pagefooter@^0.0.1","3-ba","theme-default"],"styles":{"website":"styles/website.css","pdf":"styles/pdf.css","epub":"styles/epub.css","mobi":"styles/mobi.css","ebook":"styles/ebook.css","print":"styles/print.css"},"pluginsConfig":{"tbfed-pagefooter":{"copyright":"Copyright © jimmysong.io 2017","modify_label":"Updated:","modify_format":"YYYY-MM-DD HH:mm:ss"},"github":{"url":"https://github.com/rootsongjc/kubernetes-handbook"},"editlink":{"label":"编辑本页","multilingual":false,"base":"https://github.com/rootsongjc/kubernetes-handbook/blob/master/"},"splitter":{},"codesnippet":{},"fontsettings":{"theme":"white","family":"sans","size":2},"highlight":{},"favicon":{"shortcut":"favicon.ico","bookmark":"favicon.ico"},"page-toc-button":{},"back-to-top-button":{},"github-buttons":{"repo":"rootsongjc/kubernetes-handbook","types":["star"],"size":"small"},"3-ba":{"configuration":"auto","token":"11f7d254cfa4e0ca44b175c66d379ecc"},"sharing":{"facebook":true,"twitter":true,"google":false,"weibo":false,"instapaper":false,"vk":false,"all":["facebook","google","twitter","weibo","instapaper"]},"theme-default":{"styles":{"website":"styles/website.css","pdf":"styles/pdf.css","epub":"styles/epub.css","mobi":"styles/mobi.css","ebook":"styles/ebook.css","print":"styles/print.css"},"showLevel":true},"search-plus":{},"image-captions":{"caption":"图片 - _CAPTION_","variable_name":"_pictures"}},"theme":"default","author":"Jimmy Song","pdf":{"pageNumbers":true,"fontSize":12,"fontFamily":"Arial","paperSize":"a4","chapterMark":"pagebreak","pageBreaksBefore":"/","margin":{"right":62,"left":62,"top":56,"bottom":56}},"structure":{"langs":"LANGS.md","readme":"README.md","glossary":"GLOSSARY.md","summary":"SUMMARY.md"},"variables":{"_pictures":[{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.1","level":"2.1","list_caption":"Figure: 云计算演进历程","alt":"云计算演进历程","nro":1,"url":"../images/cloud-computing-evolution-road.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"云计算演进历程","attributes":{},"skip":false,"key":"2.1.1"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.2","level":"2.1","list_caption":"Figure: 来自Twitter @MarcWilczek","alt":"来自Twitter @MarcWilczek","nro":2,"url":"../images/cloud-native-comes-of-age.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"来自Twitter @MarcWilczek","attributes":{},"skip":false,"key":"2.1.2"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.3","level":"2.1","list_caption":"Figure: Cloud native思维导图","alt":"Cloud native思维导图","nro":3,"url":"../images/cloud-native-architecutre-mindnode.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"Cloud native思维导图","attributes":{},"skip":false,"key":"2.1.3"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.4","level":"2.1","list_caption":"Figure: 十二因素应用","alt":"十二因素应用","nro":4,"url":"../images/12-factor-app.png","index":4,"caption_template":"图片 - _CAPTION_","label":"十二因素应用","attributes":{},"skip":false,"key":"2.1.4"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.5","level":"2.1","list_caption":"Figure: 容器生态","alt":"容器生态","nro":5,"url":"../images/container-ecosystem.png","index":5,"caption_template":"图片 - _CAPTION_","label":"容器生态","attributes":{},"skip":false,"key":"2.1.5"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.6","level":"2.1","list_caption":"Figure: 使用Jenkins进行持续集成与发布流程图","alt":"使用Jenkins进行持续集成与发布流程图","nro":6,"url":"../images/kubernetes-jenkins-ci-cd.png","index":6,"caption_template":"图片 - _CAPTION_","label":"使用Jenkins进行持续集成与发布流程图","attributes":{},"skip":false,"key":"2.1.6"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.7","level":"2.1","list_caption":"Figure: filebeat日志收集架构图","alt":"filebeat日志收集架构图","nro":7,"url":"../images/filebeat-log-collector-arch.png","index":7,"caption_template":"图片 - _CAPTION_","label":"filebeat日志收集架构图","attributes":{},"skip":false,"key":"2.1.7"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.8","level":"2.1","list_caption":"Figure: API文档","alt":"API文档","nro":8,"url":"../images/k8s-app-monitor-test-api-doc.jpg","index":8,"caption_template":"图片 - _CAPTION_","label":"API文档","attributes":{},"skip":false,"key":"2.1.8"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.9","level":"2.1","list_caption":"Figure: 迁移步骤示意图","alt":"迁移步骤示意图","nro":9,"url":"../images/migrating-hadoop-yarn-to-kubernetes.png","index":9,"caption_template":"图片 - _CAPTION_","label":"迁移步骤示意图","attributes":{},"skip":false,"key":"2.1.9"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.10","level":"2.1","list_caption":"Figure: service mesh架构图","alt":"service mesh架构图","nro":10,"url":"../images/serivce-mesh-control-plane.png","index":10,"caption_template":"图片 - _CAPTION_","label":"service mesh架构图","attributes":{},"skip":false,"key":"2.1.10"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.11","level":"2.1","list_caption":"Figure: kibana界面","alt":"kibana界面","nro":11,"url":"../images/filebeat-docker-test.jpg","index":11,"caption_template":"图片 - _CAPTION_","label":"kibana界面","attributes":{},"skip":false,"key":"2.1.11"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.12","level":"2.1","list_caption":"Figure: Grafana界面示意图1","alt":"Grafana界面示意图1","nro":12,"url":"../images/kubernetes-devops-example-grafana-1.png","index":12,"caption_template":"图片 - _CAPTION_","label":"Grafana界面示意图1","attributes":{},"skip":false,"key":"2.1.12"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.13","level":"2.1","list_caption":"Figure: Grafana界面示意图2","alt":"Grafana界面示意图2","nro":13,"url":"../images/kubernetes-devops-example-grafana-2.png","index":13,"caption_template":"图片 - _CAPTION_","label":"Grafana界面示意图2","attributes":{},"skip":false,"key":"2.1.13"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.14","level":"2.1","list_caption":"Figure: Grafana界面示意图3","alt":"Grafana界面示意图3","nro":14,"url":"../images/kubernetes-devops-example-grafana-3.png","index":14,"caption_template":"图片 - _CAPTION_","label":"Grafana界面示意图3","attributes":{},"skip":false,"key":"2.1.14"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.15","level":"2.1","list_caption":"Figure: dashboard","alt":"dashboard","nro":15,"url":"../images/spark-job-on-kubernetes-example-1.jpg","index":15,"caption_template":"图片 - _CAPTION_","label":"dashboard","attributes":{},"skip":false,"key":"2.1.15"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.16","level":"2.1","list_caption":"Figure: Grafana","alt":"Grafana","nro":16,"url":"../images/spark-job-on-kubernetes-example-2.jpg","index":16,"caption_template":"图片 - _CAPTION_","label":"Grafana","attributes":{},"skip":false,"key":"2.1.16"},{"backlink":"concepts/index.html#fig3.1.1","level":"3.1","list_caption":"Figure: Borg架构","alt":"Borg架构","nro":17,"url":"../images/borg.png","index":1,"caption_template":"图片 - _CAPTION_","label":"Borg架构","attributes":{},"skip":false,"key":"3.1.1"},{"backlink":"concepts/index.html#fig3.1.2","level":"3.1","list_caption":"Figure: Kubernetes架构","alt":"Kubernetes架构","nro":18,"url":"../images/architecture.png","index":2,"caption_template":"图片 - _CAPTION_","label":"Kubernetes架构","attributes":{},"skip":false,"key":"3.1.2"},{"backlink":"concepts/index.html#fig3.1.3","level":"3.1","list_caption":"Figure: Kuberentes架构（图片来自于网络）","alt":"Kuberentes架构（图片来自于网络）","nro":19,"url":"../images/kubernetes-high-level-component-archtecture.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"Kuberentes架构（图片来自于网络）","attributes":{},"skip":false,"key":"3.1.3"},{"backlink":"concepts/index.html#fig3.1.4","level":"3.1","list_caption":"Figure: kubernetes整体架构示意图","alt":"kubernetes整体架构示意图","nro":20,"url":"../images/kubernetes-whole-arch.png","index":4,"caption_template":"图片 - _CAPTION_","label":"kubernetes整体架构示意图","attributes":{},"skip":false,"key":"3.1.4"},{"backlink":"concepts/index.html#fig3.1.5","level":"3.1","list_caption":"Figure: Kubernetes master架构示意图","alt":"Kubernetes master架构示意图","nro":21,"url":"../images/kubernetes-master-arch.png","index":5,"caption_template":"图片 - _CAPTION_","label":"Kubernetes master架构示意图","attributes":{},"skip":false,"key":"3.1.5"},{"backlink":"concepts/index.html#fig3.1.6","level":"3.1","list_caption":"Figure: kubernetes node架构示意图","alt":"kubernetes node架构示意图","nro":22,"url":"../images/kubernetes-node-arch.png","index":6,"caption_template":"图片 - _CAPTION_","label":"kubernetes node架构示意图","attributes":{},"skip":false,"key":"3.1.6"},{"backlink":"concepts/index.html#fig3.1.7","level":"3.1","list_caption":"Figure: Kubernetes分层架构示意图","alt":"Kubernetes分层架构示意图","nro":23,"url":"../images/kubernetes-layers-arch.jpg","index":7,"caption_template":"图片 - _CAPTION_","label":"Kubernetes分层架构示意图","attributes":{},"skip":false,"key":"3.1.7"},{"backlink":"concepts/concepts.html#fig3.2.1","level":"3.2","list_caption":"Figure: 分层架构示意图","alt":"分层架构示意图","nro":24,"url":"../images/kubernetes-layers-arch.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"分层架构示意图","attributes":{},"skip":false,"key":"3.2.1"},{"backlink":"concepts/pod-overview.html#fig3.3.1.1","level":"3.3.1","list_caption":"Figure: pod diagram","alt":"pod diagram","nro":25,"url":"../images/pod-overview.png","index":1,"caption_template":"图片 - _CAPTION_","label":"pod diagram","attributes":{},"skip":false,"key":"3.3.1.1"},{"backlink":"concepts/pod.html#fig3.3.1.1.1","level":"3.3.1.1","list_caption":"Figure: Pod示意图","alt":"Pod示意图","nro":26,"url":"../images/pod-overview.png","index":1,"caption_template":"图片 - _CAPTION_","label":"Pod示意图","attributes":{},"skip":false,"key":"3.3.1.1.1"},{"backlink":"concepts/pod.html#fig3.3.1.1.2","level":"3.3.1.1","list_caption":"Figure: Pod Cheatsheet","alt":"Pod Cheatsheet","nro":27,"url":"../images/kubernetes-pod-cheatsheet.png","index":2,"caption_template":"图片 - _CAPTION_","label":"Pod Cheatsheet","attributes":{},"skip":false,"key":"3.3.1.1.2"},{"backlink":"concepts/service.html#fig3.3.4.1","level":"3.3.4","list_caption":"Figure: userspace代理模式下Service概览图","alt":"userspace代理模式下Service概览图","nro":28,"url":"../images/services-userspace-overview.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"userspace代理模式下Service概览图","attributes":{},"skip":false,"key":"3.3.4.1"},{"backlink":"concepts/service.html#fig3.3.4.2","level":"3.3.4","list_caption":"Figure: iptables代理模式下Service概览图","alt":"iptables代理模式下Service概览图","nro":29,"url":"../images/services-iptables-overview.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"iptables代理模式下Service概览图","attributes":{},"skip":false,"key":"3.3.4.2"},{"backlink":"concepts/deployment.html#fig3.3.6.1","level":"3.3.6","list_caption":"Figure: kubernetes deployment cheatsheet","alt":"kubernetes deployment cheatsheet","nro":30,"url":"../images/deployment-cheatsheet.png","index":1,"caption_template":"图片 - _CAPTION_","label":"kubernetes deployment cheatsheet","attributes":{},"skip":false,"key":"3.3.6.1"},{"backlink":"concepts/horizontal-pod-autoscaling.html#fig3.3.16.1","level":"3.3.16","list_caption":"Figure: horizontal-pod-autoscaler","alt":"horizontal-pod-autoscaler","nro":31,"url":"../images/horizontal-pod-autoscaler.png","index":1,"caption_template":"图片 - _CAPTION_","label":"horizontal-pod-autoscaler","attributes":{},"skip":false,"key":"3.3.16.1"},{"backlink":"concepts/label.html#fig3.3.17.1","level":"3.3.17","list_caption":"Figure: label示意图","alt":"label示意图","nro":32,"url":"../images/labels.png","index":1,"caption_template":"图片 - _CAPTION_","label":"label示意图","attributes":{},"skip":false,"key":"3.3.17.1"},{"backlink":"guide/using-kubectl.html#fig4.3.2.1","level":"4.3.2","list_caption":"Figure: kubectl cheatsheet","alt":"kubectl cheatsheet","nro":33,"url":"../images/kubernetes-kubectl-cheatsheet.png","index":1,"caption_template":"图片 - _CAPTION_","label":"kubectl cheatsheet","attributes":{},"skip":false,"key":"4.3.2.1"},{"backlink":"guide/using-kubectl.html#fig4.3.2.2","level":"4.3.2","list_caption":"Figure: kube-shell页面","alt":"kube-shell页面","nro":34,"url":"../images/kube-shell.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"kube-shell页面","attributes":{},"skip":false,"key":"4.3.2.2"},{"backlink":"guide/ip-masq-agent.html#fig4.4.6.1","level":"4.4.6","list_caption":"Figure: IP伪装代理示意图","alt":"IP伪装代理示意图","nro":35,"url":"../images/ip-masq.png","index":1,"caption_template":"图片 - _CAPTION_","label":"IP伪装代理示意图","attributes":{},"skip":false,"key":"4.4.6.1"},{"backlink":"guide/auth-with-kubeconfig-or-token.html#fig4.4.7.1","level":"4.4.7","list_caption":"Figure: kubeconfig文件","alt":"kubeconfig文件","nro":36,"url":"../images/brand-kubeconfig-yaml.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"kubeconfig文件","attributes":{},"skip":false,"key":"4.4.7.1"},{"backlink":"guide/authentication.html#fig4.4.8.1","level":"4.4.8","list_caption":"Figure: Kubernetes OpenID Connect Flow","alt":"Kubernetes OpenID Connect Flow","nro":37,"url":"../images/kubernetes-oidc-login.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"Kubernetes OpenID Connect Flow","attributes":{},"skip":false,"key":"4.4.8.1"},{"backlink":"guide/carbin-mobile-dashboard-for-kubernetes.html#fig4.5.6.1","level":"4.5.6","list_caption":"Figure: App Store","alt":"App Store","nro":38,"url":"../images/carbin-kubernetes-mobile-dashboard-1.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"App Store","attributes":{},"skip":false,"key":"4.5.6.1"},{"backlink":"guide/carbin-mobile-dashboard-for-kubernetes.html#fig4.5.6.2","level":"4.5.6","list_caption":"Figure: 在手机上操作Kubernetes集群","alt":"在手机上操作Kubernetes集群","nro":39,"url":"../images/carbin-kubernetes-mobile-dashboard-4.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"在手机上操作Kubernetes集群","attributes":{},"skip":false,"key":"4.5.6.2"},{"backlink":"guide/kubernetes-desktop-client.html#fig4.5.7.1","level":"4.5.7","list_caption":"Figure: Kubernetic客户端","alt":"Kubernetic客户端","nro":40,"url":"../images/kubernetic-desktop-ui.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"Kubernetic客户端","attributes":{},"skip":false,"key":"4.5.7.1"},{"backlink":"guide/deploy-applications-in-kubernetes.html#fig4.6.1.1","level":"4.6.1","list_caption":"Figure: API","alt":"API","nro":41,"url":"../images/k8s-app-monitor-test-api-doc.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"API","attributes":{},"skip":false,"key":"4.6.1.1"},{"backlink":"guide/deploy-applications-in-kubernetes.html#fig4.6.1.2","level":"4.6.1","list_caption":"Figure: wercker","alt":"wercker","nro":42,"url":"../images/k8s-app-monitor-agent-wercker.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"wercker","attributes":{},"skip":false,"key":"4.6.1.2"},{"backlink":"guide/deploy-applications-in-kubernetes.html#fig4.6.1.3","level":"4.6.1","list_caption":"Figure: 图表","alt":"图表","nro":43,"url":"../images/k8s-app-monitor-agent.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"图表","attributes":{},"skip":false,"key":"4.6.1.3"},{"backlink":"guide/migrating-hadoop-yarn-to-kubernetes.html#fig4.6.2.1","level":"4.6.2","list_caption":"Figure: spark on yarn with kubernetes","alt":"spark on yarn with kubernetes","nro":44,"url":"../images/spark-on-yarn-with-kubernetes.png","index":1,"caption_template":"图片 - _CAPTION_","label":"spark on yarn with kubernetes","attributes":{},"skip":false,"key":"4.6.2.1"},{"backlink":"guide/migrating-hadoop-yarn-to-kubernetes.html#fig4.6.2.2","level":"4.6.2","list_caption":"Figure: Terms","alt":"Terms","nro":45,"url":"../images/terms-in-kubernetes-app-deployment.png","index":2,"caption_template":"图片 - _CAPTION_","label":"Terms","attributes":{},"skip":false,"key":"4.6.2.2"},{"backlink":"guide/migrating-hadoop-yarn-to-kubernetes.html#fig4.6.2.3","level":"4.6.2","list_caption":"Figure: 分解步骤解析","alt":"分解步骤解析","nro":46,"url":"../images/migrating-hadoop-yarn-to-kubernetes.png","index":3,"caption_template":"图片 - _CAPTION_","label":"分解步骤解析","attributes":{},"skip":false,"key":"4.6.2.3"},{"backlink":"practice/node-installation.html#fig5.2.7.1","level":"5.2.7","list_caption":"Figure: welcome nginx","alt":"welcome nginx","nro":47,"url":"../images/kubernetes-installation-test-nginx.png","index":1,"caption_template":"图片 - _CAPTION_","label":"welcome nginx","attributes":{},"skip":false,"key":"5.2.7.1"},{"backlink":"practice/dashboard-addon-installation.html#fig5.2.9.1","level":"5.2.9","list_caption":"Figure: kubernetes dashboard","alt":"kubernetes dashboard","nro":48,"url":"../images/kubernetes-dashboard-raw.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"kubernetes dashboard","attributes":{},"skip":false,"key":"5.2.9.1"},{"backlink":"practice/dashboard-addon-installation.html#fig5.2.9.2","level":"5.2.9","list_caption":"Figure: V1.6.3版本的dashboard界面","alt":"V1.6.3版本的dashboard界面","nro":49,"url":"../images/dashboard-v163.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"V1.6.3版本的dashboard界面","attributes":{},"skip":false,"key":"5.2.9.2"},{"backlink":"practice/heapster-addon-installation.html#fig5.2.10.1","level":"5.2.10","list_caption":"Figure: dashboard-heapster","alt":"dashboard-heapster","nro":50,"url":"../images/kubernetes-dashboard-with-heapster.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"dashboard-heapster","attributes":{},"skip":false,"key":"5.2.10.1"},{"backlink":"practice/heapster-addon-installation.html#fig5.2.10.2","level":"5.2.10","list_caption":"Figure: grafana","alt":"grafana","nro":51,"url":"../images/kubernetes-heapster-grafana.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"grafana","attributes":{},"skip":false,"key":"5.2.10.2"},{"backlink":"practice/heapster-addon-installation.html#fig5.2.10.3","level":"5.2.10","list_caption":"Figure: kubernetes-influxdb-heapster","alt":"kubernetes-influxdb-heapster","nro":52,"url":"../images/kubernetes-influxdb-heapster.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"kubernetes-influxdb-heapster","attributes":{},"skip":false,"key":"5.2.10.3"},{"backlink":"practice/heapster-addon-installation.html#fig5.2.10.4","level":"5.2.10","list_caption":"Figure: 修改grafana模板","alt":"修改grafana模板","nro":53,"url":"../images/grafana-dashboard-setting.jpg","index":4,"caption_template":"图片 - _CAPTION_","label":"修改grafana模板","attributes":{},"skip":false,"key":"5.2.10.4"},{"backlink":"practice/efk-addon-installation.html#fig5.2.11.1","level":"5.2.11","list_caption":"Figure: es-setting","alt":"es-setting","nro":54,"url":"../images/es-setting.png","index":1,"caption_template":"图片 - _CAPTION_","label":"es-setting","attributes":{},"skip":false,"key":"5.2.11.1"},{"backlink":"practice/efk-addon-installation.html#fig5.2.11.2","level":"5.2.11","list_caption":"Figure: es-home","alt":"es-home","nro":55,"url":"../images/kubernetes-efk-kibana.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"es-home","attributes":{},"skip":false,"key":"5.2.11.2"},{"backlink":"practice/traefik-ingress-installation.html#fig5.3.1.1","level":"5.3.1","list_caption":"Figure: kubernetes-dashboard","alt":"kubernetes-dashboard","nro":56,"url":"../images/traefik-dashboard.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"kubernetes-dashboard","attributes":{},"skip":false,"key":"5.3.1.1"},{"backlink":"practice/traefik-ingress-installation.html#fig5.3.1.2","level":"5.3.1","list_caption":"Figure: traefik-nginx","alt":"traefik-nginx","nro":57,"url":"../images/traefik-nginx.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"traefik-nginx","attributes":{},"skip":false,"key":"5.3.1.2"},{"backlink":"practice/traefik-ingress-installation.html#fig5.3.1.3","level":"5.3.1","list_caption":"Figure: traefik-guestbook","alt":"traefik-guestbook","nro":58,"url":"../images/traefik-guestbook.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"traefik-guestbook","attributes":{},"skip":false,"key":"5.3.1.3"},{"backlink":"practice/distributed-load-test.html#fig5.3.2.1","level":"5.3.2","list_caption":"Figure: 使用dashboard来扩容","alt":"使用dashboard来扩容","nro":59,"url":"../images/dashbaord-scale.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"使用dashboard来扩容","attributes":{},"skip":false,"key":"5.3.2.1"},{"backlink":"practice/distributed-load-test.html#fig5.3.2.2","level":"5.3.2","list_caption":"Figure: Traefik的UI","alt":"Traefik的UI","nro":60,"url":"../images/traefik-dashboard-locust.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"Traefik的UI","attributes":{},"skip":false,"key":"5.3.2.2"},{"backlink":"practice/distributed-load-test.html#fig5.3.2.3","level":"5.3.2","list_caption":"Figure: Locust启动界面","alt":"Locust启动界面","nro":61,"url":"../images/locust-start-swarming.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"Locust启动界面","attributes":{},"skip":false,"key":"5.3.2.3"},{"backlink":"practice/distributed-load-test.html#fig5.3.2.4","level":"5.3.2","list_caption":"Figure: Dashboard查看页面","alt":"Dashboard查看页面","nro":62,"url":"../images/sample-webapp-rc.jpg","index":4,"caption_template":"图片 - _CAPTION_","label":"Dashboard查看页面","attributes":{},"skip":false,"key":"5.3.2.4"},{"backlink":"practice/distributed-load-test.html#fig5.3.2.5","level":"5.3.2","list_caption":"Figure: Locust测试结果页面","alt":"Locust测试结果页面","nro":63,"url":"../images/locust-dashboard.jpg","index":5,"caption_template":"图片 - _CAPTION_","label":"Locust测试结果页面","attributes":{},"skip":false,"key":"5.3.2.5"},{"backlink":"practice/network-and-cluster-perfermance-test.html#fig5.3.3.1","level":"5.3.3","list_caption":"Figure: kubernetes-dashboard","alt":"kubernetes-dashboard","nro":64,"url":"../images/kubenetes-e2e-test.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"kubernetes-dashboard","attributes":{},"skip":false,"key":"5.3.3.1"},{"backlink":"practice/network-and-cluster-perfermance-test.html#fig5.3.3.2","level":"5.3.3","list_caption":"Figure: locust测试页面","alt":"locust测试页面","nro":65,"url":"../images/kubernetes-locust-test.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"locust测试页面","attributes":{},"skip":false,"key":"5.3.3.2"},{"backlink":"practice/edge-node-configuration.html#fig5.3.4.1","level":"5.3.4","list_caption":"Figure: 边缘节点架构","alt":"边缘节点架构","nro":66,"url":"../images/kubernetes-edge-node-architecture.png","index":1,"caption_template":"图片 - _CAPTION_","label":"边缘节点架构","attributes":{},"skip":false,"key":"5.3.4.1"},{"backlink":"practice/app-log-collection.html#fig5.4.2.1","level":"5.4.2","list_caption":"Figure: filebeat日志收集架构图","alt":"filebeat日志收集架构图","nro":67,"url":"../images/filebeat-log-collector.png","index":1,"caption_template":"图片 - _CAPTION_","label":"filebeat日志收集架构图","attributes":{},"skip":false,"key":"5.4.2.1"},{"backlink":"practice/app-log-collection.html#fig5.4.2.2","level":"5.4.2","list_caption":"Figure: Kibana页面","alt":"Kibana页面","nro":68,"url":"../images/filebeat-docker-test.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"Kibana页面","attributes":{},"skip":false,"key":"5.4.2.2"},{"backlink":"practice/app-log-collection.html#fig5.4.2.3","level":"5.4.2","list_caption":"Figure: filebeat收集的日志详细信息","alt":"filebeat收集的日志详细信息","nro":69,"url":"../images/kubernetes-filebeat-detail.png","index":3,"caption_template":"图片 - _CAPTION_","label":"filebeat收集的日志详细信息","attributes":{},"skip":false,"key":"5.4.2.3"},{"backlink":"practice/monitor.html#fig5.4.4.1","level":"5.4.4","list_caption":"Figure: Kubernetes集群中的监控","alt":"Kubernetes集群中的监控","nro":70,"url":"../images/monitoring-in-kubernetes.png","index":1,"caption_template":"图片 - _CAPTION_","label":"Kubernetes集群中的监控","attributes":{},"skip":false,"key":"5.4.4.1"},{"backlink":"practice/monitor.html#fig5.4.4.2","level":"5.4.4","list_caption":"Figure: kubernetes的容器命名规则示意图","alt":"kubernetes的容器命名规则示意图","nro":71,"url":"../images/kubernetes-container-naming-rule.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"kubernetes的容器命名规则示意图","attributes":{},"skip":false,"key":"5.4.4.2"},{"backlink":"practice/monitor.html#fig5.4.4.3","level":"5.4.4","list_caption":"Figure: Heapster架构图（改进版）","alt":"Heapster架构图（改进版）","nro":72,"url":"../images/kubernetes-heapster-monitoring.png","index":3,"caption_template":"图片 - _CAPTION_","label":"Heapster架构图（改进版）","attributes":{},"skip":false,"key":"5.4.4.3"},{"backlink":"practice/monitor.html#fig5.4.4.4","level":"5.4.4","list_caption":"Figure: 应用监控架构图","alt":"应用监控架构图","nro":73,"url":"../images/kubernetes-app-monitoring.png","index":4,"caption_template":"图片 - _CAPTION_","label":"应用监控架构图","attributes":{},"skip":false,"key":"5.4.4.4"},{"backlink":"practice/monitor.html#fig5.4.4.5","level":"5.4.4","list_caption":"Figure: 应用拓扑图","alt":"应用拓扑图","nro":74,"url":"../images/weave-scope-service-topology.jpg","index":5,"caption_template":"图片 - _CAPTION_","label":"应用拓扑图","attributes":{},"skip":false,"key":"5.4.4.5"},{"backlink":"practice/data-persistence-problem.html#fig5.4.5.1","level":"5.4.5","list_caption":"Figure: 日志持久化收集解决方案示意图","alt":"日志持久化收集解决方案示意图","nro":75,"url":"../images/log-persistence-logstash.png","index":1,"caption_template":"图片 - _CAPTION_","label":"日志持久化收集解决方案示意图","attributes":{},"skip":false,"key":"5.4.5.1"},{"backlink":"practice/storage-for-containers-using-glusterfs-with-openshift.html#fig5.5.1.2.1","level":"5.5.1.2","list_caption":"Figure: 创建存储","alt":"创建存储","nro":76,"url":"../images/create-gluster-storage.png","index":1,"caption_template":"图片 - _CAPTION_","label":"创建存储","attributes":{},"skip":false,"key":"5.5.1.2.1"},{"backlink":"practice/storage-for-containers-using-glusterfs-with-openshift.html#fig5.5.1.2.2","level":"5.5.1.2","list_caption":"Figure: Screen Shot 2017-03-24 at 11.09.34.png","alt":"Screen Shot 2017-03-24 at 11.09.34.png","nro":77,"url":"https://keithtenzer.files.wordpress.com/2017/03/screen-shot-2017-03-24-at-11-09-341.png?w=440","index":2,"caption_template":"图片 - _CAPTION_","label":"Screen Shot 2017-03-24 at 11.09.34.png","attributes":{},"skip":false,"key":"5.5.1.2.2"},{"backlink":"practice/using-heapster-to-get-object-metrics.html#fig5.6.1.1.1","level":"5.6.1.1","list_caption":"Figure: Heapster架构图","alt":"Heapster架构图","nro":78,"url":"../images/heapster-architecture.png","index":1,"caption_template":"图片 - _CAPTION_","label":"Heapster架构图","attributes":{},"skip":false,"key":"5.6.1.1.1"},{"backlink":"practice/prometheus.html#fig5.6.2.1","level":"5.6.2","list_caption":"Figure: Prometheus 架构图","alt":"Prometheus 架构图","nro":79,"url":"../images/prometheus-architecture.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"Prometheus 架构图","attributes":{},"skip":false,"key":"5.6.2.1"},{"backlink":"practice/using-prometheus-to-monitor-kuberentes-cluster.html#fig5.6.2.1.1","level":"5.6.2.1","list_caption":"Figure: Grafana页面","alt":"Grafana页面","nro":80,"url":"../images/kubernetes-prometheus-monitoring.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"Grafana页面","attributes":{},"skip":false,"key":"5.6.2.1.1"},{"backlink":"practice/helm.html#fig5.7.1.1","level":"5.7.1","list_caption":"Figure: Helm chart源","alt":"Helm chart源","nro":81,"url":"../images/helm-charts-repository.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"Helm chart源","attributes":{},"skip":false,"key":"5.7.1.1"},{"backlink":"practice/helm.html#fig5.7.1.2","level":"5.7.1","list_caption":"Figure: TODO应用的Web页面","alt":"TODO应用的Web页面","nro":82,"url":"../images/helm-mean-todo-aholic.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"TODO应用的Web页面","attributes":{},"skip":false,"key":"5.7.1.2"},{"backlink":"practice/create-private-charts-repo.html#fig5.7.2.1","level":"5.7.2","list_caption":"Figure: Helm monocular界面","alt":"Helm monocular界面","nro":83,"url":"../images/helm-monocular-jimmysong.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"Helm monocular界面","attributes":{},"skip":false,"key":"5.7.2.1"},{"backlink":"practice/jenkins-ci-cd.html#fig5.8.1.1","level":"5.8.1","list_caption":"Figure: 基于Jenkins的持续集成与发布","alt":"基于Jenkins的持续集成与发布","nro":84,"url":"../images/kubernetes-jenkins-ci-cd.png","index":1,"caption_template":"图片 - _CAPTION_","label":"基于Jenkins的持续集成与发布","attributes":{},"skip":false,"key":"5.8.1.1"},{"backlink":"practice/drone-ci-cd.html#fig5.8.2.1","level":"5.8.2","list_caption":"Figure: OAuth注册","alt":"OAuth注册","nro":85,"url":"../images/github-oauth-register.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"OAuth注册","attributes":{},"skip":false,"key":"5.8.2.1"},{"backlink":"practice/drone-ci-cd.html#fig5.8.2.2","level":"5.8.2","list_caption":"Figure: OAuth key","alt":"OAuth key","nro":86,"url":"../images/github-oauth-drone-key.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"OAuth key","attributes":{},"skip":false,"key":"5.8.2.2"},{"backlink":"practice/drone-ci-cd.html#fig5.8.2.3","level":"5.8.2","list_caption":"Figure: Drone登陆界面","alt":"Drone登陆界面","nro":87,"url":"../images/drone-login-github.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"Drone登陆界面","attributes":{},"skip":false,"key":"5.8.2.3"},{"backlink":"practice/drone-ci-cd.html#fig5.8.2.4","level":"5.8.2","list_caption":"Figure: Github启用repo设置","alt":"Github启用repo设置","nro":88,"url":"../images/drone-github-active.jpg","index":4,"caption_template":"图片 - _CAPTION_","label":"Github启用repo设置","attributes":{},"skip":false,"key":"5.8.2.4"},{"backlink":"practice/drone-ci-cd.html#fig5.8.2.5","level":"5.8.2","list_caption":"Figure: Github单个repo设置","alt":"Github单个repo设置","nro":89,"url":"../images/drone-github-repo-setting.jpg","index":5,"caption_template":"图片 - _CAPTION_","label":"Github单个repo设置","attributes":{},"skip":false,"key":"5.8.2.5"},{"backlink":"practice/dashboard-upgrade.html#fig5.9.2.1","level":"5.9.2","list_caption":"Figure: 登陆界面","alt":"登陆界面","nro":90,"url":"../images/kubernetes-dashboard-1.7.1-login.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"登陆界面","attributes":{},"skip":false,"key":"5.9.2.1"},{"backlink":"practice/dashboard-upgrade.html#fig5.9.2.2","level":"5.9.2","list_caption":"Figure: 首页","alt":"首页","nro":91,"url":"../images/kubernetes-dashboard-1.7.1-default-page.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"首页","attributes":{},"skip":false,"key":"5.9.2.2"},{"backlink":"practice/dashboard-upgrade.html#fig5.9.2.3","level":"5.9.2","list_caption":"Figure: 用户空间","alt":"用户空间","nro":92,"url":"../images/kubernetes-dashboard-1.7.1-brand.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"用户空间","attributes":{},"skip":false,"key":"5.9.2.3"},{"backlink":"practice/dashboard-upgrade.html#fig5.9.2.4","level":"5.9.2","list_caption":"Figure: kubeconfig文件","alt":"kubeconfig文件","nro":93,"url":"../images/brand-kubeconfig-yaml.jpg","index":4,"caption_template":"图片 - _CAPTION_","label":"kubeconfig文件","attributes":{},"skip":false,"key":"5.9.2.4"},{"backlink":"usecases/microservices.html#fig6.2.1","level":"6.2","list_caption":"Figure: 微服务关注的部分","alt":"微服务关注的部分","nro":94,"url":"../images/microservices-concerns.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"微服务关注的部分","attributes":{},"skip":false,"key":"6.2.1"},{"backlink":"usecases/service-discovery-in-microservices.html#fig6.2.1.1","level":"6.2.1","list_caption":"Figure: 微服务中的服务发现","alt":"微服务中的服务发现","nro":95,"url":"../images/service-discovery-in-microservices.png","index":1,"caption_template":"图片 - _CAPTION_","label":"微服务中的服务发现","attributes":{},"skip":false,"key":"6.2.1.1"},{"backlink":"usecases/service-mesh.html#fig6.3.1","level":"6.3","list_caption":"Figure: 下一代异构微服务架构","alt":"下一代异构微服务架构","nro":96,"url":"../images/polyglot-microservices-serivce-mesh.png","index":1,"caption_template":"图片 - _CAPTION_","label":"下一代异构微服务架构","attributes":{},"skip":false,"key":"6.3.1"},{"backlink":"usecases/service-mesh.html#fig6.3.2","level":"6.3","list_caption":"Figure: Service Mesh 架构图","alt":"Service Mesh 架构图","nro":97,"url":"../images/serivce-mesh-control-plane.png","index":2,"caption_template":"图片 - _CAPTION_","label":"Service Mesh 架构图","attributes":{},"skip":false,"key":"6.3.2"},{"backlink":"usecases/service-mesh.html#fig6.3.3","level":"6.3","list_caption":"Figure: Istio vs linkerd","alt":"Istio vs linkerd","nro":98,"url":"../images/istio-vs-linkerd.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"Istio vs linkerd","attributes":{},"skip":false,"key":"6.3.3"},{"backlink":"usecases/istio.html#fig6.3.1.1","level":"6.3.1","list_caption":"Figure: Istio架构图","alt":"Istio架构图","nro":99,"url":"../images/istio-arch.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"Istio架构图","attributes":{},"skip":false,"key":"6.3.1.1"},{"backlink":"usecases/istio-installation.html#fig6.3.1.1.1","level":"6.3.1.1","list_caption":"Figure: BookInfo Sample应用架构图","alt":"BookInfo Sample应用架构图","nro":100,"url":"../images/bookinfo-sample-arch.png","index":1,"caption_template":"图片 - _CAPTION_","label":"BookInfo Sample应用架构图","attributes":{},"skip":false,"key":"6.3.1.1.1"},{"backlink":"usecases/istio-installation.html#fig6.3.1.1.2","level":"6.3.1.1","list_caption":"Figure: BookInfo Sample页面","alt":"BookInfo Sample页面","nro":101,"url":"../images/bookinfo-sample.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"BookInfo Sample页面","attributes":{},"skip":false,"key":"6.3.1.1.2"},{"backlink":"usecases/istio-installation.html#fig6.3.1.1.3","level":"6.3.1.1","list_caption":"Figure: Istio Grafana界面","alt":"Istio Grafana界面","nro":102,"url":"../images/istio-grafana.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"Istio Grafana界面","attributes":{},"skip":false,"key":"6.3.1.1.3"},{"backlink":"usecases/istio-installation.html#fig6.3.1.1.4","level":"6.3.1.1","list_caption":"Figure: Prometheus页面","alt":"Prometheus页面","nro":103,"url":"../images/istio-prometheus.jpg","index":4,"caption_template":"图片 - _CAPTION_","label":"Prometheus页面","attributes":{},"skip":false,"key":"6.3.1.1.4"},{"backlink":"usecases/istio-installation.html#fig6.3.1.1.5","level":"6.3.1.1","list_caption":"Figure: Zipkin页面","alt":"Zipkin页面","nro":104,"url":"../images/istio-zipkin.jpg","index":5,"caption_template":"图片 - _CAPTION_","label":"Zipkin页面","attributes":{},"skip":false,"key":"6.3.1.1.5"},{"backlink":"usecases/istio-installation.html#fig6.3.1.1.6","level":"6.3.1.1","list_caption":"Figure: ServiceGraph页面","alt":"ServiceGraph页面","nro":105,"url":"../images/istio-servicegraph.jpg","index":6,"caption_template":"图片 - _CAPTION_","label":"ServiceGraph页面","attributes":{},"skip":false,"key":"6.3.1.1.6"},{"backlink":"usecases/install-and-expand-istio-mesh.html#fig6.3.1.3.1","level":"6.3.1.3","list_caption":"Figure: BookInfo","alt":"BookInfo","nro":106,"url":"../images/noistio.png","index":1,"caption_template":"图片 - _CAPTION_","label":"BookInfo","attributes":{},"skip":false,"key":"6.3.1.3.1"},{"backlink":"usecases/install-and-expand-istio-mesh.html#fig6.3.1.3.2","level":"6.3.1.3","list_caption":"Figure: BookInfo","alt":"BookInfo","nro":107,"url":"../images/noistio.png","index":2,"caption_template":"图片 - _CAPTION_","label":"BookInfo","attributes":{},"skip":false,"key":"6.3.1.3.2"},{"backlink":"usecases/integrating-vms.html#fig6.3.1.4.1","level":"6.3.1.4","list_caption":"Figure: Bookinfo应用的拓展Mesh","alt":"Bookinfo应用的拓展Mesh","nro":108,"url":"../images/istio-mesh-expansion.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"Bookinfo应用的拓展Mesh","attributes":{},"skip":false,"key":"6.3.1.4.1"},{"backlink":"usecases/linkerd.html#fig6.3.2.1","level":"6.3.2","list_caption":"Figure: source https://linkerd.io","alt":"source https://linkerd.io","nro":109,"url":"../images/diagram-individual-instance.png","index":1,"caption_template":"图片 - _CAPTION_","label":"source https://linkerd.io","attributes":{},"skip":false,"key":"6.3.2.1"},{"backlink":"usecases/linkerd-user-guide.html#fig6.3.2.1.1","level":"6.3.2.1","list_caption":"Figure: Jenkins pipeline","alt":"Jenkins pipeline","nro":110,"url":"../images/linkerd-jenkins-pipeline.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"Jenkins pipeline","attributes":{},"skip":false,"key":"6.3.2.1.1"},{"backlink":"usecases/linkerd-user-guide.html#fig6.3.2.1.2","level":"6.3.2.1","list_caption":"Figure: Jenkins config","alt":"Jenkins config","nro":111,"url":"../images/linkerd-jenkins.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"Jenkins config","attributes":{},"skip":false,"key":"6.3.2.1.2"},{"backlink":"usecases/linkerd-user-guide.html#fig6.3.2.1.3","level":"6.3.2.1","list_caption":"Figure: namerd","alt":"namerd","nro":112,"url":"../images/namerd-internal.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"namerd","attributes":{},"skip":false,"key":"6.3.2.1.3"},{"backlink":"usecases/linkerd-user-guide.html#fig6.3.2.1.4","level":"6.3.2.1","list_caption":"Figure: linkerd监控","alt":"linkerd监控","nro":113,"url":"../images/linkerd-helloworld-outgoing.jpg","index":4,"caption_template":"图片 - _CAPTION_","label":"linkerd监控","attributes":{},"skip":false,"key":"6.3.2.1.4"},{"backlink":"usecases/linkerd-user-guide.html#fig6.3.2.1.5","level":"6.3.2.1","list_caption":"Figure: linkerd监控","alt":"linkerd监控","nro":114,"url":"../images/linkerd-helloworld-incoming.jpg","index":5,"caption_template":"图片 - _CAPTION_","label":"linkerd监控","attributes":{},"skip":false,"key":"6.3.2.1.5"},{"backlink":"usecases/linkerd-user-guide.html#fig6.3.2.1.6","level":"6.3.2.1","list_caption":"Figure: linkerd性能监控","alt":"linkerd性能监控","nro":115,"url":"../images/linkerd-grafana.png","index":6,"caption_template":"图片 - _CAPTION_","label":"linkerd性能监控","attributes":{},"skip":false,"key":"6.3.2.1.6"},{"backlink":"usecases/linkerd-user-guide.html#fig6.3.2.1.7","level":"6.3.2.1","list_caption":"Figure: Linkerd ingress controller","alt":"Linkerd ingress controller","nro":116,"url":"../images/linkerd-ingress-controller.jpg","index":7,"caption_template":"图片 - _CAPTION_","label":"Linkerd ingress controller","attributes":{},"skip":false,"key":"6.3.2.1.7"},{"backlink":"usecases/conduit-installation.html#fig6.3.3.2.1","level":"6.3.3.2","list_caption":"Figure: Conduit dashboard","alt":"Conduit dashboard","nro":117,"url":"../images/conduit-dashboard.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"Conduit dashboard","attributes":{},"skip":false,"key":"6.3.3.2.1"},{"backlink":"usecases/big-data.html#fig6.4.1","level":"6.4","list_caption":"Figure: Spark on yarn with kubernetes","alt":"Spark on yarn with kubernetes","nro":118,"url":"../images/spark-on-yarn-with-kubernetes.png","index":1,"caption_template":"图片 - _CAPTION_","label":"Spark on yarn with kubernetes","attributes":{},"skip":false,"key":"6.4.1"},{"backlink":"usecases/big-data.html#fig6.4.2","level":"6.4","list_caption":"Figure: 在kubernetes上使用多种调度方式","alt":"在kubernetes上使用多种调度方式","nro":119,"url":"../images/spark-on-kubernetes-with-different-schedulers.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"在kubernetes上使用多种调度方式","attributes":{},"skip":false,"key":"6.4.2"},{"backlink":"usecases/spark-standalone-on-kubernetes.html#fig6.4.1.1","level":"6.4.1","list_caption":"Figure: spark master ui","alt":"spark master ui","nro":120,"url":"../images/spark-ui.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"spark master ui","attributes":{},"skip":false,"key":"6.4.1.1"},{"backlink":"usecases/spark-standalone-on-kubernetes.html#fig6.4.1.2","level":"6.4.1","list_caption":"Figure: zeppelin ui","alt":"zeppelin ui","nro":121,"url":"../images/zeppelin-ui.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"zeppelin ui","attributes":{},"skip":false,"key":"6.4.1.2"},{"backlink":"usecases/serverless.html#fig6.5.1","level":"6.5","list_caption":"Figure: 从物理机到函数计算","alt":"从物理机到函数计算","nro":122,"url":"../images/from-bare-metal-to-functions.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"从物理机到函数计算","attributes":{},"skip":false,"key":"6.5.1"},{"backlink":"usecases/serverless.html#fig6.5.2","level":"6.5","list_caption":"Figure: FaaS Landscape","alt":"FaaS Landscape","nro":123,"url":"../images/redpoint-faas-landscape.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"FaaS Landscape","attributes":{},"skip":false,"key":"6.5.2"},{"backlink":"usecases/understanding-serverless.html#fig6.5.1.1","level":"6.5.1","list_caption":"Figure: 服务端软件的运行环境","alt":"服务端软件的运行环境","nro":124,"url":"../images/serverless-server-side-software.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"服务端软件的运行环境","attributes":{},"skip":false,"key":"6.5.1.1"},{"backlink":"usecases/understanding-serverless.html#fig6.5.1.2","level":"6.5.1","list_caption":"Figure: FaaS应用架构","alt":"FaaS应用架构","nro":125,"url":"../images/serverless-faas-platform.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"FaaS应用架构","attributes":{},"skip":false,"key":"6.5.1.2"},{"backlink":"usecases/understanding-serverless.html#fig6.5.1.3","level":"6.5.1","list_caption":"Figure: 传统应用程序架构","alt":"传统应用程序架构","nro":126,"url":"../images/non-serverless-game-arch.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"传统应用程序架构","attributes":{},"skip":false,"key":"6.5.1.3"},{"backlink":"usecases/understanding-serverless.html#fig6.5.1.4","level":"6.5.1","list_caption":"Figure: Serverless架构","alt":"Serverless架构","nro":127,"url":"../images/serverless-game-arch.jpg","index":4,"caption_template":"图片 - _CAPTION_","label":"Serverless架构","attributes":{},"skip":false,"key":"6.5.1.4"},{"backlink":"usecases/openfaas.html#fig6.5.2.1.1","level":"6.5.2.1","list_caption":"Figure: OpenFaaS架构","alt":"OpenFaaS架构","nro":128,"url":"../images/openfaas-arch.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"OpenFaaS架构","attributes":{},"skip":false,"key":"6.5.2.1.1"},{"backlink":"usecases/openfaas.html#fig6.5.2.1.2","level":"6.5.2.1","list_caption":"Figure: OpenFaaS Prometheus","alt":"OpenFaaS Prometheus","nro":129,"url":"../images/openfaas-prometheus.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"OpenFaaS Prometheus","attributes":{},"skip":false,"key":"6.5.2.1.2"},{"backlink":"usecases/openfaas.html#fig6.5.2.1.3","level":"6.5.2.1","list_caption":"Figure: OpenFaas Grafana监控","alt":"OpenFaas Grafana监控","nro":130,"url":"../images/openfaas-grafana.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"OpenFaas Grafana监控","attributes":{},"skip":false,"key":"6.5.2.1.3"},{"backlink":"usecases/openfaas.html#fig6.5.2.1.4","level":"6.5.2.1","list_caption":"Figure: OpenFaas Dashboard","alt":"OpenFaas Dashboard","nro":131,"url":"../images/openfaas-deploy-a-function.jpg","index":4,"caption_template":"图片 - _CAPTION_","label":"OpenFaas Dashboard","attributes":{},"skip":false,"key":"6.5.2.1.4"},{"backlink":"usecases/openfaas.html#fig6.5.2.1.5","level":"6.5.2.1","list_caption":"Figure: NodeInfo执行结果","alt":"NodeInfo执行结果","nro":132,"url":"../images/openfaas-nodeinfo.jpg","index":5,"caption_template":"图片 - _CAPTION_","label":"NodeInfo执行结果","attributes":{},"skip":false,"key":"6.5.2.1.5"},{"backlink":"develop/sigs-and-working-group.html#fig7.2.1","level":"7.2","list_caption":"Figure: Kubernetes SIG","alt":"Kubernetes SIG","nro":133,"url":"../images/kubernetes-sigs.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"Kubernetes SIG","attributes":{},"skip":false,"key":"7.2.1"},{"backlink":"develop/client-go-sample.html#fig7.5.1","level":"7.5","list_caption":"Figure: 使用kubernetes dashboard进行故障排查","alt":"使用kubernetes dashboard进行故障排查","nro":134,"url":"../images/kubernetes-client-go-sample-update.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"使用kubernetes dashboard进行故障排查","attributes":{},"skip":false,"key":"7.5.1"},{"backlink":"appendix/issues.html#fig8.6.1","level":"8.6","list_caption":"Figure: pvc-storage-limit","alt":"pvc-storage-limit","nro":135,"url":"../images/pvc-storage-limit.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"pvc-storage-limit","attributes":{},"skip":false,"key":"8.6.1"}]},"title":"Kubernetes Handbook - jimmysong.io","language":"zh-hans","links":{"sidebar":{"Jimmy Song":"https://jimmysong.io"}},"gitbook":"*","description":"Kubernetes中文指南/实践手册|kubernetes handbook|jimmysong.io|宋净超"},"file":{"path":"practice/master-installation.md","mtime":"2017-12-15T07:17:07.278Z","type":"markdown"},"gitbook":{"version":"3.2.2","time":"2017-12-20T10:50:52.717Z"},"basePath":"..","book":{"language":""}}); }); diff --git a/practice/monitor.html b/practice/monitor.html index be9fd9e8c..41ad8e513 100644 --- a/practice/monitor.html +++ b/practice/monitor.html @@ -2958,7 +2958,7 @@ diff --git a/practice/monitoring.html b/practice/monitoring.html index d8eed06fb..234cb88e3 100644 --- a/practice/monitoring.html +++ b/practice/monitoring.html @@ -2783,7 +2783,7 @@ diff --git a/practice/network-and-cluster-perfermance-test.html b/practice/network-and-cluster-perfermance-test.html index a9e3ebf95..4c3e10394 100644 --- a/practice/network-and-cluster-perfermance-test.html +++ b/practice/network-and-cluster-perfermance-test.html @@ -3403,7 +3403,7 @@ Test Suite Passed diff --git a/practice/nginx-ingress-installation.html b/practice/nginx-ingress-installation.html index 9a5c6601f..5f6a8715e 100644 --- a/practice/nginx-ingress-installation.html +++ b/practice/nginx-ingress-installation.html @@ -2929,7 +2929,7 @@ curl -v --insecure http://172.20.0.113:30484/ diff --git a/practice/node-installation.html b/practice/node-installation.html index 55ddb1ba7..0a78788e5 100644 --- a/practice/node-installation.html +++ b/practice/node-installation.html @@ -3076,7 +3076,7 @@ Commercial support is available at diff --git a/practice/operation.html b/practice/operation.html index 9bcb515a8..a010c4392 100644 --- a/practice/operation.html +++ b/practice/operation.html @@ -2785,7 +2785,7 @@ diff --git a/practice/prometheus.html b/practice/prometheus.html index c42a4d684..4cf524162 100644 --- a/practice/prometheus.html +++ b/practice/prometheus.html @@ -2815,7 +2815,7 @@ diff --git a/practice/service-discovery-and-loadbalancing.html b/practice/service-discovery-and-loadbalancing.html index f20e94815..66277f6ec 100644 --- a/practice/service-discovery-and-loadbalancing.html +++ b/practice/service-discovery-and-loadbalancing.html @@ -2887,7 +2887,7 @@ bar.foo.com --| |-> bar.foo.com s2:80 diff --git a/practice/service-rolling-update.html b/practice/service-rolling-update.html index 130e91197..958cbf475 100644 --- a/practice/service-rolling-update.html +++ b/practice/service-rolling-update.html @@ -2942,7 +2942,7 @@ replicationcontroller "zeppelin-controller" rolling updated diff --git a/practice/services-management-tool.html b/practice/services-management-tool.html index 95f7ebc83..ff14ed244 100644 --- a/practice/services-management-tool.html +++ b/practice/services-management-tool.html @@ -2780,7 +2780,7 @@ diff --git a/practice/storage-for-containers-using-glusterfs-with-openshift.html b/practice/storage-for-containers-using-glusterfs-with-openshift.html index 4c3fdd981..4b9336068 100644 --- a/practice/storage-for-containers-using-glusterfs-with-openshift.html +++ b/practice/storage-for-containers-using-glusterfs-with-openshift.html @@ -3279,7 +3279,7 @@ drwxr-s---. 2 1000070000 2001 8192 Mar 24 12:12 diff --git a/practice/storage.html b/practice/storage.html index 13f0e4cf7..7588e7962 100644 --- a/practice/storage.html +++ b/practice/storage.html @@ -2778,7 +2778,7 @@ diff --git a/practice/traefik-ingress-installation.html b/practice/traefik-ingress-installation.html index 4e53f1b9a..9790b95d4 100644 --- a/practice/traefik-ingress-installation.html +++ b/practice/traefik-ingress-installation.html @@ -2956,7 +2956,7 @@ Commercial support is available at diff --git a/practice/update-and-upgrade.html b/practice/update-and-upgrade.html index 4a3d3400c..3df39800b 100644 --- a/practice/update-and-upgrade.html +++ b/practice/update-and-upgrade.html @@ -2779,7 +2779,7 @@ diff --git a/practice/using-ceph-for-persistent-storage.html b/practice/using-ceph-for-persistent-storage.html index 14bb0b036..d0066345e 100644 --- a/practice/using-ceph-for-persistent-storage.html +++ b/practice/using-ceph-for-persistent-storage.html @@ -3106,7 +3106,7 @@ Sep 4 15:25:36 bj-xg-oam-kubernetes-001 kube-controller-manager: E0904 15:25:36 diff --git a/practice/using-glusterfs-for-persistent-storage.html b/practice/using-glusterfs-for-persistent-storage.html index 2a4e7b16d..b60f143b9 100644 --- a/practice/using-glusterfs-for-persistent-storage.html +++ b/practice/using-glusterfs-for-persistent-storage.html @@ -3080,7 +3080,7 @@ index.html diff --git a/practice/using-heapster-to-get-object-metrics.html b/practice/using-heapster-to-get-object-metrics.html index 089013ab8..2754e75f5 100644 --- a/practice/using-heapster-to-get-object-metrics.html +++ b/practice/using-heapster-to-get-object-metrics.html @@ -3025,7 +3025,7 @@ Heapster is running at https://172.20.0.113:6443/api/v1/proxy/namespaces/kube-sy diff --git a/practice/using-prometheus-to-monitor-kuberentes-cluster.html b/practice/using-prometheus-to-monitor-kuberentes-cluster.html index 348375b34..0be276caf 100644 --- a/practice/using-prometheus-to-monitor-kuberentes-cluster.html +++ b/practice/using-prometheus-to-monitor-kuberentes-cluster.html @@ -2878,7 +2878,7 @@ kubectl create clusterrolebinding prometheus --clusterrole=cluster-admin --servi diff --git a/search_plus_index.json b/search_plus_index.json index eb4c2daab..928da583e 100644 --- a/search_plus_index.json +++ b/search_plus_index.json @@ -1 +1 @@ -{"./":{"url":"./","title":"序言","keywords":"","body":"Kubernetes Handbook Kubernetes是Google基于Borg开源的容器编排调度引擎，作为CNCF（Cloud Native Computing Foundation）最重要的组件之一，它的目标不仅仅是一个编排系统，而是提供一个规范，可以让你来描述集群的架构，定义服务的最终状态，kubernetes可以帮你将系统自动得达到和维持在这个状态。Kubernetes作为云原生应用的基石，相当于一个云操作系统，其重要性不言而喻。 本书记录了本人从零开始学习和使用Kubernetes的心路历程，着重于经验分享和总结，同时也会有相关的概念解析，希望能够帮助大家少踩坑，少走弯路，还会指引大家关于关注kubernetes生态周边，如微服务构建、DevOps、大数据应用、Service Mesh、Cloud Native等领域。 起初写作本书时，安装的所有组件、所用示例和操作等皆基于Kubernetes1.6.0 版本，同时我们也将密切关注kubernetes的版本更新，随着它的版本更新升级，本书中的kubernetes版本和示例也将随之更新。 GitHub 地址：https://github.com/rootsongjc/kubernetes-handbook Gitbook 在线浏览：https://jimmysong.io/kubernetes-handbook/ 如何使用本书 在线浏览 访问 https://jimmysong.io/kubernetes-handbook/ 注意：文中涉及的配置文件和代码链接在网页中将无法访问，请下载GitHub源码后，在Markdown编辑器中打开，点击链接将跳转到你的本地目录，推荐使用typora，或者直接登录github 查看。 本地查看 将代码克隆到本地 安装 gitbook：Setup and Installation of GitBook 执行 gitbook serve 在浏览器中访问 http://localhost:4000 生成的文档在 _book 目录下 下载 PDF/ePub/Mobi 格式文档本地查看 访问 gitbook 可以看到下载地址，可以下载根据最新文档生成的 PDF/ePub/Mobi 格式文档（文档的注脚中注明了更新时间），同时也可以直接在 gitbook 中阅读，不过 gitbook 不太稳定打开速度较慢，建议大家直接在 https://jimmysong.io/kubernetes-handbook/ 浏览。 生成 pdf 下载Calibre On Mac 在Mac下安装后，使用该命令创建链接 ln -s /Applications/calibre.app/Contents/MacOS/ebook-convert /usr/local/bin 在该项目目录下执行以下命令生成kubernetes-handbook.pdf文档。 gitbook pdf . ./kubernetes-handbook.pdf On Windows 需要用到的工具：calibre，phantomjs 将上述2个安装，calibre 默认安装的路径 C:\\Program Files\\Calibre2 为你解压路径； 并将其目录均加入到系统变量 path 中,参考:目录添加到系统变量 path 中； 在 cmd 打开你需要转 pdf 的文件夹,输入gitbook pdf即可； 生成单个章节的pdf 使用pandoc和latex来生成pdf格式文档。 pandoc --latex-engine=xelatex --template=pm-template input.md -o output.pdf 如何贡献 提 issue 如果你发现文档中的错误，或者有好的建议、不明白的问题、不要犹豫，欢迎提交issue。 发起 Pull Request 当你发现文章中明确的错误或者逻辑问题，在你自己的fork的分支中，创建一个新的branch，修改错误，push 到你的branch，然后在提交issue 后直接发起Pull Request。 贡献文档 文档的组织规则 如果要创建一个大的主题就在最顶层创建一个目录； 全书五大主题，每个主题一个目录，其下不再设二级目录； 所有的图片都放在最顶层的 images 目录下，原则上文章中用到的图片都保存在本地； 所有的文档的文件名使用英文命名，可以包含数字和中划线； etc、manifests目录专门用来保存配置文件和文档中用到的其他相关文件； 添加文档 在该文章相关主题的目录下创建文档； 在 SUMMARY.md 中在相应的章节下添加文章链接； 执行 gitbook serve 测试是否报错，访问 http://localhost:4000 查看该文档是否出现在相应主题的目录下； 提交PR 关于 贡献者列表 加入K8S🤘Cloud Native实战群 扫描二维码，Jimmy Song，或直接搜索微信号jimmysong后拉您入群，请增加备注。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-15 19:49:07 "},"cloud-native/kubernetes-and-cloud-native-app-overview.html":{"url":"cloud-native/kubernetes-and-cloud-native-app-overview.html","title":"Kubernetes与云原生应用概览","keywords":"","body":"Kubernetes与云原生应用概览 几个月前Mesos已经宣布支持kubernetes，而在2017年10月份的DockerCon EU上Docker公司宣布同时官方支持Swarm和Kubernetes容器编排，kubernetes已然成为容器编排调度的标准。 作为全书的开头，首先从历史、生态和应用角度介绍一下kubernetes与云原生应用，深入浅出，高屋建瓴，没有深入到具体细节，主要是为了给初次接触kubernetes的小白扫盲，具体细节请参考链接。 从云计算到微服务再到云原生计算 下面将从云计算的发展历程引入云原生计算，请先看下图： 图片 - 云计算演进历程 云原生应用到2020年将比目前至少翻一番，下图是Marc Wilczek的调查报告。 图片 - 来自Twitter @MarcWilczek 云计算介绍 云计算包含的内容十分繁杂，也有很多技术和公司牵强附会说自己是云计算公司，说自己是做云的，实际上可能风马牛不相及。说白了，云计算就是一种配置资源的方式，根据资源配置方式的不同我们可以把云计算从宏观上分为以下三种类型： IaaS：这是为了想要建立自己的商业模式并进行自定义的客户，例如亚马逊的EC2、S3存储、Rackspace虚拟机等都是IaaS。 PaaS：工具和服务的集合，对于想用它来构建自己的应用程序或者想快速得将应用程序部署到生产环境而不必关心底层硬件的用户和开发者来说是特别有用的，比如Cloud Foundry、Google App Engine、Heroku等。 SaaS：终端用户可以直接使用的应用程序。这个就太多，我们生活中用到的很多软件都是SaaS服务，只要基于互联网来提供的服务基本都是SaaS服务，有的服务是免费的，比如Google Docs，还有更多的是根据我们购买的Plan和使用量付费，比如GitHub、各种云存储。 微服务介绍 微服务（Microservices）这个词比较新颖，但是其实这种架构设计理念早就有了。微服务是一种分布式架构设计理念，为了推动细粒度服务的使用，这些服务要能协同工作，每个服务都有自己的生命周期。一个微服务就是一个独立的实体，可以独立的部署在PAAS平台上，也可以作为一个独立的进程在主机中运行。服务之间通过API访问，修改一个服务不会影响其它服务。 要想了解微服务的详细内容推荐阅读《微服务设计》（Sam Newman著），我写过这本书的读书笔记 - 微服务设计读书笔记。 下文中会谈到kubernetes与微服务的关系，其中kubernetes的service天生就适合与微服务。 云原生概念介绍 下面是Cloud Native概念思维导图 图片 - Cloud native思维导图 云原生准确来说是一种文化，更是一种潮流，它是云计算的一个必然导向。它的意义在于让云成为云化战略成功的基石，而不是阻碍，如果业务应用上云之后开发和运维人员比原先还痛苦，成本高企的话，这样的云我们宁愿不不上。 自从云的概念开始普及，许多公司都部署了实施云化的策略，纷纷搭建起云平台，希望完成传统应用到云端的迁移。但是这个过程中会遇到一些技术难题，上云以后，效率并没有变得奇高，故障也没有迅速定位。 为了解决传统应用升级缓慢、架构臃肿、不能快速迭代、故障不能快速定位、问题无法快速解决等问题，云原生这一概念横空出世。云原生可以改进应用开发的效率，改变企业的组织结构，甚至会在文化层面上直接影响一个公司的决策。 另外，云原生也很好地解释了云上运行的应用应该具备什么样的架构特性——敏捷性、可扩展性、故障可恢复性。 综上所述，云原生应用应该具备以下几个关键词： 敏捷 可靠 高弹性 易扩展 故障隔离保护 不中断业务持续更新 以上特性也是云原生区别于传统云应用的优势特点。 从宏观概念上讲，云原生是不同思想的集合，集目前各种热门技术之大成，具体包括如下图所示的几个部分。 Kubernetes与云原生的关系 Kuberentes可以说是乘着docker和微服务的东风，已经推出便迅速蹿红，它的很多设计思想都契合了微服务和云原生应用的设计法则，这其中最著名的就是开发了Heroku PaaS平台的工程师们总结的 Twelve-factor App了。 下面我将讲解Kubernetes设计时是如何按照了十二因素应用法则，并给出kubernetes中的应用示例，并附上一句话简短的介绍。 Kubernetes介绍 Kubernetes是Google基于Borg开源的容器编排调度引擎，作为CNCF（Cloud Native Computing Foundation）最重要的组件之一，它的目标不仅仅是一个编排系统，而是提供一个规范，可以让你来描述集群的架构，定义服务的最终状态，kubernetes可以帮你将系统自动得达到和维持在这个状态。 更直白的说，Kubernetes用户可以通过编写一个yaml或者json格式的配置文件，也可以通过工具/代码生成或直接请求kubernetes API创建应用，该配置文件中包含了用户想要应用程序保持的状态，不论整个kubernetes集群中的个别主机发生什么问题，都不会影响应用程序的状态，你还可以通过改变该配置文件或请求kubernetes API来改变应用程序的状态。 12因素应用 12因素应用提出已经有几年的时间了，每个人对其可能都有自己的理解，切不可生搬硬套，也不一定所有云原生应用都必须符合这12条法则，其中有几条法则可能还有点争议，有人对其的解释和看法不同。 大家不要孤立的来看这每一个因素，将其与自己软件开发流程联系起来，这12个因素大致就是按照软件从开发到交付的流程顺序来写的。 图片 - 十二因素应用 1.基准代码 每个代码仓库（repo）都生成docker image保存到镜像仓库中，并使用唯一的ID管理，在Jenkins中使用编译时的ID。 2.依赖 显式得声明代码中的依赖，使用软件包管理工具声明，比如Go中的Glide。 3.配置 将配置与代码分离，应用部署到kubernete中可以使用容器的环境变量或ConfigMap挂载到容器中。 4.后端服务 把后端服务当作附加资源，实质上是计算存储分离和降低服务耦合，分解单体应用。 5.构建、发布、运行 严格分离构建和运行，每次修改代码生成新的镜像，重新发布，不能直接修改运行时的代码和配置。 6.进程 应用程序进程应该是无状态的，这意味着再次重启后还可以计算出原先的状态。 7.端口绑定 在kubernetes中每个Pod都有独立的IP，每个运行在Pod中的应用不必关心端口是否重复，只需在service中指定端口，集群内的service通过配置互相发现。 8.并发 每个容器都是一个进程，通过增加容器的副本数实现并发。 9.易处理 快速启动和优雅终止可最大化健壮性，kuberentes优秀的Pod生存周期控制。 10.开发环境与线上环境等价 在kubernetes中可以创建多个namespace，使用相同的镜像可以很方便的复制一套环境出来，镜像的使用可以很方便的部署一个后端服务。 11.日志 把日志当作事件流，使用stdout输出并收集汇聚起来，例如到ES中统一查看。 12.管理进程 后台管理任务当作一次性进程运行，kubectl exec进入容器内部操作。 另外，Cloud Native Go 这本书的作者，CapitalOne公司的Kevin Hoffman在TalkingData T11峰会上的High Level Cloud Native的演讲中讲述了云原生应用的15个因素，在原先的12因素应用的基础上又增加了如下三个因素： API优先 服务间的合约 团队协作的规约 文档化、规范化 RESTful或RPC 监控 实时监控远程应用 应用性能监控（APM） 应用健康监控 系统日志 不建议在线Debug 认证授权 不要等最后才去考虑应用的安全性 详细设计、明确声明、文档化 Bearer token、OAuth、OIDC认证 操作审计 详见High Level Cloud Native From Kevin Hoffman。 Kubernetes中的资源管理与容器设计模式 Kubernetes通过声明式配置，真正让开发人员能够理解应用的状态，并通过同一份配置可以立马启动一个一模一样的环境，大大提高了应用开发和部署的效率，其中kubernetes设计的多种资源类型可以帮助我们定义应用的运行状态，并使用资源配置来细粒度得明确限制应用的资源使用。 而容器生态的成熟是 Kubernetes 诞生的前提，在谈到容器的设计模式之前我们先来了解下容器生态，请看下图： 图片 - 容器生态 关于 Docker 容器的更多内容请参考 Docker最佳实践。 容器的设计模式 Kubernetes提供了多种资源对象，用户可以根据自己应用的特性加以选择。这些对象有： 类别 名称 资源对象 Pod、ReplicaSet、ReplicationController、Deployment、StatefulSet、DaemonSet、Job、CronJob、HorizontalPodAutoscaling 配置对象 Node、Namespace、Service、Secret、ConfigMap、Ingress、Label、ThirdPartyResource、 ServiceAccount 存储对象 Volume、Persistent Volume 策略对象 SecurityContext、ResourceQuota、LimitRange 在 Kubernetes 系统中，Kubernetes 对象 是持久化的条目。Kubernetes 使用这些条目去表示整个集群的状态。特别地，它们描述了如下信息： 什么容器化应用在运行（以及在哪个 Node 上） 可以被应用使用的资源 关于应用如何表现的策略，比如重启策略、升级策略，以及容错策略 Kubernetes 对象是 “目标性记录” —— 一旦创建对象，Kubernetes 系统将持续工作以确保对象存在。通过创建对象，可以有效地告知 Kubernetes 系统，所需要的集群工作负载看起来是什么样子的，这就是 Kubernetes 集群的 期望状态。 详见Kubernetes Handbook - Objects。 资源限制与配额 两层的资源限制与配置 Pod级别，最小的资源调度单位 Namespace级别，限制资源配额和每个Pod的资源使用区间 请参考Kubernetes中的ResourceQuota和LimitRange配置资源限额 管理Kubernetes集群 手工部署Kubernetes是一个很艰巨的活，你需要了解网络配置、docker的安装与使用、镜像仓库的构建、角色证书的创建、kubernetes的基本原理和构成、kubernetes应用程序的yaml文件编写等。 我编写了一本kubernetes-handbook可供大家免费阅读，该书记录了本人从零开始学习和使用Kubernetes的心路历程，着重于经验分享和总结，同时也会有相关的概念解析，希望能够帮助大家少踩坑，少走弯路。 部署Kubernetes集群 使用二进制部署 kubernetes 集群的所有组件和插件，而不是使用 kubeadm 等自动化方式来部署集群，同时开启了集群的TLS安全认证，这样可以帮助我们解系统各组件的交互原理，进而能快速解决实际问题。详见Kubernetes Handbook - 在CentOS上部署kubernetes1.6集群。 集群详情 Kubernetes 1.6.0 Docker 1.12.5（使用yum安装） Etcd 3.1.5 Flanneld 0.7 vxlan 网络 TLS 认证通信 (所有组件，如 etcd、kubernetes master 和 node) RBAC 授权 kublet TLS BootStrapping kubedns、dashboard、heapster(influxdb、grafana)、EFK(elasticsearch、fluentd、kibana) 集群插件 私有docker镜像仓库harbor（请自行部署，harbor提供离线安装包，直接使用docker-compose启动即可） 步骤介绍 创建 TLS 证书和秘钥 创建kubeconfig文件 创建高可用etcd集群 安装kubectl命令行工具 部署master节点 安装flannel网络插件 部署node节点 安装kubedns插件 安装dashboard插件 安装heapster插件 安装EFK插件 服务发现与负载均衡 Kubernetes在设计之初就充分考虑了针对容器的服务发现与负载均衡机制，提供了Service资源，并通过kube-proxy配合cloud provider来适应不同的应用场景。随着kubernetes用户的激增，用户场景的不断丰富，又产生了一些新的负载均衡机制。目前，kubernetes中的负载均衡大致可以分为以下几种机制，每种机制都有其特定的应用场景： Service：直接用Service提供cluster内部的负载均衡，并借助cloud provider提供的LB提供外部访问 Ingress：还是用Service提供cluster内部的负载均衡，但是通过自定义LB提供外部访问 Service Load Balancer：把load balancer直接跑在容器中，实现Bare Metal的Service Load Balancer Custom Load Balancer：自定义负载均衡，并替代kube-proxy，一般在物理部署Kubernetes时使用，方便接入公司已有的外部服务 详见Kubernetes Handbook - 服务发现与负载均衡。 持续集成与发布 图片 - 使用Jenkins进行持续集成与发布流程图 应用构建和发布流程说明： 用户向Gitlab提交代码，代码中必须包含Dockerfile 将代码提交到远程仓库 用户在发布应用时需要填写git仓库地址和分支、服务类型、服务名称、资源数量、实例个数，确定后触发Jenkins自动构建 Jenkins的CI流水线自动编译代码并打包成docker镜像推送到Harbor镜像仓库 Jenkins的CI流水线中包括了自定义脚本，根据我们已准备好的kubernetes的YAML模板，将其中的变量替换成用户输入的选项 生成应用的kubernetes YAML配置文件 更新Ingress的配置，根据新部署的应用的名称，在ingress的配置文件中增加一条路由信息 更新PowerDNS，向其中插入一条DNS记录，IP地址是边缘节点的IP地址。关于边缘节点，请查看边缘节点配置 Jenkins调用kubernetes的API，部署应用 日志收集与监控 基于现有的ELK日志收集方案，稍作改造，选用filebeat来收集日志，可以作为sidecar的形式跟应用运行在同一个Pod中，比较轻量级消耗资源比较少。 图片 - filebeat日志收集架构图 详见Kubernetes Handbook - 应用日志收集。 安全性与权限管理 Kubernetes是一个多租户的云平台，因此必须对用户的权限加以限制，对用户空间进行隔离。Kubernetes中的隔离主要包括这几种： 网络隔离：需要使用网络插件，比如calico。 资源隔离：kubernetes原生支持资源隔离，pod就是资源就是隔离和调度的最小单位，同时使用namespace限制用户空间和资源限额。 身份隔离：使用RBAC-基于角色的访问控制，多租户的身份认证和权限控制。 如何开发Kubernetes原生应用步骤介绍 当我们有了一个kubernetes集群后，如何在上面开发和部署应用，应该遵循怎样的流程？下面我将展示如何使用go语言开发和部署一个kubernetes native应用，使用wercker进行持续集成与持续发布，我将以一个很简单的前后端访问，获取伪造数据并展示的例子来说明。 云原生应用开发示例 我们将按照如下步骤来开发部署一个kubernetes原生应用并将它部署到kubernetes集群上开放给集群外访问： 服务API的定义 使用Go语言开发kubernetes原生应用 一个持续构建与发布工具与环境 使用traefik和VIP做边缘节点提供外部访问路由 我写了两个示例用于演示，开发部署一个伪造的 metric 并显示在 web 页面上，包括两个service： k8s-app-monitor-test：生成模拟的监控数据，发送http请求，获取json返回值 K8s-app-monitor-agent：获取监控数据并绘图，访问浏览器获取图表 定义API生成API文档 使用API blueprint格式，定义API文档，格式类似于markdown，再使用aglio生成HTML文档。 图片 - API文档 详见：如何开发部署kubernetes native应用。 如何迁移到云原生应用架构 Pivotal 是云原生应用的提出者，并推出了 Pivotal Cloud Foundry 云原生应用平台和 Spring 开源 Java 开发框架，成为云原生应用架构中先驱者和探路者。 原书作于2015年，其中的示例主要针对 Java 应用，实际上也适用于任何应用类型，云原生应用架构适用于异构语言的程序开发，不仅仅是针对 Java 语言的程序开发。截止到本人翻译本书时，云原生应用生态系统已经初具规模，CNCF 成员不断发展壮大，基于 Cloud Native 的创业公司不断涌现，kubernetes 引领容器编排潮流，和 Service Mesh 技术（如 Linkerd 和 Istio） 的出现，Go 语言的兴起（参考另一本书 Cloud Native Go）等为我们将应用迁移到云原生架构的提供了更多的方案选择。 迁移到云原生应用架构指南 指出了迁移到云原生应用架构需要做出的企业文化、组织架构和技术变革，并给出了迁移指南。 主要讨论的应用程序架构包括： 十二因素应用程序：云原生应用程序架构模式的集合 微服务：独立部署的服务，只做一件事情 自助服务的敏捷基础设施：快速，可重复和一致地提供应用环境和后台服务的平台 基于API的协作：发布和版本化的API，允许在云原生应用程序架构中的服务之间进行交互 抗压性：根据压力变强的系统 详见：迁移到云原生应用架构 迁移案例解析 迁移步骤示意图如下： 图片 - 迁移步骤示意图 步骤说明： 将原有应用拆解为服务 容器化、制作镜像 准备应用配置文件 准备kubernetes YAML文件 编写bootstarp脚本 创建ConfigMaps 详见：迁移传统应用到Kubernetes步骤详解——以Hadoop YARN为例。 Service mesh基本原理和示例介绍 Service mesh现在一般被翻译作服务网格，目前主流的Service mesh有如下两款： Istio：IBM、Google、Lyft共同开源，详细文档见Istio官方文档中文版 Linkerd：原Twitter工程师开发，现为CNCF中的项目之一 什么是Service mesh 如果用一句话来解释什么是 Service Mesh，可以将它比作是应用程序或者说微服务间的 TCP/IP，负责服务之间的网络调用、限流、熔断和监控。对于编写应用程序来说一般无须关心 TCP/IP 这一层（比如通过 HTTP 协议的 RESTful 应用），同样使用 Service Mesh 也就无须关系服务之间的那些原来是通过应用程序或者其他框架实现的事情，比如 Spring Cloud、OSS，现在只要交给 Service Mesh 就可以了。 图片 - service mesh架构图 详见什么是 service mesh - jimmysong.io。 Service mesh使用指南 两款Service mesh各有千秋，我分别写了他们的使用案例指南： 微服务管理框架service mesh——Linkerd安装试用笔记 微服务管理框架service mesh——Istio安装试用笔记 更多关于 Service Mesh 的内容请访问 Service Mesh 中文网。 使用案例 Kubernetes作为云原生计算的基本组件之一，开源2年时间以来热度与日俱增，它可以跟我们的生产结合，擦出很多火花，比如FaaS和Serverless类应用，都很适合运行在kubernetes上。 关于Cloud Native开源软件生态请参考 Awesome Cloud Native - jimmysong.io。 DevOps 下面是社区中kubernetes开源爱好者的分享内容，我觉得是对kubernetes在DevOps中应用的很好的形式值得大家借鉴。 真正践行DevOps，让开发人员在掌握自己的开发和测试环境，让环境一致，让开发效率提升，让运维没有堆积如山的tickets，让监控更加精准，从kubernetes平台开始。 行动指南 根据环境（比如开发、测试、生产）划分namespace，也可以根据项目来划分 再为每个用户划分一个namespace、创建一个serviceaccount和kubeconfig文件，不同namespace间的资源隔离，目前不隔离网络，不同namespace间的服务可以互相访问 创建yaml模板，降低编写kubernetes yaml文件编写难度 在kubectl命令上再封装一层，增加用户身份设置和环境初始化操作，简化kubectl命令和常用功能 管理员通过dashboard查看不同namespace的状态，也可以使用它来使操作更便捷 所有应用的日志统一收集到ElasticSearch中，统一日志访问入口 可以通过Grafana查看所有namespace中的应用的状态和kubernetes集群本身的状态 需要持久化的数据保存在分布式存储中，例如GlusterFS或Ceph中 使用Kibana查看日志 日志字段中包括了应用的标签、容器名称、主机名称、宿主机名称、IP地址、时间、 图片 - kibana界面 使用Grafana查看应用状态 注：感谢【K8S🤘Cloud Native实战群】尊贵的黄金会员小刚同学提供下面的Grafana监控图🙏 监控分类示意图： 图片 - Grafana界面示意图1 Kubernetes集群全局监控图1 该监控图可以看到集群硬件使用情况。 图片 - Grafana界面示意图2 Kubernetes全局监控图2 该监控可以看到单个用户的namespace下的所有资源的使用情况。 图片 - Grafana界面示意图3 Spark on Kubernetes TL;DR https://jimmysong.io/spark-on-k8s Spark原生支持standalone、mesos和YARN资源调度，现已支持Kubernetes原生调度，详见运行支持kubernetes原生调度的spark程序-Spark on Kubernetes。 为何要使用spark on kubernetes 使用kubernetes原生调度的spark on kubernetes是对原先的spark on yarn和yarn on docker的改变是革命性的，主要表现在以下几点： Kubernetes原生调度：不再需要二层调度，直接使用kubernetes的资源调度功能，跟其他应用共用整个kubernetes管理的资源池； 资源隔离，粒度更细：原先yarn中的queue在spark on kubernetes中已不存在，取而代之的是kubernetes中原生的namespace，可以为每个用户分别指定一个namespace，限制用户的资源quota； 细粒度的资源分配：可以给每个spark任务指定资源限制，实际指定多少资源就使用多少资源，因为没有了像yarn那样的二层调度（圈地式的），所以可以更高效和细粒度的使用资源； 监控的变革：因为做到了细粒度的资源分配，所以可以对用户提交的每一个任务做到资源使用的监控，从而判断用户的资源使用情况，所有的metric都记录在数据库中，甚至可以为每个用户的每次任务提交计量； 日志的变革：用户不再通过yarn的web页面来查看任务状态，而是通过pod的log来查看，可将所有的kuberentes中的应用的日志等同看待收集起来，然后可以根据标签查看对应应用的日志； 如何提交任务 仍然使用spark-submit提交spark任务，可以直接指定kubernetes API server地址，下面的命令提交本地jar包到kubernetes集群上运行，同时指定了运行任务的用户、提交命名的用户、运行的excutor实例数、driver和executor的资源限制、使用的spark版本等信息。 详细使用说明见Apache Spark on Kubernetes用户指南 - jimmysong.io。 ./spark-submit \\ --deploy-mode cluster \\ --class com.talkingdata.alluxio.hadooptest \\ --master k8s://https://172.20.0.113:6443 \\ --kubernetes-namespace spark-cluster \\ --conf spark.kubernetes.driverEnv.SPARK_USER=hadoop \\ --conf spark.kubernetes.driverEnv.HADOOP_USER_NAME=hadoop \\ --conf spark.executorEnv.HADOOP_USER_NAME=hadoop \\ --conf spark.executorEnv.SPARK_USER=hadoop \\ --conf spark.kubernetes.authenticate.driver.serviceAccountName=spark \\ --conf spark.driver.memory=100G \\ --conf spark.executor.memory=10G \\ --conf spark.driver.cores=30 \\ --conf spark.executor.cores=2 \\ --conf spark.driver.maxResultSize=10240m \\ --conf spark.kubernetes.driver.limit.cores=32 \\ --conf spark.kubernetes.executor.limit.cores=3 \\ --conf spark.kubernetes.executor.memoryOverhead=2g \\ --conf spark.executor.instances=5 \\ --conf spark.app.name=spark-pi \\ --conf spark.kubernetes.driver.docker.image=sz-pg-oam-docker-hub-001.tendcloud.com/library/spark-driver:v2.1.0-kubernetes-0.3.1-1 \\ --conf spark.kubernetes.executor.docker.image=sz-pg-oam-docker-hub-001.tendcloud.com/library/spark-executor:v2.1.0-kubernetes-0.3.1-1 \\ --conf spark.kubernetes.initcontainer.docker.image=sz-pg-oam-docker-hub-001.tendcloud.com/library/spark-init:v2.1.0-kubernetes-0.3.1-1 \\ --conf spark.kubernetes.resourceStagingServer.uri=http://172.20.0.114:31000 \\ ~/Downloads/tendcloud_2.10-1.0.jar 监控 下图是从Kubernetes dashboard上看到的spark-cluster这个namespace上运行的应用情况。 图片 - dashboard 下图是从Grafana监控页面上查看到的某个executor资源占用情况。 图片 - Grafana 参考 迁移到云原生应用架构指南 Cloud Native Go - 已由电子工业出版社出版 Cloud Native Python - 将由电子工业出版社出版 Istio Service Mesh 中文文档 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-20 17:10:30 "},"concepts/":{"url":"concepts/","title":"Kubernetes架构","keywords":"","body":"Kubernetes架构 Kubernetes最初源于谷歌内部的Borg，提供了面向应用的容器集群部署和管理系统。Kubernetes的目标旨在消除编排物理/虚拟计算，网络和存储基础设施的负担，并使应用程序运营商和开发人员完全将重点放在以容器为中心的原语上进行自助运营。Kubernetes 也提供稳定、兼容的基础（平台），用于构建定制化的workflows 和更高级的自动化任务。 Kubernetes 具备完善的集群管理能力，包括多层次的安全防护和准入机制、多租户应用支撑能力、透明的服务注册和服务发现机制、内建负载均衡器、故障发现和自我修复能力、服务滚动升级和在线扩容、可扩展的资源自动调度机制、多粒度的资源配额管理能力。 Kubernetes 还提供完善的管理工具，涵盖开发、部署测试、运维监控等各个环节。 Borg简介 Borg是谷歌内部的大规模集群管理系统，负责对谷歌内部很多核心服务的调度和管理。Borg的目的是让用户能够不必操心资源管理的问题，让他们专注于自己的核心业务，并且做到跨多个数据中心的资源利用率最大化。 Borg主要由BorgMaster、Borglet、borgcfg和Scheduler组成，如下图所示 图片 - Borg架构 BorgMaster是整个集群的大脑，负责维护整个集群的状态，并将数据持久化到Paxos存储中； Scheduer负责任务的调度，根据应用的特点将其调度到具体的机器上去； Borglet负责真正运行任务（在容器中）； borgcfg是Borg的命令行工具，用于跟Borg系统交互，一般通过一个配置文件来提交任务。 Kubernetes架构 Kubernetes借鉴了Borg的设计理念，比如Pod、Service、Labels和单Pod单IP等。Kubernetes的整体架构跟Borg非常像，如下图所示 图片 - Kubernetes架构 Kubernetes主要由以下几个核心组件组成： etcd保存了整个集群的状态； apiserver提供了资源操作的唯一入口，并提供认证、授权、访问控制、API注册和发现等机制； controller manager负责维护集群的状态，比如故障检测、自动扩展、滚动更新等； scheduler负责资源的调度，按照预定的调度策略将Pod调度到相应的机器上； kubelet负责维护容器的生命周期，同时也负责Volume（CVI）和网络（CNI）的管理； Container runtime负责镜像管理以及Pod和容器的真正运行（CRI）； kube-proxy负责为Service提供cluster内部的服务发现和负载均衡； 除了核心组件，还有一些推荐的Add-ons： kube-dns负责为整个集群提供DNS服务 Ingress Controller为服务提供外网入口 Heapster提供资源监控 Dashboard提供GUI Federation提供跨可用区的集群 Kubernetes架构示意图 整体架构 下图清晰表明了kubernetes的架构设计以及组件之间的通信协议。 图片 - Kuberentes架构（图片来自于网络） 下面是更抽象的一个视图： 图片 - kubernetes整体架构示意图 Master架构 图片 - Kubernetes master架构示意图 Node架构 图片 - kubernetes node架构示意图 分层架构 Kubernetes设计理念和功能其实就是一个类似Linux的分层架构，如下图所示 图片 - Kubernetes分层架构示意图 核心层：Kubernetes最核心的功能，对外提供API构建高层的应用，对内提供插件式应用执行环境 应用层：部署（无状态应用、有状态应用、批处理任务、集群应用等）和路由（服务发现、DNS解析等） 管理层：系统度量（如基础设施、容器和网络的度量），自动化（如自动扩展、动态Provision等）以及策略管理（RBAC、Quota、PSP、NetworkPolicy等） 接口层：kubectl命令行工具、客户端SDK以及集群联邦 生态系统：在接口层之上的庞大容器集群管理调度的生态系统，可以划分为两个范畴 Kubernetes外部：日志、监控、配置管理、CI、CD、Workflow、FaaS、OTS应用、ChatOps等 Kubernetes内部：CRI、CNI、CVI、镜像仓库、Cloud Provider、集群自身的配置和管理等 关于分层架构，可以关注下Kubernetes社区正在推进的Kbernetes architectual roadmap和slide。 参考文档 Kubernetes design and architecture http://queue.acm.org/detail.cfm?id=2898444 http://static.googleusercontent.com/media/research.google.com/zh-CN//pubs/archive/43438.pdf http://thenewstack.io/kubernetes-an-overview Kbernetes architectual roadmap和slide Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-08 22:04:06 "},"concepts/concepts.html":{"url":"concepts/concepts.html","title":"设计理念","keywords":"","body":"Kubernetes的设计理念 Kubernetes设计理念与分布式系统 分析和理解Kubernetes的设计理念可以使我们更深入地了解Kubernetes系统，更好地利用它管理分布式部署的云原生应用，另一方面也可以让我们借鉴其在分布式系统设计方面的经验。 分层架构 Kubernetes设计理念和功能其实就是一个类似Linux的分层架构，如下图所示 图片 - 分层架构示意图 核心层：Kubernetes最核心的功能，对外提供API构建高层的应用，对内提供插件式应用执行环境 应用层：部署（无状态应用、有状态应用、批处理任务、集群应用等）和路由（服务发现、DNS解析等） 管理层：系统度量（如基础设施、容器和网络的度量），自动化（如自动扩展、动态Provision等）以及策略管理（RBAC、Quota、PSP、NetworkPolicy等） 接口层：kubectl命令行工具、客户端SDK以及集群联邦 生态系统：在接口层之上的庞大容器集群管理调度的生态系统，可以划分为两个范畴 Kubernetes外部：日志、监控、配置管理、CI、CD、Workflow、FaaS、OTS应用、ChatOps等 Kubernetes内部：CRI、CNI、CVI、镜像仓库、Cloud Provider、集群自身的配置和管理等 API设计原则 对于云计算系统，系统API实际上处于系统设计的统领地位，正如本文前面所说，kubernetes集群系统每支持一项新功能，引入一项新技术，一定会新引入对应的API对象，支持对该功能的管理操作，理解掌握的API，就好比抓住了kubernetes系统的牛鼻子。Kubernetes系统API的设计有以下几条原则： 所有API应该是声明式的。正如前文所说，声明式的操作，相对于命令式操作，对于重复操作的效果是稳定的，这对于容易出现数据丢失或重复的分布式环境来说是很重要的。另外，声明式操作更容易被用户使用，可以使系统向用户隐藏实现的细节，隐藏实现的细节的同时，也就保留了系统未来持续优化的可能性。此外，声明式的API，同时隐含了所有的API对象都是名词性质的，例如Service、Volume这些API都是名词，这些名词描述了用户所期望得到的一个目标分布式对象。 API对象是彼此互补而且可组合的。这里面实际是鼓励API对象尽量实现面向对象设计时的要求，即“高内聚，松耦合”，对业务相关的概念有一个合适的分解，提高分解出来的对象的可重用性。事实上，Kubernetes这种分布式系统管理平台，也是一种业务系统，只不过它的业务就是调度和管理容器服务。 高层API以操作意图为基础设计。如何能够设计好API，跟如何能用面向对象的方法设计好应用系统有相通的地方，高层设计一定是从业务出发，而不是过早的从技术实现出发。因此，针对Kubernetes的高层API设计，一定是以Kubernetes的业务为基础出发，也就是以系统调度管理容器的操作意图为基础设计。 低层API根据高层API的控制需要设计。设计实现低层API的目的，是为了被高层API使用，考虑减少冗余、提高重用性的目的，低层API的设计也要以需求为基础，要尽量抵抗受技术实现影响的诱惑。 尽量避免简单封装，不要有在外部API无法显式知道的内部隐藏的机制。简单的封装，实际没有提供新的功能，反而增加了对所封装API的依赖性。内部隐藏的机制也是非常不利于系统维护的设计方式，例如PetSet和ReplicaSet，本来就是两种Pod集合，那么Kubernetes就用不同API对象来定义它们，而不会说只用同一个ReplicaSet，内部通过特殊的算法再来区分这个ReplicaSet是有状态的还是无状态。 API操作复杂度与对象数量成正比。这一条主要是从系统性能角度考虑，要保证整个系统随着系统规模的扩大，性能不会迅速变慢到无法使用，那么最低的限定就是API的操作复杂度不能超过O(N)，N是对象的数量，否则系统就不具备水平伸缩性了。 API对象状态不能依赖于网络连接状态。由于众所周知，在分布式环境下，网络连接断开是经常发生的事情，因此要保证API对象状态能应对网络的不稳定，API对象的状态就不能依赖于网络连接状态。 尽量避免让操作机制依赖于全局状态，因为在分布式系统中要保证全局状态的同步是非常困难的。 控制机制设计原则 控制逻辑应该只依赖于当前状态。这是为了保证分布式系统的稳定可靠，对于经常出现局部错误的分布式系统，如果控制逻辑只依赖当前状态，那么就非常容易将一个暂时出现故障的系统恢复到正常状态，因为你只要将该系统重置到某个稳定状态，就可以自信的知道系统的所有控制逻辑会开始按照正常方式运行。 假设任何错误的可能，并做容错处理。在一个分布式系统中出现局部和临时错误是大概率事件。错误可能来自于物理系统故障，外部系统故障也可能来自于系统自身的代码错误，依靠自己实现的代码不会出错来保证系统稳定其实也是难以实现的，因此要设计对任何可能错误的容错处理。 尽量避免复杂状态机，控制逻辑不要依赖无法监控的内部状态。因为分布式系统各个子系统都是不能严格通过程序内部保持同步的，所以如果两个子系统的控制逻辑如果互相有影响，那么子系统就一定要能互相访问到影响控制逻辑的状态，否则，就等同于系统里存在不确定的控制逻辑。 假设任何操作都可能被任何操作对象拒绝，甚至被错误解析。由于分布式系统的复杂性以及各子系统的相对独立性，不同子系统经常来自不同的开发团队，所以不能奢望任何操作被另一个子系统以正确的方式处理，要保证出现错误的时候，操作级别的错误不会影响到系统稳定性。 每个模块都可以在出错后自动恢复。由于分布式系统中无法保证系统各个模块是始终连接的，因此每个模块要有自我修复的能力，保证不会因为连接不到其他模块而自我崩溃。 每个模块都可以在必要时优雅地降级服务。所谓优雅地降级服务，是对系统鲁棒性的要求，即要求在设计实现模块时划分清楚基本功能和高级功能，保证基本功能不会依赖高级功能，这样同时就保证了不会因为高级功能出现故障而导致整个模块崩溃。根据这种理念实现的系统，也更容易快速地增加新的高级功能，以为不必担心引入高级功能影响原有的基本功能。 Kubernetes的核心技术概念和API对象 API对象是Kubernetes集群中的管理操作单元。Kubernetes集群系统每支持一项新功能，引入一项新技术，一定会新引入对应的API对象，支持对该功能的管理操作。例如副本集Replica Set对应的API对象是RS。 每个API对象都有3大类属性：元数据metadata、规范spec和状态status。元数据是用来标识API对象的，每个对象都至少有3个元数据：namespace，name和uid；除此以外还有各种各样的标签labels用来标识和匹配不同的对象，例如用户可以用标签env来标识区分不同的服务部署环境，分别用env=dev、env=testing、env=production来标识开发、测试、生产的不同服务。规范描述了用户期望Kubernetes集群中的分布式系统达到的理想状态（Desired State），例如用户可以通过复制控制器Replication Controller设置期望的Pod副本数为3；status描述了系统实际当前达到的状态（Status），例如系统当前实际的Pod副本数为2；那么复制控制器当前的程序逻辑就是自动启动新的Pod，争取达到副本数为3。 Kubernetes中所有的配置都是通过API对象的spec去设置的，也就是用户通过配置系统的理想状态来改变系统，这是Kubernetes重要设计理念之一，即所有的操作都是声明式（Declarative）的而不是命令式（Imperative）的。声明式操作在分布式系统中的好处是稳定，不怕丢操作或运行多次，例如设置副本数为3的操作运行多次也还是一个结果，而给副本数加1的操作就不是声明式的，运行多次结果就错了。 Pod Kubernetes有很多技术概念，同时对应很多API对象，最重要的也是最基础的是Pod。Pod是在Kubernetes集群中运行部署应用或服务的最小单元，它是可以支持多容器的。Pod的设计理念是支持多个容器在一个Pod中共享网络地址和文件系统，可以通过进程间通信和文件共享这种简单高效的方式组合完成服务。Pod对多容器的支持是K8最基础的设计理念。比如你运行一个操作系统发行版的软件仓库，一个Nginx容器用来发布软件，另一个容器专门用来从源仓库做同步，这两个容器的镜像不太可能是一个团队开发的，但是他们一块儿工作才能提供一个微服务；这种情况下，不同的团队各自开发构建自己的容器镜像，在部署的时候组合成一个微服务对外提供服务。 Pod是Kubernetes集群中所有业务类型的基础，可以看作运行在K8集群中的小机器人，不同类型的业务就需要不同类型的小机器人去执行。目前Kubernetes中的业务主要可以分为长期伺服型（long-running）、批处理型（batch）、节点后台支撑型（node-daemon）和有状态应用型（stateful application）；分别对应的小机器人控制器为Deployment、Job、DaemonSet和PetSet，本文后面会一一介绍。 副本控制器（Replication Controller，RC） RC是Kubernetes集群中最早的保证Pod高可用的API对象。通过监控运行中的Pod来保证集群中运行指定数目的Pod副本。指定的数目可以是多个也可以是1个；少于指定数目，RC就会启动运行新的Pod副本；多于指定数目，RC就会杀死多余的Pod副本。即使在指定数目为1的情况下，通过RC运行Pod也比直接运行Pod更明智，因为RC也可以发挥它高可用的能力，保证永远有1个Pod在运行。RC是Kubernetes较早期的技术概念，只适用于长期伺服型的业务类型，比如控制小机器人提供高可用的Web服务。 副本集（Replica Set，RS） RS是新一代RC，提供同样的高可用能力，区别主要在于RS后来居上，能支持更多种类的匹配模式。副本集对象一般不单独使用，而是作为Deployment的理想状态参数使用。 部署（Deployment） 部署表示用户对Kubernetes集群的一次更新操作。部署是一个比RS应用模式更广的API对象，可以是创建一个新的服务，更新一个新的服务，也可以是滚动升级一个服务。滚动升级一个服务，实际是创建一个新的RS，然后逐渐将新RS中副本数增加到理想状态，将旧RS中的副本数减小到0的复合操作；这样一个复合操作用一个RS是不太好描述的，所以用一个更通用的Deployment来描述。以Kubernetes的发展方向，未来对所有长期伺服型的的业务的管理，都会通过Deployment来管理。 服务（Service） RC、RS和Deployment只是保证了支撑服务的微服务Pod的数量，但是没有解决如何访问这些服务的问题。一个Pod只是一个运行服务的实例，随时可能在一个节点上停止，在另一个节点以一个新的IP启动一个新的Pod，因此不能以确定的IP和端口号提供服务。要稳定地提供服务需要服务发现和负载均衡能力。服务发现完成的工作，是针对客户端访问的服务，找到对应的的后端服务实例。在K8集群中，客户端需要访问的服务就是Service对象。每个Service会对应一个集群内部有效的虚拟IP，集群内部通过虚拟IP访问一个服务。在Kubernetes集群中微服务的负载均衡是由Kube-proxy实现的。Kube-proxy是Kubernetes集群内部的负载均衡器。它是一个分布式代理服务器，在Kubernetes的每个节点上都有一个；这一设计体现了它的伸缩性优势，需要访问服务的节点越多，提供负载均衡能力的Kube-proxy就越多，高可用节点也随之增多。与之相比，我们平时在服务器端做个反向代理做负载均衡，还要进一步解决反向代理的负载均衡和高可用问题。 任务（Job） Job是Kubernetes用来控制批处理型任务的API对象。批处理业务与长期伺服业务的主要区别是批处理业务的运行有头有尾，而长期伺服业务在用户不停止的情况下永远运行。Job管理的Pod根据用户的设置把任务成功完成就自动退出了。成功完成的标志根据不同的spec.completions策略而不同：单Pod型任务有一个Pod成功就标志完成；定数成功型任务保证有N个任务全部成功；工作队列型任务根据应用确认的全局成功而标志成功。 后台支撑服务集（DaemonSet） 长期伺服型和批处理型服务的核心在业务应用，可能有些节点运行多个同类业务的Pod，有些节点上又没有这类Pod运行；而后台支撑型服务的核心关注点在Kubernetes集群中的节点（物理机或虚拟机），要保证每个节点上都有一个此类Pod运行。节点可能是所有集群节点也可能是通过nodeSelector选定的一些特定节点。典型的后台支撑型服务包括，存储，日志和监控等在每个节点上支持Kubernetes集群运行的服务。 有状态服务集（PetSet） Kubernetes在1.3版本里发布了Alpha版的PetSet功能。在云原生应用的体系里，有下面两组近义词；第一组是无状态（stateless）、牲畜（cattle）、无名（nameless）、可丢弃（disposable）；第二组是有状态（stateful）、宠物（pet）、有名（having name）、不可丢弃（non-disposable）。RC和RS主要是控制提供无状态服务的，其所控制的Pod的名字是随机设置的，一个Pod出故障了就被丢弃掉，在另一个地方重启一个新的Pod，名字变了、名字和启动在哪儿都不重要，重要的只是Pod总数；而PetSet是用来控制有状态服务，PetSet中的每个Pod的名字都是事先确定的，不能更改。PetSet中Pod的名字的作用，并不是《千与千寻》的人性原因，而是关联与该Pod对应的状态。 对于RC和RS中的Pod，一般不挂载存储或者挂载共享存储，保存的是所有Pod共享的状态，Pod像牲畜一样没有分别（这似乎也确实意味着失去了人性特征）；对于PetSet中的Pod，每个Pod挂载自己独立的存储，如果一个Pod出现故障，从其他节点启动一个同样名字的Pod，要挂在上原来Pod的存储继续以它的状态提供服务。 适合于PetSet的业务包括数据库服务MySQL和PostgreSQL，集群化管理服务Zookeeper、etcd等有状态服务。PetSet的另一种典型应用场景是作为一种比普通容器更稳定可靠的模拟虚拟机的机制。传统的虚拟机正是一种有状态的宠物，运维人员需要不断地维护它，容器刚开始流行时，我们用容器来模拟虚拟机使用，所有状态都保存在容器里，而这已被证明是非常不安全、不可靠的。使用PetSet，Pod仍然可以通过漂移到不同节点提供高可用，而存储也可以通过外挂的存储来提供高可靠性，PetSet做的只是将确定的Pod与确定的存储关联起来保证状态的连续性。PetSet还只在Alpha阶段，后面的设计如何演变，我们还要继续观察。 集群联邦（Federation） Kubernetes在1.3版本里发布了beta版的Federation功能。在云计算环境中，服务的作用距离范围从近到远一般可以有：同主机（Host，Node）、跨主机同可用区（Available Zone）、跨可用区同地区（Region）、跨地区同服务商（Cloud Service Provider）、跨云平台。Kubernetes的设计定位是单一集群在同一个地域内，因为同一个地区的网络性能才能满足Kubernetes的调度和计算存储连接要求。而联合集群服务就是为提供跨Region跨服务商Kubernetes集群服务而设计的。 每个Kubernetes Federation有自己的分布式存储、API Server和Controller Manager。用户可以通过Federation的API Server注册该Federation的成员Kubernetes Cluster。当用户通过Federation的API Server创建、更改API对象时，Federation API Server会在自己所有注册的子Kubernetes Cluster都创建一份对应的API对象。在提供业务请求服务时，Kubernetes Federation会先在自己的各个子Cluster之间做负载均衡，而对于发送到某个具体Kubernetes Cluster的业务请求，会依照这个Kubernetes Cluster独立提供服务时一样的调度模式去做Kubernetes Cluster内部的负载均衡。而Cluster之间的负载均衡是通过域名服务的负载均衡来实现的。 所有的设计都尽量不影响Kubernetes Cluster现有的工作机制，这样对于每个子Kubernetes集群来说，并不需要更外层的有一个Kubernetes Federation，也就是意味着所有现有的Kubernetes代码和机制不需要因为Federation功能有任何变化。 存储卷（Volume） Kubernetes集群中的存储卷跟Docker的存储卷有些类似，只不过Docker的存储卷作用范围为一个容器，而Kubernetes的存储卷的生命周期和作用范围是一个Pod。每个Pod中声明的存储卷由Pod中的所有容器共享。Kubernetes支持非常多的存储卷类型，特别的，支持多种公有云平台的存储，包括AWS，Google和Azure云；支持多种分布式存储包括GlusterFS和Ceph；也支持较容易使用的主机本地目录hostPath和NFS。Kubernetes还支持使用Persistent Volume Claim即PVC这种逻辑存储，使用这种存储，使得存储的使用者可以忽略后台的实际存储技术（例如AWS，Google或GlusterFS和Ceph），而将有关存储实际技术的配置交给存储管理员通过Persistent Volume来配置。 持久存储卷（Persistent Volume，PV）和持久存储卷声明（Persistent Volume Claim，PVC） PV和PVC使得Kubernetes集群具备了存储的逻辑抽象能力，使得在配置Pod的逻辑里可以忽略对实际后台存储技术的配置，而把这项配置的工作交给PV的配置者，即集群的管理者。存储的PV和PVC的这种关系，跟计算的Node和Pod的关系是非常类似的；PV和Node是资源的提供者，根据集群的基础设施变化而变化，由Kubernetes集群管理员配置；而PVC和Pod是资源的使用者，根据业务服务的需求变化而变化，有Kubernetes集群的使用者即服务的管理员来配置。 节点（Node） Kubernetes集群中的计算能力由Node提供，最初Node称为服务节点Minion，后来改名为Node。Kubernetes集群中的Node也就等同于Mesos集群中的Slave节点，是所有Pod运行所在的工作主机，可以是物理机也可以是虚拟机。不论是物理机还是虚拟机，工作主机的统一特征是上面要运行kubelet管理节点上运行的容器。 密钥对象（Secret） Secret是用来保存和传递密码、密钥、认证凭证这些敏感信息的对象。使用Secret的好处是可以避免把敏感信息明文写在配置文件里。在Kubernetes集群中配置和使用服务不可避免的要用到各种敏感信息实现登录、认证等功能，例如访问AWS存储的用户名密码。为了避免将类似的敏感信息明文写在所有需要使用的配置文件中，可以将这些信息存入一个Secret对象，而在配置文件中通过Secret对象引用这些敏感信息。这种方式的好处包括：意图明确，避免重复，减少暴漏机会。 用户帐户（User Account）和服务帐户（Service Account） 顾名思义，用户帐户为人提供账户标识，而服务账户为计算机进程和Kubernetes集群中运行的Pod提供账户标识。用户帐户和服务帐户的一个区别是作用范围；用户帐户对应的是人的身份，人的身份与服务的namespace无关，所以用户账户是跨namespace的；而服务帐户对应的是一个运行中程序的身份，与特定namespace是相关的。 命名空间（Namespace） 命名空间为Kubernetes集群提供虚拟的隔离作用，Kubernetes集群初始有两个命名空间，分别是默认命名空间default和系统命名空间kube-system，除此以外，管理员可以可以创建新的命名空间满足需要。 RBAC访问授权 Kubernetes在1.3版本中发布了alpha版的基于角色的访问控制（Role-based Access Control，RBAC）的授权模式。相对于基于属性的访问控制（Attribute-based Access Control，ABAC），RBAC主要是引入了角色（Role）和角色绑定（RoleBinding）的抽象概念。在ABAC中，Kubernetes集群中的访问策略只能跟用户直接关联；而在RBAC中，访问策略可以跟某个角色关联，具体的用户在跟一个或多个角色相关联。显然，RBAC像其他新功能一样，每次引入新功能，都会引入新的API对象，从而引入新的概念抽象，而这一新的概念抽象一定会使集群服务管理和使用更容易扩展和重用。 总结 从Kubernetes的系统架构、技术概念和设计理念，我们可以看到Kubernetes系统最核心的两个设计理念：一个是容错性，一个是易扩展性。容错性实际是保证Kubernetes系统稳定性和安全性的基础，易扩展性是保证Kubernetes对变更友好，可以快速迭代增加新功能的基础。 按照分布式系统一致性算法Paxos发明人计算机科学家Leslie Lamport的理念，一个分布式系统有两类特性：安全性Safety和活性Liveness。安全性保证系统的稳定，保证系统不会崩溃，不会出现业务错误，不会做坏事，是严格约束的；活性使得系统可以提供功能，提高性能，增加易用性，让系统可以在用户“看到的时间内”做些好事，是尽力而为的。Kubernetes系统的设计理念正好与Lamport安全性与活性的理念不谋而合，也正是因为Kubernetes在引入功能和技术的时候，非常好地划分了安全性和活性，才可以让Kubernetes能有这么快版本迭代，快速引入像RBAC、Federation和PetSet这种新功能。 原文地址：《Kubernetes与云原生应用》系列之Kubernetes的系统架构与设计理念 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-21 11:59:28 "},"concepts/objects.html":{"url":"concepts/objects.html","title":"资源对象与基本概念解析","keywords":"","body":"Objects 以下列举的内容都是 kubernetes 中的 Object，这些对象都可以在 yaml 文件中作为一种 API 类型来配置。 Pod Node Namespace Service Volume PersistentVolume Deployment Secret StatefulSet DaemonSet ServiceAccount ReplicationController ReplicaSet Job CronJob SecurityContext ResourceQuota LimitRange HorizontalPodAutoscaling Ingress ConfigMap Label ThirdPartyResources 我将它们简单的分类为以下几种资源对象： 类别 名称 资源对象 Pod、ReplicaSet、ReplicationController、Deployment、StatefulSet、DaemonSet、Job、CronJob、HorizontalPodAutoscaling 配置对象 Node、Namespace、Service、Secret、ConfigMap、Ingress、Label、ThirdPartyResource、 ServiceAccount 存储对象 Volume、Persistent Volume 策略对象 SecurityContext、ResourceQuota、LimitRange 理解 kubernetes 中的对象 在 Kubernetes 系统中，Kubernetes 对象 是持久化的条目。Kubernetes 使用这些条目去表示整个集群的状态。特别地，它们描述了如下信息： 什么容器化应用在运行（以及在哪个 Node 上） 可以被应用使用的资源 关于应用如何表现的策略，比如重启策略、升级策略，以及容错策略 Kubernetes 对象是 “目标性记录” —— 一旦创建对象，Kubernetes 系统将持续工作以确保对象存在。通过创建对象，可以有效地告知 Kubernetes 系统，所需要的集群工作负载看起来是什么样子的，这就是 Kubernetes 集群的 期望状态。 与 Kubernetes 对象工作 —— 是否创建、修改，或者删除 —— 需要使用 Kubernetes API。当使用 kubectl 命令行接口时，比如，CLI 会使用必要的 Kubernetes API 调用，也可以在程序中直接使用 Kubernetes API。为了实现该目标，Kubernetes 当前提供了一个 golang 客户端库 ，其它语言库（例如Python）也正在开发中。 对象 Spec 与状态 每个 Kubernetes 对象包含两个嵌套的对象字段，它们负责管理对象的配置：对象 spec 和 对象 status。spec 必须提供，它描述了对象的 期望状态—— 希望对象所具有的特征。status 描述了对象的 实际状态，它是由 Kubernetes 系统提供和更新。在任何时刻，Kubernetes 控制平面一直处于活跃状态，管理着对象的实际状态以与我们所期望的状态相匹配。 例如，Kubernetes Deployment 对象能够表示运行在集群中的应用。当创建 Deployment 时，可能需要设置 Deployment 的 spec，以指定该应用需要有 3 个副本在运行。Kubernetes 系统读取 Deployment spec，启动我们所期望的该应用的 3 个实例 —— 更新状态以与 spec 相匹配。如果那些实例中有失败的（一种状态变更），Kubernetes 系统通过修正来响应 spec 和状态之间的不一致 —— 这种情况，启动一个新的实例来替换。 关于对象 spec、status 和 metadata 更多信息，查看 Kubernetes API Conventions。 描述 Kubernetes 对象 当创建 Kubernetes 对象时，必须提供对象的 spec，用来描述该对象的期望状态，以及关于对象的一些基本信息（例如，名称）。当使用 Kubernetes API 创建对象时（或者直接创建，或者基于kubectl），API 请求必须在请求体中包含 JSON 格式的信息。更常用的是，需要在 .yaml 文件中为 kubectl 提供这些信息。 kubectl 在执行 API 请求时，将这些信息转换成 JSON 格式。 这里有一个 .yaml 示例文件，展示了 Kubernetes Deployment 的必需字段和对象 spec： apiVersion: apps/v1beta1 kind: Deployment metadata: name: nginx-deployment spec: replicas: 3 template: metadata: labels: app: nginx spec: containers: - name: nginx image: nginx:1.7.9 ports: - containerPort: 80 一种创建 Deployment 的方式，类似上面使用 .yaml 文件，是使用 kubectl 命令行接口（CLI）中的 kubectl create 命令，传递 .yaml 作为参数。下面是一个示例： $ kubectl create -f docs/user-guide/nginx-deployment.yaml --record 输出类似如下这样： deployment \"nginx-deployment\" created 必需字段 在想要创建的 Kubernetes 对象对应的 .yaml 文件中，需要配置如下的字段： apiVersion - 创建该对象所使用的 Kubernetes API 的版本 kind - 想要创建的对象的类型 metadata - 帮助识别对象唯一性的数据，包括一个 name 字符串、UID 和可选的 namespace 也需要提供对象的 spec 字段。对象 spec 的精确格式对每个 Kubernetes 对象来说是不同的，包含了特定于该对象的嵌套字段。Kubernetes API 参考能够帮助我们找到任何我们想创建的对象的 spec 格式。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-22 17:00:47 "},"concepts/pod-overview.html":{"url":"concepts/pod-overview.html","title":"Pod","keywords":"","body":"Pod概览 理解Pod Pod是kubernetes中你可以创建和部署的最小也是最简单位。一个Pod代表着集群中运行的一个进程。 Pod中封装着应用的容器（有的情况下是好几个容器），存储、独立的网络IP，管理容器如何运行的策略选项。Pod代表着部署的一个单位：kubernetes中应用的一个实例，可能由一个或者多个容器组合在一起共享资源。 Docker是kubernetes中最常用的容器运行时，但是Pod也支持其他容器运行时。 Pods are employed a number of ways in a Kubernetes cluster, including: 在Kubrenetes集群中Pod有如下两种使用方式： 一个Pod中运行一个容器。“每个Pod中一个容器”的模式是最常见的用法；在这种使用方式中，你可以把Pod想象成是单个容器的封装，kuberentes管理的是Pod而不是直接管理容器。 在一个Pod中同时运行多个容器。一个Pod中也可以同时封装几个需要紧密耦合互相协作的容器，它们之间共享资源。这些在同一个Pod中的容器可以互相协作成为一个service单位——一个容器共享文件，另一个“sidecar”容器来更新这些文件。Pod将这些容器的存储资源作为一个实体来管理。 Kubernetes Blog 有关于Pod用例的详细信息，查看： The Distributed System Toolkit: Patterns for Composite Containers Container Design Patterns 每个Pod都是应用的一个实例。如果你想平行扩展应用的话（运行多个实例），你应该运行多个Pod，每个Pod都是一个应用实例。在Kubernetes中，这通常被称为replication。 Pod中如何管理多个容器 Pod中可以同时运行多个进程（作为容器运行）协同工作。同一个Pod中的容器会自动的分配到同一个 node 上。同一个Pod中的容器共享资源、网络环境和依赖，它们总是被同时调度。 注意在一个Pod中同时运行多个容器是一种比较高级的用法。只有当你的容器需要紧密配合协作的时候才考虑用这种模式。例如，你有一个容器作为web服务器运行，需要用到共享的volume，有另一个“sidecar”容器来从远端获取资源更新这些文件，如下图所示： 图片 - pod diagram Pod中可以共享两种资源：网络和存储。 网络 每个Pod都会被分配一个唯一的IP地址。Pod中的所有容器共享网络空间，包括IP地址和端口。Pod内部的容器可以使用localhost互相通信。Pod中的容器与外界通信时，必须分配共享网络资源（例如使用宿主机的端口映射）。 存储 可以Pod指定多个共享的Volume。Pod中的所有容器都可以访问共享的volume。Volume也可以用来持久化Pod中的存储资源，以防容器重启后文件丢失。 使用Pod 你很少会直接在kubernetes中创建单个Pod。因为Pod的生命周期是短暂的，用后即焚的实体。当Pod被创建后（不论是由你直接创建还是被其他Controller），都会被Kuberentes调度到集群的Node上。直到Pod的进程终止、被删掉、因为缺少资源而被驱逐、或者Node故障之前这个Pod都会一直保持在那个Node上。 注意：重启Pod中的容器跟重启Pod不是一回事。Pod只提供容器的运行环境并保持容器的运行状态，重启容器不会造成Pod重启。 Pod不会自愈。如果Pod运行的Node故障，或者是调度器本身故障，这个Pod就会被删除。同样的，如果Pod所在Node缺少资源或者Pod处于维护状态，Pod也会被驱逐。Kubernetes使用更高级的称为Controller的抽象层，来管理Pod实例。虽然可以直接使用Pod，但是在Kubernetes中通常是使用Controller来管理Pod的。 Pod和Controller Controller可以创建和管理多个Pod，提供副本管理、滚动升级和集群级别的自愈能力。例如，如果一个Node故障，Controller就能自动将该节点上的Pod调度到其他健康的Node上。 包含一个或者多个Pod的Controller示例： Deployment StatefulSet DaemonSet 通常，Controller会用你提供的Pod Template来创建相应的Pod。 Pod Templates Pod模版是包含了其他object的Pod定义，例如Replication Controllers，Jobs和 DaemonSets。Controller根据Pod模板来创建实际的Pod。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-22 17:00:47 "},"concepts/pod.html":{"url":"concepts/pod.html","title":"Pod解析","keywords":"","body":"Pod解析 Pod是kubernetes中可以创建的最小部署单元。 V1 core版本的Pod的配置模板见Pod template。 什么是Pod？ Pod就像是豌豆荚一样，它由一个或者多个容器组成（例如Docker容器），它们共享容器存储、网络和容器运行配置项。Pod中的容器总是被同时调度，有共同的运行环境。你可以把单个Pod想象成是运行独立应用的“逻辑主机”——其中运行着一个或者多个紧密耦合的应用容器——在有容器之前，这些应用都是运行在几个相同的物理机或者虚拟机上。 尽管kubernetes支持多种容器运行时，但是Docker依然是最常用的运行时环境，我们可以使用Docker的术语和规则来定义Pod。 Pod中共享的环境包括Linux的namespace，cgroup和其他可能的隔绝环境，这一点跟Docker容器一致。在Pod的环境中，每个容器中可能还有更小的子隔离环境。 Pod中的容器共享IP地址和端口号，它们之间可以通过localhost互相发现。它们之间可以通过进程间通信，例如SystemV信号或者POSIX共享内存。不同Pod之间的容器具有不同的IP地址，不能直接通过IPC通信。 Pod中的容器也有访问共享volume的权限，这些volume会被定义成pod的一部分并挂载到应用容器的文件系统中。 根据Docker的结构，Pod中的容器共享namespace和volume，不支持共享PID的namespace。 就像每个应用容器，pod被认为是临时（非持久的）实体。在Pod的生命周期中讨论过，pod被创建后，被分配一个唯一的ID（UID），调度到节点上，并一致维持期望的状态直到被终结（根据重启策略）或者被删除。如果node死掉了，分配到了这个node上的pod，在经过一个超时时间后会被重新调度到其他node节点上。一个给定的pod（如UID定义的）不会被“重新调度”到新的节点上，而是被一个同样的pod取代，如果期望的话甚至可以是相同的名字，但是会有一个新的UID（查看replication controller获取详情）。（未来，一个更高级别的API将支持pod迁移）。 Volume跟pod有相同的生命周期（当其UID存在的时候）。当Pod因为某种原因被删除或者被新创建的相同的Pod取代，它相关的东西（例如volume）也会被销毁和再创建一个新的volume。 图片 - Pod示意图 A multi-container pod that contains a file puller and a web server that uses a persistent volume for shared storage between the containers. Pod的动机 管理 Pod是一个服务的多个进程的聚合单位，pod提供这种模型能够简化应用部署管理，通过提供一个更高级别的抽象的方式。Pod作为一个独立的部署单位，支持横向扩展和复制。共生（协同调度），命运共同体（例如被终结），协同复制，资源共享，依赖管理，Pod都会自动的为容器处理这些问题。 资源共享和通信 Pod中的应用可以共享网络空间（IP地址和端口），因此可以通过localhost互相发现。因此，pod中的应用必须协调端口占用。每个pod都有一个唯一的IP地址，跟物理机和其他pod都处于一个扁平的网络空间中，它们之间可以直接连通。 Pod中应用容器的hostname被设置成Pod的名字。 Pod中的应用容器可以共享volume。Volume能够保证pod重启时使用的数据不丢失。 Pod的使用 Pod也可以用于垂直应用栈（例如LAMP），这样使用的主要动机是为了支持共同调度和协调管理应用程序，例如： content management systems, file and data loaders, local cache managers, etc. log and checkpoint backup, compression, rotation, snapshotting, etc. data change watchers, log tailers, logging and monitoring adapters, event publishers, etc. proxies, bridges, and adapters controllers, managers, configurators, and updaters 通常单个pod中不会同时运行一个应用的多个实例。 详细说明请看： The Distributed System ToolKit: Patterns for Composite Containers. 其他替代选择 为什么不直接在一个容器中运行多个应用程序呢？ 透明。让Pod中的容器对基础设施可见，以便基础设施能够为这些容器提供服务，例如进程管理和资源监控。这可以为用户带来极大的便利。 解耦软件依赖。每个容器都可以进行版本管理，独立的编译和发布。未来kubernetes甚至可能支持单个容器的在线升级。 使用方便。用户不必运行自己的进程管理器，还要担心错误信号传播等。 效率。因为由基础架构提供更多的职责，所以容器可以变得更加轻量级。 为什么不支持容器的亲和性的协同调度？ 这种方法可以提供容器的协同定位，能够根据容器的亲和性进行调度，但是无法实现使用pod带来的大部分好处，例如资源共享，IPC，保持状态一致性和简化管理等。 Pod的持久性（或者说缺乏持久性） Pod在设计支持就不是作为持久化实体的。在调度失败、节点故障、缺少资源或者节点维护的状态下都会死掉会被驱逐。 通常，用户不需要手动直接创建Pod，而是应该使用controller（例如Deployments），即使是在创建单个Pod的情况下。Controller可以提供集群级别的自愈功能、复制和升级管理。 The use of collective APIs as the primary user-facing primitive is relatively common among cluster scheduling systems, including Borg, Marathon, Aurora, and Tupperware. Pod is exposed as a primitive in order to facilitate: scheduler and controller pluggability support for pod-level operations without the need to \"proxy\" them via controller APIs decoupling of pod lifetime from controller lifetime, such as for bootstrapping decoupling of controllers and services — the endpoint controller just watches pods clean composition of Kubelet-level functionality with cluster-level functionality — Kubelet is effectively the \"pod controller\" high-availability applications, which will expect pods to be replaced in advance of their termination and certainly in advance of deletion, such as in the case of planned evictions, image prefetching, or live pod migration #3949 StatefulSet controller（目前还是beta状态）支持有状态的Pod。在1.4版本中被称为PetSet。在kubernetes之前的版本中创建有状态pod的最佳方式是创建一个replica为1的replication controller。 Pod的终止 因为Pod作为在集群的节点上运行的进程，所以在不再需要的时候能够优雅的终止掉是十分必要的（比起使用发送KILL信号这种暴力的方式）。用户需要能够放松删除请求，并且知道它们何时会被终止，是否被正确的删除。用户想终止程序时发送删除pod的请求，在pod可以被强制删除前会有一个优雅删除的时间，会发送一个TERM请求到每个容器的主进程。一旦超时，将向主进程发送KILL信号并从API server中删除。如果kubelet或者container manager在等待进程终止的过程中重启，在重启后仍然会重试完整的优雅删除阶段。 示例流程如下： 用户发送删除pod的命令，默认优雅删除时期是30秒； 在Pod超过该优雅删除期限后API server就会更新Pod的状态为“dead”； 在客户端命令行上显示的Pod状态为“terminating”； 跟第三步同时，当kubelet发现pod被标记为“terminating”状态时，开始停止pod进程： 如果在pod中定义了preStop hook，在停止pod前会被调用。如果在优雅删除期限过期后，preStop hook依然在运行，第二步会再增加2秒的优雅时间； 向Pod中的进程发送TERM信号； 跟第三步同时，该Pod将从该service的端点列表中删除，不再是replication controller的一部分。关闭的慢的pod将继续处理load balancer转发的流量； 过了优雅周期后，将向Pod中依然运行的进程发送SIGKILL信号而杀掉进程。 Kublete会在API server中完成Pod的的删除，通过将优雅周期设置为0（立即删除）。Pod在API中消失，并且在客户端也不可见。 删除优雅周期默认是30秒。 kubectl delete命令支持 —grace-period= 选项，允许用户设置自己的优雅周期时间。如果设置为0将强制删除pod。在kubectl>=1.5版本的命令中，你必须同时使用 --force 和 --grace-period=0 来强制删除pod。 强制删除Pod Pod的强制删除是通过在集群和etcd中将其定义为删除状态。当执行强制删除命令时，API server不会等待该pod所运行在节点上的kubelet确认，就会立即将该pod从API server中移除，这时就可以创建跟原pod同名的pod了。这时，在节点上的pod会被立即设置为terminating状态，不过在被强制删除之前依然有一小段优雅删除周期。 强制删除对于某些pod具有潜在危险性，请谨慎使用。使用StatefulSet pod的情况下，请参考删除StatefulSet中的pod文章。 Pod中容器的特权模式 从kubernetes1.1版本开始，pod中的容器就可以开启previleged模式，在容器定义文件的 SecurityContext 下使用 privileged flag。 这在使用Linux的网络操作和访问设备的能力时是很用的。同时开启的特权模式的Pod中的容器也可以访问到容器外的进程和应用。在不需要修改和重新编译kubelet的情况下就可以使用pod来开发节点的网络和存储插件。 如果master节点运行的是kuberentes1.1或更高版本，而node节点的版本低于1.1版本，则API server将也可以接受新的特权模式的pod，但是无法启动，pod将处于pending状态。 执行 kubectl describe pod FooPodName，可以看到为什么pod处于pending状态。输出的event列表中将显示： Error validating pod \"FooPodName\".\"FooPodNamespace\" from api, ignoring: spec.containers[0].securityContext.privileged: forbidden '(0xc2089d3248)true' 如果master节点的版本低于1.1，无法创建特权模式的pod。如果你仍然试图去创建的话，你得到如下错误： The Pod \"FooPodName\" is invalid. spec.containers[0].securityContext.privileged: forbidden '(0xc20b222db0)true' API Object Pod是kubernetes REST API中的顶级资源类型。 在kuberentes1.6的V1 core API版本中的Pod的数据结构如下图所示： 图片 - Pod Cheatsheet Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-22 17:00:47 "},"concepts/init-containers.html":{"url":"concepts/init-containers.html","title":"Init容器","keywords":"","body":"Init 容器 该特性在 1.6 版本已经推出 beta 版本。Init 容器可以在 PodSpec 中同应用程序的 containers 数组一起来指定。 beta 注解的值将仍需保留，并覆盖 PodSpec 字段值。 本文讲解 Init 容器的基本概念，它是一种专用的容器，在应用程序容器启动之前运行，并包括一些应用镜像中不存在的实用工具和安装脚本。 理解 Init 容器 Pod 能够具有多个容器，应用运行在容器里面，但是它也可能有一个或多个先于应用容器启动的 Init 容器。 Init 容器与普通的容器非常像，除了如下两点： Init 容器总是运行到成功完成为止。 每个 Init 容器都必须在下一个 Init 容器启动之前成功完成。 如果 Pod 的 Init 容器失败，Kubernetes 会不断地重启该 Pod，直到 Init 容器成功为止。然而，如果 Pod 对应的 restartPolicy 为 Never，它不会重新启动。 指定容器为 Init 容器，在 PodSpec 中添加 initContainers 字段，以 v1.Container 类型对象的 JSON 数组的形式，还有 app 的 containers 数组。 Init 容器的状态在 status.initContainerStatuses 字段中以容器状态数组的格式返回（类似 status.containerStatuses 字段）。 与普通容器的不同之处 Init 容器支持应用容器的全部字段和特性，包括资源限制、数据卷和安全设置。 然而，Init 容器对资源请求和限制的处理稍有不同，在下面 资源 处有说明。 而且 Init 容器不支持 Readiness Probe，因为它们必须在 Pod 就绪之前运行完成。 如果为一个 Pod 指定了多个 Init 容器，那些容器会按顺序一次运行一个。 每个 Init 容器必须运行成功，下一个才能够运行。 当所有的 Init 容器运行完成时，Kubernetes 初始化 Pod 并像平常一样运行应用容器。 Init 容器能做什么？ 因为 Init 容器具有与应用程序容器分离的单独镜像，所以它们的启动相关代码具有如下优势： 它们可以包含并运行实用工具，但是出于安全考虑，是不建议在应用程序容器镜像中包含这些实用工具的。 它们可以包含使用工具和定制化代码来安装，但是不能出现在应用程序镜像中。例如，创建镜像没必要 FROM 另一个镜像，只需要在安装过程中使用类似 sed、 awk、 python 或 dig 这样的工具。 应用程序镜像可以分离出创建和部署的角色，而没有必要联合它们构建一个单独的镜像。 Init 容器使用 Linux Namespace，所以相对应用程序容器来说具有不同的文件系统视图。因此，它们能够具有访问 Secret 的权限，而应用程序容器则不能。 它们必须在应用程序容器启动之前运行完成，而应用程序容器是并行运行的，所以 Init 容器能够提供了一种简单的阻塞或延迟应用容器的启动的方法，直到满足了一组先决条件。 示例 下面是一些如何使用 Init 容器的想法： 等待一个 Service 创建完成，通过类似如下 shell 命令： for i in {1..100}; do sleep 1; if dig myservice; then exit 0; fi; exit 1 将 Pod 注册到远程服务器，通过在命令中调用 API，类似如下： curl -X POST http://$MANAGEMENT_SERVICE_HOST:$MANAGEMENT_SERVICE_PORT/register -d 'instance=$()&ip=$()' 在启动应用容器之前等一段时间，使用类似 sleep 60 的命令。 克隆 Git 仓库到数据卷。 将配置值放到配置文件中，运行模板工具为主应用容器动态地生成配置文件。例如，在配置文件中存放 POD_IP 值，并使用 Jinja 生成主应用配置文件。 更多详细用法示例，可以在 StatefulSet 文档 和 生产环境 Pod 指南 中找到。 使用 Init 容器 下面是 Kubernetes 1.5 版本 yaml 文件，展示了一个具有 2 个 Init 容器的简单 Pod。 第一个等待 myservice 启动，第二个等待 mydb 启动。 一旦这两个 Service 都启动完成，Pod 将开始启动。 apiVersion: v1 kind: Pod metadata: name: myapp-pod labels: app: myapp annotations: pod.beta.kubernetes.io/init-containers: '[ { \"name\": \"init-myservice\", \"image\": \"busybox\", \"command\": [\"sh\", \"-c\", \"until nslookup myservice; do echo waiting for myservice; sleep 2; done;\"] }, { \"name\": \"init-mydb\", \"image\": \"busybox\", \"command\": [\"sh\", \"-c\", \"until nslookup mydb; do echo waiting for mydb; sleep 2; done;\"] } ]' spec: containers: - name: myapp-container image: busybox command: ['sh', '-c', 'echo The app is running! && sleep 3600'] 这是 Kubernetes 1.6 版本的新语法，尽管老的 annotation 语法仍然可以使用。我们已经把 Init 容器的声明移到 spec 中： apiVersion: v1 kind: Pod metadata: name: myapp-pod labels: app: myapp spec: containers: - name: myapp-container image: busybox command: ['sh', '-c', 'echo The app is running! && sleep 3600'] initContainers: - name: init-myservice image: busybox command: ['sh', '-c', 'until nslookup myservice; do echo waiting for myservice; sleep 2; done;'] - name: init-mydb image: busybox command: ['sh', '-c', 'until nslookup mydb; do echo waiting for mydb; sleep 2; done;'] 1.5 版本的语法在 1.6 版本仍然可以使用，但是我们推荐使用 1.6 版本的新语法。 在 Kubernetes 1.6 版本中，Init 容器在 API 中新建了一个字段。 虽然期望使用 beta 版本的 annotation，但在未来发行版将会被废弃掉。 下面的 yaml 文件展示了 mydb 和 myservice 两个 Service： kind: Service apiVersion: v1 metadata: name: myservice spec: ports: - protocol: TCP port: 80 targetPort: 9376 --- kind: Service apiVersion: v1 metadata: name: mydb spec: ports: - protocol: TCP port: 80 targetPort: 9377 这个 Pod 可以使用下面的命令进行启动和调试： $ kubectl create -f myapp.yaml pod \"myapp-pod\" created $ kubectl get -f myapp.yaml NAME READY STATUS RESTARTS AGE myapp-pod 0/1 Init:0/2 0 6m $ kubectl describe -f myapp.yaml Name: myapp-pod Namespace: default [...] Labels: app=myapp Status: Pending [...] Init Containers: init-myservice: [...] State: Running [...] init-mydb: [...] State: Waiting Reason: PodInitializing Ready: False [...] Containers: myapp-container: [...] State: Waiting Reason: PodInitializing Ready: False [...] Events: FirstSeen LastSeen Count From SubObjectPath Type Reason Message --------- -------- ----- ---- ------------- -------- ------ ------- 16s 16s 1 {default-scheduler } Normal Scheduled Successfully assigned myapp-pod to 172.17.4.201 16s 16s 1 {kubelet 172.17.4.201} spec.initContainers{init-myservice} Normal Pulling pulling image \"busybox\" 13s 13s 1 {kubelet 172.17.4.201} spec.initContainers{init-myservice} Normal Pulled Successfully pulled image \"busybox\" 13s 13s 1 {kubelet 172.17.4.201} spec.initContainers{init-myservice} Normal Created Created container with docker id 5ced34a04634; Security:[seccomp=unconfined] 13s 13s 1 {kubelet 172.17.4.201} spec.initContainers{init-myservice} Normal Started Started container with docker id 5ced34a04634 $ kubectl logs myapp-pod -c init-myservice # Inspect the first init container $ kubectl logs myapp-pod -c init-mydb # Inspect the second init container 一旦我们启动了 mydb 和 myservice 这两个 Service，我们能够看到 Init 容器完成，并且 myapp-pod 被创建： $ kubectl create -f services.yaml service \"myservice\" created service \"mydb\" created $ kubectl get -f myapp.yaml NAME READY STATUS RESTARTS AGE myapp-pod 1/1 Running 0 9m 这个例子非常简单，但是应该能够为我们创建自己的 Init 容器提供一些启发。 具体行为 在 Pod 启动过程中，Init 容器会按顺序在网络和数据卷初始化之后启动。 每个容器必须在下一个容器启动之前成功退出。 如果由于运行时或失败退出，导致容器启动失败，它会根据 Pod 的 restartPolicy 指定的策略进行重试。 然而，如果 Pod 的 restartPolicy 设置为 Always，Init 容器失败时会使用 RestartPolicy 策略。 在所有的 Init 容器没有成功之前，Pod 将不会变成 Ready 状态。 Init 容器的端口将不会在 Service 中进行聚集。 正在初始化中的 Pod 处于 Pending 状态，但应该会将条件 Initializing 设置为 true。 如果 Pod 重启，所有 Init 容器必须重新执行。 对 Init 容器 spec 的修改，被限制在容器 image 字段中。 更改 Init 容器的 image 字段，等价于重启该 Pod。 因为 Init 容器可能会被重启、重试或者重新执行，所以 Init 容器的代码应该是幂等的。 特别地，被写到 EmptyDirs 中文件的代码，应该对输出文件可能已经存在做好准备。 Init 容器具有应用容器的所有字段。 除了 readinessProbe，因为 Init 容器无法定义不同于完成（completion）的就绪（readiness）的之外的其他状态。 这会在验证过程中强制执行。 在 Pod 上使用 activeDeadlineSeconds，在容器上使用 livenessProbe，这样能够避免 Init 容器一直失败。 这就为 Init 容器活跃设置了一个期限。 在 Pod 中的每个 app 和 Init 容器的名称必须唯一；与任何其它容器共享同一个名称，会在验证时抛出错误。 资源 为 Init 容器指定顺序和执行逻辑，下面对资源使用的规则将被应用： 在所有 Init 容器上定义的，任何特殊资源请求或限制的最大值，是 有效初始请求/限制 Pod 对资源的有效请求/限制要高于： 所有应用容器对某个资源的请求/限制之和 对某个资源的有效初始请求/限制 基于有效请求/限制完成调度，这意味着 Init 容器能够为初始化预留资源，这些资源在 Pod 生命周期过程中并没有被使用。 Pod 的 有效 QoS 层，是 Init 容器和应用容器相同的 QoS 层。 基于有效 Pod 请求和限制来应用配额和限制。 Pod 级别的 cgroups 是基于有效 Pod 请求和限制，和调度器相同。 Pod 重启的原因 Pod 能够重启，会导致 Init 容器重新执行，主要有如下几个原因： 用户更新 PodSpec 导致 Init 容器镜像发生改变。应用容器镜像的变更只会重启应用容器。 Pod 基础设施容器被重启。这不多见，但某些具有 root 权限可访问 Node 的人可能会这样做。 当 restartPolicy 设置为 Always，Pod 中所有容器会终止，强制重启，由于垃圾收集导致 Init 容器完整的记录丢失。 支持与兼容性 Apiserver 版本为 1.6 或更高版本的集群，通过使用 spec.initContainers 字段来支持 Init 容器。 之前的版本可以使用 alpha 和 beta 注解支持 Init 容器。 spec.initContainers 字段也被加入到 alpha 和 beta 注解中，所以 Kubernetes 1.3.0 版本或更高版本可以执行 Init 容器，并且 1.6 版本的 apiserver 能够安全地回退到 1.5.x 版本，而不会使存在的已创建 Pod 失去 Init 容器的功能。 原文地址：https://k8smeetup.github.io/docs/concepts/workloads/pods/init-containers/ 译者：shirdrn Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-22 17:00:47 "},"concepts/pod-security-policy.html":{"url":"concepts/pod-security-policy.html","title":"Pod安全策略","keywords":"","body":"Pod 安全策略 PodSecurityPolicy 类型的对象能够控制，是否可以向 Pod 发送请求，该 Pod 能够影响被应用到 Pod 和容器的 SecurityContext。 查看 Pod 安全策略建议 获取更多信息。 什么是 Pod 安全策略？ Pod 安全策略 是集群级别的资源，它能够控制 Pod 运行的行为，以及它具有访问什么的能力。 PodSecurityPolicy对象定义了一组条件，指示 Pod 必须按系统所能接受的顺序运行。 它们允许管理员控制如下方面： 控制面 字段名称 已授权容器的运行 privileged 为容器添加默认的一组能力 defaultAddCapabilities 为容器去掉某些能力 requiredDropCapabilities 容器能够请求添加某些能力 allowedCapabilities 控制卷类型的使用 volumes 主机网络的使用 hostNetwork 主机端口的使用 hostPorts 主机 PID namespace 的使用 hostPID 主机 IPC namespace 的使用 hostIPC 主机路径的使用 allowedHostPaths 容器的 SELinux 上下文 seLinux 用户 ID runAsUser 配置允许的补充组 supplementalGroups 分配拥有 Pod 数据卷的 FSGroup fsGroup 必须使用一个只读的 root 文件系统 readOnlyRootFilesystem Pod 安全策略 由设置和策略组成，它们能够控制 Pod 访问的安全特征。这些设置分为如下三类： 基于布尔值控制：这种类型的字段默认为最严格限制的值。 基于被允许的值集合控制：这种类型的字段会与这组值进行对比，以确认值被允许。 基于策略控制：设置项通过一种策略提供的机制来生成该值，这种机制能够确保指定的值落在被允许的这组值中。 RunAsUser MustRunAs - 必须配置一个 range。使用该范围内的第一个值作为默认值。验证是否不在配置的该范围内。 MustRunAsNonRoot - 要求提交的 Pod 具有非零 runAsUser 值，或在镜像中定义了 USER 环境变量。不提供默认值。 RunAsAny - 没有提供默认值。允许指定任何 runAsUser 。 SELinux MustRunAs - 如果没有使用预分配的值，必须配置 seLinuxOptions。默认使用 seLinuxOptions。验证 seLinuxOptions。 RunAsAny - 没有提供默认值。允许任意指定的 seLinuxOptions ID。 SupplementalGroups MustRunAs - 至少需要指定一个范围。默认使用第一个范围的最小值。验证所有范围的值。 RunAsAny - 没有提供默认值。允许任意指定的 supplementalGroups ID。 FSGroup MustRunAs - 至少需要指定一个范围。默认使用第一个范围的最小值。验证在第一个范围内的第一个 ID。 RunAsAny - 没有提供默认值。允许任意指定的 fsGroup ID。 控制卷 通过设置 PSP 卷字段，能够控制具体卷类型的使用。当创建一个卷的时候，与该字段相关的已定义卷可以允许设置如下值： azureFile azureDisk flocker flexVolume hostPath emptyDir gcePersistentDisk awsElasticBlockStore gitRepo secret nfs iscsi glusterfs persistentVolumeClaim rbd cinder cephFS downwardAPI fc configMap vsphereVolume quobyte photonPersistentDisk projected portworxVolume scaleIO storageos * (allow all volumes) 对新的 PSP，推荐允许的卷的最小集合包括：configMap、downwardAPI、emptyDir、persistentVolumeClaim、secret 和 projected。 主机网络 HostPorts， 默认为 empty。HostPortRange 列表通过 min(包含) and max(包含) 来定义，指定了被允许的主机端口。 允许的主机路径 AllowedHostPaths 是一个被允许的主机路径前缀的白名单。空值表示所有的主机路径都可以使用。 许可 包含 PodSecurityPolicy 的 许可控制，允许控制集群资源的创建和修改，基于这些资源在集群范围内被许可的能力。 许可使用如下的方式为 Pod 创建最终的安全上下文： 检索所有可用的 PSP。 生成在请求中没有指定的安全上下文设置的字段值。 基于可用的策略，验证最终的设置。 如果某个策略能够匹配上，该 Pod 就被接受。如果请求与 PSP 不匹配，则 Pod 被拒绝。 Pod 必须基于 PSP 验证每个字段。 创建 Pod 安全策略 下面是一个 Pod 安全策略的例子，所有字段的设置都被允许： apiVersion: extensions/v1beta1 kind: PodSecurityPolicy metadata: name: permissive spec: seLinux: rule: RunAsAny supplementalGroups: rule: RunAsAny runAsUser: rule: RunAsAny fsGroup: rule: RunAsAny hostPorts: - min: 8000 max: 8080 volumes: - '*' 下载示例文件可以创建该策略，然后执行如下命令： $ kubectl create -f ./psp.yaml podsecuritypolicy \"permissive\" created 获取 Pod 安全策略列表 获取已存在策略列表，使用 kubectl get： $ kubectl get psp NAME PRIV CAPS SELINUX RUNASUSER FSGROUP SUPGROUP READONLYROOTFS VOLUMES permissive false [] RunAsAny RunAsAny RunAsAny RunAsAny false [*] privileged true [] RunAsAny RunAsAny RunAsAny RunAsAny false [*] restricted false [] RunAsAny MustRunAsNonRoot RunAsAny RunAsAny false [emptyDir secret downwardAPI configMap persistentVolumeClaim projected] 修改 Pod 安全策略 通过交互方式修改策略，使用 kubectl edit： $ kubectl edit psp permissive 该命令将打开一个默认文本编辑器，在这里能够修改策略。 删除 Pod 安全策略 一旦不再需要一个策略，很容易通过 kubectl 删除它： $ kubectl delete psp permissive podsecuritypolicy \"permissive\" deleted 启用 Pod 安全策略 为了能够在集群中使用 Pod 安全策略，必须确保如下： 启用 API 类型 extensions/v1beta1/podsecuritypolicy（仅对 1.6 之前的版本） 启用许可控制器 PodSecurityPolicy 定义自己的策略 使用 RBAC 在 Kubernetes 1.5 或更新版本，可以使用 PodSecurityPolicy 来控制，对基于用户角色和组的已授权容器的访问。访问不同的 PodSecurityPolicy 对象，可以基于认证来控制。基于 Deployment、ReplicaSet 等创建的 Pod，限制访问 PodSecurityPolicy 对象，Controller Manager 必须基于安全 API 端口运行，并且不能够具有超级用户权限。 PodSecurityPolicy 认证使用所有可用的策略，包括创建 Pod 的用户，Pod 上指定的服务账户（service acount）。当 Pod 基于 Deployment、ReplicaSet 创建时，它是创建 Pod 的 Controller Manager，所以如果基于非安全 API 端口运行，允许所有的 PodSecurityPolicy 对象，并且不能够有效地实现细分权限。用户访问给定的 PSP 策略有效，仅当是直接部署 Pod 的情况。更多详情，查看 PodSecurityPolicy RBAC 示例，当直接部署 Pod 时，应用 PodSecurityPolicy 控制基于角色和组的已授权容器的访问 。 原文地址：https://k8smeetup.github.io/docs/concepts/policy/pod-security-policy/ 译者：shirdrn Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-21 11:59:28 "},"concepts/pod-lifecycle.html":{"url":"concepts/pod-lifecycle.html","title":"Pod的生命周期","keywords":"","body":"Pod 的生命周期 Pod phase Pod 的 status 在信息保存在 PodStatus 中定义，其中有一个 phase 字段。 Pod 的相位（phase）是 Pod 在其生命周期中的简单宏观概述。该阶段并不是对容器或 Pod 的综合汇总，也不是为了做为综合状态机。 Pod 相位的数量和含义是严格指定的。除了本文档中列举的状态外，不应该再假定 Pod 有其他的 phase 值。 下面是 phase 可能的值： 挂起（Pending）：Pod 已被 Kubernetes 系统接受，但有一个或者多个容器镜像尚未创建。等待时间包括调度 Pod 的时间和通过网络下载镜像的时间，这可能需要花点时间。 运行中（Running）：该 Pod 已经绑定到了一个节点上，Pod 中所有的容器都已被创建。至少有一个容器正在运行，或者正处于启动或重启状态。 成功（Successed）：Pod 中的所有容器都被成功终止，并且不会再重启。 失败（Failed）：Pod 中的所有容器都已终止了，并且至少有一个容器是因为失败终止。也就是说，容器以非0状态退出或者被系统终止。 未知（Unkonwn）：因为某些原因无法取得 Pod 的状态，通常是因为与 Pod 所在主机通信失败。 Pod 状态 Pod 有一个 PodStatus 对象，其中包含一个 PodCondition 数组。 PodCondition 数组的每个元素都有一个 type 字段和一个 status 字段。type 字段是字符串，可能的值有 PodScheduled、Ready、Initialized 和 Unschedulable。status 字段是一个字符串，可能的值有 True、False 和 Unknown。 容器探针 探针 是由 kubelet 对容器执行的定期诊断。要执行诊断，kubelet 调用由容器实现的 Handler。有三种类型的处理程序： ExecAction：在容器内执行指定命令。如果命令退出时返回码为 0 则认为诊断成功。 TCPSocketAction：对指定端口上的容器的 IP 地址进行 TCP 检查。如果端口打开，则诊断被认为是成功的。 HTTPGetAction：对指定的端口和路径上的容器的 IP 地址执行 HTTP Get 请求。如果响应的状态码大于等于200 且小于 400，则诊断被认为是成功的。 每次探测都将获得以下三种结果之一： 成功：容器通过了诊断。 失败：容器未通过诊断。 未知：诊断失败，因此不会采取任何行动。 Kubelet 可以选择是否执行在容器上运行的两种探针执行和做出反应： livenessProbe：指示容器是否正在运行。如果存活探测失败，则 kubelet 会杀死容器，并且容器将受到其 重启策略 的影响。如果容器不提供存活探针，则默认状态为 Success。 readinessProbe：指示容器是否准备好服务请求。如果就绪探测失败，端点控制器将从与 Pod 匹配的所有 Service 的端点中删除该 Pod 的 IP 地址。初始延迟之前的就绪状态默认为 Failure。如果容器不提供就绪探针，则默认状态为 Success。 该什么时候使用存活（liveness）和就绪（readiness）探针? 如果容器中的进程能够在遇到问题或不健康的情况下自行崩溃，则不一定需要存活探针; kubelet 将根据 Pod 的restartPolicy 自动执行正确的操作。 如果您希望容器在探测失败时被杀死并重新启动，那么请指定一个存活探针，并指定restartPolicy 为 Always 或 OnFailure。 如果要仅在探测成功时才开始向 Pod 发送流量，请指定就绪探针。在这种情况下，就绪探针可能与存活探针相同，但是 spec 中的就绪探针的存在意味着 Pod 将在没有接收到任何流量的情况下启动，并且只有在探针探测成功后才开始接收流量。 如果您希望容器能够自行维护，您可以指定一个就绪探针，该探针检查与存活探针不同的端点。 请注意，如果您只想在 Pod 被删除时能够排除请求，则不一定需要使用就绪探针；在删除 Pod 时，Pod 会自动将自身置于未完成状态，无论就绪探针是否存在。当等待 Pod 中的容器停止时，Pod 仍处于未完成状态。 Pod 和容器状态 有关 Pod 容器状态的详细信息，请参阅 PodStatus 和 ContainerStatus。请注意，报告的 Pod 状态信息取决于当前的 ContainerState。 重启策略 PodSpec 中有一个 restartPolicy 字段，可能的值为 Always、OnFailure 和 Never。默认为 Always。 restartPolicy 适用于 Pod 中的所有容器。restartPolicy 仅指通过同一节点上的 kubelet 重新启动容器。失败的容器由 kubelet 以五分钟为上限的指数退避延迟（10秒，20秒，40秒...）重新启动，并在成功执行十分钟后重置。如 Pod 文档 中所述，一旦绑定到一个节点，Pod 将永远不会重新绑定到另一个节点。 Pod 的生命 一般来说，Pod 不会消失，直到人为销毁他们。这可能是一个人或控制器。这个规则的唯一例外是成功或失败的 phase 超过一段时间（由 master 确定）的Pod将过期并被自动销毁。 有三种可用的控制器： 使用 Job 运行预期会终止的 Pod，例如批量计算。Job 仅适用于重启策略为 OnFailure 或 Never 的 Pod。 对预期不会终止的 Pod 使用 ReplicationController、ReplicaSet 和 Deployment ，例如 Web 服务器。 ReplicationController 仅适用于具有 restartPolicy 为 Always 的 Pod。 提供特定于机器的系统服务，使用 DaemonSet 为每台机器运行一个 Pod 。 所有这三种类型的控制器都包含一个 PodTemplate。建议创建适当的控制器，让它们来创建 Pod，而不是直接自己创建 Pod。这是因为单独的 Pod 在机器故障的情况下没有办法自动复原，而控制器却可以。 如果节点死亡或与集群的其余部分断开连接，则 Kubernetes 将应用一个策略将丢失节点上的所有 Pod 的 phase 设置为 Failed。 示例 高级 liveness 探针示例 存活探针由 kubelet 来执行，因此所有的请求都在 kubelet 的网络命名空间中进行。 apiVersion: v1 kind: Pod metadata: labels: test: liveness name: liveness-http spec: containers: - args: - /server image: gcr.io/google_containers/liveness livenessProbe: httpGet: # when \"host\" is not defined, \"PodIP\" will be used # host: my-host # when \"scheme\" is not defined, \"HTTP\" scheme will be used. Only \"HTTP\" and \"HTTPS\" are allowed # scheme: HTTPS path: /healthz port: 8080 httpHeaders: - name: X-Custom-Header value: Awesome initialDelaySeconds: 15 timeoutSeconds: 1 name: liveness 状态示例 Pod 中只有一个容器并且正在运行。容器成功退出。 记录完成事件。 如果 restartPolicy 为： Always：重启容器；Pod phase 仍为 Running。 OnFailure：Pod phase 变成 Succeeded。 Never：Pod phase 变成 Succeeded。 Pod 中只有一个容器并且正在运行。容器退出失败。 记录失败事件。 如果 restartPolicy 为： Always：重启容器；Pod phase 仍为 Running。 OnFailure：重启容器；Pod phase 仍为 Running。 Never：Pod phase 变成 Failed。 Pod 中有两个容器并且正在运行。有一个容器退出失败。 记录失败事件。 如果 restartPolicy 为： Always：重启容器；Pod phase 仍为 Running。 OnFailure：重启容器；Pod phase 仍为 Running。 Never：不重启容器；Pod phase 仍为 Running。 如果有一个容器没有处于运行状态，并且两个容器退出： 记录失败事件。 如果 restartPolicy 为： Always：重启容器；Pod phase 仍为 Running。 OnFailure：重启容器；Pod phase 仍为 Running。 Never：Pod phase 变成 Failed。 Pod 中只有一个容器并处于运行状态。容器运行时内存超出限制： 容器以失败状态终止。 记录 OOM 事件。 如果 restartPolicy 为： Always：重启容器；Pod phase 仍为 Running。 OnFailure：重启容器；Pod phase 仍为 Running。 Never: 记录失败事件；Pod phase 仍为 Failed。 Pod 正在运行，磁盘故障： 杀掉所有容器。 记录适当事件。 Pod phase 变成 Failed。 如果使用控制器来运行，Pod 将在别处重建。 Pod 正在运行，其节点被分段。 节点控制器等待直到超时。 节点控制器将 Pod phase 设置为 Failed。 如果是用控制器来运行，Pod 将在别处重建。 原文地址：https://kubernetes.io/docs/concepts/workloads/pods/pod-lifecycle/ 翻译：rootsongjc Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-09-19 15:45:02 "},"concepts/pod-hook.html":{"url":"concepts/pod-hook.html","title":"Pod hook","keywords":"","body":"Pod hook Pod hook（钩子）是由Kubernetes管理的kubelet发起的，当容器中的进程启动前或者容器中的进程终止之前运行，这是包含在容器的生命周期之中。可以同时为Pod中的所有容器都配置hook。 Hook的类型包括两种： exec：执行一段命令 HTTP：发送HTTP请求。 参考下面的配置： apiVersion: v1 kind: Pod metadata: name: lifecycle-demo spec: containers: - name: lifecycle-demo-container image: nginx lifecycle: postStart: exec: command: [\"/bin/sh\", \"-c\", \"echo Hello from the postStart handler > /usr/share/message\"] preStop: exec: command: [\"/usr/sbin/nginx\",\"-s\",\"quit\"] 在容器创建之后，容器的Entrypoint执行之前，这时候Pod已经被调度到某台node上，被某个kubelet管理了，这时候kubelet会调用postStart操作，该操作跟容器的启动命令是在异步执行的，也就是说在postStart操作执行完成之前，kubelet会锁住容器，不让应用程序的进程启动，只有在 postStart操作完成之后容器的状态才会被设置成为RUNNING。 如果postStart或者preStop hook失败，将会终止容器。 调试hook Hook调用的日志没有暴露个给Pod的event，所以只能通过describe命令来获取，如果有错误将可以看到FailedPostStartHook或FailedPreStopHook这样的event。 参考 Attach Handlers to Container Lifecycle Events Container Lifecycle Hooks Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-23 16:33:50 "},"concepts/pod-preset.html":{"url":"concepts/pod-preset.html","title":"Pod Preset","keywords":"","body":"Pod Preset 注意：PodPreset 资源对象只有 kubernetes 1.8 以上版本才支持。 Preset 就是预设，有时候想要让一批容器在启动的时候就注入一些信息，比如 secret、volume、volume mount 和环境变量，而又不想一个一个的改这些 Pod 的 tmeplate，这时候就可以用到 PodPreset 这个资源对象了。 本页是关于 PodPreset 的概述，该对象用来在 Pod 创建的时候向 Pod 中注入某些特定信息。该信息可以包括 secret、volume、volume mount 和环境变量。 理解 Pod Preset Pod Preset 是用来在 Pod 被创建的时候向其中注入额外的运行时需求的 API 资源。 您可以使用 label selector 来指定为哪些 Pod 应用 Pod Preset。 使用 Pod Preset 使得 pod 模板的作者可以不必为每个 Pod 明确提供所有信息。这样一来，pod 模板的作者就不需要知道关于该服务的所有细节。 关于该背景的更多信息，请参阅 PodPreset 的设计方案。 如何工作 Kubernetes 提供了一个准入控制器（PodPreset），当其启用时，Pod Preset 会将应用创建请求传入到该控制器上。当有 Pod 创建请求发生时，系统将执行以下操作： 检索所有可用的 PodPresets。 检查 PodPreset 标签选择器上的标签，看看其是否能够匹配正在创建的 Pod 上的标签。 尝试将由 PodPreset 定义的各种资源合并到正在创建的 Pod 中。 出现错误时，在该 Pod 上引发记录合并错误的事件，PodPreset 不会注入任何资源到创建的 Pod 中。 注释刚生成的修改过的 Pod spec，以表明它已被 PodPreset 修改过。注释的格式为 podpreset.admission.kubernetes.io/podpreset-\": \"\"。 每个 Pod 可以匹配零个或多个 Pod Prestet；并且每个 PodPreset 可以应用于零个或多个 Pod。 PodPreset 应用于一个或多个 Pod 时，Kubernetes 会修改 Pod Spec。对于 Env、EnvFrom 和 VolumeMounts 的更改，Kubernetes 修改 Pod 中所有容器的容器 spec；对于 Volume 的更改，Kubernetes 修改 Pod Spec。 注意：Pod Preset 可以在适当的时候修改 Pod spec 中的 spec.containers 字段。Pod Preset 中的资源定义将不会应用于 initContainers 字段。 禁用特定 Pod 的 Pod Preset 在某些情况下，您可能不希望 Pod 被任何 Pod Preset 所改变。在这些情况下，您可以在 Pod 的 Pod Spec 中添加注释：podpreset.admission.kubernetes.io/exclude：\"true\"。 启用 Pod Preset 为了在群集中使用 Pod Preset，您必须确保以下内容： 您已启用 settings.k8s.io/v1alpha1/podpreset API 类型。例如，可以通过在 API server 的 --runtime-config 选项中包含 settings.k8s.io/v1alpha1=true 来完成此操作。 您已启用 PodPreset 准入控制器。 一种方法是将 PodPreset 包含在为 API server 指定的 --admission-control 选项值中。 您已经在要使用的命名空间中通过创建 PodPreset 对象来定义 PodPreset。 更多资料 使用 PodPreset 向 Pod 中注入数据 本文为 Kubernetes 官方中文文档，地址：https://kubernetes.io/cn/docs/concepts/workloads/pods/podpreset/ 翻译: rootsongjc Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-12 10:49:11 "},"concepts/node.html":{"url":"concepts/node.html","title":"Node","keywords":"","body":"Node Node是kubernetes集群的工作节点，可以是物理机也可以是虚拟机。 Node的状态 Node包括如下状态信息： Address HostName：可以被kubelet中的--hostname-override参数替代。 ExternalIP：可以被集群外部路由到的IP地址。 InternalIP：集群内部使用的IP，集群外部无法访问。 Condition OutOfDisk：磁盘空间不足时为True Ready：Node controller 40秒内没有收到node的状态报告为Unknown，健康为True，否则为False。 MemoryPressure：当node没有内存压力时为True，否则为False。 DiskPressure：当node没有磁盘压力时为True，否则为False。 Capacity CPU 内存 可运行的最大Pod个数 Info：节点的一些版本信息，如OS、kubernetes、docker等 Node管理 禁止pod调度到该节点上 kubectl cordon 驱逐该节点上的所有pod kubectl drain 该命令会删除该节点上的所有Pod（DaemonSet除外），在其他node上重新启动它们，通常该节点需要维护时使用该命令。直接使用该命令会自动调用kubectl cordon 命令。当该节点维护完成，启动了kubelet后，再使用kubectl uncordon 即可将该节点添加到kubernetes集群中。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-10-16 10:50:25 "},"concepts/namespace.html":{"url":"concepts/namespace.html","title":"Namespace","keywords":"","body":"Namespace 在一个Kubernetes集群中可以使用namespace创建多个“虚拟集群”，这些namespace之间可以完全隔离，也可以通过某种方式，让一个namespace中的service可以访问到其他的namespace中的服务，我们在CentOS中部署kubernetes1.6集群的时候就用到了好几个跨越namespace的服务，比如Traefik ingress和kube-systemnamespace下的service就可以为整个集群提供服务，这些都需要通过RBAC定义集群级别的角色来实现。 哪些情况下适合使用多个namesapce 因为namespace可以提供独立的命名空间，因此可以实现部分的环境隔离。当你的项目和人员众多的时候可以考虑根据项目属性，例如生产、测试、开发划分不同的namespace。 Namespace使用 获取集群中有哪些namespace kubectl get ns 集群中默认会有default和kube-system这两个namespace。 在执行kubectl命令时可以使用-n指定操作的namespace。 用户的普通应用默认是在default下，与集群管理相关的为整个集群提供服务的应用一般部署在kube-system的namespace下，例如我们在安装kubernetes集群时部署的kubedns、heapseter、EFK等都是在这个namespace下面。 另外，并不是所有的资源对象都会对应namespace，node和persistentVolume就不属于任何namespace。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-19 17:50:18 "},"concepts/service.html":{"url":"concepts/service.html","title":"Service","keywords":"","body":"Service Kubernetes Pod 是有生命周期的，它们可以被创建，也可以被销毁，然而一旦被销毁生命就永远结束。 通过 ReplicationController 能够动态地创建和销毁 Pod（例如，需要进行扩缩容，或者执行 滚动升级）。 每个 Pod 都会获取它自己的 IP 地址，即使这些 IP 地址不总是稳定可依赖的。 这会导致一个问题：在 Kubernetes 集群中，如果一组 Pod（称为 backend）为其它 Pod （称为 frontend）提供服务，那么那些 frontend 该如何发现，并连接到这组 Pod 中的哪些 backend 呢？ 关于 Service Kubernetes Service 定义了这样一种抽象：一个 Pod 的逻辑分组，一种可以访问它们的策略 —— 通常称为微服务。 这一组 Pod 能够被 Service 访问到，通常是通过 Label Selector（查看下面了解，为什么可能需要没有 selector 的 Service）实现的。 举个例子，考虑一个图片处理 backend，它运行了3个副本。这些副本是可互换的 —— frontend 不需要关心它们调用了哪个 backend 副本。 然而组成这一组 backend 程序的 Pod 实际上可能会发生变化，frontend 客户端不应该也没必要知道，而且也不需要跟踪这一组 backend 的状态。 Service 定义的抽象能够解耦这种关联。 对 Kubernetes 集群中的应用，Kubernetes 提供了简单的 Endpoints API，只要 Service 中的一组 Pod 发生变更，应用程序就会被更新。 对非 Kubernetes 集群中的应用，Kubernetes 提供了基于 VIP 的网桥的方式访问 Service，再由 Service 重定向到 backend Pod。 定义 Service 一个 Service 在 Kubernetes 中是一个 REST 对象，和 Pod 类似。 像所有的 REST 对象一样， Service 定义可以基于 POST 方式，请求 apiserver 创建新的实例。 例如，假定有一组 Pod，它们对外暴露了 9376 端口，同时还被打上 \"app=MyApp\" 标签。 kind: Service apiVersion: v1 metadata: name: my-service spec: selector: app: MyApp ports: - protocol: TCP port: 80 targetPort: 9376 上述配置将创建一个名称为 “my-service” 的 Service 对象，它会将请求代理到使用 TCP 端口 9376，并且具有标签 \"app=MyApp\" 的 Pod 上。 这个 Service 将被指派一个 IP 地址（通常称为 “Cluster IP”），它会被服务的代理使用（见下面）。 该 Service 的 selector 将会持续评估，处理结果将被 POST 到一个名称为 “my-service” 的 Endpoints 对象上。 需要注意的是， Service 能够将一个接收端口映射到任意的 targetPort。 默认情况下，targetPort 将被设置为与 port 字段相同的值。 可能更有趣的是，targetPort 可以是一个字符串，引用了 backend Pod 的一个端口的名称。 但是，实际指派给该端口名称的端口号，在每个 backend Pod 中可能并不相同。 对于部署和设计 Service ，这种方式会提供更大的灵活性。 例如，可以在 backend 软件下一个版本中，修改 Pod 暴露的端口，并不会中断客户端的调用。 Kubernetes Service 能够支持 TCP 和 UDP 协议，默认 TCP 协议。 没有 selector 的 Service Servcie 抽象了该如何访问 Kubernetes Pod，但也能够抽象其它类型的 backend，例如： 希望在生产环境中使用外部的数据库集群，但测试环境使用自己的数据库。 希望服务指向另一个 Namespace 中或其它集群中的服务。 正在将工作负载转移到 Kubernetes 集群，和运行在 Kubernetes 集群之外的 backend。 在任何这些场景中，都能够定义没有 selector 的 Service ： kind: Service apiVersion: v1 metadata: name: my-service spec: ports: - protocol: TCP port: 80 targetPort: 9376 由于这个 Service 没有 selector，就不会创建相关的 Endpoints 对象。可以手动将 Service 映射到指定的 Endpoints： kind: Endpoints apiVersion: v1 metadata: name: my-service subsets: - addresses: - ip: 1.2.3.4 ports: - port: 9376 注意：Endpoint IP 地址不能是 loopback（127.0.0.0/8）、 link-local（169.254.0.0/16）、或者 link-local 多播（224.0.0.0/24）。 访问没有 selector 的 Service，与有 selector 的 Service 的原理相同。请求将被路由到用户定义的 Endpoint（该示例中为 1.2.3.4:9376）。 ExternalName Service 是 Service 的特例，它没有 selector，也没有定义任何的端口和 Endpoint。 相反地，对于运行在集群外部的服务，它通过返回该外部服务的别名这种方式来提供服务。 kind: Service apiVersion: v1 metadata: name: my-service namespace: prod spec: type: ExternalName externalName: my.database.example.com 当查询主机 my-service.prod.svc.CLUSTER时，集群的 DNS 服务将返回一个值为 my.database.example.com 的 CNAME 记录。 访问这个服务的工作方式与其它的相同，唯一不同的是重定向发生在 DNS 层，而且不会进行代理或转发。 如果后续决定要将数据库迁移到 Kubernetes 集群中，可以启动对应的 Pod，增加合适的 Selector 或 Endpoint，修改 Service 的 type。 VIP 和 Service 代理 在 Kubernetes 集群中，每个 Node 运行一个 kube-proxy 进程。kube-proxy 负责为 Service 实现了一种 VIP（虚拟 IP）的形式，而不是 ExternalName 的形式。 在 Kubernetes v1.0 版本，代理完全在 userspace。在 Kubernetes v1.1 版本，新增了 iptables 代理，但并不是默认的运行模式。 从 Kubernetes v1.2 起，默认就是 iptables 代理。 在 Kubernetes v1.0 版本，Service 是 “4层”（TCP/UDP over IP）概念。 在 Kubernetes v1.1 版本，新增了 Ingress API（beta 版），用来表示 “7层”（HTTP）服务。 userspace 代理模式 这种模式，kube-proxy 会监视 Kubernetes master 对 Service 对象和 Endpoints 对象的添加和移除。 对每个 Service，它会在本地 Node 上打开一个端口（随机选择）。 任何连接到“代理端口”的请求，都会被代理到 Service 的backend Pods 中的某个上面（如 Endpoints 所报告的一样）。 使用哪个 backend Pod，是基于 Service 的 SessionAffinity 来确定的。 最后，它安装 iptables 规则，捕获到达该 Service 的 clusterIP（是虚拟 IP）和 Port 的请求，并重定向到代理端口，代理端口再代理请求到 backend Pod。 网络返回的结果是，任何到达 Service 的 IP:Port 的请求，都会被代理到一个合适的 backend，不需要客户端知道关于 Kubernetes、Service、或 Pod 的任何信息。 默认的策略是，通过 round-robin 算法来选择 backend Pod。 实现基于客户端 IP 的会话亲和性，可以通过设置 service.spec.sessionAffinity 的值为 \"ClientIP\" （默认值为 \"None\"）。 图片 - userspace代理模式下Service概览图 iptables 代理模式 这种模式，kube-proxy 会监视 Kubernetes master 对 Service 对象和 Endpoints 对象的添加和移除。 对每个 Service，它会安装 iptables 规则，从而捕获到达该 Service 的 clusterIP（虚拟 IP）和端口的请求，进而将请求重定向到 Service 的一组 backend 中的某个上面。 对于每个 Endpoints 对象，它也会安装 iptables 规则，这个规则会选择一个 backend Pod。 默认的策略是，随机选择一个 backend。 实现基于客户端 IP 的会话亲和性，可以将 service.spec.sessionAffinity 的值设置为 \"ClientIP\" （默认值为 \"None\"）。 和 userspace 代理类似，网络返回的结果是，任何到达 Service 的 IP:Port 的请求，都会被代理到一个合适的 backend，不需要客户端知道关于 Kubernetes、Service、或 Pod 的任何信息。 这应该比 userspace 代理更快、更可靠。然而，不像 userspace 代理，如果初始选择的 Pod 没有响应，iptables 代理能够自动地重试另一个 Pod，所以它需要依赖 readiness probes。 图片 - iptables代理模式下Service概览图 多端口 Service 很多 Service 需要暴露多个端口。对于这种情况，Kubernetes 支持在 Service 对象中定义多个端口。 当使用多个端口时，必须给出所有的端口的名称，这样 Endpoint 就不会产生歧义，例如： kind: Service apiVersion: v1 metadata: name: my-service spec: selector: app: MyApp ports: - name: http protocol: TCP port: 80 targetPort: 9376 - name: https protocol: TCP port: 443 targetPort: 9377 选择自己的 IP 地址 在 Service 创建的请求中，可以通过设置 spec.clusterIP 字段来指定自己的集群 IP 地址。 比如，希望替换一个已经已存在的 DNS 条目，或者遗留系统已经配置了一个固定的 IP 且很难重新配置。 用户选择的 IP 地址必须合法，并且这个 IP 地址在 service-cluster-ip-range CIDR 范围内，这对 API Server 来说是通过一个标识来指定的。 如果 IP 地址不合法，API Server 会返回 HTTP 状态码 422，表示值不合法。 为何不使用 round-robin DNS？ 一个不时出现的问题是，为什么我们都使用 VIP 的方式，而不使用标准的 round-robin DNS，有如下几个原因： 长久以来，DNS 库都没能认真对待 DNS TTL、缓存域名查询结果 很多应用只查询一次 DNS 并缓存了结果 就算应用和库能够正确查询解析，每个客户端反复重解析造成的负载也是非常难以管理的 我们尽力阻止用户做那些对他们没有好处的事情，如果很多人都来问这个问题，我们可能会选择实现它。 服务发现 Kubernetes 支持2种基本的服务发现模式 —— 环境变量和 DNS。 环境变量 当 Pod 运行在 Node 上，kubelet 会为每个活跃的 Service 添加一组环境变量。 它同时支持 Docker links 兼容 变量（查看 makeLinkVariables）、简单的 {SVCNAME}_SERVICE_HOST 和 {SVCNAME}_SERVICE_PORT 变量，这里 Service 的名称需大写，横线被转换成下划线。 举个例子，一个名称为 \"redis-master\" 的 Service 暴露了 TCP 端口 6379，同时给它分配了 Cluster IP 地址 10.0.0.11，这个 Service 生成了如下环境变量： REDIS_MASTER_SERVICE_HOST=10.0.0.11 REDIS_MASTER_SERVICE_PORT=6379 REDIS_MASTER_PORT=tcp://10.0.0.11:6379 REDIS_MASTER_PORT_6379_TCP=tcp://10.0.0.11:6379 REDIS_MASTER_PORT_6379_TCP_PROTO=tcp REDIS_MASTER_PORT_6379_TCP_PORT=6379 REDIS_MASTER_PORT_6379_TCP_ADDR=10.0.0.11 这意味着需要有顺序的要求 —— Pod 想要访问的任何 Service 必须在 Pod 自己之前被创建，否则这些环境变量就不会被赋值。DNS 并没有这个限制。 DNS 一个可选（尽管强烈推荐）集群插件 是 DNS 服务器。 DNS 服务器监视着创建新 Service 的 Kubernetes API，从而为每一个 Service 创建一组 DNS 记录。 如果整个集群的 DNS 一直被启用，那么所有的 Pod 应该能够自动对 Service 进行名称解析。 例如，有一个名称为 \"my-service\" 的 Service，它在 Kubernetes 集群中名为 \"my-ns\" 的 Namespace 中，为 \"my-service.my-ns\" 创建了一条 DNS 记录。 在名称为 \"my-ns\" 的 Namespace 中的 Pod 应该能够简单地通过名称查询找到 \"my-service\"。 在另一个 Namespace 中的 Pod 必须限定名称为 \"my-service.my-ns\"。 这些名称查询的结果是 Cluster IP。 Kubernetes 也支持对端口名称的 DNS SRV（Service）记录。 如果名称为 \"my-service.my-ns\" 的 Service 有一个名为 \"http\" 的 TCP 端口，可以对 \"_http._tcp.my-service.my-ns\" 执行 DNS SRV 查询，得到 \"http\" 的端口号。 Kubernetes DNS 服务器是唯一的一种能够访问 ExternalName 类型的 Service 的方式。 更多信息可以查看 DNS Pod 和 Service。 Headless Service 有时不需要或不想要负载均衡，以及单独的 Service IP。 遇到这种情况，可以通过指定 Cluster IP（spec.clusterIP）的值为 \"None\" 来创建 Headless Service。 这个选项允许开发人员自由寻找他们自己的方式，从而降低与 Kubernetes 系统的耦合性。 应用仍然可以使用一种自注册的模式和适配器，对其它需要发现机制的系统能够很容易地基于这个 API 来构建。 对这类 Service 并不会分配 Cluster IP，kube-proxy 不会处理它们，而且平台也不会为它们进行负载均衡和路由。 DNS 如何实现自动配置，依赖于 Service 是否定义了 selector。 配置 Selector 对定义了 selector 的 Headless Service，Endpoint 控制器在 API 中创建了 Endpoints 记录，并且修改 DNS 配置返回 A 记录（地址），通过这个地址直接到达 Service 的后端 Pod 上。 不配置 Selector 对没有定义 selector 的 Headless Service，Endpoint 控制器不会创建 Endpoints 记录。 然而 DNS 系统会查找和配置，无论是： ExternalName 类型 Service 的 CNAME 记录 记录：与 Service 共享一个名称的任何 Endpoints，以及所有其它类型 发布服务 —— 服务类型 对一些应用（如 Frontend）的某些部分，可能希望通过外部（Kubernetes 集群外部）IP 地址暴露 Service。 Kubernetes ServiceTypes 允许指定一个需要的类型的 Service，默认是 ClusterIP 类型。 Type 的取值以及行为如下： ClusterIP：通过集群的内部 IP 暴露服务，选择该值，服务只能够在集群内部可以访问，这也是默认的 ServiceType。 NodePort：通过每个 Node 上的 IP 和静态端口（NodePort）暴露服务。NodePort 服务会路由到 ClusterIP 服务，这个 ClusterIP 服务会自动创建。通过请求 :，可以从集群的外部访问一个 NodePort 服务。 LoadBalancer：使用云提供商的负载均衡器，可以向外部暴露服务。外部的负载均衡器可以路由到 NodePort 服务和 ClusterIP 服务。 ExternalName：通过返回 CNAME 和它的值，可以将服务映射到 externalName 字段的内容（例如， foo.bar.example.com）。 没有任何类型代理被创建，这只有 Kubernetes 1.7 或更高版本的 kube-dns 才支持。 NodePort 类型 如果设置 type 的值为 \"NodePort\"，Kubernetes master 将从给定的配置范围内（默认：30000-32767）分配端口，每个 Node 将从该端口（每个 Node 上的同一端口）代理到 Service。该端口将通过 Service 的 spec.ports[*].nodePort 字段被指定。 如果需要指定的端口号，可以配置 nodePort 的值，系统将分配这个端口，否则调用 API 将会失败（比如，需要关心端口冲突的可能性）。 这可以让开发人员自由地安装他们自己的负载均衡器，并配置 Kubernetes 不能完全支持的环境参数，或者直接暴露一个或多个 Node 的 IP 地址。 需要注意的是，Service 将能够通过 :spec.ports[*].nodePort 和 spec.clusterIp:spec.ports[*].port 而对外可见。 LoadBalancer 类型 使用支持外部负载均衡器的云提供商的服务，设置 type 的值为 \"LoadBalancer\"，将为 Service 提供负载均衡器。 负载均衡器是异步创建的，关于被提供的负载均衡器的信息将会通过 Service 的 status.loadBalancer 字段被发布出去。 kind: Service apiVersion: v1 metadata: name: my-service spec: selector: app: MyApp ports: - protocol: TCP port: 80 targetPort: 9376 nodePort: 30061 clusterIP: 10.0.171.239 loadBalancerIP: 78.11.24.19 type: LoadBalancer status: loadBalancer: ingress: - ip: 146.148.47.155 来自外部负载均衡器的流量将直接打到 backend Pod 上，不过实际它们是如何工作的，这要依赖于云提供商。 在这些情况下，将根据用户设置的 loadBalancerIP 来创建负载均衡器。 某些云提供商允许设置 loadBalancerIP。如果没有设置 loadBalancerIP，将会给负载均衡器指派一个临时 IP。 如果设置了 loadBalancerIP，但云提供商并不支持这种特性，那么设置的 loadBalancerIP 值将会被忽略掉。 AWS 内部负载均衡器 在混合云环境中，有时从虚拟私有云（VPC）环境中的服务路由流量是非常有必要的。 可以通过在 Service 中增加 annotation 来实现，如下所示： [...] metadata: name: my-service annotations: service.beta.kubernetes.io/aws-load-balancer-internal: 0.0.0.0/0 [...] 在水平分割的 DNS 环境中，需要两个 Service 来将外部和内部的流量路由到 Endpoint 上。 AWS SSL 支持 对运行在 AWS 上部分支持 SSL 的集群，从 1.3 版本开始，可以为 LoadBalancer 类型的 Service 增加两个 annotation： metadata: name: my-service annotations: service.beta.kubernetes.io/aws-load-balancer-ssl-cert: arn:aws:acm:us-east-1:123456789012:certificate/12345678-1234-1234-1234-123456789012 第一个 annotation 指定了使用的证书。它可以是第三方发行商发行的证书，这个证书或者被上传到 IAM，或者由 AWS 的证书管理器创建。 metadata: name: my-service annotations: service.beta.kubernetes.io/aws-load-balancer-backend-protocol: (https|http|ssl|tcp) 第二个 annotation 指定了 Pod 使用的协议。 对于 HTTPS 和 SSL，ELB 将期望该 Pod 基于加密的连接来认证自身。 HTTP 和 HTTPS 将选择7层代理：ELB 将中断与用户的连接，当转发请求时，会解析 Header 信息并添加上用户的 IP 地址（Pod 将只能在连接的另一端看到该 IP 地址）。 TCP 和 SSL 将选择4层代理：ELB 将转发流量，并不修改 Header 信息。 外部 IP 如果外部的 IP 路由到集群中一个或多个 Node 上，Kubernetes Service 会被暴露给这些 externalIPs。 通过外部 IP（作为目的 IP 地址）进入到集群，打到 Service 的端口上的流量，将会被路由到 Service 的 Endpoint 上。 externalIPs 不会被 Kubernetes 管理，它属于集群管理员的职责范畴。 根据 Service 的规定，externalIPs 可以同任意的 ServiceType 来一起指定。 在下面的例子中，my-service 可以在 80.11.12.10:80（外部 IP:端口）上被客户端访问。 kind: Service apiVersion: v1 metadata: name: my-service spec: selector: app: MyApp ports: - name: http protocol: TCP port: 80 targetPort: 9376 externalIPs: - 80.11.12.10 不足之处 为 VIP 使用 userspace 代理，将只适合小型到中型规模的集群，不能够扩展到上千 Service 的大型集群。 查看 最初设计方案 获取更多细节。 使用 userspace 代理，隐藏了访问 Service 的数据包的源 IP 地址。 这使得一些类型的防火墙无法起作用。 iptables 代理不会隐藏 Kubernetes 集群内部的 IP 地址，但却要求客户端请求必须通过一个负载均衡器或 Node 端口。 Type 字段支持嵌套功能 —— 每一层需要添加到上一层里面。 不会严格要求所有云提供商（例如，GCE 就没必要为了使一个 LoadBalancer 能工作而分配一个 NodePort，但是 AWS 需要 ），但当前 API 是强制要求的。 未来工作 未来我们能预见到，代理策略可能会变得比简单的 round-robin 均衡策略有更多细微的差别，比如 master 选举或分片。 我们也能想到，某些 Service 将具有 “真正” 的负载均衡器，这种情况下 VIP 将简化数据包的传输。 我们打算为 L7（HTTP）Service 改进我们对它的支持。 我们打算为 Service 实现更加灵活的请求进入模式，这些 Service 包含当前 ClusterIP、NodePort 和 LoadBalancer 模式，或者更多。 VIP 的那些骇人听闻的细节 对很多想使用 Service 的人来说，前面的信息应该足够了。 然而，有很多内部原理性的内容，还是值去理解的。 避免冲突 Kubernetes 最主要的哲学之一，是用户不应该暴露那些能够导致他们操作失败、但又不是他们的过错的场景。 这种场景下，让我们来看一下网络端口 —— 用户不应该必须选择一个端口号，而且该端口还有可能与其他用户的冲突。 这就是说，在彼此隔离状态下仍然会出现失败。 为了使用户能够为他们的 Service 选择一个端口号，我们必须确保不能有2个 Service 发生冲突。 我们可以通过为每个 Service 分配它们自己的 IP 地址来实现。 为了保证每个 Service 被分配到一个唯一的 IP，需要一个内部的分配器能够原子地更新 etcd 中的一个全局分配映射表，这个更新操作要先于创建每一个 Service。 为了使 Service 能够获取到 IP，这个映射表对象必须在注册中心存在，否则创建 Service 将会失败，指示一个 IP 不能被分配。 一个后台 Controller 的职责是创建映射表（从 Kubernetes 的旧版本迁移过来，旧版本中是通过在内存中加锁的方式实现），并检查由于管理员干预和清除任意 IP 造成的不合理分配，这些 IP 被分配了但当前没有 Service 使用它们。 IP 和 VIP 不像 Pod 的 IP 地址，它实际路由到一个固定的目的地，Service 的 IP 实际上不能通过单个主机来进行应答。 相反，我们使用 iptables（Linux 中的数据包处理逻辑）来定义一个虚拟IP地址（VIP），它可以根据需要透明地进行重定向。 当客户端连接到 VIP 时，它们的流量会自动地传输到一个合适的 Endpoint。 环境变量和 DNS，实际上会根据 Service 的 VIP 和端口来进行填充。 Userspace 作为一个例子，考虑前面提到的图片处理应用程序。 当创建 backend Service 时，Kubernetes master 会给它指派一个虚拟 IP 地址，比如 10.0.0.1。 假设 Service 的端口是 1234，该 Service 会被集群中所有的 kube-proxy 实例观察到。 当代理看到一个新的 Service， 它会打开一个新的端口，建立一个从该 VIP 重定向到新端口的 iptables，并开始接收请求连接。 当一个客户端连接到一个 VIP，iptables 规则开始起作用，它会重定向该数据包到 Service代理 的端口。 Service代理 选择一个 backend，并将客户端的流量代理到 backend 上。 这意味着 Service 的所有者能够选择任何他们想使用的端口，而不存在冲突的风险。 客户端可以简单地连接到一个 IP 和端口，而不需要知道实际访问了哪些 Pod。 Iptables 再次考虑前面提到的图片处理应用程序。 当创建 backend Service 时，Kubernetes master 会给它指派一个虚拟 IP 地址，比如 10.0.0.1。 假设 Service 的端口是 1234，该 Service 会被集群中所有的 kube-proxy 实例观察到。 当代理看到一个新的 Service， 它会安装一系列的 iptables 规则，从 VIP 重定向到 per-Service 规则。 该 per-Service 规则连接到 per-Endpoint 规则，该 per-Endpoint 规则会重定向（目标 NAT）到 backend。 当一个客户端连接到一个 VIP，iptables 规则开始起作用。一个 backend 会被选择（或者根据会话亲和性，或者随机），数据包被重定向到这个 backend。 不像 userspace 代理，数据包从来不拷贝到用户空间，kube-proxy 不是必须为该 VIP 工作而运行，并且客户端 IP 是不可更改的。 当流量打到 Node 的端口上，或通过负载均衡器，会执行相同的基本流程，但是在那些案例中客户端 IP 是可以更改的。 API 对象 在 Kubernetes REST API 中，Service 是 top-level 资源。关于 API 对象的更多细节可以查看：Service API 对象。 更多信息 使用 Service 连接 Frontend 到 Backend Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-10-17 11:18:59 "},"concepts/volume.html":{"url":"concepts/volume.html","title":"Volume和Persistent Volume","keywords":"","body":"Kubernetes存储卷 我们知道默认情况下容器的数据都是非持久化的，在容器消亡以后数据也跟着丢失，所以Docker提供了Volume机制以便将数据持久化存储。类似的，Kubernetes提供了更强大的Volume机制和丰富的插件，解决了容器数据持久化和容器间共享数据的问题。 Volume 目前，Kubernetes支持以下Volume类型： emptyDir hostPath gcePersistentDisk awsElasticBlockStore nfs iscsi flocker glusterfs rbd cephfs gitRepo secret persistentVolumeClaim downwardAPI azureFileVolume vsphereVolume flexvolume 注意，这些volume并非全部都是持久化的，比如emptyDir、secret、gitRepo等，这些volume会随着Pod的消亡而消失。 PersistentVolume 对于持久化的Volume，PersistentVolume (PV)和PersistentVolumeClaim (PVC)提供了更方便的管理卷的方法：PV提供网络存储资源，而PVC请求存储资源。这样，设置持久化的工作流包括配置底层文件系统或者云数据卷、创建持久性数据卷、最后创建claim来将pod跟数据卷关联起来。PV和PVC可以将pod和数据卷解耦，pod不需要知道确切的文件系统或者支持它的持久化引擎。 PV PersistentVolume（PV）是集群之中的一块网络存储。跟 Node 一样，也是集群的资源。PV 跟 Volume (卷) 类似，不过会有独立于 Pod 的生命周期。比如一个NFS的PV可以定义为 apiVersion: v1 kind: PersistentVolume metadata: name: pv0003 spec: capacity: storage: 5Gi accessModes: - ReadWriteOnce persistentVolumeReclaimPolicy: Recycle nfs: path: /tmp server: 172.17.0.2 PV的访问模式有三种： 第一种，ReadWriteOnce：是最基本的方式，可读可写，但只支持被单个Pod挂载。 第二种，ReadOnlyMany：可以以只读的方式被多个Pod挂载。 第三种，ReadWriteMany：这种存储可以以读写的方式被多个Pod共享。不是每一种存储都支持这三种方式，像共享方式，目前支持的还比较少，比较常用的是NFS。在PVC绑定PV时通常根据两个条件来绑定，一个是存储的大小，另一个就是访问模式。 StorageClass 上面通过手动的方式创建了一个NFS Volume，这在管理很多Volume的时候不太方便。Kubernetes还提供了StorageClass来动态创建PV，不仅节省了管理员的时间，还可以封装不同类型的存储供PVC选用。 GCE的例子： kind: StorageClass apiVersion: storage.k8s.io/v1beta1 metadata: name: slow provisioner: kubernetes.io/gce-pd parameters: type: pd-standard zone: us-central1-a Ceph RBD的例子： apiVersion: storage.k8s.io/v1beta1 kind: StorageClass metadata: name: fast provisioner: kubernetes.io/rbd parameters: monitors: 10.16.153.105:6789 adminId: kube adminSecretName: ceph-secret adminSecretNamespace: kube-system pool: kube userId: kube userSecretName: ceph-secret-user PVC PV是存储资源，而PersistentVolumeClaim (PVC) 是对PV的请求。PVC跟Pod类似：Pod消费Node的源，而PVC消费PV资源；Pod能够请求CPU和内存资源，而PVC请求特定大小和访问模式的数据卷。 kind: PersistentVolumeClaim apiVersion: v1 metadata: name: myclaim spec: accessModes: - ReadWriteOnce resources: requests: storage: 8Gi selector: matchLabels: release: \"stable\" matchExpressions: - {key: environment, operator: In, values: [dev]} PVC可以直接挂载到Pod中： kind: Pod apiVersion: v1 metadata: name: mypod spec: containers: - name: myfrontend image: dockerfile/nginx volumeMounts: - mountPath: \"/var/www/html\" name: mypd volumes: - name: mypd persistentVolumeClaim: claimName: myclaim emptyDir 如果Pod配置了emptyDir类型Volume， Pod 被分配到Node上时候，会创建emptyDir，只要Pod运行在Node上，emptyDir都会存在（容器挂掉不会导致emptyDir丢失数据），但是如果Pod从Node上被删除（Pod被删除，或者Pod发生迁移），emptyDir也会被删除，并且永久丢失。 apiVersion: v1 kind: Pod metadata: name: test-pd spec: containers: - image: gcr.io/google_containers/test-webserver name: test-container volumeMounts: - mountPath: /test-pd name: test-volume volumes: - name: test-volume hostPath: # directory location on host path: /data 其他Volume说明 hostPath hostPath允许挂载Node上的文件系统到Pod里面去。如果Pod有需要使用Node上的文件，可以使用hostPath。 - hostPath: path: /tmp/data name: data NFS NFS 是Network File System的缩写，即网络文件系统。Kubernetes中通过简单地配置就可以挂载NFS到Pod中，而NFS中的数据是可以永久保存的，同时NFS支持同时写操作。 volumes: - name: nfs nfs: # FIXME: use the right hostname server: 10.254.234.223 path: \"/\" FlexVolume 注意要把volume plugin放到/usr/libexec/kubernetes/kubelet-plugins/volume/exec//，plugin要实现init/attach/detach/mount/umount等命令（可参考lvm的示例）。 - name: test flexVolume: driver: \"kubernetes.io/lvm\" fsType: \"ext4\" options: volumeID: \"vol1\" size: \"1000m\" volumegroup: \"kube_vg\" Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-08-21 18:23:34 "},"concepts/deployment.html":{"url":"concepts/deployment.html","title":"Deployment","keywords":"","body":"Deployment 简述 Deployment 为 Pod 和 ReplicaSet 提供了一个声明式定义(declarative)方法，用来替代以前的ReplicationController 来方便的管理应用。典型的应用场景包括： 定义Deployment来创建Pod和ReplicaSet 滚动升级和回滚应用 扩容和缩容 暂停和继续Deployment 比如一个简单的nginx应用可以定义为 apiVersion: extensions/v1beta1 kind: Deployment metadata: name: nginx-deployment spec: replicas: 3 template: metadata: labels: app: nginx spec: containers: - name: nginx image: nginx:1.7.9 ports: - containerPort: 80 扩容： kubectl scale deployment nginx-deployment --replicas 10 如果集群支持 horizontal pod autoscaling 的话，还可以为Deployment设置自动扩展： kubectl autoscale deployment nginx-deployment --min=10 --max=15 --cpu-percent=80 更新镜像也比较简单: kubectl set image deployment/nginx-deployment nginx=nginx:1.9.1 回滚： kubectl rollout undo deployment/nginx-deployment Deployment 结构示意图 参考：https://kubernetes.io/docs/api-reference/v1.6/#deploymentspec-v1beta1-apps 图片 - kubernetes deployment cheatsheet Deployment 概念详细解析 本文翻译自kubernetes官方文档：https://kubernetes.io/docs/concepts/workloads/controllers/deployment.md 根据2017年5月10日的Commit 8481c02 翻译。 Deployment 是什么？ Deployment为Pod和Replica Set（下一代Replication Controller）提供声明式更新。 您只需要在 Deployment 中描述您想要的目标状态是什么，Deployment controller 就会帮您将 Pod 和ReplicaSet 的实际状态改变到您的目标状态。您可以定义一个全新的 Deployment 来创建 ReplicaSet 或者删除已有的 Deployment 并创建一个新的来替换。 注意：您不该手动管理由 Deployment 创建的 ReplicaSet，否则您就篡越了 Deployment controller 的职责！下文罗列了 Deployment 对象中已经覆盖了所有的用例。如果未有覆盖您所有需要的用例，请直接在 Kubernetes 的代码库中提 issue。 典型的用例如下： 使用Deployment来创建ReplicaSet。ReplicaSet在后台创建pod。检查启动状态，看它是成功还是失败。 然后，通过更新Deployment的PodTemplateSpec字段来声明Pod的新状态。这会创建一个新的ReplicaSet，Deployment会按照控制的速率将pod从旧的ReplicaSet移动到新的ReplicaSet中。 如果当前状态不稳定，回滚到之前的Deployment revision。每次回滚都会更新Deployment的revision。 扩容Deployment以满足更高的负载。 暂停Deployment来应用PodTemplateSpec的多个修复，然后恢复上线。 根据Deployment 的状态判断上线是否hang住了。 清除旧的不必要的 ReplicaSet。 创建 Deployment 下面是一个 Deployment 示例，它创建了一个 ReplicaSet 来启动3个 nginx pod。 下载示例文件并执行命令： $ kubectl create -f https://kubernetes.io/docs/user-guide/nginx-deployment.yaml --record deployment \"nginx-deployment\" created 将kubectl的 --record 的 flag 设置为 true可以在 annotation 中记录当前命令创建或者升级了该资源。这在未来会很有用，例如，查看在每个 Deployment revision 中执行了哪些命令。 然后立即执行 get 将获得如下结果： $ kubectl get deployments NAME DESIRED CURRENT UP-TO-DATE AVAILABLE AGE nginx-deployment 3 0 0 0 1s 输出结果表明我们希望的repalica数是3（根据deployment中的.spec.replicas配置）当前replica数（ .status.replicas）是0, 最新的replica数（.status.updatedReplicas）是0，可用的replica数（.status.availableReplicas）是0。 过几秒后再执行get命令，将获得如下输出： $ kubectl get deployments NAME DESIRED CURRENT UP-TO-DATE AVAILABLE AGE nginx-deployment 3 3 3 3 18s 我们可以看到Deployment已经创建了3个 replica，所有的 replica 都已经是最新的了（包含最新的pod template），可用的（根据Deployment中的.spec.minReadySeconds声明，处于已就绪状态的pod的最少个数）。执行kubectl get rs和kubectl get pods会显示Replica Set（RS）和Pod已创建。 $ kubectl get rs NAME DESIRED CURRENT READY AGE nginx-deployment-2035384211 3 3 0 18s 您可能会注意到 ReplicaSet 的名字总是-。 $ kubectl get pods --show-labels NAME READY STATUS RESTARTS AGE LABELS nginx-deployment-2035384211-7ci7o 1/1 Running 0 18s app=nginx,pod-template-hash=2035384211 nginx-deployment-2035384211-kzszj 1/1 Running 0 18s app=nginx,pod-template-hash=2035384211 nginx-deployment-2035384211-qqcnn 1/1 Running 0 18s app=nginx,pod-template-hash=2035384211 刚创建的Replica Set将保证总是有3个 nginx 的 pod 存在。 注意： 您必须在 Deployment 中的 selector 指定正确的 pod template label（在该示例中是 app = nginx），不要跟其他的 controller 的 selector 中指定的 pod template label 搞混了（包括 Deployment、Replica Set、Replication Controller 等）。Kubernetes 本身并不会阻止您任意指定 pod template label ，但是如果您真的这么做了，这些 controller 之间会相互打架，并可能导致不正确的行为。 Pod-template-hash label 注意：这个 label 不是用户指定的！ 注意上面示例输出中的 pod label 里的 pod-template-hash label。当 Deployment 创建或者接管 ReplicaSet 时，Deployment controller 会自动为 Pod 添加 pod-template-hash label。这样做的目的是防止 Deployment 的子ReplicaSet 的 pod 名字重复。通过将 ReplicaSet 的 PodTemplate 进行哈希散列，使用生成的哈希值作为 label 的值，并添加到 ReplicaSet selector 里、 pod template label 和 ReplicaSet 管理中的 Pod 上。 更新Deployment 注意： Deployment 的 rollout 当且仅当 Deployment 的 pod template（例如.spec.template）中的label更新或者镜像更改时被触发。其他更新，例如扩容Deployment不会触发 rollout。 假如我们现在想要让 nginx pod 使用nginx:1.9.1的镜像来代替原来的nginx:1.7.9的镜像。 $ kubectl set image deployment/nginx-deployment nginx=nginx:1.9.1 deployment \"nginx-deployment\" image updated 我们可以使用edit命令来编辑 Deployment，修改 .spec.template.spec.containers[0].image ，将nginx:1.7.9 改写成 nginx:1.9.1。 $ kubectl edit deployment/nginx-deployment deployment \"nginx-deployment\" edited 查看 rollout 的状态，只要执行： $ kubectl rollout status deployment/nginx-deployment Waiting for rollout to finish: 2 out of 3 new replicas have been updated... deployment \"nginx-deployment\" successfully rolled out Rollout 成功后，get Deployment： $ kubectl get deployments NAME DESIRED CURRENT UP-TO-DATE AVAILABLE AGE nginx-deployment 3 3 3 3 36s UP-TO-DATE 的 replica 的数目已经达到了配置中要求的数目。 CURRENT 的 replica 数表示 Deployment 管理的 replica 数量，AVAILABLE 的 replica 数是当前可用的replica数量。 我们通过执行kubectl get rs可以看到 Deployment 更新了Pod，通过创建一个新的 ReplicaSet 并扩容了3个 replica，同时将原来的 ReplicaSet 缩容到了0个 replica。 $ kubectl get rs NAME DESIRED CURRENT READY AGE nginx-deployment-1564180365 3 3 0 6s nginx-deployment-2035384211 0 0 0 36s 执行 get pods只会看到当前的新的 pod: $ kubectl get pods NAME READY STATUS RESTARTS AGE nginx-deployment-1564180365-khku8 1/1 Running 0 14s nginx-deployment-1564180365-nacti 1/1 Running 0 14s nginx-deployment-1564180365-z9gth 1/1 Running 0 14s 下次更新这些 pod 的时候，只需要更新 Deployment 中的 pod 的 template 即可。 Deployment 可以保证在升级时只有一定数量的 Pod 是 down 的。默认的，它会确保至少有比期望的Pod数量少一个是up状态（最多一个不可用）。 Deployment 同时也可以确保只创建出超过期望数量的一定数量的 Pod。默认的，它会确保最多比期望的Pod数量多一个的 Pod 是 up 的（最多1个 surge ）。 在未来的 Kuberentes 版本中，将从1-1变成25%-25%。 例如，如果您自己看下上面的 Deployment，您会发现，开始创建一个新的 Pod，然后删除一些旧的 Pod 再创建一个新的。当新的Pod创建出来之前不会杀掉旧的Pod。这样能够确保可用的 Pod 数量至少有2个，Pod的总数最多4个。 $ kubectl describe deployments Name: nginx-deployment Namespace: default CreationTimestamp: Tue, 15 Mar 2016 12:01:06 -0700 Labels: app=nginx Selector: app=nginx Replicas: 3 updated | 3 total | 3 available | 0 unavailable StrategyType: RollingUpdate MinReadySeconds: 0 RollingUpdateStrategy: 1 max unavailable, 1 max surge OldReplicaSets: NewReplicaSet: nginx-deployment-1564180365 (3/3 replicas created) Events: FirstSeen LastSeen Count From SubobjectPath Type Reason Message --------- -------- ----- ---- ------------- -------- ------ ------- 36s 36s 1 {deployment-controller } Normal ScalingReplicaSet Scaled up replica set nginx-deployment-2035384211 to 3 23s 23s 1 {deployment-controller } Normal ScalingReplicaSet Scaled up replica set nginx-deployment-1564180365 to 1 23s 23s 1 {deployment-controller } Normal ScalingReplicaSet Scaled down replica set nginx-deployment-2035384211 to 2 23s 23s 1 {deployment-controller } Normal ScalingReplicaSet Scaled up replica set nginx-deployment-1564180365 to 2 21s 21s 1 {deployment-controller } Normal ScalingReplicaSet Scaled down replica set nginx-deployment-2035384211 to 0 21s 21s 1 {deployment-controller } Normal ScalingReplicaSet Scaled up replica set nginx-deployment-1564180365 to 3 我们可以看到当我们刚开始创建这个 Deployment 的时候，创建了一个 ReplicaSet（nginx-deployment-2035384211），并直接扩容到了3个 replica。 当我们更新这个 Deployment 的时候，它会创建一个新的 ReplicaSet（nginx-deployment-1564180365），将它扩容到1个replica，然后缩容原先的 ReplicaSet 到2个 replica，此时满足至少2个 Pod 是可用状态，同一时刻最多有4个 Pod 处于创建的状态。 接着继续使用相同的 rolling update 策略扩容新的 ReplicaSet 和缩容旧的 ReplicaSet。最终，将会在新的 ReplicaSet 中有3个可用的 replica，旧的 ReplicaSet 的 replica 数目变成0。 Rollover（多个rollout并行） 每当 Deployment controller 观测到有新的 deployment 被创建时，如果没有已存在的 ReplicaSet 来创建期望个数的 Pod 的话，就会创建出一个新的 ReplicaSet 来做这件事。已存在的 ReplicaSet 控制 label 与.spec.selector匹配但是 template 跟.spec.template不匹配的 Pod 缩容。最终，新的 ReplicaSet 将会扩容出.spec.replicas指定数目的 Pod，旧的 ReplicaSet 会缩容到0。 如果您更新了一个的已存在并正在进行中的 Deployment，每次更新 Deployment都会创建一个新的 ReplicaSet并扩容它，同时回滚之前扩容的 ReplicaSet ——将它添加到旧的 ReplicaSet 列表中，开始缩容。 例如，假如您创建了一个有5个niginx:1.7.9 replica的 Deployment，但是当还只有3个nginx:1.7.9的 replica 创建出来的时候您就开始更新含有5个nginx:1.9.1 replica 的 Deployment。在这种情况下，Deployment 会立即杀掉已创建的3个nginx:1.7.9的 Pod，并开始创建nginx:1.9.1的 Pod。它不会等到所有的5个nginx:1.7.9的 Pod 都创建完成后才开始改变航道。 Label selector 更新 我们通常不鼓励更新 label selector，我们建议实现规划好您的 selector。 任何情况下，只要您想要执行 label selector 的更新，请一定要谨慎并确认您已经预料到所有可能因此导致的后果。 增添 selector 需要同时在 Deployment 的 spec 中更新新的 label，否则将返回校验错误。此更改是不可覆盖的，这意味着新的 selector 不会选择使用旧 selector 创建的 ReplicaSet 和 Pod，从而导致所有旧版本的 ReplicaSet 都被丢弃，并创建新的 ReplicaSet。 更新 selector，即更改 selector key 的当前值，将导致跟增添 selector 同样的后果。 删除 selector，即删除 Deployment selector 中的已有的 key，不需要对 Pod template label 做任何更改，现有的 ReplicaSet 也不会成为孤儿，但是请注意，删除的 label 仍然存在于现有的 Pod 和 ReplicaSet 中。 回退Deployment 有时候您可能想回退一个 Deployment，例如，当 Deployment 不稳定时，比如一直 crash looping。 默认情况下，kubernetes 会在系统中保存前两次的 Deployment 的 rollout 历史记录，以便您可以随时回退（您可以修改revision history limit来更改保存的revision数）。 注意： 只要 Deployment 的 rollout 被触发就会创建一个 revision。也就是说当且仅当 Deployment 的 Pod template（如.spec.template）被更改，例如更新template 中的 label 和容器镜像时，就会创建出一个新的 revision。 其他的更新，比如扩容 Deployment 不会创建 revision——因此我们可以很方便的手动或者自动扩容。这意味着当您回退到历史 revision 是，直有 Deployment 中的 Pod template 部分才会回退。 假设我们在更新 Deployment 的时候犯了一个拼写错误，将镜像的名字写成了nginx:1.91，而正确的名字应该是nginx:1.9.1： $ kubectl set image deployment/nginx-deployment nginx=nginx:1.91 deployment \"nginx-deployment\" image updated Rollout 将会卡住。 $ kubectl rollout status deployments nginx-deployment Waiting for rollout to finish: 2 out of 3 new replicas have been updated... 按住 Ctrl-C 停止上面的 rollout 状态监控。 您会看到旧的 replica（nginx-deployment-1564180365 和 nginx-deployment-2035384211）和新的 replica （nginx-deployment-3066724191）数目都是2个。 $ kubectl get rs NAME DESIRED CURRENT READY AGE nginx-deployment-1564180365 2 2 0 25s nginx-deployment-2035384211 0 0 0 36s nginx-deployment-3066724191 2 2 2 6s 看下创建 Pod，您会看到有两个新的 ReplicaSet 创建的 Pod 处于 ImagePullBackOff 状态，循环拉取镜像。 $ kubectl get pods NAME READY STATUS RESTARTS AGE nginx-deployment-1564180365-70iae 1/1 Running 0 25s nginx-deployment-1564180365-jbqqo 1/1 Running 0 25s nginx-deployment-3066724191-08mng 0/1 ImagePullBackOff 0 6s nginx-deployment-3066724191-eocby 0/1 ImagePullBackOff 0 6s 注意，Deployment controller会自动停止坏的 rollout，并停止扩容新的 ReplicaSet。 $ kubectl describe deployment Name: nginx-deployment Namespace: default CreationTimestamp: Tue, 15 Mar 2016 14:48:04 -0700 Labels: app=nginx Selector: app=nginx Replicas: 2 updated | 3 total | 2 available | 2 unavailable StrategyType: RollingUpdate MinReadySeconds: 0 RollingUpdateStrategy: 1 max unavailable, 1 max surge OldReplicaSets: nginx-deployment-1564180365 (2/2 replicas created) NewReplicaSet: nginx-deployment-3066724191 (2/2 replicas created) Events: FirstSeen LastSeen Count From SubobjectPath Type Reason Message --------- -------- ----- ---- ------------- -------- ------ ------- 1m 1m 1 {deployment-controller } Normal ScalingReplicaSet Scaled up replica set nginx-deployment-2035384211 to 3 22s 22s 1 {deployment-controller } Normal ScalingReplicaSet Scaled up replica set nginx-deployment-1564180365 to 1 22s 22s 1 {deployment-controller } Normal ScalingReplicaSet Scaled down replica set nginx-deployment-2035384211 to 2 22s 22s 1 {deployment-controller } Normal ScalingReplicaSet Scaled up replica set nginx-deployment-1564180365 to 2 21s 21s 1 {deployment-controller } Normal ScalingReplicaSet Scaled down replica set nginx-deployment-2035384211 to 0 21s 21s 1 {deployment-controller } Normal ScalingReplicaSet Scaled up replica set nginx-deployment-1564180365 to 3 13s 13s 1 {deployment-controller } Normal ScalingReplicaSet Scaled up replica set nginx-deployment-3066724191 to 1 13s 13s 1 {deployment-controller } Normal ScalingReplicaSet Scaled down replica set nginx-deployment-1564180365 to 2 13s 13s 1 {deployment-controller } Normal ScalingReplicaSet Scaled up replica set nginx-deployment-3066724191 to 2 为了修复这个问题，我们需要回退到稳定的 Deployment revision。 检查 Deployment 升级的历史记录 首先，检查下 Deployment 的 revision： $ kubectl rollout history deployment/nginx-deployment deployments \"nginx-deployment\": REVISION CHANGE-CAUSE 1 kubectl create -f https://kubernetes.io/docs/user-guide/nginx-deployment.yaml--record 2 kubectl set image deployment/nginx-deployment nginx=nginx:1.9.1 3 kubectl set image deployment/nginx-deployment nginx=nginx:1.91 因为我们创建 Deployment 的时候使用了--record参数可以记录命令，我们可以很方便的查看每次 revision 的变化。 查看单个revision 的详细信息： $ kubectl rollout history deployment/nginx-deployment --revision=2 deployments \"nginx-deployment\" revision 2 Labels: app=nginx pod-template-hash=1159050644 Annotations: kubernetes.io/change-cause=kubectl set image deployment/nginx-deployment nginx=nginx:1.9.1 Containers: nginx: Image: nginx:1.9.1 Port: 80/TCP QoS Tier: cpu: BestEffort memory: BestEffort Environment Variables: No volumes. 回退到历史版本 现在，我们可以决定回退当前的 rollout 到之前的版本： $ kubectl rollout undo deployment/nginx-deployment deployment \"nginx-deployment\" rolled back 也可以使用 --revision参数指定某个历史版本： $ kubectl rollout undo deployment/nginx-deployment --to-revision=2 deployment \"nginx-deployment\" rolled back 与 rollout 相关的命令详细文档见kubectl rollout。 该 Deployment 现在已经回退到了先前的稳定版本。如您所见，Deployment controller产生了一个回退到revison 2的DeploymentRollback的 event。 $ kubectl get deployment NAME DESIRED CURRENT UP-TO-DATE AVAILABLE AGE nginx-deployment 3 3 3 3 30m $ kubectl describe deployment Name: nginx-deployment Namespace: default CreationTimestamp: Tue, 15 Mar 2016 14:48:04 -0700 Labels: app=nginx Selector: app=nginx Replicas: 3 updated | 3 total | 3 available | 0 unavailable StrategyType: RollingUpdate MinReadySeconds: 0 RollingUpdateStrategy: 1 max unavailable, 1 max surge OldReplicaSets: NewReplicaSet: nginx-deployment-1564180365 (3/3 replicas created) Events: FirstSeen LastSeen Count From SubobjectPath Type Reason Message --------- -------- ----- ---- ------------- -------- ------ ------- 30m 30m 1 {deployment-controller } Normal ScalingReplicaSet Scaled up replica set nginx-deployment-2035384211 to 3 29m 29m 1 {deployment-controller } Normal ScalingReplicaSet Scaled up replica set nginx-deployment-1564180365 to 1 29m 29m 1 {deployment-controller } Normal ScalingReplicaSet Scaled down replica set nginx-deployment-2035384211 to 2 29m 29m 1 {deployment-controller } Normal ScalingReplicaSet Scaled up replica set nginx-deployment-1564180365 to 2 29m 29m 1 {deployment-controller } Normal ScalingReplicaSet Scaled down replica set nginx-deployment-2035384211 to 0 29m 29m 1 {deployment-controller } Normal ScalingReplicaSet Scaled up replica set nginx-deployment-3066724191 to 2 29m 29m 1 {deployment-controller } Normal ScalingReplicaSet Scaled up replica set nginx-deployment-3066724191 to 1 29m 29m 1 {deployment-controller } Normal ScalingReplicaSet Scaled down replica set nginx-deployment-1564180365 to 2 2m 2m 1 {deployment-controller } Normal ScalingReplicaSet Scaled down replica set nginx-deployment-3066724191 to 0 2m 2m 1 {deployment-controller } Normal DeploymentRollback Rolled back deployment \"nginx-deployment\" to revision 2 29m 2m 2 {deployment-controller } Normal ScalingReplicaSet Scaled up replica set nginx-deployment-1564180365 to 3 清理 Policy 您可以通过设置.spec.revisonHistoryLimit项来指定 deployment 最多保留多少 revision 历史记录。默认的会保留所有的 revision；如果将该项设置为0，Deployment就不允许回退了。 Deployment 扩容 您可以使用以下命令扩容 Deployment： $ kubectl scale deployment nginx-deployment --replicas 10 deployment \"nginx-deployment\" scaled 假设您的集群中启用了horizontal pod autoscaling，您可以给 Deployment 设置一个 autoscaler，基于当前 Pod的 CPU 利用率选择最少和最多的 Pod 数。 $ kubectl autoscale deployment nginx-deployment --min=10 --max=15 --cpu-percent=80 deployment \"nginx-deployment\" autoscaled 比例扩容 RollingUpdate Deployment 支持同时运行一个应用的多个版本。或者 autoscaler 扩 容 RollingUpdate Deployment 的时候，正在中途的 rollout（进行中或者已经暂停的），为了降低风险，Deployment controller 将会平衡已存在的活动中的 ReplicaSet（有 Pod 的 ReplicaSet）和新加入的 replica。这被称为比例扩容。 例如，您正在运行中含有10个 replica 的 Deployment。maxSurge=3，maxUnavailable=2。 $ kubectl get deploy NAME DESIRED CURRENT UP-TO-DATE AVAILABLE AGE nginx-deployment 10 10 10 10 50s 您更新了一个镜像，而在集群内部无法解析。 $ kubectl set image deploy/nginx-deployment nginx=nginx:sometag deployment \"nginx-deployment\" image updated 镜像更新启动了一个包含ReplicaSet nginx-deployment-1989198191的新的rollout，但是它被阻塞了，因为我们上面提到的maxUnavailable。 $ kubectl get rs NAME DESIRED CURRENT READY AGE nginx-deployment-1989198191 5 5 0 9s nginx-deployment-618515232 8 8 8 1m 然后发起了一个新的Deployment扩容请求。autoscaler将Deployment的repllica数目增加到了15个。Deployment controller需要判断在哪里增加这5个新的replica。如果我们没有谁用比例扩容，所有的5个replica都会加到一个新的ReplicaSet中。如果使用比例扩容，新添加的replica将传播到所有的ReplicaSet中。大的部分加入replica数最多的ReplicaSet中，小的部分加入到replica数少的ReplciaSet中。0个replica的ReplicaSet不会被扩容。 在我们上面的例子中，3个replica将添加到旧的ReplicaSet中，2个replica将添加到新的ReplicaSet中。rollout进程最终会将所有的replica移动到新的ReplicaSet中，假设新的replica成为健康状态。 $ kubectl get deploy NAME DESIRED CURRENT UP-TO-DATE AVAILABLE AGE nginx-deployment 15 18 7 8 7m $ kubectl get rs NAME DESIRED CURRENT READY AGE nginx-deployment-1989198191 7 7 0 7m nginx-deployment-618515232 11 11 11 7m 暂停和恢复Deployment 您可以在发出一次或多次更新前暂停一个 Deployment，然后再恢复它。这样您就能多次暂停和恢复 Deployment，在此期间进行一些修复工作，而不会发出不必要的 rollout。 例如使用刚刚创建 Deployment： $ kubectl get deploy NAME DESIRED CURRENT UP-TO-DATE AVAILABLE AGE nginx 3 3 3 3 1m [mkargaki@dhcp129-211 kubernetes]$ kubectl get rs NAME DESIRED CURRENT READY AGE nginx-2142116321 3 3 3 1m 使用以下命令暂停 Deployment： $ kubectl rollout pause deployment/nginx-deployment deployment \"nginx-deployment\" paused 然后更新 Deplyment中的镜像： $ kubectl set image deploy/nginx nginx=nginx:1.9.1 deployment \"nginx-deployment\" image updated 注意新的 rollout 启动了： $ kubectl rollout history deploy/nginx deployments \"nginx\" REVISION CHANGE-CAUSE 1 $ kubectl get rs NAME DESIRED CURRENT READY AGE nginx-2142116321 3 3 3 2m 您可以进行任意多次更新，例如更新使用的资源： $ kubectl set resources deployment nginx -c=nginx --limits=cpu=200m,memory=512Mi deployment \"nginx\" resource requirements updated Deployment 暂停前的初始状态将继续它的功能，而不会对 Deployment 的更新产生任何影响，只要 Deployment是暂停的。 最后，恢复这个 Deployment，观察完成更新的 ReplicaSet 已经创建出来了： $ kubectl rollout resume deploy nginx deployment \"nginx\" resumed $ KUBECTL get rs -w NAME DESIRED CURRENT READY AGE nginx-2142116321 2 2 2 2m nginx-3926361531 2 2 0 6s nginx-3926361531 2 2 1 18s nginx-2142116321 1 2 2 2m nginx-2142116321 1 2 2 2m nginx-3926361531 3 2 1 18s nginx-3926361531 3 2 1 18s nginx-2142116321 1 1 1 2m nginx-3926361531 3 3 1 18s nginx-3926361531 3 3 2 19s nginx-2142116321 0 1 1 2m nginx-2142116321 0 1 1 2m nginx-2142116321 0 0 0 2m nginx-3926361531 3 3 3 20s ^C $ KUBECTL get rs NAME DESIRED CURRENT READY AGE nginx-2142116321 0 0 0 2m nginx-3926361531 3 3 3 28s 注意： 在恢复 Deployment 之前您无法回退一个已经暂停的 Deployment。 Deployment 状态 Deployment 在生命周期中有多种状态。在创建一个新的 ReplicaSet 的时候它可以是 progressing 状态， complete 状态，或者 fail to progress 状态。 进行中的 Deployment Kubernetes 将执行过下列任务之一的 Deployment 标记为 progressing 状态： Deployment 正在创建新的ReplicaSet过程中。 Deployment 正在扩容一个已有的 ReplicaSet。 Deployment 正在缩容一个已有的 ReplicaSet。 有新的可用的 pod 出现。 您可以使用kubectl rollout status命令监控 Deployment 的进度。 完成的 Deployment Kubernetes 将包括以下特性的 Deployment 标记为 complete 状态： Deployment 最小可用。最小可用意味着 Deployment 的可用 replica 个数等于或者超过 Deployment 策略中的期望个数。 所有与该 Deployment 相关的replica都被更新到了您指定版本，也就说更新完成。 该 Deployment 中没有旧的 Pod 存在。 您可以用kubectl rollout status命令查看 Deployment 是否完成。如果 rollout 成功完成，kubectl rollout status将返回一个0值的 Exit Code。 $ kubectl rollout status deploy/nginx Waiting for rollout to finish: 2 of 3 updated replicas are available... deployment \"nginx\" successfully rolled out $ echo $? 0 失败的 Deployment 您的 Deployment 在尝试部署新的 ReplicaSet 的时候可能卡住，用于也不会完成。这可能是因为以下几个因素引起的： 无效的引用 不可读的 probe failure 镜像拉取错误 权限不够 范围限制 程序运行时配置错误 探测这种情况的一种方式是，在您的 Deployment spec 中指定spec.progressDeadlineSeconds。spec.progressDeadlineSeconds 表示 Deployment controller 等待多少秒才能确定（通过 Deployment status）Deployment进程是卡住的。 下面的kubectl命令设置progressDeadlineSeconds 使 controller 在 Deployment 在进度卡住10分钟后报告： $ kubectl patch deployment/nginx-deployment -p '{\"spec\":{\"progressDeadlineSeconds\":600}}' \"nginx-deployment\" patched 当超过截止时间后，Deployment controller 会在 Deployment 的 status.conditions中增加一条DeploymentCondition，它包括如下属性： Type=Progressing Status=False Reason=ProgressDeadlineExceeded 浏览 Kubernetes API conventions 查看关于status conditions的更多信息。 注意： kubernetes除了报告Reason=ProgressDeadlineExceeded状态信息外不会对卡住的 Deployment 做任何操作。更高层次的协调器可以利用它并采取相应行动，例如，回滚 Deployment 到之前的版本。 注意： 如果您暂停了一个 Deployment，在暂停的这段时间内kubernetnes不会检查您指定的 deadline。您可以在 Deployment 的 rollout 途中安全的暂停它，然后再恢复它，这不会触发超过deadline的状态。 您可能在使用 Deployment 的时候遇到一些短暂的错误，这些可能是由于您设置了太短的 timeout，也有可能是因为各种其他错误导致的短暂错误。例如，假设您使用了无效的引用。当您 Describe Deployment 的时候可能会注意到如下信息： $ kubectl describe deployment nginx-deployment Conditions: Type Status Reason ---- ------ ------ Available True MinimumReplicasAvailable Progressing True ReplicaSetUpdated ReplicaFailure True FailedCreate 执行 kubectl get deployment nginx-deployment -o yaml，Deployement 的状态可能看起来像这个样子： status: availableReplicas: 2 conditions: - lastTransitionTime: 2016-10-04T12:25:39Z lastUpdateTime: 2016-10-04T12:25:39Z message: Replica set \"nginx-deployment-4262182780\" is progressing. reason: ReplicaSetUpdated status: \"True\" type: Progressing - lastTransitionTime: 2016-10-04T12:25:42Z lastUpdateTime: 2016-10-04T12:25:42Z message: Deployment has minimum availability. reason: MinimumReplicasAvailable status: \"True\" type: Available - lastTransitionTime: 2016-10-04T12:25:39Z lastUpdateTime: 2016-10-04T12:25:39Z message: 'Error creating: pods \"nginx-deployment-4262182780-\" is forbidden: exceeded quota: object-counts, requested: pods=1, used: pods=3, limited: pods=2' reason: FailedCreate status: \"True\" type: ReplicaFailure observedGeneration: 3 replicas: 2 unavailableReplicas: 2 最终，一旦超过 Deployment 进程的 deadline，kuberentes 会更新状态和导致 Progressing 状态的原因： Conditions: Type Status Reason ---- ------ ------ Available True MinimumReplicasAvailable Progressing False ProgressDeadlineExceeded ReplicaFailure True FailedCreate 您可以通过缩容 Deployment的方式解决配额不足的问题，或者增加您的 namespace 的配额。如果您满足了配额条件后，Deployment controller 就会完成您的 Deployment rollout，您将看到 Deployment 的状态更新为成功状态（Status=True并且Reason=NewReplicaSetAvailable）。 Conditions: Type Status Reason ---- ------ ------ Available True MinimumReplicasAvailable Progressing True NewReplicaSetAvailable Type=Available、 Status=True 以为这您的Deployment有最小可用性。 最小可用性是在Deployment策略中指定的参数。Type=Progressing 、 Status=True意味着您的Deployment 或者在部署过程中，或者已经成功部署，达到了期望的最少的可用replica数量（查看特定状态的Reason——在我们的例子中Reason=NewReplicaSetAvailable 意味着Deployment已经完成）。 您可以使用kubectl rollout status命令查看Deployment进程是否失败。当Deployment过程超过了deadline，kubectl rollout status将返回非0的exit code。 $ kubectl rollout status deploy/nginx Waiting for rollout to finish: 2 out of 3 new replicas have been updated... error: deployment \"nginx\" exceeded its progress deadline $ echo $? 1 操作失败的 Deployment 所有对完成的 Deployment 的操作都适用于失败的 Deployment。您可以对它扩/缩容，回退到历史版本，您甚至可以多次暂停它来应用 Deployment pod template。 清理Policy 您可以设置 Deployment 中的 .spec.revisionHistoryLimit 项来指定保留多少旧的 ReplicaSet。 余下的将在后台被当作垃圾收集。默认的，所有的 revision 历史就都会被保留。在未来的版本中，将会更改为2。 注意： 将该值设置为0，将导致所有的 Deployment 历史记录都会被清除，该 Deployment 就无法再回退了。 用例 金丝雀 Deployment 如果您想要使用 Deployment 对部分用户或服务器发布 release，您可以创建多个 Deployment，每个 Deployment 对应一个 release，参照 managing resources 中对金丝雀模式的描述。 编写 Deployment Spec 在所有的 Kubernetes 配置中，Deployment 也需要apiVersion，kind和metadata这些配置项。配置文件的通用使用说明查看 部署应用，配置容器，和 使用 kubectl 管理资源 文档。 Deployment也需要 .spec section. Pod Template .spec.template 是 .spec中唯一要求的字段。 .spec.template 是 pod template. 它跟 Pod有一模一样的schema，除了它是嵌套的并且不需要apiVersion 和 kind字段。 另外为了划分Pod的范围，Deployment中的pod template必须指定适当的label（不要跟其他controller重复了，参考selector）和适当的重启策略。 .spec.template.spec.restartPolicy 可以设置为 Always , 如果不指定的话这就是默认配置。 Replicas .spec.replicas 是可以选字段，指定期望的pod数量，默认是1。 Selector .spec.selector是可选字段，用来指定 label selector ，圈定Deployment管理的pod范围。 如果被指定， .spec.selector 必须匹配 .spec.template.metadata.labels，否则它将被API拒绝。如果 .spec.selector 没有被指定， .spec.selector.matchLabels 默认是 .spec.template.metadata.labels。 在Pod的template跟.spec.template不同或者数量超过了.spec.replicas规定的数量的情况下，Deployment会杀掉label跟selector不同的Pod。 注意： 您不应该再创建其他label跟这个selector匹配的pod，或者通过其他Deployment，或者通过其他Controller，例如ReplicaSet和ReplicationController。否则该Deployment会被把它们当成都是自己创建的。Kubernetes不会阻止您这么做。 如果您有多个controller使用了重复的selector，controller们就会互相打架并导致不正确的行为。 策略 .spec.strategy 指定新的Pod替换旧的Pod的策略。 .spec.strategy.type 可以是\"Recreate\"或者是 \"RollingUpdate\"。\"RollingUpdate\"是默认值。 Recreate Deployment .spec.strategy.type==Recreate时，在创建出新的Pod之前会先杀掉所有已存在的Pod。 Rolling Update Deployment .spec.strategy.type==RollingUpdate时，Deployment使用rolling update 的方式更新Pod 。您可以指定maxUnavailable 和 maxSurge 来控制 rolling update 进程。 Max Unavailable .spec.strategy.rollingUpdate.maxUnavailable 是可选配置项，用来指定在升级过程中不可用Pod的最大数量。该值可以是一个绝对值（例如5），也可以是期望Pod数量的百分比（例如10%）。通过计算百分比的绝对值向下取整。如果.spec.strategy.rollingUpdate.maxSurge 为0时，这个值不可以为0。默认值是1。 例如，该值设置成30%，启动rolling update后旧的ReplicatSet将会立即缩容到期望的Pod数量的70%。新的Pod ready后，随着新的ReplicaSet的扩容，旧的ReplicaSet会进一步缩容，确保在升级的所有时刻可以用的Pod数量至少是期望Pod数量的70%。 Max Surge .spec.strategy.rollingUpdate.maxSurge 是可选配置项，用来指定可以超过期望的Pod数量的最大个数。该值可以是一个绝对值（例如5）或者是期望的Pod数量的百分比（例如10%）。当MaxUnavailable为0时该值不可以为0。通过百分比计算的绝对值向上取整。默认值是1。 例如，该值设置成30%，启动rolling update后新的ReplicatSet将会立即扩容，新老Pod的总数不能超过期望的Pod数量的130%。旧的Pod被杀掉后，新的ReplicaSet将继续扩容，旧的ReplicaSet会进一步缩容，确保在升级的所有时刻所有的Pod数量和不会超过期望Pod数量的130%。 Progress Deadline Seconds .spec.progressDeadlineSeconds 是可选配置项，用来指定在系统报告Deployment的failed progressing ——表现为resource的状态中type=Progressing、Status=False、 Reason=ProgressDeadlineExceeded前可以等待的Deployment进行的秒数。Deployment controller会继续重试该Deployment。未来，在实现了自动回滚后， deployment controller在观察到这种状态时就会自动回滚。 如果设置该参数，该值必须大于 .spec.minReadySeconds。 Min Ready Seconds .spec.minReadySeconds是一个可选配置项，用来指定没有任何容器crash的Pod并被认为是可用状态的最小秒数。默认是0（Pod在ready后就会被认为是可用状态）。进一步了解什么什么后Pod会被认为是ready状态，参阅 Container Probes。 Rollback To .spec.rollbackTo 是一个可以选配置项，用来配置Deployment回退的配置。设置该参数将触发回退操作，每次回退完成后，该值就会被清除。 Revision .spec.rollbackTo.revision是一个可选配置项，用来指定回退到的revision。默认是0，意味着回退到上一个revision。 Revision History Limit Deployment revision history存储在它控制的ReplicaSets中。 .spec.revisionHistoryLimit 是一个可选配置项，用来指定可以保留的旧的ReplicaSet数量。该理想值取决于心Deployment的频率和稳定性。如果该值没有设置的话，默认所有旧的Replicaset或会被保留，将资源存储在etcd中，是用kubectl get rs查看输出。每个Deployment的该配置都保存在ReplicaSet中，然而，一旦您删除的旧的RepelicaSet，您的Deployment就无法再回退到那个revison了。 如果您将该值设置为0，所有具有0个replica的ReplicaSet都会被删除。在这种情况下，新的Deployment rollout无法撤销，因为revision history都被清理掉了。 Paused .spec.paused是可以可选配置项，boolean值。用来指定暂停和恢复Deployment。Paused和没有paused的Deployment之间的唯一区别就是，所有对paused deployment中的PodTemplateSpec的修改都不会触发新的rollout。Deployment被创建之后默认是非paused。 Deployment 的替代选择 kubectl rolling update Kubectl rolling update 虽然使用类似的方式更新Pod和ReplicationController。但是我们推荐使用Deployment，因为它是声明式的，客户端侧，具有附加特性，例如即使滚动升级结束后也可以回滚到任何历史版本。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-14 13:44:36 "},"concepts/secret.html":{"url":"concepts/secret.html","title":"Secret","keywords":"","body":"Secret Secret解决了密码、token、密钥等敏感数据的配置问题，而不需要把这些敏感数据暴露到镜像或者Pod Spec中。Secret可以以Volume或者环境变量的方式使用。 Secret有三种类型： Service Account ：用来访问Kubernetes API，由Kubernetes自动创建，并且会自动挂载到Pod的/run/secrets/kubernetes.io/serviceaccount目录中； Opaque ：base64编码格式的Secret，用来存储密码、密钥等； kubernetes.io/dockerconfigjson ：用来存储私有docker registry的认证信息。 Opaque Secret Opaque类型的数据是一个map类型，要求value是base64编码格式： $ echo -n \"admin\" | base64 YWRtaW4= $ echo -n \"1f2d1e2e67df\" | base64 MWYyZDFlMmU2N2Rm secrets.yml apiVersion: v1 kind: Secret metadata: name: mysecret type: Opaque data: password: MWYyZDFlMmU2N2Rm username: YWRtaW4= 接着，就可以创建secret了：kubectl create -f secrets.yml。 创建好secret之后，有两种方式来使用它： 以Volume方式 以环境变量方式 将Secret挂载到Volume中 apiVersion: v1 kind: Pod metadata: labels: name: db name: db spec: volumes: - name: secrets secret: secretName: mysecret containers: - image: gcr.io/my_project_id/pg:v1 name: db volumeMounts: - name: secrets mountPath: \"/etc/secrets\" readOnly: true ports: - name: cp containerPort: 5432 hostPort: 5432 将Secret导出到环境变量中 apiVersion: extensions/v1beta1 kind: Deployment metadata: name: wordpress-deployment spec: replicas: 2 strategy: type: RollingUpdate template: metadata: labels: app: wordpress visualize: \"true\" spec: containers: - name: \"wordpress\" image: \"wordpress\" ports: - containerPort: 80 env: - name: WORDPRESS_DB_USER valueFrom: secretKeyRef: name: mysecret key: username - name: WORDPRESS_DB_PASSWORD valueFrom: secretKeyRef: name: mysecret key: password kubernetes.io/dockerconfigjson 可以直接用kubectl命令来创建用于docker registry认证的secret： $ kubectl create secret docker-registry myregistrykey --docker-server=DOCKER_REGISTRY_SERVER --docker-username=DOCKER_USER --docker-password=DOCKER_PASSWORD --docker-email=DOCKER_EMAIL secret \"myregistrykey\" created. 也可以直接读取~/.docker/config.json的内容来创建： $ cat ~/.docker/config.json | base64 $ cat > myregistrykey.yaml 在创建Pod的时候，通过imagePullSecrets来引用刚创建的myregistrykey: apiVersion: v1 kind: Pod metadata: name: foo spec: containers: - name: foo image: janedoe/awesomeapp:v1 imagePullSecrets: - name: myregistrykey Service Account Service Account用来访问Kubernetes API，由Kubernetes自动创建，并且会自动挂载到Pod的/run/secrets/kubernetes.io/serviceaccount目录中。 $ kubectl run nginx --image nginx deployment \"nginx\" created $ kubectl get pods NAME READY STATUS RESTARTS AGE nginx-3137573019-md1u2 1/1 Running 0 13s $ kubectl exec nginx-3137573019-md1u2 ls /run/secrets/kubernetes.io/serviceaccount ca.crt namespace token Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-08-21 18:23:34 "},"concepts/statefulset.html":{"url":"concepts/statefulset.html","title":"StatefulSet","keywords":"","body":"StatefulSet StatefulSet 作为 Controller 为 Pod 提供唯一的标识。它可以保证部署和 scale 的顺序。 使用案例参考：kubernetes contrib - statefulsets，其中包含zookeeper和kakfa的statefulset设置和使用说明。 StatefulSet是为了解决有状态服务的问题（对应Deployments和ReplicaSets是为无状态服务而设计），其应用场景包括： 稳定的持久化存储，即Pod重新调度后还是能访问到相同的持久化数据，基于PVC来实现 稳定的网络标志，即Pod重新调度后其PodName和HostName不变，基于Headless Service（即没有Cluster IP的Service）来实现 有序部署，有序扩展，即Pod是有顺序的，在部署或者扩展的时候要依据定义的顺序依次依次进行（即从0到N-1，在下一个Pod运行之前所有之前的Pod必须都是Running和Ready状态），基于init containers来实现 有序收缩，有序删除（即从N-1到0） 从上面的应用场景可以发现，StatefulSet由以下几个部分组成： 用于定义网络标志（DNS domain）的Headless Service 用于创建PersistentVolumes的volumeClaimTemplates 定义具体应用的StatefulSet StatefulSet中每个Pod的DNS格式为statefulSetName-{0..N-1}.serviceName.namespace.svc.cluster.local，其中 serviceName为Headless Service的名字 0..N-1为Pod所在的序号，从0开始到N-1 statefulSetName为StatefulSet的名字 namespace为服务所在的namespace，Headless Servic和StatefulSet必须在相同的namespace .cluster.local为Cluster Domain 使用 StatefulSet StatefulSet 适用于有以下某个或多个需求的应用： 稳定，唯一的网络标志。 稳定，持久化存储。 有序，优雅地部署和 scale。 有序，优雅地删除和终止。 有序，自动的滚动升级。 在上文中，稳定是 Pod （重新）调度中持久性的代名词。 如果应用程序不需要任何稳定的标识符、有序部署、删除和 scale，则应该使用提供一组无状态副本的 controller 来部署应用程序，例如 Deployment 或 ReplicaSet 可能更适合您的无状态需求。 限制 StatefulSet 是 beta 资源，Kubernetes 1.5 以前版本不支持。 对于所有的 alpha/beta 的资源，您都可以通过在 apiserver 中设置 --runtime-config 选项来禁用。 给定 Pod 的存储必须由 PersistentVolume Provisioner 根据请求的 storage class 进行配置，或由管理员预先配置。 删除或 scale StatefulSet 将不会删除与 StatefulSet 相关联的 volume。 这样做是为了确保数据安全性，这通常比自动清除所有相关 StatefulSet 资源更有价值。 StatefulSets 目前要求 Headless Service 负责 Pod 的网络身份。 您有责任创建此服务。 组件 下面的示例中描述了 StatefulSet 中的组件。 一个名为 nginx 的 headless service，用于控制网络域。 一个名为 web 的 StatefulSet，它的 Spec 中指定在有 3 个运行 nginx 容器的 Pod。 volumeClaimTemplates 使用 PersistentVolume Provisioner 提供的 PersistentVolumes 作为稳定存储。 apiVersion: v1 kind: Service metadata: name: nginx labels: app: nginx spec: ports: - port: 80 name: web clusterIP: None selector: app: nginx --- apiVersion: apps/v1beta1 kind: StatefulSet metadata: name: web spec: serviceName: \"nginx\" replicas: 3 template: metadata: labels: app: nginx spec: terminationGracePeriodSeconds: 10 containers: - name: nginx image: gcr.io/google_containers/nginx-slim:0.8 ports: - containerPort: 80 name: web volumeMounts: - name: www mountPath: /usr/share/nginx/html volumeClaimTemplates: - metadata: name: www annotations: volume.beta.kubernetes.io/storage-class: anything spec: accessModes: [ \"ReadWriteOnce\" ] resources: requests: storage: 1Gi Pod 身份 StatefulSet Pod 具有唯一的身份，包括序数，稳定的网络身份和稳定的存储。 身份绑定到 Pod 上，不管它（重新）调度到哪个节点上。 序数 对于一个有 N 个副本的 StatefulSet，每个副本都会被指定一个整数序数，在 [0,N)之间，且唯一。 稳定的网络 ID StatefulSet 中的每个 Pod 从 StatefulSet 的名称和 Pod 的序数派生其主机名。构造的主机名的模式是$（statefulset名称)-$(序数)。 上面的例子将创建三个名为web-0，web-1，web-2的 Pod。 StatefulSet 可以使用 Headless Service 来控制其 Pod 的域。此服务管理的域的格式为：$(服务名称).$(namespace).svc.cluster.local，其中 “cluster.local” 是 集群域。 在创建每个Pod时，它将获取一个匹配的 DNS 子域，采用以下形式：$(pod 名称).$(管理服务域)，其中管理服务由 StatefulSet 上的 serviceName 字段定义。 以下是 Cluster Domain，服务名称，StatefulSet 名称以及如何影响 StatefulSet 的 Pod 的 DNS 名称的一些示例。 Cluster Domain Service (ns/name) StatefulSet (ns/name) StatefulSet Domain Pod DNS Pod Hostname cluster.local default/nginx default/web nginx.default.svc.cluster.local web-{0..N-1}.nginx.default.svc.cluster.local web-{0..N-1} cluster.local foo/nginx foo/web nginx.foo.svc.cluster.local web-{0..N-1}.nginx.foo.svc.cluster.local web-{0..N-1} kube.local foo/nginx foo/web nginx.foo.svc.kube.local web-{0..N-1}.nginx.foo.svc.kube.local web-{0..N-1} 注意 Cluster Domain 将被设置成 cluster.local 除非进行了 其他配置。 稳定存储 Kubernetes 为每个 VolumeClaimTemplate 创建一个 PersistentVolume。上面的 nginx 的例子中，每个 Pod 将具有一个由 anything 存储类创建的 1 GB 存储的 PersistentVolume。当该 Pod （重新）调度到节点上，volumeMounts 将挂载与 PersistentVolume Claim 相关联的 PersistentVolume。请注意，与 PersistentVolume Claim 相关联的 PersistentVolume 在 产出 Pod 或 StatefulSet 的时候不会被删除。这必须手动完成。 部署和 Scale 保证 对于有 N 个副本的 StatefulSet，Pod 将按照 {0..N-1} 的顺序被创建和部署。 当 删除 Pod 的时候，将按照逆序来终结，从{N-1..0} 对 Pod 执行 scale 操作之前，它所有的前任必须处于 Running 和 Ready 状态。 在终止 Pod 前，它所有的继任者必须处于完全关闭状态。 不应该将 StatefulSet 的 pod.Spec.TerminationGracePeriodSeconds 设置为 0。这样是不安全的且强烈不建议您这样做。进一步解释，请参阅 强制删除 StatefulSet Pod。 上面的 nginx 示例创建后，3 个 Pod 将按照如下顺序创建 web-0，web-1，web-2。在 web-0 处于 运行并就绪 状态之前，web-1 将不会被部署，同样当 web-1 处于运行并就绪状态之前 web-2也不会被部署。如果在 web-1 运行并就绪后，web-2 启动之前， web-0 失败了，web-2 将不会启动，直到 web-0 成果重启并处于运行并就绪状态。 如果用户通过修补 StatefulSet 来 scale 部署的示例，以使 replicas=1，则 web-2 将首先被终止。 在 web-2 完全关闭和删除之前，web-1 不会被终止。 如果 web-0 在 web-2 终止并且完全关闭之后，但是在 web-1 终止之前失败，则 web-1 将不会终止，除非 web-0 正在运行并准备就绪。 Pod 管理策略 在 Kubernetes 1.7 和之后版本，StatefulSet 允许您放开顺序保证，同时通过 .spec.podManagementPolicy 字段保证身份的唯一性。 OrderedReady Pod 管理 StatefulSet 中默认使用的是 OrderedReady pod 管理。它实现了 如上 所述的行为。 并行 Pod 管理 Parallel pod 管理告诉 StatefulSet controller 并行的启动和终止 Pod，在启动和终止其他 Pod 之前不会等待 Pod 变成 运行并就绪或完全终止状态。 更新策略 在 kubernetes 1.7 和以上版本中，StatefulSet 的 .spec.updateStrategy 字段允许您配置和禁用 StatefulSet 中的容器、label、resource request/limit、annotation 的滚动更新。 删除 OnDelete 更新策略实现了遗留（1.6和以前）的行为。 当 spec.updateStrategy 未指定时，这是默认策略。 当StatefulSet 的 .spec.updateStrategy.type 设置为 OnDelete 时，StatefulSet 控制器将不会自动更新 StatefulSet 中的 Pod。 用户必须手动删除 Pod 以使控制器创建新的 Pod，以反映对StatefulSet的 .spec.template 进行的修改。 滚动更新 RollingUpdate 更新策略在 StatefulSet 中实现 Pod 的自动滚动更新。 当StatefulSet的 .spec.updateStrategy.type 设置为 RollingUpdate 时，StatefulSet 控制器将在 StatefulSet 中删除并重新创建每个 Pod。 它将以与 Pod 终止相同的顺序进行（从最大的序数到最小的序数），每次更新一个 Pod。 在更新其前身之前，它将等待正在更新的 Pod 状态变成正在运行并就绪。 分区 可以通过指定 .spec.updateStrategy.rollingUpdate.partition 来对 RollingUpdate 更新策略进行分区。如果指定了分区，则当 StatefulSet 的 .spec.template 更新时，具有大于或等于分区序数的所有 Pod 将被更新。具有小于分区的序数的所有 Pod 将不会被更新，即使删除它们也将被重新创建。如果 StatefulSet 的 .spec.updateStrategy.rollingUpdate.partition 大于其 .spec.replicas，则其 .spec.template 的更新将不会传播到 Pod。 在大多数情况下，您不需要使用分区，但如果您想要进行分阶段更新，使用金丝雀发布或执行分阶段发布，它们将非常有用。 简单示例 以一个简单的nginx服务web.yaml为例： --- apiVersion: v1 kind: Service metadata: name: nginx labels: app: nginx spec: ports: - port: 80 name: web clusterIP: None selector: app: nginx --- apiVersion: apps/v1beta1 kind: StatefulSet metadata: name: web spec: serviceName: \"nginx\" replicas: 2 template: metadata: labels: app: nginx spec: containers: - name: nginx image: gcr.io/google_containers/nginx-slim:0.8 ports: - containerPort: 80 name: web volumeMounts: - name: www mountPath: /usr/share/nginx/html volumeClaimTemplates: - metadata: name: www annotations: volume.alpha.kubernetes.io/storage-class: anything spec: accessModes: [ \"ReadWriteOnce\" ] resources: requests: storage: 1Gi $ kubectl create -f web.yaml service \"nginx\" created statefulset \"web\" created # 查看创建的headless service和statefulset $ kubectl get service nginx NAME CLUSTER-IP EXTERNAL-IP PORT(S) AGE nginx None 80/TCP 1m $ kubectl get statefulset web NAME DESIRED CURRENT AGE web 2 2 2m # 根据volumeClaimTemplates自动创建PVC（在GCE中会自动创建kubernetes.io/gce-pd类型的volume） $ kubectl get pvc NAME STATUS VOLUME CAPACITY ACCESSMODES AGE www-web-0 Bound pvc-d064a004-d8d4-11e6-b521-42010a800002 1Gi RWO 16s www-web-1 Bound pvc-d06a3946-d8d4-11e6-b521-42010a800002 1Gi RWO 16s # 查看创建的Pod，他们都是有序的 $ kubectl get pods -l app=nginx NAME READY STATUS RESTARTS AGE web-0 1/1 Running 0 5m web-1 1/1 Running 0 4m # 使用nslookup查看这些Pod的DNS $ kubectl run -i --tty --image busybox dns-test --restart=Never --rm /bin/sh / # nslookup web-0.nginx Server: 10.0.0.10 Address 1: 10.0.0.10 kube-dns.kube-system.svc.cluster.local Name: web-0.nginx Address 1: 10.244.2.10 / # nslookup web-1.nginx Server: 10.0.0.10 Address 1: 10.0.0.10 kube-dns.kube-system.svc.cluster.local Name: web-1.nginx Address 1: 10.244.3.12 / # nslookup web-0.nginx.default.svc.cluster.local Server: 10.0.0.10 Address 1: 10.0.0.10 kube-dns.kube-system.svc.cluster.local Name: web-0.nginx.default.svc.cluster.local Address 1: 10.244.2.10 还可以进行其他的操作 # 扩容 $ kubectl scale statefulset web --replicas=5 # 缩容 $ kubectl patch statefulset web -p '{\"spec\":{\"replicas\":3}}' # 镜像更新（目前还不支持直接更新image，需要patch来间接实现） $ kubectl patch statefulset web --type='json' -p='[{\"op\": \"replace\", \"path\": \"/spec/template/spec/containers/0/image\", \"value\":\"gcr.io/google_containers/nginx-slim:0.7\"}]' # 删除StatefulSet和Headless Service $ kubectl delete statefulset web $ kubectl delete service nginx # StatefulSet删除后PVC还会保留着，数据不再使用的话也需要删除 $ kubectl delete pvc www-web-0 www-web-1 zookeeper 另外一个更能说明StatefulSet强大功能的示例为zookeeper.yaml，这个例子仅为讲解，实际可用的配置请使用 https://github.com/kubernetes/contrib/tree/master/statefulsets 中的配置。 --- apiVersion: v1 kind: Service metadata: name: zk-headless labels: app: zk-headless spec: ports: - port: 2888 name: server - port: 3888 name: leader-election clusterIP: None selector: app: zk --- apiVersion: v1 kind: ConfigMap metadata: name: zk-config data: ensemble: \"zk-0;zk-1;zk-2\" jvm.heap: \"2G\" tick: \"2000\" init: \"10\" sync: \"5\" client.cnxns: \"60\" snap.retain: \"3\" purge.interval: \"1\" --- apiVersion: policy/v1beta1 kind: PodDisruptionBudget metadata: name: zk-budget spec: selector: matchLabels: app: zk minAvailable: 2 --- apiVersion: apps/v1beta1 kind: StatefulSet metadata: name: zk spec: serviceName: zk-headless replicas: 3 template: metadata: labels: app: zk annotations: pod.alpha.kubernetes.io/initialized: \"true\" scheduler.alpha.kubernetes.io/affinity: > { \"podAntiAffinity\": { \"requiredDuringSchedulingRequiredDuringExecution\": [{ \"labelSelector\": { \"matchExpressions\": [{ \"key\": \"app\", \"operator\": \"In\", \"values\": [\"zk-headless\"] }] }, \"topologyKey\": \"kubernetes.io/hostname\" }] } } spec: containers: - name: k8szk imagePullPolicy: Always image: gcr.io/google_samples/k8szk:v1 resources: requests: memory: \"4Gi\" cpu: \"1\" ports: - containerPort: 2181 name: client - containerPort: 2888 name: server - containerPort: 3888 name: leader-election env: - name : ZK_ENSEMBLE valueFrom: configMapKeyRef: name: zk-config key: ensemble - name : ZK_HEAP_SIZE valueFrom: configMapKeyRef: name: zk-config key: jvm.heap - name : ZK_TICK_TIME valueFrom: configMapKeyRef: name: zk-config key: tick - name : ZK_INIT_LIMIT valueFrom: configMapKeyRef: name: zk-config key: init - name : ZK_SYNC_LIMIT valueFrom: configMapKeyRef: name: zk-config key: tick - name : ZK_MAX_CLIENT_CNXNS valueFrom: configMapKeyRef: name: zk-config key: client.cnxns - name: ZK_SNAP_RETAIN_COUNT valueFrom: configMapKeyRef: name: zk-config key: snap.retain - name: ZK_PURGE_INTERVAL valueFrom: configMapKeyRef: name: zk-config key: purge.interval - name: ZK_CLIENT_PORT value: \"2181\" - name: ZK_SERVER_PORT value: \"2888\" - name: ZK_ELECTION_PORT value: \"3888\" command: - sh - -c - zkGenConfig.sh && zkServer.sh start-foreground readinessProbe: exec: command: - \"zkOk.sh\" initialDelaySeconds: 15 timeoutSeconds: 5 livenessProbe: exec: command: - \"zkOk.sh\" initialDelaySeconds: 15 timeoutSeconds: 5 volumeMounts: - name: datadir mountPath: /var/lib/zookeeper securityContext: runAsUser: 1000 fsGroup: 1000 volumeClaimTemplates: - metadata: name: datadir annotations: volume.alpha.kubernetes.io/storage-class: anything spec: accessModes: [ \"ReadWriteOnce\" ] resources: requests: storage: 20Gi kubectl create -f zookeeper.yaml 详细的使用说明见zookeeper stateful application。 关于StatefulSet的更多示例请参阅 github.com/kubernetes/contrib - statefulsets，其中包括了zookeeper和kafka。 集群外部访问StatefulSet的Pod 我们设想一下这样的场景：在kubernetes集群外部调试StatefulSet中有序的Pod，那么如何访问这些的pod呢？ 方法是为pod设置label，然后用kubectl expose将其以NodePort的方式暴露到集群外部，以上面的zookeeper的例子来说明，下面使用命令的方式来暴露其中的两个zookeeper节点，也可以写一个serivce配置yaml文件。 kubectl label pod zk-0 zkInst=0 kubectl label pod zk-1 zkInst=1 kubectl expose po zk-0 --port=2181 --target-port=2181 --name=zk-0 --selector=zkInst=0 --type=NodePort kubectl expose po zk-1 --port=2181 --target-port=2181 --name=zk-1 --selector=zkInst=1 --type=NodePort 这样在kubernetes集群外部就可以根据pod所在的主机所映射的端口来访问了。 查看zk-0这个service可以看到如下结果： NAME CLUSTER-IP EXTERNAL-IP PORT(S) AGE zk-0 10.254.98.14 2181:31693/TCP 5m 集群外部就可以使用所有的node中的任何一个IP:31693来访问这个zookeeper实例。 参考 https://kubernetes.io/docs/concepts/workloads/controllers/statefulset/ kubernetes contrib - statefulsets Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-10-23 13:32:31 "},"concepts/daemonset.html":{"url":"concepts/daemonset.html","title":"DaemonSet","keywords":"","body":"DaemonSet 什么是 DaemonSet？ DaemonSet 确保全部（或者一些）Node 上运行一个 Pod 的副本。当有 Node 加入集群时，也会为他们新增一个 Pod 。当有 Node 从集群移除时，这些 Pod 也会被回收。删除 DaemonSet 将会删除它创建的所有 Pod。 使用 DaemonSet 的一些典型用法： 运行集群存储 daemon，例如在每个 Node 上运行 glusterd、ceph。 在每个 Node 上运行日志收集 daemon，例如fluentd、logstash。 在每个 Node 上运行监控 daemon，例如 Prometheus Node Exporter、collectd、Datadog 代理、New Relic 代理，或 Ganglia gmond。 一个简单的用法是，在所有的 Node 上都存在一个 DaemonSet，将被作为每种类型的 daemon 使用。 一个稍微复杂的用法可能是，对单独的每种类型的 daemon 使用多个 DaemonSet，但具有不同的标志，和/或对不同硬件类型具有不同的内存、CPU要求。 编写 DaemonSet Spec 必需字段 和其它所有 Kubernetes 配置一样，DaemonSet 需要 apiVersion、kind 和 metadata字段。有关配置文件的通用信息，详见文档 部署应用、配置容器 和 资源管理 。 DaemonSet 也需要一个 .spec 配置段。 Pod 模板 .spec 唯一必需的字段是 .spec.template。 .spec.template 是一个 Pod 模板。 它与 Pod 具有相同的 schema，除了它是嵌套的，而且不具有 apiVersion 或 kind 字段。 Pod 除了必须字段外，在 DaemonSet 中的 Pod 模板必须指定合理的标签（查看 pod selector）。 在 DaemonSet 中的 Pod 模板必需具有一个值为 Always 的 RestartPolicy，或者未指定它的值，默认是 Always。 Pod Selector .spec.selector 字段表示 Pod Selector，它与 Job 或其它资源的 .sper.selector 的原理是相同的。 spec.selector 表示一个对象，它由如下两个字段组成： matchLabels - 与 ReplicationController 的 .spec.selector 的原理相同。 matchExpressions - 允许构建更加复杂的 Selector，可以通过指定 key、value 列表，以及与 key 和 value 列表的相关的操作符。 当上述两个字段都指定时，结果表示的是 AND 关系。 如果指定了 .spec.selector，必须与 .spec.template.metadata.labels 相匹配。如果没有指定，它们默认是等价的。如果与它们配置的不匹配，则会被 API 拒绝。 如果 Pod 的 label 与 selector 匹配，或者直接基于其它的 DaemonSet、或者 Controller（例如 ReplicationController），也不可以创建任何 Pod。 否则 DaemonSet Controller 将认为那些 Pod 是它创建的。Kubernetes 不会阻止这样做。一个场景是，可能希望在一个具有不同值的、用来测试用的 Node 上手动创建 Pod。 仅在相同的 Node 上运行 Pod 如果指定了 .spec.template.spec.nodeSelector，DaemonSet Controller 将在能够匹配上 Node Selector 的 Node 上创建 Pod。 类似这种情况，可以指定 .spec.template.spec.affinity，然后 DaemonSet Controller 将在能够匹配上 Node Affinity 的 Node 上创建 Pod。 如果根本就没有指定，则 DaemonSet Controller 将在所有 Node 上创建 Pod。 如果调度 Daemon Pod 正常情况下，Pod 运行在哪个机器上是由 Kubernetes 调度器进行选择的。然而，由 Daemon Controller 创建的 Pod 已经确定了在哪个机器上（Pod 创建时指定了 .spec.nodeName），因此： DaemonSet Controller 并不关心一个 Node 的 unschedulable 字段。 DaemonSet Controller 可以创建 Pod，即使调度器还没有被启动，这对集群启动是非常有帮助的。 Daemon Pod 关心 Taint 和 Toleration，它们会为没有指定 tolerationSeconds 的 node.alpha.kubernetes.io/notReady 和 node.alpha.kubernetes.io/unreachable 的 Taint，而创建具有 NoExecute 的 Toleration。这确保了当 alpha 特性的 TaintBasedEvictions 被启用，当 Node 出现故障，比如网络分区，这时它们将不会被清除掉（当 TaintBasedEvictions 特性没有启用，在这些场景下也不会被清除，但会因为 NodeController 的硬编码行为而被清除，Toleration 是不会的）。 与 Daemon Pod 通信 与 DaemonSet 中的 Pod 进行通信，几种可能的模式如下： Push：配置 DaemonSet 中的 Pod 向其它 Service 发送更新，例如统计数据库。它们没有客户端。 NodeIP 和已知端口：DaemonSet 中的 Pod 可以使用 hostPort，从而可以通过 Node IP 访问到 Pod。客户端能通过某种方法知道 Node IP 列表，并且基于此也可以知道端口。 DNS：创建具有相同 Pod Selector 的 Headless Service，然后通过使用 endpoints 资源或从 DNS 检索到多个 A 记录来发现 DaemonSet。 Service：创建具有相同 Pod Selector 的 Service，并使用该 Service 访问到某个随机 Node 上的 daemon。（没有办法访问到特定 Node） 更新 DaemonSet 如果修改了 Node Label，DaemonSet 将立刻向新匹配上的 Node 添加 Pod，同时删除新近无法匹配上的 Node 上的 Pod。 可以修改 DaemonSet 创建的 Pod。然而，不允许对 Pod 的所有字段进行更新。当下次 Node（即使具有相同的名称）被创建时，DaemonSet Controller 还会使用最初的模板。 可以删除一个 DaemonSet。如果使用 kubectl 并指定 --cascade=false 选项，则 Pod 将被保留在 Node 上。然后可以创建具有不同模板的新 DaemonSet。具有不同模板的新 DaemonSet 将鞥能够通过 Label 匹配识别所有已经存在的 Pod。它不会修改或删除它们，即使是错误匹配了 Pod 模板。通过删除 Pod 或者 删除 Node，可以强制创建新的 Pod。 在 Kubernetes 1.6 或以后版本，可以在 DaemonSet 上 执行滚动升级。 init 脚本 很可能通过直接在一个 Node 上启动 daemon 进程（例如，使用 init、upstartd、或 systemd）。这非常好，然而基于 DaemonSet 来运行这些进程有如下一些好处： 像对待应用程序一样，具备为 daemon 提供监控和管理日志的能力。 为 daemon 和应用程序使用相同的配置语言和工具（如 Pod 模板、kubectl）。 Kubernetes 未来版本可能会支持对 DaemonSet 创建 Pod 与 Node升级工作流进行集成。 在资源受限的容器中运行 daemon，能够增加 daemon 和应用容器的隔离性。然而这也实现了在容器中运行 daemon，但却不能在 Pod 中运行（例如，直接基于 Docker 启动）。 裸 Pod 可能要直接创建 Pod，同时指定其运行在特定的 Node 上。 然而，DaemonSet 替换了由于任何原因被删除或终止的 Pod，例如 Node 失败、例行节点维护，比如内核升级。由于这个原因，我们应该使用 DaemonSet 而不是单独创建 Pod。 静态 Pod 很可能，通过在一个指定目录下编写文件来创建 Pod，该目录受 Kubelet 所监视。这些 Pod 被称为 静态 Pod。 不像 DaemonSet，静态 Pod 不受 kubectl 和 其它 Kubernetes API 客户端管理。静态 Pod 不依赖于 apiserver，这使得它们在集群启动的情况下非常有用。 而且，未来静态 Pod 可能会被废弃掉。 Replication Controller DaemonSet 与 Replication Controller 非常类似，它们都能创建 Pod，这些 Pod 都具有不期望被终止的进程（例如，Web 服务器、存储服务器）。 为无状态的 Service 使用 Replication Controller，像 frontend，实现对副本的数量进行扩缩容、平滑升级，比之于精确控制 Pod 运行在某个主机上要重要得多。需要 Pod 副本总是运行在全部或特定主机上，并需要先于其他 Pod 启动，当这被认为非常重要时，应该使用 Daemon Controller。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-22 17:00:47 "},"concepts/serviceaccount.html":{"url":"concepts/serviceaccount.html","title":"ServiceAccount","keywords":"","body":"Service Account Service account为Pod中的进程提供身份信息。 本文是关于 Service Account 的用户指南，管理指南另见 Service Account 的集群管理指南 。 注意：本文档描述的关于 Serivce Account 的行为只有当您按照 Kubernetes 项目建议的方式搭建起集群的情况下才有效。您的集群管理员可能在您的集群中有自定义配置，这种情况下该文档可能并不适用。 当您（真人用户）访问集群（例如使用kubectl命令）时，apiserver 会将您认证为一个特定的 User Account（目前通常是admin，除非您的系统管理员自定义了集群配置）。Pod 容器中的进程也可以与 apiserver 联系。 当它们在联系 apiserver 的时候，它们会被认证为一个特定的 Service Account（例如default）。 使用默认的 Service Account 访问 API server 当您创建 pod 的时候，如果您没有指定一个 service account，系统会自动得在与该pod 相同的 namespace 下为其指派一个default service account。如果您获取刚创建的 pod 的原始 json 或 yaml 信息（例如使用kubectl get pods/podename -o yaml命令），您将看到spec.serviceAccountName字段已经被设置为 automatically set 。 您可以在 pod 中使用自动挂载的 service account 凭证来访问 API，如 Accessing the Cluster 中所描述。 Service account 是否能够取得访问 API 的许可取决于您使用的 授权插件和策略。 在 1.6 以上版本中，您可以选择取消为 serivce account 自动挂载 API 凭证，只需在 service account 中设置 automountServiceAccountToken: false： apiVersion: v1 kind: ServiceAccount metadata: name: build-robot automountServiceAccountToken: false ... 在 1.6 以上版本中，您也可以选择只取消单个 pod 的 API 凭证自动挂载： apiVersion: v1 kind: Pod metadata: name: my-pod spec: serviceAccountName: build-robot automountServiceAccountToken: false ... 如果在 pod 和 service account 中同时设置了 automountServiceAccountToken , pod 设置中的优先级更高。 使用多个Service Account 每个 namespace 中都有一个默认的叫做 default 的 service account 资源。 您可以使用以下命令列出 namespace 下的所有 serviceAccount 资源。 $ kubectl get serviceAccounts NAME SECRETS AGE default 1 1d 您可以像这样创建一个 ServiceAccount 对象： $ cat > /tmp/serviceaccount.yaml 如果您看到如下的 service account 对象的完整输出信息： $ kubectl get serviceaccounts/build-robot -o yaml apiVersion: v1 kind: ServiceAccount metadata: creationTimestamp: 2015-06-16T00:12:59Z name: build-robot namespace: default resourceVersion: \"272500\" selfLink: /api/v1/namespaces/default/serviceaccounts/build-robot uid: 721ab723-13bc-11e5-aec2-42010af0021e secrets: - name: build-robot-token-bvbk5 然后您将看到有一个 token 已经被自动创建，并被 service account 引用。 您可以使用授权插件来 设置 service account 的权限 。 设置非默认的 service account，只需要在 pod 的spec.serviceAccountName 字段中将name设置为您想要用的 service account 名字即可。 在 pod 创建之初 service account 就必须已经存在，否则创建将被拒绝。 您不能更新已创建的 pod 的 service account。 您可以清理 service account，如下所示： $ kubectl delete serviceaccount/build-robot 手动创建 service account 的 API token 假设我们已经有了一个如上文提到的名为 ”build-robot“ 的 service account，我们手动创建一个新的 secret。 $ cat > /tmp/build-robot-secret.yaml 现在您可以确认下新创建的 secret 取代了 “build-robot” 这个 service account 原来的 API token。 所有已不存在的 service account 的 token 将被 token controller 清理掉。 $ kubectl describe secrets/build-robot-secret Name: build-robot-secret Namespace: default Labels: Annotations: kubernetes.io/service-account.name=build-robot,kubernetes.io/service-account.uid=870ef2a5-35cf-11e5-8d06-005056b45392 Type: kubernetes.io/service-account-token Data ==== ca.crt: 1220 bytes token: ... namespace: 7 bytes 注意该内容中的token被省略了。 为 service account 添加 ImagePullSecret 首先，创建一个 imagePullSecret，详见这里。 然后，确认已创建。如： $ kubectl get secrets myregistrykey NAME TYPE DATA AGE myregistrykey kubernetes.io/.dockerconfigjson 1 1d 然后，修改 namespace 中的默认 service account 使用该 secret 作为 imagePullSecret。 kubectl patch serviceaccount default -p '{\"imagePullSecrets\": [{\"name\": \"myregistrykey\"}]}' Vi 交互过程中需要手动编辑： $ kubectl get serviceaccounts default -o yaml > ./sa.yaml $ cat sa.yaml apiVersion: v1 kind: ServiceAccount metadata: creationTimestamp: 2015-08-07T22:02:39Z name: default namespace: default resourceVersion: \"243024\" selfLink: /api/v1/namespaces/default/serviceaccounts/default uid: 052fb0f4-3d50-11e5-b066-42010af0d7b6 secrets: - name: default-token-uudge $ vi sa.yaml [editor session not shown] [delete line with key \"resourceVersion\"] [add lines with \"imagePullSecret:\"] $ cat sa.yaml apiVersion: v1 kind: ServiceAccount metadata: creationTimestamp: 2015-08-07T22:02:39Z name: default namespace: default selfLink: /api/v1/namespaces/default/serviceaccounts/default uid: 052fb0f4-3d50-11e5-b066-42010af0d7b6 secrets: - name: default-token-uudge imagePullSecrets: - name: myregistrykey $ kubectl replace serviceaccount default -f ./sa.yaml serviceaccounts/default 现在，所有当前 namespace 中新创建的 pod 的 spec 中都会增加如下内容： spec: imagePullSecrets: - name: myregistrykey 参考 https://kubernetes.io/docs/tasks/configure-pod-container/configure-service-account/ Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-08-29 12:22:11 "},"concepts/replicaset.html":{"url":"concepts/replicaset.html","title":"ReplicationController和ReplicaSet","keywords":"","body":"ReplicationController和ReplicaSet ReplicationCtronller用来确保容器应用的副本数始终保持在用户定义的副本数，即如果有容器异常退出，会自动创建新的Pod来替代；而如果异常多出来的容器也会自动回收。 在新版本的Kubernetes中建议使用ReplicaSet来取代ReplicationCtronller。ReplicaSet跟ReplicationCtronller没有本质的不同，只是名字不一样，并且ReplicaSet支持集合式的selector。 虽然ReplicaSet可以独立使用，但一般还是建议使用 Deployment 来自动管理ReplicaSet，这样就无需担心跟其他机制的不兼容问题（比如ReplicaSet不支持rolling-update但Deployment支持）。 ReplicaSet示例： apiVersion: extensions/v1beta1 kind: ReplicaSet metadata: name: frontend # these labels can be applied automatically # from the labels in the pod template if not set # labels: # app: guestbook # tier: frontend spec: # this replicas value is default # modify it according to your case replicas: 3 # selector can be applied automatically # from the labels in the pod template if not set, # but we are specifying the selector here to # demonstrate its usage. selector: matchLabels: tier: frontend matchExpressions: - {key: tier, operator: In, values: [frontend]} template: metadata: labels: app: guestbook tier: frontend spec: containers: - name: php-redis image: gcr.io/google_samples/gb-frontend:v3 resources: requests: cpu: 100m memory: 100Mi env: - name: GET_HOSTS_FROM value: dns # If your cluster config does not include a dns service, then to # instead access environment variables to find service host # info, comment out the 'value: dns' line above, and uncomment the # line below. # value: env ports: - containerPort: 80 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-08-21 18:23:34 "},"concepts/job.html":{"url":"concepts/job.html","title":"Job","keywords":"","body":"Job Job负责批处理任务，即仅执行一次的任务，它保证批处理任务的一个或多个Pod成功结束。 Job Spec格式 spec.template格式同Pod RestartPolicy仅支持Never或OnFailure 单个Pod时，默认Pod成功运行后Job即结束 .spec.completions标志Job结束需要成功运行的Pod个数，默认为1 .spec.parallelism标志并行运行的Pod的个数，默认为1 spec.activeDeadlineSeconds标志失败Pod的重试最大时间，超过这个时间不会继续重试 一个简单的例子： apiVersion: batch/v1 kind: Job metadata: name: pi spec: template: metadata: name: pi spec: containers: - name: pi image: perl command: [\"perl\", \"-Mbignum=bpi\", \"-wle\", \"print bpi(2000)\"] restartPolicy: Never $ kubectl create -f ./job.yaml job \"pi\" created $ pods=$(kubectl get pods --selector=job-name=pi --output=jsonpath={.items..metadata.name}) $ kubectl logs $pods 3.141592653589793238462643383279502... Bare Pods 所谓Bare Pods是指直接用PodSpec来创建的Pod（即不在ReplicaSets或者ReplicationController的管理之下的Pods）。这些Pod在Node重启后不会自动重启，但Job则会创建新的Pod继续任务。所以，推荐使用Job来替代Bare Pods，即便是应用只需要一个Pod。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-08-21 18:23:34 "},"concepts/cronjob.html":{"url":"concepts/cronjob.html","title":"CronJob","keywords":"","body":"CronJob Cron Job 管理基于时间的 Job，即： 在给定时间点只运行一次 周期性地在给定时间点运行 一个 CronJob 对象类似于 crontab （cron table）文件中的一行。它根据指定的预定计划周期性地运行一个 Job，格式可以参考 Cron 。 前提条件 当使用的 Kubernetes 集群，版本 >= 1.4（对 ScheduledJob），>= 1.5（对 CronJob），当启动 API Server（参考 为集群开启或关闭 API 版本 获取更多信息）时，通过传递选项 --runtime-config=batch/v2alpha1=true 可以开启 batch/v2alpha1 API。 典型的用法如下所示： 在给定的时间点调度 Job 运行 创建周期性运行的 Job，例如：数据库备份、发送邮件。 CronJob Spec .spec.schedule：调度，必需字段，指定任务运行周期，格式同 Cron .spec.jobTemplate：Job 模板，必需字段，指定需要运行的任务，格式同 Job .spec.startingDeadlineSeconds ：启动 Job 的期限（秒级别），该字段是可选的。如果因为任何原因而错过了被调度的时间，那么错过执行时间的 Job 将被认为是失败的。如果没有指定，则没有期限 .spec.concurrencyPolicy：并发策略，该字段也是可选的。它指定了如何处理被 Cron Job 创建的 Job 的并发执行。只允许指定下面策略中的一种： Allow（默认）：允许并发运行 Job Forbid：禁止并发运行，如果前一个还没有完成，则直接跳过下一个 Replace：取消当前正在运行的 Job，用一个新的来替换 注意，当前策略只能应用于同一个 Cron Job 创建的 Job。如果存在多个 Cron Job，它们创建的 Job 之间总是允许并发运行。 .spec.suspend ：挂起，该字段也是可选的。如果设置为 true，后续所有执行都会被挂起。它对已经开始执行的 Job 不起作用。默认值为 false。 .spec.successfulJobsHistoryLimit 和 .spec.failedJobsHistoryLimit ：历史限制，是可选的字段。它们指定了可以保留多少完成和失败的 Job。 默认没有限制，所有成功和失败的 Job 都会被保留。然而，当运行一个 Cron Job 时，Job 可以很快就堆积很多，推荐设置这两个字段的值。设置限制的值为 0，相关类型的 Job 完成后将不会被保留。 apiVersion: batch/v2alpha1 kind: CronJob metadata: name: hello spec: schedule: \"*/1 * * * *\" jobTemplate: spec: template: spec: containers: - name: hello image: busybox args: - /bin/sh - -c - date; echo Hello from the Kubernetes cluster restartPolicy: OnFailure $ kubectl create -f cronjob.yaml cronjob \"hello\" created 当然，也可以用kubectl run来创建一个CronJob： kubectl run hello --schedule=\"*/1 * * * *\" --restart=OnFailure --image=busybox -- /bin/sh -c \"date; echo Hello from the Kubernetes cluster\" $ kubectl get cronjob NAME SCHEDULE SUSPEND ACTIVE LAST-SCHEDULE hello */1 * * * * False 0 $ kubectl get jobs NAME DESIRED SUCCESSFUL AGE hello-1202039034 1 1 49s $ pods=$(kubectl get pods --selector=job-name=hello-1202039034 --output=jsonpath={.items..metadata.name} -a) $ kubectl logs $pods Mon Aug 29 21:34:09 UTC 2016 Hello from the Kubernetes cluster # 注意，删除cronjob的时候不会自动删除job，这些job可以用kubectl delete job来删除 $ kubectl delete cronjob hello cronjob \"hello\" deleted Cron Job 限制 Cron Job 在每次调度运行时间内 大概 会创建一个 Job 对象。我们之所以说 大概 ，是因为在特定的环境下可能会创建两个 Job，或者一个 Job 都没创建。我们尝试少发生这种情况，但却不能完全避免。因此，创建 Job 操作应该是 幂等的。 Job 根据它所创建的 Pod 的并行度，负责重试创建 Pod，并就决定这一组 Pod 的成功或失败。Cron Job 根本就不会去检查 Pod。 删除 Cron Job 一旦不再需要 Cron Job，简单地可以使用 kubectl 命令删除它： $ kubectl delete cronjob hello cronjob \"hello\" deleted 这将会终止正在创建的 Job。然而，运行中的 Job 将不会被终止，不会删除 Job 或 它们的 Pod。为了清理那些 Job 和 Pod，需要列出该 Cron Job 创建的全部 Job，然后删除它们： $ kubectl get jobs NAME DESIRED SUCCESSFUL AGE hello-1201907962 1 1 11m hello-1202039034 1 1 8m ... $ kubectl delete jobs hello-1201907962 hello-1202039034 ... job \"hello-1201907962\" deleted job \"hello-1202039034\" deleted ... 一旦 Job 被删除，由 Job 创建的 Pod 也会被删除。注意，所有由名称为 “hello” 的 Cron Job 创建的 Job 会以前缀字符串 “hello-” 进行命名。如果想要删除当前 Namespace 中的所有 Job，可以通过命令 kubectl delete jobs --all 立刻删除它们。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-09-03 14:01:40 "},"concepts/ingress.html":{"url":"concepts/ingress.html","title":"Ingress","keywords":"","body":"Ingress解析 前言 这是kubernete官方文档中Ingress Resource的翻译，后面的章节会讲到使用Traefik来做Ingress controller，文章末尾给出了几个相关链接。 术语 在本篇文章中你将会看到一些在其他地方被交叉使用的术语，为了防止产生歧义，我们首先来澄清下。 节点：Kubernetes集群中的一台物理机或者虚拟机。 集群：位于Internet防火墙后的节点，这是kubernetes管理的主要计算资源。 边界路由器：为集群强制执行防火墙策略的路由器。 这可能是由云提供商或物理硬件管理的网关。 集群网络：一组逻辑或物理链接，可根据Kubernetes网络模型实现群集内的通信。 集群网络的实现包括Overlay模型的 flannel 和基于SDN的OVS。 服务：使用标签选择器标识一组pod成为的Kubernetes服务。 除非另有说明，否则服务假定在集群网络内仅可通过虚拟IP访问。 什么是Ingress？ 通常情况下，service和pod仅可在集群内部网络中通过IP地址访问。所有到达边界路由器的流量或被丢弃或被转发到其他地方。从概念上讲，可能像下面这样： internet | ------------ [ Services ] Ingress是授权入站连接到达集群服务的规则集合。 internet | [ Ingress ] --|-----|-- [ Services ] 你可以给Ingress配置提供外部可访问的URL、负载均衡、SSL、基于名称的虚拟主机等。用户通过POST Ingress资源到API server的方式来请求ingress。 Ingress controller负责实现Ingress，通常使用负载平衡器，它还可以配置边界路由和其他前端，这有助于以HA方式处理流量。 先决条件 在使用Ingress resource之前，有必要先了解下面几件事情。Ingress是beta版本的resource，在kubernetes1.1之前还没有。你需要一个Ingress Controller来实现Ingress，单纯的创建一个Ingress没有任何意义。 GCE/GKE会在master节点上部署一个ingress controller。你可以在一个pod中部署任意个自定义的ingress controller。你必须正确地annotate每个ingress，比如 运行多个ingress controller 和 关闭glbc. 确定你已经阅读了Ingress controller的beta版本限制。在非GCE/GKE的环境中，你需要在pod中部署一个controller。 Ingress Resource 最简化的Ingress配置： 1: apiVersion: extensions/v1beta1 2: kind: Ingress 3: metadata: 4: name: test-ingress 5: spec: 6: rules: 7: - http: 8: paths: 9: - path: /testpath 10: backend: 11: serviceName: test 12: servicePort: 80 如果你没有配置Ingress controller就将其POST到API server不会有任何用处 配置说明 1-4行：跟Kubernetes的其他配置一样，ingress的配置也需要apiVersion，kind和metadata字段。配置文件的详细说明请查看部署应用, 配置容器和 使用resources. 5-7行: Ingress spec 中包含配置一个loadbalancer或proxy server的所有信息。最重要的是，它包含了一个匹配所有入站请求的规则列表。目前ingress只支持http规则。 8-9行：每条http规则包含以下信息：一个host配置项（比如for.bar.com，在这个例子中默认是*），path列表（比如：/testpath），每个path都关联一个backend(比如test:80)。在loadbalancer将流量转发到backend之前，所有的入站请求都要先匹配host和path。 10-12行：正如 services doc中描述的那样，backend是一个service:port的组合。Ingress的流量被转发到它所匹配的backend。 全局参数：为了简单起见，Ingress示例中没有全局参数，请参阅资源完整定义的api参考。 在所有请求都不能跟spec中的path匹配的情况下，请求被发送到Ingress controller的默认后端，可以指定全局缺省backend。 Ingress controllers 为了使Ingress正常工作，集群中必须运行Ingress controller。 这与其他类型的控制器不同，其他类型的控制器通常作为kube-controller-manager二进制文件的一部分运行，在集群启动时自动启动。 你需要选择最适合自己集群的Ingress controller或者自己实现一个。 示例和说明可以在这里找到。 在你开始前 以下文档描述了Ingress资源中公开的一组跨平台功能。 理想情况下，所有的Ingress controller都应该符合这个规范，但是我们还没有实现。 GCE和nginx控制器的文档分别在这里和这里。确保您查看控制器特定的文档，以便您了解每个文档的注意事项。 Ingress类型 单Service Ingress Kubernetes中已经存在一些概念可以暴露单个service（查看替代方案），但是你仍然可以通过Ingress来实现，通过指定一个没有rule的默认backend的方式。 ingress.yaml定义文件： apiVersion: extensions/v1beta1 kind: Ingress metadata: name: test-ingress spec: backend: serviceName: testsvc servicePort: 80 使用kubectl create -f命令创建，然后查看ingress： $ kubectl get ing NAME RULE BACKEND ADDRESS test-ingress - testsvc:80 107.178.254.228 107.178.254.228就是Ingress controller为了实现Ingress而分配的IP地址。RULE列表示所有发送给该IP的流量都被转发到了BACKEND所列的Kubernetes service上。 简单展开 如前面描述的那样，kubernete pod中的IP只在集群网络内部可见，我们需要在边界设置一个东西，让它能够接收ingress的流量并将它们转发到正确的端点上。这个东西一般是高可用的loadbalancer。使用Ingress能够允许你将loadbalancer的个数降低到最少，例如，假如你想要创建这样的一个设置： foo.bar.com -> 178.91.123.132 -> / foo s1:80 / bar s2:80 你需要一个这样的ingress： apiVersion: extensions/v1beta1 kind: Ingress metadata: name: test spec: rules: - host: foo.bar.com http: paths: - path: /foo backend: serviceName: s1 servicePort: 80 - path: /bar backend: serviceName: s2 servicePort: 80 使用kubectl create -f创建完ingress后： $ kubectl get ing NAME RULE BACKEND ADDRESS test - foo.bar.com /foo s1:80 /bar s2:80 只要服务（s1，s2）存在，Ingress controller就会将提供一个满足该Ingress的特定loadbalancer实现。 这一步完成后，您将在Ingress的最后一列看到loadbalancer的地址。 基于名称的虚拟主机 Name-based的虚拟主机在同一个IP地址下拥有多个主机名。 foo.bar.com --| |-> foo.bar.com s1:80 | 178.91.123.132 | bar.foo.com --| |-> bar.foo.com s2:80 下面这个ingress说明基于Host header的后端loadbalancer的路由请求： apiVersion: extensions/v1beta1 kind: Ingress metadata: name: test spec: rules: - host: foo.bar.com http: paths: - backend: serviceName: s1 servicePort: 80 - host: bar.foo.com http: paths: - backend: serviceName: s2 servicePort: 80 默认backend：一个没有rule的ingress，如前面章节中所示，所有流量都将发送到一个默认backend。你可以用该技巧通知loadbalancer如何找到你网站的404页面，通过制定一些列rule和一个默认backend的方式。如果请求header中的host不能跟ingress中的host匹配，并且/或请求的URL不能与任何一个path匹配，则流量将路由到你的默认backend。 TLS 你可以通过指定包含TLS私钥和证书的secret来加密Ingress。 目前，Ingress仅支持单个TLS端口443，并假定TLS termination。 如果Ingress中的TLS配置部分指定了不同的主机，则它们将根据通过SNI TLS扩展指定的主机名（假如Ingress controller支持SNI）在多个相同端口上进行复用。 TLS secret中必须包含名为tls.crt和tls.key的密钥，这里面包含了用于TLS的证书和私钥，例如： apiVersion: v1 data: tls.crt: base64 encoded cert tls.key: base64 encoded key kind: Secret metadata: name: testsecret namespace: default type: Opaque 在Ingress中引用这个secret将通知Ingress controller使用TLS加密从将客户端到loadbalancer的channel： apiVersion: extensions/v1beta1 kind: Ingress metadata: name: no-rules-map spec: tls: - secretName: testsecret backend: serviceName: s1 servicePort: 80 请注意，各种Ingress controller支持的TLS功能之间存在差距。 请参阅有关nginx，GCE或任何其他平台特定Ingress controller的文档，以了解TLS在你的环境中的工作原理。 Ingress controller启动时附带一些适用于所有Ingress的负载平衡策略设置，例如负载均衡算法，后端权重方案等。更高级的负载平衡概念（例如持久会话，动态权重）尚未在Ingress中公开。 你仍然可以通过service loadbalancer获取这些功能。 随着时间的推移，我们计划将适用于跨平台的负载平衡模式加入到Ingress资源中。 还值得注意的是，尽管健康检查不直接通过Ingress公开，但Kubernetes中存在并行概念，例如准备探查，可以使你达成相同的最终结果。 请查看特定控制器的文档，以了解他们如何处理健康检查（nginx，GCE）。 更新Ingress 假如你想要向已有的ingress中增加一个新的Host，你可以编辑和更新该ingress： $ kubectl get ing NAME RULE BACKEND ADDRESS test - 178.91.123.132 foo.bar.com /foo s1:80 $ kubectl edit ing test 这会弹出一个包含已有的yaml文件的编辑器，修改它，增加新的Host配置。 spec: rules: - host: foo.bar.com http: paths: - backend: serviceName: s1 servicePort: 80 path: /foo - host: bar.baz.com http: paths: - backend: serviceName: s2 servicePort: 80 path: /foo .. 保存它会更新API server中的资源，这会触发ingress controller重新配置loadbalancer。 $ kubectl get ing NAME RULE BACKEND ADDRESS test - 178.91.123.132 foo.bar.com /foo s1:80 bar.baz.com /foo s2:80 在一个修改过的ingress yaml文件上调用kubectl replace -f命令一样可以达到同样的效果。 跨可用域故障 在不通云供应商之间，跨故障域的流量传播技术有所不同。 有关详细信息，请查看相关Ingress controller的文档。 有关在federation集群中部署Ingress的详细信息，请参阅federation文档。 未来计划 多样化的HTTPS/TLS模型支持（如SNI，re-encryption） 通过声明来请求IP或者主机名 结合L4和L7 Ingress 更多的Ingress controller 请跟踪L7和Ingress的proposal，了解有关资源演进的更多细节，以及Ingress repository，了解有关各种Ingress controller演进的更多详细信息。 替代方案 你可以通过很多种方式暴露service而不必直接使用ingress： 使用Service.Type=LoadBalancer 使用Service.Type=NodePort 使用Port Proxy 部署一个Service loadbalancer 这允许你在多个service之间共享单个IP，并通过Service Annotations实现更高级的负载平衡。 参考 Kubernetes Ingress Resource 使用NGINX Plus负载均衡Kubernetes服务 使用 NGINX 和 NGINX Plus 的 Ingress Controller 进行 Kubernetes 的负载均衡 Kubernetes : Ingress Controller with Træfɪk and Let's Encrypt Kubernetes : Træfɪk and Let's Encrypt at scale Kubernetes Ingress Controller-Træfɪk Kubernetes 1.2 and simplifying advanced networking with Ingress Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-08-21 18:23:34 "},"concepts/traefik-ingress-controller.html":{"url":"concepts/traefik-ingress-controller.html","title":"Traefik Ingress Controller","keywords":"","body":"Traefik Ingress Controller 我们在前面部署了 Traefik 作为Ingress Controller，如果集群外部直接访问Kubenretes内部服务的话，可以直接创建Ingress如下所示： apiVersion: extensions/v1beta1 kind: Ingress metadata: name: traefik-ingress namespace: default spec: rules: - host: traefik.nginx.io http: paths: - path: / backend: serviceName: my-nginx servicePort: 80 Traefik Ingress Controller 当我们处于迁移应用到kuberentes上的阶段时，可能有部分服务实例不在kubernetes上，服务的路由使用nginx配置，这时处于nginx和ingress共存的状态。参考下面的配置： apiVersion: extensions/v1beta1 kind: Ingress metadata: name: td-ingress namespace: default annotations: traefik.frontend.rule.type: PathPrefixStrip kubernetes.io/ingress.class: traefik spec: rules: - host: \"*.tendcloud.com\" http: paths: - path: /docGenerate backend: serviceName: td-sdmk-docgenerate servicePort: 80 注意annotation的配置： traefik.frontend.rule.type: PathPrefixStrip：表示将截掉URL中的path kubernetes.io/ingress.class：表示使用的ingress类型 关于Ingress annotation的更多信息请参考：Ingress Annotations - kubernetes.io。 在nginx中增加配置： upstream docGenerate { server 172.20.0.119:80; keepalive 200; } 172.20.0.119是我们的边缘节点的VIP，见边缘节点配置。 参考 Kubernetes Ingress Backend - traefik.io Kubernetes Ingress Controller - traefik.io Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-30 16:57:20 "},"concepts/configmap.html":{"url":"concepts/configmap.html","title":"ConfigMap","keywords":"","body":"ConfigMap 其实ConfigMap功能在Kubernetes1.2版本的时候就有了，许多应用程序会从配置文件、命令行参数或环境变量中读取配置信息。这些配置信息需要与docker image解耦，你总不能每修改一个配置就重做一个image吧？ConfigMap API给我们提供了向容器中注入配置信息的机制，ConfigMap可以被用来保存单个属性，也可以用来保存整个配置文件或者JSON二进制大对象。 ConfigMap概览 ConfigMap API资源用来保存key-value pair配置数据，这个数据可以在pods里使用，或者被用来为像controller一样的系统组件存储配置数据。虽然ConfigMap跟Secrets类似，但是ConfigMap更方便的处理不含敏感信息的字符串。 注意：ConfigMaps不是属性配置文件的替代品。ConfigMaps只是作为多个properties文件的引用。你可以把它理解为Linux系统中的/etc目录，专门用来存储配置文件的目录。下面举个例子，使用ConfigMap配置来创建Kuberntes Volumes，ConfigMap中的每个data项都会成为一个新文件。 kind: ConfigMap apiVersion: v1 metadata: creationTimestamp: 2016-02-18T19:14:38Z name: example-config namespace: default data: example.property.1: hello example.property.2: world example.property.file: |- property.1=value-1 property.2=value-2 property.3=value-3 data一栏包括了配置数据，ConfigMap可以被用来保存单个属性，也可以用来保存一个配置文件。 配置数据可以通过很多种方式在Pods里被使用。ConfigMaps可以被用来： 设置环境变量的值 在容器里设置命令行参数 在数据卷里面创建config文件 用户和系统组件两者都可以在ConfigMap里面存储配置数据。 其实不用看下面的文章，直接从kubectl create configmap -h的帮助信息中就可以对ConfigMap究竟如何创建略知一二了。 Examples: # Create a new configmap named my-config based on folder bar kubectl create configmap my-config --from-file=path/to/bar # Create a new configmap named my-config with specified keys instead of file basenames on disk kubectl create configmap my-config --from-file=key1=/path/to/bar/file1.txt --from-file=key2=/path/to/bar/file2.txt # Create a new configmap named my-config with key1=config1 and key2=config2 kubectl create configmap my-config --from-literal=key1=config1 --from-literal=key2=config2 创建ConfigMaps 可以使用该命令，用给定值、文件或目录来创建ConfigMap。 kubectl create configmap 使用目录创建 比如我们已经有个了包含一些配置文件，其中包含了我们想要设置的ConfigMap的值： $ ls docs/user-guide/configmap/kubectl/ game.properties ui.properties $ cat docs/user-guide/configmap/kubectl/game.properties enemies=aliens lives=3 enemies.cheat=true enemies.cheat.level=noGoodRotten secret.code.passphrase=UUDDLRLRBABAS secret.code.allowed=true secret.code.lives=30 $ cat docs/user-guide/configmap/kubectl/ui.properties color.good=purple color.bad=yellow allow.textmode=true how.nice.to.look=fairlyNice 使用下面的命令可以创建一个包含目录中所有文件的ConfigMap。 $ kubectl create configmap game-config --from-file=docs/user-guide/configmap/kubectl —from-file指定在目录下的所有文件都会被用在ConfigMap里面创建一个键值对，键的名字就是文件名，值就是文件的内容。 让我们来看一下这个命令创建的ConfigMap： $ kubectl describe configmaps game-config Name: game-config Namespace: default Labels: Annotations: Data ==== game.properties: 158 bytes ui.properties: 83 bytes 我们可以看到那两个key是从kubectl指定的目录中的文件名。这些key的内容可能会很大，所以在kubectl describe的输出中，只能够看到键的名字和他们的大小。 如果想要看到键的值的话，可以使用kubectl get： $ kubectl get configmaps game-config -o yaml 我们以yaml格式输出配置。 apiVersion: v1 data: game.properties: | enemies=aliens lives=3 enemies.cheat=true enemies.cheat.level=noGoodRotten secret.code.passphrase=UUDDLRLRBABAS secret.code.allowed=true secret.code.lives=30 ui.properties: | color.good=purple color.bad=yellow allow.textmode=true how.nice.to.look=fairlyNice kind: ConfigMap metadata: creationTimestamp: 2016-02-18T18:34:05Z name: game-config namespace: default resourceVersion: \"407\" selfLink: /api/v1/namespaces/default/configmaps/game-config uid: 30944725-d66e-11e5-8cd0-68f728db1985 使用文件创建 刚才使用目录创建的时候我们—from-file指定的是一个目录，只要指定为一个文件就可以从单个文件中创建ConfigMap。 $ kubectl create configmap game-config-2 --from-file=docs/user-guide/configmap/kubectl/game.properties $ kubectl get configmaps game-config-2 -o yaml apiVersion: v1 data: game-special-key: | enemies=aliens lives=3 enemies.cheat=true enemies.cheat.level=noGoodRotten secret.code.passphrase=UUDDLRLRBABAS secret.code.allowed=true secret.code.lives=30 kind: ConfigMap metadata: creationTimestamp: 2016-02-18T18:54:22Z name: game-config-3 namespace: default resourceVersion: \"530\" selfLink: /api/v1/namespaces/default/configmaps/game-config-3 uid: 05f8da22-d671-11e5-8cd0-68f728db1985 —from-file这个参数可以使用多次，你可以使用两次分别指定上个实例中的那两个配置文件，效果就跟指定整个目录是一样的。 使用字面值创建 使用文字值创建，利用—from-literal参数传递配置信息，该参数可以使用多次，格式如下； $ kubectl create configmap special-config --from-literal=special.how=very --from-literal=special.type=charm $ kubectl get configmaps special-config -o yaml apiVersion: v1 data: special.how: very special.type: charm kind: ConfigMap metadata: creationTimestamp: 2016-02-18T19:14:38Z name: special-config namespace: default resourceVersion: \"651\" selfLink: /api/v1/namespaces/default/configmaps/special-config uid: dadce046-d673-11e5-8cd0-68f728db1985 Pod中使用ConfigMap 使用ConfigMap来替代环境变量 ConfigMap可以被用来填入环境变量。看下下面的ConfigMap。 apiVersion: v1 kind: ConfigMap metadata: name: special-config namespace: default data: special.how: very special.type: charm apiVersion: v1 kind: ConfigMap metadata: name: env-config namespace: default data: log_level: INFO 我们可以在Pod中这样使用ConfigMap： apiVersion: v1 kind: Pod metadata: name: dapi-test-pod spec: containers: - name: test-container image: gcr.io/google_containers/busybox command: [ \"/bin/sh\", \"-c\", \"env\" ] env: - name: SPECIAL_LEVEL_KEY valueFrom: configMapKeyRef: name: special-config key: special.how - name: SPECIAL_TYPE_KEY valueFrom: configMapKeyRef: name: special-config key: special.type envFrom: - configMapRef: name: env-config restartPolicy: Never 这个Pod运行后会输出如下几行： SPECIAL_LEVEL_KEY=very SPECIAL_TYPE_KEY=charm log_level=INFO 用ConfigMap设置命令行参数 ConfigMap也可以被使用来设置容器中的命令或者参数值。它使用的是Kubernetes的$(VAR_NAME)替换语法。我们看下下面这个ConfigMap。 apiVersion: v1 kind: ConfigMap metadata: name: special-config namespace: default data: special.how: very special.type: charm 为了将ConfigMap中的值注入到命令行的参数里面，我们还要像前面那个例子一样使用环境变量替换语法${VAR_NAME)。（其实这个东西就是给Docker容器设置环境变量，以前我创建镜像的时候经常这么玩，通过docker run的时候指定-e参数修改镜像里的环境变量，然后docker的CMD命令再利用该$(VAR_NAME)通过sed来修改配置文件或者作为命令行启动参数。） apiVersion: v1 kind: Pod metadata: name: dapi-test-pod spec: containers: - name: test-container image: gcr.io/google_containers/busybox command: [ \"/bin/sh\", \"-c\", \"echo $(SPECIAL_LEVEL_KEY) $(SPECIAL_TYPE_KEY)\" ] env: - name: SPECIAL_LEVEL_KEY valueFrom: configMapKeyRef: name: special-config key: special.how - name: SPECIAL_TYPE_KEY valueFrom: configMapKeyRef: name: special-config key: special.type restartPolicy: Never 运行这个Pod后会输出： very charm 通过数据卷插件使用ConfigMap ConfigMap也可以在数据卷里面被使用。还是这个ConfigMap。 apiVersion: v1 kind: ConfigMap metadata: name: special-config namespace: default data: special.how: very special.type: charm 在数据卷里面使用这个ConfigMap，有不同的选项。最基本的就是将文件填入数据卷，在这个文件中，键就是文件名，键值就是文件内容： apiVersion: v1 kind: Pod metadata: name: dapi-test-pod spec: containers: - name: test-container image: gcr.io/google_containers/busybox command: [ \"/bin/sh\", \"-c\", \"cat /etc/config/special.how\" ] volumeMounts: - name: config-volume mountPath: /etc/config volumes: - name: config-volume configMap: name: special-config restartPolicy: Never 运行这个Pod的输出是very。 我们也可以在ConfigMap值被映射的数据卷里控制路径。 apiVersion: v1 kind: Pod metadata: name: dapi-test-pod spec: containers: - name: test-container image: gcr.io/google_containers/busybox command: [ \"/bin/sh\",\"-c\",\"cat /etc/config/path/to/special-key\" ] volumeMounts: - name: config-volume mountPath: /etc/config volumes: - name: config-volume configMap: name: special-config items: - key: special.how path: path/to/special-key restartPolicy: Never 运行这个Pod后的结果是very。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-08 16:03:37 "},"concepts/configmap-hot-update.html":{"url":"concepts/configmap-hot-update.html","title":"ConfigMap的热更新","keywords":"","body":"ConfigMap的热更新 ConfigMap是用来存储配置文件的kubernetes资源对象，所有的配置内容都存储在etcd中，下文主要是探究 ConfigMap 的创建和更新流程，以及对 ConfigMap 更新后容器内挂载的内容是否同步更新的测试。 测试示例 假设我们在 default namespace 下有一个名为 nginx-config 的 ConfigMap，可以使用 kubectl命令来获取： $ kubectl get configmap nginx-config NAME DATA AGE nginx-config 1 99d 获取该ConfigMap的内容。 kubectl get configmap nginx-config -o yaml apiVersion: v1 data: nginx.conf: |- worker_processes 1; events { worker_connections 1024; } http { sendfile on; server { listen 80; # a test endpoint that returns http 200s location / { proxy_pass http://httpstat.us/200; proxy_set_header X-Real-IP $remote_addr; } } server { listen 80; server_name api.hello.world; location / { proxy_pass http://l5d.default.svc.cluster.local; proxy_set_header Host $host; proxy_set_header Connection \"\"; proxy_http_version 1.1; more_clear_input_headers 'l5d-ctx-*' 'l5d-dtab' 'l5d-sample'; } } server { listen 80; server_name www.hello.world; location / { # allow 'employees' to perform dtab overrides if ($cookie_special_employee_cookie != \"letmein\") { more_clear_input_headers 'l5d-ctx-*' 'l5d-dtab' 'l5d-sample'; } # add a dtab override to get people to our beta, world-v2 set $xheader \"\"; if ($cookie_special_employee_cookie ~* \"dogfood\") { set $xheader \"/host/world => /srv/world-v2;\"; } proxy_set_header 'l5d-dtab' $xheader; proxy_pass http://l5d.default.svc.cluster.local; proxy_set_header Host $host; proxy_set_header Connection \"\"; proxy_http_version 1.1; } } } kind: ConfigMap metadata: creationTimestamp: 2017-08-01T06:53:17Z name: nginx-config namespace: default resourceVersion: \"14925806\" selfLink: /api/v1/namespaces/default/configmaps/nginx-config uid: 18d70527-7686-11e7-bfbd-8af1e3a7c5bd ConfigMap中的内容是存储到etcd中的，然后查询etcd： ETCDCTL_API=3 etcdctl get /registry/configmaps/default/nginx-config -w json|python -m json.tool 注意使用 v3 版本的 etcdctl API，下面是输出结果： { \"count\": 1, \"header\": { \"cluster_id\": 12091028579527406772, \"member_id\": 16557816780141026208, \"raft_term\": 36, \"revision\": 29258723 }, \"kvs\": [ { \"create_revision\": 14925806, \"key\": \"L3JlZ2lzdHJ5L2NvbmZpZ21hcHMvZGVmYXVsdC9uZ2lueC1jb25maWc=\", \"mod_revision\": 14925806, \"value\": \"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\", \"version\": 1 } ] } 其中的value就是 nginx.conf 配置文件的内容。 可以使用base64解码查看具体值，关于etcdctl的使用请参考使用etcdctl访问kuberentes数据。 代码 ConfigMap 结构体的定义： // ConfigMap holds configuration data for pods to consume. type ConfigMap struct { metav1.TypeMeta `json:\",inline\"` // Standard object's metadata. // More info: http://releases.k8s.io/HEAD/docs/devel/api-conventions.md#metadata // +optional metav1.ObjectMeta `json:\"metadata,omitempty\" protobuf:\"bytes,1,opt,name=metadata\"` // Data contains the configuration data. // Each key must be a valid DNS_SUBDOMAIN with an optional leading dot. // +optional Data map[string]string `json:\"data,omitempty\" protobuf:\"bytes,2,rep,name=data\"` } 在 staging/src/k8s.io/client-go/kubernetes/typed/core/v1/configmap.go 中ConfigMap 的接口定义： // ConfigMapInterface has methods to work with ConfigMap resources. type ConfigMapInterface interface { Create(*v1.ConfigMap) (*v1.ConfigMap, error) Update(*v1.ConfigMap) (*v1.ConfigMap, error) Delete(name string, options *meta_v1.DeleteOptions) error DeleteCollection(options *meta_v1.DeleteOptions, listOptions meta_v1.ListOptions) error Get(name string, options meta_v1.GetOptions) (*v1.ConfigMap, error) List(opts meta_v1.ListOptions) (*v1.ConfigMapList, error) Watch(opts meta_v1.ListOptions) (watch.Interface, error) Patch(name string, pt types.PatchType, data []byte, subresources ...string) (result *v1.ConfigMap, err error) ConfigMapExpansion } 在 staging/src/k8s.io/client-go/kubernetes/typed/core/v1/configmap.go 中创建 ConfigMap 的方法如下: // Create takes the representation of a configMap and creates it. Returns the server's representation of the configMap, and an error, if there is any. func (c *configMaps) Create(configMap *v1.ConfigMap) (result *v1.ConfigMap, err error) { result = &v1.ConfigMap{} err = c.client.Post(). Namespace(c.ns). Resource(\"configmaps\"). Body(configMap). Do(). Into(result) return } 通过 RESTful 请求在 etcd 中存储 ConfigMap 的配置，该方法中设置了资源对象的 namespace 和 HTTP 请求中的 body，执行后将请求结果保存到 result 中返回给调用者。 注意 Body 的结构 // Body makes the request use obj as the body. Optional. // If obj is a string, try to read a file of that name. // If obj is a []byte, send it directly. // If obj is an io.Reader, use it directly. // If obj is a runtime.Object, marshal it correctly, and set Content-Type header. // If obj is a runtime.Object and nil, do nothing. // Otherwise, set an error. 创建 ConfigMap RESTful 请求中的的 Body 中包含 ObjectMeta 和 namespace。 HTTP 请求中的结构体： // Request allows for building up a request to a server in a chained fashion. // Any errors are stored until the end of your call, so you only have to // check once. type Request struct { // required client HTTPClient verb string baseURL *url.URL content ContentConfig serializers Serializers // generic components accessible via method setters pathPrefix string subpath string params url.Values headers http.Header // structural elements of the request that are part of the Kubernetes API conventions namespace string namespaceSet bool resource string resourceName string subresource string timeout time.Duration // output err error body io.Reader // This is only used for per-request timeouts, deadlines, and cancellations. ctx context.Context backoffMgr BackoffManager throttle flowcontrol.RateLimiter } 测试 分别测试使用 ConfigMap 挂载 Env 和 Volume 的情况。 更新使用ConfigMap挂载的Env 使用下面的配置创建 nginx 容器测试更新 ConfigMap 后容器内的环境变量是否也跟着更新。 apiVersion: extensions/v1beta1 kind: Deployment metadata: name: my-nginx spec: replicas: 1 template: metadata: labels: run: my-nginx spec: containers: - name: my-nginx image: sz-pg-oam-docker-hub-001.tendcloud.com/library/nginx:1.9 ports: - containerPort: 80 envFrom: - configMapRef: name: env-config --- apiVersion: v1 kind: ConfigMap metadata: name: env-config namespace: default data: log_level: INFO 获取环境变量的值 $ kubectl exec `kubectl get pods -l run=my-nginx -o=name|cut -d \"/\" -f2` env|grep log_level log_level=INFO 修改 ConfigMap $ kubectl edit configmap env-config 修改 log_level 的值为 DEBUG。 再次查看环境变量的值。 $ kubectl exec `kubectl get pods -l run=my-nginx -o=name|cut -d \"/\" -f2` env|grep log_level log_level=INFO 实践证明修改 ConfigMap 无法更新容器中已注入的环境变量信息。 更新使用ConfigMap挂载的Volume 使用下面的配置创建 nginx 容器测试更新 ConfigMap 后容器内挂载的文件是否也跟着更新。 apiVersion: extensions/v1beta1 kind: Deployment metadata: name: my-nginx spec: replicas: 1 template: metadata: labels: run: my-nginx spec: containers: - name: my-nginx image: sz-pg-oam-docker-hub-001.tendcloud.com/library/nginx:1.9 ports: - containerPort: 80 volumeMounts: - name: config-volume mountPath: /etc/config volumes: - name: config-volume configMap: name: special-config --- apiVersion: v1 kind: ConfigMap metadata: name: special-config namespace: default data: log_level: INFO $ kubectl exec `kubectl get pods -l run=my-nginx -o=name|cut -d \"/\" -f2` cat /tmp/log_level INFO 修改 ConfigMap $ kubectl edit configmap special-config 修改 log_level 的值为 DEBUG。 等待大概10秒钟时间，再次查看环境变量的值。 $ kubectl exec `kubectl get pods -l run=my-nginx -o=name|cut -d \"/\" -f2` cat /tmp/log_level DEBUG 我们可以看到使用 ConfigMap 方式挂载的 Volume 的文件中的内容已经变成了 DEBUG。 总结 更新 ConfigMap 后： 使用该 ConfigMap 挂载的 Env 不会同步更新 使用该 ConfigMap 挂载的 Volume 中的数据需要一段时间（实测大概10秒）才能同步更新 ENV 是在容器启动的时候注入的，启动之后 kubernetes 就不会再改变环境变量的值，且同一个 namespace 中的 pod 的环境变量是不断累加的，参考 Kubernetes中的服务发现与docker容器间的环境变量传递源码探究。为了更新容器中使用 ConfigMap 挂载的配置，可以通过滚动更新 pod 的方式来强制重新挂载 ConfigMap，也可以在更新了 ConfigMap 后，先将副本数设置为 0，然后再扩容。 参考 Kubernetes 1.7 security in practice ConfigMap | kubernetes handbook - jimmysong.io 创建高可用ectd集群 | Kubernetes handbook - jimmysong.io Kubernetes中的服务发现与docker容器间的环境变量传递源码探究 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-23 13:48:42 "},"concepts/horizontal-pod-autoscaling.html":{"url":"concepts/horizontal-pod-autoscaling.html","title":"Horizontal Pod Autoscaling","keywords":"","body":"Horizontal Pod Autoscaling 应用的资源使用率通常都有高峰和低谷的时候，如何削峰填谷，提高集群的整体资源利用率，让service中的Pod个数自动调整呢？这就有赖于Horizontal Pod Autoscaling了，顾名思义，使Pod水平自动缩放。这个Object（跟Pod、Deployment一样都是API resource）也是最能体现kubernetes之于传统运维价值的地方，不再需要手动扩容了，终于实现自动化了，还可以自定义指标，没准未来还可以通过人工智能自动进化呢！ HPA属于Kubernetes中的autoscaling SIG（Special Interest Group），其下有两个feature： Arbitrary/Custom Metrics in the Horizontal Pod Autoscaler#117 Monitoring Pipeline Metrics HPA API #118 Kubernetes自1.2版本引入HPA机制，到1.6版本之前一直是通过kubelet来获取监控指标来判断是否需要扩缩容，1.6版本之后必须通过API server、Heapseter或者kube-aggregator来获取监控指标。 对于1.6以前版本中开启自定义HPA请参考Kubernetes autoscaling based on custom metrics without using a host port，对于1.7及以上版本请参考Configure Kubernetes Autoscaling With Custom Metrics in Kubernetes 1.7 - Bitnami。 HPA解析 Horizontal Pod Autoscaling仅适用于Deployment和ReplicationController（ReplicaSet已经被ReplicationController取代），在V1版本中仅支持根据Pod的CPU利用率扩所容，在v1alpha版本中，支持根据内存和用户自定义的metric扩缩容。 如果你不想看下面的文章可以直接看下面的示例图，组件交互、组件的配置、命令示例，都画在图上了。 Horizontal Pod Autoscaling由API server和controller共同实现。 图片 - horizontal-pod-autoscaler Metrics支持 在不同版本的API中，HPA autoscale时可以根据以下指标来判断： autoscaling/v1 CPU autoscaling/v2alpha1 内存 自定义metrics kubernetes1.6起支持自定义metrics，但是必须在kube-controller-manager中配置如下两项： --horizontal-pod-autoscaler-use-rest-clients=true --api-server指向kube-aggregator，也可以使用heapster来实现，通过在启动heapster的时候指定--api-server=true。查看kubernetes metrics 多种metrics组合 HPA会根据每个metric的值计算出scale的值，并将最大的那个指作为扩容的最终结果。 使用kubectl管理 Horizontal Pod Autoscaling作为API resource也可以像Pod、Deployment一样使用kubeclt命令管理，使用方法跟它们一样，资源名称为hpa。 kubectl create hpa kubectl get hpa kubectl describe hpa kubectl delete hpa 有一点不同的是，可以直接使用kubectl autoscale直接通过命令行的方式创建Horizontal Pod Autoscaler。 用法如下： kubectl autoscale (-f FILENAME | TYPE NAME | TYPE/NAME) [--min=MINPODS] --max=MAXPODS [--cpu-percent=CPU] [flags] [options] 举个例子： kubectl autoscale deployment foo --min=2 --max=5 --cpu-percent=80 为Deployment foo创建 一个autoscaler，当Pod的CPU利用率达到80%的时候，RC的replica数在2到5之间。该命令的详细使用文档见https://kubernetes.io/docs/user-guide/kubectl/v1.6/#autoscale 。 注意 ：如果为ReplicationController创建HPA的话，无法使用rolling update，但是对于Deployment来说是可以的，因为Deployment在执行rolling update的时候会自动创建新的ReplicationController。 什么是 Horizontal Pod Autoscaling？ 利用 Horizontal Pod Autoscaling，kubernetes 能够根据监测到的 CPU 利用率（或者在 alpha 版本中支持的应用提供的 metric）自动的扩容 replication controller，deployment 和 replica set。 Horizontal Pod Autoscaler 作为 kubernetes API resource 和 controller 的实现。Resource 确定 controller 的行为。Controller 会根据监测到用户指定的目标的 CPU 利用率周期性得调整 replication controller 或 deployment 的 replica 数量。 Horizontal Pod Autoscaler 如何工作？ Horizontal Pod Autoscaler 由一个控制循环实现，循环周期由 controller manager 中的 --horizontal-pod-autoscaler-sync-period 标志指定（默认是 30 秒）。 在每个周期内，controller manager 会查询 HorizontalPodAutoscaler 中定义的 metric 的资源利用率。Controller manager 从 resource metric API（每个 pod 的 resource metric）或者自定义 metric API（所有的metric）中获取 metric。 每个 Pod 的 resource metric（例如 CPU），controller 通过 resource metric API 获取 HorizontalPodAutoscaler 中定义的每个 Pod 中的 metric。然后，如果设置了目标利用率，controller 计算利用的值与每个 Pod 的容器里的 resource request 值的百分比。如果设置了目标原始值，将直接使用该原始 metric 值。然后 controller 计算所有目标 Pod 的利用率或原始值（取决于所指定的目标类型）的平均值，产生一个用于缩放所需 replica 数量的比率。 请注意，如果某些 Pod 的容器没有设置相关的 resource request ，则不会定义 Pod 的 CPU 利用率，并且 Aucoscaler 也不会对该 metric 采取任何操作。 有关自动缩放算法如何工作的更多细节，请参阅 自动缩放算法设计文档。 对于每个 Pod 自定义的 metric，controller 功能类似于每个 Pod 的 resource metric，只是它使用原始值而不是利用率值。 对于 object metric，获取单个度量（描述有问题的对象），并与目标值进行比较，以产生如上所述的比率。 HorizontalPodAutoscaler 控制器可以以两种不同的方式获取 metric ：直接的 Heapster 访问和 REST 客户端访问。 当使用直接的 Heapster 访问时，HorizontalPodAutoscaler 直接通过 API 服务器的服务代理子资源查询 Heapster。需要在集群上部署 Heapster 并在 kube-system namespace 中运行。 有关 REST 客户端访问的详细信息，请参阅 支持自定义度量。 Autoscaler 访问相应的 replication controller，deployment 或 replica set 来缩放子资源。 Scale 是一个允许您动态设置副本数并检查其当前状态的接口。 有关缩放子资源的更多细节可以在 这里 找到。 API Object Horizontal Pod Autoscaler 是 kubernetes 的 autoscaling API 组中的 API 资源。当前的稳定版本中，只支持 CPU 自动扩缩容，可以在autoscaling/v1 API 版本中找到。 在 alpha 版本中支持根据内存和自定义 metric 扩缩容，可以在autoscaling/v2alpha1 中找到。autoscaling/v2alpha1 中引入的新字段在autoscaling/v1 中是做为 annotation 而保存的。 关于该 API 对象的更多信息，请参阅 HorizontalPodAutoscaler Object。 在 kubectl 中支持 Horizontal Pod Autoscaling Horizontal Pod Autoscaler 和其他的所有 API 资源一样，通过 kubectl 以标准的方式支持。 我们可以使用kubectl create命令创建一个新的 autoscaler。 我们可以使用kubectl get hpa列出所有的 autoscaler，使用kubectl describe hpa获取其详细信息。 最后我们可以使用kubectl delete hpa删除 autoscaler。 另外，可以使用kubectl autoscale命令，很轻易的就可以创建一个 Horizontal Pod Autoscaler。 例如，执行kubectl autoscale rc foo —min=2 —max=5 —cpu-percent=80命令将为 replication controller foo 创建一个 autoscaler，目标的 CPU 利用率是80%，replica 的数量介于 2 和 5 之间。 关于kubectl autoscale的更多信息请参阅 这里。 滚动更新期间的自动扩缩容 目前在Kubernetes中，可以通过直接管理 replication controller 或使用 deployment 对象来执行 滚动更新，该 deployment 对象为您管理基础 replication controller。 Horizontal Pod Autoscaler 仅支持后一种方法：Horizontal Pod Autoscaler 被绑定到 deployment 对象，它设置 deployment 对象的大小，deployment 负责设置底层 replication controller 的大小。 Horizontal Pod Autoscaler 不能使用直接操作 replication controller 进行滚动更新，即不能将 Horizontal Pod Autoscaler 绑定到 replication controller，并进行滚动更新（例如使用kubectl rolling-update）。 这不行的原因是，当滚动更新创建一个新的 replication controller 时，Horizontal Pod Autoscaler 将不会绑定到新的 replication controller 上。 支持多个 metric Kubernetes 1.6 中增加了支持基于多个 metric 的扩缩容。您可以使用autoscaling/v2alpha1 API 版本来为 Horizontal Pod Autoscaler 指定多个 metric。然后 Horizontal Pod Autoscaler controller 将权衡每一个 metric，并根据该 metric 提议一个新的 scale。在所有提议里最大的那个 scale 将作为最终的 scale。 支持自定义 metric 注意： Kubernetes 1.2 根据特定于应用程序的 metric ，通过使用特殊注释的方式，增加了对缩放的 alpha 支持。 在 Kubernetes 1.6中删除了对这些注释的支持，有利于autoscaling/v2alpha1 API。 虽然旧的收集自定义 metric 的旧方法仍然可用，但是这些 metric 将不可供 Horizontal Pod Autoscaler 使用，并且用于指定要缩放的自定义 metric 的以前的注释也不在受 Horizontal Pod Autoscaler 认可。 Kubernetes 1.6增加了在 Horizontal Pod Autoscale r中使用自定义 metric 的支持。 您可以为autoscaling/v2alpha1 API 中使用的 Horizontal Pod Autoscaler 添加自定义 metric 。 Kubernetes 然后查询新的自定义 metric API 来获取相应自定义 metric 的值。 前提条件 为了在 Horizontal Pod Autoscaler 中使用自定义 metric，您必须在您集群的 controller manager 中将 --horizontal-pod-autoscaler-use-rest-clients 标志设置为 true。然后，您必须通过将 controller manager 的目标 API server 设置为 API server aggregator（使用--apiserver标志），配置您的 controller manager 通过 API server aggregator 与API server 通信。 Resource metric API和自定义 metric API 也必须向 API server aggregator 注册，并且必须由集群上运行的 API server 提供。 您可以使用 Heapster 实现 resource metric API，方法是将 --api-server 标志设置为 true 并运行 Heapster。 单独的组件必须提供自定义 metric API（有关自定义metric API的更多信息，可从 k8s.io/metrics repository 获得）。 进一步阅读 设计文档：Horizontal Pod Autoscaling kubectl autoscale 命令：kubectl autoscale 使用例子： Horizontal Pod Autoscaler 参考 HPA设计文档：https://github.com/kubernetes/community/blob/master/contributors/design-proposals/horizontal-pod-autoscaler.md HPA说明：https://kubernetes.io/docs/tasks/run-application/horizontal-pod-autoscale/ HPA详解：https://kubernetes.io/docs/tasks/run-application/horizontal-pod-autoscale-walkthrough/ kubectl autoscale命令详细使用说明：https://kubernetes.io/docs/user-guide/kubectl/v1.6/#autoscale 自定义metrics开发：https://github.com/kubernetes/metrics 1.7版本的kubernetes中启用自定义HPA：Configure Kubernetes Autoscaling With Custom Metrics in Kubernetes 1.7 - Bitnami Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-27 13:12:02 "},"concepts/custom-metrics-hpa.html":{"url":"concepts/custom-metrics-hpa.html","title":"自定义指标HPA","keywords":"","body":"自定义指标HPA Kubernetes中不仅支持CPU、内存为指标的HPA，还支持自定义指标的HPA，例如QPS。 本文中使用的yaml文件见manifests/HPA。 设置自定义指标 kubernetes1.6 在kubernetes1.6集群中配置自定义指标的HPA的说明已废弃。 在设置定义指标HPA之前需要先进行如下配置： 将heapster的启动参数 --api-server 设置为 true 启用custom metric API 将kube-controller-manager的启动参数中--horizontal-pod-autoscaler-use-rest-clients设置为true，并指定--master为API server地址，如--master=http://172.20.0.113:8080 在kubernetes1.5以前很容易设置，参考1.6以前版本的kubernetes中开启自定义HPA，而在1.6中因为取消了原来的annotation方式设置custom metric，只能通过API server和kube-aggregator来获取custom metric，因为只有两种方式来设置了，一是直接通过API server获取heapster的metrics，二是部署kube-aggragator来实现。 我们将在kubernetes1.8版本的kubernetes中，使用聚合的API server来实现自定义指标的HPA。 kuberentes1.7+ 确认您的kubernetes版本在1.7或以上，修改以下配置： 将kube-controller-manager的启动参数中--horizontal-pod-autoscaler-use-rest-clients设置为true，并指定--master为API server地址，如`--master=http://172.20.0.113:8080 修改kube-apiserver的配置文件apiserver，增加一条配置--requestheader-client-ca-file=/etc/kubernetes/ssl/ca.pem --requestheader-allowed-names=aggregator --requestheader-extra-headers-prefix=X-Remote-Extra- --requestheader-group-headers=X-Remote-Group --requestheader-username-headers=X-Remote-User --proxy-client-cert-file=/etc/kubernetes/ssl/kubernetes.pem --proxy-client-key-file=/etc/kubernetes/ssl/kubernetes-key.pem，用来配置aggregator的CA证书。 已经内置了apiregistration.k8s.io/v1beta1 API，可以直接定义APIService，如： apiVersion: apiregistration.k8s.io/v1 kind: APIService metadata: name: v1.custom-metrics.metrics.k8s.io spec: insecureSkipTLSVerify: true group: custom-metrics.metrics.k8s.io groupPriorityMinimum: 1000 versionPriority: 5 service: name: api namespace: custom-metrics version: v1alpha1 部署Prometheus 使用prometheus-operator.yaml文件部署Prometheus operator。 注意：将镜像修改为你自己的镜像仓库地址。 这产生一个自定义的API：http://172.20.0.113:8080/apis/custom-metrics.metrics.k8s.io/v1alpha1，可以通过浏览器访问，还可以使用下面的命令可以检查该API： $ kubectl get --raw=apis/custom-metrics.metrics.k8s.io/v1alpha1 {\"kind\":\"APIResourceList\",\"apiVersion\":\"v1\",\"groupVersion\":\"custom-metrics.metrics.k8s.io/v1alpha1\",\"resources\":[{\"name\":\"jobs.batch/http_requests\",\"singularName\":\"\",\"namespaced\":true,\"kind\":\"MetricValueList\",\"verbs\":[\"get\"]},{\"name\":\"namespaces/http_requests\",\"singularName\":\"\",\"namespaced\":false,\"kind\":\"MetricValueList\",\"verbs\":[\"get\"]},{\"name\":\"jobs.batch/up\",\"singularName\":\"\",\"namespaced\":true,\"kind\":\"MetricValueList\",\"verbs\":[\"get\"]},{\"name\":\"pods/up\",\"singularName\":\"\",\"namespaced\":true,\"kind\":\"MetricValueList\",\"verbs\":[\"get\"]},{\"name\":\"services/scrape_samples_scraped\",\"singularName\":\"\",\"namespaced\":true,\"kind\":\"MetricValueList\",\"verbs\":[\"get\"]},{\"name\":\"namespaces/scrape_samples_scraped\",\"singularName\":\"\",\"namespaced\":false,\"kind\":\"MetricValueList\",\"verbs\":[\"get\"]},{\"name\":\"pods/scrape_duration_seconds\",\"singularName\":\"\",\"namespaced\":true,\"kind\":\"MetricValueList\",\"verbs\":[\"get\"]},{\"name\":\"services/scrape_duration_seconds\",\"singularName\":\"\",\"namespaced\":true,\"kind\":\"MetricValueList\",\"verbs\":[\"get\"]},{\"name\":\"pods/http_requests\",\"singularName\":\"\",\"namespaced\":true,\"kind\":\"MetricValueList\",\"verbs\":[\"get\"]},{\"name\":\"pods/scrape_samples_post_metric_relabeling\",\"singularName\":\"\",\"namespaced\":true,\"kind\":\"MetricValueList\",\"verbs\":[\"get\"]},{\"name\":\"jobs.batch/scrape_samples_scraped\",\"singularName\":\"\",\"namespaced\":true,\"kind\":\"MetricValueList\",\"verbs\":[\"get\"]},{\"name\":\"jobs.batch/scrape_duration_seconds\",\"singularName\":\"\",\"namespaced\":true,\"kind\":\"MetricValueList\",\"verbs\":[\"get\"]},{\"name\":\"namespaces/scrape_duration_seconds\",\"singularName\":\"\",\"namespaced\":false,\"kind\":\"MetricValueList\",\"verbs\":[\"get\"]},{\"name\":\"namespaces/scrape_samples_post_metric_relabeling\",\"singularName\":\"\",\"namespaced\":false,\"kind\":\"MetricValueList\",\"verbs\":[\"get\"]},{\"name\":\"services/scrape_samples_post_metric_relabeling\",\"singularName\":\"\",\"namespaced\":true,\"kind\":\"MetricValueList\",\"verbs\":[\"get\"]},{\"name\":\"services/up\",\"singularName\":\"\",\"namespaced\":true,\"kind\":\"MetricValueList\",\"verbs\":[\"get\"]},{\"name\":\"pods/scrape_samples_scraped\",\"singularName\":\"\",\"namespaced\":true,\"kind\":\"MetricValueList\",\"verbs\":[\"get\"]},{\"name\":\"services/http_requests\",\"singularName\":\"\",\"namespaced\":true,\"kind\":\"MetricValueList\",\"verbs\":[\"get\"]},{\"name\":\"jobs.batch/scrape_samples_post_metric_relabeling\",\"singularName\":\"\",\"namespaced\":true,\"kind\":\"MetricValueList\",\"verbs\":[\"get\"]},{\"name\":\"namespaces/up\",\"singularName\":\"\",\"namespaced\":false,\"kind\":\"MetricValueList\",\"verbs\":[\"get\"]}]} 参考 1.6以前版本的kubernetes中开启自定义HPA 1.7版本的kubernetes中启用自定义HPA Horizontal Pod Autoscaler Walkthrough Kubernetes 1.8: Now with 100% Daily Value of Custom Metrics Arbitrary/Custom Metrics in the Horizontal Pod Autoscaler#117 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-14 19:03:40 "},"concepts/label.html":{"url":"concepts/label.html","title":"Label","keywords":"","body":"Label Label是附着到object上（例如Pod）的键值对。可以在创建object的时候指定，也可以在object创建后随时指定。Labels的值对系统本身并没有什么含义，只是对用户才有意义。 \"labels\": { \"key1\" : \"value1\", \"key2\" : \"value2\" } Kubernetes最终将对labels最终索引和反向索引用来优化查询和watch，在UI和命令行中会对它们排序。不要在label中使用大型、非标识的结构化数据，记录这样的数据应该用annotation。 动机 Label能够将组织架构映射到系统架构上（就像是康威定律），这样能够更便于微服务的管理，你可以给object打上如下类型的label： \"release\" : \"stable\", \"release\" : \"canary\" \"environment\" : \"dev\", \"environment\" : \"qa\", \"environment\" : \"production\" \"tier\" : \"frontend\", \"tier\" : \"backend\", \"tier\" : \"cache\" \"partition\" : \"customerA\", \"partition\" : \"customerB\" \"track\" : \"daily\", \"track\" : \"weekly\" \"team\" : \"teamA\",\"team:\" : \"teamB\" 语法和字符集 Label key的组成： 不得超过63个字符 可以使用前缀，使用/分隔，前缀必须是DNS子域，不得超过253个字符，系统中的自动化组件创建的label必须指定前缀，kubernetes.io/由kubernetes保留 起始必须是字母（大小写都可以）或数字，中间可以有连字符、下划线和点 Label value的组成： 不得超过63个字符 起始必须是字母（大小写都可以）或数字，中间可以有连字符、下划线和点 Label selector Label不是唯一的，很多object可能有相同的label。 通过label selector，客户端／用户可以指定一个object集合，通过label selector对object的集合进行操作。 Label selector有两种类型： equality-based ：可以使用=、==、!=操作符，可以使用逗号分隔多个表达式 set-based ：可以使用in、notin、!操作符，另外还可以没有操作符，直接写出某个label的key，表示过滤有某个key的object而不管该key的value是何值，!表示没有该label的object 示例 $ kubectl get pods -l environment=production,tier=frontend $ kubectl get pods -l 'environment in (production),tier in (frontend)' $ kubectl get pods -l 'environment in (production, qa)' $ kubectl get pods -l 'environment,environment notin (frontend)' 在API object中设置label selector 在service、replicationcontroller等object中有对pod的label selector，使用方法只能使用等于操作，例如： selector: component: redis 在Job、Deployment、ReplicaSet和DaemonSet这些object中，支持set-based的过滤，例如： selector: matchLabels: component: redis matchExpressions: - {key: tier, operator: In, values: [cache]} - {key: environment, operator: NotIn, values: [dev]} 如Service通过label selector将同一类型的pod作为一个服务expose出来。 图片 - label示意图 另外在node affinity和pod affinity中的label selector的语法又有些许不同，示例如下： affinity: nodeAffinity: requiredDuringSchedulingIgnoredDuringExecution: nodeSelectorTerms: - matchExpressions: - key: kubernetes.io/e2e-az-name operator: In values: - e2e-az1 - e2e-az2 preferredDuringSchedulingIgnoredDuringExecution: - weight: 1 preference: matchExpressions: - key: another-node-label-key operator: In values: - another-node-label-value 详见node selection。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-08 16:19:53 "},"concepts/garbage-collection.html":{"url":"concepts/garbage-collection.html","title":"垃圾收集","keywords":"","body":"垃圾收集 Kubernetes 垃圾收集器的角色是删除指定的对象，这些对象曾经有但以后不再拥有 Owner 了。 注意：垃圾收集是 beta 特性，在 Kubernetes 1.4 及以上版本默认启用。 Owner 和 Dependent 一些 Kubernetes 对象是其它一些的 Owner。例如，一个 ReplicaSet 是一组 Pod 的 Owner。具有 Owner 的对象被称为是 Owner 的 Dependent。每个 Dependent 对象具有一个指向其所属对象的 metadata.ownerReferences 字段。 有时，Kubernetes 会自动设置 ownerReference 的值。例如，当创建一个 ReplicaSet 时，Kubernetes 自动设置 ReplicaSet 中每个 Pod 的 ownerReference 字段值。在 1.6 版本，Kubernetes 会自动为一些对象设置 ownerReference 的值，这些对象是由 ReplicationController、ReplicaSet、StatefulSet、DaemonSet 和 Deployment 所创建或管理。 也可以通过手动设置 ownerReference 的值，来指定 Owner 和 Dependent 之间的关系。 这有一个配置文件，表示一个具有 3 个 Pod 的 ReplicaSet： apiVersion: extensions/v1beta1 kind: ReplicaSet metadata: name: my-repset spec: replicas: 3 selector: matchLabels: pod-is-for: garbage-collection-example template: metadata: labels: pod-is-for: garbage-collection-example spec: containers: - name: nginx image: nginx 如果创建该 ReplicaSet，然后查看 Pod 的 metadata 字段，能够看到 OwnerReferences 字段： kubectl create -f https://k8s.io/docs/concepts/abstractions/controllers/my-repset.yaml kubectl get pods --output=yaml 输出显示了 Pod 的 Owner 是名为 my-repset 的 ReplicaSet： apiVersion: v1 kind: Pod metadata: ... ownerReferences: - apiVersion: extensions/v1beta1 controller: true blockOwnerDeletion: true kind: ReplicaSet name: my-repset uid: d9607e19-f88f-11e6-a518-42010a800195 ... 控制垃圾收集器删除 Dependent 当删除对象时，可以指定是否该对象的 Dependent 也自动删除掉。自动删除 Dependent 也称为 级联删除。Kubernetes 中有两种 级联删除 的模式：background 模式和 foreground 模式。 如果删除对象时，不自动删除它的 Dependent，这些 Dependent 被称作是原对象的 孤儿。 Background 级联删除 在 background 级联删除 模式下，Kubernetes 会立即删除 Owner 对象，然后垃圾收集器会在后台删除这些 Dependent。 Foreground 级联删除 在 foreground 级联删除 模式下，根对象首先进入 “删除中” 状态。在 “删除中” 状态会有如下的情况： 对象仍然可以通过 REST API 可见 会设置对象的 deletionTimestamp 字段 对象的 metadata.finalizers 字段包含了值 “foregroundDeletion” 一旦被设置为 “删除中” 状态，垃圾收集器会删除对象的所有 Dependent。垃圾收集器删除了所有 “Blocking” 的 Dependent（对象的 ownerReference.blockOwnerDeletion=true）之后，它会删除 Owner 对象。 注意，在 “foreground 删除” 模式下，Dependent 只有通过 ownerReference.blockOwnerDeletion 才能阻止删除 Owner 对象。在 Kubernetes 1.7 版本中将增加 admission controller，基于 Owner 对象上的删除权限来控制用户去设置 blockOwnerDeletion 的值为 true，所以未授权的 Dependent 不能够延迟 Owner 对象的删除。 如果一个对象的ownerReferences 字段被一个 Controller（例如 Deployment 或 ReplicaSet）设置，blockOwnerDeletion 会被自动设置，没必要手动修改这个字段。 设置级联删除策略 通过为 Owner 对象设置 deleteOptions.propagationPolicy 字段，可以控制级联删除策略。可能的取值包括：“orphan”、“Foreground” 或 “Background”。 对很多 Controller 资源，包括 ReplicationController、ReplicaSet、StatefulSet、DaemonSet 和 Deployment，默认的垃圾收集策略是 orphan。因此，除非指定其它的垃圾收集策略，否则所有 Dependent 对象使用的都是 orphan 策略。 下面是一个在后台删除 Dependent 对象的例子： kubectl proxy --port=8080 curl -X DELETE localhost:8080/apis/extensions/v1beta1/namespaces/default/replicasets/my-repset \\ -d '{\"kind\":\"DeleteOptions\",\"apiVersion\":\"v1\",\"propagationPolicy\":\"Background\"}' \\ -H \"Content-Type: application/json\" 下面是一个在前台删除 Dependent 对象的例子： kubectl proxy --port=8080 curl -X DELETE localhost:8080/apis/extensions/v1beta1/namespaces/default/replicasets/my-repset \\ -d '{\"kind\":\"DeleteOptions\",\"apiVersion\":\"v1\",\"propagationPolicy\":\"Foreground\"}' \\ -H \"Content-Type: application/json\" 下面是一个孤儿 Dependent 的例子： kubectl proxy --port=8080 curl -X DELETE localhost:8080/apis/extensions/v1beta1/namespaces/default/replicasets/my-repset \\ -d '{\"kind\":\"DeleteOptions\",\"apiVersion\":\"v1\",\"propagationPolicy\":\"Orphan\"}' \\ -H \"Content-Type: application/json\" kubectl 也支持级联删除。 通过设置 --cascade 为 true，可以使用 kubectl 自动删除 Dependent 对象。设置 --cascade 为 false，会使 Dependent 对象成为孤儿 Dependent 对象。--cascade 的默认值是 true。 下面是一个例子，使一个 ReplicaSet 的 Dependent 对象成为孤儿 Dependent： kubectl delete replicaset my-repset --cascade=false 已知的问题 1.7 版本，垃圾收集不支持 自定义资源，比如那些通过 CustomResourceDefinition 新增，或者通过 API server 聚集而成的资源对象。 其它已知的问题 原文地址：https://k8smeetup.github.io/docs/concepts/workloads/controllers/garbage-collection/ 译者：shirdrn Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-09-03 13:45:06 "},"concepts/network-policy.html":{"url":"concepts/network-policy.html","title":"NetworkPolicy","keywords":"","body":"Network Policy 网络策略说明一组 Pod 之间是如何被允许互相通信，以及如何与其它网络 Endpoint 进行通信。 NetworkPolicy 资源使用标签来选择 Pod，并定义了一些规则，这些规则指明允许什么流量进入到选中的 Pod 上。 前提条件 网络策略通过网络插件来实现，所以必须使用一种支持 NetworkPolicy 的网络方案 —— 非 Controller 创建的资源，是不起作用的。 隔离的与未隔离的 Pod 默认 Pod 是未隔离的，它们可以从任何的源接收请求。 具有一个可以选择 Pod 的网络策略后，Pod 就会变成隔离的。 一旦 Namespace 中配置的网络策略能够选择一个特定的 Pod，这个 Pod 将拒绝任何该网络策略不允许的连接。（Namespace 中其它未被网络策略选中的 Pod 将继续接收所有流量） NetworkPolicy 资源 查看 API参考 可以获取该资源的完整定义。 下面是一个 NetworkPolicy 的例子： apiVersion: networking.k8s.io/v1 kind: NetworkPolicy metadata: name: test-network-policy namespace: default spec: podSelector: matchLabels: role: db ingress: - from: - namespaceSelector: matchLabels: project: myproject - podSelector: matchLabels: role: frontend ports: - protocol: TCP port: 6379 将上面配置 POST 到 API Server 将不起任何作用，除非选择的网络方案支持网络策略。 必选字段：像所有其它 Kubernetes 配置一样， NetworkPolicy 需要 apiVersion、kind 和 metadata 这三个字段，关于如何使用配置文件的基本信息，可以查看 这里，这里 和 这里。 spec：NetworkPolicy spec 具有在给定 Namespace 中定义特定网络的全部信息。 podSelector：每个 NetworkPolicy 包含一个 podSelector，它可以选择一组应用了网络策略的 Pod。由于 NetworkPolicy 当前只支持定义 ingress 规则，这个 podSelector 实际上为该策略定义了一组 “目标Pod”。示例中的策略选择了标签为 “role=db” 的 Pod。一个空的 podSelector 选择了该 Namespace 中的所有 Pod。 ingress：每个NetworkPolicy 包含了一个白名单 ingress 规则列表。每个规则只允许能够匹配上 from 和 ports配置段的流量。示例策略包含了单个规则，它从这两个源中匹配在单个端口上的流量，第一个是通过namespaceSelector 指定的，第二个是通过 podSelector 指定的。 因此，上面示例的 NetworkPolicy： 在 “default” Namespace中 隔离了标签 “role=db” 的 Pod（如果他们还没有被隔离） 在 “default” Namespace中，允许任何具有 “role=frontend” 的 Pod，连接到标签为 “role=db” 的 Pod 的 TCP 端口 6379 允许在 Namespace 中任何具有标签 “project=myproject” 的 Pod，连接到 “default” Namespace 中标签为 “role=db” 的 Pod 的 TCP 端口 6379 查看 NetworkPolicy 入门指南 给出的更进一步的例子。 默认策略 通过创建一个可以选择所有 Pod 但不允许任何流量的 NetworkPolicy，你可以为一个 Namespace 创建一个 “默认的” 隔离策略，如下所示： apiVersion: networking.k8s.io/v1 kind: NetworkPolicy metadata: name: default-deny spec: podSelector: 这确保了即使是没有被任何 NetworkPolicy 选中的 Pod，将仍然是被隔离的。 可选地，在 Namespace 中，如果你想允许所有的流量进入到所有的 Pod（即使已经添加了某些策略，使一些 Pod 被处理为 “隔离的”），你可以通过创建一个策略来显式地指定允许所有流量： apiVersion: networking.k8s.io/v1 kind: NetworkPolicy metadata: name: allow-all spec: podSelector: ingress: - {} 原文地址：https://k8smeetup.github.io/docs/concepts/services-networking/network-policies/ 译者：shirdrn Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-09-03 14:17:05 "},"concepts/annotation.html":{"url":"concepts/annotation.html","title":"Annotation","keywords":"","body":"Annotation Annotation，顾名思义，就是注解。Annotation可以将Kubernetes资源对象关联到任意的非标识性元数据。使用客户端（如工具和库）可以检索到这些元数据。 关联元数据到对象 Label和Annotation都可以将元数据关联到Kubernetes资源对象。Label主要用于选择对象，可以挑选出满足特定条件的对象。相比之下，annotation 不能用于标识及选择对象。annotation中的元数据可多可少，可以是结构化的或非结构化的，也可以包含label中不允许出现的字符。 annotation和label一样都是key/value键值对映射结构： \"annotations\": { \"key1\" : \"value1\", \"key2\" : \"value2\" } 以下列出了一些可以记录在 annotation 中的对象信息： 声明配置层管理的字段。使用annotation关联这类字段可以用于区分以下几种配置来源：客户端或服务器设置的默认值，自动生成的字段或自动生成的 auto-scaling 和 auto-sizing 系统配置的字段。 创建信息、版本信息或镜像信息。例如时间戳、版本号、git分支、PR序号、镜像哈希值以及仓库地址。 记录日志、监控、分析或审计存储仓库的指针 可以用于debug的客户端（库或工具）信息，例如名称、版本和创建信息。 用户信息，以及工具或系统来源信息、例如来自非Kubernetes生态的相关对象的URL信息。 轻量级部署工具元数据，例如配置或检查点。 负责人的电话或联系方式，或能找到相关信息的目录条目信息，例如团队网站。 如果不使用annotation，您也可以将以上类型的信息存放在外部数据库或目录中，但这样做不利于创建用于部署、管理、内部检查的共享工具和客户端库。 示例 如 Istio 的 Deployment 配置中就使用到了 annotation： apiVersion: extensions/v1beta1 kind: Deployment metadata: name: istio-manager spec: replicas: 1 template: metadata: annotations: alpha.istio.io/sidecar: ignore labels: istio: manager spec: serviceAccountName: istio-manager-service-account containers: - name: discovery image: sz-pg-oam-docker-hub-001.tendcloud.com/library/manager:0.1.5 imagePullPolicy: Always args: [\"discovery\", \"-v\", \"2\"] ports: - containerPort: 8080 env: - name: POD_NAMESPACE valueFrom: fieldRef: apiVersion: v1 fieldPath: metadata.namespace - name: apiserver image: sz-pg-oam-docker-hub-001.tendcloud.com/library/manager:0.1.5 imagePullPolicy: Always args: [\"apiserver\", \"-v\", \"2\"] ports: - containerPort: 8081 env: - name: POD_NAMESPACE valueFrom: fieldRef: apiVersion: v1 fieldPath: metadata.namespace alpha.istio.io/sidecar 注解就是用来控制是否自动向 pod 中注入 sidecar 的。参考：安装 Istio sidecar - istio.doczh.cn。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-09 16:42:47 "},"concepts/aggregated-api-server.html":{"url":"concepts/aggregated-api-server.html","title":"Aggregated API Server","keywords":"","body":"Aggregated API Server Aggregated（聚合的）API server是为了将原来的API server这个巨石（monolithic）应用给拆分成，为了方便用户开发自己的API server集成进来，而不用直接修改kubernetes官方仓库的代码，这样一来也能将API server解耦，方便用户使用实验特性。这些API server可以跟core API server无缝衔接，使用kubectl也可以管理它们。 架构 我们需要创建一个新的组件，名为kube-aggregator，它需要负责以下几件事： 提供用于注册API server的API 汇总所有的API server信息 代理所有的客户端到API server的请求 注意：这里说的API server是一组“API Server”，而不是说我们安装集群时候的那个API server，而且这组API server是可以横向扩展的。 关于聚合的API server的更多信息请参考：Aggregated API Server 安装配置聚合的API server 有两种方式来启用kube-aggregator： 使用test mode/single-user mode，作为一个独立的进程来运行 使用gateway mode，kube-apiserver将嵌入到kbe-aggregator组件中，它将作为一个集群的gateway，用来聚合所有apiserver。 kube-aggregator二进制文件已经包含在kubernetes release里面了。 参考 Aggregated API Servers - kuberentes design-proposals Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-14 18:52:36 "},"concepts/custom-resource.html":{"url":"concepts/custom-resource.html","title":"使用自定义资源扩展API","keywords":"","body":"使用自定义资源扩展API 注意：TPR已经停止维护，kubernetes 1.7及以上版本请使用CRD。 自定义资源是对Kubernetes API的扩展，kubernetes中的每个资源都是一个API对象的集合，例如我们在YAML文件里定义的那些spec都是对kubernetes中的资源对象的定义，所有的自定义资源可以跟kubernetes中内建的资源一样使用kubectl操作。 自定义资源 Kubernetes1.6版本中包含一个内建的资源叫做TPR（ThirdPartyResource），可以用它来创建自定义资源，但该资源在kubernetes1.7中版本已被CRD（CustomResourceDefinition）取代。 扩展API 自定义资源实际上是为了扩展kubernetes的API，向kubenetes API中增加新类型，可以使用以下三种方式： 修改kubenetes的源码，显然难度比较高，也不太合适 创建自定义API server并聚合到API中 1.7以下版本编写TPR，kubernetes1.7及以上版本用CRD 编写自定义资源是扩展kubernetes API的最简单的方式，是否编写自定义资源来扩展API请参考Should I add a custom resource to my Kubernetes Cluster?，行动前请先考虑以下几点： 你的API是否属于声明式的 是否想使用kubectl命令来管理 是否要作为kubenretes中的对象类型来管理，同时显示在kuberetes dashboard上 是否可以遵守kubernetes的API规则限制，例如URL和API group、namespace限制 是否可以接受该API只能作用于集群或者namespace范围 想要复用kubernetes API的公共功能，比如CRUD、watch、内置的认证和授权等 如果这些都不是你想要的，那么你可以开发一个独立的API。 TPR 注意：TPR已经停止维护，kubernetes 1.7及以上版本请使用CRD。 假如我们要创建一个名为cron-tab.stable.example.com的TPR，yaml文件定义如下： apiVersion: extensions/v1beta1 kind: ThirdPartyResource metadata: name: cron-tab.stable.example.com description: \"A specification of a Pod to run on a cron style schedule\" versions: - name: v1 然后使用kubectl create命令创建该资源，这样就可以创建出一个API端点/apis/stable.example.com/v1/namespaces//crontabs/...。 下面是在Linkerd中的一个实际应用，Linkerd中的一个名为namerd的组件使用了TPR，定义如下： --- kind: ThirdPartyResource apiVersion: extensions/v1beta1 metadata: name: d-tab.l5d.io description: stores dtabs used by namerd versions: - name: v1alpha1 CRD 参考下面的CRD，resourcedefinition.yaml： apiVersion: apiextensions.k8s.io/v1beta1 kind: CustomResourceDefinition metadata: # 名称必须符合下面的格式：. name: crontabs.stable.example.com spec: # REST API使用的组名称：/apis// group: stable.example.com # REST API使用的版本号：/apis// version: v1 # Namespaced或Cluster scope: Namespaced names: # URL中使用的复数名称: /apis/// plural: crontabs # CLI中使用的单数名称 singular: crontab # CamelCased格式的单数类型。在清单文件中使用 kind: CronTab # CLI中使用的资源简称 shortNames: - ct 创建该CRD： kubectl create -f resourcedefinition.yaml 访问RESTful API端点如http://172.20.0.113:8080将看到如下API端点已创建： /apis/stable.example.com/v1/namespaces/*/crontabs/... 创建自定义对象 如下所示： apiVersion: \"stable.example.com/v1\" kind: CronTab metadata: name: my-new-cron-object spec: cronSpec: \"* * * * /5\" image: my-awesome-cron-image 引用该自定义资源的API创建对象。 终止器 可以为自定义对象添加一个终止器，如下所示： apiVersion: \"stable.example.com/v1\" kind: CronTab metadata: finalizers: - finalizer.stable.example.com 删除自定义对象前，异步执行的钩子。对于具有终止器的一个对象，删除请求仅仅是为metadata.deletionTimestamp字段设置一个值，而不是删除它，这将触发监控该对象的控制器执行他们所能处理的任意终止器。 详情参考：Extend the Kubernetes API with CustomResourceDefinitions 使用kubernetes1.7及以上版本请参考Migrate a ThirdPartyResource to CustomResourceDefinition。 自定义控制器 单纯设置了自定义资源，并没有什么用，只有跟自定义控制器结合起来，才能讲资源对象中的声明式API翻译成用户所期望的状态。自定义控制器可以用来管理任何资源类型，但是一般是跟自定义资源结合使用。 请参考使用Operator模式，该模式可以让开发者将自己的领域知识转换成特定的kubenretes API扩展。 API server聚合 Aggregated（聚合的）API server是为了将原来的API server这个巨石（monolithic）应用给拆分成，为了方便用户开发自己的API server集成进来，而不用直接修改kubernetes官方仓库的代码，这样一来也能将API server解耦，方便用户使用实验特性。这些API server可以跟core API server无缝衔接，试用kubectl也可以管理它们。 详情参考Aggregated API Server。 参考 Custom Resources Extend the Kubernetes API with CustomResourceDefinitions Introducing Operators: Putting Operational Knowledge into Software Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-14 18:39:25 "},"concepts/apiservice.html":{"url":"concepts/apiservice.html","title":"APIService","keywords":"","body":"APIService APIService是用来表示一个特定的GroupVersion的中的server，它的结构定义位于代码staging/src/k8s.io/kube-aggregator/pkg/apis/apiregistration/types.go中。 下面是一个APIService的示例配置： apiVersion: apiregistration.k8s.io/v1beta1 kind: APIService metadata: name: v1alpha1.custom-metrics.metrics.k8s.io spec: insecureSkipTLSVerify: true group: custom-metrics.metrics.k8s.io groupPriorityMinimum: 1000 versionPriority: 5 service: name: api namespace: custom-metrics version: v1alpha1 APIService详解 使用apiregistration.k8s.io/v1beta1 版本的APIService，在metadata.name中定义该API的名字。 使用上面的yaml的创建v1alpha1.custom-metrics.metrics.k8s.io APIService。 insecureSkipTLSVerify：当与该服务通信时，禁用TLS证书认证。强加建议不要设置这个参数，默认为 false。应该使用CABundle代替。 service：与该APIService通信时引用的service，其中要注明service的名字和所属的namespace，如果为空的话，则所有的服务都会该API groupversion将在本地443端口处理所有通信。 groupPriorityMinimum：该组API的处理优先级，主要排序是基于groupPriorityMinimum，该数字越大表明优先级越高，客户端就会与其通信处理请求。次要排序是基于字母表顺序，例如v1.bar比v1.foo的优先级更高。 versionPriority：VersionPriority控制其组内的API版本的顺序。必须大于零。主要排序基于VersionPriority，从最高到最低（20大于10）排序。次要排序是基于对象名称的字母比较。 （v1.foo在v1.bar之前）由于它们都是在一个组内，因此数字可能很小，一般都小于10。 查看我们使用上面的yaml文件创建的APIService。 kubectl get apiservice v1alpha1.custom-metrics.metrics.k8s.io -o yaml apiVersion: apiregistration.k8s.io/v1beta1 kind: APIService metadata: creationTimestamp: 2017-12-14T08:27:35Z name: v1alpha1.custom-metrics.metrics.k8s.io resourceVersion: \"35194598\" selfLink: /apis/apiregistration.k8s.io/v1beta1/apiservices/v1alpha1.custom-metrics.metrics.k8s.io uid: a31a3412-e0a8-11e7-9fa4-f4e9d49f8ed0 spec: caBundle: null group: custom-metrics.metrics.k8s.io groupPriorityMinimum: 1000 insecureSkipTLSVerify: true service: name: api namespace: custom-metrics version: v1alpha1 versionPriority: 5 status: conditions: - lastTransitionTime: 2017-12-14T08:27:38Z message: all checks passed reason: Passed status: \"True\" type: Available 查看集群支持的APISerivce 作为Kubernetes中的一种资源对象，可以使用kubectl get apiservice来查看。 例如查看集群中所有的APIService： $ kubectl get apiservice NAME AGE v1. 2d v1.authentication.k8s.io 2d v1.authorization.k8s.io 2d v1.autoscaling 2d v1.batch 2d v1.monitoring.coreos.com 1d v1.networking.k8s.io 2d v1.rbac.authorization.k8s.io 2d v1.storage.k8s.io 2d v1alpha1.custom-metrics.metrics.k8s.io 2h v1beta1.apiextensions.k8s.io 2d v1beta1.apps 2d v1beta1.authentication.k8s.io 2d v1beta1.authorization.k8s.io 2d v1beta1.batch 2d v1beta1.certificates.k8s.io 2d v1beta1.extensions 2d v1beta1.policy 2d v1beta1.rbac.authorization.k8s.io 2d v1beta1.storage.k8s.io 2d v1beta2.apps 2d v2beta1.autoscaling 2d 另外查看当前kubernetes集群支持的API版本还可以使用kubectl api-version： $ kubectl api-versions apiextensions.k8s.io/v1beta1 apiregistration.k8s.io/v1beta1 apps/v1beta1 apps/v1beta2 authentication.k8s.io/v1 authentication.k8s.io/v1beta1 authorization.k8s.io/v1 authorization.k8s.io/v1beta1 autoscaling/v1 autoscaling/v2beta1 batch/v1 batch/v1beta1 certificates.k8s.io/v1beta1 custom-metrics.metrics.k8s.io/v1alpha1 extensions/v1beta1 monitoring.coreos.com/v1 networking.k8s.io/v1 policy/v1beta1 rbac.authorization.k8s.io/v1 rbac.authorization.k8s.io/v1beta1 storage.k8s.io/v1 storage.k8s.io/v1beta1 v1 参考 API Conventions Kuberentes1.8 reference doc Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-14 19:32:56 "},"guide/":{"url":"guide/","title":"用户指南","keywords":"","body":"用户指南 该章节主要记录kubernetes使用过程中的一些配置技巧和操作。 配置Pod的liveness和readiness探针 管理集群中的TLS Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-15 19:40:21 "},"guide/resource-configuration.html":{"url":"guide/resource-configuration.html","title":"资源对象配置","keywords":"","body":"资源配置 Kubernetes 中的各个 Object 的配置指南。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-08-21 18:23:34 "},"guide/configure-liveness-readiness-probes.html":{"url":"guide/configure-liveness-readiness-probes.html","title":"配置Pod的liveness和readiness探针","keywords":"","body":"配置Pod的liveness和readiness探针 当你使用kuberentes的时候，有没有遇到过Pod在启动后一会就挂掉然后又重新启动这样的恶性循环？你有没有想过kubernetes是如何检测pod是否还存活？虽然容器已经启动，但是kubernetes如何知道容器的进程是否准备好对外提供服务了呢？让我们通过kuberentes官网的这篇文章Configure Liveness and Readiness Probes，来一探究竟。 本文将向展示如何配置容器的存活和可读性探针。 Kubelet使用liveness probe（存活探针）来确定何时重启容器。例如，当应用程序处于运行状态但无法做进一步操作，liveness探针将捕获到deadlock，重启处于该状态下的容器，使应用程序在存在bug的情况下依然能够继续运行下去（谁的程序还没几个bug呢）。 Kubelet使用readiness probe（就绪探针）来确定容器是否已经就绪可以接受流量。只有当Pod中的容器都处于就绪状态时kubelet才会认定该Pod处于就绪状态。该信号的作用是控制哪些Pod应该作为service的后端。如果Pod处于非就绪状态，那么它们将会被从service的load balancer中移除。 定义 liveness命令 许多长时间运行的应用程序最终会转换到broken状态，除非重新启动，否则无法恢复。Kubernetes提供了liveness probe来检测和补救这种情况。 在本次练习将基于 gcr.io/google_containers/busybox镜像创建运行一个容器的Pod。以下是Pod的配置文件exec-liveness.yaml： apiVersion: v1 kind: Pod metadata: labels: test: liveness name: liveness-exec spec: containers: - name: liveness args: - /bin/sh - -c - touch /tmp/healthy; sleep 30; rm -rf /tmp/healthy; sleep 600 image: gcr.io/google_containers/busybox livenessProbe: exec: command: - cat - /tmp/healthy initialDelaySeconds: 5 periodSeconds: 5 该配置文件给Pod配置了一个容器。periodSeconds 规定kubelet要每隔5秒执行一次liveness probe。 initialDelaySeconds 告诉kubelet在第一次执行probe之前要的等待5秒钟。探针检测命令是在容器中执行 cat /tmp/healthy 命令。如果命令执行成功，将返回0，kubelet就会认为该容器是活着的并且很健康。如果返回非0值，kubelet就会杀掉这个容器并重启它。 容器启动时，执行该命令： /bin/sh -c \"touch /tmp/healthy; sleep 30; rm -rf /tmp/healthy; sleep 600\" 在容器生命的最初30秒内有一个 /tmp/healthy 文件，在这30秒内 cat /tmp/healthy命令会返回一个成功的返回码。30秒后， cat /tmp/healthy 将返回失败的返回码。 创建Pod： kubectl create -f https://k8s.io/docs/tasks/configure-pod-container/exec-liveness.yaml 在30秒内，查看Pod的event： kubectl describe pod liveness-exec 结果显示没有失败的liveness probe： FirstSeen LastSeen Count From SubobjectPath Type Reason Message --------- -------- ----- ---- ------------- -------- ------ ------- 24s 24s 1 {default-scheduler } Normal Scheduled Successfully assigned liveness-exec to worker0 23s 23s 1 {kubelet worker0} spec.containers{liveness} Normal Pulling pulling image \"gcr.io/google_containers/busybox\" 23s 23s 1 {kubelet worker0} spec.containers{liveness} Normal Pulled Successfully pulled image \"gcr.io/google_containers/busybox\" 23s 23s 1 {kubelet worker0} spec.containers{liveness} Normal Created Created container with docker id 86849c15382e; Security:[seccomp=unconfined] 23s 23s 1 {kubelet worker0} spec.containers{liveness} Normal Started Started container with docker id 86849c15382e 启动35秒后，再次查看pod的event： kubectl describe pod liveness-exec 在最下面有一条信息显示liveness probe失败，容器被删掉并重新创建。 FirstSeen LastSeen Count From SubobjectPath Type Reason Message --------- -------- ----- ---- ------------- -------- ------ ------- 37s 37s 1 {default-scheduler } Normal Scheduled Successfully assigned liveness-exec to worker0 36s 36s 1 {kubelet worker0} spec.containers{liveness} Normal Pulling pulling image \"gcr.io/google_containers/busybox\" 36s 36s 1 {kubelet worker0} spec.containers{liveness} Normal Pulled Successfully pulled image \"gcr.io/google_containers/busybox\" 36s 36s 1 {kubelet worker0} spec.containers{liveness} Normal Created Created container with docker id 86849c15382e; Security:[seccomp=unconfined] 36s 36s 1 {kubelet worker0} spec.containers{liveness} Normal Started Started container with docker id 86849c15382e 2s 2s 1 {kubelet worker0} spec.containers{liveness} Warning Unhealthy Liveness probe failed: cat: can't open '/tmp/healthy': No such file or directory 再等30秒，确认容器已经重启： kubectl get pod liveness-exec 从输出结果来RESTARTS值加1了。 NAME READY STATUS RESTARTS AGE liveness-exec 1/1 Running 1 1m 定义一个liveness HTTP请求 我们还可以使用HTTP GET请求作为liveness probe。下面是一个基于gcr.io/google_containers/liveness镜像运行了一个容器的Pod的例子http-liveness.yaml： apiVersion: v1 kind: Pod metadata: labels: test: liveness name: liveness-http spec: containers: - name: liveness args: - /server image: gcr.io/google_containers/liveness livenessProbe: httpGet: path: /healthz port: 8080 httpHeaders: - name: X-Custom-Header value: Awesome initialDelaySeconds: 3 periodSeconds: 3 该配置文件只定义了一个容器，livenessProbe 指定kubelete需要每隔3秒执行一次liveness probe。initialDelaySeconds 指定kubelet在该执行第一次探测之前需要等待3秒钟。该探针将向容器中的server的8080端口发送一个HTTP GET请求。如果server的/healthz路径的handler返回一个成功的返回码，kubelet就会认定该容器是活着的并且很健康。如果返回失败的返回码，kubelet将杀掉该容器并重启它。 任何大于200小于400的返回码都会认定是成功的返回码。其他返回码都会被认为是失败的返回码。 查看该server的源码：server.go. 最开始的10秒该容器是活着的， /healthz handler返回200的状态码。这之后将返回500的返回码。 http.HandleFunc(\"/healthz\", func(w http.ResponseWriter, r *http.Request) { duration := time.Now().Sub(started) if duration.Seconds() > 10 { w.WriteHeader(500) w.Write([]byte(fmt.Sprintf(\"error: %v\", duration.Seconds()))) } else { w.WriteHeader(200) w.Write([]byte(\"ok\")) } }) 容器启动3秒后，kubelet开始执行健康检查。第一次健康监测会成功，但是10秒后，健康检查将失败，kubelet将杀掉和重启容器。 创建一个Pod来测试一下HTTP liveness检测： kubectl create -f https://k8s.io/docs/tasks/configure-pod-container/http-liveness.yaml After 10 seconds, view Pod events to verify that liveness probes have failed and the Container has been restarted: 10秒后，查看Pod的event，确认liveness probe失败并重启了容器。 kubectl describe pod liveness-http 定义TCP liveness探针 第三种liveness probe使用TCP Socket。 使用此配置，kubelet将尝试在指定端口上打开容器的套接字。 如果可以建立连接，容器被认为是健康的，如果不能就认为是失败的。 apiVersion: v1 kind: Pod metadata: name: goproxy labels: app: goproxy spec: containers: - name: goproxy image: gcr.io/google_containers/goproxy:0.1 ports: - containerPort: 8080 readinessProbe: tcpSocket: port: 8080 initialDelaySeconds: 5 periodSeconds: 10 livenessProbe: tcpSocket: port: 8080 initialDelaySeconds: 15 periodSeconds: 20 如您所见，TCP检查的配置与HTTP检查非常相似。 此示例同时使用了readiness和liveness probe。 容器启动后5秒钟，kubelet将发送第一个readiness probe。 这将尝试连接到端口8080上的goproxy容器。如果探测成功，则该pod将被标记为就绪。Kubelet将每隔10秒钟执行一次该检查。 除了readiness probe之外，该配置还包括liveness probe。 容器启动15秒后，kubelet将运行第一个liveness probe。 就像readiness probe一样，这将尝试连接到goproxy容器上的8080端口。如果liveness probe失败，容器将重新启动。 使用命名的端口 可以使用命名的ContainerPort作为HTTP或TCP liveness检查： ports: - name: liveness-port containerPort: 8080 hostPort: 8080 livenessProbe: httpGet: path: /healthz port: liveness-port 定义readiness探针 有时，应用程序暂时无法对外部流量提供服务。 例如，应用程序可能需要在启动期间加载大量数据或配置文件。 在这种情况下，你不想杀死应用程序，但你也不想发送请求。 Kubernetes提供了readiness probe来检测和减轻这些情况。 Pod中的容器可以报告自己还没有准备，不能处理Kubernetes服务发送过来的流量。 Readiness probe的配置跟liveness probe很像。唯一的不同是使用 readinessProbe而不是livenessProbe。 readinessProbe: exec: command: - cat - /tmp/healthy initialDelaySeconds: 5 periodSeconds: 5 Readiness probe的HTTP和TCP的探测器配置跟liveness probe一样。 Readiness和livenss probe可以并行用于同一容器。 使用两者可以确保流量无法到达未准备好的容器，并且容器在失败时重新启动。 配置Probe Probe中有很多精确和详细的配置，通过它们你能准确的控制liveness和readiness检查： initialDelaySeconds：容器启动后第一次执行探测是需要等待多少秒。 periodSeconds：执行探测的频率。默认是10秒，最小1秒。 timeoutSeconds：探测超时时间。默认1秒，最小1秒。 successThreshold：探测失败后，最少连续探测成功多少次才被认定为成功。默认是1。对于liveness必须是1。最小值是1。 failureThreshold：探测成功后，最少连续探测失败多少次才被认定为失败。默认是3。最小值是1。 HTTP probe中可以给 httpGet设置其他配置项： host：连接的主机名，默认连接到pod的IP。你可能想在http header中设置\"Host\"而不是使用IP。 scheme：连接使用的schema，默认HTTP。 path: 访问的HTTP server的path。 httpHeaders：自定义请求的header。HTTP运行重复的header。 port：访问的容器的端口名字或者端口号。端口号必须介于1和65525之间。 对于HTTP探测器，kubelet向指定的路径和端口发送HTTP请求以执行检查。 Kubelet将probe发送到容器的IP地址，除非地址被httpGet中的可选host字段覆盖。 在大多数情况下，你不想设置主机字段。 有一种情况下你可以设置它。 假设容器在127.0.0.1上侦听，并且Pod的hostNetwork字段为true。 然后，在httpGet下的host应该设置为127.0.0.1。 如果你的pod依赖于虚拟主机，这可能是更常见的情况，你不应该是用host，而是应该在httpHeaders中设置Host头。 参考 关于 Container Probes 的更多信息 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-07 10:22:34 "},"guide/configure-pod-service-account.html":{"url":"guide/configure-pod-service-account.html","title":"配置Pod的Service Account","keywords":"","body":"配置 Pod 的 Service Account Service account 为 Pod 中的进程提供身份信息。 本文是关于 Service Account 的用户指南，管理指南另见 Service Account 的集群管理指南 。 注意：本文档描述的关于 Serivce Account 的行为只有当您按照 Kubernetes 项目建议的方式搭建起集群的情况下才有效。您的集群管理员可能在您的集群中有自定义配置，这种情况下该文档可能并不适用。 当您（真人用户）访问集群（例如使用kubectl命令）时，apiserver 会将您认证为一个特定的 User Account（目前通常是admin，除非您的系统管理员自定义了集群配置）。Pod 容器中的进程也可以与 apiserver 联系。 当它们在联系 apiserver 的时候，它们会被认证为一个特定的 Service Account（例如default）。 使用默认的 Service Account 访问 API server 当您创建 pod 的时候，如果您没有指定一个 service account，系统会自动得在与该pod 相同的 namespace 下为其指派一个default service account。如果您获取刚创建的 pod 的原始 json 或 yaml 信息（例如使用kubectl get pods/podename -o yaml命令），您将看到spec.serviceAccountName字段已经被设置为 automatically set 。 您可以在 pod 中使用自动挂载的 service account 凭证来访问 API，如 Accessing the Cluster 中所描述。 Service account 是否能够取得访问 API 的许可取决于您使用的 授权插件和策略。 在 1.6 以上版本中，您可以选择取消为 service account 自动挂载 API 凭证，只需在 service account 中设置 automountServiceAccountToken: false： apiVersion: v1 kind: ServiceAccount metadata: name: build-robot automountServiceAccountToken: false ... 在 1.6 以上版本中，您也可以选择只取消单个 pod 的 API 凭证自动挂载： apiVersion: v1 kind: Pod metadata: name: my-pod spec: serviceAccountName: build-robot automountServiceAccountToken: false ... 如果在 pod 和 service account 中同时设置了 automountServiceAccountToken , pod 设置中的优先级更高。 使用多个Service Account 每个 namespace 中都有一个默认的叫做 default 的 service account 资源。 您可以使用以下命令列出 namespace 下的所有 serviceAccount 资源。 $ kubectl get serviceAccounts NAME SECRETS AGE default 1 1d 您可以像这样创建一个 ServiceAccount 对象： $ cat > /tmp/serviceaccount.yaml 如果您看到如下的 service account 对象的完整输出信息： $ kubectl get serviceaccounts/build-robot -o yaml apiVersion: v1 kind: ServiceAccount metadata: creationTimestamp: 2015-06-16T00:12:59Z name: build-robot namespace: default resourceVersion: \"272500\" selfLink: /api/v1/namespaces/default/serviceaccounts/build-robot uid: 721ab723-13bc-11e5-aec2-42010af0021e secrets: - name: build-robot-token-bvbk5 然后您将看到有一个 token 已经被自动创建，并被 service account 引用。 您可以使用授权插件来 设置 service account 的权限 。 设置非默认的 service account，只需要在 pod 的spec.serviceAccountName 字段中将name设置为您想要用的 service account 名字即可。 在 pod 创建之初 service account 就必须已经存在，否则创建将被拒绝。 您不能更新已创建的 pod 的 service account。 您可以清理 service account，如下所示： $ kubectl delete serviceaccount/build-robot 手动创建 service account 的 API token 假设我们已经有了一个如上文提到的名为 ”build-robot“ 的 service account，我们手动创建一个新的 secret。 $ cat > /tmp/build-robot-secret.yaml 现在您可以确认下新创建的 secret 取代了 \"build-robot\" 这个 service account 原来的 API token。 所有已不存在的 service account 的 token 将被 token controller 清理掉。 $ kubectl describe secrets/build-robot-secret Name: build-robot-secret Namespace: default Labels: Annotations: kubernetes.io/service-account.name=build-robot,kubernetes.io/service-account.uid=870ef2a5-35cf-11e5-8d06-005056b45392 Type: kubernetes.io/service-account-token Data ==== ca.crt: 1220 bytes token: ... namespace: 7 bytes 注意该内容中的token被省略了。 为 service account 添加 ImagePullSecret 首先，创建一个 imagePullSecret，详见这里。 然后，确认已创建。如： $ kubectl get secrets myregistrykey NAME TYPE DATA AGE myregistrykey kubernetes.io/.dockerconfigjson 1 1d 然后，修改 namespace 中的默认 service account 使用该 secret 作为 imagePullSecret。 kubectl patch serviceaccount default -p '{\"imagePullSecrets\": [{\"name\": \"myregistrykey\"}]}' Vi 交互过程中需要手动编辑： $ kubectl get serviceaccounts default -o yaml > ./sa.yaml $ cat sa.yaml apiVersion: v1 kind: ServiceAccount metadata: creationTimestamp: 2015-08-07T22:02:39Z name: default namespace: default resourceVersion: \"243024\" selfLink: /api/v1/namespaces/default/serviceaccounts/default uid: 052fb0f4-3d50-11e5-b066-42010af0d7b6 secrets: - name: default-token-uudge $ vi sa.yaml [editor session not shown] [delete line with key \"resourceVersion\"] [add lines with \"imagePullSecret:\"] $ cat sa.yaml apiVersion: v1 kind: ServiceAccount metadata: creationTimestamp: 2015-08-07T22:02:39Z name: default namespace: default selfLink: /api/v1/namespaces/default/serviceaccounts/default uid: 052fb0f4-3d50-11e5-b066-42010af0d7b6 secrets: - name: default-token-uudge imagePullSecrets: - name: myregistrykey $ kubectl replace serviceaccount default -f ./sa.yaml serviceaccounts/default 现在，所有当前 namespace 中新创建的 pod 的 spec 中都会增加如下内容： spec: imagePullSecrets: - name: myregistrykey Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-08-21 18:23:34 "},"guide/secret-configuration.html":{"url":"guide/secret-configuration.html","title":"Secret配置","keywords":"","body":"Secret 配置 Secret 对象类型用来保存敏感信息，例如密码、OAuth 令牌和 ssh key。将这些信息放在 secret 中比放在 pod 的定义中或者 docker 镜像中来说更加安全和灵活。参阅 Secret 设计文档 获取更多详细信息。 Secret 概览 Secret 是一种包含少量敏感信息例如密码、token 或 key 的对象。这样的信息可能会被放在 Pod spec 中或者镜像中；将其放在一个 secret 对象中可以更好地控制它的用途，并降低意外暴露的风险。 用户可以创建 secret，同时系统也创建了一些 secret。 要使用 secret，pod 需要引用 secret。Pod 可以用两种方式使用 secret：作为 volume 中的文件被挂载到 pod 中的一个或者多个容器里，或者当 kubelet 为 pod 拉取镜像时使用。 内置 secret Service Account 使用 API 凭证自动创建和附加 secret Kubernetes 自动创建包含访问 API 凭据的 secret，并自动修改您的 pod 以使用此类型的 secret。 如果需要，可以禁用或覆盖自动创建和使用API凭据。但是，如果您需要的只是安全地访问 apiserver，我们推荐这样的工作流程。 参阅 Service Account 文档获取关于 Service Account 如何工作的更多信息。 创建您自己的 Secret 使用 kubectl 创建 Secret 假设有些 pod 需要访问数据库。这些 pod 需要使用的用户名和密码在您本地机器的 ./username.txt 和 ./password.txt 文件里。 # Create files needed for rest of example. $ echo -n \"admin\" > ./username.txt $ echo -n \"1f2d1e2e67df\" > ./password.txt kubectl create secret 命令将这些文件打包到一个 Secret 中并在 API server 中创建了一个对象。 $ kubectl create secret generic db-user-pass --from-file=./username.txt --from-file=./password.txt secret \"db-user-pass\" created 您可以这样检查刚创建的 secret： $ kubectl get secrets NAME TYPE DATA AGE db-user-pass Opaque 2 51s $ kubectl describe secrets/db-user-pass Name: db-user-pass Namespace: default Labels: Annotations: Type: Opaque Data ==== password.txt: 12 bytes username.txt: 5 bytes 请注意，默认情况下，get 和 describe 命令都不会显示文件的内容。这是为了防止将 secret 中的内容被意外暴露给从终端日志记录中刻意寻找它们的人。 请参阅 解码 secret 了解如何查看它们的内容。 手动创建 Secret 您也可以先以 json 或 yaml 格式在文件中创建一个 secret 对象，然后创建该对象。 每一项必须是 base64 编码： $ echo -n \"admin\" | base64 YWRtaW4= $ echo -n \"1f2d1e2e67df\" | base64 MWYyZDFlMmU2N2Rm 现在可以像这样写一个 secret 对象： apiVersion: v1 kind: Secret metadata: name: mysecret type: Opaque data: username: YWRtaW4= password: MWYyZDFlMmU2N2Rm 数据字段是一个映射。它的键必须匹配 DNS_SUBDOMAIN，前导点也是可以的。这些值可以是任意数据，使用 base64 进行编码。 使用 kubectl create 创建 secret： $ kubectl create -f ./secret.yaml secret \"mysecret\" created 编码注意： secret 数据的序列化 JSON 和 YAML 值使用 base64 编码成字符串。换行符在这些字符串中无效，必须省略。当在Darwin/OS X上使用 base64 实用程序时，用户应避免使用 -b 选项来拆分长行。另外，对于 Linux用户如果 -w 选项不可用的话，应该添加选项 -w 0 到 base64 命令或管道 base64 | tr -d '\\n' 。 解码 Secret 可以使用 kubectl get secret 命令获取 secret。例如，获取在上一节中创建的 secret： $ kubectl get secret mysecret -o yaml apiVersion: v1 data: username: YWRtaW4= password: MWYyZDFlMmU2N2Rm kind: Secret metadata: creationTimestamp: 2016-01-22T18:41:56Z name: mysecret namespace: default resourceVersion: \"164619\" selfLink: /api/v1/namespaces/default/secrets/mysecret uid: cfee02d6-c137-11e5-8d73-42010af00002 type: Opaque 解码密码字段： $ echo \"MWYyZDFlMmU2N2Rm\" | base64 --decode 1f2d1e2e67df 使用 Secret Secret 可以作为数据卷被挂载，或作为环境变量暴露出来以供 pod 中的容器使用。它们也可以被系统的其他部分使用，而不直接暴露在 pod 内。例如，它们可以保存凭据，系统的其他部分应该用它来代表您与外部系统进行交互。 在 Pod 中使用 Secret 文件 在 Pod 中的 volume 里使用 Secret： 创建一个 secret 或者使用已有的 secret。多个 pod 可以引用同一个 secret。 修改您的 pod 的定义在 spec.volumes[] 下增加一个 volume。可以给这个 volume 随意命名，它的 spec.volumes[].secret.secretName 必须等于 secret 对象的名字。 将 spec.containers[].volumeMounts[] 加到需要用到该 secret 的容器中。指定 spec.containers[].volumeMounts[].readOnly = true 和 spec.containers[].volumeMounts[].mountPath 为您想要该 secret 出现的尚未使用的目录。 修改您的镜像并且／或者命令行让程序从该目录下寻找文件。Secret 的 data 映射中的每一个键都成为了 mountPath 下的一个文件名。 这是一个在 pod 中使用 volume 挂在 secret 的例子： apiVersion: v1 kind: Pod metadata: name: mypod spec: containers: - name: mypod image: redis volumeMounts: - name: foo mountPath: \"/etc/foo\" readOnly: true volumes: - name: foo secret: secretName: mysecret 您想要用的每个 secret 都需要在 spec.volumes 中指明。 如果 pod 中有多个容器，每个容器都需要自己的 volumeMounts 配置块，但是每个 secret 只需要一个 spec.volumes。 您可以打包多个文件到一个 secret 中，或者使用的多个 secret，怎样方便就怎样来。 向特性路径映射 secret 密钥 我们还可以控制 Secret key 映射在 volume 中的路径。您可以使用 spec.volumes[].secret.items 字段修改每个 key 的目标路径： apiVersion: v1 kind: Pod metadata: name: mypod spec: containers: - name: mypod image: redis volumeMounts: - name: foo mountPath: \"/etc/foo\" readOnly: true volumes: - name: foo secret: secretName: mysecret items: - key: username path: my-group/my-username 将会发生什么呢： username secret 存储在 /etc/foo/my-group/my-username 文件中而不是 /etc/foo/username 中。 password secret 没有被影射 如果使用了 spec.volumes[].secret.items，只有在 items 中指定的 key 被影射。要使用 secret 中所有的 key，所有这些都必须列在 items 字段中。所有列出的密钥必须存在于相应的 secret 中。否则，不会创建卷。 Secret 文件权限 您还可以指定 secret 将拥有的权限模式位文件。如果不指定，默认使用 0644。您可以为整个保密卷指定默认模式，如果需要，可以覆盖每个密钥。 apiVersion: v1 kind: Pod metadata: name: mypod spec: containers: - name: mypod image: redis volumeMounts: - name: foo mountPath: \"/etc/foo\" volumes: - name: foo secret: secretName: mysecret defaultMode: 256 然后，secret 将被挂载到 /etc/foo 目录，所有通过该 secret volume 挂载创建的文件的权限都是 0400。 请注意，JSON 规范不支持八进制符号，因此使用 256 值作为 0400 权限。如果您使用 yaml 而不是 json 作为 pod，则可以使用八进制符号以更自然的方式指定权限。 您还可以是用映射，如上一个示例，并为不同的文件指定不同的权限，如下所示： apiVersion: v1 kind: Pod metadata: name: mypod spec: containers: - name: mypod image: redis volumeMounts: - name: foo mountPath: \"/etc/foo\" volumes: - name: foo secret: secretName: mysecret items: - key: username path: my-group/my-username mode: 511 在这种情况下，导致 /etc/foo/my-group/my-username 的文件的权限值为 0777。由于 JSON 限制，必须以十进制格式指定模式。 请注意，如果稍后阅读此权限值可能会以十进制格式显示。 从 Volume 中消费 secret 值 在挂载的 secret volume 的容器内，secret key 将作为文件，并且 secret 的值使用 base-64 解码并存储在这些文件中。这是在上面的示例容器内执行的命令的结果： $ ls /etc/foo/ username password $ cat /etc/foo/username admin $ cat /etc/foo/password 1f2d1e2e67df 容器中的程序负责从文件中读取 secret。 挂载的 secret 被自动更新 当已经在 volume 中消被消费的 secret 被更新时，被映射的 key 也将被更新。 Kubelet 在周期性同步时检查被挂载的 secret 是不是最新的。但是，它正在使用其基于本地 ttl 的缓存来获取当前的 secret 值。结果是，当 secret 被更新的时刻到将新的 secret 映射到 pod 的时刻的总延迟可以与 kubelet 中的secret 缓存的 kubelet sync period + ttl 一样长。 Secret 作为环境变量 将 secret 作为 pod 中的环境变量使用： 创建一个 secret 或者使用一个已存在的 secret。多个 pod 可以引用同一个 secret。 在每个容器中修改您想要使用 secret key 的 Pod 定义，为要使用的每个 secret key 添加一个环境变量。消费secret key 的环境变量应填充 secret 的名称，并键入 env[x].valueFrom.secretKeyRef。 修改镜像并／或者命令行，以便程序在指定的环境变量中查找值。 apiVersion: v1 kind: Pod metadata: name: secret-env-pod spec: containers: - name: mycontainer image: redis env: - name: SECRET_USERNAME valueFrom: secretKeyRef: name: mysecret key: username - name: SECRET_PASSWORD valueFrom: secretKeyRef: name: mysecret key: password restartPolicy: Never 消费环境变量里的 Secret 值 在一个消耗环境变量 secret 的容器中，secret key 作为包含 secret 数据的 base-64 解码值的常规环境变量。这是从上面的示例在容器内执行的命令的结果： $ echo $SECRET_USERNAME admin $ echo $SECRET_PASSWORD 1f2d1e2e67df 使用 imagePullSecret imagePullSecret 是将包含 Docker（或其他）镜像注册表密码的 secret 传递给 Kubelet 的一种方式，因此可以代表您的 pod 拉取私有镜像。 手动指定 imagePullSecret imagePullSecret 的使用在 镜像文档 中说明。 安排 imagePullSecrets 自动附加 您可以手动创建 imagePullSecret，并从 serviceAccount 引用它。使用该 serviceAccount 创建的任何 pod 和默认使用该 serviceAccount 的 pod 将会将其的 imagePullSecret 字段设置为服务帐户的 imagePullSecret 字段。有关该过程的详细说明，请参阅 将 ImagePullSecrets 添加到服务帐户。 自动挂载手动创建的 Secret 手动创建的 secret（例如包含用于访问 github 帐户的令牌）可以根据其服务帐户自动附加到 pod。请参阅 使用 PodPreset 向 Pod 中注入信息 以获取该进程的详细说明。 详细 限制 验证 secret volume 来源确保指定的对象引用实际上指向一个类型为 Secret 的对象。因此，需要在依赖于它的任何 pod 之前创建一个 secret。 Secret API 对象驻留在命名空间中。它们只能由同一命名空间中的 pod 引用。 每个 secret 的大小限制为1MB。这是为了防止创建非常大的 secret 会耗尽 apiserver 和 kubelet 的内存。然而，创建许多较小的 secret 也可能耗尽内存。更全面得限制 secret 对内存使用的更全面的限制是计划中的功能。 Kubelet 仅支持从 API server 获取的 Pod 使用 secret。这包括使用 kubectl 创建的任何 pod，或间接通过 replication controller 创建的 pod。它不包括通过 kubelet --manifest-url 标志，其 --config 标志或其 REST API 创建的pod（这些不是创建 pod 的常用方法）。 必须先创建 secret，除非将它们标记为可选项，否则必须在将其作为环境变量在 pod 中使用之前创建 secret。对不存在的 secret 的引用将阻止其启动。 通过 secretKeyRef 对不存在于命名的 key 中的 key 进行引用将阻止该启动。 用于通过 envFrom 填充环境变量的 secret，这些环境变量具有被认为是无效环境变量名称的 key 将跳过这些键。该 pod 将被允许启动。将会有一个事件，其原因是 InvalidVariableNames，该消息将包含被跳过的无效键的列表。该示例显示一个 pod，它指的是包含2个无效键，1badkey 和 2alsobad 的默认/mysecret ConfigMap。 $ kubectl get events LASTSEEN FIRSTSEEN COUNT NAME KIND SUBOBJECT TYPE REASON 0s 0s 1 dapi-test-pod Pod Warning InvalidEnvironmentVariableNames kubelet, 127.0.0.1 Keys [1badkey, 2alsobad] from the EnvFrom secret default/mysecret were skipped since they are considered invalid environment variable names. Secret 与 Pod 生命周期的联系 通过 API 创建的 Pod 时，不会检查应用的 secret 是否存在。一旦 Pod 被调度，kubelet 就会尝试获取该 secret 的值。如果获取不到该 secret，或者暂时无法与 API server 建立连接，kubelet 将会定期重试。Kubelet 将会报告关于 pod 的事件，并解释它无法启动的原因。一旦获取的 secret，kubelet将创建并装载一个包含它的卷。在安装所有pod的卷之前，都不会启动 pod 的容器。 使用案例 使用案例：包含 ssh 密钥的 pod 创建一个包含 ssh key 的 secret： $ kubectl create secret generic ssh-key-secret --from-file=ssh-privatekey=/path/to/.ssh/id_rsa --from-file=ssh-publickey=/path/to/.ssh/id_rsa.pub 安全性注意事项：发送自己的 ssh 密钥之前要仔细思考：集群的其他用户可能有权访问该密钥。使用您想要共享 Kubernetes 群集的所有用户可以访问的服务帐户，如果它们遭到入侵，可以撤销。 现在我们可以创建一个使用 ssh 密钥引用 secret 的pod，并在一个卷中使用它： kind: Pod apiVersion: v1 metadata: name: secret-test-pod labels: name: secret-test spec: volumes: - name: secret-volume secret: secretName: ssh-key-secret containers: - name: ssh-test-container image: mySshImage volumeMounts: - name: secret-volume readOnly: true mountPath: \"/etc/secret-volume\" 当容器中的命令运行时，密钥的片段将可在以下目录： /etc/secret-volume/ssh-publickey /etc/secret-volume/ssh-privatekey 然后容器可以自由使用密钥数据建立一个 ssh 连接。 使用案例：包含 prod/test 凭据的 pod 下面的例子说明一个 pod 消费一个包含 prod 凭据的 secret，另一个 pod 使用测试环境凭据消费 secret。 创建 secret： $ kubectl create secret generic prod-db-secret --from-literal=username=produser --from-literal=password=Y4nys7f11 secret \"prod-db-secret\" created $ kubectl create secret generic test-db-secret --from-literal=username=testuser --from-literal=password=iluvtests secret \"test-db-secret\" created 创建 pod ： apiVersion: v1 kind: List items: - kind: Pod apiVersion: v1 metadata: name: prod-db-client-pod labels: name: prod-db-client spec: volumes: - name: secret-volume secret: secretName: prod-db-secret containers: - name: db-client-container image: myClientImage volumeMounts: - name: secret-volume readOnly: true mountPath: \"/etc/secret-volume\" - kind: Pod apiVersion: v1 metadata: name: test-db-client-pod labels: name: test-db-client spec: volumes: - name: secret-volume secret: secretName: test-db-secret containers: - name: db-client-container image: myClientImage volumeMounts: - name: secret-volume readOnly: true mountPath: \"/etc/secret-volume\" 这两个容器将在其文件系统上显示以下文件，其中包含每个容器环境的值： /etc/secret-volume/username /etc/secret-volume/password 请注意，两个 pod 的 spec 配置中仅有一个字段有所不同；这有助于使用普通的 pod 配置模板创建具有不同功能的 pod。您可以使用两个 service account 进一步简化基本 pod spec：一个名为 prod-user 拥有 prod-db-secret ，另一个称为 test-user 拥有 test-db-secret 。然后，pod spec 可以缩短为，例如： kind: Pod apiVersion: v1 metadata: name: prod-db-client-pod labels: name: prod-db-client spec: serviceAccount: prod-db-client containers: - name: db-client-container image: myClientImage 使用案例：secret 卷中以点号开头的文件 为了将数据“隐藏”起来（即文件名以点号开头的文件），简单地说让该键以一个点开始。例如，当如下 secret 被挂载到卷中： kind: Secret apiVersion: v1 metadata: name: dotfile-secret data: .secret-file: dmFsdWUtMg0KDQo= --- kind: Pod apiVersion: v1 metadata: name: secret-dotfiles-pod spec: volumes: - name: secret-volume secret: secretName: dotfile-secret containers: - name: dotfile-test-container image: gcr.io/google_containers/busybox command: - ls - \"-l\" - \"/etc/secret-volume\" volumeMounts: - name: secret-volume readOnly: true mountPath: \"/etc/secret-volume\" Secret-volume 将包含一个单独的文件，叫做 .secret-file，dotfile-test-container 的 /etc/secret-volume/.secret-file路径下将有该文件。 注意 以点号开头的文件在 ls -l 的输出中被隐藏起来了；列出目录内容时，必须使用 ls -la 才能查看它们。 使用案例：Secret 仅对 pod 中的一个容器可见 考虑以下一个需要处理 HTTP 请求的程序，执行一些复杂的业务逻辑，然后使用 HMAC 签署一些消息。因为它具有复杂的应用程序逻辑，所以在服务器中可能会出现一个未被注意的远程文件读取漏洞，这可能会将私钥暴露给攻击者。 这可以在两个容器中分为两个进程：前端容器，用于处理用户交互和业务逻辑，但无法看到私钥；以及可以看到私钥的签名者容器，并且响应来自前端的简单签名请求（例如通过本地主机网络）。 使用这种分割方法，攻击者现在必须欺骗应用程序服务器才能进行任意的操作，这可能比使其读取文件更难。 最佳实践 客户端使用 secret API 当部署与 secret API 交互的应用程序时，应使用诸如 RBAC 之类的 授权策略 来限制访问。 Secret 的重要性通常不尽相同，其中许多可能只对 Kubernetes 集群内（例如 service account 令牌）和对外部系统造成影响。即使一个应用程序可以理解其期望的与之交互的 secret 的权力，但是同一命名空间中的其他应用程序也可以使这些假设无效。 由于这些原因，在命名空间中 watch 和 list secret 的请求是非常强大的功能，应该避免这样的行为，因为列出 secret 可以让客户端检查所有 secret 是否在该命名空间中。在群集中watch 和 list 所有 secret 的能力应该只保留给最有特权的系统级组件。 需要访问 secrets API 的应用程序应该根据他们需要的 secret 执行 get 请求。这允许管理员限制对所有 secret 的访问，同时设置 白名单访问 应用程序需要的各个实例。 为了提高循环获取的性能，客户端可以设计引用 secret 的资源，然后 watch 资源，在引用更改时重新请求 secret。此外，还提出了一种 ”批量监控“ API 来让客户端 watch 每个资源，该功能可能会在将来的 Kubernetes 版本中提供。 安全属性 保护 因为 secret 对象可以独立于使用它们的 pod 而创建，所以在创建、查看和编辑 pod 的流程中 secret 被暴露的风险较小。系统还可以对 secret 对象采取额外的预防措施，例如避免将其写入到磁盘中可能的位置。 只有当节点上的 pod 需要用到该 secret 时，该 secret 才会被发送到该节点上。它不会被写入磁盘，而是存储在 tmpfs 中。一旦依赖于它的 pod 被删除，它就被删除。 在大多数 Kubernetes 项目维护的发行版中，用户与 API server 之间的通信以及从 API server 到 kubelet 的通信都受到 SSL/TLS 的保护。通过这些通道传输时，secret 受到保护。 节点上的 secret 数据存储在 tmpfs 卷中，因此不会传到节点上的其他磁盘。 同一节点上的很多个 pod 可能拥有多个 secret。但是，只有 pod 请求的 secret 在其容器中才是可见的。因此，一个 pod 不能访问另一个 Pod 的 secret。 Pod 中有多个容器。但是，pod 中的每个容器必须请求其挂载卷中的 secret 卷才能在容器内可见。这可以用于 在 Pod 级别构建安全分区。 风险 API server 的 secret 数据以纯文本的方式存储在 etcd 中；因此： 管理员应该限制 admin 用户访问 etcd； API server 中的 secret 数据位于 etcd 使用的磁盘上；管理员可能希望在不再使用时擦除/粉碎 etcd 使用的磁盘 如果您将 secret 数据编码为 base64 的清单（JSON 或 YAML）文件，共享该文件或将其检入代码库，这样的话该密码将会被泄露。 Base64 编码不是一种加密方式，一样也是纯文本。 应用程序在从卷中读取 secret 后仍然需要保护 secret 的值，例如不会意外记录或发送给不信任方。 可以创建和使用 secret 的 pod 的用户也可以看到该 secret 的值。即使 API server 策略不允许用户读取 secret 对象，用户也可以运行暴露 secret 的 pod。 如果运行了多个副本，那么这些 secret 将在它们之间共享。默认情况下，etcd 不能保证与 SSL/TLS 的对等通信，尽管可以进行配置。 目前，任何节点的 root 用户都可以通过模拟 kubelet 来读取 API server 中的任何 secret。只有向实际需要它们的节点发送 secret 才能限制单个节点的根漏洞的影响，该功能还在计划中。 原文地址：https://github.com/rootsongjc/kubernetes.github.io/blob/master/docs/concepts/configuration/secret.md Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-09-28 21:19:22 "},"guide/resource-quota-management.html":{"url":"guide/resource-quota-management.html","title":"管理namespace中的资源配额","keywords":"","body":"管理namespace中的资源配额 当用多个团队或者用户共用同一个集群的时候难免会有资源竞争的情况发生，这时候就需要对不同团队或用户的资源使用配额做出限制。 开启资源配额限制功能 目前有两种资源分配管理相关的控制策略插件 ResourceQuota 和 LimitRange。 要启用它们只要 API Server 的启动配置的 KUBE_ADMISSION_CONTROL 参数中加入了 ResourceQuota 的设置，这样就给集群开启了资源配额限制功能，加入 LimitRange 可以用来限制一个资源申请的范围限制，参考 为 namesapce 配置默认的内存请求与限额 和 在 namespace 中配置默认的CPU请求与限额。 两种控制策略的作用范围都是对于某一 namespace，ResourceQuota 用来限制 namespace 中所有的 Pod 占用的总的资源 request 和 limit，而 LimitRange 是用来设置 namespace 中 Pod 的默认的资源 request 和 limit 值。 资源配额分为三种类型： 计算资源配额 存储资源配额 对象数量配额 关于资源配额的详细信息请参考 kubernetes 官方文档 资源配额。 示例 我们为 spark-cluster 这个 namespace 设置 ResouceQuota 和 LimitRange。 以下 yaml 文件可以在 kubernetes-handbook 的 manifests/spark-with-kubernetes-native-scheduler 目录下找到。 配置计算资源配额 配置文件：spark-compute-resources.yaml apiVersion: v1 kind: ResourceQuota metadata: name: compute-resources namespace: spark-cluster spec: hard: pods: \"20\" requests.cpu: \"20\" requests.memory: 100Gi limits.cpu: \"40\" limits.memory: 200Gi 要想查看该配置只要执行： kubectl -n spark-cluster describe resourcequota compute-resources 配置对象数量限制 配置文件：spark-object-counts.yaml apiVersion: v1 kind: ResourceQuota metadata: name: object-counts namespace: spark-cluster spec: hard: configmaps: \"10\" persistentvolumeclaims: \"4\" replicationcontrollers: \"20\" secrets: \"10\" services: \"10\" services.loadbalancers: \"2\" 配置CPU和内存LimitRange 配置文件：spark-limit-range.yaml apiVersion: v1 kind: LimitRange metadata: name: mem-limit-range spec: limits: - default: memory: 50Gi cpu: 5 defaultRequest: memory: 1Gi cpu: 1 type: Container default 即 limit 的值 defaultRequest 即 request 的值 参考 资源配额 为命名空间配置默认的内存请求与限额 在命名空间中配置默认的CPU请求与限额 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-10-10 14:50:27 "},"guide/command-usage.html":{"url":"guide/command-usage.html","title":"命令使用","keywords":"","body":"命令使用 Kubernetes 中的 kubectl 及其他管理命令使用。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-08-21 18:23:34 "},"guide/docker-cli-to-kubectl.html":{"url":"guide/docker-cli-to-kubectl.html","title":"docker用户过度到kubectl命令行指南","keywords":"","body":"docker用户过度到kubectl命令行指南 对于没有使用过 kubernetes 的 docker 用户，如何快速掌握 kubectl 命令？ 在本文中，我们将向 docker-cli 用户介绍 Kubernetes 命令行如何与 api 进行交互。该命令行工具——kubectl，被设计成 docker-cli 用户所熟悉的样子，但是它们之间又存在一些必要的差异。该文档将向您展示每个 docker 子命令和 kubectl 与其等效的命令。 在使用 kubernetes 集群的时候，docker 命令通常情况是不需要用到的，只有在调试程序或者容器的时候用到，我们基本上使用 kubectl 命令即可，所以在操作 kubernetes 的时候我们抛弃原先使用 docker 时的一些观念。 docker run 如何运行一个 nginx Deployment 并将其暴露出来？ 查看 kubectl run 。 使用 docker 命令： $ docker run -d --restart=always -e DOMAIN=cluster --name nginx-app -p 80:80 nginx a9ec34d9878748d2f33dc20cb25c714ff21da8d40558b45bfaec9955859075d0 $ docker ps CONTAINER ID IMAGE COMMAND CREATED STATUS PORTS NAMES a9ec34d98787 nginx \"nginx -g 'daemon of 2 seconds ago Up 2 seconds 0.0.0.0:80->80/tcp, 443/tcp nginx-app 使用 kubectl 命令： # start the pod running nginx $ kubectl run --image=nginx nginx-app --port=80 --env=\"DOMAIN=cluster\" deployment \"nginx-app\" created 在大于等于 1.2 版本 Kubernetes 集群中，使用kubectl run 命令将创建一个名为 \"nginx-app\" 的 Deployment。如果您运行的是老版本，将会创建一个 replication controller。 如果您想沿用旧的行为，使用 --generation=run/v1 参数，这样就会创建 replication controller。查看 kubectl run 获取更多详细信息。 # expose a port through with a service $ kubectl expose deployment nginx-app --port=80 --name=nginx-http service \"nginx-http\" exposed 在 kubectl 命令中，我们创建了一个 Deployment，这将保证有 N 个运行 nginx 的 pod（N 代表 spec 中声明的 replica 数，默认为 1）。我们还创建了一个 service，使用 selector 匹配具有相应的 selector 的 Deployment。查看 快速开始 获取更多信息。 默认情况下镜像会在后台运行，与docker run -d ... 类似，如果您想在前台运行，使用： kubectl run [-i] [--tty] --attach --image= 与 docker run ... 不同的是，如果指定了 --attach ，我们将连接到 stdin，stdout 和 stderr，而不能控制具体连接到哪个输出流（docker -a ...）。 因为我们使用 Deployment 启动了容器，如果您终止了连接到的进程（例如 ctrl-c），容器将会重启，这跟 docker run -it不同。 如果想销毁该 Deployment（和它的 pod），您需要运行 kubeclt delete deployment 。 docker ps 如何列出哪些正在运行？查看 kubectl get。 使用 docker 命令： $ docker ps CONTAINER ID IMAGE COMMAND CREATED STATUS PORTS NAMES a9ec34d98787 nginx \"nginx -g 'daemon of About an hour ago Up About an hour 0.0.0.0:80->80/tcp, 443/tcp nginx-app 使用 kubectl 命令： $ kubectl get po NAME READY STATUS RESTARTS AGE nginx-app-5jyvm 1/1 Running 0 1h docker attach 如何连接到已经运行在容器中的进程？查看 kubectl attach。 使用 docker 命令： $ docker ps CONTAINER ID IMAGE COMMAND CREATED STATUS PORTS NAMES a9ec34d98787 nginx \"nginx -g 'daemon of 8 minutes ago Up 8 minutes 0.0.0.0:80->80/tcp, 443/tcp nginx-app $ docker attach a9ec34d98787 ... 使用 kubectl 命令： $ kubectl get pods NAME READY STATUS RESTARTS AGE nginx-app-5jyvm 1/1 Running 0 10m $ kubectl attach -it nginx-app-5jyvm ... docker exec 如何在容器中执行命令？查看 kubectl exec。 使用 docker 命令： $ docker ps CONTAINER ID IMAGE COMMAND CREATED STATUS PORTS NAMES a9ec34d98787 nginx \"nginx -g 'daemon of 8 minutes ago Up 8 minutes 0.0.0.0:80->80/tcp, 443/tcp nginx-app $ docker exec a9ec34d98787 cat /etc/hostname a9ec34d98787 使用 kubectl 命令： $ kubectl get po NAME READY STATUS RESTARTS AGE nginx-app-5jyvm 1/1 Running 0 10m $ kubectl exec nginx-app-5jyvm -- cat /etc/hostname nginx-app-5jyvm 执行交互式命令怎么办？ 使用 docker 命令： $ docker exec -ti a9ec34d98787 /bin/sh # exit 使用 kubectl 命令： $ kubectl exec -ti nginx-app-5jyvm -- /bin/sh # exit 更多信息请查看 获取运行中容器的 Shell 环境。 docker logs 如何查看运行中进程的 stdout/stderr？查看 kubectl logs。 使用 docker 命令： $ docker logs -f a9e 192.168.9.1 - - [14/Jul/2015:01:04:02 +0000] \"GET / HTTP/1.1\" 200 612 \"-\" \"curl/7.35.0\" \"-\" 192.168.9.1 - - [14/Jul/2015:01:04:03 +0000] \"GET / HTTP/1.1\" 200 612 \"-\" \"curl/7.35.0\" \"-\" 使用 kubectl 命令： $ kubectl logs -f nginx-app-zibvs 10.240.63.110 - - [14/Jul/2015:01:09:01 +0000] \"GET / HTTP/1.1\" 200 612 \"-\" \"curl/7.26.0\" \"-\" 10.240.63.110 - - [14/Jul/2015:01:09:02 +0000] \"GET / HTTP/1.1\" 200 612 \"-\" \"curl/7.26.0\" \"-\" 现在是时候提一下 pod 和容器之间的细微差别了；默认情况下如果 pod 中的进程退出 pod 也不会终止，相反它将会重启该进程。这类似于 docker run 时的 --restart=always 选项， 这是主要差别。在 docker 中，进程的每个调用的输出都是被连接起来的，但是对于 kubernetes，每个调用都是分开的。要查看以前在 kubernetes 中执行的输出，请执行以下操作： $ kubectl logs --previous nginx-app-zibvs 10.240.63.110 - - [14/Jul/2015:01:09:01 +0000] \"GET / HTTP/1.1\" 200 612 \"-\" \"curl/7.26.0\" \"-\" 10.240.63.110 - - [14/Jul/2015:01:09:02 +0000] \"GET / HTTP/1.1\" 200 612 \"-\" \"curl/7.26.0\" \"-\" 查看 记录和监控集群活动 获取更多信息。 docker stop 和 docker rm 如何停止和删除运行中的进程？查看 kubectl delete。 使用 docker 命令： $ docker ps CONTAINER ID IMAGE COMMAND CREATED STATUS PORTS NAMES a9ec34d98787 nginx \"nginx -g 'daemon of 22 hours ago Up 22 hours 0.0.0.0:80->80/tcp, 443/tcp nginx-app $ docker stop a9ec34d98787 a9ec34d98787 $ docker rm a9ec34d98787 a9ec34d98787 使用 kubectl 命令： $ kubectl get deployment nginx-app NAME DESIRED CURRENT UP-TO-DATE AVAILABLE AGE nginx-app 1 1 1 1 2m $ kubectl get po -l run=nginx-app NAME READY STATUS RESTARTS AGE nginx-app-2883164633-aklf7 1/1 Running 0 2m $ kubectl delete deployment nginx-app deployment \"nginx-app\" deleted $ kubectl get po -l run=nginx-app # Return nothing 请注意，我们不直接删除 pod。使用 kubectl 命令，我们要删除拥有该 pod 的 Deployment。如果我们直接删除pod，Deployment 将会重新创建该 pod。 docker login 在 kubectl 中没有对 docker login 的直接模拟。如果您有兴趣在私有镜像仓库中使用 Kubernetes，请参阅 使用私有镜像仓库。 docker version 如何查看客户端和服务端的版本？查看 kubectl version。 使用 docker 命令： $ docker version Client version: 1.7.0 Client API version: 1.19 Go version (client): go1.4.2 Git commit (client): 0baf609 OS/Arch (client): linux/amd64 Server version: 1.7.0 Server API version: 1.19 Go version (server): go1.4.2 Git commit (server): 0baf609 OS/Arch (server): linux/amd64 使用 kubectl 命令： $ kubectl version Client Version: version.Info{Major:\"1\", Minor:\"6\", GitVersion:\"v1.6.9+a3d1dfa6f4335\", GitCommit:\"9b77fed11a9843ce3780f70dd251e92901c43072\", GitTreeState:\"dirty\", BuildDate:\"2017-08-29T20:32:58Z\", OpenPaasKubernetesVersion:\"v1.03.02\", GoVersion:\"go1.7.5\", Compiler:\"gc\", Platform:\"linux/amd64\"} Server Version: version.Info{Major:\"1\", Minor:\"6\", GitVersion:\"v1.6.9+a3d1dfa6f4335\", GitCommit:\"9b77fed11a9843ce3780f70dd251e92901c43072\", GitTreeState:\"dirty\", BuildDate:\"2017-08-29T20:32:58Z\", OpenPaasKubernetesVersion:\"v1.03.02\", GoVersion:\"go1.7.5\", Compiler:\"gc\", Platform:\"linux/amd64\"} docker info 如何获取有关环境和配置的各种信息？查看 kubectl cluster-info。 使用 docker 命令： $ docker info Containers: 40 Images: 168 Storage Driver: aufs Root Dir: /usr/local/google/docker/aufs Backing Filesystem: extfs Dirs: 248 Dirperm1 Supported: false Execution Driver: native-0.2 Logging Driver: json-file Kernel Version: 3.13.0-53-generic Operating System: Ubuntu 14.04.2 LTS CPUs: 12 Total Memory: 31.32 GiB Name: k8s-is-fun.mtv.corp.google.com ID: ADUV:GCYR:B3VJ:HMPO:LNPQ:KD5S:YKFQ:76VN:IANZ:7TFV:ZBF4:BYJO WARNING: No swap limit support 使用 kubectl 命令： $ kubectl cluster-info Kubernetes master is running at https://108.59.85.141 KubeDNS is running at https://108.59.85.141/api/v1/namespaces/kube-system/services/kube-dns/proxy KubeUI is running at https://108.59.85.141/api/v1/namespaces/kube-system/services/kube-ui/proxy Grafana is running at https://108.59.85.141/api/v1/namespaces/kube-system/services/monitoring-grafana/proxy Heapster is running at https://108.59.85.141/api/v1/namespaces/kube-system/services/monitoring-heapster/proxy InfluxDB is running at https://108.59.85.141/api/v1/namespaces/kube-system/services/monitoring-influxdb/proxy 原文地址：https://github.com/rootsongjc/kubernetes.github.io/blob/master/docs/user-guide/docker-cli-to-kubectl.md Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-09-16 20:53:17 "},"guide/using-kubectl.html":{"url":"guide/using-kubectl.html","title":"kubectl命令概览","keywords":"","body":"Kubctl命令概览 Kubernetes提供的kubectl命令是与集群交互最直接的方式，v1.6版本的kubectl命令参考图如下： 图片 - kubectl cheatsheet Kubectl的子命令主要分为8个类别： 基础命令（初学者都会使用的） 基础命令（中级） 部署命令 集群管理命令 故障排查和调试命令 高级命令 设置命令 其他命令 熟悉这些命令有助于大家来操作和管理kubernetes集群。 kube-shell 开源项目kube-shell可以为kubectl提供自动的命令提示和补全，使用起来特别方便，推荐给大家。 Kube-shell有以下特性： 命令提示，给出命令的使用说明 自动补全，列出可选命令并可以通过tab键自动补全，支持模糊搜索 高亮 使用tab键可以列出可选的对象 vim模式 Mac下安装 pip install kube-shell --user -U Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-03 17:44:53 "},"guide/kubectl-cheatsheet.html":{"url":"guide/kubectl-cheatsheet.html","title":"kubectl命令技巧大全","keywords":"","body":"kubectl 命令技巧大全 Kubctl 命令是操作 kubernetes 集群的最直接和最 skillful 的途径，这个60多MB大小的二进制文件，到底有啥能耐呢？请看下文： Kubectl 自动补全 $ source Kubectl 上下文和配置 设置 kubectl 命令交互的 kubernetes 集群并修改配置信息。参阅 使用 kubeconfig 文件进行跨集群验证 获取关于配置文件的详细信息。 $ kubectl config view # 显示合并后的 kubeconfig 配置 # 同时使用多个 kubeconfig 文件并查看合并后的配置 $ KUBECONFIG=~/.kube/config:~/.kube/kubconfig2 kubectl config view # 获取 e2e 用户的密码 $ kubectl config view -o jsonpath='{.users[?(@.name == \"e2e\")].user.password}' $ kubectl config current-context # 显示当前的上下文 $ kubectl config use-context my-cluster-name # 设置默认上下文为 my-cluster-name # 向 kubeconf 中增加支持基本认证的新集群 $ kubectl config set-credentials kubeuser/foo.kubernetes.com --username=kubeuser --password=kubepassword # 使用指定的用户名和 namespace 设置上下文 $ kubectl config set-context gce --user=cluster-admin --namespace=foo \\ && kubectl config use-context gce 创建对象 Kubernetes 的清单文件可以使用 json 或 yaml 格式定义。可以以 .yaml、.yml、或者 .json 为扩展名。 $ kubectl create -f ./my-manifest.yaml # 创建资源 $ kubectl create -f ./my1.yaml -f ./my2.yaml # 使用多个文件创建资源 $ kubectl create -f ./dir # 使用目录下的所有清单文件来创建资源 $ kubectl create -f https://git.io/vPieo # 使用 url 来创建资源 $ kubectl run nginx --image=nginx # 启动一个 nginx 实例 $ kubectl explain pods,svc # 获取 pod 和 svc 的文档 # 从 stdin 输入中创建多个 YAML 对象 $ cat 显示和查找资源 # Get commands with basic output $ kubectl get services # 列出所有 namespace 中的所有 service $ kubectl get pods --all-namespaces # 列出所有 namespace 中的所有 pod $ kubectl get pods -o wide # 列出所有 pod 并显示详细信息 $ kubectl get deployment my-dep # 列出指定 deployment $ kubectl get pods --include-uninitialized # 列出该 namespace 中的所有 pod 包括未初始化的 # 使用详细输出来描述命令 $ kubectl describe nodes my-node $ kubectl describe pods my-pod $ kubectl get services --sort-by=.metadata.name # List Services Sorted by Name # 根据重启次数排序列出 pod $ kubectl get pods --sort-by='.status.containerStatuses[0].restartCount' # 获取所有具有 app=cassandra 的 pod 中的 version 标签 $ kubectl get pods --selector=app=cassandra rc -o \\ jsonpath='{.items[*].metadata.labels.version}' # 获取所有节点的 ExternalIP $ kubectl get nodes -o jsonpath='{.items[*].status.addresses[?(@.type==\"ExternalIP\")].address}' # 列出属于某个 PC 的 Pod 的名字 # “jq”命令用于转换复杂的 jsonpath，参考 https://stedolan.github.io/jq/ $ sel=${$(kubectl get rc my-rc --output=json | jq -j '.spec.selector | to_entries | .[] | \"\\(.key)=\\(.value),\"')%?} $ echo $(kubectl get pods --selector=$sel --output=jsonpath={.items..metadata.name}) # 查看哪些节点已就绪 $ JSONPATH='{range .items[*]}{@.metadata.name}:{range @.status.conditions[*]}{@.type}={@.status};{end}{end}' \\ && kubectl get nodes -o jsonpath=\"$JSONPATH\" | grep \"Ready=True\" # 列出当前 Pod 中使用的 Secret $ kubectl get pods -o json | jq '.items[].spec.containers[].env[]?.valueFrom.secretKeyRef.name' | grep -v null | sort | uniq 更新资源 $ kubectl rolling-update frontend-v1 -f frontend-v2.json # 滚动更新 pod frontend-v1 $ kubectl rolling-update frontend-v1 frontend-v2 --image=image:v2 # 更新资源名称并更新镜像 $ kubectl rolling-update frontend --image=image:v2 # 更新 frontend pod 中的镜像 $ kubectl rolling-update frontend-v1 frontend-v2 --rollback # 退出已存在的进行中的滚动更新 $ cat pod.json | kubectl replace -f - # 基于 stdin 输入的 JSON 替换 pod # 强制替换，删除后重新创建资源。会导致服务中断。 $ kubectl replace --force -f ./pod.json # 为 nginx RC 创建服务，启用本地 80 端口连接到容器上的 8000 端口 $ kubectl expose rc nginx --port=80 --target-port=8000 # 更新单容器 pod 的镜像版本（tag）到 v4 $ kubectl get pod mypod -o yaml | sed 's/\\(image: myimage\\):.*$/\\1:v4/' | kubectl replace -f - $ kubectl label pods my-pod new-label=awesome # 添加标签 $ kubectl annotate pods my-pod icon-url=http://goo.gl/XXBTWq # 添加注解 $ kubectl autoscale deployment foo --min=2 --max=10 # 自动扩展 deployment “foo” 修补资源 使用策略合并补丁并修补资源。 $ kubectl patch node k8s-node-1 -p '{\"spec\":{\"unschedulable\":true}}' # 部分更新节点 # 更新容器镜像； spec.containers[*].name 是必须的，因为这是合并的关键字 $ kubectl patch pod valid-pod -p '{\"spec\":{\"containers\":[{\"name\":\"kubernetes-serve-hostname\",\"image\":\"new image\"}]}}' # 使用具有位置数组的 json 补丁更新容器镜像 $ kubectl patch pod valid-pod --type='json' -p='[{\"op\": \"replace\", \"path\": \"/spec/containers/0/image\", \"value\":\"new image\"}]' # 使用具有位置数组的 json 补丁禁用 deployment 的 livenessProbe $ kubectl patch deployment valid-deployment --type json -p='[{\"op\": \"remove\", \"path\": \"/spec/template/spec/containers/0/livenessProbe\"}]' 编辑资源 在编辑器中编辑任何 API 资源。 $ kubectl edit svc/docker-registry # 编辑名为 docker-registry 的 service $ KUBE_EDITOR=\"nano\" kubectl edit svc/docker-registry # 使用其它编辑器 Scale 资源 $ kubectl scale --replicas=3 rs/foo # Scale a replicaset named 'foo' to 3 $ kubectl scale --replicas=3 -f foo.yaml # Scale a resource specified in \"foo.yaml\" to 3 $ kubectl scale --current-replicas=2 --replicas=3 deployment/mysql # If the deployment named mysql's current size is 2, scale mysql to 3 $ kubectl scale --replicas=5 rc/foo rc/bar rc/baz # Scale multiple replication controllers 删除资源 $ kubectl delete -f ./pod.json # 删除 pod.json 文件中定义的类型和名称的 pod $ kubectl delete pod,service baz foo # 删除名为“baz”的 pod 和名为“foo”的 service $ kubectl delete pods,services -l name=myLabel # 删除具有 name=myLabel 标签的 pod 和 serivce $ kubectl delete pods,services -l name=myLabel --include-uninitialized # 删除具有 name=myLabel 标签的 pod 和 service，包括尚未初始化的 $ kubectl -n my-ns delete po,svc --all # 删除 my-ns namespace 下的所有 pod 和 serivce，包括尚未初始化的 与运行中的 Pod 交互 $ kubectl logs my-pod # dump 输出 pod 的日志（stdout） $ kubectl logs my-pod -c my-container # dump 输出 pod 中容器的日志（stdout，pod 中有多个容器的情况下使用） $ kubectl logs -f my-pod # 流式输出 pod 的日志（stdout） $ kubectl logs -f my-pod -c my-container # 流式输出 pod 中容器的日志（stdout，pod 中有多个容器的情况下使用） $ kubectl run -i --tty busybox --image=busybox -- sh # 交互式 shell 的方式运行 pod $ kubectl attach my-pod -i # 连接到运行中的容器 $ kubectl port-forward my-pod 5000:6000 # 转发 pod 中的 6000 端口到本地的 5000 端口 $ kubectl exec my-pod -- ls / # 在已存在的容器中执行命令（只有一个容器的情况下） $ kubectl exec my-pod -c my-container -- ls / # 在已存在的容器中执行命令（pod 中有多个容器的情况下） $ kubectl top pod POD_NAME --containers # 显示指定 pod 和容器的指标度量 与节点和集群交互 $ kubectl cordon my-node # 标记 my-node 不可调度 $ kubectl drain my-node # 清空 my-node 以待维护 $ kubectl uncordon my-node # 标记 my-node 可调度 $ kubectl top node my-node # 显示 my-node 的指标度量 $ kubectl cluster-info # 显示 master 和服务的地址 $ kubectl cluster-info dump # 将当前集群状态输出到 stdout $ kubectl cluster-info dump --output-directory=/path/to/cluster-state # 将当前集群状态输出到 /path/to/cluster-state # 如果该键和影响的污点（taint）已存在，则使用指定的值替换 $ kubectl taint nodes foo dedicated=special-user:NoSchedule 资源类型 下表列出的是 kubernetes 中所有支持的类型和缩写的别名。 资源类型 缩写别名 clusters componentstatuses cs configmaps cm daemonsets ds deployments deploy endpoints ep event ev horizontalpodautoscalers hpa ingresses ing jobs limitranges limits namespaces ns networkpolicies nodes no statefulsets persistentvolumeclaims pvc persistentvolumes pv pods po podsecuritypolicies psp podtemplates replicasets rs replicationcontrollers rc resourcequotas quota cronjob secrets serviceaccount sa services svc storageclasses thirdpartyresources 格式化输出 要以特定的格式向终端窗口输出详细信息，可以在 kubectl 命令中添加 -o 或者 -output 标志。 输出格式 描述 -o=custom-columns= 使用逗号分隔的自定义列列表打印表格 -o=custom-columns-file= 使用 文件中的自定义列模板打印表格 -o=json 输出 JSON 格式的 API 对象 -o=jsonpath= 打印 jsonpath 表达式中定义的字段 -o=jsonpath-file= 打印由 文件中的 jsonpath 表达式定义的字段 -o=name 仅打印资源名称 -o=wide 以纯文本格式输出任何附加信息，对于 Pod ，包含节点名称 -o=yaml 输出 YAML 格式的 API 对象 Kubectl 详细输出和调试 使用 -v 或 --v 标志跟着一个整数来指定日志级别。这里 描述了通用的 kubernetes 日志约定和相关的日志级别。 详细等级 描述 --v=0 总是对操作人员可见。 --v=1 合理的默认日志级别，如果您不需要详细输出。 --v=2 可能与系统的重大变化相关的，有关稳定状态的信息和重要的日志信息。这是对大多数系统推荐的日志级别。 --v=3 有关更改的扩展信息。 --v=4 调试级别详细输出。 --v=6 显示请求的资源。 --v=7 显示HTTP请求的header。 --v=8 显示HTTP请求的内容。 参考 Kubectl 概览 JsonPath 手册 本文是对官方文档的中文翻译，原文地址：https://kubernetes.io/docs/user-guide/kubectl-cheatsheet/。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-03 17:45:30 "},"guide/using-etcdctl-to-access-kubernetes-data.html":{"url":"guide/using-etcdctl-to-access-kubernetes-data.html","title":"使用etcdctl访问kuberentes数据","keywords":"","body":"使用etcdctl访问kuberentes数据 Kubenretes1.6中使用etcd V3版本的API，使用etcdctl直接ls的话只能看到/kube-centos一个路径。需要在命令前加上ETCDCTL_API=3这个环境变量才能看到kuberentes在etcd中保存的数据。 ETCDCTL_API=3 etcdctl get /registry/namespaces/default -w=json|python -m json.tool -w指定输出格式 将得到这样的json的结果： { \"count\": 1, \"header\": { \"cluster_id\": 12091028579527406772, \"member_id\": 16557816780141026208, \"raft_term\": 36, \"revision\": 29253467 }, \"kvs\": [ { \"create_revision\": 5, \"key\": \"L3JlZ2lzdHJ5L25hbWVzcGFjZXMvZGVmYXVsdA==\", \"mod_revision\": 5, \"value\": \"azhzAAoPCgJ2MRIJTmFtZXNwYWNlEmIKSAoHZGVmYXVsdBIAGgAiACokZTU2YzMzMDgtMWVhOC0xMWU3LThjZDctZjRlOWQ0OWY4ZWQwMgA4AEILCIn4sscFEKOg9xd6ABIMCgprdWJlcm5ldGVzGggKBkFjdGl2ZRoAIgA=\", \"version\": 1 } ] } 使用--prefix可以看到所有的子目录，如查看集群中的namespace： ETCDCTL_API=3 etcdctl get /registry/namespaces --prefix -w=json|python -m json.tool 输出结果中可以看到所有的namespace。 { \"count\": 8, \"header\": { \"cluster_id\": 12091028579527406772, \"member_id\": 16557816780141026208, \"raft_term\": 36, \"revision\": 29253722 }, \"kvs\": [ { \"create_revision\": 24310883, \"key\": \"L3JlZ2lzdHJ5L25hbWVzcGFjZXMvYXV0b21vZGVs\", \"mod_revision\": 24310883, \"value\": \"azhzAAoPCgJ2MRIJTmFtZXNwYWNlEmQKSgoJYXV0b21vZGVsEgAaACIAKiQ1MjczOTU1ZC1iMzEyLTExZTctOTcwYy1mNGU5ZDQ5ZjhlZDAyADgAQgsI7fSWzwUQ6Jv1Z3oAEgwKCmt1YmVybmV0ZXMaCAoGQWN0aXZlGgAiAA==\", \"version\": 1 }, { \"create_revision\": 21387676, \"key\": \"L3JlZ2lzdHJ5L25hbWVzcGFjZXMvYnJhbmQ=\", \"mod_revision\": 21387676, \"value\": \"azhzAAoPCgJ2MRIJTmFtZXNwYWNlEmEKRwoFYnJhbmQSABoAIgAqJGNkZmQ1Y2NmLWExYzktMTFlNy05NzBjLWY0ZTlkNDlmOGVkMDIAOABCDAjR9qLOBRDYn83XAXoAEgwKCmt1YmVybmV0ZXMaCAoGQWN0aXZlGgAiAA==\", \"version\": 1 }, { \"create_revision\": 5, \"key\": \"L3JlZ2lzdHJ5L25hbWVzcGFjZXMvZGVmYXVsdA==\", \"mod_revision\": 5, \"value\": \"azhzAAoPCgJ2MRIJTmFtZXNwYWNlEmIKSAoHZGVmYXVsdBIAGgAiACokZTU2YzMzMDgtMWVhOC0xMWU3LThjZDctZjRlOWQ0OWY4ZWQwMgA4AEILCIn4sscFEKOg9xd6ABIMCgprdWJlcm5ldGVzGggKBkFjdGl2ZRoAIgA=\", \"version\": 1 }, { \"create_revision\": 18504694, \"key\": \"L3JlZ2lzdHJ5L25hbWVzcGFjZXMvZGV2\", \"mod_revision\": 24310213, \"value\": \"azhzAAoPCgJ2MRIJTmFtZXNwYWNlEmwKUgoDZGV2EgAaACIAKiQyOGRlMGVjNS04ZTEzLTExZTctOTcwYy1mNGU5ZDQ5ZjhlZDAyADgAQgwI89CezQUQ0v2fuQNaCwoEbmFtZRIDZGV2egASDAoKa3ViZXJuZXRlcxoICgZBY3RpdmUaACIA\", \"version\": 4 }, { \"create_revision\": 10, \"key\": \"L3JlZ2lzdHJ5L25hbWVzcGFjZXMva3ViZS1wdWJsaWM=\", \"mod_revision\": 10, \"value\": \"azhzAAoPCgJ2MRIJTmFtZXNwYWNlEmcKTQoLa3ViZS1wdWJsaWMSABoAIgAqJGU1ZjhkY2I1LTFlYTgtMTFlNy04Y2Q3LWY0ZTlkNDlmOGVkMDIAOABCDAiJ+LLHBRDdrsDPA3oAEgwKCmt1YmVybmV0ZXMaCAoGQWN0aXZlGgAiAA==\", \"version\": 1 }, { \"create_revision\": 2, \"key\": \"L3JlZ2lzdHJ5L25hbWVzcGFjZXMva3ViZS1zeXN0ZW0=\", \"mod_revision\": 2, \"value\": \"azhzAAoPCgJ2MRIJTmFtZXNwYWNlEmYKTAoLa3ViZS1zeXN0ZW0SABoAIgAqJGU1NmFhMDVkLTFlYTgtMTFlNy04Y2Q3LWY0ZTlkNDlmOGVkMDIAOABCCwiJ+LLHBRDoq9ASegASDAoKa3ViZXJuZXRlcxoICgZBY3RpdmUaACIA\", \"version\": 1 }, { \"create_revision\": 3774247, \"key\": \"L3JlZ2lzdHJ5L25hbWVzcGFjZXMvc3BhcmstY2x1c3Rlcg==\", \"mod_revision\": 3774247, \"value\": \"azhzAAoPCgJ2MRIJTmFtZXNwYWNlEoABCmYKDXNwYXJrLWNsdXN0ZXISABoAIgAqJDMyNjY3ZDVjLTM0YWMtMTFlNy1iZmJkLThhZjFlM2E3YzViZDIAOABCDAiA1cbIBRDU3YuAAVoVCgRuYW1lEg1zcGFyay1jbHVzdGVyegASDAoKa3ViZXJuZXRlcxoICgZBY3RpdmUaACIA\", \"version\": 1 }, { \"create_revision\": 15212191, \"key\": \"L3JlZ2lzdHJ5L25hbWVzcGFjZXMveWFybi1jbHVzdGVy\", \"mod_revision\": 15212191, \"value\": \"azhzAAoPCgJ2MRIJTmFtZXNwYWNlEn0KYwoMeWFybi1jbHVzdGVyEgAaACIAKiQ2YWNhNjk1Yi03N2Y5LTExZTctYmZiZC04YWYxZTNhN2M1YmQyADgAQgsI1qiKzAUQkoqxDloUCgRuYW1lEgx5YXJuLWNsdXN0ZXJ6ABIMCgprdWJlcm5ldGVzGggKBkFjdGl2ZRoAIgA=\", \"version\": 1 } ] } key的值是经过base64编码，需要解码后才能看到实际值，如： $ echo L3JlZ2lzdHJ5L25hbWVzcGFjZXMvYXV0b21vZGVs|base64 -d /registry/namespaces/automodel 参考 etcd中文文档 etcd官方文档 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-23 13:01:50 "},"guide/cluster-security-management.html":{"url":"guide/cluster-security-management.html","title":"集群安全性管理","keywords":"","body":"集群安全性管理 Kuberentes 支持多租户，这就需要对集群的安全性进行管理。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-09-07 11:46:45 "},"guide/managing-tls-in-a-cluster.html":{"url":"guide/managing-tls-in-a-cluster.html","title":"管理集群中的TLS","keywords":"","body":"管理集群中的TLS 在本书的最佳实践部分，我们在CentOS上部署了kuberentes集群，其中最开始又重要的一步就是创建TLS认证的，查看创建TLS证书和秘钥。很多人在进行到这一步是都会遇到各种各样千奇百怪的问题，这一步是创建集群的基础，我们有必要详细了解一下背后的流程和原理。 概览 每个Kubernetes集群都有一个集群根证书颁发机构（CA）。 集群中的组件通常使用CA来验证API server的证书，由API服务器验证kubelet客户端证书等。为了支持这一点，CA证书包被分发到集群中的每个节点，并作为一个sercret附加分发到默认service account上。 或者，你的workload可以使用此CA建立信任。 你的应用程序可以使用类似于ACME草案的协议，使用certificates.k8s.io API请求证书签名。 集群中的TLS信任 让Pod中运行的应用程序信任集群根CA通常需要一些额外的应用程序配置。 您将需要将CA证书包添加到TLS客户端或服务器信任的CA证书列表中。 例如，您可以使用golang TLS配置通过解析证书链并将解析的证书添加到tls.Config结构中的Certificates字段中，CA证书捆绑包将使用默认服务账户自动加载到pod中，路径为/var/run/secrets/kubernetes.io/serviceaccount/ca.crt。 如果您没有使用默认服务账户，请请求集群管理员构建包含您有权访问使用的证书包的configmap。 请求认证 以下部分演示如何为通过DNS访问的Kubernetes服务创建TLS证书。 步骤0. 下载安装SSL 下载cfssl工具：https://pkg.cfssl.org/. 步骤1. 创建证书签名请求 通过运行以下命令生成私钥和证书签名请求（或CSR）： $ cat 172.168.0.24 是 service 的 cluster IP，my-svc.my-namespace.svc.cluster.local 是 service 的 DNS 名称， 10.0.34.2 是 Pod 的 IP， my-pod.my-namespace.pod.cluster.local 是pod 的 DNS 名称，你可以看到以下输出： 2017/03/21 06:48:17 [INFO] generate received request 2017/03/21 06:48:17 [INFO] received CSR 2017/03/21 06:48:17 [INFO] generating key: ecdsa-256 2017/03/21 06:48:17 [INFO] encoded CSR 此命令生成两个文件; 它生成包含PEM编码的pkcs #10认证请求的server.csr，以及包含仍然要创建的证书的PEM编码密钥的server-key.pem。 步骤2. 创建证书签名请求对象以发送到Kubernetes API 使用以下命令创建CSR yaml文件，并发送到API server： $ cat 请注意，在步骤1中创建的server.csr文件是base64编码并存储在.spec.request字段中。 我们还要求提供“数字签名”，“密钥加密”和“服务器身份验证”密钥用途的证书。 我们这里支持列出的所有关键用途和扩展的关键用途，以便您可以使用相同的API请求客户端证书和其他证书。 在API server中可以看到这些CSR处于pending状态。执行下面的命令你将可以看到： $ kubectl describe csr my-svc.my-namespace Name: my-svc.my-namespace Labels: Annotations: CreationTimestamp: Tue, 21 Mar 2017 07:03:51 -0700 Requesting User: yourname@example.com Status: Pending Subject: Common Name: my-svc.my-namespace.svc.cluster.local Serial Number: Subject Alternative Names: DNS Names: my-svc.my-namespace.svc.cluster.local IP Addresses: 172.168.0.24 10.0.34.2 Events: 步骤3. 获取证书签名请求 批准证书签名请求是通过自动批准过程完成的，或由集群管理员一次完成。 有关这方面涉及的更多信息，请参见下文。 步骤4. 下载签名并使用 一旦CSR被签署并获得批准，您应该看到以下内容： $ kubectl get csr NAME AGE REQUESTOR CONDITION my-svc.my-namespace 10m yourname@example.com Approved,Issued 你可以通过运行以下命令下载颁发的证书并将其保存到server.crt文件中： $ kubectl get csr my-svc.my-namespace -o jsonpath='{.status.certificate}' \\ | base64 -d > server.crt 现在你可以用server.crt和server-key.pem来做为keypair来启动HTTPS server。 批准证书签名请求 Kubernetes 管理员（具有适当权限）可以使用 kubectl certificate approve 和kubectl certificate deny 命令手动批准（或拒绝）证书签名请求。但是，如果您打算大量使用此 API，则可以考虑编写自动化的证书控制器。 如果上述机器或人类使用 kubectl，批准者的作用是验证 CSR 满足如下两个要求： CSR 的主体控制用于签署 CSR 的私钥。这解决了伪装成授权主体的第三方的威胁。在上述示例中，此步骤将验证该 pod 控制了用于生成 CSR 的私钥。 CSR 的主体被授权在请求的上下文中执行。这解决了我们加入群集的我们不期望的主体的威胁。在上述示例中，此步骤将是验证该 pod 是否被允许加入到所请求的服务中。 当且仅当满足这两个要求时，审批者应该批准 CSR，否则拒绝 CSR。 给集群管理员的一个建议 本教程假设将signer设置为服务证书API。Kubernetes controller manager提供了一个signer的默认实现。 要启用它，请将--cluster-signing-cert-file和--cluster-signing-key-file参数传递给controller manager，并配置具有证书颁发机构的密钥对的路径。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-09 14:09:42 "},"guide/kubelet-authentication-authorization.html":{"url":"guide/kubelet-authentication-authorization.html","title":"kubelet的认证授权","keywords":"","body":"Kublet的认证授权 概览 Kubelet 的 HTTPS 端点对外暴露了用于访问不同敏感程度数据的 API，并允许您在节点或者容器内执行不同权限级别的操作。 本文档向您描述如何通过认证授权来访问 kubelet 的 HTTPS 端点。 Kubelet 认证 默认情况下，所有未被配置的其他身份验证方法拒绝的，对 kubelet 的 HTTPS 端点的请求将被视为匿名请求，并被授予 system:anonymous 用户名和 system:unauthenticated 组。 如果要禁用匿名访问并发送 401 Unauthorized 的未经身份验证的请求的响应： 启动 kubelet 时指定 --anonymous-auth=false 标志 如果要对 kubelet 的 HTTPS 端点启用 X509 客户端证书身份验证： 启动 kubelet 时指定 --client-ca-file 标志，提供 CA bundle 以验证客户端证书 启动 apiserver 时指定 --kubelet-client-certificate 和 --kubelet-client-key 标志 参阅 apiserver 认证文档 获取更多详细信息。 启用 API bearer token（包括 service account token）用于向 kubelet 的 HTTPS 端点进行身份验证： 确保在 API server 中开启了 authentication.k8s.io/v1beta1 API 组。 启动 kubelet 时指定 --authentication-token-webhook， --kubeconfig 和 --require-kubeconfig 标志 Kubelet 在配置的 API server 上调用 TokenReview API 以确定来自 bearer token 的用户信息 Kubelet 授权 接着对任何成功验证的请求（包括匿名请求）授权。默认授权模式为 AlwaysAllow，允许所有请求。 细分访问 kubelet API 有很多原因： 启用匿名认证，但匿名用户调用 kubelet API 的能力应受到限制 启动 bearer token 认证，但是 API 用户（如 service account）调用 kubelet API 的能力应受到限制 客户端证书身份验证已启用，但只有那些配置了 CA 签名的客户端证书的用户才可以使用 kubelet API 如果要细分访问 kubelet API，将授权委托给 API server： 确保 API server 中启用了 authorization.k8s.io/v1beta1 API 组 启动 kubelet 时指定 --authorization-mode=Webhook、 --kubeconfig 和 --require-kubeconfig 标志 kubelet 在配置的 API server 上调用 SubjectAccessReview API，以确定每个请求是否被授权 kubelet 使用与 apiserver 相同的 请求属性 方法来授权 API 请求。 Verb（动词）是根据传入的请求的 HTTP 动词确定的： HTTP 动词 request 动词 POST create GET, HEAD get PUT update PATCH patch DELETE delete 资源和子资源根据传入请求的路径确定： Kubelet API 资源 子资源 /stats/* nodes stats /metrics/* nodes metrics /logs/* nodes log /spec/* nodes spec all others nodes proxy Namespace 和 API 组属性总是空字符串，资源的名字总是 kubelet 的 Node API 对象的名字。 当以该模式运行时，请确保用户为 apiserver 指定了 --kubelet-client-certificate 和 --kubelet-client-key 标志并授权了如下属性： verb=*, resource=nodes, subresource=proxy verb=*, resource=nodes, subresource=stats verb=*, resource=nodes, subresource=log verb=*, resource=nodes, subresource=spec verb=*, resource=nodes, subresource=metrics Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-08-21 18:23:34 "},"guide/tls-bootstrapping.html":{"url":"guide/tls-bootstrapping.html","title":"TLS bootstrap","keywords":"","body":"TLS Bootstrap 本文档介绍如何为 kubelet 设置 TLS 客户端证书引导（bootstrap）。 Kubernetes 1.4 引入了一个用于从集群级证书颁发机构（CA）请求证书的 API。此 API 的原始目的是为 kubelet 提供 TLS 客户端证书。可以在 这里 找到该提议，在 feature #43 追踪该功能的进度。 kube-apiserver 配置 您必须提供一个 token 文件，该文件中指定了至少一个分配给 kubelet 特定 bootstrap 组的 “bootstrap token”。 该组将作为 controller manager 配置中的默认批准控制器而用于审批。随着此功能的成熟，您应该确保 token 被绑定到基于角色的访问控制（RBAC）策略上，该策略严格限制了与证书配置相关的客户端请求（使用 bootstrap token）。使用 RBAC，将 token 范围划分为组可以带来很大的灵活性（例如，当您配置完成节点后，您可以禁用特定引导组的访问）。 Token 认证文件 Token 可以是任意的，但应该可以表示为从安全随机数生成器（例如大多数现代操作系统中的 /dev/urandom）导出的至少128位熵。生成 token 有很多中方式。例如： head -c 16 /dev/urandom | od -An -t x | tr -d ' ' 产生的 token 类似于这样： 02b50b05283e98dd0fd71db496ef01e8。 Token 文件应该类似于以下示例，其中前三个值可以是任何值，引用的组名称应如下所示： 02b50b05283e98dd0fd71db496ef01e8,kubelet-bootstrap,10001,\"system:kubelet-bootstrap\" 在 kube-apiserver 命令中添加 --token-auth-file=FILENAME 标志（可能在您的 systemd unit 文件中）来启用 token 文件。 查看 该文档 获取更多详细信息。 客户端证书 CA 包 在 kube-apiserver 命令中添加 --client-ca-file=FILENAME 标志启用客户端证书认证，指定包含签名证书的证书颁发机构包（例如 --client-ca-file=/var/lib/kubernetes/ca.pem）。 kube-controller-manager 配置 请求证书的 API 向 Kubernetes controller manager 中添加证书颁发控制循环。使用磁盘上的 cfssl 本地签名文件的形式。目前，所有发型的证书均为一年有效期和并具有一系列关键用途。 签名文件 您必须提供证书颁发机构，这样才能提供颁发证书所需的密码资料。 kube-apiserver 通过指定的 --client-ca-file=FILENAME 标志来认证和采信该 CA。CA 的管理超出了本文档的范围，但建议您为 Kubernetes 生成专用的 CA。 假定证书和密钥都是 PEM 编码的。 Kube-controller-manager 标志为： --cluster-signing-cert-file=\"/etc/path/to/kubernetes/ca/ca.crt\" --cluster-signing-key-file=\"/etc/path/to/kubernetes/ca/ca.key\" 审批控制器 在 kubernetes 1.7 版本中，实验性的 “组自动批准” 控制器被弃用，新的 csrapproving 控制器将作为 kube-controller-manager 的一部分，被默认启用。 控制器使用 SubjectAccessReview API 来确定给定用户是否已被授权允许请求 CSR，然后根据授权结果进行批准。为了防止与其他批准者冲突，内置审批者没有明确地拒绝 CSR，只是忽略未经授权的请求。 控制器将 CSR 分为三个子资源： nodeclient ：用户的客户端认证请求 O=system:nodes， CN=system:node:(node name)。 selfnodeclient：更新具有相同 O 和 CN 的客户端证书的节点。 selfnodeserver：更新服务证书的节点（ALPHA，需要 feature gate）。 当前，确定 CSR 是否为 selfnodeserver 请求的检查与 kubelet 的凭据轮换实现（Alpha 功能）相关联。因此，selfnodeserver 的定义将来可能会改变，并且需要 Controller Manager 上的RotateKubeletServerCertificate feature gate。该功能的进展可以在 kubernetes/feature/#267 上追踪。 --feature-gates=RotateKubeletServerCertificate=true 以下 RBAC ClusterRoles 代表 nodeClient、selfnodeclient 和 selfnodeserver 功能。在以后的版本中可能会自动创建类似的角色。 # A ClusterRole which instructs the CSR approver to approve a user requesting # node client credentials. kind: ClusterRole apiVersion: rbac.authorization.k8s.io/v1beta1 metadata: name: approve-node-client-csr rules: - apiGroups: [\"certificates.k8s.io\"] resources: [\"certificatesigningrequests/nodeclient\"] verbs: [\"create\"] --- # A ClusterRole which instructs the CSR approver to approve a node renewing its # own client credentials. kind: ClusterRole apiVersion: rbac.authorization.k8s.io/v1beta1 metadata: name: approve-node-client-renewal-csr rules: - apiGroups: [\"certificates.k8s.io\"] resources: [\"certificatesigningrequests/selfnodeclient\"] verbs: [\"create\"] --- # A ClusterRole which instructs the CSR approver to approve a node requesting a # serving cert matching its client cert. kind: ClusterRole apiVersion: rbac.authorization.k8s.io/v1beta1 metadata: name: approve-node-server-renewal-csr rules: - apiGroups: [\"certificates.k8s.io\"] resources: [\"certificatesigningrequests/selfnodeserver\"] verbs: [\"create\"] 这些权力可以授予给凭证，如 bootstrap token。例如，要复制由已被移除的自动批准标志提供的行为，由单个组批准所有的 CSR： # REMOVED: This flag no longer works as of 1.7. --insecure-experimental-approve-all-kubelet-csrs-for-group=\"kubelet-bootstrap-token\" 管理员将创建一个 ClusterRoleBinding 来定位该组。 # Approve all CSRs for the group \"kubelet-bootstrap-token\" kind: ClusterRoleBinding apiVersion: rbac.authorization.k8s.io/v1beta1 metadata: name: auto-approve-csrs-for-group subjects: - kind: Group name: kubelet-bootstrap-token apiGroup: rbac.authorization.k8s.io roleRef: kind: ClusterRole name: approve-node-client-csr apiGroup: rbac.authorization.k8s.io 要让节点更新自己的凭据，管理员可以构造一个 ClusterRoleBinding 来定位该节点的凭据。 kind: ClusterRoleBinding apiVersion: rbac.authorization.k8s.io/v1beta1 metadata: name: node1-client-cert-renewal subjects: - kind: User name: system:node:node-1 # Let \"node-1\" renew its client certificate. apiGroup: rbac.authorization.k8s.io roleRef: kind: ClusterRole name: approve-node-client-renewal-csr apiGroup: rbac.authorization.k8s.io 删除该绑定将会阻止节点更新客户端凭据，一旦其证书到期，实际上就会将其从集群中删除。 kubelet 配置 要向 kube-apiserver 请求客户端证书，kubelet 首先需要一个包含 bootstrap 身份验证 token 的 kubeconfig 文件路径。您可以使用 kubectl config set-cluster，set-credentials 和 set-context 来构建此 kubeconfig 文件。为 kubectl config set-credentials 提供 kubelet-bootstrap 的名称，并包含 --token = ，如下所示： kubectl config set-credentials kubelet-bootstrap --token=${BOOTSTRAP_TOKEN} --kubeconfig=bootstrap.kubeconfig 启动 kubelet 时，如果 --kubeconfig 指定的文件不存在，则使用 bootstrap kubeconfig 向 API server 请求客户端证书。在批准 kubelet 的证书请求和回执时，将包含了生成的密钥和证书的 kubeconfig 文件写入由 -kubeconfig 指定的路径。证书和密钥文件将被放置在由 --cert-dir 指定的目录中。 启动 kubelet 时启用 bootstrap 用到的标志： --experimental-bootstrap-kubeconfig=\"/path/to/bootstrap/kubeconfig\" 此外，在1.7中，kubelet 实现了 Alpha 功能，使其客户端和/或服务器都能轮转提供证书。 可以分别通过 kubelet 中的 RotateKubeletClientCertificate 和 RotateKubeletServerCertificate 功能标志启用此功能，但在未来版本中可能会以向后兼容的方式发生变化。 --feature-gates=RotateKubeletClientCertificate=true,RotateKubeletServerCertificate=true RotateKubeletClientCertificate 可以让 kubelet 在其现有凭据到期时通过创建新的 CSR 来轮换其客户端证书。 RotateKubeletServerCertificate 可以让 kubelet 在其引导客户端凭据后还可以请求服务证书，并轮换该证书。服务证书目前不要求 DNS 或 IP SANs。 kubectl 审批 签名控制器不会立即签署所有证书请求。相反，它会一直等待直到适当特权的用户被标记为 “已批准” 状态。这最终将是由外部审批控制器来处理的自动化过程，但是对于 alpha 版本的 API 来说，可以由集群管理员通过 kubectl 命令手动完成。 管理员可以使用 kubectl get csr 命令列出所有的 CSR，使用 kubectl describe csr 命令描述某个 CSR的详细信息。在 1.6 版本以前，没有直接的批准/拒绝命令 ，因此审批者需要直接更新 Status 信息（查看如何实现）。此后的 Kubernetes 版本中提供了 kubectl certificate approve 和 kubectl certificate deny 命令。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-14 13:44:25 "},"guide/kubectl-user-authentication-authorization.html":{"url":"guide/kubectl-user-authentication-authorization.html","title":"创建用户认证授权的kubeconfig文件","keywords":"","body":"创建用户认证授权的kubeconfig文件 当我们安装好集群后，如果想要把 kubectl 命令交给用户使用，就不得不对用户的身份进行认证和对其权限做出限制。 下面以创建一个 devuser 用户并将其绑定到 dev 和 test 两个 namespace 为例说明。 创建 CA 证书和秘钥 创建 devuser-csr.json 文件 { \"CN\": \"devuser\", \"hosts\": [], \"key\": { \"algo\": \"rsa\", \"size\": 2048 }, \"names\": [ { \"C\": \"CN\", \"ST\": \"BeiJing\", \"L\": \"BeiJing\", \"O\": \"k8s\", \"OU\": \"System\" } ] } 生成 CA 证书和私钥 在 创建 TLS 证书和秘钥 一节中我们将生成的证书和秘钥放在了所有节点的 /etc/kubernetes/ssl 目录下，下面我们再在 master 节点上为 devuser 创建证书和秘钥，在 /etc/kubernetes/ssl 目录下执行以下命令： 执行该命令前请先确保该目录下已经包含如下文件： ca-key.pem ca.pem ca-config.json devuser-csr.json $ cfssl gencert -ca=ca.pem -ca-key=ca-key.pem -config=ca-config.json -profile=kubernetes devuser-csr.json | cfssljson -bare devuser 2017/08/31 13:31:54 [INFO] generate received request 2017/08/31 13:31:54 [INFO] received CSR 2017/08/31 13:31:54 [INFO] generating key: rsa-2048 2017/08/31 13:31:55 [INFO] encoded CSR 2017/08/31 13:31:55 [INFO] signed certificate with serial number 43372632012323103879829229080989286813242051309 2017/08/31 13:31:55 [WARNING] This certificate lacks a \"hosts\" field. This makes it unsuitable for websites. For more information see the Baseline Requirements for the Issuance and Management of Publicly-Trusted Certificates, v.1.1.6, from the CA/Browser Forum (https://cabforum.org); specifically, section 10.2.3 (\"Information Requirements\"). 这将生成如下文件： devuser.csr devuser-key.pem devuser.pem 创建 kubeconfig 文件 # 设置集群参数 export KUBE_APISERVER=\"https://172.20.0.113:6443\" kubectl config set-cluster kubernetes \\ --certificate-authority=/etc/kubernetes/ssl/ca.pem \\ --embed-certs=true \\ --server=${KUBE_APISERVER} \\ --kubeconfig=devuser.kubeconfig # 设置客户端认证参数 kubectl config set-credentials devuser \\ --client-certificate=/etc/kubernetes/ssl/devuser.pem \\ --client-key=/etc/kubernetes/ssl/devuser-key.pem \\ --embed-certs=true \\ --kubeconfig=devuser.kubeconfig # 设置上下文参数 kubectl config set-context kubernetes \\ --cluster=kubernetes \\ --user=devuser \\ --namespace=dev \\ --kubeconfig=devuser.kubeconfig # 设置默认上下文 kubectl config use-context kubernetes --kubeconfig=devuser.kubeconfig 我们现在查看 kubectl 的 context： kubectl config get-contexts CURRENT NAME CLUSTER AUTHINFO NAMESPACE * kubernetes kubernetes admin default-context default-cluster default-admin 显示的用户仍然是 admin，这是因为 kubectl 使用了 $HOME/.kube/config 文件作为了默认的 context 配置，我们只需要将其用刚生成的 devuser.kubeconfig 文件替换即可。 cp -f ./devuser.kubeconfig /root/.kube/config 关于 kubeconfig 文件的更多信息请参考 使用 kubeconfig 文件配置跨集群认证。 ClusterRoleBinding 如果我们想限制 devuser 用户的行为，需要使用 RBAC 将该用户的行为限制在某个或某几个 namespace 空间范围内，例如： kubectl create rolebinding devuser-admin-binding --clusterrole=admin --user=devuser --namespace=dev kubectl create rolebinding devuser-admin-binding --clusterrole=admin --user=devuser --namespace=test 这样 devuser 用户对 dev 和 test 两个 namespace 具有完全访问权限。 让我们来验证以下，现在我们在执行： # 获取当前的 context kubectl config get-contexts CURRENT NAME CLUSTER AUTHINFO NAMESPACE * kubernetes kubernetes devuser dev * kubernetes kubernetes devuser test # 无法访问 default namespace kubectl get pods --namespace default Error from server (Forbidden): User \"devuser\" cannot list pods in the namespace \"default\". (get pods) # 默认访问的是 dev namespace，您也可以重新设置 context 让其默认访问 test namespace kubectl get pods No resources found. 现在 kubectl 命令默认使用的 context 就是 devuser 了，且该用户只能操作 dev 和 test 这两个 namespace，并拥有完全的访问权限。 可以使用我写的create-user.sh脚本来创建namespace和用户并授权，参考说明。 关于角色绑定的更多信息请参考 RBAC——基于角色的访问控制。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-30 19:17:01 "},"guide/rbac.html":{"url":"guide/rbac.html","title":"RBAC——基于角色的访问控制","keywords":"","body":"RBAC——基于角色的访问控制 以下内容是 xingzhou 对 kubernetes 官方文档的翻译，原文地址 https://k8smeetup.github.io/docs/admin/authorization/rbac/ 基于角色的访问控制（Role-Based Access Control, 即”RBAC”）使用”rbac.authorization.k8s.io” API Group实现授权决策，允许管理员通过Kubernetes API动态配置策略。 截至Kubernetes 1.6，RBAC模式处于beta版本。 要启用RBAC，请使用--authorization-mode=RBAC启动API Server。 API概述 本节将介绍RBAC API所定义的四种顶级类型。用户可以像使用其他Kubernetes API资源一样 （例如通过kubectl、API调用等）与这些资源进行交互。例如，命令kubectl create -f (resource).yml 可以被用于以下所有的例子，当然，读者在尝试前可能需要先阅读以下相关章节的内容。 Role与ClusterRole 在RBAC API中，一个角色包含了一套表示一组权限的规则。 权限以纯粹的累加形式累积（没有”否定”的规则）。 角色可以由命名空间（namespace）内的Role对象定义，而整个Kubernetes集群范围内有效的角色则通过ClusterRole对象实现。 一个Role对象只能用于授予对某一单一命名空间中资源的访问权限。 以下示例描述了”default”命名空间中的一个Role对象的定义，用于授予对pod的读访问权限： kind: Role apiVersion: rbac.authorization.k8s.io/v1beta1 metadata: namespace: default name: pod-reader rules: - apiGroups: [\"\"] # 空字符串\"\"表明使用core API group resources: [\"pods\"] verbs: [\"get\", \"watch\", \"list\"] ClusterRole对象可以授予与Role对象相同的权限，但由于它们属于集群范围对象， 也可以使用它们授予对以下几种资源的访问权限： 集群范围资源（例如节点，即node） 非资源类型endpoint（例如”/healthz”） 跨所有命名空间的命名空间范围资源（例如pod，需要运行命令kubectl get pods --all-namespaces来查询集群中所有的pod） 下面示例中的ClusterRole定义可用于授予用户对某一特定命名空间，或者所有命名空间中的secret（取决于其绑定方式）的读访问权限： kind: ClusterRole apiVersion: rbac.authorization.k8s.io/v1beta1 metadata: # 鉴于ClusterRole是集群范围对象，所以这里不需要定义\"namespace\"字段 name: secret-reader rules: - apiGroups: [\"\"] resources: [\"secrets\"] verbs: [\"get\", \"watch\", \"list\"] RoleBinding与ClusterRoleBinding 角色绑定将一个角色中定义的各种权限授予一个或者一组用户。 角色绑定包含了一组相关主体（即subject, 包括用户——User、用户组——Group、或者服务账户——Service Account）以及对被授予角色的引用。 在命名空间中可以通过RoleBinding对象授予权限，而集群范围的权限授予则通过ClusterRoleBinding对象完成。 RoleBinding可以引用在同一命名空间内定义的Role对象。 下面示例中定义的RoleBinding对象在”default”命名空间中将”pod-reader”角色授予用户”jane”。 这一授权将允许用户”jane”从”default”命名空间中读取pod。 # 以下角色绑定定义将允许用户\"jane\"从\"default\"命名空间中读取pod。 kind: RoleBinding apiVersion: rbac.authorization.k8s.io/v1beta1 metadata: name: read-pods namespace: default subjects: - kind: User name: jane apiGroup: rbac.authorization.k8s.io roleRef: kind: Role name: pod-reader apiGroup: rbac.authorization.k8s.io RoleBinding对象也可以引用一个ClusterRole对象用于在RoleBinding所在的命名空间内授予用户对所引用的ClusterRole中 定义的命名空间资源的访问权限。这一点允许管理员在整个集群范围内首先定义一组通用的角色，然后再在不同的命名空间中复用这些角色。 例如，尽管下面示例中的RoleBinding引用的是一个ClusterRole对象，但是用户”dave”（即角色绑定主体）还是只能读取”development” 命名空间中的secret（即RoleBinding所在的命名空间）。 # 以下角色绑定允许用户\"dave\"读取\"development\"命名空间中的secret。 kind: RoleBinding apiVersion: rbac.authorization.k8s.io/v1beta1 metadata: name: read-secrets namespace: development # 这里表明仅授权读取\"development\"命名空间中的资源。 subjects: - kind: User name: dave apiGroup: rbac.authorization.k8s.io roleRef: kind: ClusterRole name: secret-reader apiGroup: rbac.authorization.k8s.io 最后，可以使用ClusterRoleBinding在集群级别和所有命名空间中授予权限。下面示例中所定义的ClusterRoleBinding 允许在用户组”manager”中的任何用户都可以读取集群中任何命名空间中的secret。 # 以下`ClusterRoleBinding`对象允许在用户组\"manager\"中的任何用户都可以读取集群中任何命名空间中的secret。 kind: ClusterRoleBinding apiVersion: rbac.authorization.k8s.io/v1beta1 metadata: name: read-secrets-global subjects: - kind: Group name: manager apiGroup: rbac.authorization.k8s.io roleRef: kind: ClusterRole name: secret-reader apiGroup: rbac.authorization.k8s.io 对资源的引用 大多数资源由代表其名字的字符串表示，例如”pods”，就像它们出现在相关API endpoint的URL中一样。然而，有一些Kubernetes API还 包含了”子资源”，比如pod的logs。在Kubernetes中，pod logs endpoint的URL格式为： GET /api/v1/namespaces/{namespace}/pods/{name}/log 在这种情况下，”pods”是命名空间资源，而”log”是pods的子资源。为了在RBAC角色中表示出这一点，我们需要使用斜线来划分资源 与子资源。如果需要角色绑定主体读取pods以及pod log，您需要定义以下角色： kind: Role apiVersion: rbac.authorization.k8s.io/v1beta1 metadata: namespace: default name: pod-and-pod-logs-reader rules: - apiGroups: [\"\"] resources: [\"pods\", \"pods/log\"] verbs: [\"get\", \"list\"] 通过resourceNames列表，角色可以针对不同种类的请求根据资源名引用资源实例。当指定了resourceNames列表时，不同动作 种类的请求的权限，如使用”get”、”delete”、”update”以及”patch”等动词的请求，将被限定到资源列表中所包含的资源实例上。 例如，如果需要限定一个角色绑定主体只能”get”或者”update”一个configmap时，您可以定义以下角色： kind: Role apiVersion: rbac.authorization.k8s.io/v1beta1 metadata: namespace: default name: configmap-updater rules: - apiGroups: [\"\"] resources: [\"configmap\"] resourceNames: [\"my-configmap\"] verbs: [\"update\", \"get\"] 值得注意的是，如果设置了resourceNames，则请求所使用的动词不能是list、watch、create或者deletecollection。 由于资源名不会出现在create、list、watch和deletecollection等API请求的URL中，所以这些请求动词不会被设置了resourceNames 的规则所允许，因为规则中的resourceNames部分不会匹配这些请求。 一些角色定义的例子 在以下示例中，我们仅截取展示了rules部分的定义。 允许读取core API Group中定义的资源”pods”： rules: - apiGroups: [\"\"] resources: [\"pods\"] verbs: [\"get\", \"list\", \"watch\"] 允许读写在”extensions”和”apps” API Group中定义的”deployments”： rules: - apiGroups: [\"extensions\", \"apps\"] resources: [\"deployments\"] verbs: [\"get\", \"list\", \"watch\", \"create\", \"update\", \"patch\", \"delete\"] 允许读取”pods”以及读写”jobs”： rules: - apiGroups: [\"\"] resources: [\"pods\"] verbs: [\"get\", \"list\", \"watch\"] - apiGroups: [\"batch\", \"extensions\"] resources: [\"jobs\"] verbs: [\"get\", \"list\", \"watch\", \"create\", \"update\", \"patch\", \"delete\"] 允许读取一个名为”my-config”的ConfigMap实例（需要将其通过RoleBinding绑定从而限制针对某一个命名空间中定义的一个ConfigMap实例的访问）： rules: - apiGroups: [\"\"] resources: [\"configmaps\"] resourceNames: [\"my-config\"] verbs: [\"get\"] 允许读取core API Group中的”nodes”资源（由于Node是集群级别资源，所以此ClusterRole定义需要与一个ClusterRoleBinding绑定才能有效）： rules: - apiGroups: [\"\"] resources: [\"nodes\"] verbs: [\"get\", \"list\", \"watch\"] 允许对非资源endpoint “/healthz”及其所有子路径的”GET”和”POST”请求（此ClusterRole定义需要与一个ClusterRoleBinding绑定才能有效）： rules: - nonResourceURLs: [\"/healthz\", \"/healthz/*\"] # 在非资源URL中，'*'代表后缀通配符 verbs: [\"get\", \"post\"] 对角色绑定主体（Subject）的引用 RoleBinding或者ClusterRoleBinding将角色绑定到角色绑定主体（Subject）。 角色绑定主体可以是用户组（Group）、用户（User）或者服务账户（Service Accounts）。 用户由字符串表示。可以是纯粹的用户名，例如”alice”、电子邮件风格的名字，如 “bob@example.com” 或者是用字符串表示的数字id。由Kubernetes管理员配置认证模块 以产生所需格式的用户名。对于用户名，RBAC授权系统不要求任何特定的格式。然而，前缀system:是 为Kubernetes系统使用而保留的，所以管理员应该确保用户名不会意外地包含这个前缀。 Kubernetes中的用户组信息由授权模块提供。用户组与用户一样由字符串表示。Kubernetes对用户组 字符串没有格式要求，但前缀system:同样是被系统保留的。 服务账户拥有包含 system:serviceaccount:前缀的用户名，并属于拥有system:serviceaccounts:前缀的用户组。 角色绑定的一些例子 以下示例中，仅截取展示了RoleBinding的subjects字段。 一个名为”alice@example.com”的用户： subjects: - kind: User name: \"alice@example.com\" apiGroup: rbac.authorization.k8s.io 一个名为”frontend-admins”的用户组： subjects: - kind: Group name: \"frontend-admins\" apiGroup: rbac.authorization.k8s.io kube-system命名空间中的默认服务账户： subjects: - kind: ServiceAccount name: default namespace: kube-system 名为”qa”命名空间中的所有服务账户： subjects: - kind: Group name: system:serviceaccounts:qa apiGroup: rbac.authorization.k8s.io 在集群中的所有服务账户： subjects: - kind: Group name: system:serviceaccounts apiGroup: rbac.authorization.k8s.io 所有认证过的用户（version 1.5+）： subjects: - kind: Group name: system:authenticated apiGroup: rbac.authorization.k8s.io 所有未认证的用户（version 1.5+）： subjects: - kind: Group name: system:unauthenticated apiGroup: rbac.authorization.k8s.io 所有用户（version 1.5+）： subjects: - kind: Group name: system:authenticated apiGroup: rbac.authorization.k8s.io - kind: Group name: system:unauthenticated apiGroup: rbac.authorization.k8s.io 默认角色与默认角色绑定 API Server会创建一组默认的ClusterRole和ClusterRoleBinding对象。 这些默认对象中有许多包含system:前缀，表明这些资源由Kubernetes基础组件”拥有”。 对这些资源的修改可能导致非功能性集群（non-functional cluster）。一个例子是system:node ClusterRole对象。 这个角色定义了kubelets的权限。如果这个角色被修改，可能会导致kubelets无法正常工作。 所有默认的ClusterRole和ClusterRoleBinding对象都会被标记为kubernetes.io/bootstrapping=rbac-defaults。 自动更新 每次启动时，API Server都会更新默认ClusterRole所缺乏的各种权限，并更新默认ClusterRoleBinding所缺乏的各个角色绑定主体。 这种自动更新机制允许集群修复一些意外的修改。由于权限和角色绑定主体在新的Kubernetes释出版本中可能变化，这也能够保证角色和角色 绑定始终保持是最新的。 如果需要禁用自动更新，请将默认ClusterRole以及ClusterRoleBinding的rbac.authorization.kubernetes.io/autoupdate 设置成为false。 请注意，缺乏默认权限和角色绑定主体可能会导致非功能性集群问题。 自Kubernetes 1.6+起，当集群RBAC授权器（RBAC Authorizer）处于开启状态时，可以启用自动更新功能. 发现类角色 默认ClusterRole 默认ClusterRoleBinding 描述 system:basic-user system:authenticated and system:unauthenticatedgroups 允许用户只读访问有关自己的基本信息。 system:discovery system:authenticated and system:unauthenticatedgroups 允许只读访问API discovery endpoints, 用于在API级别进行发现和协商。 面向用户的角色 一些默认角色并不包含system:前缀，它们是面向用户的角色。 这些角色包含超级用户角色（cluster-admin），即旨在利用ClusterRoleBinding（cluster-status）在集群范围内授权的角色， 以及那些使用RoleBinding（admin、edit和view）在特定命名空间中授权的角色。 默认ClusterRole 默认ClusterRoleBinding 描述 cluster-admin system:masters group 超级用户权限，允许对任何资源执行任何操作。 在ClusterRoleBinding中使用时，可以完全控制集群和所有命名空间中的所有资源。 在RoleBinding中使用时，可以完全控制RoleBinding所在命名空间中的所有资源，包括命名空间自己。 admin None 管理员权限，利用RoleBinding在某一命名空间内部授予。 在RoleBinding中使用时，允许针对命名空间内大部分资源的读写访问， 包括在命名空间内创建角色与角色绑定的能力。 但不允许对资源配额（resource quota）或者命名空间本身的写访问。 edit None 允许对某一个命名空间内大部分对象的读写访问，但不允许查看或者修改角色或者角色绑定。 view None 允许对某一个命名空间内大部分对象的只读访问。 不允许查看角色或者角色绑定。 由于可扩散性等原因，不允许查看secret资源。 Core Component Roles 核心组件角色 默认ClusterRole 默认ClusterRoleBinding 描述 system:kube-scheduler system:kube-scheduler user 允许访问kube-scheduler组件所需要的资源。 system:kube-controller-manager system:kube-controller-manager user 允许访问kube-controller-manager组件所需要的资源。 单个控制循环所需要的权限请参阅控制器（controller）角色. system:node system:nodes group (deprecated in 1.7) 允许对kubelet组件所需要的资源的访问，包括读取所有secret和对所有pod的写访问。 自Kubernetes 1.7开始, 相比较于这个角色，更推荐使用Node authorizer 以及NodeRestriction admission plugin， 并允许根据调度运行在节点上的pod授予kubelets API访问的权限。 自Kubernetes 1.7开始，当启用Node授权模式时，对system:nodes用户组的绑定将不会被自动创建。 system:node-proxier system:kube-proxy user 允许对kube-proxy组件所需要资源的访问。 其它组件角色 默认ClusterRole 默认ClusterRoleBinding 描述 system:auth-delegator None 允许委托认证和授权检查。 通常由附加API Server用于统一认证和授权。 system:heapster None Heapster组件的角色。 system:kube-aggregator None kube-aggregator组件的角色。 system:kube-dns kube-dns service account in the kube-systemnamespace kube-dns组件的角色。 system:node-bootstrapper None 允许对执行Kubelet TLS引导（Kubelet TLS bootstrapping）所需要资源的访问. system:node-problem-detector None node-problem-detector组件的角色。 system:persistent-volume-provisioner None 允许对大部分动态存储卷创建组件（dynamic volume provisioner）所需要资源的访问。 控制器（Controller）角色 Kubernetes controller manager负责运行核心控制循环。 当使用--use-service-account-credentials选项运行controller manager时，每个控制循环都将使用单独的服务账户启动。 而每个控制循环都存在对应的角色，前缀名为system:controller:。 如果不使用--use-service-account-credentials选项时，controller manager将会使用自己的凭证运行所有控制循环，而这些凭证必须被授予相关的角色。 这些角色包括： system:controller:attachdetach-controller system:controller:certificate-controller system:controller:cronjob-controller system:controller:daemon-set-controller system:controller:deployment-controller system:controller:disruption-controller system:controller:endpoint-controller system:controller:generic-garbage-collector system:controller:horizontal-pod-autoscaler system:controller:job-controller system:controller:namespace-controller system:controller:node-controller system:controller:persistent-volume-binder system:controller:pod-garbage-collector system:controller:replicaset-controller system:controller:replication-controller system:controller:resourcequota-controller system:controller:route-controller system:controller:service-account-controller system:controller:service-controller system:controller:statefulset-controller system:controller:ttl-controller 初始化与预防权限升级 RBAC API会阻止用户通过编辑角色或者角色绑定来升级权限。 由于这一点是在API级别实现的，所以在RBAC授权器（RBAC authorizer）未启用的状态下依然可以正常工作。 用户只有在拥有了角色所包含的所有权限的条件下才能创建／更新一个角色，这些操作还必须在角色所处的相同范围内进行（对于ClusterRole来说是集群范围，对于Role来说是在与角色相同的命名空间或者集群范围）。 例如，如果用户”user-1”没有权限读取集群范围内的secret列表，那么他也不能创建包含这种权限的ClusterRole。为了能够让用户创建／更新角色，需要： 授予用户一个角色以允许他们根据需要创建／更新Role或者ClusterRole对象。 授予用户一个角色包含他们在Role或者ClusterRole中所能够设置的所有权限。如果用户尝试创建或者修改Role或者ClusterRole以设置那些他们未被授权的权限时，这些API请求将被禁止。 用户只有在拥有所引用的角色中包含的所有权限时才可以创建／更新角色绑定（这些操作也必须在角色绑定所处的相同范围内进行）或者用户被明确授权可以在所引用的角色上执行绑定操作。 例如，如果用户”user-1”没有权限读取集群范围内的secret列表，那么他将不能创建ClusterRole来引用那些授予了此项权限的角色。为了能够让用户创建／更新角色绑定，需要： 授予用户一个角色以允许他们根据需要创建／更新RoleBinding或者ClusterRoleBinding对象。 授予用户绑定某一特定角色所需要的权限： 隐式地，通过授予用户所有所引用的角色中所包含的权限 显式地，通过授予用户在特定Role（或者ClusterRole）对象上执行bind操作的权限 例如，下面例子中的ClusterRole和RoleBinding将允许用户”user-1”授予其它用户”user-1-namespace”命名空间内的admin、edit和view等角色和角色绑定。 apiVersion: rbac.authorization.k8s.io/v1beta1 kind: ClusterRole metadata: name: role-grantor rules: - apiGroups: [\"rbac.authorization.k8s.io\"] resources: [\"rolebindings\"] verbs: [\"create\"] - apiGroups: [\"rbac.authorization.k8s.io\"] resources: [\"clusterroles\"] verbs: [\"bind\"] resourceNames: [\"admin\",\"edit\",\"view\"] --- apiVersion: rbac.authorization.k8s.io/v1beta1 kind: RoleBinding metadata: name: role-grantor-binding namespace: user-1-namespace roleRef: apiGroup: rbac.authorization.k8s.io kind: ClusterRole name: role-grantor subjects: - apiGroup: rbac.authorization.k8s.io kind: User name: user-1 当初始化第一个角色和角色绑定时，初始用户需要能够授予他们尚未拥有的权限。 初始化初始角色和角色绑定时需要： 使用包含system：masters用户组的凭证，该用户组通过默认绑定绑定到cluster-admin超级用户角色。 如果您的API Server在运行时启用了非安全端口（--insecure-port），您也可以通过这个没有施行认证或者授权的端口发送角色或者角色绑定请求。 一些命令行工具 有两个kubectl命令可以用于在命名空间内或者整个集群内授予角色。 kubectl create rolebinding 在某一特定命名空间内授予Role或者ClusterRole。示例如下： 在名为”acme”的命名空间中将admin ClusterRole授予用户”bob”： kubectl create rolebinding bob-admin-binding --clusterrole=admin --user=bob --namespace=acme 在名为”acme”的命名空间中将view ClusterRole授予服务账户”myapp”： kubectl create rolebinding myapp-view-binding --clusterrole=view --serviceaccount=acme:myapp --namespace=acme kubectl create clusterrolebinding 在整个集群中授予ClusterRole，包括所有命名空间。示例如下： 在整个集群范围内将cluster-admin ClusterRole授予用户”root”： kubectl create clusterrolebinding root-cluster-admin-binding --clusterrole=cluster-admin --user=root 在整个集群范围内将system:node ClusterRole授予用户”kubelet”： kubectl create clusterrolebinding kubelet-node-binding --clusterrole=system:node --user=kubelet 在整个集群范围内将view ClusterRole授予命名空间”acme”内的服务账户”myapp”： kubectl create clusterrolebinding myapp-view-binding --clusterrole=view --serviceaccount=acme:myapp 请参阅CLI帮助文档以获得上述命令的详细用法 服务账户（Service Account）权限 默认的RBAC策略将授予控制平面组件（control-plane component）、节点（node）和控制器（controller）一组范围受限的权限， 但对于”kube-system”命名空间以外的服务账户，则不授予任何权限（超出授予所有认证用户的发现权限）。 这一点允许您根据需要向特定服务账号授予特定权限。 细粒度的角色绑定将提供更好的安全性，但需要更多精力管理。 更粗粒度的授权可能授予服务账号不需要的API访问权限（甚至导致潜在授权扩散），但更易于管理。 从最安全到最不安全可以排序以下方法： 对某一特定应用程序的服务账户授予角色（最佳实践） 要求应用程序在其pod规范（pod spec）中指定serviceAccountName字段，并且要创建相应服务账户（例如通过API、应用程序清单或者命令kubectl create serviceaccount等）。 例如，在”my-namespace”命名空间中授予服务账户”my-sa”只读权限： kubectl create rolebinding my-sa-view \\ --clusterrole=view \\ --serviceaccount=my-namespace:my-sa \\ --namespace=my-namespace 在某一命名空间中授予”default”服务账号一个角色 如果一个应用程序没有在其pod规范中指定serviceAccountName，它将默认使用”default”服务账号。 注意：授予”default”服务账号的权限将可用于命名空间内任何没有指定serviceAccountName的pod。 下面的例子将在”my-namespace”命名空间内授予”default”服务账号只读权限： kubectl create rolebinding default-view \\ --clusterrole=view \\ --serviceaccount=my-namespace:default \\ --namespace=my-namespace 目前，许多[加载项（addon）]（/ docs / concepts / cluster-administration / addons /）作为”kube-system”命名空间中的”default”服务帐户运行。 要允许这些加载项使用超级用户访问权限，请将cluster-admin权限授予”kube-system”命名空间中的”default”服务帐户。 注意：启用上述操作意味着”kube-system”命名空间将包含允许超级用户访问API的秘钥。 kubectl create clusterrolebinding add-on-cluster-admin \\ --clusterrole=cluster-admin \\ --serviceaccount=kube-system:default 为命名空间中所有的服务账号授予角色 如果您希望命名空间内的所有应用程序都拥有同一个角色，无论它们使用什么服务账户，您可以为该命名空间的服务账户用户组授予角色。 下面的例子将授予”my-namespace”命名空间中的所有服务账户只读权限： kubectl create rolebinding serviceaccounts-view \\ --clusterrole=view \\ --group=system:serviceaccounts:my-namespace \\ --namespace=my-namespace 对集群范围内的所有服务账户授予一个受限角色（不鼓励） 如果您不想管理每个命名空间的权限，则可以将集群范围角色授予所有服务帐户。 下面的例子将所有命名空间中的只读权限授予集群中的所有服务账户： kubectl create clusterrolebinding serviceaccounts-view \\ --clusterrole=view \\ --group=system:serviceaccounts 授予超级用户访问权限给集群范围内的所有服务帐户（强烈不鼓励） 如果您根本不关心权限分块，您可以对所有服务账户授予超级用户访问权限。 警告：这种做法将允许任何具有读取权限的用户访问secret或者通过创建一个容器的方式来访问超级用户的凭据。 kubectl create clusterrolebinding serviceaccounts-cluster-admin \\ --clusterrole=cluster-admin \\ --group=system:serviceaccounts 从版本1.5升级 在Kubernetes 1.6之前，许多部署使用非常宽泛的ABAC策略，包括授予对所有服务帐户的完整API访问权限。 默认的RBAC策略将授予控制平面组件（control-plane components）、节点（nodes）和控制器（controller）一组范围受限的权限， 但对于”kube-system”命名空间以外的服务账户，则不授予任何权限（超出授予所有认证用户的发现权限）。 虽然安全性更高，但这可能会影响到期望自动接收API权限的现有工作负载。 以下是管理此转换的两种方法： 并行授权器（authorizer） 同时运行RBAC和ABAC授权器，并包括旧版ABAC策略： --authorization-mode=RBAC,ABAC --authorization-policy-file=mypolicy.jsonl RBAC授权器将尝试首先授权请求。如果RBAC授权器拒绝API请求，则ABAC授权器将被运行。这意味着RBAC策略或者ABAC策略所允许的任何请求都是可通过的。 当以日志级别为2或更高（--v = 2）运行时，您可以在API Server日志中看到RBAC拒绝请求信息（以RBAC DENY:为前缀）。 您可以使用该信息来确定哪些角色需要授予哪些用户，用户组或服务帐户。 一旦授予服务帐户角色，并且服务器日志中没有RBAC拒绝消息的工作负载正在运行，您可以删除ABAC授权器。 宽泛的RBAC权限 您可以使用RBAC角色绑定来复制一个宽泛的策略。 警告：以下政策略允许所有服务帐户作为集群管理员。 运行在容器中的任何应用程序都会自动接收服务帐户凭据，并且可以对API执行任何操作，包括查看secret和修改权限。 因此，并不推荐使用这种策略。 kubectl create clusterrolebinding permissive-binding \\ --clusterrole=cluster-admin \\ --user=admin \\ --user=kubelet \\ --group=system:serviceaccounts Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-07 10:23:27 "},"guide/ip-masq-agent.html":{"url":"guide/ip-masq-agent.html","title":"IP伪装代理","keywords":"","body":"IP 伪装代理 本文将讲述如何配置和启用 ip-masq-agent。 创建 ip-masq-agent 要创建 ip-masq-agent，运行下面的 kubectl 命令： kubectl create -f https://raw.githubusercontent.com/kubernetes-incubator/ip-masq-agent/master/ip-masq-agent.yaml 关于 ip-masq-agent 的更多信息请参考 该文档。 在大多数情况下，默认的一套规则应该是足够的；但是，如果内置的规则不适用于您的集群，您可以创建并应用 ConfigMap 来自定义受影响的 IP 范围。例如，为了仅允许 ip-masq-agent 考虑 10.0.0.0/8，您可以在名为 “config” 的文件中创建以下 ConfigMap。 nonMasqueradeCIDRs: - 10.0.0.0/8 resyncInterval: 60s 注意：重要的是，该文件被命名为 config，因为默认情况下，该文件将被用作 ip-masq-agent 查找的关键字。 运行下列命令将 ConfigMap 添加到您的集群中： kubectl create configmap ip-masq-agent --from-file=config --namespace=kube-system 这将会更新 /etc/config/ip-masq-agent 文件，并每隔 resyscInterval 时间段检查一遍该文件，将配置应用到集群的节点中。 iptables -t nat -L IP-MASQ-AGENT Chain IP-MASQ-AGENT (1 references) target prot opt source destination RETURN all -- anywhere 169.254.0.0/16 /* ip-masq-agent: cluster-local traffic should not be subject to MASQUERADE */ ADDRTYPE match dst-type !LOCAL RETURN all -- anywhere 10.0.0.0/8 /* ip-masq-agent: cluster-local MASQUERADE all -- anywhere anywhere /* ip-masq-agent: outbound traffic should be subject to MASQUERADE (this match must come after cluster-local CIDR matches) */ ADDRTYPE match dst-type !LOCAL 默认情况下，本地链路范围（169.254.0.0/16）也由 ip-masq 代理处理，该代理设置相应的 iptables 规则。想要让 ip-masq-agent 忽略本地链路，您可以在 ConfigMap 中将 masqLinkLocal 设置为 true。 nonMasqueradeCIDRs: - 10.0.0.0/8 resyncInterval: 60s masqLinkLocal: true IP 伪装代理用户指南 ip-masq-agent 用户配置 iptables 规则将 Pod 的 IP 地址隐藏在集群 node 节点的 IP 地址后面。这通常在将流量发送到群集的 pod CIDR 范围之外的目的地时执行。 关键术语 NAT（网络地址转换） 是一种通过修改 IP 头中的源和/或目标地址信息来将一个 IP 地址重映射到另一个 IP 地址的方法。通常由执行 IP 路由的设备完成。 Masquerading（伪装） NAT 的一种形式，通常用于执行多个地址转换，其中多个源 IP 地址被掩盖在单个地址之后，通常是由某设备进行 IP 路由。在 kubernetes 中，这是 Node 的 IP 地址。 CIDR（无类域内路由选择） 基于可变长度子网掩码，允许指定任意长度的前缀。CIDR 引入了一种新的 IP 地址表示方法，现在通常被称为 CIDR 表示法，将地址或路由前缀的比特位数作为后缀，例如 192.168.2.0/24。 本地链路 本地链路地址是仅能在主机连接的网段或广播域内进行有效通信的网络地址。IPv4 的链路本地地址在 CIDR 表示法定义的地址块是 169.254.0.0/16。 Ip-masq-agent 在将流量发送到集群 node 节点的 IP 和 Cluster IP 范围之外的目的地时，会配置 iptables 规则来处理伪装的 node/pod IP 地址。这基本上将 pod 的 IP 地址隐藏在了集群 node 节点的 IP 地址后面。在某些环境中，到 “外部” 地址的流量必须来自已知的机器地址。例如，在 Google Cloud 中，到互联网的任何流量必须来自虚拟机的 IP。当使用容器时，如在GKE中，Pod IP 将被拒绝作为出口。为了避免这种情况，我们必须将 Pod IP 隐藏在 VM 自己的 IP 地址之后——通常被称为 “伪装”。默认情况下，配置代理将 RFC 1918指定的三个专用 IP 范围视为非伪装 CIDR。范围包括 10.0.0.0/8、172.16.0.0/12 和 192.168.0.0/16。默认情况下，代理还将本地链路（169.254.0.0/16）视为非伪装 CIDR。代理配置为每隔60秒从 /etc/config/ip-masq-agent 位置重新加载其配置，这也是可配置的。 图片 - IP伪装代理示意图 代理的配置文件必须使用 yaml 或 json 语法，并且包含以下三个可选的 key： nonMasqueradeCIDRs：使用 CIDR 表示法指定的非伪装范围的字符串列表。 masqLinkLocal：一个布尔值（true/false），表示是否将流量伪装成本地链路前缀 169.254.0.0/16。默认为false。 resyncInterval：代理尝试从磁盘重新加载配置的时间间隔。例如 ’30s’ 其中 ‘s’ 是秒，’ms’ 是毫秒等… 到 10.0.0.0/8、172.16.0.0/12 和 192.168.0.0/16 范围的流量将不会被伪装。任何其他流量（假定是互联网）将被伪装。这里有个例子，来自 pod 的本地目的地址可以是其节点的 IP 地址、其他节点的地址或 Cluster IP 范围中的一个 IP 地址。其他任何流量都将默认伪装。以下条目显示 ip-masq-agent 应用的默认规则集： iptables -t nat -L IP-MASQ-AGENT RETURN all -- anywhere 169.254.0.0/16 /* ip-masq-agent: cluster-local traffic should not be subject to MASQUERADE */ ADDRTYPE match dst-type !LOCAL RETURN all -- anywhere 10.0.0.0/8 /* ip-masq-agent: cluster-local traffic should not be subject to MASQUERADE */ ADDRTYPE match dst-type !LOCAL RETURN all -- anywhere 172.16.0.0/12 /* ip-masq-agent: cluster-local traffic should not be subject to MASQUERADE */ ADDRTYPE match dst-type !LOCAL RETURN all -- anywhere 192.168.0.0/16 /* ip-masq-agent: cluster-local traffic should not be subject to MASQUERADE */ ADDRTYPE match dst-type !LOCAL MASQUERADE all -- anywhere anywhere /* ip-masq-agent: outbound traffic should be subject to MASQUERADE (this match must come after cluster-local CIDR matches) */ ADDRTYPE match dst-type !LOCAL 默认情况下，在 GCE/GKE 中将启动 kubernetes 1.7.0 版本，ip-masq-agent 已经在集群中运行。如果您在其他环境中运行 kubernetes，那么您可以将 ip-masq-agent 以 DaemonSet 的方式在集群中运行。 原文地址：https://k8smeetup.github.io/docs/tasks/administer-cluster/ip-masq-agent/ 译者：rootsongjc Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-09-07 14:09:13 "},"guide/auth-with-kubeconfig-or-token.html":{"url":"guide/auth-with-kubeconfig-or-token.html","title":"使用kubeconfig或token进行用户身份认证","keywords":"","body":"使用 kubeconfig 或 token 进行用户身份认证 在开启了 TLS 的集群中，每当与集群交互的时候少不了的是身份认证，使用 kubeconfig（即证书） 和 token 两种认证方式是最简单也最通用的认证方式，在 dashboard 的登录功能就可以使用这两种登录功能。 下文分两块以示例的方式来讲解两种登陆认证方式： 为 brand 命名空间下的 brand 用户创建 kubeconfig 文件 为集群的管理员（拥有所有命名空间的 amdin 权限）创建 token 使用 kubeconfig 如何生成kubeconfig文件请参考创建用户认证授权的kubeconfig文件。 注意我们生成的 kubeconfig 文件中没有 token 字段，需要手动添加该字段。 比如我们为 brand namespace 下的 brand 用户生成了名为 brand.kubeconfig 的 kubeconfig 文件，还要再该文件中追加一行 token 的配置（如何生成 token 将在下文介绍），如下所示： 图片 - kubeconfig文件 对于访问 dashboard 时候的使用 kubeconfig 文件如brand.kubeconfig 必须追到 token 字段，否则认证不会通过。而使用 kubectl 命令时的用的 kubeconfig 文件则不需要包含 token 字段。 生成 token 需要创建一个admin用户并授予admin角色绑定，使用下面的yaml文件创建admin用户并赋予他管理员权限，然后可以通过token访问kubernetes，该文件见admin-role.yaml。 kind: ClusterRoleBinding apiVersion: rbac.authorization.k8s.io/v1beta1 metadata: name: admin annotations: rbac.authorization.kubernetes.io/autoupdate: \"true\" roleRef: kind: ClusterRole name: cluster-admin apiGroup: rbac.authorization.k8s.io subjects: - kind: ServiceAccount name: admin namespace: kube-system --- apiVersion: v1 kind: ServiceAccount metadata: name: admin namespace: kube-system labels: kubernetes.io/cluster-service: \"true\" addonmanager.kubernetes.io/mode: Reconcile 然后执行下面的命令创建 serviceaccount 和角色绑定，对于其他命名空间的其他用户只要修改上述 yaml 中的 name 和 namespace 字段即可： kubectl create -f admin-role.yaml 创建完成后获取secret和token的值。 # 获取admin-token的secret名字 $ kubectl -n kube-system get secret|grep admin-token admin-token-nwphb kubernetes.io/service-account-token 3 6m # 获取token的值 $ kubectl -n kube-system describe secret admin-token-nwphb Name: admin-token-nwphb Namespace: kube-system Labels: Annotations: kubernetes.io/service-account.name=admin kubernetes.io/service-account.uid=f37bd044-bfb3-11e7-87c0-f4e9d49f8ed0 Type: kubernetes.io/service-account-token Data ==== namespace: 11 bytes token: 非常长的字符串 ca.crt: 1310 bytes 也可以使用 jsonpath 的方式直接获取 token 的值，如： kubectl -n kube-system get secret admin-token-nwphb -o jsonpath={.data.token}|base64 -d 注意：yaml 输出里的那个 token 值是进行 base64 编码后的结果，一定要将 kubectl 的输出中的 token 值进行 base64 解码，在线解码工具 base64decode，Linux 和 Mac 有自带的 base64 命令也可以直接使用，输入 base64 是进行编码，Linux 中base64 -d 表示解码，Mac 中使用 base64 -D。 我们使用了 base64 对其重新解码，因为 secret 都是经过 base64 编码的，如果直接使用 kubectl 中查看到的 token 值会认证失败，详见 secret 配置。关于 JSONPath 的使用请参考 JSONPath 手册。 参考 JSONPath 手册 Kubernetes 中的认证 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-11 01:40:46 "},"guide/authentication.html":{"url":"guide/authentication.html","title":" kubernetes中的用户与身份认证授权","keywords":"","body":"Kubernetes 中的用户与身份认证授权 在安装集群的时候我们在 master 节点上生成了一堆证书、token，还在 kubelet 的配置中用到了 bootstrap token，安装各种应用时，为了能够与 API server 通信创建了各种 service account，在 Dashboard 中使用了 kubeconfig 或 token 登陆，那么这些都属于什么认证方式？如何区分用户的？我特地翻译了下这篇官方文档，想你看了之后你将找到答案。 重点查看 bearer token 和 HTTP 认证中的 token 使用，我们已经有所应用，如 使用kubeconfig或token进行用户身份认证。 认识 Kubernetes 中的用户 Kubernetes 集群中包含两类用户：一类是由 Kubernetes 管理的 service account，另一类是普通用户。 普通用户被假定为由外部独立服务管理。管理员分发私钥，用户存储（如 Keystone 或 Google 帐户），甚至包含用户名和密码列表的文件。在这方面，Kubernetes 没有代表普通用户帐户的对象。无法通过 API 调用的方式向集群中添加普通用户。 相对的，service account 是由 Kubernetes API 管理的帐户。它们都绑定到了特定的 namespace，并由 API server 自动创建，或者通过 API 调用手动创建。Service account 关联了一套凭证，存储在 Secret，这些凭证同时被挂载到 pod 中，从而允许 pod 与 kubernetes API 之间的调用。 API 请求被绑定到普通用户或 serivce account 上，或者作为匿名请求对待。这意味着集群内部或外部的每个进程，无论从在工作站上输入 kubectl 的人类用户到节点上的 kubelet，到控制平面的成员，都必须在向 API Server 发出请求时进行身份验证，或者被视为匿名用户。 认证策略 Kubernetes 使用客户端证书、bearer token、身份验证代理或者 HTTP 基本身份验证等身份认证插件来对 API 请求进行身份验证。当有 HTTP 请求发送到 API server 时，插件会尝试将以下属性关联到请求上： 用户名：标识最终用户的字符串。常用值可能是 kube-admin 或 jane@example.com。 UID：标识最终用户的字符串，比用户名更加一致且唯一。 组：一组将用户和常规用户组相关联的字符串。 额外字段：包含其他有用认证信息的字符串列表的映射。 所有的值对于认证系统都是不透明的，只有 授权人 才能解释这些值的重要含义。 您可以一次性启用多种身份验证方式。通常使用至少以下两种认证方式： 服务帐户的 service account token 至少一种其他的用户认证的方式 当启用了多个认证模块时，第一个认证模块成功认证后将短路请求，不会进行第二个模块的认证。API server 不会保证认证的顺序。 system:authenticated 组包含在所有已验证用户的组列表中。 与其他身份验证协议（LDAP、SAML、Kerberos、x509 方案等）的集成可以使用 身份验证代理 或 身份验证 webhook 来实现。 X509 客户端证书 通过将 --client-ca-file=SOMEFILE 选项传递给 API server 来启用客户端证书认证。引用的文件必须包含一个或多个证书颁发机构，用于验证提交给 API server 的客户端证书。如果客户端证书已提交并验证，则使用 subject 的 Common Name（CN）作为请求的用户名。从 Kubernetes 1.4开始，客户端证书还可以使用证书的 organization 字段来指示用户的组成员身份。要为用户包含多个组成员身份，请在证书中包含多个 organization 字段。 例如，使用 openssl 命令工具生成用于签名认证请求的证书： openssl req -new -key jbeda.pem -out jbeda-csr.pem -subj \"/CN=jbeda/O=app1/O=app2\" 这将为一个用户名为 ”jbeda“ 的 CSR，属于两个组“app1”和“app2”。 如何生成客户端证书请参阅 附录。 静态 Token 文件 当在命令行上指定 --token-auth-file=SOMEFILE 选项时，API server 从文件读取 bearer token。目前，token 会无限期地持续下去，并且不重新启动 API server 的话就无法更改令牌列表。 token 文件是一个 csv 文件，每行至少包含三列：token、用户名、用户 uid，其次是可选的组名。请注意，如果您有多个组，则该列必须使用双引号。 token,user,uid,\"group1,group2,group3\" 在请求中放置 Bearer Token 当使用来自 http 客户端的 bearer token 时，API server 期望 Authorization header 中包含 Bearer token 的值。Bearer token 必须是一个字符串序列，只需使用 HTTP 的编码和引用功能就可以将其放入到 HTTP header 中。例如：如果 bearer token 是 31ada4fd-adec-460c-809a-9e56ceb75269，那么它将出现在 HTTP header 中，如下所示： Authorization: Bearer 31ada4fd-adec-460c-809a-9e56ceb75269 Bootstrap Token 该功能仍处于 alpha 版本。 为了简化新集群的初始化引导过程，Kubernetes 中包含了一个名为 Bootstrap Token 的动态管理的 bearer token。这些 token 使用 Secret 存储在 kube-system namespace 中，在那里它们可以被动态管理和创建。Controller Manager 中包含了一个 TokenCleaner 控制器，用于在 bootstrap token 过期时删除将其删除。 这些 token 的形式是 [a-z0-9]{6}.[a-z0-9]{16}。第一部分是 Token ID，第二部分是 Token Secret。您在 HTTP header 中指定的 token 如下所示： Authorization: Bearer 781292.db7bc3a58fc5f07e 在 API server 的启动参数中加上 --experimental-bootstrap-token-auth 标志以启用 Bootstrap Token Authenticator。您必须通过 Controller Manager 上的 --controllers 标志启用 TokenCleaner 控制器，如 --controllers=*,tokencleaner。如果您使用它来引导集群， kubeadm 会为您完成。 认证者认证为 system:bootstrap: 。被包含在 system:bootstrappers 组中。命名和组是有意限制用户使用过去的 bootstap token。可以使用用户名和组（kubeadm 使用）来制定适当的授权策略以支持引导集群。 有关 Bootstrap Token 身份验证器和控制器的更深入的文档，以及如何使用 kubeadm 管理这些令牌，请参阅 Bootstrap Token。 静态密码文件 通过将 --basic-auth-file=SOMEFILE 选项传递给 API server 来启用基本身份验证。目前，基本身份验证凭证将无限期地保留，并且密码在不重新启动API服务器的情况下无法更改。请注意，为了方便起见，目前支持基本身份验证，而上述模式更安全更容易使用。 基本身份认证是一个 csv 文件，至少包含3列：密码、用户名和用户 ID。在 Kubernetes 1.6 和更高版本中，可以指定包含以逗号分隔的组名称的可选第四列。如果您有多个组，则必须将第四列值用双引号（“）括起来，请参阅以下示例： password,user,uid,\"group1,group2,group3\" 当使用来自 HTTP 客户端的基本身份验证时，API server 需要Authorization header 中包含 Basic BASE64ENCODED(USER:PASSWORD) 的值。 Service Account Token Service account 是一个自动启用的验证器，它使用签名的 bearer token 来验证请求。该插件包括两个可选的标志： --service-account-key-file 一个包含签名 bearer token 的 PEM 编码文件。如果未指定，将使用 API server 的 TLS 私钥。 --service-account-lookup 如果启用，从 API 中删除掉的 token 将被撤销。 Service account 通常 API server 自动创建，并通过 ServiceAccount 注入控制器 关联到集群中运行的 Pod 上。Bearer token 挂载到 pod 中众所周知的位置，并允许集群进程与 API server 通信。 帐户可以使用 PodSpec 的 serviceAccountName 字段显式地与Pod关联。 注意： serviceAccountName 通常被省略，因为这会自动生成。 apiVersion: apps/v1beta2 kind: Deployment metadata: name: nginx-deployment namespace: default spec: replicas: 3 template: metadata: # ... spec: containers: - name: nginx image: nginx:1.7.9 serviceAccountName: bob-the-bot Service account bearer token 在集群外使用也是完全有效的，并且可以用于为希望与 Kubernetes 通信的长期运行作业创建身份。要手动创建 service account，只需要使用 kubectl create serviceaccount (NAME) 命令。这将在当前的 namespace 和相关连的 secret 中创建一个 service account。 $ kubectl create serviceaccount jenkins serviceaccount \"jenkins\" created $ kubectl get serviceaccounts jenkins -o yaml apiVersion: v1 kind: ServiceAccount metadata: # ... secrets: - name: jenkins-token-1yvwg 创建出的 secret 中拥有 API server 的公共 CA 和前面的 JSON Web Token（JWT）。 $ kubectl get secret jenkins-token-1yvwg -o yaml apiVersion: v1 data: ca.crt: (APISERVER'S CA BASE64 ENCODED) namespace: ZGVmYXVsdA== token: (BEARER TOKEN BASE64 ENCODED) kind: Secret metadata: # ... type: kubernetes.io/service-account-token 注意：所有值是基于 base64 编码的，因为 secret 总是基于 base64 编码。 经过签名的 JWT 可以用作 bearer token 与给定的 service account 进行身份验证。请参阅 上面 关于如何在请求中放置 bearer token。通常情况下，这些 secret 被挂载到 pod 中，以便对集群内的 API server 进行访问，但也可以从集群外访问。 Service account 验证时用户名 system:serviceaccount:(NAMESPACE):(SERVICEACCOUNT)，被指定到组 system:serviceaccounts 和 system:serviceaccounts:(NAMESPACE)。 注意：由于 service account 的 token 存储在 secret 中，所以具有对这些 secret 的读取权限的任何用户都可以作为 service account 进行身份验证。授予 service account 权限和读取 secret 功能时要谨慎。 OpenID Connect Token OpenID Connect 是由 OAuth2 供应商提供的 OAuth2，特别是 Azure Active Directory、Salesforce 和 Google。对 OAuth2 协议的主要扩展是返回一个称作 ID Token 的格外字段。该 token 是一个 JSON Web Token (JWT) ，有服务器签名，具有众所周知的字段，如用户的电子邮件。 为了识别用户，认证者使用 OAuth2 token 响应 中的 id_token（而不是 access_token）作为 bearer token。请参阅 上面 的关于将 token 置于请求中。 图片 - Kubernetes OpenID Connect Flow 登陆到您的身份提供商 您的身份提供商将为您提供一个 access_token，一个 id_token 和一个 refresh_token 当使用 kubectl 时，使用 --token 标志和 id_token ，或者直接加入到您的 kubeconfig 文件中 kubectl 在调用 API server 时将 id_token 置于 HTTP header 中 API server 将通过检查配置中指定的证书来确保 JWT 签名有效 检查以确保 id_token 没有过期 确保用户已授权 授权 API server 后向 kubectl kubectl 向用户提供反馈 由于所有需要验证您身份的数据都在 id_token 中，Kubernetes 不需要向身份提供商 “phone home”。在每个请求都是无状态的模型中，这为认证提供了非常可扩展的解决方案。它确实提供了一些挑战： Kubernetes 没有 ”web 接口“ 来出发验证进程。没有浏览器或界面来收集凭据，这就是为什么您需要首先认证您的身份提供商。 id_token 无法撤销，就像一个证书，所以它应该是短暂的（只有几分钟），所以每隔几分钟就得到一个新的令牌是非常烦人的。 没有使用 kubectl proxy 命令或注入 id_token 的反向代理，无法简单地对 Kubernetes dashboard 进行身份验证。 配置 API Server 要启用该插件，需要在 API server 中配置如下标志： 参数 描述 示例 是否必需 --oidc-issuer-url 允许 API server 发现公共签名密钥的提供者的 URL。只接受使用 https:// 的方案。通常是提供商的 URL 地址，不包含路径，例如“https://accounts.google.com” 或者 “https://login.salesforce.com”。这个 URL 应该指向下面的 .well-known/openid-configuration 如果发现 URL 是 https://accounts.google.com/.well-known/openid-configuration，值应该是https://accounts.google.com 是 --oidc-client-id 所有的 token 必须为其颁发的客户端 ID kubernetes 是 --oidc-username-claim JWT声明使用的用户名。默认情况下，sub 是最终用户的唯一标识符。管理员可以选择其他声明，如email 或 name，具体取决于他们的提供者。不过，email 以外的其他声明将以发行者的 URL 作为前缀，以防止与其他插件命名冲突。 sub 否 --oidc-groups-claim JWT声明使用的用户组。如果生命存在，它必须是一个字符串数组。 groups 否 --oidc-ca-file 用来签名您的身份提供商的网络 CA 证书的路径。默认为主机的跟 CA。 /etc/kubernetes/ssl/kc-ca.pem 否 如果为 --oidc-username-claim 选择了除 email 以外的其他声明，则该值将以 --oidc-issuer-url 作为前缀，以防止与现有 Kubernetes 名称（例如 system:users）冲突。例如，如果提供商网址是 https://accounts.google.com，而用户名声明映射到 jane，则插件会将用户身份验证为： https://accounts.google.com#jane 重要的是，API server 不是 OAuth2 客户端，而只能配置为信任单个发行者。这允许使用 Google 等公共提供者，而不必信任第三方发行的凭据。希望利用多个 OAuth 客户端的管理员应该探索支持 azp（授权方）声明的提供者，这是允许一个客户端代表另一个客户端发放令牌的机制。 Kubernetes不提供 OpenID Connect 身份提供商。您可以使用现有的公共 OpenID Connect 标识提供程序（例如Google 或 其他）。或者，您可以运行自己的身份提供程序，例如 CoreOS dex、Keycloak、CloudFoundry UAA 或 Tremolo Security 的 OpenUnison。 对于身份提供商能够适用于 Kubernetes，必须满足如下条件：Kubernetes it must: 支持 OpenID connect 发现；不必是全部。 使用非过时密码在TLS中运行 拥有 CA 签名证书（即使 CA 不是商业 CA 或自签名） 有关上述要求3的说明，需要 CA 签名证书。如果您部署自己的身份提供商（而不是像 Google 或 Microsoft 之类的云提供商），则必须让您的身份提供商的 Web 服务器证书由 CA 标志设置为 TRUE 的证书签名，即使是自签名的。这是由于 GoLang 的 TLS 客户端实现对证书验证的标准非常严格。如果您没有 CA，可以使用 CoreOS 团队的 这个脚本 创建一个简单的 CA 和一个签名的证书和密钥对。 或者你可以使用 这个类似的脚本 来生成更长寿命和更长的 SHA256 证书密钥。 针对特定系统的安装说明： UAA Dex OpenUnison 使用 kubectl 选项 1 - OIDC 身份验证器 第一个选项是使用 oidc 身份验证器。此身份验证程序将您的 id_token、refresh_token 和您的 OIDC client_secret 自动刷新您的 token。一旦您对身份提供者进行了身份验证： kubectl config set-credentials USER_NAME \\ --auth-provider=oidc \\ --auth-provider-arg=idp-issuer-url=( issuer url ) \\ --auth-provider-arg=client-id=( your client id ) \\ --auth-provider-arg=client-secret=( your client secret ) \\ --auth-provider-arg=refresh-token=( your refresh token ) \\ --auth-provider-arg=idp-certificate-authority=( path to your ca certificate ) \\ --auth-provider-arg=id-token=( your id_token ) \\ --auth-provider-arg=extra-scopes=( comma separated list of scopes to add to \"openid email profile\", optional ) 例如，在向身份提供者进行身份验证之后运行以下命令： kubectl config set-credentials mmosley \\ --auth-provider=oidc \\ --auth-provider-arg=idp-issuer-url=https://oidcidp.tremolo.lan:8443/auth/idp/OidcIdP \\ --auth-provider-arg=client-id=kubernetes \\ --auth-provider-arg=client-secret=1db158f6-177d-4d9c-8a8b-d36869918ec5 \\ --auth-provider-arg=refresh-token=q1bKLFOyUiosTfawzA93TzZIDzH2TNa2SMm0zEiPKTUwME6BkEo6Sql5yUWVBSWpKUGphaWpxSVAfekBOZbBhaEW+VlFUeVRGcluyVF5JT4+haZmPsluFoFu5XkpXk5BXqHega4GAXlF+ma+vmYpFcHe5eZR+slBFpZKtQA= \\ --auth-provider-arg=idp-certificate-authority=/root/ca.pem \\ --auth-provider-arg=extra-scopes=groups \\ --auth-provider-arg=id-token=eyJraWQiOiJDTj1vaWRjaWRwLnRyZW1vbG8ubGFuLCBPVT1EZW1vLCBPPVRybWVvbG8gU2VjdXJpdHksIEw9QXJsaW5ndG9uLCBTVD1WaXJnaW5pYSwgQz1VUy1DTj1rdWJlLWNhLTEyMDIxNDc5MjEwMzYwNzMyMTUyIiwiYWxnIjoiUlMyNTYifQ.eyJpc3MiOiJodHRwczovL29pZGNpZHAudHJlbW9sby5sYW46ODQ0My9hdXRoL2lkcC9PaWRjSWRQIiwiYXVkIjoia3ViZXJuZXRlcyIsImV4cCI6MTQ4MzU0OTUxMSwianRpIjoiMm96US15TXdFcHV4WDlHZUhQdy1hZyIsImlhdCI6MTQ4MzU0OTQ1MSwibmJmIjoxNDgzNTQ5MzMxLCJzdWIiOiI0YWViMzdiYS1iNjQ1LTQ4ZmQtYWIzMC0xYTAxZWU0MWUyMTgifQ.w6p4J_6qQ1HzTG9nrEOrubxIMb9K5hzcMPxc9IxPx2K4xO9l-oFiUw93daH3m5pluP6K7eOE6txBuRVfEcpJSwlelsOsW8gb8VJcnzMS9EnZpeA0tW_p-mnkFc3VcfyXuhe5R3G7aa5d8uHv70yJ9Y3-UhjiN9EhpMdfPAoEB9fYKKkJRzF7utTTIPGrSaSU6d2pcpfYKaxIwePzEkT4DfcQthoZdy9ucNvvLoi1DIC-UocFD8HLs8LYKEqSxQvOcvnThbObJ9af71EwmuE21fO5KzMW20KtAeget1gnldOosPtz1G5EwvaQ401-RPQzPGMVBld0_zMCAwZttJ4knw 将产生下面的配置： users: - name: mmosley user: auth-provider: config: client-id: kubernetes client-secret: 1db158f6-177d-4d9c-8a8b-d36869918ec5 extra-scopes: groups id-token: eyJraWQiOiJDTj1vaWRjaWRwLnRyZW1vbG8ubGFuLCBPVT1EZW1vLCBPPVRybWVvbG8gU2VjdXJpdHksIEw9QXJsaW5ndG9uLCBTVD1WaXJnaW5pYSwgQz1VUy1DTj1rdWJlLWNhLTEyMDIxNDc5MjEwMzYwNzMyMTUyIiwiYWxnIjoiUlMyNTYifQ.eyJpc3MiOiJodHRwczovL29pZGNpZHAudHJlbW9sby5sYW46ODQ0My9hdXRoL2lkcC9PaWRjSWRQIiwiYXVkIjoia3ViZXJuZXRlcyIsImV4cCI6MTQ4MzU0OTUxMSwianRpIjoiMm96US15TXdFcHV4WDlHZUhQdy1hZyIsImlhdCI6MTQ4MzU0OTQ1MSwibmJmIjoxNDgzNTQ5MzMxLCJzdWIiOiI0YWViMzdiYS1iNjQ1LTQ4ZmQtYWIzMC0xYTAxZWU0MWUyMTgifQ.w6p4J_6qQ1HzTG9nrEOrubxIMb9K5hzcMPxc9IxPx2K4xO9l-oFiUw93daH3m5pluP6K7eOE6txBuRVfEcpJSwlelsOsW8gb8VJcnzMS9EnZpeA0tW_p-mnkFc3VcfyXuhe5R3G7aa5d8uHv70yJ9Y3-UhjiN9EhpMdfPAoEB9fYKKkJRzF7utTTIPGrSaSU6d2pcpfYKaxIwePzEkT4DfcQthoZdy9ucNvvLoi1DIC-UocFD8HLs8LYKEqSxQvOcvnThbObJ9af71EwmuE21fO5KzMW20KtAeget1gnldOosPtz1G5EwvaQ401-RPQzPGMVBld0_zMCAwZttJ4knw idp-certificate-authority: /root/ca.pem idp-issuer-url: https://oidcidp.tremolo.lan:8443/auth/idp/OidcIdP refresh-token: q1bKLFOyUiosTfawzA93TzZIDzH2TNa2SMm0zEiPKTUwME6BkEo6Sql5yUWVBSWpKUGphaWpxSVAfekBOZbBhaEW+VlFUeVRGcluyVF5JT4+haZmPsluFoFu5XkpXk5BXq name: oidc 一旦您的 id_token 过期，kubectl 将使用 refresh_token 刷新 id_token，然后在 kube/.config 文件的client_secret 中存储 id_token 的值和refresh_token 的新值。 选项 2 - 使用 --token 选项 可以在 kubectl 命令的 --token 选项中传入 token。简单的拷贝和复制 id_token 到该选项中： kubectl --token=eyJhbGciOiJSUzI1NiJ9.eyJpc3MiOiJodHRwczovL21sYi50cmVtb2xvLmxhbjo4MDQzL2F1dGgvaWRwL29pZGMiLCJhdWQiOiJrdWJlcm5ldGVzIiwiZXhwIjoxNDc0NTk2NjY5LCJqdGkiOiI2RDUzNXoxUEpFNjJOR3QxaWVyYm9RIiwiaWF0IjoxNDc0NTk2MzY5LCJuYmYiOjE0NzQ1OTYyNDksInN1YiI6Im13aW5kdSIsInVzZXJfcm9sZSI6WyJ1c2VycyIsIm5ldy1uYW1lc3BhY2Utdmlld2VyIl0sImVtYWlsIjoibXdpbmR1QG5vbW9yZWplZGkuY29tIn0.f2As579n9VNoaKzoF-dOQGmXkFKf1FMyNV0-va_B63jn-_n9LGSCca_6IVMP8pO-Zb4KvRqGyTP0r3HkHxYy5c81AnIh8ijarruczl-TK_yF5akjSTHFZD-0gRzlevBDiH8Q79NAr-ky0P4iIXS8lY9Vnjch5MF74Zx0c3alKJHJUnnpjIACByfF2SCaYzbWFMUNat-K1PaUk5-ujMBG7yYnr95xD-63n8CO8teGUAAEMx6zRjzfhnhbzX-ajwZLGwGUBT4WqjMs70-6a7_8gZmLZb2az1cZynkFRj2BaCkVT3A2RrjeEwZEtGXlMqKJ1_I2ulrOVsYx01_yD35-rw get nodes Webhook Token 认证 Webhook 认证是用来认证 bearer token 的 hook。 --authentication-token-webhook-config-file 是一个用来描述如何访问远程 webhook 服务的 kubeconfig 文件。 --authentication-token-webhook-cache-ttl 缓存身份验证策略的时间。默认为两分钟。 配置文件使用 kubeconfig 文件格式。文件中的 ”user“ 指的是 API server 的 webhook，”clusters“ 是指远程服务。见下面的例子： # clusters refers to the remote service. clusters: - name: name-of-remote-authn-service cluster: certificate-authority: /path/to/ca.pem # CA for verifying the remote service. server: https://authn.example.com/authenticate # URL of remote service to query. Must use 'https'. # users refers to the API server's webhook configuration. users: - name: name-of-api-server user: client-certificate: /path/to/cert.pem # cert for the webhook plugin to use client-key: /path/to/key.pem # key matching the cert # kubeconfig files require a context. Provide one for the API server. current-context: webhook contexts: - context: cluster: name-of-remote-authn-service user: name-of-api-sever name: webhook 当客户端尝试使用 bearer token 与API server 进行认证是，如 上 论述，认证 webhook 用饱含该 token 的对象查询远程服务。Kubernetes 不会挑战缺少该 header 的请求。 请注意，webhook API对象与其他 Kubernetes API 对象具有相同的 版本控制兼容性规则。实现者应该意识到 Beta 对象的宽松兼容性承诺，并检查请求的 “apiVersion” 字段以确保正确的反序列化。此外，API server 必须启用 authentication.k8s.io/v1beta1 API 扩展组（--runtime config =authentication.k8s.io/v1beta1=true）。 The request body will be of the following format: { \"apiVersion\": \"authentication.k8s.io/v1beta1\", \"kind\": \"TokenReview\", \"spec\": { \"token\": \"(BEARERTOKEN)\" } } 预计远程服务将填写请求的 status 字段以指示登录成功。响应主体的 spec 字段被忽略，可以省略。成功验证后的 bearer token 将返回： { \"apiVersion\": \"authentication.k8s.io/v1beta1\", \"kind\": \"TokenReview\", \"status\": { \"authenticated\": true, \"user\": { \"username\": \"janedoe@example.com\", \"uid\": \"42\", \"groups\": [ \"developers\", \"qa\" ], \"extra\": { \"extrafield1\": [ \"extravalue1\", \"extravalue2\" ] } } } } 未成功的请求将返回： { \"apiVersion\": \"authentication.k8s.io/v1beta1\", \"kind\": \"TokenReview\", \"status\": { \"authenticated\": false } } HTTP状态代码可以用来提供额外的错误上下文。 认证代理 可以配置 API server 从请求 header 的值中识别用户，例如 X-Remote-User。这样的设计是用来与请求 header 值的验证代理结合使用。 --requestheader-username-headers 必需，大小写敏感。按 header 名称和顺序检查用户标识。包含值的第一个 header 将被作为用户名。 --requestheader-group-headers 1.6 以上版本。可选。大小写敏感。建议为 “X-Remote-Group”。按 header 名称和顺序检查用户组。所有指定的 header 中的所有值都将作为组名。 --requestheader-extra-headers-prefix 1.6 以上版本。可选，大小写敏感。建议为 “X-Remote-Extra-”。标题前缀可用于查找有关用户的额外信息（通常由配置的授权插件使用）。以任何指定的前缀开头的 header 都会删除前缀，header 名称的其余部分将成为额外的键值，而 header 值则是额外的值。 例如下面的配置： --requestheader-username-headers=X-Remote-User --requestheader-group-headers=X-Remote-Group --requestheader-extra-headers-prefix=X-Remote-Extra- 该请求： GET / HTTP/1.1 X-Remote-User: fido X-Remote-Group: dogs X-Remote-Group: dachshunds X-Remote-Extra-Scopes: openid X-Remote-Extra-Scopes: profile 将产生如下的用户信息： name: fido groups: - dogs - dachshunds extra: scopes: - openid - profile 为了防止 header 欺骗，验证代理需要在验证请求 header 之前向 API server 提供有效的客户端证书，以对照指定的 CA 进行验证。 --requestheader-client-ca-file 必需。PEM 编码的证书包。在检查用户名的请求 header 之前，必须针对指定文件中的证书颁发机构提交并验证有效的客户端证书。 --requestheader-allowed-names 可选。Common Name （cn）列表。如果设置了，则在检查用户名的请求 header 之前， 必须提供指定列表中 Common Name（cn）的有效客户端证书。如果为空，则允许使用任何 Common Name。 Keystone 密码 通过在启动过程中将 --experimental-keystone-url= 选项传递给 API server 来启用 Keystone 认证。该插件在 plugin/pkg/auth/authenticator/password/keystone/keystone.go 中实现，目前使用基本身份验证通过用户名和密码验证用户。 如果您为 Keystone 服务器配置了自签名证书，则在启动 Kubernetes API server 时可能需要设置 --experimental-keystone-ca-file=SOMEFILE 选项。如果您设置了该选项，Keystone 服务器的证书将由experimental-keystone-ca-file 中的某个权威机构验证。否则，证书由主机的根证书颁发机构验证。 有关如何使用 keystone 来管理项目和用户的详细信息，请参阅 Keystone 文档。请注意，这个插件仍处于试验阶段，正在积极开发之中，并可能在后续版本中进行更改。 请参考 讨论、蓝图 和 提出的改变 获取更多信息。 匿名请求 启用时，未被其他已配置身份验证方法拒绝的请求将被视为匿名请求，并给予 system:anonymous 的用户名和system:unuthenticated 的组名。 例如，在配置了令牌认证和启用了匿名访问的服务器上，提供无效的 bearer token 的请求将收到 401 Unauthorized 错误。提供 bearer token 的请求将被视为匿名请求。 在 1.5.1 - 1.5.x 版本中，默认情况下命名访问是被禁用的，可以通过传递 --anonymous-auth=false 选项给 API server 来启用。 在 1.6+ 版本中，如果使用 AlwaysAllow 以外的授权模式，则默认启用匿名访问，并且可以通过将 --anonymous-auth=false选项传递给API服务器来禁用。从 1.6 开始，ABAC 和 RBAC 授权人需要明确授权 system:annoymous 或 system:unauthenticated 组，因此授予对 * 用户或 * 组访问权的传统策略规则不包括匿名用户。 用户模拟 用户可以通过模拟 header 充当另一个用户。该请求会覆盖请求认证的用户信息。例如，管理员可以使用此功能通过暂时模拟其他用户并查看请求是否被拒绝来调试授权策略。 模拟请求首先认证为请求用户，然后切换到模拟的用户信息。 用户使用他们的凭证和模拟 header 进行 API 调用。 API server 认证用户 API server 确保经过身份验证的用户具有模拟权限。 请求用户的信息被替换为模拟值 请求被评估，授权作用于模拟的用户信息。 以下 HTTP header 可用户执行模拟请求： Impersonate-User：充当的用户名 Impersonate-Group：作为组名。可以多次使用来设置多个组。可选的，需要 “Impersonate-User” Impersonate-Extra-( extra name )：用于将额外字段与用户关联的动态 header。可选。需要 “Impersonate-User” 一组示例 header： Impersonate-User: jane.doe@example.com Impersonate-Group: developers Impersonate-Group: admins Impersonate-Extra-dn: cn=jane,ou=engineers,dc=example,dc=com Impersonate-Extra-scopes: view Impersonate-Extra-scopes: development 当使用 kubectl 的 --as 标志来配置 Impersonate-User header 时，可以使用 --as-group 标志来配置 Impersonate-Group header。 $ kubectl drain mynode Error from server (Forbidden): User \"clark\" cannot get nodes at the cluster scope. (get nodes mynode) $ kubectl drain mynode --as=superman --as-group=system:masters node \"mynode\" cordoned node \"mynode\" draine 为模仿用户、组或设置额外字段，模拟用户必须能够对正在模拟的属性的种类（“用户”，“组”等）执行“模拟”动词。对于启用了 RBAC 授权插件的集群，以下 ClusterRole 包含设置用户和组模拟 header 所需的规则： apiVersion: rbac.authorization.k8s.io/v1 kind: ClusterRole metadata: name: impersonator rules: - apiGroups: [\"\"] resources: [\"users\", \"groups\", \"serviceaccounts\"] verbs: [\"impersonate\"] 额外的字段被评估为资源 “userextras” 的子资源。为了允许用户使用额外字段 “scope” 的模拟 header，应授予用户以下角色： apiVersion: rbac.authorization.k8s.io/v1 kind: ClusterRole metadata: name: scopes-impersonator # Can set \"Impersonate-Extra-scopes\" header. - apiGroups: [\"authentication.k8s.io\"] resources: [\"userextras/scopes\"] verbs: [\"impersonate\"] 模拟 header 的可用值可以通过设置 resourceNames 可以使用的资源来限制。 apiVersion: rbac.authorization.k8s.io/v1 kind: ClusterRole metadata: name: limited-impersonator rules: # Can impersonate the user \"jane.doe@example.com\" - apiGroups: [\"\"] resources: [\"users\"] verbs: [\"impersonate\"] resourceNames: [\"jane.doe@example.com\"] # Can impersonate the groups \"developers\" and \"admins\" - apiGroups: [\"\"] resources: [\"groups\"] - verbs: [\"impersonate\"] resourceNames: [\"developers\",\"admins\"] # Can impersonate the extras field \"scopes\" with the values \"view\" and \"development\" - apiGroups: [\"authentication.k8s.io\"] resources: [\"userextras/scopes\"] verbs: [\"impersonate\"] resourceNames: [\"view\", \"development\"] 附录 创建证书 使用客户端证书进行身份验证时，可以使用现有的部署脚本或通过 easyrsa 或 openssl 手动生成证书。 使用已有的部署脚本 已有的部署脚本 在 cluster/saltbase/salt/generate-cert/make-ca-cert.sh。 执行该脚本时需要传递两个参数。第一个参数是 API server 的 IP地址。第二个参数是 IP 形式的主题备用名称列表： IP: 或 DNS:。 该脚本将生成三个文件： ca.crt、server.crt 和 server.key。 最后，将一下参数添加到 API server 的启动参数中： --client-ca-file=/srv/kubernetes/ca.crt --tls-cert-file=/srv/kubernetes/server.crt --tls-private-key-file=/srv/kubernetes/server.key easyrsa easyrsa 可以用来手动为集群生成证书。 下载，解压，并初始化修补版本的easyrsa3。 curl -L -O https://storage.googleapis.com/kubernetes-release/easy-rsa/easy-rsa.tar.gz tar xzf easy-rsa.tar.gz cd easy-rsa-master/easyrsa3 ./easyrsa init-pki 生成 CA（使用 --batch 设置为自动模式。使用 --req-cn 设置默认的 CN） ./easyrsa --batch \"--req-cn=${MASTER_IP}@`date +%s`\" build-ca nopass 生成服务器证书和密钥。（build-server-full [文件名]：生成一个键值对，在本地为客户端和服务器签名。） ./easyrsa --subject-alt-name=\"IP:${MASTER_IP}\" build-server-full server nopass 复制 pki/ca.crt, pki/issued/server.crt 和 pki/private/server.key 到您的目录下。 将以下参数添加到 API server 的启动参数中： --client-ca-file=/yourdirectory/ca.crt --tls-cert-file=/yourdirectory/server.crt --tls-private-key-file=/yourdirectory/server.key openssl openssl 可以用来手动为集群生成证书。 生成一个 2048 bit 的 ca.key： openssl genrsa -out ca.key 2048 根据 ca.key 生成一个 ca.crt（使用 -days 设置证书的有效时间）： openssl req -x509 -new -nodes -key ca.key -subj \"/CN=${MASTER_IP}\" -days 10000 -out ca.crt 生成一个 2048 bit 的 server.key： openssl genrsa -out server.key 2048 根据 server.key 生成一个 server.csr： openssl req -new -key server.key -subj \"/CN=${MASTER_IP}\" -out server.csr 根据 ca.key、ca.crt 和 server.csr 生成 server.crt： openssl x509 -req -in server.csr -CA ca.crt -CAkey ca.key -CAcreateserial -out server.crt -days 10000 查看证书： openssl x509 -noout -text -in ./server.crt 最后，不要忘了向 API server 的启动参数中增加配置。 认证 API 您可以使用 certificates.k8s.io API将 x509 证书配置为用于身份验证，如 此处 所述。 官方v1.6文档地址：https://v1-6.docs.kubernetes.io/docs/admin/authentication/ 官方最新文档地址：https://kubernetes.io/docs/admin/authentication/ 译者：Jimmy Song Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-17 15:54:27 "},"guide/access-kubernetes-cluster.html":{"url":"guide/access-kubernetes-cluster.html","title":"访问Kubernetes集群","keywords":"","body":"访问 Kubernetes 集群 根据用户部署和暴露服务的方式不同，有很多种方式可以用来访问 kubernetes 集群。 最简单也是最直接的方式是使用 kubectl 命令。 其次可以使用 kubeconfig 文件来认证授权访问 API server。 通过各种 proxy 经过端口转发访问 kubernetes 集群中的服务 使用 Ingress，在集群外访问 kubernetes 集群内的 service Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-08-21 18:23:34 "},"guide/access-cluster.html":{"url":"guide/access-cluster.html","title":"访问集群","keywords":"","body":"访问集群 第一次使用 kubectl 访问 如果您是第一次访问 Kubernetes API 的话，我们建议您使用 Kubernetes 命令行工具：kubectl。 为了访问集群，您需要知道集群的地址，并且需要有访问它的凭证。通常，如果您完成了 入门指南 那么这些将会自动设置，或者其他人为您部署的集群提供并给您凭证和集群地址。 使用下面的命令检查 kubectl 已知的集群的地址和凭证： $ kubectl config view 关于 kubectl 命令使用的更多 示例 和完整文档可以在这里找到：kubectl 手册 直接访问 REST API Kubectl 处理对 apiserver 的定位和认证。如果您想直接访问 REST API，可以使用像 curl、wget 或浏览器这样的 http 客户端，有以下几种方式来定位和认证： 以 proxy 模式运行 kubectl。 推荐方法。 使用已保存的 apiserver 位置信息。 使用自签名证书验证 apiserver 的身份。 没有 MITM（中间人攻击）的可能。 认证到 apiserver。 将来，可能会做智能的客户端负载均衡和故障转移。 直接向 http 客户端提供位置和凭据。 替代方法。 适用于通过使用代理而混淆的某些类型的客户端代码。 需要将根证书导入浏览器以防止 MITM。 使用 kubectl proxy 以下命令作为反向代理的模式运行 kubectl。 它处理对 apiserver 的定位并进行认证。 像这样运行： $ kubectl proxy --port=8080 & 查看关于 kubectl proxy 的更多细节。 然后您可以使用 curl、wget 或者浏览器来访问 API，如下所示： $ curl http://localhost:8080/api/ { \"versions\": [ \"v1\" ] } 不使用 kubectl proxy（1.3.x 以前版本） 通过将认证 token 直接传递给 apiserver 的方式，可以避免使用 kubectl proxy，如下所示： $ APISERVER=$(kubectl config view | grep server | cut -f 2- -d \":\" | tr -d \" \") $ TOKEN=$(kubectl config view | grep token | cut -f 2 -d \":\" | tr -d \" \") $ curl $APISERVER/api --header \"Authorization: Bearer $TOKEN\" --insecure { \"versions\": [ \"v1\" ] } 不使用 kubectl proxy（1.3.x 以后版本） 在 Kubernetes 1.3 或更高版本中，kubectl config view 不再显示 token。 使用 kubectl describe secret … 获取 default service account 的 token，如下所示： $ APISERVER=$(kubectl config view | grep server | cut -f 2- -d \":\" | tr -d \" \") $ TOKEN=$(kubectl describe secret $(kubectl get secrets | grep default | cut -f1 -d ' ') | grep -E '^token' | cut -f2 -d':' | tr -d '\\t') $ curl $APISERVER/api --header \"Authorization: Bearer $TOKEN\" --insecure { \"kind\": \"APIVersions\", \"versions\": [ \"v1\" ], \"serverAddressByClientCIDRs\": [ { \"clientCIDR\": \"0.0.0.0/0\", \"serverAddress\": \"10.0.1.149:443\" } ] } 以上示例使用--insecure 标志。 这使得它容易受到 MITM 攻击。 当 kubectl 访问集群时，它使用存储的根证书和客户端证书来访问服务器。 （这些安装在~/.kube目录中）。 由于集群证书通常是自签名的，因此可能需要特殊配置才能让您的 http 客户端使用根证书。 对于某些群集，apiserver 可能不需要身份验证；可以选择在本地主机上服务，或者使用防火墙保护。 对此还没有一个标准。配置对API的访问 描述了群集管理员如何配置此操作。 这种方法可能与未来的高可用性支持相冲突。 编程访问 API Kubernetes 支持 Go 和 Python 客户端库。 Go 客户端 要获取该库，请运行以下命令：go get k8s.io/client-go//kubernetes 请参阅 https://github.com/kubernetes/client-go 以查看支持哪些版本。 使用 client-go 客户端编程。请注意，client-go 定义了自己的 API 对象，因此如果需要，请从 client-go 而不是从主存储库导入 API 定义，例如导入 k8s.io/client-go/1.4/pkg/api/v1 是正确的。 Go 客户端可以使用与 kubectl 命令行工具相同的 kubeconfig 文件 来定位和验证 apiserver。参考官方 示例 和 client-go 示例。 如果应用程序在群集中以 Pod 的形式部署，请参考 下一节。 Python 客户端 要使用 Python client，请运行以下命令：pip install kubernetes。查看 Python 客户端库页面 获取更多的安装选择。 Python 客户端可以使用与 kubectl 命令行工具相同的 kubeconfig 文件 来定位和验证 apiserver。参考该 示例。 其他语言 还有更多的 客户端库 可以用来访问 API。有关其他库的验证方式，请参阅文档。 在 Pod 中访问 API 在 Pod 中访问 API 时，定位和认证到 API server 的方式有所不同。在 Pod 中找到 apiserver 地址的推荐方法是使用kubernetes DNS 名称，将它解析为服务 IP，后者又将被路由到 apiserver。 向 apiserver 认证的推荐方法是使用 service account 凭据。通过 kube-system，pod 与 service account 相关联，并且将该 service account 的凭据（token）放入该 pod 中每个容器的文件系统树中，位于 /var/run/secrets/kubernetes.io/serviceaccount/token。 如果可用，证书包将位于每个容器的文件系统树的 /var/run/secrets/kubernetes.io/serviceaccount/ca.crt 位置，并用于验证 apiserver 的服务证书。 最后，用于 namespace API 操作的默认 namespace 放在每个容器中的 /var/run/secrets/kubernetes.io/serviceaccount/namespace 中。 在 pod 中，连接到 API 的推荐方法是： 将 kubectl proxy 作为 pod 中的一个容器来运行，或作为在容器内运行的后台进程。它将 Kubernetes API 代理到 pod 的本地主机接口，以便其他任何 pod 中的容器内的进程都可以访问它。请参阅 在 pod 中使用 kubectl proxy 的示例。 使用 Go 客户端库，并使用 rest.InClusterConfig() 和 kubernetes.NewForConfig() 函数创建一个客户端。 他们处理对 apiserver 的定位和认证。示例 在以上的几种情况下，都需要使用 pod 的凭据与 apiserver 进行安全通信。 访问集群中运行的 service 上一节是关于连接到 kubernetes API server。这一节是关于连接到 kubernetes 集群中运行的 service。在 Kubernetes 中，node、 pod 和 services 都有它们自己的 IP。很多情况下，集群中 node 的 IP、Pod 的 IP、service 的 IP 都是不可路由的，因此在集群外面的机器就无法访问到它们，例如从您自己的笔记本电脑。 连接的方式 您可以选择以下几种方式从集群外部连接到 node、pod 和 service： 通过 public IP 访问 service。 使用 NodePort 和 LoadBalancer 类型的 service，以使 service 能够在集群外部被访问到。请查看 service 和 kubectl expose 文档。 根据您的群集环境，这可能会将服务暴露给您的公司网络，或者可能会将其暴露在互联网上。想想暴露的服务是否安全。它是否自己进行身份验证？ 将 pod 放在服务后面。 要从一组副本（例如为了调试）访问一个特定的 pod，请在 pod 上放置一个唯一的 label，并创建一个选择该 label 的新服务。 在大多数情况下，应用程序开发人员不需要通过 node IP 直接访问节点。 通过 Proxy 规则访问 service、node、pod。 在访问远程服务之前，请执行 apiserver 认证和授权。 如果服务不够安全，无法暴露给互联网，或者为了访问节点 IP 上的端口或进行调试，请使用这种方式。 代理可能会导致某些 Web 应用程序出现问题。 仅适用于 HTTP/HTTPS。 见此描述。 在集群内访问 node 和 pod。 运行一个 pod，然后使用 kubectl exec 命令连接到 shell。从该 shell 中连接到其他 node、pod 和 service。 有些集群可能允许 ssh 到集群上的某个节点。 从那个节点您可以访问到集群中的服务。这是一个非标准的方法，它可能将在某些集群上奏效，而在某些集群不行。这些节点上可能安装了浏览器和其他工具也可能没有。群集 DNS 可能无法正常工作。 访问内置服务 通常集群内会有几个在 kube-system 中启动的服务。使用 kubectl cluster-info 命令获取该列表： $ kubectl cluster-info Kubernetes master is running at https://104.197.5.247 elasticsearch-logging is running at https://104.197.5.247/api/v1/namespaces/kube-system/services/elasticsearch-logging/proxy kibana-logging is running at https://104.197.5.247/api/v1/namespaces/kube-system/services/kibana-logging/proxy kube-dns is running at https://104.197.5.247/api/v1/namespaces/kube-system/services/kube-dns/proxy grafana is running at https://104.197.5.247/api/v1/namespaces/kube-system/services/monitoring-grafana/proxy heapster is running at https://104.197.5.247/api/v1/namespaces/kube-system/services/monitoring-heapster/proxy 这显示了访问每个服务的代理 URL。 例如，此集群启用了集群级日志记录（使用Elasticsearch），如果传入合适的凭据，可以在该地址 https://104.197.5.247/api/v1/namespaces/kube-system/services/elasticsearch-logging/proxy/ 访问到，或通过 kubectl 代理，例如：http://localhost:8080/api/v1/namespaces/kube-system/services/elasticsearch-logging/proxy/。 （有关如何传递凭据和使用 kubectl 代理，请 参阅上文） 手动构建 apiserver 代理 URL 如上所述，您可以使用 kubectl cluster-info 命令来检索服务的代理 URL。要创建包含服务端点、后缀和参数的代理 URL，您只需附加到服务的代理URL： http://kubernetes_master_address/api/v1/namespaces/namespace_name/services/service_name[:port_name]/proxy 如果您没有指定 port 的名字，那么您不必在 URL 里指定 port_name。 示例 要想访问 Elasticsearch 的服务端点 _search?q=user:kimchy，您需要使用：http://104.197.5.247/api/v1/namespaces/kube-system/services/elasticsearch-logging/proxy/_search?q=user:kimchy 要想访问 Elasticsearch 的集群健康信息 _cluster/health?pretty=true，您需要使用：https://104.197.5.247/api/v1/namespaces/kube-system/services/elasticsearch-logging/proxy/_cluster/health?pretty=true { \"cluster_name\" : \"kubernetes_logging\", \"status\" : \"yellow\", \"timed_out\" : false, \"number_of_nodes\" : 1, \"number_of_data_nodes\" : 1, \"active_primary_shards\" : 5, \"active_shards\" : 5, \"relocating_shards\" : 0, \"initializing_shards\" : 0, \"unassigned_shards\" : 5 } 使用 web 浏览器来访问集群中运行的服务 您可以将 apiserver 代理网址放在浏览器的地址栏中。 然而： Web 浏览器通常不能传递 token，因此您可能需要使用基本（密码）认证。 Apiserver 可以配置为接受基本认证，但您的集群可能未配置为接受基本认证。 某些网络应用程序可能无法正常工作，特别是那些在不知道代理路径前缀的情况下构造 URL 的客户端 JavaScript。 请求重定向 重定向功能已被弃用和删除。 请改用代理（见下文）。 多种代理 在使用 kubernetes 的时候您可能会遇到许多种不同的代理： kubectl 代理： 在用户桌面或 pod 中运行 从 localhost 地址到 Kubernetes apiserver 的代理 客户端到代理使用 HTTP apiserver 的代理使用 HTTPS 定位 apiserver 添加身份验证 header apiserver 代理： 将一个堡垒机作为 apiserver 将群集之外的用户连接到群集IP，否则可能无法访问 在 apiserver 进程中运行 客户端到代理使用 HTTPS（或 http，如果 apiserver 如此配置） 根据代理目标的可用信息由代理选择使用 HTTP 或 HTTPS 可用于访问 node、pod 或 service 用于访问 service 时进行负载均衡 kube 代理： 在每个节点上运行 代理 UDP 和 TCP 不支持 HTTP 提供负载均衡 只是用来访问 service apiserver 前面的代理/负载均衡器： 存在和实现因群集而异（例如 nginx） 位于所有客户端和一个或多个 apiserver 之间 作为负载均衡器，如果有多个 apiserver 外部服务的云负载均衡器： 由一些云提供商提供（例如 AWS ELB，Google Cloud Load Balancer） 当 Kubernetes service 类型为 LoadBalancer 时，会自动创建 仅使用 UDP/TCP 实施方式因云提供商而异 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-08-21 18:23:34 "},"guide/authenticate-across-clusters-kubeconfig.html":{"url":"guide/authenticate-across-clusters-kubeconfig.html","title":"使用kubeconfig文件配置跨集群认证","keywords":"","body":"使用 kubeconfig 文件配置跨集群认证 Kubernetes 的认证方式对于不同的人来说可能有所不同。 运行 kubelet 可能有一种认证方式（即证书）。 用户可能有不同的认证方式（即令牌）。 管理员可能具有他们为个人用户提供的证书列表。 我们可能有多个集群，并希望在同一个地方将其全部定义——这样用户就能使用自己的证书并重用相同的全局配置。 所以为了能够让用户轻松地在多个集群之间切换，对于多个用户的情况下，我们将其定义在了一个 kubeconfig 文件中。 此文件包含一系列与昵称相关联的身份验证机制和集群连接信息。它还引入了一个（用户）认证信息元组和一个被称为上下文的与昵称相关联的集群连接信息的概念。 如果明确指定，则允许使用多个 kubeconfig 文件。在运行时，它们与命令行中指定的覆盖选项一起加载并合并（参见下面的 规则）。 相关讨论 http://issue.k8s.io/1755 Kubeconfig 文件的组成 Kubeconifg 文件示例 current-context: federal-context apiVersion: v1 clusters: - cluster: api-version: v1 server: http://cow.org:8080 name: cow-cluster - cluster: certificate-authority: path/to/my/cafile server: https://horse.org:4443 name: horse-cluster - cluster: insecure-skip-tls-verify: true server: https://pig.org:443 name: pig-cluster contexts: - context: cluster: horse-cluster namespace: chisel-ns user: green-user name: federal-context - context: cluster: pig-cluster namespace: saw-ns user: black-user name: queen-anne-context kind: Config preferences: colors: true users: - name: blue-user user: token: blue-token - name: green-user user: client-certificate: path/to/my/client/cert client-key: path/to/my/client/key 各个组件的拆解/释意 Cluster clusters: - cluster: certificate-authority: path/to/my/cafile server: https://horse.org:4443 name: horse-cluster - cluster: insecure-skip-tls-verify: true server: https://pig.org:443 name: pig-cluster cluster 中包含 kubernetes 集群的端点数据，包括 kubernetes apiserver 的完整 url 以及集群的证书颁发机构或者当集群的服务证书未被系统信任的证书颁发机构签名时，设置insecure-skip-tls-verify: true。 cluster 的名称（昵称）作为该 kubeconfig 文件中的集群字典的 key。 您可以使用 kubectl config set-cluster 添加或修改 cluster 条目。 user users: - name: blue-user user: token: blue-token - name: green-user user: client-certificate: path/to/my/client/cert client-key: path/to/my/client/key user 定义用于向 kubernetes 集群进行身份验证的客户端凭据。在加载/合并 kubeconfig 之后，user 将有一个名称（昵称）作为用户条目列表中的 key。 可用凭证有 client-certificate、client-key、token 和 username/password。 username/password 和 token 是二者只能选择一个，但 client-certificate 和 client-key 可以分别与它们组合。 您可以使用 kubectl config set-credentials 添加或者修改 user 条目。 context contexts: - context: cluster: horse-cluster namespace: chisel-ns user: green-user name: federal-context context 定义了一个命名的 cluster、user、namespace 元组，用于使用提供的认证信息和命名空间将请求发送到指定的集群。 三个都是可选的；仅使用 cluster、user、namespace 之一指定上下文，或指定 none。 未指定的值或在加载的 kubeconfig 中没有相应条目的命名值（例如，如果为上述 kubeconfig 文件指定了 pink-user 的上下文）将被替换为默认值。 有关覆盖/合并行为，请参阅下面的 加载和合并规则。 您可以使用 kubectl config set-context 添加或修改上下文条目。 current-context current-context: federal-context current-context is the nickname or 'key' for the cluster,user,namespace tuple that kubectl will use by default when loading config from this file. You can override any of the values in kubectl from the commandline, by passing --context=CONTEXT, --cluster=CLUSTER, --user=USER, and/or --namespace=NAMESPACE respectively. You can change the current-context with kubectl config use-context. current-context 是昵称或者说是作为 cluster、user、namespace 元组的 ”key“，当 kubectl 从该文件中加载配置的时候会被默认使用。您可以在 kubectl 命令行里覆盖这些值，通过分别传入 —context=CONTEXT、 —cluster=CLUSTER、--user=USER 和 --namespace=NAMESPACE 。 您可以使用 kubectl config use-context 更改 current-context。 apiVersion: v1 kind: Config preferences: colors: true 杂项 apiVersion 和 kind 标识客户端解析器的版本和模式，不应手动编辑。 preferences 指定可选（和当前未使用）的 kubectl 首选项。 查看 kubeconfig 文件 kubectl config view 命令可以展示当前的 kubeconfig 设置。默认将为您展示所有的 kubeconfig 设置；您可以通过传入 —minify 参数，将视图过滤到与 current-context 有关的配额设置。有关其他选项，请参阅 kubectl config view。 构建您自己的 kubeconfig 文件 您可以使用上文 示例 kubeconfig 文件 作为 注意： 如果您是通过 kube-up.sh 脚本部署的 kubernetes 集群，不需要自己创建 kubeconfig 文件——该脚本已经为您创建过了。 {:.note} 当 api server 启动的时候使用了 —token-auth-file=tokens.csv 选项时，上述文件将会与 API server 相关联，tokens.csv 文件看起来会像这个样子： blue-user,blue-user,1 mister-red,mister-red,2 注意： 启动 API server 时有很多 可用选项。请您一定要确保理解您使用的选项。 上述示例 kubeconfig 文件提供了 green-user 的客户端凭证。因为用户的 current-user 是 green-user ，任何该 API server 的客户端使用该示例 kubeconfig 文件时都可以成功登录。同样，我们可以通过修改 current-context 的值以 blue-user 的身份操作。 在上面的示例中，green-user 通过提供凭据登录，blue-user 使用的是 token。使用 kubectl config set-credentials 指定登录信息。想了解更多信息，请访问 \"示例文件相关操作命令\"。 加载和合并规则 加载和合并 kubeconfig 文件的规则很简单，但有很多。最终的配置按照以下顺序构建： 从磁盘中获取 kubeconfig。这将通过以下层次结构和合并规则完成： 如果设置了 CommandLineLocation （kubeconfig 命令行参数的值），将会只使用该文件，而不会进行合并。该参数在一条命令中只允许指定一次。 或者，如果设置了 EnvVarLocation （$KUBECONFIG 的值），其将会被作为应合并的文件列表，并根据以下规则合并文件。空文件名被忽略。非串行内容的文件将产生错误。设置特定值或 map key 的第一个文件将优先使用，并且值或 map key 也永远不会更改。 这意味着设置 CurrentContext 的第一个文件将保留其上下文。 这也意味着如果两个文件同时指定一个 red-user，那么将只使用第一个文件中的 red-user 的值。 即使第二个文件的 red-user 中有非冲突条目也被丢弃。 另外，使用 Home 目录位置（~/.kube/config）将不会合并。 根据此链中的第一个命中确定要使用的上下文 命令行参数——context 命令行选项的值 来自合并后的 kubeconfig 文件的 current-context 在这个阶段允许空 确定要使用的群集信息和用户。此时，我们可能有也可能没有上下文。他们是基于这个链中的第一次命中。 （运行两次，一次为用户，一次为集群） 命令行参数——user 指定用户，cluster 指定集群名称 如果上下文存在，则使用上下文的值 允许空 确定要使用的实际群集信息。此时，我们可能有也可能没有集群信息。根据链条构建每个集群信息（第一次命中胜出）： 命令行参数——server，api-version，certificate-authority 和 insecure-skip-tls-verify 如果存在集群信息，并且存在该属性的值，请使用它。 如果没有服务器位置，则产生错误。 确定要使用的实际用户信息。用户使用与集群信息相同的规则构建，除非，您的每个用户只能使用一种认证技术。 负载优先级为1）命令行标志 2）来自 kubeconfig 的用户字段 命令行标志是：client-certificate、client-key、username、password 和 token 如果有两种冲突的技术，则失败。 对于任何仍然缺少的信息，将使用默认值，并可能会提示验证信息 Kubeconfig 文件中的所有文件引用都相对于 kubeconfig 文件本身的位置进行解析。当命令行上显示文件引用时，它们将相对于当前工作目录进行解析。当路径保存在 ~/.kube/config 中时，相对路径使用相对存储，绝对路径使用绝对存储。 Kubeconfig 文件中的任何路径都相对于 kubeconfig 文件本身的位置进行解析。 使用 kubectl config 操作 kubeconfig kubectl config 有一些列的子命令可以帮助我们更方便的操作 kubeconfig 文件。 请参阅 kubectl/kubectl_config。 Example $ kubectl config set-credentials myself --username=admin --password=secret $ kubectl config set-cluster local-server --server=http://localhost:8080 $ kubectl config set-context default-context --cluster=local-server --user=myself $ kubectl config use-context default-context $ kubectl config set contexts.default-context.namespace the-right-prefix $ kubectl config view 产生如下输出： apiVersion: v1 clusters: - cluster: server: http://localhost:8080 name: local-server contexts: - context: cluster: local-server namespace: the-right-prefix user: myself name: default-context current-context: default-context kind: Config preferences: {} users: - name: myself user: password: secret username: admin Kubeconfig 文件会像这样子： apiVersion: v1 clusters: - cluster: server: http://localhost:8080 name: local-server contexts: - context: cluster: local-server namespace: the-right-prefix user: myself name: default-context current-context: default-context kind: Config preferences: {} users: - name: myself user: password: secret username: admin 示例文件相关操作命令 $ kubectl config set preferences.colors true $ kubectl config set-cluster cow-cluster --server=http://cow.org:8080 --api-version=v1 $ kubectl config set-cluster horse-cluster --server=https://horse.org:4443 --certificate-authority=path/to/my/cafile $ kubectl config set-cluster pig-cluster --server=https://pig.org:443 --insecure-skip-tls-verify=true $ kubectl config set-credentials blue-user --token=blue-token $ kubectl config set-credentials green-user --client-certificate=path/to/my/client/cert --client-key=path/to/my/client/key $ kubectl config set-context queen-anne-context --cluster=pig-cluster --user=black-user --namespace=saw-ns $ kubectl config set-context federal-context --cluster=horse-cluster --user=green-user --namespace=chisel-ns $ kubectl config use-context federal-context 最后将它们捆绑在一起 所以，将这一切绑在一起，快速创建自己的 kubeconfig 文件： 仔细看一下，了解您的 api-server 的启动方式：在设计 kubeconfig 文件以方便身份验证之前，您需要知道您自己的安全要求和策略。 将上面的代码段替换为您的集群的 api-server 端点的信息。 确保您的 api-server 至少能够以提供一个用户（即 green-user）凭据的方式启动。 当然您必须查看 api-server 文档，以了解当前关于身份验证细节方面的最新技术。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-08-21 18:23:34 "},"guide/connecting-to-applications-port-forward.html":{"url":"guide/connecting-to-applications-port-forward.html","title":"通过端口转发访问集群中的应用程序","keywords":"","body":"通过端口转发访问集群中的应用程序 本页向您展示如何使用 kubectl port-forward 命令连接到运行在 Kubernetes 集群中的 Redis 服务器。这种类型的连接对于数据库调试很有帮助。 创建一个 Pod 来运行 Redis 服务器 创建一个 Pod： kubectl create -f https://k8s.io/docs/tasks/access-application-cluster/redis-master.yaml 命令运行成功后将有以下输出验证该 Pod 是否已经创建： pod \"redis-master\" created 检查 Pod 是否正在运行且处于就绪状态： kubectl get pods 当 Pod 就绪，输出显示 Running 的状态： NAME READY STATUS RESTARTS AGE redis-master 2/2 Running 0 41s 验证 Redis 服务器是否已在 Pod 中运行，并监听 6379 端口： kubectl get pods redis-master --template='{{(index (index .spec.containers 0).ports 0).containerPort}}{{\"\\n\"}}' 端口输出如下： 6379 将本地端口转发到 Pod 中的端口 将本地工作站上的 6379 端口转发到 redis-master pod 的 6379 端口： kubectl port-forward redis-master 6379:6379 输出类似于： I0710 14:43:38.274550 3655 portforward.go:225] Forwarding from 127.0.0.1:6379 -> 6379 I0710 14:43:38.274797 3655 portforward.go:225] Forwarding from [::1]:6379 -> 6379 启动 Redis 命令行界面 redis-cli 在 Redis 命令行提示符下，输入 ping 命令： 127.0.0.1:6379>ping Ping 请求成功返回 PONG。 讨论 创建连接，将本地的 6379 端口转发到运行在 Pod 中的 Redis 服务器的 6379 端口。有了这个连接您就可以在本地工作站中调试运行在 Pod 中的数据库。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-08-21 18:23:34 "},"guide/service-access-application-cluster.html":{"url":"guide/service-access-application-cluster.html","title":"使用service访问群集中的应用程序","keywords":"","body":"使用 service 访问群集中的应用程序 本文向您展示如何创建 Kubernetes Service 对象，外部客户端可以使用它来访问集群中运行的应用程序。该 Service 可以为具有两个运行实例的应用程序提供负载均衡。 目的 运行 Hello World 应用程序的两个实例。 创建一个暴露 node 节点端口的 Service 对象。 使用 Service 对象访问正在运行的应用程序。 为在两个 pod 中运行的应用程序创建 service 在集群中运行 Hello World 应用程序： kubectl run hello-world --replicas=2 --labels=\"run=load-balancer-example\" --image=gcr.io/google-samples/node-hello:1.0 --port=8080 上述命令创建一个 Deployment 对象和一个相关联的 ReplicaSet 对象。该 ReplicaSet 有两个 Pod，每个 Pod 中都运行一个 Hello World 应用程序。 显示关于该 Deployment 的信息： kubectl get deployments hello-world kubectl describe deployments hello-world 显示 ReplicaSet 的信息： kubectl get replicasets kubectl describe replicasets 创建一个暴露该 Deployment 的 Service 对象： kubectl expose deployment hello-world --type=NodePort --name=example-service 显示该 Service 的信息： kubectl describe services example-service 输出类似于： Name: example-service Namespace: default Labels: run=load-balancer-example Selector: run=load-balancer-example Type: NodePort IP: 10.32.0.16 Port: 8080/TCP NodePort: 31496/TCP Endpoints: 10.200.1.4:8080,10.200.2.5:8080 Session Affinity: None No events. 记下服务的 NodePort 值。例如，在前面的输出中，NodePort 值为 31496。 列出运行 Hello World 应用程序的 Pod： kubectl get pods --selector=\"run=load-balancer-example\" --output=wide 输出类似于： NAME READY STATUS ... IP NODE hello-world-2895499144-bsbk5 1/1 Running ... 10.200.1.4 worker1 hello-world-2895499144-m1pwt 1/1 Running ... 10.200.2.5 worker2 获取正在运行 Hello World 应用程序的 Pod 的其中一个节点的 public IP 地址。如何得到这个地址取决于您的集群设置。例如，如果您使用 Minikube，可以通过运行 kubectl cluster-info 查看节点地址。如果您是使用 Google Compute Engine 实例，可以使用 gcloud compute instances list 命令查看您的公共地址节点。 在您选择的节点上，在您的节点端口上例如创建允许 TCP 流量的防火墙规则，如果您的服务 NodePort 值为 31568，创建防火墙规则，允许端口 31568 上的TCP流量。 使用节点地址和节点端口访问 Hello World 应用程序： curl http://: 其中 是您节点的 public IP地址，而 是您服务的 NodePort 值。 对成功请求的响应是一个 hello 消息： Hello Kubernetes! 使用 Service 配置文件 作为使用 kubectl expose 的替代方法，您可以使用 service 配置文件 来创建 Service。 要删除 Service，输入以下命令： kubectl delete services example-service 删除 Deployment、ReplicaSet 和正运行在 Pod 中的 Hello World 应用程序，输入以下命令： kubectl delete deployment hello-world 了解更多关于 使用 service 连接应用程序。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-08-21 18:23:34 "},"guide/accessing-kubernetes-pods-from-outside-of-the-cluster.html":{"url":"guide/accessing-kubernetes-pods-from-outside-of-the-cluster.html","title":"从外部访问Kubernetes中的Pod","keywords":"","body":"从外部访问Kubernetes中的Pod 前面几节讲到如何访问kubneretes集群，本文主要讲解访问kubenretes中的Pod和Serivce的集中方式，包括如下几种： hostNetwork hostPort NodePort LoadBalancer Ingress 说是暴露Pod其实跟暴露Service是一回事，因为Pod就是Service的backend。 hostNetwork: true 这是一种直接定义Pod网络的方式。 如果在Pod中使用hostNotwork:true配置的话，在这种pod中运行的应用程序可以直接看到pod启动的主机的网络接口。在主机的所有网络接口上都可以访问到该应用程序。以下是使用主机网络的pod的示例定义： apiVersion: v1 kind: Pod metadata: name: influxdb spec: hostNetwork: true containers: - name: influxdb image: influxdb 部署该Pod： $ kubectl create -f influxdb-hostnetwork.yml 访问该pod所在主机的8086端口： curl -v http://$POD_IP:8086/ping 将看到204 No Content的204返回码，说明可以正常访问。 注意每次启动这个Pod的时候都可能被调度到不同的节点上，所有外部访问Pod的IP也是变化的，而且调度Pod的时候还需要考虑是否与宿主机上的端口冲突，因此一般情况下除非您知道需要某个特定应用占用特定宿主机上的特定端口时才使用hostNetwork: true的方式。 这种Pod的网络模式有一个用处就是可以将网络插件包装在Pod中然后部署在每个宿主机上，这样该Pod就可以控制该宿主机上的所有网络。 hostPort 这是一种直接定义Pod网络的方式。 hostPort是直接将容器的端口与所调度的节点上的端口路由，这样用户就可以通过宿主机的IP加上来访问Pod了，如:。 apiVersion: v1 kind: Pod metadata: name: influxdb spec: containers: - name: influxdb image: influxdb ports: - containerPort: 8086 hostPort: 8086 这样做有个缺点，因为Pod重新调度的时候该Pod被调度到的宿主机可能会变动，这样就变化了，用户必须自己维护一个Pod与所在宿主机的对应关系。 这种网络方式可以用来做 nginx Ingress controller。外部流量都需要通过kubenretes node节点的80和443端口。 NodePort NodePort在kubenretes里是一个广泛应用的服务暴露方式。Kubernetes中的service默认情况下都是使用的ClusterIP这种类型，这样的service会产生一个ClusterIP，这个IP只能在集群内部访问，要想让外部能够直接访问service，需要将service type修改为 nodePort。 apiVersion: v1 kind: Pod metadata: name: influxdb labels: name: influxdb spec: containers: - name: influxdb image: influxdb ports: - containerPort: 8086 同时还可以给service指定一个nodePort值，范围是30000-32767，这个值在API server的配置文件中，用--service-node-port-range定义。 kind: Service apiVersion: v1 metadata: name: influxdb spec: type: NodePort ports: - port: 8086 nodePort: 30000 selector: name: influxdb 集群外就可以使用kubernetes任意一个节点的IP加上30000端口访问该服务了。kube-proxy会自动将流量以round-robin的方式转发给该service的每一个pod。 这种服务暴露方式，无法让你指定自己想要的应用常用端口，不过可以在集群上再部署一个反向代理作为流量入口。 LoadBalancer LoadBalancer 只能在service上定义。这是公有云提供的负载均衡器，如AWS、Azure、CloudStack、GCE等。 kind: Service apiVersion: v1 metadata: name: influxdb spec: type: LoadBalancer ports: - port: 8086 selector: name: influxdb 查看服务： $ kubectl get svc influxdb NAME CLUSTER-IP EXTERNAL-IP PORT(S) AGE influxdb 10.97.121.42 10.13.242.236 8086:30051/TCP 39s 内部可以使用ClusterIP加端口来访问服务，如19.97.121.42:8086。 外部可以用以下两种方式访问该服务： 使用任一节点的IP加30051端口访问该服务 使用EXTERNAL-IP来访问，这是一个VIP，是云供应商提供的负载均衡器IP，如10.13.242.236:8086。 Ingress Ingress是自kubernetes1.1版本后引入的资源类型。必须要部署Ingress controller才能创建Ingress资源，Ingress controller是以一种插件的形式提供。Ingress controller 是部署在Kubernetes之上的Docker容器。它的Docker镜像包含一个像nginx或HAProxy的负载均衡器和一个控制器守护进程。控制器守护程序从Kubernetes接收所需的Ingress配置。它会生成一个nginx或HAProxy配置文件，并重新启动负载平衡器进程以使更改生效。换句话说，Ingress controller是由Kubernetes管理的负载均衡器。 Kubernetes Ingress提供了负载平衡器的典型特性：HTTP路由，粘性会话，SSL终止，SSL直通，TCP和UDP负载平衡等。目前并不是所有的Ingress controller都实现了这些功能，需要查看具体的Ingress controller文档。 apiVersion: extensions/v1beta1 kind: Ingress metadata: name: influxdb spec: rules: - host: influxdb.kube.example.com http: paths: - backend: serviceName: influxdb servicePort: 8086 外部访问URL http://influxdb.kube.example.com/ping 访问该服务，入口就是80端口，然后Ingress controller直接将流量转发给后端Pod，不需再经过kube-proxy的转发，比LoadBalancer方式更高效。 总结 总的来说Ingress是一个非常灵活和越来越得到厂商支持的服务暴露方式，包括Nginx、HAProxy、Traefik，还有各种Service Mesh，而其它服务暴露方式可以更适用于服务调试、特殊应用的部署。 参考 Accessing Kubernetes Pods from Outside of the Cluster - alesnosek.com Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-21 20:08:06 "},"guide/carbin-mobile-dashboard-for-kubernetes.html":{"url":"guide/carbin-mobile-dashboard-for-kubernetes.html","title":"Carbin - Kuberentes手机客户端","keywords":"","body":"Carbin - Kubernetes手机客户端 Carbin是由bitnami开源的手机管理Kubernetes集群的客户端，目前提供iOS和安卓版本，代码开源在GitHub上：https://bitnami.com/ 为了方便移动办公，可以使用Carbin这个kuberntes手机客户端，可以链接GKE和任何Kubernetes集群，可以使用以下三种认证方式： 证书 token kubeconfig文件 所有功能跟kubernetes dashboard相同，还可以支持使用Helm chart部署应用，可以配置自定义的chart仓库地址。 iPhone用户可以在App Store中搜索Carbin即可找到。 图片 - App Store 可以很方便的在手机上操作自己的kubernetes集群，还可以登录到容器中操作，只要是kubernetes API支持的功能，都可以在该移动客户端上实现。 图片 - 在手机上操作Kubernetes集群 更多详细信息请参考：https://github.com/bitnami/cabin Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-22 18:38:12 "},"guide/kubernetes-desktop-client.html":{"url":"guide/kubernetes-desktop-client.html","title":"Kubernetic - Kubernetes桌面客户端","keywords":"","body":"Kubernetic - Kubernetes桌面客户端 Kubernetic - 一款kubenretes桌面客户端，https://kubernetic.com/，支持以下特性： 实时展示集群状态 多集群，多个namespace管理 原生kubernetes支持 支持使用chart安装应用 使用kubeconfig登陆认证 图片 - Kubernetic客户端 该客户端支持Mac和Windows系统，beta版本免费使用，stable版本需要付费。beta版本某些功能不完善，比如无法在应用内修改ingress配置，无法监控应用的状态等。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-30 20:08:19 "},"guide/application-development-deployment-flow.html":{"url":"guide/application-development-deployment-flow.html","title":"在kubernetes中开发部署应用","keywords":"","body":"在 Kubernetes 中开发部署应用 理论上只要可以使用主机名做服务注册的应用都可以迁移到 kubernetes 集群上。看到这里你可能不禁要问，为什么使用 IP 地址做服务注册发现的应用不适合迁移到 kubernetes 集群？因为这样的应用不适合自动故障恢复，因为目前 kubernetes 中不支持固定 Pod 的 IP 地址，当 Pod 故障后自动转移到其他 Node 的时候该 Pod 的 IP 地址也随之变化。 将传统应用迁移到 kubernetes 中可能还有很长的路要走，但是直接开发 Cloud native 应用，kubernetes 就是最佳运行时环境了。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-08-21 18:23:34 "},"guide/deploy-applications-in-kubernetes.html":{"url":"guide/deploy-applications-in-kubernetes.html","title":"适用于kubernetes的应用开发部署流程","keywords":"","body":"适用于kubernetes的应用开发部署流程 为了讲解详细流程，我特意写了两个示例程序放在GitHub中，模拟监控流程： k8s-app-monitor-test：生成模拟的监控数据，发送http请求，获取json返回值 K8s-app-monitor-agent：获取监控数据并绘图，访问浏览器获取图表 API文档见k8s-app-monitor-test中的api.html文件，该文档在API blueprint中定义，使用aglio生成，打开后如图所示： 图片 - API 关于服务发现 K8s-app-monitor-agent服务需要访问k8s-app-monitor-test服务，这就涉及到服务发现的问题，我们在代码中直接写死了要访问的服务的内网DNS地址（kubedns中的地址，即k8s-app-monitor-test.default.svc.cluster.local）。 我们知道Kubernetes在启动Pod的时候为容器注入环境变量，这些环境变量在所有的 namespace 中共享（环境变量是不断追加的，新启动的Pod中将拥有老的Pod中所有的环境变量，而老的Pod中的环境变量不变）。但是既然使用这些环境变量就已经可以访问到对应的service，那么获取应用的地址信息，究竟是使用变量呢？还是直接使用DNS解析来发现？ 答案是使用DNS，详细说明见Kubernetes中的服务发现与Docker容器间的环境变量传递源码探究。 打包镜像 因为我使用wercker自动构建，构建完成后自动打包成docker镜像并上传到docker hub中（需要现在docker hub中创建repo）。 构建流程见：https://app.wercker.com/jimmysong/k8s-app-monitor-agent/ 图片 - wercker 生成了如下两个docker镜像： jimmysong/k8s-app-monitor-test:latest jimmysong/k8s-app-monitor-agent:latest 启动服务 所有的kubernetes应用启动所用的yaml配置文件都保存在那两个GitHub仓库的manifest.yaml文件中。 分别在两个GitHub目录下执行kubectl create -f manifest.yaml即可启动服务。 外部访问 服务启动后需要更新ingress配置，在ingress.yaml文件中增加以下几行： - host: k8s-app-monitor-agent.jimmysong.io http: paths: - path: / backend: serviceName: k8s-app-monitor-agent servicePort: 8080 保存后，然后执行kubectl replace -f ingress.yaml即可刷新ingress。 修改本机的/etc/hosts文件，在其中加入以下一行： 172.20.0.119 k8s-app-monitor-agent.jimmysong.io 当然你也可以加入到DNS中，为了简单起见我使用hosts。 详见边缘节点配置。 在浏览器中访问http://k8s-app-monitor-agent.jimmysong.io Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-07 10:17:15 "},"guide/migrating-hadoop-yarn-to-kubernetes.html":{"url":"guide/migrating-hadoop-yarn-to-kubernetes.html","title":"迁移传统应用到kubernetes中——以Hadoop YARN为例","keywords":"","body":"迁移传统应用到Kubernetes步骤详解——以Hadoop YARN为例 本文档不是说明如何在 kubernetes 中开发和部署应用程序，如果您想要直接开发应用程序在 kubernetes 中运行可以参考 适用于kubernetes的应用开发部署流程。 本文旨在说明如何将已有的应用程序尤其是传统的分布式应用程序迁移到 kubernetes 中。如果该类应用程序符合云原生应用规范（如12因素法则）的话，那么迁移会比较顺利，否则会遇到一些麻烦甚至是阻碍。具体请参考 迁移至云原生应用架构。 接下来我们将以 Spark on YARN with kubernetes 为例来说明，该例子足够复杂也很有典型性，了解了这个例子可以帮助大家将自己的应用迁移到 kubernetes 集群上去，代码和配置文件可以在 这里 找到（本文中加入 Spark 的配置，代码中并没有包含，读者可以自己配置）。 下图即整个架构的示意图，代码和详细配置文件请参考 kube-yarn（不包含 ingress、spark 配置），所有的进程管理和容器扩容直接使用 Makefile。 图片 - spark on yarn with kubernetes 注意： 该例子仅用来说明具体的步骤划分和复杂性，在生产环境应用还有待验证，请谨慎使用。 术语 对于为曾接触过 kubernetes 或对云平台的技术细节不太了解的人来说，如何将应用迁移到 kubernetes 中可能是个头疼的问题，在行动之前有必要先了解整个过程中需要用到哪些概念和术语，有助于大家在行动中达成共识。 过程中可能用到的概念和术语初步整理如下： 图片 - Terms 为了讲解整改过程和具体细节，我们所有操作都是通过命令手动完成，不使用自动化工具。当您充分了解到其中的细节后可以通过自动化工具来优化该过程，以使其更加自动和高效，同时减少因为人为操作失误导致的迁移失败。 迁移应用 图片 - 分解步骤解析 整个迁移过程分为如下几个步骤： 将原有应用拆解为服务 我们不是一上来就开始做镜像，写配置，而是应该先梳理下要迁移的应用中有哪些可以作为服务运行，哪些是变的，哪些是不变的部分。 服务划分的原则是最小可变原则，这个同样适用于镜像制作，将服务中不变的部分编译到同一个镜像中。 对于像 Spark on YARN 这样复杂的应用，可以将其划分为三大类服务： ResourceManager NodeManager Spark client 制作镜像 根据拆解出来的服务，我们需要制作两个镜像： Hadoop Spark (From hadoop docker image) 因为我们运行的是 Spark on YARN，因此 Spark 依赖与 Hadoop 镜像，我们在 Spark 的基础上包装了一个 web service 作为服务启动。 镜像制作过程中不需要在 Dockerfile 中指定 Entrypoint 和 CMD，这些都是在 kubernetes 的 YAML 文件中指定的。 Hadoop YARN 的 Dockerfile 参考如下配置。 FROM my-docker-repo/jdk:7u80 # Add native libs ARG HADOOP_VERSION=2.6.0-cdh5.5.2 ## Prefer to download from server not use local storage ADD hadoop-${HADOOP_VERSION}.tar.gz /usr/local ADD ./lib/* /usr/local/hadoop-${HADOOP_VERSION}/lib/native/ ADD ./jars/* /usr/local/hadoop-${HADOOP_VERSION}/share/hadoop/yarn/ ENV HADOOP_PREFIX=/usr/local/hadoop \\ HADOOP_COMMON_HOME=/usr/local/hadoop \\ HADOOP_HDFS_HOME=/usr/local/hadoop \\ HADOOP_MAPRED_HOME=/usr/local/hadoop \\ HADOOP_YARN_HOME=/usr/local/hadoop \\ HADOOP_CONF_DIR=/usr/local/hadoop/etc/hadoop \\ YARN_CONF_DIR=/usr/local/hadoop/etc/hadoop \\ PATH=${PATH}:/usr/local/hadoop/bin RUN \\ cd /usr/local && ln -s ./hadoop-${HADOOP_VERSION} hadoop && \\ rm -f ${HADOOP_PREFIX}/logs/* WORKDIR $HADOOP_PREFIX # Hdfs ports EXPOSE 50010 50020 50070 50075 50090 8020 9000 # Mapred ports EXPOSE 19888 #Yarn ports EXPOSE 8030 8031 8032 8033 8040 8042 8088 #Other ports EXPOSE 49707 2122 准备应用的配置文件 因为我们只制作了一个 Hadoop 的镜像，而需要启动两个服务，这就要求在服务启动的时候必须加载不同的配置文件，现在我们只需要准备两个服务中需要同时用的的配置的部分。 YARN 依赖的配置在 artifacts 目录下，包含以下文件： bootstrap.sh capacity-scheduler.xml container-executor.cfg core-site.xml hadoop-env.sh hdfs-site.xml log4j.properties mapred-site.xml nodemanager_exclude.txt slaves start-yarn-nm.sh start-yarn-rm.sh yarn-env.sh yarn-site.xml 其中作为 bootstrap 启动脚本的 bootstrap.sh 也包含在该目录下，该脚本的如何编写请见下文。 Kubernetes YAML 文件 根据业务的特性选择最适合的 kubernetes 的资源对象来运行，因为在 YARN 中 NodeManager 需要使用主机名向 ResourceManger 注册，因此需要沿用 YARN 原有的服务发现方式，使用 headless service 和 StatefulSet 资源。更多资料请参考 StatefulSet。 所有的 Kubernetes YAML 配置文件存储在 manifest 目录下，包括如下配置： yarn-cluster 的 namespace 配置 Spark、ResourceManager、NodeManager 的 headless service 和 StatefulSet 配置 需要暴露到 kubernetes 集群外部的 ingress 配置（ResourceManager 的 Web） kube-yarn-ingress.yaml spark-statefulset.yaml yarn-cluster-namespace.yaml yarn-nm-statefulset.yaml yarn-rm-statefulset.yaml Bootstrap 脚本 Bootstrap 脚本的作用是在启动时根据 Pod 的环境变量、主机名或其他可以区分不同 Pod 和将启动角色的变量来修改配置文件和启动服务应用。 该脚本同时将原来 YARN 的日志使用 stdout 输出，便于使用 kubectl logs 查看日志或其他日志收集工具进行日志收集。 启动脚本 bootstrap.sh 跟 Hadoop 的配置文件同时保存在 artifacts 目录下。 该脚本根据 Pod 的主机名，决定如何修改 Hadoop 的配置文件和启动何种服务。bootstrap.sh 文件的部分代码如下： if [[ \"${HOSTNAME}\" =~ \"yarn-nm\" ]]; then sed -i '//d' $HADOOP_PREFIX/etc/hadoop/yarn-site.xml cat >> $HADOOP_PREFIX/etc/hadoop/yarn-site.xml yarn.nodemanager.resource.memory-mb ${MY_MEM_LIMIT:-2048} yarn.nodemanager.resource.cpu-vcores ${MY_CPU_LIMIT:-2} EOM echo '' >> $HADOOP_PREFIX/etc/hadoop/yarn-site.xml cp ${CONFIG_DIR}/start-yarn-nm.sh $HADOOP_PREFIX/sbin/ cd $HADOOP_PREFIX/sbin chmod +x start-yarn-nm.sh ./start-yarn-nm.sh fi if [[ $1 == \"-d\" ]]; then until find ${HADOOP_PREFIX}/logs -mmin -1 | egrep -q '.*'; echo \"`date`: Waiting for logs...\" ; do sleep 2 ; done tail -F ${HADOOP_PREFIX}/logs/* & while true; do sleep 1000; done fi 从这部分中代码中可以看到，如果 Pod 的主机名中包含 yarn-nm 字段则向 yarn-site.xml 配置文件中增加如下内容： yarn.nodemanager.resource.memory-mb ${MY_MEM_LIMIT:-2048} yarn.nodemanager.resource.cpu-vcores ${MY_CPU_LIMIT:-2} 其中 MY_MEM_LIMIT 和 MY_CPU_LIMIT 是 kubernetes YAML 中定义的环境变量，该环境变量又是引用的 Resource limit。 所有的配置准备完成后，执行 start-yarn-nm.sh 脚本启动 NodeManager。 如果 kubernetes YAML 中的 container CMD args 中包含 -d 则在后台运行 NodeManger 并 tail 输出 NodeManager 的日志到标准输出。 ConfigMaps 将 Hadoop 的配置文件和 bootstrap 脚本作为 ConfigMap 资源保存，用作 Pod 启动时挂载的 volume。 kubectl create configmap hadoop-config \\ --from-file=artifacts/hadoop/bootstrap.sh \\ --from-file=artifacts/hadoop/start-yarn-rm.sh \\ --from-file=artifacts/hadoop/start-yarn-nm.sh \\ --from-file=artifacts/hadoop/slaves \\ --from-file=artifacts/hadoop/core-site.xml \\ --from-file=artifacts/hadoop/hdfs-site.xml \\ --from-file=artifacts/hadoop/mapred-site.xml \\ --from-file=artifacts/hadoop/yarn-site.xml \\ --from-file=artifacts/hadoop/capacity-scheduler.xml \\ --from-file=artifacts/hadoop/container-executor.cfg \\ --from-file=artifacts/hadoop/hadoop-env.sh \\ --from-file=artifacts/hadoop/log4j.properties \\ --from-file=artifacts/hadoop/nodemanager_exclude.txt \\ --from-file=artifacts/hadoop/yarn-env.sh kubectl create configmap spark-config \\ --from-file=artifacts/spark/spark-bootstrap.sh \\ --from-file=artifacts/spark/spark-env.sh \\ --from-file=artifacts/spark/spark-defaults.conf 所有的配置完成后，可以可以使用 kubectl 命令来启动和管理集群了，我们编写了 Makefile，您可以直接使用该 Makefile 封装的命令实现部分的自动化。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-08-21 18:23:34 "},"guide/using-statefulset.html":{"url":"guide/using-statefulset.html","title":"使用StatefulSet部署用状态应用","keywords":"","body":"使用StatefulSet部署有状态应用 StatefulSet 这个对象是专门用来部署用状态应用的，可以为Pod提供稳定的身份标识，包括hostname、启动顺序、DNS名称等。 下面以在kubernetes1.6版本中部署zookeeper和kafka为例讲解StatefulSet的使用，其中kafka依赖于zookeeper。 Dockerfile和配置文件见 zookeeper 和 kafka。 注：所有的镜像基于CentOS系统的JDK制作，为我的私人镜像，外部无法访问，yaml中没有配置持久化存储。 部署Zookeeper Dockerfile中从远程获取zookeeper的安装文件，然后在定义了三个脚本： zkGenConfig.sh：生成zookeeper配置文件 zkMetrics.sh：获取zookeeper的metrics zkOk.sh：用来做ReadinessProb 我们在来看下这三个脚本的执行结果： zkGenConfig.sh zkMetrics.sh脚本实际上执行的是下面的命令： $ echo mntr | nc localhost $ZK_CLIENT_PORT >& 1 zk_version 3.4.6-1569965, built on 02/20/2014 09:09 GMT zk_avg_latency 0 zk_max_latency 5 zk_min_latency 0 zk_packets_received 427879 zk_packets_sent 427890 zk_num_alive_connections 3 zk_outstanding_requests 0 zk_server_state leader zk_znode_count 18 zk_watch_count 3 zk_ephemerals_count 4 zk_approximate_data_size 613 zk_open_file_descriptor_count 29 zk_max_file_descriptor_count 1048576 zk_followers 1 zk_synced_followers 1 zk_pending_syncs 0 zkOk.sh脚本实际上执行的是下面的命令： $ echo ruok | nc 127.0.0.1 $ZK_CLIENT_PORT imok zookeeper.yaml 下面是启动三个zookeeper实例的yaml配置文件： --- apiVersion: v1 kind: Service metadata: name: zk-svc labels: app: zk-svc spec: ports: - port: 2888 name: server - port: 3888 name: leader-election clusterIP: None selector: app: zk --- apiVersion: v1 kind: ConfigMap metadata: name: zk-cm data: jvm.heap: \"1G\" tick: \"2000\" init: \"10\" sync: \"5\" client.cnxns: \"60\" snap.retain: \"3\" purge.interval: \"0\" --- apiVersion: policy/v1beta1 kind: PodDisruptionBudget metadata: name: zk-pdb spec: selector: matchLabels: app: zk minAvailable: 2 --- apiVersion: apps/v1beta1 kind: StatefulSet metadata: name: zk spec: serviceName: zk-svc replicas: 3 template: metadata: labels: app: zk spec: affinity: podAntiAffinity: requiredDuringSchedulingIgnoredDuringExecution: - labelSelector: matchExpressions: - key: \"app\" operator: In values: - zk topologyKey: \"kubernetes.io/hostname\" containers: - name: k8szk imagePullPolicy: Always image: sz-pg-oam-docker-hub-001.tendcloud.com/library/zookeeper:3.4.6 resources: requests: memory: \"2Gi\" cpu: \"500m\" ports: - containerPort: 2181 name: client - containerPort: 2888 name: server - containerPort: 3888 name: leader-election env: - name : ZK_REPLICAS value: \"3\" - name : ZK_HEAP_SIZE valueFrom: configMapKeyRef: name: zk-cm key: jvm.heap - name : ZK_TICK_TIME valueFrom: configMapKeyRef: name: zk-cm key: tick - name : ZK_INIT_LIMIT valueFrom: configMapKeyRef: name: zk-cm key: init - name : ZK_SYNC_LIMIT valueFrom: configMapKeyRef: name: zk-cm key: tick - name : ZK_MAX_CLIENT_CNXNS valueFrom: configMapKeyRef: name: zk-cm key: client.cnxns - name: ZK_SNAP_RETAIN_COUNT valueFrom: configMapKeyRef: name: zk-cm key: snap.retain - name: ZK_PURGE_INTERVAL valueFrom: configMapKeyRef: name: zk-cm key: purge.interval - name: ZK_CLIENT_PORT value: \"2181\" - name: ZK_SERVER_PORT value: \"2888\" - name: ZK_ELECTION_PORT value: \"3888\" command: - sh - -c - zkGenConfig.sh && zkServer.sh start-foreground readinessProbe: exec: command: - \"zkOk.sh\" initialDelaySeconds: 10 timeoutSeconds: 5 livenessProbe: exec: command: - \"zkOk.sh\" initialDelaySeconds: 10 timeoutSeconds: 5 securityContext: runAsUser: 1000 fsGroup: 1000 我们再主要下上面那三个脚本的用途。 部署kafka Kafka的docker镜像制作跟zookeeper类似，都是从远程下载安装包后，解压安装。 与zookeeper不同的是，只要一个脚本，但是又依赖于我们上一步安装的zookeeper，kafkaGenConfig.sh用来生成kafka的配置文件。 我们来看下这个脚本。 #!/bin/bash HOST=`hostname -s` if [[ $HOST =~ (.*)-([0-9]+)$ ]]; then NAME=${BASH_REMATCH[1]} ORD=${BASH_REMATCH[2]} else echo \"Failed to extract ordinal from hostname $HOST\" exit 1 fi MY_ID=$((ORD+1)) sed -i s\"/broker.id=0/broker.id=$MY_ID/g\" /opt/kafka/config/server.properties sed -i s'/zookeeper.connect=localhost:2181/zookeeper.connect=zk-0.zk-svc.brand.svc:2181,zk-1.zk-svc.brand.svc:2181,zk-2.zk-svc.brand.svc:2181/g' /opt/kafka/config/server.properties 该脚本根据statefulset生成的pod的hostname的后半截数字部分作为broker ID，同时再替换zookeeper的地址。 Kafka.yaml 下面是创建3个kafka实例的yaml配置。 --- apiVersion: v1 kind: Service metadata: name: kafka-svc labels: app: kafka spec: ports: - port: 9093 name: server clusterIP: None selector: app: kafka --- apiVersion: policy/v1beta1 kind: PodDisruptionBudget metadata: name: kafka-pdb spec: selector: matchLabels: app: kafka minAvailable: 2 --- apiVersion: apps/v1beta1 kind: StatefulSet metadata: name: kafka spec: serviceName: kafka-svc replicas: 3 template: metadata: labels: app: kafka spec: affinity: podAntiAffinity: requiredDuringSchedulingIgnoredDuringExecution: - labelSelector: matchExpressions: - key: \"app\" operator: In values: - kafka topologyKey: \"kubernetes.io/hostname\" podAffinity: preferredDuringSchedulingIgnoredDuringExecution: - weight: 1 podAffinityTerm: labelSelector: matchExpressions: - key: \"app\" operator: In values: - zk topologyKey: \"kubernetes.io/hostname\" terminationGracePeriodSeconds: 300 containers: - name: k8skafka imagePullPolicy: Always image: sz-pg-oam-docker-hub-001.tendcloud.com/library/kafka:2.10-0.8.2.1 resources: requests: memory: \"1Gi\" cpu: 500m env: - name: KF_REPLICAS value: \"3\" ports: - containerPort: 9093 name: server command: - /bin/bash - -c - \"/opt/kafka/bin/kafkaGenConfig.sh && /opt/kafka/bin/kafka-server-start.sh /opt/kafka/config/server.properties\" env: - name: KAFKA_HEAP_OPTS value : \"-Xmx512M -Xms512M\" - name: KAFKA_OPTS value: \"-Dlogging.level=DEBUG\" readinessProbe: tcpSocket: port: 9092 initialDelaySeconds: 15 timeoutSeconds: 1 参考 https://kubernetes.io/docs/concepts/workloads/controllers/statefulset/ kubernetes contrib - statefulsets http://blog.kubernetes.io/2017/01/running-mongodb-on-kubernetes-with-statefulsets.html Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-10-23 16:22:50 "},"practice/":{"url":"practice/","title":"最佳实践概览","keywords":"","body":"最佳实践概览 本章节从零开始创建我们自己的kubernetes集群，并在该集群的基础上，配置服务发现、负载均衡和日志收集等功能，使我们的集群能够成为一个真正线上可用、功能完整的集群。 第一部分 在CentOS上部署kubernetes集群中介绍了如何通过二进制文件在CentOS物理机（也可以是公有云主机）上快速部署一个kubernetes集群。 第二部分介绍如何在kubernetes中的服务发现与负载均衡。 第三部分介绍如何运维kubernetes集群。 第四部分介绍kubernetes中的存储管理。 第五部分关于kubernetes集群和应用的监控。 第六部分介绍kuberentes中的服务编排与管理。 第七部分介绍如何基于kubernetes做持续集成与发布。 第八部分是kubernetes集群与插件的更新升级。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-19 17:55:03 "},"practice/install-kubernetes-on-centos.html":{"url":"practice/install-kubernetes-on-centos.html","title":"在CentOS上部署kubernetes集群","keywords":"","body":"在CentOS上部署kubernetes集群 本文档最初是基于kubenetes1.6版本编写的，对于kuberentes1.8版本同样适用，只是个别位置有稍许变动，变动的地方我将特别注明。 本系列文档介绍使用二进制部署 kubernetes 集群的所有步骤，而不是使用 kubeadm 等自动化方式来部署集群，同时开启了集群的TLS安全认证，该安装步骤适用于所有bare metal环境、on-premise环境和公有云环境。 在部署的过程中，将详细列出各组件的启动参数，给出配置文件，详解它们的含义和可能遇到的问题。 部署完成后，你将理解系统各组件的交互原理，进而能快速解决实际问题。 所以本文档主要适合于那些有一定 kubernetes 基础，想通过一步步部署的方式来学习和了解系统配置、运行原理的人。 注：本文档中不包括docker和私有镜像仓库的安装，安装说明中使用的镜像来自 Google Cloud Platform，为了方便国内用户下载，我将其克隆并上传到了 时速云镜像市场，供大家免费下载。 欲下载最新版本的官方镜像请访问 Google 云平台容器注册表。 提供所有的配置文件 集群安装时所有组件用到的配置文件，包含在以下目录中： etc：service的环境变量配置文件 manifest：kubernetes应用的yaml文件 systemd ：systemd serivce配置文件 集群详情 OS：CentOS Linux release 7.3.1611 (Core) 3.10.0-514.16.1.el7.x86_64 Kubernetes 1.6.0 Docker 1.12.5（使用yum安装） Etcd 3.1.5 Flannel 0.7.1 vxlan或者host-gw 网络 TLS 认证通信 (所有组件，如 etcd、kubernetes master 和 node) RBAC 授权 kubelet TLS BootStrapping kubedns、dashboard、heapster(influxdb、grafana)、EFK(elasticsearch、fluentd、kibana) 集群插件 私有docker镜像仓库harbor（请自行部署，harbor提供离线安装包，直接使用docker-compose启动即可） 环境说明 在下面的步骤中，我们将在三台CentOS系统的物理机上部署具有三个节点的kubernetes1.6.0集群。 角色分配如下： Master：172.20.0.113 Node：172.20.0.113、172.20.0.114、172.20.0.115 注意：172.20.0.113这台主机master和node复用。所有生成证书、执行kubectl命令的操作都在这台节点上执行。一旦node加入到kubernetes集群之后就不需要再登陆node节点了。 安装前的准备 在node节点上安装docker1.12.5 直接使用yum install docker 关闭所有节点的SELinux 永久方法 – 需要重启服务器 修改/etc/selinux/config文件中设置SELINUX=disabled ，然后重启服务器。 临时方法 – 设置系统参数 使用命令setenforce 0 附： setenforce 1 设置SELinux 成为enforcing模式 setenforce 0 设置SELinux 成为permissive模式 准备harbor私有镜像仓库 参考：https://github.com/vmware/harbor 步骤介绍 创建 TLS 证书和秘钥 创建kubeconfig 文件 创建高可用etcd集群 安装kubectl命令行工具 部署master节点 安装flannel网络插件 部署node节点 安装kubedns插件 安装dashboard插件 安装heapster插件 安装EFK插件 提醒 由于启用了 TLS 双向认证、RBAC 授权等严格的安全机制，建议从头开始部署，而不要从中间开始，否则可能会认证、授权等失败！ 部署过程中需要有很多证书的操作，请大家耐心操作，不明白的操作可以参考本书中的其他章节的解释。 该部署操作仅是搭建成了一个可用 kubernetes 集群，而很多地方还需要进行优化，heapster 插件、EFK 插件不一定会用于真实的生产环境中，但是通过部署这些插件，可以让大家了解到如何部署应用到集群上。 注：本安装文档参考了 opsnull 跟我一步步部署 kubernetes 集群 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-19 17:29:57 "},"practice/create-tls-and-secret-key.html":{"url":"practice/create-tls-and-secret-key.html","title":"创建TLS证书和秘钥","keywords":"","body":"创建TLS证书和秘钥 前言 执行下列步骤前建议你先阅读以下内容： 管理集群中的TLS：教您如何创建TLS证书 kubelet的认证授权：向您描述如何通过认证授权来访问 kubelet 的 HTTPS 端点。 TLS bootstrap：介绍如何为 kubelet 设置 TLS 客户端证书引导（bootstrap）。 注意：这一步是在安装配置kubernetes的所有步骤中最容易出错也最难于排查问题的一步，而这却刚好是第一步，万事开头难，不要因为这点困难就望而却步。 如果您足够有信心在完全不了解自己在做什么的情况下能够成功地完成了这一步的配置，那么您可以尽管跳过上面的几篇文章直接进行下面的操作。 kubernetes 系统的各组件需要使用 TLS 证书对通信进行加密，本文档使用 CloudFlare 的 PKI 工具集 cfssl 来生成 Certificate Authority (CA) 和其它证书； 生成的 CA 证书和秘钥文件如下： ca-key.pem ca.pem kubernetes-key.pem kubernetes.pem kube-proxy.pem kube-proxy-key.pem admin.pem admin-key.pem 使用证书的组件如下： etcd：使用 ca.pem、kubernetes-key.pem、kubernetes.pem； kube-apiserver：使用 ca.pem、kubernetes-key.pem、kubernetes.pem； kubelet：使用 ca.pem； kube-proxy：使用 ca.pem、kube-proxy-key.pem、kube-proxy.pem； kubectl：使用 ca.pem、admin-key.pem、admin.pem； kube-controller-manager：使用 ca-key.pem、ca.pem 注意：以下操作都在 master 节点即 172.20.0.113 这台主机上执行，证书只需要创建一次即可，以后在向集群中添加新节点时只要将 /etc/kubernetes/ 目录下的证书拷贝到新节点上即可。 安装 CFSSL 方式一：直接使用二进制源码包安装 wget https://pkg.cfssl.org/R1.2/cfssl_linux-amd64 chmod +x cfssl_linux-amd64 mv cfssl_linux-amd64 /usr/local/bin/cfssl wget https://pkg.cfssl.org/R1.2/cfssljson_linux-amd64 chmod +x cfssljson_linux-amd64 mv cfssljson_linux-amd64 /usr/local/bin/cfssljson wget https://pkg.cfssl.org/R1.2/cfssl-certinfo_linux-amd64 chmod +x cfssl-certinfo_linux-amd64 mv cfssl-certinfo_linux-amd64 /usr/local/bin/cfssl-certinfo export PATH=/usr/local/bin:$PATH 方式二：使用go命令安装 我们的系统中安装了Go1.7.5，使用以下命令安装更快捷： $ go get -u github.com/cloudflare/cfssl/cmd/... $ echo $GOPATH /usr/local $ls /usr/local/bin/cfssl* cfssl cfssl-bundle cfssl-certinfo cfssljson cfssl-newkey cfssl-scan 在$GOPATH/bin目录下得到以cfssl开头的几个命令。 注意：以下文章中出现的cat的文件名如果不存在需要手工创建。 创建 CA (Certificate Authority) 创建 CA 配置文件 mkdir /root/ssl cd /root/ssl cfssl print-defaults config > config.json cfssl print-defaults csr > csr.json # 根据config.json文件的格式创建如下的ca-config.json文件 # 过期时间设置成了 87600h cat > ca-config.json 字段说明 ca-config.json：可以定义多个 profiles，分别指定不同的过期时间、使用场景等参数；后续在签名证书时使用某个 profile； signing：表示该证书可用于签名其它证书；生成的 ca.pem 证书中 CA=TRUE； server auth：表示client可以用该 CA 对server提供的证书进行验证； client auth：表示server可以用该CA对client提供的证书进行验证； 创建 CA 证书签名请求 创建 ca-csr.json 文件，内容如下： { \"CN\": \"kubernetes\", \"key\": { \"algo\": \"rsa\", \"size\": 2048 }, \"names\": [ { \"C\": \"CN\", \"ST\": \"BeiJing\", \"L\": \"BeiJing\", \"O\": \"k8s\", \"OU\": \"System\" } ] } \"CN\"：Common Name，kube-apiserver 从证书中提取该字段作为请求的用户名 (User Name)；浏览器使用该字段验证网站是否合法； \"O\"：Organization，kube-apiserver 从证书中提取该字段作为请求用户所属的组 (Group)； 生成 CA 证书和私钥 $ cfssl gencert -initca ca-csr.json | cfssljson -bare ca $ ls ca* ca-config.json ca.csr ca-csr.json ca-key.pem ca.pem 创建 kubernetes 证书 创建 kubernetes 证书签名请求文件 kubernetes-csr.json： { \"CN\": \"kubernetes\", \"hosts\": [ \"127.0.0.1\", \"172.20.0.112\", \"172.20.0.113\", \"172.20.0.114\", \"172.20.0.115\", \"10.254.0.1\", \"kubernetes\", \"kubernetes.default\", \"kubernetes.default.svc\", \"kubernetes.default.svc.cluster\", \"kubernetes.default.svc.cluster.local\" ], \"key\": { \"algo\": \"rsa\", \"size\": 2048 }, \"names\": [ { \"C\": \"CN\", \"ST\": \"BeiJing\", \"L\": \"BeiJing\", \"O\": \"k8s\", \"OU\": \"System\" } ] } 如果 hosts 字段不为空则需要指定授权使用该证书的 IP 或域名列表，由于该证书后续被 etcd 集群和 kubernetes master 集群使用，所以上面分别指定了 etcd 集群、kubernetes master 集群的主机 IP 和 kubernetes 服务的服务 IP（一般是 kube-apiserver 指定的 service-cluster-ip-range 网段的第一个IP，如 10.254.0.1。 hosts 中的内容可以为空，即使按照上面的配置，向集群中增加新节点后也不需要重新生成证书。 生成 kubernetes 证书和私钥 $ cfssl gencert -ca=ca.pem -ca-key=ca-key.pem -config=ca-config.json -profile=kubernetes kubernetes-csr.json | cfssljson -bare kubernetes $ ls kubernetes* kubernetes.csr kubernetes-csr.json kubernetes-key.pem kubernetes.pem 或者直接在命令行上指定相关参数： echo '{\"CN\":\"kubernetes\",\"hosts\":[\"\"],\"key\":{\"algo\":\"rsa\",\"size\":2048}}' | cfssl gencert -ca=ca.pem -ca-key=ca-key.pem -config=ca-config.json -profile=kubernetes -hostname=\"127.0.0.1,172.20.0.112,172.20.0.113,172.20.0.114,172.20.0.115,kubernetes,kubernetes.default\" - | cfssljson -bare kubernetes 创建 admin 证书 创建 admin 证书签名请求文件 admin-csr.json： { \"CN\": \"admin\", \"hosts\": [], \"key\": { \"algo\": \"rsa\", \"size\": 2048 }, \"names\": [ { \"C\": \"CN\", \"ST\": \"BeiJing\", \"L\": \"BeiJing\", \"O\": \"system:masters\", \"OU\": \"System\" } ] } 后续 kube-apiserver 使用 RBAC 对客户端(如 kubelet、kube-proxy、Pod)请求进行授权； kube-apiserver 预定义了一些 RBAC 使用的 RoleBindings，如 cluster-admin 将 Group system:masters 与 Role cluster-admin 绑定，该 Role 授予了调用kube-apiserver 的所有 API的权限； OU 指定该证书的 Group 为 system:masters，kubelet 使用该证书访问 kube-apiserver 时 ，由于证书被 CA 签名，所以认证通过，同时由于证书用户组为经过预授权的 system:masters，所以被授予访问所有 API 的权限； 生成 admin 证书和私钥 $ cfssl gencert -ca=ca.pem -ca-key=ca-key.pem -config=ca-config.json -profile=kubernetes admin-csr.json | cfssljson -bare admin $ ls admin* admin.csr admin-csr.json admin-key.pem admin.pem 创建 kube-proxy 证书 创建 kube-proxy 证书签名请求文件 kube-proxy-csr.json： { \"CN\": \"system:kube-proxy\", \"hosts\": [], \"key\": { \"algo\": \"rsa\", \"size\": 2048 }, \"names\": [ { \"C\": \"CN\", \"ST\": \"BeiJing\", \"L\": \"BeiJing\", \"O\": \"k8s\", \"OU\": \"System\" } ] } CN 指定该证书的 User 为 system:kube-proxy； kube-apiserver 预定义的 RoleBinding cluster-admin 将User system:kube-proxy 与 Role system:node-proxier 绑定，该 Role 授予了调用 kube-apiserver Proxy 相关 API 的权限； 生成 kube-proxy 客户端证书和私钥 $ cfssl gencert -ca=ca.pem -ca-key=ca-key.pem -config=ca-config.json -profile=kubernetes kube-proxy-csr.json | cfssljson -bare kube-proxy $ ls kube-proxy* kube-proxy.csr kube-proxy-csr.json kube-proxy-key.pem kube-proxy.pem 校验证书 以 kubernetes 证书为例 使用 opsnssl 命令 $ openssl x509 -noout -text -in kubernetes.pem ... Signature Algorithm: sha256WithRSAEncryption Issuer: C=CN, ST=BeiJing, L=BeiJing, O=k8s, OU=System, CN=Kubernetes Validity Not Before: Apr 5 05:36:00 2017 GMT Not After : Apr 5 05:36:00 2018 GMT Subject: C=CN, ST=BeiJing, L=BeiJing, O=k8s, OU=System, CN=kubernetes ... X509v3 extensions: X509v3 Key Usage: critical Digital Signature, Key Encipherment X509v3 Extended Key Usage: TLS Web Server Authentication, TLS Web Client Authentication X509v3 Basic Constraints: critical CA:FALSE X509v3 Subject Key Identifier: DD:52:04:43:10:13:A9:29:24:17:3A:0E:D7:14:DB:36:F8:6C:E0:E0 X509v3 Authority Key Identifier: keyid:44:04:3B:60:BD:69:78:14:68:AF:A0:41:13:F6:17:07:13:63:58:CD X509v3 Subject Alternative Name: DNS:kubernetes, DNS:kubernetes.default, DNS:kubernetes.default.svc, DNS:kubernetes.default.svc.cluster, DNS:kubernetes.default.svc.cluster.local, IP Address:127.0.0.1, IP Address:172.20.0.112, IP Address:172.20.0.113, IP Address:172.20.0.114, IP Address:172.20.0.115, IP Address:10.254.0.1 ... 确认 Issuer 字段的内容和 ca-csr.json 一致； 确认 Subject 字段的内容和 kubernetes-csr.json 一致； 确认 X509v3 Subject Alternative Name 字段的内容和 kubernetes-csr.json 一致； 确认 X509v3 Key Usage、Extended Key Usage 字段的内容和 ca-config.json 中 kubernetes profile 一致； 使用 cfssl-certinfo 命令 $ cfssl-certinfo -cert kubernetes.pem ... { \"subject\": { \"common_name\": \"kubernetes\", \"country\": \"CN\", \"organization\": \"k8s\", \"organizational_unit\": \"System\", \"locality\": \"BeiJing\", \"province\": \"BeiJing\", \"names\": [ \"CN\", \"BeiJing\", \"BeiJing\", \"k8s\", \"System\", \"kubernetes\" ] }, \"issuer\": { \"common_name\": \"Kubernetes\", \"country\": \"CN\", \"organization\": \"k8s\", \"organizational_unit\": \"System\", \"locality\": \"BeiJing\", \"province\": \"BeiJing\", \"names\": [ \"CN\", \"BeiJing\", \"BeiJing\", \"k8s\", \"System\", \"Kubernetes\" ] }, \"serial_number\": \"174360492872423263473151971632292895707129022309\", \"sans\": [ \"kubernetes\", \"kubernetes.default\", \"kubernetes.default.svc\", \"kubernetes.default.svc.cluster\", \"kubernetes.default.svc.cluster.local\", \"127.0.0.1\", \"10.64.3.7\", \"10.254.0.1\" ], \"not_before\": \"2017-04-05T05:36:00Z\", \"not_after\": \"2018-04-05T05:36:00Z\", \"sigalg\": \"SHA256WithRSA\", ... 分发证书 将生成的证书和秘钥文件（后缀名为.pem）拷贝到所有机器的 /etc/kubernetes/ssl 目录下备用； mkdir -p /etc/kubernetes/ssl cp *.pem /etc/kubernetes/ssl 参考 Generate self-signed certificates Setting up a Certificate Authority and Creating TLS Certificates Client Certificates V/s Server Certificates 数字证书及 CA 的扫盲介绍 TLS bootstrap 引导程序 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-09 17:50:24 "},"practice/create-kubeconfig.html":{"url":"practice/create-kubeconfig.html","title":"创建kubeconfig文件","keywords":"","body":"创建 kubeconfig 文件 注意：请先参考 安装kubectl命令行工具，先在 master 节点上安装 kubectl 然后再进行下面的操作。 kubelet、kube-proxy 等 Node 机器上的进程与 Master 机器的 kube-apiserver 进程通信时需要认证和授权； kubernetes 1.4 开始支持由 kube-apiserver 为客户端生成 TLS 证书的 TLS Bootstrapping 功能，这样就不需要为每个客户端生成证书了；该功能当前仅支持为 kubelet 生成证书； 因为我的master节点和node节点复用，所有在这一步其实已经安装了kubectl。参考安装kubectl命令行工具。 以下操作只需要在master节点上执行，生成的*.kubeconfig文件可以直接拷贝到node节点的/etc/kubernetes目录下。 创建 TLS Bootstrapping Token Token auth file Token可以是任意的包涵128 bit的字符串，可以使用安全的随机数发生器生成。 export BOOTSTRAP_TOKEN=$(head -c 16 /dev/urandom | od -An -t x | tr -d ' ') cat > token.csv 后三行是一句，直接复制上面的脚本运行即可。 注意：在进行后续操作前请检查 token.csv 文件，确认其中的 ${BOOTSTRAP_TOKEN} 环境变量已经被真实的值替换。 BOOTSTRAP_TOKEN 将被写入到 kube-apiserver 使用的 token.csv 文件和 kubelet 使用的 bootstrap.kubeconfig 文件，如果后续重新生成了 BOOTSTRAP_TOKEN，则需要： 更新 token.csv 文件，分发到所有机器 (master 和 node）的 /etc/kubernetes/ 目录下，分发到node节点上非必需； 重新生成 bootstrap.kubeconfig 文件，分发到所有 node 机器的 /etc/kubernetes/ 目录下； 重启 kube-apiserver 和 kubelet 进程； 重新 approve kubelet 的 csr 请求； cp token.csv /etc/kubernetes/ 创建 kubelet bootstrapping kubeconfig 文件 执行下面的命令时需要先安装kubectl命令，请参考安装kubectl命令行工具。 cd /etc/kubernetes export KUBE_APISERVER=\"https://172.20.0.113:6443\" # 设置集群参数 kubectl config set-cluster kubernetes \\ --certificate-authority=/etc/kubernetes/ssl/ca.pem \\ --embed-certs=true \\ --server=${KUBE_APISERVER} \\ --kubeconfig=bootstrap.kubeconfig # 设置客户端认证参数 kubectl config set-credentials kubelet-bootstrap \\ --token=${BOOTSTRAP_TOKEN} \\ --kubeconfig=bootstrap.kubeconfig # 设置上下文参数 kubectl config set-context default \\ --cluster=kubernetes \\ --user=kubelet-bootstrap \\ --kubeconfig=bootstrap.kubeconfig # 设置默认上下文 kubectl config use-context default --kubeconfig=bootstrap.kubeconfig --embed-certs 为 true 时表示将 certificate-authority 证书写入到生成的 bootstrap.kubeconfig 文件中； 设置客户端认证参数时没有指定秘钥和证书，后续由 kube-apiserver 自动生成； 创建 kube-proxy kubeconfig 文件 export KUBE_APISERVER=\"https://172.20.0.113:6443\" # 设置集群参数 kubectl config set-cluster kubernetes \\ --certificate-authority=/etc/kubernetes/ssl/ca.pem \\ --embed-certs=true \\ --server=${KUBE_APISERVER} \\ --kubeconfig=kube-proxy.kubeconfig # 设置客户端认证参数 kubectl config set-credentials kube-proxy \\ --client-certificate=/etc/kubernetes/ssl/kube-proxy.pem \\ --client-key=/etc/kubernetes/ssl/kube-proxy-key.pem \\ --embed-certs=true \\ --kubeconfig=kube-proxy.kubeconfig # 设置上下文参数 kubectl config set-context default \\ --cluster=kubernetes \\ --user=kube-proxy \\ --kubeconfig=kube-proxy.kubeconfig # 设置默认上下文 kubectl config use-context default --kubeconfig=kube-proxy.kubeconfig 设置集群参数和客户端认证参数时 --embed-certs 都为 true，这会将 certificate-authority、client-certificate 和 client-key 指向的证书文件内容写入到生成的 kube-proxy.kubeconfig 文件中； kube-proxy.pem 证书中 CN 为 system:kube-proxy，kube-apiserver 预定义的 RoleBinding cluster-admin 将User system:kube-proxy 与 Role system:node-proxier 绑定，该 Role 授予了调用 kube-apiserver Proxy 相关 API 的权限； 分发 kubeconfig 文件 将两个 kubeconfig 文件分发到所有 Node 机器的 /etc/kubernetes/ 目录 cp bootstrap.kubeconfig kube-proxy.kubeconfig /etc/kubernetes/ 参考 关于 kubeconfig 文件的更多信息请参考 使用 kubeconfig 文件配置跨集群认证。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-15 12:24:51 "},"practice/etcd-cluster-installation.html":{"url":"practice/etcd-cluster-installation.html","title":"创建高可用etcd集群","keywords":"","body":"创建高可用 etcd 集群 kuberntes 系统使用 etcd 存储所有数据，本文档介绍部署一个三节点高可用 etcd 集群的步骤，这三个节点复用 kubernetes master 机器，分别命名为sz-pg-oam-docker-test-001.tendcloud.com、sz-pg-oam-docker-test-002.tendcloud.com、sz-pg-oam-docker-test-003.tendcloud.com： sz-pg-oam-docker-test-001.tendcloud.com：172.20.0.113 sz-pg-oam-docker-test-002.tendcloud.com：172.20.0.114 sz-pg-oam-docker-test-003.tendcloud.com：172.20.0.115 TLS 认证文件 需要为 etcd 集群创建加密通信的 TLS 证书，这里复用以前创建的 kubernetes 证书 cp ca.pem kubernetes-key.pem kubernetes.pem /etc/kubernetes/ssl kubernetes 证书的 hosts 字段列表中包含上面三台机器的 IP，否则后续证书校验会失败； 下载二进制文件 到 https://github.com/coreos/etcd/releases 页面下载最新版本的二进制文件 wget https://github.com/coreos/etcd/releases/download/v3.1.5/etcd-v3.1.5-linux-amd64.tar.gz tar -xvf etcd-v3.1.5-linux-amd64.tar.gz mv etcd-v3.1.5-linux-amd64/etcd* /usr/local/bin 或者直接使用yum命令安装： yum install etcd 若使用yum安装，默认etcd命令将在/usr/bin目录下，注意修改下面的etcd.service文件中的启动命令地址为/usr/bin/etcd。 创建 etcd 的 systemd unit 文件 注意替换IP地址为你自己的etcd集群的主机IP。 [Unit] Description=Etcd Server After=network.target After=network-online.target Wants=network-online.target Documentation=https://github.com/coreos [Service] Type=notify WorkingDirectory=/var/lib/etcd/ EnvironmentFile=-/etc/etcd/etcd.conf ExecStart=/usr/local/bin/etcd \\ --name ${ETCD_NAME} \\ --cert-file=/etc/kubernetes/ssl/kubernetes.pem \\ --key-file=/etc/kubernetes/ssl/kubernetes-key.pem \\ --peer-cert-file=/etc/kubernetes/ssl/kubernetes.pem \\ --peer-key-file=/etc/kubernetes/ssl/kubernetes-key.pem \\ --trusted-ca-file=/etc/kubernetes/ssl/ca.pem \\ --peer-trusted-ca-file=/etc/kubernetes/ssl/ca.pem \\ --initial-advertise-peer-urls ${ETCD_INITIAL_ADVERTISE_PEER_URLS} \\ --listen-peer-urls ${ETCD_LISTEN_PEER_URLS} \\ --listen-client-urls ${ETCD_LISTEN_CLIENT_URLS},http://127.0.0.1:2379 \\ --advertise-client-urls ${ETCD_ADVERTISE_CLIENT_URLS} \\ --initial-cluster-token ${ETCD_INITIAL_CLUSTER_TOKEN} \\ --initial-cluster infra1=https://172.20.0.113:2380,infra2=https://172.20.0.114:2380,infra3=https://172.20.0.115:2380 \\ --initial-cluster-state new \\ --data-dir=${ETCD_DATA_DIR} Restart=on-failure RestartSec=5 LimitNOFILE=65536 [Install] WantedBy=multi-user.target 指定 etcd 的工作目录为 /var/lib/etcd，数据目录为 /var/lib/etcd，需在启动服务前创建这个目录，否则启动服务的时候会报错“Failed at step CHDIR spawning /usr/bin/etcd: No such file or directory”； 为了保证通信安全，需要指定 etcd 的公私钥(cert-file和key-file)、Peers 通信的公私钥和 CA 证书(peer-cert-file、peer-key-file、peer-trusted-ca-file)、客户端的CA证书（trusted-ca-file）； 创建 kubernetes.pem 证书时使用的 kubernetes-csr.json 文件的 hosts 字段包含所有 etcd 节点的IP，否则证书校验会出错； --initial-cluster-state 值为 new 时，--name 的参数值必须位于 --initial-cluster 列表中； 完整 unit 文件见：etcd.service 环境变量配置文件/etc/etcd/etcd.conf。 # [member] ETCD_NAME=infra1 ETCD_DATA_DIR=\"/var/lib/etcd\" ETCD_LISTEN_PEER_URLS=\"https://172.20.0.113:2380\" ETCD_LISTEN_CLIENT_URLS=\"https://172.20.0.113:2379\" #[cluster] ETCD_INITIAL_ADVERTISE_PEER_URLS=\"https://172.20.0.113:2380\" ETCD_INITIAL_CLUSTER_TOKEN=\"etcd-cluster\" ETCD_ADVERTISE_CLIENT_URLS=\"https://172.20.0.113:2379\" 这是172.20.0.113节点的配置，其他两个etcd节点只要将上面的IP地址改成相应节点的IP地址即可。ETCD_NAME换成对应节点的infra1/2/3。 启动 etcd 服务 mv etcd.service /usr/lib/system/systemd/ systemctl daemon-reload systemctl enable etcd systemctl start etcd systemctl status etcd 在所有的 kubernetes master 节点重复上面的步骤，直到所有机器的 etcd 服务都已启动。 验证服务 在任一 kubernetes master 机器上执行如下命令： $ etcdctl \\ --ca-file=/etc/kubernetes/ssl/ca.pem \\ --cert-file=/etc/kubernetes/ssl/kubernetes.pem \\ --key-file=/etc/kubernetes/ssl/kubernetes-key.pem \\ cluster-health 2017-04-11 15:17:09.082250 I | warning: ignoring ServerName for user-provided CA for backwards compatibility is deprecated 2017-04-11 15:17:09.083681 I | warning: ignoring ServerName for user-provided CA for backwards compatibility is deprecated member 9a2ec640d25672e5 is healthy: got healthy result from https://172.20.0.115:2379 member bc6f27ae3be34308 is healthy: got healthy result from https://172.20.0.114:2379 member e5c92ea26c4edba0 is healthy: got healthy result from https://172.20.0.113:2379 cluster is healthy 结果最后一行为 cluster is healthy 时表示集群服务正常。 更多资料 关于如何在etcd中查看kubernetes的数据，请参考使用etcdctl访问kuberentes数据。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-15 19:44:13 "},"practice/kubectl-installation.html":{"url":"practice/kubectl-installation.html","title":"安装kubectl命令行工具","keywords":"","body":"安装kubectl命令行工具 本文档介绍下载和配置 kubernetes 集群命令行工具 kubelet 的步骤。 下载 kubectl wget https://dl.k8s.io/v1.6.0/kubernetes-client-linux-amd64.tar.gz tar -xzvf kubernetes-client-linux-amd64.tar.gz cp kubernetes/client/bin/kube* /usr/bin/ chmod a+x /usr/bin/kube* 创建 kubectl kubeconfig 文件 export KUBE_APISERVER=\"https://172.20.0.113:6443\" # 设置集群参数 kubectl config set-cluster kubernetes \\ --certificate-authority=/etc/kubernetes/ssl/ca.pem \\ --embed-certs=true \\ --server=${KUBE_APISERVER} # 设置客户端认证参数 kubectl config set-credentials admin \\ --client-certificate=/etc/kubernetes/ssl/admin.pem \\ --embed-certs=true \\ --client-key=/etc/kubernetes/ssl/admin-key.pem # 设置上下文参数 kubectl config set-context kubernetes \\ --cluster=kubernetes \\ --user=admin # 设置默认上下文 kubectl config use-context kubernetes admin.pem 证书 OU 字段值为 system:masters，kube-apiserver 预定义的 RoleBinding cluster-admin 将 Group system:masters 与 Role cluster-admin 绑定，该 Role 授予了调用kube-apiserver 相关 API 的权限； 生成的 kubeconfig 被保存到 ~/.kube/config 文件； 注意：~/.kube/config文件拥有对该集群的最高权限，请妥善保管。 更多资料 kubectl命令概览 kubectl命令技巧大全 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-15 19:45:42 "},"practice/master-installation.html":{"url":"practice/master-installation.html","title":"部署master节点","keywords":"","body":"部署master节点 kubernetes master 节点包含的组件： kube-apiserver kube-scheduler kube-controller-manager 目前这三个组件需要部署在同一台机器上。 kube-scheduler、kube-controller-manager 和 kube-apiserver 三者的功能紧密相关； 同时只能有一个 kube-scheduler、kube-controller-manager 进程处于工作状态，如果运行多个，则需要通过选举产生一个 leader； 本文档记录部署一个三个节点的高可用 kubernetes master 集群步骤。（后续创建一个 load balancer 来代理访问 kube-apiserver 的请求） 注： 暂时未实现master节点的高可用 master节点上没有部署flannel网络插件，如果想要在master节点上也能访问ClusterIP，请参考下一节部署ndoe节点中的配置Flanneld部分。 TLS 证书文件 以下pem证书文件我们在创建TLS证书和秘钥这一步中已经创建过了，token.csv文件在创建kubeconfig文件的时候创建。我们再检查一下。 $ ls /etc/kubernetes/ssl admin-key.pem admin.pem ca-key.pem ca.pem kube-proxy-key.pem kube-proxy.pem kubernetes-key.pem kubernetes.pem 下载最新版本的二进制文件 有两种下载方式 方式一 从 github release 页面 下载发布版 tarball，解压后再执行下载脚本 wget https://github.com/kubernetes/kubernetes/releases/download/v1.6.0/kubernetes.tar.gz tar -xzvf kubernetes.tar.gz cd kubernetes ./cluster/get-kube-binaries.sh 方式二 从 CHANGELOG页面 下载 client 或 server tarball 文件 server 的 tarball kubernetes-server-linux-amd64.tar.gz 已经包含了 client(kubectl) 二进制文件，所以不用单独下载kubernetes-client-linux-amd64.tar.gz文件； # wget https://dl.k8s.io/v1.6.0/kubernetes-client-linux-amd64.tar.gz wget https://dl.k8s.io/v1.6.0/kubernetes-server-linux-amd64.tar.gz tar -xzvf kubernetes-server-linux-amd64.tar.gz cd kubernetes tar -xzvf kubernetes-src.tar.gz 将二进制文件拷贝到指定路径 cp -r server/bin/{kube-apiserver,kube-controller-manager,kube-scheduler,kubectl,kube-proxy,kubelet} /usr/local/bin/ 配置和启动 kube-apiserver 创建 kube-apiserver的service配置文件 service配置文件/usr/lib/systemd/system/kube-apiserver.service内容： [Unit] Description=Kubernetes API Service Documentation=https://github.com/GoogleCloudPlatform/kubernetes After=network.target After=etcd.service [Service] EnvironmentFile=-/etc/kubernetes/config EnvironmentFile=-/etc/kubernetes/apiserver ExecStart=/usr/local/bin/kube-apiserver \\ $KUBE_LOGTOSTDERR \\ $KUBE_LOG_LEVEL \\ $KUBE_ETCD_SERVERS \\ $KUBE_API_ADDRESS \\ $KUBE_API_PORT \\ $KUBELET_PORT \\ $KUBE_ALLOW_PRIV \\ $KUBE_SERVICE_ADDRESSES \\ $KUBE_ADMISSION_CONTROL \\ $KUBE_API_ARGS Restart=on-failure Type=notify LimitNOFILE=65536 [Install] WantedBy=multi-user.target /etc/kubernetes/config文件的内容为： ### # kubernetes system config # # The following values are used to configure various aspects of all # kubernetes services, including # # kube-apiserver.service # kube-controller-manager.service # kube-scheduler.service # kubelet.service # kube-proxy.service # logging to stderr means we get it in the systemd journal KUBE_LOGTOSTDERR=\"--logtostderr=true\" # journal message level, 0 is debug KUBE_LOG_LEVEL=\"--v=0\" # Should this cluster be allowed to run privileged docker containers KUBE_ALLOW_PRIV=\"--allow-privileged=true\" # How the controller-manager, scheduler, and proxy find the apiserver #KUBE_MASTER=\"--master=http://sz-pg-oam-docker-test-001.tendcloud.com:8080\" KUBE_MASTER=\"--master=http://172.20.0.113:8080\" 该配置文件同时被kube-apiserver、kube-controller-manager、kube-scheduler、kubelet、kube-proxy使用。 apiserver配置文件/etc/kubernetes/apiserver内容为： ### ## kubernetes system config ## ## The following values are used to configure the kube-apiserver ## # ## The address on the local server to listen to. #KUBE_API_ADDRESS=\"--insecure-bind-address=sz-pg-oam-docker-test-001.tendcloud.com\" KUBE_API_ADDRESS=\"--advertise-address=172.20.0.113 --bind-address=172.20.0.113 --insecure-bind-address=172.20.0.113\" # ## The port on the local server to listen on. #KUBE_API_PORT=\"--port=8080\" # ## Port minions listen on #KUBELET_PORT=\"--kubelet-port=10250\" # ## Comma separated list of nodes in the etcd cluster KUBE_ETCD_SERVERS=\"--etcd-servers=https://172.20.0.113:2379,https://172.20.0.114:2379,https://172.20.0.115:2379\" # ## Address range to use for services KUBE_SERVICE_ADDRESSES=\"--service-cluster-ip-range=10.254.0.0/16\" # ## default admission control policies KUBE_ADMISSION_CONTROL=\"--admission-control=ServiceAccount,NamespaceLifecycle,NamespaceExists,LimitRanger,ResourceQuota\" # ## Add your own! KUBE_API_ARGS=\"--authorization-mode=RBAC --runtime-config=rbac.authorization.k8s.io/v1beta1 --kubelet-https=true --experimental-bootstrap-token-auth --token-auth-file=/etc/kubernetes/token.csv --service-node-port-range=30000-32767 --tls-cert-file=/etc/kubernetes/ssl/kubernetes.pem --tls-private-key-file=/etc/kubernetes/ssl/kubernetes-key.pem --client-ca-file=/etc/kubernetes/ssl/ca.pem --service-account-key-file=/etc/kubernetes/ssl/ca-key.pem --etcd-cafile=/etc/kubernetes/ssl/ca.pem --etcd-certfile=/etc/kubernetes/ssl/kubernetes.pem --etcd-keyfile=/etc/kubernetes/ssl/kubernetes-key.pem --enable-swagger-ui=true --apiserver-count=3 --audit-log-maxage=30 --audit-log-maxbackup=3 --audit-log-maxsize=100 --audit-log-path=/var/lib/audit.log --event-ttl=1h\" --authorization-mode=RBAC 指定在安全端口使用 RBAC 授权模式，拒绝未通过授权的请求； kube-scheduler、kube-controller-manager 一般和 kube-apiserver 部署在同一台机器上，它们使用非安全端口和 kube-apiserver通信; kubelet、kube-proxy、kubectl 部署在其它 Node 节点上，如果通过安全端口访问 kube-apiserver，则必须先通过 TLS 证书认证，再通过 RBAC 授权； kube-proxy、kubectl 通过在使用的证书里指定相关的 User、Group 来达到通过 RBAC 授权的目的； 如果使用了 kubelet TLS Boostrap 机制，则不能再指定 --kubelet-certificate-authority、--kubelet-client-certificate 和 --kubelet-client-key 选项，否则后续 kube-apiserver 校验 kubelet 证书时出现 ”x509: certificate signed by unknown authority“ 错误； --admission-control 值必须包含 ServiceAccount； --bind-address 不能为 127.0.0.1； runtime-config配置为rbac.authorization.k8s.io/v1beta1，表示运行时的apiVersion； --service-cluster-ip-range 指定 Service Cluster IP 地址段，该地址段不能路由可达； 缺省情况下 kubernetes 对象保存在 etcd /registry 路径下，可以通过 --etcd-prefix 参数进行调整； 完整 unit 见 kube-apiserver.service 启动kube-apiserver systemctl daemon-reload systemctl enable kube-apiserver systemctl start kube-apiserver systemctl status kube-apiserver 配置和启动 kube-controller-manager 创建 kube-controller-manager的serivce配置文件 文件路径/usr/lib/systemd/system/kube-controller-manager.service [Unit] Description=Kubernetes Controller Manager Documentation=https://github.com/GoogleCloudPlatform/kubernetes [Service] EnvironmentFile=-/etc/kubernetes/config EnvironmentFile=-/etc/kubernetes/controller-manager ExecStart=/usr/local/bin/kube-controller-manager \\ $KUBE_LOGTOSTDERR \\ $KUBE_LOG_LEVEL \\ $KUBE_MASTER \\ $KUBE_CONTROLLER_MANAGER_ARGS Restart=on-failure LimitNOFILE=65536 [Install] WantedBy=multi-user.target 配置文件/etc/kubernetes/controller-manager。 ### # The following values are used to configure the kubernetes controller-manager # defaults from config and apiserver should be adequate # Add your own! KUBE_CONTROLLER_MANAGER_ARGS=\"--address=127.0.0.1 --service-cluster-ip-range=10.254.0.0/16 --cluster-name=kubernetes --cluster-signing-cert-file=/etc/kubernetes/ssl/ca.pem --cluster-signing-key-file=/etc/kubernetes/ssl/ca-key.pem --service-account-private-key-file=/etc/kubernetes/ssl/ca-key.pem --root-ca-file=/etc/kubernetes/ssl/ca.pem --leader-elect=true\" --service-cluster-ip-range 参数指定 Cluster 中 Service 的CIDR范围，该网络在各 Node 间必须路由不可达，必须和 kube-apiserver 中的参数一致； --cluster-signing-* 指定的证书和私钥文件用来签名为 TLS BootStrap 创建的证书和私钥； --root-ca-file 用来对 kube-apiserver 证书进行校验，指定该参数后，才会在Pod 容器的 ServiceAccount 中放置该 CA 证书文件； --address 值必须为 127.0.0.1，因为当前 kube-apiserver 期望 scheduler 和 controller-manager 在同一台机器，否则： $ kubectl get componentstatuses NAME STATUS MESSAGE ERROR scheduler Unhealthy Get http://127.0.0.1:10251/healthz: dial tcp 127.0.0.1:10251: getsockopt: connection refused controller-manager Healthy ok etcd-2 Healthy {\"health\": \"true\"} etcd-0 Healthy {\"health\": \"true\"} etcd-1 Healthy {\"health\": \"true\"} 如果有组件report unhealthy请参考：https://github.com/kubernetes-incubator/bootkube/issues/64 完整 unit 见 kube-controller-manager.service 启动 kube-controller-manager systemctl daemon-reload systemctl enable kube-controller-manager systemctl start kube-controller-manager 配置和启动 kube-scheduler 创建 kube-scheduler的serivce配置文件 文件路径/usr/lib/systemd/system/kube-scheduler.service。 [Unit] Description=Kubernetes Scheduler Plugin Documentation=https://github.com/GoogleCloudPlatform/kubernetes [Service] EnvironmentFile=-/etc/kubernetes/config EnvironmentFile=-/etc/kubernetes/scheduler ExecStart=/usr/local/bin/kube-scheduler \\ $KUBE_LOGTOSTDERR \\ $KUBE_LOG_LEVEL \\ $KUBE_MASTER \\ $KUBE_SCHEDULER_ARGS Restart=on-failure LimitNOFILE=65536 [Install] WantedBy=multi-user.target 配置文件/etc/kubernetes/scheduler。 ### # kubernetes scheduler config # default config should be adequate # Add your own! KUBE_SCHEDULER_ARGS=\"--leader-elect=true --address=127.0.0.1\" --address 值必须为 127.0.0.1，因为当前 kube-apiserver 期望 scheduler 和 controller-manager 在同一台机器； 完整 unit 见 kube-scheduler.service 启动 kube-scheduler systemctl daemon-reload systemctl enable kube-scheduler systemctl start kube-scheduler 验证 master 节点功能 $ kubectl get componentstatuses NAME STATUS MESSAGE ERROR scheduler Healthy ok controller-manager Healthy ok etcd-0 Healthy {\"health\": \"true\"} etcd-1 Healthy {\"health\": \"true\"} etcd-2 Healthy {\"health\": \"true\"} Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-15 15:17:07 "},"practice/flannel-installation.html":{"url":"practice/flannel-installation.html","title":"安装flannel网络插件","keywords":"","body":"安装flannel网络插件 所有的node节点都需要安装网络插件才能让所有的Pod加入到同一个局域网中，本文是安装flannel网络插件的参考文档。 建议直接使用yum安装flanneld，除非对版本有特殊需求，默认安装的是0.7.1版本的flannel。 yum install -y flannel service配置文件/usr/lib/systemd/system/flanneld.service。 [Unit] Description=Flanneld overlay address etcd agent After=network.target After=network-online.target Wants=network-online.target After=etcd.service Before=docker.service [Service] Type=notify EnvironmentFile=/etc/sysconfig/flanneld EnvironmentFile=-/etc/sysconfig/docker-network ExecStart=/usr/bin/flanneld-start \\ -etcd-endpoints=${ETCD_ENDPOINTS} \\ -etcd-prefix=${ETCD_PREFIX} \\ $FLANNEL_OPTIONS ExecStartPost=/usr/libexec/flannel/mk-docker-opts.sh -k DOCKER_NETWORK_OPTIONS -d /run/flannel/docker Restart=on-failure [Install] WantedBy=multi-user.target RequiredBy=docker.service /etc/sysconfig/flanneld配置文件： # Flanneld configuration options # etcd url location. Point this to the server where etcd runs ETCD_ENDPOINTS=\"https://172.20.0.113:2379,https://172.20.0.114:2379,https://172.20.0.115:2379\" # etcd config key. This is the configuration key that flannel queries # For address range assignment ETCD_PREFIX=\"/kube-centos/network\" # Any additional options that you want to pass FLANNEL_OPTIONS=\"-etcd-cafile=/etc/kubernetes/ssl/ca.pem -etcd-certfile=/etc/kubernetes/ssl/kubernetes.pem -etcd-keyfile=/etc/kubernetes/ssl/kubernetes-key.pem\" 在etcd中创建网络配置 执行下面的命令为docker分配IP地址段。 etcdctl --endpoints=https://172.20.0.113:2379,https://172.20.0.114:2379,https://172.20.0.115:2379 \\ --ca-file=/etc/kubernetes/ssl/ca.pem \\ --cert-file=/etc/kubernetes/ssl/kubernetes.pem \\ --key-file=/etc/kubernetes/ssl/kubernetes-key.pem \\ mkdir /kube-centos/network etcdctl --endpoints=https://172.20.0.113:2379,https://172.20.0.114:2379,https://172.20.0.115:2379 \\ --ca-file=/etc/kubernetes/ssl/ca.pem \\ --cert-file=/etc/kubernetes/ssl/kubernetes.pem \\ --key-file=/etc/kubernetes/ssl/kubernetes-key.pem \\ mk /kube-centos/network/config '{\"Network\":\"172.30.0.0/16\",\"SubnetLen\":24,\"Backend\":{\"Type\":\"vxlan\"}}' 如果你要使用host-gw模式，可以直接将vxlan改成host-gw即可。 注：参考网络和集群性能测试那节，最终我们使用的host-gw模式，关于flannel支持的backend模式见：https://github.com/coreos/flannel/blob/master/Documentation/backends.md。 启动flannel systemctl daemon-reload systemctl enable flanneld systemctl start flanneld systemctl status flanneld 现在查询etcd中的内容可以看到： $etcdctl --endpoints=${ETCD_ENDPOINTS} \\ --ca-file=/etc/kubernetes/ssl/ca.pem \\ --cert-file=/etc/kubernetes/ssl/kubernetes.pem \\ --key-file=/etc/kubernetes/ssl/kubernetes-key.pem \\ ls /kube-centos/network/subnets /kube-centos/network/subnets/172.30.14.0-24 /kube-centos/network/subnets/172.30.38.0-24 /kube-centos/network/subnets/172.30.46.0-24 $etcdctl --endpoints=${ETCD_ENDPOINTS} \\ --ca-file=/etc/kubernetes/ssl/ca.pem \\ --cert-file=/etc/kubernetes/ssl/kubernetes.pem \\ --key-file=/etc/kubernetes/ssl/kubernetes-key.pem \\ get /kube-centos/network/config { \"Network\": \"172.30.0.0/16\", \"SubnetLen\": 24, \"Backend\": { \"Type\": \"vxlan\" } } $etcdctl get /kube-centos/network/subnets/172.30.14.0-24 {\"PublicIP\":\"172.20.0.114\",\"BackendType\":\"vxlan\",\"BackendData\":{\"VtepMAC\":\"56:27:7d:1c:08:22\"}} $etcdctl get /kube-centos/network/subnets/172.30.38.0-24 {\"PublicIP\":\"172.20.0.115\",\"BackendType\":\"vxlan\",\"BackendData\":{\"VtepMAC\":\"12:82:83:59:cf:b8\"}} $etcdctl get /kube-centos/network/subnets/172.30.46.0-24 {\"PublicIP\":\"172.20.0.113\",\"BackendType\":\"vxlan\",\"BackendData\":{\"VtepMAC\":\"e6:b2:fd:f6:66:96\"}} 如果可以查看到以上内容证明flannel已经安装完成，下一步是在node节点上安装和配置docker、kubelet、kube-proxy等，请参考下一节部署node节点。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-19 17:48:43 "},"practice/node-installation.html":{"url":"practice/node-installation.html","title":"部署node节点","keywords":"","body":"部署node节点 Kubernetes node节点包含如下组件： Flanneld：参考我之前写的文章Kubernetes基于Flannel的网络配置，之前没有配置TLS，现在需要在service配置文件中增加TLS配置，安装过程请参考上一节安装flannel网络插件。 Docker1.12.5：docker的安装很简单，这里也不说了，但是需要注意docker的配置。 kubelet：直接用二进制文件安装 kube-proxy：直接用二进制文件安装 注意：每台 node 上都需要安装 flannel，master 节点上可以不安装。 步骤简介 确认在上一步中我们安装配置的网络插件flannel已启动且运行正常 安装配置docker后启动 安装配置kubelet、kube-proxy后启动 验证 目录和文件 我们再检查一下三个节点上，经过前几步操作我们已经创建了如下的证书和配置文件。 $ ls /etc/kubernetes/ssl admin-key.pem admin.pem ca-key.pem ca.pem kube-proxy-key.pem kube-proxy.pem kubernetes-key.pem kubernetes.pem $ ls /etc/kubernetes/ apiserver bootstrap.kubeconfig config controller-manager kubelet kube-proxy.kubeconfig proxy scheduler ssl token.csv 配置Docker 如果您使用yum的方式安装的flannel则不需要执行mk-docker-opts.sh文件这一步，参考Flannel官方文档中的Docker Integration。 如果你不是使用yum安装的flannel，那么需要下载flannel github release中的tar包，解压后会获得一个mk-docker-opts.sh文件，到flannel release页面下载对应版本的安装包，该脚本见mk-docker-opts.sh，因为我们使用yum安装所以不需要执行这一步。 这个文件是用来Generate Docker daemon options based on flannel env file。 使用systemctl命令启动flanneld后，会自动执行./mk-docker-opts.sh -i生成如下两个文件环境变量文件： /run/flannel/subnet.env FLANNEL_NETWORK=172.30.0.0/16 FLANNEL_SUBNET=172.30.46.1/24 FLANNEL_MTU=1450 FLANNEL_IPMASQ=false /run/docker_opts.env DOCKER_OPT_BIP=\"--bip=172.30.46.1/24\" DOCKER_OPT_IPMASQ=\"--ip-masq=true\" DOCKER_OPT_MTU=\"--mtu=1450\" Docker将会读取这两个环境变量文件作为容器启动参数。 注意：不论您用什么方式安装的flannel，下面这一步是必不可少的。 yum方式安装的flannel 修改docker的配置文件/usr/lib/systemd/system/docker.service，增加一条环境变量配置： EnvironmentFile=-/run/flannel/docker /run/flannel/docker文件是flannel启动后自动生成的，其中包含了docker启动时需要的参数。 二进制方式安装的flannel 修改docker的配置文件/usr/lib/systemd/system/docker.service，增加如下几条环境变量配置： EnvironmentFile=-/run/docker_opts.env EnvironmentFile=-/run/flannel/subnet.env 这两个文件是mk-docker-opts.sh脚本生成环境变量文件默认的保存位置，docker启动的时候需要加载这几个配置文件才可以加入到flannel创建的虚拟网络里。 所以不论您使用何种方式安装的flannel，将以下配置加入到docker.service中可确保万无一失。 EnvironmentFile=-/run/flannel/docker EnvironmentFile=-/run/docker_opts.env EnvironmentFile=-/run/flannel/subnet.env EnvironmentFile=-/etc/sysconfig/docker EnvironmentFile=-/etc/sysconfig/docker-storage EnvironmentFile=-/etc/sysconfig/docker-network EnvironmentFile=-/run/docker_opts.env 请参考docker.service中的配置。 启动docker 重启了docker后还要重启kubelet，这时又遇到问题，kubelet启动失败。报错： Mar 31 16:44:41 sz-pg-oam-docker-test-002.tendcloud.com kubelet[81047]: error: failed to run Kubelet: failed to create kubelet: misconfiguration: kubelet cgroup driver: \"cgroupfs\" is different from docker cgroup driver: \"systemd\" 这是kubelet与docker的cgroup driver不一致导致的，kubelet启动的时候有个—cgroup-driver参数可以指定为\"cgroupfs\"或者“systemd”。 --cgroup-driver string Driver that the kubelet uses to manipulate cgroups on the host. Possible values: 'cgroupfs', 'systemd' (default \"cgroupfs\") 配置docker的service配置文件/usr/lib/systemd/system/docker.service，设置ExecStart中的--exec-opt native.cgroupdriver=systemd。 安装和配置kubelet kubernets1.8 相对于kubernetes1.6集群必须进行的配置有： 对于kuberentes1.8集群，必须关闭swap，否则kubelet启动将失败。 修改/etc/fstab将，swap系统注释掉。 kubelet 启动时向 kube-apiserver 发送 TLS bootstrapping 请求，需要先将 bootstrap token 文件中的 kubelet-bootstrap 用户赋予 system:node-bootstrapper cluster 角色(role)， 然后 kubelet 才能有权限创建认证请求(certificate signing requests)： cd /etc/kubernetes kubectl create clusterrolebinding kubelet-bootstrap \\ --clusterrole=system:node-bootstrapper \\ --user=kubelet-bootstrap --user=kubelet-bootstrap 是在 /etc/kubernetes/token.csv 文件中指定的用户名，同时也写入了 /etc/kubernetes/bootstrap.kubeconfig 文件； 下载最新的kubelet和kube-proxy二进制文件 wget https://dl.k8s.io/v1.6.0/kubernetes-server-linux-amd64.tar.gz tar -xzvf kubernetes-server-linux-amd64.tar.gz cd kubernetes tar -xzvf kubernetes-src.tar.gz cp -r ./server/bin/{kube-proxy,kubelet} /usr/local/bin/ 创建kubelet的service配置文件 文件位置/usr/lib/systemd/system/kubelet.service。 [Unit] Description=Kubernetes Kubelet Server Documentation=https://github.com/GoogleCloudPlatform/kubernetes After=docker.service Requires=docker.service [Service] WorkingDirectory=/var/lib/kubelet EnvironmentFile=-/etc/kubernetes/config EnvironmentFile=-/etc/kubernetes/kubelet ExecStart=/usr/local/bin/kubelet \\ $KUBE_LOGTOSTDERR \\ $KUBE_LOG_LEVEL \\ $KUBELET_API_SERVER \\ $KUBELET_ADDRESS \\ $KUBELET_PORT \\ $KUBELET_HOSTNAME \\ $KUBE_ALLOW_PRIV \\ $KUBELET_POD_INFRA_CONTAINER \\ $KUBELET_ARGS Restart=on-failure [Install] WantedBy=multi-user.target kubelet的配置文件/etc/kubernetes/kubelet。其中的IP地址更改为你的每台node节点的IP地址。 注意：在启动kubelet之前，需要先手动创建/var/lib/kubelet目录。 下面是kubelet的配置文件/etc/kubernetes/kubelet: kubernetes1.8 相对于kubenrete1.6的配置变动： 对于kuberentes1.8集群中的kubelet配置，取消了KUBELET_API_SERVER的配置，而改用kubeconfig文件来定义master地址。 ### ## kubernetes kubelet (minion) config # ## The address for the info server to serve on (set to 0.0.0.0 or \"\" for all interfaces) KUBELET_ADDRESS=\"--address=172.20.0.113\" # ## The port for the info server to serve on #KUBELET_PORT=\"--port=10250\" # ## You may leave this blank to use the actual hostname KUBELET_HOSTNAME=\"--hostname-override=172.20.0.113\" # ## location of the api-server KUBELET_API_SERVER=\"--api-servers=http://172.20.0.113:8080\" # ## pod infrastructure container KUBELET_POD_INFRA_CONTAINER=\"--pod-infra-container-image=sz-pg-oam-docker-hub-001.tendcloud.com/library/pod-infrastructure:rhel7\" # ## Add your own! KUBELET_ARGS=\"--cgroup-driver=systemd --cluster-dns=10.254.0.2 --experimental-bootstrap-kubeconfig=/etc/kubernetes/bootstrap.kubeconfig --kubeconfig=/etc/kubernetes/kubelet.kubeconfig --require-kubeconfig --cert-dir=/etc/kubernetes/ssl --cluster-domain=cluster.local --hairpin-mode promiscuous-bridge --serialize-image-pulls=false\" --address 不能设置为 127.0.0.1，否则后续 Pods 访问 kubelet 的 API 接口时会失败，因为 Pods 访问的 127.0.0.1 指向自己而不是 kubelet； 如果设置了 --hostname-override 选项，则 kube-proxy 也需要设置该选项，否则会出现找不到 Node 的情况； \"--cgroup-driver 配置成 systemd，不要使用cgroup，否则在 CentOS 系统中 kubelet 将启动失败（保持docker和kubelet中的cgroup driver配置一致即可，不一定非使用systemd）。 --experimental-bootstrap-kubeconfig 指向 bootstrap kubeconfig 文件，kubelet 使用该文件中的用户名和 token 向 kube-apiserver 发送 TLS Bootstrapping 请求； 管理员通过了 CSR 请求后，kubelet 自动在 --cert-dir 目录创建证书和私钥文件(kubelet-client.crt 和 kubelet-client.key)，然后写入 --kubeconfig 文件； 建议在 --kubeconfig 配置文件中指定 kube-apiserver 地址，如果未指定 --api-servers 选项，则必须指定 --require-kubeconfig 选项后才从配置文件中读取 kube-apiserver 的地址，否则 kubelet 启动后将找不到 kube-apiserver (日志中提示未找到 API Server），kubectl get nodes 不会返回对应的 Node 信息; --cluster-dns 指定 kubedns 的 Service IP(可以先分配，后续创建 kubedns 服务时指定该 IP)，--cluster-domain 指定域名后缀，这两个参数同时指定后才会生效； --cluster-domain 指定 pod 启动时 /etc/resolve.conf 文件中的 search domain ，起初我们将其配置成了 cluster.local.，这样在解析 service 的 DNS 名称时是正常的，可是在解析 headless service 中的 FQDN pod name 的时候却错误，因此我们将其修改为 cluster.local，去掉嘴后面的 ”点号“ 就可以解决该问题，关于 kubernetes 中的域名/服务名称解析请参见我的另一篇文章。 --kubeconfig=/etc/kubernetes/kubelet.kubeconfig中指定的kubelet.kubeconfig文件在第一次启动kubelet之前并不存在，请看下文，当通过CSR请求后会自动生成kubelet.kubeconfig文件，如果你的节点上已经生成了~/.kube/config文件，你可以将该文件拷贝到该路径下，并重命名为kubelet.kubeconfig，所有node节点可以共用同一个kubelet.kubeconfig文件，这样新添加的节点就不需要再创建CSR请求就能自动添加到kubernetes集群中。同样，在任意能够访问到kubernetes集群的主机上使用kubectl --kubeconfig命令操作集群时，只要使用~/.kube/config文件就可以通过权限认证，因为这里面已经有认证信息并认为你是admin用户，对集群拥有所有权限。 KUBELET_POD_INFRA_CONTAINER 是基础镜像容器，这里我用的是私有镜像仓库地址，大家部署的时候需要修改为自己的镜像。我上传了一个到时速云上，可以直接 docker pull index.tenxcloud.com/jimmy/pod-infrastructure 下载。 完整 unit 见 kubelet.service 启动kublet systemctl daemon-reload systemctl enable kubelet systemctl start kubelet systemctl status kubelet 通过 kublet 的 TLS 证书请求 kubelet 首次启动时向 kube-apiserver 发送证书签名请求，必须通过后 kubernetes 系统才会将该 Node 加入到集群。 查看未授权的 CSR 请求 $ kubectl get csr NAME AGE REQUESTOR CONDITION csr-2b308 4m kubelet-bootstrap Pending $ kubectl get nodes No resources found. 通过 CSR 请求 $ kubectl certificate approve csr-2b308 certificatesigningrequest \"csr-2b308\" approved $ kubectl get nodes NAME STATUS AGE VERSION 10.64.3.7 Ready 49m v1.6.1 自动生成了 kubelet kubeconfig 文件和公私钥 $ ls -l /etc/kubernetes/kubelet.kubeconfig -rw------- 1 root root 2284 Apr 7 02:07 /etc/kubernetes/kubelet.kubeconfig $ ls -l /etc/kubernetes/ssl/kubelet* -rw-r--r-- 1 root root 1046 Apr 7 02:07 /etc/kubernetes/ssl/kubelet-client.crt -rw------- 1 root root 227 Apr 7 02:04 /etc/kubernetes/ssl/kubelet-client.key -rw-r--r-- 1 root root 1103 Apr 7 02:07 /etc/kubernetes/ssl/kubelet.crt -rw------- 1 root root 1675 Apr 7 02:07 /etc/kubernetes/ssl/kubelet.key 假如你更新kubernetes的证书，只要没有更新token.csv，当重启kubelet后，该node就会自动加入到kuberentes集群中，而不会重新发送certificaterequest，也不需要在master节点上执行kubectl certificate approve操作。前提是不要删除node节点上的/etc/kubernetes/ssl/kubelet*和/etc/kubernetes/kubelet.kubeconfig文件。否则kubelet启动时会提示找不到证书而失败。 注意：如果启动kubelet的时候见到证书相关的报错，有个trick可以解决这个问题，可以将master节点上的~/.kube/config文件（该文件在安装kubectl命令行工具这一步中将会自动生成）拷贝到node节点的/etc/kubernetes/kubelet.kubeconfig位置，这样就不需要通过CSR，当kubelet启动后就会自动加入的集群中。 配置 kube-proxy 安装conntrack yum install -y conntrack-tools 创建 kube-proxy 的service配置文件 文件路径/usr/lib/systemd/system/kube-proxy.service。 [Unit] Description=Kubernetes Kube-Proxy Server Documentation=https://github.com/GoogleCloudPlatform/kubernetes After=network.target [Service] EnvironmentFile=-/etc/kubernetes/config EnvironmentFile=-/etc/kubernetes/proxy ExecStart=/usr/local/bin/kube-proxy \\ $KUBE_LOGTOSTDERR \\ $KUBE_LOG_LEVEL \\ $KUBE_MASTER \\ $KUBE_PROXY_ARGS Restart=on-failure LimitNOFILE=65536 [Install] WantedBy=multi-user.target kube-proxy配置文件/etc/kubernetes/proxy。 ### # kubernetes proxy config # default config should be adequate # Add your own! KUBE_PROXY_ARGS=\"--bind-address=172.20.0.113 --hostname-override=172.20.0.113 --kubeconfig=/etc/kubernetes/kube-proxy.kubeconfig --cluster-cidr=10.254.0.0/16\" --hostname-override 参数值必须与 kubelet 的值一致，否则 kube-proxy 启动后会找不到该 Node，从而不会创建任何 iptables 规则； kube-proxy 根据 --cluster-cidr 判断集群内部和外部流量，指定 --cluster-cidr 或 --masquerade-all 选项后 kube-proxy 才会对访问 Service IP 的请求做 SNAT； --kubeconfig 指定的配置文件嵌入了 kube-apiserver 的地址、用户名、证书、秘钥等请求和认证信息； 预定义的 RoleBinding cluster-admin 将User system:kube-proxy 与 Role system:node-proxier 绑定，该 Role 授予了调用 kube-apiserver Proxy 相关 API 的权限； 完整 unit 见 kube-proxy.service 启动 kube-proxy systemctl daemon-reload systemctl enable kube-proxy systemctl start kube-proxy systemctl status kube-proxy 验证测试 我们创建一个nginx的service试一下集群是否可用。 $ kubectl run nginx --replicas=2 --labels=\"run=load-balancer-example\" --image=sz-pg-oam-docker-hub-001.tendcloud.com/library/nginx:1.9 --port=80 deployment \"nginx\" created $ kubectl expose deployment nginx --type=NodePort --name=example-service service \"example-service\" exposed $ kubectl describe svc example-service Name: example-service Namespace: default Labels: run=load-balancer-example Annotations: Selector: run=load-balancer-example Type: NodePort IP: 10.254.62.207 Port: 80/TCP NodePort: 32724/TCP Endpoints: 172.30.60.2:80,172.30.94.2:80 Session Affinity: None Events: $ curl \"10.254.62.207:80\" Welcome to nginx! body { width: 35em; margin: 0 auto; font-family: Tahoma, Verdana, Arial, sans-serif; } Welcome to nginx! If you see this page, the nginx web server is successfully installed and working. Further configuration is required. For online documentation and support please refer to nginx.org. Commercial support is available at nginx.com. Thank you for using nginx. 提示：上面的测试示例中使用的nginx是我的私有镜像仓库中的镜像sz-pg-oam-docker-hub-001.tendcloud.com/library/nginx:1.9，大家在测试过程中请换成自己的nginx镜像地址。 访问172.20.0.113:32724或172.20.0.114:32724或者172.20.0.115:32724都可以得到nginx的页面。 图片 - welcome nginx 参考 Kubelet 的认证授权 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-19 17:45:18 "},"practice/kubedns-addon-installation.html":{"url":"practice/kubedns-addon-installation.html","title":"安装kubedns插件","keywords":"","body":"安装kubedns插件 官方的yaml文件目录：kubernetes/cluster/addons/dns。 该插件直接使用kubernetes部署，官方的配置文件中包含以下镜像： gcr.io/google_containers/k8s-dns-dnsmasq-nanny-amd64:1.14.1 gcr.io/google_containers/k8s-dns-kube-dns-amd64:1.14.1 gcr.io/google_containers/k8s-dns-sidecar-amd64:1.14.1 我clone了上述镜像，上传到我的私有镜像仓库： sz-pg-oam-docker-hub-001.tendcloud.com/library/k8s-dns-dnsmasq-nanny-amd64:1.14.1 sz-pg-oam-docker-hub-001.tendcloud.com/library/k8s-dns-kube-dns-amd64:1.14.1 sz-pg-oam-docker-hub-001.tendcloud.com/library/k8s-dns-sidecar-amd64:1.14.1 同时上传了一份到时速云备份： index.tenxcloud.com/jimmy/k8s-dns-dnsmasq-nanny-amd64:1.14.1 index.tenxcloud.com/jimmy/k8s-dns-kube-dns-amd64:1.14.1 index.tenxcloud.com/jimmy/k8s-dns-sidecar-amd64:1.14.1 以下yaml配置文件中使用的是私有镜像仓库中的镜像。 kubedns-cm.yaml kubedns-sa.yaml kubedns-controller.yaml kubedns-svc.yaml 已经修改好的 yaml 文件见：../manifests/kubedns 系统预定义的 RoleBinding 预定义的 RoleBinding system:kube-dns 将 kube-system 命名空间的 kube-dns ServiceAccount 与 system:kube-dns Role 绑定， 该 Role 具有访问 kube-apiserver DNS 相关 API 的权限； $ kubectl get clusterrolebindings system:kube-dns -o yaml apiVersion: rbac.authorization.k8s.io/v1beta1 kind: ClusterRoleBinding metadata: annotations: rbac.authorization.kubernetes.io/autoupdate: \"true\" creationTimestamp: 2017-04-11T11:20:42Z labels: kubernetes.io/bootstrapping: rbac-defaults name: system:kube-dns resourceVersion: \"58\" selfLink: /apis/rbac.authorization.k8s.io/v1beta1/clusterrolebindingssystem%3Akube-dns uid: e61f4d92-1ea8-11e7-8cd7-f4e9d49f8ed0 roleRef: apiGroup: rbac.authorization.k8s.io kind: ClusterRole name: system:kube-dns subjects: - kind: ServiceAccount name: kube-dns namespace: kube-system kubedns-controller.yaml 中定义的 Pods 时使用了 kubedns-sa.yaml 文件定义的 kube-dns ServiceAccount，所以具有访问 kube-apiserver DNS 相关 API 的权限。 配置 kube-dns ServiceAccount 无需修改。 配置 kube-dns 服务 $ diff kubedns-svc.yaml.base kubedns-svc.yaml 30c30 clusterIP: 10.254.0.2 spec.clusterIP = 10.254.0.2，即明确指定了 kube-dns Service IP，这个 IP 需要和 kubelet 的 --cluster-dns 参数值一致； 配置 kube-dns Deployment $ diff kubedns-controller.yaml.base kubedns-controller.yaml 58c58 image: sz-pg-oam-docker-hub-001.tendcloud.com/library/k8s-dns-kube-dns-amd64:v1.14.1 88c88 - --domain=cluster.local. 92c92 #__PILLAR__FEDERATIONS__DOMAIN__MAP__ 110c110 image: sz-pg-oam-docker-hub-001.tendcloud.com/library/k8s-dns-dnsmasq-nanny-amd64:v1.14.1 129c129 - --server=/cluster.local./127.0.0.1#10053 148c148 image: sz-pg-oam-docker-hub-001.tendcloud.com/library/k8s-dns-sidecar-amd64:v1.14.1 161,162c161,162 - --probe=kubedns,127.0.0.1:10053,kubernetes.default.svc.cluster.local.,5,A > - --probe=dnsmasq,127.0.0.1:53,kubernetes.default.svc.cluster.local.,5,A 使用系统已经做了 RoleBinding 的 kube-dns ServiceAccount，该账户具有访问 kube-apiserver DNS 相关 API 的权限； 执行所有定义文件 $ pwd /root/kubedns $ ls *.yaml kubedns-cm.yaml kubedns-controller.yaml kubedns-sa.yaml kubedns-svc.yaml $ kubectl create -f . 检查 kubedns 功能 新建一个 Deployment $ cat my-nginx.yaml apiVersion: extensions/v1beta1 kind: Deployment metadata: name: my-nginx spec: replicas: 2 template: metadata: labels: run: my-nginx spec: containers: - name: my-nginx image: sz-pg-oam-docker-hub-001.tendcloud.com/library/nginx:1.9 ports: - containerPort: 80 $ kubectl create -f my-nginx.yaml Export 该 Deployment, 生成 my-nginx 服务 $ kubectl expose deploy my-nginx $ kubectl get services --all-namespaces |grep my-nginx default my-nginx 10.254.179.239 80/TCP 42m 创建另一个 Pod，查看 /etc/resolv.conf 是否包含 kubelet 配置的 --cluster-dns 和 --cluster-domain，是否能够将服务 my-nginx 解析到 Cluster IP 10.254.179.239。 $ kubectl create -f nginx-pod.yaml $ kubectl exec nginx -i -t -- /bin/bash root@nginx:/# cat /etc/resolv.conf nameserver 10.254.0.2 search default.svc.cluster.local. svc.cluster.local. cluster.local. tendcloud.com options ndots:5 root@nginx:/# ping my-nginx PING my-nginx.default.svc.cluster.local (10.254.179.239): 56 data bytes 76 bytes from 119.147.223.109: Destination Net Unreachable ^C--- my-nginx.default.svc.cluster.local ping statistics --- root@nginx:/# ping kubernetes PING kubernetes.default.svc.cluster.local (10.254.0.1): 56 data bytes ^C--- kubernetes.default.svc.cluster.local ping statistics --- 11 packets transmitted, 0 packets received, 100% packet loss root@nginx:/# ping kube-dns.kube-system.svc.cluster.local PING kube-dns.kube-system.svc.cluster.local (10.254.0.2): 56 data bytes ^C--- kube-dns.kube-system.svc.cluster.local ping statistics --- 6 packets transmitted, 0 packets received, 100% packet loss 从结果来看，service名称可以正常解析。 注意：直接ping ClusterIP是ping不通的，ClusterIP是根据IPtables路由到服务的endpoint上，只有结合ClusterIP加端口才能访问到对应的服务。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-10-12 10:11:37 "},"practice/dashboard-addon-installation.html":{"url":"practice/dashboard-addon-installation.html","title":"安装dashboard插件","keywords":"","body":"安装dashboard插件 官方文件目录：kubernetes/cluster/addons/dashboard 我们使用的文件如下： $ ls *.yaml dashboard-controller.yaml dashboard-service.yaml dashboard-rbac.yaml 已经修改好的 yaml 文件见：../manifests/dashboard 由于 kube-apiserver 启用了 RBAC 授权，而官方源码目录的 dashboard-controller.yaml 没有定义授权的 ServiceAccount，所以后续访问 API server 的 API 时会被拒绝，web中提示： Forbidden (403) User \"system:serviceaccount:kube-system:default\" cannot list jobs.batch in the namespace \"default\". (get jobs.batch) 增加了一个dashboard-rbac.yaml文件，定义一个名为 dashboard 的 ServiceAccount，然后将它和 Cluster Role view 绑定。 配置dashboard-service $ diff dashboard-service.yaml.orig dashboard-service.yaml 10a11 > type: NodePort 指定端口类型为 NodePort，这样外界可以通过地址 nodeIP:nodePort 访问 dashboard； 配置dashboard-controller $ diff dashboard-controller.yaml.orig dashboard-controller.yaml 23c23 image: sz-pg-oam-docker-hub-001.tendcloud.com/library/kubernetes-dashboard-amd64:v1.6.0 执行所有定义文件 $ pwd /root/kubernetes/cluster/addons/dashboard $ ls *.yaml dashboard-controller.yaml dashboard-service.yaml $ kubectl create -f . service \"kubernetes-dashboard\" created deployment \"kubernetes-dashboard\" created 检查执行结果 查看分配的 NodePort $ kubectl get services kubernetes-dashboard -n kube-system NAME CLUSTER-IP EXTERNAL-IP PORT(S) AGE kubernetes-dashboard 10.254.224.130 80:30312/TCP 25s NodePort 30312映射到 dashboard pod 80端口； 检查 controller $ kubectl get deployment kubernetes-dashboard -n kube-system NAME DESIRED CURRENT UP-TO-DATE AVAILABLE AGE kubernetes-dashboard 1 1 1 1 3m $ kubectl get pods -n kube-system | grep dashboard kubernetes-dashboard-1339745653-pmn6z 1/1 Running 0 4m 访问dashboard 有以下三种方式： kubernetes-dashboard 服务暴露了 NodePort，可以使用 http://NodeIP:nodePort 地址访问 dashboard 通过 API server 访问 dashboard（https 6443端口和http 8080端口方式） 通过 kubectl proxy 访问 dashboard 通过 kubectl proxy 访问 dashboard 启动代理 $ kubectl proxy --address='172.20.0.113' --port=8086 --accept-hosts='^*$' Starting to serve on 172.20.0.113:8086 需要指定 --accept-hosts 选项，否则浏览器访问 dashboard 页面时提示 “Unauthorized”； 浏览器访问 URL：http://172.20.0.113:8086/ui 自动跳转到：http://172.20.0.113:8086/api/v1/proxy/namespaces/kube-system/services/kubernetes-dashboard/#/workload?namespace=default 通过 API server 访问dashboard 获取集群服务地址列表 $ kubectl cluster-info Kubernetes master is running at https://172.20.0.113:6443 KubeDNS is running at https://172.20.0.113:6443/api/v1/proxy/namespaces/kube-system/services/kube-dns kubernetes-dashboard is running at https://172.20.0.113:6443/api/v1/proxy/namespaces/kube-system/services/kubernetes-dashboard 浏览器访问 URL：https://172.20.0.113:6443/api/v1/proxy/namespaces/kube-system/services/kubernetes-dashboard（浏览器会提示证书验证，因为通过加密通道，以改方式访问的话，需要提前导入证书到你的计算机中）。这是我当时在这遇到的坑：[通过 kube-apiserver 访问dashboard，提示User \"system:anonymous\" cannot proxy services in the namespace \"kube-system\". #5](https://github.com/opsnull/follow-me-install-kubernetes-cluster/issues/5)，已经解决。 导入证书 将生成的admin.pem证书转换格式 openssl pkcs12 -export -in admin.pem -out admin.p12 -inkey admin-key.pem 将生成的admin.p12证书导入的你的电脑，导出的时候记住你设置的密码，导入的时候还要用到。 如果你不想使用https的话，可以直接访问insecure port 8080端口：http://172.20.0.113:8080/api/v1/proxy/namespaces/kube-system/services/kubernetes-dashboard 图片 - kubernetes dashboard 由于缺少 Heapster 插件，当前 dashboard 不能展示 Pod、Nodes 的 CPU、内存等 metric 图形。 更新 Kubernetes 1.6 版本的 dashboard 的镜像已经到了 v1.6.3 版本，我们可以使用下面的方式更新。 修改 dashboard-controller.yaml 文件中的镜像的版本将 v1.6.0 更改为 v1.6.3。 image: sz-pg-oam-docker-hub-001.tendcloud.com/library/kubernetes-dashboard-amd64:v1.6.3 然后执行下面的命令： kubectl apply -f dashboard-controller.yaml 即可在线更新 dashboard 的版本。 监听 dashboard Pod 的状态可以看到： kubernetes-dashboard-215087767-2jsgd 0/1 Pending 0 0s kubernetes-dashboard-3966630548-0jj1j 1/1 Terminating 0 1d kubernetes-dashboard-215087767-2jsgd 0/1 Pending 0 0s kubernetes-dashboard-3966630548-0jj1j 1/1 Terminating 0 1d kubernetes-dashboard-215087767-2jsgd 0/1 ContainerCreating 0 0s kubernetes-dashboard-3966630548-0jj1j 0/1 Terminating 0 1d kubernetes-dashboard-3966630548-0jj1j 0/1 Terminating 0 1d kubernetes-dashboard-215087767-2jsgd 1/1 Running 0 6s kubernetes-dashboard-3966630548-0jj1j 0/1 Terminating 0 1d kubernetes-dashboard-3966630548-0jj1j 0/1 Terminating 0 1d kubernetes-dashboard-3966630548-0jj1j 0/1 Terminating 0 1d 新的 Pod 的启动了，旧的 Pod 被终结了。 Dashboard 的访问地址不变，重新访问 http://172.20.0.113:8080/api/v1/proxy/namespaces/kube-system/services/kubernetes-dashboard，可以看到新版的界面： 图片 - V1.6.3版本的dashboard界面 新版本中最大的变化是增加了进入容器内部的入口，可以在页面上进入到容器内部操作，同时又增加了一个搜索框。 关于如何将dashboard从1.6版本升级到1.7版本请参考升级dashboard。 参考 WebUI(Dashboard) 文档 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-15 21:36:09 "},"practice/heapster-addon-installation.html":{"url":"practice/heapster-addon-installation.html","title":"安装heapster插件","keywords":"","body":"安装heapster插件 准备镜像 官方镜像保存在 gcr.io 中需要翻墙才能下载，为了方便大家使用，我下载后放到了时速云中，为公有镜像供大家下载。 index.tenxcloud.com/jimmy/heapster-amd64:v1.3.0-beta.1 index.tenxcloud.com/jimmy/heapster-influxdb-amd64:v1.1.1 index.tenxcloud.com/jimmy/heapster-grafana-amd64:v4.0.2 准备YAML文件 到 heapster release 页面 下载最新版本的 heapster。 wget https://github.com/kubernetes/heapster/archive/v1.3.0.zip unzip v1.3.0.zip mv v1.3.0.zip heapster-1.3.0 文件目录： heapster-1.3.0/deploy/kube-config/influxdb $ cd heapster-1.3.0/deploy/kube-config/influxdb $ ls *.yaml grafana-deployment.yaml grafana-service.yaml heapster-deployment.yaml heapster-service.yaml influxdb-deployment.yaml influxdb-service.yaml heapster-rbac.yaml 我们自己创建了heapster的rbac配置heapster-rbac.yaml。 已经修改好的 yaml 文件见：../manifests/heapster 配置 grafana-deployment $ diff grafana-deployment.yaml.orig grafana-deployment.yaml 16c16 image: sz-pg-oam-docker-hub-001.tendcloud.com/library/heapster-grafana-amd64:v4.0.2 40,41c40,41 value: /api/v1/proxy/namespaces/kube-system/services/monitoring-grafana/ > #value: / 如果后续使用 kube-apiserver 或者 kubectl proxy 访问 grafana dashboard，则必须将 GF_SERVER_ROOT_URL 设置为 /api/v1/proxy/namespaces/kube-system/services/monitoring-grafana/，否则后续访问grafana时访问时提示找不到http://172.20.0.113:8086/api/v1/proxy/namespaces/kube-system/services/monitoring-grafana/api/dashboards/home 页面； 配置 heapster-deployment $ diff heapster-deployment.yaml.orig heapster-deployment.yaml 16c16 image: sz-pg-oam-docker-hub-001.tendcloud.com/library/heapster-amd64:v1.3.0-beta.1 配置 influxdb-deployment influxdb 官方建议使用命令行或 HTTP API 接口来查询数据库，从 v1.1.0 版本开始默认关闭 admin UI，将在后续版本中移除 admin UI 插件。 开启镜像中 admin UI的办法如下：先导出镜像中的 influxdb 配置文件，开启 admin 插件后，再将配置文件内容写入 ConfigMap，最后挂载到镜像中，达到覆盖原始配置的目的： 注意：manifests 目录已经提供了 修改后的 ConfigMap 定义文件 $ # 导出镜像中的 influxdb 配置文件 $ docker run --rm --entrypoint 'cat' -ti lvanneo/heapster-influxdb-amd64:v1.1.1 /etc/config.toml >config.toml.orig $ cp config.toml.orig config.toml $ # 修改：启用 admin 接口 $ vim config.toml $ diff config.toml.orig config.toml 35c35 enabled = true $ # 将修改后的配置写入到 ConfigMap 对象中 $ kubectl create configmap influxdb-config --from-file=config.toml -n kube-system configmap \"influxdb-config\" created $ # 将 ConfigMap 中的配置文件挂载到 Pod 中，达到覆盖原始配置的目的 $ diff influxdb-deployment.yaml.orig influxdb-deployment.yaml 16c16 image: sz-pg-oam-docker-hub-001.tendcloud.com/library/heapster-influxdb-amd64:v1.1.1 19a20,21 > - mountPath: /etc/ > name: influxdb-config 22a25,27 > - name: influxdb-config > configMap: > name: influxdb-config 配置 monitoring-influxdb Service $ diff influxdb-service.yaml.orig influxdb-service.yaml 12a13 > type: NodePort 15a17,20 > name: http > - port: 8083 > targetPort: 8083 > name: admin 定义端口类型为 NodePort，额外增加了 admin 端口映射，用于后续浏览器访问 influxdb 的 admin UI 界面； 执行所有定义文件 $ pwd /root/heapster-1.3.0/deploy/kube-config/influxdb $ ls *.yaml grafana-service.yaml heapster-rbac.yaml influxdb-cm.yaml influxdb-service.yaml grafana-deployment.yaml heapster-deployment.yaml heapster-service.yaml influxdb-deployment.yaml $ kubectl create -f . deployment \"monitoring-grafana\" created service \"monitoring-grafana\" created deployment \"heapster\" created serviceaccount \"heapster\" created clusterrolebinding \"heapster\" created service \"heapster\" created configmap \"influxdb-config\" created deployment \"monitoring-influxdb\" created service \"monitoring-influxdb\" created 检查执行结果 检查 Deployment $ kubectl get deployments -n kube-system | grep -E 'heapster|monitoring' heapster 1 1 1 1 2m monitoring-grafana 1 1 1 1 2m monitoring-influxdb 1 1 1 1 2m 检查 Pods $ kubectl get pods -n kube-system | grep -E 'heapster|monitoring' heapster-110704576-gpg8v 1/1 Running 0 2m monitoring-grafana-2861879979-9z89f 1/1 Running 0 2m monitoring-influxdb-1411048194-lzrpc 1/1 Running 0 2m 检查 kubernets dashboard 界面，看是显示各 Nodes、Pods 的 CPU、内存、负载等利用率曲线图； 图片 - dashboard-heapster 访问 grafana 通过 kube-apiserver 访问： 获取 monitoring-grafana 服务 URL $ kubectl cluster-info Kubernetes master is running at https://172.20.0.113:6443 Heapster is running at https://172.20.0.113:6443/api/v1/proxy/namespaces/kube-system/services/heapster KubeDNS is running at https://172.20.0.113:6443/api/v1/proxy/namespaces/kube-system/services/kube-dns kubernetes-dashboard is running at https://172.20.0.113:6443/api/v1/proxy/namespaces/kube-system/services/kubernetes-dashboard monitoring-grafana is running at https://172.20.0.113:6443/api/v1/proxy/namespaces/kube-system/services/monitoring-grafana monitoring-influxdb is running at https://172.20.0.113:6443/api/v1/proxy/namespaces/kube-system/services/monitoring-influxdb To further debug and diagnose cluster problems, use 'kubectl cluster-info dump'. 浏览器访问 URL： http://172.20.0.113:8080/api/v1/proxy/namespaces/kube-system/services/monitoring-grafana 通过 kubectl proxy 访问： 创建代理 $ kubectl proxy --address='172.20.0.113' --port=8086 --accept-hosts='^*$' Starting to serve on 172.20.0.113:8086 浏览器访问 URL：http://172.20.0.113:8086/api/v1/proxy/namespaces/kube-system/services/monitoring-grafana 图片 - grafana 访问 influxdb admin UI 获取 influxdb http 8086 映射的 NodePort $ kubectl get svc -n kube-system|grep influxdb monitoring-influxdb 10.254.22.46 8086:32299/TCP,8083:30269/TCP 9m 通过 kube-apiserver 的非安全端口访问 influxdb 的 admin UI 界面： http://172.20.0.113:8080/api/v1/proxy/namespaces/kube-system/services/monitoring-influxdb:8083/ 在页面的 “Connection Settings” 的 Host 中输入 node IP， Port 中输入 8086 映射的 nodePort 如上面的 32299，点击 “Save” 即可（我的集群中的地址是172.20.0.113:32299）： 图片 - kubernetes-influxdb-heapster 注意 在安装好 Grafana 之后我们使用的是默认的 template 配置，页面上的 namespace 选择里只有 default 和 kube-system，并不是说其他的 namespace 里的指标没有得到监控，只是我们没有在 Grafana 中开启他它们的显示而已。见 Cannot see other namespaces except, kube-system and default #1279。 图片 - 修改grafana模板 将 Templating 中的 namespace 的 Data source 设置为 influxdb-datasource，Refresh 设置为 on Dashboard Load 保存设置，刷新浏览器，即可看到其他 namespace 选项。 参考 使用Heapster获取集群对象的metric数据 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-21 11:57:43 "},"practice/efk-addon-installation.html":{"url":"practice/efk-addon-installation.html","title":"安装EFK插件","keywords":"","body":"安装EFK插件 我们通过在每台node上部署一个以DaemonSet方式运行的fluentd来收集每台node上的日志。Fluentd将docker日志目录/var/lib/docker/containers和/var/log目录挂载到Pod中，然后Pod会在node节点的/var/log/pods目录中创建新的目录，可以区别不同的容器日志输出，该目录下有一个日志文件链接到/var/lib/docker/contianers目录下的容器日志输出。 官方文件目录：cluster/addons/fluentd-elasticsearch $ ls *.yaml es-controller.yaml es-service.yaml fluentd-es-ds.yaml kibana-controller.yaml kibana-service.yaml efk-rbac.yaml 同样EFK服务也需要一个efk-rbac.yaml文件，配置serviceaccount为efk。 已经修改好的 yaml 文件见：../manifests/EFK 配置 es-controller.yaml $ diff es-controller.yaml.orig es-controller.yaml 24c24 - image: sz-pg-oam-docker-hub-001.tendcloud.com/library/elasticsearch:v2.4.1-2 配置 es-service.yaml 无需配置； 配置 fluentd-es-ds.yaml $ diff fluentd-es-ds.yaml.orig fluentd-es-ds.yaml 26c26 image: sz-pg-oam-docker-hub-001.tendcloud.com/library/fluentd-elasticsearch:1.22 配置 kibana-controller.yaml $ diff kibana-controller.yaml.orig kibana-controller.yaml 22c22 image: sz-pg-oam-docker-hub-001.tendcloud.com/library/kibana:v4.6.1-1 给 Node 设置标签 定义 DaemonSet fluentd-es-v1.22 时设置了 nodeSelector beta.kubernetes.io/fluentd-ds-ready=true ，所以需要在期望运行 fluentd 的 Node 上设置该标签； $ kubectl get nodes NAME STATUS AGE VERSION 172.20.0.113 Ready 1d v1.6.0 $ kubectl label nodes 172.20.0.113 beta.kubernetes.io/fluentd-ds-ready=true node \"172.20.0.113\" labeled 给其他两台node打上同样的标签。 执行定义文件 $ kubectl create -f . serviceaccount \"efk\" created clusterrolebinding \"efk\" created replicationcontroller \"elasticsearch-logging-v1\" created service \"elasticsearch-logging\" created daemonset \"fluentd-es-v1.22\" created deployment \"kibana-logging\" created service \"kibana-logging\" created 检查执行结果 $ kubectl get deployment -n kube-system|grep kibana kibana-logging 1 1 1 1 2m $ kubectl get pods -n kube-system|grep -E 'elasticsearch|fluentd|kibana' elasticsearch-logging-v1-mlstp 1/1 Running 0 1m elasticsearch-logging-v1-nfbbf 1/1 Running 0 1m fluentd-es-v1.22-31sm0 1/1 Running 0 1m fluentd-es-v1.22-bpgqs 1/1 Running 0 1m fluentd-es-v1.22-qmn7h 1/1 Running 0 1m kibana-logging-1432287342-0gdng 1/1 Running 0 1m $ kubectl get service -n kube-system|grep -E 'elasticsearch|kibana' elasticsearch-logging 10.254.77.62 9200/TCP 2m kibana-logging 10.254.8.113 5601/TCP 2m kibana Pod 第一次启动时会用较长时间(10-20分钟)来优化和 Cache 状态页面，可以 tailf 该 Pod 的日志观察进度： $ kubectl logs kibana-logging-1432287342-0gdng -n kube-system -f ELASTICSEARCH_URL=http://elasticsearch-logging:9200 server.basePath: /api/v1/proxy/namespaces/kube-system/services/kibana-logging {\"type\":\"log\",\"@timestamp\":\"2017-04-12T13:08:06Z\",\"tags\":[\"info\",\"optimize\"],\"pid\":7,\"message\":\"Optimizing and caching bundles for kibana and statusPage. This may take a few minutes\"} {\"type\":\"log\",\"@timestamp\":\"2017-04-12T13:18:17Z\",\"tags\":[\"info\",\"optimize\"],\"pid\":7,\"message\":\"Optimization of bundles for kibana and statusPage complete in 610.40 seconds\"} {\"type\":\"log\",\"@timestamp\":\"2017-04-12T13:18:17Z\",\"tags\":[\"status\",\"plugin:kibana@1.0.0\",\"info\"],\"pid\":7,\"state\":\"green\",\"message\":\"Status changed from uninitialized to green - Ready\",\"prevState\":\"uninitialized\",\"prevMsg\":\"uninitialized\"} {\"type\":\"log\",\"@timestamp\":\"2017-04-12T13:18:18Z\",\"tags\":[\"status\",\"plugin:elasticsearch@1.0.0\",\"info\"],\"pid\":7,\"state\":\"yellow\",\"message\":\"Status changed from uninitialized to yellow - Waiting for Elasticsearch\",\"prevState\":\"uninitialized\",\"prevMsg\":\"uninitialized\"} {\"type\":\"log\",\"@timestamp\":\"2017-04-12T13:18:19Z\",\"tags\":[\"status\",\"plugin:kbn_vislib_vis_types@1.0.0\",\"info\"],\"pid\":7,\"state\":\"green\",\"message\":\"Status changed from uninitialized to green - Ready\",\"prevState\":\"uninitialized\",\"prevMsg\":\"uninitialized\"} {\"type\":\"log\",\"@timestamp\":\"2017-04-12T13:18:19Z\",\"tags\":[\"status\",\"plugin:markdown_vis@1.0.0\",\"info\"],\"pid\":7,\"state\":\"green\",\"message\":\"Status changed from uninitialized to green - Ready\",\"prevState\":\"uninitialized\",\"prevMsg\":\"uninitialized\"} {\"type\":\"log\",\"@timestamp\":\"2017-04-12T13:18:19Z\",\"tags\":[\"status\",\"plugin:metric_vis@1.0.0\",\"info\"],\"pid\":7,\"state\":\"green\",\"message\":\"Status changed from uninitialized to green - Ready\",\"prevState\":\"uninitialized\",\"prevMsg\":\"uninitialized\"} {\"type\":\"log\",\"@timestamp\":\"2017-04-12T13:18:19Z\",\"tags\":[\"status\",\"plugin:spyModes@1.0.0\",\"info\"],\"pid\":7,\"state\":\"green\",\"message\":\"Status changed from uninitialized to green - Ready\",\"prevState\":\"uninitialized\",\"prevMsg\":\"uninitialized\"} {\"type\":\"log\",\"@timestamp\":\"2017-04-12T13:18:19Z\",\"tags\":[\"status\",\"plugin:statusPage@1.0.0\",\"info\"],\"pid\":7,\"state\":\"green\",\"message\":\"Status changed from uninitialized to green - Ready\",\"prevState\":\"uninitialized\",\"prevMsg\":\"uninitialized\"} {\"type\":\"log\",\"@timestamp\":\"2017-04-12T13:18:19Z\",\"tags\":[\"status\",\"plugin:table_vis@1.0.0\",\"info\"],\"pid\":7,\"state\":\"green\",\"message\":\"Status changed from uninitialized to green - Ready\",\"prevState\":\"uninitialized\",\"prevMsg\":\"uninitialized\"} {\"type\":\"log\",\"@timestamp\":\"2017-04-12T13:18:19Z\",\"tags\":[\"listening\",\"info\"],\"pid\":7,\"message\":\"Server running at http://0.0.0.0:5601\"} {\"type\":\"log\",\"@timestamp\":\"2017-04-12T13:18:24Z\",\"tags\":[\"status\",\"plugin:elasticsearch@1.0.0\",\"info\"],\"pid\":7,\"state\":\"yellow\",\"message\":\"Status changed from yellow to yellow - No existing Kibana index found\",\"prevState\":\"yellow\",\"prevMsg\":\"Waiting for Elasticsearch\"} {\"type\":\"log\",\"@timestamp\":\"2017-04-12T13:18:29Z\",\"tags\":[\"status\",\"plugin:elasticsearch@1.0.0\",\"info\"],\"pid\":7,\"state\":\"green\",\"message\":\"Status changed from yellow to green - Kibana index ready\",\"prevState\":\"yellow\",\"prevMsg\":\"No existing Kibana index found\"} 访问 kibana 通过 kube-apiserver 访问： 获取 monitoring-grafana 服务 URL $ kubectl cluster-info Kubernetes master is running at https://172.20.0.113:6443 Elasticsearch is running at https://172.20.0.113:6443/api/v1/proxy/namespaces/kube-system/services/elasticsearch-logging Heapster is running at https://172.20.0.113:6443/api/v1/proxy/namespaces/kube-system/services/heapster Kibana is running at https://172.20.0.113:6443/api/v1/proxy/namespaces/kube-system/services/kibana-logging KubeDNS is running at https://172.20.0.113:6443/api/v1/proxy/namespaces/kube-system/services/kube-dns kubernetes-dashboard is running at https://172.20.0.113:6443/api/v1/proxy/namespaces/kube-system/services/kubernetes-dashboard monitoring-grafana is running at https://172.20.0.113:6443/api/v1/proxy/namespaces/kube-system/services/monitoring-grafana monitoring-influxdb is running at https://172.20.0.113:6443/api/v1/proxy/namespaces/kube-system/services/monitoring-influxdb 浏览器访问 URL： https://172.20.0.113:6443/api/v1/proxy/namespaces/kube-system/services/kibana-logging/app/kibana 通过 kubectl proxy 访问： 创建代理 $ kubectl proxy --address='172.20.0.113' --port=8086 --accept-hosts='^*$' Starting to serve on 172.20.0.113:8086 浏览器访问 URL：http://172.20.0.113:8086/api/v1/proxy/namespaces/kube-system/services/kibana-logging 在 Settings -> Indices 页面创建一个 index（相当于 mysql 中的一个 database），选中 Index contains time-based events，使用默认的 logstash-* pattern，点击 Create ; 可能遇到的问题 如果你在这里发现Create按钮是灰色的无法点击，且Time-filed name中没有选项，fluentd要读取/var/log/containers/目录下的log日志，这些日志是从/var/lib/docker/containers/${CONTAINER_ID}/${CONTAINER_ID}-json.log链接过来的，查看你的docker配置，—log-dirver需要设置为json-file格式，默认的可能是journald，参考docker logging。 图片 - es-setting 创建Index后，可以在 Discover 下看到 ElasticSearch logging 中汇聚的日志； Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-07 10:18:25 "},"practice/service-discovery-and-loadbalancing.html":{"url":"practice/service-discovery-and-loadbalancing.html","title":"服务发现与负载均衡","keywords":"","body":"服务发现与负载均衡 Kubernetes在设计之初就充分考虑了针对容器的服务发现与负载均衡机制，提供了Service资源，并通过kube-proxy配合cloud provider来适应不同的应用场景。随着kubernetes用户的激增，用户场景的不断丰富，又产生了一些新的负载均衡机制。目前，kubernetes中的负载均衡大致可以分为以下几种机制，每种机制都有其特定的应用场景： Service：直接用Service提供cluster内部的负载均衡，并借助cloud provider提供的LB提供外部访问 Ingress Controller：还是用Service提供cluster内部的负载均衡，但是通过自定义LB提供外部访问 Service Load Balancer：把load balancer直接跑在容器中，实现Bare Metal的Service Load Balancer Custom Load Balancer：自定义负载均衡，并替代kube-proxy，一般在物理部署Kubernetes时使用，方便接入公司已有的外部服务 Service Service是对一组提供相同功能的Pods的抽象，并为它们提供一个统一的入口。借助Service，应用可以方便的实现服务发现与负载均衡，并实现应用的零宕机升级。Service通过标签来选取服务后端，一般配合Replication Controller或者Deployment来保证后端容器的正常运行。 Service有三种类型： ClusterIP：默认类型，自动分配一个仅cluster内部可以访问的虚拟IP NodePort：在ClusterIP基础上为Service在每台机器上绑定一个端口，这样就可以通过:NodePort来访问该服务 LoadBalancer：在NodePort的基础上，借助cloud provider创建一个外部的负载均衡器，并将请求转发到:NodePort 另外，也可以将已有的服务以Service的形式加入到Kubernetes集群中来，只需要在创建Service的时候不指定Label selector，而是在Service创建好后手动为其添加endpoint。 Ingress Controller Service虽然解决了服务发现和负载均衡的问题，但它在使用上还是有一些限制，比如 对外访问的时候，NodePort类型需要在外部搭建额外的负载均衡，而LoadBalancer要求kubernetes必须跑在支持的cloud provider上面 Ingress就是为了解决这些限制而引入的新资源，主要用来将服务暴露到cluster外面，并且可以自定义服务的访问策略。比如想要通过负载均衡器实现不同子域名到不同服务的访问： foo.bar.com --| |-> foo.bar.com s1:80 | 178.91.123.132 | bar.foo.com --| |-> bar.foo.com s2:80 可以这样来定义Ingress： apiVersion: extensions/v1beta1 kind: Ingress metadata: name: test spec: rules: - host: foo.bar.com http: paths: - backend: serviceName: s1 servicePort: 80 - host: bar.foo.com http: paths: - backend: serviceName: s2 servicePort: 80 注意： Ingress本身并不会自动创建负载均衡器，cluster中需要运行一个ingress controller来根据Ingress的定义来管理负载均衡器。目前社区提供了nginx和gce的参考实现。 Traefik提供了易用的Ingress Controller，使用方法见https://docs.traefik.io/user-guide/kubernetes/。 Service Load Balancer 在Ingress出现以前，Service Load Balancer是推荐的解决Service局限性的方式。Service Load Balancer将haproxy跑在容器中，并监控service和endpoint的变化，通过容器IP对外提供4层和7层负载均衡服务。 社区提供的Service Load Balancer支持四种负载均衡协议：TCP、HTTP、HTTPS和SSL TERMINATION，并支持ACL访问控制。 Custom Load Balancer 虽然Kubernetes提供了丰富的负载均衡机制，但在实际使用的时候，还是会碰到一些复杂的场景是它不能支持的，比如： 接入已有的负载均衡设备 多租户网络情况下，容器网络和主机网络是隔离的，这样kube-proxy就不能正常工作 这个时候就可以自定义组件，并代替kube-proxy来做负载均衡。基本的思路是监控kubernetes中service和endpoints的变化，并根据这些变化来配置负载均衡器。比如weave flux、nginx plus、kube2haproxy等。 Endpoints 有几种情况下需要用到没有selector的service。 使用kubernetes集群外部的数据库时 service中用到了其他namespace或kubernetes集群中的service 在kubernetes的工作负载与集群外的后端之间互相迁移 可以这样定义一个没有selector的service。 kind: Service apiVersion: v1 metadata: name: my-service spec: ports: - protocol: TCP port: 80 targetPort: 9376 定义一个Endpoints来对应该service。 kind: Endpoints apiVersion: v1 metadata: name: my-service subsets: - addresses: - ip: 1.2.3.4 ports: - port: 9376 访问没有selector的service跟访问有selector的service时没有任何区别。 使用kubernetes时有一个很常见的需求，就是当数据库部署在kubernetes集群之外的时候，集群内的service如何访问数据库呢？当然你可以直接使用数据库的IP地址和端口号来直接访问，有没有什么优雅的方式呢？你需要用到ExternalName Service。 kind: Service apiVersion: v1 metadata: name: my-service namespace: prod spec: type: ExternalName externalName: my.database.example.com ports: - port: 12345 这个例子中，在kubernetes集群内访问my-service实际上会重定向到my.database.example.com:12345这个地址。 参考资料 https://kubernetes.io/docs/concepts/services-networking/service/ http://kubernetes.io/docs/user-guide/ingress/ https://github.com/kubernetes/contrib/tree/master/service-loadbalancer https://www.nginx.com/blog/load-balancing-kubernetes-services-nginx-plus/ https://github.com/weaveworks/flux https://github.com/AdoHe/kube2haproxy Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-17 15:54:27 "},"practice/traefik-ingress-installation.html":{"url":"practice/traefik-ingress-installation.html","title":"安装Traefik ingress","keywords":"","body":"安装traefik ingress Ingress简介 如果你还不了解，ingress是什么，可以先看下我翻译的Kubernetes官网上ingress的介绍Kubernetes Ingress解析。 理解Ingress 简单的说，ingress就是从kubernetes集群外访问集群的入口，将用户的URL请求转发到不同的service上。Ingress相当于nginx、apache等负载均衡方向代理服务器，其中还包括规则定义，即URL的路由信息，路由信息得的刷新由Ingress controller来提供。 理解Ingress Controller Ingress Controller 实质上可以理解为是个监视器，Ingress Controller 通过不断地跟 kubernetes API 打交道，实时的感知后端 service、pod 等变化，比如新增和减少 pod，service 增加与减少等；当得到这些变化信息后，Ingress Controller 再结合下文的 Ingress 生成配置，然后更新反向代理负载均衡器，并刷新其配置，达到服务发现的作用。 部署Traefik 介绍traefik Traefik是一款开源的反向代理与负载均衡工具。它最大的优点是能够与常见的微服务系统直接整合，可以实现自动化动态配置。目前支持Docker, Swarm, Mesos/Marathon, Mesos, Kubernetes, Consul, Etcd, Zookeeper, BoltDB, Rest API等等后端模型。 以下配置文件可以在kubernetes-handbookGitHub仓库中的../manifests/traefik-ingress/目录下找到。 创建ingress-rbac.yaml 将用于service account验证。 apiVersion: v1 kind: ServiceAccount metadata: name: ingress namespace: kube-system --- kind: ClusterRoleBinding apiVersion: rbac.authorization.k8s.io/v1beta1 metadata: name: ingress subjects: - kind: ServiceAccount name: ingress namespace: kube-system roleRef: kind: ClusterRole name: cluster-admin apiGroup: rbac.authorization.k8s.io 创建名为traefik-ingress的ingress，文件名traefik.yaml apiVersion: extensions/v1beta1 kind: Ingress metadata: name: traefik-ingress namespace: default spec: rules: - host: traefik.nginx.io http: paths: - path: / backend: serviceName: my-nginx servicePort: 80 - host: traefik.frontend.io http: paths: - path: / backend: serviceName: frontend servicePort: 80 这其中的backend中要配置default namespace中启动的service名字，如果你没有配置namespace名字，默认使用default namespace，如果你在其他namespace中创建服务想要暴露到kubernetes集群外部，可以创建新的ingress.yaml文件，同时在文件中指定该namespace，其他配置与上面的文件格式相同。。path就是URL地址后的路径，如traefik.frontend.io/path，service将会接受path这个路径，host最好使用service-name.filed1.filed2.domain-name这种类似主机名称的命名方式，方便区分服务。 根据你自己环境中部署的service的名字和端口自行修改，有新service增加时，修改该文件后可以使用kubectl replace -f traefik.yaml来更新。 我们现在集群中已经有两个service了，一个是nginx，另一个是官方的guestbook例子。 创建Depeloyment apiVersion: extensions/v1beta1 kind: Deployment metadata: name: traefik-ingress-lb namespace: kube-system labels: k8s-app: traefik-ingress-lb spec: template: metadata: labels: k8s-app: traefik-ingress-lb name: traefik-ingress-lb spec: terminationGracePeriodSeconds: 60 hostNetwork: true restartPolicy: Always serviceAccountName: ingress containers: - image: traefik name: traefik-ingress-lb resources: limits: cpu: 200m memory: 30Mi requests: cpu: 100m memory: 20Mi ports: - name: http containerPort: 80 hostPort: 80 - name: admin containerPort: 8580 hostPort: 8580 args: - --web - --web.address=:8580 - --kubernetes 注意我们这里用的是Deploy类型，没有限定该pod运行在哪个主机上。Traefik的端口是8580。 Traefik UI apiVersion: v1 kind: Service metadata: name: traefik-web-ui namespace: kube-system spec: selector: k8s-app: traefik-ingress-lb ports: - name: web port: 80 targetPort: 8580 --- apiVersion: extensions/v1beta1 kind: Ingress metadata: name: traefik-web-ui namespace: kube-system spec: rules: - host: traefik-ui.local http: paths: - path: / backend: serviceName: traefik-web-ui servicePort: web 配置完成后就可以启动treafik ingress了。 kubectl create -f . 我查看到traefik的pod在172.20.0.115这台节点上启动了。 访问该地址http://172.20.0.115:8580/将可以看到dashboard。 图片 - kubernetes-dashboard 左侧黄色部分部分列出的是所有的rule，右侧绿色部分是所有的backend。 测试 在集群的任意一个节点上执行。假如现在我要访问nginx的\"/\"路径。 $ curl -H Host:traefik.nginx.io http://172.20.0.115/ Welcome to nginx! body { width: 35em; margin: 0 auto; font-family: Tahoma, Verdana, Arial, sans-serif; } Welcome to nginx! If you see this page, the nginx web server is successfully installed and working. Further configuration is required. For online documentation and support please refer to nginx.org. Commercial support is available at nginx.com. Thank you for using nginx. 如果你需要在kubernetes集群以外访问就需要设置DNS，或者修改本机的hosts文件。 在其中加入： 172.20.0.115 traefik.nginx.io 172.20.0.115 traefik.frontend.io 所有访问这些地址的流量都会发送给172.20.0.115这台主机，就是我们启动traefik的主机。 Traefik会解析http请求header里的Host参数将流量转发给Ingress配置里的相应service。 修改hosts后就就可以在kubernetes集群外访问以上两个service，如下图： 图片 - traefik-nginx 图片 - traefik-guestbook 参考 Traefik-kubernetes 初试 Traefik简介 Guestbook example Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-10-11 23:39:48 "},"practice/distributed-load-test.html":{"url":"practice/distributed-load-test.html","title":"分布式负载测试","keywords":"","body":"分布式负载测试 该教程描述如何在Kubernetes中进行分布式负载均衡测试，包括一个web应用、docker镜像和Kubernetes controllers/services。关于分布式负载测试的更多资料请查看Distributed Load Testing Using Kubernetes 。 准备 不需要GCE及其他组件，你只需要有一个kubernetes集群即可。 如果你还没有kubernetes集群，可以参考kubernetes-handbook部署一个。 部署Web应用 本文中使用的镜像、kubernetes应用的yaml配置来自我的另一个项目，请参考：https://github.com/rootsongjc/distributed-load-testing-using-kubernetes sample-webapp 目录下包含一个简单的web测试应用。我们将其构建为docker镜像，在kubernetes中运行。你可以自己构建，也可以直接用这个我构建好的镜像index.tenxcloud.com/jimmy/k8s-sample-webapp:latest。 在kubernetes上部署sample-webapp。 $ git clone https://github.com/rootsongjc/distributed-load-testing-using-kubernetes.git $ cd kubernetes-config $ kubectl create -f sample-webapp-controller.yaml $ kubectl create -f sample-webapp-service.yaml 部署Locust的Controller和Service locust-master和locust-work使用同样的docker镜像，修改cotnroller中spec.template.spec.containers.env字段中的value为你sample-webapp service的名字。 - name: TARGET_HOST value: http://sample-webapp:8000 创建Controller Docker镜像（可选） locust-master和locust-work controller使用的都是locust-tasks docker镜像。你可以直接下载gcr.io/cloud-solutions-images/locust-tasks，也可以自己编译。自己编译大概要花几分钟时间，镜像大小为820M。 $ docker build -t index.tenxcloud.com/jimmy/locust-tasks:latest . $ docker push index.tenxcloud.com/jimmy/locust-tasks:latest 注意：我使用的是时速云的镜像仓库。 每个controller的yaml的spec.template.spec.containers.image 字段指定的是我的镜像： image: index.tenxcloud.com/jimmy/locust-tasks:latest 部署locust-master $ kubectl create -f locust-master-controller.yaml $ kubectl create -f locust-master-service.yaml 部署locust-worker Now deploy locust-worker-controller: $ kubectl create -f locust-worker-controller.yaml 你可以很轻易的给work扩容，通过命令行方式： $ kubectl scale --replicas=20 replicationcontrollers locust-worker 当然你也可以通过WebUI：Dashboard - Workloads - Replication Controllers - ServiceName - Scale来扩容。 图片 - 使用dashboard来扩容 配置Traefik 参考kubernetes的traefik ingress安装，在ingress.yaml中加入如下配置： - host: traefik.locust.io http: paths: - path: / backend: serviceName: locust-master servicePort: 8089 然后执行kubectl replace -f ingress.yaml即可更新traefik。 通过Traefik的dashboard就可以看到刚增加的traefik.locust.io节点。 图片 - Traefik的UI 执行测试 打开http://traefik.locust.io页面，点击Edit输入伪造的用户数和用户每秒发送的请求个数，点击Start Swarming就可以开始测试了。 图片 - Locust启动界面 在测试过程中调整sample-webapp的pod个数（默认设置了1个pod），观察pod的负载变化情况。 图片 - Dashboard查看页面 从一段时间的观察中可以看到负载被平均分配给了3个pod。 在locust的页面中可以实时观察也可以下载测试结果。 图片 - Locust测试结果页面 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-10-14 01:23:23 "},"practice/network-and-cluster-perfermance-test.html":{"url":"practice/network-and-cluster-perfermance-test.html","title":"网络和集群性能测试","keywords":"","body":"Kubernetes网络和集群性能测试 准备 测试环境 在以下几种环境下进行测试： Kubernetes集群node节点上通过Cluster IP方式访问 Kubernetes集群内部通过service访问 Kubernetes集群外部通过traefik ingress暴露的地址访问 测试地址 Cluster IP: 10.254.149.31 Service Port：8000 Ingress Host：traefik.sample-webapp.io 测试工具 Locust：一个简单易用的用户负载测试工具，用来测试web或其他系统能够同时处理的并发用户数。 curl kubemark 测试程序：sample-webapp，源码见Github kubernetes的分布式负载测试 测试说明 通过向sample-webapp发送curl请求获取响应时间，直接curl后的结果为： $ curl \"http://10.254.149.31:8000/\" Welcome to the \"Distributed Load Testing Using Kubernetes\" sample web app 网络延迟测试 场景一、 Kubernetes集群node节点上通过Cluster IP访问 测试命令 curl -o /dev/null -s -w '%{time_connect} %{time_starttransfer} %{time_total}' \"http://10.254.149.31:8000/\" 10组测试结果 No time_connect time_starttransfer time_total 1 0.000 0.003 0.003 2 0.000 0.002 0.002 3 0.000 0.002 0.002 4 0.000 0.002 0.002 5 0.000 0.002 0.002 6 0.000 0.002 0.002 7 0.000 0.002 0.002 8 0.000 0.002 0.002 9 0.000 0.002 0.002 10 0.000 0.002 0.002 平均响应时间：2ms 时间指标说明 单位：秒 time_connect：建立到服务器的 TCP 连接所用的时间 time_starttransfer：在发出请求之后，Web 服务器返回数据的第一个字节所用的时间 time_total：完成请求所用的时间 场景二、Kubernetes集群内部通过service访问 测试命令 curl -o /dev/null -s -w '%{time_connect} %{time_starttransfer} %{time_total}' \"http://sample-webapp:8000/\" 10组测试结果 No time_connect time_starttransfer time_total 1 0.004 0.006 0.006 2 0.004 0.006 0.006 3 0.004 0.006 0.006 4 0.004 0.006 0.006 5 0.004 0.006 0.006 6 0.004 0.006 0.006 7 0.004 0.006 0.006 8 0.004 0.006 0.006 9 0.004 0.006 0.006 10 0.004 0.006 0.006 平均响应时间：6ms 场景三、在公网上通过traefik ingress访问 测试命令 curl -o /dev/null -s -w '%{time_connect} %{time_starttransfer} %{time_total}' \"http://traefik.sample-webapp.io\" >>result 10组测试结果 No time_connect time_starttransfer time_total 1 0.043 0.085 0.085 2 0.052 0.093 0.093 3 0.043 0.082 0.082 4 0.051 0.093 0.093 5 0.068 0.188 0.188 6 0.049 0.089 0.089 7 0.051 0.113 0.113 8 0.055 0.120 0.120 9 0.065 0.126 0.127 10 0.050 0.111 0.111 平均响应时间：110ms 测试结果 在这三种场景下的响应时间测试结果如下： Kubernetes集群node节点上通过Cluster IP方式访问：2ms Kubernetes集群内部通过service访问：6ms Kubernetes集群外部通过traefik ingress暴露的地址访问：110ms 注意：执行测试的node节点/Pod与serivce所在的pod的距离（是否在同一台主机上），对前两个场景可以能会有一定影响。 网络性能测试 网络使用flannel的vxlan模式。 使用iperf进行测试。 服务端命令： iperf -s -p 12345 -i 1 -M 客户端命令： iperf -c ${server-ip} -p 12345 -i 1 -t 10 -w 20K 场景一、主机之间 [ ID] Interval Transfer Bandwidth [ 3] 0.0- 1.0 sec 598 MBytes 5.02 Gbits/sec [ 3] 1.0- 2.0 sec 637 MBytes 5.35 Gbits/sec [ 3] 2.0- 3.0 sec 664 MBytes 5.57 Gbits/sec [ 3] 3.0- 4.0 sec 657 MBytes 5.51 Gbits/sec [ 3] 4.0- 5.0 sec 641 MBytes 5.38 Gbits/sec [ 3] 5.0- 6.0 sec 639 MBytes 5.36 Gbits/sec [ 3] 6.0- 7.0 sec 628 MBytes 5.26 Gbits/sec [ 3] 7.0- 8.0 sec 649 MBytes 5.44 Gbits/sec [ 3] 8.0- 9.0 sec 638 MBytes 5.35 Gbits/sec [ 3] 9.0-10.0 sec 652 MBytes 5.47 Gbits/sec [ 3] 0.0-10.0 sec 6.25 GBytes 5.37 Gbits/sec 场景二、不同主机的Pod之间(使用flannel的vxlan模式) [ ID] Interval Transfer Bandwidth [ 3] 0.0- 1.0 sec 372 MBytes 3.12 Gbits/sec [ 3] 1.0- 2.0 sec 345 MBytes 2.89 Gbits/sec [ 3] 2.0- 3.0 sec 361 MBytes 3.03 Gbits/sec [ 3] 3.0- 4.0 sec 397 MBytes 3.33 Gbits/sec [ 3] 4.0- 5.0 sec 405 MBytes 3.40 Gbits/sec [ 3] 5.0- 6.0 sec 410 MBytes 3.44 Gbits/sec [ 3] 6.0- 7.0 sec 404 MBytes 3.39 Gbits/sec [ 3] 7.0- 8.0 sec 408 MBytes 3.42 Gbits/sec [ 3] 8.0- 9.0 sec 451 MBytes 3.78 Gbits/sec [ 3] 9.0-10.0 sec 387 MBytes 3.25 Gbits/sec [ 3] 0.0-10.0 sec 3.85 GBytes 3.30 Gbits/sec 场景三、Node与非同主机的Pod之间（使用flannel的vxlan模式） [ ID] Interval Transfer Bandwidth [ 3] 0.0- 1.0 sec 372 MBytes 3.12 Gbits/sec [ 3] 1.0- 2.0 sec 420 MBytes 3.53 Gbits/sec [ 3] 2.0- 3.0 sec 434 MBytes 3.64 Gbits/sec [ 3] 3.0- 4.0 sec 409 MBytes 3.43 Gbits/sec [ 3] 4.0- 5.0 sec 382 MBytes 3.21 Gbits/sec [ 3] 5.0- 6.0 sec 408 MBytes 3.42 Gbits/sec [ 3] 6.0- 7.0 sec 403 MBytes 3.38 Gbits/sec [ 3] 7.0- 8.0 sec 423 MBytes 3.55 Gbits/sec [ 3] 8.0- 9.0 sec 376 MBytes 3.15 Gbits/sec [ 3] 9.0-10.0 sec 451 MBytes 3.78 Gbits/sec [ 3] 0.0-10.0 sec 3.98 GBytes 3.42 Gbits/sec 场景四、不同主机的Pod之间（使用flannel的host-gw模式） [ ID] Interval Transfer Bandwidth [ 5] 0.0- 1.0 sec 530 MBytes 4.45 Gbits/sec [ 5] 1.0- 2.0 sec 576 MBytes 4.84 Gbits/sec [ 5] 2.0- 3.0 sec 631 MBytes 5.29 Gbits/sec [ 5] 3.0- 4.0 sec 580 MBytes 4.87 Gbits/sec [ 5] 4.0- 5.0 sec 627 MBytes 5.26 Gbits/sec [ 5] 5.0- 6.0 sec 578 MBytes 4.85 Gbits/sec [ 5] 6.0- 7.0 sec 584 MBytes 4.90 Gbits/sec [ 5] 7.0- 8.0 sec 571 MBytes 4.79 Gbits/sec [ 5] 8.0- 9.0 sec 564 MBytes 4.73 Gbits/sec [ 5] 9.0-10.0 sec 572 MBytes 4.80 Gbits/sec [ 5] 0.0-10.0 sec 5.68 GBytes 4.88 Gbits/sec 场景五、Node与非同主机的Pod之间（使用flannel的host-gw模式） [ ID] Interval Transfer Bandwidth [ 3] 0.0- 1.0 sec 570 MBytes 4.78 Gbits/sec [ 3] 1.0- 2.0 sec 552 MBytes 4.63 Gbits/sec [ 3] 2.0- 3.0 sec 598 MBytes 5.02 Gbits/sec [ 3] 3.0- 4.0 sec 580 MBytes 4.87 Gbits/sec [ 3] 4.0- 5.0 sec 590 MBytes 4.95 Gbits/sec [ 3] 5.0- 6.0 sec 594 MBytes 4.98 Gbits/sec [ 3] 6.0- 7.0 sec 598 MBytes 5.02 Gbits/sec [ 3] 7.0- 8.0 sec 606 MBytes 5.08 Gbits/sec [ 3] 8.0- 9.0 sec 596 MBytes 5.00 Gbits/sec [ 3] 9.0-10.0 sec 604 MBytes 5.07 Gbits/sec [ 3] 0.0-10.0 sec 5.75 GBytes 4.94 Gbits/sec 网络性能对比综述 使用Flannel的vxlan模式实现每个pod一个IP的方式，会比宿主机直接互联的网络性能损耗30%～40%，符合网上流传的测试结论。而flannel的host-gw模式比起宿主机互连的网络性能损耗大约是10%。 Vxlan会有一个封包解包的过程，所以会对网络性能造成较大的损耗，而host-gw模式是直接使用路由信息，网络损耗小，关于host-gw的架构请访问Flannel host-gw architecture。 Kubernete的性能测试 参考Kubernetes集群性能测试中的步骤，对kubernetes的性能进行测试。 我的集群版本是Kubernetes1.6.0，首先克隆代码，将kubernetes目录复制到$GOPATH/src/k8s.io/下然后执行： $ ./hack/generate-bindata.sh /usr/local/src/k8s.io/kubernetes /usr/local/src/k8s.io/kubernetes Generated bindata file : test/e2e/generated/bindata.go has 13498 test/e2e/generated/bindata.go lines of lovely automated artifacts No changes in generated bindata file: pkg/generated/bindata.go /usr/local/src/k8s.io/kubernetes $ make WHAT=\"test/e2e/e2e.test\" ... +++ [0425 17:01:34] Generating bindata: test/e2e/generated/gobindata_util.go /usr/local/src/k8s.io/kubernetes /usr/local/src/k8s.io/kubernetes/test/e2e/generated /usr/local/src/k8s.io/kubernetes/test/e2e/generated +++ [0425 17:01:34] Building go targets for linux/amd64: test/e2e/e2e.test $ make ginkgo +++ [0425 17:05:57] Building the toolchain targets: k8s.io/kubernetes/hack/cmd/teststale k8s.io/kubernetes/vendor/github.com/jteeuwen/go-bindata/go-bindata +++ [0425 17:05:57] Generating bindata: test/e2e/generated/gobindata_util.go /usr/local/src/k8s.io/kubernetes /usr/local/src/k8s.io/kubernetes/test/e2e/generated /usr/local/src/k8s.io/kubernetes/test/e2e/generated +++ [0425 17:05:58] Building go targets for linux/amd64: vendor/github.com/onsi/ginkgo/ginkgo $ export KUBERNETES_PROVIDER=local $ export KUBECTL_PATH=/usr/bin/kubectl $ go run hack/e2e.go -v -test --test_args=\"--host=http://172.20.0.113:8080 --ginkgo.focus=\\[Feature:Performance\\]\" >>log.txt 测试结果 Apr 25 18:27:31.461: INFO: API calls latencies: { \"apicalls\": [ { \"resource\": \"pods\", \"verb\": \"POST\", \"latency\": { \"Perc50\": 2148000, \"Perc90\": 13772000, \"Perc99\": 14436000, \"Perc100\": 0 } }, { \"resource\": \"services\", \"verb\": \"DELETE\", \"latency\": { \"Perc50\": 9843000, \"Perc90\": 11226000, \"Perc99\": 12391000, \"Perc100\": 0 } }, ... Apr 25 18:27:31.461: INFO: [Result:Performance] { \"version\": \"v1\", \"dataItems\": [ { \"data\": { \"Perc50\": 2.148, \"Perc90\": 13.772, \"Perc99\": 14.436 }, \"unit\": \"ms\", \"labels\": { \"Resource\": \"pods\", \"Verb\": \"POST\" } }, ... 2.857: INFO: Running AfterSuite actions on all node Apr 26 10:35:32.857: INFO: Running AfterSuite actions on node 1 Ran 2 of 606 Specs in 268.371 seconds SUCCESS! -- 2 Passed | 0 Failed | 0 Pending | 604 Skipped PASS Ginkgo ran 1 suite in 4m28.667870101s Test Suite Passed 从kubemark输出的日志中可以看到API calls latencies和Performance。 日志里显示，创建90个pod用时40秒以内，平均创建每个pod耗时0.44秒。 不同type的资源类型API请求耗时分布 Resource Verb 50% 90% 99% services DELETE 8.472ms 9.841ms 38.226ms endpoints PUT 1.641ms 3.161ms 30.715ms endpoints GET 931µs 10.412ms 27.97ms nodes PATCH 4.245ms 11.117ms 18.63ms pods PUT 2.193ms 2.619ms 17.285ms 从log.txt日志中还可以看到更多详细请求的测试指标。 图片 - kubernetes-dashboard 注意事项 测试过程中需要用到docker镜像存储在GCE中，需要翻墙下载，我没看到哪里配置这个镜像的地址。该镜像副本已上传时速云： 用到的镜像有如下两个： gcr.io/google_containers/pause-amd64:3.0 gcr.io/google_containers/serve_hostname:v1.4 时速云镜像地址： index.tenxcloud.com/jimmy/pause-amd64:3.0 index.tenxcloud.com/jimmy/serve_hostname:v1.4 将镜像pull到本地后重新打tag。 Locust测试 请求统计 Method Name # requests # failures Median response time Average response time Min response time Max response time Average Content Size Requests/s POST /login 5070 78 59000 80551 11218 202140 54 1.17 POST /metrics 5114232 85879 63000 82280 29518 331330 94 1178.77 None Total 5119302 85957 63000 82279 11218 331330 94 1179.94 响应时间分布 Name # requests 50% 66% 75% 80% 90% 95% 98% 99% 100% POST /login 5070 59000 125000 140000 148000 160000 166000 174000 176000 202140 POST /metrics 5114993 63000 127000 142000 149000 160000 166000 172000 176000 331330 None Total 5120063 63000 127000 142000 149000 160000 166000 172000 176000 331330 以上两个表格都是瞬时值。请求失败率在2%左右。 Sample-webapp起了48个pod。 Locust模拟10万用户，每秒增长100个。 图片 - locust测试页面 关于Locust的使用请参考Github：https://github.com/rootsongjc/distributed-load-testing-using-kubernetes 参考 基于 Python 的性能测试工具 locust (与 LR 的简单对比) Locust docs python用户负载测试工具：locust Kubernetes集群性能测试 CoreOS是如何将Kubernetes的性能提高10倍的 Kubernetes 1.3 的性能和弹性 —— 2000 节点，60,0000 Pod 的集群 运用Kubernetes进行分布式负载测试 Kubemark User Guide Flannel host-gw architecture Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-15 21:34:41 "},"practice/edge-node-configuration.html":{"url":"practice/edge-node-configuration.html","title":"边缘节点配置","keywords":"","body":"边缘节点配置 前言 为了配置kubernetes中的traefik ingress的高可用，对于kubernetes集群以外只暴露一个访问入口，需要使用keepalived排除单点问题。本文参考了kube-keepalived-vip，但并没有使用容器方式安装，而是直接在node节点上安装。 定义 首先解释下什么叫边缘节点（Edge Node），所谓的边缘节点即集群内部用来向集群外暴露服务能力的节点，集群外部的服务通过该节点来调用集群内部的服务，边缘节点是集群内外交流的一个Endpoint。 边缘节点要考虑两个问题 边缘节点的高可用，不能有单点故障，否则整个kubernetes集群将不可用 对外的一致暴露端口，即只能有一个外网访问IP和端口 架构 为了满足边缘节点的以上需求，我们使用keepalived来实现。 在Kubernetes中添加了service的同时，在DNS中增加一个记录，这条记录需要跟ingress中的host字段相同，IP地址即VIP的地址，本示例中是172.20.0.119，这样集群外部就可以通过service的DNS名称来访问服务了。 选择Kubernetes的三个node作为边缘节点，并安装keepalived，下图展示了边缘节点的配置，同时展示了向Kubernetes中添加服务的过程。 图片 - 边缘节点架构 准备 复用kubernetes测试集群的三台主机。 172.20.0.113 172.20.0.114 172.20.0.115 安装 使用keepalived管理VIP，VIP是使用IPVS创建的，IPVS已经成为linux内核的模块，不需要安装 LVS的工作原理请参考：http://www.cnblogs.com/codebean/archive/2011/07/25/2116043.html 不使用镜像方式安装了，直接手动安装，指定三个节点为边缘节点（Edge node）。 因为我们的测试集群一共只有三个node，所有在在三个node上都要安装keepalived和ipvsadmin。 yum install keepalived ipvsadm 配置说明 需要对原先的traefik ingress进行改造，从以Deployment方式启动改成DeamonSet。还需要指定一个与node在同一网段的IP地址作为VIP，我们指定成172.20.0.119，配置keepalived前需要先保证这个IP没有被分配。。 Traefik以DaemonSet的方式启动 通过nodeSelector选择边缘节点 通过hostPort暴露端口 当前VIP漂移到了172.20.0.115上 Traefik根据访问的host和path配置，将流量转发到相应的service上 配置keepalived 参考基于keepalived 实现VIP转移，lvs，nginx的高可用，配置keepalived。 keepalived的官方配置文档见：http://keepalived.org/pdf/UserGuide.pdf 配置文件/etc/keepalived/keepalived.conf文件内容如下： ! Configuration File for keepalived global_defs { notification_email { root@localhost } notification_email_from kaadmin@localhost smtp_server 127.0.0.1 smtp_connect_timeout 30 router_id LVS_DEVEL } vrrp_instance VI_1 { state MASTER interface eth0 virtual_router_id 51 priority 100 advert_int 1 authentication { auth_type PASS auth_pass 1111 } virtual_ipaddress { 172.20.0.119 } } virtual_server 172.20.0.119 80{ delay_loop 6 lb_algo loadbalance lb_kind DR nat_mask 255.255.255.0 persistence_timeout 0 protocol TCP real_server 172.20.0.113 80{ weight 1 TCP_CHECK { connect_timeout 3 } } real_server 172.20.0.114 80{ weight 1 TCP_CHECK { connect_timeout 3 } } real_server 172.20.0.115 80{ weight 1 TCP_CHECK { connect_timeout 3 } } } Realserver的IP和端口即traefik供外网访问的IP和端口。 将以上配置分别拷贝到另外两台node的/etc/keepalived目录下。 我们使用转发效率最高的lb_kind DR直接路由方式转发，使用TCP_CHECK来检测real_server的health。 设置keepalived为开机自启动： chkconfig keepalived on 启动keepalived systemctl start keepalived 三台node都启动了keepalived后，观察eth0的IP，会在三台node的某一台上发现一个VIP是172.20.0.119。 $ ip addr show eth0 2: eth0: mtu 1500 qdisc mq state UP qlen 1000 link/ether f4:e9:d4:9f:6b:a0 brd ff:ff:ff:ff:ff:ff inet 172.20.0.115/17 brd 172.20.127.255 scope global eth0 valid_lft forever preferred_lft forever inet 172.20.0.119/32 scope global eth0 valid_lft forever preferred_lft forever 关掉拥有这个VIP主机上的keepalived，观察VIP是否漂移到了另外两台主机的其中之一上。 改造Traefik 在这之前我们启动的traefik使用的是deployment，只启动了一个pod，无法保证高可用（即需要将pod固定在某一台主机上，这样才能对外提供一个唯一的访问地址），现在使用了keepalived就可以通过VIP来访问traefik，同时启动多个traefik的pod保证高可用。 配置文件traefik.yaml内容如下： apiVersion: extensions/v1beta1 kind: DaemonSet metadata: name: traefik-ingress-lb namespace: kube-system labels: k8s-app: traefik-ingress-lb spec: template: metadata: labels: k8s-app: traefik-ingress-lb name: traefik-ingress-lb spec: terminationGracePeriodSeconds: 60 hostNetwork: true restartPolicy: Always serviceAccountName: ingress containers: - image: traefik name: traefik-ingress-lb resources: limits: cpu: 200m memory: 30Mi requests: cpu: 100m memory: 20Mi ports: - name: http containerPort: 80 hostPort: 80 - name: admin containerPort: 8580 hostPort: 8580 args: - --web - --web.address=:8580 - --kubernetes nodeSelector: edgenode: \"true\" 注意，我们使用了nodeSelector选择边缘节点来调度traefik-ingress-lb运行在它上面，所有你需要使用： kubectl label nodes 172.20.0.113 edgenode=true kubectl label nodes 172.20.0.114 edgenode=true kubectl label nodes 172.20.0.115 edgenode=true 给三个node打标签。 查看DaemonSet的启动情况： $ kubectl -n kube-system get ds NAME DESIRED CURRENT READY UP-TO-DATE AVAILABLE NODE-SELECTOR AGE traefik-ingress-lb 3 3 3 3 3 edgenode=true 2h 现在就可以在外网通过172.20.0.119:80来访问到traefik ingress了。 参考 kube-keepalived-vip http://www.keepalived.org/ keepalived工作原理与配置说明 LVS简介及使用 基于keepalived 实现VIP转移，lvs，nginx的高可用 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-05 19:33:27 "},"practice/nginx-ingress-installation.html":{"url":"practice/nginx-ingress-installation.html","title":"安装Nginx ingress","keywords":"","body":"安装Nginx ingress Nginx ingress 使用ConfigMap来管理Nginx配置，nginx是大家熟知的代理和负载均衡软件，比起Traefik来说功能更加强大. 我们使用helm来部署，chart保存在私有的仓库中，请确保您已经安装和配置好helm，helm安装使用见使用Helm管理kubernetes应用。 镜像准备 安装时需要用到的镜像有： sophos/nginx-vts-exporter:v0.6 gcr.io/google_containers/nginx-ingress-controller:0.9.0-beta.15 gcr.io/google_containers/defaultbackend:1.3 gcr.io中的那个两个镜像我复制了一份到时速云，可供大家下载： index.tenxcloud.com/jimmy/defaultbackend:1.3 index.tenxcloud.com/jimmy/nginx-ingress-controller:0.9.0-beta.15 Docker hub上的那个镜像可以直接下载，所有的安装时需要的配置保存在../manifests/nginx-ingress目录下。 步骤详解 安装nginx-ingress chart到本地repo中 修改values.yaml配置，启用RBAC支持，相关配置见nginx-ingress chart。 helm package . 查看niginx-ingress $ helm search nginx-ingress NAME VERSION DESCRIPTION local/nginx-ingress 0.8.9 An nginx Ingress controller that uses ConfigMap... stable/nginx-ingress 0.8.9 An nginx Ingress controller that uses ConfigMap... stable/nginx-lego 0.3.0 Chart for nginx-ingress-controller and kube-lego 使用helm部署nginx-ingress $ helm install --name nginx-ingress local/nginx-ingress NAME: nginx-ingress LAST DEPLOYED: Fri Oct 27 18:26:58 2017 NAMESPACE: default STATUS: DEPLOYED RESOURCES: ==> rbac.authorization.k8s.io/v1beta1/Role NAME KIND nginx-ingress-nginx-ingress Role.v1beta1.rbac.authorization.k8s.io ==> rbac.authorization.k8s.io/v1beta1/RoleBinding nginx-ingress-nginx-ingress RoleBinding.v1beta1.rbac.authorization.k8s.io ==> v1/Service NAME CLUSTER-IP EXTERNAL-IP PORT(S) AGE nginx-ingress-nginx-ingress-controller 10.254.100.108 80:30484/TCP,443:31053/TCP 1s nginx-ingress-nginx-ingress-default-backend 10.254.58.156 80/TCP 1s ==> extensions/v1beta1/Deployment NAME DESIRED CURRENT UP-TO-DATE AVAILABLE AGE nginx-ingress-nginx-ingress-default-backend 1 1 1 0 1s nginx-ingress-nginx-ingress-controller 1 1 1 0 1s ==> v1/ConfigMap NAME DATA AGE nginx-ingress-nginx-ingress-controller 1 1s ==> v1/ServiceAccount NAME SECRETS AGE nginx-ingress-nginx-ingress 1 1s ==> rbac.authorization.k8s.io/v1beta1/ClusterRole NAME KIND nginx-ingress-nginx-ingress ClusterRole.v1beta1.rbac.authorization.k8s.io ==> rbac.authorization.k8s.io/v1beta1/ClusterRoleBinding nginx-ingress-nginx-ingress ClusterRoleBinding.v1beta1.rbac.authorization.k8s.io NOTES: The nginx-ingress controller has been installed. Get the application URL by running these commands: export HTTP_NODE_PORT=$(kubectl --namespace default get services -o jsonpath=\"{.spec.ports[0].nodePort}\" nginx-ingress-nginx-ingress-controller) export HTTPS_NODE_PORT=$(kubectl --namespace default get services -o jsonpath=\"{.spec.ports[1].nodePort}\" nginx-ingress-nginx-ingress-controller) export NODE_IP=$(kubectl --namespace default get nodes -o jsonpath=\"{.items[0].status.addresses[1].address}\") echo \"Visit http://$NODE_IP:$HTTP_NODE_PORT to access your application via HTTP.\" echo \"Visit https://$NODE_IP:$HTTPS_NODE_PORT to access your application via HTTPS.\" An example Ingress that makes use of the controller: apiVersion: extensions/v1beta1 kind: Ingress metadata: annotations: kubernetes.io/ingress.class: nginx name: example namespace: foo spec: rules: - host: www.example.com http: paths: - backend: serviceName: exampleService servicePort: 80 path: / # This section is only required if TLS is to be enabled for the Ingress tls: - hosts: - www.example.com secretName: example-tls If TLS is enabled for the Ingress, a Secret containing the certificate and key must also be provided: apiVersion: v1 kind: Secret metadata: name: example-tls namespace: foo data: tls.crt: tls.key: type: kubernetes.io/tls 访问Nginx 首先获取Nginx的地址，从我们使用helm安装nginx-ingress命令的输出中那个可以看到提示，根据提示执行可以看到nginx的http和https地址： export HTTP_NODE_PORT=$(kubectl --namespace default get services -o jsonpath=\"{.spec.ports[0].nodePort}\" nginx-ingress-nginx-ingress-controller) export HTTPS_NODE_PORT=$(kubectl --namespace default get services -o jsonpath=\"{.spec.ports[1].nodePort}\" nginx-ingress-nginx-ingress-controller) export NODE_IP=$(kubectl --namespace default get nodes -o jsonpath=\"{.items[0].status.addresses[1].address}\") echo \"Visit http://$NODE_IP:$HTTP_NODE_PORT to access your application via HTTP.\" echo \"Visit https://$NODE_IP:$HTTPS_NODE_PORT to access your application via HTTPS.\" Visit http://172.20.0.113:30484 to access your application via HTTP. Visit https://172.20.0.113:31053 to access your application via HTTPS. http地址：http://172.20.0.113:30484 https地址：https://172.20.0.113:31053 我们分别在http和https地址上测试一下： /healthz返回200 /返回404错误 curl -v http://172.20.0.113:30484/healthz # 返回200 curl -v http://172.20.0.113:30484/ # 返回404 curl -v --insecure http://172.20.0.113:30484/healthz # 返回200 curl -v --insecure http://172.20.0.113:30484/ # 返回404 删除nginx-ingress helm delete --purge nginx-ingress 使用--purge参数可以彻底删除release不留下记录，否则下一次部署的时候不能使用重名的release。 参考 Ingress-nginx github Nginx chart configuration 使用Helm管理kubernetes应用 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-10-27 19:39:00 "},"practice/operation.html":{"url":"practice/operation.html","title":"运维管理","keywords":"","body":"运维管理 将集群部署到生产环境后就不得不考虑运维管理问题。运维管理问题主要包括如下几个方面： 监控：包括 kubernetes 本身组件和 Pod、应用的监控 日志收集：包括 kubernetes 本身组件的日志，应用的日志 审计：用户对集群操作的审计 安全：用户权限的管理和镜像漏洞扫描 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-08-28 18:37:34 "},"practice/service-rolling-update.html":{"url":"practice/service-rolling-update.html","title":"服务滚动升级","keywords":"","body":"服务滚动升级 当有镜像发布新版本，新版本服务上线时如何实现服务的滚动和平滑升级？ 如果你使用ReplicationController创建的pod可以使用kubectl rollingupdate命令滚动升级，如果使用的是Deployment创建的Pod可以直接修改yaml文件后执行kubectl apply即可。 Deployment已经内置了RollingUpdate strategy，因此不用再调用kubectl rollingupdate命令，升级的过程是先创建新版的pod将流量导入到新pod上后销毁原来的旧的pod。 Rolling Update适用于Deployment、Replication Controller，官方推荐使用Deployment而不再使用Replication Controller。 使用ReplicationController时的滚动升级请参考官网说明：https://kubernetes.io/docs/tasks/run-application/rolling-update-replication-controller/ ReplicationController与Deployment的关系 ReplicationController和Deployment的RollingUpdate命令有些不同，但是实现的机制是一样的，关于这两个kind的关系我引用了ReplicationController与Deployment的区别中的部分内容如下，详细区别请查看原文。 ReplicationController Replication Controller为Kubernetes的一个核心内容，应用托管到Kubernetes之后，需要保证应用能够持续的运行，Replication Controller就是这个保证的key，主要的功能如下： 确保pod数量：它会确保Kubernetes中有指定数量的Pod在运行。如果少于指定数量的pod，Replication Controller会创建新的，反之则会删除掉多余的以保证Pod数量不变。 确保pod健康：当pod不健康，运行出错或者无法提供服务时，Replication Controller也会杀死不健康的pod，重新创建新的。 弹性伸缩 ：在业务高峰或者低峰期的时候，可以通过Replication Controller动态的调整pod的数量来提高资源的利用率。同时，配置相应的监控功能（Hroizontal Pod Autoscaler），会定时自动从监控平台获取Replication Controller关联pod的整体资源使用情况，做到自动伸缩。 滚动升级：滚动升级为一种平滑的升级方式，通过逐步替换的策略，保证整体系统的稳定，在初始化升级的时候就可以及时发现和解决问题，避免问题不断扩大。 Deployment Deployment同样为Kubernetes的一个核心内容，主要职责同样是为了保证pod的数量和健康，90%的功能与Replication Controller完全一样，可以看做新一代的Replication Controller。但是，它又具备了Replication Controller之外的新特性： Replication Controller全部功能：Deployment继承了上面描述的Replication Controller全部功能。 事件和状态查看：可以查看Deployment的升级详细进度和状态。 回滚：当升级pod镜像或者相关参数的时候发现问题，可以使用回滚操作回滚到上一个稳定的版本或者指定的版本。 版本记录: 每一次对Deployment的操作，都能保存下来，给予后续可能的回滚使用。 暂停和启动：对于每一次升级，都能够随时暂停和启动。 多种升级方案：Recreate：删除所有已存在的pod,重新创建新的; RollingUpdate：滚动升级，逐步替换的策略，同时滚动升级时，支持更多的附加参数，例如设置最大不可用pod数量，最小升级间隔时间等等。 创建测试镜像 我们来创建一个特别简单的web服务，当你访问网页时，将输出一句版本信息。通过区分这句版本信息输出我们就可以断定升级是否完成。 所有配置和代码见../manifests/test/rolling-update-test目录。 Web服务的代码main.go package main import ( \"fmt\" \"log\" \"net/http\" ) func sayhello(w http.ResponseWriter, r *http.Request) { fmt.Fprintf(w, \"This is version 1.\") //这个写入到w的是输出到客户端的 } func main() { http.HandleFunc(\"/\", sayhello) //设置访问的路由 log.Println(\"This is version 1.\") err := http.ListenAndServe(\":9090\", nil) //设置监听的端口 if err != nil { log.Fatal(\"ListenAndServe: \", err) } } 创建Dockerfile FROM alpine:3.5 MAINTAINER Jimmy Song ADD hellov2 / ENTRYPOINT [\"/hellov2\"] 注意修改添加的文件的名称。 创建Makefile 修改镜像仓库的地址为你自己的私有镜像仓库地址。 修改Makefile中的TAG为新的版本号。 all: build push clean .PHONY: build push clean TAG = v1 # Build for linux amd64 build: GOOS=linux GOARCH=amd64 go build -o hello${TAG} main.go docker build -t sz-pg-oam-docker-hub-001.tendcloud.com/library/hello:${TAG} . # Push to tenxcloud push: docker push sz-pg-oam-docker-hub-001.tendcloud.com/library/hello:${TAG} # Clean clean: rm -f hello${TAG} 编译 make all 分别修改main.go中的输出语句、Dockerfile中的文件名称和Makefile中的TAG，创建两个版本的镜像。 测试 我们使用Deployment部署服务来测试。 配置文件rolling-update-test.yaml： apiVersion: extensions/v1beta1 kind: Deployment metadata: name: rolling-update-test spec: replicas: 3 template: metadata: labels: app: rolling-update-test spec: containers: - name: rolling-update-test image: sz-pg-oam-docker-hub-001.tendcloud.com/library/hello:v1 ports: - containerPort: 9090 --- apiVersion: v1 kind: Service metadata: name: rolling-update-test labels: app: rolling-update-test spec: ports: - port: 9090 protocol: TCP name: http selector: app: rolling-update-test 部署service kubectl create -f rolling-update-test.yaml 修改traefik ingress配置 在ingress.yaml文件中增加新service的配置。 - host: rolling-update-test.traefik.io http: paths: - path: / backend: serviceName: rolling-update-test servicePort: 9090 修改本地的host配置，增加一条配置： 172.20.0.119 rolling-update-test.traefik.io 注意：172.20.0.119是我们之前使用keepalived创建的VIP。 打开浏览器访问http://rolling-update-test.traefik.io将会看到以下输出： This is version 1. 滚动升级 只需要将rolling-update-test.yaml文件中的image改成新版本的镜像名，然后执行： kubectl apply -f rolling-update-test.yaml 也可以参考Kubernetes Deployment Concept中的方法，直接设置新的镜像。 kubectl set image deployment/rolling-update-test rolling-update-test=sz-pg-oam-docker-hub-001.tendcloud.com/library/hello:v2 或者使用kubectl edit deployment/rolling-update-test修改镜像名称后保存。 使用以下命令查看升级进度： kubectl rollout status deployment/rolling-update-test 升级完成后在浏览器中刷新http://rolling-update-test.traefik.io将会看到以下输出： This is version 2. 说明滚动升级成功。 使用ReplicationController创建的Pod如何RollingUpdate 以上讲解使用Deployment创建的Pod的RollingUpdate方式，那么如果使用传统的ReplicationController创建的Pod如何Update呢？ 举个例子： $ kubectl -n spark-cluster rolling-update zeppelin-controller --image sz-pg-oam-docker-hub-001.tendcloud.com/library/zeppelin:0.7.1 Created zeppelin-controller-99be89dbbe5cd5b8d6feab8f57a04a8b Scaling up zeppelin-controller-99be89dbbe5cd5b8d6feab8f57a04a8b from 0 to 1, scaling down zeppelin-controller from 1 to 0 (keep 1 pods available, don't exceed 2 pods) Scaling zeppelin-controller-99be89dbbe5cd5b8d6feab8f57a04a8b up to 1 Scaling zeppelin-controller down to 0 Update succeeded. Deleting old controller: zeppelin-controller Renaming zeppelin-controller-99be89dbbe5cd5b8d6feab8f57a04a8b to zeppelin-controller replicationcontroller \"zeppelin-controller\" rolling updated 只需要指定新的镜像即可，当然你可以配置RollingUpdate的策略。 参考 Rolling update机制解析 Running a Stateless Application Using a Deployment Simple Rolling Update 使用kubernetes的deployment进行RollingUpdate Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-10-11 23:40:19 "},"practice/app-log-collection.html":{"url":"practice/app-log-collection.html","title":"应用日志收集","keywords":"","body":"应用日志收集 前言 在进行日志收集的过程中，我们首先想到的是使用Logstash，因为它是ELK stack中的重要成员，但是在测试过程中发现，Logstash是基于JDK的，在没有产生日志的情况单纯启动Logstash就大概要消耗500M内存，在每个Pod中都启动一个日志收集组件的情况下，使用logstash有点浪费系统资源，经人推荐我们选择使用Filebeat替代，经测试单独启动Filebeat容器大约会消耗12M内存，比起logstash相当轻量级。 方案选择 Kubernetes官方提供了EFK的日志收集解决方案，但是这种方案并不适合所有的业务场景，它本身就有一些局限性，例如： 所有日志都必须是out前台输出，真实业务场景中无法保证所有日志都在前台输出 只能有一个日志输出文件，而真实业务场景中往往有多个日志输出文件 Fluentd并不是常用的日志收集工具，我们更习惯用logstash，现使用filebeat替代 我们已经有自己的ELK集群且有专人维护，没有必要再在kubernetes上做一个日志收集服务 基于以上几个原因，我们决定使用自己的ELK集群。 Kubernetes集群中的日志收集解决方案 编号 方案 优点 缺点 1 每个app的镜像中都集成日志收集组件 部署方便，kubernetes的yaml文件无须特别配置，可以为每个app自定义日志收集配置 强耦合，不方便应用和日志收集组件升级和维护且会导致镜像过大 2 单独创建一个日志收集组件跟app的容器一起运行在同一个pod中 低耦合，扩展性强，方便维护和升级 需要对kubernetes的yaml文件进行单独配置，略显繁琐 3 将所有的Pod的日志都挂载到宿主机上，每台主机上单独起一个日志收集Pod 完全解耦，性能最高，管理起来最方便 需要统一日志收集规则，目录和输出方式 综合以上优缺点，我们选择使用方案二。 该方案在扩展性、个性化、部署和后期维护方面都能做到均衡，因此选择该方案。 图片 - filebeat日志收集架构图 我们创建了自己的filebeat镜像。创建过程和使用方式见https://github.com/rootsongjc/docker-images 镜像地址：index.tenxcloud.com/jimmy/filebeat:5.4.0 测试 我们部署一个应用filebeat来收集日志的功能测试。 创建应用yaml文件filebeat-test.yaml。 apiVersion: extensions/v1beta1 kind: Deployment metadata: name: filebeat-test namespace: default spec: replicas: 3 template: metadata: labels: k8s-app: filebeat-test spec: containers: - image: sz-pg-oam-docker-hub-001.tendcloud.com/library/filebeat:5.4.0 name: filebeat volumeMounts: - name: app-logs mountPath: /log - name: filebeat-config mountPath: /etc/filebeat/ - image: sz-pg-oam-docker-hub-001.tendcloud.com/library/analytics-docker-test:Build_8 name : app ports: - containerPort: 80 volumeMounts: - name: app-logs mountPath: /usr/local/TalkingData/logs volumes: - name: app-logs emptyDir: {} - name: filebeat-config configMap: name: filebeat-config --- apiVersion: v1 kind: Service metadata: name: filebeat-test labels: app: filebeat-test spec: ports: - port: 80 protocol: TCP name: http selector: run: filebeat-test --- apiVersion: v1 kind: ConfigMap metadata: name: filebeat-config data: filebeat.yml: | filebeat.prospectors: - input_type: log paths: - \"/log/*\" - \"/log/usermange/common/*\" output.elasticsearch: hosts: [\"172.23.5.255:9200\"] username: \"elastic\" password: \"changeme\" index: \"filebeat-docker-test\" 说明 该文件中包含了配置文件filebeat的配置文件的ConfigMap，因此不需要再定义环境变量。 当然你也可以不同ConfigMap，通过传统的传递环境变量的方式来配置filebeat。 例如对filebeat的容器进行如下配置： containers: - image: sz-pg-oam-docker-hub-001.tendcloud.com/library/filebeat:5.4.0 name: filebeat volumeMounts: - name: app-logs mountPath: /log env: - name: PATHS value: \"/log/*\" - name: ES_SERVER value: 172.23.5.255:9200 - name: INDEX value: logstash-docker - name: INPUT_TYPE value: log 目前使用这种方式会有个问题，及时PATHS只能传递单个目录，如果想传递多个目录需要修改filebeat镜像的docker-entrypoint.sh脚本，对该环境变量进行解析增加filebeat.yml文件中的PATHS列表。 推荐使用ConfigMap，这样filebeat的配置就能够更灵活。 注意事项 将app的/usr/local/TalkingData/logs目录挂载到filebeat的/log目录下。 该文件可以在manifests/test/filebeat-test.yaml找到。 我使用了自己的私有镜像仓库，测试时请换成自己的应用镜像。 Filebeat的环境变量的值配置请参考https://github.com/rootsongjc/docker-images 创建应用 部署Deployment kubectl create -f filebeat-test.yaml 查看http://172.23.5.255:9200/_cat/indices将可以看到列表有这样的indices： green open filebeat-docker-test 7xPEwEbUQRirk8oDX36gAA 5 1 2151 0 1.6mb 841.8kb 访问Kibana的web页面，查看filebeat-2017.05.17的索引，可以看到filebeat收集到了app日志。 图片 - Kibana页面 点开没个日志条目，可以看到以下详细字段： 图片 - filebeat收集的日志详细信息 _index值即我们在YAML文件的configMap中配置的index值 beat.hostname和beat.name即pod的名称 source表示filebeat容器中的日志目录 我们可以通过人为得使index = service name，这样就可以方便的收集和查看每个service的日志。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-10-11 22:22:55 "},"practice/configuration-best-practice.html":{"url":"practice/configuration-best-practice.html","title":"配置最佳实践","keywords":"","body":"配置最佳实践 本文档旨在汇总和强调用户指南、快速开始文档和示例中的最佳实践。该文档会很活跃并持续更新中。如果你觉得很有用的最佳实践但是本文档中没有包含，欢迎给我们提Pull Request。 通用配置建议 定义配置文件的时候，指定最新的稳定API版本（目前是V1）。 在配置文件push到集群之前应该保存在版本控制系统中。这样当需要的时候能够快速回滚，必要的时候也可以快速的创建集群。 使用YAML格式而不是JSON格式的配置文件。在大多数场景下它们都可以作为数据交换格式，但是YAML格式比起JSON更易读和配置。 尽量将相关的对象放在同一个配置文件里。这样比分成多个文件更容易管理。参考guestbook-all-in-one.yaml文件中的配置（注意，尽管你可以在使用kubectl命令时指定配置文件目录，你也可以在配置文件目录下执行kubectl create——查看下面的详细信息）。 为了简化和最小化配置，也为了防止错误发生，不要指定不必要的默认配置。例如，省略掉ReplicationController的selector和label，如果你希望它们跟podTemplate中的label一样的话，因为那些配置默认是podTemplate的label产生的。更多信息请查看 guestbook app 的yaml文件和 examples 。 将资源对象的描述放在一个annotation中可以更好的内省。 裸的Pods vs Replication Controllers和 Jobs 如果有其他方式替代“裸的“ pod（如没有绑定到replication controller 上的pod），那么就使用其他选择。在node节点出现故障时，裸奔的pod不会被重新调度。Replication Controller总是会重新创建pod，除了明确指定了restartPolicy: Never 的场景。Job 也许是比较合适的选择。 Services 通常最好在创建相关的replication controllers之前先创建service ，你也可以在创建Replication Controller的时候不指定replica数量（默认是1），创建service后，在通过Replication Controller来扩容。这样可以在扩容很多个replica之前先确认pod是正常的。 除非十分必要的情况下（如运行一个node daemon），不要使用hostPort（用来指定暴露在主机上的端口号）。当你给Pod绑定了一个hostPort，该pod可被调度到的主机的受限了，因为端口冲突。如果是为了调试目的来通过端口访问的话，你可以使用 kubectl proxy and apiserver proxy 或者 kubectl port-forward。你可使用 Service 来对外暴露服务。如果你确实需要将pod的端口暴露到主机上，考虑使用 NodePort service。 跟hostPort一样的原因，避免使用 hostNetwork。 如果你不需要kube-proxy的负载均衡的话，可以考虑使用使用headless services。 使用Label 定义 labels 来指定应用或Deployment的 semantic attributes 。例如，不是将label附加到一组pod来显式表示某些服务（例如，service:myservice），或者显式地表示管理pod的replication controller（例如，controller:mycontroller），附加label应该是标示语义属性的标签， 例如{app:myapp,tier:frontend,phase:test,deployment:v3}。 这将允许您选择适合上下文的对象组——例如，所有的”tier:frontend“pod的服务或app是“myapp”的所有“测试”阶段组件。 有关此方法的示例，请参阅guestbook应用程序。 可以通过简单地从其service的选择器中省略特定于发行版本的标签，而不是更新服务的选择器来完全匹配replication controller的选择器，来实现跨越多个部署的服务，例如滚动更新。 为了滚动升级的方便，在Replication Controller的名字中包含版本信息，例如作为名字的后缀。设置一个version标签页是很有用的。滚动更新创建一个新的controller而不是修改现有的controller。因此，version含混不清的controller名字就可能带来问题。查看Rolling Update Replication Controller文档获取更多关于滚动升级命令的信息。 注意 Deployment 对象不需要再管理 replication controller 的版本名。Deployment 中描述了对象的期望状态，如果对spec的更改被应用了话，Deployment controller 会以控制的速率来更改实际状态到期望状态。（Deployment目前是 extensions API Group的一部分）。 利用label做调试。因为Kubernetes replication controller和service使用label来匹配pods，这允许你通过移除pod中的label的方式将其从一个controller或者service中移除，原来的controller会创建一个新的pod来取代移除的pod。这是一个很有用的方式，帮你在一个隔离的环境中调试之前的“活着的” pod。查看 kubectl label 命令。 容器镜像 默认容器镜像拉取策略 是 IfNotPresent, 当本地已存在该镜像的时候 Kubelet 不会再从镜像仓库拉取。如果你希望总是从镜像仓库中拉取镜像的话，在yaml文件中指定镜像拉取策略为Always（ imagePullPolicy: Always）或者指定镜像的tag为 :latest 。 如果你没有将镜像标签指定为:latest，例如指定为myimage:v1，当该标签的镜像进行了更新，kubelet也不会拉取该镜像。你可以在每次镜像更新后都生成一个新的tag（例如myimage:v2），在配置文件中明确指定该版本。 注意： 在生产环境下部署容器应该尽量避免使用:latest标签，因为这样很难追溯到底运行的是哪个版本的容器和回滚。 使用kubectl 尽量使用 kubectl create -f 。kubeclt会自动查找该目录下的所有后缀名为.yaml、.yml和.json文件并将它们传递给create命令。 使用 kubectl delete 而不是 stop. Delete 是 stop的超集，stop 已经被弃用。 使用 kubectl bulk 操作（通过文件或者label）来get和delete。查看label selectors 和 using labels effectively。 使用 kubectl run 和 expose 命令快速创建只有单个容器的Deployment。查看 quick start guide中的示例。 参考 Configuration Best Practices Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-08-21 18:23:34 "},"practice/monitor.html":{"url":"practice/monitor.html","title":"集群及应用监控","keywords":"","body":"集群及应用监控 在前面的安装heapster插件章节，我们已经谈到Kubernetes本身提供了监控插件作为集群和容器监控的选择，但是在实际使用中，因为种种原因，再考虑到跟我们自身的监控系统集成，我们准备重新造轮子。 针对kubernetes集群和应用的监控，相较于传统的虚拟机和物理机的监控有很多不同，因此对于传统监控需要有很多改造的地方，需要关注以下三个方面： Kubernetes集群本身的监控，主要是kubernetes的各个组件 kubernetes集群中Pod的监控，Pod的CPU、内存、网络、磁盘等监控 集群内部应用的监控，针对应用本身的监控 Kubernetes集群中的监控 图片 - Kubernetes集群中的监控 跟物理机器和虚拟机的监控不同，在kuberntes集群中的监控复杂度更高一些，因为多了一个虚拟化层，当然这个跟直接监控docker容器又不一样，kubernetes在docker之上又抽象了一层service的概念。 在kubernetes中的监控需要考虑到这几个方面： 应该给Pod打上哪些label，这些label将成为监控的metrics。 当应用的Pod漂移了之后怎么办？因为要考虑到Pod的生命周期比虚拟机和物理机短的多，如何持续监控应用的状态？ 更多的监控项，kubernetes本身、容器、应用等。 监控指标的来源，是通过heapster收集后汇聚还是直接从每台主机的docker上取？ 容器的命名规则 首先我们需要清楚使用cAdvisor收集的数据的格式和字段信息。 当我们通过cAdvisor获取到了容器的信息后，例如访问${NODE_IP}:4194/api/v1.3/docker获取的json结果中的某个容器包含如下字段： \"labels\": { \"annotation.io.kubernetes.container.hash\": \"f47f0602\", \"annotation.io.kubernetes.container.ports\": \"[{\\\"containerPort\\\":80,\\\"protocol\\\":\\\"TCP\\\"}]\", \"annotation.io.kubernetes.container.restartCount\": \"0\", \"annotation.io.kubernetes.container.terminationMessagePath\": \"/dev/termination-log\", \"annotation.io.kubernetes.container.terminationMessagePolicy\": \"File\", \"annotation.io.kubernetes.pod.terminationGracePeriod\": \"30\", \"io.kubernetes.container.logpath\": \"/var/log/pods/d8a2e995-3617-11e7-a4b0-ecf4bbe5d414/php-redis_0.log\", \"io.kubernetes.container.name\": \"php-redis\", \"io.kubernetes.docker.type\": \"container\", \"io.kubernetes.pod.name\": \"frontend-2337258262-771lz\", \"io.kubernetes.pod.namespace\": \"default\", \"io.kubernetes.pod.uid\": \"d8a2e995-3617-11e7-a4b0-ecf4bbe5d414\", \"io.kubernetes.sandbox.id\": \"843a0f018c0cef2a5451434713ea3f409f0debc2101d2264227e814ca0745677\" }, 这些信息其实都是kubernetes创建容器时给docker container打的Labels，使用docker inspect $conainer_name命令同样可以看到上述信息。 你是否想过这些label跟容器的名字有什么关系？当你在node节点上执行docker ps看到的容器名字又对应哪个应用的Pod呢？ 在kubernetes代码中pkg/kubelet/dockertools/docker.go中的BuildDockerName方法定义了容器的名称规范。 这段容器名称定义代码如下： // Creates a name which can be reversed to identify both full pod name and container name. // This function returns stable name, unique name and a unique id. // Although rand.Uint32() is not really unique, but it's enough for us because error will // only occur when instances of the same container in the same pod have the same UID. The // chance is really slim. func BuildDockerName(dockerName KubeletContainerName, container *v1.Container) (string, string, string) { containerName := dockerName.ContainerName + \".\" + strconv.FormatUint(kubecontainer.HashContainerLegacy(container), 16) stableName := fmt.Sprintf(\"%s_%s_%s_%s\", containerNamePrefix, containerName, dockerName.PodFullName, dockerName.PodUID) UID := fmt.Sprintf(\"%08x\", rand.Uint32()) return stableName, fmt.Sprintf(\"%s_%s\", stableName, UID), UID } // Unpacks a container name, returning the pod full name and container name we would have used to // construct the docker name. If we are unable to parse the name, an error is returned. func ParseDockerName(name string) (dockerName *KubeletContainerName, hash uint64, err error) { // For some reason docker appears to be appending '/' to names. // If it's there, strip it. name = strings.TrimPrefix(name, \"/\") parts := strings.Split(name, \"_\") if len(parts) == 0 || parts[0] != containerNamePrefix { err = fmt.Errorf(\"failed to parse Docker container name %q into parts\", name) return nil, 0, err } if len(parts) 1 { hash, err = strconv.ParseUint(nameParts[1], 16, 32) if err != nil { glog.Warningf(\"invalid container hash %q in container %q\", nameParts[1], name) } } podFullName := parts[2] + \"_\" + parts[3] podUID := types.UID(parts[4]) return &KubeletContainerName{podFullName, podUID, containerName}, hash, nil } 我们可以看到容器名称中包含如下几个字段，中间用下划线隔开，至少有6个字段，未来可能添加更多字段。 下面的是四个基本字段。 containerNamePrefix_containerName_PodFullName_PodUID 所有kubernetes启动的容器的containerNamePrefix都是k8s。 Kubernetes启动的docker容器的容器名称规范，下面以官方示例guestbook为例，Deployment 名为 frontend中启动的名为php-redis的docker容器的副本书为3。 Deployment frontend的配置如下： apiVersion: extensions/v1beta1 kind: Deployment metadata: name: frontend spec: template: metadata: labels: app: guestbook tier: frontend spec: containers: - name: php-redis image: bj-xg-oam-docker-hub-001.tendcloud.com/library/gb-frontend:v4 resources: requests: cpu: 100m memory: 100Mi env: - name: GET_HOSTS_FROM value: dns ports: - containerPort: 80 我们选取三个实例中的一个运行php-redis的docker容器。 k8s_php-redis_frontend-2337258262-154p7_default_d8a2e2dd-3617-11e7-a4b0-ecf4bbe5d414_0 containerNamePrefix：k8s containerName：php-redis podFullName：frontend-2337258262-154p7 computeHash：154p7 deploymentName：frontend replicaSetName：frontend-2337258262 namespace：default podUID：d8a2e2dd-3617-11e7-a4b0-ecf4bbe5d414 kubernetes容器命名规则解析，见下图所示。 图片 - kubernetes的容器命名规则示意图 使用Heapster进行集群监控 Heapster是kubernetes官方提供的监控方案，我们在前面的章节中已经讲解了如何部署和使用heapster，见安装Heapster插件。 但是Grafana显示的指标只根据Namespace和Pod两层来分类，实在有些单薄，我们希望通过应用的label增加service这一层分类。架构图如下： 图片 - Heapster架构图（改进版） 在不改变原有架构的基础上，通过应用的label来区分不同应用的pod。 应用监控 Kubernetes中应用的监控架构如图： 图片 - 应用监控架构图 这种方式有以下几个要点： 访问kubernetes API获取应用Pod的IP和端口 Pod labels作为监控metric的tag 直接访问应用的Pod的IP和端口获取应用监控数据 metrics发送到OWL中存储和展示 应用拓扑状态图 对于复杂的应用编排和依赖关系，我们希望能够有清晰的图标一览应用状态和拓扑关系，因此我们用到了Weaveworks开源的scope。 安装scope 我们在kubernetes集群上使用standalone方式安装，详情参考Installing Weave Scope。 使用scope.yaml文件安装scope，该服务安装在kube-system namespace下。 $ kubectl apply -f scope.yaml 创建一个新的Ingress：kube-system.yaml，配置如下： apiVersion: extensions/v1beta1 kind: Ingress metadata: name: traefik-ingress namespace: kube-system spec: rules: - host: scope.weave.io http: paths: - path: / backend: serviceName: weave-scope-app servicePort: 80 执行kubectl apply -f kube-system.yaml后在你的主机上的/etc/hosts文件中添加一条记录： 172.20.0.119 scope.weave.io 在浏览器中访问scope.weave.io就可以访问到scope了，详见边缘节点配置。 图片 - 应用拓扑图 如上图所示，scope可以监控kubernetes集群中的一系列资源的状态、资源使用情况、应用拓扑、scale、还可以直接通过浏览器进入容器内部调试等。 参考 Monitoring in the Kubernetes Era Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-10-11 23:41:27 "},"practice/data-persistence-problem.html":{"url":"practice/data-persistence-problem.html","title":"数据持久化问题","keywords":"","body":"数据落盘问题的由来 这本质上是数据持久化问题，对于有些应用依赖持久化数据，比如应用自身产生的日志需要持久化存储的情况，需要保证容器里的数据不丢失，在Pod挂掉后，其他应用依然可以访问到这些数据，因此我们需要将数据持久化存储起来。 数据落盘问题解决方案 下面以一个应用的日志收集为例，该日志需要持久化收集到ElasticSearch集群中，如果不考虑数据丢失的情形，可以直接使用前面提到的应用日志收集一节中的方法，但考虑到Pod挂掉时logstash（或filebeat）并没有收集完该pod内日志的情形，我们想到了如下这种解决方案，示意图如下： 图片 - 日志持久化收集解决方案示意图 首先需要给数据落盘的应用划分node，即这些应用只调用到若干台主机上 给这若干台主机增加label 使用deamonset方式在这若干台主机上启动logstash的Pod（使用nodeSelector来限定在这几台主机上，我们在边缘节点启动的treafik也是这种模式） 将应用的数据通过volume挂载到宿主机上 Logstash（或者filebeat）收集宿主机上的数据，数据持久化不会丢失 Side-effect 首先kubernetes本身就提供了数据持久化的解决方案statefulset，不过需要用到公有云的存储货其他分布式存储，这一点在我们的私有云环境里被否定了。 需要管理主机的label，增加运维复杂度，但是具体问题具体对待 必须保证应用启动顺序，需要先启动logstash 为主机打label使用nodeSelector的方式限制了资源调度的范围 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-08-21 18:23:34 "},"practice/manage-compute-resources-container.html":{"url":"practice/manage-compute-resources-container.html","title":"管理容器的计算资源","keywords":"","body":"管理容器的计算资源 当您定义 Pod 的时候可以选择为每个容器指定需要的 CPU 和内存（RAM）大小。当为容器指定了资源请求后，调度器就能够更好的判断出将容器调度到哪个节点上。如果您还为容器指定了资源限制，节点上的资源就可以按照指定的方式做竞争。关于资源请求和限制的不同点和更多资料请参考 Resource QoS。 资源类型 CPU 和 memory 都是 资源类型。资源类型具有基本单位。CPU 的单位是 core，memory 的单位是 byte。 CPU和内存统称为计算资源，也可以称为资源。计算资源的数量是可以被请求、分配和消耗的可测量的。它们与 API 资源 不同。 API 资源（如 Pod 和 Service）是可通过 Kubernetes API server 读取和修改的对象。 Pod 和 容器的资源请求和限制 Pod 中的每个容器都可以指定以下的一个或者多个值： spec.containers[].resources.limits.cpu spec.containers[].resources.limits.memory spec.containers[].resources.requests.cpu spec.containers[].resources.requests.memory 尽管只能在个别容器上指定请求和限制，但是我们可以方便地计算出 Pod 资源请求和限制。特定资源类型的Pod 资源请求/限制是 Pod 中每个容器的该类型的资源请求/限制的总和。 CPU 的含义 CPU 资源的限制和请求以 cpu 为单位。 Kubernetes 中的一个 cpu 等于： 1 AWS vCPU 1 GCP Core 1 Azure vCore 1 Hyperthread 在带有超线程的裸机 Intel 处理器上 允许浮点数请求。具有 spec.containers[].resources.requests.cpu 为 0.5 的容器保证了一半 CPU 要求 1 CPU的一半。表达式 0.1 等价于表达式 100m，可以看作 “100 millicpu”。有些人说成是“一百毫 cpu”，其实说的是同样的事情。具有小数点（如 0.1）的请求由 API 转换为100m，精度不超过 1m。因此，可能会优先选择 100m 的形式。 CPU 总是要用绝对数量，不可以使用相对数量；0.1 的 CPU 在单核、双核、48核的机器中的意义是一样的。 内存的含义 内存的限制和请求以字节为单位。您可以使用以下后缀之一作为平均整数或定点整数表示内存：E，P，T，G，M，K。您还可以使用两个字母的等效的幂数：Ei，Pi，Ti ，Gi，Mi，Ki。例如，以下代表大致相同的值： 128974848, 129e6, 129M, 123Mi 下面是个例子。 以下 Pod 有两个容器。每个容器的请求为 0.25 cpu 和 64MiB（226 字节）内存，每个容器的限制为 0.5 cpu 和 128MiB 内存。您可以说该 Pod 请求 0.5 cpu 和 128 MiB 的内存，限制为 1 cpu 和 256MiB 的内存。 apiVersion: v1 kind: Pod metadata: name: frontend spec: containers: - name: db image: mysql resources: requests: memory: \"64Mi\" cpu: \"250m\" limits: memory: \"128Mi\" cpu: \"500m\" - name: wp image: wordpress resources: requests: memory: \"64Mi\" cpu: \"250m\" limits: memory: \"128Mi\" cpu: \"500m\" 具有资源请求的 Pod 如何调度 当您创建一个 Pod 时，Kubernetes 调度程序将为 Pod 选择一个节点。每个节点具有每种资源类型的最大容量：可为 Pod 提供的 CPU 和内存量。调度程序确保对于每种资源类型，调度的容器的资源请求的总和小于节点的容量。请注意，尽管节点上的实际内存或 CPU 资源使用量非常低，但如果容量检查失败，则调度程序仍然拒绝在该节点上放置 Pod。当资源使用量稍后增加时，例如在请求率的每日峰值期间，这可以防止节点上的资源短缺。 具有资源限制的 Pod 如何运行 当 kubelet 启动一个 Pod 的容器时，它会将 CPU 和内存限制传递到容器运行时。 当使用 Docker 时： spec.containers[].resources.requests.cpu 的值将转换成 millicore 值，这是个浮点数，并乘以1024，这个数字中的较大者或2用作 docker run 命令中的 --cpu-shares 标志的值。 spec.containers[].resources.limits.cpu 被转换成 millicore 值。被乘以 100000 然后 除以 1000。这个数字用作 docker run 命令中的 --cpu-quota 标志的值。[--cpu-quota ] 标志被设置成了 100000，表示测量配额使用的默认100ms 周期。如果 [--cpu-cfs-quota] 标志设置为 true，则 kubelet 会强制执行 cpu 限制。从 Kubernetes 1.2 版本起，此标志默认为 true。 spec.containers[].resources.limits.memory 被转换为整型，作为 docker run 命令中的 --memory 标志的值。 如果容器超过其内存限制，则可能会被终止。如果可重新启动，则与所有其他类型的运行时故障一样，kubelet 将重新启动它。 如果一个容器超过其内存请求，那么当节点内存不足时，它的 Pod 可能被逐出。 容器可能被允许也可能不被允许超过其 CPU 限制时间。但是，由于 CPU 使用率过高，不会被杀死。 要确定容器是否由于资源限制而无法安排或被杀死，请参阅 疑难解答 部分。 监控计算资源使用 Pod 的资源使用情况被报告为 Pod 状态的一部分。 如果为集群配置了 可选监控，则可以从监控系统检索 Pod 资源的使用情况。 疑难解答 我的 Pod 处于 pending 状态且事件信息显示 failedScheduling 如果调度器找不到任何该 Pod 可以匹配的节点，则该 Pod 将保持不可调度状态，直到找到一个可以被调度到的位置。每当调度器找不到 Pod 可以调度的地方时，会产生一个事件，如下所示： $ kubectl describe pod frontend | grep -A 3 Events Events: FirstSeen LastSeen Count From Subobject PathReason Message 36s 5s 6 {scheduler } FailedScheduling Failed for reason PodExceedsFreeCPU and possibly others 在上述示例中，由于节点上的 CPU 资源不足，名为 “frontend” 的 Pod 将无法调度。由于内存不足（PodExceedsFreeMemory），类似的错误消息也可能会导致失败。一般来说，如果有这种类型的消息而处于 pending 状态，您可以尝试如下几件事情： $ kubectl describe nodes e2e-test-minion-group-4lw4 Name: e2e-test-minion-group-4lw4 [ ... lines removed for clarity ...] Capacity: alpha.kubernetes.io/nvidia-gpu: 0 cpu: 2 memory: 7679792Ki pods: 110 Allocatable: alpha.kubernetes.io/nvidia-gpu: 0 cpu: 1800m memory: 7474992Ki pods: 110 [ ... lines removed for clarity ...] Non-terminated Pods: (5 in total) Namespace Name CPU Requests CPU Limits Memory Requests Memory Limits --------- ---- ------------ ---------- --------------- ------------- kube-system fluentd-gcp-v1.38-28bv1 100m (5%) 0 (0%) 200Mi (2%) 200Mi (2%) kube-system kube-dns-3297075139-61lj3 260m (13%) 0 (0%) 100Mi (1%) 170Mi (2%) kube-system kube-proxy-e2e-test-... 100m (5%) 0 (0%) 0 (0%) 0 (0%) kube-system monitoring-influxdb-grafana-v4-z1m12 200m (10%) 200m (10%) 600Mi (8%) 600Mi (8%) kube-system node-problem-detector-v0.1-fj7m3 20m (1%) 200m (10%) 20Mi (0%) 100Mi (1%) Allocated resources: (Total limits may be over 100 percent, i.e., overcommitted.) CPU Requests CPU Limits Memory Requests Memory Limits ------------ ---------- --------------- ------------- 680m (34%) 400m (20%) 920Mi (12%) 1070Mi (14%) 我的容器被终结了 您的容器可能因为资源枯竭而被终结了。要查看容器是否因为遇到资源限制而被杀死，请在相关的 Pod 上调用 kubectl describe pod： [12:54:41] $ kubectl describe pod simmemleak-hra99 Name: simmemleak-hra99 Namespace: default Image(s): saadali/simmemleak Node: kubernetes-node-tf0f/10.240.216.66 Labels: name=simmemleak Status: Running Reason: Message: IP: 10.244.2.75 Replication Controllers: simmemleak (1/1 replicas created) Containers: simmemleak: Image: saadali/simmemleak Limits: cpu: 100m memory: 50Mi State: Running Started: Tue, 07 Jul 2015 12:54:41 -0700 Last Termination State: Terminated Exit Code: 1 Started: Fri, 07 Jul 2015 12:54:30 -0700 Finished: Fri, 07 Jul 2015 12:54:33 -0700 Ready: False Restart Count: 5 Conditions: Type Status Ready False Events: FirstSeen LastSeen Count From SubobjectPath Reason Message Tue, 07 Jul 2015 12:53:51 -0700 Tue, 07 Jul 2015 12:53:51 -0700 1 {scheduler } scheduled Successfully assigned simmemleak-hra99 to kubernetes-node-tf0f Tue, 07 Jul 2015 12:53:51 -0700 Tue, 07 Jul 2015 12:53:51 -0700 1 {kubelet kubernetes-node-tf0f} implicitly required container POD pulled Pod container image \"gcr.io/google_containers/pause:0.8.0\" already present on machine Tue, 07 Jul 2015 12:53:51 -0700 Tue, 07 Jul 2015 12:53:51 -0700 1 {kubelet kubernetes-node-tf0f} implicitly required container POD created Created with docker id 6a41280f516d Tue, 07 Jul 2015 12:53:51 -0700 Tue, 07 Jul 2015 12:53:51 -0700 1 {kubelet kubernetes-node-tf0f} implicitly required container POD started Started with docker id 6a41280f516d Tue, 07 Jul 2015 12:53:51 -0700 Tue, 07 Jul 2015 12:53:51 -0700 1 {kubelet kubernetes-node-tf0f} spec.containers{simmemleak} created Created with docker id 87348f12526a 在上面的例子中，Restart Count: 5 意味着 Pod 中的 simmemleak 容器被终止并重启了五次。 您可以使用 kubectl get pod 命令加上 -o go-template=... 选项来获取之前终止容器的状态。 [13:59:01] $ kubectl get pod -o go-template='{{range.status.containerStatuses}}{{\"Container Name: \"}}{{.name}}{{\"\\r\\nLastState: \"}}{{.lastState}}{{end}}' simmemleak-60xbc Container Name: simmemleak LastState: map[terminated:map[exitCode:137 reason:OOM Killed startedAt:2015-07-07T20:58:43Z finishedAt:2015-07-07T20:58:43Z containerID:docker://0e4095bba1feccdfe7ef9fb6ebffe972b4b14285d5acdec6f0d3ae8a22fad8b2]] 您可以看到容器因为 reason:OOM killed 被终止，OOM 表示 Out Of Memory。 不透明整型资源（Alpha功能） Kubernetes 1.5 版本中引入不透明整型资源。不透明的整数资源允许集群运维人员发布新的节点级资源，否则系统将不了解这些资源。 用户可以在 Pod 的 spec 中消费这些资源，就像 CPU 和内存一样。调度器负责资源计量，以便在不超过可用量的同时分配给 Pod。 注意： 不透明整型资源在 kubernetes 1.5 中还是 Alpha 版本。只实现了资源计量，节点级别的隔离还处于积极的开发阶段。 不透明整型资源是以 pod.alpha.kubernetes.io/opaque-int-resource- 为前缀的资源。API server 将限制这些资源的数量为整数。有效 数量的例子有 3、3000m 和 3Ki。无效数量的例子有 0.5 和 1500m。 申请使用不透明整型资源需要两步。首先，集群运维人员必须在一个或多个节点上通告每个节点不透明的资源。然后，用户必须在 Pod 中请求不透明资源。 要发布新的不透明整型资源，集群运维人员应向 API server 提交 PATCH HTTP请求，以指定集群中节点的status.capacity 的可用数量。在此操作之后，节点的 status.capacity 将包括一个新的资源。 status.allocatable 字段由 kubelet 异步地使用新资源自动更新。请注意，由于调度器在评估 Pod 适应度时使用节点 status.allocatable 值，所以在使用新资源修补节点容量和请求在该节点上调度资源的第一个 pod 之间可能会有短暂的延迟。 示例 这是一个 HTTP 请求，master 节点是 k8s-master，在 k8s-node-1 节点上通告 5 个 “foo” 资源。 PATCH /api/v1/nodes/k8s-node-1/status HTTP/1.1 Accept: application/json Content-Type: application/json-patch+json Host: k8s-master:8080 [ { \"op\": \"add\", \"path\": \"/status/capacity/pod.alpha.kubernetes.io~1opaque-int-resource-foo\", \"value\": \"5\" } ] curl --header \"Content-Type: application/json-patch+json\" \\ --request PATCH \\ --data '[{\"op\": \"add\", \"path\": \"/status/capacity/pod.alpha.kubernetes.io~1opaque-int-resource-foo\", \"value\": \"5\"}]' \\ http://k8s-master:8080/api/v1/nodes/k8s-node-1/status 注意： 在前面的请求中，~1 是 patch 路径中 / 字符的编码。JSON-Patch 中的操作路径值被解释为 JSON-Pointer。更多详细信息请参阅 IETF RFC 6901, section 3。 apiVersion: v1 kind: Pod metadata: name: my-pod spec: containers: - name: my-container image: myimage resources: requests: cpu: 2 pod.alpha.kubernetes.io/opaque-int-resource-foo: 1 计划改进 在 kubernetes 1.5 版本中仅允许在容器上指定资源量。计划改进对所有容器在 Pod 中共享资源的计量，如 emptyDir volume。 在 kubernetes 1.5 版本中仅支持容器对 CPU 和内存的申请和限制。计划增加新的资源类型，包括节点磁盘空间资源和一个可支持自定义 资源类型 的框架。 Kubernetes 通过支持通过多级别的 服务质量 来支持资源的过度使用。 在 kubernetes 1.5 版本中，一个 CPU 单位在不同的云提供商和同一云提供商的不同机器类型中的意味都不同。例如，在 AWS 上，节点的容量报告为 ECU，而在 GCE 中报告为逻辑内核。我们计划修改 cpu 资源的定义，以便在不同的提供商和平台之间保持一致。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-09-28 17:36:25 "},"practice/storage.html":{"url":"practice/storage.html","title":"存储管理","keywords":"","body":"存储管理 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-08-21 18:23:35 "},"practice/glusterfs.html":{"url":"practice/glusterfs.html","title":"GlusterFS","keywords":"","body":"GlusterFS GlusterFS是Scale-Out存储解决方案Gluster的核心，它是一个开源的分布式文件系统，具有强大的横向扩展能力，通过扩展能够支持数PB存储容量和处理数千客户端。GlusterFS借助TCP/IP或InfiniBand RDMA网络将物理分布的存储资源聚集在一起，使用单一全局命名空间来管理数据。GlusterFS基于可堆叠的用户空间设计，可为各种不同的数据负载提供优异的性能。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-08-21 18:23:34 "},"practice/using-glusterfs-for-persistent-storage.html":{"url":"practice/using-glusterfs-for-persistent-storage.html","title":"使用GlusterFS做持久化存储","keywords":"","body":"使用glusterfs做持久化存储 我们复用kubernetes的三台主机做glusterfs存储。 以下步骤参考自：https://www.xf80.com/2017/04/21/kubernetes-glusterfs/（该网站已无法访问） 安装glusterfs 我们直接在物理机上使用yum安装，如果你选择在kubernetes上安装，请参考：https://github.com/gluster/gluster-kubernetes/blob/master/docs/setup-guide.md # 先安装 gluster 源 $ yum install centos-release-gluster -y # 安装 glusterfs 组件 $ yum install -y glusterfs glusterfs-server glusterfs-fuse glusterfs-rdma glusterfs-geo-replication glusterfs-devel ## 创建 glusterfs 目录 $ mkdir /opt/glusterd ## 修改 glusterd 目录 $ sed -i 's/var\\/lib/opt/g' /etc/glusterfs/glusterd.vol # 启动 glusterfs $ systemctl start glusterd.service # 设置开机启动 $ systemctl enable glusterd.service #查看状态 $ systemctl status glusterd.service 配置 glusterfs # 配置 hosts $ vi /etc/hosts 172.20.0.113 sz-pg-oam-docker-test-001.tendcloud.com 172.20.0.114 sz-pg-oam-docker-test-002.tendcloud.com 172.20.0.115 sz-pg-oam-docker-test-003.tendcloud.com # 开放端口 $ iptables -I INPUT -p tcp --dport 24007 -j ACCEPT # 创建存储目录 $ mkdir /opt/gfs_data # 添加节点到 集群 # 执行操作的本机不需要probe 本机 [root@sz-pg-oam-docker-test-001 ~]# gluster peer probe sz-pg-oam-docker-test-002.tendcloud.com gluster peer probe sz-pg-oam-docker-test-003.tendcloud.com # 查看集群状态 $ gluster peer status Number of Peers: 2 Hostname: sz-pg-oam-docker-test-002.tendcloud.com Uuid: f25546cc-2011-457d-ba24-342554b51317 State: Peer in Cluster (Connected) Hostname: sz-pg-oam-docker-test-003.tendcloud.com Uuid: 42b6cad1-aa01-46d0-bbba-f7ec6821d66d State: Peer in Cluster (Connected) 配置 volume GlusterFS中的volume的模式有很多中，包括以下几种： 分布卷（默认模式）：即DHT, 也叫 分布卷: 将文件已hash算法随机分布到 一台服务器节点中存储。 复制模式：即AFR, 创建volume 时带 replica x 数量: 将文件复制到 replica x 个节点中。 条带模式：即Striped, 创建volume 时带 stripe x 数量： 将文件切割成数据块，分别存储到 stripe x 个节点中 ( 类似raid 0 )。 分布式条带模式：最少需要4台服务器才能创建。 创建volume 时 stripe 2 server = 4 个节点： 是DHT 与 Striped 的组合型。 分布式复制模式：最少需要4台服务器才能创建。 创建volume 时 replica 2 server = 4 个节点：是DHT 与 AFR 的组合型。 条带复制卷模式：最少需要4台服务器才能创建。 创建volume 时 stripe 2 replica 2 server = 4 个节点： 是 Striped 与 AFR 的组合型。 三种模式混合： 至少需要8台 服务器才能创建。 stripe 2 replica 2 , 每4个节点 组成一个 组。 这几种模式的示例图参考：CentOS7安装GlusterFS。 因为我们只有三台主机，在此我们使用默认的分布卷模式。请勿在生产环境上使用该模式，容易导致数据丢失。 # 创建分布卷 $ gluster volume create k8s-volume transport tcp sz-pg-oam-docker-test-001.tendcloud.com:/opt/gfs_data sz-pg-oam-docker-test-002.tendcloud.com:/opt/gfs_data sz-pg-oam-docker-test-003.tendcloud.com:/opt/gfs_data force # 查看volume状态 $ gluster volume info Volume Name: k8s-volume Type: Distribute Volume ID: 9a3b0710-4565-4eb7-abae-1d5c8ed625ac Status: Created Snapshot Count: 0 Number of Bricks: 3 Transport-type: tcp Bricks: Brick1: sz-pg-oam-docker-test-001.tendcloud.com:/opt/gfs_data Brick2: sz-pg-oam-docker-test-002.tendcloud.com:/opt/gfs_data Brick3: sz-pg-oam-docker-test-003.tendcloud.com:/opt/gfs_data Options Reconfigured: transport.address-family: inet nfs.disable: on # 启动 分布卷 $ gluster volume start k8s-volume Glusterfs调优 # 开启 指定 volume 的配额 $ gluster volume quota k8s-volume enable # 限制 指定 volume 的配额 $ gluster volume quota k8s-volume limit-usage / 1TB # 设置 cache 大小, 默认32MB $ gluster volume set k8s-volume performance.cache-size 4GB # 设置 io 线程, 太大会导致进程崩溃 $ gluster volume set k8s-volume performance.io-thread-count 16 # 设置 网络检测时间, 默认42s $ gluster volume set k8s-volume network.ping-timeout 10 # 设置 写缓冲区的大小, 默认1M $ gluster volume set k8s-volume performance.write-behind-window-size 1024MB Kubernetes中配置glusterfs 官方的文档见：https://github.com/kubernetes/kubernetes/tree/master/examples/volumes/glusterfs 以下用到的所有yaml和json配置文件可以在../manifests/glusterfs中找到。注意替换其中私有镜像地址为你自己的镜像地址。 kubernetes安装客户端 # 在所有 k8s node 中安装 glusterfs 客户端 $ yum install -y glusterfs glusterfs-fuse # 配置 hosts $ vi /etc/hosts 172.20.0.113 sz-pg-oam-docker-test-001.tendcloud.com 172.20.0.114 sz-pg-oam-docker-test-002.tendcloud.com 172.20.0.115 sz-pg-oam-docker-test-003.tendcloud.com 因为我们glusterfs跟kubernetes集群复用主机，因为此这一步可以省去。 配置 endpoints $ curl -O https://raw.githubusercontent.com/kubernetes/kubernetes/master/examples/volumes/glusterfs/glusterfs-endpoints.json # 修改 endpoints.json ，配置 glusters 集群节点ip # 每一个 addresses 为一个 ip 组 { \"addresses\": [ { \"ip\": \"172.22.0.113\" } ], \"ports\": [ { \"port\": 1990 } ] }, # 导入 glusterfs-endpoints.json $ kubectl apply -f glusterfs-endpoints.json # 查看 endpoints 信息 $ kubectl get ep 配置 service $ curl -O https://raw.githubusercontent.com/kubernetes/kubernetes/master/examples/volumes/glusterfs/glusterfs-service.json # service.json 里面查找的是 enpointes 的名称与端口，端口默认配置为 1，我改成了1990 # 导入 glusterfs-service.json $ kubectl apply -f glusterfs-service.json # 查看 service 信息 $ kubectl get svc 创建测试 pod $ curl -O https://raw.githubusercontent.com/kubernetes/kubernetes/master/examples/volumes/glusterfs/glusterfs-pod.json # 编辑 glusterfs-pod.json # 修改 volumes 下的 path 为上面创建的 volume 名称 \"path\": \"k8s-volume\" # 导入 glusterfs-pod.json $ kubectl apply -f glusterfs-pod.json # 查看 pods 状态 $ kubectl get pods NAME READY STATUS RESTARTS AGE glusterfs 1/1 Running 0 1m # 查看 pods 所在 node $ kubectl describe pods/glusterfs # 登陆 node 物理机，使用 df 可查看挂载目录 $ df -h 172.20.0.113:k8s-volume 1073741824 0 1073741824 0% 172.20.0.113:k8s-volume 1.0T 0 1.0T 0% /var/lib/kubelet/pods/3de9fc69-30b7-11e7-bfbd-8af1e3a7c5bd/volumes/kubernetes.io~glusterfs/glusterfsvol 配置PersistentVolume PersistentVolume（PV）和 PersistentVolumeClaim（PVC）是kubernetes提供的两种API资源，用于抽象存储细节。管理员关注于如何通过pv提供存储功能而无需关注用户如何使用，同样的用户只需要挂载PVC到容器中而不需要关注存储卷采用何种技术实现。 PVC和PV的关系跟pod和node关系类似，前者消耗后者的资源。PVC可以向PV申请指定大小的存储资源并设置访问模式。 PV属性 storage容量 读写属性：分别为ReadWriteOnce：单个节点读写； ReadOnlyMany：多节点只读 ； ReadWriteMany：多节点读写 $ cat glusterfs-pv.yaml apiVersion: v1 kind: PersistentVolume metadata: name: gluster-dev-volume spec: capacity: storage: 8Gi accessModes: - ReadWriteMany glusterfs: endpoints: \"glusterfs-cluster\" path: \"k8s-volume\" readOnly: false # 导入PV $ kubectl apply -f glusterfs-pv.yaml # 查看 pv $ kubectl get pv NAME CAPACITY ACCESSMODES RECLAIMPOLICY STATUS CLAIM STORAGECLASS REASON AGE gluster-dev-volume 8Gi RWX Retain Available 3s PVC属性 访问属性与PV相同 容量：向PV申请的容量 配置PVC $ cat glusterfs-pvc.yaml kind: PersistentVolumeClaim apiVersion: v1 metadata: name: glusterfs-nginx spec: accessModes: - ReadWriteMany resources: requests: storage: 8Gi # 导入 pvc $ kubectl apply -f glusterfs-pvc.yaml # 查看 pvc $ kubectl get pv NAME STATUS VOLUME CAPACITY ACCESSMODES STORAGECLASS AGE glusterfs-nginx Bound gluster-dev-volume 8Gi RWX 4s 创建 nginx deployment 挂载 volume $ vi nginx-deployment.yaml apiVersion: extensions/v1beta1 kind: Deployment metadata: name: nginx-dm spec: replicas: 2 template: metadata: labels: name: nginx spec: containers: - name: nginx image: nginx:alpine imagePullPolicy: IfNotPresent ports: - containerPort: 80 volumeMounts: - name: gluster-dev-volume mountPath: \"/usr/share/nginx/html\" volumes: - name: gluster-dev-volume persistentVolumeClaim: claimName: glusterfs-nginx # 导入 deployment $ kubectl apply -f nginx-deployment.yaml # 查看 deployment $ kubectl get pods |grep nginx-dm nginx-dm-3698525684-g0mvt 1/1 Running 0 6s nginx-dm-3698525684-hbzq1 1/1 Running 0 6s # 查看 挂载 $ kubectl exec -it nginx-dm-3698525684-g0mvt -- df -h|grep k8s-volume 172.20.0.113:k8s-volume 1.0T 0 1.0T 0% /usr/share/nginx/html # 创建文件 测试 $ kubectl exec -it nginx-dm-3698525684-g0mvt -- touch /usr/share/nginx/html/index.html $ kubectl exec -it nginx-dm-3698525684-g0mvt -- ls -lt /usr/share/nginx/html/index.html -rw-r--r-- 1 root root 0 May 4 11:36 /usr/share/nginx/html/index.html # 验证 glusterfs # 因为我们使用分布卷，所以可以看到某个节点中有文件 [root@sz-pg-oam-docker-test-001 ~] ls /opt/gfs_data/ [root@sz-pg-oam-docker-test-002 ~] ls /opt/gfs_data/ index.html [root@sz-pg-oam-docker-test-003 ~] ls /opt/gfs_data/ 参考 CentOS 7 安装 GlusterFS GlusterFS with kubernetes Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-10-19 18:27:14 "},"practice/storage-for-containers-using-glusterfs-with-openshift.html":{"url":"practice/storage-for-containers-using-glusterfs-with-openshift.html","title":"在OpenShift中使用GlusterFS做持久化存储","keywords":"","body":"在OpenShift中使用GlusterFS做持久化存储 概述 本文由Daniel Messer（Technical Marketing Manager Storage @RedHat）和Keith Tenzer（Solutions Architect @RedHat）共同撰写。 Storage for Containers Overview – Part I Storage for Containers using Gluster – Part II Storage for Containers using Container Native Storage – Part III Storage for Containers using Ceph – Part IV Storage for Containers using NetApp ONTAP NAS – Part V Storage for Containers using NetApp SolidFire – Part VI Gluster作为Container-Ready Storage(CRS) 在本文中，我们将介绍容器存储的首选以及如何部署它。 Kusternet和OpenShift支持GlusterFS已经有一段时间了。 GlusterFS的适用性很好，可用于所有的部署场景：裸机、虚拟机、内部部署和公共云。 在容器中运行GlusterFS的新特性将在本系列后面讨论。 GlusterFS是一个分布式文件系统，内置了原生协议（GlusterFS）和各种其他协议（NFS，SMB，...）。 为了与OpenShift集成，节点将通过FUSE使用原生协议，将GlusterFS卷挂在到节点本身上，然后将它们绑定到目标容器中。 OpenShift / Kubernetes具有实现请求、释放和挂载、卸载GlusterFS卷的原生程序。 CRS概述 在存储方面，根据OpenShift / Kubernetes的要求，还有一个额外的组件管理集群，称为“heketi”。 这实际上是一个用于GlusterFS的REST API，它还提供CLI版本。 在以下步骤中，我们将在3个GlusterFS节点中部署heketi，使用它来部署GlusterFS存储池，将其连接到OpenShift，并使用它来通过PersistentVolumeClaims为容器配置存储。 我们将总共部署4台虚拟机。 一个用于OpenShift（实验室设置），另一个用于GlusterFS。 注意：您的系统应至少需要有四核CPU，16GB RAM和20 GB可用磁盘空间。 部署OpenShift 首先你需要先部署OpenShift。最有效率的方式是直接在虚拟机中部署一个All-in-One环境，部署指南见 the “OpenShift Enterprise 3.4 all-in-one Lab Environment” article.。 确保你的OpenShift虚拟机可以解析外部域名。编辑/etc/dnsmasq.conf文件，增加下面的Google DNS： server=8.8.8.8 重启： # systemctl restart dnsmasq # ping -c1 google.com 部署Gluster GlusterFS至少需要有以下配置的3台虚拟机： RHEL 7.3 2 CPUs 2 GB内存 30 GB磁盘存储给操作系统 10 GB磁盘存储给GlusterFS bricks 修改/etc/hosts文件，定义三台虚拟机的主机名。 例如（主机名可以根据你自己的环境自由调整） # cat /etc/hosts 127.0.0.1 localhost localhost.localdomain localhost4 localhost4.localdomain4 ::1 localhost localhost.localdomain localhost6 localhost6.localdomain6 172.16.99.144 ocp-master.lab ocp-master 172.16.128.7 crs-node1.lab crs-node1 172.16.128.8 crs-node2.lab crs-node2 172.16.128.9 crs-node3.lab crs-node3 在3台GlusterFS虚拟机上都执行以下步骤： # subscription-manager repos --disable=\"*\" # subscription-manager repos --enable=rhel-7-server-rpms 如果你已经订阅了GlusterFS那么可以直接使用，开启rh-gluster-3-for-rhel-7-server-rpms的yum源。 如果你没有的话，那么可以通过EPEL使用非官方支持的GlusterFS的社区源。 # yum -y install http://dl.fedoraproject.org/pub/epel/epel-release-latest-7.noarch.rpm # rpm --import http://dl.fedoraproject.org/pub/epel/RPM-GPG-KEY-EPEL-7 在/etc/yum.repos.d/目录下创建glusterfs-3.10.repo文件： [glusterfs-3.10] name=glusterfs-3.10 description=\"GlusterFS 3.10 Community Version\" baseurl=https://buildlogs.centos.org/centos/7/storage/x86_64/gluster-3.10/ gpgcheck=0 enabled=1 验证源已经被激活。 # yum repolist 现在可以开始安装GlusterFS了。 # yum -y install glusterfs-server 需要为GlusterFS peers打开几个基本TCP端口，以便与OpenShift进行通信并提供存储： # firewall-cmd --add-port=24007-24008/tcp --add-port=49152-49664/tcp --add-port=2222/tcp # firewall-cmd --runtime-to-permanent 现在我们可以启动GlusterFS的daemon进程了： # systemctl enable glusterd # systemctl start glusterd 完成。GlusterFS已经启动并正在运行。其他配置将通过heketi完成。 在GlusterFS的一台虚拟机上安装heketi [root@crs-node1 ~]# yum -y install heketi heketi-client 更新EPEL 如果你没有Red Hat Gluster Storage订阅的话，你可以从EPEL中获取heketi。 在撰写本文时，2016年10月那时候还是3.0.0-1.el7版本，它不适用于OpenShift 3.4。 你将需要更新到更新的版本： [root@crs-node1 ~]# yum -y install wget [root@crs-node1 ~]# wget https://github.com/heketi/heketi/releases/download/v4.0.0/heketi-v4.0.0.linux.amd64.tar.gz [root@crs-node1 ~]# tar -xzf heketi-v4.0.0.linux.amd64.tar.gz [root@crs-node1 ~]# systemctl stop heketi [root@crs-node1 ~]# cp heketi/heketi* /usr/bin/ [root@crs-node1 ~]# chown heketi:heketi /usr/bin/heketi* 在/etc/systemd/system/heketi.service中创建v4版本的heketi二进制文件的更新语法文件： [Unit] Description=Heketi Server [Service] Type=simple WorkingDirectory=/var/lib/heketi EnvironmentFile=-/etc/heketi/heketi.json User=heketi ExecStart=/usr/bin/heketi --config=/etc/heketi/heketi.json Restart=on-failure StandardOutput=syslog StandardError=syslog [Install] WantedBy=multi-user.target [root@crs-node1 ~]# systemctl daemon-reload [root@crs-node1 ~]# systemctl start heketi Heketi使用SSH来配置GlusterFS的所有节点。创建SSH密钥对，将公钥拷贝到所有3个节点上（包括你登陆的第一个节点）： [root@crs-node1 ~]# ssh-keygen -f /etc/heketi/heketi_key -t rsa -N '' [root@crs-node1 ~]# ssh-copy-id -i /etc/heketi/heketi_key.pub root@crs-node1.lab [root@crs-node1 ~]# ssh-copy-id -i /etc/heketi/heketi_key.pub root@crs-node2.lab [root@crs-node1 ~]# ssh-copy-id -i /etc/heketi/heketi_key.pub root@crs-node3.lab [root@crs-node1 ~]# chown heketi:heketi /etc/heketi/heketi_key* 剩下唯一要做的事情就是配置heketi来使用SSH。 编辑/etc/heketi/heketi.json文件使它看起来像下面这个样子（改变的部分突出显示下划线）： { \"_port_comment\":\"Heketi Server Port Number\", \"port\":\"8080\", \"_use_auth\":\"Enable JWT authorization. Please enable for deployment\", \"use_auth\":false, \"_jwt\":\"Private keys for access\", \"jwt\":{ \"_admin\":\"Admin has access to all APIs\", \"admin\":{ \"key\":\"My Secret\" }, \"_user\":\"User only has access to /volumes endpoint\", \"user\":{ \"key\":\"My Secret\" } }, \"_glusterfs_comment\":\"GlusterFS Configuration\", \"glusterfs\":{ \"_executor_comment\":[ \"Execute plugin. Possible choices: mock, ssh\", \"mock: This setting is used for testing and development.\", \" It will not send commands to any node.\", \"ssh: This setting will notify Heketi to ssh to the nodes.\", \" It will need the values in sshexec to be configured.\", \"kubernetes: Communicate with GlusterFS containers over\", \" Kubernetes exec api.\" ], \"executor\":\"ssh\", \"_sshexec_comment\":\"SSH username and private key file information\", \"sshexec\":{ \"keyfile\":\"/etc/heketi/heketi_key\", \"user\":\"root\", \"port\":\"22\", \"fstab\":\"/etc/fstab\" }, \"_kubeexec_comment\":\"Kubernetes configuration\", \"kubeexec\":{ \"host\":\"https://kubernetes.host:8443\", \"cert\":\"/path/to/crt.file\", \"insecure\":false, \"user\":\"kubernetes username\", \"password\":\"password for kubernetes user\", \"namespace\":\"OpenShift project or Kubernetes namespace\", \"fstab\":\"Optional: Specify fstab file on node. Default is /etc/fstab\" }, \"_db_comment\":\"Database file name\", \"db\":\"/var/lib/heketi/heketi.db\", \"_loglevel_comment\":[ \"Set log level. Choices are:\", \" none, critical, error, warning, info, debug\", \"Default is warning\" ], \"loglevel\":\"debug\" } } 完成。heketi将监听8080端口，我们来确认下防火墙规则允许它监听该端口： # firewall-cmd --add-port=8080/tcp # firewall-cmd --runtime-to-permanent 重启heketi： # systemctl enable heketi # systemctl restart heketi 测试它是否在运行： # curl http://crs-node1.lab:8080/hello Hello from Heketi 很好。heketi上场的时候到了。 我们将使用它来配置我们的GlusterFS存储池。 该软件已经在我们所有的虚拟机上运行，但并未被配置。 要将其改造为满足我们需求的存储系统，需要在拓扑文件中描述我们所需的GlusterFS存储池，如下所示： # vi topology.json { \"clusters\": [ { \"nodes\": [ { \"node\": { \"hostnames\": { \"manage\": [ \"crs-node1.lab\" ], \"storage\": [ \"172.16.128.7\" ] }, \"zone\": 1 }, \"devices\": [ \"/dev/sdb\" ] }, { \"node\": { \"hostnames\": { \"manage\": [ \"crs-node2.lab\" ], \"storage\": [ \"172.16.128.8\" ] }, \"zone\": 1 }, \"devices\": [ \"/dev/sdb\" ] }, { \"node\": { \"hostnames\": { \"manage\": [ \"crs-node3.lab\" ], \"storage\": [ \"172.16.128.9\" ] }, \"zone\": 1 }, \"devices\": [ \"/dev/sdb\" ] } ] } ] } 该文件格式比较简单，基本上是告诉heketi要创建一个3节点的集群，其中每个节点包含的配置有FQDN，IP地址以及至少一个将用作GlusterFS块的备用块设备。 现在将该文件发送给heketi： # export HEKETI_CLI_SERVER=http://crs-node1.lab:8080 # heketi-cli topology load --json=topology.json Creating cluster ... ID: 78cdb57aa362f5284bc95b2549bc7e7d Creating node crs-node1.lab ... ID: ffd7671c0083d88aeda9fd1cb40b339b Adding device /dev/sdb ... OK Creating node crs-node2.lab ... ID: 8220975c0a4479792e684584153050a9 Adding device /dev/sdb ... OK Creating node crs-node3.lab ... ID: b94f14c4dbd8850f6ac589ac3b39cc8e Adding device /dev/sdb ... OK 现在heketi已经配置了3个节点的GlusterFS存储池。很简单！你现在可以看到3个虚拟机都已经成功构成了GlusterFS中的可信存储池（Trusted Stroage Pool）。 [root@crs-node1 ~]# gluster peer status Number of Peers: 2 Hostname: crs-node2.lab Uuid: 93b34946-9571-46a8-983c-c9f128557c0e State: Peer in Cluster (Connected) Other names: crs-node2.lab Hostname: 172.16.128.9 Uuid: e3c1f9b0-be97-42e5-beda-f70fc05f47ea State: Peer in Cluster (Connected) 现在回到OpenShift！ 将Gluster与OpenShift集成 为了集成OpenShift，需要两样东西：一个动态的Kubernetes Storage Provisioner和一个StorageClass。 Provisioner在OpenShift中开箱即用。 实际上关键的是如何将存储挂载到容器上。 StorageClass是OpenShift中的用户可以用来实现的PersistentVolumeClaims的实体，它反过来能够触发一个Provisioner实现实际的配置，并将结果表示为Kubernetes PersistentVolume（PV）。 就像OpenShift中的其他组件一样，StorageClass也简单的用YAML文件定义： # cat crs-storageclass.yaml kind: StorageClass apiVersion: storage.k8s.io/v1beta1 metadata: name: container-ready-storage annotations: storageclass.beta.kubernetes.io/is-default-class: \"true\" provisioner: kubernetes.io/glusterfs parameters: resturl: \"http://crs-node1.lab:8080\" restauthenabled: \"false\" 我们的provisioner是kubernetes.io/glusterfs，将它指向我们的heketi实例。 我们将类命名为“container-ready-storage”，同时使其成为所有没有显示指定StorageClass的PersistentVolumeClaim的默认StorageClass。 为你的GlusterFS池创建StorageClass： # oc create -f crs-storageclass.yaml 在OpenShift中使用Gluster 我们来看下如何在OpenShift中使用GlusterFS。首先在OpenShift虚拟机中创建一个测试项目。 # oc new-project crs-storage --display-name=\"Container-Ready Storage\" 这会向Kubernetes/OpenShift发出storage请求，请求一个PersistentVolumeClaim（PVC）。 这是一个简单的对象，它描述最少需要多少容量和应该提供哪种访问模式（非共享，共享，只读）。 它通常是应用程序模板的一部分，但我们只需创建一个独立的PVC： # cat crs-claim.yaml apiVersion: v1 kind: PersistentVolumeClaim metadata: name: my-crs-storage namespace: crs-storage spec: accessModes: - ReadWriteOnce resources: requests: storage: 1Gi 发送该请求： # oc create -f crs-claim.yaml 观察在OpenShfit中，PVC正在以动态创建volume的方式实现： # oc get pvc NAME STATUS VOLUME CAPACITY ACCESSMODES AGE my-crs-storage Bound pvc-41ad5adb-107c-11e7-afae-000c2949cce7 1Gi RWO 58s 太棒了！ 你现在可以在OpenShift中使用存储容量，而不需要直接与存储系统进行任何交互。 我们来看看创建的volume： # oc get pv/pvc-41ad5adb-107c-11e7-afae-000c2949cce7 Name: pvc-41ad5adb-107c-11e7-afae-000c2949cce7 Labels: StorageClass: container-ready-storage Status: Bound Claim: crs-storage/my-crs-storage Reclaim Policy: Delete Access Modes: RWO Capacity: 1Gi Message: Source: Type: Glusterfs (a Glusterfs mount on the host that shares a pod's lifetime) EndpointsName: gluster-dynamic-my-crs-storage Path: vol_85e444ee3bc154de084976a9aef16025 ReadOnly: false What happened in the background was that when the PVC reached the system, our default StorageClass reached out to the GlusterFS Provisioner with the volume specs from the PVC. The provisioner in turn communicates with our heketi instance which facilitates the creation of the GlusterFS volume, which we can trace in it’s log messages: 该volume是根据PVC中的定义特别创建的。 在PVC中，我们没有明确指定要使用哪个StorageClass，因为heketi的GlusterFS StorageClass已经被定义为系统范围的默认值。 在后台发生的情况是，当PVC到达系统时，默认的StorageClass请求具有该PVC中volume声明规格的GlusterFS Provisioner。 Provisioner又与我们的heketi实例通信，这有助于创建GlusterFS volume，我们可以在其日志消息中追踪： [root@crs-node1 ~]# journalctl -l -u heketi.service ... Mar 24 11:25:52 crs-node1.lab heketi[2598]: [heketi] DEBUG 2017/03/24 11:25:52 /src/github.com/heketi/heketi/apps/glusterfs/volume_entry.go:298: Volume to be created on cluster e Mar 24 11:25:52 crs-node1.lab heketi[2598]: [heketi] INFO 2017/03/24 11:25:52 Creating brick 9e791b1daa12af783c9195941fe63103 Mar 24 11:25:52 crs-node1.lab heketi[2598]: [heketi] INFO 2017/03/24 11:25:52 Creating brick 3e06af2f855bef521a95ada91680d14b Mar 24 11:25:52 crs-node1.lab heketi[2598]: [heketi] INFO 2017/03/24 11:25:52 Creating brick e4daa240f1359071e3f7ea22618cfbab ... Mar 24 11:25:52 crs-node1.lab heketi[2598]: [sshexec] INFO 2017/03/24 11:25:52 Creating volume vol_85e444ee3bc154de084976a9aef16025 replica 3 ... Mar 24 11:25:53 crs-node1.lab heketi[2598]: Result: volume create: vol_85e444ee3bc154de084976a9aef16025: success: please start the volume to access data ... Mar 24 11:25:55 crs-node1.lab heketi[2598]: Result: volume start: vol_85e444ee3bc154de084976a9aef16025: success ... Mar 24 11:25:55 crs-node1.lab heketi[2598]: [asynchttp] INFO 2017/03/24 11:25:55 Completed job c3d6c4f9fc74796f4a5262647dc790fe in 3.176522702s ... 成功！ 大约用了3秒钟，GlusterFS池就配置完成了，并配置了一个volume。 默认值是replica 3，这意味着数据将被复制到3个不同节点的3个块上（用GlusterFS作为后端存储）。 该过程是通过Heketi在OpenShift进行编排的。 你也可以从GlusterFS的角度看到有关volume的信息： [root@crs-node1 ~]# gluster volume list vol_85e444ee3bc154de084976a9aef16025 [root@crs-node1 ~]# gluster volume info vol_85e444ee3bc154de084976a9aef16025 Volume Name: vol_85e444ee3bc154de084976a9aef16025 Type: Replicate Volume ID: a32168c8-858e-472a-b145-08c20192082b Status: Started Snapshot Count: 0 Number of Bricks: 1 x 3 = 3 Transport-type: tcp Bricks: Brick1: 172.16.128.8:/var/lib/heketi/mounts/vg_147b43f6f6903be8b23209903b7172ae/brick_9e791b1daa12af783c9195941fe63103/brick Brick2: 172.16.128.9:/var/lib/heketi/mounts/vg_72c0f520b0c57d807be21e9c90312f85/brick_3e06af2f855bef521a95ada91680d14b/brick Brick3: 172.16.128.7:/var/lib/heketi/mounts/vg_67314f879686de975f9b8936ae43c5c5/brick_e4daa240f1359071e3f7ea22618cfbab/brick Options Reconfigured: transport.address-family: inet nfs.disable: on 请注意，GlusterFS中的卷名称如何对应于OpenShift中Kubernetes Persistent Volume的“路径”。 或者，你也可以使用OpenShift UI来配置存储，这样可以很方便地在系统中的所有已知的StorageClasses中进行选择： 图片 - 创建存储 图片 - Screen Shot 2017-03-24 at 11.09.34.png 让我们做点更有趣的事情，在OpenShift中运行工作负载。 在仍运行着crs-storage项目的OpenShift虚拟机中执行： # oc get templates -n openshift 你应该可以看到一个应用程序和数据库模板列表，这个列表将方便你更轻松的使用OpenShift来部署你的应用程序项目。 我们将使用MySQL来演示如何在OpenShift上部署具有持久化和弹性存储的有状态应用程序。 Mysql-persistent模板包含一个用于MySQL数据库目录的1G空间的PVC。 为了演示目的，可以直接使用默认值。 # oc process mysql-persistent -n openshift | oc create -f - 等待部署完成。你可以通过UI或者命令行观察部署进度： # oc get pods NAME READY STATUS RESTARTS AGE mysql-1-h4afb 1/1 Running 0 2m 好了。我们已经使用这个模板创建了一个service，secrets、PVC和pod。我们来使用它（你的pod名字将跟我的不同）： # oc rsh mysql-1-h4afb 你已经成功的将它挂载到MySQL的pod上。我们连接一下数据库试试： sh-4.2$ mysql -u $MYSQL_USER -p$MYSQL_PASSWORD -h $HOSTNAME $MYSQL_DATABASE 这点很方便，所有重要的配置，如MySQL凭据，数据库名称等都是pod模板中的环境变量的一部分，因此可以在pod中作为shell的环境变量。 我们来创建一些数据： mysql> show databases; +--------------------+ | Database | +--------------------+ | information_schema | | sampledb | +--------------------+ 2 rows in set (0.02 sec) mysql> \\u sampledb Database changed mysql> CREATE TABLE IF NOT EXISTS equipment ( -> equip_id int(5) NOT NULL AUTO_INCREMENT, -> type varchar(50) DEFAULT NULL, -> install_date DATE DEFAULT NULL, -> color varchar(20) DEFAULT NULL, -> working bool DEFAULT NULL, -> location varchar(250) DEFAULT NULL, -> PRIMARY KEY(equip_id) -> ); Query OK, 0 rows affected (0.13 sec) mysql> INSERT INTO equipment (type, install_date, color, working, location) -> VALUES -> (\"Slide\", Now(), \"blue\", 1, \"Southwest Corner\"); Query OK, 1 row affected, 1 warning (0.01 sec) mysql> SELECT * FROM equipment; +----------+-------+--------------+-------+---------+------------------+ | equip_id | type | install_date | color | working | location | +----------+-------+--------------+-------+---------+------------------+ | 1 | Slide | 2017-03-24 | blue | 1 | Southwest Corner | +----------+-------+--------------+-------+---------+------------------+ 1 row in set (0.00 sec) 很好，数据库运行正常。 你想看下数据存储在哪里吗？很简单！查看刚使用模板创建的mysql volume： # oc get pvc/mysql NAME STATUS VOLUME CAPACITY ACCESSMODES AGE mysql Bound pvc-a678b583-1082-11e7-afae-000c2949cce7 1Gi RWO 11m # oc describe pv/pvc-a678b583-1082-11e7-afae-000c2949cce7 Name: pvc-a678b583-1082-11e7-afae-000c2949cce7 Labels: StorageClass: container-ready-storage Status: Bound Claim: crs-storage/mysql Reclaim Policy: Delete Access Modes: RWO Capacity: 1Gi Message: Source: Type: Glusterfs (a Glusterfs mount on the host that shares a pod's lifetime) EndpointsName: gluster-dynamic-mysql Path: vol_6299fc74eee513119dafd43f8a438db1 ReadOnly: false GlusterFS的volume名字是vol_6299fc74eee513119dafd43f8a438db1。回到你的GlusterFS虚拟机中，输入： # gluster volume info vol_6299fc74eee513119dafd43f8a438db Volume Name: vol_6299fc74eee513119dafd43f8a438db1 Type: Replicate Volume ID: 4115918f-28f7-4d4a-b3f5-4b9afe5b391f Status: Started Snapshot Count: 0 Number of Bricks: 1 x 3 = 3 Transport-type: tcp Bricks: Brick1: 172.16.128.7:/var/lib/heketi/mounts/vg_67314f879686de975f9b8936ae43c5c5/brick_f264a47aa32be5d595f83477572becf8/brick Brick2: 172.16.128.8:/var/lib/heketi/mounts/vg_147b43f6f6903be8b23209903b7172ae/brick_f5731fe7175cbe6e6567e013c2591343/brick Brick3: 172.16.128.9:/var/lib/heketi/mounts/vg_72c0f520b0c57d807be21e9c90312f85/brick_ac6add804a6a467cd81cd1404841bbf1/brick Options Reconfigured: transport.address-family: inet nfs.disable: on 你可以看到数据是如何被复制到3个GlusterFS块的。我们从中挑一个（最好挑选你刚登陆的那台虚拟机并查看目录）： # ll /var/lib/heketi/mounts/vg_67314f879686de975f9b8936ae43c5c5/brick_f264a47aa32be5d595f83477572becf8/brick total 180300 -rw-r-----. 2 1000070000 2001 56 Mar 24 12:11 auto.cnf -rw-------. 2 1000070000 2001 1676 Mar 24 12:11 ca-key.pem -rw-r--r--. 2 1000070000 2001 1075 Mar 24 12:11 ca.pem -rw-r--r--. 2 1000070000 2001 1079 Mar 24 12:12 client-cert.pem -rw-------. 2 1000070000 2001 1680 Mar 24 12:12 client-key.pem -rw-r-----. 2 1000070000 2001 352 Mar 24 12:12 ib_buffer_pool -rw-r-----. 2 1000070000 2001 12582912 Mar 24 12:20 ibdata1 -rw-r-----. 2 1000070000 2001 79691776 Mar 24 12:20 ib_logfile0 -rw-r-----. 2 1000070000 2001 79691776 Mar 24 12:11 ib_logfile1 -rw-r-----. 2 1000070000 2001 12582912 Mar 24 12:12 ibtmp1 drwxr-s---. 2 1000070000 2001 8192 Mar 24 12:12 mysql -rw-r-----. 2 1000070000 2001 2 Mar 24 12:12 mysql-1-h4afb.pid drwxr-s---. 2 1000070000 2001 8192 Mar 24 12:12 performance_schema -rw-------. 2 1000070000 2001 1676 Mar 24 12:12 private_key.pem -rw-r--r--. 2 1000070000 2001 452 Mar 24 12:12 public_key.pem drwxr-s---. 2 1000070000 2001 62 Mar 24 12:20 sampledb -rw-r--r--. 2 1000070000 2001 1079 Mar 24 12:11 server-cert.pem -rw-------. 2 1000070000 2001 1676 Mar 24 12:11 server-key.pem drwxr-s---. 2 1000070000 2001 8192 Mar 24 12:12 sys 你可以在这里看到MySQL数据库目录。 它使用GlusterFS作为后端存储，并作为绑定挂载给MySQL容器使用。 如果你检查OpenShift VM上的mount表，你将会看到GlusterFS的mount。 总结 在这里我们是在OpenShift之外创建了一个简单但功能强大的GlusterFS存储池。 该池可以独立于应用程序扩展和收缩。 该池的整个生命周期由一个简单的称为heketi的前端管理，你只需要在部署增长时进行手动干预。 对于日常配置操作，使用它的API与OpenShifts动态配置器交互，无需开发人员直接与基础架构团队进行交互。 o这就是我们如何将存储带入DevOps世界 - 无痛苦，并在OpenShift PaaS系统的开发人员工具中直接提供。 GlusterFS和OpenShift可跨越所有环境：裸机，虚拟机，私有和公共云（Azure，Google Cloud，AWS ...），确保应用程序可移植性，并避免云供应商锁定。 祝你愉快在容器中使用GlusterFS！ (c) 2017 Keith Tenzer 原文链接：https://keithtenzer.com/2017/03/24/storage-for-containers-using-gluster-part-ii/ Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-15 21:43:07 "},"practice/cephfs.html":{"url":"practice/cephfs.html","title":"CephFS","keywords":"","body":"CephFS Cephfs 是一个基于 ceph 集群且兼容POSIX标准的文件系统。创建 cephfs 文件系统时需要在 ceph 集群中添加 mds 服务，该服务负责处理 POSIX 文件系统中的 metadata 部分，实际的数据部分交由 ceph 集群中的 OSDs 处理。cephfs 支持以内核模块方式加载也支持 fuse 方式加载。无论是内核模式还是 fuse 模式，都是通过调用 libcephfs 库来实现 cephfs 文件系统的加载，而 libcephfs 库又调用 librados 库与 ceph 集群进行通信，从而实现 cephfs 的加载。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-09-01 17:25:47 "},"practice/using-ceph-for-persistent-storage.html":{"url":"practice/using-ceph-for-persistent-storage.html","title":"使用Ceph做持久化存储","keywords":"","body":"使用Ceph做持久化存储创建MySQL集群 本文中用到的 yaml 文件可以在 ../manifests/mariadb-cluster 目录下找到。 下面我们以部署一个高可用的 MySQL 集群为例，讲解如何使用 Ceph 做数据持久化，其中使用 StorageClass 动态创建 PV，Ceph 集群我们使用 kubernetes 集群外部的已有的集群，我们没有必要重新部署了。 在 1.4 以后，kubernetes 提供了一种更加方便的动态创建 PV 的方式；也就是说使用 StoragaClass 时无需预先创建固定大小的 PV，等待使用者创建 PVC 来使用；而是直接创建 PVC 即可分配使用。 使用 kubernetes 集群外部的 Ceph 存储 在部署 kubernetes 之前我们就已经有了 Ceph 集群，因此我们可以直接拿来用。但是 kubernetes 的所有节点（尤其是 master 节点）上依然需要安装 ceph 客户端。 yum install -y ceph-common 还需要将 ceph 的配置文件 ceph.conf 放在所有节点的 /etc/ceph 目录下。 Kubernetes 使用 ceph 存储需要用到如下配置： Monitors: Ceph montors 列表 Path：作为挂载的根路径，默认是 / User：RADOS用户名，默认是 admin secretFile：keyring 文件路径，默认是 /etc/ceph/user.secret，我们 Ceph 集群提供的文件是 ceph.client.admin.keyring，将在下面用到 secretRef：Ceph 认证 secret 的引用，如果配置了将会覆盖 secretFile。 readOnly：该文件系统是否只读。 Galera Cluster介绍 Galera是一个MySQL(也支持MariaDB，Percona)的同步多主集群软件。 从用户视角看，一组Galera集群可以看作一个具有多入口的MySQL库，用户可以同时从多个IP读写这个库。目前Galera已经得到广泛应用，例如Openstack中，在集群规模不大的情况下，稳定性已经得到了实践考验。真正的multi-master，即所有节点可以同时读写数据库。 详细步骤 以下步骤包括创建 Ceph 的配置 和 MySQL 的配置两部分。 配置 Ceph 关于 Ceph 的 yaml 文件可以在 ../manifest/mariadb-cluster 目录下找到。 1. 生成 Ceph secret 使用 Ceph 管理员提供给你的 ceph.client.admin.keyring 文件，我们将它放在了 /etc/ceph 目录下，用来生成 secret。 grep key /etc/ceph/ceph.client.admin.keyring |awk '{printf \"%s\", $NF}'|base64 将获得加密后的 key：QVFDWDA2aFo5TG5TQnhBQVl1b0lUL2V3YlRSaEtwVEhPWkxvUlE9PQ==，我们将在后面用到。 2. 创建租户namespace 创建 galera-namespace.yaml 文件内容为： apiVersion: v1 kind: Namespace metadata: name: galera 3. 创建 Ceph secret 创建 ceph-secret.yaml 文件内容为： apiVersion: v1 kind: Secret metadata: name: ceph-secret namespace: galera type: \"kubernetes.io/rbd\" data: key: QVFDWDA2aFo5TG5TQnhBQVl1b0lUL2V3YlRSaEtwVEhPWkxvUlE9PQ== 4. 创建 StorageClass 创建 ceph-class.yaml 文件内容为： apiVersion: storage.k8s.io/v1 kind: StorageClass metadata: name: ceph-web provisioner: kubernetes.io/rbd parameters: monitors: 172.28.7.98,172.28.7.99,172.28.7.100 adminId: admin adminSecretName: ceph-secret adminSecretNamespace: galera pool: rbd #此处默认是rbd池，生产上建议自己创建存储池隔离 userId: admin userSecretName: ceph-secret 此配置请参考 kubernetes 官方文档：https://kubernetes.io/docs/concepts/storage/persistent-volumes/#ceph-rbd 配置 MySQL 1. 创建 MySQL 配置文件 创建 mysql-config.yaml 文件内容为： apiVersion: v1 kind: ConfigMap metadata: name: mysql-config-vol namespace: galera labels: app: mysql data: mariadb.cnf: | [client] default-character-set = utf8 [mysqld] character-set-server = utf8 collation-server = utf8_general_ci # InnoDB optimizations innodb_log_file_size = 64M galera.cnf: | [galera] user = mysql bind-address = 0.0.0.0 # Optimizations innodb_flush_log_at_trx_commit = 0 sync_binlog = 0 expire_logs_days = 7 # Required settings default_storage_engine = InnoDB binlog_format = ROW innodb_autoinc_lock_mode = 2 query_cache_size = 0 query_cache_type = 0 # MariaDB Galera settings #wsrep_debug=ON wsrep_on=ON wsrep_provider=/usr/lib/galera/libgalera_smm.so wsrep_sst_method=rsync # Cluster settings (automatically updated) wsrep_cluster_address=gcomm:// wsrep_cluster_name=galera wsrep_node_address=127.0.0.1 2. 创建 MySQL root 用户和密码 创建加密密码 $ echo -n jimmysong|base64 amltbXlzb25n 注意：一定要用-n 去掉换行符，不然会报错。 创建 root 用户 $ echo -n root |base64 cm9vdA== 创建 MySQL secret 创建 mysql-secret.yaml 文件内容为： apiVersion: v1 kind: Secret metadata: name: mysql-secrets namespace: galera labels: app: mysql data: # Root password: changeit run echo -n jimmysong|base64 root-password: amltbXlzb25n # Root user: root root-user: cm9vdA== 3. 创建 yaml 配置文件 创建 MySQL 的 yaml 文件 galera-mariadb.yaml 内容为： apiVersion: v1 kind: Service metadata: annotations: service.alpha.kubernetes.io/tolerate-unready-endpoints: \"true\" name: mysql namespace: galera labels: app: mysql tier: data spec: ports: - port: 3306 name: mysql clusterIP: None selector: app: mysql --- apiVersion: apps/v1beta1 kind: StatefulSet metadata: name: mysql namespace: galera spec: serviceName: \"mysql\" replicas: 3 template: metadata: labels: app: mysql tier: data annotations: pod.beta.kubernetes.io/init-containers: '[ { \"name\": \"galera-init\", \"image\": \"sz-pg-oam-docker-hub-001.tendcloud.com/library/k8s-galera-init:latest\", \"args\": [\"-service=mysql\"], \"env\": [ { \"name\": \"POD_NAMESPACE\", \"valueFrom\": { \"fieldRef\": { \"apiVersion\": \"v1\", \"fieldPath\": \"metadata.namespace\" } } }, { \"name\": \"SAFE_TO_BOOTSTRAP\", \"value\": \"1\" }, { \"name\": \"DEBUG\", \"value\": \"1\" } ], \"volumeMounts\": [ { \"name\": \"config\", \"mountPath\": \"/etc/mysql/conf.d\" }, { \"name\": \"data\", \"mountPath\": \"/var/lib/mysql\" } ] } ]' spec: terminationGracePeriodSeconds: 10 containers: - name: mysql image: sz-pg-oam-docker-hub-001.tendcloud.com/library/mariadb:10.1 imagePullPolicy: IfNotPresent ports: - containerPort: 3306 name: mysql - containerPort: 4444 name: sst - containerPort: 4567 name: replication - containerPort: 4568 name: ist env: - name: MYSQL_ROOT_PASSWORD valueFrom: secretKeyRef: name: mysql-secrets key: root-password - name: MYSQL_ROOT_USER valueFrom: secretKeyRef: name: mysql-secrets key: root-user - name: MYSQL_INITDB_SKIP_TZINFO value: \"yes\" livenessProbe: exec: command: [\"sh\", \"-c\", \"mysql -u\\\"${MYSQL_ROOT_USER:-root}\\\" -p\\\"${MYSQL_ROOT_PASSWORD}\\\" -e 'show databases;'\"] initialDelaySeconds: 60 timeoutSeconds: 5 readinessProbe: exec: command: [\"sh\", \"-c\", \"mysql -u\\\"${MYSQL_ROOT_USER:-root}\\\" -p\\\"${MYSQL_ROOT_PASSWORD}\\\" -e 'show databases;'\"] initialDelaySeconds: 20 timeoutSeconds: 5 volumeMounts: - name: config mountPath: /etc/mysql/conf.d - name: data mountPath: /var/lib/mysql volumes: - name: config configMap: name: mysql-config-vol imagePullSecrets: - name: \"registrykey\" volumeClaimTemplates: - metadata: name: data annotations: volume.beta.kubernetes.io/storage-class: \"ceph-web\" #引用ceph class 的类 spec: accessModes: [ \"ReadWriteOnce\" ] resources: requests: storage: 3Gi 部署 MySQL 集群 在 /etc/mariadb-cluster 目录下执行： kubectl create -f . 验证 存在 issue，参考 Error creating rbd image: executable file not found in $PATH#38923 问题记录 如果没有安装 ceph-common 的话，kubernetes 在创建 PVC 的时候会有如下报错信息： Events: FirstSeen LastSeen Count From SubObjectPath Type Reason Message --------- -------- ----- ---- ------------- -------- ------ ------- 1h 12s 441 {persistentvolume-controller } Warning ProvisioningFailed Failed to provision volume with StorageClass \"ceph-web\": failed to create rbd image: executable file not found in $PATH, command output: 检查 kube-controller-manager 的日志将看到如下错误信息： journalctl -xe -u kube-controller-manager ... rbd_util.go:364] failed to create rbd image, output ... rbd.go:317] rbd: create volume failed, err: failed to create rbd image: executable file not found in $PATH, command output: 这是因为 kube-controller-manager 主机上没有安装 ceph-common 的缘故。 但是安装了 ceph-common 之后依然有问题： Sep 4 15:25:36 bj-xg-oam-kubernetes-001 kube-controller-manager: W0904 15:25:36.032128 13211 rbd_util.go:364] failed to create rbd image, output Sep 4 15:25:36 bj-xg-oam-kubernetes-001 kube-controller-manager: W0904 15:25:36.032201 13211 rbd_util.go:364] failed to create rbd image, output Sep 4 15:25:36 bj-xg-oam-kubernetes-001 kube-controller-manager: W0904 15:25:36.032252 13211 rbd_util.go:364] failed to create rbd image, output Sep 4 15:25:36 bj-xg-oam-kubernetes-001 kube-controller-manager: E0904 15:25:36.032276 13211 rbd.go:317] rbd: create volume failed, err: failed to create rbd image: fork/exec /usr/bin/rbd: invalid argument, command output: 该问题尚未解决，参考 Error creating rbd image: executable file not found in $PATH#38923 从日志记录来看追查到 pkg/volume/rbd/rbd.go 的 func (r *rbdVolumeProvisioner) Provision() (*v1.PersistentVolume, error) { 方法对 ceph-class.yaml 中的参数进行了验证和处理后调用了 pkg/volume/rbd/rdb_utils.go 文件第 344 行 CreateImage 方法（kubernetes v1.6.1版本）： func (util *RBDUtil) CreateImage(p *rbdVolumeProvisioner) (r *v1.RBDVolumeSource, size int, err error) { var output []byte capacity := p.options.PVC.Spec.Resources.Requests[v1.ResourceName(v1.ResourceStorage)] volSizeBytes := capacity.Value() // convert to MB that rbd defaults on sz := int(volume.RoundUpSize(volSizeBytes, 1024*1024)) volSz := fmt.Sprintf(\"%d\", sz) // rbd create l := len(p.rbdMounter.Mon) // pick a mon randomly start := rand.Int() % l // iterate all monitors until create succeeds. for i := start; i 该方法调用失败。 参考 https://github.com/kubernetes/examples/blob/master/staging/volumes/cephfs/README.md k8s-ceph-statefulsets-storageclass-nfs 动态卷有状态应用实践 Kubernetes persistent storage with Ceph https://kubernetes.io/docs/concepts/storage/persistent-volumes/#ceph-rbd Error creating rbd image: executable file not found in $PATH#38923 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-07 10:26:02 "},"practice/monitoring.html":{"url":"practice/monitoring.html","title":"集群与应用监控","keywords":"","body":"监控 Kubernetes 使得管理复杂环境变得更简单，但是对 kubernetes 本身的各种组件还有运行在 kubernetes 集群上的各种应用程序做到很好的洞察就很难了。Kubernetes 本身对应用程序的做了很多抽象，在生产环境下对这些不同的抽象组件的健康就是迫在眉睫的事情。 我们在安装 kubernetes 集群的时候，默认安装了 kubernetes 官方提供的 heapster 插件，可以对 kubernetes 集群上的应用进行简单的监控，获取 pod 级别的内存、CPU和网络监控信息，同时还能够通过 API 监控 kubernetes 中的基本资源监控指标。 然而，Prometheus 的出现让人眼前一亮，与 kubernetes 一样同样为 CNCF 中的项目，而且是第一个加入到 CNCF 中的项目。 Prometheus 是由 SoundCloud 开源监控告警解决方案，从 2012 年开始编写代码，再到 2015 年 GitHub 上开源以来，已经吸引了 9k+ 关注，以及很多大公司的使用；2016 年 Prometheus 成为继 k8s 后，第二名 CNCF(Cloud Native Computing Foundation) 成员。 作为新一代开源解决方案，很多理念与 Google SRE 运维之道不谋而合。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-09 10:53:54 "},"practice/heapster.html":{"url":"practice/heapster.html","title":"Heapster","keywords":"","body":"Heapster Heapster作为kubernetes安装过程中默认安装的一个插件，见安装heapster插件。这对于集群监控十分有用，同时在Horizontal Pod Autoscaling中也用到了，HPA将Heapster作为Resource Metrics API，向其获取metric，做法是在kube-controller-manager 中配置--api-server指向kube-aggregator，也可以使用heapster来实现，通过在启动heapster的时候指定--api-server=true。 Heapster可以收集Node节点上的cAdvisor数据，还可以按照kubernetes的资源类型来集合资源，比如Pod、Namespace域，可以分别获取它们的CPU、内存、网络和磁盘的metric。默认的metric数据聚合时间间隔是1分钟。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-09 10:47:20 "},"practice/using-heapster-to-get-object-metrics.html":{"url":"practice/using-heapster-to-get-object-metrics.html","title":"使用Heapster获取集群和对象的metric数据","keywords":"","body":"使用Heapster获取集群对象的metric数据 Heapster作为kubernetes安装过程中默认安装的一个插件，见安装heapster插件。这对于集群监控十分有用，同时在Horizontal Pod Autoscaling中也用到了，HPA将Heapster作为Resource Metrics API，向其获取metric，做法是在kube-controller-manager 中配置--api-server指向kube-aggregator，也可以使用heapster来实现，通过在启动heapster的时候指定--api-server=true。 Heapster可以收集Node节点上的cAdvisor数据，还可以按照kubernetes的资源类型来集合资源，比如Pod、Namespace域，可以分别获取它们的CPU、内存、网络和磁盘的metric。默认的metric数据聚合时间间隔是1分钟。 架构 下面是Heapster架构图： 图片 - Heapster架构图 Heapser是用Go语言开发Kubernetes集群计算资源使用情况的数据采集工具，编译后可以直接以一个二进制文件运行，通过向heapster传递的参数来指定数据采集行为，这些数据可以选择多种sink方式，例如Graphite、influxDB、OpenTSDB、ElasticSearch、Kafka等。 使用案例 Heapster使用起来很简单，本身就是二进制文件，直接使用命令行启动，也可以放在容器里运行，在作为kubernetes插件运行时，我们是直接放在容器中的，见安装heapster插件。 运行 下面是heapster的启动参数： Flag Description --allowed-users string comma-separated list of allowed users --alsologtostderr log to standard error as well as files --api-server Enable API server for the Metrics API. If set, the Metrics API will be served on --insecure-port (internally) and --secure-port (externally). --authentication-kubeconfig string kubeconfig file pointing at the 'core' kubernetes server with enough rights to create tokenaccessreviews.authentication.k8s.io. --authentication-token-webhook-cache-ttl duration The duration to cache responses from the webhook token authenticator. (default 10s) --authorization-kubeconfig string kubeconfig file pointing at the 'core' kubernetes server with enough rights to create subjectaccessreviews.authorization.k8s.io. --authorization-webhook-cache-authorized-ttl duration The duration to cache 'authorized' responses from the webhook authorizer. (default 10s) --authorization-webhook-cache-unauthorized-ttl duration The duration to cache 'unauthorized' responses from the webhook authorizer. (default 10s) --bind-address ip The IP address on which to listen for the --secure-port port. The associated interface(s) must be reachable by the rest of the cluster, and by CLI/web clients. If blank, all interfaces will be used (0.0.0.0). (default 0.0.0.0) --cert-dir string The directory where the TLS certs are located (by default /var/run/kubernetes). If --tls-cert-file and --tls-private-key-file are provided, this flag will be ignored. (default \"/var/run/kubernetes\") --client-ca-file string If set, any request presenting a client certificate signed by one of the authorities in the client-ca-file is authenticated with an identity corresponding to the CommonName of the client certificate. --contention-profiling Enable contention profiling. Requires --profiling to be set to work. --disable-export Disable exporting metrics in api/v1/metric-export --enable-swagger-ui Enables swagger ui on the apiserver at /swagger-ui --heapster-port int port used by the Heapster-specific APIs (default 8082) --historical-source string which source type to use for the historical API (should be exactly the same as one of the sink URIs), or empty to disable the historical API --label-seperator string seperator used for joining labels (default \",\") --listen-ip string IP to listen on, defaults to all IPs --log-backtrace-at traceLocation when logging hits line file:N, emit a stack trace (default :0) --log-dir string If non-empty, write log files in this directory --log-flush-frequency duration Maximum number of seconds between log flushes (default 5s) --logtostderr log to standard error instead of files (default true) --max-procs int max number of CPUs that can be used simultaneously. Less than 1 for default (number of cores) --metric-resolution duration The resolution at which heapster will retain metrics. (default 1m0s) --profiling Enable profiling via web interface host:port/debug/pprof/ (default true) --requestheader-allowed-names stringSlice List of client certificate common names to allow to provide usernames in headers specified by --requestheader-username-headers. If empty, any client certificate validated by the authorities in --requestheader-client-ca-file is allowed. --requestheader-client-ca-file string Root certificate bundle to use to verify client certificates on incoming requests before trusting usernames in headers specified by --requestheader-username-headers --requestheader-extra-headers-prefix stringSlice List of request header prefixes to inspect. X-Remote-Extra- is suggested. (default [x-remote-extra-]) --requestheader-group-headers stringSlice List of request headers to inspect for groups. X-Remote-Group is suggested. (default [x-remote-group]) --requestheader-username-headers stringSlice List of request headers to inspect for usernames. X-Remote-User is common. (default [x-remote-user]) --secure-port int The port on which to serve HTTPS with authentication and authorization. If 0, don't serve HTTPS at all. (default 6443) --sink *flags.Uris external sink(s) that receive data (default []) --source *flags.Uris source(s) to watch (default []) --stderrthreshold severity logs at or above this threshold go to stderr (default 2) --tls-ca-file string If set, this certificate authority will used for secure access from Admission Controllers. This must be a valid PEM-encoded CA bundle. Altneratively, the certificate authority can be appended to the certificate provided by --tls-cert-file. --tls-cert string file containing TLS certificate --tls-cert-file string File containing the default x509 Certificate for HTTPS. (CA cert, if any, concatenated after server cert). If HTTPS serving is enabled, and --tls-cert-file and --tls-private-key-file are not provided, a self-signed certificate and key are generated for the public address and saved to /var/run/kubernetes. --tls-client-ca string file containing TLS client CA for client cert validation --tls-key string file containing TLS key --tls-private-key-file string File containing the default x509 private key matching --tls-cert-file. --tls-sni-cert-key namedCertKey A pair of x509 certificate and private key file paths, optionally suffixed with a list of domain patterns which are fully qualified domain names, possibly with prefixed wildcard segments. If no domain patterns are provided, the names of the certificate are extracted. Non-wildcard matches trump over wildcard matches, explicit domain patterns trump over extracted names. For multiple key/certificate pairs, use the --tls-sni-cert-key multiple times. Examples: \"example.key,example.crt\" or \"*.foo.com,foo.com:foo.key,foo.crt\". (default []) --v Level log level for V logs --version print version info and exit --vmodule moduleSpec comma-separated list of pattern=N settings for file-filtered logging Version version: v1.4.0 commit: 546ab66f API使用 Heapster提供RESTful API接口，下面以获取spark-cluster namespace的memory usage为例讲解Heapster API的使用。 构造URL地址 https://172.20.0.113:6443/api/v1/proxy/namespaces/kube-system/services/heapster/api/v1/model/namespaces/spark-cluster/metrics/memory/usage?start=2017-10-16T09:14:00Z&end=2017-10-16T09:16:00Z 结果 访问该地址获取的结果是这样的： { \"metrics\": [ { \"timestamp\": \"2017-10-16T09:14:00Z\", \"value\": 322592768 }, { \"timestamp\": \"2017-10-16T09:15:00Z\", \"value\": 322592768 }, { \"timestamp\": \"2017-10-16T09:16:00Z\", \"value\": 322592768 } ], \"latestTimestamp\": \"2017-10-16T09:16:00Z\" } 注意：Heapster中查询的所有值都是以最小单位为单位，比如CPU为1milicore，内存为B。 第一部分：Heapster API地址 https://172.20.0.113:6443/api/v1/proxy/namespaces/kube-system/services/heapster/ 可以使用下面的命令获取: $ kubectl cluster-info Heapster is running at https://172.20.0.113:6443/api/v1/proxy/namespaces/kube-system/services/heapster ... 第二部分：Heapster API参数 /api/v1/model/namespaces/spark-cluster/metrics/memory/usage 表示查询的是spark-cluster namespace中的memory/usage的metrics。 第三部分：时间片 ?start=2017-10-16T09:14:00Z&end=2017-10-16T09:16:00Z 查询参数为时间片：包括start和end。使用RFC-3339时间格式，在Linux系统中可以这样获取： $ date --rfc-3339=\"seconds\" 2017-10-16 17:23:20+08:00 该时间中的空格替换成T，最后的+08:00替换成Z代表时区。可以只指定start时间，end时间自动设置为当前时间。 参考 kubernetes metrics Heapster metric model Heapster storage schema Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-10-16 17:41:04 "},"practice/prometheus.html":{"url":"practice/prometheus.html","title":"Prometheus","keywords":"","body":"Prometheus Prometheus 是由 SoundCloud 开源监控告警解决方案，从 2012 年开始编写代码，再到 2015 年 github 上开源以来，已经吸引了 9k+ 关注，以及很多大公司的使用；2016 年 Prometheus 成为继 k8s 后，第二名 CNCF(Cloud Native Computing Foundation) 成员。 作为新一代开源解决方案，很多理念与 Google SRE 运维之道不谋而合。 主要功能 多维 数据模型（时序由 metric 名字和 k/v 的 labels 构成）。 灵活的查询语句（PromQL）。 无依赖存储，支持 local 和 remote 不同模型。 采用 http 协议，使用 pull 模式，拉取数据，简单易懂。 监控目标，可以采用服务发现或静态配置的方式。 支持多种统计数据模型，图形化友好。 核心组件 Prometheus Server， 主要用于抓取数据和存储时序数据，另外还提供查询和 Alert Rule 配置管理。 client libraries，用于对接 Prometheus Server, 可以查询和上报数据。 push gateway ，用于批量，短期的监控数据的汇总节点，主要用于业务数据汇报等。 各种汇报数据的 exporters ，例如汇报机器数据的 node_exporter, 汇报 MongoDB 信息的 MongoDB exporter 等等。 用于告警通知管理的 alertmanager 。 基础架构 一图胜千言，先来张官方的架构图： 图片 - Prometheus 架构图 从这个架构图，也可以看出 Prometheus 的主要模块包含， Server, Exporters, Pushgateway, PromQL, Alertmanager, WebUI 等。 它大致使用逻辑是这样： Prometheus server 定期从静态配置的 targets 或者服务发现的 targets 拉取数据。 当新拉取的数据大于配置内存缓存区的时候，Prometheus 会将数据持久化到磁盘（如果使用 remote storage 将持久化到云端）。 Prometheus 可以配置 rules，然后定时查询数据，当条件触发的时候，会将 alert 推送到配置的 Alertmanager。 Alertmanager 收到警告的时候，可以根据配置，聚合，去重，降噪，最后发送警告。 可以使用 API， Prometheus Console 或者 Grafana 查询和聚合数据。 注意 Prometheus 的数据是基于时序的 float64 的值，如果你的数据值有更多类型，无法满足。 Prometheus 不适合做审计计费，因为它的数据是按一定时间采集的，关注的更多是系统的运行瞬时状态以及趋势，即使有少量数据没有采集也能容忍，但是审计计费需要记录每个请求，并且数据长期存储，这个和 Prometheus 无法满足，可能需要采用专门的审计系统。 以上介绍来自 https://github.com/songjiayang/prometheus_practice/ Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-11 10:12:28 "},"practice/using-prometheus-to-monitor-kuberentes-cluster.html":{"url":"practice/using-prometheus-to-monitor-kuberentes-cluster.html","title":"使用Prometheus监控kubernetes集群","keywords":"","body":"使用Prometheus监控kubernetes集群 我们使用 Giantswarm 开源的 kubernetes-promethues 来监控 kubernetes 集群，所有的 YAML 文件可以在 ../manifests/prometheus 目录下找到。 需要用到的镜像有： sz-pg-oam-docker-hub-001.tendcloud.com/library/prometheus-alertmanager:v0.7.1 sz-pg-oam-docker-hub-001.tendcloud.com/library/grafana:4.2.0 sz-pg-oam-docker-hub-001.tendcloud.com/library/giantswarm-tiny-tools:latest sz-pg-oam-docker-hub-001.tendcloud.com/library/prom-prometheus:v1.7.0 sz-pg-oam-docker-hub-001.tendcloud.com/library/kube-state-metrics:v1.0.1 sz-pg-oam-docker-hub-001.tendcloud.com/library/dockermuenster-caddy:0.9.3 sz-pg-oam-docker-hub-001.tendcloud.com/library/prom-node-exporter:v0.14.0 同时备份到时速云： index.tenxcloud.com/jimmy/prometheus-alertmanager:v0.7.1 index.tenxcloud.com/jimmy/grafana:4.2.0 index.tenxcloud.com/jimmy/giantswarm-tiny-tools:latest index.tenxcloud.com/jimmy/prom-prometheus:v1.7.0 index.tenxcloud.com/jimmy/kube-state-metrics:v1.0.1 index.tenxcloud.com/jimmy/dockermuenster-caddy:0.9.3 index.tenxcloud.com/jimmy/prom-node-exporter:v0.14.0 注：所有镜像都是从官方镜像仓库下载下。 部署 我将部署时需要用的的配置文件分成了 namespace、serviceaccount、configmaps、clusterrolebinding 和最后的部署 prometheus、grafana 的过程。 ## 创建 monitoring namespaece kubectl create -f prometheus-monitoring-ns.yaml ## 创建 serviceaccount kubectl create -f prometheus-monitoring-serviceaccount.yaml ## 创建 configmaps kubectl create -f prometheus-configmaps.yaml ## 创建 clusterrolebinding kubectl create clusterrolebinding kube-state-metrics --clusterrole=cluster-admin --serviceaccount=monitoring:kube-state-metrics kubectl create clusterrolebinding prometheus --clusterrole=cluster-admin --serviceaccount=monitoring:prometheus ## 部署 Prometheus kubectl create -f prometheus-monitoring.yaml 访问 kubernetes 任何一个 node 上的 Grafana service 的 nodeport： 图片 - Grafana页面 该图中的数据显示明显有问题，还需要修正。 prometheus-monitoring.yaml 文件中有一个 Job 就是用来导入 grafana dashboard 配置信息的，如果该 Job 执行失败，可以单独在在 monitoring 的 namespace 中启动一个容器，将 manifests/prometheus 目录下的 json 文件复制到容器中，然后进入容器 json 文件的目录下执行： for file in *-datasource.json ; do if [ -e \"$file\" ] ; then echo \"importing $file\" && curl --silent --fail --show-error \\ --request POST http://admin:admin@grafana:3000/api/datasources \\ --header \"Content-Type: application/json\" \\ --data-binary \"@$file\" ; echo \"\" ; fi done ; for file in *-dashboard.json ; do if [ -e \"$file\" ] ; then echo \"importing $file\" && ( echo '{\"dashboard\":'; \\ cat \"$file\"; \\ echo ',\"overwrite\":true,\"inputs\":[{\"name\":\"DS_PROMETHEUS\",\"type\":\"datasource\",\"pluginId\":\"prometheus\",\"value\":\"prometheus\"}]}' ) \\ | jq -c '.' \\ | curl --silent --fail --show-error \\ --request POST http://admin:admin@grafana:3000/api/dashboards/import \\ --header \"Content-Type: application/json\" \\ --data-binary \"@-\" ; echo \"\" ; fi done 这样也可以向 grafana 中导入 dashboard。 存在的问题 该项目的代码中存在几个问题。 1. RBAC 角色授权问题 需要用到两个 clusterrolebinding： kube-state-metrics，对应的serviceaccount是kube-state-metrics prometheus，对应的 serviceaccount是 prometheus-k8s 在部署 Prometheus 之前应该先创建 serviceaccount、clusterrole、clusterrolebinding 等对象，否则在安装过程中可能因为权限问题而导致各种错误，所以这些配置应该写在一个单独的文件中，而不应该跟其他部署写在一起，即使要写在一个文件中，也应该写在文件的最前面，因为使用 kubectl 部署的时候，kubectl 不会判断 YAML 文件中的资源依赖关系，只是简单的从头部开始执行部署，因此写在文件前面的对象会先部署。 解决方法 也可以绕过复杂的 RBAC 设置，直接使用下面的命令将对应的 serviceaccount 设置成 admin 权限，如下： kubectl create clusterrolebinding kube-state-metrics --clusterrole=cluster-admin --serviceaccount=monitoring:kube-state-metrics kubectl create clusterrolebinding prometheus --clusterrole=cluster-admin --serviceaccount=monitoring:prometheus 参考 RBAC——基于角色的访问控制 2. API 兼容问题 从 kube-state-metrics 日志中可以看出用户 kube-state-metrics 没有权限访问如下资源类型： *v1.Job *v1.PersistentVolumeClaim *v1beta1.StatefulSet *v2alpha1.CronJob 而在我们使用的 kubernetes 1.6.0 版本的集群中 API 路径跟 kube-state-metrics 中不同，无法 list 以上三种资源对象的资源。详情见：https://github.com/giantswarm/kubernetes-prometheus/issues/77 3. Job 中的权限认证问题 在 grafana-import-dashboards 这个 job 中有个 init-containers 其中指定的 command 执行错误，应该使用 curl -sX GET -H \"Authorization:bearer `cat /var/run/secrets/kubernetes.io/serviceaccount/token`\" -k https://kubernetes.default/api/v1/namespaces/monitoring/endpoints/grafana 不需要指定 csr 文件，只需要 token 即可。 参考 wait-for-endpoints init-containers fails to load with k8s 1.6.0 #56 参考 Kubernetes Setup for Prometheus and Grafana RBAC——基于角色的访问控制 wait-for-endpoints init-containers fails to load with k8s 1.6.0 #56 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-10-11 23:43:48 "},"practice/services-management-tool.html":{"url":"practice/services-management-tool.html","title":"服务编排管理","keywords":"","body":"服务编排管理 Kubernetes虽然提供了多种容器编排对象，例如Deployment、StatefulSet、DeamonSet、Job等，还有多种基础资源封装例如ConfigMap、Secret、Serivce等，但是一个应用往往有多个服务，有的可能还要依赖持久化存储，当这些服务之间直接互相依赖，需要有一定的组合的情况下，使用YAML文件的方式配置应用往往十分繁琐还容易出错，这时候就需要服务编排工具。 服务编排管理工具就是构建在kubernetes的基础object之上，统筹各个服务之间的关系和依赖的。目前常用到的工具是 Helm。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-10-19 15:29:43 "},"practice/helm.html":{"url":"practice/helm.html","title":"使用Helm管理kubernetes应用","keywords":"","body":"使用Helm管理kubernetes应用 读完本文后您应该可以自己创建chart，并创建自己的私有chart仓库。 Helm是一个kubernetes应用的包管理工具，用来管理charts——预先配置好的安装包资源，有点类似于Ubuntu的APT和CentOS中的yum。 Helm chart是用来封装kubernetes原生应用程序的yaml文件，可以在你部署应用的时候自定义应用程序的一些metadata，便与应用程序的分发。 Helm和charts的主要作用： 应用程序封装 版本管理 依赖检查 便于应用程序分发 安装Helm 前提要求 Kubernetes1.5以上版本 集群可访问到的镜像仓库 执行helm命令的主机可以访问到kubernetes集群 安装步骤 首先需要安装helm客户端 curl https://raw.githubusercontent.com/kubernetes/helm/master/scripts/get > get_helm.sh chmod 700 get_helm.sh ./get_helm.sh 创建tiller的serviceaccount和clusterrolebinding kubectl create serviceaccount --namespace kube-system tiller kubectl create clusterrolebinding tiller-cluster-rule --clusterrole=cluster-admin --serviceaccount=kube-system:tiller 然后安装helm服务端tiller helm init -i sz-pg-oam-docker-hub-001.tendcloud.com/library/kubernetes-helm-tiller:v2.3.1 我们使用-i指定自己的镜像，因为官方的镜像因为某些原因无法拉取。 为应用程序设置serviceAccount： kubectl patch deploy --namespace kube-system tiller-deploy -p '{\"spec\":{\"template\":{\"spec\":{\"serviceAccount\":\"tiller\"}}}}' 检查是否安装成功： $ kubectl -n kube-system get pods|grep tiller tiller-deploy-2372561459-f6p0z 1/1 Running 0 1h $ helm version Client: &version.Version{SemVer:\"v2.3.1\", GitCommit:\"32562a3040bb5ca690339b9840b6f60f8ce25da4\", GitTreeState:\"clean\"} Server: &version.Version{SemVer:\"v2.3.1\", GitCommit:\"32562a3040bb5ca690339b9840b6f60f8ce25da4\", GitTreeState:\"clean\"} 创建自己的chart 我们创建一个名为mychart的chart，看一看chart的文件结构。 $ helm create mongodb $ tree mongodb mongodb ├── Chart.yaml #Chart本身的版本和配置信息 ├── charts #依赖的chart ├── templates #配置模板目录 │ ├── NOTES.txt #helm提示信息 │ ├── _helpers.tpl #用于修改kubernetes objcet配置的模板 │ ├── deployment.yaml #kubernetes Deployment object │ └── service.yaml #kubernetes Serivce └── values.yaml #kubernetes object configuration 2 directories, 6 files 模板 Templates目录下是yaml文件的模板，遵循Go template语法。使用过Hugo的静态网站生成工具的人应该对此很熟悉。 我们查看下deployment.yaml文件的内容。 apiVersion: extensions/v1beta1 kind: Deployment metadata: name: {{ template \"fullname\" . }} labels: chart: \"{{ .Chart.Name }}-{{ .Chart.Version | replace \"+\" \"_\" }}\" spec: replicas: {{ .Values.replicaCount }} template: metadata: labels: app: {{ template \"fullname\" . }} spec: containers: - name: {{ .Chart.Name }} image: \"{{ .Values.image.repository }}:{{ .Values.image.tag }}\" imagePullPolicy: {{ .Values.image.pullPolicy }} ports: - containerPort: {{ .Values.service.internalPort }} livenessProbe: httpGet: path: / port: {{ .Values.service.internalPort }} readinessProbe: httpGet: path: / port: {{ .Values.service.internalPort }} resources: {{ toYaml .Values.resources | indent 12 }} 这是该应用的Deployment的yaml配置文件，其中的双大括号包扩起来的部分是Go template，其中的Values是在values.yaml文件中定义的： # Default values for mychart. # This is a YAML-formatted file. # Declare variables to be passed into your templates. replicaCount: 1 image: repository: nginx tag: stable pullPolicy: IfNotPresent service: name: nginx type: ClusterIP externalPort: 80 internalPort: 80 resources: limits: cpu: 100m memory: 128Mi requests: cpu: 100m memory: 128Mi 比如在Deployment.yaml中定义的容器镜像image: \"{{ .Values.image.repository }}:{{ .Values.image.tag }}\"其中的： .Values.image.repository就是nginx .Values.image.tag就是stable 以上两个变量值是在create chart的时候自动生成的默认值。 我们将默认的镜像地址和tag改成我们自己的镜像sz-pg-oam-docker-hub-001.tendcloud.com/library/nginx:1.9。 检查配置和模板是否有效 当使用kubernetes部署应用的时候实际上讲templates渲染成最终的kubernetes能够识别的yaml格式。 使用helm install --dry-run --debug 命令来验证chart配置。该输出中包含了模板的变量配置与最终渲染的yaml文件。 $ helm install --dry-run --debug mychart Created tunnel using local port: '58406' SERVER: \"localhost:58406\" CHART PATH: /Users/jimmy/Workspace/github/bitnami/charts/incubator/mean/charts/mychart NAME: filled-seahorse REVISION: 1 RELEASED: Tue Oct 24 18:57:13 2017 CHART: mychart-0.1.0 USER-SUPPLIED VALUES: {} COMPUTED VALUES: image: pullPolicy: IfNotPresent repository: sz-pg-oam-docker-hub-001.tendcloud.com/library/nginx tag: 1.9 replicaCount: 1 resources: limits: cpu: 100m memory: 128Mi requests: cpu: 100m memory: 128Mi service: externalPort: 80 internalPort: 80 name: nginx type: ClusterIP HOOKS: MANIFEST: --- # Source: mychart/templates/service.yaml apiVersion: v1 kind: Service metadata: name: filled-seahorse-mychart labels: chart: \"mychart-0.1.0\" spec: type: ClusterIP ports: - port: 80 targetPort: 80 protocol: TCP name: nginx selector: app: filled-seahorse-mychart --- # Source: mychart/templates/deployment.yaml apiVersion: extensions/v1beta1 kind: Deployment metadata: name: filled-seahorse-mychart labels: chart: \"mychart-0.1.0\" spec: replicas: 1 template: metadata: labels: app: filled-seahorse-mychart spec: containers: - name: mychart image: \"sz-pg-oam-docker-hub-001.tendcloud.com/library/nginx:1.9\" imagePullPolicy: IfNotPresent ports: - containerPort: 80 livenessProbe: httpGet: path: / port: 80 readinessProbe: httpGet: path: / port: 80 resources: limits: cpu: 100m memory: 128Mi requests: cpu: 100m memory: 128Mi 我们可以看到Deployment和Service的名字前半截由两个随机的单词组成，最后才是我们在values.yaml中配置的值。 部署到kubernetes 在mychart目录下执行下面的命令将nginx部署到kubernetes集群上。 helm install . NAME: eating-hound LAST DEPLOYED: Wed Oct 25 14:58:15 2017 NAMESPACE: default STATUS: DEPLOYED RESOURCES: ==> v1/Service NAME CLUSTER-IP EXTERNAL-IP PORT(S) AGE eating-hound-mychart 10.254.135.68 80/TCP 0s ==> extensions/v1beta1/Deployment NAME DESIRED CURRENT UP-TO-DATE AVAILABLE AGE eating-hound-mychart 1 1 1 0 0s NOTES: 1. Get the application URL by running these commands: export POD_NAME=$(kubectl get pods --namespace default -l \"app=eating-hound-mychart\" -o jsonpath=\"{.items[0].metadata.name}\") echo \"Visit http://127.0.0.1:8080 to use your application\" kubectl port-forward $POD_NAME 8080:80 现在nginx已经部署到kubernetes集群上，本地执行提示中的命令在本地主机上访问到nginx实例。 export POD_NAME=$(kubectl get pods --namespace default -l \"app=eating-hound-mychart\" -o jsonpath=\"{.items[0].metadata.name}\") echo \"Visit http://127.0.0.1:8080 to use your application\" kubectl port-forward $POD_NAME 8080:80 在本地访问http://127.0.0.1:8080即可访问到nginx。 查看部署的relaese $ helm list NAME REVISION UPDATED STATUS CHART NAMESPACE eating-hound 1 Wed Oct 25 14:58:15 2017 DEPLOYED mychart-0.1.0 default 删除部署的release $ helm delete eating-hound release \"eating-hound\" deleted 打包分享 我们可以修改Chart.yaml中的helm chart配置信息，然后使用下列命令将chart打包成一个压缩文件。 helm package . 打包出mychart-0.1.0.tgz文件。 依赖 我们可以在requirement.yaml中定义应用所依赖的chart，例如定义对mariadb的依赖： dependencies: - name: mariadb version: 0.6.0 repository: https://kubernetes-charts.storage.googleapis.com 使用helm lint .命令可以检查依赖和模板配置是否正确。 安装源 我们在前面安装chart可以通过HTTP server的方式提供。 $ helm serve Regenerating index. This may take a moment. Now serving you on 127.0.0.1:8879 访问http://localhost:8879可以看到刚刚安装的chart。 图片 - Helm chart源 点击链接即可以下载chart的压缩包。 部署MEAN测试案例 MEAN是用来构建网站和web应用的免费开源的JavaScript软件栈，该软件栈包括MongoDB、Express.js、Angular和Node.js。 下载charts $ git clone https://github.com/bitnami/charts.git $ cd charts/incubator/mean $ helm dep list NAME VERSION REPOSITORY STATUS mongodb 0.4.x https://kubernetes-charts.storage.googleapis.com/ missing 缺少mongodb的依赖，需要更新一下chart。 注：https://kubernetes-charts.storage.googleapis.com/是Google维护的chart库，访问该地址可以看到所有的chart列表。 $ helm dep update Hang tight while we grab the latest from your chart repositories... ...Unable to get an update from the \"local\" chart repository (http://127.0.0.1:8879/charts): Get http://127.0.0.1:8879/charts/index.yaml: dial tcp 127.0.0.1:8879: getsockopt: connection refused ...Successfully got an update from the \"stable\" chart repository Update Complete. ⎈Happy Helming!⎈ Saving 1 charts Downloading mongodb from repo https://kubernetes-charts.storage.googleapis.com/ 所有的image都在 values.yaml 文件中配置。 下载缺失的chart。 $ helm dep build Hang tight while we grab the latest from your chart repositories... ...Unable to get an update from the \"local\" chart repository (http://127.0.0.1:8879/charts): Get http://127.0.0.1:8879/charts/index.yaml: dial tcp 127.0.0.1:8879: getsockopt: connection refused ...Successfully got an update from the \"stable\" chart repository Update Complete. ⎈Happy Helming!⎈ Saving 1 charts Downloading mongodb from repo https://kubernetes-charts.storage.googleapis.com/ 修改mongodb chart配置 将刚才下载的charts/mongodb-0.4.17.tgz给解压后，修改其中的配置： 将persistence下的enabled设置为false 将image修改为我们的私有镜像：sz-pg-oam-docker-hub-001.tendcloud.com/library/bitnami-mongodb:3.4.9-r1 执行helm install --dry-run --debug .确定模板无误。 将修改后的mongodb chart打包，在mongodb的目录下执行： helm package . 现在再访问前面启动的helm server http://localhost:8879将可以在页面上看到mongodb-0.4.17这个chart。 我们对官方chart配置做了如下修改后推送到了自己的chart仓库： requirements.yaml和requirements.lock文件中的repository为http://localhost:8879 将values.yaml中的storageClass设置为null 将values.yaml中的Image都改为私有镜像 repositroy都设置为http://localhost:8879 注：因为我们没有使用PVC所以将所有的关于持久化存储的配置都设置为false了。 部署MEAN 在mean目录下执行： helm install . NAME: orbiting-platypus LAST DEPLOYED: Wed Oct 25 16:21:48 2017 NAMESPACE: default STATUS: DEPLOYED RESOURCES: ==> v1/Secret NAME TYPE DATA AGE orbiting-platypus-mongodb Opaque 2 2s ==> v1/ConfigMap NAME DATA AGE orbiting-platypus-mean 1 2s ==> v1/Service NAME CLUSTER-IP EXTERNAL-IP PORT(S) AGE orbiting-platypus-mongodb 10.254.144.208 27017/TCP 2s orbiting-platypus-mean 10.254.165.23 80/TCP 2s ==> extensions/v1beta1/Deployment NAME DESIRED CURRENT UP-TO-DATE AVAILABLE AGE orbiting-platypus-mean 1 1 1 0 2s orbiting-platypus-mongodb 1 1 1 0 2s NOTES: Get the URL of your Node app by running: export POD_NAME=$(kubectl get pods --namespace default -l \"app=orbiting-platypus-mean\" -o jsonpath=\"{.items[0].metadata.name}\") echo http://127.0.0.1:8080/ kubectl port-forward $POD_NAME 8080:80 这样MEAN软件栈就部署到你的kuberentes集群里面了（默认是在default namespace下）。 验证检查 为了验证MEAN是否安装成功过，可以使用kubectl get pods查看pod是否启动完成，会先启动mongodb的pod，然后启动MEAN中的4步init。 访问Web UI 在Ingress中增加如下配置： - host: mean.jimmysong.io http: paths: - backend: serviceName: orbiting-platypus-mean servicePort: 80 path: / 然后在页面中更新ingress: kubectl repalce -f ingress.yaml 关于Ingress配置请参考：边缘节点配置 然后在本地的/etc/hosts文件中增加一条配置： 172.20.0.119 mean.jimmysong.io 注：172.20.0.119即边缘节点的VIP。 因为该页面需要加载google的angularjs、还有两个css在国内无法访问，可以使用curl测试： curl mean.jimmysong.io 将会返回HTML内容： Node/Angular Todo App html { overflow-y: scroll; } body { padding-top: 50px; } #todo-list { margin-bottom: 30px; } #todo-form { margin-bottom: 50px; } I'm a Todo-aholic {{ todos.length }} {{ todo.text }} Add A demo by Scotch. Read the tutorial. 访问 http://mean.jimmysong.io 可以看到如下界面，我在其中添加几条todo： 图片 - TODO应用的Web页面 注：Todo中的文字来自What does the fox say？ 测试完成后可以使用下面的命令将mean chart推送的本地chart仓库中。 在mean目录下执行： helm package . 再次刷新http://localhost:8879将可以看到如下三个chart： mean mean-0.1.3 mongodb mongodb-0.4.17 mychart mychart-0.1.0 参考 Deploy, Scale And Upgrade An Application On Kubernetes With Helm Helm charts Go template Helm docs How To Create Your First Helm Chart Speed deployment on Kubernetes with Helm Chart – Quick YAML example from scratch Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-01 23:39:54 "},"practice/create-private-charts-repo.html":{"url":"practice/create-private-charts-repo.html","title":"构建私有Chart仓库","keywords":"","body":"构建私有Chart仓库 使用Chart便于封装和管理kubernetes中的应用，因此当企业内部的应用多了以后，互相依赖、部署环境复杂之后，原先的直接使用yaml文件的管理方式已经不再适应生产的需要，因此我们有必要构建自己的chart仓库。本文中我们将使用 GitHub Pages 来构建我们自己的 chart 仓库。 目的 我们需要构建一个GitHub pages存储所有chart的压缩文件，最好还要有一个前端来展示和搜索chart。 什么是Chart Chart是helm管理的应用的打包格式。它包括如下特征： Chart中包括一系列的yaml格式的描述文件。 一个Chart只用来部署单个的应用的，不应该过于复杂，不应该包含多个依赖，相当于一个微服务。 Chart有特定的目录结构，可以打包起来进行版本控制。 Chart的组成结构 我们以nginx的chart为例，讲解chart的组成结构。 nginx/ Chart.yaml # 必须：一个包含chart的名称、版本和启用条件信息等的YAML文件 LICENSE # 可选: chart的许可证 README.md # 可选: 使用说明 requirements.yaml # 可选: 该chart的依赖配置 values.yaml # 必须：该chart的默认配置值 charts/ # 可选: 包含该chart依赖的chart templates/ # 可选：kubernetes manifest文件模板，用于生成kubernetes yaml文件 templates/NOTES.txt # 可选: 该chart的使用说明和提示信息文本文件，作为helm install后的提示信息 Chart的安装方式 安装chart主要分为安装本地定义的chart和远程chart仓库中的chart两种方式。 安装本地chart 指定本地chart目录：helm install . 指定本地chart压缩包：helm install nginx-1.2.3.tgz 安装chart仓库中的chart 使用默认的远程仓库：helm install stable/nginx 使用指定的仓库：helm install localhost:8879/nginx-1.2.3.tgz 实际上可以将chart打包后作为静态文件托管到web服务器上，例如GitHub pages作为chart仓库也可以。 依赖管理 有两种方式来管理chart的依赖。 直接在本的chart的charts目录下定义 通过在requirements.yaml文件中定义依赖的chart 在每个chart的charts目录下可以定义依赖的子chart。子chart有如下特点： 无法访问父chart中的配置 父chart可以覆盖子chart中的配置 Chart仓库 Chart 仓库（repository）是一个用来托管index.yaml文件和打包好的chart文件的web服务器。当前chart仓库本身没有设置身份和权限验证，查看此链接获取该问题的最新进展。 因为chart仓库只是一个HTTP服务，通过HTTP GET获取YAML文件和chart的压缩包，所以可以将这些文件存储在web服务器中，例如GCS、Amazon S3、GitHub Pages等。 关于chart仓库的更多信息请参考Helm chart文档。 使用GitHub pages托管charts 我们在上文中说到，chart可以使用GitHub pages做存储，接下来我们将会把之前够够构建的chart上传到GitHub pages并在helm中新增一个repo。 构建Monocular UI 参考 Monocular UI 构建UI。 克隆项目到本地 git clone https://github.com/kubernetes-helm/monocular.git 依赖环境 Angular 2 angular/cli Typescript Sass Webpack Bootstrap 在monoclar/scr/ui目录下执行以下命令安装依赖： yarn install npm install -g @angular/cli npm install -g typescript npm install -g webpack 运行 使用docker-compose 最简单的运行方式使用docker-compose： docker-comopse up 该命令需要用到如下镜像： bitnami/mongodb:3 bitnami/node:8 quay.io/deis/go-dev:v1.5.0 会有一个很长的build过程，构建失败。 使用helm 首先需要已在本地安装了helm，并在kubernetes集群中安装了tiller，见使用helm管理kubernetes应用。 # 需要安装nginx ingress $ helm install stable/nginx-ingress $ helm repo add monocular https://kubernetes-helm.github.io/monocular $ helm install monocular/monocular 图片 - Helm monocular界面 因为nginx ingress配置问题，官方的chart中api与ui使用的是同样的domain name，我使用的是traefik ingress，api访问不到，所以加载不了chart。 参考 Monocular UI Helm Chart - GitHub 简化Kubernetes应用部署工具-Helm之应用部署 Speed deployment on Kubernetes with Helm Chart – Quick YAML example from scratch Using a private github repo as helm chart repo (https access) Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-15 18:47:40 "},"practice/ci-cd.html":{"url":"practice/ci-cd.html","title":"持续集成与发布","keywords":"","body":"持续集成与发布 持续集成与发布，简称CI/CD，是微服务构建的重要环节，也是DevOps中推崇的方法论。如何在kubernetes中使用持续构建与发布工具？可以既可以与企业内部原有的持续构建集成，例如Jenkins，也可以在kubernetes中部署一套新的持续构建与发布工具，例如Drone。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-10-27 11:40:14 "},"practice/jenkins-ci-cd.html":{"url":"practice/jenkins-ci-cd.html","title":"使用Jenkins进行持续集成与发布","keywords":"","body":"使用Jenkins进行持续集成与发布 我们基于Jenkins的CI/CD流程如下所示。 图片 - 基于Jenkins的持续集成与发布 流程说明 应用构建和发布流程说明。 用户向Gitlab提交代码，代码中必须包含Dockerfile 将代码提交到远程仓库 用户在发布应用时需要填写git仓库地址和分支、服务类型、服务名称、资源数量、实例个数，确定后触发Jenkins自动构建 Jenkins的CI流水线自动编译代码并打包成docker镜像推送到Harbor镜像仓库 Jenkins的CI流水线中包括了自定义脚本，根据我们已准备好的kubernetes的YAML模板，将其中的变量替换成用户输入的选项 生成应用的kubernetes YAML配置文件 更新Ingress的配置，根据新部署的应用的名称，在ingress的配置文件中增加一条路由信息 更新PowerDNS，向其中插入一条DNS记录，IP地址是边缘节点的IP地址。关于边缘节点，请查看边缘节点配置 Jenkins调用kubernetes的API，部署应用 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-08-21 18:23:35 "},"practice/drone-ci-cd.html":{"url":"practice/drone-ci-cd.html","title":"使用Drone进行持续集成与发布","keywords":"","body":"使用Drone进行持续构建与发布 Drone是一个用Go语言开发的基于容器运行的持续集成软件。 配置GitHub 使用Drone对GitHub上的代码进行持续构建与发布，需要首先在GitHub上设置一个OAuth，如下： 1. 在Github上创建一个新的OAtuh应用 访问https://github.com/settings/applications/new，创建新的OAuth应用。 图片 - OAuth注册 填写应用程序的地址，因为是在本地与行，所以我们都填http://localhost。 2. 获取OAtuh Client ID和Client Secret 在注册完成后就可以获得如下图所示的OAuth Client ID和Client Secret，保存下来，我们后面要用到。 图片 - OAuth key 使用docker-compose单机运行 我们在本地环境，使用docker-compose，按照Drone官方安装文档安装配置Drone。 我们将代码托管在Github上，需要Drone可以持续集成和发布Github的代码，因此需要修改docker-compoe.yaml文件中的GitHub配置。 version: '2' services: drone-server: image: drone/drone:0.7 ports: - 80:8000 volumes: - /var/lib/drone:/var/lib/drone/ restart: always environment: - DRONE_OPEN=true - DRONE_ADMIN=jimmy - DRONE_HOST=localhost - DRONE_GITHUB=true - DRONE_GITHUB_CLIENT=${DRONE_GITHUB_CLIENT} - DRONE_GITHUB_SECRET=${DRONE_GITHUB_SECRET} - DRONE_SECRET=admin drone-agent: image: drone/drone:0.7 command: agent restart: always depends_on: - drone-server volumes: - /var/run/docker.sock:/var/run/docker.sock environment: - DRONE_SERVER=ws://drone-server:8000/ws/broker - DRONE_SECRET=admin /var/lib/drone是在本地挂载的目录，请确保该目录已存在，且可以被docker访问到，Mac下可以在docker的共享目录中配置。 DRONE_SECRET可以是一个随机的字符串，要确保drone-server与drone-client的DRONE_SECRET相同。 DRONE_GITHUB_CLIENT和DRONE_GITHUB_SECRET即在前面申请的OAuth的Client ID和Client Secret。 启动Drone 使用下面的命令在本地启动drone： docker-compose up 这样是在前台启动，加上-d参数就可以在后台启动。 访问http://localhost可以看到登陆画面。 图片 - Drone登陆界面 授权后可以看到GitHub repo设置。 图片 - Github启用repo设置 图片 - Github单个repo设置 参考 Drone Installation Github - Drone Drone 搭配 Kubernetes 升級應用程式版本 - blog.wu-boy.com Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-03 19:08:34 "},"practice/update-and-upgrade.html":{"url":"practice/update-and-upgrade.html","title":"更新与升级","keywords":"","body":"更新与升级 Kubernetes到目前为止基本保持三个月发行一个新版本的节奏，更新节奏可以说非常快，这一部分将主要跟踪kubernetes及其相关组件的更新与升级。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-02 16:29:33 "},"practice/manually-upgrade.html":{"url":"practice/manually-upgrade.html","title":"手动升级kubernetes集群","keywords":"","body":"手动升级kubernetes集群 在我最开始写作本书的时候，kubernetes刚发布1.6.0版本，而kubernetes基本按照每三个月发布一个大版本的速度迭代，为了使用新特性和只支持新版本kubernetes的配套软件，升级kubernetes就迫在眉睫，在此我们使用替换kubernets的旧的二进制文件这种暴力的方式来升级测试集群，若升级生产集群还望三思。 另外，自kubernetes1.6版本之后发布的1.7和1.8版本又增加了一些新特性，参考： Kubernetes1.7更新日志 Kubernetes1.8更新日志 目前kubernetes的官方文档上并没有详细的手动安装的集群如何升级的参考资料，只有两篇关于kubernetes集群升级的文档。 在ubuntu上如何使用juju升级：https://kubernetes.io/docs/getting-started-guides/ubuntu/upgrades/ 使用kubeadm升级：https://kubernetes.io/docs/tasks/administer-cluster/kubeadm-upgrade-1-7/ 手动升级的还没有详细的方案，大多是基于管理工具部署和升级，比如juju、kubeadm、kops、kubespray等。 manual upgrade/downgrade testing for Kubernetes 1.6 - google group，在这个Google group中讨论了kubernetes手动升级的问题，并给出了参考建议。 升级步骤 注意：该升级步骤是实验性的，建议在测试集群上使用，无法保证线上服务不中断，实际升级完成后无需对线上服务做任何操作。 大体上的升级步骤是，先升级master节点，然后再一次升级每台node节点。 准备 备份kubernetes原先的二进制文件和配置文件。 下载最新版本的kubernetes二进制包，如1.8.5版本，查看changelog，下载二进制包，我们使用的是kubernetes-server-linux-amd64.tar.gz，分发到集群的每个节点上。 升级master节点 停止master节点的进程 systemctl stop kube-apiserver systemctl stop kube-scheduler systemctl stop kube-controller-manager systemctl stop kube-proxy systemctl stop kubelet 使用新版本的kubernetes二进制文件替换原来老版本的文件，然后启动master节点上的进程： systemctl start kube-apiserver systemctl start kube-scheduler systemctl start kube-controller-manager 因为我们的master节点同时也作为node节点，所有还要执行下面的”升级node节点“中的步骤。 升级node节点 关闭swap # 临时关闭 swapoff -a # 永久关闭，注释掉swap分区即可 vim /etc/fstab #UUID=65c9f92d-4828-4d46-bf19-fb78a38d2fd1 swap swap defaults 0 0 修改kubelet的配置文件 将kubelet的配置文件/etc/kubernetes/kublet配置文件中的KUBELET_API_SERVER=\"--api-servers=http://172.20.0.113:8080\"行注释掉。 注意：：kubernetes1.7及以上版本已经没有该配置了，API server的地址写在了kubeconfig文件中。 停止node节点上的kubernetes进程： systemctl stop kubelet systemctl stop kube-proxy 使用新版本的kubernetes二进制文件替换原来老版本的文件，然后启动node节点上的进程： systemctl start kubelet systemctl start kube-proxy 启动新版本的kube-proxy报错找不到conntrack命令，使用yum install -y conntrack-tools命令安装后重启kube-proxy即可。 检查 到此升级完成，在master节点上检查节点状态： NAME STATUS ROLES AGE VERSION 172.20.0.113 Ready 244d v1.8.5 172.20.0.114 Ready 244d v1.8.5 172.20.0.115 Ready 244d v1.8.5 所有节点的状态都正常，再检查下原先的运行在kubernetes之上的服务是否正常，如果服务正常的话说明这次升级无误。 参考 Cluster Upgrade #2524 Upgrading self-hosted Kubernetes Upgrading Kubernetes - kops Upgrading kubeadm clusters from 1.6 to 1.7 How to Upgrade a Kubernetes Cluster With No Downtime manual upgrade/downgrade testing for Kubernetes 1.6 - google group Notes/Instructions for Manual Upgrade Testing1.5 -> 1.6 Upgrading Kubernetes in Kubespray Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-13 13:02:14 "},"practice/dashboard-upgrade.html":{"url":"practice/dashboard-upgrade.html","title":"升级dashboard","keywords":"","body":"升级Dashboard 我们在kubernetes1.6的时候同时安装了dashboard插件，该插件也是基于kubernetes1.6版本开发的。如今kubernetes1.8版本业已发布，如何升级dashboard以获取新版中功能呢？ Dashboard的升级比较简单，因为它仅仅是一个前端应用，用来展现集群信息和与后端API交互，理论上只需要更新原先dashboard的yaml配置文件中的镜像就可以了，但是为了使用dashboard1.7版本中的用户登陆功能，还需要做一些额外的操作。 dashboard的更新日志请见release note，当前的最新版本为v1.7.1，下面将介绍将dashboard从v1.6.3升级到v1.7.1并开启用户登陆认证的详细步骤。 升级步骤 删除原来的版本 首先删除原来的dashboard资源： kubectl delete -f dashboard/ 将dashboard目录下的所有yaml文件中的资源全部删除，包括Deployment、service和角色绑定等。 部署新版本 我们使用官方的配置文件来安装，首先下载官方配置： wget https://raw.githubusercontent.com/kubernetes/dashboard/master/src/deploy/recommended/kubernetes-dashboard.yaml 修改其中的两个镜像地址为我们的私有地址。 gcr.io/google_containers/kubernetes-dashboard-init-amd64:v1.0.1 gcr.io/google_containers/kubernetes-dashboard-amd64:v1.7.1 这个两个镜像可以同时从时速云上获取： index.tenxcloud.com/jimmy/kubernetes-dashboard-amd64:v1.7.1 index.tenxcloud.com/jimmy/kubernetes-dashboard-init-amd64:v1.0.1 将service type设置为NodePort，修改后的yaml文件见kubernetes-dashboard.yaml，然后就可以部署新版本的dashboard了。 kubectl create -f kubernetes-dashboard.yaml 获取dashboard的外网访问端口： kubectl -n kube-system get svc kubernetes-dashboard NAME CLUSTER-IP EXTERNAL-IP PORT(S) AGE kubernetes-dashboard 10.254.177.181 443:32324/TCP 49m 访问集群中的任何一个节点，即可打开dashboard登陆页面，如https://172.20.0.113:32324/（请使用https访问），支持使用`kubeconfig`和`token`两种的认证方式： 图片 - 登陆界面 选择本地的kubeconfig文件以登陆集群，kubeconfig文件中包括登陆的用户名、证书和token信息。 登陆之后首先看到的界面是这样的： 图片 - 首页 这是因为该用户没有对default命名空间的访问权限。 修改URL地址中的namespace字段为该用户有权限访问的命名空间如brand：https://172.20.0.113:32324/#!/overview?namespace=brand： 图片 - 用户空间 设置界面的语言 我们看到现在 dashboard 的页面都已经被汉化了，当前支持英文、中文简体、中文繁体、日语，根据浏览器的语言自动切换的。如果想要强制设置 dashboard 中显示的语言，需要在 dahsboard 的 Deployment yaml 配置中增加如下配置： env: - name: ACCEPT_LANGUAGE value: english 关于 i18n 的设计文档请参考：https://github.com/kubernetes/dashboard/blob/master/docs/design/i18n.md 更简单的方式是，如果您使用的Chrome浏览器，则在浏览器中的配置中设置语言的顺序后刷新网页，dashboard将以您在Chrome中配置的首选语言显示。 身份认证 登陆 dashboard 的时候支持 kubeconfig 和 token 两种认证方式，kubeconfig 中也依赖 token 字段，所以生成 token 这一步是必不可少的。 下文分两块来讲解两种登陆认证方式： 为 brand 命名空间下的 brand 用户创建 kubeconfig 文件 为集群的管理员（拥有所有命名空间的 amdin 权限）创建 token 使用 kubeconfig 登陆dashboard的时候可以指定kubeconfig文件来认证用户权限，如何生成登陆dashboard时指定的kubeconfig文件请参考创建用户认证授权的kubeconfig文件。 注意我们生成的 kubeconfig 文件中没有 token 字段，需要手动添加该字段。 比如我们为 brand namespace 下的 brand 用户生成了名为 brand.kubeconfig 的 kubeconfig 文件，还要再该文件中追加一行 token 的配置（如何生成 token 将在下文介绍），如下所示： 图片 - kubeconfig文件 这样就可以使用brand.kubeconfig文件来登陆dashboard了，而且只能访问和操作brand命名空间下的对象。 生成 token 需要创建一个admin用户并授予admin角色绑定，使用下面的yaml文件创建admin用户并赋予他管理员权限，然后可以通过token登陆dashbaord，该文件见admin-role.yaml。这种认证方式本质上是通过 Service Account 的身份认证加上 Bearer token 请求 API server 的方式实现，参考 Kubernetes 中的认证。 kind: ClusterRoleBinding apiVersion: rbac.authorization.k8s.io/v1beta1 metadata: name: admin annotations: rbac.authorization.kubernetes.io/autoupdate: \"true\" roleRef: kind: ClusterRole name: cluster-admin apiGroup: rbac.authorization.k8s.io subjects: - kind: ServiceAccount name: admin namespace: kube-system --- apiVersion: v1 kind: ServiceAccount metadata: name: admin namespace: kube-system labels: kubernetes.io/cluster-service: \"true\" addonmanager.kubernetes.io/mode: Reconcile 然后执行下面的命令创建 serviceaccount 和角色绑定，对于其他命名空间的其他用户只要修改上述 yaml 中的 name 和 namespace 字段即可： kubectl create -f admin-role.yaml 创建完成后获取secret和token的值。 # 获取admin-token的secret名字 $ kubectl -n kube-system get secret|grep admin-token admin-token-nwphb kubernetes.io/service-account-token 3 6m # 获取token的值 $ kubectl -n kube-system describe secret admin-token-nwphb Name: admin-token-nwphb Namespace: kube-system Labels: Annotations: kubernetes.io/service-account.name=admin kubernetes.io/service-account.uid=f37bd044-bfb3-11e7-87c0-f4e9d49f8ed0 Type: kubernetes.io/service-account-token Data ==== namespace: 11 bytes token: 非常长的字符串 ca.crt: 1310 bytes 在 dashboard 登录页面上使用上面输出中的那个非常长的字符串进行 base64 解码后作为 token 登录，即可以拥有管理员权限操作整个kubernetes集群中的对象。当然您也可以将这串 token 进行 base64 解码后，加到 admin 用户的kubeconfig文件中，继续使用kubeconfig登录，两种认证方式任您选择。 注意：一定要将 kubectl 的输出中的 token 值进行 base64 解码，在线解码工具 base64decode，Linux 和 Mac 有自带的 base64 命令也可以直接使用，输入 base64 是进行编码，Linux 中base64 -d 表示解码，Mac 中使用 base64 -D。 也可以使用 jsonpath 的方式直接获取 token 的值，如： kubectl -n kube-system get secret admin-token-nwphb -o jsonpath={.data.token}|base64 -d 注意我们使用了 base64 对其重新解码，因为 secret 都是经过 base64 编码的，如果直接使用 kubectl 中查看到的 token 值会认证失败，详见 secret 配置。关于 JSONPath 的使用请参考 JSONPath 手册。 参考 Dashboard log in mechanism #2093 Accessing Dashboard 1.7.X and above Kubernetes dashboard UX for Role-Based Access Control How to sign in kubernetes dashboard? - StackOverflow JSONPath 手册 Kubernetes 中的认证 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-13 16:28:52 "},"usecases/":{"url":"usecases/","title":"领域应用概览","keywords":"","body":"领域应用 Kubernetes 和云原生应用在各个领域中的实践。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-15 20:05:51 "},"usecases/microservices.html":{"url":"usecases/microservices.html","title":"微服务架构","keywords":"","body":"微服务架构 Kubernetes 设计之初就是按照 Cloud Native 的理念设计的，Cloud Native 中有个重要概念就是微服务的架构设计，当将单体应用拆分微服务后， 随着服务数量的增多，如何微服务进行管理以保证服务的 SLA 呢？为了从架构层面上解决这个问题，解放程序员的创造性，避免繁琐的服务发现、监控、分布式追踪等事务，Service mesh 应运而生。 微服务 下图是Bilgin Ibryam给出的微服务中应该关心的主题，图片来自RedHat Developers。 图片 - 微服务关注的部分 当前最成熟最完整的微服务框架可以说非Spring莫属，而Spring又仅限于Java语言开发，其架构本身又跟Kubernetes存在很多重合的部分，如何探索将Kubernetes作为微服务架构平台就成为一个热点话题。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-20 16:43:03 "},"usecases/service-discovery-in-microservices.html":{"url":"usecases/service-discovery-in-microservices.html","title":"微服务中的服务发现","keywords":"","body":"微服务中的服务发现 在单体架构时，因为服务不会经常和动态迁移，所有服务地址可以直接在配置文件中配置，所以也不会有服务发现的问题。但是对于微服务来说，应用的拆分，服务之间的解耦，和服务动态扩展带来的服务迁移，服务发现就成了微服务中的一个关键问题。 服务发现分为客户端服务发现和服务端服务发现两种，架构如下图所示。 图片 - 微服务中的服务发现 这两种架构都各有利弊，我们拿客户端服务发现软件Eureka和服务端服务发现架构Kubernetes/SkyDNS+Ingress LB+Traefik+PowerDNS为例说明。 服务发现方案 Pros Cons Eureka 使用简单，适用于java语言开发的项目，比服务端服务发现少一次网络跳转 对非Java语言的支持不够好，Consumer需要内置特定的服务发现客户端和发现逻辑 Kubernetes Consumer无需关注服务发现具体细节，只需知道服务的DNS域名即可，支持异构语言开发 需要基础设施支撑，多了一次网络跳转，可能有性能损失 Eureka 也不是单独使用的，一般会配合 ribbon 一起使用，ribbon 作为路由和负载均衡。 Ribbon提供一组丰富的功能集： 多种内建的负载均衡规则： Round-robin 轮询负载均衡 平均加权响应时间负载均衡 随机负载均衡 可用性过滤负载均衡（避免跳闸线路和高并发链接数） 自定义负载均衡插件系统 与服务发现解决方案的可拔插集成（包括Eureka） 云原生智能，例如可用区亲和性和不健康区规避 内建的故障恢复能力 参考 谈服务发现的背景、架构以及落地方案 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-09-16 16:38:32 "},"usecases/service-mesh.html":{"url":"usecases/service-mesh.html","title":"Service Mesh 服务网格","keywords":"","body":"Service Mesh 服务网格 Service mesh 又译作 ”服务网格“，作为服务间通信的基础设施层。Buoyant 公司的 CEO Willian Morgan 在他的这篇文章 WHAT’S A SERVICE MESH? AND WHY DO I NEED ONE? 中解释了什么是 Service Mesh，为什么云原生应用需要 Service Mesh。 A service mesh is a dedicated infrastructure layer for handling service-to-service communication. It’s responsible for the reliable delivery of requests through the complex topology of services that comprise a modern, cloud native application. In practice, the service mesh is typically implemented as an array of lightweight network proxies that are deployed alongside application code, without the application needing to be aware. 今年来以 Istio 和 Linkderd 为代表的 Service Mesh 蓬勃发展，大有成为下一代语言异构微服务架构的王者之范，今天又碰巧看到了 Red Hat 的 Burr Sutter 提出了8 Steps to Becoming Awesome with Kubernetes，整个PPT一共60多页，很有建设性，点此跳转到我的GitHub上下载，我将其归档到cloud-native-slides-share中了。 图片 - 下一代异构微服务架构 自我6月份初接触Istio依赖就发觉service mesh很好的解决了异构语言中的很多问题，而且是kuberentes service 上层不可或缺的服务间代理。关于istio的更多内容请参考 istio中文文档。 什么是 service mesh？ Service mesh 有如下几个特点： 应用程序间通讯的中间层 轻量级网络代理 应用程序无感知 解耦应用程序的重试/超时、监控、追踪和服务发现 目前两款流行的 service mesh 开源软件 Istio 和 Linkerd 都可以直接在 kubernetes 中集成，其中 Linkerd 已经成为 CNCF 成员。 理解 Service Mesh 如果用一句话来解释什么是 Service Mesh，可以将它比作是应用程序或者说微服务间的 TCP/IP，负责服务之间的网络调用、限流、熔断和监控。对于编写应用程序来说一般无须关心 TCP/IP 这一层（比如通过 HTTP 协议的 RESTful 应用），同样使用 Service Mesh 也就无须关系服务之间的那些原来是通过应用程序或者其他框架实现的事情，比如 Spring Cloud、OSS，现在只要交给 Service Mesh 就可以了。 Phil Calçado 在他的这篇博客 Pattern: Service Mesh 中详细解释了 Service Mesh 的来龙去脉： 从最原始的主机之间直接使用网线相连 网络层的出现 集成到应用程序内部的控制流 分解到应用程序外部的控制流 应用程序的中集成服务发现和断路器 出现了专门用于服务发现和断路器的软件包/库，如 Twitter 的 Finagle 和 Facebook 的 Proxygen，这时候还是集成在应用程序内部 出现了专门用于服务发现和断路器的开源软件，如 Netflix OSS、Airbnb 的 synapse 和 nerve 最后作为微服务的中间层 service mesh 出现 Service mesh 的架构如下图所示： 图片 - Service Mesh 架构图 图片来自：Pattern: Service Mesh Service mesh 作为 sidecar 运行，对应用程序来说是透明，所有应用程序间的流量都会通过它，所以对应用程序流量的控制都可以在 serivce mesh 中实现。 Service mesh如何工作？ 下面以 Linkerd 为例讲解 service mesh 如何工作，Istio 作为 service mesh 的另一种实现原理与 linkerd 基本类似，后续文章将会详解 Istio 和 Linkerd 如何在 kubernetes 中工作。 Linkerd 将服务请求路由到目的地址，根据中的参数判断是到生产环境、测试环境还是 staging 环境中的服务（服务可能同时部署在这三个环境中），是路由到本地环境还是公有云环境？所有的这些路由信息可以动态配置，可以是全局配置也可以为某些服务单独配置。 当 Linkerd 确认了目的地址后，将流量发送到相应服务发现端点，在 kubernetes 中是 service，然后 service 会将服务转发给后端的实例。 Linkerd 根据它观测到最近请求的延迟时间，选择出所有应用程序的实例中响应最快的实例。 Linkerd 将请求发送给该实例，同时记录响应类型和延迟数据。 如果该实例挂了、不响应了或者进程不工作了，Linkerd 将把请求发送到其他实例上重试。 如果该实例持续返回 error，Linkerd 会将该实例从负载均衡池中移除，稍后再周期性得重试。 如果请求的截止时间已过，Linkerd 主动失败该请求，而不是再次尝试添加负载。 Linkerd 以 metric 和分布式追踪的形式捕获上述行为的各个方面，这些追踪信息将发送到集中 metric 系统。 为何使用 service mesh？ Service mesh 并没有给我们带来新功能，它是用于解决其他工具已经解决过的问题，只不过这次是在 Cloud Native 的 kubernetes 环境下的实现。 在传统的 MVC 三层 Web 应用程序架构下，服务之间的通讯并不复杂，在应用程序内部自己管理即可，但是在现今的复杂的大型网站情况下，单体应用被分解为众多的微服务，服务之间的依赖和通讯十分复杂，出现了 twitter 开发的 Finagle、Netflix 开发的 Hystrix 和 Google 的 Stubby 这样的 ”胖客户端“ 库，这些就是早期的 service mesh，但是它们都近适用于特定的环境和特定的开发语言，并不能作为平台级的 service mesh 支持。 在 Cloud Native 架构下，容器的使用给予了异构应用程序的更多可行性，kubernetes 增强的应用的横向扩容能力，用户可以快速的编排出复杂环境、复杂依赖关系的应用程序，同时开发者又无须过分关心应用程序的监控、扩展性、服务发现和分布式追踪这些繁琐的事情而专注于程序开发，赋予开发者更多的创造性。 Istio VS Linkerd 当前的Service Mesh实现主要有两大阵营，要给是Linkerd（也是最初提出该概念的），另一个是Istio，当然还有很多其他号称也是Service Mesh，比如Nginx出品的Nginmesh。 Feature Istio Linkerd 部署架构 Envoy/Sidecar DaemonSets 易用性 复杂 简单 支持平台 kuberentes kubernetes/mesos/Istio/local 当前版本 0.3.0 1.3.3 是否已有生产部署 否 是 下图是Istio和Linkerd架构的不同，Istio是使用Sidecar模式，将Envoy植入到Pod中，而Linkerd则是在每台node上都以DaemonSet的方式运行。 图片 - Istio vs linkerd 关于Istio和Linkerd的详细信息请参考 安装并试用Istio service mesh 与 Linkerd 使用指南。 另外出品Linkerd的公司buoyant又推出了conduit，这是一种更轻量级的Service Mesh。 参考 WHAT’S A SERVICE MESH? AND WHY DO I NEED ONE? So what even is a Service Mesh? Hot take on Istio and Linkerd linkerd: A service mesh for AWS ECS Introducing Istio: A robust service mesh for microservices Application Network Functions With ESBs, API Management, and Now.. Service Mesh? Pattern: Service Mesh Istio官方文档中文版 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-07 20:09:18 "},"usecases/istio.html":{"url":"usecases/istio.html","title":"Istio","keywords":"","body":"Istio简介 前言 Istio是由Google、IBM和Lyft开源的微服务管理、保护和监控框架。Istio为希腊语，意思是”起航“。 简介 使用istio可以很简单的创建具有负载均衡、服务间认证、监控等功能的服务网络，而不需要对服务的代码进行任何修改。你只需要在部署环境中，例如Kubernetes的pod里注入一个特别的sidecar proxy来增加对istio的支持，用来截获微服务之间的网络流量。 目前版本的istio只支持kubernetes，未来计划支持其他其他环境。 另外，Istio的前身是IBM开源的Amalgam8，追本溯源，我们来看下它的特性。 Amalgam8 Amalgam8的网站上说，它是一个Content-based Routing Fabric for Polyglot Microservices，简单、强大且开源。 Amalgam8是一款基于内容和版本的路由布局，用于集成多语言异构体微服务。 其control plane API可用于动态编程规则，用于在正在运行的应用程序中跨微服务进行路由和操作请求。 以内容/版本感知方式路由请求的能力简化了DevOps任务，如金丝雀和红/黑发布，A/B Test和系统地测试弹性微服务。 可以使用Amalgam8平台与受欢迎的容器运行时（如Docker，Kubernetes，Marathon / Mesos）或其他云计算提供商（如IBM Bluemix，Google Cloud Platform或Amazon AWS）。 特性 使用istio的进行微服务管理有如下特性： 流量管理：控制服务间的流量和API调用流，使调用更可靠，增强不同环境下的网络鲁棒性。 可观测性：了解服务之间的依赖关系和它们之间的性质和流量，提供快速识别定位问题的能力。 策略实施：通过配置mesh而不是以改变代码的方式来控制服务之间的访问策略。 服务识别和安全：提供在mesh里的服务可识别性和安全性保护。 未来将支持多种平台，不论是kubernetes、Mesos、还是云。同时可以集成已有的ACL、日志、监控、配额、审计等。 架构 图片 - Istio架构图 Istio架构分为控制层和数据层。 数据层：由一组智能代理（Envoy）作为sidecar部署，协调和控制所有microservices之间的网络通信。 控制层：负责管理和配置代理路由流量，以及在运行时执行的政策。 Envoy Istio使用Envoy代理的扩展版本，该代理是以C++开发的高性能代理，用于调解service mesh中所有服务的所有入站和出站流量。 Istio利用了Envoy的许多内置功能，例如动态服务发现，负载平衡，TLS终止，HTTP/2＆gRPC代理，断路器，运行状况检查，基于百分比的流量拆分分阶段上线，故障注入和丰富指标。 Envoy在kubernetes中作为pod的sidecar来部署。 这允许Istio将大量关于流量行为的信号作为属性提取出来，这些属性又可以在Mixer中用于执行策略决策，并发送给监控系统以提供有关整个mesh的行为的信息。 Sidecar代理模型还允许你将Istio功能添加到现有部署中，无需重新构建或重写代码。 更多信息参见设计目标。 Mixer Mixer负责在service mesh上执行访问控制和使用策略，并收集Envoy代理和其他服务的遥测数据。代理提取请求级属性，发送到mixer进行评估。有关此属性提取和策略评估的更多信息，请参见Mixer配置。 混音器包括一个灵活的插件模型，使其能够与各种主机环境和基础架构后端进行接口，从这些细节中抽象出Envoy代理和Istio管理的服务。 Istio Manager Istio-Manager用作用户和Istio之间的接口，收集和验证配置，并将其传播到各种Istio组件。它从Mixer和Envoy中抽取环境特定的实现细节，为他们提供独立于底层平台的用户服务的抽象表示。 此外，流量管理规则（即通用4层规则和七层HTTP/gRPC路由规则）可以在运行时通过Istio-Manager进行编程。 Istio-auth Istio-Auth提供强大的服务间和最终用户认证，使用相互TLS，内置身份和凭据管理。它可用于升级service mesh中的未加密流量，并为运营商提供基于服务身份而不是网络控制的策略的能力。 Istio的未来版本将增加细粒度的访问控制和审计，以使用各种访问控制机制（包括属性和基于角色的访问控制以及授权hook）来控制和监控访问你服务、API或资源的人员。 参考 Istio开源平台发布，Google、IBM和Lyft分别承担什么角色？ Istio：用于微服务的服务啮合层 Istio Overview Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-08-21 18:23:35 "},"usecases/istio-installation.html":{"url":"usecases/istio-installation.html","title":"安装并试用Istio service mesh","keywords":"","body":"安装并试用Istio service mesh 本文根据官网的文档整理而成，步骤包括安装istio 0.1.5并创建一个bookinfo的微服务来测试istio的功能。 文中使用的yaml文件可以在kubernetes-handbook的manifests/istio目录中找到，所有的镜像都换成了我的私有镜像仓库地址，请根据官网的镜像自行修改。 安装环境 CentOS 7.3.1611 Docker 1.12.6 Kubernetes 1.6.0 安装 1.下载安装包 下载地址：https://github.com/istio/istio/releases 下载Linux版本的当前最新版安装包 wget https://github.com/istio/istio/releases/download/0.1.5/istio-0.1.5-linux.tar.gz 2.解压 解压后，得到的目录结构如下： . ├── bin │ └── istioctl ├── install │ └── kubernetes │ ├── addons │ │ ├── grafana.yaml │ │ ├── prometheus.yaml │ │ ├── servicegraph.yaml │ │ └── zipkin.yaml │ ├── istio-auth.yaml │ ├── istio-rbac-alpha.yaml │ ├── istio-rbac-beta.yaml │ ├── istio.yaml │ ├── README.md │ └── templates │ ├── istio-auth │ │ ├── istio-auth-with-cluster-ca.yaml │ │ ├── istio-cluster-ca.yaml │ │ ├── istio-egress-auth.yaml │ │ ├── istio-ingress-auth.yaml │ │ └── istio-namespace-ca.yaml │ ├── istio-egress.yaml │ ├── istio-ingress.yaml │ ├── istio-manager.yaml │ └── istio-mixer.yaml ├── istio.VERSION ├── LICENSE └── samples ├── apps │ ├── bookinfo │ │ ├── bookinfo.yaml │ │ ├── cleanup.sh │ │ ├── destination-ratings-test-delay.yaml │ │ ├── loadbalancing-policy-reviews.yaml │ │ ├── mixer-rule-additional-telemetry.yaml │ │ ├── mixer-rule-empty-rule.yaml │ │ ├── mixer-rule-ratings-denial.yaml │ │ ├── mixer-rule-ratings-ratelimit.yaml │ │ ├── README.md │ │ ├── route-rule-all-v1.yaml │ │ ├── route-rule-delay.yaml │ │ ├── route-rule-reviews-50-v3.yaml │ │ ├── route-rule-reviews-test-v2.yaml │ │ ├── route-rule-reviews-v2-v3.yaml │ │ └── route-rule-reviews-v3.yaml │ ├── httpbin │ │ ├── httpbin.yaml │ │ └── README.md │ └── sleep │ ├── README.md │ └── sleep.yaml └── README.md 11 directories, 41 files 从文件里表中可以看到，安装包中包括了kubernetes的yaml文件，示例应用和安装模板。 3.安装istioctl 将./bin/istioctl拷贝到你的$PATH目录下。 4.检查RBAC 因为我们安装的kuberentes版本是1.6.0默认支持RBAC，这一步可以跳过。如果你使用的其他版本的kubernetes，请参考官方文档操作。 执行以下命令，正确的输出是这样的： $ kubectl api-versions | grep rbac rbac.authorization.k8s.io/v1alpha1 rbac.authorization.k8s.io/v1beta1 5.创建角色绑定 $ kubectl create -f install/kubernetes/istio-rbac-beta.yaml clusterrole \"istio-manager\" created clusterrole \"istio-ca\" created clusterrole \"istio-sidecar\" created clusterrolebinding \"istio-manager-admin-role-binding\" created clusterrolebinding \"istio-ca-role-binding\" created clusterrolebinding \"istio-ingress-admin-role-binding\" created clusterrolebinding \"istio-sidecar-role-binding\" created 注意：官网的安装包中的该文件中存在RoleBinding错误，应该是集群级别的clusterrolebinding，而release里的代码只是普通的rolebinding，查看该Issue Istio manager cannot list of create k8s TPR when RBAC enabled #327。 6.安装istio核心组件 用到的镜像有： docker.io/istio/mixer:0.1.5 docker.io/istio/manager:0.1.5 docker.io/istio/proxy_debug:0.1.5 我们暂时不开启Istio Auth。 本文中用到的所有yaml文件中的type: LoadBalancer去掉，使用默认的ClusterIP，然后配置Traefik ingress，就可以在集群外部访问。请参考安装Traefik ingress。 kubectl apply -f install/kubernetes/istio.yaml 7.安装监控插件 用到的镜像有： docker.io/istio/grafana:0.1.5 quay.io/coreos/prometheus:v1.1.1 gcr.io/istio-testing/servicegraph:latest docker.io/openzipkin/zipkin:latest 为了方便下载，其中两个镜像我备份到了时速云： index.tenxcloud.com/jimmy/prometheus:v1.1.1 index.tenxcloud.com/jimmy/servicegraph:latest 安装插件 kubectl apply -f install/kubernetes/addons/prometheus.yaml kubectl apply -f install/kubernetes/addons/grafana.yaml kubectl apply -f install/kubernetes/addons/servicegraph.yaml kubectl apply -f install/kubernetes/addons/zipkin.yaml 在traefik ingress中增加增加以上几个服务的配置，同时增加istio-ingress配置。 - host: grafana.istio.io http: paths: - path: / backend: serviceName: grafana servicePort: 3000 - host: servicegraph.istio.io http: paths: - path: / backend: serviceName: servicegraph servicePort: 8088 - host: prometheus.istio.io http: paths: - path: / backend: serviceName: prometheus servicePort: 9090 - host: zipkin.istio.io http: paths: - path: / backend: serviceName: zipkin servicePort: 9411 - host: ingress.istio.io http: paths: - path: / backend: serviceName: istio-ingress servicePort: 80 测试 我们使用Istio提供的测试应用bookinfo微服务来进行测试。 该微服务用到的镜像有： istio/examples-bookinfo-details-v1 istio/examples-bookinfo-ratings-v1 istio/examples-bookinfo-reviews-v1 istio/examples-bookinfo-reviews-v2 istio/examples-bookinfo-reviews-v3 istio/examples-bookinfo-productpage-v1 该应用架构图如下： 图片 - BookInfo Sample应用架构图 部署应用 kubectl create -f Istio kube-inject命令会在bookinfo.yaml文件中增加Envoy sidecar信息。参考：https://istio.io/docs/reference/commands/istioctl.html#istioctl-kube-inject 在本机的/etc/hosts下增加VIP节点和ingress.istio.io的对应信息。具体步骤参考：边缘节点配置 在浏览器中访问http://ingress.istio.io/productpage 图片 - BookInfo Sample页面 多次刷新页面，你会发现有的页面上的评论里有星级打分，有的页面就没有，这是因为我们部署了三个版本的应用，有的应用里包含了评分，有的没有。Istio根据默认策略随机将流量分配到三个版本的应用上。 查看部署的bookinfo应用中的productpage-v1 service和deployment，查看productpage-v1的pod的详细json信息可以看到这样的结构： $ kubectl get productpage-v1-944450470-bd530 -o json 见productpage-v1-istio.json文件。从详细输出中可以看到这个Pod中实际有两个容器，这里面包括了initContainer，作为istio植入到kubernetes deployment中的sidecar。 \"initContainers\": [ { \"args\": [ \"-p\", \"15001\", \"-u\", \"1337\" ], \"image\": \"docker.io/istio/init:0.1\", \"imagePullPolicy\": \"Always\", \"name\": \"init\", \"resources\": {}, \"securityContext\": { \"capabilities\": { \"add\": [ \"NET_ADMIN\" ] } }, \"terminationMessagePath\": \"/dev/termination-log\", \"terminationMessagePolicy\": \"File\", \"volumeMounts\": [ { \"mountPath\": \"/var/run/secrets/kubernetes.io/serviceaccount\", \"name\": \"default-token-3l9f0\", \"readOnly\": true } ] }, { \"args\": [ \"-c\", \"sysctl -w kernel.core_pattern=/tmp/core.%e.%p.%t \\u0026\\u0026 ulimit -c unlimited\" ], \"command\": [ \"/bin/sh\" ], \"image\": \"alpine\", \"imagePullPolicy\": \"Always\", \"name\": \"enable-core-dump\", \"resources\": {}, \"securityContext\": { \"privileged\": true }, \"terminationMessagePath\": \"/dev/termination-log\", \"terminationMessagePolicy\": \"File\", \"volumeMounts\": [ { \"mountPath\": \"/var/run/secrets/kubernetes.io/serviceaccount\", \"name\": \"default-token-3l9f0\", \"readOnly\": true } ] } ], 监控 不断刷新productpage页面，将可以在以下几个监控中看到如下界面。 Grafana页面 http://grafana.istio.io 图片 - Istio Grafana界面 Prometheus页面 http://prometheus.istio.io 图片 - Prometheus页面 Zipkin页面 http://zipkin.istio.io 图片 - Zipkin页面 ServiceGraph页面 http://servicegraph.istio.io/dotviz 可以用来查看服务间的依赖关系。 访问 http://servicegraph.istio.io/graph 可以获得json格式的返回结果。 图片 - ServiceGraph页面 更进一步 BookInfo示例中有三个版本的reviews，可以使用istio来配置路由请求，将流量分发到不同版本的应用上。参考Configuring Request Routing。 还有一些更高级的功能，我们后续将进一步探索。 参考 Installing Istio BookInfo sample Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-21 21:18:30 "},"usecases/configuring-request-routing.html":{"url":"usecases/configuring-request-routing.html","title":"配置请求的路由规则","keywords":"","body":"配置请求的路由规则 在上一节安装istio中我们创建BookInfo的示例，熟悉了Istio的基本功能，现在我们再来看一下istio的高级特性——配置请求的路由规则。 使用istio我们可以根据权重和HTTP headers来动态配置请求路由。 基于内容的路由 因为BookInfo示例部署了3个版本的评论微服务，我们需要设置一个默认路由。 否则，当你多次访问应用程序时，会注意到有时输出包含星级，有时候又没有。 这是因为没有明确的默认版本集，Istio将以随机方式将请求路由到服务的所有可用版本。 注意：假定您尚未设置任何路由。如果您已经为示例创建了冲突的路由规则，则需要在以下命令中使用replace而不是create。 下面这个例子能够根据网站的不同登陆用户，将流量划分到服务的不同版本和实例。跟kubernetes中的应用一样，所有的路由规则都是通过声明式的yaml配置。关于reviews:v1和reviews:v2的唯一区别是，v1没有调用评分服务，productpage页面上不会显示评分星标。 将微服务的默认版本设置成v1。 istioctl create -f samples/apps/bookinfo/route-rule-all-v1.yaml 使用以下命令查看定义的路由规则。 istioctl get route-rules -o yaml type: route-rule name: details-default namespace: default spec: destination: details.default.svc.cluster.local precedence: 1 route: - tags: version: v1 --- type: route-rule name: productpage-default namespace: default spec: destination: productpage.default.svc.cluster.local precedence: 1 route: - tags: version: v1 --- type: route-rule name: reviews-default namespace: default spec: destination: reviews.default.svc.cluster.local precedence: 1 route: - tags: version: v1 --- type: route-rule name: ratings-default namespace: default spec: destination: ratings.default.svc.cluster.local precedence: 1 route: - tags: version: v1 --- 由于对代理的规则传播是异步的，因此在尝试访问应用程序之前，需要等待几秒钟才能将规则传播到所有pod。 在浏览器中打开BookInfo URL（http://$GATEWAY_URL/productpage ，我们在上一节中使用的是 http://ingress.istio.io/productpage ）您应该会看到BookInfo应用程序的产品页面显示。 注意，产品页面上没有评分星，因为reviews:v1不访问评级服务。 将特定用户路由到reviews:v2。 为测试用户jason启用评分服务，将productpage的流量路由到reviews:v2实例上。 istioctl create -f samples/apps/bookinfo/route-rule-reviews-test-v2.yaml 确认规则生效： istioctl get route-rule reviews-test-v2 destination: reviews.default.svc.cluster.local match: httpHeaders: cookie: regex: ^(.*?;)?(user=jason)(;.*)?$ precedence: 2 route: - tags: version: v2 使用jason用户登陆productpage页面。 你可以看到每个刷新页面时，页面上都有一个1到5颗星的评级。如果你使用其他用户登陆的话，将因继续使用reviews:v1而看不到星标评分。 内部实现 在这个例子中，一开始使用istio将100%的流量发送到BookInfo服务的reviews:v1的实例上。然后又根据请求的header（例如用户的cookie）将流量选择性的发送到reviews:v2实例上。 验证了v2实例的功能后，就可以将全部用户的流量发送到v2实例上，或者逐步的迁移流量，如10%、20%直到100%。 如果你看了故障注入这一节，你会发现v2版本中有个bug，而在v3版本中修复了，你想将流量迁移到reviews:v1迁移到reviews:v3版本上，只需要运行如下命令： 将50%的流量从reviews:v1转移到reviews:v3上。 istioctl replace -f samples/apps/bookinfo/route-rule-reviews-50-v3.yaml 注意这次使用的是replace命令，而不是create，因为该rule已经在前面创建过了。 登出jason用户，或者删除测试规则，可以看到新的规则已经生效。 删除测试规则。 istioctl delete route-rule reviews-test-v2 istioctl delete route-rule ratings-test-delay 现在的规则就是刷新productpage页面，50%的概率看到红色星标的评论，50%的概率看不到星标。 注意：因为使用Envoy sidecar的实现，你需要刷新页面很多次才能看到接近规则配置的概率分布，你可以将v3的概率修改为90%，这样刷新页面时，看到红色星标的概率更高。 当v3版本的微服务稳定以后，就可以将100%的流量分摊到reviews:v3上了。 istioctl replace -f samples/apps/bookinfo/route-rule-reviews-v3.yaml 现在不论你使用什么用户登陆productpage页面，你都可以看到带红色星标评分的评论了。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-22 17:00:47 "},"usecases/install-and-expand-istio-mesh.html":{"url":"usecases/install-and-expand-istio-mesh.html","title":"安装和拓展Istio service mesh","keywords":"","body":"安装和拓展 Istio mesh 前置条件 下面的操作说明需要您可以访问 kubernetes 1.7.3 后更高版本 的集群，并且启用了 RBAC (基于角色的访问控制)。您需要安装了 1.7.3 或更高版本 的 kubectl 命令。如果您希望启用 自动注入 sidecar，您需要启用 kubernetes 集群的 alpha 功能。 注意：如果您安装了 Istio 0.1.x，在安装新版本前请先 卸载 它们（包括已启用 Istio 应用程序 Pod 中的 sidecar）。 取决于您的 kubernetes 提供商： 本地安装 Istio，安装最新版本的 Minikube (version 0.22.1 或者更高)。 Google Container Engine 使用 kubectl 获取证书 （使用您自己的集群的名字替换 ，使用集群实际所在的位置替换 ）： gcloud container clusters get-credentials --zone --project 将集群管理员权限授予当前用户（需要管理员权限才能为Istio创建必要的RBAC规则）： kubectl create clusterrolebinding cluster-admin-binding --clusterrole=cluster-admin --user=$(gcloud config get-value core/account) IBM Bluemix Container Service 使用 kubectl 获取证书 （使用您自己的集群的名字替换）： $(bx cs cluster-config |grep \"export KUBECONFIG\") Openshift Origin 3.7 或者以上版本： 默认情况下，Openshift 不允许以 UID 0运行容器。为 Istio 的入口（ingress）和出口（egress）service account 启用使用UID 0运行的容器： oc adm policy add-scc-to-user anyuid -z istio-ingress-service-account -n istio-system oc adm policy add-scc-to-user anyuid -z istio-egress-service-account -n istio-system oc adm policy add-sc-to-user anyuid -z default -n istio-system 运行应用程序 Pod 的 service account 需要特权安全性上下文限制，以此作为 sidecar 注入的一部分: oc adm policy add-scc-to-user privileged -z default -n 安装或升级 Kubernetes 命令行工具 kubectl 以匹配您的集群版本（1.7或以上版本）。 安装步骤 不论对于哪个 Istio 发行版，都安装到 istio-system namespace 下，即可以管理所有其它 namespace 下的微服务。 到 Istio release 页面上，根据您的操作系统下载对应的发行版。如果您使用的是 MacOS 或者 Linux 系统，可以使用下面的额命令自动下载和解压最新的发行版： curl -L https://git.io/getLatestIstio | sh - 解压安装文件，切换到文件所在目录。安装文件目录下包含： install/ 目录下是 kubernetes 使用的 .yaml 安装文件 samples/ 目录下是示例程序 istioctl 客户端二进制文件在 bin 目录下。istioctl 文件用户手动注入 Envoy sidecar 代理、创建路由和策略等。 istio.VERSION 配置文件 切换到 istio 包的解压目录。例如 istio-0.2.7： cd istio-0.2.7 将 istioctl 客户端二进制文件加到 PATH 中。 例如，在 MacOS 或 Linux 系统上执行下面的命令： export PATH=$PWD/bin:$PATH 安装 Istio 的核心部分。选择面两个 互斥 选项中的之一： a) 安装 Istio 的时候不启用 sidecar 之间的 TLS 双向认证： 为具有现在应用程序的集群选择该选项，使用 Istio sidecar 的服务需要能够与非 Istio Kubernetes 服务以及使用 liveliness 和 readiness 探针、headless service 和 StatefulSet 的应用程序通信。 kubectl apply -f install/kubernetes/istio.yaml 或者 b) 安装 Istio 的时候启用 sidecar 之间的 TLS 双向认证： kubectl apply -f install/kubernetes/istio-auth.yaml 这两个选项都会创建 istio-system 命名空间以及所需的 RBAC 权限，并部署 Istio-Pilot、Istio-Mixer、Istio-Ingress、Istio-Egress 和 Istio-CA（证书颁发机构）。 可选的：如果您的 kubernetes 集群开启了 alpha 功能，并想要启用 自动注入 sidecar，需要安装 Istio-Initializer： kubectl apply -f install/kubernetes/istio-initializer.yaml 验证安装 确认系列 kubernetes 服务已经部署了： istio-pilot、 istio-mixer、istio-ingress、 istio-egress： kubectl get svc -n istio-system NAME CLUSTER-IP EXTERNAL-IP PORT(S) AGE istio-egress 10.83.247.89 80/TCP 5h istio-ingress 10.83.245.171 35.184.245.62 80:32730/TCP,443:30574/TCP 5h istio-pilot 10.83.251.173 8080/TCP,8081/TCP 5h istio-mixer 10.83.244.253 9091/TCP,9094/TCP,42422/TCP 5h 注意：如果您运行的集群不支持外部负载均衡器（如 minikube）， istio-ingress 服务的 EXTERNAL-IP 显示。你必须改为使用 NodePort service 或者 端口转发方式来访问应用程序。 确认对应的 Kubernetes pod 已部署并且所有的容器都启动并运行： istio-pilot-*、 istio-mixer-*、 istio-ingress-*、 istio-egress-*、istio-ca-*， istio-initializer-* 是可以选的。 kubectl get pods -n istio-system istio-ca-3657790228-j21b9 1/1 Running 0 5h istio-egress-1684034556-fhw89 1/1 Running 0 5h istio-ingress-1842462111-j3vcs 1/1 Running 0 5h istio-initializer-184129454-zdgf5 1/1 Running 0 5h istio-pilot-2275554717-93c43 1/1 Running 0 5h istio-mixer-2104784889-20rm8 2/2 Running 0 5h 部署应用 您可以部署自己的应用或者示例应用程序如 BookInfo。 注意：应用程序必须使用 HTTP/1.1 或 HTTP/2.0 协议来传输 HTTP 流量，因为 HTTP/1.0 已经不再支持。 如果您启动了 Istio-Initializer，如上所示，您可以使用 kubectl create 直接部署应用。Istio-Initializer 会向应用程序的 pod 中自动注入 Envoy 容器： kubectl create -f .yaml 如果您没有安装 Istio-initializer 的话，您必须使用 istioctl kube-inject 命令在部署应用之前向应用程序的 pod 中手动注入 Envoy 容器： kubectl create -f .yaml) 卸载 卸载 Istio initializer: 如果您安装 Isto 的时候启用了 initializer，请卸载它： kubectl delete -f install/kubernetes/istio-initializer.yaml 卸载 Istio 核心组件。对于某一 Istio 版本，删除 RBAC 权限，istio-system namespace，和该命名空间的下的各层级资源。 不必理会在层级删除过程中的各种报错，因为这些资源可能已经被删除的。 a) 如果您在安装 Istio 的时候关闭了 TLS 双向认证： kubectl delete -f install/kubernetes/istio.yaml 或者 b) 如果您在安装 Istio 的时候启用到了 TLS 双向认证： kubectl delete -f install/kubernetes/istio-auth.yaml 安装 Istio Sidecar Pod Spec 需满足的条件 为了成为 Service Mesh 中的一部分，kubernetes 集群中的每个 Pod 都必须满足如下条件： Service 注解：每个 pod 都必须只属于某一个 Kubernetes Service （当前不支持一个 pod 同时属于多个 service）。 命名的端口：Service 的端口必须命名。端口的名字必须遵循如下格式 [-]，可以是http、http2、 grpc、 mongo、 或者 redis 作为 ，这样才能使用 Istio 的路由功能。例如name: http2-foo 和 name: http 都是有效的端口名称，而 name: http2foo 不是。如果端口的名称是不可识别的前缀或者未命名，那么该端口上的流量就会作为普通的 TCP 流量（除非使用 Protocol: UDP 明确声明使用 UDP 端口）。 带有 app label 的 Deployment：我们建议 kubernetes 的Deploymenet 资源的配置文件中为 Pod 明确指定 app label。每个Deployment 的配置中都需要有个不同的有意义的 app 标签。app label 用于在分布式坠重中添加上下文信息。 Mesh 中的每个 pod 里都有一个 Sidecar： 最后，Mesh 中的每个 pod 都必须运行与 Istio 兼容的sidecar。下面部分介绍了将 sidecar 注入到 pod 中的两种方法：使用istioctl 命令行工具手动注入，或者使用 istio initializer 自动注入。注意 sidecar 不涉及到容器间的流量，因为它们都在同一个 pod 中。 手动注入 sidecar istioctl 命令行中有一个称为 kube-inject 的便利工具，使用它可以将 Istio 的 sidecar 规范添加到 kubernetes 工作负载的规范配置中。与 Initializer 程序不同，kube-inject 只是将 YAML 规范转换成包含 Istio sidecar 的规范。您需要使用标准的工具如 kubectl 来部署修改后的 YAML。例如，以下命令将 sidecar 添加到 sleep.yaml 文件中指定的 pod 中，并将修改后的规范提交给 kubernetes： kubectl apply -f 示例 我们来试一试将 Istio sidecar 注入到 sleep 服务中去。 kubectl apply -f Kube-inject 子命令将 Istio sidecar 和 init 容器注入到 deployment 配置中，转换后的输出如下所示： ... 略过 ... --- apiVersion: extensions/v1beta1 kind: Deployment metadata: annotations: sidecar.istio.io/status: injected-version-root@69916ebba0fc-0.2.6-081ffece00c82cb9de33cd5617682999aee5298d name: sleep spec: replicas: 1 template: metadata: annotations: sidecar.istio.io/status: injected-version-root@69916ebba0fc-0.2.6-081ffece00c82cb9de33cd5617682999aee5298d labels: app: sleep spec: containers: - name: sleep image: tutum/curl command: [\"/bin/sleep\",\"infinity\"] imagePullPolicy: IfNotPresent - name: istio-proxy image: docker.io/istio/proxy_debug:0.2.6 args: ... 略过 ... initContainers: - name: istio-init image: docker.io/istio/proxy_init:0.2.6 imagePullPolicy: IfNotPresent args: ... 略过 ... --- 注入 sidecar 的关键在于 initContainers 和 istio-proxy 容器。为了简洁起见，上述输出有所省略。 验证 sleep deployment 中包含 sidecar。injected-version 对应于注入的 sidecar 镜像的版本和镜像的 TAG。在您的设置的可能会有所不同。 echo $(kubectl get deployment sleep -o jsonpath='{.metadata.annotations.sidecar\\.istio\\.io\\/status}') injected-version-9c7c291eab0a522f8033decd0f5b031f5ed0e126 你可以查看包含注入的容器和挂载的 volume 的完整 deployment 信息。 kubectl get deployment sleep -o yaml 自动注入 sidecar Istio sidecar 可以在部署之前使用 Kubernetes 中一个名为 Initializer 的 Alpha 功能自动注入到 Pod 中。 注意：Kubernetes InitializerConfiguration没有命名空间，适用于整个集群的工作负载。不要在共享测试环境中启用此功能。 前置条件 Initializer 需要在集群设置期间显示启用，如 此处 所述。 假设集群中已启用RBAC，则可以在不同环境中启用初始化程序，如下所示： GKE gcloud container clusters create NAME \\ --enable-kubernetes-alpha \\ --machine-type=n1-standard-2 \\ --num-nodes=4 \\ --no-enable-legacy-authorization \\ --zone=ZONE IBM Bluemix kubernetes v1.7.4 或更高版本的集群已默认启用 initializer。 Minikube Minikube v0.22.1 或更高版本需要为 GenericAdmissionWebhook 功能配置适当的证书。获取最新版本： https://github.com/kubernetes/minikube/releases. minikube start \\ --extra-config=apiserver.Admission.PluginNames=\"Initializers,NamespaceLifecycle,LimitRanger,ServiceAccount,DefaultStorageClass,GenericAdmissionWebhook,ResourceQuota\" \\ --kubernetes-version=v1.7.5 安装 您现在可以从 Istio 安装根目录设置 Istio Initializer。 kubectl apply -f install/kubernetes/istio-initializer.yaml 将会创建下列资源： istio-sidecar InitializerConfiguration 资源，指定 Istio sidecar 注入的资源。默认情况下 Istio sidecar 将被注入到 deployment、 statefulset、 job 和 daemonset中。 istio-inject ConfigMap，initializer 的默认注入策略，一组初始化 namespace，以及注入时使用的模版参数。这些配置的详细说明请参考 配置选项。 istio-initializer Deployment，运行 initializer 控制器。 istio-initializer-service-account ServiceAccount，用于 istio-initializer deployment。ClusterRole 和 ClusterRoleBinding 在 install/kubernetes/istio.yaml 中定义。注意所有的资源类型都需要有 initialize 和 patch 。正式处于这个原因，initializer 要作为 deployment 的一部分来运行而不是嵌入到其它控制器中，例如 istio-pilot。 验证 为了验证 sidecar 是否成功注入，为上面的 sleep 服务创建 deployment 和 service。 kubectl apply -f samples/sleep/sleep.yaml 验证 sleep deployment 中包含 sidecar。injected-version 对应于注入的 sidecar 镜像的版本和镜像的 TAG。在您的设置的可能会有所不同。 $ echo $(kubectl get deployment sleep -o jsonpath='{.metadata.annotations.sidecar\\.istio\\.io\\/status}') injected-version-9c7c291eab0a522f8033decd0f5b031f5ed0e126 你可以查看包含注入的容器和挂载的 volume 的完整 deployment 信息。 kubectl get deployment sleep -o yaml 了解发生了什么 以下是将工作负载提交给 Kubernetes 后发生的情况： 1) kubernetes 将 sidecar.initializer.istio.io 添加到工作负载的 pending initializer 列表中。 2) istio-initializer 控制器观察到有一个新的未初始化的工作负载被创建了。pending initializer 列表中的第一个个将作为 sidecar.initializer.istio.io 的名称。 3) istio-initializer 检查它是否负责初始化 namespace 中的工作负载。如果没有为该 namespace 配置 initializer，则不需要做进一步的工作，而且 initializer 会忽略工作负载。默认情况下，initializer 负责所有的 namespace（参考 配置选项）。 4) istio-initializer 将自己从 pending initializer 中移除。如果 pending initializer 列表非空，则 Kubernetes 不回结束工作负载的创建。错误配置的 initializer 意味着破损的集群。 5) istio-initializer 检查 mesh 的默认注入策略，并检查所有单个工作负载的策略负载值，以确定是否需要注入 sidecar。 6) istio-initializer 向工作负载中注入 sidecar 模板，然后通过 PATCH 向 kubernetes 提交。 7) kubernetes 正常的完成了工作负载的创建，并且工作负载中已经包含了注入的 sidecar。 配置选项 istio-initializer 具有用于注入的全局默认策略以及每个工作负载覆盖配置。全局策略由 istio-inject ConfigMap 配置（请参见下面的示例）。Initializer pod 必须重新启动以采用新的配置更改。 apiVersion: v1 kind: ConfigMap metadata: name: istio-inject namespace: istio-system data: config: |- policy: \"enabled\" namespaces: [\"\"] # everything, aka v1.NamepsaceAll, aka cluster-wide # excludeNamespaces: [\"ns1\", \"ns2\"] initializerName: \"sidecar.initializer.istio.io\" params: initImage: docker.io/istio/proxy_init:0.2.6 proxyImage: docker.io/istio/proxy:0.2.6 verbosity: 2 version: 0.2.6 meshConfigMapName: istio imagePullPolicy: IfNotPresent 下面是配置中的关键参数： policy off - 禁用 initializer 修改资源。pending 的 sidecar.initializer.istio.io initializer 将被删除以避免创建阻塞资源。 disable - initializer 不会注入 sidecar 到 watch 的所有 namespace 的资源中。启用 sidecar 注入请将 sidecar.istio.io/inject 注解的值设置为 true。 enable - initializer 将会注入 sidecar 到 watch 的所有 namespace 的资源中。禁用 sidecar 注入请将 sidecar.istio.io/inject 注解的值设置为 false。 namespaces 要 watch 和初始化的 namespace 列表。特殊的 \"\" namespace 对应于 v1.NamespaceAll 并配置初始化程序以初始化所有 namespace。kube-system、kube-publice 和 istio-system 被免除初始化。 excludeNamespaces 从 Istio initializer 中排除的 namespace 列表。不可以定义为 v1.NamespaceAll 或者与 namespaces 一起定义。 initializerName 这必须与 InitializerConfiguration 中 initializer 设定项的名称相匹配。Initializer 只处理匹配其配置名称的工作负载。 params 这些参数允许您对注入的 sidecar 进行有限的更改。更改这些值不会影响已部署的工作负载。 重写自动注入 单个工作负载可以通过使用 sidecar.istio.io/inject 注解重写全局策略。如果注解被省略，则使用全局策略。 如果注解的值是 true，则不管全局策略如何，sidecar 都将被注入。 如果注解的值是 false，则不管全局策略如何，sidecar 都不会被注入。 下表显示全局策略和每个工作负载覆盖的组合。 policy workload annotation injected off N/A no disabled omitted no disabled false no disabled true yes enabled omitted yes enabled false no enabled true yes 例如，即使全局策略是 disable，下面的 deployment 也会被注入sidecar。 apiVersion: extensions/v1beta1 kind: Deployment metadata: name: myapp annotations: sidecar.istio.io/inject: \"true\" spec: replicas: 1 template: ... 这是在包含 Istio 和非 Istio 服务的混合群集中使用自动注入的好方法。 卸载 Initializer 运行下面的命令，删除 Istio initializer： kubectl delete -f install/kubernetes/istio-initializer.yaml 注意上述命令并不会删除已注入到 Pod 中的 sidecar。要想删除这些 sidecar，需要在不使用 initializer 的情况下重新部署这些 pod。 拓展 Istio Mesh 将虚拟机或裸机集成到部署在 kubernetes 集群上的 Istio mesh 中的说明。 前置条件 按照 安装指南 在 kubernetes 集群上安装 Istio service mesh。 机器必须具有到 mesh 端点的 IP 地址连接。这通常需要一个 VPC 或者 VPN，以及一个向端点提供直接路由（没有 NAT 或者防火墙拒绝）的容器网络。及其不需要访问有 Kubernetes 分配的 cluster IP。 虚拟机必须可以访问到 Istio 控制平面服务（如Pilot、Mixer、CA）和 Kubernetes DNS 服务器。通常使用 内部负载均衡器 来实现。 您也可以使用 NodePort，在虚拟机上运行 Istio 的组件，或者使用自定义网络配置，有几个单独的文档会涵盖这些高级配置。 安装步骤 安装过程包括准备用于拓展的 mesh 和安装和配置虚拟机。 install/tools/setupMeshEx.sh ：这是一个帮助大家设置 kubernetes 环境的示例脚本。检查脚本内容和支持的环境变量（如 GCP_OPTS）。 install/tools/setupIstioVM.sh：这是一个用于配置主机环境的示例脚本。 您应该根据您的配置工具和DNS要求对其进行自定义。 准备要拓展的 Kubernetes 集群： 为 Kube DNS、Pilot、Mixer 和 CA 安装内部负载均衡器（ILB）。每个云供应商的配置都有所不同，根据具体情况修改注解。 0.2.7 版本的 YAML 文件的 DNS ILB 的 namespace 配置不正确。 使用 这一个 替代。 setupMeshEx.sh 中也有错误。使用上面链接中的最新文件或者从 GitHub.com/istio/istio 克隆。 kubectl apply -f install/kubernetes/mesh-expansion.yaml 生成要部署到虚拟机上的 Istio cluster.env 配置。该文件中包含要拦截的 cluster IP 地址范围。 export GCP_OPTS=\"--zone MY_ZONE --project MY_PROJECT\" install/tools/setupMeshEx.sh generateClusterEnv MY_CLUSTER_NAME 该示例生成的文件： cat cluster.env ISTIO_SERVICE_CIDR=10.63.240.0/20 产生虚拟机使用的 DNS 配置文件。这样可以让虚拟机上的应用程序解析到集群中的服务名称，这些名称将被 sidecar 拦截和转发。 # Make sure your kubectl context is set to your cluster install/tools/setupMeshEx.sh generateDnsmasq 该示例生成的文件： cat kubedns server=/svc.cluster.local/10.150.0.7 address=/istio-mixer/10.150.0.8 address=/istio-pilot/10.150.0.6 address=/istio-ca/10.150.0.9 address=/istio-mixer.istio-system/10.150.0.8 address=/istio-pilot.istio-system/10.150.0.6 address=/istio-ca.istio-system/10.150.0.9 设置机器 例如，您可以使用下面的“一条龙”脚本复制和安装配置： # 检查该脚本看看它是否满足您的需求 # 在 Mac 上，使用 brew install base64 或者 set BASE64_DECODE=\"/usr/bin/base64 -D\" export GCP_OPTS=\"--zone MY_ZONE --project MY_PROJECT\" install/tools/setupMeshEx.sh machineSetup VM_NAME 或者等效得手动安装步骤如下： ------ 手动安装步骤开始 ------ 将配置文件和 Istio 的 Debian 文件复制到要加入到集群的每台机器上。重命名为 /etc/dnsmasq.d/kubedns 和/var/lib/istio/envoy/cluster.env。 配置和验证 DNS 配置。需要安装 dnsmasq 或者直接将其添加到 /etc/resolv.conf 中，或者通过 DHCP 脚本。验证配置是否有效，检查虚拟机是否可以解析和连接到 pilot，例如： 在虚拟机或外部主机上： host istio-pilot.istio-system 产生的消息示例： # Verify you get the same address as shown as \"EXTERNAL-IP\" in 'kubectl get svc -n istio-system istio-pilot-ilb' istio-pilot.istio-system has address 10.150.0.6 检查是否可以解析 cluster IP。实际地址取决您的 deployment： host istio-pilot.istio-system.svc.cluster.local. 该示例产生的消息： istio-pilot.istio-system.svc.cluster.local has address 10.63.247.248 同样检查 istio-ingress： host istio-ingress.istio-system.svc.cluster.local. 该示例产生的消息： istio-ingress.istio-system.svc.cluster.local has address 10.63.243.30 验证连接性，检查迅即是否可以连接到 Pilot 的端点： curl 'http://istio-pilot.istio-system:8080/v1/registration/istio-pilot.istio-system.svc.cluster.local|http-discovery' { \"hosts\": [ { \"ip_address\": \"10.60.1.4\", \"port\": 8080 } ] } # 在虚拟机上使用上面的地址。将直接连接到运行 istio-pilot 的 pod。 curl 'http://10.60.1.4:8080/v1/registration/istio-pilot.istio-system.svc.cluster.local|http-discovery' 提取出实话 Istio 认证的 secret 并将它复制到机器上。Istio 的默认安装中包括 CA，即使是禁用了自动 mTLS 设置（她为每个 service account 创建 secret，secret 命名为 istio.）也会生成 Istio secret。建议您执行此步骤，以便日后启用 mTLS，并升级到默认启用 mTLS 的未来版本。 # ACCOUNT 默认是 'default'，SERVICE_ACCOUNT 是环境变量 # NAMESPACE 默认为当前 namespace，SERVICE_NAMESPACE 是环境变量 # （这一步由 machineSetup 完成） # 在 Mac 上执行 brew install base64 或者 set BASE64_DECODE=\"/usr/bin/base64 -D\" install/tools/setupMeshEx.sh machineCerts ACCOUNT NAMESPACE 生成的文件（key.pem, root-cert.pem, cert-chain.pem）必须拷贝到每台主机的 /etc/certs 目录，并且让 istio-proxy 可读。 安装 Istio Debian 文件，启动 istio 和 istio-auth-node-agent 服务。 从 github releases 获取 Debian 安装包： # 注意：在软件源配置好后，下面的额命令可以使用 'apt-get' 命令替代。 source istio.VERSION # defines version and URLs env var curl -L ${PILOT_DEBIAN_URL}/istio-agent.deb > ${ISTIO_STAGING}/istio-agent.deb curl -L ${AUTH_DEBIAN_URL}/istio-auth-node-agent.deb > ${ISTIO_STAGING}/istio-auth-node-agent.deb curl -L ${PROXY_DEBIAN_URL}/istio-proxy.deb > ${ISTIO_STAGING}/istio-proxy.deb dpkg -i istio-proxy-envoy.deb dpkg -i istio-agent.deb dpkg -i istio-auth-node-agent.deb systemctl start istio systemctl start istio-auth-node-agent ------ 手动安装步骤结束 ------ 安装完成后，机器就能访问运行在 Kubernetes 集群上的服务或者其他的 mesh 拓展的机器。 # 假设您在 'bookinfo' namespace 下安装的 bookinfo curl productpage.bookinfo.svc.cluster.local:9080 ... html content ... 检查进程是否正在运行： ps aux |grep istio root 6941 0.0 0.2 75392 16820 ? Ssl 21:32 0:00 /usr/local/istio/bin/node_agent --logtostderr root 6955 0.0 0.0 49344 3048 ? Ss 21:32 0:00 su -s /bin/bash -c INSTANCE_IP=10.150.0.5 POD_NAME=demo-vm-1 POD_NAMESPACE=default exec /usr/local/bin/pilot-agent proxy > /var/log/istio/istio.log istio-proxy istio-p+ 7016 0.0 0.1 215172 12096 ? Ssl 21:32 0:00 /usr/local/bin/pilot-agent proxy istio-p+ 7094 4.0 0.3 69540 24800 ? Sl 21:32 0:37 /usr/local/bin/envoy -c /etc/istio/proxy/envoy-rev1.json --restart-epoch 1 --drain-time-s 2 --parent-shutdown-time-s 3 --service-cluster istio-proxy --service-node sidecar~10.150.0.5~demo-vm-1.default~default.svc.cluster.local 检查 Istio auth-node-agent 是否健康： sudo systemctl status istio-auth-node-agent ● istio-auth-node-agent.service - istio-auth-node-agent: The Istio auth node agent Loaded: loaded (/lib/systemd/system/istio-auth-node-agent.service; disabled; vendor preset: enabled) Active: active (running) since Fri 2017-10-13 21:32:29 UTC; 9s ago Docs: http://istio.io/ Main PID: 6941 (node_agent) Tasks: 5 Memory: 5.9M CPU: 92ms CGroup: /system.slice/istio-auth-node-agent.service └─6941 /usr/local/istio/bin/node_agent --logtostderr Oct 13 21:32:29 demo-vm-1 systemd[1]: Started istio-auth-node-agent: The Istio auth node agent. Oct 13 21:32:29 demo-vm-1 node_agent[6941]: I1013 21:32:29.469314 6941 main.go:66] Starting Node Agent Oct 13 21:32:29 demo-vm-1 node_agent[6941]: I1013 21:32:29.469365 6941 nodeagent.go:96] Node Agent starts successfully. Oct 13 21:32:29 demo-vm-1 node_agent[6941]: I1013 21:32:29.483324 6941 nodeagent.go:112] Sending CSR (retrial #0) ... Oct 13 21:32:29 demo-vm-1 node_agent[6941]: I1013 21:32:29.862575 6941 nodeagent.go:128] CSR is approved successfully. Will renew cert in 29m59.137732603s 在拓展的 mesh 中的机器上运行服务 配置 sidecar 拦截端口。在 /var/lib/istio/envoy/sidecar.env 中通过 ISTIO_INBOUND_PORTS 环境变量配置。 例如（运行服务的虚拟机）： echo \"ISTIO_INBOUND_PORTS=27017,3306,8080\" > /var/lib/istio/envoy/sidecar.env systemctl restart istio 手动配置 selector-less 的 service 和 endpoint。“selector-less” service 用于那些不依托 Kubernetes pod 的 service。 例如，在有权限的机器上修改 Kubernetes 中的 service： # istioctl register servicename machine-ip portname:port istioctl -n onprem register mysql 1.2.3.4 3306 istioctl -n onprem register svc1 1.2.3.4 http:7000 安装完成后，Kubernetes pod 和其它 mesh 扩展将能够访问集群上运行的服务。 整合到一起 请参阅 拓展 BookInfo Mesh 指南。 部署 bookinfo 示例应用 该示例部署由四个单独的微服务组成的简单应用程序，用于演示Istio服务网格的各种功能。 概况 在本示例中，我们将部署一个简单的应用程序，显示书籍的信息，类似于网上书店的书籍条目。在页面上有书籍的描述、详细信息（ISBN、页数等）和书评。 BookInfo 应用程序包括四个独立的微服务： productpage：productpage(产品页面)微服务，调用 details 和 reviews 微服务来填充页面。 details：details 微服务包含书籍的详细信息。 reviews：reviews 微服务包含书籍的点评。它也调用 ratings 微服务。 ratings：ratings 微服务包含随书评一起出现的评分信息。 有3个版本的 reviews 微服务： 版本v1不调用 ratings 服务。 版本v2调用 ratings ，并将每个评级显示为1到5个黑色星。 版本v3调用 ratings ，并将每个评级显示为1到5个红色星。 应用程序的端到端架构如下所示。 图片 - BookInfo 该应用程序是多语言构建的，即这些微服务是用不同的语言编写的。值得注意的是，这些服务与 Istio 没有任何依赖关系，单这是个有趣的 Service Mesh 示例，特别是因为评论服务和众多的语言和版本。 开始之前 如果您还没有这样做，请按照与您的平台 安装指南 对应的说明安装Istio。 部署应用程序 使用 Istio 运行应用程序示例不需要修改应用程序本身。相反，我们只需要在支持 Istio 的环境中配置和运行服务， Envoy sidecar 将会注入到每个服务中。所需的命令和配置根据运行时环境的不同而有所不同，但在所有情况下，生成的部署将如下所示： 图片 - BookInfo 所有的微服务都将与一个 Envoy sidecar 一起打包，拦截这些服务的入站和出站的调用请求，提供通过 Istio 控制平面从外部控制整个应用的路由，遥测收集和策略执行所需的 hook。 要启动该应用程序，请按照以下对应于您的 Istio 运行时环境的说明进行操作。 在 Kubernetes 中运行 注意：如果您使用 GKE，清确保您的集群至少有 4 个标准的 GKE 节点。如果您使用 Minikube，请确保您至少有 4GB 内存。 将目录更改为 Istio 安装目录的根目录。 构建应用程序容器： 如果您使用 自动注入 sidecar 的方式部署的集群，那么只需要使用 kubectl 命令部署服务： kubectl apply -f samples/bookinfo/kube/bookinfo.yaml 如果您使用 手动注入 sidecar 的方式部署的集群，清使用下面的命令： kubectl apply -f 请注意，该 istioctl kube-inject 命令用于在创建部署之前修改 bookinfo.yaml 文件。这将把 Envoy 注入到 Kubernetes 资源。 上述命令启动四个微服务并创建网关入口资源，如下图所示。3 个版本的评论的服务 v1、v2、v3 都已启动。 请注意在实际部署中，随着时间的推移部署新版本的微服务，而不是同时部署所有版本。 确认所有服务和 pod 已正确定义并运行： kubectl get services 这将产生以下输出： NAME CLUSTER-IP EXTERNAL-IP PORT(S) AGE details 10.0.0.31 9080/TCP 6m istio-ingress 10.0.0.122 80:31565/TCP 8m istio-pilot 10.0.0.189 8080/TCP 8m istio-mixer 10.0.0.132 9091/TCP,42422/TCP 8m kubernetes 10.0.0.1 443/TCP 14d productpage 10.0.0.120 9080/TCP 6m ratings 10.0.0.15 9080/TCP 6m reviews 10.0.0.170 9080/TCP 6m 而且 kubectl get pods 将产生: NAME READY STATUS RESTARTS AGE details-v1-1520924117-48z17 2/2 Running 0 6m istio-ingress-3181829929-xrrk5 1/1 Running 0 8m istio-pilot-175173354-d6jm7 2/2 Running 0 8m istio-mixer-3883863574-jt09j 2/2 Running 0 8m productpage-v1-560495357-jk1lz 2/2 Running 0 6m ratings-v1-734492171-rnr5l 2/2 Running 0 6m reviews-v1-874083890-f0qf0 2/2 Running 0 6m reviews-v2-1343845940-b34q5 2/2 Running 0 6m reviews-v3-1813607990-8ch52 2/2 Running 0 6m 确定 ingress IP 和端口 如果您的 kubernetes 集群环境支持外部负载均衡器的话，可以使用下面的命令获取 ingress 的IP地址： kubectl get ingress -o wide 输出如下所示： NAME HOSTS ADDRESS PORTS AGE gateway * 130.211.10.121 80 1d Ingress 服务的地址是： export GATEWAY_URL=130.211.10.121:80 GKE：如果服务无法获取外部 IP，kubectl get ingress -o wide 会显示工作节点的列表。在这种情况下，您可以使用任何地址以及 NodePort 访问入口。但是，如果集群具有防火墙，则还需要创建防火墙规则以允许TCP流量到NodePort，您可以使用以下命令创建防火墙规则： export GATEWAY_URL=:$(kubectl get svc istio-ingress -n istio-system -o jsonpath='{.spec.ports[0].nodePort}') gcloud compute firewall-rules create allow-book --allow tcp:$(kubectl get svc istio-ingress -n istio-system -o jsonpath='{.spec.ports[0].nodePort}') IBM Bluemix Free Tier：在免费版的 Bluemix 的 kubernetes 集群中不支持外部负载均衡器。您可以使用工作节点的公共 IP，并通过 NodePort 来访问 ingress。工作节点的公共 IP可以通过如下命令获取： bx cs workers export GATEWAY_URL=:$(kubectl get svc istio-ingress -n istio-system -o jsonpath='{.spec.ports[0].nodePort}') Minikube：Minikube 不支持外部负载均衡器。您可以使用 ingress 服务的主机 IP 和 NodePort 来访问 ingress： export GATEWAY_URL=$(kubectl get po -l istio=ingress -o 'jsonpath={.items[0].status.hostIP}'):$(kubectl get svc istio-ingress -o 'jsonpath={.spec.ports[0].nodePort}') 在 Consul 或 Eureka 环境下使用 Docker 运行 切换到 Istio 的安装根目录下。 启动应用程序容器。 执行下面的命令测试 Consul： docker-compose -f samples/bookinfo/consul/bookinfo.yaml up -d 执行下面的命令测试 Eureka： docker-compose -f samples/bookinfo/eureka/bookinfo.yaml up -d 确认所有容器都在运行： docker ps -a 如果 Istio Pilot 容器终止了，重新执行上面的命令重新运行。 设置 GATEWAY_URL: export GATEWAY_URL=localhost:9081 下一步 使用以下 curl 命令确认 BookInfo 应用程序正在运行: curl -o /dev/null -s -w \"%{http_code}\\n\" http://${GATEWAY_URL}/productpage 200 你也可以通过在浏览器中打开 http://$GATEWAY_URL/productpage 页面访问 Bookinfo 网页。如果您多次刷新浏览器将在 productpage 中看到评论的不同的版本，它们会按照 round robin（红星、黑星、没有星星）的方式展现，因为我们还没有使用 Istio 来控制版本的路由。 现在，您可以使用此示例来尝试 Istio 的流量路由、故障注入、速率限制等功能。要继续的话，请参阅 Istio 指南，具体取决于您的兴趣。智能路由 是初学者入门的好方式。 清理 在完成 BookInfo 示例后，您可以卸载它，如下所示： 卸载 Kubernetes 环境 删除路由规则，终止应用程序 pod samples/bookinfo/kube/cleanup.sh 确认关闭 istioctl get routerules #-- there should be no more routing rules kubectl get pods #-- the BookInfo pods should be deleted 卸载 docker 环境 删除路由规则和应用程序容器 若使用 Consul 环境安装，执行下面的命令： samples/bookinfo/consul/cleanup.sh 若使用 Eureka 环境安装，执行下面的命令： samples/bookinfo/eureka/cleanup.sh 确认清理完成： istioctl get routerules #-- there should be no more routing rules docker ps -a #-- the BookInfo containers should be delete Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-21 21:18:30 "},"usecases/integrating-vms.html":{"url":"usecases/integrating-vms.html","title":"集成虚拟机","keywords":"","body":"集成虚拟机 该示例跨越 Kubernetes 集群和一组虚拟机上部署 Bookinfo 服务，描述如何使用 Istio service mesh 将此基础架构以单一 mesh 的方式操控。 注意：本文档还在建设中，并且只在 Google Cloud Platform 上进行过测试。 在 IBM Bluemix 或其它平台上，pod 的 overlay 网络跟虚拟机的网络是隔离的。 即使使用 Istio，虚拟机也不能直接与 Kubernetes Pod 进行通信。 概览 图片 - Bookinfo应用的拓展Mesh 开始之前 按照 安装指南 上的步骤部署 Istio。 部署 BookInfo 示例应用程序（在 bookinfo namespace 下）。 在 Istio 集群相同的项目下创建名为 vm-1 的虚拟机，并 加入到 Mesh。 在虚拟机上运行 mysql 我们将首先在虚拟机上安装 mysql，将其配置成评分服务的后台存储。 在虚拟机上执行： sudo apt-get update && sudo apt-get install -y mariadb-server sudo mysql # 授权 root 用户访问 GRANT ALL PRIVILEGES ON *.* TO 'root'@'localhost' IDENTIFIED BY 'password' WITH GRANT OPTION; quit; sudo systemctl restart mysql 关于 Mysql 的详细配置请见： Mysql。 在虚拟机上执行下面的命令，向 mysql 中添加评分数据库。 # 向 mysql 数据库中添加评分数据库 curl -q https://raw.githubusercontent.com/istio/istio/master/samples/bookinfo/src/mysql/mysqldb-init.sql | mysql -u root -ppassword 为了便于直观地查看bookinfo应用程序输出的差异，可以更改使用以下命令生成的评分： # 查看评分 mysql -u root -ppassword test -e \"select * from ratings;\" +----------+--------+ | ReviewID | Rating | +----------+--------+ | 1 | 5 | | 2 | 4 | +----------+--------+ # 修改评分 mysql -u root -ppassword test -e \"update ratings set rating=1 where reviewid=1;select * from ratings;\" +----------+--------+ | ReviewID | Rating | +----------+--------+ | 1 | 1 | | 2 | 4 | +----------+--------+ 找出将添加到 mesh 中的虚拟机的 IP 地址 在虚拟机上执行： hostname -I 将 mysql 服务注册到 mesh 中 在一台可以访问 istioctl 命令的主机上，注册该虚拟机和 mysql db service： istioctl register -n vm mysqldb 3306 # 示例输出 $ istioctl register mysqldb 192.168.56.112 3306 I1015 22:24:33.846492 15465 register.go:44] Registering for service 'mysqldb' ip '192.168.56.112', ports list [{3306 mysql}] I1015 22:24:33.846550 15465 register.go:49] 0 labels ([]) and 1 annotations ([alpha.istio.io/kubernetes-serviceaccounts=default]) W1015 22:24:33.866410 15465 register.go:123] Got 'services \"mysqldb\" not found' looking up svc 'mysqldb' in namespace 'default', attempting to create it W1015 22:24:33.904162 15465 register.go:139] Got 'endpoints \"mysqldb\" not found' looking up endpoints for 'mysqldb' in namespace 'default', attempting to create them I1015 22:24:33.910707 15465 register.go:180] No pre existing exact matching ports list found, created new subset {[{192.168.56.112 nil}] [] [{mysql 3306 }]} I1015 22:24:33.921195 15465 register.go:191] Successfully updated mysqldb, now with 1 endpoints 集群管理 如果你之前在 kubernetes 上运行过 mysql，您需要将 kubernetes 的 mysql service 移除： kubectl delete service mysql 执行 istioctl 来配置 service（在您的 admin 机器上）： istioctl register mysql IP mysql:PORT 注意：mysqldb 虚拟机不需要也不应该有特别的 kubernetes 权限。 使用 mysql 服务 bookinfo 中的评分服务将使用机器上的数据库。要验证它是否有效，请创建使用虚拟机上的 mysql db 的评分服务的 v2 版本。然后指定强制评论服务使用评分 v2 版本的路由规则。 # 创建使用 mysql 后端的评分服务版本 istioctl kube-inject -n bookinfo -f samples/bookinfo/kube/bookinfo-ratings-v2-mysql-vm.yaml | kubectl apply -n bookinfo -f - # 强制 bookinfo 使用评分后端的路有规则 istioctl create -n bookinfo -f samples/bookinfo/kube/route-rule-ratings-mysql-vm.yaml 您可以验证 bookinfo 应用程序的输出结果，显示来自 Reviewer1 的 1 星级和来自 Reviewer2 的 4 星级，或者更改虚拟机上的评分并查看结果。 同时，您还可以在 RawVM MySQL 的文档中找到一些故障排查和其它信息。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-07 23:54:19 "},"usecases/linkerd.html":{"url":"usecases/linkerd.html","title":"Linkerd","keywords":"","body":"Linkerd简介 Linkerd是一个用于云原生应用的开源、可扩展的service mesh（一般翻译成服务网格，还有一种说法叫”服务啮合层“，见Istio：用于微服务的服务啮合层）。 Linkerd是什么 Linkerd的出现是为了解决像twitter、google这类超大规模生产系统的复杂性问题。Linkerd不是通过控制服务之间的通信机制来解决这个问题，而是通过在服务实例之上添加一个抽象层来解决的。 图片 - source https://linkerd.io Linkerd负责跨服务通信中最困难、易出错的部分，包括延迟感知、负载平衡、连接池、TLS、仪表盘、请求路由等——这些都会影响应用程序伸缩性、性能和弹性。 如何运行 Linkerd作为独立代理运行，无需特定的语言和库支持。应用程序通常会在已知位置运行linkerd实例，然后通过这些实例代理服务调用——即不是直接连接到目标服务，服务连接到它们对应的linkerd实例，并将它们视为目标服务。 在该层上，linkerd应用路由规则，与现有服务发现机制通信，对目标实例做负载均衡——与此同时调整通信并报告指标。 通过延迟调用linkerd的机制，应用程序代码与以下内容解耦： 生产拓扑 服务发现机制 负载均衡和连接管理逻辑 应用程序也将从一致的全局流量控制系统中受益。这对于多语言应用程序尤其重要，因为通过库来实现这种一致性是非常困难的。 Linkerd实例可以作为sidecar（既为每个应用实体或每个主机部署一个实例）来运行。 由于linkerd实例是无状态和独立的，因此它们可以轻松适应现有的部署拓扑。它们可以与各种配置的应用程序代码一起部署，并且基本不需要去协调它们。 参考 Buoyant发布服务网格Linkerd的1.0版本 Linkerd documentation Istio：用于微服务的服务啮合层 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-10-22 22:32:51 "},"usecases/linkerd-user-guide.html":{"url":"usecases/linkerd-user-guide.html","title":"Linkerd 使用指南","keywords":"","body":"Linkerd 使用指南 前言 Linkerd 作为一款 service mesh 与kubernetes 结合后主要有以下几种用法： 作为服务网关，可以监控 kubernetes 中的服务和实例 使用 TLS 加密服务 通过流量转移到持续交付 开发测试环境（Eat your own dog food）、Ingress 和边缘路由 给微服务做 staging 分布式 tracing 作为 Ingress controller 使用 gRPC 更方便 以下我们着重讲解在 kubernetes 中如何使用 linkerd 作为 kubernetes 的 Ingress controller，并作为边缘节点代替 Traefik 的功能，详见 边缘节点的配置。 准备 安装测试时需要用到的镜像有： buoyantio/helloworld:0.1.4 buoyantio/jenkins-plus:2.60.1 buoyantio/kubectl:v1.4.0 buoyantio/linkerd:1.1.2 buoyantio/namerd:1.1.2 buoyantio/nginx:1.10.2 linkerd/namerctl:0.8.6 openzipkin/zipkin:1.20 tutum/dnsutils:latest 这些镜像可以直接通过 Docker Hub 获取，我将它们下载下来并上传到了自己的私有镜像仓库 sz-pg-oam-docker-hub-001.tendcloud.com 中，下文中用到的镜像皆来自我的私有镜像仓库，yaml 配置见 linkerd 目录，并在使用时将配置中的镜像地址修改为你自己的。 部署 首先需要先创建 RBAC，因为使用 namerd 和 ingress 时需要用到。 $ kubectl create -f linkerd-rbac-beta.yml Linkerd 提供了 Jenkins 示例，在部署的时候使用以下命令： $ kubectl create -f jenkins-rbac-beta.yml $ kubectl create -f jenkins.yml 访问 http://jenkins.jimmysong.io 图片 - Jenkins pipeline 图片 - Jenkins config 注意：要访问 Jenkins 需要在 Ingress 中增加配置，下文会提到。 在 kubernetes 中使用 Jenkins 的时候需要注意 Pipeline 中的配置： def currentVersion = getCurrentVersion() def newVersion = getNextVersion(currentVersion) def frontendIp = kubectl(\"get svc l5d -o jsonpath=\\\"{.status.loadBalancer.ingress[0].*}\\\"\").trim() def originalDst = getDst(getDtab()) frontendIP 的地址要配置成 service 的 Cluster IP ，因为我们没有用到LoadBalancer。 需要安装 namerd，namerd 负责 dtab 信息的存储，当然也可以存储在 etcd、consul中。dtab 保存的是路由规则信息，支持递归解析，详见 dtab。 流量切换主要是通过 dtab 来实现的，通过在 HTTP 请求的 header 中增加 l5d-dtab 和 Host 信息可以对流量分离到 kubernetes 中的不同 service 上。 遇到的问题 Failed with the following error(s) Error signal dtab is already marked as being deployed! 因为该 dtab entry 已经存在，需要删除后再运行。 访问 http://namerd.jimmysong.io 图片 - namerd dtab 保存在 namerd 中，该页面中的更改不会生效，需要使用命令行来操作。 使用 namerctl 来操作。 $ namerctl --base-url http://namerd-backend.jimmysong.io dtab update internal file 注意：update 时需要将更新文本先写入文件中。 部署 Linkerd 直接使用 yaml 文件部署，注意修改镜像仓库地址。 # 创建 namerd $ kubectl create -f namerd.yaml # 创建 ingress $ kubectl create -f linkerd-ingress.yml # 创建测试服务 hello-world $ kubectl create -f hello-world.yml # 创建 API 服务 $ kubectl create -f api.yml # 创建测试服务 world-v2 $ kubectl create -f world-v2.yml 为了在本地调试 linkerd，我们将 linkerd 的 service 加入到 ingress 中，详见 边缘节点配置。 在 Ingress 中增加如下内容： - host: linkerd.jimmysong.io http: paths: - path: / backend: serviceName: l5d servicePort: 9990 - host: linkerd-viz.jimmysong.io http: paths: - path: / backend: serviceName: linkerd-viz servicePort: 80 - host: l5d.jimmysong.io http: paths: - path: / backend: serviceName: l5d servicePort: 4141 - host: jenkins.jimmysong.io http: paths: - path: / backend: serviceName: jenkins servicePort: 80 在本地/etc/hosts中添加如下内容： 172.20.0.119 linkerd.jimmysong.io 172.20.0.119 linkerd-viz.jimmysong.io 172.20.0.119 l5d.jimmysong.io 测试路由功能 使用 curl 简单测试。 单条测试 $ curl -s -H \"Host: www.hello.world\" 172.20.0.120:4141 Hello (172.30.60.14) world (172.30.71.19)!!% 请注意请求返回的结果，表示访问的是 world-v1 service。 $ for i in $(seq 0 10000);do echo $i;curl -s -H \"Host: www.hello.world\" 172.20.0.120:4141;done 使用 ab test。 $ ab -c 4 -n 10000 -H \"Host: www.hello.world\" http://172.20.0.120:4141/ This is ApacheBench, Version 2.3 Copyright 1996 Adam Twiss, Zeus Technology Ltd, http://www.zeustech.net/ Licensed to The Apache Software Foundation, http://www.apache.org/ Benchmarking 172.20.0.120 (be patient) Completed 1000 requests Completed 2000 requests Completed 3000 requests Completed 4000 requests Completed 5000 requests Completed 6000 requests Completed 7000 requests Completed 8000 requests Completed 9000 requests Completed 10000 requests Finished 10000 requests Server Software: Server Hostname: 172.20.0.120 Server Port: 4141 Document Path: / Document Length: 43 bytes Concurrency Level: 4 Time taken for tests: 262.505 seconds Complete requests: 10000 Failed requests: 0 Total transferred: 2210000 bytes HTML transferred: 430000 bytes Requests per second: 38.09 [#/sec] (mean) Time per request: 105.002 [ms] (mean) Time per request: 26.250 [ms] (mean, across all concurrent requests) Transfer rate: 8.22 [Kbytes/sec] received Connection Times (ms) min mean[+/-sd] median max Connect: 36 51 91.1 39 2122 Processing: 39 54 29.3 46 585 Waiting: 39 52 20.3 46 362 Total: 76 105 96.3 88 2216 Percentage of the requests served within a certain time (ms) 50% 88 66% 93 75% 99 80% 103 90% 119 95% 146 98% 253 99% 397 100% 2216 (longest request) 监控 kubernets 中的服务与实例 访问 http://linkerd.jimmysong.io 查看流量情况 Outcoming 图片 - linkerd监控 Incoming 图片 - linkerd监控 访问 http://linkerd-viz.jimmysong.io 查看应用 metric 监控 图片 - linkerd性能监控 测试路由 测试在 http header 中增加 dtab 规则。 $ curl -H \"Host: www.hello.world\" -H \"l5d-dtab:/host/world => /srv/world-v2;\" 172.20.0.120:4141 Hello (172.30.60.14) earth (172.30.94.40)!! 请注意调用返回的结果，表示调用的是 world-v2 的 service。 另外再对比 ab test 的结果与 linkerd-viz 页面上的结果，可以看到结果一致。 但是我们可能不想把该功能暴露给所有人，所以可以在前端部署一个 nginx 来过滤 header 中的 l5d-dtab 打头的字段，并通过设置 cookie 的方式来替代 header 里的 l5d-dtab 字段。 $ http_proxy=http://172.20.0.120:4141 curl -s http:/hello Hello (172.30.60.14) world (172.30.71.19)!! 将 Linkerd 作为 Ingress controller 将 Linkerd 作为 kubernetes ingress controller 的方式跟将 Treafik 作为 ingress controller 的过程过程完全一样，可以直接参考 边缘节点配置。 架构如下图所示。 图片 - Linkerd ingress controller (图片来自 A Service Mesh for Kubernetes - Buoyant.io) 当然可以绕过 kubernetes ingress controller 直接使用 linkerd 作为边界路由，通过 dtab 和 linkerd 前面的 nginx 来路由流量。 参考 https://github.com/linkerd/linkerd-examples/ A Service Mesh for Kubernetes dtab Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-08-21 18:23:35 "},"usecases/conduit.html":{"url":"usecases/conduit.html","title":"Conduit","keywords":"","body":"Conduit - 基于Kubernetes的轻量级Service Mesh 2017年12月在得克萨斯州的Asdin，KubeCon和CloudNativeCon上，创造了Service Mesh这个词汇并开源了Linkerd的公司Buoyant，又开源了一款针对Kubernetes的超轻量Service Sesh——Conduit。它可以透明得管理服务运行时之间的通信，使得在Kubernetes上运行服务更加安全和可靠；它还具有不用修改任何应用程序代码即可改进应用程序的可观测性、可靠性及安全性等方面的特性。 Condiut与Linkerd的设计方式不同，它跟Istio一样使用的是Sidecar模式，但架构又没Istio那么复杂。Conduit只支持Kubernetes，且只支持HTTP2（包括gRPC）协议。 Conduit使用Rust和Go语言开发，GitHub地址https://github.com/runconduit/conduit 安装Conduit必须使用Kubernetes1.8以上版本。 参考 Conduit GitHub：https://github.com/runconduit/conduit 关于Conduit的更多资源请参考官方网站：https://conduit.io/ 另外，我们翻译Conduit的官方文档见：https://github.com/doczhcn/conduit 关于Service Mesh的更多内容请访问Service Mesh中国：http://www.servicemesh.cn/ Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-10 21:53:24 "},"usecases/conduit-overview.html":{"url":"usecases/conduit-overview.html","title":"Condiut概览","keywords":"","body":"Conduit 概览 Conduit是一款针对Kubernetes的超轻量级的service mesh。它可以透明得管理服务运行时之间的通信，使得在Kubernetes上运行服务更加安全和可靠；它还具有不用修改任何应用程序代码即可改进应用程序的可观测性、可靠性及安全性等方面的特性。 本文档将从一个较高层次介绍Conduit及其是如何工作的。如果不熟悉service mesh模型，或许可以先阅读William Morgan的概览文章 什么是service mesh？为什么需要它？ Conduit架构 Conduit service mesh部署到Kubernetes集群时有两个基本组件：数据平面和控制平面。数据平面承载服务实例间实际的应用请求流量，而控制平面则驱动数据平面并修改其行为（及访问聚合指标）提供API。Conduit CLI和Web UI使用此API并为人类提供符合人体工程学的控制。 让我们依次认识这些组件。 Conduit的数据平面由轻量级的代理组成，这些代理作为sidecar容器与每个服务代码的实例部署在一起。如果将服务“添加”到Conduit servie mesh中，必须重新部署该服务的pod，以便在每个pod中包含一个数据平面代理。（conduit inject 命令可以完成这个任务，以及透明地从每个实例通过代理汇集流量所需的配置工作。） 这些代理透明地拦截进出每个pod的通信，并增加诸如重试和超时、检测及加密（TLS）等特性，并根据相关策略来允许和拒绝请求。 这些代理并未设计成通过手动方式配置；相反，它们的行为是由控制平面驱动的。 Conduit控制平面是一组运行在专用Kubernetes名称空间（默认情况下为conduit）的服务。这些服务完成各种事情 - 聚合遥测数据，提供面向用户的API，向数据平面代理提供控制数据等。它们一起驱动数据平面的行为。 使用Conduit 为了支持Conduit的人机交互，可以使用Conduit CLI及web UI（也可以通过相关工具比如 kubectl）。CLI 和 web UI通过API驱动控制平面，而控制平面相应地驱动数据平面的行为。 控制平面API设计得足够通用，以便能基于此构建其他工具。比如，你可能希望从另外一个CI/CD系统来驱动API。 运行 conduit --help 可查看关于CLI功能的简短概述。 Conduit 与 Kubernetes Conduit设计用于无缝地融入现有的Kubernetes系统。该设计有几个重要特征。 第一，Conduit CLI（conduit）设计成尽可能地与 kubectl 一起使用。比如，conduit install 和 conduit inject 生成的Kubernetes配置，被设计成直接送入kubectl。这是为了在service mesh和编排系统之间提供一个明确的分工，并且使得Conduit适配现有的Kubernetes工作流程。 第二，Kubernetes中Conduit的核心词是Deployment，而不是Service。举个例子，conduit inject 增加一个Deployment，Conduit web UI会显示这些Deployment，并且每个Deployment都会给出聚合的性能指标。这是因为单个pod可以是任意数量Service的一部分，而这会导致流量与pod之间出现复杂的映射。相比之下，Deployment要求单个pod只能属于一个Deployment的一部分。通过基于Deployment而不是Service来构建，流量与pod间的映射就总是清晰的。 这两个设计特性能很好地组合。比如，conduit inject可用于一个运行的Deployment，因为当它更新Deployment时， Kubernetes会回滚pod以包括数据平面代理。 扩展Conduit的行为 Conduit控制平面还为构建自定义功能提供了一个便捷的入口。Conduit最初发布时并不支持这一点，在不远的将来，通过编写gRPC插件来作为控制平面的一部分运行，将能扩展Conduit的功能，而无需重新编译Conduit。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-11 14:19:23 "},"usecases/conduit-installation.html":{"url":"usecases/conduit-installation.html","title":"安装Conduit","keywords":"","body":"安装Conduit 本文档指导您如何在kubernetes上安装Conduit service mesh。 前提条件 kubernetes版本为1.8或以上 用到的镜像如下： buoyantio/kubectl:v1.6.2 gcr.io/runconduit/controller:v0.1.0 gcr.io/runconduit/web:v0.1.0 prom/prometheus:v1.8.1 其中位于gcr.io的镜像我备份到了DockerHub： jimmysong/runconduit-web:v0.1.0 jimmysong/runconduit-controller:v0.1.0 另外两个镜像本身就可以从DockerHub上下载。 部署 到release页面上下载conduit的二进制文件。 使用conduit install命令生成了用于部署到kubernetes中yaml文件，然后修改文件中的镜像仓库地址为你自己的镜像地址。 conduit install>conduit-0.1.0.yaml # 修改完镜像地址执行 kubectl apply -f conduit-0.1.0.yaml 修改后的yaml文件见：conduit-0.1.0.yaml。 注意：Conduit官方给出的yaml文件中不包括RBAC授权，我重新修改了，增加了RBAC和ServiceAccount。 使用kubectl proxy来开放外网访问conduit dashboard： kubectl proxy --address='172.20.0.113' --port=8001 --accept-hosts='^*$' 在浏览器中访问http://172.20.0.113:8001/api/v1/namespaces/conduit/services/web:http/proxy/servicemesh将看到如下页面： 图片 - Conduit dashboard Conduit inject Conduit注入的时候需要用到如下两个镜像： gcr.io/runconduit/proxy:v0.1.0 gcr.io/runconduit/proxy-init:v0.1.0 我将其备份到了DockerHub： jimmysong/runconduit-proxy:v0.1.0 jimmysong/runconduit-proxy-init:v0.1.0 查看conduit向yaml文件中注入了哪些配置，我们使用my-nginx.yaml为例： conduit inject --init-image sz-pg-oam-docker-hub-001.tendcloud.com/library/runconduit-proxy-init --proxy-image sz-pg-oam-docker-hub-001.tendcloud.com/library/runconduit-proxy my-nginx.yaml|kubectl apply -f - 注意：只需要指定镜像名称即可，tag与使用的conduit server版本相同，会自动注入。 my-nginx.yaml的内容如下： apiVersion: extensions/v1beta1 kind: Deployment metadata: name: my-nginx spec: replicas: 2 template: metadata: labels: run: my-nginx spec: containers: - name: my-nginx image: sz-pg-oam-docker-hub-001.tendcloud.com/library/nginx:1.9 ports: - containerPort: 80 --- apiVersion: v1 kind: Service metadata: name: my-nginx labels: app: my-nginx spec: ports: - port: 80 protocol: TCP name: http selector: run: my-nginx Conduit自动注入后生成的新的yaml文件内容如下： apiVersion: extensions/v1beta1 kind: Deployment metadata: creationTimestamp: null name: my-nginx spec: replicas: 2 strategy: {} template: metadata: annotations: conduit.io/created-by: conduit/cli v0.1.0 conduit.io/proxy-version: v0.1.0 creationTimestamp: null labels: conduit.io/controller: conduit conduit.io/plane: data run: my-nginx spec: containers: - image: sz-pg-oam-docker-hub-001.tendcloud.com/library/nginx:1.9 name: my-nginx ports: - containerPort: 80 resources: {} - env: - name: CONDUIT_PROXY_LOG value: trace,h2=debug,mio=info,tokio_core=info - name: CONDUIT_PROXY_CONTROL_URL value: tcp://proxy-api.conduit.svc.cluster.local:8086 - name: CONDUIT_PROXY_CONTROL_LISTENER value: tcp://0.0.0.0:4190 - name: CONDUIT_PROXY_PRIVATE_LISTENER value: tcp://127.0.0.1:4140 - name: CONDUIT_PROXY_PUBLIC_LISTENER value: tcp://0.0.0.0:4143 - name: CONDUIT_PROXY_NODE_NAME valueFrom: fieldRef: fieldPath: spec.nodeName - name: CONDUIT_PROXY_POD_NAME valueFrom: fieldRef: fieldPath: metadata.name - name: CONDUIT_PROXY_POD_NAMESPACE valueFrom: fieldRef: fieldPath: metadata.namespace image: sz-pg-oam-docker-hub-001.tendcloud.com/library/runconduit-proxy:v0.1.0 imagePullPolicy: IfNotPresent name: conduit-proxy ports: - containerPort: 4143 name: conduit-proxy resources: {} securityContext: runAsUser: 2102 initContainers: - args: - -p - \"4143\" - -o - \"4140\" - -i - \"4190\" - -u - \"2102\" image: sz-pg-oam-docker-hub-001.tendcloud.com/library/runconduit-proxy-init:v0.1.0 imagePullPolicy: IfNotPresent name: conduit-init resources: {} securityContext: capabilities: add: - NET_ADMIN privileged: false status: {} --- apiVersion: v1 kind: Service metadata: name: my-nginx labels: app: my-nginx spec: ports: - port: 80 protocol: TCP name: http selector: run: my-nginx --- 部署示例应用 使用下面的命令部署官方提供的示例应用： curl https://raw.githubusercontent.com/rootsongjc/kubernetes-handbook/master/manifests/conduit/emojivoto.yml | conduit inject --init-image sz-pg-oam-docker-hub-001.tendcloud.com/library/runconduit-proxy-init --proxy-image sz-pg-oam-docker-hub-001.tendcloud.com/library/runconduit-proxy - --skip-inbound-ports=80 | kubectl apply -f - 注意：其中使用的镜像地址已经改为我的私有镜像仓库地址，大家使用时请注意修改。 参考 Getting started - conduit.io Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-13 21:33:43 "},"usecases/big-data.html":{"url":"usecases/big-data.html","title":"大数据","keywords":"","body":"大数据 Kubernetes community中已经有了一个Big data SIG，大家可以通过这个SIG了解kubernetes结合大数据的应用。 在Swarm、Mesos、kubernetes这三种流行的容器编排调度架构中，Mesos对于大数据应用支持是最好的，spark原生就是运行在mesos上的，当然也可以容器化运行在kubernetes上。当前在kubernetes上运行大数据应用主要是spark应用。 Spark on Kubernetes Spark原生支持standalone、mesos和YARN的调度方式，当前kubernetes社区正在支持kubernetes的原生调度来运行spark - 。 当然您也可以在kubernetes直接部署spark on yarn或者spark standalone模式，仍然沿用已有的 Spark Standalone 使用spark standalone模式在kubernetes上运行，kubernetes不负责spark任务的调度。参考：Spark standalone on Kubernetes 这种模式中使用的spark本身负责任务调度，kubernetes只是作为一个spark的部署平台。 Spark on Yarn 使用StatefulSet和Headless serverless来实现，请参考 Spark on Yarn 这种模式中kubernetes依然不负责spark应用的调度，而只是将Yarn换了一个部署环境而已。 下面是架构图： 图片 - Spark on yarn with kubernetes Spark on Kubernetes Spark on kubernetes，使用kubernetes作为调度引擎，spark的任务直接调度到node节点上。参考：运行支持kubernetes原生调度的Spark程序 调度方式总结 下图显示的是三种调度方式中单个kubernetes node节点上运行的spark相关容器的调度情况。 图片 - 在kubernetes上使用多种调度方式 毫无疑问，使用kubernetes原生调度的spark任务才是最节省资源的。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-06 14:45:35 "},"usecases/spark-standalone-on-kubernetes.html":{"url":"usecases/spark-standalone-on-kubernetes.html","title":"Spark standalone on Kubernetes","keywords":"","body":"Spark standalone on Kubernetes 该项目是基于 Spark standalone 模式，对资源的分配调度还有作业状态查询的功能实在有限，对于让 spark 使用真正原生的 kubernetes 资源调度推荐大家尝试 https://github.com/apache-spark-on-k8s/ 本文中使用的镜像我已编译好上传到了时速云上，大家可以直接下载。 index.tenxcloud.com/jimmy/spark:1.5.2_v1 index.tenxcloud.com/jimmy/zeppelin:0.7.1 代码和使用文档见Github地址：https://github.com/rootsongjc/spark-on-kubernetes 本文中用到的 yaml 文件可以在 ../manifests/spark-standalone 目录下找到，也可以在上面的 https://github.com/rootsongjc/spark-on-kubernetes/ 项目的 manifests 目录下找到。 注意：时速云上本来已经提供的镜像 index.tenxcloud.com/google_containers/spark:1.5.2_v1 ，但是该镜像似乎有问题，下载总是失败。 在Kubernetes上启动spark 创建名为spark-cluster的namespace，所有操作都在该namespace中进行。 所有yaml文件都在manifests目录下。 $ kubectl create -f manifests/ 将会启动一个拥有三个worker的spark集群和zeppelin。 同时在该namespace中增加ingress配置，将spark的UI和zeppelin页面都暴露出来，可以在集群外部访问。 该ingress后端使用traefik。 访问spark 通过上面对ingress的配置暴露服务，需要修改本机的/etc/hosts文件，增加以下配置，使其能够解析到上述service。 172.20.0.119 zeppelin.traefik.io 172.20.0.119 spark.traefik.io 172.20.0.119是我设置的VIP地址，VIP的设置和traefik的配置请查看kubernetes-handbook。 spark ui 访问http://spark.traefik.io 图片 - spark master ui zeppelin ui 访问http://zepellin.treafik.io Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-08-30 14:16:22 "},"usecases/running-spark-with-kubernetes-native-scheduler.html":{"url":"usecases/running-spark-with-kubernetes-native-scheduler.html","title":"运行支持kubernetes原生调度的Spark程序","keywords":"","body":"运行支持kubernetes原生调度的Spark程序 TL;DR 这个主题比较大，该开源项目也还在不断进行中，我单独做了一个 web 用来记录 spark on kubernetes 的研究和最新进展见: https://jimmysong.io/spark-on-k8s 我们之前就在 kubernetes 中运行过 standalone 方式的 spark 集群，见 Spark standalone on kubernetes。 目前运行支持 kubernetes 原生调度的 spark 程序由 Google 主导，目前运行支持 kubernetes 原生调度的 spark 程序由 Google 主导，fork 自 spark 的官方代码库，见https://github.com/apache-spark-on-k8s/spark/ ，属于Big Data SIG。 参与到该项目的公司有： Bloomberg Google Haiwen Hyperpilot Intel Palantir Pepperdata Red Hat 为何使用 spark on kubernetes 使用kubernetes原生调度的spark on kubernetes是对现有的spark on yarn/mesos的资源使用方式的革命性的改进，主要表现在以下几点： Kubernetes原生调度：不再需要二层调度，直接使用kubernetes的资源调度功能，跟其他应用共用整个kubernetes管理的资源池； 资源隔离，粒度更细：原先yarn中的queue在spark on kubernetes中已不存在，取而代之的是kubernetes中原生的namespace，可以为每个用户分别指定一个namespace，限制用户的资源quota； 细粒度的资源分配：可以给每个spark任务指定资源限制，实际指定多少资源就使用多少资源，因为没有了像yarn那样的二层调度（圈地式的），所以可以更高效和细粒度的使用资源； 监控的变革：因为做到了细粒度的资源分配，所以可以对用户提交的每一个任务做到资源使用的监控，从而判断用户的资源使用情况，所有的metric都记录在数据库中，甚至可以为每个用户的每次任务提交计量； 日志的变革：用户不再通过yarn的web页面来查看任务状态，而是通过pod的log来查看，可将所有的kuberentes中的应用的日志等同看待收集起来，然后可以根据标签查看对应应用的日志； 所有这些变革都可以让我们更高效的获取资源、更有效率的获取资源！ Spark 概念说明 Apache Spark 是一个围绕速度、易用性和复杂分析构建的大数据处理框架。最初在2009年由加州大学伯克利分校的AMPLab开发，并于2010年成为Apache的开源项目之一。 在 Spark 中包括如下组件或概念： Application：Spark Application 的概念和 Hadoop 中的 MapReduce 类似，指的是用户编写的 Spark 应用程序，包含了一个 Driver 功能的代码和分布在集群中多个节点上运行的 Executor 代码； Driver：Spark 中的 Driver 即运行上述 Application 的 main() 函数并且创建 SparkContext，其中创建 SparkContext 的目的是为了准备Spark应用程序的运行环境。在 Spark 中由 SparkContext 负责和 ClusterManager 通信，进行资源的申请、任务的分配和监控等；当 Executor 部分运行完毕后，Driver负责将SparkContext 关闭。通常用 SparkContext 代表 Driver； Executor：Application运行在Worker 节点上的一个进程，该进程负责运行Task，并且负责将数据存在内存或者磁盘上，每个Application都有各自独立的一批Executor。在Spark on Yarn模式下，其进程名称为CoarseGrainedExecutorBackend，类似于 Hadoop MapReduce 中的 YarnChild。一个 CoarseGrainedExecutorBackend 进程有且仅有一个 executor 对象，它负责将 Task 包装成 taskRunner，并从线程池中抽取出一个空闲线程运行 Task。每个 CoarseGrainedExecutorBackend 能并行运行 Task 的数量就取决于分配给它的 CPU 的个数了； Cluster Manager：指的是在集群上获取资源的外部服务，目前有： Standalone：Spark原生的资源管理，由Master负责资源的分配； Hadoop Yarn：由YARN中的ResourceManager负责资源的分配； Worker：集群中任何可以运行Application代码的节点，类似于YARN中的NodeManager节点。在Standalone模式中指的就是通过Slave文件配置的Worker节点，在Spark on Yarn模式中指的就是NodeManager节点； 作业（Job）：包含多个Task组成的并行计算，往往由Spark Action催生，一个JOB包含多个RDD及作用于相应RDD上的各种Operation； 阶段（Stage）：每个Job会被拆分很多组 Task，每组任务被称为Stage，也可称TaskSet，一个作业分为多个阶段，每一个stage的分割点是action。比如一个job是：（transformation1 -> transformation1 -> action1 -> transformation3 -> action2），这个job就会被分为两个stage，分割点是action1和action2。 任务（Task）： 被送到某个Executor上的工作任务； Context：启动spark application的时候创建，作为Spark 运行时环境。 Dynamic Allocation（动态资源分配）：一个配置选项，可以将其打开。从Spark1.2之后，对于On Yarn模式，已经支持动态资源分配（Dynamic Resource Allocation），这样，就可以根据Application的负载（Task情况），动态的增加和减少executors，这种策略非常适合在YARN上使用spark-sql做数据开发和分析，以及将spark-sql作为长服务来使用的场景。Executor 的动态分配需要在 cluster mode 下启用 \"external shuffle service\"。 动态资源分配策略：开启动态分配策略后，application会在task因没有足够资源被挂起的时候去动态申请资源，这意味着该application现有的executor无法满足所有task并行运行。spark一轮一轮的申请资源，当有task挂起或等待 spark.dynamicAllocation.schedulerBacklogTimeout (默认1s)时间的时候，会开始动态资源分配；之后会每隔 spark.dynamicAllocation.sustainedSchedulerBacklogTimeout (默认1s)时间申请一次，直到申请到足够的资源。每次申请的资源量是指数增长的，即1,2,4,8等。之所以采用指数增长，出于两方面考虑：其一，开始申请的少是考虑到可能application会马上得到满足；其次要成倍增加，是为了防止application需要很多资源，而该方式可以在很少次数的申请之后得到满足。 架构设计 关于 spark standalone 的局限性与 kubernetes native spark 架构之间的区别请参考 Anirudh Ramanathan 在 2016年10月8日提交的 issue Support Spark natively in Kubernetes #34377。 简而言之，spark standalone on kubernetes 有如下几个缺点： 无法对于多租户做隔离，每个用户都想给 pod 申请 node 节点可用的最大的资源。 Spark 的 master／worker 本来不是设计成使用 kubernetes 的资源调度，这样会存在两层的资源调度问题，不利于与 kuberentes 集成。 而 kubernetes native spark 集群中，spark 可以调用 kubernetes API 获取集群资源和调度。要实现 kubernetes native spark 需要为 spark 提供一个集群外部的 manager 可以用来跟 kubernetes API 交互。 调度器后台 使用 kubernetes 原生调度的 spark 的基本设计思路是将 spark 的 driver 和 executor 都放在 kubernetes 的 pod 中运行，另外还有两个附加的组件：ResourceStagingServer 和 KubernetesExternalShuffleService。 Spark driver 其实可以运行在 kubernetes 集群内部（cluster mode）可以运行在外部（client mode），executor 只能运行在集群内部，当有 spark 作业提交到 kubernetes 集群上时，调度器后台将会为 executor pod 设置如下属性： 使用我们预先编译好的包含 kubernetes 支持的 spark 镜像，然后调用 CoarseGrainedExecutorBackend main class 启动 JVM。 调度器后台为 executor pod 的运行时注入环境变量，例如各种 JVM 参数，包括用户在 spark-submit 时指定的那些参数。 Executor 的 CPU、内存限制根据这些注入的环境变量保存到应用程序的 SparkConf 中。 可以在配置中指定 spark 运行在指定的 namespace 中。 参考：Scheduler backend 文档 安装指南 我们可以直接使用官方已编译好的 docker 镜像来部署，下面是官方发布的镜像： 组件 镜像 Spark Driver Image kubespark/spark-driver:v2.1.0-kubernetes-0.3.1 Spark Executor Image kubespark/spark-executor:v2.1.0-kubernetes-0.3.1 Spark Initialization Image kubespark/spark-init:v2.1.0-kubernetes-0.3.1 Spark Staging Server Image kubespark/spark-resource-staging-server:v2.1.0-kubernetes-0.3.1 PySpark Driver Image kubespark/driver-py:v2.1.0-kubernetes-0.3.1 PySpark Executor Image kubespark/executor-py:v2.1.0-kubernetes-0.3.1 我将这些镜像放到了我的私有镜像仓库中了。 还需要安装支持 kubernetes 的 spark 客户端，在这里下载：https://github.com/apache-spark-on-k8s/spark/releases 根据使用的镜像版本，我下载的是 v2.1.0-kubernetes-0.3.1 运行 SparkPi 测试 我们将任务运行在 spark-cluster 的 namespace 中，启动 5 个 executor 实例。 ./bin/spark-submit \\ --deploy-mode cluster \\ --class org.apache.spark.examples.SparkPi \\ --master k8s://https://172.20.0.113:6443 \\ --kubernetes-namespace spark-cluster \\ --conf spark.executor.instances=5 \\ --conf spark.app.name=spark-pi \\ --conf spark.kubernetes.driver.docker.image=sz-pg-oam-docker-hub-001.tendcloud.com/library/kubespark-spark-driver:v2.1.0-kubernetes-0.3.1 \\ --conf spark.kubernetes.executor.docker.image=sz-pg-oam-docker-hub-001.tendcloud.com/library/kubespark-spark-executor:v2.1.0-kubernetes-0.3.1 \\ --conf spark.kubernetes.initcontainer.docker.image=sz-pg-oam-docker-hub-001.tendcloud.com/library/kubespark-spark-init:v2.1.0-kubernetes-0.3.1 \\ local:///opt/spark/examples/jars/spark-examples_2.11-2.1.0-k8s-0.3.1-SNAPSHOT.jar 关于该命令参数的介绍请参考：https://apache-spark-on-k8s.github.io/userdocs/running-on-kubernetes.html 注意： 该 jar 包实际上是 spark.kubernetes.executor.docker.image 镜像中的。 这时候提交任务运行还是失败，报错信息中可以看到两个问题： Executor 无法找到 driver pod 用户 system:serviceaccount:spark-cluster:defaul 没有权限获取 spark-cluster 中的 pod 信息。 提了个 issue Failed to run the sample spark-pi test using spark-submit on the doc #478 需要为 spark 集群创建一个 serviceaccount 和 clusterrolebinding： kubectl create serviceaccount spark --namespace spark-cluster kubectl create rolebinding spark-edit --clusterrole=edit --serviceaccount=spark-cluster:spark --namespace=spark-cluster 该 Bug 将在新版本中修复。 用户指南 编译 Fork 并克隆项目到本地： git clone https://github.com/rootsongjc/spark.git 编译前请确保你的环境中已经安装 Java8 和 Maven3。 ## 第一次编译前需要安装依赖 build/mvn install -Pkubernetes -pl resource-managers/kubernetes/core -am -DskipTests ## 编译 spark on kubernetes build/mvn compile -Pkubernetes -pl resource-managers/kubernetes/core -am -DskipTests ## 发布 dev/make-distribution.sh --tgz -Phadoop-2.7 -Pkubernetes 第一次编译和发布的过程耗时可能会比较长，请耐心等待，如果有依赖下载不下来，请自备梯子。 详细的开发指南请见：https://github.com/apache-spark-on-k8s/spark/blob/branch-2.2-kubernetes/resource-managers/kubernetes/README.md 构建镜像 使用该脚本来自动构建容器镜像：https://github.com/apache-spark-on-k8s/spark/pull/488 将该脚本放在 dist 目录下，执行： ./build-push-docker-images.sh -r sz-pg-oam-docker-hub-001.tendcloud.com/library -t v2.1.0-kubernetes-0.3.1-1 build ./build-push-docker-images.sh -r sz-pg-oam-docker-hub-001.tendcloud.com/library -t v2.1.0-kubernetes-0.3.1-1 push 注意：如果你使用的 MacOS，bash 的版本可能太低，执行改脚本将出错，请检查你的 bash 版本： bash --version GNU bash, version 3.2.57(1)-release (x86_64-apple-darwin16) Copyright (C) 2007 Free Software Foundation, Inc. 上面我在升级 bash 之前获取的版本信息，使用下面的命令升级 bash： brew install bash 升级后的 bash 版本为 4.4.12(1)-release (x86_64-apple-darwin16.3.0)。 编译并上传镜像到我的私有镜像仓库，将会构建出如下几个镜像： sz-pg-oam-docker-hub-001.tendcloud.com/library/spark-driver:v2.1.0-kubernetes-0.3.1-1 sz-pg-oam-docker-hub-001.tendcloud.com/library/spark-resource-staging-server:v2.1.0-kubernetes-0.3.1-1 sz-pg-oam-docker-hub-001.tendcloud.com/library/spark-init:v2.1.0-kubernetes-0.3.1-1 sz-pg-oam-docker-hub-001.tendcloud.com/library/spark-shuffle:v2.1.0-kubernetes-0.3.1-1 sz-pg-oam-docker-hub-001.tendcloud.com/library/spark-executor:v2.1.0-kubernetes-0.3.1-1 sz-pg-oam-docker-hub-001.tendcloud.com/library/spark-executor-py:v2.1.0-kubernetes-0.3.1-1 sz-pg-oam-docker-hub-001.tendcloud.com/library/spark-driver-py:v2.1.0-kubernetes-0.3.1-1 运行测试 在 dist/bin 目录下执行 spark-pi 测试： ./spark-submit \\ --deploy-mode cluster \\ --class org.apache.spark.examples.SparkPi \\ --master k8s://https://172.20.0.113:6443 \\ --kubernetes-namespace spark-cluster \\ --conf spark.kubernetes.authenticate.driver.serviceAccountName=spark \\ --conf spark.executor.instances=5 \\ --conf spark.app.name=spark-pi \\ --conf spark.kubernetes.driver.docker.image=sz-pg-oam-docker-hub-001.tendcloud.com/library/spark-driver:v2.1.0-kubernetes-0.3.1-1 \\ --conf spark.kubernetes.executor.docker.image=sz-pg-oam-docker-hub-001.tendcloud.com/library/spark-executor:v2.1.0-kubernetes-0.3.1-1 \\ --conf spark.kubernetes.initcontainer.docker.image=sz-pg-oam-docker-hub-001.tendcloud.com/library/spark-init:v2.1.0-kubernetes-0.3.1-1 \\ local:///opt/spark/examples/jars/spark-examples_2.11-2.2.0-k8s-0.4.0-SNAPSHOT.jar 详细的参数说明见：https://apache-spark-on-k8s.github.io/userdocs/running-on-kubernetes.html 注意：local:///opt/spark/examples/jars/spark-examples_2.11-2.2.0-k8s-0.4.0-SNAPSHOT.jar 文件是在 spark-driver 和 spark-executor 镜像里的，在上一步构建镜像时已经构建并上传到了镜像仓库中。 执行日志显示： 2017-09-14 14:59:01 INFO Client:54 - Waiting for application spark-pi to finish... 2017-09-14 14:59:01 INFO LoggingPodStatusWatcherImpl:54 - State changed, new state: pod name: spark-pi-1505372339796-driver namespace: spark-cluster labels: spark-app-selector -> spark-f4d3a5d3ad964a05a51feb6191d50357, spark-role -> driver pod uid: 304cf440-991a-11e7-970c-f4e9d49f8ed0 creation time: 2017-09-14T06:59:01Z service account name: spark volumes: spark-token-zr8wv node name: N/A start time: N/A container images: N/A phase: Pending status: [] 2017-09-14 14:59:01 INFO LoggingPodStatusWatcherImpl:54 - State changed, new state: pod name: spark-pi-1505372339796-driver namespace: spark-cluster labels: spark-app-selector -> spark-f4d3a5d3ad964a05a51feb6191d50357, spark-role -> driver pod uid: 304cf440-991a-11e7-970c-f4e9d49f8ed0 creation time: 2017-09-14T06:59:01Z service account name: spark volumes: spark-token-zr8wv node name: 172.20.0.114 start time: N/A container images: N/A phase: Pending status: [] 2017-09-14 14:59:01 INFO LoggingPodStatusWatcherImpl:54 - State changed, new state: pod name: spark-pi-1505372339796-driver namespace: spark-cluster labels: spark-app-selector -> spark-f4d3a5d3ad964a05a51feb6191d50357, spark-role -> driver pod uid: 304cf440-991a-11e7-970c-f4e9d49f8ed0 creation time: 2017-09-14T06:59:01Z service account name: spark volumes: spark-token-zr8wv node name: 172.20.0.114 start time: 2017-09-14T06:59:01Z container images: sz-pg-oam-docker-hub-001.tendcloud.com/library/spark-driver:v2.1.0-kubernetes-0.3.1-1 phase: Pending status: [ContainerStatus(containerID=null, image=sz-pg-oam-docker-hub-001.tendcloud.com/library/spark-driver:v2.1.0-kubernetes-0.3.1-1, imageID=, lastState=ContainerState(running=null, terminated=null, waiting=null, additionalProperties={}), name=spark-kubernetes-driver, ready=false, restartCount=0, state=ContainerState(running=null, terminated=null, waiting=ContainerStateWaiting(message=null, reason=ContainerCreating, additionalProperties={}), additionalProperties={}), additionalProperties={})] 2017-09-14 14:59:03 INFO LoggingPodStatusWatcherImpl:54 - State changed, new state: pod name: spark-pi-1505372339796-driver namespace: spark-cluster labels: spark-app-selector -> spark-f4d3a5d3ad964a05a51feb6191d50357, spark-role -> driver pod uid: 304cf440-991a-11e7-970c-f4e9d49f8ed0 creation time: 2017-09-14T06:59:01Z service account name: spark volumes: spark-token-zr8wv node name: 172.20.0.114 start time: 2017-09-14T06:59:01Z container images: sz-pg-oam-docker-hub-001.tendcloud.com/library/spark-driver:v2.1.0-kubernetes-0.3.1-1 phase: Running status: [ContainerStatus(containerID=docker://5c5c821c482a1e35552adccb567020532b79244392374f25754f0050e6cd4c62, image=sz-pg-oam-docker-hub-001.tendcloud.com/library/spark-driver:v2.1.0-kubernetes-0.3.1-1, imageID=docker-pullable://sz-pg-oam-docker-hub-001.tendcloud.com/library/spark-driver@sha256:beb92a3e3f178e286d9e5baebdead88b5ba76d651f347ad2864bb6f8eda26f94, lastState=ContainerState(running=null, terminated=null, waiting=null, additionalProperties={}), name=spark-kubernetes-driver, ready=true, restartCount=0, state=ContainerState(running=ContainerStateRunning(startedAt=2017-09-14T06:59:02Z, additionalProperties={}), terminated=null, waiting=null, additionalProperties={}), additionalProperties={})] 2017-09-14 14:59:12 INFO LoggingPodStatusWatcherImpl:54 - State changed, new state: pod name: spark-pi-1505372339796-driver namespace: spark-cluster labels: spark-app-selector -> spark-f4d3a5d3ad964a05a51feb6191d50357, spark-role -> driver pod uid: 304cf440-991a-11e7-970c-f4e9d49f8ed0 creation time: 2017-09-14T06:59:01Z service account name: spark volumes: spark-token-zr8wv node name: 172.20.0.114 start time: 2017-09-14T06:59:01Z container images: sz-pg-oam-docker-hub-001.tendcloud.com/library/spark-driver:v2.1.0-kubernetes-0.3.1-1 phase: Succeeded status: [ContainerStatus(containerID=docker://5c5c821c482a1e35552adccb567020532b79244392374f25754f0050e6cd4c62, image=sz-pg-oam-docker-hub-001.tendcloud.com/library/spark-driver:v2.1.0-kubernetes-0.3.1-1, imageID=docker-pullable://sz-pg-oam-docker-hub-001.tendcloud.com/library/spark-driver@sha256:beb92a3e3f178e286d9e5baebdead88b5ba76d651f347ad2864bb6f8eda26f94, lastState=ContainerState(running=null, terminated=null, waiting=null, additionalProperties={}), name=spark-kubernetes-driver, ready=false, restartCount=0, state=ContainerState(running=null, terminated=ContainerStateTerminated(containerID=docker://5c5c821c482a1e35552adccb567020532b79244392374f25754f0050e6cd4c62, exitCode=0, finishedAt=2017-09-14T06:59:11Z, message=null, reason=Completed, signal=null, startedAt=null, additionalProperties={}), waiting=null, additionalProperties={}), additionalProperties={})] 2017-09-14 14:59:12 INFO LoggingPodStatusWatcherImpl:54 - Container final statuses: Container name: spark-kubernetes-driver Container image: sz-pg-oam-docker-hub-001.tendcloud.com/library/spark-driver:v2.1.0-kubernetes-0.3.1-1 Container state: Terminated Exit code: 0 2017-09-14 14:59:12 INFO Client:54 - Application spark-pi finished. 从日志中可以看到任务运行的状态信息。 使用下面的命令可以看到 kubernetes 启动的 Pod 信息： kubectl --namespace spark-cluster get pods -w 将会看到 spark-driver 和 spark-exec 的 Pod 信息。 依赖管理 上文中我们在运行测试程序时，命令行中指定的 jar 文件已包含在 docker 镜像中，是不是说我们每次提交任务都需要重新创建一个镜像呢？非也！如果真是这样也太麻烦了。 创建 resource staging server 为了方便用户提交任务，不需要每次提交任务的时候都创建一个镜像，我们使用了 resource staging server 。 kubectl create -f conf/kubernetes-resource-staging-server.yaml 我们同样将其部署在 spark-cluster namespace 下，该 yaml 文件见 kubernetes-handbook 的 manifests/spark-with-kubernetes-native-scheduler 目录。 优化 其中有一点需要优化，在使用下面的命令提交任务时，使用 --conf spark.kubernetes.resourceStagingServer.uri 参数指定 resource staging server 地址，用户不应该关注 resource staging server 究竟运行在哪台宿主机上，可以使用下面两种方式实现： 使用 nodeSelector 将 resource staging server 固定调度到某一台机器上，该地址依然使用宿主机的 IP 地址 改变 spark-resource-staging-service service 的 type 为 ClusterIP， 然后使用 Ingress 将其暴露到集群外部，然后加入的内网 DNS 里，用户使用 DNS 名称指定 resource staging server 的地址。 然后可以执行下面的命令来提交本地的 jar 到 kubernetes 上运行。 ./spark-submit \\ --deploy-mode cluster \\ --class org.apache.spark.examples.SparkPi \\ --master k8s://https://172.20.0.113:6443 \\ --kubernetes-namespace spark-cluster \\ --conf spark.kubernetes.authenticate.driver.serviceAccountName=spark \\ --conf spark.executor.instances=5 \\ --conf spark.app.name=spark-pi \\ --conf spark.kubernetes.driver.docker.image=sz-pg-oam-docker-hub-001.tendcloud.com/library/spark-driver:v2.1.0-kubernetes-0.3.1-1 \\ --conf spark.kubernetes.executor.docker.image=sz-pg-oam-docker-hub-001.tendcloud.com/library/spark-executor:v2.1.0-kubernetes-0.3.1-1 \\ --conf spark.kubernetes.initcontainer.docker.image=sz-pg-oam-docker-hub-001.tendcloud.com/library/spark-init:v2.1.0-kubernetes-0.3.1-1 \\ --conf spark.kubernetes.resourceStagingServer.uri=http://172.20.0.114:31000 \\ ../examples/jars/spark-examples_2.11-2.2.0-k8s-0.4.0-SNAPSHOT.jar 该命令将提交本地的 ../examples/jars/spark-examples_2.11-2.2.0-k8s-0.4.0-SNAPSHOT.jar 文件到 resource staging server，executor 将从该 server 上获取 jar 包并运行，这样用户就不需要每次提交任务都编译一个镜像了。 详见：https://apache-spark-on-k8s.github.io/userdocs/running-on-kubernetes.html#dependency-management 设置 HDFS 用户 如果 Hadoop 集群没有设置 kerbros 安全认证的话，在指定 spark-submit 的时候可以通过指定如下四个环境变量， 设置 Spark 与 HDFS 通信使用的用户： --conf spark.kubernetes.driverEnv.SPARK_USER=hadoop --conf spark.kubernetes.driverEnv.HADOOP_USER_NAME=hadoop --conf spark.executorEnv.HADOOP_USER_NAME=hadoop --conf spark.executorEnv.SPARK_USER=hadoop 使用 hadoop 用户提交本地 jar 包的命令示例： ./spark-submit \\ --deploy-mode cluster \\ --class com.talkingdata.alluxio.hadooptest \\ --master k8s://https://172.20.0.113:6443 \\ --kubernetes-namespace spark-cluster \\ --conf spark.kubernetes.driverEnv.SPARK_USER=hadoop \\ --conf spark.kubernetes.driverEnv.HADOOP_USER_NAME=hadoop \\ --conf spark.executorEnv.HADOOP_USER_NAME=hadoop \\ --conf spark.executorEnv.SPARK_USER=hadoop \\ --conf spark.kubernetes.authenticate.driver.serviceAccountName=spark \\ --conf spark.executor.instances=5 \\ --conf spark.app.name=spark-pi \\ --conf spark.kubernetes.driver.docker.image=sz-pg-oam-docker-hub-001.tendcloud.com/library/spark-driver:v2.1.0-kubernetes-0.3.1-1 \\ --conf spark.kubernetes.executor.docker.image=sz-pg-oam-docker-hub-001.tendcloud.com/library/spark-executor:v2.1.0-kubernetes-0.3.1-1 \\ --conf spark.kubernetes.initcontainer.docker.image=sz-pg-oam-docker-hub-001.tendcloud.com/library/spark-init:v2.1.0-kubernetes-0.3.1-1 \\ --conf spark.kubernetes.resourceStagingServer.uri=http://172.20.0.114:31000 \\ ~/Downloads/tendcloud_2.10-1.0.jar 详见：https://github.com/apache-spark-on-k8s/spark/issues/408 限制 Driver 和 Executor 的资源使用 在执行 spark-submit 时使用如下参数设置内存和 CPU 资源限制： --conf spark.driver.memory=3G --conf spark.executor.memory=3G --conf spark.driver.cores=2 --conf spark.executor.cores=10 这几个参数中值如何传递到 Pod 的资源设置中的呢？ 比如我们设置在执行 spark-submit 的时候传递了这样的两个参数：--conf spark.driver.cores=2 和 --conf spark.driver.memory=100G 那么查看 driver pod 的 yaml 输出结果将会看到这样的资源设置： resources: limits: memory: 110Gi requests: cpu: \"2\" memory: 100Gi 以上参数是对 request 值的设置，那么 limit 的资源设置的值又是从何而来？ 可以使用 spark.kubernetes.driver.limit.cores 和 spark.kubernetes.executor.limit.cores 来设置 CPU的 hard limit。 memory limit 的值是根据 memory request 的值加上 spark.kubernetes.executor.memoryOverhead 的值计算而来的，该配置项用于设置分配给每个 executor 的超过 heap 内存的值（可以使用k、m、g单位）。该值用于虚拟机的开销、其他本地服务开销。根据 executor 的大小设置（通常是 6%到10%）。 我们可以这样来提交一个任务，同时设置 driver 和 executor 的 CPU、内存的资源 request 和 limit 值（driver 的内存 limit 值为 request 值的 110%）。 ./spark-submit \\ --deploy-mode cluster \\ --class org.apache.spark.examples.SparkPi \\ --master k8s://https://172.20.0.113:6443 \\ --kubernetes-namespace spark-cluster \\ --conf spark.kubernetes.authenticate.driver.serviceAccountName=spark \\ --conf spark.driver.memory=100G \\ --conf spark.executor.memory=10G \\ --conf spark.driver.cores=30 \\ --conf spark.executor.cores=2 \\ --conf spark.driver.maxResultSize=10240m \\ --conf spark.kubernetes.driver.limit.cores=32 \\ --conf spark.kubernetes.executor.limit.cores=3 \\ --conf spark.kubernetes.executor.memoryOverhead=2g \\ --conf spark.executor.instances=5 \\ --conf spark.app.name=spark-pi \\ --conf spark.kubernetes.driver.docker.image=sz-pg-oam-docker-hub-001.tendcloud.com/library/spark-driver:v2.1.0-kubernetes-0.3.1-1 \\ --conf spark.kubernetes.executor.docker.image=sz-pg-oam-docker-hub-001.tendcloud.com/library/spark-executor:v2.1.0-kubernetes-0.3.1-1 \\ --conf spark.kubernetes.initcontainer.docker.image=sz-pg-oam-docker-hub-001.tendcloud.com/library/spark-init:v2.1.0-kubernetes-0.3.1-1 \\ local:///opt/spark/examples/jars/spark-examples_2.11-2.2.0-k8s-0.4.0-SNAPSHOT.jar 10000000 这将启动一个包含一千万个 task 的计算 pi 的 spark 任务，任务运行过程中，drvier 的 CPU 实际消耗大约为 3 核，内存 40G，每个 executor 的 CPU 实际消耗大约不到 1 核，内存不到 4G，我们可以根据实际资源消耗不断优化资源的 request 值。 SPARK_DRIVER_MEMORY 和 SPARK_EXECUTOR_MEMORY 和分别作为 Driver 容器和 Executor 容器启动的环境变量，比如下面这个 Driver 启动的 CMD 中： CMD SPARK_CLASSPATH=\"${SPARK_HOME}/jars/*\" && \\ env | grep SPARK_JAVA_OPT_ | sed 's/[^=]*=\\(.*\\)/\\1/g' > /tmp/java_opts.txt && \\ readarray -t SPARK_DRIVER_JAVA_OPTS 我们可以看到对 SPARK_DRIVER_MEMORY 环境变量的引用。Executor 的设置与 driver 类似。 而我们可以使用这样的参数来传递环境变量的值 spark.executorEnv.[EnvironmentVariableName]，只要将 EnvironmentVariableName 替换为环境变量名称即可。 参考 Spark动态资源分配-Dynamic Resource Allocation Running Spark on Kubernetes Apache Spark Jira Issue - 18278 - SPIP: Support native submission of spark jobs to a kubernetes cluster Kubernetes Github Issue - 34377 Support Spark natively in Kubernetes Kubernetes example spark https://github.com/rootsongjc/spark-on-kubernetes Scheduler backend Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-09-27 21:03:00 "},"usecases/serverless.html":{"url":"usecases/serverless.html","title":"Serverless架构","keywords":"","body":"Serverless架构 Serverless（无服务器架构）指的是由开发者实现的服务端逻辑运行在无状态的计算容器中，它由事件触发， 完全被第三方管理，其业务层面的状态则被开发者使用的数据库和存储资源所记录。 下图来自谷歌云平台官网，是对云计算的一个很好的分层概括，其中 serverless 就是构建在虚拟机和容器之上的一层，与应用本身的关系更加密切。 图片 - 从物理机到函数计算 Serverless架构的优点 今天大多数公司在开发应用程序并将其部署在服务器上的时候，无论是选择公有云还是私有的数据中心，都需要提前了解究竟需要多少台服务器、多大容量的存储和数据库的功能等。并需要部署运行应用程序和依赖的软件到基础设施之上。假设我们不想在这些细节上花费精力，是否有一种简单的架构模型能够满足我们这种想法？这个答案已经存在，这就是今天软件架构世界中新鲜但是很热门的一个话题——Serverless（无服务器）架构。 ——AWS 费良宏 降低运营成本： Serverless是非常简单的外包解决方案。它可以让您委托服务提供商管理服务器、数据库和应用程序甚至逻辑，否则您就不得不自己来维护。由于这个服务使用者的数量会非常庞大，于是就会产生规模经济效应。在降低成本上包含了两个方面，即基础设施的成本和人员（运营/开发）的成本。 降低开发成本： IaaS和PaaS存在的前提是，服务器和操作系统管理可以商品化。Serverless作为另一种服务的结果是整个应用程序组件被商品化。 扩展能力： Serverless架构一个显而易见的优点即“横向扩展是完全自动的、有弹性的、且由服务提供者所管理”。从基本的基础设施方面受益最大的好处是，您只需支付您所需要的计算能力。 更简单的管理： Serverless架构明显比其他架构更简单。更少的组件，就意味着您的管理开销会更少。 “绿色”的计算： 按照《福布斯》杂志的统计，在商业和企业数据中心的典型服务器仅提供5%～15%的平均最大处理能力的输出。这无疑是一种资源的巨大浪费。随着Serverless架构的出现，让服务提供商提供我们的计算能力最大限度满足实时需求。这将使我们更有效地利用计算资源。 Kubernetes上的serverless 架构 目前已经有一批优秀的基于 kubernetes 的 serverless 架构（FaaS）开源项目如下： faas - 🐳 Functions as a Service - a serverless framework for Docker & Kubernetes https://blog.alexellis.io/introducing… faas-netes - Enable Kubernetes as a backend for Functions as a Service (OpenFaaS) https://github.com/alexellis/faas fn - The container native, cloud agnostic serverless platform. http://fnproject.io funktion - a CLI tool for working with funktion https://funktion.fabric8.io/ fx - Poor man's serverless framework based on Docker, Function as a Service with painless. IronFunctions - IronFunctions - the serverless microservices platform. http://iron.io kubeless - Kubernetes Native Serverless Framework http://kubeless.io OpenWhisk - Apache OpenWhisk (Incubating) is a serverless, open source cloud platform that executes functions in response to events at any scale. 以上项目收录于 awsome-cloud-native FaaS Function-as-a-Service 全景图（图片来自https://github.com/amyers1793/FunctionasaServiceLandscape) 图片 - FaaS Landscape 参考 Serverless Architectures - Martin Fowler Serverless架构综述 2017年会是Serverless爆发之年吗？ 从IaaS到FaaS—— Serverless架构的前世今生 Introducing Redpoint's FaaS Landscape Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-21 18:11:54 "},"usecases/understanding-serverless.html":{"url":"usecases/understanding-serverless.html","title":"理解Serverless","keywords":"","body":"理解Serverless No silver bullet. - The Mythical Man-Month 许多年前，我们开发的软件还是C/S（客户端/服务器）和MVC（模型-试图-控制器）的形式，再后来有了SOA，最近几年又出现了微服务架构，更新一点的有Cloud Native（云原生）应用，企业应用从单体架构，到服务化，再到更细粒度的微服务化，应用开发之初就是为了应对互联网的特有的高并发、不间断的特性，需要很高的性能和可扩展性，人们对软件开发的追求孜孜不倦，希望力求在软件开发的复杂度和效率之间达到一个平衡。但可惜的是，NO SILVER BULLET！几十年前（1975年）Fred Brooks就在The Mythical Man-Month中就写到了这句话。那么Serverlss会是那颗银弹吗？ 云改变了我们对操作系统的认知，原来一个系统的计算资源、存储和网络是可以分离配置的，而且还可以弹性扩展，但是长久以来，我们在开发应用时始终没有摆脱的服务器的束缚（或者说认知），应用必须运行在不论是实体还是虚拟的服务器上，必须经过部署、配置、初始化才可以运行，还需要对服务器和应用进行监控和管理，还需要保证数据的安全性，这些云能够帮我们简化吗？让我们只要关注自己代码的逻辑就好了，其它的东西让云帮我实现就好了。 Serverless介绍 Serverless架构是云的自然延伸，为了理解serverless，我们有必要回顾一下云计算的发展。 IaaS 2006年AWS推出EC2（Elastic Compute Cloud），作为第一代IaaS（Infrastructure as a Service），用户可以通过AWS快速的申请到计算资源，并在上面部署自己的互联网服务。IaaS从本质上讲是服务器租赁并提供基础设施外包服务。就比如我们用的水和电一样，我们不会自己去引入自来水和发电，而是直接从自来水公司和电网公司购入，并根据实际使用付费。 EC2真正对IT的改变是硬件的虚拟化（更细粒度的虚拟化），而EC2给用户带来了以下五个好处： 降低劳动力成本：减少了企业本身雇佣IT人员的成本 降低风险：不用再像自己运维物理机那样，担心各种意外风险，EC2有主机损坏，再申请一个就好了。 降低基础设施成本：可以按小时、周、月或者年为周期租用EC2。 扩展性：不必过早的预期基础设施采购，因为通过云厂商可以很快的获取。 节约时间成本：快速的获取资源开展业务实验。 以上说了是IaaS或者说基础设施外包的好处，当然其中也有弊端，我们将在后面讨论。 以上是AWS为代表的公有云IaaS，还有使用OpenStack构建的私有云也能够提供IaaS能力。 PaaS PaaS（Platform as a Service）是构建在IaaS之上的一种平台服务，提供操作系统安装、监控和服务发现等功能，用户只需要部署自己的应用即可，最早的一代是Heroku。Heroko是商业的PaaS，还有一个开源的PaaS——Cloud Foundry，用户可以基于它来构建私有PaaS，如果同时使用公有云和私有云，如果能在两者之间构建一个统一的PaaS，那就是“混合云”了。 在PaaS上最广泛使用的技术就要数docker了，因为使用容器可以很清晰的描述应用程序，并保证环境一致性。管理云上的容器，可以称为是CaaS（Container as a Service），如GCE（Google Container Engine）。也可以基于Kubernetes、Mesos这类开源软件构件自己的CaaS，不论是直接在IaaS构建还是基于PaaS。 PaaS是对软件的一个更高的抽象层次，已经接触到应用程序的运行环境本身，可以由开发者自定义，而不必接触更底层的操作系统。 Serverless的定义 Serverless不如IaaS和PaaS那么好理解，因为它通常包含了两个领域BaaS（Backend as a Service）和FaaS（Function as a Service）。 BaaS BaaS（Backend as a Service）后端即服务，一般是一个个的API调用后端或别人已经实现好的程序逻辑，比如身份验证服务Auth0，这些BaaS通常会用来管理数据，还有很多公有云上提供的我们常用的开源软件的商用服务，比如亚马逊的RDS可以替代我们自己部署的MySQL，还有各种其它数据库和存储服务。 FaaS FaaS（Functions as a Service）函数即服务，FaaS是无服务器计算的一种形式，当前使用最广泛的是AWS的Lambada。 现在当大家讨论Serverless的时候首先想到的就是FaaS，有点甚嚣尘上了。FaaS本质上是一种事件驱动的由消息触发的服务，FaaS供应商一般会集成各种同步和异步的事件源，通过订阅这些事件源，可以突发或者定期的触发函数运行。 图片 - 服务端软件的运行环境 传统的服务器端软件不同是经应用程序部署到拥有操作系统的虚拟机或者容器中，一般需要长时间驻留在操作系统中运行，而FaaS是直接将程序部署上到平台上即可，当有事件到来时触发执行，执行完了就可以卸载掉。 图片 - FaaS应用架构 总结 两者都为我们的计算资源提供了弹性的保障，BaaS其实依然是服务外包，而FaaS使我们更加关注应用程序的逻辑，两者使我们不需要关注应用程序所在的服务器，但实际上服务器依然是客观存在的。 当我们将应用程序迁移到容器和虚拟机中时，其实对于应用程序本身的体系结构并没有多少改变，只不过有些流程和规定需要遵守，比如12因素应用守则，但是serverlss对应用程序的体系结构来说就是一次颠覆了，通常我们需要考虑事件驱动模型，更加细化的不熟形式，以及在FaaS组件之外保持状态的需求。 Serverless应用 我们以一个游戏应用为例，来说明什么是serverless应用。 一款移动端游戏至少包含如下几个特性： 移动端友好的用户体验 用户管理和权限认证 关卡、升级等游戏逻辑，游戏排行，玩家的等级、任务等信息 传统的应用程序架构可能是这样的： 图片 - 传统应用程序架构 一个app前端，iOS后者安卓 用Java写的后端，使用JBoss或者Tomcat做server运行 使用关系型数据库存储用户数据，如MySQL 这样的架构可以让前端十分轻便，不需要做什么应用逻辑，只是负责渲染用户界面，将请求通过HTTP发送给后端，而所有的数据操作都是有由后端的Java程序来完成的。 这样的架构开发起来比较容易，但是维护起来确十分复杂，前端开发、后端的开发都需要十分专业的人员、环境的配置，还要有人专门维护数据库、应用的更新和升级。 图片 - Serverless架构 而在serverless架构中，我们不再需要在服务器端代码中存储任何会话状态，而是直接将它们存储在NoSQL中，这样将使应用程序无状态，有助于弹性扩展。前端可以直接利用BaaS而减少后端的编码需求，这样架构的本质上是减少了应用程序开发的人力成本，降低了自己维护基础设施的风险，而且利用云的能力更便于扩展和快速迭代。 Serverless的优势 在最前面我们提到了使用IaaS给我们带来了五点好处，FaaS当然也包括了这些好处，但是它给我们带来的最大的好处就是多快好省。减少从概念原型到实施的等待时间，比自己维护服务更省钱。 降低人力成本 不需要再自己维护服务器，操心服务器的各种性能指标和资源利用率，而是关心应用程序本身的状态和逻辑。而且serverless应用本身的部署也十分容易，我们只要上传基本的代码但愿，例如Javascript或Python的源代码的zip文件，以及基于JVM的语言的纯JAR文件。不需使用Puppet、Chef、Ansible或Docker来进行配置管理，降低了运维成本。同时，对于运维来说，也不再需要监控那些更底层的如磁盘使用量、CPU使用率等底层和长期的指标信息，而是监控应用程序本身的度量，这将更加直观和有效。 在此看来有人可能会提出“NoOps”的说法，其实这是不存在的，只要有应用存在的一天就会有Ops，只是人员的角色会有所转变，部署将变得更加自动化，监控将更加面向应用程序本身，更底层的运维依然需要专业的人员去做。 降低风险 对于组件越多越复杂的系统，出故障的风险就越大。我们使用BaaS或FaaS将它们外包出去，让专业人员来处理这些故障，有时候比我们自己来修复更可靠，利用专业人员的知识来降低停机的风险，缩短故障修复的时间，让我们的系统稳定性更高。 减少资源开销 我们在申请主机资源一般会评估一个峰值最大开销来申请资源，往往导致过度的配置，这意味着即使在主机闲置的状态下也要始终支付峰值容量的开销。对于某些应用来说这是不得已的做法，比如数据库这种很难扩展的应用，而对于普通应用这就显得不太合理了，虽然我们都觉得即使浪费了资源也比当峰值到来时应用程序因为资源不足而挂掉好。 解决这个问题最好的办法就是，不计划到底需要使用多少资源，而是根据实际需要来请求资源，当然前提必须是整个资源池是充足的（公有云显然更适合）。根据使用时间来付费，根据每次申请的计算资源来付费，让计费的粒度更小，将更有利于降低资源的开销。这是对应用程序本身的优化，例如让每次请求耗时更短，让每次消耗的资源更少将能够显著节省成本。 增加缩放的灵活性 以AWS Lamba为例，当平台接收到第一个触发函数的事件时，它将启动一个容器来运行你的代码。如果此时收到了新的事件，而第一个容器仍在处理上一个事件，平台将启动第二个代码实例来处理第二个事件。AWS lambad的这种自动的零管理水平缩放，将持续到有足够的代码实例来处理所有的工作负载。 但是，AWS仍然只会向您收取代码的执行时间，无论它需要启动多少个容器实例要满足你的负载请求。例如，假设所有事件的总执行时间是相同的，在一个容器中按顺序调用Lambda 100次与在100个不同容器中同时调用100次Lambda的成本是 一样的。当然AWS Lambada也不会无限制的扩展实例个数，如果有人对你发起了DDos攻击怎么办，那么不就会产生高昂的成本吗？AWS是有默认限制的，默认执行Lambada函数最大并发数是1000。 缩短创新周期 小团队的开发人员正可以在几天之内从头开始开发应用程序并部署到生产。使用短而简单的函数和事件来粘合强大的驱动数据存储和服务的API。完成的应用程序具有高度可用性和可扩展性，利用率高，成本低，部署速度快。 以docker为代表的容器技术仅仅是缩短了应用程序的迭代周期，而serverless技术是直接缩短了创新周期，从概念到最小可行性部署的时间，让初级开发人员也能在很短的时间内完成以前通常要经验丰富的工程师才能完成的项目。 Serverless的劣势 我们知道没有十全十美的技术，在说了serverless的那么多优势之后，我们再来探讨以下serverless的劣势，或者说局限性和适用场景。 状态管理 要想实现自由的缩放，无状态是必须的，而对于有状态的服务，使用serverless这就丧失了灵活性，有状态服务需要与存储交互就不可避免的增加了延迟和复杂性。 延迟 应用程序中不同组件的访问延迟是一个大问题，我们可以通过使用专有的网络协议、RPC调用、数据格式来优化，或者是将实例放在同一个机架内或同一个主机实例上来优化以减少延迟。 而serverless应用程序是高度分布式、低耦合的，这就意味着延迟将始终是一个问题，单纯使用serverless的应用程序是不太现实的。 本地测试 Serverless应用的本地测试困难是一个很棘手的问题。虽然可以在测试环境下使用各种数据库和消息队列来模拟生产环境，但是对于无服务应用的集成或者端到端测试尤其困难，很难在本地模拟应用程序的各种连接，并与性能和缩放的特性结合起来测试，并且serverless应用本身也是分布式的，简单的将无数的FaaS和BaaS组件粘合起来也是有挑战性的。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-19 20:06:23 "},"usecases/faas.html":{"url":"usecases/faas.html","title":"FaaS-函数即服务","keywords":"","body":"FaaS-函数即服务 FaaS（Functions as a Service）函数即服务，FaaS是无服务器计算的一种形式，当前使用最广泛的是AWS的Lambada。 现在当大家讨论Serverless的时候首先想到的就是FaaS，有点甚嚣尘上了。FaaS本质上是一种事件驱动的由消息触发的服务，FaaS供应商一般会集成各种同步和异步的事件源，通过订阅这些事件源，可以突发或者定期的触发函数运行。 当前开源的FaaS框架大部分都是基于Kuberentes来实现的，例如： faas-netes - Enable Kubernetes as a backend for Functions as a Service (OpenFaaS) https://github.com/alexellis/faas fn - The container native, cloud agnostic serverless platform. http://fnproject.io funktion - a CLI tool for working with funktion https://funktion.fabric8.io/ fx - Poor man's serverless framework based on Docker, Function as a Service with painless. IronFunctions - IronFunctions - the serverless microservices platform. http://iron.io kubeless - Kubernetes Native Serverless Framework http://kubeless.io OpenFaaS - OpenFaaS - Serverless Functions Made Simple for Docker & Kubernetes https://blog.alexellis.io/introducing-functions-as-a-service/ OpenWhisk - Apache OpenWhisk (Incubating) is a serverless, open source cloud platform that executes functions in response to events at any scale. 关于整个Cloud Native开源生态，请参考awesome-cloud-native。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-19 20:06:08 "},"usecases/openfaas.html":{"url":"usecases/openfaas.html","title":"OpenFaaS","keywords":"","body":"OpenFaaS OpenFaaS一款高人气的开源的faas框架，可以在 在Kubernetes上部署OpenFaaS十分简单，用到的镜像如下： functions/faas-netesd:0.3.4 functions/gateway:0.6.14 functions/prometheus:latest-k8s functions/alertmanager:latest-k8s 这些镜像都存储在DockerHub上。 OpenFaaS的架构如下图： 图片 - OpenFaaS架构 部署 参考Deployment guide for Kubernetes部署OpenFaaS。 如果您的Kuberentes集群可以访问DockerHub那么直接使用官方提供的YAML文件即可。 YAML文件见官方仓库：https://github.com/openfaas/faas-netes 部署同步请求 一共用到了三个YAML文件： faas.yml monitoring.yml rbac.yml OpenFaaS安装好后会启动一个Prometheus。 图片 - OpenFaaS Prometheus 在这里可以看到函数的运行情况。 同时OpenFaaS还提供了Dashboard，需要我们自己向Grafana中配置Prometheus数据源后导入，JSON配置见：https://grafana.com/dashboards/3526，可以下载后直接导入到Grafana中。 图片 - OpenFaas Grafana监控 OpenFaaS的使用 OpenFaaS提供了便捷的UI，在部署完成后就可以通过NodePort方式访问。 图片 - OpenFaas Dashboard 其中已经内置了一些函数应用可供我们试用，还可以创建自己的函数。 比如内置的NodeInfo应用，可以获取该应用所部署到的主机的信息，如下图： 图片 - NodeInfo执行结果 注意：有一些js和css文件需要翻墙才能访问，否则页面将出现格式错误。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-19 22:37:24 "},"usecases/edge-computing.html":{"url":"usecases/edge-computing.html","title":"边缘计算","keywords":"","body":"边缘计算 TBD 参考 The Birth of an Edge Orchestrator – Cloudify Meets Edge Computing K8s(Kubernetes) and SDN for Multi-access Edge Computing deployment Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-10-27 19:19:45 "},"develop/":{"url":"develop/","title":"开发指南概览","keywords":"","body":"开发指南说明 讲解如何在原生 Kubernetes 的基础上做定制开发。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-15 19:44:17 "},"develop/sigs-and-working-group.html":{"url":"develop/sigs-and-working-group.html","title":"SIG和工作组","keywords":"","body":"SIG和工作组 Kubernetes的社区是以SIG（Special Interest Group特别兴趣小组）和工作组的形式组织起来的，每个工作组都会定期召开视频会议。 所有的SIG和工作组都使用slack和邮件列表沟通。 主要SIG列表 api-machinery：所有API级别的功能，包括了API server、API注册和发现、通用的API CRUD语义，准入控制，编码/解码，转换，默认值，持久化层（etcd），OpenAPI，第三方资源，垃圾手机和客户端库的方方面面。 aws：如何在AWS上支持和使用kubernetes。 apps：在kubernetes上部署和运维应用程序。关注开发者和DevOps在kubernetes上运行应用程序的体验。 architecture：维持kubernetes在架构设计上的一致性和原则。 auth：kubernetes的认证授权、权限管理和安全性策略。 autoscaling：集群的自动缩放，pod的水平和垂直自动缩放，pod的资源初始化，pod监控和指标收集等主题。 azure：如何在Azure上支持和使用kubernetes。 big-data：在kubernetes上部署和运行大数据应用，如Spark、Kafka、Hadoop、Flink、Storm等。 CLI：kubectl和相关工具。 cluster-lifecycle：部署和升级kubernetes集群。 cluster-ops：促进kubernetes集群本身的可操作性和集群间的互操作性，使不同的运营商之间协调一致。 contributor-experience：维持良好的开发者社区。 docs：文档，流程和出版物。 GCP：在Google Cloud Platform上使用kubernetes。 instrumentation：集群可观测性的最佳实践，包括指标设置、日志收集、事件等。 multicluster：多kubernetes集群的用例和工具。 network：kubernetes集群的网络。 node：node节点、kubelet。 onprem：在非云供应商的环境下运行kubernetes，例如on premise、裸机等环境。 openstack：协调跨OpenStack和Kubernetes社区的努力。 product-management：侧重于产品管理方面。 release：发布、PR和bug提交等。 scalability：负责回答可伸缩性相关的问题。 scheduling：资源调度。 service-catalog：为CNCF service broker和Kubernetes broker实现开发API。 storage：存储和volume插件。 testing：测试。 ui：与UI相关的话题。 windows：在kubernets上运行Windows Server Container。 工作组列表 App Def：改进API中的声明性原语、客户端库、工具的用户体验。 Cloud Provider：云供应商工作组。 Cluster API：定义一个代表Kubernetes集群的可移植API。 API将包含控制平面及其配置和底层基础设施（节点，节点池等）。 Container Identity：确保容器能够获得安全的身份认证并与外部连通的解决方案。 Kubeadm Adoption：提高kubeadm工具的采用率。 Resource ManagementJ：资源隔离和提高资源利用率。 详细信息请参考 https://github.com/kubernetes/community/blob/master/sig-list.md Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-03 13:11:06 "},"develop/developing-environment.html":{"url":"develop/developing-environment.html","title":"开发环境搭建","keywords":"","body":"配置Kubernetes开发环境 我们将在Mac上使用docker环境编译kuberentes。 安装依赖 brew install gnu-tar Docker环境，至少需要给容器分配4G内存，在低于3G内存的时候可能会编译失败。 执行编译 切换目录到kuberentes源码的根目录下执行： ./build/run.sh make可以在docker中执行跨平台编译出二进制文件。 需要用的的docker镜像： gcr.io/google_containers/kube-cross:v1.7.5-2 我将该镜像备份到时速云上了，可供大家使用： index.tenxcloud.com/jimmy/kube-cross:v1.7.5-2 该镜像基于Ubuntu构建，大小2.15G，编译环境中包含以下软件： Go1.7.5 etcd protobuf g++ 其他golang依赖包 在我自己的电脑上的整个编译过程大概要半个小时。 编译完成的二进制文件在/_output/local/go/bin/目录下。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-08-21 18:23:34 "},"develop/testing.html":{"url":"develop/testing.html","title":"单元测试和集成测试","keywords":"","body":"Kubernetes测试 单元测试 单元测试仅依赖于源代码，是测试代码逻辑是否符合预期的最简单方法。 运行所有的单元测试 make test 仅测试指定的package # 单个package make test WHAT=./pkg/api # 多个packages make test WHAT=./pkg/{api,kubelet} 或者，也可以直接用go test go test -v k8s.io/kubernetes/pkg/kubelet 仅测试指定package的某个测试case # Runs TestValidatePod in pkg/api/validation with the verbose flag set make test WHAT=./pkg/api/validation KUBE_GOFLAGS=\"-v\" KUBE_TEST_ARGS='-run ^TestValidatePod$' # Runs tests that match the regex ValidatePod|ValidateConfigMap in pkg/api/validation make test WHAT=./pkg/api/validation KUBE_GOFLAGS=\"-v\" KUBE_TEST_ARGS=\"-run ValidatePod\\|ValidateConfigMap$\" 或者直接用go test go test -v k8s.io/kubernetes/pkg/api/validation -run ^TestValidatePod$ 并行测试 并行测试是root out flakes的一种有效方法： # Have 2 workers run all tests 5 times each (10 total iterations). make test PARALLEL=2 ITERATION=5 生成测试报告 make test KUBE_COVER=y Benchmark测试 go test ./pkg/apiserver -benchmem -run=XXX -bench=BenchmarkWatch 集成测试 Kubernetes集成测试需要安装etcd（只要按照即可，不需要启动），比如 hack/install-etcd.sh # Installs in ./third_party/etcd echo export PATH=\"\\$PATH:$(pwd)/third_party/etcd\" >> ~/.profile # Add to PATH 集成测试会在需要的时候自动启动etcd和kubernetes服务，并运行test/integration里面的测试。 运行所有集成测试 make test-integration # Run all integration tests. 指定集成测试用例 # Run integration test TestPodUpdateActiveDeadlineSeconds with the verbose flag set. make test-integration KUBE_GOFLAGS=\"-v\" KUBE_TEST_ARGS=\"-run ^TestPodUpdateActiveDeadlineSeconds$\" End to end (e2e)测试 End to end (e2e) 测试模拟用户行为操作Kubernetes，用来保证Kubernetes服务或集群的行为完全符合设计预期。 在开启e2e测试之前，需要先编译测试文件，并设置KUBERNETES_PROVIDER（默认为gce）： make WHAT='test/e2e/e2e.test' make ginkgo export KUBERNETES_PROVIDER=local 启动cluster，测试，最后停止cluster # build Kubernetes, up a cluster, run tests, and tear everything down go run hack/e2e.go -- -v --build --up --test --down 仅测试指定的用例 go run hack/e2e.go -v -test --test_args='--ginkgo.focus=Kubectl\\sclient\\s\\[k8s\\.io\\]\\sKubectl\\srolling\\-update\\sshould\\ssupport\\srolling\\-update\\sto\\ssame\\simage\\s\\[Conformance\\]$' 略过测试用例 go run hack/e2e.go -- -v --test --test_args=\"--ginkgo.skip=Pods.*env 并行测试 # Run tests in parallel, skip any that must be run serially GINKGO_PARALLEL=y go run hack/e2e.go --v --test --test_args=\"--ginkgo.skip=\\[Serial\\]\" # Run tests in parallel, skip any that must be run serially and keep the test namespace if test failed GINKGO_PARALLEL=y go run hack/e2e.go --v --test --test_args=\"--ginkgo.skip=\\[Serial\\] --delete-namespace-on-failure=false\" 清理测试 go run hack/e2e.go -- -v --down 有用的-ctl # -ctl can be used to quickly call kubectl against your e2e cluster. Useful for # cleaning up after a failed test or viewing logs. Use -v to avoid suppressing # kubectl output. go run hack/e2e.go -- -v -ctl='get events' go run hack/e2e.go -- -v -ctl='delete pod foobar' Fedaration e2e测试 export FEDERATION=true export E2E_ZONES=\"us-central1-a us-central1-b us-central1-f\" # or export FEDERATION_PUSH_REPO_BASE=\"quay.io/colin_hom\" export FEDERATION_PUSH_REPO_BASE=\"gcr.io/${GCE_PROJECT_NAME}\" # build container images KUBE_RELEASE_RUN_TESTS=n KUBE_FASTBUILD=true go run hack/e2e.go -- -v -build # push the federation container images build/push-federation-images.sh # Deploy federation control plane go run hack/e2e.go -- -v --up # Finally, run the tests go run hack/e2e.go -- -v --test --test_args=\"--ginkgo.focus=\\[Feature:Federation\\]\" # Don't forget to teardown everything down go run hack/e2e.go -- -v --down 可以用cluster/log-dump.sh 方便的下载相关日志，帮助排查测试中碰到的问题。 Node e2e测试 Node e2e仅测试Kubelet的相关功能，可以在本地或者集群中测试 export KUBERNETES_PROVIDER=local make test-e2e-node FOCUS=\"InitContainer\" make test_e2e_node TEST_ARGS=\"--experimental-cgroups-per-qos=true\" 补充说明 借助kubectl的模版可以方便获取想要的数据，比如查询某个container的镜像的方法为 kubectl get pods nginx-4263166205-ggst4 -o template '--template={{if (exists . \"status\" \"containerStatuses\")}}{{range .status.containerStatuses}}{{if eq .name \"nginx\"}}{{.image}}{{end}}{{end}}{{end}}' 参考文档 Kubernetes testing End-to-End Testing Node e2e test How to write e2e test Coding Conventions Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-08-21 18:23:34 "},"develop/client-go-sample.html":{"url":"develop/client-go-sample.html","title":"client-go示例","keywords":"","body":"client-go示例 访问kubernetes集群有几下几种方式： 方式 特点 支持者 Kubernetes dashboard 直接通过Web UI进行操作，简单直接，可定制化程度低 官方支持 kubectl 命令行操作，功能最全，但是比较复杂，适合对其进行进一步的分装，定制功能，版本适配最好 官方支持 client-go 从kubernetes的代码中抽离出来的客户端包，简单易用，但需要小心区分kubernetes的API版本 官方支持 client-python python客户端，kubernetes-incubator 官方支持 Java client fabric8中的一部分，kubernetes的java客户端 redhat 下面，我们基于client-go，对Deployment升级镜像的步骤进行了定制，通过命令行传递一个Deployment的名字、应用容器名和新image名字的方式来升级。代码和使用方式见 https://github.com/rootsongjc/kubernetes-client-go-sample 。 kubernetes-client-go-sample 代码如下： package main import ( \"flag\" \"fmt\" \"os\" \"path/filepath\" \"k8s.io/apimachinery/pkg/api/errors\" metav1 \"k8s.io/apimachinery/pkg/apis/meta/v1\" \"k8s.io/client-go/kubernetes\" \"k8s.io/client-go/tools/clientcmd\" ) func main() { var kubeconfig *string if home := homeDir(); home != \"\" { kubeconfig = flag.String(\"kubeconfig\", filepath.Join(home, \".kube\", \"config\"), \"(optional) absolute path to the kubeconfig file\") } else { kubeconfig = flag.String(\"kubeconfig\", \"\", \"absolute path to the kubeconfig file\") } deploymentName := flag.String(\"deployment\", \"\", \"deployment name\") imageName := flag.String(\"image\", \"\", \"new image name\") appName := flag.String(\"app\", \"app\", \"application name\") flag.Parse() if *deploymentName == \"\" { fmt.Println(\"You must specify the deployment name.\") os.Exit(0) } if *imageName == \"\" { fmt.Println(\"You must specify the new image name.\") os.Exit(0) } // use the current context in kubeconfig config, err := clientcmd.BuildConfigFromFlags(\"\", *kubeconfig) if err != nil { panic(err.Error()) } // create the clientset clientset, err := kubernetes.NewForConfig(config) if err != nil { panic(err.Error()) } deployment, err := clientset.AppsV1beta1().Deployments(\"default\").Get(*deploymentName, metav1.GetOptions{}) if err != nil { panic(err.Error()) } if errors.IsNotFound(err) { fmt.Printf(\"Deployment not found\\n\") } else if statusError, isStatus := err.(*errors.StatusError); isStatus { fmt.Printf(\"Error getting deployment%v\\n\", statusError.ErrStatus.Message) } else if err != nil { panic(err.Error()) } else { fmt.Printf(\"Found deployment\\n\") name := deployment.GetName() fmt.Println(\"name ->\", name) containers := &deployment.Spec.Template.Spec.Containers found := false for i := range *containers { c := *containers if c[i].Name == *appName { found = true fmt.Println(\"Old image ->\", c[i].Image) fmt.Println(\"New image ->\", *imageName) c[i].Image = *imageName } } if found == false { fmt.Println(\"The application container not exist in the deployment pods.\") os.Exit(0) } _, err := clientset.AppsV1beta1().Deployments(\"default\").Update(deployment) if err != nil { panic(err.Error()) } } } func homeDir() string { if h := os.Getenv(\"HOME\"); h != \"\" { return h } return os.Getenv(\"USERPROFILE\") // windows } 我们使用kubeconfig文件认证连接kubernetes集群，该文件默认的位置是$HOME/.kube/config。 该代码编译后可以直接在kubernetes集群之外，任何一个可以连接到API server的机器上运行。 编译运行 $ go get github.com/rootsongjc/kubernetes-client-go-sample $ cd $GOPATH/src/github.com/rootsongjc/kubernetes-client-go-sample $ make $ ./update-deployment-image -h Usage of ./update-deployment-image: -alsologtostderr log to standard error as well as files -app string application name (default \"app\") -deployment string deployment name -image string new image name -kubeconfig string (optional) absolute path to the kubeconfig file (default \"/Users/jimmy/.kube/config\") -log_backtrace_at value when logging hits line file:N, emit a stack trace -log_dir string If non-empty, write log files in this directory -logtostderr log to standard error instead of files -stderrthreshold value logs at or above this threshold go to stderr -v value log level for V logs -vmodule value comma-separated list of pattern=N settings for file-filtered logging 使用不存在的image更新 $ ./update-deployment-image -deployment filebeat-test -image sz-pg-oam-docker-hub-001.tendcloud.com/library/analytics-docker-test:Build_9 Found deployment name -> filebeat-test Old image -> sz-pg-oam-docker-hub-001.tendcloud.com/library/analytics-docker-test:Build_8 New image -> sz-pg-oam-docker-hub-001.tendcloud.com/library/analytics-docker-test:Build_9 查看Deployment的event。 $ kubectl describe deployment filebeat-test Name: filebeat-test Namespace: default CreationTimestamp: Fri, 19 May 2017 15:12:28 +0800 Labels: k8s-app=filebeat-test Selector: k8s-app=filebeat-test Replicas: 2 updated | 3 total | 2 available | 2 unavailable StrategyType: RollingUpdate MinReadySeconds: 0 RollingUpdateStrategy: 1 max unavailable, 1 max surge Conditions: Type Status Reason ---- ------ ------ Available True MinimumReplicasAvailable Progressing True ReplicaSetUpdated OldReplicaSets: filebeat-test-2365467882 (2/2 replicas created) NewReplicaSet: filebeat-test-2470325483 (2/2 replicas created) Events: FirstSeen LastSeen Count From SubObjectPath Type ReasoMessage --------- -------- ----- ---- ------------- -------- ------------ 2h 1m 3 {deployment-controller } Normal ScalingReplicaSet Scaled down replica set filebeat-test-2365467882 to 2 1m 1m 1 {deployment-controller } Normal ScalingReplicaSet Scaled up replica set filebeat-test-2470325483 to 1 1m 1m 1 {deployment-controller } Normal ScalingReplicaSet Scaled up replica set filebeat-test-2470325483 to 2 可以看到老的ReplicaSet从3个replica减少到了2个，有2个使用新配置的replica不可用，目前可用的replica是2个。 这是因为我们指定的镜像不存在，查看Deployment的pod的状态。 $ kubectl get pods -l k8s-app=filebeat-test NAME READY STATUS RESTARTS AGE filebeat-test-2365467882-4zwx8 2/2 Running 0 33d filebeat-test-2365467882-rqskl 2/2 Running 0 33d filebeat-test-2470325483-6vjbw 1/2 ImagePullBackOff 0 4m filebeat-test-2470325483-gc14k 1/2 ImagePullBackOff 0 4m 我们可以看到有两个pod正在拉取image。 还原为原先的镜像 将image设置为原来的镜像。 $ ./update-deployment-image -deployment filebeat-test -image sz-pg-oam-docker-hub-001.tendcloud.com/library/analytics-docker-test:Build_8 Found deployment name -> filebeat-test Old image -> sz-pg-oam-docker-hub-001.tendcloud.com/library/analytics-docker-test:Build_9 New image -> sz-pg-oam-docker-hub-001.tendcloud.com/library/analytics-docker-test:Build_8 现在再查看Deployment的状态。 $ kubectl describe deployment filebeat-test Name: filebeat-test Namespace: default CreationTimestamp: Fri, 19 May 2017 15:12:28 +0800 Labels: k8s-app=filebeat-test Selector: k8s-app=filebeat-test Replicas: 3 updated | 3 total | 3 available | 0 unavailable StrategyType: RollingUpdate MinReadySeconds: 0 RollingUpdateStrategy: 1 max unavailable, 1 max surge Conditions: Type Status Reason ---- ------ ------ Available True MinimumReplicasAvailable Progressing True NewReplicaSetAvailable OldReplicaSets: NewReplicaSet: filebeat-test-2365467882 (3/3 replicas created) Events: FirstSeen LastSeen Count From SubObjectPath Type ReasoMessage --------- -------- ----- ---- ------------- -------- ------------ 2h 8m 3 {deployment-controller } Normal ScalingReplicaSet Scaled down replica set filebeat-test-2365467882 to 2 8m 8m 1 {deployment-controller } Normal ScalingReplicaSet Scaled up replica set filebeat-test-2470325483 to 1 8m 8m 1 {deployment-controller } Normal ScalingReplicaSet Scaled up replica set filebeat-test-2470325483 to 2 2h 1m 3 {deployment-controller } Normal ScalingReplicaSet Scaled up replica set filebeat-test-2365467882 to 3 1m 1m 1 {deployment-controller } Normal ScalingReplicaSet Scaled down replica set filebeat-test-2470325483 to 0 可以看到available的replica个数恢复成3了。 其实在使用该命令的过程中，通过kubernetes dashboard的页面上查看Deployment的状态更直观，更加方便故障排查。 图片 - 使用kubernetes dashboard进行故障排查 这也是dashboard最大的优势，简单、直接、高效。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-08-21 18:23:34 "},"develop/contribute.html":{"url":"develop/contribute.html","title":"社区贡献","keywords":"","body":"Kubernetes社区贡献 Contributing guidelines Kubernetes Developer Guide Special Interest Groups Feature Tracking and Backlog Community Expectations Kubernetes 官方文档汉化项目 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-09-03 13:18:12 "},"develop/minikube.html":{"url":"develop/minikube.html","title":"Minikube","keywords":"","body":"Minikube Minikube用于在本地运行kubernetes环境，用来开发和测试。 在Mac上安装xhyve-driver docker-machine-driver-xhyve # docker-machine-driver-xhyve need root owner and uid $ sudo chown root:wheel $(brew --prefix)/opt/docker-machine-driver-xhyve/bin/docker-machine-driver-xhyve $ sudo chmod u+s $(brew --prefix)/opt/docker-machine-driver-xhyve/bin/docker-machine-driver-xhyve 到 https://github.com/kubernetes/minikube/releases 下载 minikube，我安装的是minikube v0.22.3 下载完成后修改文件名为minikube，然后chmod +x minikube，移动到$PATH目录下： mv ~/Download/minikube-darwin-adm64 /usr/loal/bin/ chmod +x /usr/local/bin/minikube 安装kubectl 参考Install and Set Up kubectl，直接使用二进制文件安装即可。 curl -LO https://storage.googleapis.com/kubernetes-release/release/`curl -s https://storage.googleapis.com/kubernetes-release/release/stable.txt`/bin/darwin/amd64/kubectl 启动Minikube minikube start Starting local Kubernetes v1.7.5 cluster... Starting VM... Downloading Minikube ISO 139.09 MB / 139.09 MB [============================================] 100.00% 0s Getting VM IP address... Moving files into cluster... Setting up certs... Connecting to cluster... Setting up kubeconfig... Starting cluster components... Kubectl is now configured to use the cluster. 这将生成默认的~/.kube/config文件，自动指向minikube。 停止Minikube minikube stop 参考 Running Kubernetes Locally via Minikube Install minikube Driver plugin installation - xhyve-driver Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-10-23 19:06:48 "},"appendix/":{"url":"appendix/","title":"附录说明","keywords":"","body":"附录说明 参考文档以及一些实用的资源链接。 Kubernetes documentation Awesome Kubernetes Kubernetes the hard way Kubernetes Bootcamp Design patterns for container-based distributed systems Awesome Cloud Native Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-15 19:43:33 "},"appendix/debug-kubernetes-services.html":{"url":"appendix/debug-kubernetes-services.html","title":"Kubernetes中的应用故障排查","keywords":"","body":"Kubernetes service中的故障排查 查看某个资源的定义和用法 kubectl explain 查看Pod的状态 kubectl get pods kubectl describe pods my-pod 监控Pod状态的变化 kubectl get pod -w 可以看到一个 namespace 中所有的 pod 的 phase 变化，请参考 Pod 的生命周期。 查看 Pod 的日志 kubectl logs my-pod kubectl logs my-pod -c my-container kubectl logs -f my-pod kubectl logs -f my-pod -c my-container -f 参数可以 follow 日志输出。 交互式 debug kubectl exec my-pod -it /bin/bash kubectl top pod POD_NAME --containers Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-09-21 14:58:20 "},"appendix/material-share.html":{"url":"appendix/material-share.html","title":"Kubernetes相关资讯和情报链接","keywords":"","body":"Kubernetes相关资讯和情报链接 授人以鱼，不如授人以渔。下面的资料将有助于大家了解kubernetes生态圈当前发展状况和发展趋势，我特此整理相关资料如下。 生态环境 包括kubernetes和cloud native相关的开源软件、工具和全景图。 awesome-kubernetes - A curated list for awesome kubernetes sources 🚢🎉 https://ramitsurana.github.io/awesome… awesome-cloud-native - A curated list for awesome cloud native architectures https://jimmysong.io/awesome-cloud-native/ cloud native landscape - Cloud Native Landscape https://cncf.io 开源书籍和教程 arun-gupta/kubernetes-aws-workshop arun-gupta/kubernetes-java-sample feiskyer/kubernetes-handbook kelseyhightower/kubernetes-the-hard-way ks - A series of Kubernetes walk-throughs opsnull/follow-me-install-kubernetes-cluster rootsongjc/kubernetes-handbook 幻灯片、图书和情报资料分享 cloud-native-slides-share - Cloud Native 相关meetup、会议PPT、图书资料分享 博客与网站 Kubernetes和Cloud Native相关网站、专栏、博客等。 网站与专栏 thenewstack.io k8sport.org giantswarm blog k8smeetup.com dockone.io Cloud Native知乎专栏 kubernetes.org.cn 博客 apcera aporeto applatix apprenda bitnami buoyant cisco cncf codeship containership coreos coscale deis fabric8 grafana gravitational heptio istio jimmysong kubernetes moby openshift pivotal platform9 prometheus rancher sysdig spinnaker supergiant thecodeteam twistlock vamp weave wercker Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-21 18:51:19 "},"appendix/docker-best-practice.html":{"url":"appendix/docker-best-practice.html","title":"Docker最佳实践","keywords":"","body":"Docker最佳实践 本文档旨在实验Docker1.13新特性和帮助大家了解docker集群的管理和使用。 环境配置 Docker1.13环境配置 docker源码编译 网络管理 网络配置和管理是容器使用中的的一个重点和难点，对比我们之前使用的docker版本是1.11.1，docker1.13中网络模式跟之前的变动比较大，我们会花大力气讲解。 如何创建docker network Rancher网络探讨和扁平网络实现 swarm mode的路由网络 docker扁平化网络插件Shrike（基于docker1.11） 存储管理 Docker存储插件 infinit 被docker公司收购的法国团队开发 convoy rancher开发的docker volume plugin torus 已废弃 flocker ClusterHQ开发 日志管理 Docker提供了一系列log drivers，如fluentd、journald、syslog等。 需要配置docker engine的启动参数。 创建应用 官方文档：Docker swarm sample app overview 基于docker1.13手把手教你创建swarm app swarm集群应用管理 使用docker-compose创建应用 集群管理 我们使用docker内置的swarm来管理docker集群。 swarm mode介绍 我们推荐使用开源的docker集群管理配置方案： Crane：由数人云开源的基于swarmkit的容器管理软件，可以作为docker和go语言开发的一个不错入门项目 Rancher:Rancher是一个企业级的容器管理平台，可以使用Kubernetes、swarm和rancher自研的cattle来管理集群。 Crane的部署和使用 Rancher的部署和使用 资源限制 内存资源限制 CPU资源限制 IO资源限制 服务发现 下面罗列一些列常见的服务发现工具。 Etcd:服务发现/全局分布式键值对存储。这是CoreOS的创建者提供的工具，面向容器和宿主机提供服务发现和全局配置存储功能。它在每个宿主机有基于http协议的API和命令行的客户端。https://github.com/docker/etcd Cousul：服务发现/全局分布式键值对存储。这个服务发现平台有很多高级的特性，使得它能够脱颖而出，例如：配置健康检查、ACL功能、HAProxy配置等等。 Zookeeper：诞生于Hadoop生态系统里的组件，Apache的开源项目。服务发现/全局分布式键值对存储。这个工具较上面两个都比较老，提供一个更加成熟的平台和一些新特性。 Crypt：加密etcd条目的项目。Crypt允许组建通过采用公钥加密的方式来保护它们的信息。需要读取数据的组件或被分配密钥，而其他组件则不能读取数据。 Confd：观测键值对存储变更和新值的触发器重新配置服务。Confd项目旨在基于服务发现的变化，而动态重新配置任意应用程序。该系统包含了一个工具来检测节点中的变化、一个模版系统能够重新加载受影响的应用。 Vulcand：vulcand在组件中作为负载均衡使用。它使用etcd作为后端，并基于检测变更来调整它的配置。 Marathon：虽然marathon主要是调度器，它也实现了一个基本的重新加载HAProxy的功能，当发现变更时，它来协调可用的服务。 Frontrunner：这个项目嵌入在marathon中对HAproxy的更新提供一个稳定的解决方案。 Synapse：由Airbnb出品的，Ruby语言开发，这个项目引入嵌入式的HAProxy组件，它能够发送流量给各个组件。http://bruth.github.io/synapse/docs/ Nerve：它被用来与synapse结合在一起来为各个组件提供健康检查，如果组件不可用，nerve将更新synapse并将该组件移除出去。 插件开发 插件开发示例-sshfs 我的docker插件开发文章 Docker17.03-CE插件开发举例 网络插件 Contiv 思科出的Docker网络插件，趟坑全记录，目前还无法上生产，1.0正式版还没出，密切关注中。 Calico 产品化做的不错，已经有人用在生产上了。 存储插件 业界使用案例 京东从OpenStack切换到Kubernetes的经验之谈 美团点评容器平台介绍 阿里超大规模docker化之路 TalkingData-容器技术在大数据场景下的应用Yarn on Docker 乐视云基于Kubernetes的PaaS平台建设 资源编排 建议使用kuberentes，虽然比较复杂，但是专业的工具做专业的事情，将编排这么重要的生产特性绑定到docker上的风险还是很大的，我已经转投到kubernetes怀抱了，那么你呢？ 我的kubernetes探险之旅 相关资源 容器技术工具与资源 容器技术2016年总结 关于 Author: Jimmy Song rootsongjc@gmail.com 更多关于Docker、MicroServices、Big Data、DevOps、Deep Learning的内容请关注Jimmy Song's Blog，将不定期更新。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-10 18:13:03 "},"appendix/tricks.html":{"url":"appendix/tricks.html","title":"使用技巧","keywords":"","body":"1. 在容器中获取 Pod 的IP 通过环境变量来实现，该环境变量直接引用 resource 的状态字段，示例如下： apiVersion: v1 kind: ReplicationController metadata: name: world-v2 spec: replicas: 3 selector: app: world-v2 template: metadata: labels: app: world-v2 spec: containers: - name: service image: test env: - name: POD_IP valueFrom: fieldRef: fieldPath: status.podIP ports: - name: service containerPort: 777 容器中可以直接使用 POD_IP 环境变量获取容器的 IP。 2. 指定容器的启动参数 我们可以在 Pod 中为容器使用 command 为容器指定启动参数： command: [\"/bin/bash\",\"-c\",\"bootstrap.sh\"] 看似很简单，使用数组的方式定义，所有命令使用跟 Dockerfile 中的 CMD 配置是一样的，但是有一点不同的是，bootsttap.sh 必须具有可执行权限，否则容器启动时会出错。 3. 让Pod调用宿主机的docker能力 我们可以想象一下这样的场景，让 Pod 来调用宿主机的 docker 能力，只需要将宿主机的 docker 命令和 docker.sock 文件挂载到 Pod 里面即可，如下： apiVersion: v1 kind: Pod metadata: name: busybox-cloudbomb spec: containers: - image: busybox command: - /bin/sh - \"-c\" - \"while true; \\ do \\ docker run -d --name BOOM_$(cat /dev/urandom | tr -cd 'a-f0-9' | head -c 6) nginx ; \\ done\" name: cloudbomb volumeMounts: - mountPath: /var/run/docker.sock name: docker-socket - mountPath: /bin/docker name: docker-binary volumes: - name: docker-socket hostPath: path: /var/run/docker.sock - name: docker-binary hostPath: path: /bin/docker 参考：Architecture Patterns for Microservices in Kubernetes 4. 使用Init container初始化应用配置 Init container可以在应用程序的容器启动前先按顺序执行一批初始化容器，只有所有Init容器都启动成功后，Pod才算启动成功。看下下面这个例子（来源：kubernetes: mounting volume from within init container - Stack Overflow）： apiVersion: v1 kind: Pod metadata: name: init labels: app: init annotations: pod.beta.kubernetes.io/init-containers: '[ { \"name\": \"download\", \"image\": \"axeclbr/git\", \"command\": [ \"git\", \"clone\", \"https://github.com/mdn/beginner-html-site-scripted\", \"/var/lib/data\" ], \"volumeMounts\": [ { \"mountPath\": \"/var/lib/data\", \"name\": \"git\" } ] } ]' spec: containers: - name: run image: docker.io/centos/httpd ports: - containerPort: 80 volumeMounts: - mountPath: /var/www/html name: git volumes: - emptyDir: {} name: git 这个例子就是用来再应用程序启动前首先从GitHub中拉取代码并存储到共享目录下。 关于Init容器的更详细说明请参考 init容器。 5. 使容器内时间与宿主机同步 我们下载的很多容器内的时区都是格林尼治时间，与北京时间差8小时，这将导致容器内的日志和文件创建时间与实际时区不符，有两种方式解决这个问题： 修改镜像中的时区配置文件 将宿主机的时区配置文件/etc/localtime使用volume方式挂载到容器中 第二种方式比较简单，不需要重做镜像，只要在应用的yaml文件中增加如下配置： volumeMounts: - name: host-time mountPath: /etc/localtime readOnly: true volumes: - name: host-time hostPath: path: /etc/localtime 6. 在Pod中获取宿主机的主机名、namespace等 这条技巧补充了第一条获取 podIP 的内容，方法都是一样的，只不过列出了更多的引用字段。 参考下面的 pod 定义，每个 pod 里都有一个 {.spec.nodeName} 字段，通过 fieldRef 和环境变量，就可以在Pod中获取宿主机的主机名（访问环境变量MY_NODE_NAME）。 apiVersion: v1 kind: Pod metadata: name: dapi-test-pod spec: containers: - name: test-container image: busybox command: [ \"/bin/sh\", \"-c\", \"env\" ] env: - name: MY_NODE_NAME valueFrom: fieldRef: fieldPath: spec.nodeName - name: MY_POD_NAME valueFrom: fieldRef: fieldPath: metadata.name - name: MY_POD_NAMESPACE valueFrom: fieldRef: fieldPath: metadata.namespace - name: MY_POD_IP valueFrom: fieldRef: fieldPath: status.podIP - name: MY_POD_SERVICE_ACCOUNT valueFrom: fieldRef: fieldPath: spec.serviceAccountName restartPolicy: Never 7. 配置Pod使用外部DNS 修改kube-dns的使用的ConfigMap。 apiVersion: v1 kind: ConfigMap metadata: name: kube-dns namespace: kube-system data: stubDomains: | {\"k8s.com\": [\"192.168.10.10\"]} upstreamNameservers: | [\"8.8.8.8\", \"8.8.4.4\"] upstreamNameservers 即使用的外部DNS，参考：Configuring Private DNS Zones and Upstream Nameservers in Kubernetes Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-21 16:44:45 "},"appendix/issues.html":{"url":"appendix/issues.html","title":"问题记录","keywords":"","body":"问题记录 安装、使用kubernetes的过程中遇到的所有问题的记录。 推荐直接在Kubernetes的GitHub上提issue，在此记录所提交的issue。 1.Failed to start ContainerManager failed to initialise top level QOS containers #43856 重启kubelet时报错，目前的解决方法是： 1.在docker.service配置中增加的--exec-opt native.cgroupdriver=systemd配置。 2.手动删除slice（貌似不管用） 3.重启主机，这招最管用😄 for i in $(systemctl list-unit-files —no-legend —no-pager -l | grep —color=never -o .*.slice | grep kubepod);do systemctl stop $i;done 上面的几种方法在该bug修复前只有重启主机管用，该bug已于2017年4月27日修复，merge到了master分支，见https://github.com/kubernetes/kubernetes/pull/44940 2.High Availability of Kube-apiserver #19816 API server的HA如何实现？或者说这个master节点上的服务api-server、scheduler、controller 如何实现HA？目前的解决方案是什么？ 目前的解决方案是api-server是无状态的可以启动多个，然后在前端再加一个nginx或者ha-proxy。而scheduler和controller都是直接用容器的方式启动的。 3.Kubelet启动时Failed to start ContainerManager systemd version does not support ability to start a slice as transient unit CentOS系统版本7.2.1511 kubelet启动时报错systemd版本不支持start a slice as transient unit。 尝试升级CentOS版本到7.3，看看是否可以修复该问题。 与kubeadm init waiting for the control plane to become ready on CentOS 7.2 with kubeadm 1.6.1 #228类似。 另外有一个使用systemd管理kubelet的proposal。 4.kube-proxy报错kube-proxy[2241]: E0502 15:55:13.889842 2241 conntrack.go:42] conntrack returned error: error looking for path of conntrack: exec: \"conntrack\": executable file not found in $PATH 导致的现象 kubedns启动成功，运行正常，但是service之间无法解析，kubernetes中的DNS解析异常 解决方法 CentOS中安装conntrack-tools包后重启kubernetes集群即可。 5. Pod stucks in terminating if it has a privileged container but has been scheduled to a node which doesn't allow privilege issue#42568 当pod被调度到无法权限不足的node上时，pod一直处于pending状态，且无法删除pod，删除时一直处于terminating状态。 kubelet中的报错信息 Error validating pod kube-keepalived-vip-1p62d_default(5d79ccc0-3173-11e7-bfbd-8af1e3a7c5bd) from api, ignoring: spec.containers[0].securityContext.privileged: Forbidden: disallowed by cluster policy 6.PVC中对Storage的容量设置不生效 使用glusterfs做持久化存储文档中我们构建了PV和PVC，当时给glusterfs-nginx的PVC设置了8G的存储限额，nginx-dm这个Deployment使用了该PVC，进入该Deployment中的Pod执行测试： dd if=/dev/zero of=test bs=1G count=10 图片 - pvc-storage-limit 从截图中可以看到创建了9个size为1G的block后无法继续创建了，已经超出了8G的限额。 7. 使用 Headless service 的时候 kubedns 解析不生效 kubelet 的配置文件 /etc/kubernetes/kubelet 中的配置中将集群 DNS 的 domain name 配置成了 ––cluster-domain=cluster.local. ，虽然对于 service 的名字能够正常的完成 DNS 解析，但是对于 headless service 中的 pod 名字解析不了，查看 pod 的 /etc/resolv.conf 文件可以看到以下内容： nameserver 10.0.254.2 search default.svc.cluster.local. svc.cluster.local. cluster.local. tendcloud.com options ndots:5 修改 /etc/kubernetes/kubelet 文件中的 ––cluster-domain=cluster.local. 将 local 后面的点去掉后重启所有的 kubelet，这样新创建的 pod 中的 /etc/resolv.conf文件的 DNS 配置和解析就正常了。 8. kubernetes 集成 ceph 存储 rbd 命令组装问题 kubernetes 使用 ceph 创建 PVC 的时候会有如下报错信息： Events: FirstSeen LastSeen Count From SubObjectPath Type Reason Message --------- -------- ----- ---- ------------- -------- ------ ------- 1h 12s 441 {persistentvolume-controller } Warning ProvisioningFailed Failed to provision volume with StorageClass \"ceph-web\": failed to create rbd image: executable file not found in $PATH, command output: 检查 kube-controller-manager 的日志将看到如下错误信息： Sep 4 15:25:36 bj-xg-oam-kubernetes-001 kube-controller-manager: W0904 15:25:36.032128 13211 rbd_util.go:364] failed to create rbd image, output Sep 4 15:25:36 bj-xg-oam-kubernetes-001 kube-controller-manager: W0904 15:25:36.032201 13211 rbd_util.go:364] failed to create rbd image, output Sep 4 15:25:36 bj-xg-oam-kubernetes-001 kube-controller-manager: W0904 15:25:36.032252 13211 rbd_util.go:364] failed to create rbd image, output Sep 4 15:25:36 bj-xg-oam-kubernetes-001 kube-controller-manager: E0904 15:25:36.032276 13211 rbd.go:317] rbd: create volume failed, err: failed to create rbd image: fork/exec /usr/bin/rbd: invalid argument, command output: 该问题尚未解决，参考 Error creating rbd image: executable file not found in $PATH#38923 9. Helm: Error: no available release name found 在开启了RBAC的kubernetes集群中，当使用helm部署应用，执行helm install的时候，会报着个错误： Error: no available release name found Error: the server does not allow access to the requested resource (get configmaps) 这是因为我们使用的2.3.1版本的helm init的时候没有为tiller创建serviceaccount和clusterrolebiding的缘故导致的。 kubectl create serviceaccount --namespace kube-system tiller kubectl create clusterrolebinding tiller-cluster-rule --clusterrole=cluster-admin --serviceaccount=kube-system:tiller # helm init -i sz-pg-oam-docker-hub-001.tendcloud.com/library/kubernetes-helm-tiller:v2.3.1 kubectl patch deploy --namespace kube-system tiller-deploy -p '{\"spec\":{\"template\":{\"spec\":{\"serviceAccount\":\"tiller\"}}}}' 参考 Helm: Error: no available release name found - StackOverflow Helm 2.2.3 not working properly with kubeadm 1.6.1 default RBAC rules #2224 参考 Persistent Volume Resource Design Proposals Helm: Error: no available release name found Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-10-24 19:04:57 "},"appendix/kubernetes-changelog.html":{"url":"appendix/kubernetes-changelog.html","title":"kubernetes版本更新日志","keywords":"","body":"Kubernetes版本更新日志 刚开始写作本书时kubernetes1.6刚刚发布，随后kubernetes基本以每3个月发布一个版本的速度不断迭代，为了追踪不同版本的新特性，我们有必要在此记录一下。 每个kubernetes版本的详细更新日志请参考：https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG.md 发布记录 2017年6月29日 Kubernetes1.7发布 2017年9月28日 Kubernetes1.8发布 2017年12月15日 Kubernetes1.9发布 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-16 10:11:41 "},"appendix/kubernetes-1.7-changelog.html":{"url":"appendix/kubernetes-1.7-changelog.html","title":"Kubernetes1.7更新日志","keywords":"","body":"Kubernetes1.7更新日志 2017年6月29日，kuberentes1.7发布。该版本的kubernetes在安全性、存储和可扩展性方面有了很大的提升。改版本的详细更新文档请查看Changelog。 这些新特性中包含了安全性更高的加密的secret、pod间通讯的网络策略，限制kubelet访问的节点授权程序以及客户端/服务器TLS证书轮换。 对于那些在Kubernetes上运行横向扩展数据库的人来说，这个版本有一个主要的特性，可以为StatefulSet添加自动更新并增强DaemonSet的更新。我们还宣布了对本地存储的Alpha支持，以及用于更快地缩放StatefulSets的突发模式。 此外，对于高级用户，此发行版中的API聚合允许使用用于自定义的API与API server同时运行。其他亮点包括支持可扩展的准入控制器，可插拔云供应商程序和容器运行时接口（CRI）增强功能。 新功能 安全 Network Policy API 提升为稳定版本。用户可以通过使用网络插件实现的网络策略来控制哪些Pod之间能够互相通信。 节点授权和准入控制插件是新增加的功能，可以用于限制kubelet可以访问的secret、pod和其它基于节点的对象。 加密的Secret和etcd中的其它资源，现在是alpha版本。 Kubelet TLS bootstrapping现在支持客户端和服务器端的证书轮换。 由API server存储的审计日志现在更具可定制性和可扩展性，支持事件过滤和webhook。它们还为系统审计提供更丰富的数据。 有状态负载 StatefulSet更新是1.7版本的beta功能，它允许使用包括滚动更新在内的一系列更新策略自动更新诸如Kafka，Zookeeper和etcd等有状态应用程序。 StatefulSets现在还支持对不需要通过Pod管理策略进行排序的应用程序进行快速扩展和启动。这可以是主要的性能改进。 本地存储（alpha）是有状态应用程序最常用的功能之一。用户现在可以通过标准的PVC/PV接口和StatefulSet中的StorageClass访问本地存储卷。 DaemonSet——为每个节点创建一个Pod，现在有了更新功能，在1.7中增加了智能回滚和历史记录功能。 新的StorageOS Volume插件可以使用本地或附加节点存储中以提供高可用的集群范围的持久卷。 可扩展性 运行时的API聚合是此版本中最强大的扩展功能，允许高级用户将Kubernetes风格的预先构建的第三方或用户创建的API添加到其集群中。 容器运行时接口（CRI）已经增强，可以使用新的RPC调用从运行时检索容器度量。 CRI的验证测试已经发布，与containerd进行了Alpha集成，现在支持基本的生命周期和镜像管理。参考深入介绍CRI的文章。 其它功能 引入了对外部准入控制器的Alpha支持，提供了两个选项，用于向API server添加自定义业务逻辑，以便在创建对象和验证策略时对其进行修改。 基于策略的联合资源布局提供Alpha版本，用于根据自定义需求（如法规、定价或性能）为联合（federated）集群提供布局策略。 弃用 第三方资源（TPR）已被自定义资源定义（Custom Resource Definitions，CRD）取代，后者提供了一个更清晰的API，并解决了TPR测试期间引发的问题和案例。如果您使用TPR测试版功能，则建议您迁移，因为它将在Kubernetes 1.8中被移除。 以上是Kubernetes1.7中的主要新特性，详细更新文档请查看Changelog。 参考 Kuberentes 1.7: Security Hardening, Stateful Application Updates and Extensibility Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-16 10:10:15 "},"appendix/kubernetes-1.8-changelog.html":{"url":"appendix/kubernetes-1.8-changelog.html","title":"Kubernetes1.8更新日志","keywords":"","body":"Kubernetes1.8更新日志 2017年9月28日，kubernetes1.8版本发布。该版本中包括了一些功能改进和增强，并增加了项目的成熟度，将强了kubernetes的治理模式，这些都将有利于kubernetes项目的持续发展。 聚焦安全性 Kubernetes1.8的基于角色的访问控制（RBAC）成为stable支持。RBAC允许集群管理员动态定义角色对于Kubernetes API的访问策略。通过网络策略筛选出站流量的Beta支持，增强了对入站流量进行过滤的现有支持。 RBAC和网络策略是强化Kubernetes内组织和监管安全要求的两个强大工具。 Kubelet的传输层安全性（TLS）证书轮换成为beta版。自动证书轮换减轻了集群安全性运维的负担。 聚焦工作负载支持 Kubernetes 1.8通过apps/v1beta2组和版本推动核心工作负载API的beta版本。Beta版本包含当前版本的Deployment、DaemonSet、ReplicaSet和StatefulSet。 工作负载API是将现有工作负载迁移到Kubernetes以及开发基于Kubernetes的云原生应用程序提供了基石。 对于那些考虑在Kubernetes上运行大数据任务的，现在的工作负载API支持运行kubernetes原生支持的Apache Spark。 批量工作负载，比如夜间ETL工作，将从CronJobs的beta版本中受益。 自定义资源定义（CRD）在Kubernetes 1.8中依然为测试版。CRD提供了一个强大的机制来扩展Kubernetes和用户定义的API对象。 CRD的一个用例是通过Operator Pattern自动执行复杂的有状态应用，例如键值存储、数据库和存储引擎。随着稳定性的继续推进，预计将继续加强对CRD的验证。 更多 Volume快照、PV调整大小、自动taint、pod优先级、kubectl插件等！ 除了稳定现有的功能，Kubernetes 1.8还提供了许多预览新功能的alpha版本。 社区中的每个特别兴趣小组（SIG）都在继续为所在领域的用户提供更多的功能。有关完整列表，请访问发行说明。 参考 Kubernetes 1.8: Security, Workloads and Feature Depth Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-12 13:44:48 "},"appendix/kubernetes-1.9-changelog.html":{"url":"appendix/kubernetes-1.9-changelog.html","title":"Kubernetes1.9更新日志","keywords":"","body":"Kubernetes1.9更新日志 2017年12月15日，kubernetes1.9版本发布。Kubernetes依然按照每三个月一个大版本发布的速度稳定迭代，这是今年发布的第四个版本，也是今年的最后一个版本，该版本最大的改进是Apps Workloads API成为稳定版本，这消除了很多潜在用户对于该功能稳定性的担忧。还有一个重大更新，就是测试支持了Windows了，这打开了在kubernetes中运行Windows工作负载的大门。 Workloads API GA apps/v1 Workloads API成为GA（General Availability），且默认启用。 Apps Workloads API将DaemonSet、Deployment、ReplicaSet和StatefulSet API组合在一起，作为Kubernetes中长时间运行的无状态和有状态工作负载的基础。 Deployment和ReplicaSet是Kubernetes中最常用的两个对象，经过一年多的实际使用和反馈后，现在已经趋于稳定。SIG apps同时将这些经验应用到另外的两个对象上，使得DaemonSet和StatefulSet也能顺利毕业走向成熟。v1（GA）意味着已经生产可用，并保证长期的向后兼容。 Windows支持（beta） Kubernetes最初是为Linux系统开发的，但是用户逐渐意识到容器编排的好处，我们看到有人需要在Kubernetes上运行Windows工作负载。在12个月前，我们开始认真考虑在Kubernetes上支持Windows Server的工作。 SIG-Windows现在已经将这个功能推广到beta版本，这意味着我们可以评估它的使用情况。 增强存储 kubernetes从第一个版本开始就支持多种持久化数据存储，包括常用的NFS或iSCSI，以及对主要公共云和私有云提供商的存储解决方案的原生支持。随着项目和生态系统的发展，Kubernetes的存储选择越来越多。然而，为新的存储系统添加volume插件一直是一个挑战。 容器存储接口（CSI）是一个跨行业标准计划，旨在降低云原生存储开发的障碍并确保兼容性。 SIG-Storage和CSI社区正在合作提供一个单一接口，用于配置、附着和挂载与Kubernetes兼容的存储。 Kubernetes 1.9引入了容器存储接口（CSI）的alpha实现，这将使挂载新的volume插件就像部署一个pod一样简单，并且第三方存储提供商在开发他们的解决方案时也无需修改kubernetes的核心代码。 由于该功能在1.9版本中为alpha，因此必须明确启用该功能，不建议用于生产使用，但它为更具扩展性和基于标准的Kubernetes存储生态系统提供了清晰的路线图。 其它功能 自定义资源定义（CRD）校验，现在已经成为beta，默认情况下已启用，可以用来帮助CRD作者对于无效对象定义给出清晰和即时的反馈。 SIG Node硬件加速器转向alpha，启用GPU，从而实现机器学习和其他高性能工作负载。 CoreDNS alpha可以使用标准工具来安装CoreDNS。 kube-proxy的IPVS模式进入beta版，为大型集群提供更好的可扩展性和性能。 社区中的每个特别兴趣小组（SIG）继续提供其所在领域的用户最迫切需要的功能。有关完整列表，请访问发行说明。 获取 Kubernetes1.9已经可以通过GitHub下载。 参考 Kubernetes 1.9: Apps Workloads GA and Expanded Ecosystem Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-16 10:47:44 "}} \ No newline at end of file +{"./":{"url":"./","title":"序言","keywords":"","body":"Kubernetes Handbook Kubernetes是Google基于Borg开源的容器编排调度引擎，作为CNCF（Cloud Native Computing Foundation）最重要的组件之一，它的目标不仅仅是一个编排系统，而是提供一个规范，可以让你来描述集群的架构，定义服务的最终状态，kubernetes可以帮你将系统自动得达到和维持在这个状态。Kubernetes作为云原生应用的基石，相当于一个云操作系统，其重要性不言而喻。 本书记录了本人从零开始学习和使用Kubernetes的心路历程，着重于经验分享和总结，同时也会有相关的概念解析，希望能够帮助大家少踩坑，少走弯路，还会指引大家关于关注kubernetes生态周边，如微服务构建、DevOps、大数据应用、Service Mesh、Cloud Native等领域。 起初写作本书时，安装的所有组件、所用示例和操作等皆基于Kubernetes1.6.0 版本，同时我们也将密切关注kubernetes的版本更新，随着它的版本更新升级，本书中的kubernetes版本和示例也将随之更新。 GitHub 地址：https://github.com/rootsongjc/kubernetes-handbook Gitbook 在线浏览：https://jimmysong.io/kubernetes-handbook/ 如何使用本书 在线浏览 访问 https://jimmysong.io/kubernetes-handbook/ 注意：文中涉及的配置文件和代码链接在网页中将无法访问，请下载GitHub源码后，在Markdown编辑器中打开，点击链接将跳转到你的本地目录，推荐使用typora，或者直接登录github 查看。 本地查看 将代码克隆到本地 安装 gitbook：Setup and Installation of GitBook 执行 gitbook serve 在浏览器中访问 http://localhost:4000 生成的文档在 _book 目录下 下载 PDF/ePub/Mobi 格式文档本地查看 访问 gitbook 可以看到下载地址，可以下载根据最新文档生成的 PDF/ePub/Mobi 格式文档（文档的注脚中注明了更新时间），同时也可以直接在 gitbook 中阅读，不过 gitbook 不太稳定打开速度较慢，建议大家直接在 https://jimmysong.io/kubernetes-handbook/ 浏览。 生成 pdf 下载Calibre On Mac 在Mac下安装后，使用该命令创建链接 ln -s /Applications/calibre.app/Contents/MacOS/ebook-convert /usr/local/bin 在该项目目录下执行以下命令生成kubernetes-handbook.pdf文档。 gitbook pdf . ./kubernetes-handbook.pdf On Windows 需要用到的工具：calibre，phantomjs 将上述2个安装，calibre 默认安装的路径 C:\\Program Files\\Calibre2 为你解压路径； 并将其目录均加入到系统变量 path 中,参考:目录添加到系统变量 path 中； 在 cmd 打开你需要转 pdf 的文件夹,输入gitbook pdf即可； 生成单个章节的pdf 使用pandoc和latex来生成pdf格式文档。 pandoc --latex-engine=xelatex --template=pm-template input.md -o output.pdf 如何贡献 提 issue 如果你发现文档中的错误，或者有好的建议、不明白的问题、不要犹豫，欢迎提交issue。 发起 Pull Request 当你发现文章中明确的错误或者逻辑问题，在你自己的fork的分支中，创建一个新的branch，修改错误，push 到你的branch，然后在提交issue 后直接发起Pull Request。 贡献文档 文档的组织规则 如果要创建一个大的主题就在最顶层创建一个目录； 全书五大主题，每个主题一个目录，其下不再设二级目录； 所有的图片都放在最顶层的 images 目录下，原则上文章中用到的图片都保存在本地； 所有的文档的文件名使用英文命名，可以包含数字和中划线； etc、manifests目录专门用来保存配置文件和文档中用到的其他相关文件； 添加文档 在该文章相关主题的目录下创建文档； 在 SUMMARY.md 中在相应的章节下添加文章链接； 执行 gitbook serve 测试是否报错，访问 http://localhost:4000 查看该文档是否出现在相应主题的目录下； 提交PR 关于 贡献者列表 加入K8S🤘Cloud Native实战群 扫描二维码，Jimmy Song，或直接搜索微信号jimmysong后拉您入群，请增加备注。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-15 19:49:07 "},"cloud-native/kubernetes-and-cloud-native-app-overview.html":{"url":"cloud-native/kubernetes-and-cloud-native-app-overview.html","title":"Kubernetes与云原生应用概览","keywords":"","body":"Kubernetes与云原生应用概览 几个月前Mesos已经宣布支持kubernetes，而在2017年10月份的DockerCon EU上Docker公司宣布同时官方支持Swarm和Kubernetes容器编排，kubernetes已然成为容器编排调度的标准。 作为全书的开头，首先从历史、生态和应用角度介绍一下kubernetes与云原生应用，深入浅出，高屋建瓴，没有深入到具体细节，主要是为了给初次接触kubernetes的小白扫盲，具体细节请参考链接。 从云计算到微服务再到云原生计算 下面将从云计算的发展历程引入云原生计算，请先看下图： 图片 - 云计算演进历程 云原生应用到2020年将比目前至少翻一番，下图是Marc Wilczek的调查报告。 图片 - 来自Twitter @MarcWilczek 云计算介绍 云计算包含的内容十分繁杂，也有很多技术和公司牵强附会说自己是云计算公司，说自己是做云的，实际上可能风马牛不相及。说白了，云计算就是一种配置资源的方式，根据资源配置方式的不同我们可以把云计算从宏观上分为以下三种类型： IaaS：这是为了想要建立自己的商业模式并进行自定义的客户，例如亚马逊的EC2、S3存储、Rackspace虚拟机等都是IaaS。 PaaS：工具和服务的集合，对于想用它来构建自己的应用程序或者想快速得将应用程序部署到生产环境而不必关心底层硬件的用户和开发者来说是特别有用的，比如Cloud Foundry、Google App Engine、Heroku等。 SaaS：终端用户可以直接使用的应用程序。这个就太多，我们生活中用到的很多软件都是SaaS服务，只要基于互联网来提供的服务基本都是SaaS服务，有的服务是免费的，比如Google Docs，还有更多的是根据我们购买的Plan和使用量付费，比如GitHub、各种云存储。 微服务介绍 微服务（Microservices）这个词比较新颖，但是其实这种架构设计理念早就有了。微服务是一种分布式架构设计理念，为了推动细粒度服务的使用，这些服务要能协同工作，每个服务都有自己的生命周期。一个微服务就是一个独立的实体，可以独立的部署在PAAS平台上，也可以作为一个独立的进程在主机中运行。服务之间通过API访问，修改一个服务不会影响其它服务。 要想了解微服务的详细内容推荐阅读《微服务设计》（Sam Newman著），我写过这本书的读书笔记 - 微服务设计读书笔记。 下文中会谈到kubernetes与微服务的关系，其中kubernetes的service天生就适合与微服务。 云原生概念介绍 下面是Cloud Native概念思维导图 图片 - Cloud native思维导图 云原生准确来说是一种文化，更是一种潮流，它是云计算的一个必然导向。它的意义在于让云成为云化战略成功的基石，而不是阻碍，如果业务应用上云之后开发和运维人员比原先还痛苦，成本高企的话，这样的云我们宁愿不不上。 自从云的概念开始普及，许多公司都部署了实施云化的策略，纷纷搭建起云平台，希望完成传统应用到云端的迁移。但是这个过程中会遇到一些技术难题，上云以后，效率并没有变得奇高，故障也没有迅速定位。 为了解决传统应用升级缓慢、架构臃肿、不能快速迭代、故障不能快速定位、问题无法快速解决等问题，云原生这一概念横空出世。云原生可以改进应用开发的效率，改变企业的组织结构，甚至会在文化层面上直接影响一个公司的决策。 另外，云原生也很好地解释了云上运行的应用应该具备什么样的架构特性——敏捷性、可扩展性、故障可恢复性。 综上所述，云原生应用应该具备以下几个关键词： 敏捷 可靠 高弹性 易扩展 故障隔离保护 不中断业务持续更新 以上特性也是云原生区别于传统云应用的优势特点。 从宏观概念上讲，云原生是不同思想的集合，集目前各种热门技术之大成，具体包括如下图所示的几个部分。 Kubernetes与云原生的关系 Kuberentes可以说是乘着docker和微服务的东风，已经推出便迅速蹿红，它的很多设计思想都契合了微服务和云原生应用的设计法则，这其中最著名的就是开发了Heroku PaaS平台的工程师们总结的 Twelve-factor App了。 下面我将讲解Kubernetes设计时是如何按照了十二因素应用法则，并给出kubernetes中的应用示例，并附上一句话简短的介绍。 Kubernetes介绍 Kubernetes是Google基于Borg开源的容器编排调度引擎，作为CNCF（Cloud Native Computing Foundation）最重要的组件之一，它的目标不仅仅是一个编排系统，而是提供一个规范，可以让你来描述集群的架构，定义服务的最终状态，kubernetes可以帮你将系统自动得达到和维持在这个状态。 更直白的说，Kubernetes用户可以通过编写一个yaml或者json格式的配置文件，也可以通过工具/代码生成或直接请求kubernetes API创建应用，该配置文件中包含了用户想要应用程序保持的状态，不论整个kubernetes集群中的个别主机发生什么问题，都不会影响应用程序的状态，你还可以通过改变该配置文件或请求kubernetes API来改变应用程序的状态。 12因素应用 12因素应用提出已经有几年的时间了，每个人对其可能都有自己的理解，切不可生搬硬套，也不一定所有云原生应用都必须符合这12条法则，其中有几条法则可能还有点争议，有人对其的解释和看法不同。 大家不要孤立的来看这每一个因素，将其与自己软件开发流程联系起来，这12个因素大致就是按照软件从开发到交付的流程顺序来写的。 图片 - 十二因素应用 1.基准代码 每个代码仓库（repo）都生成docker image保存到镜像仓库中，并使用唯一的ID管理，在Jenkins中使用编译时的ID。 2.依赖 显式得声明代码中的依赖，使用软件包管理工具声明，比如Go中的Glide。 3.配置 将配置与代码分离，应用部署到kubernete中可以使用容器的环境变量或ConfigMap挂载到容器中。 4.后端服务 把后端服务当作附加资源，实质上是计算存储分离和降低服务耦合，分解单体应用。 5.构建、发布、运行 严格分离构建和运行，每次修改代码生成新的镜像，重新发布，不能直接修改运行时的代码和配置。 6.进程 应用程序进程应该是无状态的，这意味着再次重启后还可以计算出原先的状态。 7.端口绑定 在kubernetes中每个Pod都有独立的IP，每个运行在Pod中的应用不必关心端口是否重复，只需在service中指定端口，集群内的service通过配置互相发现。 8.并发 每个容器都是一个进程，通过增加容器的副本数实现并发。 9.易处理 快速启动和优雅终止可最大化健壮性，kuberentes优秀的Pod生存周期控制。 10.开发环境与线上环境等价 在kubernetes中可以创建多个namespace，使用相同的镜像可以很方便的复制一套环境出来，镜像的使用可以很方便的部署一个后端服务。 11.日志 把日志当作事件流，使用stdout输出并收集汇聚起来，例如到ES中统一查看。 12.管理进程 后台管理任务当作一次性进程运行，kubectl exec进入容器内部操作。 另外，Cloud Native Go 这本书的作者，CapitalOne公司的Kevin Hoffman在TalkingData T11峰会上的High Level Cloud Native的演讲中讲述了云原生应用的15个因素，在原先的12因素应用的基础上又增加了如下三个因素： API优先 服务间的合约 团队协作的规约 文档化、规范化 RESTful或RPC 监控 实时监控远程应用 应用性能监控（APM） 应用健康监控 系统日志 不建议在线Debug 认证授权 不要等最后才去考虑应用的安全性 详细设计、明确声明、文档化 Bearer token、OAuth、OIDC认证 操作审计 详见High Level Cloud Native From Kevin Hoffman。 Kubernetes中的资源管理与容器设计模式 Kubernetes通过声明式配置，真正让开发人员能够理解应用的状态，并通过同一份配置可以立马启动一个一模一样的环境，大大提高了应用开发和部署的效率，其中kubernetes设计的多种资源类型可以帮助我们定义应用的运行状态，并使用资源配置来细粒度得明确限制应用的资源使用。 而容器生态的成熟是 Kubernetes 诞生的前提，在谈到容器的设计模式之前我们先来了解下容器生态，请看下图： 图片 - 容器生态 关于 Docker 容器的更多内容请参考 Docker最佳实践。 容器的设计模式 Kubernetes提供了多种资源对象，用户可以根据自己应用的特性加以选择。这些对象有： 类别 名称 资源对象 Pod、ReplicaSet、ReplicationController、Deployment、StatefulSet、DaemonSet、Job、CronJob、HorizontalPodAutoscaling 配置对象 Node、Namespace、Service、Secret、ConfigMap、Ingress、Label、ThirdPartyResource、 ServiceAccount 存储对象 Volume、Persistent Volume 策略对象 SecurityContext、ResourceQuota、LimitRange 在 Kubernetes 系统中，Kubernetes 对象 是持久化的条目。Kubernetes 使用这些条目去表示整个集群的状态。特别地，它们描述了如下信息： 什么容器化应用在运行（以及在哪个 Node 上） 可以被应用使用的资源 关于应用如何表现的策略，比如重启策略、升级策略，以及容错策略 Kubernetes 对象是 “目标性记录” —— 一旦创建对象，Kubernetes 系统将持续工作以确保对象存在。通过创建对象，可以有效地告知 Kubernetes 系统，所需要的集群工作负载看起来是什么样子的，这就是 Kubernetes 集群的 期望状态。 详见Kubernetes Handbook - Objects。 资源限制与配额 两层的资源限制与配置 Pod级别，最小的资源调度单位 Namespace级别，限制资源配额和每个Pod的资源使用区间 请参考Kubernetes中的ResourceQuota和LimitRange配置资源限额 管理Kubernetes集群 手工部署Kubernetes是一个很艰巨的活，你需要了解网络配置、docker的安装与使用、镜像仓库的构建、角色证书的创建、kubernetes的基本原理和构成、kubernetes应用程序的yaml文件编写等。 我编写了一本kubernetes-handbook可供大家免费阅读，该书记录了本人从零开始学习和使用Kubernetes的心路历程，着重于经验分享和总结，同时也会有相关的概念解析，希望能够帮助大家少踩坑，少走弯路。 部署Kubernetes集群 使用二进制部署 kubernetes 集群的所有组件和插件，而不是使用 kubeadm 等自动化方式来部署集群，同时开启了集群的TLS安全认证，这样可以帮助我们解系统各组件的交互原理，进而能快速解决实际问题。详见Kubernetes Handbook - 在CentOS上部署kubernetes1.6集群。 集群详情 Kubernetes 1.6.0 Docker 1.12.5（使用yum安装） Etcd 3.1.5 Flanneld 0.7 vxlan 网络 TLS 认证通信 (所有组件，如 etcd、kubernetes master 和 node) RBAC 授权 kublet TLS BootStrapping kubedns、dashboard、heapster(influxdb、grafana)、EFK(elasticsearch、fluentd、kibana) 集群插件 私有docker镜像仓库harbor（请自行部署，harbor提供离线安装包，直接使用docker-compose启动即可） 步骤介绍 创建 TLS 证书和秘钥 创建kubeconfig文件 创建高可用etcd集群 安装kubectl命令行工具 部署master节点 安装flannel网络插件 部署node节点 安装kubedns插件 安装dashboard插件 安装heapster插件 安装EFK插件 服务发现与负载均衡 Kubernetes在设计之初就充分考虑了针对容器的服务发现与负载均衡机制，提供了Service资源，并通过kube-proxy配合cloud provider来适应不同的应用场景。随着kubernetes用户的激增，用户场景的不断丰富，又产生了一些新的负载均衡机制。目前，kubernetes中的负载均衡大致可以分为以下几种机制，每种机制都有其特定的应用场景： Service：直接用Service提供cluster内部的负载均衡，并借助cloud provider提供的LB提供外部访问 Ingress：还是用Service提供cluster内部的负载均衡，但是通过自定义LB提供外部访问 Service Load Balancer：把load balancer直接跑在容器中，实现Bare Metal的Service Load Balancer Custom Load Balancer：自定义负载均衡，并替代kube-proxy，一般在物理部署Kubernetes时使用，方便接入公司已有的外部服务 详见Kubernetes Handbook - 服务发现与负载均衡。 持续集成与发布 图片 - 使用Jenkins进行持续集成与发布流程图 应用构建和发布流程说明： 用户向Gitlab提交代码，代码中必须包含Dockerfile 将代码提交到远程仓库 用户在发布应用时需要填写git仓库地址和分支、服务类型、服务名称、资源数量、实例个数，确定后触发Jenkins自动构建 Jenkins的CI流水线自动编译代码并打包成docker镜像推送到Harbor镜像仓库 Jenkins的CI流水线中包括了自定义脚本，根据我们已准备好的kubernetes的YAML模板，将其中的变量替换成用户输入的选项 生成应用的kubernetes YAML配置文件 更新Ingress的配置，根据新部署的应用的名称，在ingress的配置文件中增加一条路由信息 更新PowerDNS，向其中插入一条DNS记录，IP地址是边缘节点的IP地址。关于边缘节点，请查看边缘节点配置 Jenkins调用kubernetes的API，部署应用 日志收集与监控 基于现有的ELK日志收集方案，稍作改造，选用filebeat来收集日志，可以作为sidecar的形式跟应用运行在同一个Pod中，比较轻量级消耗资源比较少。 图片 - filebeat日志收集架构图 详见Kubernetes Handbook - 应用日志收集。 安全性与权限管理 Kubernetes是一个多租户的云平台，因此必须对用户的权限加以限制，对用户空间进行隔离。Kubernetes中的隔离主要包括这几种： 网络隔离：需要使用网络插件，比如calico。 资源隔离：kubernetes原生支持资源隔离，pod就是资源就是隔离和调度的最小单位，同时使用namespace限制用户空间和资源限额。 身份隔离：使用RBAC-基于角色的访问控制，多租户的身份认证和权限控制。 如何开发Kubernetes原生应用步骤介绍 当我们有了一个kubernetes集群后，如何在上面开发和部署应用，应该遵循怎样的流程？下面我将展示如何使用go语言开发和部署一个kubernetes native应用，使用wercker进行持续集成与持续发布，我将以一个很简单的前后端访问，获取伪造数据并展示的例子来说明。 云原生应用开发示例 我们将按照如下步骤来开发部署一个kubernetes原生应用并将它部署到kubernetes集群上开放给集群外访问： 服务API的定义 使用Go语言开发kubernetes原生应用 一个持续构建与发布工具与环境 使用traefik和VIP做边缘节点提供外部访问路由 我写了两个示例用于演示，开发部署一个伪造的 metric 并显示在 web 页面上，包括两个service： k8s-app-monitor-test：生成模拟的监控数据，发送http请求，获取json返回值 K8s-app-monitor-agent：获取监控数据并绘图，访问浏览器获取图表 定义API生成API文档 使用API blueprint格式，定义API文档，格式类似于markdown，再使用aglio生成HTML文档。 图片 - API文档 详见：如何开发部署kubernetes native应用。 如何迁移到云原生应用架构 Pivotal 是云原生应用的提出者，并推出了 Pivotal Cloud Foundry 云原生应用平台和 Spring 开源 Java 开发框架，成为云原生应用架构中先驱者和探路者。 原书作于2015年，其中的示例主要针对 Java 应用，实际上也适用于任何应用类型，云原生应用架构适用于异构语言的程序开发，不仅仅是针对 Java 语言的程序开发。截止到本人翻译本书时，云原生应用生态系统已经初具规模，CNCF 成员不断发展壮大，基于 Cloud Native 的创业公司不断涌现，kubernetes 引领容器编排潮流，和 Service Mesh 技术（如 Linkerd 和 Istio） 的出现，Go 语言的兴起（参考另一本书 Cloud Native Go）等为我们将应用迁移到云原生架构的提供了更多的方案选择。 迁移到云原生应用架构指南 指出了迁移到云原生应用架构需要做出的企业文化、组织架构和技术变革，并给出了迁移指南。 主要讨论的应用程序架构包括： 十二因素应用程序：云原生应用程序架构模式的集合 微服务：独立部署的服务，只做一件事情 自助服务的敏捷基础设施：快速，可重复和一致地提供应用环境和后台服务的平台 基于API的协作：发布和版本化的API，允许在云原生应用程序架构中的服务之间进行交互 抗压性：根据压力变强的系统 详见：迁移到云原生应用架构 迁移案例解析 迁移步骤示意图如下： 图片 - 迁移步骤示意图 步骤说明： 将原有应用拆解为服务 容器化、制作镜像 准备应用配置文件 准备kubernetes YAML文件 编写bootstarp脚本 创建ConfigMaps 详见：迁移传统应用到Kubernetes步骤详解——以Hadoop YARN为例。 Service mesh基本原理和示例介绍 Service mesh现在一般被翻译作服务网格，目前主流的Service mesh有如下两款： Istio：IBM、Google、Lyft共同开源，详细文档见Istio官方文档中文版 Linkerd：原Twitter工程师开发，现为CNCF中的项目之一 什么是Service mesh 如果用一句话来解释什么是 Service Mesh，可以将它比作是应用程序或者说微服务间的 TCP/IP，负责服务之间的网络调用、限流、熔断和监控。对于编写应用程序来说一般无须关心 TCP/IP 这一层（比如通过 HTTP 协议的 RESTful 应用），同样使用 Service Mesh 也就无须关系服务之间的那些原来是通过应用程序或者其他框架实现的事情，比如 Spring Cloud、OSS，现在只要交给 Service Mesh 就可以了。 图片 - service mesh架构图 详见什么是 service mesh - jimmysong.io。 Service mesh使用指南 两款Service mesh各有千秋，我分别写了他们的使用案例指南： 微服务管理框架service mesh——Linkerd安装试用笔记 微服务管理框架service mesh——Istio安装试用笔记 更多关于 Service Mesh 的内容请访问 Service Mesh 中文网。 使用案例 Kubernetes作为云原生计算的基本组件之一，开源2年时间以来热度与日俱增，它可以跟我们的生产结合，擦出很多火花，比如FaaS和Serverless类应用，都很适合运行在kubernetes上。 关于Cloud Native开源软件生态请参考 Awesome Cloud Native - jimmysong.io。 DevOps 下面是社区中kubernetes开源爱好者的分享内容，我觉得是对kubernetes在DevOps中应用的很好的形式值得大家借鉴。 真正践行DevOps，让开发人员在掌握自己的开发和测试环境，让环境一致，让开发效率提升，让运维没有堆积如山的tickets，让监控更加精准，从kubernetes平台开始。 行动指南 根据环境（比如开发、测试、生产）划分namespace，也可以根据项目来划分 再为每个用户划分一个namespace、创建一个serviceaccount和kubeconfig文件，不同namespace间的资源隔离，目前不隔离网络，不同namespace间的服务可以互相访问 创建yaml模板，降低编写kubernetes yaml文件编写难度 在kubectl命令上再封装一层，增加用户身份设置和环境初始化操作，简化kubectl命令和常用功能 管理员通过dashboard查看不同namespace的状态，也可以使用它来使操作更便捷 所有应用的日志统一收集到ElasticSearch中，统一日志访问入口 可以通过Grafana查看所有namespace中的应用的状态和kubernetes集群本身的状态 需要持久化的数据保存在分布式存储中，例如GlusterFS或Ceph中 使用Kibana查看日志 日志字段中包括了应用的标签、容器名称、主机名称、宿主机名称、IP地址、时间、 图片 - kibana界面 使用Grafana查看应用状态 注：感谢【K8S🤘Cloud Native实战群】尊贵的黄金会员小刚同学提供下面的Grafana监控图🙏 监控分类示意图： 图片 - Grafana界面示意图1 Kubernetes集群全局监控图1 该监控图可以看到集群硬件使用情况。 图片 - Grafana界面示意图2 Kubernetes全局监控图2 该监控可以看到单个用户的namespace下的所有资源的使用情况。 图片 - Grafana界面示意图3 Spark on Kubernetes TL;DR https://jimmysong.io/spark-on-k8s Spark原生支持standalone、mesos和YARN资源调度，现已支持Kubernetes原生调度，详见运行支持kubernetes原生调度的spark程序-Spark on Kubernetes。 为何要使用spark on kubernetes 使用kubernetes原生调度的spark on kubernetes是对原先的spark on yarn和yarn on docker的改变是革命性的，主要表现在以下几点： Kubernetes原生调度：不再需要二层调度，直接使用kubernetes的资源调度功能，跟其他应用共用整个kubernetes管理的资源池； 资源隔离，粒度更细：原先yarn中的queue在spark on kubernetes中已不存在，取而代之的是kubernetes中原生的namespace，可以为每个用户分别指定一个namespace，限制用户的资源quota； 细粒度的资源分配：可以给每个spark任务指定资源限制，实际指定多少资源就使用多少资源，因为没有了像yarn那样的二层调度（圈地式的），所以可以更高效和细粒度的使用资源； 监控的变革：因为做到了细粒度的资源分配，所以可以对用户提交的每一个任务做到资源使用的监控，从而判断用户的资源使用情况，所有的metric都记录在数据库中，甚至可以为每个用户的每次任务提交计量； 日志的变革：用户不再通过yarn的web页面来查看任务状态，而是通过pod的log来查看，可将所有的kuberentes中的应用的日志等同看待收集起来，然后可以根据标签查看对应应用的日志； 如何提交任务 仍然使用spark-submit提交spark任务，可以直接指定kubernetes API server地址，下面的命令提交本地jar包到kubernetes集群上运行，同时指定了运行任务的用户、提交命名的用户、运行的excutor实例数、driver和executor的资源限制、使用的spark版本等信息。 详细使用说明见Apache Spark on Kubernetes用户指南 - jimmysong.io。 ./spark-submit \\ --deploy-mode cluster \\ --class com.talkingdata.alluxio.hadooptest \\ --master k8s://https://172.20.0.113:6443 \\ --kubernetes-namespace spark-cluster \\ --conf spark.kubernetes.driverEnv.SPARK_USER=hadoop \\ --conf spark.kubernetes.driverEnv.HADOOP_USER_NAME=hadoop \\ --conf spark.executorEnv.HADOOP_USER_NAME=hadoop \\ --conf spark.executorEnv.SPARK_USER=hadoop \\ --conf spark.kubernetes.authenticate.driver.serviceAccountName=spark \\ --conf spark.driver.memory=100G \\ --conf spark.executor.memory=10G \\ --conf spark.driver.cores=30 \\ --conf spark.executor.cores=2 \\ --conf spark.driver.maxResultSize=10240m \\ --conf spark.kubernetes.driver.limit.cores=32 \\ --conf spark.kubernetes.executor.limit.cores=3 \\ --conf spark.kubernetes.executor.memoryOverhead=2g \\ --conf spark.executor.instances=5 \\ --conf spark.app.name=spark-pi \\ --conf spark.kubernetes.driver.docker.image=sz-pg-oam-docker-hub-001.tendcloud.com/library/spark-driver:v2.1.0-kubernetes-0.3.1-1 \\ --conf spark.kubernetes.executor.docker.image=sz-pg-oam-docker-hub-001.tendcloud.com/library/spark-executor:v2.1.0-kubernetes-0.3.1-1 \\ --conf spark.kubernetes.initcontainer.docker.image=sz-pg-oam-docker-hub-001.tendcloud.com/library/spark-init:v2.1.0-kubernetes-0.3.1-1 \\ --conf spark.kubernetes.resourceStagingServer.uri=http://172.20.0.114:31000 \\ ~/Downloads/tendcloud_2.10-1.0.jar 监控 下图是从Kubernetes dashboard上看到的spark-cluster这个namespace上运行的应用情况。 图片 - dashboard 下图是从Grafana监控页面上查看到的某个executor资源占用情况。 图片 - Grafana 参考 迁移到云原生应用架构指南 Cloud Native Go - 已由电子工业出版社出版 Cloud Native Python - 将由电子工业出版社出版 Istio Service Mesh 中文文档 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-20 17:10:30 "},"concepts/":{"url":"concepts/","title":"Kubernetes架构","keywords":"","body":"Kubernetes架构 Kubernetes最初源于谷歌内部的Borg，提供了面向应用的容器集群部署和管理系统。Kubernetes的目标旨在消除编排物理/虚拟计算，网络和存储基础设施的负担，并使应用程序运营商和开发人员完全将重点放在以容器为中心的原语上进行自助运营。Kubernetes 也提供稳定、兼容的基础（平台），用于构建定制化的workflows 和更高级的自动化任务。 Kubernetes 具备完善的集群管理能力，包括多层次的安全防护和准入机制、多租户应用支撑能力、透明的服务注册和服务发现机制、内建负载均衡器、故障发现和自我修复能力、服务滚动升级和在线扩容、可扩展的资源自动调度机制、多粒度的资源配额管理能力。 Kubernetes 还提供完善的管理工具，涵盖开发、部署测试、运维监控等各个环节。 Borg简介 Borg是谷歌内部的大规模集群管理系统，负责对谷歌内部很多核心服务的调度和管理。Borg的目的是让用户能够不必操心资源管理的问题，让他们专注于自己的核心业务，并且做到跨多个数据中心的资源利用率最大化。 Borg主要由BorgMaster、Borglet、borgcfg和Scheduler组成，如下图所示 图片 - Borg架构 BorgMaster是整个集群的大脑，负责维护整个集群的状态，并将数据持久化到Paxos存储中； Scheduer负责任务的调度，根据应用的特点将其调度到具体的机器上去； Borglet负责真正运行任务（在容器中）； borgcfg是Borg的命令行工具，用于跟Borg系统交互，一般通过一个配置文件来提交任务。 Kubernetes架构 Kubernetes借鉴了Borg的设计理念，比如Pod、Service、Labels和单Pod单IP等。Kubernetes的整体架构跟Borg非常像，如下图所示 图片 - Kubernetes架构 Kubernetes主要由以下几个核心组件组成： etcd保存了整个集群的状态； apiserver提供了资源操作的唯一入口，并提供认证、授权、访问控制、API注册和发现等机制； controller manager负责维护集群的状态，比如故障检测、自动扩展、滚动更新等； scheduler负责资源的调度，按照预定的调度策略将Pod调度到相应的机器上； kubelet负责维护容器的生命周期，同时也负责Volume（CVI）和网络（CNI）的管理； Container runtime负责镜像管理以及Pod和容器的真正运行（CRI）； kube-proxy负责为Service提供cluster内部的服务发现和负载均衡； 除了核心组件，还有一些推荐的Add-ons： kube-dns负责为整个集群提供DNS服务 Ingress Controller为服务提供外网入口 Heapster提供资源监控 Dashboard提供GUI Federation提供跨可用区的集群 Kubernetes架构示意图 整体架构 下图清晰表明了kubernetes的架构设计以及组件之间的通信协议。 图片 - Kuberentes架构（图片来自于网络） 下面是更抽象的一个视图： 图片 - kubernetes整体架构示意图 Master架构 图片 - Kubernetes master架构示意图 Node架构 图片 - kubernetes node架构示意图 分层架构 Kubernetes设计理念和功能其实就是一个类似Linux的分层架构，如下图所示 图片 - Kubernetes分层架构示意图 核心层：Kubernetes最核心的功能，对外提供API构建高层的应用，对内提供插件式应用执行环境 应用层：部署（无状态应用、有状态应用、批处理任务、集群应用等）和路由（服务发现、DNS解析等） 管理层：系统度量（如基础设施、容器和网络的度量），自动化（如自动扩展、动态Provision等）以及策略管理（RBAC、Quota、PSP、NetworkPolicy等） 接口层：kubectl命令行工具、客户端SDK以及集群联邦 生态系统：在接口层之上的庞大容器集群管理调度的生态系统，可以划分为两个范畴 Kubernetes外部：日志、监控、配置管理、CI、CD、Workflow、FaaS、OTS应用、ChatOps等 Kubernetes内部：CRI、CNI、CVI、镜像仓库、Cloud Provider、集群自身的配置和管理等 关于分层架构，可以关注下Kubernetes社区正在推进的Kbernetes architectual roadmap和slide。 参考文档 Kubernetes design and architecture http://queue.acm.org/detail.cfm?id=2898444 http://static.googleusercontent.com/media/research.google.com/zh-CN//pubs/archive/43438.pdf http://thenewstack.io/kubernetes-an-overview Kbernetes architectual roadmap和slide Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-08 22:04:06 "},"concepts/concepts.html":{"url":"concepts/concepts.html","title":"设计理念","keywords":"","body":"Kubernetes的设计理念 Kubernetes设计理念与分布式系统 分析和理解Kubernetes的设计理念可以使我们更深入地了解Kubernetes系统，更好地利用它管理分布式部署的云原生应用，另一方面也可以让我们借鉴其在分布式系统设计方面的经验。 分层架构 Kubernetes设计理念和功能其实就是一个类似Linux的分层架构，如下图所示 图片 - 分层架构示意图 核心层：Kubernetes最核心的功能，对外提供API构建高层的应用，对内提供插件式应用执行环境 应用层：部署（无状态应用、有状态应用、批处理任务、集群应用等）和路由（服务发现、DNS解析等） 管理层：系统度量（如基础设施、容器和网络的度量），自动化（如自动扩展、动态Provision等）以及策略管理（RBAC、Quota、PSP、NetworkPolicy等） 接口层：kubectl命令行工具、客户端SDK以及集群联邦 生态系统：在接口层之上的庞大容器集群管理调度的生态系统，可以划分为两个范畴 Kubernetes外部：日志、监控、配置管理、CI、CD、Workflow、FaaS、OTS应用、ChatOps等 Kubernetes内部：CRI、CNI、CVI、镜像仓库、Cloud Provider、集群自身的配置和管理等 API设计原则 对于云计算系统，系统API实际上处于系统设计的统领地位，正如本文前面所说，kubernetes集群系统每支持一项新功能，引入一项新技术，一定会新引入对应的API对象，支持对该功能的管理操作，理解掌握的API，就好比抓住了kubernetes系统的牛鼻子。Kubernetes系统API的设计有以下几条原则： 所有API应该是声明式的。正如前文所说，声明式的操作，相对于命令式操作，对于重复操作的效果是稳定的，这对于容易出现数据丢失或重复的分布式环境来说是很重要的。另外，声明式操作更容易被用户使用，可以使系统向用户隐藏实现的细节，隐藏实现的细节的同时，也就保留了系统未来持续优化的可能性。此外，声明式的API，同时隐含了所有的API对象都是名词性质的，例如Service、Volume这些API都是名词，这些名词描述了用户所期望得到的一个目标分布式对象。 API对象是彼此互补而且可组合的。这里面实际是鼓励API对象尽量实现面向对象设计时的要求，即“高内聚，松耦合”，对业务相关的概念有一个合适的分解，提高分解出来的对象的可重用性。事实上，Kubernetes这种分布式系统管理平台，也是一种业务系统，只不过它的业务就是调度和管理容器服务。 高层API以操作意图为基础设计。如何能够设计好API，跟如何能用面向对象的方法设计好应用系统有相通的地方，高层设计一定是从业务出发，而不是过早的从技术实现出发。因此，针对Kubernetes的高层API设计，一定是以Kubernetes的业务为基础出发，也就是以系统调度管理容器的操作意图为基础设计。 低层API根据高层API的控制需要设计。设计实现低层API的目的，是为了被高层API使用，考虑减少冗余、提高重用性的目的，低层API的设计也要以需求为基础，要尽量抵抗受技术实现影响的诱惑。 尽量避免简单封装，不要有在外部API无法显式知道的内部隐藏的机制。简单的封装，实际没有提供新的功能，反而增加了对所封装API的依赖性。内部隐藏的机制也是非常不利于系统维护的设计方式，例如PetSet和ReplicaSet，本来就是两种Pod集合，那么Kubernetes就用不同API对象来定义它们，而不会说只用同一个ReplicaSet，内部通过特殊的算法再来区分这个ReplicaSet是有状态的还是无状态。 API操作复杂度与对象数量成正比。这一条主要是从系统性能角度考虑，要保证整个系统随着系统规模的扩大，性能不会迅速变慢到无法使用，那么最低的限定就是API的操作复杂度不能超过O(N)，N是对象的数量，否则系统就不具备水平伸缩性了。 API对象状态不能依赖于网络连接状态。由于众所周知，在分布式环境下，网络连接断开是经常发生的事情，因此要保证API对象状态能应对网络的不稳定，API对象的状态就不能依赖于网络连接状态。 尽量避免让操作机制依赖于全局状态，因为在分布式系统中要保证全局状态的同步是非常困难的。 控制机制设计原则 控制逻辑应该只依赖于当前状态。这是为了保证分布式系统的稳定可靠，对于经常出现局部错误的分布式系统，如果控制逻辑只依赖当前状态，那么就非常容易将一个暂时出现故障的系统恢复到正常状态，因为你只要将该系统重置到某个稳定状态，就可以自信的知道系统的所有控制逻辑会开始按照正常方式运行。 假设任何错误的可能，并做容错处理。在一个分布式系统中出现局部和临时错误是大概率事件。错误可能来自于物理系统故障，外部系统故障也可能来自于系统自身的代码错误，依靠自己实现的代码不会出错来保证系统稳定其实也是难以实现的，因此要设计对任何可能错误的容错处理。 尽量避免复杂状态机，控制逻辑不要依赖无法监控的内部状态。因为分布式系统各个子系统都是不能严格通过程序内部保持同步的，所以如果两个子系统的控制逻辑如果互相有影响，那么子系统就一定要能互相访问到影响控制逻辑的状态，否则，就等同于系统里存在不确定的控制逻辑。 假设任何操作都可能被任何操作对象拒绝，甚至被错误解析。由于分布式系统的复杂性以及各子系统的相对独立性，不同子系统经常来自不同的开发团队，所以不能奢望任何操作被另一个子系统以正确的方式处理，要保证出现错误的时候，操作级别的错误不会影响到系统稳定性。 每个模块都可以在出错后自动恢复。由于分布式系统中无法保证系统各个模块是始终连接的，因此每个模块要有自我修复的能力，保证不会因为连接不到其他模块而自我崩溃。 每个模块都可以在必要时优雅地降级服务。所谓优雅地降级服务，是对系统鲁棒性的要求，即要求在设计实现模块时划分清楚基本功能和高级功能，保证基本功能不会依赖高级功能，这样同时就保证了不会因为高级功能出现故障而导致整个模块崩溃。根据这种理念实现的系统，也更容易快速地增加新的高级功能，以为不必担心引入高级功能影响原有的基本功能。 Kubernetes的核心技术概念和API对象 API对象是Kubernetes集群中的管理操作单元。Kubernetes集群系统每支持一项新功能，引入一项新技术，一定会新引入对应的API对象，支持对该功能的管理操作。例如副本集Replica Set对应的API对象是RS。 每个API对象都有3大类属性：元数据metadata、规范spec和状态status。元数据是用来标识API对象的，每个对象都至少有3个元数据：namespace，name和uid；除此以外还有各种各样的标签labels用来标识和匹配不同的对象，例如用户可以用标签env来标识区分不同的服务部署环境，分别用env=dev、env=testing、env=production来标识开发、测试、生产的不同服务。规范描述了用户期望Kubernetes集群中的分布式系统达到的理想状态（Desired State），例如用户可以通过复制控制器Replication Controller设置期望的Pod副本数为3；status描述了系统实际当前达到的状态（Status），例如系统当前实际的Pod副本数为2；那么复制控制器当前的程序逻辑就是自动启动新的Pod，争取达到副本数为3。 Kubernetes中所有的配置都是通过API对象的spec去设置的，也就是用户通过配置系统的理想状态来改变系统，这是Kubernetes重要设计理念之一，即所有的操作都是声明式（Declarative）的而不是命令式（Imperative）的。声明式操作在分布式系统中的好处是稳定，不怕丢操作或运行多次，例如设置副本数为3的操作运行多次也还是一个结果，而给副本数加1的操作就不是声明式的，运行多次结果就错了。 Pod Kubernetes有很多技术概念，同时对应很多API对象，最重要的也是最基础的是Pod。Pod是在Kubernetes集群中运行部署应用或服务的最小单元，它是可以支持多容器的。Pod的设计理念是支持多个容器在一个Pod中共享网络地址和文件系统，可以通过进程间通信和文件共享这种简单高效的方式组合完成服务。Pod对多容器的支持是K8最基础的设计理念。比如你运行一个操作系统发行版的软件仓库，一个Nginx容器用来发布软件，另一个容器专门用来从源仓库做同步，这两个容器的镜像不太可能是一个团队开发的，但是他们一块儿工作才能提供一个微服务；这种情况下，不同的团队各自开发构建自己的容器镜像，在部署的时候组合成一个微服务对外提供服务。 Pod是Kubernetes集群中所有业务类型的基础，可以看作运行在K8集群中的小机器人，不同类型的业务就需要不同类型的小机器人去执行。目前Kubernetes中的业务主要可以分为长期伺服型（long-running）、批处理型（batch）、节点后台支撑型（node-daemon）和有状态应用型（stateful application）；分别对应的小机器人控制器为Deployment、Job、DaemonSet和PetSet，本文后面会一一介绍。 副本控制器（Replication Controller，RC） RC是Kubernetes集群中最早的保证Pod高可用的API对象。通过监控运行中的Pod来保证集群中运行指定数目的Pod副本。指定的数目可以是多个也可以是1个；少于指定数目，RC就会启动运行新的Pod副本；多于指定数目，RC就会杀死多余的Pod副本。即使在指定数目为1的情况下，通过RC运行Pod也比直接运行Pod更明智，因为RC也可以发挥它高可用的能力，保证永远有1个Pod在运行。RC是Kubernetes较早期的技术概念，只适用于长期伺服型的业务类型，比如控制小机器人提供高可用的Web服务。 副本集（Replica Set，RS） RS是新一代RC，提供同样的高可用能力，区别主要在于RS后来居上，能支持更多种类的匹配模式。副本集对象一般不单独使用，而是作为Deployment的理想状态参数使用。 部署（Deployment） 部署表示用户对Kubernetes集群的一次更新操作。部署是一个比RS应用模式更广的API对象，可以是创建一个新的服务，更新一个新的服务，也可以是滚动升级一个服务。滚动升级一个服务，实际是创建一个新的RS，然后逐渐将新RS中副本数增加到理想状态，将旧RS中的副本数减小到0的复合操作；这样一个复合操作用一个RS是不太好描述的，所以用一个更通用的Deployment来描述。以Kubernetes的发展方向，未来对所有长期伺服型的的业务的管理，都会通过Deployment来管理。 服务（Service） RC、RS和Deployment只是保证了支撑服务的微服务Pod的数量，但是没有解决如何访问这些服务的问题。一个Pod只是一个运行服务的实例，随时可能在一个节点上停止，在另一个节点以一个新的IP启动一个新的Pod，因此不能以确定的IP和端口号提供服务。要稳定地提供服务需要服务发现和负载均衡能力。服务发现完成的工作，是针对客户端访问的服务，找到对应的的后端服务实例。在K8集群中，客户端需要访问的服务就是Service对象。每个Service会对应一个集群内部有效的虚拟IP，集群内部通过虚拟IP访问一个服务。在Kubernetes集群中微服务的负载均衡是由Kube-proxy实现的。Kube-proxy是Kubernetes集群内部的负载均衡器。它是一个分布式代理服务器，在Kubernetes的每个节点上都有一个；这一设计体现了它的伸缩性优势，需要访问服务的节点越多，提供负载均衡能力的Kube-proxy就越多，高可用节点也随之增多。与之相比，我们平时在服务器端做个反向代理做负载均衡，还要进一步解决反向代理的负载均衡和高可用问题。 任务（Job） Job是Kubernetes用来控制批处理型任务的API对象。批处理业务与长期伺服业务的主要区别是批处理业务的运行有头有尾，而长期伺服业务在用户不停止的情况下永远运行。Job管理的Pod根据用户的设置把任务成功完成就自动退出了。成功完成的标志根据不同的spec.completions策略而不同：单Pod型任务有一个Pod成功就标志完成；定数成功型任务保证有N个任务全部成功；工作队列型任务根据应用确认的全局成功而标志成功。 后台支撑服务集（DaemonSet） 长期伺服型和批处理型服务的核心在业务应用，可能有些节点运行多个同类业务的Pod，有些节点上又没有这类Pod运行；而后台支撑型服务的核心关注点在Kubernetes集群中的节点（物理机或虚拟机），要保证每个节点上都有一个此类Pod运行。节点可能是所有集群节点也可能是通过nodeSelector选定的一些特定节点。典型的后台支撑型服务包括，存储，日志和监控等在每个节点上支持Kubernetes集群运行的服务。 有状态服务集（PetSet） Kubernetes在1.3版本里发布了Alpha版的PetSet功能。在云原生应用的体系里，有下面两组近义词；第一组是无状态（stateless）、牲畜（cattle）、无名（nameless）、可丢弃（disposable）；第二组是有状态（stateful）、宠物（pet）、有名（having name）、不可丢弃（non-disposable）。RC和RS主要是控制提供无状态服务的，其所控制的Pod的名字是随机设置的，一个Pod出故障了就被丢弃掉，在另一个地方重启一个新的Pod，名字变了、名字和启动在哪儿都不重要，重要的只是Pod总数；而PetSet是用来控制有状态服务，PetSet中的每个Pod的名字都是事先确定的，不能更改。PetSet中Pod的名字的作用，并不是《千与千寻》的人性原因，而是关联与该Pod对应的状态。 对于RC和RS中的Pod，一般不挂载存储或者挂载共享存储，保存的是所有Pod共享的状态，Pod像牲畜一样没有分别（这似乎也确实意味着失去了人性特征）；对于PetSet中的Pod，每个Pod挂载自己独立的存储，如果一个Pod出现故障，从其他节点启动一个同样名字的Pod，要挂在上原来Pod的存储继续以它的状态提供服务。 适合于PetSet的业务包括数据库服务MySQL和PostgreSQL，集群化管理服务Zookeeper、etcd等有状态服务。PetSet的另一种典型应用场景是作为一种比普通容器更稳定可靠的模拟虚拟机的机制。传统的虚拟机正是一种有状态的宠物，运维人员需要不断地维护它，容器刚开始流行时，我们用容器来模拟虚拟机使用，所有状态都保存在容器里，而这已被证明是非常不安全、不可靠的。使用PetSet，Pod仍然可以通过漂移到不同节点提供高可用，而存储也可以通过外挂的存储来提供高可靠性，PetSet做的只是将确定的Pod与确定的存储关联起来保证状态的连续性。PetSet还只在Alpha阶段，后面的设计如何演变，我们还要继续观察。 集群联邦（Federation） Kubernetes在1.3版本里发布了beta版的Federation功能。在云计算环境中，服务的作用距离范围从近到远一般可以有：同主机（Host，Node）、跨主机同可用区（Available Zone）、跨可用区同地区（Region）、跨地区同服务商（Cloud Service Provider）、跨云平台。Kubernetes的设计定位是单一集群在同一个地域内，因为同一个地区的网络性能才能满足Kubernetes的调度和计算存储连接要求。而联合集群服务就是为提供跨Region跨服务商Kubernetes集群服务而设计的。 每个Kubernetes Federation有自己的分布式存储、API Server和Controller Manager。用户可以通过Federation的API Server注册该Federation的成员Kubernetes Cluster。当用户通过Federation的API Server创建、更改API对象时，Federation API Server会在自己所有注册的子Kubernetes Cluster都创建一份对应的API对象。在提供业务请求服务时，Kubernetes Federation会先在自己的各个子Cluster之间做负载均衡，而对于发送到某个具体Kubernetes Cluster的业务请求，会依照这个Kubernetes Cluster独立提供服务时一样的调度模式去做Kubernetes Cluster内部的负载均衡。而Cluster之间的负载均衡是通过域名服务的负载均衡来实现的。 所有的设计都尽量不影响Kubernetes Cluster现有的工作机制，这样对于每个子Kubernetes集群来说，并不需要更外层的有一个Kubernetes Federation，也就是意味着所有现有的Kubernetes代码和机制不需要因为Federation功能有任何变化。 存储卷（Volume） Kubernetes集群中的存储卷跟Docker的存储卷有些类似，只不过Docker的存储卷作用范围为一个容器，而Kubernetes的存储卷的生命周期和作用范围是一个Pod。每个Pod中声明的存储卷由Pod中的所有容器共享。Kubernetes支持非常多的存储卷类型，特别的，支持多种公有云平台的存储，包括AWS，Google和Azure云；支持多种分布式存储包括GlusterFS和Ceph；也支持较容易使用的主机本地目录hostPath和NFS。Kubernetes还支持使用Persistent Volume Claim即PVC这种逻辑存储，使用这种存储，使得存储的使用者可以忽略后台的实际存储技术（例如AWS，Google或GlusterFS和Ceph），而将有关存储实际技术的配置交给存储管理员通过Persistent Volume来配置。 持久存储卷（Persistent Volume，PV）和持久存储卷声明（Persistent Volume Claim，PVC） PV和PVC使得Kubernetes集群具备了存储的逻辑抽象能力，使得在配置Pod的逻辑里可以忽略对实际后台存储技术的配置，而把这项配置的工作交给PV的配置者，即集群的管理者。存储的PV和PVC的这种关系，跟计算的Node和Pod的关系是非常类似的；PV和Node是资源的提供者，根据集群的基础设施变化而变化，由Kubernetes集群管理员配置；而PVC和Pod是资源的使用者，根据业务服务的需求变化而变化，有Kubernetes集群的使用者即服务的管理员来配置。 节点（Node） Kubernetes集群中的计算能力由Node提供，最初Node称为服务节点Minion，后来改名为Node。Kubernetes集群中的Node也就等同于Mesos集群中的Slave节点，是所有Pod运行所在的工作主机，可以是物理机也可以是虚拟机。不论是物理机还是虚拟机，工作主机的统一特征是上面要运行kubelet管理节点上运行的容器。 密钥对象（Secret） Secret是用来保存和传递密码、密钥、认证凭证这些敏感信息的对象。使用Secret的好处是可以避免把敏感信息明文写在配置文件里。在Kubernetes集群中配置和使用服务不可避免的要用到各种敏感信息实现登录、认证等功能，例如访问AWS存储的用户名密码。为了避免将类似的敏感信息明文写在所有需要使用的配置文件中，可以将这些信息存入一个Secret对象，而在配置文件中通过Secret对象引用这些敏感信息。这种方式的好处包括：意图明确，避免重复，减少暴漏机会。 用户帐户（User Account）和服务帐户（Service Account） 顾名思义，用户帐户为人提供账户标识，而服务账户为计算机进程和Kubernetes集群中运行的Pod提供账户标识。用户帐户和服务帐户的一个区别是作用范围；用户帐户对应的是人的身份，人的身份与服务的namespace无关，所以用户账户是跨namespace的；而服务帐户对应的是一个运行中程序的身份，与特定namespace是相关的。 命名空间（Namespace） 命名空间为Kubernetes集群提供虚拟的隔离作用，Kubernetes集群初始有两个命名空间，分别是默认命名空间default和系统命名空间kube-system，除此以外，管理员可以可以创建新的命名空间满足需要。 RBAC访问授权 Kubernetes在1.3版本中发布了alpha版的基于角色的访问控制（Role-based Access Control，RBAC）的授权模式。相对于基于属性的访问控制（Attribute-based Access Control，ABAC），RBAC主要是引入了角色（Role）和角色绑定（RoleBinding）的抽象概念。在ABAC中，Kubernetes集群中的访问策略只能跟用户直接关联；而在RBAC中，访问策略可以跟某个角色关联，具体的用户在跟一个或多个角色相关联。显然，RBAC像其他新功能一样，每次引入新功能，都会引入新的API对象，从而引入新的概念抽象，而这一新的概念抽象一定会使集群服务管理和使用更容易扩展和重用。 总结 从Kubernetes的系统架构、技术概念和设计理念，我们可以看到Kubernetes系统最核心的两个设计理念：一个是容错性，一个是易扩展性。容错性实际是保证Kubernetes系统稳定性和安全性的基础，易扩展性是保证Kubernetes对变更友好，可以快速迭代增加新功能的基础。 按照分布式系统一致性算法Paxos发明人计算机科学家Leslie Lamport的理念，一个分布式系统有两类特性：安全性Safety和活性Liveness。安全性保证系统的稳定，保证系统不会崩溃，不会出现业务错误，不会做坏事，是严格约束的；活性使得系统可以提供功能，提高性能，增加易用性，让系统可以在用户“看到的时间内”做些好事，是尽力而为的。Kubernetes系统的设计理念正好与Lamport安全性与活性的理念不谋而合，也正是因为Kubernetes在引入功能和技术的时候，非常好地划分了安全性和活性，才可以让Kubernetes能有这么快版本迭代，快速引入像RBAC、Federation和PetSet这种新功能。 原文地址：《Kubernetes与云原生应用》系列之Kubernetes的系统架构与设计理念 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-21 11:59:28 "},"concepts/objects.html":{"url":"concepts/objects.html","title":"资源对象与基本概念解析","keywords":"","body":"Objects 以下列举的内容都是 kubernetes 中的 Object，这些对象都可以在 yaml 文件中作为一种 API 类型来配置。 Pod Node Namespace Service Volume PersistentVolume Deployment Secret StatefulSet DaemonSet ServiceAccount ReplicationController ReplicaSet Job CronJob SecurityContext ResourceQuota LimitRange HorizontalPodAutoscaling Ingress ConfigMap Label ThirdPartyResources 我将它们简单的分类为以下几种资源对象： 类别 名称 资源对象 Pod、ReplicaSet、ReplicationController、Deployment、StatefulSet、DaemonSet、Job、CronJob、HorizontalPodAutoscaling 配置对象 Node、Namespace、Service、Secret、ConfigMap、Ingress、Label、ThirdPartyResource、 ServiceAccount 存储对象 Volume、Persistent Volume 策略对象 SecurityContext、ResourceQuota、LimitRange 理解 kubernetes 中的对象 在 Kubernetes 系统中，Kubernetes 对象 是持久化的条目。Kubernetes 使用这些条目去表示整个集群的状态。特别地，它们描述了如下信息： 什么容器化应用在运行（以及在哪个 Node 上） 可以被应用使用的资源 关于应用如何表现的策略，比如重启策略、升级策略，以及容错策略 Kubernetes 对象是 “目标性记录” —— 一旦创建对象，Kubernetes 系统将持续工作以确保对象存在。通过创建对象，可以有效地告知 Kubernetes 系统，所需要的集群工作负载看起来是什么样子的，这就是 Kubernetes 集群的 期望状态。 与 Kubernetes 对象工作 —— 是否创建、修改，或者删除 —— 需要使用 Kubernetes API。当使用 kubectl 命令行接口时，比如，CLI 会使用必要的 Kubernetes API 调用，也可以在程序中直接使用 Kubernetes API。为了实现该目标，Kubernetes 当前提供了一个 golang 客户端库 ，其它语言库（例如Python）也正在开发中。 对象 Spec 与状态 每个 Kubernetes 对象包含两个嵌套的对象字段，它们负责管理对象的配置：对象 spec 和 对象 status。spec 必须提供，它描述了对象的 期望状态—— 希望对象所具有的特征。status 描述了对象的 实际状态，它是由 Kubernetes 系统提供和更新。在任何时刻，Kubernetes 控制平面一直处于活跃状态，管理着对象的实际状态以与我们所期望的状态相匹配。 例如，Kubernetes Deployment 对象能够表示运行在集群中的应用。当创建 Deployment 时，可能需要设置 Deployment 的 spec，以指定该应用需要有 3 个副本在运行。Kubernetes 系统读取 Deployment spec，启动我们所期望的该应用的 3 个实例 —— 更新状态以与 spec 相匹配。如果那些实例中有失败的（一种状态变更），Kubernetes 系统通过修正来响应 spec 和状态之间的不一致 —— 这种情况，启动一个新的实例来替换。 关于对象 spec、status 和 metadata 更多信息，查看 Kubernetes API Conventions。 描述 Kubernetes 对象 当创建 Kubernetes 对象时，必须提供对象的 spec，用来描述该对象的期望状态，以及关于对象的一些基本信息（例如，名称）。当使用 Kubernetes API 创建对象时（或者直接创建，或者基于kubectl），API 请求必须在请求体中包含 JSON 格式的信息。更常用的是，需要在 .yaml 文件中为 kubectl 提供这些信息。 kubectl 在执行 API 请求时，将这些信息转换成 JSON 格式。 这里有一个 .yaml 示例文件，展示了 Kubernetes Deployment 的必需字段和对象 spec： apiVersion: apps/v1beta1 kind: Deployment metadata: name: nginx-deployment spec: replicas: 3 template: metadata: labels: app: nginx spec: containers: - name: nginx image: nginx:1.7.9 ports: - containerPort: 80 一种创建 Deployment 的方式，类似上面使用 .yaml 文件，是使用 kubectl 命令行接口（CLI）中的 kubectl create 命令，传递 .yaml 作为参数。下面是一个示例： $ kubectl create -f docs/user-guide/nginx-deployment.yaml --record 输出类似如下这样： deployment \"nginx-deployment\" created 必需字段 在想要创建的 Kubernetes 对象对应的 .yaml 文件中，需要配置如下的字段： apiVersion - 创建该对象所使用的 Kubernetes API 的版本 kind - 想要创建的对象的类型 metadata - 帮助识别对象唯一性的数据，包括一个 name 字符串、UID 和可选的 namespace 也需要提供对象的 spec 字段。对象 spec 的精确格式对每个 Kubernetes 对象来说是不同的，包含了特定于该对象的嵌套字段。Kubernetes API 参考能够帮助我们找到任何我们想创建的对象的 spec 格式。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-22 17:00:47 "},"concepts/pod-overview.html":{"url":"concepts/pod-overview.html","title":"Pod","keywords":"","body":"Pod概览 理解Pod Pod是kubernetes中你可以创建和部署的最小也是最简单位。一个Pod代表着集群中运行的一个进程。 Pod中封装着应用的容器（有的情况下是好几个容器），存储、独立的网络IP，管理容器如何运行的策略选项。Pod代表着部署的一个单位：kubernetes中应用的一个实例，可能由一个或者多个容器组合在一起共享资源。 Docker是kubernetes中最常用的容器运行时，但是Pod也支持其他容器运行时。 Pods are employed a number of ways in a Kubernetes cluster, including: 在Kubrenetes集群中Pod有如下两种使用方式： 一个Pod中运行一个容器。“每个Pod中一个容器”的模式是最常见的用法；在这种使用方式中，你可以把Pod想象成是单个容器的封装，kuberentes管理的是Pod而不是直接管理容器。 在一个Pod中同时运行多个容器。一个Pod中也可以同时封装几个需要紧密耦合互相协作的容器，它们之间共享资源。这些在同一个Pod中的容器可以互相协作成为一个service单位——一个容器共享文件，另一个“sidecar”容器来更新这些文件。Pod将这些容器的存储资源作为一个实体来管理。 Kubernetes Blog 有关于Pod用例的详细信息，查看： The Distributed System Toolkit: Patterns for Composite Containers Container Design Patterns 每个Pod都是应用的一个实例。如果你想平行扩展应用的话（运行多个实例），你应该运行多个Pod，每个Pod都是一个应用实例。在Kubernetes中，这通常被称为replication。 Pod中如何管理多个容器 Pod中可以同时运行多个进程（作为容器运行）协同工作。同一个Pod中的容器会自动的分配到同一个 node 上。同一个Pod中的容器共享资源、网络环境和依赖，它们总是被同时调度。 注意在一个Pod中同时运行多个容器是一种比较高级的用法。只有当你的容器需要紧密配合协作的时候才考虑用这种模式。例如，你有一个容器作为web服务器运行，需要用到共享的volume，有另一个“sidecar”容器来从远端获取资源更新这些文件，如下图所示： 图片 - pod diagram Pod中可以共享两种资源：网络和存储。 网络 每个Pod都会被分配一个唯一的IP地址。Pod中的所有容器共享网络空间，包括IP地址和端口。Pod内部的容器可以使用localhost互相通信。Pod中的容器与外界通信时，必须分配共享网络资源（例如使用宿主机的端口映射）。 存储 可以Pod指定多个共享的Volume。Pod中的所有容器都可以访问共享的volume。Volume也可以用来持久化Pod中的存储资源，以防容器重启后文件丢失。 使用Pod 你很少会直接在kubernetes中创建单个Pod。因为Pod的生命周期是短暂的，用后即焚的实体。当Pod被创建后（不论是由你直接创建还是被其他Controller），都会被Kuberentes调度到集群的Node上。直到Pod的进程终止、被删掉、因为缺少资源而被驱逐、或者Node故障之前这个Pod都会一直保持在那个Node上。 注意：重启Pod中的容器跟重启Pod不是一回事。Pod只提供容器的运行环境并保持容器的运行状态，重启容器不会造成Pod重启。 Pod不会自愈。如果Pod运行的Node故障，或者是调度器本身故障，这个Pod就会被删除。同样的，如果Pod所在Node缺少资源或者Pod处于维护状态，Pod也会被驱逐。Kubernetes使用更高级的称为Controller的抽象层，来管理Pod实例。虽然可以直接使用Pod，但是在Kubernetes中通常是使用Controller来管理Pod的。 Pod和Controller Controller可以创建和管理多个Pod，提供副本管理、滚动升级和集群级别的自愈能力。例如，如果一个Node故障，Controller就能自动将该节点上的Pod调度到其他健康的Node上。 包含一个或者多个Pod的Controller示例： Deployment StatefulSet DaemonSet 通常，Controller会用你提供的Pod Template来创建相应的Pod。 Pod Templates Pod模版是包含了其他object的Pod定义，例如Replication Controllers，Jobs和 DaemonSets。Controller根据Pod模板来创建实际的Pod。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-22 17:00:47 "},"concepts/pod.html":{"url":"concepts/pod.html","title":"Pod解析","keywords":"","body":"Pod解析 Pod是kubernetes中可以创建的最小部署单元。 V1 core版本的Pod的配置模板见Pod template。 什么是Pod？ Pod就像是豌豆荚一样，它由一个或者多个容器组成（例如Docker容器），它们共享容器存储、网络和容器运行配置项。Pod中的容器总是被同时调度，有共同的运行环境。你可以把单个Pod想象成是运行独立应用的“逻辑主机”——其中运行着一个或者多个紧密耦合的应用容器——在有容器之前，这些应用都是运行在几个相同的物理机或者虚拟机上。 尽管kubernetes支持多种容器运行时，但是Docker依然是最常用的运行时环境，我们可以使用Docker的术语和规则来定义Pod。 Pod中共享的环境包括Linux的namespace，cgroup和其他可能的隔绝环境，这一点跟Docker容器一致。在Pod的环境中，每个容器中可能还有更小的子隔离环境。 Pod中的容器共享IP地址和端口号，它们之间可以通过localhost互相发现。它们之间可以通过进程间通信，例如SystemV信号或者POSIX共享内存。不同Pod之间的容器具有不同的IP地址，不能直接通过IPC通信。 Pod中的容器也有访问共享volume的权限，这些volume会被定义成pod的一部分并挂载到应用容器的文件系统中。 根据Docker的结构，Pod中的容器共享namespace和volume，不支持共享PID的namespace。 就像每个应用容器，pod被认为是临时（非持久的）实体。在Pod的生命周期中讨论过，pod被创建后，被分配一个唯一的ID（UID），调度到节点上，并一致维持期望的状态直到被终结（根据重启策略）或者被删除。如果node死掉了，分配到了这个node上的pod，在经过一个超时时间后会被重新调度到其他node节点上。一个给定的pod（如UID定义的）不会被“重新调度”到新的节点上，而是被一个同样的pod取代，如果期望的话甚至可以是相同的名字，但是会有一个新的UID（查看replication controller获取详情）。（未来，一个更高级别的API将支持pod迁移）。 Volume跟pod有相同的生命周期（当其UID存在的时候）。当Pod因为某种原因被删除或者被新创建的相同的Pod取代，它相关的东西（例如volume）也会被销毁和再创建一个新的volume。 图片 - Pod示意图 A multi-container pod that contains a file puller and a web server that uses a persistent volume for shared storage between the containers. Pod的动机 管理 Pod是一个服务的多个进程的聚合单位，pod提供这种模型能够简化应用部署管理，通过提供一个更高级别的抽象的方式。Pod作为一个独立的部署单位，支持横向扩展和复制。共生（协同调度），命运共同体（例如被终结），协同复制，资源共享，依赖管理，Pod都会自动的为容器处理这些问题。 资源共享和通信 Pod中的应用可以共享网络空间（IP地址和端口），因此可以通过localhost互相发现。因此，pod中的应用必须协调端口占用。每个pod都有一个唯一的IP地址，跟物理机和其他pod都处于一个扁平的网络空间中，它们之间可以直接连通。 Pod中应用容器的hostname被设置成Pod的名字。 Pod中的应用容器可以共享volume。Volume能够保证pod重启时使用的数据不丢失。 Pod的使用 Pod也可以用于垂直应用栈（例如LAMP），这样使用的主要动机是为了支持共同调度和协调管理应用程序，例如： content management systems, file and data loaders, local cache managers, etc. log and checkpoint backup, compression, rotation, snapshotting, etc. data change watchers, log tailers, logging and monitoring adapters, event publishers, etc. proxies, bridges, and adapters controllers, managers, configurators, and updaters 通常单个pod中不会同时运行一个应用的多个实例。 详细说明请看： The Distributed System ToolKit: Patterns for Composite Containers. 其他替代选择 为什么不直接在一个容器中运行多个应用程序呢？ 透明。让Pod中的容器对基础设施可见，以便基础设施能够为这些容器提供服务，例如进程管理和资源监控。这可以为用户带来极大的便利。 解耦软件依赖。每个容器都可以进行版本管理，独立的编译和发布。未来kubernetes甚至可能支持单个容器的在线升级。 使用方便。用户不必运行自己的进程管理器，还要担心错误信号传播等。 效率。因为由基础架构提供更多的职责，所以容器可以变得更加轻量级。 为什么不支持容器的亲和性的协同调度？ 这种方法可以提供容器的协同定位，能够根据容器的亲和性进行调度，但是无法实现使用pod带来的大部分好处，例如资源共享，IPC，保持状态一致性和简化管理等。 Pod的持久性（或者说缺乏持久性） Pod在设计支持就不是作为持久化实体的。在调度失败、节点故障、缺少资源或者节点维护的状态下都会死掉会被驱逐。 通常，用户不需要手动直接创建Pod，而是应该使用controller（例如Deployments），即使是在创建单个Pod的情况下。Controller可以提供集群级别的自愈功能、复制和升级管理。 The use of collective APIs as the primary user-facing primitive is relatively common among cluster scheduling systems, including Borg, Marathon, Aurora, and Tupperware. Pod is exposed as a primitive in order to facilitate: scheduler and controller pluggability support for pod-level operations without the need to \"proxy\" them via controller APIs decoupling of pod lifetime from controller lifetime, such as for bootstrapping decoupling of controllers and services — the endpoint controller just watches pods clean composition of Kubelet-level functionality with cluster-level functionality — Kubelet is effectively the \"pod controller\" high-availability applications, which will expect pods to be replaced in advance of their termination and certainly in advance of deletion, such as in the case of planned evictions, image prefetching, or live pod migration #3949 StatefulSet controller（目前还是beta状态）支持有状态的Pod。在1.4版本中被称为PetSet。在kubernetes之前的版本中创建有状态pod的最佳方式是创建一个replica为1的replication controller。 Pod的终止 因为Pod作为在集群的节点上运行的进程，所以在不再需要的时候能够优雅的终止掉是十分必要的（比起使用发送KILL信号这种暴力的方式）。用户需要能够放松删除请求，并且知道它们何时会被终止，是否被正确的删除。用户想终止程序时发送删除pod的请求，在pod可以被强制删除前会有一个优雅删除的时间，会发送一个TERM请求到每个容器的主进程。一旦超时，将向主进程发送KILL信号并从API server中删除。如果kubelet或者container manager在等待进程终止的过程中重启，在重启后仍然会重试完整的优雅删除阶段。 示例流程如下： 用户发送删除pod的命令，默认优雅删除时期是30秒； 在Pod超过该优雅删除期限后API server就会更新Pod的状态为“dead”； 在客户端命令行上显示的Pod状态为“terminating”； 跟第三步同时，当kubelet发现pod被标记为“terminating”状态时，开始停止pod进程： 如果在pod中定义了preStop hook，在停止pod前会被调用。如果在优雅删除期限过期后，preStop hook依然在运行，第二步会再增加2秒的优雅时间； 向Pod中的进程发送TERM信号； 跟第三步同时，该Pod将从该service的端点列表中删除，不再是replication controller的一部分。关闭的慢的pod将继续处理load balancer转发的流量； 过了优雅周期后，将向Pod中依然运行的进程发送SIGKILL信号而杀掉进程。 Kublete会在API server中完成Pod的的删除，通过将优雅周期设置为0（立即删除）。Pod在API中消失，并且在客户端也不可见。 删除优雅周期默认是30秒。 kubectl delete命令支持 —grace-period= 选项，允许用户设置自己的优雅周期时间。如果设置为0将强制删除pod。在kubectl>=1.5版本的命令中，你必须同时使用 --force 和 --grace-period=0 来强制删除pod。 强制删除Pod Pod的强制删除是通过在集群和etcd中将其定义为删除状态。当执行强制删除命令时，API server不会等待该pod所运行在节点上的kubelet确认，就会立即将该pod从API server中移除，这时就可以创建跟原pod同名的pod了。这时，在节点上的pod会被立即设置为terminating状态，不过在被强制删除之前依然有一小段优雅删除周期。 强制删除对于某些pod具有潜在危险性，请谨慎使用。使用StatefulSet pod的情况下，请参考删除StatefulSet中的pod文章。 Pod中容器的特权模式 从kubernetes1.1版本开始，pod中的容器就可以开启previleged模式，在容器定义文件的 SecurityContext 下使用 privileged flag。 这在使用Linux的网络操作和访问设备的能力时是很用的。同时开启的特权模式的Pod中的容器也可以访问到容器外的进程和应用。在不需要修改和重新编译kubelet的情况下就可以使用pod来开发节点的网络和存储插件。 如果master节点运行的是kuberentes1.1或更高版本，而node节点的版本低于1.1版本，则API server将也可以接受新的特权模式的pod，但是无法启动，pod将处于pending状态。 执行 kubectl describe pod FooPodName，可以看到为什么pod处于pending状态。输出的event列表中将显示： Error validating pod \"FooPodName\".\"FooPodNamespace\" from api, ignoring: spec.containers[0].securityContext.privileged: forbidden '(0xc2089d3248)true' 如果master节点的版本低于1.1，无法创建特权模式的pod。如果你仍然试图去创建的话，你得到如下错误： The Pod \"FooPodName\" is invalid. spec.containers[0].securityContext.privileged: forbidden '(0xc20b222db0)true' API Object Pod是kubernetes REST API中的顶级资源类型。 在kuberentes1.6的V1 core API版本中的Pod的数据结构如下图所示： 图片 - Pod Cheatsheet Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-22 17:00:47 "},"concepts/init-containers.html":{"url":"concepts/init-containers.html","title":"Init容器","keywords":"","body":"Init 容器 该特性在 1.6 版本已经推出 beta 版本。Init 容器可以在 PodSpec 中同应用程序的 containers 数组一起来指定。 beta 注解的值将仍需保留，并覆盖 PodSpec 字段值。 本文讲解 Init 容器的基本概念，它是一种专用的容器，在应用程序容器启动之前运行，并包括一些应用镜像中不存在的实用工具和安装脚本。 理解 Init 容器 Pod 能够具有多个容器，应用运行在容器里面，但是它也可能有一个或多个先于应用容器启动的 Init 容器。 Init 容器与普通的容器非常像，除了如下两点： Init 容器总是运行到成功完成为止。 每个 Init 容器都必须在下一个 Init 容器启动之前成功完成。 如果 Pod 的 Init 容器失败，Kubernetes 会不断地重启该 Pod，直到 Init 容器成功为止。然而，如果 Pod 对应的 restartPolicy 为 Never，它不会重新启动。 指定容器为 Init 容器，在 PodSpec 中添加 initContainers 字段，以 v1.Container 类型对象的 JSON 数组的形式，还有 app 的 containers 数组。 Init 容器的状态在 status.initContainerStatuses 字段中以容器状态数组的格式返回（类似 status.containerStatuses 字段）。 与普通容器的不同之处 Init 容器支持应用容器的全部字段和特性，包括资源限制、数据卷和安全设置。 然而，Init 容器对资源请求和限制的处理稍有不同，在下面 资源 处有说明。 而且 Init 容器不支持 Readiness Probe，因为它们必须在 Pod 就绪之前运行完成。 如果为一个 Pod 指定了多个 Init 容器，那些容器会按顺序一次运行一个。 每个 Init 容器必须运行成功，下一个才能够运行。 当所有的 Init 容器运行完成时，Kubernetes 初始化 Pod 并像平常一样运行应用容器。 Init 容器能做什么？ 因为 Init 容器具有与应用程序容器分离的单独镜像，所以它们的启动相关代码具有如下优势： 它们可以包含并运行实用工具，但是出于安全考虑，是不建议在应用程序容器镜像中包含这些实用工具的。 它们可以包含使用工具和定制化代码来安装，但是不能出现在应用程序镜像中。例如，创建镜像没必要 FROM 另一个镜像，只需要在安装过程中使用类似 sed、 awk、 python 或 dig 这样的工具。 应用程序镜像可以分离出创建和部署的角色，而没有必要联合它们构建一个单独的镜像。 Init 容器使用 Linux Namespace，所以相对应用程序容器来说具有不同的文件系统视图。因此，它们能够具有访问 Secret 的权限，而应用程序容器则不能。 它们必须在应用程序容器启动之前运行完成，而应用程序容器是并行运行的，所以 Init 容器能够提供了一种简单的阻塞或延迟应用容器的启动的方法，直到满足了一组先决条件。 示例 下面是一些如何使用 Init 容器的想法： 等待一个 Service 创建完成，通过类似如下 shell 命令： for i in {1..100}; do sleep 1; if dig myservice; then exit 0; fi; exit 1 将 Pod 注册到远程服务器，通过在命令中调用 API，类似如下： curl -X POST http://$MANAGEMENT_SERVICE_HOST:$MANAGEMENT_SERVICE_PORT/register -d 'instance=$()&ip=$()' 在启动应用容器之前等一段时间，使用类似 sleep 60 的命令。 克隆 Git 仓库到数据卷。 将配置值放到配置文件中，运行模板工具为主应用容器动态地生成配置文件。例如，在配置文件中存放 POD_IP 值，并使用 Jinja 生成主应用配置文件。 更多详细用法示例，可以在 StatefulSet 文档 和 生产环境 Pod 指南 中找到。 使用 Init 容器 下面是 Kubernetes 1.5 版本 yaml 文件，展示了一个具有 2 个 Init 容器的简单 Pod。 第一个等待 myservice 启动，第二个等待 mydb 启动。 一旦这两个 Service 都启动完成，Pod 将开始启动。 apiVersion: v1 kind: Pod metadata: name: myapp-pod labels: app: myapp annotations: pod.beta.kubernetes.io/init-containers: '[ { \"name\": \"init-myservice\", \"image\": \"busybox\", \"command\": [\"sh\", \"-c\", \"until nslookup myservice; do echo waiting for myservice; sleep 2; done;\"] }, { \"name\": \"init-mydb\", \"image\": \"busybox\", \"command\": [\"sh\", \"-c\", \"until nslookup mydb; do echo waiting for mydb; sleep 2; done;\"] } ]' spec: containers: - name: myapp-container image: busybox command: ['sh', '-c', 'echo The app is running! && sleep 3600'] 这是 Kubernetes 1.6 版本的新语法，尽管老的 annotation 语法仍然可以使用。我们已经把 Init 容器的声明移到 spec 中： apiVersion: v1 kind: Pod metadata: name: myapp-pod labels: app: myapp spec: containers: - name: myapp-container image: busybox command: ['sh', '-c', 'echo The app is running! && sleep 3600'] initContainers: - name: init-myservice image: busybox command: ['sh', '-c', 'until nslookup myservice; do echo waiting for myservice; sleep 2; done;'] - name: init-mydb image: busybox command: ['sh', '-c', 'until nslookup mydb; do echo waiting for mydb; sleep 2; done;'] 1.5 版本的语法在 1.6 版本仍然可以使用，但是我们推荐使用 1.6 版本的新语法。 在 Kubernetes 1.6 版本中，Init 容器在 API 中新建了一个字段。 虽然期望使用 beta 版本的 annotation，但在未来发行版将会被废弃掉。 下面的 yaml 文件展示了 mydb 和 myservice 两个 Service： kind: Service apiVersion: v1 metadata: name: myservice spec: ports: - protocol: TCP port: 80 targetPort: 9376 --- kind: Service apiVersion: v1 metadata: name: mydb spec: ports: - protocol: TCP port: 80 targetPort: 9377 这个 Pod 可以使用下面的命令进行启动和调试： $ kubectl create -f myapp.yaml pod \"myapp-pod\" created $ kubectl get -f myapp.yaml NAME READY STATUS RESTARTS AGE myapp-pod 0/1 Init:0/2 0 6m $ kubectl describe -f myapp.yaml Name: myapp-pod Namespace: default [...] Labels: app=myapp Status: Pending [...] Init Containers: init-myservice: [...] State: Running [...] init-mydb: [...] State: Waiting Reason: PodInitializing Ready: False [...] Containers: myapp-container: [...] State: Waiting Reason: PodInitializing Ready: False [...] Events: FirstSeen LastSeen Count From SubObjectPath Type Reason Message --------- -------- ----- ---- ------------- -------- ------ ------- 16s 16s 1 {default-scheduler } Normal Scheduled Successfully assigned myapp-pod to 172.17.4.201 16s 16s 1 {kubelet 172.17.4.201} spec.initContainers{init-myservice} Normal Pulling pulling image \"busybox\" 13s 13s 1 {kubelet 172.17.4.201} spec.initContainers{init-myservice} Normal Pulled Successfully pulled image \"busybox\" 13s 13s 1 {kubelet 172.17.4.201} spec.initContainers{init-myservice} Normal Created Created container with docker id 5ced34a04634; Security:[seccomp=unconfined] 13s 13s 1 {kubelet 172.17.4.201} spec.initContainers{init-myservice} Normal Started Started container with docker id 5ced34a04634 $ kubectl logs myapp-pod -c init-myservice # Inspect the first init container $ kubectl logs myapp-pod -c init-mydb # Inspect the second init container 一旦我们启动了 mydb 和 myservice 这两个 Service，我们能够看到 Init 容器完成，并且 myapp-pod 被创建： $ kubectl create -f services.yaml service \"myservice\" created service \"mydb\" created $ kubectl get -f myapp.yaml NAME READY STATUS RESTARTS AGE myapp-pod 1/1 Running 0 9m 这个例子非常简单，但是应该能够为我们创建自己的 Init 容器提供一些启发。 具体行为 在 Pod 启动过程中，Init 容器会按顺序在网络和数据卷初始化之后启动。 每个容器必须在下一个容器启动之前成功退出。 如果由于运行时或失败退出，导致容器启动失败，它会根据 Pod 的 restartPolicy 指定的策略进行重试。 然而，如果 Pod 的 restartPolicy 设置为 Always，Init 容器失败时会使用 RestartPolicy 策略。 在所有的 Init 容器没有成功之前，Pod 将不会变成 Ready 状态。 Init 容器的端口将不会在 Service 中进行聚集。 正在初始化中的 Pod 处于 Pending 状态，但应该会将条件 Initializing 设置为 true。 如果 Pod 重启，所有 Init 容器必须重新执行。 对 Init 容器 spec 的修改，被限制在容器 image 字段中。 更改 Init 容器的 image 字段，等价于重启该 Pod。 因为 Init 容器可能会被重启、重试或者重新执行，所以 Init 容器的代码应该是幂等的。 特别地，被写到 EmptyDirs 中文件的代码，应该对输出文件可能已经存在做好准备。 Init 容器具有应用容器的所有字段。 除了 readinessProbe，因为 Init 容器无法定义不同于完成（completion）的就绪（readiness）的之外的其他状态。 这会在验证过程中强制执行。 在 Pod 上使用 activeDeadlineSeconds，在容器上使用 livenessProbe，这样能够避免 Init 容器一直失败。 这就为 Init 容器活跃设置了一个期限。 在 Pod 中的每个 app 和 Init 容器的名称必须唯一；与任何其它容器共享同一个名称，会在验证时抛出错误。 资源 为 Init 容器指定顺序和执行逻辑，下面对资源使用的规则将被应用： 在所有 Init 容器上定义的，任何特殊资源请求或限制的最大值，是 有效初始请求/限制 Pod 对资源的有效请求/限制要高于： 所有应用容器对某个资源的请求/限制之和 对某个资源的有效初始请求/限制 基于有效请求/限制完成调度，这意味着 Init 容器能够为初始化预留资源，这些资源在 Pod 生命周期过程中并没有被使用。 Pod 的 有效 QoS 层，是 Init 容器和应用容器相同的 QoS 层。 基于有效 Pod 请求和限制来应用配额和限制。 Pod 级别的 cgroups 是基于有效 Pod 请求和限制，和调度器相同。 Pod 重启的原因 Pod 能够重启，会导致 Init 容器重新执行，主要有如下几个原因： 用户更新 PodSpec 导致 Init 容器镜像发生改变。应用容器镜像的变更只会重启应用容器。 Pod 基础设施容器被重启。这不多见，但某些具有 root 权限可访问 Node 的人可能会这样做。 当 restartPolicy 设置为 Always，Pod 中所有容器会终止，强制重启，由于垃圾收集导致 Init 容器完整的记录丢失。 支持与兼容性 Apiserver 版本为 1.6 或更高版本的集群，通过使用 spec.initContainers 字段来支持 Init 容器。 之前的版本可以使用 alpha 和 beta 注解支持 Init 容器。 spec.initContainers 字段也被加入到 alpha 和 beta 注解中，所以 Kubernetes 1.3.0 版本或更高版本可以执行 Init 容器，并且 1.6 版本的 apiserver 能够安全地回退到 1.5.x 版本，而不会使存在的已创建 Pod 失去 Init 容器的功能。 原文地址：https://k8smeetup.github.io/docs/concepts/workloads/pods/init-containers/ 译者：shirdrn Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-22 17:00:47 "},"concepts/pod-security-policy.html":{"url":"concepts/pod-security-policy.html","title":"Pod安全策略","keywords":"","body":"Pod 安全策略 PodSecurityPolicy 类型的对象能够控制，是否可以向 Pod 发送请求，该 Pod 能够影响被应用到 Pod 和容器的 SecurityContext。 查看 Pod 安全策略建议 获取更多信息。 什么是 Pod 安全策略？ Pod 安全策略 是集群级别的资源，它能够控制 Pod 运行的行为，以及它具有访问什么的能力。 PodSecurityPolicy对象定义了一组条件，指示 Pod 必须按系统所能接受的顺序运行。 它们允许管理员控制如下方面： 控制面 字段名称 已授权容器的运行 privileged 为容器添加默认的一组能力 defaultAddCapabilities 为容器去掉某些能力 requiredDropCapabilities 容器能够请求添加某些能力 allowedCapabilities 控制卷类型的使用 volumes 主机网络的使用 hostNetwork 主机端口的使用 hostPorts 主机 PID namespace 的使用 hostPID 主机 IPC namespace 的使用 hostIPC 主机路径的使用 allowedHostPaths 容器的 SELinux 上下文 seLinux 用户 ID runAsUser 配置允许的补充组 supplementalGroups 分配拥有 Pod 数据卷的 FSGroup fsGroup 必须使用一个只读的 root 文件系统 readOnlyRootFilesystem Pod 安全策略 由设置和策略组成，它们能够控制 Pod 访问的安全特征。这些设置分为如下三类： 基于布尔值控制：这种类型的字段默认为最严格限制的值。 基于被允许的值集合控制：这种类型的字段会与这组值进行对比，以确认值被允许。 基于策略控制：设置项通过一种策略提供的机制来生成该值，这种机制能够确保指定的值落在被允许的这组值中。 RunAsUser MustRunAs - 必须配置一个 range。使用该范围内的第一个值作为默认值。验证是否不在配置的该范围内。 MustRunAsNonRoot - 要求提交的 Pod 具有非零 runAsUser 值，或在镜像中定义了 USER 环境变量。不提供默认值。 RunAsAny - 没有提供默认值。允许指定任何 runAsUser 。 SELinux MustRunAs - 如果没有使用预分配的值，必须配置 seLinuxOptions。默认使用 seLinuxOptions。验证 seLinuxOptions。 RunAsAny - 没有提供默认值。允许任意指定的 seLinuxOptions ID。 SupplementalGroups MustRunAs - 至少需要指定一个范围。默认使用第一个范围的最小值。验证所有范围的值。 RunAsAny - 没有提供默认值。允许任意指定的 supplementalGroups ID。 FSGroup MustRunAs - 至少需要指定一个范围。默认使用第一个范围的最小值。验证在第一个范围内的第一个 ID。 RunAsAny - 没有提供默认值。允许任意指定的 fsGroup ID。 控制卷 通过设置 PSP 卷字段，能够控制具体卷类型的使用。当创建一个卷的时候，与该字段相关的已定义卷可以允许设置如下值： azureFile azureDisk flocker flexVolume hostPath emptyDir gcePersistentDisk awsElasticBlockStore gitRepo secret nfs iscsi glusterfs persistentVolumeClaim rbd cinder cephFS downwardAPI fc configMap vsphereVolume quobyte photonPersistentDisk projected portworxVolume scaleIO storageos * (allow all volumes) 对新的 PSP，推荐允许的卷的最小集合包括：configMap、downwardAPI、emptyDir、persistentVolumeClaim、secret 和 projected。 主机网络 HostPorts， 默认为 empty。HostPortRange 列表通过 min(包含) and max(包含) 来定义，指定了被允许的主机端口。 允许的主机路径 AllowedHostPaths 是一个被允许的主机路径前缀的白名单。空值表示所有的主机路径都可以使用。 许可 包含 PodSecurityPolicy 的 许可控制，允许控制集群资源的创建和修改，基于这些资源在集群范围内被许可的能力。 许可使用如下的方式为 Pod 创建最终的安全上下文： 检索所有可用的 PSP。 生成在请求中没有指定的安全上下文设置的字段值。 基于可用的策略，验证最终的设置。 如果某个策略能够匹配上，该 Pod 就被接受。如果请求与 PSP 不匹配，则 Pod 被拒绝。 Pod 必须基于 PSP 验证每个字段。 创建 Pod 安全策略 下面是一个 Pod 安全策略的例子，所有字段的设置都被允许： apiVersion: extensions/v1beta1 kind: PodSecurityPolicy metadata: name: permissive spec: seLinux: rule: RunAsAny supplementalGroups: rule: RunAsAny runAsUser: rule: RunAsAny fsGroup: rule: RunAsAny hostPorts: - min: 8000 max: 8080 volumes: - '*' 下载示例文件可以创建该策略，然后执行如下命令： $ kubectl create -f ./psp.yaml podsecuritypolicy \"permissive\" created 获取 Pod 安全策略列表 获取已存在策略列表，使用 kubectl get： $ kubectl get psp NAME PRIV CAPS SELINUX RUNASUSER FSGROUP SUPGROUP READONLYROOTFS VOLUMES permissive false [] RunAsAny RunAsAny RunAsAny RunAsAny false [*] privileged true [] RunAsAny RunAsAny RunAsAny RunAsAny false [*] restricted false [] RunAsAny MustRunAsNonRoot RunAsAny RunAsAny false [emptyDir secret downwardAPI configMap persistentVolumeClaim projected] 修改 Pod 安全策略 通过交互方式修改策略，使用 kubectl edit： $ kubectl edit psp permissive 该命令将打开一个默认文本编辑器，在这里能够修改策略。 删除 Pod 安全策略 一旦不再需要一个策略，很容易通过 kubectl 删除它： $ kubectl delete psp permissive podsecuritypolicy \"permissive\" deleted 启用 Pod 安全策略 为了能够在集群中使用 Pod 安全策略，必须确保如下： 启用 API 类型 extensions/v1beta1/podsecuritypolicy（仅对 1.6 之前的版本） 启用许可控制器 PodSecurityPolicy 定义自己的策略 使用 RBAC 在 Kubernetes 1.5 或更新版本，可以使用 PodSecurityPolicy 来控制，对基于用户角色和组的已授权容器的访问。访问不同的 PodSecurityPolicy 对象，可以基于认证来控制。基于 Deployment、ReplicaSet 等创建的 Pod，限制访问 PodSecurityPolicy 对象，Controller Manager 必须基于安全 API 端口运行，并且不能够具有超级用户权限。 PodSecurityPolicy 认证使用所有可用的策略，包括创建 Pod 的用户，Pod 上指定的服务账户（service acount）。当 Pod 基于 Deployment、ReplicaSet 创建时，它是创建 Pod 的 Controller Manager，所以如果基于非安全 API 端口运行，允许所有的 PodSecurityPolicy 对象，并且不能够有效地实现细分权限。用户访问给定的 PSP 策略有效，仅当是直接部署 Pod 的情况。更多详情，查看 PodSecurityPolicy RBAC 示例，当直接部署 Pod 时，应用 PodSecurityPolicy 控制基于角色和组的已授权容器的访问 。 原文地址：https://k8smeetup.github.io/docs/concepts/policy/pod-security-policy/ 译者：shirdrn Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-21 11:59:28 "},"concepts/pod-lifecycle.html":{"url":"concepts/pod-lifecycle.html","title":"Pod的生命周期","keywords":"","body":"Pod 的生命周期 Pod phase Pod 的 status 在信息保存在 PodStatus 中定义，其中有一个 phase 字段。 Pod 的相位（phase）是 Pod 在其生命周期中的简单宏观概述。该阶段并不是对容器或 Pod 的综合汇总，也不是为了做为综合状态机。 Pod 相位的数量和含义是严格指定的。除了本文档中列举的状态外，不应该再假定 Pod 有其他的 phase 值。 下面是 phase 可能的值： 挂起（Pending）：Pod 已被 Kubernetes 系统接受，但有一个或者多个容器镜像尚未创建。等待时间包括调度 Pod 的时间和通过网络下载镜像的时间，这可能需要花点时间。 运行中（Running）：该 Pod 已经绑定到了一个节点上，Pod 中所有的容器都已被创建。至少有一个容器正在运行，或者正处于启动或重启状态。 成功（Successed）：Pod 中的所有容器都被成功终止，并且不会再重启。 失败（Failed）：Pod 中的所有容器都已终止了，并且至少有一个容器是因为失败终止。也就是说，容器以非0状态退出或者被系统终止。 未知（Unkonwn）：因为某些原因无法取得 Pod 的状态，通常是因为与 Pod 所在主机通信失败。 Pod 状态 Pod 有一个 PodStatus 对象，其中包含一个 PodCondition 数组。 PodCondition 数组的每个元素都有一个 type 字段和一个 status 字段。type 字段是字符串，可能的值有 PodScheduled、Ready、Initialized 和 Unschedulable。status 字段是一个字符串，可能的值有 True、False 和 Unknown。 容器探针 探针 是由 kubelet 对容器执行的定期诊断。要执行诊断，kubelet 调用由容器实现的 Handler。有三种类型的处理程序： ExecAction：在容器内执行指定命令。如果命令退出时返回码为 0 则认为诊断成功。 TCPSocketAction：对指定端口上的容器的 IP 地址进行 TCP 检查。如果端口打开，则诊断被认为是成功的。 HTTPGetAction：对指定的端口和路径上的容器的 IP 地址执行 HTTP Get 请求。如果响应的状态码大于等于200 且小于 400，则诊断被认为是成功的。 每次探测都将获得以下三种结果之一： 成功：容器通过了诊断。 失败：容器未通过诊断。 未知：诊断失败，因此不会采取任何行动。 Kubelet 可以选择是否执行在容器上运行的两种探针执行和做出反应： livenessProbe：指示容器是否正在运行。如果存活探测失败，则 kubelet 会杀死容器，并且容器将受到其 重启策略 的影响。如果容器不提供存活探针，则默认状态为 Success。 readinessProbe：指示容器是否准备好服务请求。如果就绪探测失败，端点控制器将从与 Pod 匹配的所有 Service 的端点中删除该 Pod 的 IP 地址。初始延迟之前的就绪状态默认为 Failure。如果容器不提供就绪探针，则默认状态为 Success。 该什么时候使用存活（liveness）和就绪（readiness）探针? 如果容器中的进程能够在遇到问题或不健康的情况下自行崩溃，则不一定需要存活探针; kubelet 将根据 Pod 的restartPolicy 自动执行正确的操作。 如果您希望容器在探测失败时被杀死并重新启动，那么请指定一个存活探针，并指定restartPolicy 为 Always 或 OnFailure。 如果要仅在探测成功时才开始向 Pod 发送流量，请指定就绪探针。在这种情况下，就绪探针可能与存活探针相同，但是 spec 中的就绪探针的存在意味着 Pod 将在没有接收到任何流量的情况下启动，并且只有在探针探测成功后才开始接收流量。 如果您希望容器能够自行维护，您可以指定一个就绪探针，该探针检查与存活探针不同的端点。 请注意，如果您只想在 Pod 被删除时能够排除请求，则不一定需要使用就绪探针；在删除 Pod 时，Pod 会自动将自身置于未完成状态，无论就绪探针是否存在。当等待 Pod 中的容器停止时，Pod 仍处于未完成状态。 Pod 和容器状态 有关 Pod 容器状态的详细信息，请参阅 PodStatus 和 ContainerStatus。请注意，报告的 Pod 状态信息取决于当前的 ContainerState。 重启策略 PodSpec 中有一个 restartPolicy 字段，可能的值为 Always、OnFailure 和 Never。默认为 Always。 restartPolicy 适用于 Pod 中的所有容器。restartPolicy 仅指通过同一节点上的 kubelet 重新启动容器。失败的容器由 kubelet 以五分钟为上限的指数退避延迟（10秒，20秒，40秒...）重新启动，并在成功执行十分钟后重置。如 Pod 文档 中所述，一旦绑定到一个节点，Pod 将永远不会重新绑定到另一个节点。 Pod 的生命 一般来说，Pod 不会消失，直到人为销毁他们。这可能是一个人或控制器。这个规则的唯一例外是成功或失败的 phase 超过一段时间（由 master 确定）的Pod将过期并被自动销毁。 有三种可用的控制器： 使用 Job 运行预期会终止的 Pod，例如批量计算。Job 仅适用于重启策略为 OnFailure 或 Never 的 Pod。 对预期不会终止的 Pod 使用 ReplicationController、ReplicaSet 和 Deployment ，例如 Web 服务器。 ReplicationController 仅适用于具有 restartPolicy 为 Always 的 Pod。 提供特定于机器的系统服务，使用 DaemonSet 为每台机器运行一个 Pod 。 所有这三种类型的控制器都包含一个 PodTemplate。建议创建适当的控制器，让它们来创建 Pod，而不是直接自己创建 Pod。这是因为单独的 Pod 在机器故障的情况下没有办法自动复原，而控制器却可以。 如果节点死亡或与集群的其余部分断开连接，则 Kubernetes 将应用一个策略将丢失节点上的所有 Pod 的 phase 设置为 Failed。 示例 高级 liveness 探针示例 存活探针由 kubelet 来执行，因此所有的请求都在 kubelet 的网络命名空间中进行。 apiVersion: v1 kind: Pod metadata: labels: test: liveness name: liveness-http spec: containers: - args: - /server image: gcr.io/google_containers/liveness livenessProbe: httpGet: # when \"host\" is not defined, \"PodIP\" will be used # host: my-host # when \"scheme\" is not defined, \"HTTP\" scheme will be used. Only \"HTTP\" and \"HTTPS\" are allowed # scheme: HTTPS path: /healthz port: 8080 httpHeaders: - name: X-Custom-Header value: Awesome initialDelaySeconds: 15 timeoutSeconds: 1 name: liveness 状态示例 Pod 中只有一个容器并且正在运行。容器成功退出。 记录完成事件。 如果 restartPolicy 为： Always：重启容器；Pod phase 仍为 Running。 OnFailure：Pod phase 变成 Succeeded。 Never：Pod phase 变成 Succeeded。 Pod 中只有一个容器并且正在运行。容器退出失败。 记录失败事件。 如果 restartPolicy 为： Always：重启容器；Pod phase 仍为 Running。 OnFailure：重启容器；Pod phase 仍为 Running。 Never：Pod phase 变成 Failed。 Pod 中有两个容器并且正在运行。有一个容器退出失败。 记录失败事件。 如果 restartPolicy 为： Always：重启容器；Pod phase 仍为 Running。 OnFailure：重启容器；Pod phase 仍为 Running。 Never：不重启容器；Pod phase 仍为 Running。 如果有一个容器没有处于运行状态，并且两个容器退出： 记录失败事件。 如果 restartPolicy 为： Always：重启容器；Pod phase 仍为 Running。 OnFailure：重启容器；Pod phase 仍为 Running。 Never：Pod phase 变成 Failed。 Pod 中只有一个容器并处于运行状态。容器运行时内存超出限制： 容器以失败状态终止。 记录 OOM 事件。 如果 restartPolicy 为： Always：重启容器；Pod phase 仍为 Running。 OnFailure：重启容器；Pod phase 仍为 Running。 Never: 记录失败事件；Pod phase 仍为 Failed。 Pod 正在运行，磁盘故障： 杀掉所有容器。 记录适当事件。 Pod phase 变成 Failed。 如果使用控制器来运行，Pod 将在别处重建。 Pod 正在运行，其节点被分段。 节点控制器等待直到超时。 节点控制器将 Pod phase 设置为 Failed。 如果是用控制器来运行，Pod 将在别处重建。 原文地址：https://kubernetes.io/docs/concepts/workloads/pods/pod-lifecycle/ 翻译：rootsongjc Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-09-19 15:45:02 "},"concepts/pod-hook.html":{"url":"concepts/pod-hook.html","title":"Pod hook","keywords":"","body":"Pod hook Pod hook（钩子）是由Kubernetes管理的kubelet发起的，当容器中的进程启动前或者容器中的进程终止之前运行，这是包含在容器的生命周期之中。可以同时为Pod中的所有容器都配置hook。 Hook的类型包括两种： exec：执行一段命令 HTTP：发送HTTP请求。 参考下面的配置： apiVersion: v1 kind: Pod metadata: name: lifecycle-demo spec: containers: - name: lifecycle-demo-container image: nginx lifecycle: postStart: exec: command: [\"/bin/sh\", \"-c\", \"echo Hello from the postStart handler > /usr/share/message\"] preStop: exec: command: [\"/usr/sbin/nginx\",\"-s\",\"quit\"] 在容器创建之后，容器的Entrypoint执行之前，这时候Pod已经被调度到某台node上，被某个kubelet管理了，这时候kubelet会调用postStart操作，该操作跟容器的启动命令是在异步执行的，也就是说在postStart操作执行完成之前，kubelet会锁住容器，不让应用程序的进程启动，只有在 postStart操作完成之后容器的状态才会被设置成为RUNNING。 如果postStart或者preStop hook失败，将会终止容器。 调试hook Hook调用的日志没有暴露个给Pod的event，所以只能通过describe命令来获取，如果有错误将可以看到FailedPostStartHook或FailedPreStopHook这样的event。 参考 Attach Handlers to Container Lifecycle Events Container Lifecycle Hooks Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-23 16:33:50 "},"concepts/pod-preset.html":{"url":"concepts/pod-preset.html","title":"Pod Preset","keywords":"","body":"Pod Preset 注意：PodPreset 资源对象只有 kubernetes 1.8 以上版本才支持。 Preset 就是预设，有时候想要让一批容器在启动的时候就注入一些信息，比如 secret、volume、volume mount 和环境变量，而又不想一个一个的改这些 Pod 的 tmeplate，这时候就可以用到 PodPreset 这个资源对象了。 本页是关于 PodPreset 的概述，该对象用来在 Pod 创建的时候向 Pod 中注入某些特定信息。该信息可以包括 secret、volume、volume mount 和环境变量。 理解 Pod Preset Pod Preset 是用来在 Pod 被创建的时候向其中注入额外的运行时需求的 API 资源。 您可以使用 label selector 来指定为哪些 Pod 应用 Pod Preset。 使用 Pod Preset 使得 pod 模板的作者可以不必为每个 Pod 明确提供所有信息。这样一来，pod 模板的作者就不需要知道关于该服务的所有细节。 关于该背景的更多信息，请参阅 PodPreset 的设计方案。 如何工作 Kubernetes 提供了一个准入控制器（PodPreset），当其启用时，Pod Preset 会将应用创建请求传入到该控制器上。当有 Pod 创建请求发生时，系统将执行以下操作： 检索所有可用的 PodPresets。 检查 PodPreset 标签选择器上的标签，看看其是否能够匹配正在创建的 Pod 上的标签。 尝试将由 PodPreset 定义的各种资源合并到正在创建的 Pod 中。 出现错误时，在该 Pod 上引发记录合并错误的事件，PodPreset 不会注入任何资源到创建的 Pod 中。 注释刚生成的修改过的 Pod spec，以表明它已被 PodPreset 修改过。注释的格式为 podpreset.admission.kubernetes.io/podpreset-\": \"\"。 每个 Pod 可以匹配零个或多个 Pod Prestet；并且每个 PodPreset 可以应用于零个或多个 Pod。 PodPreset 应用于一个或多个 Pod 时，Kubernetes 会修改 Pod Spec。对于 Env、EnvFrom 和 VolumeMounts 的更改，Kubernetes 修改 Pod 中所有容器的容器 spec；对于 Volume 的更改，Kubernetes 修改 Pod Spec。 注意：Pod Preset 可以在适当的时候修改 Pod spec 中的 spec.containers 字段。Pod Preset 中的资源定义将不会应用于 initContainers 字段。 禁用特定 Pod 的 Pod Preset 在某些情况下，您可能不希望 Pod 被任何 Pod Preset 所改变。在这些情况下，您可以在 Pod 的 Pod Spec 中添加注释：podpreset.admission.kubernetes.io/exclude：\"true\"。 启用 Pod Preset 为了在群集中使用 Pod Preset，您必须确保以下内容： 您已启用 settings.k8s.io/v1alpha1/podpreset API 类型。例如，可以通过在 API server 的 --runtime-config 选项中包含 settings.k8s.io/v1alpha1=true 来完成此操作。 您已启用 PodPreset 准入控制器。 一种方法是将 PodPreset 包含在为 API server 指定的 --admission-control 选项值中。 您已经在要使用的命名空间中通过创建 PodPreset 对象来定义 PodPreset。 更多资料 使用 PodPreset 向 Pod 中注入数据 本文为 Kubernetes 官方中文文档，地址：https://kubernetes.io/cn/docs/concepts/workloads/pods/podpreset/ 翻译: rootsongjc Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-12 10:49:11 "},"concepts/node.html":{"url":"concepts/node.html","title":"Node","keywords":"","body":"Node Node是kubernetes集群的工作节点，可以是物理机也可以是虚拟机。 Node的状态 Node包括如下状态信息： Address HostName：可以被kubelet中的--hostname-override参数替代。 ExternalIP：可以被集群外部路由到的IP地址。 InternalIP：集群内部使用的IP，集群外部无法访问。 Condition OutOfDisk：磁盘空间不足时为True Ready：Node controller 40秒内没有收到node的状态报告为Unknown，健康为True，否则为False。 MemoryPressure：当node没有内存压力时为True，否则为False。 DiskPressure：当node没有磁盘压力时为True，否则为False。 Capacity CPU 内存 可运行的最大Pod个数 Info：节点的一些版本信息，如OS、kubernetes、docker等 Node管理 禁止pod调度到该节点上 kubectl cordon 驱逐该节点上的所有pod kubectl drain 该命令会删除该节点上的所有Pod（DaemonSet除外），在其他node上重新启动它们，通常该节点需要维护时使用该命令。直接使用该命令会自动调用kubectl cordon 命令。当该节点维护完成，启动了kubelet后，再使用kubectl uncordon 即可将该节点添加到kubernetes集群中。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-10-16 10:50:25 "},"concepts/namespace.html":{"url":"concepts/namespace.html","title":"Namespace","keywords":"","body":"Namespace 在一个Kubernetes集群中可以使用namespace创建多个“虚拟集群”，这些namespace之间可以完全隔离，也可以通过某种方式，让一个namespace中的service可以访问到其他的namespace中的服务，我们在CentOS中部署kubernetes1.6集群的时候就用到了好几个跨越namespace的服务，比如Traefik ingress和kube-systemnamespace下的service就可以为整个集群提供服务，这些都需要通过RBAC定义集群级别的角色来实现。 哪些情况下适合使用多个namesapce 因为namespace可以提供独立的命名空间，因此可以实现部分的环境隔离。当你的项目和人员众多的时候可以考虑根据项目属性，例如生产、测试、开发划分不同的namespace。 Namespace使用 获取集群中有哪些namespace kubectl get ns 集群中默认会有default和kube-system这两个namespace。 在执行kubectl命令时可以使用-n指定操作的namespace。 用户的普通应用默认是在default下，与集群管理相关的为整个集群提供服务的应用一般部署在kube-system的namespace下，例如我们在安装kubernetes集群时部署的kubedns、heapseter、EFK等都是在这个namespace下面。 另外，并不是所有的资源对象都会对应namespace，node和persistentVolume就不属于任何namespace。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-19 17:50:18 "},"concepts/service.html":{"url":"concepts/service.html","title":"Service","keywords":"","body":"Service Kubernetes Pod 是有生命周期的，它们可以被创建，也可以被销毁，然而一旦被销毁生命就永远结束。 通过 ReplicationController 能够动态地创建和销毁 Pod（例如，需要进行扩缩容，或者执行 滚动升级）。 每个 Pod 都会获取它自己的 IP 地址，即使这些 IP 地址不总是稳定可依赖的。 这会导致一个问题：在 Kubernetes 集群中，如果一组 Pod（称为 backend）为其它 Pod （称为 frontend）提供服务，那么那些 frontend 该如何发现，并连接到这组 Pod 中的哪些 backend 呢？ 关于 Service Kubernetes Service 定义了这样一种抽象：一个 Pod 的逻辑分组，一种可以访问它们的策略 —— 通常称为微服务。 这一组 Pod 能够被 Service 访问到，通常是通过 Label Selector（查看下面了解，为什么可能需要没有 selector 的 Service）实现的。 举个例子，考虑一个图片处理 backend，它运行了3个副本。这些副本是可互换的 —— frontend 不需要关心它们调用了哪个 backend 副本。 然而组成这一组 backend 程序的 Pod 实际上可能会发生变化，frontend 客户端不应该也没必要知道，而且也不需要跟踪这一组 backend 的状态。 Service 定义的抽象能够解耦这种关联。 对 Kubernetes 集群中的应用，Kubernetes 提供了简单的 Endpoints API，只要 Service 中的一组 Pod 发生变更，应用程序就会被更新。 对非 Kubernetes 集群中的应用，Kubernetes 提供了基于 VIP 的网桥的方式访问 Service，再由 Service 重定向到 backend Pod。 定义 Service 一个 Service 在 Kubernetes 中是一个 REST 对象，和 Pod 类似。 像所有的 REST 对象一样， Service 定义可以基于 POST 方式，请求 apiserver 创建新的实例。 例如，假定有一组 Pod，它们对外暴露了 9376 端口，同时还被打上 \"app=MyApp\" 标签。 kind: Service apiVersion: v1 metadata: name: my-service spec: selector: app: MyApp ports: - protocol: TCP port: 80 targetPort: 9376 上述配置将创建一个名称为 “my-service” 的 Service 对象，它会将请求代理到使用 TCP 端口 9376，并且具有标签 \"app=MyApp\" 的 Pod 上。 这个 Service 将被指派一个 IP 地址（通常称为 “Cluster IP”），它会被服务的代理使用（见下面）。 该 Service 的 selector 将会持续评估，处理结果将被 POST 到一个名称为 “my-service” 的 Endpoints 对象上。 需要注意的是， Service 能够将一个接收端口映射到任意的 targetPort。 默认情况下，targetPort 将被设置为与 port 字段相同的值。 可能更有趣的是，targetPort 可以是一个字符串，引用了 backend Pod 的一个端口的名称。 但是，实际指派给该端口名称的端口号，在每个 backend Pod 中可能并不相同。 对于部署和设计 Service ，这种方式会提供更大的灵活性。 例如，可以在 backend 软件下一个版本中，修改 Pod 暴露的端口，并不会中断客户端的调用。 Kubernetes Service 能够支持 TCP 和 UDP 协议，默认 TCP 协议。 没有 selector 的 Service Servcie 抽象了该如何访问 Kubernetes Pod，但也能够抽象其它类型的 backend，例如： 希望在生产环境中使用外部的数据库集群，但测试环境使用自己的数据库。 希望服务指向另一个 Namespace 中或其它集群中的服务。 正在将工作负载转移到 Kubernetes 集群，和运行在 Kubernetes 集群之外的 backend。 在任何这些场景中，都能够定义没有 selector 的 Service ： kind: Service apiVersion: v1 metadata: name: my-service spec: ports: - protocol: TCP port: 80 targetPort: 9376 由于这个 Service 没有 selector，就不会创建相关的 Endpoints 对象。可以手动将 Service 映射到指定的 Endpoints： kind: Endpoints apiVersion: v1 metadata: name: my-service subsets: - addresses: - ip: 1.2.3.4 ports: - port: 9376 注意：Endpoint IP 地址不能是 loopback（127.0.0.0/8）、 link-local（169.254.0.0/16）、或者 link-local 多播（224.0.0.0/24）。 访问没有 selector 的 Service，与有 selector 的 Service 的原理相同。请求将被路由到用户定义的 Endpoint（该示例中为 1.2.3.4:9376）。 ExternalName Service 是 Service 的特例，它没有 selector，也没有定义任何的端口和 Endpoint。 相反地，对于运行在集群外部的服务，它通过返回该外部服务的别名这种方式来提供服务。 kind: Service apiVersion: v1 metadata: name: my-service namespace: prod spec: type: ExternalName externalName: my.database.example.com 当查询主机 my-service.prod.svc.CLUSTER时，集群的 DNS 服务将返回一个值为 my.database.example.com 的 CNAME 记录。 访问这个服务的工作方式与其它的相同，唯一不同的是重定向发生在 DNS 层，而且不会进行代理或转发。 如果后续决定要将数据库迁移到 Kubernetes 集群中，可以启动对应的 Pod，增加合适的 Selector 或 Endpoint，修改 Service 的 type。 VIP 和 Service 代理 在 Kubernetes 集群中，每个 Node 运行一个 kube-proxy 进程。kube-proxy 负责为 Service 实现了一种 VIP（虚拟 IP）的形式，而不是 ExternalName 的形式。 在 Kubernetes v1.0 版本，代理完全在 userspace。在 Kubernetes v1.1 版本，新增了 iptables 代理，但并不是默认的运行模式。 从 Kubernetes v1.2 起，默认就是 iptables 代理。 在 Kubernetes v1.0 版本，Service 是 “4层”（TCP/UDP over IP）概念。 在 Kubernetes v1.1 版本，新增了 Ingress API（beta 版），用来表示 “7层”（HTTP）服务。 userspace 代理模式 这种模式，kube-proxy 会监视 Kubernetes master 对 Service 对象和 Endpoints 对象的添加和移除。 对每个 Service，它会在本地 Node 上打开一个端口（随机选择）。 任何连接到“代理端口”的请求，都会被代理到 Service 的backend Pods 中的某个上面（如 Endpoints 所报告的一样）。 使用哪个 backend Pod，是基于 Service 的 SessionAffinity 来确定的。 最后，它安装 iptables 规则，捕获到达该 Service 的 clusterIP（是虚拟 IP）和 Port 的请求，并重定向到代理端口，代理端口再代理请求到 backend Pod。 网络返回的结果是，任何到达 Service 的 IP:Port 的请求，都会被代理到一个合适的 backend，不需要客户端知道关于 Kubernetes、Service、或 Pod 的任何信息。 默认的策略是，通过 round-robin 算法来选择 backend Pod。 实现基于客户端 IP 的会话亲和性，可以通过设置 service.spec.sessionAffinity 的值为 \"ClientIP\" （默认值为 \"None\"）。 图片 - userspace代理模式下Service概览图 iptables 代理模式 这种模式，kube-proxy 会监视 Kubernetes master 对 Service 对象和 Endpoints 对象的添加和移除。 对每个 Service，它会安装 iptables 规则，从而捕获到达该 Service 的 clusterIP（虚拟 IP）和端口的请求，进而将请求重定向到 Service 的一组 backend 中的某个上面。 对于每个 Endpoints 对象，它也会安装 iptables 规则，这个规则会选择一个 backend Pod。 默认的策略是，随机选择一个 backend。 实现基于客户端 IP 的会话亲和性，可以将 service.spec.sessionAffinity 的值设置为 \"ClientIP\" （默认值为 \"None\"）。 和 userspace 代理类似，网络返回的结果是，任何到达 Service 的 IP:Port 的请求，都会被代理到一个合适的 backend，不需要客户端知道关于 Kubernetes、Service、或 Pod 的任何信息。 这应该比 userspace 代理更快、更可靠。然而，不像 userspace 代理，如果初始选择的 Pod 没有响应，iptables 代理能够自动地重试另一个 Pod，所以它需要依赖 readiness probes。 图片 - iptables代理模式下Service概览图 多端口 Service 很多 Service 需要暴露多个端口。对于这种情况，Kubernetes 支持在 Service 对象中定义多个端口。 当使用多个端口时，必须给出所有的端口的名称，这样 Endpoint 就不会产生歧义，例如： kind: Service apiVersion: v1 metadata: name: my-service spec: selector: app: MyApp ports: - name: http protocol: TCP port: 80 targetPort: 9376 - name: https protocol: TCP port: 443 targetPort: 9377 选择自己的 IP 地址 在 Service 创建的请求中，可以通过设置 spec.clusterIP 字段来指定自己的集群 IP 地址。 比如，希望替换一个已经已存在的 DNS 条目，或者遗留系统已经配置了一个固定的 IP 且很难重新配置。 用户选择的 IP 地址必须合法，并且这个 IP 地址在 service-cluster-ip-range CIDR 范围内，这对 API Server 来说是通过一个标识来指定的。 如果 IP 地址不合法，API Server 会返回 HTTP 状态码 422，表示值不合法。 为何不使用 round-robin DNS？ 一个不时出现的问题是，为什么我们都使用 VIP 的方式，而不使用标准的 round-robin DNS，有如下几个原因： 长久以来，DNS 库都没能认真对待 DNS TTL、缓存域名查询结果 很多应用只查询一次 DNS 并缓存了结果 就算应用和库能够正确查询解析，每个客户端反复重解析造成的负载也是非常难以管理的 我们尽力阻止用户做那些对他们没有好处的事情，如果很多人都来问这个问题，我们可能会选择实现它。 服务发现 Kubernetes 支持2种基本的服务发现模式 —— 环境变量和 DNS。 环境变量 当 Pod 运行在 Node 上，kubelet 会为每个活跃的 Service 添加一组环境变量。 它同时支持 Docker links 兼容 变量（查看 makeLinkVariables）、简单的 {SVCNAME}_SERVICE_HOST 和 {SVCNAME}_SERVICE_PORT 变量，这里 Service 的名称需大写，横线被转换成下划线。 举个例子，一个名称为 \"redis-master\" 的 Service 暴露了 TCP 端口 6379，同时给它分配了 Cluster IP 地址 10.0.0.11，这个 Service 生成了如下环境变量： REDIS_MASTER_SERVICE_HOST=10.0.0.11 REDIS_MASTER_SERVICE_PORT=6379 REDIS_MASTER_PORT=tcp://10.0.0.11:6379 REDIS_MASTER_PORT_6379_TCP=tcp://10.0.0.11:6379 REDIS_MASTER_PORT_6379_TCP_PROTO=tcp REDIS_MASTER_PORT_6379_TCP_PORT=6379 REDIS_MASTER_PORT_6379_TCP_ADDR=10.0.0.11 这意味着需要有顺序的要求 —— Pod 想要访问的任何 Service 必须在 Pod 自己之前被创建，否则这些环境变量就不会被赋值。DNS 并没有这个限制。 DNS 一个可选（尽管强烈推荐）集群插件 是 DNS 服务器。 DNS 服务器监视着创建新 Service 的 Kubernetes API，从而为每一个 Service 创建一组 DNS 记录。 如果整个集群的 DNS 一直被启用，那么所有的 Pod 应该能够自动对 Service 进行名称解析。 例如，有一个名称为 \"my-service\" 的 Service，它在 Kubernetes 集群中名为 \"my-ns\" 的 Namespace 中，为 \"my-service.my-ns\" 创建了一条 DNS 记录。 在名称为 \"my-ns\" 的 Namespace 中的 Pod 应该能够简单地通过名称查询找到 \"my-service\"。 在另一个 Namespace 中的 Pod 必须限定名称为 \"my-service.my-ns\"。 这些名称查询的结果是 Cluster IP。 Kubernetes 也支持对端口名称的 DNS SRV（Service）记录。 如果名称为 \"my-service.my-ns\" 的 Service 有一个名为 \"http\" 的 TCP 端口，可以对 \"_http._tcp.my-service.my-ns\" 执行 DNS SRV 查询，得到 \"http\" 的端口号。 Kubernetes DNS 服务器是唯一的一种能够访问 ExternalName 类型的 Service 的方式。 更多信息可以查看 DNS Pod 和 Service。 Headless Service 有时不需要或不想要负载均衡，以及单独的 Service IP。 遇到这种情况，可以通过指定 Cluster IP（spec.clusterIP）的值为 \"None\" 来创建 Headless Service。 这个选项允许开发人员自由寻找他们自己的方式，从而降低与 Kubernetes 系统的耦合性。 应用仍然可以使用一种自注册的模式和适配器，对其它需要发现机制的系统能够很容易地基于这个 API 来构建。 对这类 Service 并不会分配 Cluster IP，kube-proxy 不会处理它们，而且平台也不会为它们进行负载均衡和路由。 DNS 如何实现自动配置，依赖于 Service 是否定义了 selector。 配置 Selector 对定义了 selector 的 Headless Service，Endpoint 控制器在 API 中创建了 Endpoints 记录，并且修改 DNS 配置返回 A 记录（地址），通过这个地址直接到达 Service 的后端 Pod 上。 不配置 Selector 对没有定义 selector 的 Headless Service，Endpoint 控制器不会创建 Endpoints 记录。 然而 DNS 系统会查找和配置，无论是： ExternalName 类型 Service 的 CNAME 记录 记录：与 Service 共享一个名称的任何 Endpoints，以及所有其它类型 发布服务 —— 服务类型 对一些应用（如 Frontend）的某些部分，可能希望通过外部（Kubernetes 集群外部）IP 地址暴露 Service。 Kubernetes ServiceTypes 允许指定一个需要的类型的 Service，默认是 ClusterIP 类型。 Type 的取值以及行为如下： ClusterIP：通过集群的内部 IP 暴露服务，选择该值，服务只能够在集群内部可以访问，这也是默认的 ServiceType。 NodePort：通过每个 Node 上的 IP 和静态端口（NodePort）暴露服务。NodePort 服务会路由到 ClusterIP 服务，这个 ClusterIP 服务会自动创建。通过请求 :，可以从集群的外部访问一个 NodePort 服务。 LoadBalancer：使用云提供商的负载均衡器，可以向外部暴露服务。外部的负载均衡器可以路由到 NodePort 服务和 ClusterIP 服务。 ExternalName：通过返回 CNAME 和它的值，可以将服务映射到 externalName 字段的内容（例如， foo.bar.example.com）。 没有任何类型代理被创建，这只有 Kubernetes 1.7 或更高版本的 kube-dns 才支持。 NodePort 类型 如果设置 type 的值为 \"NodePort\"，Kubernetes master 将从给定的配置范围内（默认：30000-32767）分配端口，每个 Node 将从该端口（每个 Node 上的同一端口）代理到 Service。该端口将通过 Service 的 spec.ports[*].nodePort 字段被指定。 如果需要指定的端口号，可以配置 nodePort 的值，系统将分配这个端口，否则调用 API 将会失败（比如，需要关心端口冲突的可能性）。 这可以让开发人员自由地安装他们自己的负载均衡器，并配置 Kubernetes 不能完全支持的环境参数，或者直接暴露一个或多个 Node 的 IP 地址。 需要注意的是，Service 将能够通过 :spec.ports[*].nodePort 和 spec.clusterIp:spec.ports[*].port 而对外可见。 LoadBalancer 类型 使用支持外部负载均衡器的云提供商的服务，设置 type 的值为 \"LoadBalancer\"，将为 Service 提供负载均衡器。 负载均衡器是异步创建的，关于被提供的负载均衡器的信息将会通过 Service 的 status.loadBalancer 字段被发布出去。 kind: Service apiVersion: v1 metadata: name: my-service spec: selector: app: MyApp ports: - protocol: TCP port: 80 targetPort: 9376 nodePort: 30061 clusterIP: 10.0.171.239 loadBalancerIP: 78.11.24.19 type: LoadBalancer status: loadBalancer: ingress: - ip: 146.148.47.155 来自外部负载均衡器的流量将直接打到 backend Pod 上，不过实际它们是如何工作的，这要依赖于云提供商。 在这些情况下，将根据用户设置的 loadBalancerIP 来创建负载均衡器。 某些云提供商允许设置 loadBalancerIP。如果没有设置 loadBalancerIP，将会给负载均衡器指派一个临时 IP。 如果设置了 loadBalancerIP，但云提供商并不支持这种特性，那么设置的 loadBalancerIP 值将会被忽略掉。 AWS 内部负载均衡器 在混合云环境中，有时从虚拟私有云（VPC）环境中的服务路由流量是非常有必要的。 可以通过在 Service 中增加 annotation 来实现，如下所示： [...] metadata: name: my-service annotations: service.beta.kubernetes.io/aws-load-balancer-internal: 0.0.0.0/0 [...] 在水平分割的 DNS 环境中，需要两个 Service 来将外部和内部的流量路由到 Endpoint 上。 AWS SSL 支持 对运行在 AWS 上部分支持 SSL 的集群，从 1.3 版本开始，可以为 LoadBalancer 类型的 Service 增加两个 annotation： metadata: name: my-service annotations: service.beta.kubernetes.io/aws-load-balancer-ssl-cert: arn:aws:acm:us-east-1:123456789012:certificate/12345678-1234-1234-1234-123456789012 第一个 annotation 指定了使用的证书。它可以是第三方发行商发行的证书，这个证书或者被上传到 IAM，或者由 AWS 的证书管理器创建。 metadata: name: my-service annotations: service.beta.kubernetes.io/aws-load-balancer-backend-protocol: (https|http|ssl|tcp) 第二个 annotation 指定了 Pod 使用的协议。 对于 HTTPS 和 SSL，ELB 将期望该 Pod 基于加密的连接来认证自身。 HTTP 和 HTTPS 将选择7层代理：ELB 将中断与用户的连接，当转发请求时，会解析 Header 信息并添加上用户的 IP 地址（Pod 将只能在连接的另一端看到该 IP 地址）。 TCP 和 SSL 将选择4层代理：ELB 将转发流量，并不修改 Header 信息。 外部 IP 如果外部的 IP 路由到集群中一个或多个 Node 上，Kubernetes Service 会被暴露给这些 externalIPs。 通过外部 IP（作为目的 IP 地址）进入到集群，打到 Service 的端口上的流量，将会被路由到 Service 的 Endpoint 上。 externalIPs 不会被 Kubernetes 管理，它属于集群管理员的职责范畴。 根据 Service 的规定，externalIPs 可以同任意的 ServiceType 来一起指定。 在下面的例子中，my-service 可以在 80.11.12.10:80（外部 IP:端口）上被客户端访问。 kind: Service apiVersion: v1 metadata: name: my-service spec: selector: app: MyApp ports: - name: http protocol: TCP port: 80 targetPort: 9376 externalIPs: - 80.11.12.10 不足之处 为 VIP 使用 userspace 代理，将只适合小型到中型规模的集群，不能够扩展到上千 Service 的大型集群。 查看 最初设计方案 获取更多细节。 使用 userspace 代理，隐藏了访问 Service 的数据包的源 IP 地址。 这使得一些类型的防火墙无法起作用。 iptables 代理不会隐藏 Kubernetes 集群内部的 IP 地址，但却要求客户端请求必须通过一个负载均衡器或 Node 端口。 Type 字段支持嵌套功能 —— 每一层需要添加到上一层里面。 不会严格要求所有云提供商（例如，GCE 就没必要为了使一个 LoadBalancer 能工作而分配一个 NodePort，但是 AWS 需要 ），但当前 API 是强制要求的。 未来工作 未来我们能预见到，代理策略可能会变得比简单的 round-robin 均衡策略有更多细微的差别，比如 master 选举或分片。 我们也能想到，某些 Service 将具有 “真正” 的负载均衡器，这种情况下 VIP 将简化数据包的传输。 我们打算为 L7（HTTP）Service 改进我们对它的支持。 我们打算为 Service 实现更加灵活的请求进入模式，这些 Service 包含当前 ClusterIP、NodePort 和 LoadBalancer 模式，或者更多。 VIP 的那些骇人听闻的细节 对很多想使用 Service 的人来说，前面的信息应该足够了。 然而，有很多内部原理性的内容，还是值去理解的。 避免冲突 Kubernetes 最主要的哲学之一，是用户不应该暴露那些能够导致他们操作失败、但又不是他们的过错的场景。 这种场景下，让我们来看一下网络端口 —— 用户不应该必须选择一个端口号，而且该端口还有可能与其他用户的冲突。 这就是说，在彼此隔离状态下仍然会出现失败。 为了使用户能够为他们的 Service 选择一个端口号，我们必须确保不能有2个 Service 发生冲突。 我们可以通过为每个 Service 分配它们自己的 IP 地址来实现。 为了保证每个 Service 被分配到一个唯一的 IP，需要一个内部的分配器能够原子地更新 etcd 中的一个全局分配映射表，这个更新操作要先于创建每一个 Service。 为了使 Service 能够获取到 IP，这个映射表对象必须在注册中心存在，否则创建 Service 将会失败，指示一个 IP 不能被分配。 一个后台 Controller 的职责是创建映射表（从 Kubernetes 的旧版本迁移过来，旧版本中是通过在内存中加锁的方式实现），并检查由于管理员干预和清除任意 IP 造成的不合理分配，这些 IP 被分配了但当前没有 Service 使用它们。 IP 和 VIP 不像 Pod 的 IP 地址，它实际路由到一个固定的目的地，Service 的 IP 实际上不能通过单个主机来进行应答。 相反，我们使用 iptables（Linux 中的数据包处理逻辑）来定义一个虚拟IP地址（VIP），它可以根据需要透明地进行重定向。 当客户端连接到 VIP 时，它们的流量会自动地传输到一个合适的 Endpoint。 环境变量和 DNS，实际上会根据 Service 的 VIP 和端口来进行填充。 Userspace 作为一个例子，考虑前面提到的图片处理应用程序。 当创建 backend Service 时，Kubernetes master 会给它指派一个虚拟 IP 地址，比如 10.0.0.1。 假设 Service 的端口是 1234，该 Service 会被集群中所有的 kube-proxy 实例观察到。 当代理看到一个新的 Service， 它会打开一个新的端口，建立一个从该 VIP 重定向到新端口的 iptables，并开始接收请求连接。 当一个客户端连接到一个 VIP，iptables 规则开始起作用，它会重定向该数据包到 Service代理 的端口。 Service代理 选择一个 backend，并将客户端的流量代理到 backend 上。 这意味着 Service 的所有者能够选择任何他们想使用的端口，而不存在冲突的风险。 客户端可以简单地连接到一个 IP 和端口，而不需要知道实际访问了哪些 Pod。 Iptables 再次考虑前面提到的图片处理应用程序。 当创建 backend Service 时，Kubernetes master 会给它指派一个虚拟 IP 地址，比如 10.0.0.1。 假设 Service 的端口是 1234，该 Service 会被集群中所有的 kube-proxy 实例观察到。 当代理看到一个新的 Service， 它会安装一系列的 iptables 规则，从 VIP 重定向到 per-Service 规则。 该 per-Service 规则连接到 per-Endpoint 规则，该 per-Endpoint 规则会重定向（目标 NAT）到 backend。 当一个客户端连接到一个 VIP，iptables 规则开始起作用。一个 backend 会被选择（或者根据会话亲和性，或者随机），数据包被重定向到这个 backend。 不像 userspace 代理，数据包从来不拷贝到用户空间，kube-proxy 不是必须为该 VIP 工作而运行，并且客户端 IP 是不可更改的。 当流量打到 Node 的端口上，或通过负载均衡器，会执行相同的基本流程，但是在那些案例中客户端 IP 是可以更改的。 API 对象 在 Kubernetes REST API 中，Service 是 top-level 资源。关于 API 对象的更多细节可以查看：Service API 对象。 更多信息 使用 Service 连接 Frontend 到 Backend Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-10-17 11:18:59 "},"concepts/volume.html":{"url":"concepts/volume.html","title":"Volume和Persistent Volume","keywords":"","body":"Kubernetes存储卷 我们知道默认情况下容器的数据都是非持久化的，在容器消亡以后数据也跟着丢失，所以Docker提供了Volume机制以便将数据持久化存储。类似的，Kubernetes提供了更强大的Volume机制和丰富的插件，解决了容器数据持久化和容器间共享数据的问题。 Volume 目前，Kubernetes支持以下Volume类型： emptyDir hostPath gcePersistentDisk awsElasticBlockStore nfs iscsi flocker glusterfs rbd cephfs gitRepo secret persistentVolumeClaim downwardAPI azureFileVolume vsphereVolume flexvolume 注意，这些volume并非全部都是持久化的，比如emptyDir、secret、gitRepo等，这些volume会随着Pod的消亡而消失。 PersistentVolume 对于持久化的Volume，PersistentVolume (PV)和PersistentVolumeClaim (PVC)提供了更方便的管理卷的方法：PV提供网络存储资源，而PVC请求存储资源。这样，设置持久化的工作流包括配置底层文件系统或者云数据卷、创建持久性数据卷、最后创建claim来将pod跟数据卷关联起来。PV和PVC可以将pod和数据卷解耦，pod不需要知道确切的文件系统或者支持它的持久化引擎。 PV PersistentVolume（PV）是集群之中的一块网络存储。跟 Node 一样，也是集群的资源。PV 跟 Volume (卷) 类似，不过会有独立于 Pod 的生命周期。比如一个NFS的PV可以定义为 apiVersion: v1 kind: PersistentVolume metadata: name: pv0003 spec: capacity: storage: 5Gi accessModes: - ReadWriteOnce persistentVolumeReclaimPolicy: Recycle nfs: path: /tmp server: 172.17.0.2 PV的访问模式有三种： 第一种，ReadWriteOnce：是最基本的方式，可读可写，但只支持被单个Pod挂载。 第二种，ReadOnlyMany：可以以只读的方式被多个Pod挂载。 第三种，ReadWriteMany：这种存储可以以读写的方式被多个Pod共享。不是每一种存储都支持这三种方式，像共享方式，目前支持的还比较少，比较常用的是NFS。在PVC绑定PV时通常根据两个条件来绑定，一个是存储的大小，另一个就是访问模式。 StorageClass 上面通过手动的方式创建了一个NFS Volume，这在管理很多Volume的时候不太方便。Kubernetes还提供了StorageClass来动态创建PV，不仅节省了管理员的时间，还可以封装不同类型的存储供PVC选用。 GCE的例子： kind: StorageClass apiVersion: storage.k8s.io/v1beta1 metadata: name: slow provisioner: kubernetes.io/gce-pd parameters: type: pd-standard zone: us-central1-a Ceph RBD的例子： apiVersion: storage.k8s.io/v1beta1 kind: StorageClass metadata: name: fast provisioner: kubernetes.io/rbd parameters: monitors: 10.16.153.105:6789 adminId: kube adminSecretName: ceph-secret adminSecretNamespace: kube-system pool: kube userId: kube userSecretName: ceph-secret-user PVC PV是存储资源，而PersistentVolumeClaim (PVC) 是对PV的请求。PVC跟Pod类似：Pod消费Node的源，而PVC消费PV资源；Pod能够请求CPU和内存资源，而PVC请求特定大小和访问模式的数据卷。 kind: PersistentVolumeClaim apiVersion: v1 metadata: name: myclaim spec: accessModes: - ReadWriteOnce resources: requests: storage: 8Gi selector: matchLabels: release: \"stable\" matchExpressions: - {key: environment, operator: In, values: [dev]} PVC可以直接挂载到Pod中： kind: Pod apiVersion: v1 metadata: name: mypod spec: containers: - name: myfrontend image: dockerfile/nginx volumeMounts: - mountPath: \"/var/www/html\" name: mypd volumes: - name: mypd persistentVolumeClaim: claimName: myclaim emptyDir 如果Pod配置了emptyDir类型Volume， Pod 被分配到Node上时候，会创建emptyDir，只要Pod运行在Node上，emptyDir都会存在（容器挂掉不会导致emptyDir丢失数据），但是如果Pod从Node上被删除（Pod被删除，或者Pod发生迁移），emptyDir也会被删除，并且永久丢失。 apiVersion: v1 kind: Pod metadata: name: test-pd spec: containers: - image: gcr.io/google_containers/test-webserver name: test-container volumeMounts: - mountPath: /test-pd name: test-volume volumes: - name: test-volume hostPath: # directory location on host path: /data 其他Volume说明 hostPath hostPath允许挂载Node上的文件系统到Pod里面去。如果Pod有需要使用Node上的文件，可以使用hostPath。 - hostPath: path: /tmp/data name: data NFS NFS 是Network File System的缩写，即网络文件系统。Kubernetes中通过简单地配置就可以挂载NFS到Pod中，而NFS中的数据是可以永久保存的，同时NFS支持同时写操作。 volumes: - name: nfs nfs: # FIXME: use the right hostname server: 10.254.234.223 path: \"/\" FlexVolume 注意要把volume plugin放到/usr/libexec/kubernetes/kubelet-plugins/volume/exec//，plugin要实现init/attach/detach/mount/umount等命令（可参考lvm的示例）。 - name: test flexVolume: driver: \"kubernetes.io/lvm\" fsType: \"ext4\" options: volumeID: \"vol1\" size: \"1000m\" volumegroup: \"kube_vg\" Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-08-21 18:23:34 "},"concepts/deployment.html":{"url":"concepts/deployment.html","title":"Deployment","keywords":"","body":"Deployment 简述 Deployment 为 Pod 和 ReplicaSet 提供了一个声明式定义(declarative)方法，用来替代以前的ReplicationController 来方便的管理应用。典型的应用场景包括： 定义Deployment来创建Pod和ReplicaSet 滚动升级和回滚应用 扩容和缩容 暂停和继续Deployment 比如一个简单的nginx应用可以定义为 apiVersion: extensions/v1beta1 kind: Deployment metadata: name: nginx-deployment spec: replicas: 3 template: metadata: labels: app: nginx spec: containers: - name: nginx image: nginx:1.7.9 ports: - containerPort: 80 扩容： kubectl scale deployment nginx-deployment --replicas 10 如果集群支持 horizontal pod autoscaling 的话，还可以为Deployment设置自动扩展： kubectl autoscale deployment nginx-deployment --min=10 --max=15 --cpu-percent=80 更新镜像也比较简单: kubectl set image deployment/nginx-deployment nginx=nginx:1.9.1 回滚： kubectl rollout undo deployment/nginx-deployment Deployment 结构示意图 参考：https://kubernetes.io/docs/api-reference/v1.6/#deploymentspec-v1beta1-apps 图片 - kubernetes deployment cheatsheet Deployment 概念详细解析 本文翻译自kubernetes官方文档：https://kubernetes.io/docs/concepts/workloads/controllers/deployment.md 根据2017年5月10日的Commit 8481c02 翻译。 Deployment 是什么？ Deployment为Pod和Replica Set（下一代Replication Controller）提供声明式更新。 您只需要在 Deployment 中描述您想要的目标状态是什么，Deployment controller 就会帮您将 Pod 和ReplicaSet 的实际状态改变到您的目标状态。您可以定义一个全新的 Deployment 来创建 ReplicaSet 或者删除已有的 Deployment 并创建一个新的来替换。 注意：您不该手动管理由 Deployment 创建的 ReplicaSet，否则您就篡越了 Deployment controller 的职责！下文罗列了 Deployment 对象中已经覆盖了所有的用例。如果未有覆盖您所有需要的用例，请直接在 Kubernetes 的代码库中提 issue。 典型的用例如下： 使用Deployment来创建ReplicaSet。ReplicaSet在后台创建pod。检查启动状态，看它是成功还是失败。 然后，通过更新Deployment的PodTemplateSpec字段来声明Pod的新状态。这会创建一个新的ReplicaSet，Deployment会按照控制的速率将pod从旧的ReplicaSet移动到新的ReplicaSet中。 如果当前状态不稳定，回滚到之前的Deployment revision。每次回滚都会更新Deployment的revision。 扩容Deployment以满足更高的负载。 暂停Deployment来应用PodTemplateSpec的多个修复，然后恢复上线。 根据Deployment 的状态判断上线是否hang住了。 清除旧的不必要的 ReplicaSet。 创建 Deployment 下面是一个 Deployment 示例，它创建了一个 ReplicaSet 来启动3个 nginx pod。 下载示例文件并执行命令： $ kubectl create -f https://kubernetes.io/docs/user-guide/nginx-deployment.yaml --record deployment \"nginx-deployment\" created 将kubectl的 --record 的 flag 设置为 true可以在 annotation 中记录当前命令创建或者升级了该资源。这在未来会很有用，例如，查看在每个 Deployment revision 中执行了哪些命令。 然后立即执行 get 将获得如下结果： $ kubectl get deployments NAME DESIRED CURRENT UP-TO-DATE AVAILABLE AGE nginx-deployment 3 0 0 0 1s 输出结果表明我们希望的repalica数是3（根据deployment中的.spec.replicas配置）当前replica数（ .status.replicas）是0, 最新的replica数（.status.updatedReplicas）是0，可用的replica数（.status.availableReplicas）是0。 过几秒后再执行get命令，将获得如下输出： $ kubectl get deployments NAME DESIRED CURRENT UP-TO-DATE AVAILABLE AGE nginx-deployment 3 3 3 3 18s 我们可以看到Deployment已经创建了3个 replica，所有的 replica 都已经是最新的了（包含最新的pod template），可用的（根据Deployment中的.spec.minReadySeconds声明，处于已就绪状态的pod的最少个数）。执行kubectl get rs和kubectl get pods会显示Replica Set（RS）和Pod已创建。 $ kubectl get rs NAME DESIRED CURRENT READY AGE nginx-deployment-2035384211 3 3 0 18s 您可能会注意到 ReplicaSet 的名字总是-。 $ kubectl get pods --show-labels NAME READY STATUS RESTARTS AGE LABELS nginx-deployment-2035384211-7ci7o 1/1 Running 0 18s app=nginx,pod-template-hash=2035384211 nginx-deployment-2035384211-kzszj 1/1 Running 0 18s app=nginx,pod-template-hash=2035384211 nginx-deployment-2035384211-qqcnn 1/1 Running 0 18s app=nginx,pod-template-hash=2035384211 刚创建的Replica Set将保证总是有3个 nginx 的 pod 存在。 注意： 您必须在 Deployment 中的 selector 指定正确的 pod template label（在该示例中是 app = nginx），不要跟其他的 controller 的 selector 中指定的 pod template label 搞混了（包括 Deployment、Replica Set、Replication Controller 等）。Kubernetes 本身并不会阻止您任意指定 pod template label ，但是如果您真的这么做了，这些 controller 之间会相互打架，并可能导致不正确的行为。 Pod-template-hash label 注意：这个 label 不是用户指定的！ 注意上面示例输出中的 pod label 里的 pod-template-hash label。当 Deployment 创建或者接管 ReplicaSet 时，Deployment controller 会自动为 Pod 添加 pod-template-hash label。这样做的目的是防止 Deployment 的子ReplicaSet 的 pod 名字重复。通过将 ReplicaSet 的 PodTemplate 进行哈希散列，使用生成的哈希值作为 label 的值，并添加到 ReplicaSet selector 里、 pod template label 和 ReplicaSet 管理中的 Pod 上。 更新Deployment 注意： Deployment 的 rollout 当且仅当 Deployment 的 pod template（例如.spec.template）中的label更新或者镜像更改时被触发。其他更新，例如扩容Deployment不会触发 rollout。 假如我们现在想要让 nginx pod 使用nginx:1.9.1的镜像来代替原来的nginx:1.7.9的镜像。 $ kubectl set image deployment/nginx-deployment nginx=nginx:1.9.1 deployment \"nginx-deployment\" image updated 我们可以使用edit命令来编辑 Deployment，修改 .spec.template.spec.containers[0].image ，将nginx:1.7.9 改写成 nginx:1.9.1。 $ kubectl edit deployment/nginx-deployment deployment \"nginx-deployment\" edited 查看 rollout 的状态，只要执行： $ kubectl rollout status deployment/nginx-deployment Waiting for rollout to finish: 2 out of 3 new replicas have been updated... deployment \"nginx-deployment\" successfully rolled out Rollout 成功后，get Deployment： $ kubectl get deployments NAME DESIRED CURRENT UP-TO-DATE AVAILABLE AGE nginx-deployment 3 3 3 3 36s UP-TO-DATE 的 replica 的数目已经达到了配置中要求的数目。 CURRENT 的 replica 数表示 Deployment 管理的 replica 数量，AVAILABLE 的 replica 数是当前可用的replica数量。 我们通过执行kubectl get rs可以看到 Deployment 更新了Pod，通过创建一个新的 ReplicaSet 并扩容了3个 replica，同时将原来的 ReplicaSet 缩容到了0个 replica。 $ kubectl get rs NAME DESIRED CURRENT READY AGE nginx-deployment-1564180365 3 3 0 6s nginx-deployment-2035384211 0 0 0 36s 执行 get pods只会看到当前的新的 pod: $ kubectl get pods NAME READY STATUS RESTARTS AGE nginx-deployment-1564180365-khku8 1/1 Running 0 14s nginx-deployment-1564180365-nacti 1/1 Running 0 14s nginx-deployment-1564180365-z9gth 1/1 Running 0 14s 下次更新这些 pod 的时候，只需要更新 Deployment 中的 pod 的 template 即可。 Deployment 可以保证在升级时只有一定数量的 Pod 是 down 的。默认的，它会确保至少有比期望的Pod数量少一个是up状态（最多一个不可用）。 Deployment 同时也可以确保只创建出超过期望数量的一定数量的 Pod。默认的，它会确保最多比期望的Pod数量多一个的 Pod 是 up 的（最多1个 surge ）。 在未来的 Kuberentes 版本中，将从1-1变成25%-25%。 例如，如果您自己看下上面的 Deployment，您会发现，开始创建一个新的 Pod，然后删除一些旧的 Pod 再创建一个新的。当新的Pod创建出来之前不会杀掉旧的Pod。这样能够确保可用的 Pod 数量至少有2个，Pod的总数最多4个。 $ kubectl describe deployments Name: nginx-deployment Namespace: default CreationTimestamp: Tue, 15 Mar 2016 12:01:06 -0700 Labels: app=nginx Selector: app=nginx Replicas: 3 updated | 3 total | 3 available | 0 unavailable StrategyType: RollingUpdate MinReadySeconds: 0 RollingUpdateStrategy: 1 max unavailable, 1 max surge OldReplicaSets: NewReplicaSet: nginx-deployment-1564180365 (3/3 replicas created) Events: FirstSeen LastSeen Count From SubobjectPath Type Reason Message --------- -------- ----- ---- ------------- -------- ------ ------- 36s 36s 1 {deployment-controller } Normal ScalingReplicaSet Scaled up replica set nginx-deployment-2035384211 to 3 23s 23s 1 {deployment-controller } Normal ScalingReplicaSet Scaled up replica set nginx-deployment-1564180365 to 1 23s 23s 1 {deployment-controller } Normal ScalingReplicaSet Scaled down replica set nginx-deployment-2035384211 to 2 23s 23s 1 {deployment-controller } Normal ScalingReplicaSet Scaled up replica set nginx-deployment-1564180365 to 2 21s 21s 1 {deployment-controller } Normal ScalingReplicaSet Scaled down replica set nginx-deployment-2035384211 to 0 21s 21s 1 {deployment-controller } Normal ScalingReplicaSet Scaled up replica set nginx-deployment-1564180365 to 3 我们可以看到当我们刚开始创建这个 Deployment 的时候，创建了一个 ReplicaSet（nginx-deployment-2035384211），并直接扩容到了3个 replica。 当我们更新这个 Deployment 的时候，它会创建一个新的 ReplicaSet（nginx-deployment-1564180365），将它扩容到1个replica，然后缩容原先的 ReplicaSet 到2个 replica，此时满足至少2个 Pod 是可用状态，同一时刻最多有4个 Pod 处于创建的状态。 接着继续使用相同的 rolling update 策略扩容新的 ReplicaSet 和缩容旧的 ReplicaSet。最终，将会在新的 ReplicaSet 中有3个可用的 replica，旧的 ReplicaSet 的 replica 数目变成0。 Rollover（多个rollout并行） 每当 Deployment controller 观测到有新的 deployment 被创建时，如果没有已存在的 ReplicaSet 来创建期望个数的 Pod 的话，就会创建出一个新的 ReplicaSet 来做这件事。已存在的 ReplicaSet 控制 label 与.spec.selector匹配但是 template 跟.spec.template不匹配的 Pod 缩容。最终，新的 ReplicaSet 将会扩容出.spec.replicas指定数目的 Pod，旧的 ReplicaSet 会缩容到0。 如果您更新了一个的已存在并正在进行中的 Deployment，每次更新 Deployment都会创建一个新的 ReplicaSet并扩容它，同时回滚之前扩容的 ReplicaSet ——将它添加到旧的 ReplicaSet 列表中，开始缩容。 例如，假如您创建了一个有5个niginx:1.7.9 replica的 Deployment，但是当还只有3个nginx:1.7.9的 replica 创建出来的时候您就开始更新含有5个nginx:1.9.1 replica 的 Deployment。在这种情况下，Deployment 会立即杀掉已创建的3个nginx:1.7.9的 Pod，并开始创建nginx:1.9.1的 Pod。它不会等到所有的5个nginx:1.7.9的 Pod 都创建完成后才开始改变航道。 Label selector 更新 我们通常不鼓励更新 label selector，我们建议实现规划好您的 selector。 任何情况下，只要您想要执行 label selector 的更新，请一定要谨慎并确认您已经预料到所有可能因此导致的后果。 增添 selector 需要同时在 Deployment 的 spec 中更新新的 label，否则将返回校验错误。此更改是不可覆盖的，这意味着新的 selector 不会选择使用旧 selector 创建的 ReplicaSet 和 Pod，从而导致所有旧版本的 ReplicaSet 都被丢弃，并创建新的 ReplicaSet。 更新 selector，即更改 selector key 的当前值，将导致跟增添 selector 同样的后果。 删除 selector，即删除 Deployment selector 中的已有的 key，不需要对 Pod template label 做任何更改，现有的 ReplicaSet 也不会成为孤儿，但是请注意，删除的 label 仍然存在于现有的 Pod 和 ReplicaSet 中。 回退Deployment 有时候您可能想回退一个 Deployment，例如，当 Deployment 不稳定时，比如一直 crash looping。 默认情况下，kubernetes 会在系统中保存前两次的 Deployment 的 rollout 历史记录，以便您可以随时回退（您可以修改revision history limit来更改保存的revision数）。 注意： 只要 Deployment 的 rollout 被触发就会创建一个 revision。也就是说当且仅当 Deployment 的 Pod template（如.spec.template）被更改，例如更新template 中的 label 和容器镜像时，就会创建出一个新的 revision。 其他的更新，比如扩容 Deployment 不会创建 revision——因此我们可以很方便的手动或者自动扩容。这意味着当您回退到历史 revision 是，直有 Deployment 中的 Pod template 部分才会回退。 假设我们在更新 Deployment 的时候犯了一个拼写错误，将镜像的名字写成了nginx:1.91，而正确的名字应该是nginx:1.9.1： $ kubectl set image deployment/nginx-deployment nginx=nginx:1.91 deployment \"nginx-deployment\" image updated Rollout 将会卡住。 $ kubectl rollout status deployments nginx-deployment Waiting for rollout to finish: 2 out of 3 new replicas have been updated... 按住 Ctrl-C 停止上面的 rollout 状态监控。 您会看到旧的 replica（nginx-deployment-1564180365 和 nginx-deployment-2035384211）和新的 replica （nginx-deployment-3066724191）数目都是2个。 $ kubectl get rs NAME DESIRED CURRENT READY AGE nginx-deployment-1564180365 2 2 0 25s nginx-deployment-2035384211 0 0 0 36s nginx-deployment-3066724191 2 2 2 6s 看下创建 Pod，您会看到有两个新的 ReplicaSet 创建的 Pod 处于 ImagePullBackOff 状态，循环拉取镜像。 $ kubectl get pods NAME READY STATUS RESTARTS AGE nginx-deployment-1564180365-70iae 1/1 Running 0 25s nginx-deployment-1564180365-jbqqo 1/1 Running 0 25s nginx-deployment-3066724191-08mng 0/1 ImagePullBackOff 0 6s nginx-deployment-3066724191-eocby 0/1 ImagePullBackOff 0 6s 注意，Deployment controller会自动停止坏的 rollout，并停止扩容新的 ReplicaSet。 $ kubectl describe deployment Name: nginx-deployment Namespace: default CreationTimestamp: Tue, 15 Mar 2016 14:48:04 -0700 Labels: app=nginx Selector: app=nginx Replicas: 2 updated | 3 total | 2 available | 2 unavailable StrategyType: RollingUpdate MinReadySeconds: 0 RollingUpdateStrategy: 1 max unavailable, 1 max surge OldReplicaSets: nginx-deployment-1564180365 (2/2 replicas created) NewReplicaSet: nginx-deployment-3066724191 (2/2 replicas created) Events: FirstSeen LastSeen Count From SubobjectPath Type Reason Message --------- -------- ----- ---- ------------- -------- ------ ------- 1m 1m 1 {deployment-controller } Normal ScalingReplicaSet Scaled up replica set nginx-deployment-2035384211 to 3 22s 22s 1 {deployment-controller } Normal ScalingReplicaSet Scaled up replica set nginx-deployment-1564180365 to 1 22s 22s 1 {deployment-controller } Normal ScalingReplicaSet Scaled down replica set nginx-deployment-2035384211 to 2 22s 22s 1 {deployment-controller } Normal ScalingReplicaSet Scaled up replica set nginx-deployment-1564180365 to 2 21s 21s 1 {deployment-controller } Normal ScalingReplicaSet Scaled down replica set nginx-deployment-2035384211 to 0 21s 21s 1 {deployment-controller } Normal ScalingReplicaSet Scaled up replica set nginx-deployment-1564180365 to 3 13s 13s 1 {deployment-controller } Normal ScalingReplicaSet Scaled up replica set nginx-deployment-3066724191 to 1 13s 13s 1 {deployment-controller } Normal ScalingReplicaSet Scaled down replica set nginx-deployment-1564180365 to 2 13s 13s 1 {deployment-controller } Normal ScalingReplicaSet Scaled up replica set nginx-deployment-3066724191 to 2 为了修复这个问题，我们需要回退到稳定的 Deployment revision。 检查 Deployment 升级的历史记录 首先，检查下 Deployment 的 revision： $ kubectl rollout history deployment/nginx-deployment deployments \"nginx-deployment\": REVISION CHANGE-CAUSE 1 kubectl create -f https://kubernetes.io/docs/user-guide/nginx-deployment.yaml--record 2 kubectl set image deployment/nginx-deployment nginx=nginx:1.9.1 3 kubectl set image deployment/nginx-deployment nginx=nginx:1.91 因为我们创建 Deployment 的时候使用了--record参数可以记录命令，我们可以很方便的查看每次 revision 的变化。 查看单个revision 的详细信息： $ kubectl rollout history deployment/nginx-deployment --revision=2 deployments \"nginx-deployment\" revision 2 Labels: app=nginx pod-template-hash=1159050644 Annotations: kubernetes.io/change-cause=kubectl set image deployment/nginx-deployment nginx=nginx:1.9.1 Containers: nginx: Image: nginx:1.9.1 Port: 80/TCP QoS Tier: cpu: BestEffort memory: BestEffort Environment Variables: No volumes. 回退到历史版本 现在，我们可以决定回退当前的 rollout 到之前的版本： $ kubectl rollout undo deployment/nginx-deployment deployment \"nginx-deployment\" rolled back 也可以使用 --revision参数指定某个历史版本： $ kubectl rollout undo deployment/nginx-deployment --to-revision=2 deployment \"nginx-deployment\" rolled back 与 rollout 相关的命令详细文档见kubectl rollout。 该 Deployment 现在已经回退到了先前的稳定版本。如您所见，Deployment controller产生了一个回退到revison 2的DeploymentRollback的 event。 $ kubectl get deployment NAME DESIRED CURRENT UP-TO-DATE AVAILABLE AGE nginx-deployment 3 3 3 3 30m $ kubectl describe deployment Name: nginx-deployment Namespace: default CreationTimestamp: Tue, 15 Mar 2016 14:48:04 -0700 Labels: app=nginx Selector: app=nginx Replicas: 3 updated | 3 total | 3 available | 0 unavailable StrategyType: RollingUpdate MinReadySeconds: 0 RollingUpdateStrategy: 1 max unavailable, 1 max surge OldReplicaSets: NewReplicaSet: nginx-deployment-1564180365 (3/3 replicas created) Events: FirstSeen LastSeen Count From SubobjectPath Type Reason Message --------- -------- ----- ---- ------------- -------- ------ ------- 30m 30m 1 {deployment-controller } Normal ScalingReplicaSet Scaled up replica set nginx-deployment-2035384211 to 3 29m 29m 1 {deployment-controller } Normal ScalingReplicaSet Scaled up replica set nginx-deployment-1564180365 to 1 29m 29m 1 {deployment-controller } Normal ScalingReplicaSet Scaled down replica set nginx-deployment-2035384211 to 2 29m 29m 1 {deployment-controller } Normal ScalingReplicaSet Scaled up replica set nginx-deployment-1564180365 to 2 29m 29m 1 {deployment-controller } Normal ScalingReplicaSet Scaled down replica set nginx-deployment-2035384211 to 0 29m 29m 1 {deployment-controller } Normal ScalingReplicaSet Scaled up replica set nginx-deployment-3066724191 to 2 29m 29m 1 {deployment-controller } Normal ScalingReplicaSet Scaled up replica set nginx-deployment-3066724191 to 1 29m 29m 1 {deployment-controller } Normal ScalingReplicaSet Scaled down replica set nginx-deployment-1564180365 to 2 2m 2m 1 {deployment-controller } Normal ScalingReplicaSet Scaled down replica set nginx-deployment-3066724191 to 0 2m 2m 1 {deployment-controller } Normal DeploymentRollback Rolled back deployment \"nginx-deployment\" to revision 2 29m 2m 2 {deployment-controller } Normal ScalingReplicaSet Scaled up replica set nginx-deployment-1564180365 to 3 清理 Policy 您可以通过设置.spec.revisonHistoryLimit项来指定 deployment 最多保留多少 revision 历史记录。默认的会保留所有的 revision；如果将该项设置为0，Deployment就不允许回退了。 Deployment 扩容 您可以使用以下命令扩容 Deployment： $ kubectl scale deployment nginx-deployment --replicas 10 deployment \"nginx-deployment\" scaled 假设您的集群中启用了horizontal pod autoscaling，您可以给 Deployment 设置一个 autoscaler，基于当前 Pod的 CPU 利用率选择最少和最多的 Pod 数。 $ kubectl autoscale deployment nginx-deployment --min=10 --max=15 --cpu-percent=80 deployment \"nginx-deployment\" autoscaled 比例扩容 RollingUpdate Deployment 支持同时运行一个应用的多个版本。或者 autoscaler 扩 容 RollingUpdate Deployment 的时候，正在中途的 rollout（进行中或者已经暂停的），为了降低风险，Deployment controller 将会平衡已存在的活动中的 ReplicaSet（有 Pod 的 ReplicaSet）和新加入的 replica。这被称为比例扩容。 例如，您正在运行中含有10个 replica 的 Deployment。maxSurge=3，maxUnavailable=2。 $ kubectl get deploy NAME DESIRED CURRENT UP-TO-DATE AVAILABLE AGE nginx-deployment 10 10 10 10 50s 您更新了一个镜像，而在集群内部无法解析。 $ kubectl set image deploy/nginx-deployment nginx=nginx:sometag deployment \"nginx-deployment\" image updated 镜像更新启动了一个包含ReplicaSet nginx-deployment-1989198191的新的rollout，但是它被阻塞了，因为我们上面提到的maxUnavailable。 $ kubectl get rs NAME DESIRED CURRENT READY AGE nginx-deployment-1989198191 5 5 0 9s nginx-deployment-618515232 8 8 8 1m 然后发起了一个新的Deployment扩容请求。autoscaler将Deployment的repllica数目增加到了15个。Deployment controller需要判断在哪里增加这5个新的replica。如果我们没有谁用比例扩容，所有的5个replica都会加到一个新的ReplicaSet中。如果使用比例扩容，新添加的replica将传播到所有的ReplicaSet中。大的部分加入replica数最多的ReplicaSet中，小的部分加入到replica数少的ReplciaSet中。0个replica的ReplicaSet不会被扩容。 在我们上面的例子中，3个replica将添加到旧的ReplicaSet中，2个replica将添加到新的ReplicaSet中。rollout进程最终会将所有的replica移动到新的ReplicaSet中，假设新的replica成为健康状态。 $ kubectl get deploy NAME DESIRED CURRENT UP-TO-DATE AVAILABLE AGE nginx-deployment 15 18 7 8 7m $ kubectl get rs NAME DESIRED CURRENT READY AGE nginx-deployment-1989198191 7 7 0 7m nginx-deployment-618515232 11 11 11 7m 暂停和恢复Deployment 您可以在发出一次或多次更新前暂停一个 Deployment，然后再恢复它。这样您就能多次暂停和恢复 Deployment，在此期间进行一些修复工作，而不会发出不必要的 rollout。 例如使用刚刚创建 Deployment： $ kubectl get deploy NAME DESIRED CURRENT UP-TO-DATE AVAILABLE AGE nginx 3 3 3 3 1m [mkargaki@dhcp129-211 kubernetes]$ kubectl get rs NAME DESIRED CURRENT READY AGE nginx-2142116321 3 3 3 1m 使用以下命令暂停 Deployment： $ kubectl rollout pause deployment/nginx-deployment deployment \"nginx-deployment\" paused 然后更新 Deplyment中的镜像： $ kubectl set image deploy/nginx nginx=nginx:1.9.1 deployment \"nginx-deployment\" image updated 注意新的 rollout 启动了： $ kubectl rollout history deploy/nginx deployments \"nginx\" REVISION CHANGE-CAUSE 1 $ kubectl get rs NAME DESIRED CURRENT READY AGE nginx-2142116321 3 3 3 2m 您可以进行任意多次更新，例如更新使用的资源： $ kubectl set resources deployment nginx -c=nginx --limits=cpu=200m,memory=512Mi deployment \"nginx\" resource requirements updated Deployment 暂停前的初始状态将继续它的功能，而不会对 Deployment 的更新产生任何影响，只要 Deployment是暂停的。 最后，恢复这个 Deployment，观察完成更新的 ReplicaSet 已经创建出来了： $ kubectl rollout resume deploy nginx deployment \"nginx\" resumed $ KUBECTL get rs -w NAME DESIRED CURRENT READY AGE nginx-2142116321 2 2 2 2m nginx-3926361531 2 2 0 6s nginx-3926361531 2 2 1 18s nginx-2142116321 1 2 2 2m nginx-2142116321 1 2 2 2m nginx-3926361531 3 2 1 18s nginx-3926361531 3 2 1 18s nginx-2142116321 1 1 1 2m nginx-3926361531 3 3 1 18s nginx-3926361531 3 3 2 19s nginx-2142116321 0 1 1 2m nginx-2142116321 0 1 1 2m nginx-2142116321 0 0 0 2m nginx-3926361531 3 3 3 20s ^C $ KUBECTL get rs NAME DESIRED CURRENT READY AGE nginx-2142116321 0 0 0 2m nginx-3926361531 3 3 3 28s 注意： 在恢复 Deployment 之前您无法回退一个已经暂停的 Deployment。 Deployment 状态 Deployment 在生命周期中有多种状态。在创建一个新的 ReplicaSet 的时候它可以是 progressing 状态， complete 状态，或者 fail to progress 状态。 进行中的 Deployment Kubernetes 将执行过下列任务之一的 Deployment 标记为 progressing 状态： Deployment 正在创建新的ReplicaSet过程中。 Deployment 正在扩容一个已有的 ReplicaSet。 Deployment 正在缩容一个已有的 ReplicaSet。 有新的可用的 pod 出现。 您可以使用kubectl rollout status命令监控 Deployment 的进度。 完成的 Deployment Kubernetes 将包括以下特性的 Deployment 标记为 complete 状态： Deployment 最小可用。最小可用意味着 Deployment 的可用 replica 个数等于或者超过 Deployment 策略中的期望个数。 所有与该 Deployment 相关的replica都被更新到了您指定版本，也就说更新完成。 该 Deployment 中没有旧的 Pod 存在。 您可以用kubectl rollout status命令查看 Deployment 是否完成。如果 rollout 成功完成，kubectl rollout status将返回一个0值的 Exit Code。 $ kubectl rollout status deploy/nginx Waiting for rollout to finish: 2 of 3 updated replicas are available... deployment \"nginx\" successfully rolled out $ echo $? 0 失败的 Deployment 您的 Deployment 在尝试部署新的 ReplicaSet 的时候可能卡住，用于也不会完成。这可能是因为以下几个因素引起的： 无效的引用 不可读的 probe failure 镜像拉取错误 权限不够 范围限制 程序运行时配置错误 探测这种情况的一种方式是，在您的 Deployment spec 中指定spec.progressDeadlineSeconds。spec.progressDeadlineSeconds 表示 Deployment controller 等待多少秒才能确定（通过 Deployment status）Deployment进程是卡住的。 下面的kubectl命令设置progressDeadlineSeconds 使 controller 在 Deployment 在进度卡住10分钟后报告： $ kubectl patch deployment/nginx-deployment -p '{\"spec\":{\"progressDeadlineSeconds\":600}}' \"nginx-deployment\" patched 当超过截止时间后，Deployment controller 会在 Deployment 的 status.conditions中增加一条DeploymentCondition，它包括如下属性： Type=Progressing Status=False Reason=ProgressDeadlineExceeded 浏览 Kubernetes API conventions 查看关于status conditions的更多信息。 注意： kubernetes除了报告Reason=ProgressDeadlineExceeded状态信息外不会对卡住的 Deployment 做任何操作。更高层次的协调器可以利用它并采取相应行动，例如，回滚 Deployment 到之前的版本。 注意： 如果您暂停了一个 Deployment，在暂停的这段时间内kubernetnes不会检查您指定的 deadline。您可以在 Deployment 的 rollout 途中安全的暂停它，然后再恢复它，这不会触发超过deadline的状态。 您可能在使用 Deployment 的时候遇到一些短暂的错误，这些可能是由于您设置了太短的 timeout，也有可能是因为各种其他错误导致的短暂错误。例如，假设您使用了无效的引用。当您 Describe Deployment 的时候可能会注意到如下信息： $ kubectl describe deployment nginx-deployment Conditions: Type Status Reason ---- ------ ------ Available True MinimumReplicasAvailable Progressing True ReplicaSetUpdated ReplicaFailure True FailedCreate 执行 kubectl get deployment nginx-deployment -o yaml，Deployement 的状态可能看起来像这个样子： status: availableReplicas: 2 conditions: - lastTransitionTime: 2016-10-04T12:25:39Z lastUpdateTime: 2016-10-04T12:25:39Z message: Replica set \"nginx-deployment-4262182780\" is progressing. reason: ReplicaSetUpdated status: \"True\" type: Progressing - lastTransitionTime: 2016-10-04T12:25:42Z lastUpdateTime: 2016-10-04T12:25:42Z message: Deployment has minimum availability. reason: MinimumReplicasAvailable status: \"True\" type: Available - lastTransitionTime: 2016-10-04T12:25:39Z lastUpdateTime: 2016-10-04T12:25:39Z message: 'Error creating: pods \"nginx-deployment-4262182780-\" is forbidden: exceeded quota: object-counts, requested: pods=1, used: pods=3, limited: pods=2' reason: FailedCreate status: \"True\" type: ReplicaFailure observedGeneration: 3 replicas: 2 unavailableReplicas: 2 最终，一旦超过 Deployment 进程的 deadline，kuberentes 会更新状态和导致 Progressing 状态的原因： Conditions: Type Status Reason ---- ------ ------ Available True MinimumReplicasAvailable Progressing False ProgressDeadlineExceeded ReplicaFailure True FailedCreate 您可以通过缩容 Deployment的方式解决配额不足的问题，或者增加您的 namespace 的配额。如果您满足了配额条件后，Deployment controller 就会完成您的 Deployment rollout，您将看到 Deployment 的状态更新为成功状态（Status=True并且Reason=NewReplicaSetAvailable）。 Conditions: Type Status Reason ---- ------ ------ Available True MinimumReplicasAvailable Progressing True NewReplicaSetAvailable Type=Available、 Status=True 以为这您的Deployment有最小可用性。 最小可用性是在Deployment策略中指定的参数。Type=Progressing 、 Status=True意味着您的Deployment 或者在部署过程中，或者已经成功部署，达到了期望的最少的可用replica数量（查看特定状态的Reason——在我们的例子中Reason=NewReplicaSetAvailable 意味着Deployment已经完成）。 您可以使用kubectl rollout status命令查看Deployment进程是否失败。当Deployment过程超过了deadline，kubectl rollout status将返回非0的exit code。 $ kubectl rollout status deploy/nginx Waiting for rollout to finish: 2 out of 3 new replicas have been updated... error: deployment \"nginx\" exceeded its progress deadline $ echo $? 1 操作失败的 Deployment 所有对完成的 Deployment 的操作都适用于失败的 Deployment。您可以对它扩/缩容，回退到历史版本，您甚至可以多次暂停它来应用 Deployment pod template。 清理Policy 您可以设置 Deployment 中的 .spec.revisionHistoryLimit 项来指定保留多少旧的 ReplicaSet。 余下的将在后台被当作垃圾收集。默认的，所有的 revision 历史就都会被保留。在未来的版本中，将会更改为2。 注意： 将该值设置为0，将导致所有的 Deployment 历史记录都会被清除，该 Deployment 就无法再回退了。 用例 金丝雀 Deployment 如果您想要使用 Deployment 对部分用户或服务器发布 release，您可以创建多个 Deployment，每个 Deployment 对应一个 release，参照 managing resources 中对金丝雀模式的描述。 编写 Deployment Spec 在所有的 Kubernetes 配置中，Deployment 也需要apiVersion，kind和metadata这些配置项。配置文件的通用使用说明查看 部署应用，配置容器，和 使用 kubectl 管理资源 文档。 Deployment也需要 .spec section. Pod Template .spec.template 是 .spec中唯一要求的字段。 .spec.template 是 pod template. 它跟 Pod有一模一样的schema，除了它是嵌套的并且不需要apiVersion 和 kind字段。 另外为了划分Pod的范围，Deployment中的pod template必须指定适当的label（不要跟其他controller重复了，参考selector）和适当的重启策略。 .spec.template.spec.restartPolicy 可以设置为 Always , 如果不指定的话这就是默认配置。 Replicas .spec.replicas 是可以选字段，指定期望的pod数量，默认是1。 Selector .spec.selector是可选字段，用来指定 label selector ，圈定Deployment管理的pod范围。 如果被指定， .spec.selector 必须匹配 .spec.template.metadata.labels，否则它将被API拒绝。如果 .spec.selector 没有被指定， .spec.selector.matchLabels 默认是 .spec.template.metadata.labels。 在Pod的template跟.spec.template不同或者数量超过了.spec.replicas规定的数量的情况下，Deployment会杀掉label跟selector不同的Pod。 注意： 您不应该再创建其他label跟这个selector匹配的pod，或者通过其他Deployment，或者通过其他Controller，例如ReplicaSet和ReplicationController。否则该Deployment会被把它们当成都是自己创建的。Kubernetes不会阻止您这么做。 如果您有多个controller使用了重复的selector，controller们就会互相打架并导致不正确的行为。 策略 .spec.strategy 指定新的Pod替换旧的Pod的策略。 .spec.strategy.type 可以是\"Recreate\"或者是 \"RollingUpdate\"。\"RollingUpdate\"是默认值。 Recreate Deployment .spec.strategy.type==Recreate时，在创建出新的Pod之前会先杀掉所有已存在的Pod。 Rolling Update Deployment .spec.strategy.type==RollingUpdate时，Deployment使用rolling update 的方式更新Pod 。您可以指定maxUnavailable 和 maxSurge 来控制 rolling update 进程。 Max Unavailable .spec.strategy.rollingUpdate.maxUnavailable 是可选配置项，用来指定在升级过程中不可用Pod的最大数量。该值可以是一个绝对值（例如5），也可以是期望Pod数量的百分比（例如10%）。通过计算百分比的绝对值向下取整。如果.spec.strategy.rollingUpdate.maxSurge 为0时，这个值不可以为0。默认值是1。 例如，该值设置成30%，启动rolling update后旧的ReplicatSet将会立即缩容到期望的Pod数量的70%。新的Pod ready后，随着新的ReplicaSet的扩容，旧的ReplicaSet会进一步缩容，确保在升级的所有时刻可以用的Pod数量至少是期望Pod数量的70%。 Max Surge .spec.strategy.rollingUpdate.maxSurge 是可选配置项，用来指定可以超过期望的Pod数量的最大个数。该值可以是一个绝对值（例如5）或者是期望的Pod数量的百分比（例如10%）。当MaxUnavailable为0时该值不可以为0。通过百分比计算的绝对值向上取整。默认值是1。 例如，该值设置成30%，启动rolling update后新的ReplicatSet将会立即扩容，新老Pod的总数不能超过期望的Pod数量的130%。旧的Pod被杀掉后，新的ReplicaSet将继续扩容，旧的ReplicaSet会进一步缩容，确保在升级的所有时刻所有的Pod数量和不会超过期望Pod数量的130%。 Progress Deadline Seconds .spec.progressDeadlineSeconds 是可选配置项，用来指定在系统报告Deployment的failed progressing ——表现为resource的状态中type=Progressing、Status=False、 Reason=ProgressDeadlineExceeded前可以等待的Deployment进行的秒数。Deployment controller会继续重试该Deployment。未来，在实现了自动回滚后， deployment controller在观察到这种状态时就会自动回滚。 如果设置该参数，该值必须大于 .spec.minReadySeconds。 Min Ready Seconds .spec.minReadySeconds是一个可选配置项，用来指定没有任何容器crash的Pod并被认为是可用状态的最小秒数。默认是0（Pod在ready后就会被认为是可用状态）。进一步了解什么什么后Pod会被认为是ready状态，参阅 Container Probes。 Rollback To .spec.rollbackTo 是一个可以选配置项，用来配置Deployment回退的配置。设置该参数将触发回退操作，每次回退完成后，该值就会被清除。 Revision .spec.rollbackTo.revision是一个可选配置项，用来指定回退到的revision。默认是0，意味着回退到上一个revision。 Revision History Limit Deployment revision history存储在它控制的ReplicaSets中。 .spec.revisionHistoryLimit 是一个可选配置项，用来指定可以保留的旧的ReplicaSet数量。该理想值取决于心Deployment的频率和稳定性。如果该值没有设置的话，默认所有旧的Replicaset或会被保留，将资源存储在etcd中，是用kubectl get rs查看输出。每个Deployment的该配置都保存在ReplicaSet中，然而，一旦您删除的旧的RepelicaSet，您的Deployment就无法再回退到那个revison了。 如果您将该值设置为0，所有具有0个replica的ReplicaSet都会被删除。在这种情况下，新的Deployment rollout无法撤销，因为revision history都被清理掉了。 Paused .spec.paused是可以可选配置项，boolean值。用来指定暂停和恢复Deployment。Paused和没有paused的Deployment之间的唯一区别就是，所有对paused deployment中的PodTemplateSpec的修改都不会触发新的rollout。Deployment被创建之后默认是非paused。 Deployment 的替代选择 kubectl rolling update Kubectl rolling update 虽然使用类似的方式更新Pod和ReplicationController。但是我们推荐使用Deployment，因为它是声明式的，客户端侧，具有附加特性，例如即使滚动升级结束后也可以回滚到任何历史版本。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-14 13:44:36 "},"concepts/secret.html":{"url":"concepts/secret.html","title":"Secret","keywords":"","body":"Secret Secret解决了密码、token、密钥等敏感数据的配置问题，而不需要把这些敏感数据暴露到镜像或者Pod Spec中。Secret可以以Volume或者环境变量的方式使用。 Secret有三种类型： Service Account ：用来访问Kubernetes API，由Kubernetes自动创建，并且会自动挂载到Pod的/run/secrets/kubernetes.io/serviceaccount目录中； Opaque ：base64编码格式的Secret，用来存储密码、密钥等； kubernetes.io/dockerconfigjson ：用来存储私有docker registry的认证信息。 Opaque Secret Opaque类型的数据是一个map类型，要求value是base64编码格式： $ echo -n \"admin\" | base64 YWRtaW4= $ echo -n \"1f2d1e2e67df\" | base64 MWYyZDFlMmU2N2Rm secrets.yml apiVersion: v1 kind: Secret metadata: name: mysecret type: Opaque data: password: MWYyZDFlMmU2N2Rm username: YWRtaW4= 接着，就可以创建secret了：kubectl create -f secrets.yml。 创建好secret之后，有两种方式来使用它： 以Volume方式 以环境变量方式 将Secret挂载到Volume中 apiVersion: v1 kind: Pod metadata: labels: name: db name: db spec: volumes: - name: secrets secret: secretName: mysecret containers: - image: gcr.io/my_project_id/pg:v1 name: db volumeMounts: - name: secrets mountPath: \"/etc/secrets\" readOnly: true ports: - name: cp containerPort: 5432 hostPort: 5432 将Secret导出到环境变量中 apiVersion: extensions/v1beta1 kind: Deployment metadata: name: wordpress-deployment spec: replicas: 2 strategy: type: RollingUpdate template: metadata: labels: app: wordpress visualize: \"true\" spec: containers: - name: \"wordpress\" image: \"wordpress\" ports: - containerPort: 80 env: - name: WORDPRESS_DB_USER valueFrom: secretKeyRef: name: mysecret key: username - name: WORDPRESS_DB_PASSWORD valueFrom: secretKeyRef: name: mysecret key: password kubernetes.io/dockerconfigjson 可以直接用kubectl命令来创建用于docker registry认证的secret： $ kubectl create secret docker-registry myregistrykey --docker-server=DOCKER_REGISTRY_SERVER --docker-username=DOCKER_USER --docker-password=DOCKER_PASSWORD --docker-email=DOCKER_EMAIL secret \"myregistrykey\" created. 也可以直接读取~/.docker/config.json的内容来创建： $ cat ~/.docker/config.json | base64 $ cat > myregistrykey.yaml 在创建Pod的时候，通过imagePullSecrets来引用刚创建的myregistrykey: apiVersion: v1 kind: Pod metadata: name: foo spec: containers: - name: foo image: janedoe/awesomeapp:v1 imagePullSecrets: - name: myregistrykey Service Account Service Account用来访问Kubernetes API，由Kubernetes自动创建，并且会自动挂载到Pod的/run/secrets/kubernetes.io/serviceaccount目录中。 $ kubectl run nginx --image nginx deployment \"nginx\" created $ kubectl get pods NAME READY STATUS RESTARTS AGE nginx-3137573019-md1u2 1/1 Running 0 13s $ kubectl exec nginx-3137573019-md1u2 ls /run/secrets/kubernetes.io/serviceaccount ca.crt namespace token Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-08-21 18:23:34 "},"concepts/statefulset.html":{"url":"concepts/statefulset.html","title":"StatefulSet","keywords":"","body":"StatefulSet StatefulSet 作为 Controller 为 Pod 提供唯一的标识。它可以保证部署和 scale 的顺序。 使用案例参考：kubernetes contrib - statefulsets，其中包含zookeeper和kakfa的statefulset设置和使用说明。 StatefulSet是为了解决有状态服务的问题（对应Deployments和ReplicaSets是为无状态服务而设计），其应用场景包括： 稳定的持久化存储，即Pod重新调度后还是能访问到相同的持久化数据，基于PVC来实现 稳定的网络标志，即Pod重新调度后其PodName和HostName不变，基于Headless Service（即没有Cluster IP的Service）来实现 有序部署，有序扩展，即Pod是有顺序的，在部署或者扩展的时候要依据定义的顺序依次依次进行（即从0到N-1，在下一个Pod运行之前所有之前的Pod必须都是Running和Ready状态），基于init containers来实现 有序收缩，有序删除（即从N-1到0） 从上面的应用场景可以发现，StatefulSet由以下几个部分组成： 用于定义网络标志（DNS domain）的Headless Service 用于创建PersistentVolumes的volumeClaimTemplates 定义具体应用的StatefulSet StatefulSet中每个Pod的DNS格式为statefulSetName-{0..N-1}.serviceName.namespace.svc.cluster.local，其中 serviceName为Headless Service的名字 0..N-1为Pod所在的序号，从0开始到N-1 statefulSetName为StatefulSet的名字 namespace为服务所在的namespace，Headless Servic和StatefulSet必须在相同的namespace .cluster.local为Cluster Domain 使用 StatefulSet StatefulSet 适用于有以下某个或多个需求的应用： 稳定，唯一的网络标志。 稳定，持久化存储。 有序，优雅地部署和 scale。 有序，优雅地删除和终止。 有序，自动的滚动升级。 在上文中，稳定是 Pod （重新）调度中持久性的代名词。 如果应用程序不需要任何稳定的标识符、有序部署、删除和 scale，则应该使用提供一组无状态副本的 controller 来部署应用程序，例如 Deployment 或 ReplicaSet 可能更适合您的无状态需求。 限制 StatefulSet 是 beta 资源，Kubernetes 1.5 以前版本不支持。 对于所有的 alpha/beta 的资源，您都可以通过在 apiserver 中设置 --runtime-config 选项来禁用。 给定 Pod 的存储必须由 PersistentVolume Provisioner 根据请求的 storage class 进行配置，或由管理员预先配置。 删除或 scale StatefulSet 将不会删除与 StatefulSet 相关联的 volume。 这样做是为了确保数据安全性，这通常比自动清除所有相关 StatefulSet 资源更有价值。 StatefulSets 目前要求 Headless Service 负责 Pod 的网络身份。 您有责任创建此服务。 组件 下面的示例中描述了 StatefulSet 中的组件。 一个名为 nginx 的 headless service，用于控制网络域。 一个名为 web 的 StatefulSet，它的 Spec 中指定在有 3 个运行 nginx 容器的 Pod。 volumeClaimTemplates 使用 PersistentVolume Provisioner 提供的 PersistentVolumes 作为稳定存储。 apiVersion: v1 kind: Service metadata: name: nginx labels: app: nginx spec: ports: - port: 80 name: web clusterIP: None selector: app: nginx --- apiVersion: apps/v1beta1 kind: StatefulSet metadata: name: web spec: serviceName: \"nginx\" replicas: 3 template: metadata: labels: app: nginx spec: terminationGracePeriodSeconds: 10 containers: - name: nginx image: gcr.io/google_containers/nginx-slim:0.8 ports: - containerPort: 80 name: web volumeMounts: - name: www mountPath: /usr/share/nginx/html volumeClaimTemplates: - metadata: name: www annotations: volume.beta.kubernetes.io/storage-class: anything spec: accessModes: [ \"ReadWriteOnce\" ] resources: requests: storage: 1Gi Pod 身份 StatefulSet Pod 具有唯一的身份，包括序数，稳定的网络身份和稳定的存储。 身份绑定到 Pod 上，不管它（重新）调度到哪个节点上。 序数 对于一个有 N 个副本的 StatefulSet，每个副本都会被指定一个整数序数，在 [0,N)之间，且唯一。 稳定的网络 ID StatefulSet 中的每个 Pod 从 StatefulSet 的名称和 Pod 的序数派生其主机名。构造的主机名的模式是$（statefulset名称)-$(序数)。 上面的例子将创建三个名为web-0，web-1，web-2的 Pod。 StatefulSet 可以使用 Headless Service 来控制其 Pod 的域。此服务管理的域的格式为：$(服务名称).$(namespace).svc.cluster.local，其中 “cluster.local” 是 集群域。 在创建每个Pod时，它将获取一个匹配的 DNS 子域，采用以下形式：$(pod 名称).$(管理服务域)，其中管理服务由 StatefulSet 上的 serviceName 字段定义。 以下是 Cluster Domain，服务名称，StatefulSet 名称以及如何影响 StatefulSet 的 Pod 的 DNS 名称的一些示例。 Cluster Domain Service (ns/name) StatefulSet (ns/name) StatefulSet Domain Pod DNS Pod Hostname cluster.local default/nginx default/web nginx.default.svc.cluster.local web-{0..N-1}.nginx.default.svc.cluster.local web-{0..N-1} cluster.local foo/nginx foo/web nginx.foo.svc.cluster.local web-{0..N-1}.nginx.foo.svc.cluster.local web-{0..N-1} kube.local foo/nginx foo/web nginx.foo.svc.kube.local web-{0..N-1}.nginx.foo.svc.kube.local web-{0..N-1} 注意 Cluster Domain 将被设置成 cluster.local 除非进行了 其他配置。 稳定存储 Kubernetes 为每个 VolumeClaimTemplate 创建一个 PersistentVolume。上面的 nginx 的例子中，每个 Pod 将具有一个由 anything 存储类创建的 1 GB 存储的 PersistentVolume。当该 Pod （重新）调度到节点上，volumeMounts 将挂载与 PersistentVolume Claim 相关联的 PersistentVolume。请注意，与 PersistentVolume Claim 相关联的 PersistentVolume 在 产出 Pod 或 StatefulSet 的时候不会被删除。这必须手动完成。 部署和 Scale 保证 对于有 N 个副本的 StatefulSet，Pod 将按照 {0..N-1} 的顺序被创建和部署。 当 删除 Pod 的时候，将按照逆序来终结，从{N-1..0} 对 Pod 执行 scale 操作之前，它所有的前任必须处于 Running 和 Ready 状态。 在终止 Pod 前，它所有的继任者必须处于完全关闭状态。 不应该将 StatefulSet 的 pod.Spec.TerminationGracePeriodSeconds 设置为 0。这样是不安全的且强烈不建议您这样做。进一步解释，请参阅 强制删除 StatefulSet Pod。 上面的 nginx 示例创建后，3 个 Pod 将按照如下顺序创建 web-0，web-1，web-2。在 web-0 处于 运行并就绪 状态之前，web-1 将不会被部署，同样当 web-1 处于运行并就绪状态之前 web-2也不会被部署。如果在 web-1 运行并就绪后，web-2 启动之前， web-0 失败了，web-2 将不会启动，直到 web-0 成果重启并处于运行并就绪状态。 如果用户通过修补 StatefulSet 来 scale 部署的示例，以使 replicas=1，则 web-2 将首先被终止。 在 web-2 完全关闭和删除之前，web-1 不会被终止。 如果 web-0 在 web-2 终止并且完全关闭之后，但是在 web-1 终止之前失败，则 web-1 将不会终止，除非 web-0 正在运行并准备就绪。 Pod 管理策略 在 Kubernetes 1.7 和之后版本，StatefulSet 允许您放开顺序保证，同时通过 .spec.podManagementPolicy 字段保证身份的唯一性。 OrderedReady Pod 管理 StatefulSet 中默认使用的是 OrderedReady pod 管理。它实现了 如上 所述的行为。 并行 Pod 管理 Parallel pod 管理告诉 StatefulSet controller 并行的启动和终止 Pod，在启动和终止其他 Pod 之前不会等待 Pod 变成 运行并就绪或完全终止状态。 更新策略 在 kubernetes 1.7 和以上版本中，StatefulSet 的 .spec.updateStrategy 字段允许您配置和禁用 StatefulSet 中的容器、label、resource request/limit、annotation 的滚动更新。 删除 OnDelete 更新策略实现了遗留（1.6和以前）的行为。 当 spec.updateStrategy 未指定时，这是默认策略。 当StatefulSet 的 .spec.updateStrategy.type 设置为 OnDelete 时，StatefulSet 控制器将不会自动更新 StatefulSet 中的 Pod。 用户必须手动删除 Pod 以使控制器创建新的 Pod，以反映对StatefulSet的 .spec.template 进行的修改。 滚动更新 RollingUpdate 更新策略在 StatefulSet 中实现 Pod 的自动滚动更新。 当StatefulSet的 .spec.updateStrategy.type 设置为 RollingUpdate 时，StatefulSet 控制器将在 StatefulSet 中删除并重新创建每个 Pod。 它将以与 Pod 终止相同的顺序进行（从最大的序数到最小的序数），每次更新一个 Pod。 在更新其前身之前，它将等待正在更新的 Pod 状态变成正在运行并就绪。 分区 可以通过指定 .spec.updateStrategy.rollingUpdate.partition 来对 RollingUpdate 更新策略进行分区。如果指定了分区，则当 StatefulSet 的 .spec.template 更新时，具有大于或等于分区序数的所有 Pod 将被更新。具有小于分区的序数的所有 Pod 将不会被更新，即使删除它们也将被重新创建。如果 StatefulSet 的 .spec.updateStrategy.rollingUpdate.partition 大于其 .spec.replicas，则其 .spec.template 的更新将不会传播到 Pod。 在大多数情况下，您不需要使用分区，但如果您想要进行分阶段更新，使用金丝雀发布或执行分阶段发布，它们将非常有用。 简单示例 以一个简单的nginx服务web.yaml为例： --- apiVersion: v1 kind: Service metadata: name: nginx labels: app: nginx spec: ports: - port: 80 name: web clusterIP: None selector: app: nginx --- apiVersion: apps/v1beta1 kind: StatefulSet metadata: name: web spec: serviceName: \"nginx\" replicas: 2 template: metadata: labels: app: nginx spec: containers: - name: nginx image: gcr.io/google_containers/nginx-slim:0.8 ports: - containerPort: 80 name: web volumeMounts: - name: www mountPath: /usr/share/nginx/html volumeClaimTemplates: - metadata: name: www annotations: volume.alpha.kubernetes.io/storage-class: anything spec: accessModes: [ \"ReadWriteOnce\" ] resources: requests: storage: 1Gi $ kubectl create -f web.yaml service \"nginx\" created statefulset \"web\" created # 查看创建的headless service和statefulset $ kubectl get service nginx NAME CLUSTER-IP EXTERNAL-IP PORT(S) AGE nginx None 80/TCP 1m $ kubectl get statefulset web NAME DESIRED CURRENT AGE web 2 2 2m # 根据volumeClaimTemplates自动创建PVC（在GCE中会自动创建kubernetes.io/gce-pd类型的volume） $ kubectl get pvc NAME STATUS VOLUME CAPACITY ACCESSMODES AGE www-web-0 Bound pvc-d064a004-d8d4-11e6-b521-42010a800002 1Gi RWO 16s www-web-1 Bound pvc-d06a3946-d8d4-11e6-b521-42010a800002 1Gi RWO 16s # 查看创建的Pod，他们都是有序的 $ kubectl get pods -l app=nginx NAME READY STATUS RESTARTS AGE web-0 1/1 Running 0 5m web-1 1/1 Running 0 4m # 使用nslookup查看这些Pod的DNS $ kubectl run -i --tty --image busybox dns-test --restart=Never --rm /bin/sh / # nslookup web-0.nginx Server: 10.0.0.10 Address 1: 10.0.0.10 kube-dns.kube-system.svc.cluster.local Name: web-0.nginx Address 1: 10.244.2.10 / # nslookup web-1.nginx Server: 10.0.0.10 Address 1: 10.0.0.10 kube-dns.kube-system.svc.cluster.local Name: web-1.nginx Address 1: 10.244.3.12 / # nslookup web-0.nginx.default.svc.cluster.local Server: 10.0.0.10 Address 1: 10.0.0.10 kube-dns.kube-system.svc.cluster.local Name: web-0.nginx.default.svc.cluster.local Address 1: 10.244.2.10 还可以进行其他的操作 # 扩容 $ kubectl scale statefulset web --replicas=5 # 缩容 $ kubectl patch statefulset web -p '{\"spec\":{\"replicas\":3}}' # 镜像更新（目前还不支持直接更新image，需要patch来间接实现） $ kubectl patch statefulset web --type='json' -p='[{\"op\": \"replace\", \"path\": \"/spec/template/spec/containers/0/image\", \"value\":\"gcr.io/google_containers/nginx-slim:0.7\"}]' # 删除StatefulSet和Headless Service $ kubectl delete statefulset web $ kubectl delete service nginx # StatefulSet删除后PVC还会保留着，数据不再使用的话也需要删除 $ kubectl delete pvc www-web-0 www-web-1 zookeeper 另外一个更能说明StatefulSet强大功能的示例为zookeeper.yaml，这个例子仅为讲解，实际可用的配置请使用 https://github.com/kubernetes/contrib/tree/master/statefulsets 中的配置。 --- apiVersion: v1 kind: Service metadata: name: zk-headless labels: app: zk-headless spec: ports: - port: 2888 name: server - port: 3888 name: leader-election clusterIP: None selector: app: zk --- apiVersion: v1 kind: ConfigMap metadata: name: zk-config data: ensemble: \"zk-0;zk-1;zk-2\" jvm.heap: \"2G\" tick: \"2000\" init: \"10\" sync: \"5\" client.cnxns: \"60\" snap.retain: \"3\" purge.interval: \"1\" --- apiVersion: policy/v1beta1 kind: PodDisruptionBudget metadata: name: zk-budget spec: selector: matchLabels: app: zk minAvailable: 2 --- apiVersion: apps/v1beta1 kind: StatefulSet metadata: name: zk spec: serviceName: zk-headless replicas: 3 template: metadata: labels: app: zk annotations: pod.alpha.kubernetes.io/initialized: \"true\" scheduler.alpha.kubernetes.io/affinity: > { \"podAntiAffinity\": { \"requiredDuringSchedulingRequiredDuringExecution\": [{ \"labelSelector\": { \"matchExpressions\": [{ \"key\": \"app\", \"operator\": \"In\", \"values\": [\"zk-headless\"] }] }, \"topologyKey\": \"kubernetes.io/hostname\" }] } } spec: containers: - name: k8szk imagePullPolicy: Always image: gcr.io/google_samples/k8szk:v1 resources: requests: memory: \"4Gi\" cpu: \"1\" ports: - containerPort: 2181 name: client - containerPort: 2888 name: server - containerPort: 3888 name: leader-election env: - name : ZK_ENSEMBLE valueFrom: configMapKeyRef: name: zk-config key: ensemble - name : ZK_HEAP_SIZE valueFrom: configMapKeyRef: name: zk-config key: jvm.heap - name : ZK_TICK_TIME valueFrom: configMapKeyRef: name: zk-config key: tick - name : ZK_INIT_LIMIT valueFrom: configMapKeyRef: name: zk-config key: init - name : ZK_SYNC_LIMIT valueFrom: configMapKeyRef: name: zk-config key: tick - name : ZK_MAX_CLIENT_CNXNS valueFrom: configMapKeyRef: name: zk-config key: client.cnxns - name: ZK_SNAP_RETAIN_COUNT valueFrom: configMapKeyRef: name: zk-config key: snap.retain - name: ZK_PURGE_INTERVAL valueFrom: configMapKeyRef: name: zk-config key: purge.interval - name: ZK_CLIENT_PORT value: \"2181\" - name: ZK_SERVER_PORT value: \"2888\" - name: ZK_ELECTION_PORT value: \"3888\" command: - sh - -c - zkGenConfig.sh && zkServer.sh start-foreground readinessProbe: exec: command: - \"zkOk.sh\" initialDelaySeconds: 15 timeoutSeconds: 5 livenessProbe: exec: command: - \"zkOk.sh\" initialDelaySeconds: 15 timeoutSeconds: 5 volumeMounts: - name: datadir mountPath: /var/lib/zookeeper securityContext: runAsUser: 1000 fsGroup: 1000 volumeClaimTemplates: - metadata: name: datadir annotations: volume.alpha.kubernetes.io/storage-class: anything spec: accessModes: [ \"ReadWriteOnce\" ] resources: requests: storage: 20Gi kubectl create -f zookeeper.yaml 详细的使用说明见zookeeper stateful application。 关于StatefulSet的更多示例请参阅 github.com/kubernetes/contrib - statefulsets，其中包括了zookeeper和kafka。 集群外部访问StatefulSet的Pod 我们设想一下这样的场景：在kubernetes集群外部调试StatefulSet中有序的Pod，那么如何访问这些的pod呢？ 方法是为pod设置label，然后用kubectl expose将其以NodePort的方式暴露到集群外部，以上面的zookeeper的例子来说明，下面使用命令的方式来暴露其中的两个zookeeper节点，也可以写一个serivce配置yaml文件。 kubectl label pod zk-0 zkInst=0 kubectl label pod zk-1 zkInst=1 kubectl expose po zk-0 --port=2181 --target-port=2181 --name=zk-0 --selector=zkInst=0 --type=NodePort kubectl expose po zk-1 --port=2181 --target-port=2181 --name=zk-1 --selector=zkInst=1 --type=NodePort 这样在kubernetes集群外部就可以根据pod所在的主机所映射的端口来访问了。 查看zk-0这个service可以看到如下结果： NAME CLUSTER-IP EXTERNAL-IP PORT(S) AGE zk-0 10.254.98.14 2181:31693/TCP 5m 集群外部就可以使用所有的node中的任何一个IP:31693来访问这个zookeeper实例。 参考 https://kubernetes.io/docs/concepts/workloads/controllers/statefulset/ kubernetes contrib - statefulsets Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-10-23 13:32:31 "},"concepts/daemonset.html":{"url":"concepts/daemonset.html","title":"DaemonSet","keywords":"","body":"DaemonSet 什么是 DaemonSet？ DaemonSet 确保全部（或者一些）Node 上运行一个 Pod 的副本。当有 Node 加入集群时，也会为他们新增一个 Pod 。当有 Node 从集群移除时，这些 Pod 也会被回收。删除 DaemonSet 将会删除它创建的所有 Pod。 使用 DaemonSet 的一些典型用法： 运行集群存储 daemon，例如在每个 Node 上运行 glusterd、ceph。 在每个 Node 上运行日志收集 daemon，例如fluentd、logstash。 在每个 Node 上运行监控 daemon，例如 Prometheus Node Exporter、collectd、Datadog 代理、New Relic 代理，或 Ganglia gmond。 一个简单的用法是，在所有的 Node 上都存在一个 DaemonSet，将被作为每种类型的 daemon 使用。 一个稍微复杂的用法可能是，对单独的每种类型的 daemon 使用多个 DaemonSet，但具有不同的标志，和/或对不同硬件类型具有不同的内存、CPU要求。 编写 DaemonSet Spec 必需字段 和其它所有 Kubernetes 配置一样，DaemonSet 需要 apiVersion、kind 和 metadata字段。有关配置文件的通用信息，详见文档 部署应用、配置容器 和 资源管理 。 DaemonSet 也需要一个 .spec 配置段。 Pod 模板 .spec 唯一必需的字段是 .spec.template。 .spec.template 是一个 Pod 模板。 它与 Pod 具有相同的 schema，除了它是嵌套的，而且不具有 apiVersion 或 kind 字段。 Pod 除了必须字段外，在 DaemonSet 中的 Pod 模板必须指定合理的标签（查看 pod selector）。 在 DaemonSet 中的 Pod 模板必需具有一个值为 Always 的 RestartPolicy，或者未指定它的值，默认是 Always。 Pod Selector .spec.selector 字段表示 Pod Selector，它与 Job 或其它资源的 .sper.selector 的原理是相同的。 spec.selector 表示一个对象，它由如下两个字段组成： matchLabels - 与 ReplicationController 的 .spec.selector 的原理相同。 matchExpressions - 允许构建更加复杂的 Selector，可以通过指定 key、value 列表，以及与 key 和 value 列表的相关的操作符。 当上述两个字段都指定时，结果表示的是 AND 关系。 如果指定了 .spec.selector，必须与 .spec.template.metadata.labels 相匹配。如果没有指定，它们默认是等价的。如果与它们配置的不匹配，则会被 API 拒绝。 如果 Pod 的 label 与 selector 匹配，或者直接基于其它的 DaemonSet、或者 Controller（例如 ReplicationController），也不可以创建任何 Pod。 否则 DaemonSet Controller 将认为那些 Pod 是它创建的。Kubernetes 不会阻止这样做。一个场景是，可能希望在一个具有不同值的、用来测试用的 Node 上手动创建 Pod。 仅在相同的 Node 上运行 Pod 如果指定了 .spec.template.spec.nodeSelector，DaemonSet Controller 将在能够匹配上 Node Selector 的 Node 上创建 Pod。 类似这种情况，可以指定 .spec.template.spec.affinity，然后 DaemonSet Controller 将在能够匹配上 Node Affinity 的 Node 上创建 Pod。 如果根本就没有指定，则 DaemonSet Controller 将在所有 Node 上创建 Pod。 如果调度 Daemon Pod 正常情况下，Pod 运行在哪个机器上是由 Kubernetes 调度器进行选择的。然而，由 Daemon Controller 创建的 Pod 已经确定了在哪个机器上（Pod 创建时指定了 .spec.nodeName），因此： DaemonSet Controller 并不关心一个 Node 的 unschedulable 字段。 DaemonSet Controller 可以创建 Pod，即使调度器还没有被启动，这对集群启动是非常有帮助的。 Daemon Pod 关心 Taint 和 Toleration，它们会为没有指定 tolerationSeconds 的 node.alpha.kubernetes.io/notReady 和 node.alpha.kubernetes.io/unreachable 的 Taint，而创建具有 NoExecute 的 Toleration。这确保了当 alpha 特性的 TaintBasedEvictions 被启用，当 Node 出现故障，比如网络分区，这时它们将不会被清除掉（当 TaintBasedEvictions 特性没有启用，在这些场景下也不会被清除，但会因为 NodeController 的硬编码行为而被清除，Toleration 是不会的）。 与 Daemon Pod 通信 与 DaemonSet 中的 Pod 进行通信，几种可能的模式如下： Push：配置 DaemonSet 中的 Pod 向其它 Service 发送更新，例如统计数据库。它们没有客户端。 NodeIP 和已知端口：DaemonSet 中的 Pod 可以使用 hostPort，从而可以通过 Node IP 访问到 Pod。客户端能通过某种方法知道 Node IP 列表，并且基于此也可以知道端口。 DNS：创建具有相同 Pod Selector 的 Headless Service，然后通过使用 endpoints 资源或从 DNS 检索到多个 A 记录来发现 DaemonSet。 Service：创建具有相同 Pod Selector 的 Service，并使用该 Service 访问到某个随机 Node 上的 daemon。（没有办法访问到特定 Node） 更新 DaemonSet 如果修改了 Node Label，DaemonSet 将立刻向新匹配上的 Node 添加 Pod，同时删除新近无法匹配上的 Node 上的 Pod。 可以修改 DaemonSet 创建的 Pod。然而，不允许对 Pod 的所有字段进行更新。当下次 Node（即使具有相同的名称）被创建时，DaemonSet Controller 还会使用最初的模板。 可以删除一个 DaemonSet。如果使用 kubectl 并指定 --cascade=false 选项，则 Pod 将被保留在 Node 上。然后可以创建具有不同模板的新 DaemonSet。具有不同模板的新 DaemonSet 将鞥能够通过 Label 匹配识别所有已经存在的 Pod。它不会修改或删除它们，即使是错误匹配了 Pod 模板。通过删除 Pod 或者 删除 Node，可以强制创建新的 Pod。 在 Kubernetes 1.6 或以后版本，可以在 DaemonSet 上 执行滚动升级。 init 脚本 很可能通过直接在一个 Node 上启动 daemon 进程（例如，使用 init、upstartd、或 systemd）。这非常好，然而基于 DaemonSet 来运行这些进程有如下一些好处： 像对待应用程序一样，具备为 daemon 提供监控和管理日志的能力。 为 daemon 和应用程序使用相同的配置语言和工具（如 Pod 模板、kubectl）。 Kubernetes 未来版本可能会支持对 DaemonSet 创建 Pod 与 Node升级工作流进行集成。 在资源受限的容器中运行 daemon，能够增加 daemon 和应用容器的隔离性。然而这也实现了在容器中运行 daemon，但却不能在 Pod 中运行（例如，直接基于 Docker 启动）。 裸 Pod 可能要直接创建 Pod，同时指定其运行在特定的 Node 上。 然而，DaemonSet 替换了由于任何原因被删除或终止的 Pod，例如 Node 失败、例行节点维护，比如内核升级。由于这个原因，我们应该使用 DaemonSet 而不是单独创建 Pod。 静态 Pod 很可能，通过在一个指定目录下编写文件来创建 Pod，该目录受 Kubelet 所监视。这些 Pod 被称为 静态 Pod。 不像 DaemonSet，静态 Pod 不受 kubectl 和 其它 Kubernetes API 客户端管理。静态 Pod 不依赖于 apiserver，这使得它们在集群启动的情况下非常有用。 而且，未来静态 Pod 可能会被废弃掉。 Replication Controller DaemonSet 与 Replication Controller 非常类似，它们都能创建 Pod，这些 Pod 都具有不期望被终止的进程（例如，Web 服务器、存储服务器）。 为无状态的 Service 使用 Replication Controller，像 frontend，实现对副本的数量进行扩缩容、平滑升级，比之于精确控制 Pod 运行在某个主机上要重要得多。需要 Pod 副本总是运行在全部或特定主机上，并需要先于其他 Pod 启动，当这被认为非常重要时，应该使用 Daemon Controller。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-22 17:00:47 "},"concepts/serviceaccount.html":{"url":"concepts/serviceaccount.html","title":"ServiceAccount","keywords":"","body":"Service Account Service account为Pod中的进程提供身份信息。 本文是关于 Service Account 的用户指南，管理指南另见 Service Account 的集群管理指南 。 注意：本文档描述的关于 Serivce Account 的行为只有当您按照 Kubernetes 项目建议的方式搭建起集群的情况下才有效。您的集群管理员可能在您的集群中有自定义配置，这种情况下该文档可能并不适用。 当您（真人用户）访问集群（例如使用kubectl命令）时，apiserver 会将您认证为一个特定的 User Account（目前通常是admin，除非您的系统管理员自定义了集群配置）。Pod 容器中的进程也可以与 apiserver 联系。 当它们在联系 apiserver 的时候，它们会被认证为一个特定的 Service Account（例如default）。 使用默认的 Service Account 访问 API server 当您创建 pod 的时候，如果您没有指定一个 service account，系统会自动得在与该pod 相同的 namespace 下为其指派一个default service account。如果您获取刚创建的 pod 的原始 json 或 yaml 信息（例如使用kubectl get pods/podename -o yaml命令），您将看到spec.serviceAccountName字段已经被设置为 automatically set 。 您可以在 pod 中使用自动挂载的 service account 凭证来访问 API，如 Accessing the Cluster 中所描述。 Service account 是否能够取得访问 API 的许可取决于您使用的 授权插件和策略。 在 1.6 以上版本中，您可以选择取消为 serivce account 自动挂载 API 凭证，只需在 service account 中设置 automountServiceAccountToken: false： apiVersion: v1 kind: ServiceAccount metadata: name: build-robot automountServiceAccountToken: false ... 在 1.6 以上版本中，您也可以选择只取消单个 pod 的 API 凭证自动挂载： apiVersion: v1 kind: Pod metadata: name: my-pod spec: serviceAccountName: build-robot automountServiceAccountToken: false ... 如果在 pod 和 service account 中同时设置了 automountServiceAccountToken , pod 设置中的优先级更高。 使用多个Service Account 每个 namespace 中都有一个默认的叫做 default 的 service account 资源。 您可以使用以下命令列出 namespace 下的所有 serviceAccount 资源。 $ kubectl get serviceAccounts NAME SECRETS AGE default 1 1d 您可以像这样创建一个 ServiceAccount 对象： $ cat > /tmp/serviceaccount.yaml 如果您看到如下的 service account 对象的完整输出信息： $ kubectl get serviceaccounts/build-robot -o yaml apiVersion: v1 kind: ServiceAccount metadata: creationTimestamp: 2015-06-16T00:12:59Z name: build-robot namespace: default resourceVersion: \"272500\" selfLink: /api/v1/namespaces/default/serviceaccounts/build-robot uid: 721ab723-13bc-11e5-aec2-42010af0021e secrets: - name: build-robot-token-bvbk5 然后您将看到有一个 token 已经被自动创建，并被 service account 引用。 您可以使用授权插件来 设置 service account 的权限 。 设置非默认的 service account，只需要在 pod 的spec.serviceAccountName 字段中将name设置为您想要用的 service account 名字即可。 在 pod 创建之初 service account 就必须已经存在，否则创建将被拒绝。 您不能更新已创建的 pod 的 service account。 您可以清理 service account，如下所示： $ kubectl delete serviceaccount/build-robot 手动创建 service account 的 API token 假设我们已经有了一个如上文提到的名为 ”build-robot“ 的 service account，我们手动创建一个新的 secret。 $ cat > /tmp/build-robot-secret.yaml 现在您可以确认下新创建的 secret 取代了 “build-robot” 这个 service account 原来的 API token。 所有已不存在的 service account 的 token 将被 token controller 清理掉。 $ kubectl describe secrets/build-robot-secret Name: build-robot-secret Namespace: default Labels: Annotations: kubernetes.io/service-account.name=build-robot,kubernetes.io/service-account.uid=870ef2a5-35cf-11e5-8d06-005056b45392 Type: kubernetes.io/service-account-token Data ==== ca.crt: 1220 bytes token: ... namespace: 7 bytes 注意该内容中的token被省略了。 为 service account 添加 ImagePullSecret 首先，创建一个 imagePullSecret，详见这里。 然后，确认已创建。如： $ kubectl get secrets myregistrykey NAME TYPE DATA AGE myregistrykey kubernetes.io/.dockerconfigjson 1 1d 然后，修改 namespace 中的默认 service account 使用该 secret 作为 imagePullSecret。 kubectl patch serviceaccount default -p '{\"imagePullSecrets\": [{\"name\": \"myregistrykey\"}]}' Vi 交互过程中需要手动编辑： $ kubectl get serviceaccounts default -o yaml > ./sa.yaml $ cat sa.yaml apiVersion: v1 kind: ServiceAccount metadata: creationTimestamp: 2015-08-07T22:02:39Z name: default namespace: default resourceVersion: \"243024\" selfLink: /api/v1/namespaces/default/serviceaccounts/default uid: 052fb0f4-3d50-11e5-b066-42010af0d7b6 secrets: - name: default-token-uudge $ vi sa.yaml [editor session not shown] [delete line with key \"resourceVersion\"] [add lines with \"imagePullSecret:\"] $ cat sa.yaml apiVersion: v1 kind: ServiceAccount metadata: creationTimestamp: 2015-08-07T22:02:39Z name: default namespace: default selfLink: /api/v1/namespaces/default/serviceaccounts/default uid: 052fb0f4-3d50-11e5-b066-42010af0d7b6 secrets: - name: default-token-uudge imagePullSecrets: - name: myregistrykey $ kubectl replace serviceaccount default -f ./sa.yaml serviceaccounts/default 现在，所有当前 namespace 中新创建的 pod 的 spec 中都会增加如下内容： spec: imagePullSecrets: - name: myregistrykey 参考 https://kubernetes.io/docs/tasks/configure-pod-container/configure-service-account/ Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-08-29 12:22:11 "},"concepts/replicaset.html":{"url":"concepts/replicaset.html","title":"ReplicationController和ReplicaSet","keywords":"","body":"ReplicationController和ReplicaSet ReplicationCtronller用来确保容器应用的副本数始终保持在用户定义的副本数，即如果有容器异常退出，会自动创建新的Pod来替代；而如果异常多出来的容器也会自动回收。 在新版本的Kubernetes中建议使用ReplicaSet来取代ReplicationCtronller。ReplicaSet跟ReplicationCtronller没有本质的不同，只是名字不一样，并且ReplicaSet支持集合式的selector。 虽然ReplicaSet可以独立使用，但一般还是建议使用 Deployment 来自动管理ReplicaSet，这样就无需担心跟其他机制的不兼容问题（比如ReplicaSet不支持rolling-update但Deployment支持）。 ReplicaSet示例： apiVersion: extensions/v1beta1 kind: ReplicaSet metadata: name: frontend # these labels can be applied automatically # from the labels in the pod template if not set # labels: # app: guestbook # tier: frontend spec: # this replicas value is default # modify it according to your case replicas: 3 # selector can be applied automatically # from the labels in the pod template if not set, # but we are specifying the selector here to # demonstrate its usage. selector: matchLabels: tier: frontend matchExpressions: - {key: tier, operator: In, values: [frontend]} template: metadata: labels: app: guestbook tier: frontend spec: containers: - name: php-redis image: gcr.io/google_samples/gb-frontend:v3 resources: requests: cpu: 100m memory: 100Mi env: - name: GET_HOSTS_FROM value: dns # If your cluster config does not include a dns service, then to # instead access environment variables to find service host # info, comment out the 'value: dns' line above, and uncomment the # line below. # value: env ports: - containerPort: 80 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-08-21 18:23:34 "},"concepts/job.html":{"url":"concepts/job.html","title":"Job","keywords":"","body":"Job Job负责批处理任务，即仅执行一次的任务，它保证批处理任务的一个或多个Pod成功结束。 Job Spec格式 spec.template格式同Pod RestartPolicy仅支持Never或OnFailure 单个Pod时，默认Pod成功运行后Job即结束 .spec.completions标志Job结束需要成功运行的Pod个数，默认为1 .spec.parallelism标志并行运行的Pod的个数，默认为1 spec.activeDeadlineSeconds标志失败Pod的重试最大时间，超过这个时间不会继续重试 一个简单的例子： apiVersion: batch/v1 kind: Job metadata: name: pi spec: template: metadata: name: pi spec: containers: - name: pi image: perl command: [\"perl\", \"-Mbignum=bpi\", \"-wle\", \"print bpi(2000)\"] restartPolicy: Never $ kubectl create -f ./job.yaml job \"pi\" created $ pods=$(kubectl get pods --selector=job-name=pi --output=jsonpath={.items..metadata.name}) $ kubectl logs $pods 3.141592653589793238462643383279502... Bare Pods 所谓Bare Pods是指直接用PodSpec来创建的Pod（即不在ReplicaSets或者ReplicationController的管理之下的Pods）。这些Pod在Node重启后不会自动重启，但Job则会创建新的Pod继续任务。所以，推荐使用Job来替代Bare Pods，即便是应用只需要一个Pod。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-08-21 18:23:34 "},"concepts/cronjob.html":{"url":"concepts/cronjob.html","title":"CronJob","keywords":"","body":"CronJob Cron Job 管理基于时间的 Job，即： 在给定时间点只运行一次 周期性地在给定时间点运行 一个 CronJob 对象类似于 crontab （cron table）文件中的一行。它根据指定的预定计划周期性地运行一个 Job，格式可以参考 Cron 。 前提条件 当使用的 Kubernetes 集群，版本 >= 1.4（对 ScheduledJob），>= 1.5（对 CronJob），当启动 API Server（参考 为集群开启或关闭 API 版本 获取更多信息）时，通过传递选项 --runtime-config=batch/v2alpha1=true 可以开启 batch/v2alpha1 API。 典型的用法如下所示： 在给定的时间点调度 Job 运行 创建周期性运行的 Job，例如：数据库备份、发送邮件。 CronJob Spec .spec.schedule：调度，必需字段，指定任务运行周期，格式同 Cron .spec.jobTemplate：Job 模板，必需字段，指定需要运行的任务，格式同 Job .spec.startingDeadlineSeconds ：启动 Job 的期限（秒级别），该字段是可选的。如果因为任何原因而错过了被调度的时间，那么错过执行时间的 Job 将被认为是失败的。如果没有指定，则没有期限 .spec.concurrencyPolicy：并发策略，该字段也是可选的。它指定了如何处理被 Cron Job 创建的 Job 的并发执行。只允许指定下面策略中的一种： Allow（默认）：允许并发运行 Job Forbid：禁止并发运行，如果前一个还没有完成，则直接跳过下一个 Replace：取消当前正在运行的 Job，用一个新的来替换 注意，当前策略只能应用于同一个 Cron Job 创建的 Job。如果存在多个 Cron Job，它们创建的 Job 之间总是允许并发运行。 .spec.suspend ：挂起，该字段也是可选的。如果设置为 true，后续所有执行都会被挂起。它对已经开始执行的 Job 不起作用。默认值为 false。 .spec.successfulJobsHistoryLimit 和 .spec.failedJobsHistoryLimit ：历史限制，是可选的字段。它们指定了可以保留多少完成和失败的 Job。 默认没有限制，所有成功和失败的 Job 都会被保留。然而，当运行一个 Cron Job 时，Job 可以很快就堆积很多，推荐设置这两个字段的值。设置限制的值为 0，相关类型的 Job 完成后将不会被保留。 apiVersion: batch/v2alpha1 kind: CronJob metadata: name: hello spec: schedule: \"*/1 * * * *\" jobTemplate: spec: template: spec: containers: - name: hello image: busybox args: - /bin/sh - -c - date; echo Hello from the Kubernetes cluster restartPolicy: OnFailure $ kubectl create -f cronjob.yaml cronjob \"hello\" created 当然，也可以用kubectl run来创建一个CronJob： kubectl run hello --schedule=\"*/1 * * * *\" --restart=OnFailure --image=busybox -- /bin/sh -c \"date; echo Hello from the Kubernetes cluster\" $ kubectl get cronjob NAME SCHEDULE SUSPEND ACTIVE LAST-SCHEDULE hello */1 * * * * False 0 $ kubectl get jobs NAME DESIRED SUCCESSFUL AGE hello-1202039034 1 1 49s $ pods=$(kubectl get pods --selector=job-name=hello-1202039034 --output=jsonpath={.items..metadata.name} -a) $ kubectl logs $pods Mon Aug 29 21:34:09 UTC 2016 Hello from the Kubernetes cluster # 注意，删除cronjob的时候不会自动删除job，这些job可以用kubectl delete job来删除 $ kubectl delete cronjob hello cronjob \"hello\" deleted Cron Job 限制 Cron Job 在每次调度运行时间内 大概 会创建一个 Job 对象。我们之所以说 大概 ，是因为在特定的环境下可能会创建两个 Job，或者一个 Job 都没创建。我们尝试少发生这种情况，但却不能完全避免。因此，创建 Job 操作应该是 幂等的。 Job 根据它所创建的 Pod 的并行度，负责重试创建 Pod，并就决定这一组 Pod 的成功或失败。Cron Job 根本就不会去检查 Pod。 删除 Cron Job 一旦不再需要 Cron Job，简单地可以使用 kubectl 命令删除它： $ kubectl delete cronjob hello cronjob \"hello\" deleted 这将会终止正在创建的 Job。然而，运行中的 Job 将不会被终止，不会删除 Job 或 它们的 Pod。为了清理那些 Job 和 Pod，需要列出该 Cron Job 创建的全部 Job，然后删除它们： $ kubectl get jobs NAME DESIRED SUCCESSFUL AGE hello-1201907962 1 1 11m hello-1202039034 1 1 8m ... $ kubectl delete jobs hello-1201907962 hello-1202039034 ... job \"hello-1201907962\" deleted job \"hello-1202039034\" deleted ... 一旦 Job 被删除，由 Job 创建的 Pod 也会被删除。注意，所有由名称为 “hello” 的 Cron Job 创建的 Job 会以前缀字符串 “hello-” 进行命名。如果想要删除当前 Namespace 中的所有 Job，可以通过命令 kubectl delete jobs --all 立刻删除它们。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-09-03 14:01:40 "},"concepts/ingress.html":{"url":"concepts/ingress.html","title":"Ingress","keywords":"","body":"Ingress解析 前言 这是kubernete官方文档中Ingress Resource的翻译，后面的章节会讲到使用Traefik来做Ingress controller，文章末尾给出了几个相关链接。 术语 在本篇文章中你将会看到一些在其他地方被交叉使用的术语，为了防止产生歧义，我们首先来澄清下。 节点：Kubernetes集群中的一台物理机或者虚拟机。 集群：位于Internet防火墙后的节点，这是kubernetes管理的主要计算资源。 边界路由器：为集群强制执行防火墙策略的路由器。 这可能是由云提供商或物理硬件管理的网关。 集群网络：一组逻辑或物理链接，可根据Kubernetes网络模型实现群集内的通信。 集群网络的实现包括Overlay模型的 flannel 和基于SDN的OVS。 服务：使用标签选择器标识一组pod成为的Kubernetes服务。 除非另有说明，否则服务假定在集群网络内仅可通过虚拟IP访问。 什么是Ingress？ 通常情况下，service和pod仅可在集群内部网络中通过IP地址访问。所有到达边界路由器的流量或被丢弃或被转发到其他地方。从概念上讲，可能像下面这样： internet | ------------ [ Services ] Ingress是授权入站连接到达集群服务的规则集合。 internet | [ Ingress ] --|-----|-- [ Services ] 你可以给Ingress配置提供外部可访问的URL、负载均衡、SSL、基于名称的虚拟主机等。用户通过POST Ingress资源到API server的方式来请求ingress。 Ingress controller负责实现Ingress，通常使用负载平衡器，它还可以配置边界路由和其他前端，这有助于以HA方式处理流量。 先决条件 在使用Ingress resource之前，有必要先了解下面几件事情。Ingress是beta版本的resource，在kubernetes1.1之前还没有。你需要一个Ingress Controller来实现Ingress，单纯的创建一个Ingress没有任何意义。 GCE/GKE会在master节点上部署一个ingress controller。你可以在一个pod中部署任意个自定义的ingress controller。你必须正确地annotate每个ingress，比如 运行多个ingress controller 和 关闭glbc. 确定你已经阅读了Ingress controller的beta版本限制。在非GCE/GKE的环境中，你需要在pod中部署一个controller。 Ingress Resource 最简化的Ingress配置： 1: apiVersion: extensions/v1beta1 2: kind: Ingress 3: metadata: 4: name: test-ingress 5: spec: 6: rules: 7: - http: 8: paths: 9: - path: /testpath 10: backend: 11: serviceName: test 12: servicePort: 80 如果你没有配置Ingress controller就将其POST到API server不会有任何用处 配置说明 1-4行：跟Kubernetes的其他配置一样，ingress的配置也需要apiVersion，kind和metadata字段。配置文件的详细说明请查看部署应用, 配置容器和 使用resources. 5-7行: Ingress spec 中包含配置一个loadbalancer或proxy server的所有信息。最重要的是，它包含了一个匹配所有入站请求的规则列表。目前ingress只支持http规则。 8-9行：每条http规则包含以下信息：一个host配置项（比如for.bar.com，在这个例子中默认是*），path列表（比如：/testpath），每个path都关联一个backend(比如test:80)。在loadbalancer将流量转发到backend之前，所有的入站请求都要先匹配host和path。 10-12行：正如 services doc中描述的那样，backend是一个service:port的组合。Ingress的流量被转发到它所匹配的backend。 全局参数：为了简单起见，Ingress示例中没有全局参数，请参阅资源完整定义的api参考。 在所有请求都不能跟spec中的path匹配的情况下，请求被发送到Ingress controller的默认后端，可以指定全局缺省backend。 Ingress controllers 为了使Ingress正常工作，集群中必须运行Ingress controller。 这与其他类型的控制器不同，其他类型的控制器通常作为kube-controller-manager二进制文件的一部分运行，在集群启动时自动启动。 你需要选择最适合自己集群的Ingress controller或者自己实现一个。 示例和说明可以在这里找到。 在你开始前 以下文档描述了Ingress资源中公开的一组跨平台功能。 理想情况下，所有的Ingress controller都应该符合这个规范，但是我们还没有实现。 GCE和nginx控制器的文档分别在这里和这里。确保您查看控制器特定的文档，以便您了解每个文档的注意事项。 Ingress类型 单Service Ingress Kubernetes中已经存在一些概念可以暴露单个service（查看替代方案），但是你仍然可以通过Ingress来实现，通过指定一个没有rule的默认backend的方式。 ingress.yaml定义文件： apiVersion: extensions/v1beta1 kind: Ingress metadata: name: test-ingress spec: backend: serviceName: testsvc servicePort: 80 使用kubectl create -f命令创建，然后查看ingress： $ kubectl get ing NAME RULE BACKEND ADDRESS test-ingress - testsvc:80 107.178.254.228 107.178.254.228就是Ingress controller为了实现Ingress而分配的IP地址。RULE列表示所有发送给该IP的流量都被转发到了BACKEND所列的Kubernetes service上。 简单展开 如前面描述的那样，kubernete pod中的IP只在集群网络内部可见，我们需要在边界设置一个东西，让它能够接收ingress的流量并将它们转发到正确的端点上。这个东西一般是高可用的loadbalancer。使用Ingress能够允许你将loadbalancer的个数降低到最少，例如，假如你想要创建这样的一个设置： foo.bar.com -> 178.91.123.132 -> / foo s1:80 / bar s2:80 你需要一个这样的ingress： apiVersion: extensions/v1beta1 kind: Ingress metadata: name: test spec: rules: - host: foo.bar.com http: paths: - path: /foo backend: serviceName: s1 servicePort: 80 - path: /bar backend: serviceName: s2 servicePort: 80 使用kubectl create -f创建完ingress后： $ kubectl get ing NAME RULE BACKEND ADDRESS test - foo.bar.com /foo s1:80 /bar s2:80 只要服务（s1，s2）存在，Ingress controller就会将提供一个满足该Ingress的特定loadbalancer实现。 这一步完成后，您将在Ingress的最后一列看到loadbalancer的地址。 基于名称的虚拟主机 Name-based的虚拟主机在同一个IP地址下拥有多个主机名。 foo.bar.com --| |-> foo.bar.com s1:80 | 178.91.123.132 | bar.foo.com --| |-> bar.foo.com s2:80 下面这个ingress说明基于Host header的后端loadbalancer的路由请求： apiVersion: extensions/v1beta1 kind: Ingress metadata: name: test spec: rules: - host: foo.bar.com http: paths: - backend: serviceName: s1 servicePort: 80 - host: bar.foo.com http: paths: - backend: serviceName: s2 servicePort: 80 默认backend：一个没有rule的ingress，如前面章节中所示，所有流量都将发送到一个默认backend。你可以用该技巧通知loadbalancer如何找到你网站的404页面，通过制定一些列rule和一个默认backend的方式。如果请求header中的host不能跟ingress中的host匹配，并且/或请求的URL不能与任何一个path匹配，则流量将路由到你的默认backend。 TLS 你可以通过指定包含TLS私钥和证书的secret来加密Ingress。 目前，Ingress仅支持单个TLS端口443，并假定TLS termination。 如果Ingress中的TLS配置部分指定了不同的主机，则它们将根据通过SNI TLS扩展指定的主机名（假如Ingress controller支持SNI）在多个相同端口上进行复用。 TLS secret中必须包含名为tls.crt和tls.key的密钥，这里面包含了用于TLS的证书和私钥，例如： apiVersion: v1 data: tls.crt: base64 encoded cert tls.key: base64 encoded key kind: Secret metadata: name: testsecret namespace: default type: Opaque 在Ingress中引用这个secret将通知Ingress controller使用TLS加密从将客户端到loadbalancer的channel： apiVersion: extensions/v1beta1 kind: Ingress metadata: name: no-rules-map spec: tls: - secretName: testsecret backend: serviceName: s1 servicePort: 80 请注意，各种Ingress controller支持的TLS功能之间存在差距。 请参阅有关nginx，GCE或任何其他平台特定Ingress controller的文档，以了解TLS在你的环境中的工作原理。 Ingress controller启动时附带一些适用于所有Ingress的负载平衡策略设置，例如负载均衡算法，后端权重方案等。更高级的负载平衡概念（例如持久会话，动态权重）尚未在Ingress中公开。 你仍然可以通过service loadbalancer获取这些功能。 随着时间的推移，我们计划将适用于跨平台的负载平衡模式加入到Ingress资源中。 还值得注意的是，尽管健康检查不直接通过Ingress公开，但Kubernetes中存在并行概念，例如准备探查，可以使你达成相同的最终结果。 请查看特定控制器的文档，以了解他们如何处理健康检查（nginx，GCE）。 更新Ingress 假如你想要向已有的ingress中增加一个新的Host，你可以编辑和更新该ingress： $ kubectl get ing NAME RULE BACKEND ADDRESS test - 178.91.123.132 foo.bar.com /foo s1:80 $ kubectl edit ing test 这会弹出一个包含已有的yaml文件的编辑器，修改它，增加新的Host配置。 spec: rules: - host: foo.bar.com http: paths: - backend: serviceName: s1 servicePort: 80 path: /foo - host: bar.baz.com http: paths: - backend: serviceName: s2 servicePort: 80 path: /foo .. 保存它会更新API server中的资源，这会触发ingress controller重新配置loadbalancer。 $ kubectl get ing NAME RULE BACKEND ADDRESS test - 178.91.123.132 foo.bar.com /foo s1:80 bar.baz.com /foo s2:80 在一个修改过的ingress yaml文件上调用kubectl replace -f命令一样可以达到同样的效果。 跨可用域故障 在不通云供应商之间，跨故障域的流量传播技术有所不同。 有关详细信息，请查看相关Ingress controller的文档。 有关在federation集群中部署Ingress的详细信息，请参阅federation文档。 未来计划 多样化的HTTPS/TLS模型支持（如SNI，re-encryption） 通过声明来请求IP或者主机名 结合L4和L7 Ingress 更多的Ingress controller 请跟踪L7和Ingress的proposal，了解有关资源演进的更多细节，以及Ingress repository，了解有关各种Ingress controller演进的更多详细信息。 替代方案 你可以通过很多种方式暴露service而不必直接使用ingress： 使用Service.Type=LoadBalancer 使用Service.Type=NodePort 使用Port Proxy 部署一个Service loadbalancer 这允许你在多个service之间共享单个IP，并通过Service Annotations实现更高级的负载平衡。 参考 Kubernetes Ingress Resource 使用NGINX Plus负载均衡Kubernetes服务 使用 NGINX 和 NGINX Plus 的 Ingress Controller 进行 Kubernetes 的负载均衡 Kubernetes : Ingress Controller with Træfɪk and Let's Encrypt Kubernetes : Træfɪk and Let's Encrypt at scale Kubernetes Ingress Controller-Træfɪk Kubernetes 1.2 and simplifying advanced networking with Ingress Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-08-21 18:23:34 "},"concepts/traefik-ingress-controller.html":{"url":"concepts/traefik-ingress-controller.html","title":"Traefik Ingress Controller","keywords":"","body":"Traefik Ingress Controller 我们在前面部署了 Traefik 作为Ingress Controller，如果集群外部直接访问Kubenretes内部服务的话，可以直接创建Ingress如下所示： apiVersion: extensions/v1beta1 kind: Ingress metadata: name: traefik-ingress namespace: default spec: rules: - host: traefik.nginx.io http: paths: - path: / backend: serviceName: my-nginx servicePort: 80 Traefik Ingress Controller 当我们处于迁移应用到kuberentes上的阶段时，可能有部分服务实例不在kubernetes上，服务的路由使用nginx配置，这时处于nginx和ingress共存的状态。参考下面的配置： apiVersion: extensions/v1beta1 kind: Ingress metadata: name: td-ingress namespace: default annotations: traefik.frontend.rule.type: PathPrefixStrip kubernetes.io/ingress.class: traefik spec: rules: - host: \"*.tendcloud.com\" http: paths: - path: /docGenerate backend: serviceName: td-sdmk-docgenerate servicePort: 80 注意annotation的配置： traefik.frontend.rule.type: PathPrefixStrip：表示将截掉URL中的path kubernetes.io/ingress.class：表示使用的ingress类型 关于Ingress annotation的更多信息请参考：Ingress Annotations - kubernetes.io。 在nginx中增加配置： upstream docGenerate { server 172.20.0.119:80; keepalive 200; } 172.20.0.119是我们的边缘节点的VIP，见边缘节点配置。 参考 Kubernetes Ingress Backend - traefik.io Kubernetes Ingress Controller - traefik.io Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-30 16:57:20 "},"concepts/configmap.html":{"url":"concepts/configmap.html","title":"ConfigMap","keywords":"","body":"ConfigMap 其实ConfigMap功能在Kubernetes1.2版本的时候就有了，许多应用程序会从配置文件、命令行参数或环境变量中读取配置信息。这些配置信息需要与docker image解耦，你总不能每修改一个配置就重做一个image吧？ConfigMap API给我们提供了向容器中注入配置信息的机制，ConfigMap可以被用来保存单个属性，也可以用来保存整个配置文件或者JSON二进制大对象。 ConfigMap概览 ConfigMap API资源用来保存key-value pair配置数据，这个数据可以在pods里使用，或者被用来为像controller一样的系统组件存储配置数据。虽然ConfigMap跟Secrets类似，但是ConfigMap更方便的处理不含敏感信息的字符串。 注意：ConfigMaps不是属性配置文件的替代品。ConfigMaps只是作为多个properties文件的引用。你可以把它理解为Linux系统中的/etc目录，专门用来存储配置文件的目录。下面举个例子，使用ConfigMap配置来创建Kuberntes Volumes，ConfigMap中的每个data项都会成为一个新文件。 kind: ConfigMap apiVersion: v1 metadata: creationTimestamp: 2016-02-18T19:14:38Z name: example-config namespace: default data: example.property.1: hello example.property.2: world example.property.file: |- property.1=value-1 property.2=value-2 property.3=value-3 data一栏包括了配置数据，ConfigMap可以被用来保存单个属性，也可以用来保存一个配置文件。 配置数据可以通过很多种方式在Pods里被使用。ConfigMaps可以被用来： 设置环境变量的值 在容器里设置命令行参数 在数据卷里面创建config文件 用户和系统组件两者都可以在ConfigMap里面存储配置数据。 其实不用看下面的文章，直接从kubectl create configmap -h的帮助信息中就可以对ConfigMap究竟如何创建略知一二了。 Examples: # Create a new configmap named my-config based on folder bar kubectl create configmap my-config --from-file=path/to/bar # Create a new configmap named my-config with specified keys instead of file basenames on disk kubectl create configmap my-config --from-file=key1=/path/to/bar/file1.txt --from-file=key2=/path/to/bar/file2.txt # Create a new configmap named my-config with key1=config1 and key2=config2 kubectl create configmap my-config --from-literal=key1=config1 --from-literal=key2=config2 创建ConfigMaps 可以使用该命令，用给定值、文件或目录来创建ConfigMap。 kubectl create configmap 使用目录创建 比如我们已经有个了包含一些配置文件，其中包含了我们想要设置的ConfigMap的值： $ ls docs/user-guide/configmap/kubectl/ game.properties ui.properties $ cat docs/user-guide/configmap/kubectl/game.properties enemies=aliens lives=3 enemies.cheat=true enemies.cheat.level=noGoodRotten secret.code.passphrase=UUDDLRLRBABAS secret.code.allowed=true secret.code.lives=30 $ cat docs/user-guide/configmap/kubectl/ui.properties color.good=purple color.bad=yellow allow.textmode=true how.nice.to.look=fairlyNice 使用下面的命令可以创建一个包含目录中所有文件的ConfigMap。 $ kubectl create configmap game-config --from-file=docs/user-guide/configmap/kubectl —from-file指定在目录下的所有文件都会被用在ConfigMap里面创建一个键值对，键的名字就是文件名，值就是文件的内容。 让我们来看一下这个命令创建的ConfigMap： $ kubectl describe configmaps game-config Name: game-config Namespace: default Labels: Annotations: Data ==== game.properties: 158 bytes ui.properties: 83 bytes 我们可以看到那两个key是从kubectl指定的目录中的文件名。这些key的内容可能会很大，所以在kubectl describe的输出中，只能够看到键的名字和他们的大小。 如果想要看到键的值的话，可以使用kubectl get： $ kubectl get configmaps game-config -o yaml 我们以yaml格式输出配置。 apiVersion: v1 data: game.properties: | enemies=aliens lives=3 enemies.cheat=true enemies.cheat.level=noGoodRotten secret.code.passphrase=UUDDLRLRBABAS secret.code.allowed=true secret.code.lives=30 ui.properties: | color.good=purple color.bad=yellow allow.textmode=true how.nice.to.look=fairlyNice kind: ConfigMap metadata: creationTimestamp: 2016-02-18T18:34:05Z name: game-config namespace: default resourceVersion: \"407\" selfLink: /api/v1/namespaces/default/configmaps/game-config uid: 30944725-d66e-11e5-8cd0-68f728db1985 使用文件创建 刚才使用目录创建的时候我们—from-file指定的是一个目录，只要指定为一个文件就可以从单个文件中创建ConfigMap。 $ kubectl create configmap game-config-2 --from-file=docs/user-guide/configmap/kubectl/game.properties $ kubectl get configmaps game-config-2 -o yaml apiVersion: v1 data: game-special-key: | enemies=aliens lives=3 enemies.cheat=true enemies.cheat.level=noGoodRotten secret.code.passphrase=UUDDLRLRBABAS secret.code.allowed=true secret.code.lives=30 kind: ConfigMap metadata: creationTimestamp: 2016-02-18T18:54:22Z name: game-config-3 namespace: default resourceVersion: \"530\" selfLink: /api/v1/namespaces/default/configmaps/game-config-3 uid: 05f8da22-d671-11e5-8cd0-68f728db1985 —from-file这个参数可以使用多次，你可以使用两次分别指定上个实例中的那两个配置文件，效果就跟指定整个目录是一样的。 使用字面值创建 使用文字值创建，利用—from-literal参数传递配置信息，该参数可以使用多次，格式如下； $ kubectl create configmap special-config --from-literal=special.how=very --from-literal=special.type=charm $ kubectl get configmaps special-config -o yaml apiVersion: v1 data: special.how: very special.type: charm kind: ConfigMap metadata: creationTimestamp: 2016-02-18T19:14:38Z name: special-config namespace: default resourceVersion: \"651\" selfLink: /api/v1/namespaces/default/configmaps/special-config uid: dadce046-d673-11e5-8cd0-68f728db1985 Pod中使用ConfigMap 使用ConfigMap来替代环境变量 ConfigMap可以被用来填入环境变量。看下下面的ConfigMap。 apiVersion: v1 kind: ConfigMap metadata: name: special-config namespace: default data: special.how: very special.type: charm apiVersion: v1 kind: ConfigMap metadata: name: env-config namespace: default data: log_level: INFO 我们可以在Pod中这样使用ConfigMap： apiVersion: v1 kind: Pod metadata: name: dapi-test-pod spec: containers: - name: test-container image: gcr.io/google_containers/busybox command: [ \"/bin/sh\", \"-c\", \"env\" ] env: - name: SPECIAL_LEVEL_KEY valueFrom: configMapKeyRef: name: special-config key: special.how - name: SPECIAL_TYPE_KEY valueFrom: configMapKeyRef: name: special-config key: special.type envFrom: - configMapRef: name: env-config restartPolicy: Never 这个Pod运行后会输出如下几行： SPECIAL_LEVEL_KEY=very SPECIAL_TYPE_KEY=charm log_level=INFO 用ConfigMap设置命令行参数 ConfigMap也可以被使用来设置容器中的命令或者参数值。它使用的是Kubernetes的$(VAR_NAME)替换语法。我们看下下面这个ConfigMap。 apiVersion: v1 kind: ConfigMap metadata: name: special-config namespace: default data: special.how: very special.type: charm 为了将ConfigMap中的值注入到命令行的参数里面，我们还要像前面那个例子一样使用环境变量替换语法${VAR_NAME)。（其实这个东西就是给Docker容器设置环境变量，以前我创建镜像的时候经常这么玩，通过docker run的时候指定-e参数修改镜像里的环境变量，然后docker的CMD命令再利用该$(VAR_NAME)通过sed来修改配置文件或者作为命令行启动参数。） apiVersion: v1 kind: Pod metadata: name: dapi-test-pod spec: containers: - name: test-container image: gcr.io/google_containers/busybox command: [ \"/bin/sh\", \"-c\", \"echo $(SPECIAL_LEVEL_KEY) $(SPECIAL_TYPE_KEY)\" ] env: - name: SPECIAL_LEVEL_KEY valueFrom: configMapKeyRef: name: special-config key: special.how - name: SPECIAL_TYPE_KEY valueFrom: configMapKeyRef: name: special-config key: special.type restartPolicy: Never 运行这个Pod后会输出： very charm 通过数据卷插件使用ConfigMap ConfigMap也可以在数据卷里面被使用。还是这个ConfigMap。 apiVersion: v1 kind: ConfigMap metadata: name: special-config namespace: default data: special.how: very special.type: charm 在数据卷里面使用这个ConfigMap，有不同的选项。最基本的就是将文件填入数据卷，在这个文件中，键就是文件名，键值就是文件内容： apiVersion: v1 kind: Pod metadata: name: dapi-test-pod spec: containers: - name: test-container image: gcr.io/google_containers/busybox command: [ \"/bin/sh\", \"-c\", \"cat /etc/config/special.how\" ] volumeMounts: - name: config-volume mountPath: /etc/config volumes: - name: config-volume configMap: name: special-config restartPolicy: Never 运行这个Pod的输出是very。 我们也可以在ConfigMap值被映射的数据卷里控制路径。 apiVersion: v1 kind: Pod metadata: name: dapi-test-pod spec: containers: - name: test-container image: gcr.io/google_containers/busybox command: [ \"/bin/sh\",\"-c\",\"cat /etc/config/path/to/special-key\" ] volumeMounts: - name: config-volume mountPath: /etc/config volumes: - name: config-volume configMap: name: special-config items: - key: special.how path: path/to/special-key restartPolicy: Never 运行这个Pod后的结果是very。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-08 16:03:37 "},"concepts/configmap-hot-update.html":{"url":"concepts/configmap-hot-update.html","title":"ConfigMap的热更新","keywords":"","body":"ConfigMap的热更新 ConfigMap是用来存储配置文件的kubernetes资源对象，所有的配置内容都存储在etcd中，下文主要是探究 ConfigMap 的创建和更新流程，以及对 ConfigMap 更新后容器内挂载的内容是否同步更新的测试。 测试示例 假设我们在 default namespace 下有一个名为 nginx-config 的 ConfigMap，可以使用 kubectl命令来获取： $ kubectl get configmap nginx-config NAME DATA AGE nginx-config 1 99d 获取该ConfigMap的内容。 kubectl get configmap nginx-config -o yaml apiVersion: v1 data: nginx.conf: |- worker_processes 1; events { worker_connections 1024; } http { sendfile on; server { listen 80; # a test endpoint that returns http 200s location / { proxy_pass http://httpstat.us/200; proxy_set_header X-Real-IP $remote_addr; } } server { listen 80; server_name api.hello.world; location / { proxy_pass http://l5d.default.svc.cluster.local; proxy_set_header Host $host; proxy_set_header Connection \"\"; proxy_http_version 1.1; more_clear_input_headers 'l5d-ctx-*' 'l5d-dtab' 'l5d-sample'; } } server { listen 80; server_name www.hello.world; location / { # allow 'employees' to perform dtab overrides if ($cookie_special_employee_cookie != \"letmein\") { more_clear_input_headers 'l5d-ctx-*' 'l5d-dtab' 'l5d-sample'; } # add a dtab override to get people to our beta, world-v2 set $xheader \"\"; if ($cookie_special_employee_cookie ~* \"dogfood\") { set $xheader \"/host/world => /srv/world-v2;\"; } proxy_set_header 'l5d-dtab' $xheader; proxy_pass http://l5d.default.svc.cluster.local; proxy_set_header Host $host; proxy_set_header Connection \"\"; proxy_http_version 1.1; } } } kind: ConfigMap metadata: creationTimestamp: 2017-08-01T06:53:17Z name: nginx-config namespace: default resourceVersion: \"14925806\" selfLink: /api/v1/namespaces/default/configmaps/nginx-config uid: 18d70527-7686-11e7-bfbd-8af1e3a7c5bd ConfigMap中的内容是存储到etcd中的，然后查询etcd： ETCDCTL_API=3 etcdctl get /registry/configmaps/default/nginx-config -w json|python -m json.tool 注意使用 v3 版本的 etcdctl API，下面是输出结果： { \"count\": 1, \"header\": { \"cluster_id\": 12091028579527406772, \"member_id\": 16557816780141026208, \"raft_term\": 36, \"revision\": 29258723 }, \"kvs\": [ { \"create_revision\": 14925806, \"key\": \"L3JlZ2lzdHJ5L2NvbmZpZ21hcHMvZGVmYXVsdC9uZ2lueC1jb25maWc=\", \"mod_revision\": 14925806, \"value\": \"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\", \"version\": 1 } ] } 其中的value就是 nginx.conf 配置文件的内容。 可以使用base64解码查看具体值，关于etcdctl的使用请参考使用etcdctl访问kuberentes数据。 代码 ConfigMap 结构体的定义： // ConfigMap holds configuration data for pods to consume. type ConfigMap struct { metav1.TypeMeta `json:\",inline\"` // Standard object's metadata. // More info: http://releases.k8s.io/HEAD/docs/devel/api-conventions.md#metadata // +optional metav1.ObjectMeta `json:\"metadata,omitempty\" protobuf:\"bytes,1,opt,name=metadata\"` // Data contains the configuration data. // Each key must be a valid DNS_SUBDOMAIN with an optional leading dot. // +optional Data map[string]string `json:\"data,omitempty\" protobuf:\"bytes,2,rep,name=data\"` } 在 staging/src/k8s.io/client-go/kubernetes/typed/core/v1/configmap.go 中ConfigMap 的接口定义： // ConfigMapInterface has methods to work with ConfigMap resources. type ConfigMapInterface interface { Create(*v1.ConfigMap) (*v1.ConfigMap, error) Update(*v1.ConfigMap) (*v1.ConfigMap, error) Delete(name string, options *meta_v1.DeleteOptions) error DeleteCollection(options *meta_v1.DeleteOptions, listOptions meta_v1.ListOptions) error Get(name string, options meta_v1.GetOptions) (*v1.ConfigMap, error) List(opts meta_v1.ListOptions) (*v1.ConfigMapList, error) Watch(opts meta_v1.ListOptions) (watch.Interface, error) Patch(name string, pt types.PatchType, data []byte, subresources ...string) (result *v1.ConfigMap, err error) ConfigMapExpansion } 在 staging/src/k8s.io/client-go/kubernetes/typed/core/v1/configmap.go 中创建 ConfigMap 的方法如下: // Create takes the representation of a configMap and creates it. Returns the server's representation of the configMap, and an error, if there is any. func (c *configMaps) Create(configMap *v1.ConfigMap) (result *v1.ConfigMap, err error) { result = &v1.ConfigMap{} err = c.client.Post(). Namespace(c.ns). Resource(\"configmaps\"). Body(configMap). Do(). Into(result) return } 通过 RESTful 请求在 etcd 中存储 ConfigMap 的配置，该方法中设置了资源对象的 namespace 和 HTTP 请求中的 body，执行后将请求结果保存到 result 中返回给调用者。 注意 Body 的结构 // Body makes the request use obj as the body. Optional. // If obj is a string, try to read a file of that name. // If obj is a []byte, send it directly. // If obj is an io.Reader, use it directly. // If obj is a runtime.Object, marshal it correctly, and set Content-Type header. // If obj is a runtime.Object and nil, do nothing. // Otherwise, set an error. 创建 ConfigMap RESTful 请求中的的 Body 中包含 ObjectMeta 和 namespace。 HTTP 请求中的结构体： // Request allows for building up a request to a server in a chained fashion. // Any errors are stored until the end of your call, so you only have to // check once. type Request struct { // required client HTTPClient verb string baseURL *url.URL content ContentConfig serializers Serializers // generic components accessible via method setters pathPrefix string subpath string params url.Values headers http.Header // structural elements of the request that are part of the Kubernetes API conventions namespace string namespaceSet bool resource string resourceName string subresource string timeout time.Duration // output err error body io.Reader // This is only used for per-request timeouts, deadlines, and cancellations. ctx context.Context backoffMgr BackoffManager throttle flowcontrol.RateLimiter } 测试 分别测试使用 ConfigMap 挂载 Env 和 Volume 的情况。 更新使用ConfigMap挂载的Env 使用下面的配置创建 nginx 容器测试更新 ConfigMap 后容器内的环境变量是否也跟着更新。 apiVersion: extensions/v1beta1 kind: Deployment metadata: name: my-nginx spec: replicas: 1 template: metadata: labels: run: my-nginx spec: containers: - name: my-nginx image: sz-pg-oam-docker-hub-001.tendcloud.com/library/nginx:1.9 ports: - containerPort: 80 envFrom: - configMapRef: name: env-config --- apiVersion: v1 kind: ConfigMap metadata: name: env-config namespace: default data: log_level: INFO 获取环境变量的值 $ kubectl exec `kubectl get pods -l run=my-nginx -o=name|cut -d \"/\" -f2` env|grep log_level log_level=INFO 修改 ConfigMap $ kubectl edit configmap env-config 修改 log_level 的值为 DEBUG。 再次查看环境变量的值。 $ kubectl exec `kubectl get pods -l run=my-nginx -o=name|cut -d \"/\" -f2` env|grep log_level log_level=INFO 实践证明修改 ConfigMap 无法更新容器中已注入的环境变量信息。 更新使用ConfigMap挂载的Volume 使用下面的配置创建 nginx 容器测试更新 ConfigMap 后容器内挂载的文件是否也跟着更新。 apiVersion: extensions/v1beta1 kind: Deployment metadata: name: my-nginx spec: replicas: 1 template: metadata: labels: run: my-nginx spec: containers: - name: my-nginx image: sz-pg-oam-docker-hub-001.tendcloud.com/library/nginx:1.9 ports: - containerPort: 80 volumeMounts: - name: config-volume mountPath: /etc/config volumes: - name: config-volume configMap: name: special-config --- apiVersion: v1 kind: ConfigMap metadata: name: special-config namespace: default data: log_level: INFO $ kubectl exec `kubectl get pods -l run=my-nginx -o=name|cut -d \"/\" -f2` cat /tmp/log_level INFO 修改 ConfigMap $ kubectl edit configmap special-config 修改 log_level 的值为 DEBUG。 等待大概10秒钟时间，再次查看环境变量的值。 $ kubectl exec `kubectl get pods -l run=my-nginx -o=name|cut -d \"/\" -f2` cat /tmp/log_level DEBUG 我们可以看到使用 ConfigMap 方式挂载的 Volume 的文件中的内容已经变成了 DEBUG。 总结 更新 ConfigMap 后： 使用该 ConfigMap 挂载的 Env 不会同步更新 使用该 ConfigMap 挂载的 Volume 中的数据需要一段时间（实测大概10秒）才能同步更新 ENV 是在容器启动的时候注入的，启动之后 kubernetes 就不会再改变环境变量的值，且同一个 namespace 中的 pod 的环境变量是不断累加的，参考 Kubernetes中的服务发现与docker容器间的环境变量传递源码探究。为了更新容器中使用 ConfigMap 挂载的配置，可以通过滚动更新 pod 的方式来强制重新挂载 ConfigMap，也可以在更新了 ConfigMap 后，先将副本数设置为 0，然后再扩容。 参考 Kubernetes 1.7 security in practice ConfigMap | kubernetes handbook - jimmysong.io 创建高可用ectd集群 | Kubernetes handbook - jimmysong.io Kubernetes中的服务发现与docker容器间的环境变量传递源码探究 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-23 13:48:42 "},"concepts/horizontal-pod-autoscaling.html":{"url":"concepts/horizontal-pod-autoscaling.html","title":"Horizontal Pod Autoscaling","keywords":"","body":"Horizontal Pod Autoscaling 应用的资源使用率通常都有高峰和低谷的时候，如何削峰填谷，提高集群的整体资源利用率，让service中的Pod个数自动调整呢？这就有赖于Horizontal Pod Autoscaling了，顾名思义，使Pod水平自动缩放。这个Object（跟Pod、Deployment一样都是API resource）也是最能体现kubernetes之于传统运维价值的地方，不再需要手动扩容了，终于实现自动化了，还可以自定义指标，没准未来还可以通过人工智能自动进化呢！ HPA属于Kubernetes中的autoscaling SIG（Special Interest Group），其下有两个feature： Arbitrary/Custom Metrics in the Horizontal Pod Autoscaler#117 Monitoring Pipeline Metrics HPA API #118 Kubernetes自1.2版本引入HPA机制，到1.6版本之前一直是通过kubelet来获取监控指标来判断是否需要扩缩容，1.6版本之后必须通过API server、Heapseter或者kube-aggregator来获取监控指标。 对于1.6以前版本中开启自定义HPA请参考Kubernetes autoscaling based on custom metrics without using a host port，对于1.7及以上版本请参考Configure Kubernetes Autoscaling With Custom Metrics in Kubernetes 1.7 - Bitnami。 HPA解析 Horizontal Pod Autoscaling仅适用于Deployment和ReplicationController（ReplicaSet已经被ReplicationController取代），在V1版本中仅支持根据Pod的CPU利用率扩所容，在v1alpha版本中，支持根据内存和用户自定义的metric扩缩容。 如果你不想看下面的文章可以直接看下面的示例图，组件交互、组件的配置、命令示例，都画在图上了。 Horizontal Pod Autoscaling由API server和controller共同实现。 图片 - horizontal-pod-autoscaler Metrics支持 在不同版本的API中，HPA autoscale时可以根据以下指标来判断： autoscaling/v1 CPU autoscaling/v2alpha1 内存 自定义metrics kubernetes1.6起支持自定义metrics，但是必须在kube-controller-manager中配置如下两项： --horizontal-pod-autoscaler-use-rest-clients=true --api-server指向kube-aggregator，也可以使用heapster来实现，通过在启动heapster的时候指定--api-server=true。查看kubernetes metrics 多种metrics组合 HPA会根据每个metric的值计算出scale的值，并将最大的那个指作为扩容的最终结果。 使用kubectl管理 Horizontal Pod Autoscaling作为API resource也可以像Pod、Deployment一样使用kubeclt命令管理，使用方法跟它们一样，资源名称为hpa。 kubectl create hpa kubectl get hpa kubectl describe hpa kubectl delete hpa 有一点不同的是，可以直接使用kubectl autoscale直接通过命令行的方式创建Horizontal Pod Autoscaler。 用法如下： kubectl autoscale (-f FILENAME | TYPE NAME | TYPE/NAME) [--min=MINPODS] --max=MAXPODS [--cpu-percent=CPU] [flags] [options] 举个例子： kubectl autoscale deployment foo --min=2 --max=5 --cpu-percent=80 为Deployment foo创建 一个autoscaler，当Pod的CPU利用率达到80%的时候，RC的replica数在2到5之间。该命令的详细使用文档见https://kubernetes.io/docs/user-guide/kubectl/v1.6/#autoscale 。 注意 ：如果为ReplicationController创建HPA的话，无法使用rolling update，但是对于Deployment来说是可以的，因为Deployment在执行rolling update的时候会自动创建新的ReplicationController。 什么是 Horizontal Pod Autoscaling？ 利用 Horizontal Pod Autoscaling，kubernetes 能够根据监测到的 CPU 利用率（或者在 alpha 版本中支持的应用提供的 metric）自动的扩容 replication controller，deployment 和 replica set。 Horizontal Pod Autoscaler 作为 kubernetes API resource 和 controller 的实现。Resource 确定 controller 的行为。Controller 会根据监测到用户指定的目标的 CPU 利用率周期性得调整 replication controller 或 deployment 的 replica 数量。 Horizontal Pod Autoscaler 如何工作？ Horizontal Pod Autoscaler 由一个控制循环实现，循环周期由 controller manager 中的 --horizontal-pod-autoscaler-sync-period 标志指定（默认是 30 秒）。 在每个周期内，controller manager 会查询 HorizontalPodAutoscaler 中定义的 metric 的资源利用率。Controller manager 从 resource metric API（每个 pod 的 resource metric）或者自定义 metric API（所有的metric）中获取 metric。 每个 Pod 的 resource metric（例如 CPU），controller 通过 resource metric API 获取 HorizontalPodAutoscaler 中定义的每个 Pod 中的 metric。然后，如果设置了目标利用率，controller 计算利用的值与每个 Pod 的容器里的 resource request 值的百分比。如果设置了目标原始值，将直接使用该原始 metric 值。然后 controller 计算所有目标 Pod 的利用率或原始值（取决于所指定的目标类型）的平均值，产生一个用于缩放所需 replica 数量的比率。 请注意，如果某些 Pod 的容器没有设置相关的 resource request ，则不会定义 Pod 的 CPU 利用率，并且 Aucoscaler 也不会对该 metric 采取任何操作。 有关自动缩放算法如何工作的更多细节，请参阅 自动缩放算法设计文档。 对于每个 Pod 自定义的 metric，controller 功能类似于每个 Pod 的 resource metric，只是它使用原始值而不是利用率值。 对于 object metric，获取单个度量（描述有问题的对象），并与目标值进行比较，以产生如上所述的比率。 HorizontalPodAutoscaler 控制器可以以两种不同的方式获取 metric ：直接的 Heapster 访问和 REST 客户端访问。 当使用直接的 Heapster 访问时，HorizontalPodAutoscaler 直接通过 API 服务器的服务代理子资源查询 Heapster。需要在集群上部署 Heapster 并在 kube-system namespace 中运行。 有关 REST 客户端访问的详细信息，请参阅 支持自定义度量。 Autoscaler 访问相应的 replication controller，deployment 或 replica set 来缩放子资源。 Scale 是一个允许您动态设置副本数并检查其当前状态的接口。 有关缩放子资源的更多细节可以在 这里 找到。 API Object Horizontal Pod Autoscaler 是 kubernetes 的 autoscaling API 组中的 API 资源。当前的稳定版本中，只支持 CPU 自动扩缩容，可以在autoscaling/v1 API 版本中找到。 在 alpha 版本中支持根据内存和自定义 metric 扩缩容，可以在autoscaling/v2alpha1 中找到。autoscaling/v2alpha1 中引入的新字段在autoscaling/v1 中是做为 annotation 而保存的。 关于该 API 对象的更多信息，请参阅 HorizontalPodAutoscaler Object。 在 kubectl 中支持 Horizontal Pod Autoscaling Horizontal Pod Autoscaler 和其他的所有 API 资源一样，通过 kubectl 以标准的方式支持。 我们可以使用kubectl create命令创建一个新的 autoscaler。 我们可以使用kubectl get hpa列出所有的 autoscaler，使用kubectl describe hpa获取其详细信息。 最后我们可以使用kubectl delete hpa删除 autoscaler。 另外，可以使用kubectl autoscale命令，很轻易的就可以创建一个 Horizontal Pod Autoscaler。 例如，执行kubectl autoscale rc foo —min=2 —max=5 —cpu-percent=80命令将为 replication controller foo 创建一个 autoscaler，目标的 CPU 利用率是80%，replica 的数量介于 2 和 5 之间。 关于kubectl autoscale的更多信息请参阅 这里。 滚动更新期间的自动扩缩容 目前在Kubernetes中，可以通过直接管理 replication controller 或使用 deployment 对象来执行 滚动更新，该 deployment 对象为您管理基础 replication controller。 Horizontal Pod Autoscaler 仅支持后一种方法：Horizontal Pod Autoscaler 被绑定到 deployment 对象，它设置 deployment 对象的大小，deployment 负责设置底层 replication controller 的大小。 Horizontal Pod Autoscaler 不能使用直接操作 replication controller 进行滚动更新，即不能将 Horizontal Pod Autoscaler 绑定到 replication controller，并进行滚动更新（例如使用kubectl rolling-update）。 这不行的原因是，当滚动更新创建一个新的 replication controller 时，Horizontal Pod Autoscaler 将不会绑定到新的 replication controller 上。 支持多个 metric Kubernetes 1.6 中增加了支持基于多个 metric 的扩缩容。您可以使用autoscaling/v2alpha1 API 版本来为 Horizontal Pod Autoscaler 指定多个 metric。然后 Horizontal Pod Autoscaler controller 将权衡每一个 metric，并根据该 metric 提议一个新的 scale。在所有提议里最大的那个 scale 将作为最终的 scale。 支持自定义 metric 注意： Kubernetes 1.2 根据特定于应用程序的 metric ，通过使用特殊注释的方式，增加了对缩放的 alpha 支持。 在 Kubernetes 1.6中删除了对这些注释的支持，有利于autoscaling/v2alpha1 API。 虽然旧的收集自定义 metric 的旧方法仍然可用，但是这些 metric 将不可供 Horizontal Pod Autoscaler 使用，并且用于指定要缩放的自定义 metric 的以前的注释也不在受 Horizontal Pod Autoscaler 认可。 Kubernetes 1.6增加了在 Horizontal Pod Autoscale r中使用自定义 metric 的支持。 您可以为autoscaling/v2alpha1 API 中使用的 Horizontal Pod Autoscaler 添加自定义 metric 。 Kubernetes 然后查询新的自定义 metric API 来获取相应自定义 metric 的值。 前提条件 为了在 Horizontal Pod Autoscaler 中使用自定义 metric，您必须在您集群的 controller manager 中将 --horizontal-pod-autoscaler-use-rest-clients 标志设置为 true。然后，您必须通过将 controller manager 的目标 API server 设置为 API server aggregator（使用--apiserver标志），配置您的 controller manager 通过 API server aggregator 与API server 通信。 Resource metric API和自定义 metric API 也必须向 API server aggregator 注册，并且必须由集群上运行的 API server 提供。 您可以使用 Heapster 实现 resource metric API，方法是将 --api-server 标志设置为 true 并运行 Heapster。 单独的组件必须提供自定义 metric API（有关自定义metric API的更多信息，可从 k8s.io/metrics repository 获得）。 进一步阅读 设计文档：Horizontal Pod Autoscaling kubectl autoscale 命令：kubectl autoscale 使用例子： Horizontal Pod Autoscaler 参考 HPA设计文档：https://github.com/kubernetes/community/blob/master/contributors/design-proposals/horizontal-pod-autoscaler.md HPA说明：https://kubernetes.io/docs/tasks/run-application/horizontal-pod-autoscale/ HPA详解：https://kubernetes.io/docs/tasks/run-application/horizontal-pod-autoscale-walkthrough/ kubectl autoscale命令详细使用说明：https://kubernetes.io/docs/user-guide/kubectl/v1.6/#autoscale 自定义metrics开发：https://github.com/kubernetes/metrics 1.7版本的kubernetes中启用自定义HPA：Configure Kubernetes Autoscaling With Custom Metrics in Kubernetes 1.7 - Bitnami Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-27 13:12:02 "},"concepts/custom-metrics-hpa.html":{"url":"concepts/custom-metrics-hpa.html","title":"自定义指标HPA","keywords":"","body":"自定义指标HPA Kubernetes中不仅支持CPU、内存为指标的HPA，还支持自定义指标的HPA，例如QPS。 本文中使用的yaml文件见manifests/HPA。 设置自定义指标 kubernetes1.6 在kubernetes1.6集群中配置自定义指标的HPA的说明已废弃。 在设置定义指标HPA之前需要先进行如下配置： 将heapster的启动参数 --api-server 设置为 true 启用custom metric API 将kube-controller-manager的启动参数中--horizontal-pod-autoscaler-use-rest-clients设置为true，并指定--master为API server地址，如--master=http://172.20.0.113:8080 在kubernetes1.5以前很容易设置，参考1.6以前版本的kubernetes中开启自定义HPA，而在1.6中因为取消了原来的annotation方式设置custom metric，只能通过API server和kube-aggregator来获取custom metric，因为只有两种方式来设置了，一是直接通过API server获取heapster的metrics，二是部署kube-aggragator来实现。 我们将在kubernetes1.8版本的kubernetes中，使用聚合的API server来实现自定义指标的HPA。 kuberentes1.7+ 确认您的kubernetes版本在1.7或以上，修改以下配置： 将kube-controller-manager的启动参数中--horizontal-pod-autoscaler-use-rest-clients设置为true，并指定--master为API server地址，如`--master=http://172.20.0.113:8080 修改kube-apiserver的配置文件apiserver，增加一条配置--requestheader-client-ca-file=/etc/kubernetes/ssl/ca.pem --requestheader-allowed-names=aggregator --requestheader-extra-headers-prefix=X-Remote-Extra- --requestheader-group-headers=X-Remote-Group --requestheader-username-headers=X-Remote-User --proxy-client-cert-file=/etc/kubernetes/ssl/kubernetes.pem --proxy-client-key-file=/etc/kubernetes/ssl/kubernetes-key.pem，用来配置aggregator的CA证书。 已经内置了apiregistration.k8s.io/v1beta1 API，可以直接定义APIService，如： apiVersion: apiregistration.k8s.io/v1 kind: APIService metadata: name: v1.custom-metrics.metrics.k8s.io spec: insecureSkipTLSVerify: true group: custom-metrics.metrics.k8s.io groupPriorityMinimum: 1000 versionPriority: 5 service: name: api namespace: custom-metrics version: v1alpha1 部署Prometheus 使用prometheus-operator.yaml文件部署Prometheus operator。 注意：将镜像修改为你自己的镜像仓库地址。 这产生一个自定义的API：http://172.20.0.113:8080/apis/custom-metrics.metrics.k8s.io/v1alpha1，可以通过浏览器访问，还可以使用下面的命令可以检查该API： $ kubectl get --raw=apis/custom-metrics.metrics.k8s.io/v1alpha1 {\"kind\":\"APIResourceList\",\"apiVersion\":\"v1\",\"groupVersion\":\"custom-metrics.metrics.k8s.io/v1alpha1\",\"resources\":[{\"name\":\"jobs.batch/http_requests\",\"singularName\":\"\",\"namespaced\":true,\"kind\":\"MetricValueList\",\"verbs\":[\"get\"]},{\"name\":\"namespaces/http_requests\",\"singularName\":\"\",\"namespaced\":false,\"kind\":\"MetricValueList\",\"verbs\":[\"get\"]},{\"name\":\"jobs.batch/up\",\"singularName\":\"\",\"namespaced\":true,\"kind\":\"MetricValueList\",\"verbs\":[\"get\"]},{\"name\":\"pods/up\",\"singularName\":\"\",\"namespaced\":true,\"kind\":\"MetricValueList\",\"verbs\":[\"get\"]},{\"name\":\"services/scrape_samples_scraped\",\"singularName\":\"\",\"namespaced\":true,\"kind\":\"MetricValueList\",\"verbs\":[\"get\"]},{\"name\":\"namespaces/scrape_samples_scraped\",\"singularName\":\"\",\"namespaced\":false,\"kind\":\"MetricValueList\",\"verbs\":[\"get\"]},{\"name\":\"pods/scrape_duration_seconds\",\"singularName\":\"\",\"namespaced\":true,\"kind\":\"MetricValueList\",\"verbs\":[\"get\"]},{\"name\":\"services/scrape_duration_seconds\",\"singularName\":\"\",\"namespaced\":true,\"kind\":\"MetricValueList\",\"verbs\":[\"get\"]},{\"name\":\"pods/http_requests\",\"singularName\":\"\",\"namespaced\":true,\"kind\":\"MetricValueList\",\"verbs\":[\"get\"]},{\"name\":\"pods/scrape_samples_post_metric_relabeling\",\"singularName\":\"\",\"namespaced\":true,\"kind\":\"MetricValueList\",\"verbs\":[\"get\"]},{\"name\":\"jobs.batch/scrape_samples_scraped\",\"singularName\":\"\",\"namespaced\":true,\"kind\":\"MetricValueList\",\"verbs\":[\"get\"]},{\"name\":\"jobs.batch/scrape_duration_seconds\",\"singularName\":\"\",\"namespaced\":true,\"kind\":\"MetricValueList\",\"verbs\":[\"get\"]},{\"name\":\"namespaces/scrape_duration_seconds\",\"singularName\":\"\",\"namespaced\":false,\"kind\":\"MetricValueList\",\"verbs\":[\"get\"]},{\"name\":\"namespaces/scrape_samples_post_metric_relabeling\",\"singularName\":\"\",\"namespaced\":false,\"kind\":\"MetricValueList\",\"verbs\":[\"get\"]},{\"name\":\"services/scrape_samples_post_metric_relabeling\",\"singularName\":\"\",\"namespaced\":true,\"kind\":\"MetricValueList\",\"verbs\":[\"get\"]},{\"name\":\"services/up\",\"singularName\":\"\",\"namespaced\":true,\"kind\":\"MetricValueList\",\"verbs\":[\"get\"]},{\"name\":\"pods/scrape_samples_scraped\",\"singularName\":\"\",\"namespaced\":true,\"kind\":\"MetricValueList\",\"verbs\":[\"get\"]},{\"name\":\"services/http_requests\",\"singularName\":\"\",\"namespaced\":true,\"kind\":\"MetricValueList\",\"verbs\":[\"get\"]},{\"name\":\"jobs.batch/scrape_samples_post_metric_relabeling\",\"singularName\":\"\",\"namespaced\":true,\"kind\":\"MetricValueList\",\"verbs\":[\"get\"]},{\"name\":\"namespaces/up\",\"singularName\":\"\",\"namespaced\":false,\"kind\":\"MetricValueList\",\"verbs\":[\"get\"]}]} 参考 1.6以前版本的kubernetes中开启自定义HPA 1.7版本的kubernetes中启用自定义HPA Horizontal Pod Autoscaler Walkthrough Kubernetes 1.8: Now with 100% Daily Value of Custom Metrics Arbitrary/Custom Metrics in the Horizontal Pod Autoscaler#117 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-14 19:03:40 "},"concepts/label.html":{"url":"concepts/label.html","title":"Label","keywords":"","body":"Label Label是附着到object上（例如Pod）的键值对。可以在创建object的时候指定，也可以在object创建后随时指定。Labels的值对系统本身并没有什么含义，只是对用户才有意义。 \"labels\": { \"key1\" : \"value1\", \"key2\" : \"value2\" } Kubernetes最终将对labels最终索引和反向索引用来优化查询和watch，在UI和命令行中会对它们排序。不要在label中使用大型、非标识的结构化数据，记录这样的数据应该用annotation。 动机 Label能够将组织架构映射到系统架构上（就像是康威定律），这样能够更便于微服务的管理，你可以给object打上如下类型的label： \"release\" : \"stable\", \"release\" : \"canary\" \"environment\" : \"dev\", \"environment\" : \"qa\", \"environment\" : \"production\" \"tier\" : \"frontend\", \"tier\" : \"backend\", \"tier\" : \"cache\" \"partition\" : \"customerA\", \"partition\" : \"customerB\" \"track\" : \"daily\", \"track\" : \"weekly\" \"team\" : \"teamA\",\"team:\" : \"teamB\" 语法和字符集 Label key的组成： 不得超过63个字符 可以使用前缀，使用/分隔，前缀必须是DNS子域，不得超过253个字符，系统中的自动化组件创建的label必须指定前缀，kubernetes.io/由kubernetes保留 起始必须是字母（大小写都可以）或数字，中间可以有连字符、下划线和点 Label value的组成： 不得超过63个字符 起始必须是字母（大小写都可以）或数字，中间可以有连字符、下划线和点 Label selector Label不是唯一的，很多object可能有相同的label。 通过label selector，客户端／用户可以指定一个object集合，通过label selector对object的集合进行操作。 Label selector有两种类型： equality-based ：可以使用=、==、!=操作符，可以使用逗号分隔多个表达式 set-based ：可以使用in、notin、!操作符，另外还可以没有操作符，直接写出某个label的key，表示过滤有某个key的object而不管该key的value是何值，!表示没有该label的object 示例 $ kubectl get pods -l environment=production,tier=frontend $ kubectl get pods -l 'environment in (production),tier in (frontend)' $ kubectl get pods -l 'environment in (production, qa)' $ kubectl get pods -l 'environment,environment notin (frontend)' 在API object中设置label selector 在service、replicationcontroller等object中有对pod的label selector，使用方法只能使用等于操作，例如： selector: component: redis 在Job、Deployment、ReplicaSet和DaemonSet这些object中，支持set-based的过滤，例如： selector: matchLabels: component: redis matchExpressions: - {key: tier, operator: In, values: [cache]} - {key: environment, operator: NotIn, values: [dev]} 如Service通过label selector将同一类型的pod作为一个服务expose出来。 图片 - label示意图 另外在node affinity和pod affinity中的label selector的语法又有些许不同，示例如下： affinity: nodeAffinity: requiredDuringSchedulingIgnoredDuringExecution: nodeSelectorTerms: - matchExpressions: - key: kubernetes.io/e2e-az-name operator: In values: - e2e-az1 - e2e-az2 preferredDuringSchedulingIgnoredDuringExecution: - weight: 1 preference: matchExpressions: - key: another-node-label-key operator: In values: - another-node-label-value 详见node selection。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-08 16:19:53 "},"concepts/garbage-collection.html":{"url":"concepts/garbage-collection.html","title":"垃圾收集","keywords":"","body":"垃圾收集 Kubernetes 垃圾收集器的角色是删除指定的对象，这些对象曾经有但以后不再拥有 Owner 了。 注意：垃圾收集是 beta 特性，在 Kubernetes 1.4 及以上版本默认启用。 Owner 和 Dependent 一些 Kubernetes 对象是其它一些的 Owner。例如，一个 ReplicaSet 是一组 Pod 的 Owner。具有 Owner 的对象被称为是 Owner 的 Dependent。每个 Dependent 对象具有一个指向其所属对象的 metadata.ownerReferences 字段。 有时，Kubernetes 会自动设置 ownerReference 的值。例如，当创建一个 ReplicaSet 时，Kubernetes 自动设置 ReplicaSet 中每个 Pod 的 ownerReference 字段值。在 1.6 版本，Kubernetes 会自动为一些对象设置 ownerReference 的值，这些对象是由 ReplicationController、ReplicaSet、StatefulSet、DaemonSet 和 Deployment 所创建或管理。 也可以通过手动设置 ownerReference 的值，来指定 Owner 和 Dependent 之间的关系。 这有一个配置文件，表示一个具有 3 个 Pod 的 ReplicaSet： apiVersion: extensions/v1beta1 kind: ReplicaSet metadata: name: my-repset spec: replicas: 3 selector: matchLabels: pod-is-for: garbage-collection-example template: metadata: labels: pod-is-for: garbage-collection-example spec: containers: - name: nginx image: nginx 如果创建该 ReplicaSet，然后查看 Pod 的 metadata 字段，能够看到 OwnerReferences 字段： kubectl create -f https://k8s.io/docs/concepts/abstractions/controllers/my-repset.yaml kubectl get pods --output=yaml 输出显示了 Pod 的 Owner 是名为 my-repset 的 ReplicaSet： apiVersion: v1 kind: Pod metadata: ... ownerReferences: - apiVersion: extensions/v1beta1 controller: true blockOwnerDeletion: true kind: ReplicaSet name: my-repset uid: d9607e19-f88f-11e6-a518-42010a800195 ... 控制垃圾收集器删除 Dependent 当删除对象时，可以指定是否该对象的 Dependent 也自动删除掉。自动删除 Dependent 也称为 级联删除。Kubernetes 中有两种 级联删除 的模式：background 模式和 foreground 模式。 如果删除对象时，不自动删除它的 Dependent，这些 Dependent 被称作是原对象的 孤儿。 Background 级联删除 在 background 级联删除 模式下，Kubernetes 会立即删除 Owner 对象，然后垃圾收集器会在后台删除这些 Dependent。 Foreground 级联删除 在 foreground 级联删除 模式下，根对象首先进入 “删除中” 状态。在 “删除中” 状态会有如下的情况： 对象仍然可以通过 REST API 可见 会设置对象的 deletionTimestamp 字段 对象的 metadata.finalizers 字段包含了值 “foregroundDeletion” 一旦被设置为 “删除中” 状态，垃圾收集器会删除对象的所有 Dependent。垃圾收集器删除了所有 “Blocking” 的 Dependent（对象的 ownerReference.blockOwnerDeletion=true）之后，它会删除 Owner 对象。 注意，在 “foreground 删除” 模式下，Dependent 只有通过 ownerReference.blockOwnerDeletion 才能阻止删除 Owner 对象。在 Kubernetes 1.7 版本中将增加 admission controller，基于 Owner 对象上的删除权限来控制用户去设置 blockOwnerDeletion 的值为 true，所以未授权的 Dependent 不能够延迟 Owner 对象的删除。 如果一个对象的ownerReferences 字段被一个 Controller（例如 Deployment 或 ReplicaSet）设置，blockOwnerDeletion 会被自动设置，没必要手动修改这个字段。 设置级联删除策略 通过为 Owner 对象设置 deleteOptions.propagationPolicy 字段，可以控制级联删除策略。可能的取值包括：“orphan”、“Foreground” 或 “Background”。 对很多 Controller 资源，包括 ReplicationController、ReplicaSet、StatefulSet、DaemonSet 和 Deployment，默认的垃圾收集策略是 orphan。因此，除非指定其它的垃圾收集策略，否则所有 Dependent 对象使用的都是 orphan 策略。 下面是一个在后台删除 Dependent 对象的例子： kubectl proxy --port=8080 curl -X DELETE localhost:8080/apis/extensions/v1beta1/namespaces/default/replicasets/my-repset \\ -d '{\"kind\":\"DeleteOptions\",\"apiVersion\":\"v1\",\"propagationPolicy\":\"Background\"}' \\ -H \"Content-Type: application/json\" 下面是一个在前台删除 Dependent 对象的例子： kubectl proxy --port=8080 curl -X DELETE localhost:8080/apis/extensions/v1beta1/namespaces/default/replicasets/my-repset \\ -d '{\"kind\":\"DeleteOptions\",\"apiVersion\":\"v1\",\"propagationPolicy\":\"Foreground\"}' \\ -H \"Content-Type: application/json\" 下面是一个孤儿 Dependent 的例子： kubectl proxy --port=8080 curl -X DELETE localhost:8080/apis/extensions/v1beta1/namespaces/default/replicasets/my-repset \\ -d '{\"kind\":\"DeleteOptions\",\"apiVersion\":\"v1\",\"propagationPolicy\":\"Orphan\"}' \\ -H \"Content-Type: application/json\" kubectl 也支持级联删除。 通过设置 --cascade 为 true，可以使用 kubectl 自动删除 Dependent 对象。设置 --cascade 为 false，会使 Dependent 对象成为孤儿 Dependent 对象。--cascade 的默认值是 true。 下面是一个例子，使一个 ReplicaSet 的 Dependent 对象成为孤儿 Dependent： kubectl delete replicaset my-repset --cascade=false 已知的问题 1.7 版本，垃圾收集不支持 自定义资源，比如那些通过 CustomResourceDefinition 新增，或者通过 API server 聚集而成的资源对象。 其它已知的问题 原文地址：https://k8smeetup.github.io/docs/concepts/workloads/controllers/garbage-collection/ 译者：shirdrn Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-09-03 13:45:06 "},"concepts/network-policy.html":{"url":"concepts/network-policy.html","title":"NetworkPolicy","keywords":"","body":"Network Policy 网络策略说明一组 Pod 之间是如何被允许互相通信，以及如何与其它网络 Endpoint 进行通信。 NetworkPolicy 资源使用标签来选择 Pod，并定义了一些规则，这些规则指明允许什么流量进入到选中的 Pod 上。 前提条件 网络策略通过网络插件来实现，所以必须使用一种支持 NetworkPolicy 的网络方案 —— 非 Controller 创建的资源，是不起作用的。 隔离的与未隔离的 Pod 默认 Pod 是未隔离的，它们可以从任何的源接收请求。 具有一个可以选择 Pod 的网络策略后，Pod 就会变成隔离的。 一旦 Namespace 中配置的网络策略能够选择一个特定的 Pod，这个 Pod 将拒绝任何该网络策略不允许的连接。（Namespace 中其它未被网络策略选中的 Pod 将继续接收所有流量） NetworkPolicy 资源 查看 API参考 可以获取该资源的完整定义。 下面是一个 NetworkPolicy 的例子： apiVersion: networking.k8s.io/v1 kind: NetworkPolicy metadata: name: test-network-policy namespace: default spec: podSelector: matchLabels: role: db ingress: - from: - namespaceSelector: matchLabels: project: myproject - podSelector: matchLabels: role: frontend ports: - protocol: TCP port: 6379 将上面配置 POST 到 API Server 将不起任何作用，除非选择的网络方案支持网络策略。 必选字段：像所有其它 Kubernetes 配置一样， NetworkPolicy 需要 apiVersion、kind 和 metadata 这三个字段，关于如何使用配置文件的基本信息，可以查看 这里，这里 和 这里。 spec：NetworkPolicy spec 具有在给定 Namespace 中定义特定网络的全部信息。 podSelector：每个 NetworkPolicy 包含一个 podSelector，它可以选择一组应用了网络策略的 Pod。由于 NetworkPolicy 当前只支持定义 ingress 规则，这个 podSelector 实际上为该策略定义了一组 “目标Pod”。示例中的策略选择了标签为 “role=db” 的 Pod。一个空的 podSelector 选择了该 Namespace 中的所有 Pod。 ingress：每个NetworkPolicy 包含了一个白名单 ingress 规则列表。每个规则只允许能够匹配上 from 和 ports配置段的流量。示例策略包含了单个规则，它从这两个源中匹配在单个端口上的流量，第一个是通过namespaceSelector 指定的，第二个是通过 podSelector 指定的。 因此，上面示例的 NetworkPolicy： 在 “default” Namespace中 隔离了标签 “role=db” 的 Pod（如果他们还没有被隔离） 在 “default” Namespace中，允许任何具有 “role=frontend” 的 Pod，连接到标签为 “role=db” 的 Pod 的 TCP 端口 6379 允许在 Namespace 中任何具有标签 “project=myproject” 的 Pod，连接到 “default” Namespace 中标签为 “role=db” 的 Pod 的 TCP 端口 6379 查看 NetworkPolicy 入门指南 给出的更进一步的例子。 默认策略 通过创建一个可以选择所有 Pod 但不允许任何流量的 NetworkPolicy，你可以为一个 Namespace 创建一个 “默认的” 隔离策略，如下所示： apiVersion: networking.k8s.io/v1 kind: NetworkPolicy metadata: name: default-deny spec: podSelector: 这确保了即使是没有被任何 NetworkPolicy 选中的 Pod，将仍然是被隔离的。 可选地，在 Namespace 中，如果你想允许所有的流量进入到所有的 Pod（即使已经添加了某些策略，使一些 Pod 被处理为 “隔离的”），你可以通过创建一个策略来显式地指定允许所有流量： apiVersion: networking.k8s.io/v1 kind: NetworkPolicy metadata: name: allow-all spec: podSelector: ingress: - {} 原文地址：https://k8smeetup.github.io/docs/concepts/services-networking/network-policies/ 译者：shirdrn Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-09-03 14:17:05 "},"concepts/annotation.html":{"url":"concepts/annotation.html","title":"Annotation","keywords":"","body":"Annotation Annotation，顾名思义，就是注解。Annotation可以将Kubernetes资源对象关联到任意的非标识性元数据。使用客户端（如工具和库）可以检索到这些元数据。 关联元数据到对象 Label和Annotation都可以将元数据关联到Kubernetes资源对象。Label主要用于选择对象，可以挑选出满足特定条件的对象。相比之下，annotation 不能用于标识及选择对象。annotation中的元数据可多可少，可以是结构化的或非结构化的，也可以包含label中不允许出现的字符。 annotation和label一样都是key/value键值对映射结构： \"annotations\": { \"key1\" : \"value1\", \"key2\" : \"value2\" } 以下列出了一些可以记录在 annotation 中的对象信息： 声明配置层管理的字段。使用annotation关联这类字段可以用于区分以下几种配置来源：客户端或服务器设置的默认值，自动生成的字段或自动生成的 auto-scaling 和 auto-sizing 系统配置的字段。 创建信息、版本信息或镜像信息。例如时间戳、版本号、git分支、PR序号、镜像哈希值以及仓库地址。 记录日志、监控、分析或审计存储仓库的指针 可以用于debug的客户端（库或工具）信息，例如名称、版本和创建信息。 用户信息，以及工具或系统来源信息、例如来自非Kubernetes生态的相关对象的URL信息。 轻量级部署工具元数据，例如配置或检查点。 负责人的电话或联系方式，或能找到相关信息的目录条目信息，例如团队网站。 如果不使用annotation，您也可以将以上类型的信息存放在外部数据库或目录中，但这样做不利于创建用于部署、管理、内部检查的共享工具和客户端库。 示例 如 Istio 的 Deployment 配置中就使用到了 annotation： apiVersion: extensions/v1beta1 kind: Deployment metadata: name: istio-manager spec: replicas: 1 template: metadata: annotations: alpha.istio.io/sidecar: ignore labels: istio: manager spec: serviceAccountName: istio-manager-service-account containers: - name: discovery image: sz-pg-oam-docker-hub-001.tendcloud.com/library/manager:0.1.5 imagePullPolicy: Always args: [\"discovery\", \"-v\", \"2\"] ports: - containerPort: 8080 env: - name: POD_NAMESPACE valueFrom: fieldRef: apiVersion: v1 fieldPath: metadata.namespace - name: apiserver image: sz-pg-oam-docker-hub-001.tendcloud.com/library/manager:0.1.5 imagePullPolicy: Always args: [\"apiserver\", \"-v\", \"2\"] ports: - containerPort: 8081 env: - name: POD_NAMESPACE valueFrom: fieldRef: apiVersion: v1 fieldPath: metadata.namespace alpha.istio.io/sidecar 注解就是用来控制是否自动向 pod 中注入 sidecar 的。参考：安装 Istio sidecar - istio.doczh.cn。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-09 16:42:47 "},"concepts/aggregated-api-server.html":{"url":"concepts/aggregated-api-server.html","title":"Aggregated API Server","keywords":"","body":"Aggregated API Server Aggregated（聚合的）API server是为了将原来的API server这个巨石（monolithic）应用给拆分成，为了方便用户开发自己的API server集成进来，而不用直接修改kubernetes官方仓库的代码，这样一来也能将API server解耦，方便用户使用实验特性。这些API server可以跟core API server无缝衔接，使用kubectl也可以管理它们。 架构 我们需要创建一个新的组件，名为kube-aggregator，它需要负责以下几件事： 提供用于注册API server的API 汇总所有的API server信息 代理所有的客户端到API server的请求 注意：这里说的API server是一组“API Server”，而不是说我们安装集群时候的那个API server，而且这组API server是可以横向扩展的。 关于聚合的API server的更多信息请参考：Aggregated API Server 安装配置聚合的API server 有两种方式来启用kube-aggregator： 使用test mode/single-user mode，作为一个独立的进程来运行 使用gateway mode，kube-apiserver将嵌入到kbe-aggregator组件中，它将作为一个集群的gateway，用来聚合所有apiserver。 kube-aggregator二进制文件已经包含在kubernetes release里面了。 参考 Aggregated API Servers - kuberentes design-proposals Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-14 18:52:36 "},"concepts/custom-resource.html":{"url":"concepts/custom-resource.html","title":"使用自定义资源扩展API","keywords":"","body":"使用自定义资源扩展API 注意：TPR已经停止维护，kubernetes 1.7及以上版本请使用CRD。 自定义资源是对Kubernetes API的扩展，kubernetes中的每个资源都是一个API对象的集合，例如我们在YAML文件里定义的那些spec都是对kubernetes中的资源对象的定义，所有的自定义资源可以跟kubernetes中内建的资源一样使用kubectl操作。 自定义资源 Kubernetes1.6版本中包含一个内建的资源叫做TPR（ThirdPartyResource），可以用它来创建自定义资源，但该资源在kubernetes1.7中版本已被CRD（CustomResourceDefinition）取代。 扩展API 自定义资源实际上是为了扩展kubernetes的API，向kubenetes API中增加新类型，可以使用以下三种方式： 修改kubenetes的源码，显然难度比较高，也不太合适 创建自定义API server并聚合到API中 1.7以下版本编写TPR，kubernetes1.7及以上版本用CRD 编写自定义资源是扩展kubernetes API的最简单的方式，是否编写自定义资源来扩展API请参考Should I add a custom resource to my Kubernetes Cluster?，行动前请先考虑以下几点： 你的API是否属于声明式的 是否想使用kubectl命令来管理 是否要作为kubenretes中的对象类型来管理，同时显示在kuberetes dashboard上 是否可以遵守kubernetes的API规则限制，例如URL和API group、namespace限制 是否可以接受该API只能作用于集群或者namespace范围 想要复用kubernetes API的公共功能，比如CRUD、watch、内置的认证和授权等 如果这些都不是你想要的，那么你可以开发一个独立的API。 TPR 注意：TPR已经停止维护，kubernetes 1.7及以上版本请使用CRD。 假如我们要创建一个名为cron-tab.stable.example.com的TPR，yaml文件定义如下： apiVersion: extensions/v1beta1 kind: ThirdPartyResource metadata: name: cron-tab.stable.example.com description: \"A specification of a Pod to run on a cron style schedule\" versions: - name: v1 然后使用kubectl create命令创建该资源，这样就可以创建出一个API端点/apis/stable.example.com/v1/namespaces//crontabs/...。 下面是在Linkerd中的一个实际应用，Linkerd中的一个名为namerd的组件使用了TPR，定义如下： --- kind: ThirdPartyResource apiVersion: extensions/v1beta1 metadata: name: d-tab.l5d.io description: stores dtabs used by namerd versions: - name: v1alpha1 CRD 参考下面的CRD，resourcedefinition.yaml： apiVersion: apiextensions.k8s.io/v1beta1 kind: CustomResourceDefinition metadata: # 名称必须符合下面的格式：. name: crontabs.stable.example.com spec: # REST API使用的组名称：/apis// group: stable.example.com # REST API使用的版本号：/apis// version: v1 # Namespaced或Cluster scope: Namespaced names: # URL中使用的复数名称: /apis/// plural: crontabs # CLI中使用的单数名称 singular: crontab # CamelCased格式的单数类型。在清单文件中使用 kind: CronTab # CLI中使用的资源简称 shortNames: - ct 创建该CRD： kubectl create -f resourcedefinition.yaml 访问RESTful API端点如http://172.20.0.113:8080将看到如下API端点已创建： /apis/stable.example.com/v1/namespaces/*/crontabs/... 创建自定义对象 如下所示： apiVersion: \"stable.example.com/v1\" kind: CronTab metadata: name: my-new-cron-object spec: cronSpec: \"* * * * /5\" image: my-awesome-cron-image 引用该自定义资源的API创建对象。 终止器 可以为自定义对象添加一个终止器，如下所示： apiVersion: \"stable.example.com/v1\" kind: CronTab metadata: finalizers: - finalizer.stable.example.com 删除自定义对象前，异步执行的钩子。对于具有终止器的一个对象，删除请求仅仅是为metadata.deletionTimestamp字段设置一个值，而不是删除它，这将触发监控该对象的控制器执行他们所能处理的任意终止器。 详情参考：Extend the Kubernetes API with CustomResourceDefinitions 使用kubernetes1.7及以上版本请参考Migrate a ThirdPartyResource to CustomResourceDefinition。 自定义控制器 单纯设置了自定义资源，并没有什么用，只有跟自定义控制器结合起来，才能讲资源对象中的声明式API翻译成用户所期望的状态。自定义控制器可以用来管理任何资源类型，但是一般是跟自定义资源结合使用。 请参考使用Operator模式，该模式可以让开发者将自己的领域知识转换成特定的kubenretes API扩展。 API server聚合 Aggregated（聚合的）API server是为了将原来的API server这个巨石（monolithic）应用给拆分成，为了方便用户开发自己的API server集成进来，而不用直接修改kubernetes官方仓库的代码，这样一来也能将API server解耦，方便用户使用实验特性。这些API server可以跟core API server无缝衔接，试用kubectl也可以管理它们。 详情参考Aggregated API Server。 参考 Custom Resources Extend the Kubernetes API with CustomResourceDefinitions Introducing Operators: Putting Operational Knowledge into Software Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-14 18:39:25 "},"concepts/apiservice.html":{"url":"concepts/apiservice.html","title":"APIService","keywords":"","body":"APIService APIService是用来表示一个特定的GroupVersion的中的server，它的结构定义位于代码staging/src/k8s.io/kube-aggregator/pkg/apis/apiregistration/types.go中。 下面是一个APIService的示例配置： apiVersion: apiregistration.k8s.io/v1beta1 kind: APIService metadata: name: v1alpha1.custom-metrics.metrics.k8s.io spec: insecureSkipTLSVerify: true group: custom-metrics.metrics.k8s.io groupPriorityMinimum: 1000 versionPriority: 5 service: name: api namespace: custom-metrics version: v1alpha1 APIService详解 使用apiregistration.k8s.io/v1beta1 版本的APIService，在metadata.name中定义该API的名字。 使用上面的yaml的创建v1alpha1.custom-metrics.metrics.k8s.io APIService。 insecureSkipTLSVerify：当与该服务通信时，禁用TLS证书认证。强加建议不要设置这个参数，默认为 false。应该使用CABundle代替。 service：与该APIService通信时引用的service，其中要注明service的名字和所属的namespace，如果为空的话，则所有的服务都会该API groupversion将在本地443端口处理所有通信。 groupPriorityMinimum：该组API的处理优先级，主要排序是基于groupPriorityMinimum，该数字越大表明优先级越高，客户端就会与其通信处理请求。次要排序是基于字母表顺序，例如v1.bar比v1.foo的优先级更高。 versionPriority：VersionPriority控制其组内的API版本的顺序。必须大于零。主要排序基于VersionPriority，从最高到最低（20大于10）排序。次要排序是基于对象名称的字母比较。 （v1.foo在v1.bar之前）由于它们都是在一个组内，因此数字可能很小，一般都小于10。 查看我们使用上面的yaml文件创建的APIService。 kubectl get apiservice v1alpha1.custom-metrics.metrics.k8s.io -o yaml apiVersion: apiregistration.k8s.io/v1beta1 kind: APIService metadata: creationTimestamp: 2017-12-14T08:27:35Z name: v1alpha1.custom-metrics.metrics.k8s.io resourceVersion: \"35194598\" selfLink: /apis/apiregistration.k8s.io/v1beta1/apiservices/v1alpha1.custom-metrics.metrics.k8s.io uid: a31a3412-e0a8-11e7-9fa4-f4e9d49f8ed0 spec: caBundle: null group: custom-metrics.metrics.k8s.io groupPriorityMinimum: 1000 insecureSkipTLSVerify: true service: name: api namespace: custom-metrics version: v1alpha1 versionPriority: 5 status: conditions: - lastTransitionTime: 2017-12-14T08:27:38Z message: all checks passed reason: Passed status: \"True\" type: Available 查看集群支持的APISerivce 作为Kubernetes中的一种资源对象，可以使用kubectl get apiservice来查看。 例如查看集群中所有的APIService： $ kubectl get apiservice NAME AGE v1. 2d v1.authentication.k8s.io 2d v1.authorization.k8s.io 2d v1.autoscaling 2d v1.batch 2d v1.monitoring.coreos.com 1d v1.networking.k8s.io 2d v1.rbac.authorization.k8s.io 2d v1.storage.k8s.io 2d v1alpha1.custom-metrics.metrics.k8s.io 2h v1beta1.apiextensions.k8s.io 2d v1beta1.apps 2d v1beta1.authentication.k8s.io 2d v1beta1.authorization.k8s.io 2d v1beta1.batch 2d v1beta1.certificates.k8s.io 2d v1beta1.extensions 2d v1beta1.policy 2d v1beta1.rbac.authorization.k8s.io 2d v1beta1.storage.k8s.io 2d v1beta2.apps 2d v2beta1.autoscaling 2d 另外查看当前kubernetes集群支持的API版本还可以使用kubectl api-version： $ kubectl api-versions apiextensions.k8s.io/v1beta1 apiregistration.k8s.io/v1beta1 apps/v1beta1 apps/v1beta2 authentication.k8s.io/v1 authentication.k8s.io/v1beta1 authorization.k8s.io/v1 authorization.k8s.io/v1beta1 autoscaling/v1 autoscaling/v2beta1 batch/v1 batch/v1beta1 certificates.k8s.io/v1beta1 custom-metrics.metrics.k8s.io/v1alpha1 extensions/v1beta1 monitoring.coreos.com/v1 networking.k8s.io/v1 policy/v1beta1 rbac.authorization.k8s.io/v1 rbac.authorization.k8s.io/v1beta1 storage.k8s.io/v1 storage.k8s.io/v1beta1 v1 参考 API Conventions Kuberentes1.8 reference doc Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-14 19:32:56 "},"guide/":{"url":"guide/","title":"用户指南","keywords":"","body":"用户指南 该章节主要记录kubernetes使用过程中的一些配置技巧和操作。 配置Pod的liveness和readiness探针 管理集群中的TLS Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-15 19:40:21 "},"guide/resource-configuration.html":{"url":"guide/resource-configuration.html","title":"资源对象配置","keywords":"","body":"资源配置 Kubernetes 中的各个 Object 的配置指南。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-08-21 18:23:34 "},"guide/configure-liveness-readiness-probes.html":{"url":"guide/configure-liveness-readiness-probes.html","title":"配置Pod的liveness和readiness探针","keywords":"","body":"配置Pod的liveness和readiness探针 当你使用kuberentes的时候，有没有遇到过Pod在启动后一会就挂掉然后又重新启动这样的恶性循环？你有没有想过kubernetes是如何检测pod是否还存活？虽然容器已经启动，但是kubernetes如何知道容器的进程是否准备好对外提供服务了呢？让我们通过kuberentes官网的这篇文章Configure Liveness and Readiness Probes，来一探究竟。 本文将向展示如何配置容器的存活和可读性探针。 Kubelet使用liveness probe（存活探针）来确定何时重启容器。例如，当应用程序处于运行状态但无法做进一步操作，liveness探针将捕获到deadlock，重启处于该状态下的容器，使应用程序在存在bug的情况下依然能够继续运行下去（谁的程序还没几个bug呢）。 Kubelet使用readiness probe（就绪探针）来确定容器是否已经就绪可以接受流量。只有当Pod中的容器都处于就绪状态时kubelet才会认定该Pod处于就绪状态。该信号的作用是控制哪些Pod应该作为service的后端。如果Pod处于非就绪状态，那么它们将会被从service的load balancer中移除。 定义 liveness命令 许多长时间运行的应用程序最终会转换到broken状态，除非重新启动，否则无法恢复。Kubernetes提供了liveness probe来检测和补救这种情况。 在本次练习将基于 gcr.io/google_containers/busybox镜像创建运行一个容器的Pod。以下是Pod的配置文件exec-liveness.yaml： apiVersion: v1 kind: Pod metadata: labels: test: liveness name: liveness-exec spec: containers: - name: liveness args: - /bin/sh - -c - touch /tmp/healthy; sleep 30; rm -rf /tmp/healthy; sleep 600 image: gcr.io/google_containers/busybox livenessProbe: exec: command: - cat - /tmp/healthy initialDelaySeconds: 5 periodSeconds: 5 该配置文件给Pod配置了一个容器。periodSeconds 规定kubelet要每隔5秒执行一次liveness probe。 initialDelaySeconds 告诉kubelet在第一次执行probe之前要的等待5秒钟。探针检测命令是在容器中执行 cat /tmp/healthy 命令。如果命令执行成功，将返回0，kubelet就会认为该容器是活着的并且很健康。如果返回非0值，kubelet就会杀掉这个容器并重启它。 容器启动时，执行该命令： /bin/sh -c \"touch /tmp/healthy; sleep 30; rm -rf /tmp/healthy; sleep 600\" 在容器生命的最初30秒内有一个 /tmp/healthy 文件，在这30秒内 cat /tmp/healthy命令会返回一个成功的返回码。30秒后， cat /tmp/healthy 将返回失败的返回码。 创建Pod： kubectl create -f https://k8s.io/docs/tasks/configure-pod-container/exec-liveness.yaml 在30秒内，查看Pod的event： kubectl describe pod liveness-exec 结果显示没有失败的liveness probe： FirstSeen LastSeen Count From SubobjectPath Type Reason Message --------- -------- ----- ---- ------------- -------- ------ ------- 24s 24s 1 {default-scheduler } Normal Scheduled Successfully assigned liveness-exec to worker0 23s 23s 1 {kubelet worker0} spec.containers{liveness} Normal Pulling pulling image \"gcr.io/google_containers/busybox\" 23s 23s 1 {kubelet worker0} spec.containers{liveness} Normal Pulled Successfully pulled image \"gcr.io/google_containers/busybox\" 23s 23s 1 {kubelet worker0} spec.containers{liveness} Normal Created Created container with docker id 86849c15382e; Security:[seccomp=unconfined] 23s 23s 1 {kubelet worker0} spec.containers{liveness} Normal Started Started container with docker id 86849c15382e 启动35秒后，再次查看pod的event： kubectl describe pod liveness-exec 在最下面有一条信息显示liveness probe失败，容器被删掉并重新创建。 FirstSeen LastSeen Count From SubobjectPath Type Reason Message --------- -------- ----- ---- ------------- -------- ------ ------- 37s 37s 1 {default-scheduler } Normal Scheduled Successfully assigned liveness-exec to worker0 36s 36s 1 {kubelet worker0} spec.containers{liveness} Normal Pulling pulling image \"gcr.io/google_containers/busybox\" 36s 36s 1 {kubelet worker0} spec.containers{liveness} Normal Pulled Successfully pulled image \"gcr.io/google_containers/busybox\" 36s 36s 1 {kubelet worker0} spec.containers{liveness} Normal Created Created container with docker id 86849c15382e; Security:[seccomp=unconfined] 36s 36s 1 {kubelet worker0} spec.containers{liveness} Normal Started Started container with docker id 86849c15382e 2s 2s 1 {kubelet worker0} spec.containers{liveness} Warning Unhealthy Liveness probe failed: cat: can't open '/tmp/healthy': No such file or directory 再等30秒，确认容器已经重启： kubectl get pod liveness-exec 从输出结果来RESTARTS值加1了。 NAME READY STATUS RESTARTS AGE liveness-exec 1/1 Running 1 1m 定义一个liveness HTTP请求 我们还可以使用HTTP GET请求作为liveness probe。下面是一个基于gcr.io/google_containers/liveness镜像运行了一个容器的Pod的例子http-liveness.yaml： apiVersion: v1 kind: Pod metadata: labels: test: liveness name: liveness-http spec: containers: - name: liveness args: - /server image: gcr.io/google_containers/liveness livenessProbe: httpGet: path: /healthz port: 8080 httpHeaders: - name: X-Custom-Header value: Awesome initialDelaySeconds: 3 periodSeconds: 3 该配置文件只定义了一个容器，livenessProbe 指定kubelete需要每隔3秒执行一次liveness probe。initialDelaySeconds 指定kubelet在该执行第一次探测之前需要等待3秒钟。该探针将向容器中的server的8080端口发送一个HTTP GET请求。如果server的/healthz路径的handler返回一个成功的返回码，kubelet就会认定该容器是活着的并且很健康。如果返回失败的返回码，kubelet将杀掉该容器并重启它。 任何大于200小于400的返回码都会认定是成功的返回码。其他返回码都会被认为是失败的返回码。 查看该server的源码：server.go. 最开始的10秒该容器是活着的， /healthz handler返回200的状态码。这之后将返回500的返回码。 http.HandleFunc(\"/healthz\", func(w http.ResponseWriter, r *http.Request) { duration := time.Now().Sub(started) if duration.Seconds() > 10 { w.WriteHeader(500) w.Write([]byte(fmt.Sprintf(\"error: %v\", duration.Seconds()))) } else { w.WriteHeader(200) w.Write([]byte(\"ok\")) } }) 容器启动3秒后，kubelet开始执行健康检查。第一次健康监测会成功，但是10秒后，健康检查将失败，kubelet将杀掉和重启容器。 创建一个Pod来测试一下HTTP liveness检测： kubectl create -f https://k8s.io/docs/tasks/configure-pod-container/http-liveness.yaml After 10 seconds, view Pod events to verify that liveness probes have failed and the Container has been restarted: 10秒后，查看Pod的event，确认liveness probe失败并重启了容器。 kubectl describe pod liveness-http 定义TCP liveness探针 第三种liveness probe使用TCP Socket。 使用此配置，kubelet将尝试在指定端口上打开容器的套接字。 如果可以建立连接，容器被认为是健康的，如果不能就认为是失败的。 apiVersion: v1 kind: Pod metadata: name: goproxy labels: app: goproxy spec: containers: - name: goproxy image: gcr.io/google_containers/goproxy:0.1 ports: - containerPort: 8080 readinessProbe: tcpSocket: port: 8080 initialDelaySeconds: 5 periodSeconds: 10 livenessProbe: tcpSocket: port: 8080 initialDelaySeconds: 15 periodSeconds: 20 如您所见，TCP检查的配置与HTTP检查非常相似。 此示例同时使用了readiness和liveness probe。 容器启动后5秒钟，kubelet将发送第一个readiness probe。 这将尝试连接到端口8080上的goproxy容器。如果探测成功，则该pod将被标记为就绪。Kubelet将每隔10秒钟执行一次该检查。 除了readiness probe之外，该配置还包括liveness probe。 容器启动15秒后，kubelet将运行第一个liveness probe。 就像readiness probe一样，这将尝试连接到goproxy容器上的8080端口。如果liveness probe失败，容器将重新启动。 使用命名的端口 可以使用命名的ContainerPort作为HTTP或TCP liveness检查： ports: - name: liveness-port containerPort: 8080 hostPort: 8080 livenessProbe: httpGet: path: /healthz port: liveness-port 定义readiness探针 有时，应用程序暂时无法对外部流量提供服务。 例如，应用程序可能需要在启动期间加载大量数据或配置文件。 在这种情况下，你不想杀死应用程序，但你也不想发送请求。 Kubernetes提供了readiness probe来检测和减轻这些情况。 Pod中的容器可以报告自己还没有准备，不能处理Kubernetes服务发送过来的流量。 Readiness probe的配置跟liveness probe很像。唯一的不同是使用 readinessProbe而不是livenessProbe。 readinessProbe: exec: command: - cat - /tmp/healthy initialDelaySeconds: 5 periodSeconds: 5 Readiness probe的HTTP和TCP的探测器配置跟liveness probe一样。 Readiness和livenss probe可以并行用于同一容器。 使用两者可以确保流量无法到达未准备好的容器，并且容器在失败时重新启动。 配置Probe Probe中有很多精确和详细的配置，通过它们你能准确的控制liveness和readiness检查： initialDelaySeconds：容器启动后第一次执行探测是需要等待多少秒。 periodSeconds：执行探测的频率。默认是10秒，最小1秒。 timeoutSeconds：探测超时时间。默认1秒，最小1秒。 successThreshold：探测失败后，最少连续探测成功多少次才被认定为成功。默认是1。对于liveness必须是1。最小值是1。 failureThreshold：探测成功后，最少连续探测失败多少次才被认定为失败。默认是3。最小值是1。 HTTP probe中可以给 httpGet设置其他配置项： host：连接的主机名，默认连接到pod的IP。你可能想在http header中设置\"Host\"而不是使用IP。 scheme：连接使用的schema，默认HTTP。 path: 访问的HTTP server的path。 httpHeaders：自定义请求的header。HTTP运行重复的header。 port：访问的容器的端口名字或者端口号。端口号必须介于1和65525之间。 对于HTTP探测器，kubelet向指定的路径和端口发送HTTP请求以执行检查。 Kubelet将probe发送到容器的IP地址，除非地址被httpGet中的可选host字段覆盖。 在大多数情况下，你不想设置主机字段。 有一种情况下你可以设置它。 假设容器在127.0.0.1上侦听，并且Pod的hostNetwork字段为true。 然后，在httpGet下的host应该设置为127.0.0.1。 如果你的pod依赖于虚拟主机，这可能是更常见的情况，你不应该是用host，而是应该在httpHeaders中设置Host头。 参考 关于 Container Probes 的更多信息 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-07 10:22:34 "},"guide/configure-pod-service-account.html":{"url":"guide/configure-pod-service-account.html","title":"配置Pod的Service Account","keywords":"","body":"配置 Pod 的 Service Account Service account 为 Pod 中的进程提供身份信息。 本文是关于 Service Account 的用户指南，管理指南另见 Service Account 的集群管理指南 。 注意：本文档描述的关于 Serivce Account 的行为只有当您按照 Kubernetes 项目建议的方式搭建起集群的情况下才有效。您的集群管理员可能在您的集群中有自定义配置，这种情况下该文档可能并不适用。 当您（真人用户）访问集群（例如使用kubectl命令）时，apiserver 会将您认证为一个特定的 User Account（目前通常是admin，除非您的系统管理员自定义了集群配置）。Pod 容器中的进程也可以与 apiserver 联系。 当它们在联系 apiserver 的时候，它们会被认证为一个特定的 Service Account（例如default）。 使用默认的 Service Account 访问 API server 当您创建 pod 的时候，如果您没有指定一个 service account，系统会自动得在与该pod 相同的 namespace 下为其指派一个default service account。如果您获取刚创建的 pod 的原始 json 或 yaml 信息（例如使用kubectl get pods/podename -o yaml命令），您将看到spec.serviceAccountName字段已经被设置为 automatically set 。 您可以在 pod 中使用自动挂载的 service account 凭证来访问 API，如 Accessing the Cluster 中所描述。 Service account 是否能够取得访问 API 的许可取决于您使用的 授权插件和策略。 在 1.6 以上版本中，您可以选择取消为 service account 自动挂载 API 凭证，只需在 service account 中设置 automountServiceAccountToken: false： apiVersion: v1 kind: ServiceAccount metadata: name: build-robot automountServiceAccountToken: false ... 在 1.6 以上版本中，您也可以选择只取消单个 pod 的 API 凭证自动挂载： apiVersion: v1 kind: Pod metadata: name: my-pod spec: serviceAccountName: build-robot automountServiceAccountToken: false ... 如果在 pod 和 service account 中同时设置了 automountServiceAccountToken , pod 设置中的优先级更高。 使用多个Service Account 每个 namespace 中都有一个默认的叫做 default 的 service account 资源。 您可以使用以下命令列出 namespace 下的所有 serviceAccount 资源。 $ kubectl get serviceAccounts NAME SECRETS AGE default 1 1d 您可以像这样创建一个 ServiceAccount 对象： $ cat > /tmp/serviceaccount.yaml 如果您看到如下的 service account 对象的完整输出信息： $ kubectl get serviceaccounts/build-robot -o yaml apiVersion: v1 kind: ServiceAccount metadata: creationTimestamp: 2015-06-16T00:12:59Z name: build-robot namespace: default resourceVersion: \"272500\" selfLink: /api/v1/namespaces/default/serviceaccounts/build-robot uid: 721ab723-13bc-11e5-aec2-42010af0021e secrets: - name: build-robot-token-bvbk5 然后您将看到有一个 token 已经被自动创建，并被 service account 引用。 您可以使用授权插件来 设置 service account 的权限 。 设置非默认的 service account，只需要在 pod 的spec.serviceAccountName 字段中将name设置为您想要用的 service account 名字即可。 在 pod 创建之初 service account 就必须已经存在，否则创建将被拒绝。 您不能更新已创建的 pod 的 service account。 您可以清理 service account，如下所示： $ kubectl delete serviceaccount/build-robot 手动创建 service account 的 API token 假设我们已经有了一个如上文提到的名为 ”build-robot“ 的 service account，我们手动创建一个新的 secret。 $ cat > /tmp/build-robot-secret.yaml 现在您可以确认下新创建的 secret 取代了 \"build-robot\" 这个 service account 原来的 API token。 所有已不存在的 service account 的 token 将被 token controller 清理掉。 $ kubectl describe secrets/build-robot-secret Name: build-robot-secret Namespace: default Labels: Annotations: kubernetes.io/service-account.name=build-robot,kubernetes.io/service-account.uid=870ef2a5-35cf-11e5-8d06-005056b45392 Type: kubernetes.io/service-account-token Data ==== ca.crt: 1220 bytes token: ... namespace: 7 bytes 注意该内容中的token被省略了。 为 service account 添加 ImagePullSecret 首先，创建一个 imagePullSecret，详见这里。 然后，确认已创建。如： $ kubectl get secrets myregistrykey NAME TYPE DATA AGE myregistrykey kubernetes.io/.dockerconfigjson 1 1d 然后，修改 namespace 中的默认 service account 使用该 secret 作为 imagePullSecret。 kubectl patch serviceaccount default -p '{\"imagePullSecrets\": [{\"name\": \"myregistrykey\"}]}' Vi 交互过程中需要手动编辑： $ kubectl get serviceaccounts default -o yaml > ./sa.yaml $ cat sa.yaml apiVersion: v1 kind: ServiceAccount metadata: creationTimestamp: 2015-08-07T22:02:39Z name: default namespace: default resourceVersion: \"243024\" selfLink: /api/v1/namespaces/default/serviceaccounts/default uid: 052fb0f4-3d50-11e5-b066-42010af0d7b6 secrets: - name: default-token-uudge $ vi sa.yaml [editor session not shown] [delete line with key \"resourceVersion\"] [add lines with \"imagePullSecret:\"] $ cat sa.yaml apiVersion: v1 kind: ServiceAccount metadata: creationTimestamp: 2015-08-07T22:02:39Z name: default namespace: default selfLink: /api/v1/namespaces/default/serviceaccounts/default uid: 052fb0f4-3d50-11e5-b066-42010af0d7b6 secrets: - name: default-token-uudge imagePullSecrets: - name: myregistrykey $ kubectl replace serviceaccount default -f ./sa.yaml serviceaccounts/default 现在，所有当前 namespace 中新创建的 pod 的 spec 中都会增加如下内容： spec: imagePullSecrets: - name: myregistrykey Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-08-21 18:23:34 "},"guide/secret-configuration.html":{"url":"guide/secret-configuration.html","title":"Secret配置","keywords":"","body":"Secret 配置 Secret 对象类型用来保存敏感信息，例如密码、OAuth 令牌和 ssh key。将这些信息放在 secret 中比放在 pod 的定义中或者 docker 镜像中来说更加安全和灵活。参阅 Secret 设计文档 获取更多详细信息。 Secret 概览 Secret 是一种包含少量敏感信息例如密码、token 或 key 的对象。这样的信息可能会被放在 Pod spec 中或者镜像中；将其放在一个 secret 对象中可以更好地控制它的用途，并降低意外暴露的风险。 用户可以创建 secret，同时系统也创建了一些 secret。 要使用 secret，pod 需要引用 secret。Pod 可以用两种方式使用 secret：作为 volume 中的文件被挂载到 pod 中的一个或者多个容器里，或者当 kubelet 为 pod 拉取镜像时使用。 内置 secret Service Account 使用 API 凭证自动创建和附加 secret Kubernetes 自动创建包含访问 API 凭据的 secret，并自动修改您的 pod 以使用此类型的 secret。 如果需要，可以禁用或覆盖自动创建和使用API凭据。但是，如果您需要的只是安全地访问 apiserver，我们推荐这样的工作流程。 参阅 Service Account 文档获取关于 Service Account 如何工作的更多信息。 创建您自己的 Secret 使用 kubectl 创建 Secret 假设有些 pod 需要访问数据库。这些 pod 需要使用的用户名和密码在您本地机器的 ./username.txt 和 ./password.txt 文件里。 # Create files needed for rest of example. $ echo -n \"admin\" > ./username.txt $ echo -n \"1f2d1e2e67df\" > ./password.txt kubectl create secret 命令将这些文件打包到一个 Secret 中并在 API server 中创建了一个对象。 $ kubectl create secret generic db-user-pass --from-file=./username.txt --from-file=./password.txt secret \"db-user-pass\" created 您可以这样检查刚创建的 secret： $ kubectl get secrets NAME TYPE DATA AGE db-user-pass Opaque 2 51s $ kubectl describe secrets/db-user-pass Name: db-user-pass Namespace: default Labels: Annotations: Type: Opaque Data ==== password.txt: 12 bytes username.txt: 5 bytes 请注意，默认情况下，get 和 describe 命令都不会显示文件的内容。这是为了防止将 secret 中的内容被意外暴露给从终端日志记录中刻意寻找它们的人。 请参阅 解码 secret 了解如何查看它们的内容。 手动创建 Secret 您也可以先以 json 或 yaml 格式在文件中创建一个 secret 对象，然后创建该对象。 每一项必须是 base64 编码： $ echo -n \"admin\" | base64 YWRtaW4= $ echo -n \"1f2d1e2e67df\" | base64 MWYyZDFlMmU2N2Rm 现在可以像这样写一个 secret 对象： apiVersion: v1 kind: Secret metadata: name: mysecret type: Opaque data: username: YWRtaW4= password: MWYyZDFlMmU2N2Rm 数据字段是一个映射。它的键必须匹配 DNS_SUBDOMAIN，前导点也是可以的。这些值可以是任意数据，使用 base64 进行编码。 使用 kubectl create 创建 secret： $ kubectl create -f ./secret.yaml secret \"mysecret\" created 编码注意： secret 数据的序列化 JSON 和 YAML 值使用 base64 编码成字符串。换行符在这些字符串中无效，必须省略。当在Darwin/OS X上使用 base64 实用程序时，用户应避免使用 -b 选项来拆分长行。另外，对于 Linux用户如果 -w 选项不可用的话，应该添加选项 -w 0 到 base64 命令或管道 base64 | tr -d '\\n' 。 解码 Secret 可以使用 kubectl get secret 命令获取 secret。例如，获取在上一节中创建的 secret： $ kubectl get secret mysecret -o yaml apiVersion: v1 data: username: YWRtaW4= password: MWYyZDFlMmU2N2Rm kind: Secret metadata: creationTimestamp: 2016-01-22T18:41:56Z name: mysecret namespace: default resourceVersion: \"164619\" selfLink: /api/v1/namespaces/default/secrets/mysecret uid: cfee02d6-c137-11e5-8d73-42010af00002 type: Opaque 解码密码字段： $ echo \"MWYyZDFlMmU2N2Rm\" | base64 --decode 1f2d1e2e67df 使用 Secret Secret 可以作为数据卷被挂载，或作为环境变量暴露出来以供 pod 中的容器使用。它们也可以被系统的其他部分使用，而不直接暴露在 pod 内。例如，它们可以保存凭据，系统的其他部分应该用它来代表您与外部系统进行交互。 在 Pod 中使用 Secret 文件 在 Pod 中的 volume 里使用 Secret： 创建一个 secret 或者使用已有的 secret。多个 pod 可以引用同一个 secret。 修改您的 pod 的定义在 spec.volumes[] 下增加一个 volume。可以给这个 volume 随意命名，它的 spec.volumes[].secret.secretName 必须等于 secret 对象的名字。 将 spec.containers[].volumeMounts[] 加到需要用到该 secret 的容器中。指定 spec.containers[].volumeMounts[].readOnly = true 和 spec.containers[].volumeMounts[].mountPath 为您想要该 secret 出现的尚未使用的目录。 修改您的镜像并且／或者命令行让程序从该目录下寻找文件。Secret 的 data 映射中的每一个键都成为了 mountPath 下的一个文件名。 这是一个在 pod 中使用 volume 挂在 secret 的例子： apiVersion: v1 kind: Pod metadata: name: mypod spec: containers: - name: mypod image: redis volumeMounts: - name: foo mountPath: \"/etc/foo\" readOnly: true volumes: - name: foo secret: secretName: mysecret 您想要用的每个 secret 都需要在 spec.volumes 中指明。 如果 pod 中有多个容器，每个容器都需要自己的 volumeMounts 配置块，但是每个 secret 只需要一个 spec.volumes。 您可以打包多个文件到一个 secret 中，或者使用的多个 secret，怎样方便就怎样来。 向特性路径映射 secret 密钥 我们还可以控制 Secret key 映射在 volume 中的路径。您可以使用 spec.volumes[].secret.items 字段修改每个 key 的目标路径： apiVersion: v1 kind: Pod metadata: name: mypod spec: containers: - name: mypod image: redis volumeMounts: - name: foo mountPath: \"/etc/foo\" readOnly: true volumes: - name: foo secret: secretName: mysecret items: - key: username path: my-group/my-username 将会发生什么呢： username secret 存储在 /etc/foo/my-group/my-username 文件中而不是 /etc/foo/username 中。 password secret 没有被影射 如果使用了 spec.volumes[].secret.items，只有在 items 中指定的 key 被影射。要使用 secret 中所有的 key，所有这些都必须列在 items 字段中。所有列出的密钥必须存在于相应的 secret 中。否则，不会创建卷。 Secret 文件权限 您还可以指定 secret 将拥有的权限模式位文件。如果不指定，默认使用 0644。您可以为整个保密卷指定默认模式，如果需要，可以覆盖每个密钥。 apiVersion: v1 kind: Pod metadata: name: mypod spec: containers: - name: mypod image: redis volumeMounts: - name: foo mountPath: \"/etc/foo\" volumes: - name: foo secret: secretName: mysecret defaultMode: 256 然后，secret 将被挂载到 /etc/foo 目录，所有通过该 secret volume 挂载创建的文件的权限都是 0400。 请注意，JSON 规范不支持八进制符号，因此使用 256 值作为 0400 权限。如果您使用 yaml 而不是 json 作为 pod，则可以使用八进制符号以更自然的方式指定权限。 您还可以是用映射，如上一个示例，并为不同的文件指定不同的权限，如下所示： apiVersion: v1 kind: Pod metadata: name: mypod spec: containers: - name: mypod image: redis volumeMounts: - name: foo mountPath: \"/etc/foo\" volumes: - name: foo secret: secretName: mysecret items: - key: username path: my-group/my-username mode: 511 在这种情况下，导致 /etc/foo/my-group/my-username 的文件的权限值为 0777。由于 JSON 限制，必须以十进制格式指定模式。 请注意，如果稍后阅读此权限值可能会以十进制格式显示。 从 Volume 中消费 secret 值 在挂载的 secret volume 的容器内，secret key 将作为文件，并且 secret 的值使用 base-64 解码并存储在这些文件中。这是在上面的示例容器内执行的命令的结果： $ ls /etc/foo/ username password $ cat /etc/foo/username admin $ cat /etc/foo/password 1f2d1e2e67df 容器中的程序负责从文件中读取 secret。 挂载的 secret 被自动更新 当已经在 volume 中消被消费的 secret 被更新时，被映射的 key 也将被更新。 Kubelet 在周期性同步时检查被挂载的 secret 是不是最新的。但是，它正在使用其基于本地 ttl 的缓存来获取当前的 secret 值。结果是，当 secret 被更新的时刻到将新的 secret 映射到 pod 的时刻的总延迟可以与 kubelet 中的secret 缓存的 kubelet sync period + ttl 一样长。 Secret 作为环境变量 将 secret 作为 pod 中的环境变量使用： 创建一个 secret 或者使用一个已存在的 secret。多个 pod 可以引用同一个 secret。 在每个容器中修改您想要使用 secret key 的 Pod 定义，为要使用的每个 secret key 添加一个环境变量。消费secret key 的环境变量应填充 secret 的名称，并键入 env[x].valueFrom.secretKeyRef。 修改镜像并／或者命令行，以便程序在指定的环境变量中查找值。 apiVersion: v1 kind: Pod metadata: name: secret-env-pod spec: containers: - name: mycontainer image: redis env: - name: SECRET_USERNAME valueFrom: secretKeyRef: name: mysecret key: username - name: SECRET_PASSWORD valueFrom: secretKeyRef: name: mysecret key: password restartPolicy: Never 消费环境变量里的 Secret 值 在一个消耗环境变量 secret 的容器中，secret key 作为包含 secret 数据的 base-64 解码值的常规环境变量。这是从上面的示例在容器内执行的命令的结果： $ echo $SECRET_USERNAME admin $ echo $SECRET_PASSWORD 1f2d1e2e67df 使用 imagePullSecret imagePullSecret 是将包含 Docker（或其他）镜像注册表密码的 secret 传递给 Kubelet 的一种方式，因此可以代表您的 pod 拉取私有镜像。 手动指定 imagePullSecret imagePullSecret 的使用在 镜像文档 中说明。 安排 imagePullSecrets 自动附加 您可以手动创建 imagePullSecret，并从 serviceAccount 引用它。使用该 serviceAccount 创建的任何 pod 和默认使用该 serviceAccount 的 pod 将会将其的 imagePullSecret 字段设置为服务帐户的 imagePullSecret 字段。有关该过程的详细说明，请参阅 将 ImagePullSecrets 添加到服务帐户。 自动挂载手动创建的 Secret 手动创建的 secret（例如包含用于访问 github 帐户的令牌）可以根据其服务帐户自动附加到 pod。请参阅 使用 PodPreset 向 Pod 中注入信息 以获取该进程的详细说明。 详细 限制 验证 secret volume 来源确保指定的对象引用实际上指向一个类型为 Secret 的对象。因此，需要在依赖于它的任何 pod 之前创建一个 secret。 Secret API 对象驻留在命名空间中。它们只能由同一命名空间中的 pod 引用。 每个 secret 的大小限制为1MB。这是为了防止创建非常大的 secret 会耗尽 apiserver 和 kubelet 的内存。然而，创建许多较小的 secret 也可能耗尽内存。更全面得限制 secret 对内存使用的更全面的限制是计划中的功能。 Kubelet 仅支持从 API server 获取的 Pod 使用 secret。这包括使用 kubectl 创建的任何 pod，或间接通过 replication controller 创建的 pod。它不包括通过 kubelet --manifest-url 标志，其 --config 标志或其 REST API 创建的pod（这些不是创建 pod 的常用方法）。 必须先创建 secret，除非将它们标记为可选项，否则必须在将其作为环境变量在 pod 中使用之前创建 secret。对不存在的 secret 的引用将阻止其启动。 通过 secretKeyRef 对不存在于命名的 key 中的 key 进行引用将阻止该启动。 用于通过 envFrom 填充环境变量的 secret，这些环境变量具有被认为是无效环境变量名称的 key 将跳过这些键。该 pod 将被允许启动。将会有一个事件，其原因是 InvalidVariableNames，该消息将包含被跳过的无效键的列表。该示例显示一个 pod，它指的是包含2个无效键，1badkey 和 2alsobad 的默认/mysecret ConfigMap。 $ kubectl get events LASTSEEN FIRSTSEEN COUNT NAME KIND SUBOBJECT TYPE REASON 0s 0s 1 dapi-test-pod Pod Warning InvalidEnvironmentVariableNames kubelet, 127.0.0.1 Keys [1badkey, 2alsobad] from the EnvFrom secret default/mysecret were skipped since they are considered invalid environment variable names. Secret 与 Pod 生命周期的联系 通过 API 创建的 Pod 时，不会检查应用的 secret 是否存在。一旦 Pod 被调度，kubelet 就会尝试获取该 secret 的值。如果获取不到该 secret，或者暂时无法与 API server 建立连接，kubelet 将会定期重试。Kubelet 将会报告关于 pod 的事件，并解释它无法启动的原因。一旦获取的 secret，kubelet将创建并装载一个包含它的卷。在安装所有pod的卷之前，都不会启动 pod 的容器。 使用案例 使用案例：包含 ssh 密钥的 pod 创建一个包含 ssh key 的 secret： $ kubectl create secret generic ssh-key-secret --from-file=ssh-privatekey=/path/to/.ssh/id_rsa --from-file=ssh-publickey=/path/to/.ssh/id_rsa.pub 安全性注意事项：发送自己的 ssh 密钥之前要仔细思考：集群的其他用户可能有权访问该密钥。使用您想要共享 Kubernetes 群集的所有用户可以访问的服务帐户，如果它们遭到入侵，可以撤销。 现在我们可以创建一个使用 ssh 密钥引用 secret 的pod，并在一个卷中使用它： kind: Pod apiVersion: v1 metadata: name: secret-test-pod labels: name: secret-test spec: volumes: - name: secret-volume secret: secretName: ssh-key-secret containers: - name: ssh-test-container image: mySshImage volumeMounts: - name: secret-volume readOnly: true mountPath: \"/etc/secret-volume\" 当容器中的命令运行时，密钥的片段将可在以下目录： /etc/secret-volume/ssh-publickey /etc/secret-volume/ssh-privatekey 然后容器可以自由使用密钥数据建立一个 ssh 连接。 使用案例：包含 prod/test 凭据的 pod 下面的例子说明一个 pod 消费一个包含 prod 凭据的 secret，另一个 pod 使用测试环境凭据消费 secret。 创建 secret： $ kubectl create secret generic prod-db-secret --from-literal=username=produser --from-literal=password=Y4nys7f11 secret \"prod-db-secret\" created $ kubectl create secret generic test-db-secret --from-literal=username=testuser --from-literal=password=iluvtests secret \"test-db-secret\" created 创建 pod ： apiVersion: v1 kind: List items: - kind: Pod apiVersion: v1 metadata: name: prod-db-client-pod labels: name: prod-db-client spec: volumes: - name: secret-volume secret: secretName: prod-db-secret containers: - name: db-client-container image: myClientImage volumeMounts: - name: secret-volume readOnly: true mountPath: \"/etc/secret-volume\" - kind: Pod apiVersion: v1 metadata: name: test-db-client-pod labels: name: test-db-client spec: volumes: - name: secret-volume secret: secretName: test-db-secret containers: - name: db-client-container image: myClientImage volumeMounts: - name: secret-volume readOnly: true mountPath: \"/etc/secret-volume\" 这两个容器将在其文件系统上显示以下文件，其中包含每个容器环境的值： /etc/secret-volume/username /etc/secret-volume/password 请注意，两个 pod 的 spec 配置中仅有一个字段有所不同；这有助于使用普通的 pod 配置模板创建具有不同功能的 pod。您可以使用两个 service account 进一步简化基本 pod spec：一个名为 prod-user 拥有 prod-db-secret ，另一个称为 test-user 拥有 test-db-secret 。然后，pod spec 可以缩短为，例如： kind: Pod apiVersion: v1 metadata: name: prod-db-client-pod labels: name: prod-db-client spec: serviceAccount: prod-db-client containers: - name: db-client-container image: myClientImage 使用案例：secret 卷中以点号开头的文件 为了将数据“隐藏”起来（即文件名以点号开头的文件），简单地说让该键以一个点开始。例如，当如下 secret 被挂载到卷中： kind: Secret apiVersion: v1 metadata: name: dotfile-secret data: .secret-file: dmFsdWUtMg0KDQo= --- kind: Pod apiVersion: v1 metadata: name: secret-dotfiles-pod spec: volumes: - name: secret-volume secret: secretName: dotfile-secret containers: - name: dotfile-test-container image: gcr.io/google_containers/busybox command: - ls - \"-l\" - \"/etc/secret-volume\" volumeMounts: - name: secret-volume readOnly: true mountPath: \"/etc/secret-volume\" Secret-volume 将包含一个单独的文件，叫做 .secret-file，dotfile-test-container 的 /etc/secret-volume/.secret-file路径下将有该文件。 注意 以点号开头的文件在 ls -l 的输出中被隐藏起来了；列出目录内容时，必须使用 ls -la 才能查看它们。 使用案例：Secret 仅对 pod 中的一个容器可见 考虑以下一个需要处理 HTTP 请求的程序，执行一些复杂的业务逻辑，然后使用 HMAC 签署一些消息。因为它具有复杂的应用程序逻辑，所以在服务器中可能会出现一个未被注意的远程文件读取漏洞，这可能会将私钥暴露给攻击者。 这可以在两个容器中分为两个进程：前端容器，用于处理用户交互和业务逻辑，但无法看到私钥；以及可以看到私钥的签名者容器，并且响应来自前端的简单签名请求（例如通过本地主机网络）。 使用这种分割方法，攻击者现在必须欺骗应用程序服务器才能进行任意的操作，这可能比使其读取文件更难。 最佳实践 客户端使用 secret API 当部署与 secret API 交互的应用程序时，应使用诸如 RBAC 之类的 授权策略 来限制访问。 Secret 的重要性通常不尽相同，其中许多可能只对 Kubernetes 集群内（例如 service account 令牌）和对外部系统造成影响。即使一个应用程序可以理解其期望的与之交互的 secret 的权力，但是同一命名空间中的其他应用程序也可以使这些假设无效。 由于这些原因，在命名空间中 watch 和 list secret 的请求是非常强大的功能，应该避免这样的行为，因为列出 secret 可以让客户端检查所有 secret 是否在该命名空间中。在群集中watch 和 list 所有 secret 的能力应该只保留给最有特权的系统级组件。 需要访问 secrets API 的应用程序应该根据他们需要的 secret 执行 get 请求。这允许管理员限制对所有 secret 的访问，同时设置 白名单访问 应用程序需要的各个实例。 为了提高循环获取的性能，客户端可以设计引用 secret 的资源，然后 watch 资源，在引用更改时重新请求 secret。此外，还提出了一种 ”批量监控“ API 来让客户端 watch 每个资源，该功能可能会在将来的 Kubernetes 版本中提供。 安全属性 保护 因为 secret 对象可以独立于使用它们的 pod 而创建，所以在创建、查看和编辑 pod 的流程中 secret 被暴露的风险较小。系统还可以对 secret 对象采取额外的预防措施，例如避免将其写入到磁盘中可能的位置。 只有当节点上的 pod 需要用到该 secret 时，该 secret 才会被发送到该节点上。它不会被写入磁盘，而是存储在 tmpfs 中。一旦依赖于它的 pod 被删除，它就被删除。 在大多数 Kubernetes 项目维护的发行版中，用户与 API server 之间的通信以及从 API server 到 kubelet 的通信都受到 SSL/TLS 的保护。通过这些通道传输时，secret 受到保护。 节点上的 secret 数据存储在 tmpfs 卷中，因此不会传到节点上的其他磁盘。 同一节点上的很多个 pod 可能拥有多个 secret。但是，只有 pod 请求的 secret 在其容器中才是可见的。因此，一个 pod 不能访问另一个 Pod 的 secret。 Pod 中有多个容器。但是，pod 中的每个容器必须请求其挂载卷中的 secret 卷才能在容器内可见。这可以用于 在 Pod 级别构建安全分区。 风险 API server 的 secret 数据以纯文本的方式存储在 etcd 中；因此： 管理员应该限制 admin 用户访问 etcd； API server 中的 secret 数据位于 etcd 使用的磁盘上；管理员可能希望在不再使用时擦除/粉碎 etcd 使用的磁盘 如果您将 secret 数据编码为 base64 的清单（JSON 或 YAML）文件，共享该文件或将其检入代码库，这样的话该密码将会被泄露。 Base64 编码不是一种加密方式，一样也是纯文本。 应用程序在从卷中读取 secret 后仍然需要保护 secret 的值，例如不会意外记录或发送给不信任方。 可以创建和使用 secret 的 pod 的用户也可以看到该 secret 的值。即使 API server 策略不允许用户读取 secret 对象，用户也可以运行暴露 secret 的 pod。 如果运行了多个副本，那么这些 secret 将在它们之间共享。默认情况下，etcd 不能保证与 SSL/TLS 的对等通信，尽管可以进行配置。 目前，任何节点的 root 用户都可以通过模拟 kubelet 来读取 API server 中的任何 secret。只有向实际需要它们的节点发送 secret 才能限制单个节点的根漏洞的影响，该功能还在计划中。 原文地址：https://github.com/rootsongjc/kubernetes.github.io/blob/master/docs/concepts/configuration/secret.md Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-09-28 21:19:22 "},"guide/resource-quota-management.html":{"url":"guide/resource-quota-management.html","title":"管理namespace中的资源配额","keywords":"","body":"管理namespace中的资源配额 当用多个团队或者用户共用同一个集群的时候难免会有资源竞争的情况发生，这时候就需要对不同团队或用户的资源使用配额做出限制。 开启资源配额限制功能 目前有两种资源分配管理相关的控制策略插件 ResourceQuota 和 LimitRange。 要启用它们只要 API Server 的启动配置的 KUBE_ADMISSION_CONTROL 参数中加入了 ResourceQuota 的设置，这样就给集群开启了资源配额限制功能，加入 LimitRange 可以用来限制一个资源申请的范围限制，参考 为 namesapce 配置默认的内存请求与限额 和 在 namespace 中配置默认的CPU请求与限额。 两种控制策略的作用范围都是对于某一 namespace，ResourceQuota 用来限制 namespace 中所有的 Pod 占用的总的资源 request 和 limit，而 LimitRange 是用来设置 namespace 中 Pod 的默认的资源 request 和 limit 值。 资源配额分为三种类型： 计算资源配额 存储资源配额 对象数量配额 关于资源配额的详细信息请参考 kubernetes 官方文档 资源配额。 示例 我们为 spark-cluster 这个 namespace 设置 ResouceQuota 和 LimitRange。 以下 yaml 文件可以在 kubernetes-handbook 的 manifests/spark-with-kubernetes-native-scheduler 目录下找到。 配置计算资源配额 配置文件：spark-compute-resources.yaml apiVersion: v1 kind: ResourceQuota metadata: name: compute-resources namespace: spark-cluster spec: hard: pods: \"20\" requests.cpu: \"20\" requests.memory: 100Gi limits.cpu: \"40\" limits.memory: 200Gi 要想查看该配置只要执行： kubectl -n spark-cluster describe resourcequota compute-resources 配置对象数量限制 配置文件：spark-object-counts.yaml apiVersion: v1 kind: ResourceQuota metadata: name: object-counts namespace: spark-cluster spec: hard: configmaps: \"10\" persistentvolumeclaims: \"4\" replicationcontrollers: \"20\" secrets: \"10\" services: \"10\" services.loadbalancers: \"2\" 配置CPU和内存LimitRange 配置文件：spark-limit-range.yaml apiVersion: v1 kind: LimitRange metadata: name: mem-limit-range spec: limits: - default: memory: 50Gi cpu: 5 defaultRequest: memory: 1Gi cpu: 1 type: Container default 即 limit 的值 defaultRequest 即 request 的值 参考 资源配额 为命名空间配置默认的内存请求与限额 在命名空间中配置默认的CPU请求与限额 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-10-10 14:50:27 "},"guide/command-usage.html":{"url":"guide/command-usage.html","title":"命令使用","keywords":"","body":"命令使用 Kubernetes 中的 kubectl 及其他管理命令使用。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-08-21 18:23:34 "},"guide/docker-cli-to-kubectl.html":{"url":"guide/docker-cli-to-kubectl.html","title":"docker用户过度到kubectl命令行指南","keywords":"","body":"docker用户过度到kubectl命令行指南 对于没有使用过 kubernetes 的 docker 用户，如何快速掌握 kubectl 命令？ 在本文中，我们将向 docker-cli 用户介绍 Kubernetes 命令行如何与 api 进行交互。该命令行工具——kubectl，被设计成 docker-cli 用户所熟悉的样子，但是它们之间又存在一些必要的差异。该文档将向您展示每个 docker 子命令和 kubectl 与其等效的命令。 在使用 kubernetes 集群的时候，docker 命令通常情况是不需要用到的，只有在调试程序或者容器的时候用到，我们基本上使用 kubectl 命令即可，所以在操作 kubernetes 的时候我们抛弃原先使用 docker 时的一些观念。 docker run 如何运行一个 nginx Deployment 并将其暴露出来？ 查看 kubectl run 。 使用 docker 命令： $ docker run -d --restart=always -e DOMAIN=cluster --name nginx-app -p 80:80 nginx a9ec34d9878748d2f33dc20cb25c714ff21da8d40558b45bfaec9955859075d0 $ docker ps CONTAINER ID IMAGE COMMAND CREATED STATUS PORTS NAMES a9ec34d98787 nginx \"nginx -g 'daemon of 2 seconds ago Up 2 seconds 0.0.0.0:80->80/tcp, 443/tcp nginx-app 使用 kubectl 命令： # start the pod running nginx $ kubectl run --image=nginx nginx-app --port=80 --env=\"DOMAIN=cluster\" deployment \"nginx-app\" created 在大于等于 1.2 版本 Kubernetes 集群中，使用kubectl run 命令将创建一个名为 \"nginx-app\" 的 Deployment。如果您运行的是老版本，将会创建一个 replication controller。 如果您想沿用旧的行为，使用 --generation=run/v1 参数，这样就会创建 replication controller。查看 kubectl run 获取更多详细信息。 # expose a port through with a service $ kubectl expose deployment nginx-app --port=80 --name=nginx-http service \"nginx-http\" exposed 在 kubectl 命令中，我们创建了一个 Deployment，这将保证有 N 个运行 nginx 的 pod（N 代表 spec 中声明的 replica 数，默认为 1）。我们还创建了一个 service，使用 selector 匹配具有相应的 selector 的 Deployment。查看 快速开始 获取更多信息。 默认情况下镜像会在后台运行，与docker run -d ... 类似，如果您想在前台运行，使用： kubectl run [-i] [--tty] --attach --image= 与 docker run ... 不同的是，如果指定了 --attach ，我们将连接到 stdin，stdout 和 stderr，而不能控制具体连接到哪个输出流（docker -a ...）。 因为我们使用 Deployment 启动了容器，如果您终止了连接到的进程（例如 ctrl-c），容器将会重启，这跟 docker run -it不同。 如果想销毁该 Deployment（和它的 pod），您需要运行 kubeclt delete deployment 。 docker ps 如何列出哪些正在运行？查看 kubectl get。 使用 docker 命令： $ docker ps CONTAINER ID IMAGE COMMAND CREATED STATUS PORTS NAMES a9ec34d98787 nginx \"nginx -g 'daemon of About an hour ago Up About an hour 0.0.0.0:80->80/tcp, 443/tcp nginx-app 使用 kubectl 命令： $ kubectl get po NAME READY STATUS RESTARTS AGE nginx-app-5jyvm 1/1 Running 0 1h docker attach 如何连接到已经运行在容器中的进程？查看 kubectl attach。 使用 docker 命令： $ docker ps CONTAINER ID IMAGE COMMAND CREATED STATUS PORTS NAMES a9ec34d98787 nginx \"nginx -g 'daemon of 8 minutes ago Up 8 minutes 0.0.0.0:80->80/tcp, 443/tcp nginx-app $ docker attach a9ec34d98787 ... 使用 kubectl 命令： $ kubectl get pods NAME READY STATUS RESTARTS AGE nginx-app-5jyvm 1/1 Running 0 10m $ kubectl attach -it nginx-app-5jyvm ... docker exec 如何在容器中执行命令？查看 kubectl exec。 使用 docker 命令： $ docker ps CONTAINER ID IMAGE COMMAND CREATED STATUS PORTS NAMES a9ec34d98787 nginx \"nginx -g 'daemon of 8 minutes ago Up 8 minutes 0.0.0.0:80->80/tcp, 443/tcp nginx-app $ docker exec a9ec34d98787 cat /etc/hostname a9ec34d98787 使用 kubectl 命令： $ kubectl get po NAME READY STATUS RESTARTS AGE nginx-app-5jyvm 1/1 Running 0 10m $ kubectl exec nginx-app-5jyvm -- cat /etc/hostname nginx-app-5jyvm 执行交互式命令怎么办？ 使用 docker 命令： $ docker exec -ti a9ec34d98787 /bin/sh # exit 使用 kubectl 命令： $ kubectl exec -ti nginx-app-5jyvm -- /bin/sh # exit 更多信息请查看 获取运行中容器的 Shell 环境。 docker logs 如何查看运行中进程的 stdout/stderr？查看 kubectl logs。 使用 docker 命令： $ docker logs -f a9e 192.168.9.1 - - [14/Jul/2015:01:04:02 +0000] \"GET / HTTP/1.1\" 200 612 \"-\" \"curl/7.35.0\" \"-\" 192.168.9.1 - - [14/Jul/2015:01:04:03 +0000] \"GET / HTTP/1.1\" 200 612 \"-\" \"curl/7.35.0\" \"-\" 使用 kubectl 命令： $ kubectl logs -f nginx-app-zibvs 10.240.63.110 - - [14/Jul/2015:01:09:01 +0000] \"GET / HTTP/1.1\" 200 612 \"-\" \"curl/7.26.0\" \"-\" 10.240.63.110 - - [14/Jul/2015:01:09:02 +0000] \"GET / HTTP/1.1\" 200 612 \"-\" \"curl/7.26.0\" \"-\" 现在是时候提一下 pod 和容器之间的细微差别了；默认情况下如果 pod 中的进程退出 pod 也不会终止，相反它将会重启该进程。这类似于 docker run 时的 --restart=always 选项， 这是主要差别。在 docker 中，进程的每个调用的输出都是被连接起来的，但是对于 kubernetes，每个调用都是分开的。要查看以前在 kubernetes 中执行的输出，请执行以下操作： $ kubectl logs --previous nginx-app-zibvs 10.240.63.110 - - [14/Jul/2015:01:09:01 +0000] \"GET / HTTP/1.1\" 200 612 \"-\" \"curl/7.26.0\" \"-\" 10.240.63.110 - - [14/Jul/2015:01:09:02 +0000] \"GET / HTTP/1.1\" 200 612 \"-\" \"curl/7.26.0\" \"-\" 查看 记录和监控集群活动 获取更多信息。 docker stop 和 docker rm 如何停止和删除运行中的进程？查看 kubectl delete。 使用 docker 命令： $ docker ps CONTAINER ID IMAGE COMMAND CREATED STATUS PORTS NAMES a9ec34d98787 nginx \"nginx -g 'daemon of 22 hours ago Up 22 hours 0.0.0.0:80->80/tcp, 443/tcp nginx-app $ docker stop a9ec34d98787 a9ec34d98787 $ docker rm a9ec34d98787 a9ec34d98787 使用 kubectl 命令： $ kubectl get deployment nginx-app NAME DESIRED CURRENT UP-TO-DATE AVAILABLE AGE nginx-app 1 1 1 1 2m $ kubectl get po -l run=nginx-app NAME READY STATUS RESTARTS AGE nginx-app-2883164633-aklf7 1/1 Running 0 2m $ kubectl delete deployment nginx-app deployment \"nginx-app\" deleted $ kubectl get po -l run=nginx-app # Return nothing 请注意，我们不直接删除 pod。使用 kubectl 命令，我们要删除拥有该 pod 的 Deployment。如果我们直接删除pod，Deployment 将会重新创建该 pod。 docker login 在 kubectl 中没有对 docker login 的直接模拟。如果您有兴趣在私有镜像仓库中使用 Kubernetes，请参阅 使用私有镜像仓库。 docker version 如何查看客户端和服务端的版本？查看 kubectl version。 使用 docker 命令： $ docker version Client version: 1.7.0 Client API version: 1.19 Go version (client): go1.4.2 Git commit (client): 0baf609 OS/Arch (client): linux/amd64 Server version: 1.7.0 Server API version: 1.19 Go version (server): go1.4.2 Git commit (server): 0baf609 OS/Arch (server): linux/amd64 使用 kubectl 命令： $ kubectl version Client Version: version.Info{Major:\"1\", Minor:\"6\", GitVersion:\"v1.6.9+a3d1dfa6f4335\", GitCommit:\"9b77fed11a9843ce3780f70dd251e92901c43072\", GitTreeState:\"dirty\", BuildDate:\"2017-08-29T20:32:58Z\", OpenPaasKubernetesVersion:\"v1.03.02\", GoVersion:\"go1.7.5\", Compiler:\"gc\", Platform:\"linux/amd64\"} Server Version: version.Info{Major:\"1\", Minor:\"6\", GitVersion:\"v1.6.9+a3d1dfa6f4335\", GitCommit:\"9b77fed11a9843ce3780f70dd251e92901c43072\", GitTreeState:\"dirty\", BuildDate:\"2017-08-29T20:32:58Z\", OpenPaasKubernetesVersion:\"v1.03.02\", GoVersion:\"go1.7.5\", Compiler:\"gc\", Platform:\"linux/amd64\"} docker info 如何获取有关环境和配置的各种信息？查看 kubectl cluster-info。 使用 docker 命令： $ docker info Containers: 40 Images: 168 Storage Driver: aufs Root Dir: /usr/local/google/docker/aufs Backing Filesystem: extfs Dirs: 248 Dirperm1 Supported: false Execution Driver: native-0.2 Logging Driver: json-file Kernel Version: 3.13.0-53-generic Operating System: Ubuntu 14.04.2 LTS CPUs: 12 Total Memory: 31.32 GiB Name: k8s-is-fun.mtv.corp.google.com ID: ADUV:GCYR:B3VJ:HMPO:LNPQ:KD5S:YKFQ:76VN:IANZ:7TFV:ZBF4:BYJO WARNING: No swap limit support 使用 kubectl 命令： $ kubectl cluster-info Kubernetes master is running at https://108.59.85.141 KubeDNS is running at https://108.59.85.141/api/v1/namespaces/kube-system/services/kube-dns/proxy KubeUI is running at https://108.59.85.141/api/v1/namespaces/kube-system/services/kube-ui/proxy Grafana is running at https://108.59.85.141/api/v1/namespaces/kube-system/services/monitoring-grafana/proxy Heapster is running at https://108.59.85.141/api/v1/namespaces/kube-system/services/monitoring-heapster/proxy InfluxDB is running at https://108.59.85.141/api/v1/namespaces/kube-system/services/monitoring-influxdb/proxy 原文地址：https://github.com/rootsongjc/kubernetes.github.io/blob/master/docs/user-guide/docker-cli-to-kubectl.md Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-09-16 20:53:17 "},"guide/using-kubectl.html":{"url":"guide/using-kubectl.html","title":"kubectl命令概览","keywords":"","body":"Kubctl命令概览 Kubernetes提供的kubectl命令是与集群交互最直接的方式，v1.6版本的kubectl命令参考图如下： 图片 - kubectl cheatsheet Kubectl的子命令主要分为8个类别： 基础命令（初学者都会使用的） 基础命令（中级） 部署命令 集群管理命令 故障排查和调试命令 高级命令 设置命令 其他命令 熟悉这些命令有助于大家来操作和管理kubernetes集群。 kube-shell 开源项目kube-shell可以为kubectl提供自动的命令提示和补全，使用起来特别方便，推荐给大家。 Kube-shell有以下特性： 命令提示，给出命令的使用说明 自动补全，列出可选命令并可以通过tab键自动补全，支持模糊搜索 高亮 使用tab键可以列出可选的对象 vim模式 Mac下安装 pip install kube-shell --user -U Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-03 17:44:53 "},"guide/kubectl-cheatsheet.html":{"url":"guide/kubectl-cheatsheet.html","title":"kubectl命令技巧大全","keywords":"","body":"kubectl 命令技巧大全 Kubctl 命令是操作 kubernetes 集群的最直接和最 skillful 的途径，这个60多MB大小的二进制文件，到底有啥能耐呢？请看下文： Kubectl 自动补全 $ source Kubectl 上下文和配置 设置 kubectl 命令交互的 kubernetes 集群并修改配置信息。参阅 使用 kubeconfig 文件进行跨集群验证 获取关于配置文件的详细信息。 $ kubectl config view # 显示合并后的 kubeconfig 配置 # 同时使用多个 kubeconfig 文件并查看合并后的配置 $ KUBECONFIG=~/.kube/config:~/.kube/kubconfig2 kubectl config view # 获取 e2e 用户的密码 $ kubectl config view -o jsonpath='{.users[?(@.name == \"e2e\")].user.password}' $ kubectl config current-context # 显示当前的上下文 $ kubectl config use-context my-cluster-name # 设置默认上下文为 my-cluster-name # 向 kubeconf 中增加支持基本认证的新集群 $ kubectl config set-credentials kubeuser/foo.kubernetes.com --username=kubeuser --password=kubepassword # 使用指定的用户名和 namespace 设置上下文 $ kubectl config set-context gce --user=cluster-admin --namespace=foo \\ && kubectl config use-context gce 创建对象 Kubernetes 的清单文件可以使用 json 或 yaml 格式定义。可以以 .yaml、.yml、或者 .json 为扩展名。 $ kubectl create -f ./my-manifest.yaml # 创建资源 $ kubectl create -f ./my1.yaml -f ./my2.yaml # 使用多个文件创建资源 $ kubectl create -f ./dir # 使用目录下的所有清单文件来创建资源 $ kubectl create -f https://git.io/vPieo # 使用 url 来创建资源 $ kubectl run nginx --image=nginx # 启动一个 nginx 实例 $ kubectl explain pods,svc # 获取 pod 和 svc 的文档 # 从 stdin 输入中创建多个 YAML 对象 $ cat 显示和查找资源 # Get commands with basic output $ kubectl get services # 列出所有 namespace 中的所有 service $ kubectl get pods --all-namespaces # 列出所有 namespace 中的所有 pod $ kubectl get pods -o wide # 列出所有 pod 并显示详细信息 $ kubectl get deployment my-dep # 列出指定 deployment $ kubectl get pods --include-uninitialized # 列出该 namespace 中的所有 pod 包括未初始化的 # 使用详细输出来描述命令 $ kubectl describe nodes my-node $ kubectl describe pods my-pod $ kubectl get services --sort-by=.metadata.name # List Services Sorted by Name # 根据重启次数排序列出 pod $ kubectl get pods --sort-by='.status.containerStatuses[0].restartCount' # 获取所有具有 app=cassandra 的 pod 中的 version 标签 $ kubectl get pods --selector=app=cassandra rc -o \\ jsonpath='{.items[*].metadata.labels.version}' # 获取所有节点的 ExternalIP $ kubectl get nodes -o jsonpath='{.items[*].status.addresses[?(@.type==\"ExternalIP\")].address}' # 列出属于某个 PC 的 Pod 的名字 # “jq”命令用于转换复杂的 jsonpath，参考 https://stedolan.github.io/jq/ $ sel=${$(kubectl get rc my-rc --output=json | jq -j '.spec.selector | to_entries | .[] | \"\\(.key)=\\(.value),\"')%?} $ echo $(kubectl get pods --selector=$sel --output=jsonpath={.items..metadata.name}) # 查看哪些节点已就绪 $ JSONPATH='{range .items[*]}{@.metadata.name}:{range @.status.conditions[*]}{@.type}={@.status};{end}{end}' \\ && kubectl get nodes -o jsonpath=\"$JSONPATH\" | grep \"Ready=True\" # 列出当前 Pod 中使用的 Secret $ kubectl get pods -o json | jq '.items[].spec.containers[].env[]?.valueFrom.secretKeyRef.name' | grep -v null | sort | uniq 更新资源 $ kubectl rolling-update frontend-v1 -f frontend-v2.json # 滚动更新 pod frontend-v1 $ kubectl rolling-update frontend-v1 frontend-v2 --image=image:v2 # 更新资源名称并更新镜像 $ kubectl rolling-update frontend --image=image:v2 # 更新 frontend pod 中的镜像 $ kubectl rolling-update frontend-v1 frontend-v2 --rollback # 退出已存在的进行中的滚动更新 $ cat pod.json | kubectl replace -f - # 基于 stdin 输入的 JSON 替换 pod # 强制替换，删除后重新创建资源。会导致服务中断。 $ kubectl replace --force -f ./pod.json # 为 nginx RC 创建服务，启用本地 80 端口连接到容器上的 8000 端口 $ kubectl expose rc nginx --port=80 --target-port=8000 # 更新单容器 pod 的镜像版本（tag）到 v4 $ kubectl get pod mypod -o yaml | sed 's/\\(image: myimage\\):.*$/\\1:v4/' | kubectl replace -f - $ kubectl label pods my-pod new-label=awesome # 添加标签 $ kubectl annotate pods my-pod icon-url=http://goo.gl/XXBTWq # 添加注解 $ kubectl autoscale deployment foo --min=2 --max=10 # 自动扩展 deployment “foo” 修补资源 使用策略合并补丁并修补资源。 $ kubectl patch node k8s-node-1 -p '{\"spec\":{\"unschedulable\":true}}' # 部分更新节点 # 更新容器镜像； spec.containers[*].name 是必须的，因为这是合并的关键字 $ kubectl patch pod valid-pod -p '{\"spec\":{\"containers\":[{\"name\":\"kubernetes-serve-hostname\",\"image\":\"new image\"}]}}' # 使用具有位置数组的 json 补丁更新容器镜像 $ kubectl patch pod valid-pod --type='json' -p='[{\"op\": \"replace\", \"path\": \"/spec/containers/0/image\", \"value\":\"new image\"}]' # 使用具有位置数组的 json 补丁禁用 deployment 的 livenessProbe $ kubectl patch deployment valid-deployment --type json -p='[{\"op\": \"remove\", \"path\": \"/spec/template/spec/containers/0/livenessProbe\"}]' 编辑资源 在编辑器中编辑任何 API 资源。 $ kubectl edit svc/docker-registry # 编辑名为 docker-registry 的 service $ KUBE_EDITOR=\"nano\" kubectl edit svc/docker-registry # 使用其它编辑器 Scale 资源 $ kubectl scale --replicas=3 rs/foo # Scale a replicaset named 'foo' to 3 $ kubectl scale --replicas=3 -f foo.yaml # Scale a resource specified in \"foo.yaml\" to 3 $ kubectl scale --current-replicas=2 --replicas=3 deployment/mysql # If the deployment named mysql's current size is 2, scale mysql to 3 $ kubectl scale --replicas=5 rc/foo rc/bar rc/baz # Scale multiple replication controllers 删除资源 $ kubectl delete -f ./pod.json # 删除 pod.json 文件中定义的类型和名称的 pod $ kubectl delete pod,service baz foo # 删除名为“baz”的 pod 和名为“foo”的 service $ kubectl delete pods,services -l name=myLabel # 删除具有 name=myLabel 标签的 pod 和 serivce $ kubectl delete pods,services -l name=myLabel --include-uninitialized # 删除具有 name=myLabel 标签的 pod 和 service，包括尚未初始化的 $ kubectl -n my-ns delete po,svc --all # 删除 my-ns namespace 下的所有 pod 和 serivce，包括尚未初始化的 与运行中的 Pod 交互 $ kubectl logs my-pod # dump 输出 pod 的日志（stdout） $ kubectl logs my-pod -c my-container # dump 输出 pod 中容器的日志（stdout，pod 中有多个容器的情况下使用） $ kubectl logs -f my-pod # 流式输出 pod 的日志（stdout） $ kubectl logs -f my-pod -c my-container # 流式输出 pod 中容器的日志（stdout，pod 中有多个容器的情况下使用） $ kubectl run -i --tty busybox --image=busybox -- sh # 交互式 shell 的方式运行 pod $ kubectl attach my-pod -i # 连接到运行中的容器 $ kubectl port-forward my-pod 5000:6000 # 转发 pod 中的 6000 端口到本地的 5000 端口 $ kubectl exec my-pod -- ls / # 在已存在的容器中执行命令（只有一个容器的情况下） $ kubectl exec my-pod -c my-container -- ls / # 在已存在的容器中执行命令（pod 中有多个容器的情况下） $ kubectl top pod POD_NAME --containers # 显示指定 pod 和容器的指标度量 与节点和集群交互 $ kubectl cordon my-node # 标记 my-node 不可调度 $ kubectl drain my-node # 清空 my-node 以待维护 $ kubectl uncordon my-node # 标记 my-node 可调度 $ kubectl top node my-node # 显示 my-node 的指标度量 $ kubectl cluster-info # 显示 master 和服务的地址 $ kubectl cluster-info dump # 将当前集群状态输出到 stdout $ kubectl cluster-info dump --output-directory=/path/to/cluster-state # 将当前集群状态输出到 /path/to/cluster-state # 如果该键和影响的污点（taint）已存在，则使用指定的值替换 $ kubectl taint nodes foo dedicated=special-user:NoSchedule 资源类型 下表列出的是 kubernetes 中所有支持的类型和缩写的别名。 资源类型 缩写别名 clusters componentstatuses cs configmaps cm daemonsets ds deployments deploy endpoints ep event ev horizontalpodautoscalers hpa ingresses ing jobs limitranges limits namespaces ns networkpolicies nodes no statefulsets persistentvolumeclaims pvc persistentvolumes pv pods po podsecuritypolicies psp podtemplates replicasets rs replicationcontrollers rc resourcequotas quota cronjob secrets serviceaccount sa services svc storageclasses thirdpartyresources 格式化输出 要以特定的格式向终端窗口输出详细信息，可以在 kubectl 命令中添加 -o 或者 -output 标志。 输出格式 描述 -o=custom-columns= 使用逗号分隔的自定义列列表打印表格 -o=custom-columns-file= 使用 文件中的自定义列模板打印表格 -o=json 输出 JSON 格式的 API 对象 -o=jsonpath= 打印 jsonpath 表达式中定义的字段 -o=jsonpath-file= 打印由 文件中的 jsonpath 表达式定义的字段 -o=name 仅打印资源名称 -o=wide 以纯文本格式输出任何附加信息，对于 Pod ，包含节点名称 -o=yaml 输出 YAML 格式的 API 对象 Kubectl 详细输出和调试 使用 -v 或 --v 标志跟着一个整数来指定日志级别。这里 描述了通用的 kubernetes 日志约定和相关的日志级别。 详细等级 描述 --v=0 总是对操作人员可见。 --v=1 合理的默认日志级别，如果您不需要详细输出。 --v=2 可能与系统的重大变化相关的，有关稳定状态的信息和重要的日志信息。这是对大多数系统推荐的日志级别。 --v=3 有关更改的扩展信息。 --v=4 调试级别详细输出。 --v=6 显示请求的资源。 --v=7 显示HTTP请求的header。 --v=8 显示HTTP请求的内容。 参考 Kubectl 概览 JsonPath 手册 本文是对官方文档的中文翻译，原文地址：https://kubernetes.io/docs/user-guide/kubectl-cheatsheet/。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-03 17:45:30 "},"guide/using-etcdctl-to-access-kubernetes-data.html":{"url":"guide/using-etcdctl-to-access-kubernetes-data.html","title":"使用etcdctl访问kuberentes数据","keywords":"","body":"使用etcdctl访问kuberentes数据 Kubenretes1.6中使用etcd V3版本的API，使用etcdctl直接ls的话只能看到/kube-centos一个路径。需要在命令前加上ETCDCTL_API=3这个环境变量才能看到kuberentes在etcd中保存的数据。 ETCDCTL_API=3 etcdctl get /registry/namespaces/default -w=json|python -m json.tool -w指定输出格式 将得到这样的json的结果： { \"count\": 1, \"header\": { \"cluster_id\": 12091028579527406772, \"member_id\": 16557816780141026208, \"raft_term\": 36, \"revision\": 29253467 }, \"kvs\": [ { \"create_revision\": 5, \"key\": \"L3JlZ2lzdHJ5L25hbWVzcGFjZXMvZGVmYXVsdA==\", \"mod_revision\": 5, \"value\": \"azhzAAoPCgJ2MRIJTmFtZXNwYWNlEmIKSAoHZGVmYXVsdBIAGgAiACokZTU2YzMzMDgtMWVhOC0xMWU3LThjZDctZjRlOWQ0OWY4ZWQwMgA4AEILCIn4sscFEKOg9xd6ABIMCgprdWJlcm5ldGVzGggKBkFjdGl2ZRoAIgA=\", \"version\": 1 } ] } 使用--prefix可以看到所有的子目录，如查看集群中的namespace： ETCDCTL_API=3 etcdctl get /registry/namespaces --prefix -w=json|python -m json.tool 输出结果中可以看到所有的namespace。 { \"count\": 8, \"header\": { \"cluster_id\": 12091028579527406772, \"member_id\": 16557816780141026208, \"raft_term\": 36, \"revision\": 29253722 }, \"kvs\": [ { \"create_revision\": 24310883, \"key\": \"L3JlZ2lzdHJ5L25hbWVzcGFjZXMvYXV0b21vZGVs\", \"mod_revision\": 24310883, \"value\": \"azhzAAoPCgJ2MRIJTmFtZXNwYWNlEmQKSgoJYXV0b21vZGVsEgAaACIAKiQ1MjczOTU1ZC1iMzEyLTExZTctOTcwYy1mNGU5ZDQ5ZjhlZDAyADgAQgsI7fSWzwUQ6Jv1Z3oAEgwKCmt1YmVybmV0ZXMaCAoGQWN0aXZlGgAiAA==\", \"version\": 1 }, { \"create_revision\": 21387676, \"key\": \"L3JlZ2lzdHJ5L25hbWVzcGFjZXMvYnJhbmQ=\", \"mod_revision\": 21387676, \"value\": \"azhzAAoPCgJ2MRIJTmFtZXNwYWNlEmEKRwoFYnJhbmQSABoAIgAqJGNkZmQ1Y2NmLWExYzktMTFlNy05NzBjLWY0ZTlkNDlmOGVkMDIAOABCDAjR9qLOBRDYn83XAXoAEgwKCmt1YmVybmV0ZXMaCAoGQWN0aXZlGgAiAA==\", \"version\": 1 }, { \"create_revision\": 5, \"key\": \"L3JlZ2lzdHJ5L25hbWVzcGFjZXMvZGVmYXVsdA==\", \"mod_revision\": 5, \"value\": \"azhzAAoPCgJ2MRIJTmFtZXNwYWNlEmIKSAoHZGVmYXVsdBIAGgAiACokZTU2YzMzMDgtMWVhOC0xMWU3LThjZDctZjRlOWQ0OWY4ZWQwMgA4AEILCIn4sscFEKOg9xd6ABIMCgprdWJlcm5ldGVzGggKBkFjdGl2ZRoAIgA=\", \"version\": 1 }, { \"create_revision\": 18504694, \"key\": \"L3JlZ2lzdHJ5L25hbWVzcGFjZXMvZGV2\", \"mod_revision\": 24310213, \"value\": \"azhzAAoPCgJ2MRIJTmFtZXNwYWNlEmwKUgoDZGV2EgAaACIAKiQyOGRlMGVjNS04ZTEzLTExZTctOTcwYy1mNGU5ZDQ5ZjhlZDAyADgAQgwI89CezQUQ0v2fuQNaCwoEbmFtZRIDZGV2egASDAoKa3ViZXJuZXRlcxoICgZBY3RpdmUaACIA\", \"version\": 4 }, { \"create_revision\": 10, \"key\": \"L3JlZ2lzdHJ5L25hbWVzcGFjZXMva3ViZS1wdWJsaWM=\", \"mod_revision\": 10, \"value\": \"azhzAAoPCgJ2MRIJTmFtZXNwYWNlEmcKTQoLa3ViZS1wdWJsaWMSABoAIgAqJGU1ZjhkY2I1LTFlYTgtMTFlNy04Y2Q3LWY0ZTlkNDlmOGVkMDIAOABCDAiJ+LLHBRDdrsDPA3oAEgwKCmt1YmVybmV0ZXMaCAoGQWN0aXZlGgAiAA==\", \"version\": 1 }, { \"create_revision\": 2, \"key\": \"L3JlZ2lzdHJ5L25hbWVzcGFjZXMva3ViZS1zeXN0ZW0=\", \"mod_revision\": 2, \"value\": \"azhzAAoPCgJ2MRIJTmFtZXNwYWNlEmYKTAoLa3ViZS1zeXN0ZW0SABoAIgAqJGU1NmFhMDVkLTFlYTgtMTFlNy04Y2Q3LWY0ZTlkNDlmOGVkMDIAOABCCwiJ+LLHBRDoq9ASegASDAoKa3ViZXJuZXRlcxoICgZBY3RpdmUaACIA\", \"version\": 1 }, { \"create_revision\": 3774247, \"key\": \"L3JlZ2lzdHJ5L25hbWVzcGFjZXMvc3BhcmstY2x1c3Rlcg==\", \"mod_revision\": 3774247, \"value\": \"azhzAAoPCgJ2MRIJTmFtZXNwYWNlEoABCmYKDXNwYXJrLWNsdXN0ZXISABoAIgAqJDMyNjY3ZDVjLTM0YWMtMTFlNy1iZmJkLThhZjFlM2E3YzViZDIAOABCDAiA1cbIBRDU3YuAAVoVCgRuYW1lEg1zcGFyay1jbHVzdGVyegASDAoKa3ViZXJuZXRlcxoICgZBY3RpdmUaACIA\", \"version\": 1 }, { \"create_revision\": 15212191, \"key\": \"L3JlZ2lzdHJ5L25hbWVzcGFjZXMveWFybi1jbHVzdGVy\", \"mod_revision\": 15212191, \"value\": \"azhzAAoPCgJ2MRIJTmFtZXNwYWNlEn0KYwoMeWFybi1jbHVzdGVyEgAaACIAKiQ2YWNhNjk1Yi03N2Y5LTExZTctYmZiZC04YWYxZTNhN2M1YmQyADgAQgsI1qiKzAUQkoqxDloUCgRuYW1lEgx5YXJuLWNsdXN0ZXJ6ABIMCgprdWJlcm5ldGVzGggKBkFjdGl2ZRoAIgA=\", \"version\": 1 } ] } key的值是经过base64编码，需要解码后才能看到实际值，如： $ echo L3JlZ2lzdHJ5L25hbWVzcGFjZXMvYXV0b21vZGVs|base64 -d /registry/namespaces/automodel 参考 etcd中文文档 etcd官方文档 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-23 13:01:50 "},"guide/cluster-security-management.html":{"url":"guide/cluster-security-management.html","title":"集群安全性管理","keywords":"","body":"集群安全性管理 Kuberentes 支持多租户，这就需要对集群的安全性进行管理。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-09-07 11:46:45 "},"guide/managing-tls-in-a-cluster.html":{"url":"guide/managing-tls-in-a-cluster.html","title":"管理集群中的TLS","keywords":"","body":"管理集群中的TLS 在本书的最佳实践部分，我们在CentOS上部署了kuberentes集群，其中最开始又重要的一步就是创建TLS认证的，查看创建TLS证书和秘钥。很多人在进行到这一步是都会遇到各种各样千奇百怪的问题，这一步是创建集群的基础，我们有必要详细了解一下背后的流程和原理。 概览 每个Kubernetes集群都有一个集群根证书颁发机构（CA）。 集群中的组件通常使用CA来验证API server的证书，由API服务器验证kubelet客户端证书等。为了支持这一点，CA证书包被分发到集群中的每个节点，并作为一个sercret附加分发到默认service account上。 或者，你的workload可以使用此CA建立信任。 你的应用程序可以使用类似于ACME草案的协议，使用certificates.k8s.io API请求证书签名。 集群中的TLS信任 让Pod中运行的应用程序信任集群根CA通常需要一些额外的应用程序配置。 您将需要将CA证书包添加到TLS客户端或服务器信任的CA证书列表中。 例如，您可以使用golang TLS配置通过解析证书链并将解析的证书添加到tls.Config结构中的Certificates字段中，CA证书捆绑包将使用默认服务账户自动加载到pod中，路径为/var/run/secrets/kubernetes.io/serviceaccount/ca.crt。 如果您没有使用默认服务账户，请请求集群管理员构建包含您有权访问使用的证书包的configmap。 请求认证 以下部分演示如何为通过DNS访问的Kubernetes服务创建TLS证书。 步骤0. 下载安装SSL 下载cfssl工具：https://pkg.cfssl.org/. 步骤1. 创建证书签名请求 通过运行以下命令生成私钥和证书签名请求（或CSR）： $ cat 172.168.0.24 是 service 的 cluster IP，my-svc.my-namespace.svc.cluster.local 是 service 的 DNS 名称， 10.0.34.2 是 Pod 的 IP， my-pod.my-namespace.pod.cluster.local 是pod 的 DNS 名称，你可以看到以下输出： 2017/03/21 06:48:17 [INFO] generate received request 2017/03/21 06:48:17 [INFO] received CSR 2017/03/21 06:48:17 [INFO] generating key: ecdsa-256 2017/03/21 06:48:17 [INFO] encoded CSR 此命令生成两个文件; 它生成包含PEM编码的pkcs #10认证请求的server.csr，以及包含仍然要创建的证书的PEM编码密钥的server-key.pem。 步骤2. 创建证书签名请求对象以发送到Kubernetes API 使用以下命令创建CSR yaml文件，并发送到API server： $ cat 请注意，在步骤1中创建的server.csr文件是base64编码并存储在.spec.request字段中。 我们还要求提供“数字签名”，“密钥加密”和“服务器身份验证”密钥用途的证书。 我们这里支持列出的所有关键用途和扩展的关键用途，以便您可以使用相同的API请求客户端证书和其他证书。 在API server中可以看到这些CSR处于pending状态。执行下面的命令你将可以看到： $ kubectl describe csr my-svc.my-namespace Name: my-svc.my-namespace Labels: Annotations: CreationTimestamp: Tue, 21 Mar 2017 07:03:51 -0700 Requesting User: yourname@example.com Status: Pending Subject: Common Name: my-svc.my-namespace.svc.cluster.local Serial Number: Subject Alternative Names: DNS Names: my-svc.my-namespace.svc.cluster.local IP Addresses: 172.168.0.24 10.0.34.2 Events: 步骤3. 获取证书签名请求 批准证书签名请求是通过自动批准过程完成的，或由集群管理员一次完成。 有关这方面涉及的更多信息，请参见下文。 步骤4. 下载签名并使用 一旦CSR被签署并获得批准，您应该看到以下内容： $ kubectl get csr NAME AGE REQUESTOR CONDITION my-svc.my-namespace 10m yourname@example.com Approved,Issued 你可以通过运行以下命令下载颁发的证书并将其保存到server.crt文件中： $ kubectl get csr my-svc.my-namespace -o jsonpath='{.status.certificate}' \\ | base64 -d > server.crt 现在你可以用server.crt和server-key.pem来做为keypair来启动HTTPS server。 批准证书签名请求 Kubernetes 管理员（具有适当权限）可以使用 kubectl certificate approve 和kubectl certificate deny 命令手动批准（或拒绝）证书签名请求。但是，如果您打算大量使用此 API，则可以考虑编写自动化的证书控制器。 如果上述机器或人类使用 kubectl，批准者的作用是验证 CSR 满足如下两个要求： CSR 的主体控制用于签署 CSR 的私钥。这解决了伪装成授权主体的第三方的威胁。在上述示例中，此步骤将验证该 pod 控制了用于生成 CSR 的私钥。 CSR 的主体被授权在请求的上下文中执行。这解决了我们加入群集的我们不期望的主体的威胁。在上述示例中，此步骤将是验证该 pod 是否被允许加入到所请求的服务中。 当且仅当满足这两个要求时，审批者应该批准 CSR，否则拒绝 CSR。 给集群管理员的一个建议 本教程假设将signer设置为服务证书API。Kubernetes controller manager提供了一个signer的默认实现。 要启用它，请将--cluster-signing-cert-file和--cluster-signing-key-file参数传递给controller manager，并配置具有证书颁发机构的密钥对的路径。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-09 14:09:42 "},"guide/kubelet-authentication-authorization.html":{"url":"guide/kubelet-authentication-authorization.html","title":"kubelet的认证授权","keywords":"","body":"Kublet的认证授权 概览 Kubelet 的 HTTPS 端点对外暴露了用于访问不同敏感程度数据的 API，并允许您在节点或者容器内执行不同权限级别的操作。 本文档向您描述如何通过认证授权来访问 kubelet 的 HTTPS 端点。 Kubelet 认证 默认情况下，所有未被配置的其他身份验证方法拒绝的，对 kubelet 的 HTTPS 端点的请求将被视为匿名请求，并被授予 system:anonymous 用户名和 system:unauthenticated 组。 如果要禁用匿名访问并发送 401 Unauthorized 的未经身份验证的请求的响应： 启动 kubelet 时指定 --anonymous-auth=false 标志 如果要对 kubelet 的 HTTPS 端点启用 X509 客户端证书身份验证： 启动 kubelet 时指定 --client-ca-file 标志，提供 CA bundle 以验证客户端证书 启动 apiserver 时指定 --kubelet-client-certificate 和 --kubelet-client-key 标志 参阅 apiserver 认证文档 获取更多详细信息。 启用 API bearer token（包括 service account token）用于向 kubelet 的 HTTPS 端点进行身份验证： 确保在 API server 中开启了 authentication.k8s.io/v1beta1 API 组。 启动 kubelet 时指定 --authentication-token-webhook， --kubeconfig 和 --require-kubeconfig 标志 Kubelet 在配置的 API server 上调用 TokenReview API 以确定来自 bearer token 的用户信息 Kubelet 授权 接着对任何成功验证的请求（包括匿名请求）授权。默认授权模式为 AlwaysAllow，允许所有请求。 细分访问 kubelet API 有很多原因： 启用匿名认证，但匿名用户调用 kubelet API 的能力应受到限制 启动 bearer token 认证，但是 API 用户（如 service account）调用 kubelet API 的能力应受到限制 客户端证书身份验证已启用，但只有那些配置了 CA 签名的客户端证书的用户才可以使用 kubelet API 如果要细分访问 kubelet API，将授权委托给 API server： 确保 API server 中启用了 authorization.k8s.io/v1beta1 API 组 启动 kubelet 时指定 --authorization-mode=Webhook、 --kubeconfig 和 --require-kubeconfig 标志 kubelet 在配置的 API server 上调用 SubjectAccessReview API，以确定每个请求是否被授权 kubelet 使用与 apiserver 相同的 请求属性 方法来授权 API 请求。 Verb（动词）是根据传入的请求的 HTTP 动词确定的： HTTP 动词 request 动词 POST create GET, HEAD get PUT update PATCH patch DELETE delete 资源和子资源根据传入请求的路径确定： Kubelet API 资源 子资源 /stats/* nodes stats /metrics/* nodes metrics /logs/* nodes log /spec/* nodes spec all others nodes proxy Namespace 和 API 组属性总是空字符串，资源的名字总是 kubelet 的 Node API 对象的名字。 当以该模式运行时，请确保用户为 apiserver 指定了 --kubelet-client-certificate 和 --kubelet-client-key 标志并授权了如下属性： verb=*, resource=nodes, subresource=proxy verb=*, resource=nodes, subresource=stats verb=*, resource=nodes, subresource=log verb=*, resource=nodes, subresource=spec verb=*, resource=nodes, subresource=metrics Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-08-21 18:23:34 "},"guide/tls-bootstrapping.html":{"url":"guide/tls-bootstrapping.html","title":"TLS bootstrap","keywords":"","body":"TLS Bootstrap 本文档介绍如何为 kubelet 设置 TLS 客户端证书引导（bootstrap）。 Kubernetes 1.4 引入了一个用于从集群级证书颁发机构（CA）请求证书的 API。此 API 的原始目的是为 kubelet 提供 TLS 客户端证书。可以在 这里 找到该提议，在 feature #43 追踪该功能的进度。 kube-apiserver 配置 您必须提供一个 token 文件，该文件中指定了至少一个分配给 kubelet 特定 bootstrap 组的 “bootstrap token”。 该组将作为 controller manager 配置中的默认批准控制器而用于审批。随着此功能的成熟，您应该确保 token 被绑定到基于角色的访问控制（RBAC）策略上，该策略严格限制了与证书配置相关的客户端请求（使用 bootstrap token）。使用 RBAC，将 token 范围划分为组可以带来很大的灵活性（例如，当您配置完成节点后，您可以禁用特定引导组的访问）。 Token 认证文件 Token 可以是任意的，但应该可以表示为从安全随机数生成器（例如大多数现代操作系统中的 /dev/urandom）导出的至少128位熵。生成 token 有很多中方式。例如： head -c 16 /dev/urandom | od -An -t x | tr -d ' ' 产生的 token 类似于这样： 02b50b05283e98dd0fd71db496ef01e8。 Token 文件应该类似于以下示例，其中前三个值可以是任何值，引用的组名称应如下所示： 02b50b05283e98dd0fd71db496ef01e8,kubelet-bootstrap,10001,\"system:kubelet-bootstrap\" 在 kube-apiserver 命令中添加 --token-auth-file=FILENAME 标志（可能在您的 systemd unit 文件中）来启用 token 文件。 查看 该文档 获取更多详细信息。 客户端证书 CA 包 在 kube-apiserver 命令中添加 --client-ca-file=FILENAME 标志启用客户端证书认证，指定包含签名证书的证书颁发机构包（例如 --client-ca-file=/var/lib/kubernetes/ca.pem）。 kube-controller-manager 配置 请求证书的 API 向 Kubernetes controller manager 中添加证书颁发控制循环。使用磁盘上的 cfssl 本地签名文件的形式。目前，所有发型的证书均为一年有效期和并具有一系列关键用途。 签名文件 您必须提供证书颁发机构，这样才能提供颁发证书所需的密码资料。 kube-apiserver 通过指定的 --client-ca-file=FILENAME 标志来认证和采信该 CA。CA 的管理超出了本文档的范围，但建议您为 Kubernetes 生成专用的 CA。 假定证书和密钥都是 PEM 编码的。 Kube-controller-manager 标志为： --cluster-signing-cert-file=\"/etc/path/to/kubernetes/ca/ca.crt\" --cluster-signing-key-file=\"/etc/path/to/kubernetes/ca/ca.key\" 审批控制器 在 kubernetes 1.7 版本中，实验性的 “组自动批准” 控制器被弃用，新的 csrapproving 控制器将作为 kube-controller-manager 的一部分，被默认启用。 控制器使用 SubjectAccessReview API 来确定给定用户是否已被授权允许请求 CSR，然后根据授权结果进行批准。为了防止与其他批准者冲突，内置审批者没有明确地拒绝 CSR，只是忽略未经授权的请求。 控制器将 CSR 分为三个子资源： nodeclient ：用户的客户端认证请求 O=system:nodes， CN=system:node:(node name)。 selfnodeclient：更新具有相同 O 和 CN 的客户端证书的节点。 selfnodeserver：更新服务证书的节点（ALPHA，需要 feature gate）。 当前，确定 CSR 是否为 selfnodeserver 请求的检查与 kubelet 的凭据轮换实现（Alpha 功能）相关联。因此，selfnodeserver 的定义将来可能会改变，并且需要 Controller Manager 上的RotateKubeletServerCertificate feature gate。该功能的进展可以在 kubernetes/feature/#267 上追踪。 --feature-gates=RotateKubeletServerCertificate=true 以下 RBAC ClusterRoles 代表 nodeClient、selfnodeclient 和 selfnodeserver 功能。在以后的版本中可能会自动创建类似的角色。 # A ClusterRole which instructs the CSR approver to approve a user requesting # node client credentials. kind: ClusterRole apiVersion: rbac.authorization.k8s.io/v1beta1 metadata: name: approve-node-client-csr rules: - apiGroups: [\"certificates.k8s.io\"] resources: [\"certificatesigningrequests/nodeclient\"] verbs: [\"create\"] --- # A ClusterRole which instructs the CSR approver to approve a node renewing its # own client credentials. kind: ClusterRole apiVersion: rbac.authorization.k8s.io/v1beta1 metadata: name: approve-node-client-renewal-csr rules: - apiGroups: [\"certificates.k8s.io\"] resources: [\"certificatesigningrequests/selfnodeclient\"] verbs: [\"create\"] --- # A ClusterRole which instructs the CSR approver to approve a node requesting a # serving cert matching its client cert. kind: ClusterRole apiVersion: rbac.authorization.k8s.io/v1beta1 metadata: name: approve-node-server-renewal-csr rules: - apiGroups: [\"certificates.k8s.io\"] resources: [\"certificatesigningrequests/selfnodeserver\"] verbs: [\"create\"] 这些权力可以授予给凭证，如 bootstrap token。例如，要复制由已被移除的自动批准标志提供的行为，由单个组批准所有的 CSR： # REMOVED: This flag no longer works as of 1.7. --insecure-experimental-approve-all-kubelet-csrs-for-group=\"kubelet-bootstrap-token\" 管理员将创建一个 ClusterRoleBinding 来定位该组。 # Approve all CSRs for the group \"kubelet-bootstrap-token\" kind: ClusterRoleBinding apiVersion: rbac.authorization.k8s.io/v1beta1 metadata: name: auto-approve-csrs-for-group subjects: - kind: Group name: kubelet-bootstrap-token apiGroup: rbac.authorization.k8s.io roleRef: kind: ClusterRole name: approve-node-client-csr apiGroup: rbac.authorization.k8s.io 要让节点更新自己的凭据，管理员可以构造一个 ClusterRoleBinding 来定位该节点的凭据。 kind: ClusterRoleBinding apiVersion: rbac.authorization.k8s.io/v1beta1 metadata: name: node1-client-cert-renewal subjects: - kind: User name: system:node:node-1 # Let \"node-1\" renew its client certificate. apiGroup: rbac.authorization.k8s.io roleRef: kind: ClusterRole name: approve-node-client-renewal-csr apiGroup: rbac.authorization.k8s.io 删除该绑定将会阻止节点更新客户端凭据，一旦其证书到期，实际上就会将其从集群中删除。 kubelet 配置 要向 kube-apiserver 请求客户端证书，kubelet 首先需要一个包含 bootstrap 身份验证 token 的 kubeconfig 文件路径。您可以使用 kubectl config set-cluster，set-credentials 和 set-context 来构建此 kubeconfig 文件。为 kubectl config set-credentials 提供 kubelet-bootstrap 的名称，并包含 --token = ，如下所示： kubectl config set-credentials kubelet-bootstrap --token=${BOOTSTRAP_TOKEN} --kubeconfig=bootstrap.kubeconfig 启动 kubelet 时，如果 --kubeconfig 指定的文件不存在，则使用 bootstrap kubeconfig 向 API server 请求客户端证书。在批准 kubelet 的证书请求和回执时，将包含了生成的密钥和证书的 kubeconfig 文件写入由 -kubeconfig 指定的路径。证书和密钥文件将被放置在由 --cert-dir 指定的目录中。 启动 kubelet 时启用 bootstrap 用到的标志： --experimental-bootstrap-kubeconfig=\"/path/to/bootstrap/kubeconfig\" 此外，在1.7中，kubelet 实现了 Alpha 功能，使其客户端和/或服务器都能轮转提供证书。 可以分别通过 kubelet 中的 RotateKubeletClientCertificate 和 RotateKubeletServerCertificate 功能标志启用此功能，但在未来版本中可能会以向后兼容的方式发生变化。 --feature-gates=RotateKubeletClientCertificate=true,RotateKubeletServerCertificate=true RotateKubeletClientCertificate 可以让 kubelet 在其现有凭据到期时通过创建新的 CSR 来轮换其客户端证书。 RotateKubeletServerCertificate 可以让 kubelet 在其引导客户端凭据后还可以请求服务证书，并轮换该证书。服务证书目前不要求 DNS 或 IP SANs。 kubectl 审批 签名控制器不会立即签署所有证书请求。相反，它会一直等待直到适当特权的用户被标记为 “已批准” 状态。这最终将是由外部审批控制器来处理的自动化过程，但是对于 alpha 版本的 API 来说，可以由集群管理员通过 kubectl 命令手动完成。 管理员可以使用 kubectl get csr 命令列出所有的 CSR，使用 kubectl describe csr 命令描述某个 CSR的详细信息。在 1.6 版本以前，没有直接的批准/拒绝命令 ，因此审批者需要直接更新 Status 信息（查看如何实现）。此后的 Kubernetes 版本中提供了 kubectl certificate approve 和 kubectl certificate deny 命令。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-14 13:44:25 "},"guide/kubectl-user-authentication-authorization.html":{"url":"guide/kubectl-user-authentication-authorization.html","title":"创建用户认证授权的kubeconfig文件","keywords":"","body":"创建用户认证授权的kubeconfig文件 当我们安装好集群后，如果想要把 kubectl 命令交给用户使用，就不得不对用户的身份进行认证和对其权限做出限制。 下面以创建一个 devuser 用户并将其绑定到 dev 和 test 两个 namespace 为例说明。 创建 CA 证书和秘钥 创建 devuser-csr.json 文件 { \"CN\": \"devuser\", \"hosts\": [], \"key\": { \"algo\": \"rsa\", \"size\": 2048 }, \"names\": [ { \"C\": \"CN\", \"ST\": \"BeiJing\", \"L\": \"BeiJing\", \"O\": \"k8s\", \"OU\": \"System\" } ] } 生成 CA 证书和私钥 在 创建 TLS 证书和秘钥 一节中我们将生成的证书和秘钥放在了所有节点的 /etc/kubernetes/ssl 目录下，下面我们再在 master 节点上为 devuser 创建证书和秘钥，在 /etc/kubernetes/ssl 目录下执行以下命令： 执行该命令前请先确保该目录下已经包含如下文件： ca-key.pem ca.pem ca-config.json devuser-csr.json $ cfssl gencert -ca=ca.pem -ca-key=ca-key.pem -config=ca-config.json -profile=kubernetes devuser-csr.json | cfssljson -bare devuser 2017/08/31 13:31:54 [INFO] generate received request 2017/08/31 13:31:54 [INFO] received CSR 2017/08/31 13:31:54 [INFO] generating key: rsa-2048 2017/08/31 13:31:55 [INFO] encoded CSR 2017/08/31 13:31:55 [INFO] signed certificate with serial number 43372632012323103879829229080989286813242051309 2017/08/31 13:31:55 [WARNING] This certificate lacks a \"hosts\" field. This makes it unsuitable for websites. For more information see the Baseline Requirements for the Issuance and Management of Publicly-Trusted Certificates, v.1.1.6, from the CA/Browser Forum (https://cabforum.org); specifically, section 10.2.3 (\"Information Requirements\"). 这将生成如下文件： devuser.csr devuser-key.pem devuser.pem 创建 kubeconfig 文件 # 设置集群参数 export KUBE_APISERVER=\"https://172.20.0.113:6443\" kubectl config set-cluster kubernetes \\ --certificate-authority=/etc/kubernetes/ssl/ca.pem \\ --embed-certs=true \\ --server=${KUBE_APISERVER} \\ --kubeconfig=devuser.kubeconfig # 设置客户端认证参数 kubectl config set-credentials devuser \\ --client-certificate=/etc/kubernetes/ssl/devuser.pem \\ --client-key=/etc/kubernetes/ssl/devuser-key.pem \\ --embed-certs=true \\ --kubeconfig=devuser.kubeconfig # 设置上下文参数 kubectl config set-context kubernetes \\ --cluster=kubernetes \\ --user=devuser \\ --namespace=dev \\ --kubeconfig=devuser.kubeconfig # 设置默认上下文 kubectl config use-context kubernetes --kubeconfig=devuser.kubeconfig 我们现在查看 kubectl 的 context： kubectl config get-contexts CURRENT NAME CLUSTER AUTHINFO NAMESPACE * kubernetes kubernetes admin default-context default-cluster default-admin 显示的用户仍然是 admin，这是因为 kubectl 使用了 $HOME/.kube/config 文件作为了默认的 context 配置，我们只需要将其用刚生成的 devuser.kubeconfig 文件替换即可。 cp -f ./devuser.kubeconfig /root/.kube/config 关于 kubeconfig 文件的更多信息请参考 使用 kubeconfig 文件配置跨集群认证。 ClusterRoleBinding 如果我们想限制 devuser 用户的行为，需要使用 RBAC 将该用户的行为限制在某个或某几个 namespace 空间范围内，例如： kubectl create rolebinding devuser-admin-binding --clusterrole=admin --user=devuser --namespace=dev kubectl create rolebinding devuser-admin-binding --clusterrole=admin --user=devuser --namespace=test 这样 devuser 用户对 dev 和 test 两个 namespace 具有完全访问权限。 让我们来验证以下，现在我们在执行： # 获取当前的 context kubectl config get-contexts CURRENT NAME CLUSTER AUTHINFO NAMESPACE * kubernetes kubernetes devuser dev * kubernetes kubernetes devuser test # 无法访问 default namespace kubectl get pods --namespace default Error from server (Forbidden): User \"devuser\" cannot list pods in the namespace \"default\". (get pods) # 默认访问的是 dev namespace，您也可以重新设置 context 让其默认访问 test namespace kubectl get pods No resources found. 现在 kubectl 命令默认使用的 context 就是 devuser 了，且该用户只能操作 dev 和 test 这两个 namespace，并拥有完全的访问权限。 可以使用我写的create-user.sh脚本来创建namespace和用户并授权，参考说明。 关于角色绑定的更多信息请参考 RBAC——基于角色的访问控制。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-30 19:17:01 "},"guide/rbac.html":{"url":"guide/rbac.html","title":"RBAC——基于角色的访问控制","keywords":"","body":"RBAC——基于角色的访问控制 以下内容是 xingzhou 对 kubernetes 官方文档的翻译，原文地址 https://k8smeetup.github.io/docs/admin/authorization/rbac/ 基于角色的访问控制（Role-Based Access Control, 即”RBAC”）使用”rbac.authorization.k8s.io” API Group实现授权决策，允许管理员通过Kubernetes API动态配置策略。 截至Kubernetes 1.6，RBAC模式处于beta版本。 要启用RBAC，请使用--authorization-mode=RBAC启动API Server。 API概述 本节将介绍RBAC API所定义的四种顶级类型。用户可以像使用其他Kubernetes API资源一样 （例如通过kubectl、API调用等）与这些资源进行交互。例如，命令kubectl create -f (resource).yml 可以被用于以下所有的例子，当然，读者在尝试前可能需要先阅读以下相关章节的内容。 Role与ClusterRole 在RBAC API中，一个角色包含了一套表示一组权限的规则。 权限以纯粹的累加形式累积（没有”否定”的规则）。 角色可以由命名空间（namespace）内的Role对象定义，而整个Kubernetes集群范围内有效的角色则通过ClusterRole对象实现。 一个Role对象只能用于授予对某一单一命名空间中资源的访问权限。 以下示例描述了”default”命名空间中的一个Role对象的定义，用于授予对pod的读访问权限： kind: Role apiVersion: rbac.authorization.k8s.io/v1beta1 metadata: namespace: default name: pod-reader rules: - apiGroups: [\"\"] # 空字符串\"\"表明使用core API group resources: [\"pods\"] verbs: [\"get\", \"watch\", \"list\"] ClusterRole对象可以授予与Role对象相同的权限，但由于它们属于集群范围对象， 也可以使用它们授予对以下几种资源的访问权限： 集群范围资源（例如节点，即node） 非资源类型endpoint（例如”/healthz”） 跨所有命名空间的命名空间范围资源（例如pod，需要运行命令kubectl get pods --all-namespaces来查询集群中所有的pod） 下面示例中的ClusterRole定义可用于授予用户对某一特定命名空间，或者所有命名空间中的secret（取决于其绑定方式）的读访问权限： kind: ClusterRole apiVersion: rbac.authorization.k8s.io/v1beta1 metadata: # 鉴于ClusterRole是集群范围对象，所以这里不需要定义\"namespace\"字段 name: secret-reader rules: - apiGroups: [\"\"] resources: [\"secrets\"] verbs: [\"get\", \"watch\", \"list\"] RoleBinding与ClusterRoleBinding 角色绑定将一个角色中定义的各种权限授予一个或者一组用户。 角色绑定包含了一组相关主体（即subject, 包括用户——User、用户组——Group、或者服务账户——Service Account）以及对被授予角色的引用。 在命名空间中可以通过RoleBinding对象授予权限，而集群范围的权限授予则通过ClusterRoleBinding对象完成。 RoleBinding可以引用在同一命名空间内定义的Role对象。 下面示例中定义的RoleBinding对象在”default”命名空间中将”pod-reader”角色授予用户”jane”。 这一授权将允许用户”jane”从”default”命名空间中读取pod。 # 以下角色绑定定义将允许用户\"jane\"从\"default\"命名空间中读取pod。 kind: RoleBinding apiVersion: rbac.authorization.k8s.io/v1beta1 metadata: name: read-pods namespace: default subjects: - kind: User name: jane apiGroup: rbac.authorization.k8s.io roleRef: kind: Role name: pod-reader apiGroup: rbac.authorization.k8s.io RoleBinding对象也可以引用一个ClusterRole对象用于在RoleBinding所在的命名空间内授予用户对所引用的ClusterRole中 定义的命名空间资源的访问权限。这一点允许管理员在整个集群范围内首先定义一组通用的角色，然后再在不同的命名空间中复用这些角色。 例如，尽管下面示例中的RoleBinding引用的是一个ClusterRole对象，但是用户”dave”（即角色绑定主体）还是只能读取”development” 命名空间中的secret（即RoleBinding所在的命名空间）。 # 以下角色绑定允许用户\"dave\"读取\"development\"命名空间中的secret。 kind: RoleBinding apiVersion: rbac.authorization.k8s.io/v1beta1 metadata: name: read-secrets namespace: development # 这里表明仅授权读取\"development\"命名空间中的资源。 subjects: - kind: User name: dave apiGroup: rbac.authorization.k8s.io roleRef: kind: ClusterRole name: secret-reader apiGroup: rbac.authorization.k8s.io 最后，可以使用ClusterRoleBinding在集群级别和所有命名空间中授予权限。下面示例中所定义的ClusterRoleBinding 允许在用户组”manager”中的任何用户都可以读取集群中任何命名空间中的secret。 # 以下`ClusterRoleBinding`对象允许在用户组\"manager\"中的任何用户都可以读取集群中任何命名空间中的secret。 kind: ClusterRoleBinding apiVersion: rbac.authorization.k8s.io/v1beta1 metadata: name: read-secrets-global subjects: - kind: Group name: manager apiGroup: rbac.authorization.k8s.io roleRef: kind: ClusterRole name: secret-reader apiGroup: rbac.authorization.k8s.io 对资源的引用 大多数资源由代表其名字的字符串表示，例如”pods”，就像它们出现在相关API endpoint的URL中一样。然而，有一些Kubernetes API还 包含了”子资源”，比如pod的logs。在Kubernetes中，pod logs endpoint的URL格式为： GET /api/v1/namespaces/{namespace}/pods/{name}/log 在这种情况下，”pods”是命名空间资源，而”log”是pods的子资源。为了在RBAC角色中表示出这一点，我们需要使用斜线来划分资源 与子资源。如果需要角色绑定主体读取pods以及pod log，您需要定义以下角色： kind: Role apiVersion: rbac.authorization.k8s.io/v1beta1 metadata: namespace: default name: pod-and-pod-logs-reader rules: - apiGroups: [\"\"] resources: [\"pods\", \"pods/log\"] verbs: [\"get\", \"list\"] 通过resourceNames列表，角色可以针对不同种类的请求根据资源名引用资源实例。当指定了resourceNames列表时，不同动作 种类的请求的权限，如使用”get”、”delete”、”update”以及”patch”等动词的请求，将被限定到资源列表中所包含的资源实例上。 例如，如果需要限定一个角色绑定主体只能”get”或者”update”一个configmap时，您可以定义以下角色： kind: Role apiVersion: rbac.authorization.k8s.io/v1beta1 metadata: namespace: default name: configmap-updater rules: - apiGroups: [\"\"] resources: [\"configmap\"] resourceNames: [\"my-configmap\"] verbs: [\"update\", \"get\"] 值得注意的是，如果设置了resourceNames，则请求所使用的动词不能是list、watch、create或者deletecollection。 由于资源名不会出现在create、list、watch和deletecollection等API请求的URL中，所以这些请求动词不会被设置了resourceNames 的规则所允许，因为规则中的resourceNames部分不会匹配这些请求。 一些角色定义的例子 在以下示例中，我们仅截取展示了rules部分的定义。 允许读取core API Group中定义的资源”pods”： rules: - apiGroups: [\"\"] resources: [\"pods\"] verbs: [\"get\", \"list\", \"watch\"] 允许读写在”extensions”和”apps” API Group中定义的”deployments”： rules: - apiGroups: [\"extensions\", \"apps\"] resources: [\"deployments\"] verbs: [\"get\", \"list\", \"watch\", \"create\", \"update\", \"patch\", \"delete\"] 允许读取”pods”以及读写”jobs”： rules: - apiGroups: [\"\"] resources: [\"pods\"] verbs: [\"get\", \"list\", \"watch\"] - apiGroups: [\"batch\", \"extensions\"] resources: [\"jobs\"] verbs: [\"get\", \"list\", \"watch\", \"create\", \"update\", \"patch\", \"delete\"] 允许读取一个名为”my-config”的ConfigMap实例（需要将其通过RoleBinding绑定从而限制针对某一个命名空间中定义的一个ConfigMap实例的访问）： rules: - apiGroups: [\"\"] resources: [\"configmaps\"] resourceNames: [\"my-config\"] verbs: [\"get\"] 允许读取core API Group中的”nodes”资源（由于Node是集群级别资源，所以此ClusterRole定义需要与一个ClusterRoleBinding绑定才能有效）： rules: - apiGroups: [\"\"] resources: [\"nodes\"] verbs: [\"get\", \"list\", \"watch\"] 允许对非资源endpoint “/healthz”及其所有子路径的”GET”和”POST”请求（此ClusterRole定义需要与一个ClusterRoleBinding绑定才能有效）： rules: - nonResourceURLs: [\"/healthz\", \"/healthz/*\"] # 在非资源URL中，'*'代表后缀通配符 verbs: [\"get\", \"post\"] 对角色绑定主体（Subject）的引用 RoleBinding或者ClusterRoleBinding将角色绑定到角色绑定主体（Subject）。 角色绑定主体可以是用户组（Group）、用户（User）或者服务账户（Service Accounts）。 用户由字符串表示。可以是纯粹的用户名，例如”alice”、电子邮件风格的名字，如 “bob@example.com” 或者是用字符串表示的数字id。由Kubernetes管理员配置认证模块 以产生所需格式的用户名。对于用户名，RBAC授权系统不要求任何特定的格式。然而，前缀system:是 为Kubernetes系统使用而保留的，所以管理员应该确保用户名不会意外地包含这个前缀。 Kubernetes中的用户组信息由授权模块提供。用户组与用户一样由字符串表示。Kubernetes对用户组 字符串没有格式要求，但前缀system:同样是被系统保留的。 服务账户拥有包含 system:serviceaccount:前缀的用户名，并属于拥有system:serviceaccounts:前缀的用户组。 角色绑定的一些例子 以下示例中，仅截取展示了RoleBinding的subjects字段。 一个名为”alice@example.com”的用户： subjects: - kind: User name: \"alice@example.com\" apiGroup: rbac.authorization.k8s.io 一个名为”frontend-admins”的用户组： subjects: - kind: Group name: \"frontend-admins\" apiGroup: rbac.authorization.k8s.io kube-system命名空间中的默认服务账户： subjects: - kind: ServiceAccount name: default namespace: kube-system 名为”qa”命名空间中的所有服务账户： subjects: - kind: Group name: system:serviceaccounts:qa apiGroup: rbac.authorization.k8s.io 在集群中的所有服务账户： subjects: - kind: Group name: system:serviceaccounts apiGroup: rbac.authorization.k8s.io 所有认证过的用户（version 1.5+）： subjects: - kind: Group name: system:authenticated apiGroup: rbac.authorization.k8s.io 所有未认证的用户（version 1.5+）： subjects: - kind: Group name: system:unauthenticated apiGroup: rbac.authorization.k8s.io 所有用户（version 1.5+）： subjects: - kind: Group name: system:authenticated apiGroup: rbac.authorization.k8s.io - kind: Group name: system:unauthenticated apiGroup: rbac.authorization.k8s.io 默认角色与默认角色绑定 API Server会创建一组默认的ClusterRole和ClusterRoleBinding对象。 这些默认对象中有许多包含system:前缀，表明这些资源由Kubernetes基础组件”拥有”。 对这些资源的修改可能导致非功能性集群（non-functional cluster）。一个例子是system:node ClusterRole对象。 这个角色定义了kubelets的权限。如果这个角色被修改，可能会导致kubelets无法正常工作。 所有默认的ClusterRole和ClusterRoleBinding对象都会被标记为kubernetes.io/bootstrapping=rbac-defaults。 自动更新 每次启动时，API Server都会更新默认ClusterRole所缺乏的各种权限，并更新默认ClusterRoleBinding所缺乏的各个角色绑定主体。 这种自动更新机制允许集群修复一些意外的修改。由于权限和角色绑定主体在新的Kubernetes释出版本中可能变化，这也能够保证角色和角色 绑定始终保持是最新的。 如果需要禁用自动更新，请将默认ClusterRole以及ClusterRoleBinding的rbac.authorization.kubernetes.io/autoupdate 设置成为false。 请注意，缺乏默认权限和角色绑定主体可能会导致非功能性集群问题。 自Kubernetes 1.6+起，当集群RBAC授权器（RBAC Authorizer）处于开启状态时，可以启用自动更新功能. 发现类角色 默认ClusterRole 默认ClusterRoleBinding 描述 system:basic-user system:authenticated and system:unauthenticatedgroups 允许用户只读访问有关自己的基本信息。 system:discovery system:authenticated and system:unauthenticatedgroups 允许只读访问API discovery endpoints, 用于在API级别进行发现和协商。 面向用户的角色 一些默认角色并不包含system:前缀，它们是面向用户的角色。 这些角色包含超级用户角色（cluster-admin），即旨在利用ClusterRoleBinding（cluster-status）在集群范围内授权的角色， 以及那些使用RoleBinding（admin、edit和view）在特定命名空间中授权的角色。 默认ClusterRole 默认ClusterRoleBinding 描述 cluster-admin system:masters group 超级用户权限，允许对任何资源执行任何操作。 在ClusterRoleBinding中使用时，可以完全控制集群和所有命名空间中的所有资源。 在RoleBinding中使用时，可以完全控制RoleBinding所在命名空间中的所有资源，包括命名空间自己。 admin None 管理员权限，利用RoleBinding在某一命名空间内部授予。 在RoleBinding中使用时，允许针对命名空间内大部分资源的读写访问， 包括在命名空间内创建角色与角色绑定的能力。 但不允许对资源配额（resource quota）或者命名空间本身的写访问。 edit None 允许对某一个命名空间内大部分对象的读写访问，但不允许查看或者修改角色或者角色绑定。 view None 允许对某一个命名空间内大部分对象的只读访问。 不允许查看角色或者角色绑定。 由于可扩散性等原因，不允许查看secret资源。 Core Component Roles 核心组件角色 默认ClusterRole 默认ClusterRoleBinding 描述 system:kube-scheduler system:kube-scheduler user 允许访问kube-scheduler组件所需要的资源。 system:kube-controller-manager system:kube-controller-manager user 允许访问kube-controller-manager组件所需要的资源。 单个控制循环所需要的权限请参阅控制器（controller）角色. system:node system:nodes group (deprecated in 1.7) 允许对kubelet组件所需要的资源的访问，包括读取所有secret和对所有pod的写访问。 自Kubernetes 1.7开始, 相比较于这个角色，更推荐使用Node authorizer 以及NodeRestriction admission plugin， 并允许根据调度运行在节点上的pod授予kubelets API访问的权限。 自Kubernetes 1.7开始，当启用Node授权模式时，对system:nodes用户组的绑定将不会被自动创建。 system:node-proxier system:kube-proxy user 允许对kube-proxy组件所需要资源的访问。 其它组件角色 默认ClusterRole 默认ClusterRoleBinding 描述 system:auth-delegator None 允许委托认证和授权检查。 通常由附加API Server用于统一认证和授权。 system:heapster None Heapster组件的角色。 system:kube-aggregator None kube-aggregator组件的角色。 system:kube-dns kube-dns service account in the kube-systemnamespace kube-dns组件的角色。 system:node-bootstrapper None 允许对执行Kubelet TLS引导（Kubelet TLS bootstrapping）所需要资源的访问. system:node-problem-detector None node-problem-detector组件的角色。 system:persistent-volume-provisioner None 允许对大部分动态存储卷创建组件（dynamic volume provisioner）所需要资源的访问。 控制器（Controller）角色 Kubernetes controller manager负责运行核心控制循环。 当使用--use-service-account-credentials选项运行controller manager时，每个控制循环都将使用单独的服务账户启动。 而每个控制循环都存在对应的角色，前缀名为system:controller:。 如果不使用--use-service-account-credentials选项时，controller manager将会使用自己的凭证运行所有控制循环，而这些凭证必须被授予相关的角色。 这些角色包括： system:controller:attachdetach-controller system:controller:certificate-controller system:controller:cronjob-controller system:controller:daemon-set-controller system:controller:deployment-controller system:controller:disruption-controller system:controller:endpoint-controller system:controller:generic-garbage-collector system:controller:horizontal-pod-autoscaler system:controller:job-controller system:controller:namespace-controller system:controller:node-controller system:controller:persistent-volume-binder system:controller:pod-garbage-collector system:controller:replicaset-controller system:controller:replication-controller system:controller:resourcequota-controller system:controller:route-controller system:controller:service-account-controller system:controller:service-controller system:controller:statefulset-controller system:controller:ttl-controller 初始化与预防权限升级 RBAC API会阻止用户通过编辑角色或者角色绑定来升级权限。 由于这一点是在API级别实现的，所以在RBAC授权器（RBAC authorizer）未启用的状态下依然可以正常工作。 用户只有在拥有了角色所包含的所有权限的条件下才能创建／更新一个角色，这些操作还必须在角色所处的相同范围内进行（对于ClusterRole来说是集群范围，对于Role来说是在与角色相同的命名空间或者集群范围）。 例如，如果用户”user-1”没有权限读取集群范围内的secret列表，那么他也不能创建包含这种权限的ClusterRole。为了能够让用户创建／更新角色，需要： 授予用户一个角色以允许他们根据需要创建／更新Role或者ClusterRole对象。 授予用户一个角色包含他们在Role或者ClusterRole中所能够设置的所有权限。如果用户尝试创建或者修改Role或者ClusterRole以设置那些他们未被授权的权限时，这些API请求将被禁止。 用户只有在拥有所引用的角色中包含的所有权限时才可以创建／更新角色绑定（这些操作也必须在角色绑定所处的相同范围内进行）或者用户被明确授权可以在所引用的角色上执行绑定操作。 例如，如果用户”user-1”没有权限读取集群范围内的secret列表，那么他将不能创建ClusterRole来引用那些授予了此项权限的角色。为了能够让用户创建／更新角色绑定，需要： 授予用户一个角色以允许他们根据需要创建／更新RoleBinding或者ClusterRoleBinding对象。 授予用户绑定某一特定角色所需要的权限： 隐式地，通过授予用户所有所引用的角色中所包含的权限 显式地，通过授予用户在特定Role（或者ClusterRole）对象上执行bind操作的权限 例如，下面例子中的ClusterRole和RoleBinding将允许用户”user-1”授予其它用户”user-1-namespace”命名空间内的admin、edit和view等角色和角色绑定。 apiVersion: rbac.authorization.k8s.io/v1beta1 kind: ClusterRole metadata: name: role-grantor rules: - apiGroups: [\"rbac.authorization.k8s.io\"] resources: [\"rolebindings\"] verbs: [\"create\"] - apiGroups: [\"rbac.authorization.k8s.io\"] resources: [\"clusterroles\"] verbs: [\"bind\"] resourceNames: [\"admin\",\"edit\",\"view\"] --- apiVersion: rbac.authorization.k8s.io/v1beta1 kind: RoleBinding metadata: name: role-grantor-binding namespace: user-1-namespace roleRef: apiGroup: rbac.authorization.k8s.io kind: ClusterRole name: role-grantor subjects: - apiGroup: rbac.authorization.k8s.io kind: User name: user-1 当初始化第一个角色和角色绑定时，初始用户需要能够授予他们尚未拥有的权限。 初始化初始角色和角色绑定时需要： 使用包含system：masters用户组的凭证，该用户组通过默认绑定绑定到cluster-admin超级用户角色。 如果您的API Server在运行时启用了非安全端口（--insecure-port），您也可以通过这个没有施行认证或者授权的端口发送角色或者角色绑定请求。 一些命令行工具 有两个kubectl命令可以用于在命名空间内或者整个集群内授予角色。 kubectl create rolebinding 在某一特定命名空间内授予Role或者ClusterRole。示例如下： 在名为”acme”的命名空间中将admin ClusterRole授予用户”bob”： kubectl create rolebinding bob-admin-binding --clusterrole=admin --user=bob --namespace=acme 在名为”acme”的命名空间中将view ClusterRole授予服务账户”myapp”： kubectl create rolebinding myapp-view-binding --clusterrole=view --serviceaccount=acme:myapp --namespace=acme kubectl create clusterrolebinding 在整个集群中授予ClusterRole，包括所有命名空间。示例如下： 在整个集群范围内将cluster-admin ClusterRole授予用户”root”： kubectl create clusterrolebinding root-cluster-admin-binding --clusterrole=cluster-admin --user=root 在整个集群范围内将system:node ClusterRole授予用户”kubelet”： kubectl create clusterrolebinding kubelet-node-binding --clusterrole=system:node --user=kubelet 在整个集群范围内将view ClusterRole授予命名空间”acme”内的服务账户”myapp”： kubectl create clusterrolebinding myapp-view-binding --clusterrole=view --serviceaccount=acme:myapp 请参阅CLI帮助文档以获得上述命令的详细用法 服务账户（Service Account）权限 默认的RBAC策略将授予控制平面组件（control-plane component）、节点（node）和控制器（controller）一组范围受限的权限， 但对于”kube-system”命名空间以外的服务账户，则不授予任何权限（超出授予所有认证用户的发现权限）。 这一点允许您根据需要向特定服务账号授予特定权限。 细粒度的角色绑定将提供更好的安全性，但需要更多精力管理。 更粗粒度的授权可能授予服务账号不需要的API访问权限（甚至导致潜在授权扩散），但更易于管理。 从最安全到最不安全可以排序以下方法： 对某一特定应用程序的服务账户授予角色（最佳实践） 要求应用程序在其pod规范（pod spec）中指定serviceAccountName字段，并且要创建相应服务账户（例如通过API、应用程序清单或者命令kubectl create serviceaccount等）。 例如，在”my-namespace”命名空间中授予服务账户”my-sa”只读权限： kubectl create rolebinding my-sa-view \\ --clusterrole=view \\ --serviceaccount=my-namespace:my-sa \\ --namespace=my-namespace 在某一命名空间中授予”default”服务账号一个角色 如果一个应用程序没有在其pod规范中指定serviceAccountName，它将默认使用”default”服务账号。 注意：授予”default”服务账号的权限将可用于命名空间内任何没有指定serviceAccountName的pod。 下面的例子将在”my-namespace”命名空间内授予”default”服务账号只读权限： kubectl create rolebinding default-view \\ --clusterrole=view \\ --serviceaccount=my-namespace:default \\ --namespace=my-namespace 目前，许多[加载项（addon）]（/ docs / concepts / cluster-administration / addons /）作为”kube-system”命名空间中的”default”服务帐户运行。 要允许这些加载项使用超级用户访问权限，请将cluster-admin权限授予”kube-system”命名空间中的”default”服务帐户。 注意：启用上述操作意味着”kube-system”命名空间将包含允许超级用户访问API的秘钥。 kubectl create clusterrolebinding add-on-cluster-admin \\ --clusterrole=cluster-admin \\ --serviceaccount=kube-system:default 为命名空间中所有的服务账号授予角色 如果您希望命名空间内的所有应用程序都拥有同一个角色，无论它们使用什么服务账户，您可以为该命名空间的服务账户用户组授予角色。 下面的例子将授予”my-namespace”命名空间中的所有服务账户只读权限： kubectl create rolebinding serviceaccounts-view \\ --clusterrole=view \\ --group=system:serviceaccounts:my-namespace \\ --namespace=my-namespace 对集群范围内的所有服务账户授予一个受限角色（不鼓励） 如果您不想管理每个命名空间的权限，则可以将集群范围角色授予所有服务帐户。 下面的例子将所有命名空间中的只读权限授予集群中的所有服务账户： kubectl create clusterrolebinding serviceaccounts-view \\ --clusterrole=view \\ --group=system:serviceaccounts 授予超级用户访问权限给集群范围内的所有服务帐户（强烈不鼓励） 如果您根本不关心权限分块，您可以对所有服务账户授予超级用户访问权限。 警告：这种做法将允许任何具有读取权限的用户访问secret或者通过创建一个容器的方式来访问超级用户的凭据。 kubectl create clusterrolebinding serviceaccounts-cluster-admin \\ --clusterrole=cluster-admin \\ --group=system:serviceaccounts 从版本1.5升级 在Kubernetes 1.6之前，许多部署使用非常宽泛的ABAC策略，包括授予对所有服务帐户的完整API访问权限。 默认的RBAC策略将授予控制平面组件（control-plane components）、节点（nodes）和控制器（controller）一组范围受限的权限， 但对于”kube-system”命名空间以外的服务账户，则不授予任何权限（超出授予所有认证用户的发现权限）。 虽然安全性更高，但这可能会影响到期望自动接收API权限的现有工作负载。 以下是管理此转换的两种方法： 并行授权器（authorizer） 同时运行RBAC和ABAC授权器，并包括旧版ABAC策略： --authorization-mode=RBAC,ABAC --authorization-policy-file=mypolicy.jsonl RBAC授权器将尝试首先授权请求。如果RBAC授权器拒绝API请求，则ABAC授权器将被运行。这意味着RBAC策略或者ABAC策略所允许的任何请求都是可通过的。 当以日志级别为2或更高（--v = 2）运行时，您可以在API Server日志中看到RBAC拒绝请求信息（以RBAC DENY:为前缀）。 您可以使用该信息来确定哪些角色需要授予哪些用户，用户组或服务帐户。 一旦授予服务帐户角色，并且服务器日志中没有RBAC拒绝消息的工作负载正在运行，您可以删除ABAC授权器。 宽泛的RBAC权限 您可以使用RBAC角色绑定来复制一个宽泛的策略。 警告：以下政策略允许所有服务帐户作为集群管理员。 运行在容器中的任何应用程序都会自动接收服务帐户凭据，并且可以对API执行任何操作，包括查看secret和修改权限。 因此，并不推荐使用这种策略。 kubectl create clusterrolebinding permissive-binding \\ --clusterrole=cluster-admin \\ --user=admin \\ --user=kubelet \\ --group=system:serviceaccounts Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-07 10:23:27 "},"guide/ip-masq-agent.html":{"url":"guide/ip-masq-agent.html","title":"IP伪装代理","keywords":"","body":"IP 伪装代理 本文将讲述如何配置和启用 ip-masq-agent。 创建 ip-masq-agent 要创建 ip-masq-agent，运行下面的 kubectl 命令： kubectl create -f https://raw.githubusercontent.com/kubernetes-incubator/ip-masq-agent/master/ip-masq-agent.yaml 关于 ip-masq-agent 的更多信息请参考 该文档。 在大多数情况下，默认的一套规则应该是足够的；但是，如果内置的规则不适用于您的集群，您可以创建并应用 ConfigMap 来自定义受影响的 IP 范围。例如，为了仅允许 ip-masq-agent 考虑 10.0.0.0/8，您可以在名为 “config” 的文件中创建以下 ConfigMap。 nonMasqueradeCIDRs: - 10.0.0.0/8 resyncInterval: 60s 注意：重要的是，该文件被命名为 config，因为默认情况下，该文件将被用作 ip-masq-agent 查找的关键字。 运行下列命令将 ConfigMap 添加到您的集群中： kubectl create configmap ip-masq-agent --from-file=config --namespace=kube-system 这将会更新 /etc/config/ip-masq-agent 文件，并每隔 resyscInterval 时间段检查一遍该文件，将配置应用到集群的节点中。 iptables -t nat -L IP-MASQ-AGENT Chain IP-MASQ-AGENT (1 references) target prot opt source destination RETURN all -- anywhere 169.254.0.0/16 /* ip-masq-agent: cluster-local traffic should not be subject to MASQUERADE */ ADDRTYPE match dst-type !LOCAL RETURN all -- anywhere 10.0.0.0/8 /* ip-masq-agent: cluster-local MASQUERADE all -- anywhere anywhere /* ip-masq-agent: outbound traffic should be subject to MASQUERADE (this match must come after cluster-local CIDR matches) */ ADDRTYPE match dst-type !LOCAL 默认情况下，本地链路范围（169.254.0.0/16）也由 ip-masq 代理处理，该代理设置相应的 iptables 规则。想要让 ip-masq-agent 忽略本地链路，您可以在 ConfigMap 中将 masqLinkLocal 设置为 true。 nonMasqueradeCIDRs: - 10.0.0.0/8 resyncInterval: 60s masqLinkLocal: true IP 伪装代理用户指南 ip-masq-agent 用户配置 iptables 规则将 Pod 的 IP 地址隐藏在集群 node 节点的 IP 地址后面。这通常在将流量发送到群集的 pod CIDR 范围之外的目的地时执行。 关键术语 NAT（网络地址转换） 是一种通过修改 IP 头中的源和/或目标地址信息来将一个 IP 地址重映射到另一个 IP 地址的方法。通常由执行 IP 路由的设备完成。 Masquerading（伪装） NAT 的一种形式，通常用于执行多个地址转换，其中多个源 IP 地址被掩盖在单个地址之后，通常是由某设备进行 IP 路由。在 kubernetes 中，这是 Node 的 IP 地址。 CIDR（无类域内路由选择） 基于可变长度子网掩码，允许指定任意长度的前缀。CIDR 引入了一种新的 IP 地址表示方法，现在通常被称为 CIDR 表示法，将地址或路由前缀的比特位数作为后缀，例如 192.168.2.0/24。 本地链路 本地链路地址是仅能在主机连接的网段或广播域内进行有效通信的网络地址。IPv4 的链路本地地址在 CIDR 表示法定义的地址块是 169.254.0.0/16。 Ip-masq-agent 在将流量发送到集群 node 节点的 IP 和 Cluster IP 范围之外的目的地时，会配置 iptables 规则来处理伪装的 node/pod IP 地址。这基本上将 pod 的 IP 地址隐藏在了集群 node 节点的 IP 地址后面。在某些环境中，到 “外部” 地址的流量必须来自已知的机器地址。例如，在 Google Cloud 中，到互联网的任何流量必须来自虚拟机的 IP。当使用容器时，如在GKE中，Pod IP 将被拒绝作为出口。为了避免这种情况，我们必须将 Pod IP 隐藏在 VM 自己的 IP 地址之后——通常被称为 “伪装”。默认情况下，配置代理将 RFC 1918指定的三个专用 IP 范围视为非伪装 CIDR。范围包括 10.0.0.0/8、172.16.0.0/12 和 192.168.0.0/16。默认情况下，代理还将本地链路（169.254.0.0/16）视为非伪装 CIDR。代理配置为每隔60秒从 /etc/config/ip-masq-agent 位置重新加载其配置，这也是可配置的。 图片 - IP伪装代理示意图 代理的配置文件必须使用 yaml 或 json 语法，并且包含以下三个可选的 key： nonMasqueradeCIDRs：使用 CIDR 表示法指定的非伪装范围的字符串列表。 masqLinkLocal：一个布尔值（true/false），表示是否将流量伪装成本地链路前缀 169.254.0.0/16。默认为false。 resyncInterval：代理尝试从磁盘重新加载配置的时间间隔。例如 ’30s’ 其中 ‘s’ 是秒，’ms’ 是毫秒等… 到 10.0.0.0/8、172.16.0.0/12 和 192.168.0.0/16 范围的流量将不会被伪装。任何其他流量（假定是互联网）将被伪装。这里有个例子，来自 pod 的本地目的地址可以是其节点的 IP 地址、其他节点的地址或 Cluster IP 范围中的一个 IP 地址。其他任何流量都将默认伪装。以下条目显示 ip-masq-agent 应用的默认规则集： iptables -t nat -L IP-MASQ-AGENT RETURN all -- anywhere 169.254.0.0/16 /* ip-masq-agent: cluster-local traffic should not be subject to MASQUERADE */ ADDRTYPE match dst-type !LOCAL RETURN all -- anywhere 10.0.0.0/8 /* ip-masq-agent: cluster-local traffic should not be subject to MASQUERADE */ ADDRTYPE match dst-type !LOCAL RETURN all -- anywhere 172.16.0.0/12 /* ip-masq-agent: cluster-local traffic should not be subject to MASQUERADE */ ADDRTYPE match dst-type !LOCAL RETURN all -- anywhere 192.168.0.0/16 /* ip-masq-agent: cluster-local traffic should not be subject to MASQUERADE */ ADDRTYPE match dst-type !LOCAL MASQUERADE all -- anywhere anywhere /* ip-masq-agent: outbound traffic should be subject to MASQUERADE (this match must come after cluster-local CIDR matches) */ ADDRTYPE match dst-type !LOCAL 默认情况下，在 GCE/GKE 中将启动 kubernetes 1.7.0 版本，ip-masq-agent 已经在集群中运行。如果您在其他环境中运行 kubernetes，那么您可以将 ip-masq-agent 以 DaemonSet 的方式在集群中运行。 原文地址：https://k8smeetup.github.io/docs/tasks/administer-cluster/ip-masq-agent/ 译者：rootsongjc Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-09-07 14:09:13 "},"guide/auth-with-kubeconfig-or-token.html":{"url":"guide/auth-with-kubeconfig-or-token.html","title":"使用kubeconfig或token进行用户身份认证","keywords":"","body":"使用 kubeconfig 或 token 进行用户身份认证 在开启了 TLS 的集群中，每当与集群交互的时候少不了的是身份认证，使用 kubeconfig（即证书） 和 token 两种认证方式是最简单也最通用的认证方式，在 dashboard 的登录功能就可以使用这两种登录功能。 下文分两块以示例的方式来讲解两种登陆认证方式： 为 brand 命名空间下的 brand 用户创建 kubeconfig 文件 为集群的管理员（拥有所有命名空间的 amdin 权限）创建 token 使用 kubeconfig 如何生成kubeconfig文件请参考创建用户认证授权的kubeconfig文件。 注意我们生成的 kubeconfig 文件中没有 token 字段，需要手动添加该字段。 比如我们为 brand namespace 下的 brand 用户生成了名为 brand.kubeconfig 的 kubeconfig 文件，还要再该文件中追加一行 token 的配置（如何生成 token 将在下文介绍），如下所示： 图片 - kubeconfig文件 对于访问 dashboard 时候的使用 kubeconfig 文件如brand.kubeconfig 必须追到 token 字段，否则认证不会通过。而使用 kubectl 命令时的用的 kubeconfig 文件则不需要包含 token 字段。 生成 token 需要创建一个admin用户并授予admin角色绑定，使用下面的yaml文件创建admin用户并赋予他管理员权限，然后可以通过token访问kubernetes，该文件见admin-role.yaml。 kind: ClusterRoleBinding apiVersion: rbac.authorization.k8s.io/v1beta1 metadata: name: admin annotations: rbac.authorization.kubernetes.io/autoupdate: \"true\" roleRef: kind: ClusterRole name: cluster-admin apiGroup: rbac.authorization.k8s.io subjects: - kind: ServiceAccount name: admin namespace: kube-system --- apiVersion: v1 kind: ServiceAccount metadata: name: admin namespace: kube-system labels: kubernetes.io/cluster-service: \"true\" addonmanager.kubernetes.io/mode: Reconcile 然后执行下面的命令创建 serviceaccount 和角色绑定，对于其他命名空间的其他用户只要修改上述 yaml 中的 name 和 namespace 字段即可： kubectl create -f admin-role.yaml 创建完成后获取secret和token的值。 # 获取admin-token的secret名字 $ kubectl -n kube-system get secret|grep admin-token admin-token-nwphb kubernetes.io/service-account-token 3 6m # 获取token的值 $ kubectl -n kube-system describe secret admin-token-nwphb Name: admin-token-nwphb Namespace: kube-system Labels: Annotations: kubernetes.io/service-account.name=admin kubernetes.io/service-account.uid=f37bd044-bfb3-11e7-87c0-f4e9d49f8ed0 Type: kubernetes.io/service-account-token Data ==== namespace: 11 bytes token: 非常长的字符串 ca.crt: 1310 bytes 也可以使用 jsonpath 的方式直接获取 token 的值，如： kubectl -n kube-system get secret admin-token-nwphb -o jsonpath={.data.token}|base64 -d 注意：yaml 输出里的那个 token 值是进行 base64 编码后的结果，一定要将 kubectl 的输出中的 token 值进行 base64 解码，在线解码工具 base64decode，Linux 和 Mac 有自带的 base64 命令也可以直接使用，输入 base64 是进行编码，Linux 中base64 -d 表示解码，Mac 中使用 base64 -D。 我们使用了 base64 对其重新解码，因为 secret 都是经过 base64 编码的，如果直接使用 kubectl 中查看到的 token 值会认证失败，详见 secret 配置。关于 JSONPath 的使用请参考 JSONPath 手册。 参考 JSONPath 手册 Kubernetes 中的认证 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-11 01:40:46 "},"guide/authentication.html":{"url":"guide/authentication.html","title":" kubernetes中的用户与身份认证授权","keywords":"","body":"Kubernetes 中的用户与身份认证授权 在安装集群的时候我们在 master 节点上生成了一堆证书、token，还在 kubelet 的配置中用到了 bootstrap token，安装各种应用时，为了能够与 API server 通信创建了各种 service account，在 Dashboard 中使用了 kubeconfig 或 token 登陆，那么这些都属于什么认证方式？如何区分用户的？我特地翻译了下这篇官方文档，想你看了之后你将找到答案。 重点查看 bearer token 和 HTTP 认证中的 token 使用，我们已经有所应用，如 使用kubeconfig或token进行用户身份认证。 认识 Kubernetes 中的用户 Kubernetes 集群中包含两类用户：一类是由 Kubernetes 管理的 service account，另一类是普通用户。 普通用户被假定为由外部独立服务管理。管理员分发私钥，用户存储（如 Keystone 或 Google 帐户），甚至包含用户名和密码列表的文件。在这方面，Kubernetes 没有代表普通用户帐户的对象。无法通过 API 调用的方式向集群中添加普通用户。 相对的，service account 是由 Kubernetes API 管理的帐户。它们都绑定到了特定的 namespace，并由 API server 自动创建，或者通过 API 调用手动创建。Service account 关联了一套凭证，存储在 Secret，这些凭证同时被挂载到 pod 中，从而允许 pod 与 kubernetes API 之间的调用。 API 请求被绑定到普通用户或 serivce account 上，或者作为匿名请求对待。这意味着集群内部或外部的每个进程，无论从在工作站上输入 kubectl 的人类用户到节点上的 kubelet，到控制平面的成员，都必须在向 API Server 发出请求时进行身份验证，或者被视为匿名用户。 认证策略 Kubernetes 使用客户端证书、bearer token、身份验证代理或者 HTTP 基本身份验证等身份认证插件来对 API 请求进行身份验证。当有 HTTP 请求发送到 API server 时，插件会尝试将以下属性关联到请求上： 用户名：标识最终用户的字符串。常用值可能是 kube-admin 或 jane@example.com。 UID：标识最终用户的字符串，比用户名更加一致且唯一。 组：一组将用户和常规用户组相关联的字符串。 额外字段：包含其他有用认证信息的字符串列表的映射。 所有的值对于认证系统都是不透明的，只有 授权人 才能解释这些值的重要含义。 您可以一次性启用多种身份验证方式。通常使用至少以下两种认证方式： 服务帐户的 service account token 至少一种其他的用户认证的方式 当启用了多个认证模块时，第一个认证模块成功认证后将短路请求，不会进行第二个模块的认证。API server 不会保证认证的顺序。 system:authenticated 组包含在所有已验证用户的组列表中。 与其他身份验证协议（LDAP、SAML、Kerberos、x509 方案等）的集成可以使用 身份验证代理 或 身份验证 webhook 来实现。 X509 客户端证书 通过将 --client-ca-file=SOMEFILE 选项传递给 API server 来启用客户端证书认证。引用的文件必须包含一个或多个证书颁发机构，用于验证提交给 API server 的客户端证书。如果客户端证书已提交并验证，则使用 subject 的 Common Name（CN）作为请求的用户名。从 Kubernetes 1.4开始，客户端证书还可以使用证书的 organization 字段来指示用户的组成员身份。要为用户包含多个组成员身份，请在证书中包含多个 organization 字段。 例如，使用 openssl 命令工具生成用于签名认证请求的证书： openssl req -new -key jbeda.pem -out jbeda-csr.pem -subj \"/CN=jbeda/O=app1/O=app2\" 这将为一个用户名为 ”jbeda“ 的 CSR，属于两个组“app1”和“app2”。 如何生成客户端证书请参阅 附录。 静态 Token 文件 当在命令行上指定 --token-auth-file=SOMEFILE 选项时，API server 从文件读取 bearer token。目前，token 会无限期地持续下去，并且不重新启动 API server 的话就无法更改令牌列表。 token 文件是一个 csv 文件，每行至少包含三列：token、用户名、用户 uid，其次是可选的组名。请注意，如果您有多个组，则该列必须使用双引号。 token,user,uid,\"group1,group2,group3\" 在请求中放置 Bearer Token 当使用来自 http 客户端的 bearer token 时，API server 期望 Authorization header 中包含 Bearer token 的值。Bearer token 必须是一个字符串序列，只需使用 HTTP 的编码和引用功能就可以将其放入到 HTTP header 中。例如：如果 bearer token 是 31ada4fd-adec-460c-809a-9e56ceb75269，那么它将出现在 HTTP header 中，如下所示： Authorization: Bearer 31ada4fd-adec-460c-809a-9e56ceb75269 Bootstrap Token 该功能仍处于 alpha 版本。 为了简化新集群的初始化引导过程，Kubernetes 中包含了一个名为 Bootstrap Token 的动态管理的 bearer token。这些 token 使用 Secret 存储在 kube-system namespace 中，在那里它们可以被动态管理和创建。Controller Manager 中包含了一个 TokenCleaner 控制器，用于在 bootstrap token 过期时删除将其删除。 这些 token 的形式是 [a-z0-9]{6}.[a-z0-9]{16}。第一部分是 Token ID，第二部分是 Token Secret。您在 HTTP header 中指定的 token 如下所示： Authorization: Bearer 781292.db7bc3a58fc5f07e 在 API server 的启动参数中加上 --experimental-bootstrap-token-auth 标志以启用 Bootstrap Token Authenticator。您必须通过 Controller Manager 上的 --controllers 标志启用 TokenCleaner 控制器，如 --controllers=*,tokencleaner。如果您使用它来引导集群， kubeadm 会为您完成。 认证者认证为 system:bootstrap: 。被包含在 system:bootstrappers 组中。命名和组是有意限制用户使用过去的 bootstap token。可以使用用户名和组（kubeadm 使用）来制定适当的授权策略以支持引导集群。 有关 Bootstrap Token 身份验证器和控制器的更深入的文档，以及如何使用 kubeadm 管理这些令牌，请参阅 Bootstrap Token。 静态密码文件 通过将 --basic-auth-file=SOMEFILE 选项传递给 API server 来启用基本身份验证。目前，基本身份验证凭证将无限期地保留，并且密码在不重新启动API服务器的情况下无法更改。请注意，为了方便起见，目前支持基本身份验证，而上述模式更安全更容易使用。 基本身份认证是一个 csv 文件，至少包含3列：密码、用户名和用户 ID。在 Kubernetes 1.6 和更高版本中，可以指定包含以逗号分隔的组名称的可选第四列。如果您有多个组，则必须将第四列值用双引号（“）括起来，请参阅以下示例： password,user,uid,\"group1,group2,group3\" 当使用来自 HTTP 客户端的基本身份验证时，API server 需要Authorization header 中包含 Basic BASE64ENCODED(USER:PASSWORD) 的值。 Service Account Token Service account 是一个自动启用的验证器，它使用签名的 bearer token 来验证请求。该插件包括两个可选的标志： --service-account-key-file 一个包含签名 bearer token 的 PEM 编码文件。如果未指定，将使用 API server 的 TLS 私钥。 --service-account-lookup 如果启用，从 API 中删除掉的 token 将被撤销。 Service account 通常 API server 自动创建，并通过 ServiceAccount 注入控制器 关联到集群中运行的 Pod 上。Bearer token 挂载到 pod 中众所周知的位置，并允许集群进程与 API server 通信。 帐户可以使用 PodSpec 的 serviceAccountName 字段显式地与Pod关联。 注意： serviceAccountName 通常被省略，因为这会自动生成。 apiVersion: apps/v1beta2 kind: Deployment metadata: name: nginx-deployment namespace: default spec: replicas: 3 template: metadata: # ... spec: containers: - name: nginx image: nginx:1.7.9 serviceAccountName: bob-the-bot Service account bearer token 在集群外使用也是完全有效的，并且可以用于为希望与 Kubernetes 通信的长期运行作业创建身份。要手动创建 service account，只需要使用 kubectl create serviceaccount (NAME) 命令。这将在当前的 namespace 和相关连的 secret 中创建一个 service account。 $ kubectl create serviceaccount jenkins serviceaccount \"jenkins\" created $ kubectl get serviceaccounts jenkins -o yaml apiVersion: v1 kind: ServiceAccount metadata: # ... secrets: - name: jenkins-token-1yvwg 创建出的 secret 中拥有 API server 的公共 CA 和前面的 JSON Web Token（JWT）。 $ kubectl get secret jenkins-token-1yvwg -o yaml apiVersion: v1 data: ca.crt: (APISERVER'S CA BASE64 ENCODED) namespace: ZGVmYXVsdA== token: (BEARER TOKEN BASE64 ENCODED) kind: Secret metadata: # ... type: kubernetes.io/service-account-token 注意：所有值是基于 base64 编码的，因为 secret 总是基于 base64 编码。 经过签名的 JWT 可以用作 bearer token 与给定的 service account 进行身份验证。请参阅 上面 关于如何在请求中放置 bearer token。通常情况下，这些 secret 被挂载到 pod 中，以便对集群内的 API server 进行访问，但也可以从集群外访问。 Service account 验证时用户名 system:serviceaccount:(NAMESPACE):(SERVICEACCOUNT)，被指定到组 system:serviceaccounts 和 system:serviceaccounts:(NAMESPACE)。 注意：由于 service account 的 token 存储在 secret 中，所以具有对这些 secret 的读取权限的任何用户都可以作为 service account 进行身份验证。授予 service account 权限和读取 secret 功能时要谨慎。 OpenID Connect Token OpenID Connect 是由 OAuth2 供应商提供的 OAuth2，特别是 Azure Active Directory、Salesforce 和 Google。对 OAuth2 协议的主要扩展是返回一个称作 ID Token 的格外字段。该 token 是一个 JSON Web Token (JWT) ，有服务器签名，具有众所周知的字段，如用户的电子邮件。 为了识别用户，认证者使用 OAuth2 token 响应 中的 id_token（而不是 access_token）作为 bearer token。请参阅 上面 的关于将 token 置于请求中。 图片 - Kubernetes OpenID Connect Flow 登陆到您的身份提供商 您的身份提供商将为您提供一个 access_token，一个 id_token 和一个 refresh_token 当使用 kubectl 时，使用 --token 标志和 id_token ，或者直接加入到您的 kubeconfig 文件中 kubectl 在调用 API server 时将 id_token 置于 HTTP header 中 API server 将通过检查配置中指定的证书来确保 JWT 签名有效 检查以确保 id_token 没有过期 确保用户已授权 授权 API server 后向 kubectl kubectl 向用户提供反馈 由于所有需要验证您身份的数据都在 id_token 中，Kubernetes 不需要向身份提供商 “phone home”。在每个请求都是无状态的模型中，这为认证提供了非常可扩展的解决方案。它确实提供了一些挑战： Kubernetes 没有 ”web 接口“ 来出发验证进程。没有浏览器或界面来收集凭据，这就是为什么您需要首先认证您的身份提供商。 id_token 无法撤销，就像一个证书，所以它应该是短暂的（只有几分钟），所以每隔几分钟就得到一个新的令牌是非常烦人的。 没有使用 kubectl proxy 命令或注入 id_token 的反向代理，无法简单地对 Kubernetes dashboard 进行身份验证。 配置 API Server 要启用该插件，需要在 API server 中配置如下标志： 参数 描述 示例 是否必需 --oidc-issuer-url 允许 API server 发现公共签名密钥的提供者的 URL。只接受使用 https:// 的方案。通常是提供商的 URL 地址，不包含路径，例如“https://accounts.google.com” 或者 “https://login.salesforce.com”。这个 URL 应该指向下面的 .well-known/openid-configuration 如果发现 URL 是 https://accounts.google.com/.well-known/openid-configuration，值应该是https://accounts.google.com 是 --oidc-client-id 所有的 token 必须为其颁发的客户端 ID kubernetes 是 --oidc-username-claim JWT声明使用的用户名。默认情况下，sub 是最终用户的唯一标识符。管理员可以选择其他声明，如email 或 name，具体取决于他们的提供者。不过，email 以外的其他声明将以发行者的 URL 作为前缀，以防止与其他插件命名冲突。 sub 否 --oidc-groups-claim JWT声明使用的用户组。如果生命存在，它必须是一个字符串数组。 groups 否 --oidc-ca-file 用来签名您的身份提供商的网络 CA 证书的路径。默认为主机的跟 CA。 /etc/kubernetes/ssl/kc-ca.pem 否 如果为 --oidc-username-claim 选择了除 email 以外的其他声明，则该值将以 --oidc-issuer-url 作为前缀，以防止与现有 Kubernetes 名称（例如 system:users）冲突。例如，如果提供商网址是 https://accounts.google.com，而用户名声明映射到 jane，则插件会将用户身份验证为： https://accounts.google.com#jane 重要的是，API server 不是 OAuth2 客户端，而只能配置为信任单个发行者。这允许使用 Google 等公共提供者，而不必信任第三方发行的凭据。希望利用多个 OAuth 客户端的管理员应该探索支持 azp（授权方）声明的提供者，这是允许一个客户端代表另一个客户端发放令牌的机制。 Kubernetes不提供 OpenID Connect 身份提供商。您可以使用现有的公共 OpenID Connect 标识提供程序（例如Google 或 其他）。或者，您可以运行自己的身份提供程序，例如 CoreOS dex、Keycloak、CloudFoundry UAA 或 Tremolo Security 的 OpenUnison。 对于身份提供商能够适用于 Kubernetes，必须满足如下条件：Kubernetes it must: 支持 OpenID connect 发现；不必是全部。 使用非过时密码在TLS中运行 拥有 CA 签名证书（即使 CA 不是商业 CA 或自签名） 有关上述要求3的说明，需要 CA 签名证书。如果您部署自己的身份提供商（而不是像 Google 或 Microsoft 之类的云提供商），则必须让您的身份提供商的 Web 服务器证书由 CA 标志设置为 TRUE 的证书签名，即使是自签名的。这是由于 GoLang 的 TLS 客户端实现对证书验证的标准非常严格。如果您没有 CA，可以使用 CoreOS 团队的 这个脚本 创建一个简单的 CA 和一个签名的证书和密钥对。 或者你可以使用 这个类似的脚本 来生成更长寿命和更长的 SHA256 证书密钥。 针对特定系统的安装说明： UAA Dex OpenUnison 使用 kubectl 选项 1 - OIDC 身份验证器 第一个选项是使用 oidc 身份验证器。此身份验证程序将您的 id_token、refresh_token 和您的 OIDC client_secret 自动刷新您的 token。一旦您对身份提供者进行了身份验证： kubectl config set-credentials USER_NAME \\ --auth-provider=oidc \\ --auth-provider-arg=idp-issuer-url=( issuer url ) \\ --auth-provider-arg=client-id=( your client id ) \\ --auth-provider-arg=client-secret=( your client secret ) \\ --auth-provider-arg=refresh-token=( your refresh token ) \\ --auth-provider-arg=idp-certificate-authority=( path to your ca certificate ) \\ --auth-provider-arg=id-token=( your id_token ) \\ --auth-provider-arg=extra-scopes=( comma separated list of scopes to add to \"openid email profile\", optional ) 例如，在向身份提供者进行身份验证之后运行以下命令： kubectl config set-credentials mmosley \\ --auth-provider=oidc \\ --auth-provider-arg=idp-issuer-url=https://oidcidp.tremolo.lan:8443/auth/idp/OidcIdP \\ --auth-provider-arg=client-id=kubernetes \\ --auth-provider-arg=client-secret=1db158f6-177d-4d9c-8a8b-d36869918ec5 \\ --auth-provider-arg=refresh-token=q1bKLFOyUiosTfawzA93TzZIDzH2TNa2SMm0zEiPKTUwME6BkEo6Sql5yUWVBSWpKUGphaWpxSVAfekBOZbBhaEW+VlFUeVRGcluyVF5JT4+haZmPsluFoFu5XkpXk5BXqHega4GAXlF+ma+vmYpFcHe5eZR+slBFpZKtQA= \\ --auth-provider-arg=idp-certificate-authority=/root/ca.pem \\ --auth-provider-arg=extra-scopes=groups \\ --auth-provider-arg=id-token=eyJraWQiOiJDTj1vaWRjaWRwLnRyZW1vbG8ubGFuLCBPVT1EZW1vLCBPPVRybWVvbG8gU2VjdXJpdHksIEw9QXJsaW5ndG9uLCBTVD1WaXJnaW5pYSwgQz1VUy1DTj1rdWJlLWNhLTEyMDIxNDc5MjEwMzYwNzMyMTUyIiwiYWxnIjoiUlMyNTYifQ.eyJpc3MiOiJodHRwczovL29pZGNpZHAudHJlbW9sby5sYW46ODQ0My9hdXRoL2lkcC9PaWRjSWRQIiwiYXVkIjoia3ViZXJuZXRlcyIsImV4cCI6MTQ4MzU0OTUxMSwianRpIjoiMm96US15TXdFcHV4WDlHZUhQdy1hZyIsImlhdCI6MTQ4MzU0OTQ1MSwibmJmIjoxNDgzNTQ5MzMxLCJzdWIiOiI0YWViMzdiYS1iNjQ1LTQ4ZmQtYWIzMC0xYTAxZWU0MWUyMTgifQ.w6p4J_6qQ1HzTG9nrEOrubxIMb9K5hzcMPxc9IxPx2K4xO9l-oFiUw93daH3m5pluP6K7eOE6txBuRVfEcpJSwlelsOsW8gb8VJcnzMS9EnZpeA0tW_p-mnkFc3VcfyXuhe5R3G7aa5d8uHv70yJ9Y3-UhjiN9EhpMdfPAoEB9fYKKkJRzF7utTTIPGrSaSU6d2pcpfYKaxIwePzEkT4DfcQthoZdy9ucNvvLoi1DIC-UocFD8HLs8LYKEqSxQvOcvnThbObJ9af71EwmuE21fO5KzMW20KtAeget1gnldOosPtz1G5EwvaQ401-RPQzPGMVBld0_zMCAwZttJ4knw 将产生下面的配置： users: - name: mmosley user: auth-provider: config: client-id: kubernetes client-secret: 1db158f6-177d-4d9c-8a8b-d36869918ec5 extra-scopes: groups id-token: eyJraWQiOiJDTj1vaWRjaWRwLnRyZW1vbG8ubGFuLCBPVT1EZW1vLCBPPVRybWVvbG8gU2VjdXJpdHksIEw9QXJsaW5ndG9uLCBTVD1WaXJnaW5pYSwgQz1VUy1DTj1rdWJlLWNhLTEyMDIxNDc5MjEwMzYwNzMyMTUyIiwiYWxnIjoiUlMyNTYifQ.eyJpc3MiOiJodHRwczovL29pZGNpZHAudHJlbW9sby5sYW46ODQ0My9hdXRoL2lkcC9PaWRjSWRQIiwiYXVkIjoia3ViZXJuZXRlcyIsImV4cCI6MTQ4MzU0OTUxMSwianRpIjoiMm96US15TXdFcHV4WDlHZUhQdy1hZyIsImlhdCI6MTQ4MzU0OTQ1MSwibmJmIjoxNDgzNTQ5MzMxLCJzdWIiOiI0YWViMzdiYS1iNjQ1LTQ4ZmQtYWIzMC0xYTAxZWU0MWUyMTgifQ.w6p4J_6qQ1HzTG9nrEOrubxIMb9K5hzcMPxc9IxPx2K4xO9l-oFiUw93daH3m5pluP6K7eOE6txBuRVfEcpJSwlelsOsW8gb8VJcnzMS9EnZpeA0tW_p-mnkFc3VcfyXuhe5R3G7aa5d8uHv70yJ9Y3-UhjiN9EhpMdfPAoEB9fYKKkJRzF7utTTIPGrSaSU6d2pcpfYKaxIwePzEkT4DfcQthoZdy9ucNvvLoi1DIC-UocFD8HLs8LYKEqSxQvOcvnThbObJ9af71EwmuE21fO5KzMW20KtAeget1gnldOosPtz1G5EwvaQ401-RPQzPGMVBld0_zMCAwZttJ4knw idp-certificate-authority: /root/ca.pem idp-issuer-url: https://oidcidp.tremolo.lan:8443/auth/idp/OidcIdP refresh-token: q1bKLFOyUiosTfawzA93TzZIDzH2TNa2SMm0zEiPKTUwME6BkEo6Sql5yUWVBSWpKUGphaWpxSVAfekBOZbBhaEW+VlFUeVRGcluyVF5JT4+haZmPsluFoFu5XkpXk5BXq name: oidc 一旦您的 id_token 过期，kubectl 将使用 refresh_token 刷新 id_token，然后在 kube/.config 文件的client_secret 中存储 id_token 的值和refresh_token 的新值。 选项 2 - 使用 --token 选项 可以在 kubectl 命令的 --token 选项中传入 token。简单的拷贝和复制 id_token 到该选项中： kubectl --token=eyJhbGciOiJSUzI1NiJ9.eyJpc3MiOiJodHRwczovL21sYi50cmVtb2xvLmxhbjo4MDQzL2F1dGgvaWRwL29pZGMiLCJhdWQiOiJrdWJlcm5ldGVzIiwiZXhwIjoxNDc0NTk2NjY5LCJqdGkiOiI2RDUzNXoxUEpFNjJOR3QxaWVyYm9RIiwiaWF0IjoxNDc0NTk2MzY5LCJuYmYiOjE0NzQ1OTYyNDksInN1YiI6Im13aW5kdSIsInVzZXJfcm9sZSI6WyJ1c2VycyIsIm5ldy1uYW1lc3BhY2Utdmlld2VyIl0sImVtYWlsIjoibXdpbmR1QG5vbW9yZWplZGkuY29tIn0.f2As579n9VNoaKzoF-dOQGmXkFKf1FMyNV0-va_B63jn-_n9LGSCca_6IVMP8pO-Zb4KvRqGyTP0r3HkHxYy5c81AnIh8ijarruczl-TK_yF5akjSTHFZD-0gRzlevBDiH8Q79NAr-ky0P4iIXS8lY9Vnjch5MF74Zx0c3alKJHJUnnpjIACByfF2SCaYzbWFMUNat-K1PaUk5-ujMBG7yYnr95xD-63n8CO8teGUAAEMx6zRjzfhnhbzX-ajwZLGwGUBT4WqjMs70-6a7_8gZmLZb2az1cZynkFRj2BaCkVT3A2RrjeEwZEtGXlMqKJ1_I2ulrOVsYx01_yD35-rw get nodes Webhook Token 认证 Webhook 认证是用来认证 bearer token 的 hook。 --authentication-token-webhook-config-file 是一个用来描述如何访问远程 webhook 服务的 kubeconfig 文件。 --authentication-token-webhook-cache-ttl 缓存身份验证策略的时间。默认为两分钟。 配置文件使用 kubeconfig 文件格式。文件中的 ”user“ 指的是 API server 的 webhook，”clusters“ 是指远程服务。见下面的例子： # clusters refers to the remote service. clusters: - name: name-of-remote-authn-service cluster: certificate-authority: /path/to/ca.pem # CA for verifying the remote service. server: https://authn.example.com/authenticate # URL of remote service to query. Must use 'https'. # users refers to the API server's webhook configuration. users: - name: name-of-api-server user: client-certificate: /path/to/cert.pem # cert for the webhook plugin to use client-key: /path/to/key.pem # key matching the cert # kubeconfig files require a context. Provide one for the API server. current-context: webhook contexts: - context: cluster: name-of-remote-authn-service user: name-of-api-sever name: webhook 当客户端尝试使用 bearer token 与API server 进行认证是，如 上 论述，认证 webhook 用饱含该 token 的对象查询远程服务。Kubernetes 不会挑战缺少该 header 的请求。 请注意，webhook API对象与其他 Kubernetes API 对象具有相同的 版本控制兼容性规则。实现者应该意识到 Beta 对象的宽松兼容性承诺，并检查请求的 “apiVersion” 字段以确保正确的反序列化。此外，API server 必须启用 authentication.k8s.io/v1beta1 API 扩展组（--runtime config =authentication.k8s.io/v1beta1=true）。 The request body will be of the following format: { \"apiVersion\": \"authentication.k8s.io/v1beta1\", \"kind\": \"TokenReview\", \"spec\": { \"token\": \"(BEARERTOKEN)\" } } 预计远程服务将填写请求的 status 字段以指示登录成功。响应主体的 spec 字段被忽略，可以省略。成功验证后的 bearer token 将返回： { \"apiVersion\": \"authentication.k8s.io/v1beta1\", \"kind\": \"TokenReview\", \"status\": { \"authenticated\": true, \"user\": { \"username\": \"janedoe@example.com\", \"uid\": \"42\", \"groups\": [ \"developers\", \"qa\" ], \"extra\": { \"extrafield1\": [ \"extravalue1\", \"extravalue2\" ] } } } } 未成功的请求将返回： { \"apiVersion\": \"authentication.k8s.io/v1beta1\", \"kind\": \"TokenReview\", \"status\": { \"authenticated\": false } } HTTP状态代码可以用来提供额外的错误上下文。 认证代理 可以配置 API server 从请求 header 的值中识别用户，例如 X-Remote-User。这样的设计是用来与请求 header 值的验证代理结合使用。 --requestheader-username-headers 必需，大小写敏感。按 header 名称和顺序检查用户标识。包含值的第一个 header 将被作为用户名。 --requestheader-group-headers 1.6 以上版本。可选。大小写敏感。建议为 “X-Remote-Group”。按 header 名称和顺序检查用户组。所有指定的 header 中的所有值都将作为组名。 --requestheader-extra-headers-prefix 1.6 以上版本。可选，大小写敏感。建议为 “X-Remote-Extra-”。标题前缀可用于查找有关用户的额外信息（通常由配置的授权插件使用）。以任何指定的前缀开头的 header 都会删除前缀，header 名称的其余部分将成为额外的键值，而 header 值则是额外的值。 例如下面的配置： --requestheader-username-headers=X-Remote-User --requestheader-group-headers=X-Remote-Group --requestheader-extra-headers-prefix=X-Remote-Extra- 该请求： GET / HTTP/1.1 X-Remote-User: fido X-Remote-Group: dogs X-Remote-Group: dachshunds X-Remote-Extra-Scopes: openid X-Remote-Extra-Scopes: profile 将产生如下的用户信息： name: fido groups: - dogs - dachshunds extra: scopes: - openid - profile 为了防止 header 欺骗，验证代理需要在验证请求 header 之前向 API server 提供有效的客户端证书，以对照指定的 CA 进行验证。 --requestheader-client-ca-file 必需。PEM 编码的证书包。在检查用户名的请求 header 之前，必须针对指定文件中的证书颁发机构提交并验证有效的客户端证书。 --requestheader-allowed-names 可选。Common Name （cn）列表。如果设置了，则在检查用户名的请求 header 之前， 必须提供指定列表中 Common Name（cn）的有效客户端证书。如果为空，则允许使用任何 Common Name。 Keystone 密码 通过在启动过程中将 --experimental-keystone-url= 选项传递给 API server 来启用 Keystone 认证。该插件在 plugin/pkg/auth/authenticator/password/keystone/keystone.go 中实现，目前使用基本身份验证通过用户名和密码验证用户。 如果您为 Keystone 服务器配置了自签名证书，则在启动 Kubernetes API server 时可能需要设置 --experimental-keystone-ca-file=SOMEFILE 选项。如果您设置了该选项，Keystone 服务器的证书将由experimental-keystone-ca-file 中的某个权威机构验证。否则，证书由主机的根证书颁发机构验证。 有关如何使用 keystone 来管理项目和用户的详细信息，请参阅 Keystone 文档。请注意，这个插件仍处于试验阶段，正在积极开发之中，并可能在后续版本中进行更改。 请参考 讨论、蓝图 和 提出的改变 获取更多信息。 匿名请求 启用时，未被其他已配置身份验证方法拒绝的请求将被视为匿名请求，并给予 system:anonymous 的用户名和system:unuthenticated 的组名。 例如，在配置了令牌认证和启用了匿名访问的服务器上，提供无效的 bearer token 的请求将收到 401 Unauthorized 错误。提供 bearer token 的请求将被视为匿名请求。 在 1.5.1 - 1.5.x 版本中，默认情况下命名访问是被禁用的，可以通过传递 --anonymous-auth=false 选项给 API server 来启用。 在 1.6+ 版本中，如果使用 AlwaysAllow 以外的授权模式，则默认启用匿名访问，并且可以通过将 --anonymous-auth=false选项传递给API服务器来禁用。从 1.6 开始，ABAC 和 RBAC 授权人需要明确授权 system:annoymous 或 system:unauthenticated 组，因此授予对 * 用户或 * 组访问权的传统策略规则不包括匿名用户。 用户模拟 用户可以通过模拟 header 充当另一个用户。该请求会覆盖请求认证的用户信息。例如，管理员可以使用此功能通过暂时模拟其他用户并查看请求是否被拒绝来调试授权策略。 模拟请求首先认证为请求用户，然后切换到模拟的用户信息。 用户使用他们的凭证和模拟 header 进行 API 调用。 API server 认证用户 API server 确保经过身份验证的用户具有模拟权限。 请求用户的信息被替换为模拟值 请求被评估，授权作用于模拟的用户信息。 以下 HTTP header 可用户执行模拟请求： Impersonate-User：充当的用户名 Impersonate-Group：作为组名。可以多次使用来设置多个组。可选的，需要 “Impersonate-User” Impersonate-Extra-( extra name )：用于将额外字段与用户关联的动态 header。可选。需要 “Impersonate-User” 一组示例 header： Impersonate-User: jane.doe@example.com Impersonate-Group: developers Impersonate-Group: admins Impersonate-Extra-dn: cn=jane,ou=engineers,dc=example,dc=com Impersonate-Extra-scopes: view Impersonate-Extra-scopes: development 当使用 kubectl 的 --as 标志来配置 Impersonate-User header 时，可以使用 --as-group 标志来配置 Impersonate-Group header。 $ kubectl drain mynode Error from server (Forbidden): User \"clark\" cannot get nodes at the cluster scope. (get nodes mynode) $ kubectl drain mynode --as=superman --as-group=system:masters node \"mynode\" cordoned node \"mynode\" draine 为模仿用户、组或设置额外字段，模拟用户必须能够对正在模拟的属性的种类（“用户”，“组”等）执行“模拟”动词。对于启用了 RBAC 授权插件的集群，以下 ClusterRole 包含设置用户和组模拟 header 所需的规则： apiVersion: rbac.authorization.k8s.io/v1 kind: ClusterRole metadata: name: impersonator rules: - apiGroups: [\"\"] resources: [\"users\", \"groups\", \"serviceaccounts\"] verbs: [\"impersonate\"] 额外的字段被评估为资源 “userextras” 的子资源。为了允许用户使用额外字段 “scope” 的模拟 header，应授予用户以下角色： apiVersion: rbac.authorization.k8s.io/v1 kind: ClusterRole metadata: name: scopes-impersonator # Can set \"Impersonate-Extra-scopes\" header. - apiGroups: [\"authentication.k8s.io\"] resources: [\"userextras/scopes\"] verbs: [\"impersonate\"] 模拟 header 的可用值可以通过设置 resourceNames 可以使用的资源来限制。 apiVersion: rbac.authorization.k8s.io/v1 kind: ClusterRole metadata: name: limited-impersonator rules: # Can impersonate the user \"jane.doe@example.com\" - apiGroups: [\"\"] resources: [\"users\"] verbs: [\"impersonate\"] resourceNames: [\"jane.doe@example.com\"] # Can impersonate the groups \"developers\" and \"admins\" - apiGroups: [\"\"] resources: [\"groups\"] - verbs: [\"impersonate\"] resourceNames: [\"developers\",\"admins\"] # Can impersonate the extras field \"scopes\" with the values \"view\" and \"development\" - apiGroups: [\"authentication.k8s.io\"] resources: [\"userextras/scopes\"] verbs: [\"impersonate\"] resourceNames: [\"view\", \"development\"] 附录 创建证书 使用客户端证书进行身份验证时，可以使用现有的部署脚本或通过 easyrsa 或 openssl 手动生成证书。 使用已有的部署脚本 已有的部署脚本 在 cluster/saltbase/salt/generate-cert/make-ca-cert.sh。 执行该脚本时需要传递两个参数。第一个参数是 API server 的 IP地址。第二个参数是 IP 形式的主题备用名称列表： IP: 或 DNS:。 该脚本将生成三个文件： ca.crt、server.crt 和 server.key。 最后，将一下参数添加到 API server 的启动参数中： --client-ca-file=/srv/kubernetes/ca.crt --tls-cert-file=/srv/kubernetes/server.crt --tls-private-key-file=/srv/kubernetes/server.key easyrsa easyrsa 可以用来手动为集群生成证书。 下载，解压，并初始化修补版本的easyrsa3。 curl -L -O https://storage.googleapis.com/kubernetes-release/easy-rsa/easy-rsa.tar.gz tar xzf easy-rsa.tar.gz cd easy-rsa-master/easyrsa3 ./easyrsa init-pki 生成 CA（使用 --batch 设置为自动模式。使用 --req-cn 设置默认的 CN） ./easyrsa --batch \"--req-cn=${MASTER_IP}@`date +%s`\" build-ca nopass 生成服务器证书和密钥。（build-server-full [文件名]：生成一个键值对，在本地为客户端和服务器签名。） ./easyrsa --subject-alt-name=\"IP:${MASTER_IP}\" build-server-full server nopass 复制 pki/ca.crt, pki/issued/server.crt 和 pki/private/server.key 到您的目录下。 将以下参数添加到 API server 的启动参数中： --client-ca-file=/yourdirectory/ca.crt --tls-cert-file=/yourdirectory/server.crt --tls-private-key-file=/yourdirectory/server.key openssl openssl 可以用来手动为集群生成证书。 生成一个 2048 bit 的 ca.key： openssl genrsa -out ca.key 2048 根据 ca.key 生成一个 ca.crt（使用 -days 设置证书的有效时间）： openssl req -x509 -new -nodes -key ca.key -subj \"/CN=${MASTER_IP}\" -days 10000 -out ca.crt 生成一个 2048 bit 的 server.key： openssl genrsa -out server.key 2048 根据 server.key 生成一个 server.csr： openssl req -new -key server.key -subj \"/CN=${MASTER_IP}\" -out server.csr 根据 ca.key、ca.crt 和 server.csr 生成 server.crt： openssl x509 -req -in server.csr -CA ca.crt -CAkey ca.key -CAcreateserial -out server.crt -days 10000 查看证书： openssl x509 -noout -text -in ./server.crt 最后，不要忘了向 API server 的启动参数中增加配置。 认证 API 您可以使用 certificates.k8s.io API将 x509 证书配置为用于身份验证，如 此处 所述。 官方v1.6文档地址：https://v1-6.docs.kubernetes.io/docs/admin/authentication/ 官方最新文档地址：https://kubernetes.io/docs/admin/authentication/ 译者：Jimmy Song Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-17 15:54:27 "},"guide/access-kubernetes-cluster.html":{"url":"guide/access-kubernetes-cluster.html","title":"访问Kubernetes集群","keywords":"","body":"访问 Kubernetes 集群 根据用户部署和暴露服务的方式不同，有很多种方式可以用来访问 kubernetes 集群。 最简单也是最直接的方式是使用 kubectl 命令。 其次可以使用 kubeconfig 文件来认证授权访问 API server。 通过各种 proxy 经过端口转发访问 kubernetes 集群中的服务 使用 Ingress，在集群外访问 kubernetes 集群内的 service Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-08-21 18:23:34 "},"guide/access-cluster.html":{"url":"guide/access-cluster.html","title":"访问集群","keywords":"","body":"访问集群 第一次使用 kubectl 访问 如果您是第一次访问 Kubernetes API 的话，我们建议您使用 Kubernetes 命令行工具：kubectl。 为了访问集群，您需要知道集群的地址，并且需要有访问它的凭证。通常，如果您完成了 入门指南 那么这些将会自动设置，或者其他人为您部署的集群提供并给您凭证和集群地址。 使用下面的命令检查 kubectl 已知的集群的地址和凭证： $ kubectl config view 关于 kubectl 命令使用的更多 示例 和完整文档可以在这里找到：kubectl 手册 直接访问 REST API Kubectl 处理对 apiserver 的定位和认证。如果您想直接访问 REST API，可以使用像 curl、wget 或浏览器这样的 http 客户端，有以下几种方式来定位和认证： 以 proxy 模式运行 kubectl。 推荐方法。 使用已保存的 apiserver 位置信息。 使用自签名证书验证 apiserver 的身份。 没有 MITM（中间人攻击）的可能。 认证到 apiserver。 将来，可能会做智能的客户端负载均衡和故障转移。 直接向 http 客户端提供位置和凭据。 替代方法。 适用于通过使用代理而混淆的某些类型的客户端代码。 需要将根证书导入浏览器以防止 MITM。 使用 kubectl proxy 以下命令作为反向代理的模式运行 kubectl。 它处理对 apiserver 的定位并进行认证。 像这样运行： $ kubectl proxy --port=8080 & 查看关于 kubectl proxy 的更多细节。 然后您可以使用 curl、wget 或者浏览器来访问 API，如下所示： $ curl http://localhost:8080/api/ { \"versions\": [ \"v1\" ] } 不使用 kubectl proxy（1.3.x 以前版本） 通过将认证 token 直接传递给 apiserver 的方式，可以避免使用 kubectl proxy，如下所示： $ APISERVER=$(kubectl config view | grep server | cut -f 2- -d \":\" | tr -d \" \") $ TOKEN=$(kubectl config view | grep token | cut -f 2 -d \":\" | tr -d \" \") $ curl $APISERVER/api --header \"Authorization: Bearer $TOKEN\" --insecure { \"versions\": [ \"v1\" ] } 不使用 kubectl proxy（1.3.x 以后版本） 在 Kubernetes 1.3 或更高版本中，kubectl config view 不再显示 token。 使用 kubectl describe secret … 获取 default service account 的 token，如下所示： $ APISERVER=$(kubectl config view | grep server | cut -f 2- -d \":\" | tr -d \" \") $ TOKEN=$(kubectl describe secret $(kubectl get secrets | grep default | cut -f1 -d ' ') | grep -E '^token' | cut -f2 -d':' | tr -d '\\t') $ curl $APISERVER/api --header \"Authorization: Bearer $TOKEN\" --insecure { \"kind\": \"APIVersions\", \"versions\": [ \"v1\" ], \"serverAddressByClientCIDRs\": [ { \"clientCIDR\": \"0.0.0.0/0\", \"serverAddress\": \"10.0.1.149:443\" } ] } 以上示例使用--insecure 标志。 这使得它容易受到 MITM 攻击。 当 kubectl 访问集群时，它使用存储的根证书和客户端证书来访问服务器。 （这些安装在~/.kube目录中）。 由于集群证书通常是自签名的，因此可能需要特殊配置才能让您的 http 客户端使用根证书。 对于某些群集，apiserver 可能不需要身份验证；可以选择在本地主机上服务，或者使用防火墙保护。 对此还没有一个标准。配置对API的访问 描述了群集管理员如何配置此操作。 这种方法可能与未来的高可用性支持相冲突。 编程访问 API Kubernetes 支持 Go 和 Python 客户端库。 Go 客户端 要获取该库，请运行以下命令：go get k8s.io/client-go//kubernetes 请参阅 https://github.com/kubernetes/client-go 以查看支持哪些版本。 使用 client-go 客户端编程。请注意，client-go 定义了自己的 API 对象，因此如果需要，请从 client-go 而不是从主存储库导入 API 定义，例如导入 k8s.io/client-go/1.4/pkg/api/v1 是正确的。 Go 客户端可以使用与 kubectl 命令行工具相同的 kubeconfig 文件 来定位和验证 apiserver。参考官方 示例 和 client-go 示例。 如果应用程序在群集中以 Pod 的形式部署，请参考 下一节。 Python 客户端 要使用 Python client，请运行以下命令：pip install kubernetes。查看 Python 客户端库页面 获取更多的安装选择。 Python 客户端可以使用与 kubectl 命令行工具相同的 kubeconfig 文件 来定位和验证 apiserver。参考该 示例。 其他语言 还有更多的 客户端库 可以用来访问 API。有关其他库的验证方式，请参阅文档。 在 Pod 中访问 API 在 Pod 中访问 API 时，定位和认证到 API server 的方式有所不同。在 Pod 中找到 apiserver 地址的推荐方法是使用kubernetes DNS 名称，将它解析为服务 IP，后者又将被路由到 apiserver。 向 apiserver 认证的推荐方法是使用 service account 凭据。通过 kube-system，pod 与 service account 相关联，并且将该 service account 的凭据（token）放入该 pod 中每个容器的文件系统树中，位于 /var/run/secrets/kubernetes.io/serviceaccount/token。 如果可用，证书包将位于每个容器的文件系统树的 /var/run/secrets/kubernetes.io/serviceaccount/ca.crt 位置，并用于验证 apiserver 的服务证书。 最后，用于 namespace API 操作的默认 namespace 放在每个容器中的 /var/run/secrets/kubernetes.io/serviceaccount/namespace 中。 在 pod 中，连接到 API 的推荐方法是： 将 kubectl proxy 作为 pod 中的一个容器来运行，或作为在容器内运行的后台进程。它将 Kubernetes API 代理到 pod 的本地主机接口，以便其他任何 pod 中的容器内的进程都可以访问它。请参阅 在 pod 中使用 kubectl proxy 的示例。 使用 Go 客户端库，并使用 rest.InClusterConfig() 和 kubernetes.NewForConfig() 函数创建一个客户端。 他们处理对 apiserver 的定位和认证。示例 在以上的几种情况下，都需要使用 pod 的凭据与 apiserver 进行安全通信。 访问集群中运行的 service 上一节是关于连接到 kubernetes API server。这一节是关于连接到 kubernetes 集群中运行的 service。在 Kubernetes 中，node、 pod 和 services 都有它们自己的 IP。很多情况下，集群中 node 的 IP、Pod 的 IP、service 的 IP 都是不可路由的，因此在集群外面的机器就无法访问到它们，例如从您自己的笔记本电脑。 连接的方式 您可以选择以下几种方式从集群外部连接到 node、pod 和 service： 通过 public IP 访问 service。 使用 NodePort 和 LoadBalancer 类型的 service，以使 service 能够在集群外部被访问到。请查看 service 和 kubectl expose 文档。 根据您的群集环境，这可能会将服务暴露给您的公司网络，或者可能会将其暴露在互联网上。想想暴露的服务是否安全。它是否自己进行身份验证？ 将 pod 放在服务后面。 要从一组副本（例如为了调试）访问一个特定的 pod，请在 pod 上放置一个唯一的 label，并创建一个选择该 label 的新服务。 在大多数情况下，应用程序开发人员不需要通过 node IP 直接访问节点。 通过 Proxy 规则访问 service、node、pod。 在访问远程服务之前，请执行 apiserver 认证和授权。 如果服务不够安全，无法暴露给互联网，或者为了访问节点 IP 上的端口或进行调试，请使用这种方式。 代理可能会导致某些 Web 应用程序出现问题。 仅适用于 HTTP/HTTPS。 见此描述。 在集群内访问 node 和 pod。 运行一个 pod，然后使用 kubectl exec 命令连接到 shell。从该 shell 中连接到其他 node、pod 和 service。 有些集群可能允许 ssh 到集群上的某个节点。 从那个节点您可以访问到集群中的服务。这是一个非标准的方法，它可能将在某些集群上奏效，而在某些集群不行。这些节点上可能安装了浏览器和其他工具也可能没有。群集 DNS 可能无法正常工作。 访问内置服务 通常集群内会有几个在 kube-system 中启动的服务。使用 kubectl cluster-info 命令获取该列表： $ kubectl cluster-info Kubernetes master is running at https://104.197.5.247 elasticsearch-logging is running at https://104.197.5.247/api/v1/namespaces/kube-system/services/elasticsearch-logging/proxy kibana-logging is running at https://104.197.5.247/api/v1/namespaces/kube-system/services/kibana-logging/proxy kube-dns is running at https://104.197.5.247/api/v1/namespaces/kube-system/services/kube-dns/proxy grafana is running at https://104.197.5.247/api/v1/namespaces/kube-system/services/monitoring-grafana/proxy heapster is running at https://104.197.5.247/api/v1/namespaces/kube-system/services/monitoring-heapster/proxy 这显示了访问每个服务的代理 URL。 例如，此集群启用了集群级日志记录（使用Elasticsearch），如果传入合适的凭据，可以在该地址 https://104.197.5.247/api/v1/namespaces/kube-system/services/elasticsearch-logging/proxy/ 访问到，或通过 kubectl 代理，例如：http://localhost:8080/api/v1/namespaces/kube-system/services/elasticsearch-logging/proxy/。 （有关如何传递凭据和使用 kubectl 代理，请 参阅上文） 手动构建 apiserver 代理 URL 如上所述，您可以使用 kubectl cluster-info 命令来检索服务的代理 URL。要创建包含服务端点、后缀和参数的代理 URL，您只需附加到服务的代理URL： http://kubernetes_master_address/api/v1/namespaces/namespace_name/services/service_name[:port_name]/proxy 如果您没有指定 port 的名字，那么您不必在 URL 里指定 port_name。 示例 要想访问 Elasticsearch 的服务端点 _search?q=user:kimchy，您需要使用：http://104.197.5.247/api/v1/namespaces/kube-system/services/elasticsearch-logging/proxy/_search?q=user:kimchy 要想访问 Elasticsearch 的集群健康信息 _cluster/health?pretty=true，您需要使用：https://104.197.5.247/api/v1/namespaces/kube-system/services/elasticsearch-logging/proxy/_cluster/health?pretty=true { \"cluster_name\" : \"kubernetes_logging\", \"status\" : \"yellow\", \"timed_out\" : false, \"number_of_nodes\" : 1, \"number_of_data_nodes\" : 1, \"active_primary_shards\" : 5, \"active_shards\" : 5, \"relocating_shards\" : 0, \"initializing_shards\" : 0, \"unassigned_shards\" : 5 } 使用 web 浏览器来访问集群中运行的服务 您可以将 apiserver 代理网址放在浏览器的地址栏中。 然而： Web 浏览器通常不能传递 token，因此您可能需要使用基本（密码）认证。 Apiserver 可以配置为接受基本认证，但您的集群可能未配置为接受基本认证。 某些网络应用程序可能无法正常工作，特别是那些在不知道代理路径前缀的情况下构造 URL 的客户端 JavaScript。 请求重定向 重定向功能已被弃用和删除。 请改用代理（见下文）。 多种代理 在使用 kubernetes 的时候您可能会遇到许多种不同的代理： kubectl 代理： 在用户桌面或 pod 中运行 从 localhost 地址到 Kubernetes apiserver 的代理 客户端到代理使用 HTTP apiserver 的代理使用 HTTPS 定位 apiserver 添加身份验证 header apiserver 代理： 将一个堡垒机作为 apiserver 将群集之外的用户连接到群集IP，否则可能无法访问 在 apiserver 进程中运行 客户端到代理使用 HTTPS（或 http，如果 apiserver 如此配置） 根据代理目标的可用信息由代理选择使用 HTTP 或 HTTPS 可用于访问 node、pod 或 service 用于访问 service 时进行负载均衡 kube 代理： 在每个节点上运行 代理 UDP 和 TCP 不支持 HTTP 提供负载均衡 只是用来访问 service apiserver 前面的代理/负载均衡器： 存在和实现因群集而异（例如 nginx） 位于所有客户端和一个或多个 apiserver 之间 作为负载均衡器，如果有多个 apiserver 外部服务的云负载均衡器： 由一些云提供商提供（例如 AWS ELB，Google Cloud Load Balancer） 当 Kubernetes service 类型为 LoadBalancer 时，会自动创建 仅使用 UDP/TCP 实施方式因云提供商而异 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-08-21 18:23:34 "},"guide/authenticate-across-clusters-kubeconfig.html":{"url":"guide/authenticate-across-clusters-kubeconfig.html","title":"使用kubeconfig文件配置跨集群认证","keywords":"","body":"使用 kubeconfig 文件配置跨集群认证 Kubernetes 的认证方式对于不同的人来说可能有所不同。 运行 kubelet 可能有一种认证方式（即证书）。 用户可能有不同的认证方式（即令牌）。 管理员可能具有他们为个人用户提供的证书列表。 我们可能有多个集群，并希望在同一个地方将其全部定义——这样用户就能使用自己的证书并重用相同的全局配置。 所以为了能够让用户轻松地在多个集群之间切换，对于多个用户的情况下，我们将其定义在了一个 kubeconfig 文件中。 此文件包含一系列与昵称相关联的身份验证机制和集群连接信息。它还引入了一个（用户）认证信息元组和一个被称为上下文的与昵称相关联的集群连接信息的概念。 如果明确指定，则允许使用多个 kubeconfig 文件。在运行时，它们与命令行中指定的覆盖选项一起加载并合并（参见下面的 规则）。 相关讨论 http://issue.k8s.io/1755 Kubeconfig 文件的组成 Kubeconifg 文件示例 current-context: federal-context apiVersion: v1 clusters: - cluster: api-version: v1 server: http://cow.org:8080 name: cow-cluster - cluster: certificate-authority: path/to/my/cafile server: https://horse.org:4443 name: horse-cluster - cluster: insecure-skip-tls-verify: true server: https://pig.org:443 name: pig-cluster contexts: - context: cluster: horse-cluster namespace: chisel-ns user: green-user name: federal-context - context: cluster: pig-cluster namespace: saw-ns user: black-user name: queen-anne-context kind: Config preferences: colors: true users: - name: blue-user user: token: blue-token - name: green-user user: client-certificate: path/to/my/client/cert client-key: path/to/my/client/key 各个组件的拆解/释意 Cluster clusters: - cluster: certificate-authority: path/to/my/cafile server: https://horse.org:4443 name: horse-cluster - cluster: insecure-skip-tls-verify: true server: https://pig.org:443 name: pig-cluster cluster 中包含 kubernetes 集群的端点数据，包括 kubernetes apiserver 的完整 url 以及集群的证书颁发机构或者当集群的服务证书未被系统信任的证书颁发机构签名时，设置insecure-skip-tls-verify: true。 cluster 的名称（昵称）作为该 kubeconfig 文件中的集群字典的 key。 您可以使用 kubectl config set-cluster 添加或修改 cluster 条目。 user users: - name: blue-user user: token: blue-token - name: green-user user: client-certificate: path/to/my/client/cert client-key: path/to/my/client/key user 定义用于向 kubernetes 集群进行身份验证的客户端凭据。在加载/合并 kubeconfig 之后，user 将有一个名称（昵称）作为用户条目列表中的 key。 可用凭证有 client-certificate、client-key、token 和 username/password。 username/password 和 token 是二者只能选择一个，但 client-certificate 和 client-key 可以分别与它们组合。 您可以使用 kubectl config set-credentials 添加或者修改 user 条目。 context contexts: - context: cluster: horse-cluster namespace: chisel-ns user: green-user name: federal-context context 定义了一个命名的 cluster、user、namespace 元组，用于使用提供的认证信息和命名空间将请求发送到指定的集群。 三个都是可选的；仅使用 cluster、user、namespace 之一指定上下文，或指定 none。 未指定的值或在加载的 kubeconfig 中没有相应条目的命名值（例如，如果为上述 kubeconfig 文件指定了 pink-user 的上下文）将被替换为默认值。 有关覆盖/合并行为，请参阅下面的 加载和合并规则。 您可以使用 kubectl config set-context 添加或修改上下文条目。 current-context current-context: federal-context current-context is the nickname or 'key' for the cluster,user,namespace tuple that kubectl will use by default when loading config from this file. You can override any of the values in kubectl from the commandline, by passing --context=CONTEXT, --cluster=CLUSTER, --user=USER, and/or --namespace=NAMESPACE respectively. You can change the current-context with kubectl config use-context. current-context 是昵称或者说是作为 cluster、user、namespace 元组的 ”key“，当 kubectl 从该文件中加载配置的时候会被默认使用。您可以在 kubectl 命令行里覆盖这些值，通过分别传入 —context=CONTEXT、 —cluster=CLUSTER、--user=USER 和 --namespace=NAMESPACE 。 您可以使用 kubectl config use-context 更改 current-context。 apiVersion: v1 kind: Config preferences: colors: true 杂项 apiVersion 和 kind 标识客户端解析器的版本和模式，不应手动编辑。 preferences 指定可选（和当前未使用）的 kubectl 首选项。 查看 kubeconfig 文件 kubectl config view 命令可以展示当前的 kubeconfig 设置。默认将为您展示所有的 kubeconfig 设置；您可以通过传入 —minify 参数，将视图过滤到与 current-context 有关的配额设置。有关其他选项，请参阅 kubectl config view。 构建您自己的 kubeconfig 文件 您可以使用上文 示例 kubeconfig 文件 作为 注意： 如果您是通过 kube-up.sh 脚本部署的 kubernetes 集群，不需要自己创建 kubeconfig 文件——该脚本已经为您创建过了。 {:.note} 当 api server 启动的时候使用了 —token-auth-file=tokens.csv 选项时，上述文件将会与 API server 相关联，tokens.csv 文件看起来会像这个样子： blue-user,blue-user,1 mister-red,mister-red,2 注意： 启动 API server 时有很多 可用选项。请您一定要确保理解您使用的选项。 上述示例 kubeconfig 文件提供了 green-user 的客户端凭证。因为用户的 current-user 是 green-user ，任何该 API server 的客户端使用该示例 kubeconfig 文件时都可以成功登录。同样，我们可以通过修改 current-context 的值以 blue-user 的身份操作。 在上面的示例中，green-user 通过提供凭据登录，blue-user 使用的是 token。使用 kubectl config set-credentials 指定登录信息。想了解更多信息，请访问 \"示例文件相关操作命令\"。 加载和合并规则 加载和合并 kubeconfig 文件的规则很简单，但有很多。最终的配置按照以下顺序构建： 从磁盘中获取 kubeconfig。这将通过以下层次结构和合并规则完成： 如果设置了 CommandLineLocation （kubeconfig 命令行参数的值），将会只使用该文件，而不会进行合并。该参数在一条命令中只允许指定一次。 或者，如果设置了 EnvVarLocation （$KUBECONFIG 的值），其将会被作为应合并的文件列表，并根据以下规则合并文件。空文件名被忽略。非串行内容的文件将产生错误。设置特定值或 map key 的第一个文件将优先使用，并且值或 map key 也永远不会更改。 这意味着设置 CurrentContext 的第一个文件将保留其上下文。 这也意味着如果两个文件同时指定一个 red-user，那么将只使用第一个文件中的 red-user 的值。 即使第二个文件的 red-user 中有非冲突条目也被丢弃。 另外，使用 Home 目录位置（~/.kube/config）将不会合并。 根据此链中的第一个命中确定要使用的上下文 命令行参数——context 命令行选项的值 来自合并后的 kubeconfig 文件的 current-context 在这个阶段允许空 确定要使用的群集信息和用户。此时，我们可能有也可能没有上下文。他们是基于这个链中的第一次命中。 （运行两次，一次为用户，一次为集群） 命令行参数——user 指定用户，cluster 指定集群名称 如果上下文存在，则使用上下文的值 允许空 确定要使用的实际群集信息。此时，我们可能有也可能没有集群信息。根据链条构建每个集群信息（第一次命中胜出）： 命令行参数——server，api-version，certificate-authority 和 insecure-skip-tls-verify 如果存在集群信息，并且存在该属性的值，请使用它。 如果没有服务器位置，则产生错误。 确定要使用的实际用户信息。用户使用与集群信息相同的规则构建，除非，您的每个用户只能使用一种认证技术。 负载优先级为1）命令行标志 2）来自 kubeconfig 的用户字段 命令行标志是：client-certificate、client-key、username、password 和 token 如果有两种冲突的技术，则失败。 对于任何仍然缺少的信息，将使用默认值，并可能会提示验证信息 Kubeconfig 文件中的所有文件引用都相对于 kubeconfig 文件本身的位置进行解析。当命令行上显示文件引用时，它们将相对于当前工作目录进行解析。当路径保存在 ~/.kube/config 中时，相对路径使用相对存储，绝对路径使用绝对存储。 Kubeconfig 文件中的任何路径都相对于 kubeconfig 文件本身的位置进行解析。 使用 kubectl config 操作 kubeconfig kubectl config 有一些列的子命令可以帮助我们更方便的操作 kubeconfig 文件。 请参阅 kubectl/kubectl_config。 Example $ kubectl config set-credentials myself --username=admin --password=secret $ kubectl config set-cluster local-server --server=http://localhost:8080 $ kubectl config set-context default-context --cluster=local-server --user=myself $ kubectl config use-context default-context $ kubectl config set contexts.default-context.namespace the-right-prefix $ kubectl config view 产生如下输出： apiVersion: v1 clusters: - cluster: server: http://localhost:8080 name: local-server contexts: - context: cluster: local-server namespace: the-right-prefix user: myself name: default-context current-context: default-context kind: Config preferences: {} users: - name: myself user: password: secret username: admin Kubeconfig 文件会像这样子： apiVersion: v1 clusters: - cluster: server: http://localhost:8080 name: local-server contexts: - context: cluster: local-server namespace: the-right-prefix user: myself name: default-context current-context: default-context kind: Config preferences: {} users: - name: myself user: password: secret username: admin 示例文件相关操作命令 $ kubectl config set preferences.colors true $ kubectl config set-cluster cow-cluster --server=http://cow.org:8080 --api-version=v1 $ kubectl config set-cluster horse-cluster --server=https://horse.org:4443 --certificate-authority=path/to/my/cafile $ kubectl config set-cluster pig-cluster --server=https://pig.org:443 --insecure-skip-tls-verify=true $ kubectl config set-credentials blue-user --token=blue-token $ kubectl config set-credentials green-user --client-certificate=path/to/my/client/cert --client-key=path/to/my/client/key $ kubectl config set-context queen-anne-context --cluster=pig-cluster --user=black-user --namespace=saw-ns $ kubectl config set-context federal-context --cluster=horse-cluster --user=green-user --namespace=chisel-ns $ kubectl config use-context federal-context 最后将它们捆绑在一起 所以，将这一切绑在一起，快速创建自己的 kubeconfig 文件： 仔细看一下，了解您的 api-server 的启动方式：在设计 kubeconfig 文件以方便身份验证之前，您需要知道您自己的安全要求和策略。 将上面的代码段替换为您的集群的 api-server 端点的信息。 确保您的 api-server 至少能够以提供一个用户（即 green-user）凭据的方式启动。 当然您必须查看 api-server 文档，以了解当前关于身份验证细节方面的最新技术。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-08-21 18:23:34 "},"guide/connecting-to-applications-port-forward.html":{"url":"guide/connecting-to-applications-port-forward.html","title":"通过端口转发访问集群中的应用程序","keywords":"","body":"通过端口转发访问集群中的应用程序 本页向您展示如何使用 kubectl port-forward 命令连接到运行在 Kubernetes 集群中的 Redis 服务器。这种类型的连接对于数据库调试很有帮助。 创建一个 Pod 来运行 Redis 服务器 创建一个 Pod： kubectl create -f https://k8s.io/docs/tasks/access-application-cluster/redis-master.yaml 命令运行成功后将有以下输出验证该 Pod 是否已经创建： pod \"redis-master\" created 检查 Pod 是否正在运行且处于就绪状态： kubectl get pods 当 Pod 就绪，输出显示 Running 的状态： NAME READY STATUS RESTARTS AGE redis-master 2/2 Running 0 41s 验证 Redis 服务器是否已在 Pod 中运行，并监听 6379 端口： kubectl get pods redis-master --template='{{(index (index .spec.containers 0).ports 0).containerPort}}{{\"\\n\"}}' 端口输出如下： 6379 将本地端口转发到 Pod 中的端口 将本地工作站上的 6379 端口转发到 redis-master pod 的 6379 端口： kubectl port-forward redis-master 6379:6379 输出类似于： I0710 14:43:38.274550 3655 portforward.go:225] Forwarding from 127.0.0.1:6379 -> 6379 I0710 14:43:38.274797 3655 portforward.go:225] Forwarding from [::1]:6379 -> 6379 启动 Redis 命令行界面 redis-cli 在 Redis 命令行提示符下，输入 ping 命令： 127.0.0.1:6379>ping Ping 请求成功返回 PONG。 讨论 创建连接，将本地的 6379 端口转发到运行在 Pod 中的 Redis 服务器的 6379 端口。有了这个连接您就可以在本地工作站中调试运行在 Pod 中的数据库。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-08-21 18:23:34 "},"guide/service-access-application-cluster.html":{"url":"guide/service-access-application-cluster.html","title":"使用service访问群集中的应用程序","keywords":"","body":"使用 service 访问群集中的应用程序 本文向您展示如何创建 Kubernetes Service 对象，外部客户端可以使用它来访问集群中运行的应用程序。该 Service 可以为具有两个运行实例的应用程序提供负载均衡。 目的 运行 Hello World 应用程序的两个实例。 创建一个暴露 node 节点端口的 Service 对象。 使用 Service 对象访问正在运行的应用程序。 为在两个 pod 中运行的应用程序创建 service 在集群中运行 Hello World 应用程序： kubectl run hello-world --replicas=2 --labels=\"run=load-balancer-example\" --image=gcr.io/google-samples/node-hello:1.0 --port=8080 上述命令创建一个 Deployment 对象和一个相关联的 ReplicaSet 对象。该 ReplicaSet 有两个 Pod，每个 Pod 中都运行一个 Hello World 应用程序。 显示关于该 Deployment 的信息： kubectl get deployments hello-world kubectl describe deployments hello-world 显示 ReplicaSet 的信息： kubectl get replicasets kubectl describe replicasets 创建一个暴露该 Deployment 的 Service 对象： kubectl expose deployment hello-world --type=NodePort --name=example-service 显示该 Service 的信息： kubectl describe services example-service 输出类似于： Name: example-service Namespace: default Labels: run=load-balancer-example Selector: run=load-balancer-example Type: NodePort IP: 10.32.0.16 Port: 8080/TCP NodePort: 31496/TCP Endpoints: 10.200.1.4:8080,10.200.2.5:8080 Session Affinity: None No events. 记下服务的 NodePort 值。例如，在前面的输出中，NodePort 值为 31496。 列出运行 Hello World 应用程序的 Pod： kubectl get pods --selector=\"run=load-balancer-example\" --output=wide 输出类似于： NAME READY STATUS ... IP NODE hello-world-2895499144-bsbk5 1/1 Running ... 10.200.1.4 worker1 hello-world-2895499144-m1pwt 1/1 Running ... 10.200.2.5 worker2 获取正在运行 Hello World 应用程序的 Pod 的其中一个节点的 public IP 地址。如何得到这个地址取决于您的集群设置。例如，如果您使用 Minikube，可以通过运行 kubectl cluster-info 查看节点地址。如果您是使用 Google Compute Engine 实例，可以使用 gcloud compute instances list 命令查看您的公共地址节点。 在您选择的节点上，在您的节点端口上例如创建允许 TCP 流量的防火墙规则，如果您的服务 NodePort 值为 31568，创建防火墙规则，允许端口 31568 上的TCP流量。 使用节点地址和节点端口访问 Hello World 应用程序： curl http://: 其中 是您节点的 public IP地址，而 是您服务的 NodePort 值。 对成功请求的响应是一个 hello 消息： Hello Kubernetes! 使用 Service 配置文件 作为使用 kubectl expose 的替代方法，您可以使用 service 配置文件 来创建 Service。 要删除 Service，输入以下命令： kubectl delete services example-service 删除 Deployment、ReplicaSet 和正运行在 Pod 中的 Hello World 应用程序，输入以下命令： kubectl delete deployment hello-world 了解更多关于 使用 service 连接应用程序。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-08-21 18:23:34 "},"guide/accessing-kubernetes-pods-from-outside-of-the-cluster.html":{"url":"guide/accessing-kubernetes-pods-from-outside-of-the-cluster.html","title":"从外部访问Kubernetes中的Pod","keywords":"","body":"从外部访问Kubernetes中的Pod 前面几节讲到如何访问kubneretes集群，本文主要讲解访问kubenretes中的Pod和Serivce的集中方式，包括如下几种： hostNetwork hostPort NodePort LoadBalancer Ingress 说是暴露Pod其实跟暴露Service是一回事，因为Pod就是Service的backend。 hostNetwork: true 这是一种直接定义Pod网络的方式。 如果在Pod中使用hostNotwork:true配置的话，在这种pod中运行的应用程序可以直接看到pod启动的主机的网络接口。在主机的所有网络接口上都可以访问到该应用程序。以下是使用主机网络的pod的示例定义： apiVersion: v1 kind: Pod metadata: name: influxdb spec: hostNetwork: true containers: - name: influxdb image: influxdb 部署该Pod： $ kubectl create -f influxdb-hostnetwork.yml 访问该pod所在主机的8086端口： curl -v http://$POD_IP:8086/ping 将看到204 No Content的204返回码，说明可以正常访问。 注意每次启动这个Pod的时候都可能被调度到不同的节点上，所有外部访问Pod的IP也是变化的，而且调度Pod的时候还需要考虑是否与宿主机上的端口冲突，因此一般情况下除非您知道需要某个特定应用占用特定宿主机上的特定端口时才使用hostNetwork: true的方式。 这种Pod的网络模式有一个用处就是可以将网络插件包装在Pod中然后部署在每个宿主机上，这样该Pod就可以控制该宿主机上的所有网络。 hostPort 这是一种直接定义Pod网络的方式。 hostPort是直接将容器的端口与所调度的节点上的端口路由，这样用户就可以通过宿主机的IP加上来访问Pod了，如:。 apiVersion: v1 kind: Pod metadata: name: influxdb spec: containers: - name: influxdb image: influxdb ports: - containerPort: 8086 hostPort: 8086 这样做有个缺点，因为Pod重新调度的时候该Pod被调度到的宿主机可能会变动，这样就变化了，用户必须自己维护一个Pod与所在宿主机的对应关系。 这种网络方式可以用来做 nginx Ingress controller。外部流量都需要通过kubenretes node节点的80和443端口。 NodePort NodePort在kubenretes里是一个广泛应用的服务暴露方式。Kubernetes中的service默认情况下都是使用的ClusterIP这种类型，这样的service会产生一个ClusterIP，这个IP只能在集群内部访问，要想让外部能够直接访问service，需要将service type修改为 nodePort。 apiVersion: v1 kind: Pod metadata: name: influxdb labels: name: influxdb spec: containers: - name: influxdb image: influxdb ports: - containerPort: 8086 同时还可以给service指定一个nodePort值，范围是30000-32767，这个值在API server的配置文件中，用--service-node-port-range定义。 kind: Service apiVersion: v1 metadata: name: influxdb spec: type: NodePort ports: - port: 8086 nodePort: 30000 selector: name: influxdb 集群外就可以使用kubernetes任意一个节点的IP加上30000端口访问该服务了。kube-proxy会自动将流量以round-robin的方式转发给该service的每一个pod。 这种服务暴露方式，无法让你指定自己想要的应用常用端口，不过可以在集群上再部署一个反向代理作为流量入口。 LoadBalancer LoadBalancer 只能在service上定义。这是公有云提供的负载均衡器，如AWS、Azure、CloudStack、GCE等。 kind: Service apiVersion: v1 metadata: name: influxdb spec: type: LoadBalancer ports: - port: 8086 selector: name: influxdb 查看服务： $ kubectl get svc influxdb NAME CLUSTER-IP EXTERNAL-IP PORT(S) AGE influxdb 10.97.121.42 10.13.242.236 8086:30051/TCP 39s 内部可以使用ClusterIP加端口来访问服务，如19.97.121.42:8086。 外部可以用以下两种方式访问该服务： 使用任一节点的IP加30051端口访问该服务 使用EXTERNAL-IP来访问，这是一个VIP，是云供应商提供的负载均衡器IP，如10.13.242.236:8086。 Ingress Ingress是自kubernetes1.1版本后引入的资源类型。必须要部署Ingress controller才能创建Ingress资源，Ingress controller是以一种插件的形式提供。Ingress controller 是部署在Kubernetes之上的Docker容器。它的Docker镜像包含一个像nginx或HAProxy的负载均衡器和一个控制器守护进程。控制器守护程序从Kubernetes接收所需的Ingress配置。它会生成一个nginx或HAProxy配置文件，并重新启动负载平衡器进程以使更改生效。换句话说，Ingress controller是由Kubernetes管理的负载均衡器。 Kubernetes Ingress提供了负载平衡器的典型特性：HTTP路由，粘性会话，SSL终止，SSL直通，TCP和UDP负载平衡等。目前并不是所有的Ingress controller都实现了这些功能，需要查看具体的Ingress controller文档。 apiVersion: extensions/v1beta1 kind: Ingress metadata: name: influxdb spec: rules: - host: influxdb.kube.example.com http: paths: - backend: serviceName: influxdb servicePort: 8086 外部访问URL http://influxdb.kube.example.com/ping 访问该服务，入口就是80端口，然后Ingress controller直接将流量转发给后端Pod，不需再经过kube-proxy的转发，比LoadBalancer方式更高效。 总结 总的来说Ingress是一个非常灵活和越来越得到厂商支持的服务暴露方式，包括Nginx、HAProxy、Traefik，还有各种Service Mesh，而其它服务暴露方式可以更适用于服务调试、特殊应用的部署。 参考 Accessing Kubernetes Pods from Outside of the Cluster - alesnosek.com Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-21 20:08:06 "},"guide/carbin-mobile-dashboard-for-kubernetes.html":{"url":"guide/carbin-mobile-dashboard-for-kubernetes.html","title":"Carbin - Kuberentes手机客户端","keywords":"","body":"Carbin - Kubernetes手机客户端 Carbin是由bitnami开源的手机管理Kubernetes集群的客户端，目前提供iOS和安卓版本，代码开源在GitHub上：https://bitnami.com/ 为了方便移动办公，可以使用Carbin这个kuberntes手机客户端，可以链接GKE和任何Kubernetes集群，可以使用以下三种认证方式： 证书 token kubeconfig文件 所有功能跟kubernetes dashboard相同，还可以支持使用Helm chart部署应用，可以配置自定义的chart仓库地址。 iPhone用户可以在App Store中搜索Carbin即可找到。 图片 - App Store 可以很方便的在手机上操作自己的kubernetes集群，还可以登录到容器中操作，只要是kubernetes API支持的功能，都可以在该移动客户端上实现。 图片 - 在手机上操作Kubernetes集群 更多详细信息请参考：https://github.com/bitnami/cabin Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-22 18:38:12 "},"guide/kubernetes-desktop-client.html":{"url":"guide/kubernetes-desktop-client.html","title":"Kubernetic - Kubernetes桌面客户端","keywords":"","body":"Kubernetic - Kubernetes桌面客户端 Kubernetic - 一款kubenretes桌面客户端，https://kubernetic.com/，支持以下特性： 实时展示集群状态 多集群，多个namespace管理 原生kubernetes支持 支持使用chart安装应用 使用kubeconfig登陆认证 图片 - Kubernetic客户端 该客户端支持Mac和Windows系统，beta版本免费使用，stable版本需要付费。beta版本某些功能不完善，比如无法在应用内修改ingress配置，无法监控应用的状态等。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-30 20:08:19 "},"guide/application-development-deployment-flow.html":{"url":"guide/application-development-deployment-flow.html","title":"在kubernetes中开发部署应用","keywords":"","body":"在 Kubernetes 中开发部署应用 理论上只要可以使用主机名做服务注册的应用都可以迁移到 kubernetes 集群上。看到这里你可能不禁要问，为什么使用 IP 地址做服务注册发现的应用不适合迁移到 kubernetes 集群？因为这样的应用不适合自动故障恢复，因为目前 kubernetes 中不支持固定 Pod 的 IP 地址，当 Pod 故障后自动转移到其他 Node 的时候该 Pod 的 IP 地址也随之变化。 将传统应用迁移到 kubernetes 中可能还有很长的路要走，但是直接开发 Cloud native 应用，kubernetes 就是最佳运行时环境了。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-08-21 18:23:34 "},"guide/deploy-applications-in-kubernetes.html":{"url":"guide/deploy-applications-in-kubernetes.html","title":"适用于kubernetes的应用开发部署流程","keywords":"","body":"适用于kubernetes的应用开发部署流程 为了讲解详细流程，我特意写了两个示例程序放在GitHub中，模拟监控流程： k8s-app-monitor-test：生成模拟的监控数据，发送http请求，获取json返回值 K8s-app-monitor-agent：获取监控数据并绘图，访问浏览器获取图表 API文档见k8s-app-monitor-test中的api.html文件，该文档在API blueprint中定义，使用aglio生成，打开后如图所示： 图片 - API 关于服务发现 K8s-app-monitor-agent服务需要访问k8s-app-monitor-test服务，这就涉及到服务发现的问题，我们在代码中直接写死了要访问的服务的内网DNS地址（kubedns中的地址，即k8s-app-monitor-test.default.svc.cluster.local）。 我们知道Kubernetes在启动Pod的时候为容器注入环境变量，这些环境变量在所有的 namespace 中共享（环境变量是不断追加的，新启动的Pod中将拥有老的Pod中所有的环境变量，而老的Pod中的环境变量不变）。但是既然使用这些环境变量就已经可以访问到对应的service，那么获取应用的地址信息，究竟是使用变量呢？还是直接使用DNS解析来发现？ 答案是使用DNS，详细说明见Kubernetes中的服务发现与Docker容器间的环境变量传递源码探究。 打包镜像 因为我使用wercker自动构建，构建完成后自动打包成docker镜像并上传到docker hub中（需要现在docker hub中创建repo）。 构建流程见：https://app.wercker.com/jimmysong/k8s-app-monitor-agent/ 图片 - wercker 生成了如下两个docker镜像： jimmysong/k8s-app-monitor-test:latest jimmysong/k8s-app-monitor-agent:latest 启动服务 所有的kubernetes应用启动所用的yaml配置文件都保存在那两个GitHub仓库的manifest.yaml文件中。 分别在两个GitHub目录下执行kubectl create -f manifest.yaml即可启动服务。 外部访问 服务启动后需要更新ingress配置，在ingress.yaml文件中增加以下几行： - host: k8s-app-monitor-agent.jimmysong.io http: paths: - path: / backend: serviceName: k8s-app-monitor-agent servicePort: 8080 保存后，然后执行kubectl replace -f ingress.yaml即可刷新ingress。 修改本机的/etc/hosts文件，在其中加入以下一行： 172.20.0.119 k8s-app-monitor-agent.jimmysong.io 当然你也可以加入到DNS中，为了简单起见我使用hosts。 详见边缘节点配置。 在浏览器中访问http://k8s-app-monitor-agent.jimmysong.io Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-07 10:17:15 "},"guide/migrating-hadoop-yarn-to-kubernetes.html":{"url":"guide/migrating-hadoop-yarn-to-kubernetes.html","title":"迁移传统应用到kubernetes中——以Hadoop YARN为例","keywords":"","body":"迁移传统应用到Kubernetes步骤详解——以Hadoop YARN为例 本文档不是说明如何在 kubernetes 中开发和部署应用程序，如果您想要直接开发应用程序在 kubernetes 中运行可以参考 适用于kubernetes的应用开发部署流程。 本文旨在说明如何将已有的应用程序尤其是传统的分布式应用程序迁移到 kubernetes 中。如果该类应用程序符合云原生应用规范（如12因素法则）的话，那么迁移会比较顺利，否则会遇到一些麻烦甚至是阻碍。具体请参考 迁移至云原生应用架构。 接下来我们将以 Spark on YARN with kubernetes 为例来说明，该例子足够复杂也很有典型性，了解了这个例子可以帮助大家将自己的应用迁移到 kubernetes 集群上去，代码和配置文件可以在 这里 找到（本文中加入 Spark 的配置，代码中并没有包含，读者可以自己配置）。 下图即整个架构的示意图，代码和详细配置文件请参考 kube-yarn（不包含 ingress、spark 配置），所有的进程管理和容器扩容直接使用 Makefile。 图片 - spark on yarn with kubernetes 注意： 该例子仅用来说明具体的步骤划分和复杂性，在生产环境应用还有待验证，请谨慎使用。 术语 对于为曾接触过 kubernetes 或对云平台的技术细节不太了解的人来说，如何将应用迁移到 kubernetes 中可能是个头疼的问题，在行动之前有必要先了解整个过程中需要用到哪些概念和术语，有助于大家在行动中达成共识。 过程中可能用到的概念和术语初步整理如下： 图片 - Terms 为了讲解整改过程和具体细节，我们所有操作都是通过命令手动完成，不使用自动化工具。当您充分了解到其中的细节后可以通过自动化工具来优化该过程，以使其更加自动和高效，同时减少因为人为操作失误导致的迁移失败。 迁移应用 图片 - 分解步骤解析 整个迁移过程分为如下几个步骤： 将原有应用拆解为服务 我们不是一上来就开始做镜像，写配置，而是应该先梳理下要迁移的应用中有哪些可以作为服务运行，哪些是变的，哪些是不变的部分。 服务划分的原则是最小可变原则，这个同样适用于镜像制作，将服务中不变的部分编译到同一个镜像中。 对于像 Spark on YARN 这样复杂的应用，可以将其划分为三大类服务： ResourceManager NodeManager Spark client 制作镜像 根据拆解出来的服务，我们需要制作两个镜像： Hadoop Spark (From hadoop docker image) 因为我们运行的是 Spark on YARN，因此 Spark 依赖与 Hadoop 镜像，我们在 Spark 的基础上包装了一个 web service 作为服务启动。 镜像制作过程中不需要在 Dockerfile 中指定 Entrypoint 和 CMD，这些都是在 kubernetes 的 YAML 文件中指定的。 Hadoop YARN 的 Dockerfile 参考如下配置。 FROM my-docker-repo/jdk:7u80 # Add native libs ARG HADOOP_VERSION=2.6.0-cdh5.5.2 ## Prefer to download from server not use local storage ADD hadoop-${HADOOP_VERSION}.tar.gz /usr/local ADD ./lib/* /usr/local/hadoop-${HADOOP_VERSION}/lib/native/ ADD ./jars/* /usr/local/hadoop-${HADOOP_VERSION}/share/hadoop/yarn/ ENV HADOOP_PREFIX=/usr/local/hadoop \\ HADOOP_COMMON_HOME=/usr/local/hadoop \\ HADOOP_HDFS_HOME=/usr/local/hadoop \\ HADOOP_MAPRED_HOME=/usr/local/hadoop \\ HADOOP_YARN_HOME=/usr/local/hadoop \\ HADOOP_CONF_DIR=/usr/local/hadoop/etc/hadoop \\ YARN_CONF_DIR=/usr/local/hadoop/etc/hadoop \\ PATH=${PATH}:/usr/local/hadoop/bin RUN \\ cd /usr/local && ln -s ./hadoop-${HADOOP_VERSION} hadoop && \\ rm -f ${HADOOP_PREFIX}/logs/* WORKDIR $HADOOP_PREFIX # Hdfs ports EXPOSE 50010 50020 50070 50075 50090 8020 9000 # Mapred ports EXPOSE 19888 #Yarn ports EXPOSE 8030 8031 8032 8033 8040 8042 8088 #Other ports EXPOSE 49707 2122 准备应用的配置文件 因为我们只制作了一个 Hadoop 的镜像，而需要启动两个服务，这就要求在服务启动的时候必须加载不同的配置文件，现在我们只需要准备两个服务中需要同时用的的配置的部分。 YARN 依赖的配置在 artifacts 目录下，包含以下文件： bootstrap.sh capacity-scheduler.xml container-executor.cfg core-site.xml hadoop-env.sh hdfs-site.xml log4j.properties mapred-site.xml nodemanager_exclude.txt slaves start-yarn-nm.sh start-yarn-rm.sh yarn-env.sh yarn-site.xml 其中作为 bootstrap 启动脚本的 bootstrap.sh 也包含在该目录下，该脚本的如何编写请见下文。 Kubernetes YAML 文件 根据业务的特性选择最适合的 kubernetes 的资源对象来运行，因为在 YARN 中 NodeManager 需要使用主机名向 ResourceManger 注册，因此需要沿用 YARN 原有的服务发现方式，使用 headless service 和 StatefulSet 资源。更多资料请参考 StatefulSet。 所有的 Kubernetes YAML 配置文件存储在 manifest 目录下，包括如下配置： yarn-cluster 的 namespace 配置 Spark、ResourceManager、NodeManager 的 headless service 和 StatefulSet 配置 需要暴露到 kubernetes 集群外部的 ingress 配置（ResourceManager 的 Web） kube-yarn-ingress.yaml spark-statefulset.yaml yarn-cluster-namespace.yaml yarn-nm-statefulset.yaml yarn-rm-statefulset.yaml Bootstrap 脚本 Bootstrap 脚本的作用是在启动时根据 Pod 的环境变量、主机名或其他可以区分不同 Pod 和将启动角色的变量来修改配置文件和启动服务应用。 该脚本同时将原来 YARN 的日志使用 stdout 输出，便于使用 kubectl logs 查看日志或其他日志收集工具进行日志收集。 启动脚本 bootstrap.sh 跟 Hadoop 的配置文件同时保存在 artifacts 目录下。 该脚本根据 Pod 的主机名，决定如何修改 Hadoop 的配置文件和启动何种服务。bootstrap.sh 文件的部分代码如下： if [[ \"${HOSTNAME}\" =~ \"yarn-nm\" ]]; then sed -i '//d' $HADOOP_PREFIX/etc/hadoop/yarn-site.xml cat >> $HADOOP_PREFIX/etc/hadoop/yarn-site.xml yarn.nodemanager.resource.memory-mb ${MY_MEM_LIMIT:-2048} yarn.nodemanager.resource.cpu-vcores ${MY_CPU_LIMIT:-2} EOM echo '' >> $HADOOP_PREFIX/etc/hadoop/yarn-site.xml cp ${CONFIG_DIR}/start-yarn-nm.sh $HADOOP_PREFIX/sbin/ cd $HADOOP_PREFIX/sbin chmod +x start-yarn-nm.sh ./start-yarn-nm.sh fi if [[ $1 == \"-d\" ]]; then until find ${HADOOP_PREFIX}/logs -mmin -1 | egrep -q '.*'; echo \"`date`: Waiting for logs...\" ; do sleep 2 ; done tail -F ${HADOOP_PREFIX}/logs/* & while true; do sleep 1000; done fi 从这部分中代码中可以看到，如果 Pod 的主机名中包含 yarn-nm 字段则向 yarn-site.xml 配置文件中增加如下内容： yarn.nodemanager.resource.memory-mb ${MY_MEM_LIMIT:-2048} yarn.nodemanager.resource.cpu-vcores ${MY_CPU_LIMIT:-2} 其中 MY_MEM_LIMIT 和 MY_CPU_LIMIT 是 kubernetes YAML 中定义的环境变量，该环境变量又是引用的 Resource limit。 所有的配置准备完成后，执行 start-yarn-nm.sh 脚本启动 NodeManager。 如果 kubernetes YAML 中的 container CMD args 中包含 -d 则在后台运行 NodeManger 并 tail 输出 NodeManager 的日志到标准输出。 ConfigMaps 将 Hadoop 的配置文件和 bootstrap 脚本作为 ConfigMap 资源保存，用作 Pod 启动时挂载的 volume。 kubectl create configmap hadoop-config \\ --from-file=artifacts/hadoop/bootstrap.sh \\ --from-file=artifacts/hadoop/start-yarn-rm.sh \\ --from-file=artifacts/hadoop/start-yarn-nm.sh \\ --from-file=artifacts/hadoop/slaves \\ --from-file=artifacts/hadoop/core-site.xml \\ --from-file=artifacts/hadoop/hdfs-site.xml \\ --from-file=artifacts/hadoop/mapred-site.xml \\ --from-file=artifacts/hadoop/yarn-site.xml \\ --from-file=artifacts/hadoop/capacity-scheduler.xml \\ --from-file=artifacts/hadoop/container-executor.cfg \\ --from-file=artifacts/hadoop/hadoop-env.sh \\ --from-file=artifacts/hadoop/log4j.properties \\ --from-file=artifacts/hadoop/nodemanager_exclude.txt \\ --from-file=artifacts/hadoop/yarn-env.sh kubectl create configmap spark-config \\ --from-file=artifacts/spark/spark-bootstrap.sh \\ --from-file=artifacts/spark/spark-env.sh \\ --from-file=artifacts/spark/spark-defaults.conf 所有的配置完成后，可以可以使用 kubectl 命令来启动和管理集群了，我们编写了 Makefile，您可以直接使用该 Makefile 封装的命令实现部分的自动化。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-08-21 18:23:34 "},"guide/using-statefulset.html":{"url":"guide/using-statefulset.html","title":"使用StatefulSet部署用状态应用","keywords":"","body":"使用StatefulSet部署有状态应用 StatefulSet 这个对象是专门用来部署用状态应用的，可以为Pod提供稳定的身份标识，包括hostname、启动顺序、DNS名称等。 下面以在kubernetes1.6版本中部署zookeeper和kafka为例讲解StatefulSet的使用，其中kafka依赖于zookeeper。 Dockerfile和配置文件见 zookeeper 和 kafka。 注：所有的镜像基于CentOS系统的JDK制作，为我的私人镜像，外部无法访问，yaml中没有配置持久化存储。 部署Zookeeper Dockerfile中从远程获取zookeeper的安装文件，然后在定义了三个脚本： zkGenConfig.sh：生成zookeeper配置文件 zkMetrics.sh：获取zookeeper的metrics zkOk.sh：用来做ReadinessProb 我们在来看下这三个脚本的执行结果： zkGenConfig.sh zkMetrics.sh脚本实际上执行的是下面的命令： $ echo mntr | nc localhost $ZK_CLIENT_PORT >& 1 zk_version 3.4.6-1569965, built on 02/20/2014 09:09 GMT zk_avg_latency 0 zk_max_latency 5 zk_min_latency 0 zk_packets_received 427879 zk_packets_sent 427890 zk_num_alive_connections 3 zk_outstanding_requests 0 zk_server_state leader zk_znode_count 18 zk_watch_count 3 zk_ephemerals_count 4 zk_approximate_data_size 613 zk_open_file_descriptor_count 29 zk_max_file_descriptor_count 1048576 zk_followers 1 zk_synced_followers 1 zk_pending_syncs 0 zkOk.sh脚本实际上执行的是下面的命令： $ echo ruok | nc 127.0.0.1 $ZK_CLIENT_PORT imok zookeeper.yaml 下面是启动三个zookeeper实例的yaml配置文件： --- apiVersion: v1 kind: Service metadata: name: zk-svc labels: app: zk-svc spec: ports: - port: 2888 name: server - port: 3888 name: leader-election clusterIP: None selector: app: zk --- apiVersion: v1 kind: ConfigMap metadata: name: zk-cm data: jvm.heap: \"1G\" tick: \"2000\" init: \"10\" sync: \"5\" client.cnxns: \"60\" snap.retain: \"3\" purge.interval: \"0\" --- apiVersion: policy/v1beta1 kind: PodDisruptionBudget metadata: name: zk-pdb spec: selector: matchLabels: app: zk minAvailable: 2 --- apiVersion: apps/v1beta1 kind: StatefulSet metadata: name: zk spec: serviceName: zk-svc replicas: 3 template: metadata: labels: app: zk spec: affinity: podAntiAffinity: requiredDuringSchedulingIgnoredDuringExecution: - labelSelector: matchExpressions: - key: \"app\" operator: In values: - zk topologyKey: \"kubernetes.io/hostname\" containers: - name: k8szk imagePullPolicy: Always image: sz-pg-oam-docker-hub-001.tendcloud.com/library/zookeeper:3.4.6 resources: requests: memory: \"2Gi\" cpu: \"500m\" ports: - containerPort: 2181 name: client - containerPort: 2888 name: server - containerPort: 3888 name: leader-election env: - name : ZK_REPLICAS value: \"3\" - name : ZK_HEAP_SIZE valueFrom: configMapKeyRef: name: zk-cm key: jvm.heap - name : ZK_TICK_TIME valueFrom: configMapKeyRef: name: zk-cm key: tick - name : ZK_INIT_LIMIT valueFrom: configMapKeyRef: name: zk-cm key: init - name : ZK_SYNC_LIMIT valueFrom: configMapKeyRef: name: zk-cm key: tick - name : ZK_MAX_CLIENT_CNXNS valueFrom: configMapKeyRef: name: zk-cm key: client.cnxns - name: ZK_SNAP_RETAIN_COUNT valueFrom: configMapKeyRef: name: zk-cm key: snap.retain - name: ZK_PURGE_INTERVAL valueFrom: configMapKeyRef: name: zk-cm key: purge.interval - name: ZK_CLIENT_PORT value: \"2181\" - name: ZK_SERVER_PORT value: \"2888\" - name: ZK_ELECTION_PORT value: \"3888\" command: - sh - -c - zkGenConfig.sh && zkServer.sh start-foreground readinessProbe: exec: command: - \"zkOk.sh\" initialDelaySeconds: 10 timeoutSeconds: 5 livenessProbe: exec: command: - \"zkOk.sh\" initialDelaySeconds: 10 timeoutSeconds: 5 securityContext: runAsUser: 1000 fsGroup: 1000 我们再主要下上面那三个脚本的用途。 部署kafka Kafka的docker镜像制作跟zookeeper类似，都是从远程下载安装包后，解压安装。 与zookeeper不同的是，只要一个脚本，但是又依赖于我们上一步安装的zookeeper，kafkaGenConfig.sh用来生成kafka的配置文件。 我们来看下这个脚本。 #!/bin/bash HOST=`hostname -s` if [[ $HOST =~ (.*)-([0-9]+)$ ]]; then NAME=${BASH_REMATCH[1]} ORD=${BASH_REMATCH[2]} else echo \"Failed to extract ordinal from hostname $HOST\" exit 1 fi MY_ID=$((ORD+1)) sed -i s\"/broker.id=0/broker.id=$MY_ID/g\" /opt/kafka/config/server.properties sed -i s'/zookeeper.connect=localhost:2181/zookeeper.connect=zk-0.zk-svc.brand.svc:2181,zk-1.zk-svc.brand.svc:2181,zk-2.zk-svc.brand.svc:2181/g' /opt/kafka/config/server.properties 该脚本根据statefulset生成的pod的hostname的后半截数字部分作为broker ID，同时再替换zookeeper的地址。 Kafka.yaml 下面是创建3个kafka实例的yaml配置。 --- apiVersion: v1 kind: Service metadata: name: kafka-svc labels: app: kafka spec: ports: - port: 9093 name: server clusterIP: None selector: app: kafka --- apiVersion: policy/v1beta1 kind: PodDisruptionBudget metadata: name: kafka-pdb spec: selector: matchLabels: app: kafka minAvailable: 2 --- apiVersion: apps/v1beta1 kind: StatefulSet metadata: name: kafka spec: serviceName: kafka-svc replicas: 3 template: metadata: labels: app: kafka spec: affinity: podAntiAffinity: requiredDuringSchedulingIgnoredDuringExecution: - labelSelector: matchExpressions: - key: \"app\" operator: In values: - kafka topologyKey: \"kubernetes.io/hostname\" podAffinity: preferredDuringSchedulingIgnoredDuringExecution: - weight: 1 podAffinityTerm: labelSelector: matchExpressions: - key: \"app\" operator: In values: - zk topologyKey: \"kubernetes.io/hostname\" terminationGracePeriodSeconds: 300 containers: - name: k8skafka imagePullPolicy: Always image: sz-pg-oam-docker-hub-001.tendcloud.com/library/kafka:2.10-0.8.2.1 resources: requests: memory: \"1Gi\" cpu: 500m env: - name: KF_REPLICAS value: \"3\" ports: - containerPort: 9093 name: server command: - /bin/bash - -c - \"/opt/kafka/bin/kafkaGenConfig.sh && /opt/kafka/bin/kafka-server-start.sh /opt/kafka/config/server.properties\" env: - name: KAFKA_HEAP_OPTS value : \"-Xmx512M -Xms512M\" - name: KAFKA_OPTS value: \"-Dlogging.level=DEBUG\" readinessProbe: tcpSocket: port: 9092 initialDelaySeconds: 15 timeoutSeconds: 1 参考 https://kubernetes.io/docs/concepts/workloads/controllers/statefulset/ kubernetes contrib - statefulsets http://blog.kubernetes.io/2017/01/running-mongodb-on-kubernetes-with-statefulsets.html Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-10-23 16:22:50 "},"practice/":{"url":"practice/","title":"最佳实践概览","keywords":"","body":"最佳实践概览 本章节从零开始创建我们自己的kubernetes集群，并在该集群的基础上，配置服务发现、负载均衡和日志收集等功能，使我们的集群能够成为一个真正线上可用、功能完整的集群。 第一部分 在CentOS上部署kubernetes集群中介绍了如何通过二进制文件在CentOS物理机（也可以是公有云主机）上快速部署一个kubernetes集群。 第二部分介绍如何在kubernetes中的服务发现与负载均衡。 第三部分介绍如何运维kubernetes集群。 第四部分介绍kubernetes中的存储管理。 第五部分关于kubernetes集群和应用的监控。 第六部分介绍kuberentes中的服务编排与管理。 第七部分介绍如何基于kubernetes做持续集成与发布。 第八部分是kubernetes集群与插件的更新升级。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-19 17:55:03 "},"practice/install-kubernetes-on-centos.html":{"url":"practice/install-kubernetes-on-centos.html","title":"在CentOS上部署kubernetes集群","keywords":"","body":"在CentOS上部署kubernetes集群 本文档最初是基于kubenetes1.6版本编写的，对于kuberentes1.8版本同样适用，只是个别位置有稍许变动，变动的地方我将特别注明。 本系列文档介绍使用二进制部署 kubernetes 集群的所有步骤，而不是使用 kubeadm 等自动化方式来部署集群，同时开启了集群的TLS安全认证，该安装步骤适用于所有bare metal环境、on-premise环境和公有云环境。 在部署的过程中，将详细列出各组件的启动参数，给出配置文件，详解它们的含义和可能遇到的问题。 部署完成后，你将理解系统各组件的交互原理，进而能快速解决实际问题。 所以本文档主要适合于那些有一定 kubernetes 基础，想通过一步步部署的方式来学习和了解系统配置、运行原理的人。 注：本文档中不包括docker和私有镜像仓库的安装，安装说明中使用的镜像来自 Google Cloud Platform，为了方便国内用户下载，我将其克隆并上传到了 时速云镜像市场，供大家免费下载。 欲下载最新版本的官方镜像请访问 Google 云平台容器注册表。 提供所有的配置文件 集群安装时所有组件用到的配置文件，包含在以下目录中： etc：service的环境变量配置文件 manifest：kubernetes应用的yaml文件 systemd ：systemd serivce配置文件 集群详情 OS：CentOS Linux release 7.3.1611 (Core) 3.10.0-514.16.1.el7.x86_64 Kubernetes 1.6.0 Docker 1.12.5（使用yum安装） Etcd 3.1.5 Flannel 0.7.1 vxlan或者host-gw 网络 TLS 认证通信 (所有组件，如 etcd、kubernetes master 和 node) RBAC 授权 kubelet TLS BootStrapping kubedns、dashboard、heapster(influxdb、grafana)、EFK(elasticsearch、fluentd、kibana) 集群插件 私有docker镜像仓库harbor（请自行部署，harbor提供离线安装包，直接使用docker-compose启动即可） 环境说明 在下面的步骤中，我们将在三台CentOS系统的物理机上部署具有三个节点的kubernetes1.6.0集群。 角色分配如下： Master：172.20.0.113 Node：172.20.0.113、172.20.0.114、172.20.0.115 注意：172.20.0.113这台主机master和node复用。所有生成证书、执行kubectl命令的操作都在这台节点上执行。一旦node加入到kubernetes集群之后就不需要再登陆node节点了。 安装前的准备 在node节点上安装docker1.12.5 直接使用yum install docker 关闭所有节点的SELinux 永久方法 – 需要重启服务器 修改/etc/selinux/config文件中设置SELINUX=disabled ，然后重启服务器。 临时方法 – 设置系统参数 使用命令setenforce 0 附： setenforce 1 设置SELinux 成为enforcing模式 setenforce 0 设置SELinux 成为permissive模式 准备harbor私有镜像仓库 参考：https://github.com/vmware/harbor 步骤介绍 创建 TLS 证书和秘钥 创建kubeconfig 文件 创建高可用etcd集群 安装kubectl命令行工具 部署master节点 安装flannel网络插件 部署node节点 安装kubedns插件 安装dashboard插件 安装heapster插件 安装EFK插件 提醒 由于启用了 TLS 双向认证、RBAC 授权等严格的安全机制，建议从头开始部署，而不要从中间开始，否则可能会认证、授权等失败！ 部署过程中需要有很多证书的操作，请大家耐心操作，不明白的操作可以参考本书中的其他章节的解释。 该部署操作仅是搭建成了一个可用 kubernetes 集群，而很多地方还需要进行优化，heapster 插件、EFK 插件不一定会用于真实的生产环境中，但是通过部署这些插件，可以让大家了解到如何部署应用到集群上。 注：本安装文档参考了 opsnull 跟我一步步部署 kubernetes 集群 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-19 17:29:57 "},"practice/create-tls-and-secret-key.html":{"url":"practice/create-tls-and-secret-key.html","title":"创建TLS证书和秘钥","keywords":"","body":"创建TLS证书和秘钥 前言 执行下列步骤前建议你先阅读以下内容： 管理集群中的TLS：教您如何创建TLS证书 kubelet的认证授权：向您描述如何通过认证授权来访问 kubelet 的 HTTPS 端点。 TLS bootstrap：介绍如何为 kubelet 设置 TLS 客户端证书引导（bootstrap）。 注意：这一步是在安装配置kubernetes的所有步骤中最容易出错也最难于排查问题的一步，而这却刚好是第一步，万事开头难，不要因为这点困难就望而却步。 如果您足够有信心在完全不了解自己在做什么的情况下能够成功地完成了这一步的配置，那么您可以尽管跳过上面的几篇文章直接进行下面的操作。 kubernetes 系统的各组件需要使用 TLS 证书对通信进行加密，本文档使用 CloudFlare 的 PKI 工具集 cfssl 来生成 Certificate Authority (CA) 和其它证书； 生成的 CA 证书和秘钥文件如下： ca-key.pem ca.pem kubernetes-key.pem kubernetes.pem kube-proxy.pem kube-proxy-key.pem admin.pem admin-key.pem 使用证书的组件如下： etcd：使用 ca.pem、kubernetes-key.pem、kubernetes.pem； kube-apiserver：使用 ca.pem、kubernetes-key.pem、kubernetes.pem； kubelet：使用 ca.pem； kube-proxy：使用 ca.pem、kube-proxy-key.pem、kube-proxy.pem； kubectl：使用 ca.pem、admin-key.pem、admin.pem； kube-controller-manager：使用 ca-key.pem、ca.pem 注意：以下操作都在 master 节点即 172.20.0.113 这台主机上执行，证书只需要创建一次即可，以后在向集群中添加新节点时只要将 /etc/kubernetes/ 目录下的证书拷贝到新节点上即可。 安装 CFSSL 方式一：直接使用二进制源码包安装 wget https://pkg.cfssl.org/R1.2/cfssl_linux-amd64 chmod +x cfssl_linux-amd64 mv cfssl_linux-amd64 /usr/local/bin/cfssl wget https://pkg.cfssl.org/R1.2/cfssljson_linux-amd64 chmod +x cfssljson_linux-amd64 mv cfssljson_linux-amd64 /usr/local/bin/cfssljson wget https://pkg.cfssl.org/R1.2/cfssl-certinfo_linux-amd64 chmod +x cfssl-certinfo_linux-amd64 mv cfssl-certinfo_linux-amd64 /usr/local/bin/cfssl-certinfo export PATH=/usr/local/bin:$PATH 方式二：使用go命令安装 我们的系统中安装了Go1.7.5，使用以下命令安装更快捷： $ go get -u github.com/cloudflare/cfssl/cmd/... $ echo $GOPATH /usr/local $ls /usr/local/bin/cfssl* cfssl cfssl-bundle cfssl-certinfo cfssljson cfssl-newkey cfssl-scan 在$GOPATH/bin目录下得到以cfssl开头的几个命令。 注意：以下文章中出现的cat的文件名如果不存在需要手工创建。 创建 CA (Certificate Authority) 创建 CA 配置文件 mkdir /root/ssl cd /root/ssl cfssl print-defaults config > config.json cfssl print-defaults csr > csr.json # 根据config.json文件的格式创建如下的ca-config.json文件 # 过期时间设置成了 87600h cat > ca-config.json 字段说明 ca-config.json：可以定义多个 profiles，分别指定不同的过期时间、使用场景等参数；后续在签名证书时使用某个 profile； signing：表示该证书可用于签名其它证书；生成的 ca.pem 证书中 CA=TRUE； server auth：表示client可以用该 CA 对server提供的证书进行验证； client auth：表示server可以用该CA对client提供的证书进行验证； 创建 CA 证书签名请求 创建 ca-csr.json 文件，内容如下： { \"CN\": \"kubernetes\", \"key\": { \"algo\": \"rsa\", \"size\": 2048 }, \"names\": [ { \"C\": \"CN\", \"ST\": \"BeiJing\", \"L\": \"BeiJing\", \"O\": \"k8s\", \"OU\": \"System\" } ] } \"CN\"：Common Name，kube-apiserver 从证书中提取该字段作为请求的用户名 (User Name)；浏览器使用该字段验证网站是否合法； \"O\"：Organization，kube-apiserver 从证书中提取该字段作为请求用户所属的组 (Group)； 生成 CA 证书和私钥 $ cfssl gencert -initca ca-csr.json | cfssljson -bare ca $ ls ca* ca-config.json ca.csr ca-csr.json ca-key.pem ca.pem 创建 kubernetes 证书 创建 kubernetes 证书签名请求文件 kubernetes-csr.json： { \"CN\": \"kubernetes\", \"hosts\": [ \"127.0.0.1\", \"172.20.0.112\", \"172.20.0.113\", \"172.20.0.114\", \"172.20.0.115\", \"10.254.0.1\", \"kubernetes\", \"kubernetes.default\", \"kubernetes.default.svc\", \"kubernetes.default.svc.cluster\", \"kubernetes.default.svc.cluster.local\" ], \"key\": { \"algo\": \"rsa\", \"size\": 2048 }, \"names\": [ { \"C\": \"CN\", \"ST\": \"BeiJing\", \"L\": \"BeiJing\", \"O\": \"k8s\", \"OU\": \"System\" } ] } 如果 hosts 字段不为空则需要指定授权使用该证书的 IP 或域名列表，由于该证书后续被 etcd 集群和 kubernetes master 集群使用，所以上面分别指定了 etcd 集群、kubernetes master 集群的主机 IP 和 kubernetes 服务的服务 IP（一般是 kube-apiserver 指定的 service-cluster-ip-range 网段的第一个IP，如 10.254.0.1。 hosts 中的内容可以为空，即使按照上面的配置，向集群中增加新节点后也不需要重新生成证书。 生成 kubernetes 证书和私钥 $ cfssl gencert -ca=ca.pem -ca-key=ca-key.pem -config=ca-config.json -profile=kubernetes kubernetes-csr.json | cfssljson -bare kubernetes $ ls kubernetes* kubernetes.csr kubernetes-csr.json kubernetes-key.pem kubernetes.pem 或者直接在命令行上指定相关参数： echo '{\"CN\":\"kubernetes\",\"hosts\":[\"\"],\"key\":{\"algo\":\"rsa\",\"size\":2048}}' | cfssl gencert -ca=ca.pem -ca-key=ca-key.pem -config=ca-config.json -profile=kubernetes -hostname=\"127.0.0.1,172.20.0.112,172.20.0.113,172.20.0.114,172.20.0.115,kubernetes,kubernetes.default\" - | cfssljson -bare kubernetes 创建 admin 证书 创建 admin 证书签名请求文件 admin-csr.json： { \"CN\": \"admin\", \"hosts\": [], \"key\": { \"algo\": \"rsa\", \"size\": 2048 }, \"names\": [ { \"C\": \"CN\", \"ST\": \"BeiJing\", \"L\": \"BeiJing\", \"O\": \"system:masters\", \"OU\": \"System\" } ] } 后续 kube-apiserver 使用 RBAC 对客户端(如 kubelet、kube-proxy、Pod)请求进行授权； kube-apiserver 预定义了一些 RBAC 使用的 RoleBindings，如 cluster-admin 将 Group system:masters 与 Role cluster-admin 绑定，该 Role 授予了调用kube-apiserver 的所有 API的权限； OU 指定该证书的 Group 为 system:masters，kubelet 使用该证书访问 kube-apiserver 时 ，由于证书被 CA 签名，所以认证通过，同时由于证书用户组为经过预授权的 system:masters，所以被授予访问所有 API 的权限； 生成 admin 证书和私钥 $ cfssl gencert -ca=ca.pem -ca-key=ca-key.pem -config=ca-config.json -profile=kubernetes admin-csr.json | cfssljson -bare admin $ ls admin* admin.csr admin-csr.json admin-key.pem admin.pem 创建 kube-proxy 证书 创建 kube-proxy 证书签名请求文件 kube-proxy-csr.json： { \"CN\": \"system:kube-proxy\", \"hosts\": [], \"key\": { \"algo\": \"rsa\", \"size\": 2048 }, \"names\": [ { \"C\": \"CN\", \"ST\": \"BeiJing\", \"L\": \"BeiJing\", \"O\": \"k8s\", \"OU\": \"System\" } ] } CN 指定该证书的 User 为 system:kube-proxy； kube-apiserver 预定义的 RoleBinding cluster-admin 将User system:kube-proxy 与 Role system:node-proxier 绑定，该 Role 授予了调用 kube-apiserver Proxy 相关 API 的权限； 生成 kube-proxy 客户端证书和私钥 $ cfssl gencert -ca=ca.pem -ca-key=ca-key.pem -config=ca-config.json -profile=kubernetes kube-proxy-csr.json | cfssljson -bare kube-proxy $ ls kube-proxy* kube-proxy.csr kube-proxy-csr.json kube-proxy-key.pem kube-proxy.pem 校验证书 以 kubernetes 证书为例 使用 opsnssl 命令 $ openssl x509 -noout -text -in kubernetes.pem ... Signature Algorithm: sha256WithRSAEncryption Issuer: C=CN, ST=BeiJing, L=BeiJing, O=k8s, OU=System, CN=Kubernetes Validity Not Before: Apr 5 05:36:00 2017 GMT Not After : Apr 5 05:36:00 2018 GMT Subject: C=CN, ST=BeiJing, L=BeiJing, O=k8s, OU=System, CN=kubernetes ... X509v3 extensions: X509v3 Key Usage: critical Digital Signature, Key Encipherment X509v3 Extended Key Usage: TLS Web Server Authentication, TLS Web Client Authentication X509v3 Basic Constraints: critical CA:FALSE X509v3 Subject Key Identifier: DD:52:04:43:10:13:A9:29:24:17:3A:0E:D7:14:DB:36:F8:6C:E0:E0 X509v3 Authority Key Identifier: keyid:44:04:3B:60:BD:69:78:14:68:AF:A0:41:13:F6:17:07:13:63:58:CD X509v3 Subject Alternative Name: DNS:kubernetes, DNS:kubernetes.default, DNS:kubernetes.default.svc, DNS:kubernetes.default.svc.cluster, DNS:kubernetes.default.svc.cluster.local, IP Address:127.0.0.1, IP Address:172.20.0.112, IP Address:172.20.0.113, IP Address:172.20.0.114, IP Address:172.20.0.115, IP Address:10.254.0.1 ... 确认 Issuer 字段的内容和 ca-csr.json 一致； 确认 Subject 字段的内容和 kubernetes-csr.json 一致； 确认 X509v3 Subject Alternative Name 字段的内容和 kubernetes-csr.json 一致； 确认 X509v3 Key Usage、Extended Key Usage 字段的内容和 ca-config.json 中 kubernetes profile 一致； 使用 cfssl-certinfo 命令 $ cfssl-certinfo -cert kubernetes.pem ... { \"subject\": { \"common_name\": \"kubernetes\", \"country\": \"CN\", \"organization\": \"k8s\", \"organizational_unit\": \"System\", \"locality\": \"BeiJing\", \"province\": \"BeiJing\", \"names\": [ \"CN\", \"BeiJing\", \"BeiJing\", \"k8s\", \"System\", \"kubernetes\" ] }, \"issuer\": { \"common_name\": \"Kubernetes\", \"country\": \"CN\", \"organization\": \"k8s\", \"organizational_unit\": \"System\", \"locality\": \"BeiJing\", \"province\": \"BeiJing\", \"names\": [ \"CN\", \"BeiJing\", \"BeiJing\", \"k8s\", \"System\", \"Kubernetes\" ] }, \"serial_number\": \"174360492872423263473151971632292895707129022309\", \"sans\": [ \"kubernetes\", \"kubernetes.default\", \"kubernetes.default.svc\", \"kubernetes.default.svc.cluster\", \"kubernetes.default.svc.cluster.local\", \"127.0.0.1\", \"10.64.3.7\", \"10.254.0.1\" ], \"not_before\": \"2017-04-05T05:36:00Z\", \"not_after\": \"2018-04-05T05:36:00Z\", \"sigalg\": \"SHA256WithRSA\", ... 分发证书 将生成的证书和秘钥文件（后缀名为.pem）拷贝到所有机器的 /etc/kubernetes/ssl 目录下备用； mkdir -p /etc/kubernetes/ssl cp *.pem /etc/kubernetes/ssl 参考 Generate self-signed certificates Setting up a Certificate Authority and Creating TLS Certificates Client Certificates V/s Server Certificates 数字证书及 CA 的扫盲介绍 TLS bootstrap 引导程序 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-09 17:50:24 "},"practice/create-kubeconfig.html":{"url":"practice/create-kubeconfig.html","title":"创建kubeconfig文件","keywords":"","body":"创建 kubeconfig 文件 注意：请先参考 安装kubectl命令行工具，先在 master 节点上安装 kubectl 然后再进行下面的操作。 kubelet、kube-proxy 等 Node 机器上的进程与 Master 机器的 kube-apiserver 进程通信时需要认证和授权； kubernetes 1.4 开始支持由 kube-apiserver 为客户端生成 TLS 证书的 TLS Bootstrapping 功能，这样就不需要为每个客户端生成证书了；该功能当前仅支持为 kubelet 生成证书； 因为我的master节点和node节点复用，所有在这一步其实已经安装了kubectl。参考安装kubectl命令行工具。 以下操作只需要在master节点上执行，生成的*.kubeconfig文件可以直接拷贝到node节点的/etc/kubernetes目录下。 创建 TLS Bootstrapping Token Token auth file Token可以是任意的包涵128 bit的字符串，可以使用安全的随机数发生器生成。 export BOOTSTRAP_TOKEN=$(head -c 16 /dev/urandom | od -An -t x | tr -d ' ') cat > token.csv 后三行是一句，直接复制上面的脚本运行即可。 注意：在进行后续操作前请检查 token.csv 文件，确认其中的 ${BOOTSTRAP_TOKEN} 环境变量已经被真实的值替换。 BOOTSTRAP_TOKEN 将被写入到 kube-apiserver 使用的 token.csv 文件和 kubelet 使用的 bootstrap.kubeconfig 文件，如果后续重新生成了 BOOTSTRAP_TOKEN，则需要： 更新 token.csv 文件，分发到所有机器 (master 和 node）的 /etc/kubernetes/ 目录下，分发到node节点上非必需； 重新生成 bootstrap.kubeconfig 文件，分发到所有 node 机器的 /etc/kubernetes/ 目录下； 重启 kube-apiserver 和 kubelet 进程； 重新 approve kubelet 的 csr 请求； cp token.csv /etc/kubernetes/ 创建 kubelet bootstrapping kubeconfig 文件 执行下面的命令时需要先安装kubectl命令，请参考安装kubectl命令行工具。 cd /etc/kubernetes export KUBE_APISERVER=\"https://172.20.0.113:6443\" # 设置集群参数 kubectl config set-cluster kubernetes \\ --certificate-authority=/etc/kubernetes/ssl/ca.pem \\ --embed-certs=true \\ --server=${KUBE_APISERVER} \\ --kubeconfig=bootstrap.kubeconfig # 设置客户端认证参数 kubectl config set-credentials kubelet-bootstrap \\ --token=${BOOTSTRAP_TOKEN} \\ --kubeconfig=bootstrap.kubeconfig # 设置上下文参数 kubectl config set-context default \\ --cluster=kubernetes \\ --user=kubelet-bootstrap \\ --kubeconfig=bootstrap.kubeconfig # 设置默认上下文 kubectl config use-context default --kubeconfig=bootstrap.kubeconfig --embed-certs 为 true 时表示将 certificate-authority 证书写入到生成的 bootstrap.kubeconfig 文件中； 设置客户端认证参数时没有指定秘钥和证书，后续由 kube-apiserver 自动生成； 创建 kube-proxy kubeconfig 文件 export KUBE_APISERVER=\"https://172.20.0.113:6443\" # 设置集群参数 kubectl config set-cluster kubernetes \\ --certificate-authority=/etc/kubernetes/ssl/ca.pem \\ --embed-certs=true \\ --server=${KUBE_APISERVER} \\ --kubeconfig=kube-proxy.kubeconfig # 设置客户端认证参数 kubectl config set-credentials kube-proxy \\ --client-certificate=/etc/kubernetes/ssl/kube-proxy.pem \\ --client-key=/etc/kubernetes/ssl/kube-proxy-key.pem \\ --embed-certs=true \\ --kubeconfig=kube-proxy.kubeconfig # 设置上下文参数 kubectl config set-context default \\ --cluster=kubernetes \\ --user=kube-proxy \\ --kubeconfig=kube-proxy.kubeconfig # 设置默认上下文 kubectl config use-context default --kubeconfig=kube-proxy.kubeconfig 设置集群参数和客户端认证参数时 --embed-certs 都为 true，这会将 certificate-authority、client-certificate 和 client-key 指向的证书文件内容写入到生成的 kube-proxy.kubeconfig 文件中； kube-proxy.pem 证书中 CN 为 system:kube-proxy，kube-apiserver 预定义的 RoleBinding cluster-admin 将User system:kube-proxy 与 Role system:node-proxier 绑定，该 Role 授予了调用 kube-apiserver Proxy 相关 API 的权限； 分发 kubeconfig 文件 将两个 kubeconfig 文件分发到所有 Node 机器的 /etc/kubernetes/ 目录 cp bootstrap.kubeconfig kube-proxy.kubeconfig /etc/kubernetes/ 参考 关于 kubeconfig 文件的更多信息请参考 使用 kubeconfig 文件配置跨集群认证。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-15 12:24:51 "},"practice/etcd-cluster-installation.html":{"url":"practice/etcd-cluster-installation.html","title":"创建高可用etcd集群","keywords":"","body":"创建高可用 etcd 集群 kuberntes 系统使用 etcd 存储所有数据，本文档介绍部署一个三节点高可用 etcd 集群的步骤，这三个节点复用 kubernetes master 机器，分别命名为sz-pg-oam-docker-test-001.tendcloud.com、sz-pg-oam-docker-test-002.tendcloud.com、sz-pg-oam-docker-test-003.tendcloud.com： sz-pg-oam-docker-test-001.tendcloud.com：172.20.0.113 sz-pg-oam-docker-test-002.tendcloud.com：172.20.0.114 sz-pg-oam-docker-test-003.tendcloud.com：172.20.0.115 TLS 认证文件 需要为 etcd 集群创建加密通信的 TLS 证书，这里复用以前创建的 kubernetes 证书 cp ca.pem kubernetes-key.pem kubernetes.pem /etc/kubernetes/ssl kubernetes 证书的 hosts 字段列表中包含上面三台机器的 IP，否则后续证书校验会失败； 下载二进制文件 到 https://github.com/coreos/etcd/releases 页面下载最新版本的二进制文件 wget https://github.com/coreos/etcd/releases/download/v3.1.5/etcd-v3.1.5-linux-amd64.tar.gz tar -xvf etcd-v3.1.5-linux-amd64.tar.gz mv etcd-v3.1.5-linux-amd64/etcd* /usr/local/bin 或者直接使用yum命令安装： yum install etcd 若使用yum安装，默认etcd命令将在/usr/bin目录下，注意修改下面的etcd.service文件中的启动命令地址为/usr/bin/etcd。 创建 etcd 的 systemd unit 文件 注意替换IP地址为你自己的etcd集群的主机IP。 [Unit] Description=Etcd Server After=network.target After=network-online.target Wants=network-online.target Documentation=https://github.com/coreos [Service] Type=notify WorkingDirectory=/var/lib/etcd/ EnvironmentFile=-/etc/etcd/etcd.conf ExecStart=/usr/local/bin/etcd \\ --name ${ETCD_NAME} \\ --cert-file=/etc/kubernetes/ssl/kubernetes.pem \\ --key-file=/etc/kubernetes/ssl/kubernetes-key.pem \\ --peer-cert-file=/etc/kubernetes/ssl/kubernetes.pem \\ --peer-key-file=/etc/kubernetes/ssl/kubernetes-key.pem \\ --trusted-ca-file=/etc/kubernetes/ssl/ca.pem \\ --peer-trusted-ca-file=/etc/kubernetes/ssl/ca.pem \\ --initial-advertise-peer-urls ${ETCD_INITIAL_ADVERTISE_PEER_URLS} \\ --listen-peer-urls ${ETCD_LISTEN_PEER_URLS} \\ --listen-client-urls ${ETCD_LISTEN_CLIENT_URLS},http://127.0.0.1:2379 \\ --advertise-client-urls ${ETCD_ADVERTISE_CLIENT_URLS} \\ --initial-cluster-token ${ETCD_INITIAL_CLUSTER_TOKEN} \\ --initial-cluster infra1=https://172.20.0.113:2380,infra2=https://172.20.0.114:2380,infra3=https://172.20.0.115:2380 \\ --initial-cluster-state new \\ --data-dir=${ETCD_DATA_DIR} Restart=on-failure RestartSec=5 LimitNOFILE=65536 [Install] WantedBy=multi-user.target 指定 etcd 的工作目录为 /var/lib/etcd，数据目录为 /var/lib/etcd，需在启动服务前创建这个目录，否则启动服务的时候会报错“Failed at step CHDIR spawning /usr/bin/etcd: No such file or directory”； 为了保证通信安全，需要指定 etcd 的公私钥(cert-file和key-file)、Peers 通信的公私钥和 CA 证书(peer-cert-file、peer-key-file、peer-trusted-ca-file)、客户端的CA证书（trusted-ca-file）； 创建 kubernetes.pem 证书时使用的 kubernetes-csr.json 文件的 hosts 字段包含所有 etcd 节点的IP，否则证书校验会出错； --initial-cluster-state 值为 new 时，--name 的参数值必须位于 --initial-cluster 列表中； 完整 unit 文件见：etcd.service 环境变量配置文件/etc/etcd/etcd.conf。 # [member] ETCD_NAME=infra1 ETCD_DATA_DIR=\"/var/lib/etcd\" ETCD_LISTEN_PEER_URLS=\"https://172.20.0.113:2380\" ETCD_LISTEN_CLIENT_URLS=\"https://172.20.0.113:2379\" #[cluster] ETCD_INITIAL_ADVERTISE_PEER_URLS=\"https://172.20.0.113:2380\" ETCD_INITIAL_CLUSTER_TOKEN=\"etcd-cluster\" ETCD_ADVERTISE_CLIENT_URLS=\"https://172.20.0.113:2379\" 这是172.20.0.113节点的配置，其他两个etcd节点只要将上面的IP地址改成相应节点的IP地址即可。ETCD_NAME换成对应节点的infra1/2/3。 启动 etcd 服务 mv etcd.service /usr/lib/system/systemd/ systemctl daemon-reload systemctl enable etcd systemctl start etcd systemctl status etcd 在所有的 kubernetes master 节点重复上面的步骤，直到所有机器的 etcd 服务都已启动。 验证服务 在任一 kubernetes master 机器上执行如下命令： $ etcdctl \\ --ca-file=/etc/kubernetes/ssl/ca.pem \\ --cert-file=/etc/kubernetes/ssl/kubernetes.pem \\ --key-file=/etc/kubernetes/ssl/kubernetes-key.pem \\ cluster-health 2017-04-11 15:17:09.082250 I | warning: ignoring ServerName for user-provided CA for backwards compatibility is deprecated 2017-04-11 15:17:09.083681 I | warning: ignoring ServerName for user-provided CA for backwards compatibility is deprecated member 9a2ec640d25672e5 is healthy: got healthy result from https://172.20.0.115:2379 member bc6f27ae3be34308 is healthy: got healthy result from https://172.20.0.114:2379 member e5c92ea26c4edba0 is healthy: got healthy result from https://172.20.0.113:2379 cluster is healthy 结果最后一行为 cluster is healthy 时表示集群服务正常。 更多资料 关于如何在etcd中查看kubernetes的数据，请参考使用etcdctl访问kuberentes数据。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-15 19:44:13 "},"practice/kubectl-installation.html":{"url":"practice/kubectl-installation.html","title":"安装kubectl命令行工具","keywords":"","body":"安装kubectl命令行工具 本文档介绍下载和配置 kubernetes 集群命令行工具 kubelet 的步骤。 下载 kubectl wget https://dl.k8s.io/v1.6.0/kubernetes-client-linux-amd64.tar.gz tar -xzvf kubernetes-client-linux-amd64.tar.gz cp kubernetes/client/bin/kube* /usr/bin/ chmod a+x /usr/bin/kube* 创建 kubectl kubeconfig 文件 export KUBE_APISERVER=\"https://172.20.0.113:6443\" # 设置集群参数 kubectl config set-cluster kubernetes \\ --certificate-authority=/etc/kubernetes/ssl/ca.pem \\ --embed-certs=true \\ --server=${KUBE_APISERVER} # 设置客户端认证参数 kubectl config set-credentials admin \\ --client-certificate=/etc/kubernetes/ssl/admin.pem \\ --embed-certs=true \\ --client-key=/etc/kubernetes/ssl/admin-key.pem # 设置上下文参数 kubectl config set-context kubernetes \\ --cluster=kubernetes \\ --user=admin # 设置默认上下文 kubectl config use-context kubernetes admin.pem 证书 OU 字段值为 system:masters，kube-apiserver 预定义的 RoleBinding cluster-admin 将 Group system:masters 与 Role cluster-admin 绑定，该 Role 授予了调用kube-apiserver 相关 API 的权限； 生成的 kubeconfig 被保存到 ~/.kube/config 文件； 注意：~/.kube/config文件拥有对该集群的最高权限，请妥善保管。 更多资料 kubectl命令概览 kubectl命令技巧大全 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-15 19:45:42 "},"practice/master-installation.html":{"url":"practice/master-installation.html","title":"部署master节点","keywords":"","body":"部署master节点 kubernetes master 节点包含的组件： kube-apiserver kube-scheduler kube-controller-manager 目前这三个组件需要部署在同一台机器上。 kube-scheduler、kube-controller-manager 和 kube-apiserver 三者的功能紧密相关； 同时只能有一个 kube-scheduler、kube-controller-manager 进程处于工作状态，如果运行多个，则需要通过选举产生一个 leader； 本文档记录部署一个三个节点的高可用 kubernetes master 集群步骤。（后续创建一个 load balancer 来代理访问 kube-apiserver 的请求） 注： 暂时未实现master节点的高可用 master节点上没有部署flannel网络插件，如果想要在master节点上也能访问ClusterIP，请参考下一节部署ndoe节点中的配置Flanneld部分。 TLS 证书文件 以下pem证书文件我们在创建TLS证书和秘钥这一步中已经创建过了，token.csv文件在创建kubeconfig文件的时候创建。我们再检查一下。 $ ls /etc/kubernetes/ssl admin-key.pem admin.pem ca-key.pem ca.pem kube-proxy-key.pem kube-proxy.pem kubernetes-key.pem kubernetes.pem 下载最新版本的二进制文件 有两种下载方式 方式一 从 github release 页面 下载发布版 tarball，解压后再执行下载脚本 wget https://github.com/kubernetes/kubernetes/releases/download/v1.6.0/kubernetes.tar.gz tar -xzvf kubernetes.tar.gz cd kubernetes ./cluster/get-kube-binaries.sh 方式二 从 CHANGELOG页面 下载 client 或 server tarball 文件 server 的 tarball kubernetes-server-linux-amd64.tar.gz 已经包含了 client(kubectl) 二进制文件，所以不用单独下载kubernetes-client-linux-amd64.tar.gz文件； # wget https://dl.k8s.io/v1.6.0/kubernetes-client-linux-amd64.tar.gz wget https://dl.k8s.io/v1.6.0/kubernetes-server-linux-amd64.tar.gz tar -xzvf kubernetes-server-linux-amd64.tar.gz cd kubernetes tar -xzvf kubernetes-src.tar.gz 将二进制文件拷贝到指定路径 cp -r server/bin/{kube-apiserver,kube-controller-manager,kube-scheduler,kubectl,kube-proxy,kubelet} /usr/local/bin/ 配置和启动 kube-apiserver 创建 kube-apiserver的service配置文件 service配置文件/usr/lib/systemd/system/kube-apiserver.service内容： [Unit] Description=Kubernetes API Service Documentation=https://github.com/GoogleCloudPlatform/kubernetes After=network.target After=etcd.service [Service] EnvironmentFile=-/etc/kubernetes/config EnvironmentFile=-/etc/kubernetes/apiserver ExecStart=/usr/local/bin/kube-apiserver \\ $KUBE_LOGTOSTDERR \\ $KUBE_LOG_LEVEL \\ $KUBE_ETCD_SERVERS \\ $KUBE_API_ADDRESS \\ $KUBE_API_PORT \\ $KUBELET_PORT \\ $KUBE_ALLOW_PRIV \\ $KUBE_SERVICE_ADDRESSES \\ $KUBE_ADMISSION_CONTROL \\ $KUBE_API_ARGS Restart=on-failure Type=notify LimitNOFILE=65536 [Install] WantedBy=multi-user.target /etc/kubernetes/config文件的内容为： ### # kubernetes system config # # The following values are used to configure various aspects of all # kubernetes services, including # # kube-apiserver.service # kube-controller-manager.service # kube-scheduler.service # kubelet.service # kube-proxy.service # logging to stderr means we get it in the systemd journal KUBE_LOGTOSTDERR=\"--logtostderr=true\" # journal message level, 0 is debug KUBE_LOG_LEVEL=\"--v=0\" # Should this cluster be allowed to run privileged docker containers KUBE_ALLOW_PRIV=\"--allow-privileged=true\" # How the controller-manager, scheduler, and proxy find the apiserver #KUBE_MASTER=\"--master=http://sz-pg-oam-docker-test-001.tendcloud.com:8080\" KUBE_MASTER=\"--master=http://172.20.0.113:8080\" 该配置文件同时被kube-apiserver、kube-controller-manager、kube-scheduler、kubelet、kube-proxy使用。 apiserver配置文件/etc/kubernetes/apiserver内容为： ### ## kubernetes system config ## ## The following values are used to configure the kube-apiserver ## # ## The address on the local server to listen to. #KUBE_API_ADDRESS=\"--insecure-bind-address=sz-pg-oam-docker-test-001.tendcloud.com\" KUBE_API_ADDRESS=\"--advertise-address=172.20.0.113 --bind-address=172.20.0.113 --insecure-bind-address=172.20.0.113\" # ## The port on the local server to listen on. #KUBE_API_PORT=\"--port=8080\" # ## Port minions listen on #KUBELET_PORT=\"--kubelet-port=10250\" # ## Comma separated list of nodes in the etcd cluster KUBE_ETCD_SERVERS=\"--etcd-servers=https://172.20.0.113:2379,https://172.20.0.114:2379,https://172.20.0.115:2379\" # ## Address range to use for services KUBE_SERVICE_ADDRESSES=\"--service-cluster-ip-range=10.254.0.0/16\" # ## default admission control policies KUBE_ADMISSION_CONTROL=\"--admission-control=ServiceAccount,NamespaceLifecycle,NamespaceExists,LimitRanger,ResourceQuota\" # ## Add your own! KUBE_API_ARGS=\"--authorization-mode=RBAC --runtime-config=rbac.authorization.k8s.io/v1beta1 --kubelet-https=true --experimental-bootstrap-token-auth --token-auth-file=/etc/kubernetes/token.csv --service-node-port-range=30000-32767 --tls-cert-file=/etc/kubernetes/ssl/kubernetes.pem --tls-private-key-file=/etc/kubernetes/ssl/kubernetes-key.pem --client-ca-file=/etc/kubernetes/ssl/ca.pem --service-account-key-file=/etc/kubernetes/ssl/ca-key.pem --etcd-cafile=/etc/kubernetes/ssl/ca.pem --etcd-certfile=/etc/kubernetes/ssl/kubernetes.pem --etcd-keyfile=/etc/kubernetes/ssl/kubernetes-key.pem --enable-swagger-ui=true --apiserver-count=3 --audit-log-maxage=30 --audit-log-maxbackup=3 --audit-log-maxsize=100 --audit-log-path=/var/lib/audit.log --event-ttl=1h\" --authorization-mode=RBAC 指定在安全端口使用 RBAC 授权模式，拒绝未通过授权的请求； kube-scheduler、kube-controller-manager 一般和 kube-apiserver 部署在同一台机器上，它们使用非安全端口和 kube-apiserver通信; kubelet、kube-proxy、kubectl 部署在其它 Node 节点上，如果通过安全端口访问 kube-apiserver，则必须先通过 TLS 证书认证，再通过 RBAC 授权； kube-proxy、kubectl 通过在使用的证书里指定相关的 User、Group 来达到通过 RBAC 授权的目的； 如果使用了 kubelet TLS Boostrap 机制，则不能再指定 --kubelet-certificate-authority、--kubelet-client-certificate 和 --kubelet-client-key 选项，否则后续 kube-apiserver 校验 kubelet 证书时出现 ”x509: certificate signed by unknown authority“ 错误； --admission-control 值必须包含 ServiceAccount； --bind-address 不能为 127.0.0.1； runtime-config配置为rbac.authorization.k8s.io/v1beta1，表示运行时的apiVersion； --service-cluster-ip-range 指定 Service Cluster IP 地址段，该地址段不能路由可达； 缺省情况下 kubernetes 对象保存在 etcd /registry 路径下，可以通过 --etcd-prefix 参数进行调整； 完整 unit 见 kube-apiserver.service 启动kube-apiserver systemctl daemon-reload systemctl enable kube-apiserver systemctl start kube-apiserver systemctl status kube-apiserver 配置和启动 kube-controller-manager 创建 kube-controller-manager的serivce配置文件 文件路径/usr/lib/systemd/system/kube-controller-manager.service [Unit] Description=Kubernetes Controller Manager Documentation=https://github.com/GoogleCloudPlatform/kubernetes [Service] EnvironmentFile=-/etc/kubernetes/config EnvironmentFile=-/etc/kubernetes/controller-manager ExecStart=/usr/local/bin/kube-controller-manager \\ $KUBE_LOGTOSTDERR \\ $KUBE_LOG_LEVEL \\ $KUBE_MASTER \\ $KUBE_CONTROLLER_MANAGER_ARGS Restart=on-failure LimitNOFILE=65536 [Install] WantedBy=multi-user.target 配置文件/etc/kubernetes/controller-manager。 ### # The following values are used to configure the kubernetes controller-manager # defaults from config and apiserver should be adequate # Add your own! KUBE_CONTROLLER_MANAGER_ARGS=\"--address=127.0.0.1 --service-cluster-ip-range=10.254.0.0/16 --cluster-name=kubernetes --cluster-signing-cert-file=/etc/kubernetes/ssl/ca.pem --cluster-signing-key-file=/etc/kubernetes/ssl/ca-key.pem --service-account-private-key-file=/etc/kubernetes/ssl/ca-key.pem --root-ca-file=/etc/kubernetes/ssl/ca.pem --leader-elect=true\" --service-cluster-ip-range 参数指定 Cluster 中 Service 的CIDR范围，该网络在各 Node 间必须路由不可达，必须和 kube-apiserver 中的参数一致； --cluster-signing-* 指定的证书和私钥文件用来签名为 TLS BootStrap 创建的证书和私钥； --root-ca-file 用来对 kube-apiserver 证书进行校验，指定该参数后，才会在Pod 容器的 ServiceAccount 中放置该 CA 证书文件； --address 值必须为 127.0.0.1，因为当前 kube-apiserver 期望 scheduler 和 controller-manager 在同一台机器，否则： $ kubectl get componentstatuses NAME STATUS MESSAGE ERROR scheduler Unhealthy Get http://127.0.0.1:10251/healthz: dial tcp 127.0.0.1:10251: getsockopt: connection refused controller-manager Healthy ok etcd-2 Healthy {\"health\": \"true\"} etcd-0 Healthy {\"health\": \"true\"} etcd-1 Healthy {\"health\": \"true\"} 如果有组件report unhealthy请参考：https://github.com/kubernetes-incubator/bootkube/issues/64 完整 unit 见 kube-controller-manager.service 启动 kube-controller-manager systemctl daemon-reload systemctl enable kube-controller-manager systemctl start kube-controller-manager 配置和启动 kube-scheduler 创建 kube-scheduler的serivce配置文件 文件路径/usr/lib/systemd/system/kube-scheduler.service。 [Unit] Description=Kubernetes Scheduler Plugin Documentation=https://github.com/GoogleCloudPlatform/kubernetes [Service] EnvironmentFile=-/etc/kubernetes/config EnvironmentFile=-/etc/kubernetes/scheduler ExecStart=/usr/local/bin/kube-scheduler \\ $KUBE_LOGTOSTDERR \\ $KUBE_LOG_LEVEL \\ $KUBE_MASTER \\ $KUBE_SCHEDULER_ARGS Restart=on-failure LimitNOFILE=65536 [Install] WantedBy=multi-user.target 配置文件/etc/kubernetes/scheduler。 ### # kubernetes scheduler config # default config should be adequate # Add your own! KUBE_SCHEDULER_ARGS=\"--leader-elect=true --address=127.0.0.1\" --address 值必须为 127.0.0.1，因为当前 kube-apiserver 期望 scheduler 和 controller-manager 在同一台机器； 完整 unit 见 kube-scheduler.service 启动 kube-scheduler systemctl daemon-reload systemctl enable kube-scheduler systemctl start kube-scheduler 验证 master 节点功能 $ kubectl get componentstatuses NAME STATUS MESSAGE ERROR scheduler Healthy ok controller-manager Healthy ok etcd-0 Healthy {\"health\": \"true\"} etcd-1 Healthy {\"health\": \"true\"} etcd-2 Healthy {\"health\": \"true\"} Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-15 15:17:07 "},"practice/flannel-installation.html":{"url":"practice/flannel-installation.html","title":"安装flannel网络插件","keywords":"","body":"安装flannel网络插件 所有的node节点都需要安装网络插件才能让所有的Pod加入到同一个局域网中，本文是安装flannel网络插件的参考文档。 建议直接使用yum安装flanneld，除非对版本有特殊需求，默认安装的是0.7.1版本的flannel。 yum install -y flannel service配置文件/usr/lib/systemd/system/flanneld.service。 [Unit] Description=Flanneld overlay address etcd agent After=network.target After=network-online.target Wants=network-online.target After=etcd.service Before=docker.service [Service] Type=notify EnvironmentFile=/etc/sysconfig/flanneld EnvironmentFile=-/etc/sysconfig/docker-network ExecStart=/usr/bin/flanneld-start \\ -etcd-endpoints=${ETCD_ENDPOINTS} \\ -etcd-prefix=${ETCD_PREFIX} \\ $FLANNEL_OPTIONS ExecStartPost=/usr/libexec/flannel/mk-docker-opts.sh -k DOCKER_NETWORK_OPTIONS -d /run/flannel/docker Restart=on-failure [Install] WantedBy=multi-user.target RequiredBy=docker.service /etc/sysconfig/flanneld配置文件： # Flanneld configuration options # etcd url location. Point this to the server where etcd runs ETCD_ENDPOINTS=\"https://172.20.0.113:2379,https://172.20.0.114:2379,https://172.20.0.115:2379\" # etcd config key. This is the configuration key that flannel queries # For address range assignment ETCD_PREFIX=\"/kube-centos/network\" # Any additional options that you want to pass FLANNEL_OPTIONS=\"-etcd-cafile=/etc/kubernetes/ssl/ca.pem -etcd-certfile=/etc/kubernetes/ssl/kubernetes.pem -etcd-keyfile=/etc/kubernetes/ssl/kubernetes-key.pem\" 在etcd中创建网络配置 执行下面的命令为docker分配IP地址段。 etcdctl --endpoints=https://172.20.0.113:2379,https://172.20.0.114:2379,https://172.20.0.115:2379 \\ --ca-file=/etc/kubernetes/ssl/ca.pem \\ --cert-file=/etc/kubernetes/ssl/kubernetes.pem \\ --key-file=/etc/kubernetes/ssl/kubernetes-key.pem \\ mkdir /kube-centos/network etcdctl --endpoints=https://172.20.0.113:2379,https://172.20.0.114:2379,https://172.20.0.115:2379 \\ --ca-file=/etc/kubernetes/ssl/ca.pem \\ --cert-file=/etc/kubernetes/ssl/kubernetes.pem \\ --key-file=/etc/kubernetes/ssl/kubernetes-key.pem \\ mk /kube-centos/network/config '{\"Network\":\"172.30.0.0/16\",\"SubnetLen\":24,\"Backend\":{\"Type\":\"vxlan\"}}' 如果你要使用host-gw模式，可以直接将vxlan改成host-gw即可。 注：参考网络和集群性能测试那节，最终我们使用的host-gw模式，关于flannel支持的backend模式见：https://github.com/coreos/flannel/blob/master/Documentation/backends.md。 启动flannel systemctl daemon-reload systemctl enable flanneld systemctl start flanneld systemctl status flanneld 现在查询etcd中的内容可以看到： $etcdctl --endpoints=${ETCD_ENDPOINTS} \\ --ca-file=/etc/kubernetes/ssl/ca.pem \\ --cert-file=/etc/kubernetes/ssl/kubernetes.pem \\ --key-file=/etc/kubernetes/ssl/kubernetes-key.pem \\ ls /kube-centos/network/subnets /kube-centos/network/subnets/172.30.14.0-24 /kube-centos/network/subnets/172.30.38.0-24 /kube-centos/network/subnets/172.30.46.0-24 $etcdctl --endpoints=${ETCD_ENDPOINTS} \\ --ca-file=/etc/kubernetes/ssl/ca.pem \\ --cert-file=/etc/kubernetes/ssl/kubernetes.pem \\ --key-file=/etc/kubernetes/ssl/kubernetes-key.pem \\ get /kube-centos/network/config { \"Network\": \"172.30.0.0/16\", \"SubnetLen\": 24, \"Backend\": { \"Type\": \"vxlan\" } } $etcdctl get /kube-centos/network/subnets/172.30.14.0-24 {\"PublicIP\":\"172.20.0.114\",\"BackendType\":\"vxlan\",\"BackendData\":{\"VtepMAC\":\"56:27:7d:1c:08:22\"}} $etcdctl get /kube-centos/network/subnets/172.30.38.0-24 {\"PublicIP\":\"172.20.0.115\",\"BackendType\":\"vxlan\",\"BackendData\":{\"VtepMAC\":\"12:82:83:59:cf:b8\"}} $etcdctl get /kube-centos/network/subnets/172.30.46.0-24 {\"PublicIP\":\"172.20.0.113\",\"BackendType\":\"vxlan\",\"BackendData\":{\"VtepMAC\":\"e6:b2:fd:f6:66:96\"}} 如果可以查看到以上内容证明flannel已经安装完成，下一步是在node节点上安装和配置docker、kubelet、kube-proxy等，请参考下一节部署node节点。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-19 17:48:43 "},"practice/node-installation.html":{"url":"practice/node-installation.html","title":"部署node节点","keywords":"","body":"部署node节点 Kubernetes node节点包含如下组件： Flanneld：参考我之前写的文章Kubernetes基于Flannel的网络配置，之前没有配置TLS，现在需要在service配置文件中增加TLS配置，安装过程请参考上一节安装flannel网络插件。 Docker1.12.5：docker的安装很简单，这里也不说了，但是需要注意docker的配置。 kubelet：直接用二进制文件安装 kube-proxy：直接用二进制文件安装 注意：每台 node 上都需要安装 flannel，master 节点上可以不安装。 步骤简介 确认在上一步中我们安装配置的网络插件flannel已启动且运行正常 安装配置docker后启动 安装配置kubelet、kube-proxy后启动 验证 目录和文件 我们再检查一下三个节点上，经过前几步操作我们已经创建了如下的证书和配置文件。 $ ls /etc/kubernetes/ssl admin-key.pem admin.pem ca-key.pem ca.pem kube-proxy-key.pem kube-proxy.pem kubernetes-key.pem kubernetes.pem $ ls /etc/kubernetes/ apiserver bootstrap.kubeconfig config controller-manager kubelet kube-proxy.kubeconfig proxy scheduler ssl token.csv 配置Docker 如果您使用yum的方式安装的flannel则不需要执行mk-docker-opts.sh文件这一步，参考Flannel官方文档中的Docker Integration。 如果你不是使用yum安装的flannel，那么需要下载flannel github release中的tar包，解压后会获得一个mk-docker-opts.sh文件，到flannel release页面下载对应版本的安装包，该脚本见mk-docker-opts.sh，因为我们使用yum安装所以不需要执行这一步。 这个文件是用来Generate Docker daemon options based on flannel env file。 使用systemctl命令启动flanneld后，会自动执行./mk-docker-opts.sh -i生成如下两个文件环境变量文件： /run/flannel/subnet.env FLANNEL_NETWORK=172.30.0.0/16 FLANNEL_SUBNET=172.30.46.1/24 FLANNEL_MTU=1450 FLANNEL_IPMASQ=false /run/docker_opts.env DOCKER_OPT_BIP=\"--bip=172.30.46.1/24\" DOCKER_OPT_IPMASQ=\"--ip-masq=true\" DOCKER_OPT_MTU=\"--mtu=1450\" Docker将会读取这两个环境变量文件作为容器启动参数。 注意：不论您用什么方式安装的flannel，下面这一步是必不可少的。 yum方式安装的flannel 修改docker的配置文件/usr/lib/systemd/system/docker.service，增加一条环境变量配置： EnvironmentFile=-/run/flannel/docker /run/flannel/docker文件是flannel启动后自动生成的，其中包含了docker启动时需要的参数。 二进制方式安装的flannel 修改docker的配置文件/usr/lib/systemd/system/docker.service，增加如下几条环境变量配置： EnvironmentFile=-/run/docker_opts.env EnvironmentFile=-/run/flannel/subnet.env 这两个文件是mk-docker-opts.sh脚本生成环境变量文件默认的保存位置，docker启动的时候需要加载这几个配置文件才可以加入到flannel创建的虚拟网络里。 所以不论您使用何种方式安装的flannel，将以下配置加入到docker.service中可确保万无一失。 EnvironmentFile=-/run/flannel/docker EnvironmentFile=-/run/docker_opts.env EnvironmentFile=-/run/flannel/subnet.env EnvironmentFile=-/etc/sysconfig/docker EnvironmentFile=-/etc/sysconfig/docker-storage EnvironmentFile=-/etc/sysconfig/docker-network EnvironmentFile=-/run/docker_opts.env 请参考docker.service中的配置。 启动docker 重启了docker后还要重启kubelet，这时又遇到问题，kubelet启动失败。报错： Mar 31 16:44:41 sz-pg-oam-docker-test-002.tendcloud.com kubelet[81047]: error: failed to run Kubelet: failed to create kubelet: misconfiguration: kubelet cgroup driver: \"cgroupfs\" is different from docker cgroup driver: \"systemd\" 这是kubelet与docker的cgroup driver不一致导致的，kubelet启动的时候有个—cgroup-driver参数可以指定为\"cgroupfs\"或者“systemd”。 --cgroup-driver string Driver that the kubelet uses to manipulate cgroups on the host. Possible values: 'cgroupfs', 'systemd' (default \"cgroupfs\") 配置docker的service配置文件/usr/lib/systemd/system/docker.service，设置ExecStart中的--exec-opt native.cgroupdriver=systemd。 安装和配置kubelet kubernets1.8 相对于kubernetes1.6集群必须进行的配置有： 对于kuberentes1.8集群，必须关闭swap，否则kubelet启动将失败。 修改/etc/fstab将，swap系统注释掉。 kubelet 启动时向 kube-apiserver 发送 TLS bootstrapping 请求，需要先将 bootstrap token 文件中的 kubelet-bootstrap 用户赋予 system:node-bootstrapper cluster 角色(role)， 然后 kubelet 才能有权限创建认证请求(certificate signing requests)： cd /etc/kubernetes kubectl create clusterrolebinding kubelet-bootstrap \\ --clusterrole=system:node-bootstrapper \\ --user=kubelet-bootstrap --user=kubelet-bootstrap 是在 /etc/kubernetes/token.csv 文件中指定的用户名，同时也写入了 /etc/kubernetes/bootstrap.kubeconfig 文件； 下载最新的kubelet和kube-proxy二进制文件 wget https://dl.k8s.io/v1.6.0/kubernetes-server-linux-amd64.tar.gz tar -xzvf kubernetes-server-linux-amd64.tar.gz cd kubernetes tar -xzvf kubernetes-src.tar.gz cp -r ./server/bin/{kube-proxy,kubelet} /usr/local/bin/ 创建kubelet的service配置文件 文件位置/usr/lib/systemd/system/kubelet.service。 [Unit] Description=Kubernetes Kubelet Server Documentation=https://github.com/GoogleCloudPlatform/kubernetes After=docker.service Requires=docker.service [Service] WorkingDirectory=/var/lib/kubelet EnvironmentFile=-/etc/kubernetes/config EnvironmentFile=-/etc/kubernetes/kubelet ExecStart=/usr/local/bin/kubelet \\ $KUBE_LOGTOSTDERR \\ $KUBE_LOG_LEVEL \\ $KUBELET_API_SERVER \\ $KUBELET_ADDRESS \\ $KUBELET_PORT \\ $KUBELET_HOSTNAME \\ $KUBE_ALLOW_PRIV \\ $KUBELET_POD_INFRA_CONTAINER \\ $KUBELET_ARGS Restart=on-failure [Install] WantedBy=multi-user.target kubelet的配置文件/etc/kubernetes/kubelet。其中的IP地址更改为你的每台node节点的IP地址。 注意：在启动kubelet之前，需要先手动创建/var/lib/kubelet目录。 下面是kubelet的配置文件/etc/kubernetes/kubelet: kubernetes1.8 相对于kubenrete1.6的配置变动： 对于kuberentes1.8集群中的kubelet配置，取消了KUBELET_API_SERVER的配置，而改用kubeconfig文件来定义master地址。 ### ## kubernetes kubelet (minion) config # ## The address for the info server to serve on (set to 0.0.0.0 or \"\" for all interfaces) KUBELET_ADDRESS=\"--address=172.20.0.113\" # ## The port for the info server to serve on #KUBELET_PORT=\"--port=10250\" # ## You may leave this blank to use the actual hostname KUBELET_HOSTNAME=\"--hostname-override=172.20.0.113\" # ## location of the api-server KUBELET_API_SERVER=\"--api-servers=http://172.20.0.113:8080\" # ## pod infrastructure container KUBELET_POD_INFRA_CONTAINER=\"--pod-infra-container-image=sz-pg-oam-docker-hub-001.tendcloud.com/library/pod-infrastructure:rhel7\" # ## Add your own! KUBELET_ARGS=\"--cgroup-driver=systemd --cluster-dns=10.254.0.2 --experimental-bootstrap-kubeconfig=/etc/kubernetes/bootstrap.kubeconfig --kubeconfig=/etc/kubernetes/kubelet.kubeconfig --require-kubeconfig --cert-dir=/etc/kubernetes/ssl --cluster-domain=cluster.local --hairpin-mode promiscuous-bridge --serialize-image-pulls=false\" --address 不能设置为 127.0.0.1，否则后续 Pods 访问 kubelet 的 API 接口时会失败，因为 Pods 访问的 127.0.0.1 指向自己而不是 kubelet； 如果设置了 --hostname-override 选项，则 kube-proxy 也需要设置该选项，否则会出现找不到 Node 的情况； \"--cgroup-driver 配置成 systemd，不要使用cgroup，否则在 CentOS 系统中 kubelet 将启动失败（保持docker和kubelet中的cgroup driver配置一致即可，不一定非使用systemd）。 --experimental-bootstrap-kubeconfig 指向 bootstrap kubeconfig 文件，kubelet 使用该文件中的用户名和 token 向 kube-apiserver 发送 TLS Bootstrapping 请求； 管理员通过了 CSR 请求后，kubelet 自动在 --cert-dir 目录创建证书和私钥文件(kubelet-client.crt 和 kubelet-client.key)，然后写入 --kubeconfig 文件； 建议在 --kubeconfig 配置文件中指定 kube-apiserver 地址，如果未指定 --api-servers 选项，则必须指定 --require-kubeconfig 选项后才从配置文件中读取 kube-apiserver 的地址，否则 kubelet 启动后将找不到 kube-apiserver (日志中提示未找到 API Server），kubectl get nodes 不会返回对应的 Node 信息; --cluster-dns 指定 kubedns 的 Service IP(可以先分配，后续创建 kubedns 服务时指定该 IP)，--cluster-domain 指定域名后缀，这两个参数同时指定后才会生效； --cluster-domain 指定 pod 启动时 /etc/resolve.conf 文件中的 search domain ，起初我们将其配置成了 cluster.local.，这样在解析 service 的 DNS 名称时是正常的，可是在解析 headless service 中的 FQDN pod name 的时候却错误，因此我们将其修改为 cluster.local，去掉嘴后面的 ”点号“ 就可以解决该问题，关于 kubernetes 中的域名/服务名称解析请参见我的另一篇文章。 --kubeconfig=/etc/kubernetes/kubelet.kubeconfig中指定的kubelet.kubeconfig文件在第一次启动kubelet之前并不存在，请看下文，当通过CSR请求后会自动生成kubelet.kubeconfig文件，如果你的节点上已经生成了~/.kube/config文件，你可以将该文件拷贝到该路径下，并重命名为kubelet.kubeconfig，所有node节点可以共用同一个kubelet.kubeconfig文件，这样新添加的节点就不需要再创建CSR请求就能自动添加到kubernetes集群中。同样，在任意能够访问到kubernetes集群的主机上使用kubectl --kubeconfig命令操作集群时，只要使用~/.kube/config文件就可以通过权限认证，因为这里面已经有认证信息并认为你是admin用户，对集群拥有所有权限。 KUBELET_POD_INFRA_CONTAINER 是基础镜像容器，这里我用的是私有镜像仓库地址，大家部署的时候需要修改为自己的镜像。我上传了一个到时速云上，可以直接 docker pull index.tenxcloud.com/jimmy/pod-infrastructure 下载。 完整 unit 见 kubelet.service 启动kublet systemctl daemon-reload systemctl enable kubelet systemctl start kubelet systemctl status kubelet 通过 kublet 的 TLS 证书请求 kubelet 首次启动时向 kube-apiserver 发送证书签名请求，必须通过后 kubernetes 系统才会将该 Node 加入到集群。 查看未授权的 CSR 请求 $ kubectl get csr NAME AGE REQUESTOR CONDITION csr-2b308 4m kubelet-bootstrap Pending $ kubectl get nodes No resources found. 通过 CSR 请求 $ kubectl certificate approve csr-2b308 certificatesigningrequest \"csr-2b308\" approved $ kubectl get nodes NAME STATUS AGE VERSION 10.64.3.7 Ready 49m v1.6.1 自动生成了 kubelet kubeconfig 文件和公私钥 $ ls -l /etc/kubernetes/kubelet.kubeconfig -rw------- 1 root root 2284 Apr 7 02:07 /etc/kubernetes/kubelet.kubeconfig $ ls -l /etc/kubernetes/ssl/kubelet* -rw-r--r-- 1 root root 1046 Apr 7 02:07 /etc/kubernetes/ssl/kubelet-client.crt -rw------- 1 root root 227 Apr 7 02:04 /etc/kubernetes/ssl/kubelet-client.key -rw-r--r-- 1 root root 1103 Apr 7 02:07 /etc/kubernetes/ssl/kubelet.crt -rw------- 1 root root 1675 Apr 7 02:07 /etc/kubernetes/ssl/kubelet.key 假如你更新kubernetes的证书，只要没有更新token.csv，当重启kubelet后，该node就会自动加入到kuberentes集群中，而不会重新发送certificaterequest，也不需要在master节点上执行kubectl certificate approve操作。前提是不要删除node节点上的/etc/kubernetes/ssl/kubelet*和/etc/kubernetes/kubelet.kubeconfig文件。否则kubelet启动时会提示找不到证书而失败。 注意：如果启动kubelet的时候见到证书相关的报错，有个trick可以解决这个问题，可以将master节点上的~/.kube/config文件（该文件在安装kubectl命令行工具这一步中将会自动生成）拷贝到node节点的/etc/kubernetes/kubelet.kubeconfig位置，这样就不需要通过CSR，当kubelet启动后就会自动加入的集群中。 配置 kube-proxy 安装conntrack yum install -y conntrack-tools 创建 kube-proxy 的service配置文件 文件路径/usr/lib/systemd/system/kube-proxy.service。 [Unit] Description=Kubernetes Kube-Proxy Server Documentation=https://github.com/GoogleCloudPlatform/kubernetes After=network.target [Service] EnvironmentFile=-/etc/kubernetes/config EnvironmentFile=-/etc/kubernetes/proxy ExecStart=/usr/local/bin/kube-proxy \\ $KUBE_LOGTOSTDERR \\ $KUBE_LOG_LEVEL \\ $KUBE_MASTER \\ $KUBE_PROXY_ARGS Restart=on-failure LimitNOFILE=65536 [Install] WantedBy=multi-user.target kube-proxy配置文件/etc/kubernetes/proxy。 ### # kubernetes proxy config # default config should be adequate # Add your own! KUBE_PROXY_ARGS=\"--bind-address=172.20.0.113 --hostname-override=172.20.0.113 --kubeconfig=/etc/kubernetes/kube-proxy.kubeconfig --cluster-cidr=10.254.0.0/16\" --hostname-override 参数值必须与 kubelet 的值一致，否则 kube-proxy 启动后会找不到该 Node，从而不会创建任何 iptables 规则； kube-proxy 根据 --cluster-cidr 判断集群内部和外部流量，指定 --cluster-cidr 或 --masquerade-all 选项后 kube-proxy 才会对访问 Service IP 的请求做 SNAT； --kubeconfig 指定的配置文件嵌入了 kube-apiserver 的地址、用户名、证书、秘钥等请求和认证信息； 预定义的 RoleBinding cluster-admin 将User system:kube-proxy 与 Role system:node-proxier 绑定，该 Role 授予了调用 kube-apiserver Proxy 相关 API 的权限； 完整 unit 见 kube-proxy.service 启动 kube-proxy systemctl daemon-reload systemctl enable kube-proxy systemctl start kube-proxy systemctl status kube-proxy 验证测试 我们创建一个nginx的service试一下集群是否可用。 $ kubectl run nginx --replicas=2 --labels=\"run=load-balancer-example\" --image=sz-pg-oam-docker-hub-001.tendcloud.com/library/nginx:1.9 --port=80 deployment \"nginx\" created $ kubectl expose deployment nginx --type=NodePort --name=example-service service \"example-service\" exposed $ kubectl describe svc example-service Name: example-service Namespace: default Labels: run=load-balancer-example Annotations: Selector: run=load-balancer-example Type: NodePort IP: 10.254.62.207 Port: 80/TCP NodePort: 32724/TCP Endpoints: 172.30.60.2:80,172.30.94.2:80 Session Affinity: None Events: $ curl \"10.254.62.207:80\" Welcome to nginx! body { width: 35em; margin: 0 auto; font-family: Tahoma, Verdana, Arial, sans-serif; } Welcome to nginx! If you see this page, the nginx web server is successfully installed and working. Further configuration is required. For online documentation and support please refer to nginx.org. Commercial support is available at nginx.com. Thank you for using nginx. 提示：上面的测试示例中使用的nginx是我的私有镜像仓库中的镜像sz-pg-oam-docker-hub-001.tendcloud.com/library/nginx:1.9，大家在测试过程中请换成自己的nginx镜像地址。 访问172.20.0.113:32724或172.20.0.114:32724或者172.20.0.115:32724都可以得到nginx的页面。 图片 - welcome nginx 参考 Kubelet 的认证授权 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-19 17:45:18 "},"practice/kubedns-addon-installation.html":{"url":"practice/kubedns-addon-installation.html","title":"安装kubedns插件","keywords":"","body":"安装kubedns插件 官方的yaml文件目录：kubernetes/cluster/addons/dns。 该插件直接使用kubernetes部署，官方的配置文件中包含以下镜像： gcr.io/google_containers/k8s-dns-dnsmasq-nanny-amd64:1.14.1 gcr.io/google_containers/k8s-dns-kube-dns-amd64:1.14.1 gcr.io/google_containers/k8s-dns-sidecar-amd64:1.14.1 我clone了上述镜像，上传到我的私有镜像仓库： sz-pg-oam-docker-hub-001.tendcloud.com/library/k8s-dns-dnsmasq-nanny-amd64:1.14.1 sz-pg-oam-docker-hub-001.tendcloud.com/library/k8s-dns-kube-dns-amd64:1.14.1 sz-pg-oam-docker-hub-001.tendcloud.com/library/k8s-dns-sidecar-amd64:1.14.1 同时上传了一份到时速云备份： index.tenxcloud.com/jimmy/k8s-dns-dnsmasq-nanny-amd64:1.14.1 index.tenxcloud.com/jimmy/k8s-dns-kube-dns-amd64:1.14.1 index.tenxcloud.com/jimmy/k8s-dns-sidecar-amd64:1.14.1 以下yaml配置文件中使用的是私有镜像仓库中的镜像。 kubedns-cm.yaml kubedns-sa.yaml kubedns-controller.yaml kubedns-svc.yaml 已经修改好的 yaml 文件见：../manifests/kubedns 系统预定义的 RoleBinding 预定义的 RoleBinding system:kube-dns 将 kube-system 命名空间的 kube-dns ServiceAccount 与 system:kube-dns Role 绑定， 该 Role 具有访问 kube-apiserver DNS 相关 API 的权限； $ kubectl get clusterrolebindings system:kube-dns -o yaml apiVersion: rbac.authorization.k8s.io/v1beta1 kind: ClusterRoleBinding metadata: annotations: rbac.authorization.kubernetes.io/autoupdate: \"true\" creationTimestamp: 2017-04-11T11:20:42Z labels: kubernetes.io/bootstrapping: rbac-defaults name: system:kube-dns resourceVersion: \"58\" selfLink: /apis/rbac.authorization.k8s.io/v1beta1/clusterrolebindingssystem%3Akube-dns uid: e61f4d92-1ea8-11e7-8cd7-f4e9d49f8ed0 roleRef: apiGroup: rbac.authorization.k8s.io kind: ClusterRole name: system:kube-dns subjects: - kind: ServiceAccount name: kube-dns namespace: kube-system kubedns-controller.yaml 中定义的 Pods 时使用了 kubedns-sa.yaml 文件定义的 kube-dns ServiceAccount，所以具有访问 kube-apiserver DNS 相关 API 的权限。 配置 kube-dns ServiceAccount 无需修改。 配置 kube-dns 服务 $ diff kubedns-svc.yaml.base kubedns-svc.yaml 30c30 clusterIP: 10.254.0.2 spec.clusterIP = 10.254.0.2，即明确指定了 kube-dns Service IP，这个 IP 需要和 kubelet 的 --cluster-dns 参数值一致； 配置 kube-dns Deployment $ diff kubedns-controller.yaml.base kubedns-controller.yaml 58c58 image: sz-pg-oam-docker-hub-001.tendcloud.com/library/k8s-dns-kube-dns-amd64:v1.14.1 88c88 - --domain=cluster.local. 92c92 #__PILLAR__FEDERATIONS__DOMAIN__MAP__ 110c110 image: sz-pg-oam-docker-hub-001.tendcloud.com/library/k8s-dns-dnsmasq-nanny-amd64:v1.14.1 129c129 - --server=/cluster.local./127.0.0.1#10053 148c148 image: sz-pg-oam-docker-hub-001.tendcloud.com/library/k8s-dns-sidecar-amd64:v1.14.1 161,162c161,162 - --probe=kubedns,127.0.0.1:10053,kubernetes.default.svc.cluster.local.,5,A > - --probe=dnsmasq,127.0.0.1:53,kubernetes.default.svc.cluster.local.,5,A 使用系统已经做了 RoleBinding 的 kube-dns ServiceAccount，该账户具有访问 kube-apiserver DNS 相关 API 的权限； 执行所有定义文件 $ pwd /root/kubedns $ ls *.yaml kubedns-cm.yaml kubedns-controller.yaml kubedns-sa.yaml kubedns-svc.yaml $ kubectl create -f . 检查 kubedns 功能 新建一个 Deployment $ cat my-nginx.yaml apiVersion: extensions/v1beta1 kind: Deployment metadata: name: my-nginx spec: replicas: 2 template: metadata: labels: run: my-nginx spec: containers: - name: my-nginx image: sz-pg-oam-docker-hub-001.tendcloud.com/library/nginx:1.9 ports: - containerPort: 80 $ kubectl create -f my-nginx.yaml Export 该 Deployment, 生成 my-nginx 服务 $ kubectl expose deploy my-nginx $ kubectl get services --all-namespaces |grep my-nginx default my-nginx 10.254.179.239 80/TCP 42m 创建另一个 Pod，查看 /etc/resolv.conf 是否包含 kubelet 配置的 --cluster-dns 和 --cluster-domain，是否能够将服务 my-nginx 解析到 Cluster IP 10.254.179.239。 $ kubectl create -f nginx-pod.yaml $ kubectl exec nginx -i -t -- /bin/bash root@nginx:/# cat /etc/resolv.conf nameserver 10.254.0.2 search default.svc.cluster.local. svc.cluster.local. cluster.local. tendcloud.com options ndots:5 root@nginx:/# ping my-nginx PING my-nginx.default.svc.cluster.local (10.254.179.239): 56 data bytes 76 bytes from 119.147.223.109: Destination Net Unreachable ^C--- my-nginx.default.svc.cluster.local ping statistics --- root@nginx:/# ping kubernetes PING kubernetes.default.svc.cluster.local (10.254.0.1): 56 data bytes ^C--- kubernetes.default.svc.cluster.local ping statistics --- 11 packets transmitted, 0 packets received, 100% packet loss root@nginx:/# ping kube-dns.kube-system.svc.cluster.local PING kube-dns.kube-system.svc.cluster.local (10.254.0.2): 56 data bytes ^C--- kube-dns.kube-system.svc.cluster.local ping statistics --- 6 packets transmitted, 0 packets received, 100% packet loss 从结果来看，service名称可以正常解析。 注意：直接ping ClusterIP是ping不通的，ClusterIP是根据IPtables路由到服务的endpoint上，只有结合ClusterIP加端口才能访问到对应的服务。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-10-12 10:11:37 "},"practice/dashboard-addon-installation.html":{"url":"practice/dashboard-addon-installation.html","title":"安装dashboard插件","keywords":"","body":"安装dashboard插件 官方文件目录：kubernetes/cluster/addons/dashboard 我们使用的文件如下： $ ls *.yaml dashboard-controller.yaml dashboard-service.yaml dashboard-rbac.yaml 已经修改好的 yaml 文件见：../manifests/dashboard 由于 kube-apiserver 启用了 RBAC 授权，而官方源码目录的 dashboard-controller.yaml 没有定义授权的 ServiceAccount，所以后续访问 API server 的 API 时会被拒绝，web中提示： Forbidden (403) User \"system:serviceaccount:kube-system:default\" cannot list jobs.batch in the namespace \"default\". (get jobs.batch) 增加了一个dashboard-rbac.yaml文件，定义一个名为 dashboard 的 ServiceAccount，然后将它和 Cluster Role view 绑定。 配置dashboard-service $ diff dashboard-service.yaml.orig dashboard-service.yaml 10a11 > type: NodePort 指定端口类型为 NodePort，这样外界可以通过地址 nodeIP:nodePort 访问 dashboard； 配置dashboard-controller $ diff dashboard-controller.yaml.orig dashboard-controller.yaml 23c23 image: sz-pg-oam-docker-hub-001.tendcloud.com/library/kubernetes-dashboard-amd64:v1.6.0 执行所有定义文件 $ pwd /root/kubernetes/cluster/addons/dashboard $ ls *.yaml dashboard-controller.yaml dashboard-service.yaml $ kubectl create -f . service \"kubernetes-dashboard\" created deployment \"kubernetes-dashboard\" created 检查执行结果 查看分配的 NodePort $ kubectl get services kubernetes-dashboard -n kube-system NAME CLUSTER-IP EXTERNAL-IP PORT(S) AGE kubernetes-dashboard 10.254.224.130 80:30312/TCP 25s NodePort 30312映射到 dashboard pod 80端口； 检查 controller $ kubectl get deployment kubernetes-dashboard -n kube-system NAME DESIRED CURRENT UP-TO-DATE AVAILABLE AGE kubernetes-dashboard 1 1 1 1 3m $ kubectl get pods -n kube-system | grep dashboard kubernetes-dashboard-1339745653-pmn6z 1/1 Running 0 4m 访问dashboard 有以下三种方式： kubernetes-dashboard 服务暴露了 NodePort，可以使用 http://NodeIP:nodePort 地址访问 dashboard 通过 API server 访问 dashboard（https 6443端口和http 8080端口方式） 通过 kubectl proxy 访问 dashboard 通过 kubectl proxy 访问 dashboard 启动代理 $ kubectl proxy --address='172.20.0.113' --port=8086 --accept-hosts='^*$' Starting to serve on 172.20.0.113:8086 需要指定 --accept-hosts 选项，否则浏览器访问 dashboard 页面时提示 “Unauthorized”； 浏览器访问 URL：http://172.20.0.113:8086/ui 自动跳转到：http://172.20.0.113:8086/api/v1/proxy/namespaces/kube-system/services/kubernetes-dashboard/#/workload?namespace=default 通过 API server 访问dashboard 获取集群服务地址列表 $ kubectl cluster-info Kubernetes master is running at https://172.20.0.113:6443 KubeDNS is running at https://172.20.0.113:6443/api/v1/proxy/namespaces/kube-system/services/kube-dns kubernetes-dashboard is running at https://172.20.0.113:6443/api/v1/proxy/namespaces/kube-system/services/kubernetes-dashboard 浏览器访问 URL：https://172.20.0.113:6443/api/v1/proxy/namespaces/kube-system/services/kubernetes-dashboard（浏览器会提示证书验证，因为通过加密通道，以改方式访问的话，需要提前导入证书到你的计算机中）。这是我当时在这遇到的坑：[通过 kube-apiserver 访问dashboard，提示User \"system:anonymous\" cannot proxy services in the namespace \"kube-system\". #5](https://github.com/opsnull/follow-me-install-kubernetes-cluster/issues/5)，已经解决。 导入证书 将生成的admin.pem证书转换格式 openssl pkcs12 -export -in admin.pem -out admin.p12 -inkey admin-key.pem 将生成的admin.p12证书导入的你的电脑，导出的时候记住你设置的密码，导入的时候还要用到。 如果你不想使用https的话，可以直接访问insecure port 8080端口：http://172.20.0.113:8080/api/v1/proxy/namespaces/kube-system/services/kubernetes-dashboard 图片 - kubernetes dashboard 由于缺少 Heapster 插件，当前 dashboard 不能展示 Pod、Nodes 的 CPU、内存等 metric 图形。 更新 Kubernetes 1.6 版本的 dashboard 的镜像已经到了 v1.6.3 版本，我们可以使用下面的方式更新。 修改 dashboard-controller.yaml 文件中的镜像的版本将 v1.6.0 更改为 v1.6.3。 image: sz-pg-oam-docker-hub-001.tendcloud.com/library/kubernetes-dashboard-amd64:v1.6.3 然后执行下面的命令： kubectl apply -f dashboard-controller.yaml 即可在线更新 dashboard 的版本。 监听 dashboard Pod 的状态可以看到： kubernetes-dashboard-215087767-2jsgd 0/1 Pending 0 0s kubernetes-dashboard-3966630548-0jj1j 1/1 Terminating 0 1d kubernetes-dashboard-215087767-2jsgd 0/1 Pending 0 0s kubernetes-dashboard-3966630548-0jj1j 1/1 Terminating 0 1d kubernetes-dashboard-215087767-2jsgd 0/1 ContainerCreating 0 0s kubernetes-dashboard-3966630548-0jj1j 0/1 Terminating 0 1d kubernetes-dashboard-3966630548-0jj1j 0/1 Terminating 0 1d kubernetes-dashboard-215087767-2jsgd 1/1 Running 0 6s kubernetes-dashboard-3966630548-0jj1j 0/1 Terminating 0 1d kubernetes-dashboard-3966630548-0jj1j 0/1 Terminating 0 1d kubernetes-dashboard-3966630548-0jj1j 0/1 Terminating 0 1d 新的 Pod 的启动了，旧的 Pod 被终结了。 Dashboard 的访问地址不变，重新访问 http://172.20.0.113:8080/api/v1/proxy/namespaces/kube-system/services/kubernetes-dashboard，可以看到新版的界面： 图片 - V1.6.3版本的dashboard界面 新版本中最大的变化是增加了进入容器内部的入口，可以在页面上进入到容器内部操作，同时又增加了一个搜索框。 关于如何将dashboard从1.6版本升级到1.7版本请参考升级dashboard。 参考 WebUI(Dashboard) 文档 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-15 21:36:09 "},"practice/heapster-addon-installation.html":{"url":"practice/heapster-addon-installation.html","title":"安装heapster插件","keywords":"","body":"安装heapster插件 准备镜像 官方镜像保存在 gcr.io 中需要翻墙才能下载，为了方便大家使用，我下载后放到了时速云中，为公有镜像供大家下载。 index.tenxcloud.com/jimmy/heapster-amd64:v1.3.0-beta.1 index.tenxcloud.com/jimmy/heapster-influxdb-amd64:v1.1.1 index.tenxcloud.com/jimmy/heapster-grafana-amd64:v4.0.2 准备YAML文件 到 heapster release 页面 下载最新版本的 heapster。 wget https://github.com/kubernetes/heapster/archive/v1.3.0.zip unzip v1.3.0.zip mv v1.3.0.zip heapster-1.3.0 文件目录： heapster-1.3.0/deploy/kube-config/influxdb $ cd heapster-1.3.0/deploy/kube-config/influxdb $ ls *.yaml grafana-deployment.yaml grafana-service.yaml heapster-deployment.yaml heapster-service.yaml influxdb-deployment.yaml influxdb-service.yaml heapster-rbac.yaml 我们自己创建了heapster的rbac配置heapster-rbac.yaml。 已经修改好的 yaml 文件见：../manifests/heapster 配置 grafana-deployment $ diff grafana-deployment.yaml.orig grafana-deployment.yaml 16c16 image: sz-pg-oam-docker-hub-001.tendcloud.com/library/heapster-grafana-amd64:v4.0.2 40,41c40,41 value: /api/v1/proxy/namespaces/kube-system/services/monitoring-grafana/ > #value: / 如果后续使用 kube-apiserver 或者 kubectl proxy 访问 grafana dashboard，则必须将 GF_SERVER_ROOT_URL 设置为 /api/v1/proxy/namespaces/kube-system/services/monitoring-grafana/，否则后续访问grafana时访问时提示找不到http://172.20.0.113:8086/api/v1/proxy/namespaces/kube-system/services/monitoring-grafana/api/dashboards/home 页面； 配置 heapster-deployment $ diff heapster-deployment.yaml.orig heapster-deployment.yaml 16c16 image: sz-pg-oam-docker-hub-001.tendcloud.com/library/heapster-amd64:v1.3.0-beta.1 配置 influxdb-deployment influxdb 官方建议使用命令行或 HTTP API 接口来查询数据库，从 v1.1.0 版本开始默认关闭 admin UI，将在后续版本中移除 admin UI 插件。 开启镜像中 admin UI的办法如下：先导出镜像中的 influxdb 配置文件，开启 admin 插件后，再将配置文件内容写入 ConfigMap，最后挂载到镜像中，达到覆盖原始配置的目的： 注意：manifests 目录已经提供了 修改后的 ConfigMap 定义文件 $ # 导出镜像中的 influxdb 配置文件 $ docker run --rm --entrypoint 'cat' -ti lvanneo/heapster-influxdb-amd64:v1.1.1 /etc/config.toml >config.toml.orig $ cp config.toml.orig config.toml $ # 修改：启用 admin 接口 $ vim config.toml $ diff config.toml.orig config.toml 35c35 enabled = true $ # 将修改后的配置写入到 ConfigMap 对象中 $ kubectl create configmap influxdb-config --from-file=config.toml -n kube-system configmap \"influxdb-config\" created $ # 将 ConfigMap 中的配置文件挂载到 Pod 中，达到覆盖原始配置的目的 $ diff influxdb-deployment.yaml.orig influxdb-deployment.yaml 16c16 image: sz-pg-oam-docker-hub-001.tendcloud.com/library/heapster-influxdb-amd64:v1.1.1 19a20,21 > - mountPath: /etc/ > name: influxdb-config 22a25,27 > - name: influxdb-config > configMap: > name: influxdb-config 配置 monitoring-influxdb Service $ diff influxdb-service.yaml.orig influxdb-service.yaml 12a13 > type: NodePort 15a17,20 > name: http > - port: 8083 > targetPort: 8083 > name: admin 定义端口类型为 NodePort，额外增加了 admin 端口映射，用于后续浏览器访问 influxdb 的 admin UI 界面； 执行所有定义文件 $ pwd /root/heapster-1.3.0/deploy/kube-config/influxdb $ ls *.yaml grafana-service.yaml heapster-rbac.yaml influxdb-cm.yaml influxdb-service.yaml grafana-deployment.yaml heapster-deployment.yaml heapster-service.yaml influxdb-deployment.yaml $ kubectl create -f . deployment \"monitoring-grafana\" created service \"monitoring-grafana\" created deployment \"heapster\" created serviceaccount \"heapster\" created clusterrolebinding \"heapster\" created service \"heapster\" created configmap \"influxdb-config\" created deployment \"monitoring-influxdb\" created service \"monitoring-influxdb\" created 检查执行结果 检查 Deployment $ kubectl get deployments -n kube-system | grep -E 'heapster|monitoring' heapster 1 1 1 1 2m monitoring-grafana 1 1 1 1 2m monitoring-influxdb 1 1 1 1 2m 检查 Pods $ kubectl get pods -n kube-system | grep -E 'heapster|monitoring' heapster-110704576-gpg8v 1/1 Running 0 2m monitoring-grafana-2861879979-9z89f 1/1 Running 0 2m monitoring-influxdb-1411048194-lzrpc 1/1 Running 0 2m 检查 kubernets dashboard 界面，看是显示各 Nodes、Pods 的 CPU、内存、负载等利用率曲线图； 图片 - dashboard-heapster 访问 grafana 通过 kube-apiserver 访问： 获取 monitoring-grafana 服务 URL $ kubectl cluster-info Kubernetes master is running at https://172.20.0.113:6443 Heapster is running at https://172.20.0.113:6443/api/v1/proxy/namespaces/kube-system/services/heapster KubeDNS is running at https://172.20.0.113:6443/api/v1/proxy/namespaces/kube-system/services/kube-dns kubernetes-dashboard is running at https://172.20.0.113:6443/api/v1/proxy/namespaces/kube-system/services/kubernetes-dashboard monitoring-grafana is running at https://172.20.0.113:6443/api/v1/proxy/namespaces/kube-system/services/monitoring-grafana monitoring-influxdb is running at https://172.20.0.113:6443/api/v1/proxy/namespaces/kube-system/services/monitoring-influxdb To further debug and diagnose cluster problems, use 'kubectl cluster-info dump'. 浏览器访问 URL： http://172.20.0.113:8080/api/v1/proxy/namespaces/kube-system/services/monitoring-grafana 通过 kubectl proxy 访问： 创建代理 $ kubectl proxy --address='172.20.0.113' --port=8086 --accept-hosts='^*$' Starting to serve on 172.20.0.113:8086 浏览器访问 URL：http://172.20.0.113:8086/api/v1/proxy/namespaces/kube-system/services/monitoring-grafana 图片 - grafana 访问 influxdb admin UI 获取 influxdb http 8086 映射的 NodePort $ kubectl get svc -n kube-system|grep influxdb monitoring-influxdb 10.254.22.46 8086:32299/TCP,8083:30269/TCP 9m 通过 kube-apiserver 的非安全端口访问 influxdb 的 admin UI 界面： http://172.20.0.113:8080/api/v1/proxy/namespaces/kube-system/services/monitoring-influxdb:8083/ 在页面的 “Connection Settings” 的 Host 中输入 node IP， Port 中输入 8086 映射的 nodePort 如上面的 32299，点击 “Save” 即可（我的集群中的地址是172.20.0.113:32299）： 图片 - kubernetes-influxdb-heapster 注意 在安装好 Grafana 之后我们使用的是默认的 template 配置，页面上的 namespace 选择里只有 default 和 kube-system，并不是说其他的 namespace 里的指标没有得到监控，只是我们没有在 Grafana 中开启他它们的显示而已。见 Cannot see other namespaces except, kube-system and default #1279。 图片 - 修改grafana模板 将 Templating 中的 namespace 的 Data source 设置为 influxdb-datasource，Refresh 设置为 on Dashboard Load 保存设置，刷新浏览器，即可看到其他 namespace 选项。 参考 使用Heapster获取集群对象的metric数据 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-21 11:57:43 "},"practice/efk-addon-installation.html":{"url":"practice/efk-addon-installation.html","title":"安装EFK插件","keywords":"","body":"安装EFK插件 我们通过在每台node上部署一个以DaemonSet方式运行的fluentd来收集每台node上的日志。Fluentd将docker日志目录/var/lib/docker/containers和/var/log目录挂载到Pod中，然后Pod会在node节点的/var/log/pods目录中创建新的目录，可以区别不同的容器日志输出，该目录下有一个日志文件链接到/var/lib/docker/contianers目录下的容器日志输出。 官方文件目录：cluster/addons/fluentd-elasticsearch $ ls *.yaml es-controller.yaml es-service.yaml fluentd-es-ds.yaml kibana-controller.yaml kibana-service.yaml efk-rbac.yaml 同样EFK服务也需要一个efk-rbac.yaml文件，配置serviceaccount为efk。 已经修改好的 yaml 文件见：../manifests/EFK 配置 es-controller.yaml $ diff es-controller.yaml.orig es-controller.yaml 24c24 - image: sz-pg-oam-docker-hub-001.tendcloud.com/library/elasticsearch:v2.4.1-2 配置 es-service.yaml 无需配置； 配置 fluentd-es-ds.yaml $ diff fluentd-es-ds.yaml.orig fluentd-es-ds.yaml 26c26 image: sz-pg-oam-docker-hub-001.tendcloud.com/library/fluentd-elasticsearch:1.22 配置 kibana-controller.yaml $ diff kibana-controller.yaml.orig kibana-controller.yaml 22c22 image: sz-pg-oam-docker-hub-001.tendcloud.com/library/kibana:v4.6.1-1 给 Node 设置标签 定义 DaemonSet fluentd-es-v1.22 时设置了 nodeSelector beta.kubernetes.io/fluentd-ds-ready=true ，所以需要在期望运行 fluentd 的 Node 上设置该标签； $ kubectl get nodes NAME STATUS AGE VERSION 172.20.0.113 Ready 1d v1.6.0 $ kubectl label nodes 172.20.0.113 beta.kubernetes.io/fluentd-ds-ready=true node \"172.20.0.113\" labeled 给其他两台node打上同样的标签。 执行定义文件 $ kubectl create -f . serviceaccount \"efk\" created clusterrolebinding \"efk\" created replicationcontroller \"elasticsearch-logging-v1\" created service \"elasticsearch-logging\" created daemonset \"fluentd-es-v1.22\" created deployment \"kibana-logging\" created service \"kibana-logging\" created 检查执行结果 $ kubectl get deployment -n kube-system|grep kibana kibana-logging 1 1 1 1 2m $ kubectl get pods -n kube-system|grep -E 'elasticsearch|fluentd|kibana' elasticsearch-logging-v1-mlstp 1/1 Running 0 1m elasticsearch-logging-v1-nfbbf 1/1 Running 0 1m fluentd-es-v1.22-31sm0 1/1 Running 0 1m fluentd-es-v1.22-bpgqs 1/1 Running 0 1m fluentd-es-v1.22-qmn7h 1/1 Running 0 1m kibana-logging-1432287342-0gdng 1/1 Running 0 1m $ kubectl get service -n kube-system|grep -E 'elasticsearch|kibana' elasticsearch-logging 10.254.77.62 9200/TCP 2m kibana-logging 10.254.8.113 5601/TCP 2m kibana Pod 第一次启动时会用较长时间(10-20分钟)来优化和 Cache 状态页面，可以 tailf 该 Pod 的日志观察进度： $ kubectl logs kibana-logging-1432287342-0gdng -n kube-system -f ELASTICSEARCH_URL=http://elasticsearch-logging:9200 server.basePath: /api/v1/proxy/namespaces/kube-system/services/kibana-logging {\"type\":\"log\",\"@timestamp\":\"2017-04-12T13:08:06Z\",\"tags\":[\"info\",\"optimize\"],\"pid\":7,\"message\":\"Optimizing and caching bundles for kibana and statusPage. This may take a few minutes\"} {\"type\":\"log\",\"@timestamp\":\"2017-04-12T13:18:17Z\",\"tags\":[\"info\",\"optimize\"],\"pid\":7,\"message\":\"Optimization of bundles for kibana and statusPage complete in 610.40 seconds\"} {\"type\":\"log\",\"@timestamp\":\"2017-04-12T13:18:17Z\",\"tags\":[\"status\",\"plugin:kibana@1.0.0\",\"info\"],\"pid\":7,\"state\":\"green\",\"message\":\"Status changed from uninitialized to green - Ready\",\"prevState\":\"uninitialized\",\"prevMsg\":\"uninitialized\"} {\"type\":\"log\",\"@timestamp\":\"2017-04-12T13:18:18Z\",\"tags\":[\"status\",\"plugin:elasticsearch@1.0.0\",\"info\"],\"pid\":7,\"state\":\"yellow\",\"message\":\"Status changed from uninitialized to yellow - Waiting for Elasticsearch\",\"prevState\":\"uninitialized\",\"prevMsg\":\"uninitialized\"} {\"type\":\"log\",\"@timestamp\":\"2017-04-12T13:18:19Z\",\"tags\":[\"status\",\"plugin:kbn_vislib_vis_types@1.0.0\",\"info\"],\"pid\":7,\"state\":\"green\",\"message\":\"Status changed from uninitialized to green - Ready\",\"prevState\":\"uninitialized\",\"prevMsg\":\"uninitialized\"} {\"type\":\"log\",\"@timestamp\":\"2017-04-12T13:18:19Z\",\"tags\":[\"status\",\"plugin:markdown_vis@1.0.0\",\"info\"],\"pid\":7,\"state\":\"green\",\"message\":\"Status changed from uninitialized to green - Ready\",\"prevState\":\"uninitialized\",\"prevMsg\":\"uninitialized\"} {\"type\":\"log\",\"@timestamp\":\"2017-04-12T13:18:19Z\",\"tags\":[\"status\",\"plugin:metric_vis@1.0.0\",\"info\"],\"pid\":7,\"state\":\"green\",\"message\":\"Status changed from uninitialized to green - Ready\",\"prevState\":\"uninitialized\",\"prevMsg\":\"uninitialized\"} {\"type\":\"log\",\"@timestamp\":\"2017-04-12T13:18:19Z\",\"tags\":[\"status\",\"plugin:spyModes@1.0.0\",\"info\"],\"pid\":7,\"state\":\"green\",\"message\":\"Status changed from uninitialized to green - Ready\",\"prevState\":\"uninitialized\",\"prevMsg\":\"uninitialized\"} {\"type\":\"log\",\"@timestamp\":\"2017-04-12T13:18:19Z\",\"tags\":[\"status\",\"plugin:statusPage@1.0.0\",\"info\"],\"pid\":7,\"state\":\"green\",\"message\":\"Status changed from uninitialized to green - Ready\",\"prevState\":\"uninitialized\",\"prevMsg\":\"uninitialized\"} {\"type\":\"log\",\"@timestamp\":\"2017-04-12T13:18:19Z\",\"tags\":[\"status\",\"plugin:table_vis@1.0.0\",\"info\"],\"pid\":7,\"state\":\"green\",\"message\":\"Status changed from uninitialized to green - Ready\",\"prevState\":\"uninitialized\",\"prevMsg\":\"uninitialized\"} {\"type\":\"log\",\"@timestamp\":\"2017-04-12T13:18:19Z\",\"tags\":[\"listening\",\"info\"],\"pid\":7,\"message\":\"Server running at http://0.0.0.0:5601\"} {\"type\":\"log\",\"@timestamp\":\"2017-04-12T13:18:24Z\",\"tags\":[\"status\",\"plugin:elasticsearch@1.0.0\",\"info\"],\"pid\":7,\"state\":\"yellow\",\"message\":\"Status changed from yellow to yellow - No existing Kibana index found\",\"prevState\":\"yellow\",\"prevMsg\":\"Waiting for Elasticsearch\"} {\"type\":\"log\",\"@timestamp\":\"2017-04-12T13:18:29Z\",\"tags\":[\"status\",\"plugin:elasticsearch@1.0.0\",\"info\"],\"pid\":7,\"state\":\"green\",\"message\":\"Status changed from yellow to green - Kibana index ready\",\"prevState\":\"yellow\",\"prevMsg\":\"No existing Kibana index found\"} 访问 kibana 通过 kube-apiserver 访问： 获取 monitoring-grafana 服务 URL $ kubectl cluster-info Kubernetes master is running at https://172.20.0.113:6443 Elasticsearch is running at https://172.20.0.113:6443/api/v1/proxy/namespaces/kube-system/services/elasticsearch-logging Heapster is running at https://172.20.0.113:6443/api/v1/proxy/namespaces/kube-system/services/heapster Kibana is running at https://172.20.0.113:6443/api/v1/proxy/namespaces/kube-system/services/kibana-logging KubeDNS is running at https://172.20.0.113:6443/api/v1/proxy/namespaces/kube-system/services/kube-dns kubernetes-dashboard is running at https://172.20.0.113:6443/api/v1/proxy/namespaces/kube-system/services/kubernetes-dashboard monitoring-grafana is running at https://172.20.0.113:6443/api/v1/proxy/namespaces/kube-system/services/monitoring-grafana monitoring-influxdb is running at https://172.20.0.113:6443/api/v1/proxy/namespaces/kube-system/services/monitoring-influxdb 浏览器访问 URL： https://172.20.0.113:6443/api/v1/proxy/namespaces/kube-system/services/kibana-logging/app/kibana 通过 kubectl proxy 访问： 创建代理 $ kubectl proxy --address='172.20.0.113' --port=8086 --accept-hosts='^*$' Starting to serve on 172.20.0.113:8086 浏览器访问 URL：http://172.20.0.113:8086/api/v1/proxy/namespaces/kube-system/services/kibana-logging 在 Settings -> Indices 页面创建一个 index（相当于 mysql 中的一个 database），选中 Index contains time-based events，使用默认的 logstash-* pattern，点击 Create ; 可能遇到的问题 如果你在这里发现Create按钮是灰色的无法点击，且Time-filed name中没有选项，fluentd要读取/var/log/containers/目录下的log日志，这些日志是从/var/lib/docker/containers/${CONTAINER_ID}/${CONTAINER_ID}-json.log链接过来的，查看你的docker配置，—log-dirver需要设置为json-file格式，默认的可能是journald，参考docker logging。 图片 - es-setting 创建Index后，可以在 Discover 下看到 ElasticSearch logging 中汇聚的日志； Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-07 10:18:25 "},"practice/service-discovery-and-loadbalancing.html":{"url":"practice/service-discovery-and-loadbalancing.html","title":"服务发现与负载均衡","keywords":"","body":"服务发现与负载均衡 Kubernetes在设计之初就充分考虑了针对容器的服务发现与负载均衡机制，提供了Service资源，并通过kube-proxy配合cloud provider来适应不同的应用场景。随着kubernetes用户的激增，用户场景的不断丰富，又产生了一些新的负载均衡机制。目前，kubernetes中的负载均衡大致可以分为以下几种机制，每种机制都有其特定的应用场景： Service：直接用Service提供cluster内部的负载均衡，并借助cloud provider提供的LB提供外部访问 Ingress Controller：还是用Service提供cluster内部的负载均衡，但是通过自定义LB提供外部访问 Service Load Balancer：把load balancer直接跑在容器中，实现Bare Metal的Service Load Balancer Custom Load Balancer：自定义负载均衡，并替代kube-proxy，一般在物理部署Kubernetes时使用，方便接入公司已有的外部服务 Service Service是对一组提供相同功能的Pods的抽象，并为它们提供一个统一的入口。借助Service，应用可以方便的实现服务发现与负载均衡，并实现应用的零宕机升级。Service通过标签来选取服务后端，一般配合Replication Controller或者Deployment来保证后端容器的正常运行。 Service有三种类型： ClusterIP：默认类型，自动分配一个仅cluster内部可以访问的虚拟IP NodePort：在ClusterIP基础上为Service在每台机器上绑定一个端口，这样就可以通过:NodePort来访问该服务 LoadBalancer：在NodePort的基础上，借助cloud provider创建一个外部的负载均衡器，并将请求转发到:NodePort 另外，也可以将已有的服务以Service的形式加入到Kubernetes集群中来，只需要在创建Service的时候不指定Label selector，而是在Service创建好后手动为其添加endpoint。 Ingress Controller Service虽然解决了服务发现和负载均衡的问题，但它在使用上还是有一些限制，比如 对外访问的时候，NodePort类型需要在外部搭建额外的负载均衡，而LoadBalancer要求kubernetes必须跑在支持的cloud provider上面 Ingress就是为了解决这些限制而引入的新资源，主要用来将服务暴露到cluster外面，并且可以自定义服务的访问策略。比如想要通过负载均衡器实现不同子域名到不同服务的访问： foo.bar.com --| |-> foo.bar.com s1:80 | 178.91.123.132 | bar.foo.com --| |-> bar.foo.com s2:80 可以这样来定义Ingress： apiVersion: extensions/v1beta1 kind: Ingress metadata: name: test spec: rules: - host: foo.bar.com http: paths: - backend: serviceName: s1 servicePort: 80 - host: bar.foo.com http: paths: - backend: serviceName: s2 servicePort: 80 注意： Ingress本身并不会自动创建负载均衡器，cluster中需要运行一个ingress controller来根据Ingress的定义来管理负载均衡器。目前社区提供了nginx和gce的参考实现。 Traefik提供了易用的Ingress Controller，使用方法见https://docs.traefik.io/user-guide/kubernetes/。 Service Load Balancer 在Ingress出现以前，Service Load Balancer是推荐的解决Service局限性的方式。Service Load Balancer将haproxy跑在容器中，并监控service和endpoint的变化，通过容器IP对外提供4层和7层负载均衡服务。 社区提供的Service Load Balancer支持四种负载均衡协议：TCP、HTTP、HTTPS和SSL TERMINATION，并支持ACL访问控制。 Custom Load Balancer 虽然Kubernetes提供了丰富的负载均衡机制，但在实际使用的时候，还是会碰到一些复杂的场景是它不能支持的，比如： 接入已有的负载均衡设备 多租户网络情况下，容器网络和主机网络是隔离的，这样kube-proxy就不能正常工作 这个时候就可以自定义组件，并代替kube-proxy来做负载均衡。基本的思路是监控kubernetes中service和endpoints的变化，并根据这些变化来配置负载均衡器。比如weave flux、nginx plus、kube2haproxy等。 Endpoints 有几种情况下需要用到没有selector的service。 使用kubernetes集群外部的数据库时 service中用到了其他namespace或kubernetes集群中的service 在kubernetes的工作负载与集群外的后端之间互相迁移 可以这样定义一个没有selector的service。 kind: Service apiVersion: v1 metadata: name: my-service spec: ports: - protocol: TCP port: 80 targetPort: 9376 定义一个Endpoints来对应该service。 kind: Endpoints apiVersion: v1 metadata: name: my-service subsets: - addresses: - ip: 1.2.3.4 ports: - port: 9376 访问没有selector的service跟访问有selector的service时没有任何区别。 使用kubernetes时有一个很常见的需求，就是当数据库部署在kubernetes集群之外的时候，集群内的service如何访问数据库呢？当然你可以直接使用数据库的IP地址和端口号来直接访问，有没有什么优雅的方式呢？你需要用到ExternalName Service。 kind: Service apiVersion: v1 metadata: name: my-service namespace: prod spec: type: ExternalName externalName: my.database.example.com ports: - port: 12345 这个例子中，在kubernetes集群内访问my-service实际上会重定向到my.database.example.com:12345这个地址。 参考资料 https://kubernetes.io/docs/concepts/services-networking/service/ http://kubernetes.io/docs/user-guide/ingress/ https://github.com/kubernetes/contrib/tree/master/service-loadbalancer https://www.nginx.com/blog/load-balancing-kubernetes-services-nginx-plus/ https://github.com/weaveworks/flux https://github.com/AdoHe/kube2haproxy Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-17 15:54:27 "},"practice/traefik-ingress-installation.html":{"url":"practice/traefik-ingress-installation.html","title":"安装Traefik ingress","keywords":"","body":"安装traefik ingress Ingress简介 如果你还不了解，ingress是什么，可以先看下我翻译的Kubernetes官网上ingress的介绍Kubernetes Ingress解析。 理解Ingress 简单的说，ingress就是从kubernetes集群外访问集群的入口，将用户的URL请求转发到不同的service上。Ingress相当于nginx、apache等负载均衡方向代理服务器，其中还包括规则定义，即URL的路由信息，路由信息得的刷新由Ingress controller来提供。 理解Ingress Controller Ingress Controller 实质上可以理解为是个监视器，Ingress Controller 通过不断地跟 kubernetes API 打交道，实时的感知后端 service、pod 等变化，比如新增和减少 pod，service 增加与减少等；当得到这些变化信息后，Ingress Controller 再结合下文的 Ingress 生成配置，然后更新反向代理负载均衡器，并刷新其配置，达到服务发现的作用。 部署Traefik 介绍traefik Traefik是一款开源的反向代理与负载均衡工具。它最大的优点是能够与常见的微服务系统直接整合，可以实现自动化动态配置。目前支持Docker, Swarm, Mesos/Marathon, Mesos, Kubernetes, Consul, Etcd, Zookeeper, BoltDB, Rest API等等后端模型。 以下配置文件可以在kubernetes-handbookGitHub仓库中的../manifests/traefik-ingress/目录下找到。 创建ingress-rbac.yaml 将用于service account验证。 apiVersion: v1 kind: ServiceAccount metadata: name: ingress namespace: kube-system --- kind: ClusterRoleBinding apiVersion: rbac.authorization.k8s.io/v1beta1 metadata: name: ingress subjects: - kind: ServiceAccount name: ingress namespace: kube-system roleRef: kind: ClusterRole name: cluster-admin apiGroup: rbac.authorization.k8s.io 创建名为traefik-ingress的ingress，文件名traefik.yaml apiVersion: extensions/v1beta1 kind: Ingress metadata: name: traefik-ingress namespace: default spec: rules: - host: traefik.nginx.io http: paths: - path: / backend: serviceName: my-nginx servicePort: 80 - host: traefik.frontend.io http: paths: - path: / backend: serviceName: frontend servicePort: 80 这其中的backend中要配置default namespace中启动的service名字，如果你没有配置namespace名字，默认使用default namespace，如果你在其他namespace中创建服务想要暴露到kubernetes集群外部，可以创建新的ingress.yaml文件，同时在文件中指定该namespace，其他配置与上面的文件格式相同。。path就是URL地址后的路径，如traefik.frontend.io/path，service将会接受path这个路径，host最好使用service-name.filed1.filed2.domain-name这种类似主机名称的命名方式，方便区分服务。 根据你自己环境中部署的service的名字和端口自行修改，有新service增加时，修改该文件后可以使用kubectl replace -f traefik.yaml来更新。 我们现在集群中已经有两个service了，一个是nginx，另一个是官方的guestbook例子。 创建Depeloyment apiVersion: extensions/v1beta1 kind: Deployment metadata: name: traefik-ingress-lb namespace: kube-system labels: k8s-app: traefik-ingress-lb spec: template: metadata: labels: k8s-app: traefik-ingress-lb name: traefik-ingress-lb spec: terminationGracePeriodSeconds: 60 hostNetwork: true restartPolicy: Always serviceAccountName: ingress containers: - image: traefik name: traefik-ingress-lb resources: limits: cpu: 200m memory: 30Mi requests: cpu: 100m memory: 20Mi ports: - name: http containerPort: 80 hostPort: 80 - name: admin containerPort: 8580 hostPort: 8580 args: - --web - --web.address=:8580 - --kubernetes 注意我们这里用的是Deploy类型，没有限定该pod运行在哪个主机上。Traefik的端口是8580。 Traefik UI apiVersion: v1 kind: Service metadata: name: traefik-web-ui namespace: kube-system spec: selector: k8s-app: traefik-ingress-lb ports: - name: web port: 80 targetPort: 8580 --- apiVersion: extensions/v1beta1 kind: Ingress metadata: name: traefik-web-ui namespace: kube-system spec: rules: - host: traefik-ui.local http: paths: - path: / backend: serviceName: traefik-web-ui servicePort: web 配置完成后就可以启动treafik ingress了。 kubectl create -f . 我查看到traefik的pod在172.20.0.115这台节点上启动了。 访问该地址http://172.20.0.115:8580/将可以看到dashboard。 图片 - kubernetes-dashboard 左侧黄色部分部分列出的是所有的rule，右侧绿色部分是所有的backend。 测试 在集群的任意一个节点上执行。假如现在我要访问nginx的\"/\"路径。 $ curl -H Host:traefik.nginx.io http://172.20.0.115/ Welcome to nginx! body { width: 35em; margin: 0 auto; font-family: Tahoma, Verdana, Arial, sans-serif; } Welcome to nginx! If you see this page, the nginx web server is successfully installed and working. Further configuration is required. For online documentation and support please refer to nginx.org. Commercial support is available at nginx.com. Thank you for using nginx. 如果你需要在kubernetes集群以外访问就需要设置DNS，或者修改本机的hosts文件。 在其中加入： 172.20.0.115 traefik.nginx.io 172.20.0.115 traefik.frontend.io 所有访问这些地址的流量都会发送给172.20.0.115这台主机，就是我们启动traefik的主机。 Traefik会解析http请求header里的Host参数将流量转发给Ingress配置里的相应service。 修改hosts后就就可以在kubernetes集群外访问以上两个service，如下图： 图片 - traefik-nginx 图片 - traefik-guestbook 参考 Traefik-kubernetes 初试 Traefik简介 Guestbook example Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-10-11 23:39:48 "},"practice/distributed-load-test.html":{"url":"practice/distributed-load-test.html","title":"分布式负载测试","keywords":"","body":"分布式负载测试 该教程描述如何在Kubernetes中进行分布式负载均衡测试，包括一个web应用、docker镜像和Kubernetes controllers/services。关于分布式负载测试的更多资料请查看Distributed Load Testing Using Kubernetes 。 准备 不需要GCE及其他组件，你只需要有一个kubernetes集群即可。 如果你还没有kubernetes集群，可以参考kubernetes-handbook部署一个。 部署Web应用 本文中使用的镜像、kubernetes应用的yaml配置来自我的另一个项目，请参考：https://github.com/rootsongjc/distributed-load-testing-using-kubernetes sample-webapp 目录下包含一个简单的web测试应用。我们将其构建为docker镜像，在kubernetes中运行。你可以自己构建，也可以直接用这个我构建好的镜像index.tenxcloud.com/jimmy/k8s-sample-webapp:latest。 在kubernetes上部署sample-webapp。 $ git clone https://github.com/rootsongjc/distributed-load-testing-using-kubernetes.git $ cd kubernetes-config $ kubectl create -f sample-webapp-controller.yaml $ kubectl create -f sample-webapp-service.yaml 部署Locust的Controller和Service locust-master和locust-work使用同样的docker镜像，修改cotnroller中spec.template.spec.containers.env字段中的value为你sample-webapp service的名字。 - name: TARGET_HOST value: http://sample-webapp:8000 创建Controller Docker镜像（可选） locust-master和locust-work controller使用的都是locust-tasks docker镜像。你可以直接下载gcr.io/cloud-solutions-images/locust-tasks，也可以自己编译。自己编译大概要花几分钟时间，镜像大小为820M。 $ docker build -t index.tenxcloud.com/jimmy/locust-tasks:latest . $ docker push index.tenxcloud.com/jimmy/locust-tasks:latest 注意：我使用的是时速云的镜像仓库。 每个controller的yaml的spec.template.spec.containers.image 字段指定的是我的镜像： image: index.tenxcloud.com/jimmy/locust-tasks:latest 部署locust-master $ kubectl create -f locust-master-controller.yaml $ kubectl create -f locust-master-service.yaml 部署locust-worker Now deploy locust-worker-controller: $ kubectl create -f locust-worker-controller.yaml 你可以很轻易的给work扩容，通过命令行方式： $ kubectl scale --replicas=20 replicationcontrollers locust-worker 当然你也可以通过WebUI：Dashboard - Workloads - Replication Controllers - ServiceName - Scale来扩容。 图片 - 使用dashboard来扩容 配置Traefik 参考kubernetes的traefik ingress安装，在ingress.yaml中加入如下配置： - host: traefik.locust.io http: paths: - path: / backend: serviceName: locust-master servicePort: 8089 然后执行kubectl replace -f ingress.yaml即可更新traefik。 通过Traefik的dashboard就可以看到刚增加的traefik.locust.io节点。 图片 - Traefik的UI 执行测试 打开http://traefik.locust.io页面，点击Edit输入伪造的用户数和用户每秒发送的请求个数，点击Start Swarming就可以开始测试了。 图片 - Locust启动界面 在测试过程中调整sample-webapp的pod个数（默认设置了1个pod），观察pod的负载变化情况。 图片 - Dashboard查看页面 从一段时间的观察中可以看到负载被平均分配给了3个pod。 在locust的页面中可以实时观察也可以下载测试结果。 图片 - Locust测试结果页面 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-10-14 01:23:23 "},"practice/network-and-cluster-perfermance-test.html":{"url":"practice/network-and-cluster-perfermance-test.html","title":"网络和集群性能测试","keywords":"","body":"Kubernetes网络和集群性能测试 准备 测试环境 在以下几种环境下进行测试： Kubernetes集群node节点上通过Cluster IP方式访问 Kubernetes集群内部通过service访问 Kubernetes集群外部通过traefik ingress暴露的地址访问 测试地址 Cluster IP: 10.254.149.31 Service Port：8000 Ingress Host：traefik.sample-webapp.io 测试工具 Locust：一个简单易用的用户负载测试工具，用来测试web或其他系统能够同时处理的并发用户数。 curl kubemark 测试程序：sample-webapp，源码见Github kubernetes的分布式负载测试 测试说明 通过向sample-webapp发送curl请求获取响应时间，直接curl后的结果为： $ curl \"http://10.254.149.31:8000/\" Welcome to the \"Distributed Load Testing Using Kubernetes\" sample web app 网络延迟测试 场景一、 Kubernetes集群node节点上通过Cluster IP访问 测试命令 curl -o /dev/null -s -w '%{time_connect} %{time_starttransfer} %{time_total}' \"http://10.254.149.31:8000/\" 10组测试结果 No time_connect time_starttransfer time_total 1 0.000 0.003 0.003 2 0.000 0.002 0.002 3 0.000 0.002 0.002 4 0.000 0.002 0.002 5 0.000 0.002 0.002 6 0.000 0.002 0.002 7 0.000 0.002 0.002 8 0.000 0.002 0.002 9 0.000 0.002 0.002 10 0.000 0.002 0.002 平均响应时间：2ms 时间指标说明 单位：秒 time_connect：建立到服务器的 TCP 连接所用的时间 time_starttransfer：在发出请求之后，Web 服务器返回数据的第一个字节所用的时间 time_total：完成请求所用的时间 场景二、Kubernetes集群内部通过service访问 测试命令 curl -o /dev/null -s -w '%{time_connect} %{time_starttransfer} %{time_total}' \"http://sample-webapp:8000/\" 10组测试结果 No time_connect time_starttransfer time_total 1 0.004 0.006 0.006 2 0.004 0.006 0.006 3 0.004 0.006 0.006 4 0.004 0.006 0.006 5 0.004 0.006 0.006 6 0.004 0.006 0.006 7 0.004 0.006 0.006 8 0.004 0.006 0.006 9 0.004 0.006 0.006 10 0.004 0.006 0.006 平均响应时间：6ms 场景三、在公网上通过traefik ingress访问 测试命令 curl -o /dev/null -s -w '%{time_connect} %{time_starttransfer} %{time_total}' \"http://traefik.sample-webapp.io\" >>result 10组测试结果 No time_connect time_starttransfer time_total 1 0.043 0.085 0.085 2 0.052 0.093 0.093 3 0.043 0.082 0.082 4 0.051 0.093 0.093 5 0.068 0.188 0.188 6 0.049 0.089 0.089 7 0.051 0.113 0.113 8 0.055 0.120 0.120 9 0.065 0.126 0.127 10 0.050 0.111 0.111 平均响应时间：110ms 测试结果 在这三种场景下的响应时间测试结果如下： Kubernetes集群node节点上通过Cluster IP方式访问：2ms Kubernetes集群内部通过service访问：6ms Kubernetes集群外部通过traefik ingress暴露的地址访问：110ms 注意：执行测试的node节点/Pod与serivce所在的pod的距离（是否在同一台主机上），对前两个场景可以能会有一定影响。 网络性能测试 网络使用flannel的vxlan模式。 使用iperf进行测试。 服务端命令： iperf -s -p 12345 -i 1 -M 客户端命令： iperf -c ${server-ip} -p 12345 -i 1 -t 10 -w 20K 场景一、主机之间 [ ID] Interval Transfer Bandwidth [ 3] 0.0- 1.0 sec 598 MBytes 5.02 Gbits/sec [ 3] 1.0- 2.0 sec 637 MBytes 5.35 Gbits/sec [ 3] 2.0- 3.0 sec 664 MBytes 5.57 Gbits/sec [ 3] 3.0- 4.0 sec 657 MBytes 5.51 Gbits/sec [ 3] 4.0- 5.0 sec 641 MBytes 5.38 Gbits/sec [ 3] 5.0- 6.0 sec 639 MBytes 5.36 Gbits/sec [ 3] 6.0- 7.0 sec 628 MBytes 5.26 Gbits/sec [ 3] 7.0- 8.0 sec 649 MBytes 5.44 Gbits/sec [ 3] 8.0- 9.0 sec 638 MBytes 5.35 Gbits/sec [ 3] 9.0-10.0 sec 652 MBytes 5.47 Gbits/sec [ 3] 0.0-10.0 sec 6.25 GBytes 5.37 Gbits/sec 场景二、不同主机的Pod之间(使用flannel的vxlan模式) [ ID] Interval Transfer Bandwidth [ 3] 0.0- 1.0 sec 372 MBytes 3.12 Gbits/sec [ 3] 1.0- 2.0 sec 345 MBytes 2.89 Gbits/sec [ 3] 2.0- 3.0 sec 361 MBytes 3.03 Gbits/sec [ 3] 3.0- 4.0 sec 397 MBytes 3.33 Gbits/sec [ 3] 4.0- 5.0 sec 405 MBytes 3.40 Gbits/sec [ 3] 5.0- 6.0 sec 410 MBytes 3.44 Gbits/sec [ 3] 6.0- 7.0 sec 404 MBytes 3.39 Gbits/sec [ 3] 7.0- 8.0 sec 408 MBytes 3.42 Gbits/sec [ 3] 8.0- 9.0 sec 451 MBytes 3.78 Gbits/sec [ 3] 9.0-10.0 sec 387 MBytes 3.25 Gbits/sec [ 3] 0.0-10.0 sec 3.85 GBytes 3.30 Gbits/sec 场景三、Node与非同主机的Pod之间（使用flannel的vxlan模式） [ ID] Interval Transfer Bandwidth [ 3] 0.0- 1.0 sec 372 MBytes 3.12 Gbits/sec [ 3] 1.0- 2.0 sec 420 MBytes 3.53 Gbits/sec [ 3] 2.0- 3.0 sec 434 MBytes 3.64 Gbits/sec [ 3] 3.0- 4.0 sec 409 MBytes 3.43 Gbits/sec [ 3] 4.0- 5.0 sec 382 MBytes 3.21 Gbits/sec [ 3] 5.0- 6.0 sec 408 MBytes 3.42 Gbits/sec [ 3] 6.0- 7.0 sec 403 MBytes 3.38 Gbits/sec [ 3] 7.0- 8.0 sec 423 MBytes 3.55 Gbits/sec [ 3] 8.0- 9.0 sec 376 MBytes 3.15 Gbits/sec [ 3] 9.0-10.0 sec 451 MBytes 3.78 Gbits/sec [ 3] 0.0-10.0 sec 3.98 GBytes 3.42 Gbits/sec 场景四、不同主机的Pod之间（使用flannel的host-gw模式） [ ID] Interval Transfer Bandwidth [ 5] 0.0- 1.0 sec 530 MBytes 4.45 Gbits/sec [ 5] 1.0- 2.0 sec 576 MBytes 4.84 Gbits/sec [ 5] 2.0- 3.0 sec 631 MBytes 5.29 Gbits/sec [ 5] 3.0- 4.0 sec 580 MBytes 4.87 Gbits/sec [ 5] 4.0- 5.0 sec 627 MBytes 5.26 Gbits/sec [ 5] 5.0- 6.0 sec 578 MBytes 4.85 Gbits/sec [ 5] 6.0- 7.0 sec 584 MBytes 4.90 Gbits/sec [ 5] 7.0- 8.0 sec 571 MBytes 4.79 Gbits/sec [ 5] 8.0- 9.0 sec 564 MBytes 4.73 Gbits/sec [ 5] 9.0-10.0 sec 572 MBytes 4.80 Gbits/sec [ 5] 0.0-10.0 sec 5.68 GBytes 4.88 Gbits/sec 场景五、Node与非同主机的Pod之间（使用flannel的host-gw模式） [ ID] Interval Transfer Bandwidth [ 3] 0.0- 1.0 sec 570 MBytes 4.78 Gbits/sec [ 3] 1.0- 2.0 sec 552 MBytes 4.63 Gbits/sec [ 3] 2.0- 3.0 sec 598 MBytes 5.02 Gbits/sec [ 3] 3.0- 4.0 sec 580 MBytes 4.87 Gbits/sec [ 3] 4.0- 5.0 sec 590 MBytes 4.95 Gbits/sec [ 3] 5.0- 6.0 sec 594 MBytes 4.98 Gbits/sec [ 3] 6.0- 7.0 sec 598 MBytes 5.02 Gbits/sec [ 3] 7.0- 8.0 sec 606 MBytes 5.08 Gbits/sec [ 3] 8.0- 9.0 sec 596 MBytes 5.00 Gbits/sec [ 3] 9.0-10.0 sec 604 MBytes 5.07 Gbits/sec [ 3] 0.0-10.0 sec 5.75 GBytes 4.94 Gbits/sec 网络性能对比综述 使用Flannel的vxlan模式实现每个pod一个IP的方式，会比宿主机直接互联的网络性能损耗30%～40%，符合网上流传的测试结论。而flannel的host-gw模式比起宿主机互连的网络性能损耗大约是10%。 Vxlan会有一个封包解包的过程，所以会对网络性能造成较大的损耗，而host-gw模式是直接使用路由信息，网络损耗小，关于host-gw的架构请访问Flannel host-gw architecture。 Kubernete的性能测试 参考Kubernetes集群性能测试中的步骤，对kubernetes的性能进行测试。 我的集群版本是Kubernetes1.6.0，首先克隆代码，将kubernetes目录复制到$GOPATH/src/k8s.io/下然后执行： $ ./hack/generate-bindata.sh /usr/local/src/k8s.io/kubernetes /usr/local/src/k8s.io/kubernetes Generated bindata file : test/e2e/generated/bindata.go has 13498 test/e2e/generated/bindata.go lines of lovely automated artifacts No changes in generated bindata file: pkg/generated/bindata.go /usr/local/src/k8s.io/kubernetes $ make WHAT=\"test/e2e/e2e.test\" ... +++ [0425 17:01:34] Generating bindata: test/e2e/generated/gobindata_util.go /usr/local/src/k8s.io/kubernetes /usr/local/src/k8s.io/kubernetes/test/e2e/generated /usr/local/src/k8s.io/kubernetes/test/e2e/generated +++ [0425 17:01:34] Building go targets for linux/amd64: test/e2e/e2e.test $ make ginkgo +++ [0425 17:05:57] Building the toolchain targets: k8s.io/kubernetes/hack/cmd/teststale k8s.io/kubernetes/vendor/github.com/jteeuwen/go-bindata/go-bindata +++ [0425 17:05:57] Generating bindata: test/e2e/generated/gobindata_util.go /usr/local/src/k8s.io/kubernetes /usr/local/src/k8s.io/kubernetes/test/e2e/generated /usr/local/src/k8s.io/kubernetes/test/e2e/generated +++ [0425 17:05:58] Building go targets for linux/amd64: vendor/github.com/onsi/ginkgo/ginkgo $ export KUBERNETES_PROVIDER=local $ export KUBECTL_PATH=/usr/bin/kubectl $ go run hack/e2e.go -v -test --test_args=\"--host=http://172.20.0.113:8080 --ginkgo.focus=\\[Feature:Performance\\]\" >>log.txt 测试结果 Apr 25 18:27:31.461: INFO: API calls latencies: { \"apicalls\": [ { \"resource\": \"pods\", \"verb\": \"POST\", \"latency\": { \"Perc50\": 2148000, \"Perc90\": 13772000, \"Perc99\": 14436000, \"Perc100\": 0 } }, { \"resource\": \"services\", \"verb\": \"DELETE\", \"latency\": { \"Perc50\": 9843000, \"Perc90\": 11226000, \"Perc99\": 12391000, \"Perc100\": 0 } }, ... Apr 25 18:27:31.461: INFO: [Result:Performance] { \"version\": \"v1\", \"dataItems\": [ { \"data\": { \"Perc50\": 2.148, \"Perc90\": 13.772, \"Perc99\": 14.436 }, \"unit\": \"ms\", \"labels\": { \"Resource\": \"pods\", \"Verb\": \"POST\" } }, ... 2.857: INFO: Running AfterSuite actions on all node Apr 26 10:35:32.857: INFO: Running AfterSuite actions on node 1 Ran 2 of 606 Specs in 268.371 seconds SUCCESS! -- 2 Passed | 0 Failed | 0 Pending | 604 Skipped PASS Ginkgo ran 1 suite in 4m28.667870101s Test Suite Passed 从kubemark输出的日志中可以看到API calls latencies和Performance。 日志里显示，创建90个pod用时40秒以内，平均创建每个pod耗时0.44秒。 不同type的资源类型API请求耗时分布 Resource Verb 50% 90% 99% services DELETE 8.472ms 9.841ms 38.226ms endpoints PUT 1.641ms 3.161ms 30.715ms endpoints GET 931µs 10.412ms 27.97ms nodes PATCH 4.245ms 11.117ms 18.63ms pods PUT 2.193ms 2.619ms 17.285ms 从log.txt日志中还可以看到更多详细请求的测试指标。 图片 - kubernetes-dashboard 注意事项 测试过程中需要用到docker镜像存储在GCE中，需要翻墙下载，我没看到哪里配置这个镜像的地址。该镜像副本已上传时速云： 用到的镜像有如下两个： gcr.io/google_containers/pause-amd64:3.0 gcr.io/google_containers/serve_hostname:v1.4 时速云镜像地址： index.tenxcloud.com/jimmy/pause-amd64:3.0 index.tenxcloud.com/jimmy/serve_hostname:v1.4 将镜像pull到本地后重新打tag。 Locust测试 请求统计 Method Name # requests # failures Median response time Average response time Min response time Max response time Average Content Size Requests/s POST /login 5070 78 59000 80551 11218 202140 54 1.17 POST /metrics 5114232 85879 63000 82280 29518 331330 94 1178.77 None Total 5119302 85957 63000 82279 11218 331330 94 1179.94 响应时间分布 Name # requests 50% 66% 75% 80% 90% 95% 98% 99% 100% POST /login 5070 59000 125000 140000 148000 160000 166000 174000 176000 202140 POST /metrics 5114993 63000 127000 142000 149000 160000 166000 172000 176000 331330 None Total 5120063 63000 127000 142000 149000 160000 166000 172000 176000 331330 以上两个表格都是瞬时值。请求失败率在2%左右。 Sample-webapp起了48个pod。 Locust模拟10万用户，每秒增长100个。 图片 - locust测试页面 关于Locust的使用请参考Github：https://github.com/rootsongjc/distributed-load-testing-using-kubernetes 参考 基于 Python 的性能测试工具 locust (与 LR 的简单对比) Locust docs python用户负载测试工具：locust Kubernetes集群性能测试 CoreOS是如何将Kubernetes的性能提高10倍的 Kubernetes 1.3 的性能和弹性 —— 2000 节点，60,0000 Pod 的集群 运用Kubernetes进行分布式负载测试 Kubemark User Guide Flannel host-gw architecture Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-15 21:34:41 "},"practice/edge-node-configuration.html":{"url":"practice/edge-node-configuration.html","title":"边缘节点配置","keywords":"","body":"边缘节点配置 前言 为了配置kubernetes中的traefik ingress的高可用，对于kubernetes集群以外只暴露一个访问入口，需要使用keepalived排除单点问题。本文参考了kube-keepalived-vip，但并没有使用容器方式安装，而是直接在node节点上安装。 定义 首先解释下什么叫边缘节点（Edge Node），所谓的边缘节点即集群内部用来向集群外暴露服务能力的节点，集群外部的服务通过该节点来调用集群内部的服务，边缘节点是集群内外交流的一个Endpoint。 边缘节点要考虑两个问题 边缘节点的高可用，不能有单点故障，否则整个kubernetes集群将不可用 对外的一致暴露端口，即只能有一个外网访问IP和端口 架构 为了满足边缘节点的以上需求，我们使用keepalived来实现。 在Kubernetes中添加了service的同时，在DNS中增加一个记录，这条记录需要跟ingress中的host字段相同，IP地址即VIP的地址，本示例中是172.20.0.119，这样集群外部就可以通过service的DNS名称来访问服务了。 选择Kubernetes的三个node作为边缘节点，并安装keepalived，下图展示了边缘节点的配置，同时展示了向Kubernetes中添加服务的过程。 图片 - 边缘节点架构 准备 复用kubernetes测试集群的三台主机。 172.20.0.113 172.20.0.114 172.20.0.115 安装 使用keepalived管理VIP，VIP是使用IPVS创建的，IPVS已经成为linux内核的模块，不需要安装 LVS的工作原理请参考：http://www.cnblogs.com/codebean/archive/2011/07/25/2116043.html 不使用镜像方式安装了，直接手动安装，指定三个节点为边缘节点（Edge node）。 因为我们的测试集群一共只有三个node，所有在在三个node上都要安装keepalived和ipvsadmin。 yum install keepalived ipvsadm 配置说明 需要对原先的traefik ingress进行改造，从以Deployment方式启动改成DeamonSet。还需要指定一个与node在同一网段的IP地址作为VIP，我们指定成172.20.0.119，配置keepalived前需要先保证这个IP没有被分配。。 Traefik以DaemonSet的方式启动 通过nodeSelector选择边缘节点 通过hostPort暴露端口 当前VIP漂移到了172.20.0.115上 Traefik根据访问的host和path配置，将流量转发到相应的service上 配置keepalived 参考基于keepalived 实现VIP转移，lvs，nginx的高可用，配置keepalived。 keepalived的官方配置文档见：http://keepalived.org/pdf/UserGuide.pdf 配置文件/etc/keepalived/keepalived.conf文件内容如下： ! Configuration File for keepalived global_defs { notification_email { root@localhost } notification_email_from kaadmin@localhost smtp_server 127.0.0.1 smtp_connect_timeout 30 router_id LVS_DEVEL } vrrp_instance VI_1 { state MASTER interface eth0 virtual_router_id 51 priority 100 advert_int 1 authentication { auth_type PASS auth_pass 1111 } virtual_ipaddress { 172.20.0.119 } } virtual_server 172.20.0.119 80{ delay_loop 6 lb_algo loadbalance lb_kind DR nat_mask 255.255.255.0 persistence_timeout 0 protocol TCP real_server 172.20.0.113 80{ weight 1 TCP_CHECK { connect_timeout 3 } } real_server 172.20.0.114 80{ weight 1 TCP_CHECK { connect_timeout 3 } } real_server 172.20.0.115 80{ weight 1 TCP_CHECK { connect_timeout 3 } } } Realserver的IP和端口即traefik供外网访问的IP和端口。 将以上配置分别拷贝到另外两台node的/etc/keepalived目录下。 我们使用转发效率最高的lb_kind DR直接路由方式转发，使用TCP_CHECK来检测real_server的health。 设置keepalived为开机自启动： chkconfig keepalived on 启动keepalived systemctl start keepalived 三台node都启动了keepalived后，观察eth0的IP，会在三台node的某一台上发现一个VIP是172.20.0.119。 $ ip addr show eth0 2: eth0: mtu 1500 qdisc mq state UP qlen 1000 link/ether f4:e9:d4:9f:6b:a0 brd ff:ff:ff:ff:ff:ff inet 172.20.0.115/17 brd 172.20.127.255 scope global eth0 valid_lft forever preferred_lft forever inet 172.20.0.119/32 scope global eth0 valid_lft forever preferred_lft forever 关掉拥有这个VIP主机上的keepalived，观察VIP是否漂移到了另外两台主机的其中之一上。 改造Traefik 在这之前我们启动的traefik使用的是deployment，只启动了一个pod，无法保证高可用（即需要将pod固定在某一台主机上，这样才能对外提供一个唯一的访问地址），现在使用了keepalived就可以通过VIP来访问traefik，同时启动多个traefik的pod保证高可用。 配置文件traefik.yaml内容如下： apiVersion: extensions/v1beta1 kind: DaemonSet metadata: name: traefik-ingress-lb namespace: kube-system labels: k8s-app: traefik-ingress-lb spec: template: metadata: labels: k8s-app: traefik-ingress-lb name: traefik-ingress-lb spec: terminationGracePeriodSeconds: 60 hostNetwork: true restartPolicy: Always serviceAccountName: ingress containers: - image: traefik name: traefik-ingress-lb resources: limits: cpu: 200m memory: 30Mi requests: cpu: 100m memory: 20Mi ports: - name: http containerPort: 80 hostPort: 80 - name: admin containerPort: 8580 hostPort: 8580 args: - --web - --web.address=:8580 - --kubernetes nodeSelector: edgenode: \"true\" 注意，我们使用了nodeSelector选择边缘节点来调度traefik-ingress-lb运行在它上面，所有你需要使用： kubectl label nodes 172.20.0.113 edgenode=true kubectl label nodes 172.20.0.114 edgenode=true kubectl label nodes 172.20.0.115 edgenode=true 给三个node打标签。 查看DaemonSet的启动情况： $ kubectl -n kube-system get ds NAME DESIRED CURRENT READY UP-TO-DATE AVAILABLE NODE-SELECTOR AGE traefik-ingress-lb 3 3 3 3 3 edgenode=true 2h 现在就可以在外网通过172.20.0.119:80来访问到traefik ingress了。 参考 kube-keepalived-vip http://www.keepalived.org/ keepalived工作原理与配置说明 LVS简介及使用 基于keepalived 实现VIP转移，lvs，nginx的高可用 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-05 19:33:27 "},"practice/nginx-ingress-installation.html":{"url":"practice/nginx-ingress-installation.html","title":"安装Nginx ingress","keywords":"","body":"安装Nginx ingress Nginx ingress 使用ConfigMap来管理Nginx配置，nginx是大家熟知的代理和负载均衡软件，比起Traefik来说功能更加强大. 我们使用helm来部署，chart保存在私有的仓库中，请确保您已经安装和配置好helm，helm安装使用见使用Helm管理kubernetes应用。 镜像准备 安装时需要用到的镜像有： sophos/nginx-vts-exporter:v0.6 gcr.io/google_containers/nginx-ingress-controller:0.9.0-beta.15 gcr.io/google_containers/defaultbackend:1.3 gcr.io中的那个两个镜像我复制了一份到时速云，可供大家下载： index.tenxcloud.com/jimmy/defaultbackend:1.3 index.tenxcloud.com/jimmy/nginx-ingress-controller:0.9.0-beta.15 Docker hub上的那个镜像可以直接下载，所有的安装时需要的配置保存在../manifests/nginx-ingress目录下。 步骤详解 安装nginx-ingress chart到本地repo中 修改values.yaml配置，启用RBAC支持，相关配置见nginx-ingress chart。 helm package . 查看niginx-ingress $ helm search nginx-ingress NAME VERSION DESCRIPTION local/nginx-ingress 0.8.9 An nginx Ingress controller that uses ConfigMap... stable/nginx-ingress 0.8.9 An nginx Ingress controller that uses ConfigMap... stable/nginx-lego 0.3.0 Chart for nginx-ingress-controller and kube-lego 使用helm部署nginx-ingress $ helm install --name nginx-ingress local/nginx-ingress NAME: nginx-ingress LAST DEPLOYED: Fri Oct 27 18:26:58 2017 NAMESPACE: default STATUS: DEPLOYED RESOURCES: ==> rbac.authorization.k8s.io/v1beta1/Role NAME KIND nginx-ingress-nginx-ingress Role.v1beta1.rbac.authorization.k8s.io ==> rbac.authorization.k8s.io/v1beta1/RoleBinding nginx-ingress-nginx-ingress RoleBinding.v1beta1.rbac.authorization.k8s.io ==> v1/Service NAME CLUSTER-IP EXTERNAL-IP PORT(S) AGE nginx-ingress-nginx-ingress-controller 10.254.100.108 80:30484/TCP,443:31053/TCP 1s nginx-ingress-nginx-ingress-default-backend 10.254.58.156 80/TCP 1s ==> extensions/v1beta1/Deployment NAME DESIRED CURRENT UP-TO-DATE AVAILABLE AGE nginx-ingress-nginx-ingress-default-backend 1 1 1 0 1s nginx-ingress-nginx-ingress-controller 1 1 1 0 1s ==> v1/ConfigMap NAME DATA AGE nginx-ingress-nginx-ingress-controller 1 1s ==> v1/ServiceAccount NAME SECRETS AGE nginx-ingress-nginx-ingress 1 1s ==> rbac.authorization.k8s.io/v1beta1/ClusterRole NAME KIND nginx-ingress-nginx-ingress ClusterRole.v1beta1.rbac.authorization.k8s.io ==> rbac.authorization.k8s.io/v1beta1/ClusterRoleBinding nginx-ingress-nginx-ingress ClusterRoleBinding.v1beta1.rbac.authorization.k8s.io NOTES: The nginx-ingress controller has been installed. Get the application URL by running these commands: export HTTP_NODE_PORT=$(kubectl --namespace default get services -o jsonpath=\"{.spec.ports[0].nodePort}\" nginx-ingress-nginx-ingress-controller) export HTTPS_NODE_PORT=$(kubectl --namespace default get services -o jsonpath=\"{.spec.ports[1].nodePort}\" nginx-ingress-nginx-ingress-controller) export NODE_IP=$(kubectl --namespace default get nodes -o jsonpath=\"{.items[0].status.addresses[1].address}\") echo \"Visit http://$NODE_IP:$HTTP_NODE_PORT to access your application via HTTP.\" echo \"Visit https://$NODE_IP:$HTTPS_NODE_PORT to access your application via HTTPS.\" An example Ingress that makes use of the controller: apiVersion: extensions/v1beta1 kind: Ingress metadata: annotations: kubernetes.io/ingress.class: nginx name: example namespace: foo spec: rules: - host: www.example.com http: paths: - backend: serviceName: exampleService servicePort: 80 path: / # This section is only required if TLS is to be enabled for the Ingress tls: - hosts: - www.example.com secretName: example-tls If TLS is enabled for the Ingress, a Secret containing the certificate and key must also be provided: apiVersion: v1 kind: Secret metadata: name: example-tls namespace: foo data: tls.crt: tls.key: type: kubernetes.io/tls 访问Nginx 首先获取Nginx的地址，从我们使用helm安装nginx-ingress命令的输出中那个可以看到提示，根据提示执行可以看到nginx的http和https地址： export HTTP_NODE_PORT=$(kubectl --namespace default get services -o jsonpath=\"{.spec.ports[0].nodePort}\" nginx-ingress-nginx-ingress-controller) export HTTPS_NODE_PORT=$(kubectl --namespace default get services -o jsonpath=\"{.spec.ports[1].nodePort}\" nginx-ingress-nginx-ingress-controller) export NODE_IP=$(kubectl --namespace default get nodes -o jsonpath=\"{.items[0].status.addresses[1].address}\") echo \"Visit http://$NODE_IP:$HTTP_NODE_PORT to access your application via HTTP.\" echo \"Visit https://$NODE_IP:$HTTPS_NODE_PORT to access your application via HTTPS.\" Visit http://172.20.0.113:30484 to access your application via HTTP. Visit https://172.20.0.113:31053 to access your application via HTTPS. http地址：http://172.20.0.113:30484 https地址：https://172.20.0.113:31053 我们分别在http和https地址上测试一下： /healthz返回200 /返回404错误 curl -v http://172.20.0.113:30484/healthz # 返回200 curl -v http://172.20.0.113:30484/ # 返回404 curl -v --insecure http://172.20.0.113:30484/healthz # 返回200 curl -v --insecure http://172.20.0.113:30484/ # 返回404 删除nginx-ingress helm delete --purge nginx-ingress 使用--purge参数可以彻底删除release不留下记录，否则下一次部署的时候不能使用重名的release。 参考 Ingress-nginx github Nginx chart configuration 使用Helm管理kubernetes应用 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-10-27 19:39:00 "},"practice/operation.html":{"url":"practice/operation.html","title":"运维管理","keywords":"","body":"运维管理 将集群部署到生产环境后就不得不考虑运维管理问题。运维管理问题主要包括如下几个方面： 监控：包括 kubernetes 本身组件和 Pod、应用的监控 日志收集：包括 kubernetes 本身组件的日志，应用的日志 审计：用户对集群操作的审计 安全：用户权限的管理和镜像漏洞扫描 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-08-28 18:37:34 "},"practice/service-rolling-update.html":{"url":"practice/service-rolling-update.html","title":"服务滚动升级","keywords":"","body":"服务滚动升级 当有镜像发布新版本，新版本服务上线时如何实现服务的滚动和平滑升级？ 如果你使用ReplicationController创建的pod可以使用kubectl rollingupdate命令滚动升级，如果使用的是Deployment创建的Pod可以直接修改yaml文件后执行kubectl apply即可。 Deployment已经内置了RollingUpdate strategy，因此不用再调用kubectl rollingupdate命令，升级的过程是先创建新版的pod将流量导入到新pod上后销毁原来的旧的pod。 Rolling Update适用于Deployment、Replication Controller，官方推荐使用Deployment而不再使用Replication Controller。 使用ReplicationController时的滚动升级请参考官网说明：https://kubernetes.io/docs/tasks/run-application/rolling-update-replication-controller/ ReplicationController与Deployment的关系 ReplicationController和Deployment的RollingUpdate命令有些不同，但是实现的机制是一样的，关于这两个kind的关系我引用了ReplicationController与Deployment的区别中的部分内容如下，详细区别请查看原文。 ReplicationController Replication Controller为Kubernetes的一个核心内容，应用托管到Kubernetes之后，需要保证应用能够持续的运行，Replication Controller就是这个保证的key，主要的功能如下： 确保pod数量：它会确保Kubernetes中有指定数量的Pod在运行。如果少于指定数量的pod，Replication Controller会创建新的，反之则会删除掉多余的以保证Pod数量不变。 确保pod健康：当pod不健康，运行出错或者无法提供服务时，Replication Controller也会杀死不健康的pod，重新创建新的。 弹性伸缩 ：在业务高峰或者低峰期的时候，可以通过Replication Controller动态的调整pod的数量来提高资源的利用率。同时，配置相应的监控功能（Hroizontal Pod Autoscaler），会定时自动从监控平台获取Replication Controller关联pod的整体资源使用情况，做到自动伸缩。 滚动升级：滚动升级为一种平滑的升级方式，通过逐步替换的策略，保证整体系统的稳定，在初始化升级的时候就可以及时发现和解决问题，避免问题不断扩大。 Deployment Deployment同样为Kubernetes的一个核心内容，主要职责同样是为了保证pod的数量和健康，90%的功能与Replication Controller完全一样，可以看做新一代的Replication Controller。但是，它又具备了Replication Controller之外的新特性： Replication Controller全部功能：Deployment继承了上面描述的Replication Controller全部功能。 事件和状态查看：可以查看Deployment的升级详细进度和状态。 回滚：当升级pod镜像或者相关参数的时候发现问题，可以使用回滚操作回滚到上一个稳定的版本或者指定的版本。 版本记录: 每一次对Deployment的操作，都能保存下来，给予后续可能的回滚使用。 暂停和启动：对于每一次升级，都能够随时暂停和启动。 多种升级方案：Recreate：删除所有已存在的pod,重新创建新的; RollingUpdate：滚动升级，逐步替换的策略，同时滚动升级时，支持更多的附加参数，例如设置最大不可用pod数量，最小升级间隔时间等等。 创建测试镜像 我们来创建一个特别简单的web服务，当你访问网页时，将输出一句版本信息。通过区分这句版本信息输出我们就可以断定升级是否完成。 所有配置和代码见../manifests/test/rolling-update-test目录。 Web服务的代码main.go package main import ( \"fmt\" \"log\" \"net/http\" ) func sayhello(w http.ResponseWriter, r *http.Request) { fmt.Fprintf(w, \"This is version 1.\") //这个写入到w的是输出到客户端的 } func main() { http.HandleFunc(\"/\", sayhello) //设置访问的路由 log.Println(\"This is version 1.\") err := http.ListenAndServe(\":9090\", nil) //设置监听的端口 if err != nil { log.Fatal(\"ListenAndServe: \", err) } } 创建Dockerfile FROM alpine:3.5 MAINTAINER Jimmy Song ADD hellov2 / ENTRYPOINT [\"/hellov2\"] 注意修改添加的文件的名称。 创建Makefile 修改镜像仓库的地址为你自己的私有镜像仓库地址。 修改Makefile中的TAG为新的版本号。 all: build push clean .PHONY: build push clean TAG = v1 # Build for linux amd64 build: GOOS=linux GOARCH=amd64 go build -o hello${TAG} main.go docker build -t sz-pg-oam-docker-hub-001.tendcloud.com/library/hello:${TAG} . # Push to tenxcloud push: docker push sz-pg-oam-docker-hub-001.tendcloud.com/library/hello:${TAG} # Clean clean: rm -f hello${TAG} 编译 make all 分别修改main.go中的输出语句、Dockerfile中的文件名称和Makefile中的TAG，创建两个版本的镜像。 测试 我们使用Deployment部署服务来测试。 配置文件rolling-update-test.yaml： apiVersion: extensions/v1beta1 kind: Deployment metadata: name: rolling-update-test spec: replicas: 3 template: metadata: labels: app: rolling-update-test spec: containers: - name: rolling-update-test image: sz-pg-oam-docker-hub-001.tendcloud.com/library/hello:v1 ports: - containerPort: 9090 --- apiVersion: v1 kind: Service metadata: name: rolling-update-test labels: app: rolling-update-test spec: ports: - port: 9090 protocol: TCP name: http selector: app: rolling-update-test 部署service kubectl create -f rolling-update-test.yaml 修改traefik ingress配置 在ingress.yaml文件中增加新service的配置。 - host: rolling-update-test.traefik.io http: paths: - path: / backend: serviceName: rolling-update-test servicePort: 9090 修改本地的host配置，增加一条配置： 172.20.0.119 rolling-update-test.traefik.io 注意：172.20.0.119是我们之前使用keepalived创建的VIP。 打开浏览器访问http://rolling-update-test.traefik.io将会看到以下输出： This is version 1. 滚动升级 只需要将rolling-update-test.yaml文件中的image改成新版本的镜像名，然后执行： kubectl apply -f rolling-update-test.yaml 也可以参考Kubernetes Deployment Concept中的方法，直接设置新的镜像。 kubectl set image deployment/rolling-update-test rolling-update-test=sz-pg-oam-docker-hub-001.tendcloud.com/library/hello:v2 或者使用kubectl edit deployment/rolling-update-test修改镜像名称后保存。 使用以下命令查看升级进度： kubectl rollout status deployment/rolling-update-test 升级完成后在浏览器中刷新http://rolling-update-test.traefik.io将会看到以下输出： This is version 2. 说明滚动升级成功。 使用ReplicationController创建的Pod如何RollingUpdate 以上讲解使用Deployment创建的Pod的RollingUpdate方式，那么如果使用传统的ReplicationController创建的Pod如何Update呢？ 举个例子： $ kubectl -n spark-cluster rolling-update zeppelin-controller --image sz-pg-oam-docker-hub-001.tendcloud.com/library/zeppelin:0.7.1 Created zeppelin-controller-99be89dbbe5cd5b8d6feab8f57a04a8b Scaling up zeppelin-controller-99be89dbbe5cd5b8d6feab8f57a04a8b from 0 to 1, scaling down zeppelin-controller from 1 to 0 (keep 1 pods available, don't exceed 2 pods) Scaling zeppelin-controller-99be89dbbe5cd5b8d6feab8f57a04a8b up to 1 Scaling zeppelin-controller down to 0 Update succeeded. Deleting old controller: zeppelin-controller Renaming zeppelin-controller-99be89dbbe5cd5b8d6feab8f57a04a8b to zeppelin-controller replicationcontroller \"zeppelin-controller\" rolling updated 只需要指定新的镜像即可，当然你可以配置RollingUpdate的策略。 参考 Rolling update机制解析 Running a Stateless Application Using a Deployment Simple Rolling Update 使用kubernetes的deployment进行RollingUpdate Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-10-11 23:40:19 "},"practice/app-log-collection.html":{"url":"practice/app-log-collection.html","title":"应用日志收集","keywords":"","body":"应用日志收集 前言 在进行日志收集的过程中，我们首先想到的是使用Logstash，因为它是ELK stack中的重要成员，但是在测试过程中发现，Logstash是基于JDK的，在没有产生日志的情况单纯启动Logstash就大概要消耗500M内存，在每个Pod中都启动一个日志收集组件的情况下，使用logstash有点浪费系统资源，经人推荐我们选择使用Filebeat替代，经测试单独启动Filebeat容器大约会消耗12M内存，比起logstash相当轻量级。 方案选择 Kubernetes官方提供了EFK的日志收集解决方案，但是这种方案并不适合所有的业务场景，它本身就有一些局限性，例如： 所有日志都必须是out前台输出，真实业务场景中无法保证所有日志都在前台输出 只能有一个日志输出文件，而真实业务场景中往往有多个日志输出文件 Fluentd并不是常用的日志收集工具，我们更习惯用logstash，现使用filebeat替代 我们已经有自己的ELK集群且有专人维护，没有必要再在kubernetes上做一个日志收集服务 基于以上几个原因，我们决定使用自己的ELK集群。 Kubernetes集群中的日志收集解决方案 编号 方案 优点 缺点 1 每个app的镜像中都集成日志收集组件 部署方便，kubernetes的yaml文件无须特别配置，可以为每个app自定义日志收集配置 强耦合，不方便应用和日志收集组件升级和维护且会导致镜像过大 2 单独创建一个日志收集组件跟app的容器一起运行在同一个pod中 低耦合，扩展性强，方便维护和升级 需要对kubernetes的yaml文件进行单独配置，略显繁琐 3 将所有的Pod的日志都挂载到宿主机上，每台主机上单独起一个日志收集Pod 完全解耦，性能最高，管理起来最方便 需要统一日志收集规则，目录和输出方式 综合以上优缺点，我们选择使用方案二。 该方案在扩展性、个性化、部署和后期维护方面都能做到均衡，因此选择该方案。 图片 - filebeat日志收集架构图 我们创建了自己的filebeat镜像。创建过程和使用方式见https://github.com/rootsongjc/docker-images 镜像地址：index.tenxcloud.com/jimmy/filebeat:5.4.0 测试 我们部署一个应用filebeat来收集日志的功能测试。 创建应用yaml文件filebeat-test.yaml。 apiVersion: extensions/v1beta1 kind: Deployment metadata: name: filebeat-test namespace: default spec: replicas: 3 template: metadata: labels: k8s-app: filebeat-test spec: containers: - image: sz-pg-oam-docker-hub-001.tendcloud.com/library/filebeat:5.4.0 name: filebeat volumeMounts: - name: app-logs mountPath: /log - name: filebeat-config mountPath: /etc/filebeat/ - image: sz-pg-oam-docker-hub-001.tendcloud.com/library/analytics-docker-test:Build_8 name : app ports: - containerPort: 80 volumeMounts: - name: app-logs mountPath: /usr/local/TalkingData/logs volumes: - name: app-logs emptyDir: {} - name: filebeat-config configMap: name: filebeat-config --- apiVersion: v1 kind: Service metadata: name: filebeat-test labels: app: filebeat-test spec: ports: - port: 80 protocol: TCP name: http selector: run: filebeat-test --- apiVersion: v1 kind: ConfigMap metadata: name: filebeat-config data: filebeat.yml: | filebeat.prospectors: - input_type: log paths: - \"/log/*\" - \"/log/usermange/common/*\" output.elasticsearch: hosts: [\"172.23.5.255:9200\"] username: \"elastic\" password: \"changeme\" index: \"filebeat-docker-test\" 说明 该文件中包含了配置文件filebeat的配置文件的ConfigMap，因此不需要再定义环境变量。 当然你也可以不同ConfigMap，通过传统的传递环境变量的方式来配置filebeat。 例如对filebeat的容器进行如下配置： containers: - image: sz-pg-oam-docker-hub-001.tendcloud.com/library/filebeat:5.4.0 name: filebeat volumeMounts: - name: app-logs mountPath: /log env: - name: PATHS value: \"/log/*\" - name: ES_SERVER value: 172.23.5.255:9200 - name: INDEX value: logstash-docker - name: INPUT_TYPE value: log 目前使用这种方式会有个问题，及时PATHS只能传递单个目录，如果想传递多个目录需要修改filebeat镜像的docker-entrypoint.sh脚本，对该环境变量进行解析增加filebeat.yml文件中的PATHS列表。 推荐使用ConfigMap，这样filebeat的配置就能够更灵活。 注意事项 将app的/usr/local/TalkingData/logs目录挂载到filebeat的/log目录下。 该文件可以在manifests/test/filebeat-test.yaml找到。 我使用了自己的私有镜像仓库，测试时请换成自己的应用镜像。 Filebeat的环境变量的值配置请参考https://github.com/rootsongjc/docker-images 创建应用 部署Deployment kubectl create -f filebeat-test.yaml 查看http://172.23.5.255:9200/_cat/indices将可以看到列表有这样的indices： green open filebeat-docker-test 7xPEwEbUQRirk8oDX36gAA 5 1 2151 0 1.6mb 841.8kb 访问Kibana的web页面，查看filebeat-2017.05.17的索引，可以看到filebeat收集到了app日志。 图片 - Kibana页面 点开没个日志条目，可以看到以下详细字段： 图片 - filebeat收集的日志详细信息 _index值即我们在YAML文件的configMap中配置的index值 beat.hostname和beat.name即pod的名称 source表示filebeat容器中的日志目录 我们可以通过人为得使index = service name，这样就可以方便的收集和查看每个service的日志。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-10-11 22:22:55 "},"practice/configuration-best-practice.html":{"url":"practice/configuration-best-practice.html","title":"配置最佳实践","keywords":"","body":"配置最佳实践 本文档旨在汇总和强调用户指南、快速开始文档和示例中的最佳实践。该文档会很活跃并持续更新中。如果你觉得很有用的最佳实践但是本文档中没有包含，欢迎给我们提Pull Request。 通用配置建议 定义配置文件的时候，指定最新的稳定API版本（目前是V1）。 在配置文件push到集群之前应该保存在版本控制系统中。这样当需要的时候能够快速回滚，必要的时候也可以快速的创建集群。 使用YAML格式而不是JSON格式的配置文件。在大多数场景下它们都可以作为数据交换格式，但是YAML格式比起JSON更易读和配置。 尽量将相关的对象放在同一个配置文件里。这样比分成多个文件更容易管理。参考guestbook-all-in-one.yaml文件中的配置（注意，尽管你可以在使用kubectl命令时指定配置文件目录，你也可以在配置文件目录下执行kubectl create——查看下面的详细信息）。 为了简化和最小化配置，也为了防止错误发生，不要指定不必要的默认配置。例如，省略掉ReplicationController的selector和label，如果你希望它们跟podTemplate中的label一样的话，因为那些配置默认是podTemplate的label产生的。更多信息请查看 guestbook app 的yaml文件和 examples 。 将资源对象的描述放在一个annotation中可以更好的内省。 裸的Pods vs Replication Controllers和 Jobs 如果有其他方式替代“裸的“ pod（如没有绑定到replication controller 上的pod），那么就使用其他选择。在node节点出现故障时，裸奔的pod不会被重新调度。Replication Controller总是会重新创建pod，除了明确指定了restartPolicy: Never 的场景。Job 也许是比较合适的选择。 Services 通常最好在创建相关的replication controllers之前先创建service ，你也可以在创建Replication Controller的时候不指定replica数量（默认是1），创建service后，在通过Replication Controller来扩容。这样可以在扩容很多个replica之前先确认pod是正常的。 除非十分必要的情况下（如运行一个node daemon），不要使用hostPort（用来指定暴露在主机上的端口号）。当你给Pod绑定了一个hostPort，该pod可被调度到的主机的受限了，因为端口冲突。如果是为了调试目的来通过端口访问的话，你可以使用 kubectl proxy and apiserver proxy 或者 kubectl port-forward。你可使用 Service 来对外暴露服务。如果你确实需要将pod的端口暴露到主机上，考虑使用 NodePort service。 跟hostPort一样的原因，避免使用 hostNetwork。 如果你不需要kube-proxy的负载均衡的话，可以考虑使用使用headless services。 使用Label 定义 labels 来指定应用或Deployment的 semantic attributes 。例如，不是将label附加到一组pod来显式表示某些服务（例如，service:myservice），或者显式地表示管理pod的replication controller（例如，controller:mycontroller），附加label应该是标示语义属性的标签， 例如{app:myapp,tier:frontend,phase:test,deployment:v3}。 这将允许您选择适合上下文的对象组——例如，所有的”tier:frontend“pod的服务或app是“myapp”的所有“测试”阶段组件。 有关此方法的示例，请参阅guestbook应用程序。 可以通过简单地从其service的选择器中省略特定于发行版本的标签，而不是更新服务的选择器来完全匹配replication controller的选择器，来实现跨越多个部署的服务，例如滚动更新。 为了滚动升级的方便，在Replication Controller的名字中包含版本信息，例如作为名字的后缀。设置一个version标签页是很有用的。滚动更新创建一个新的controller而不是修改现有的controller。因此，version含混不清的controller名字就可能带来问题。查看Rolling Update Replication Controller文档获取更多关于滚动升级命令的信息。 注意 Deployment 对象不需要再管理 replication controller 的版本名。Deployment 中描述了对象的期望状态，如果对spec的更改被应用了话，Deployment controller 会以控制的速率来更改实际状态到期望状态。（Deployment目前是 extensions API Group的一部分）。 利用label做调试。因为Kubernetes replication controller和service使用label来匹配pods，这允许你通过移除pod中的label的方式将其从一个controller或者service中移除，原来的controller会创建一个新的pod来取代移除的pod。这是一个很有用的方式，帮你在一个隔离的环境中调试之前的“活着的” pod。查看 kubectl label 命令。 容器镜像 默认容器镜像拉取策略 是 IfNotPresent, 当本地已存在该镜像的时候 Kubelet 不会再从镜像仓库拉取。如果你希望总是从镜像仓库中拉取镜像的话，在yaml文件中指定镜像拉取策略为Always（ imagePullPolicy: Always）或者指定镜像的tag为 :latest 。 如果你没有将镜像标签指定为:latest，例如指定为myimage:v1，当该标签的镜像进行了更新，kubelet也不会拉取该镜像。你可以在每次镜像更新后都生成一个新的tag（例如myimage:v2），在配置文件中明确指定该版本。 注意： 在生产环境下部署容器应该尽量避免使用:latest标签，因为这样很难追溯到底运行的是哪个版本的容器和回滚。 使用kubectl 尽量使用 kubectl create -f 。kubeclt会自动查找该目录下的所有后缀名为.yaml、.yml和.json文件并将它们传递给create命令。 使用 kubectl delete 而不是 stop. Delete 是 stop的超集，stop 已经被弃用。 使用 kubectl bulk 操作（通过文件或者label）来get和delete。查看label selectors 和 using labels effectively。 使用 kubectl run 和 expose 命令快速创建只有单个容器的Deployment。查看 quick start guide中的示例。 参考 Configuration Best Practices Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-08-21 18:23:34 "},"practice/monitor.html":{"url":"practice/monitor.html","title":"集群及应用监控","keywords":"","body":"集群及应用监控 在前面的安装heapster插件章节，我们已经谈到Kubernetes本身提供了监控插件作为集群和容器监控的选择，但是在实际使用中，因为种种原因，再考虑到跟我们自身的监控系统集成，我们准备重新造轮子。 针对kubernetes集群和应用的监控，相较于传统的虚拟机和物理机的监控有很多不同，因此对于传统监控需要有很多改造的地方，需要关注以下三个方面： Kubernetes集群本身的监控，主要是kubernetes的各个组件 kubernetes集群中Pod的监控，Pod的CPU、内存、网络、磁盘等监控 集群内部应用的监控，针对应用本身的监控 Kubernetes集群中的监控 图片 - Kubernetes集群中的监控 跟物理机器和虚拟机的监控不同，在kuberntes集群中的监控复杂度更高一些，因为多了一个虚拟化层，当然这个跟直接监控docker容器又不一样，kubernetes在docker之上又抽象了一层service的概念。 在kubernetes中的监控需要考虑到这几个方面： 应该给Pod打上哪些label，这些label将成为监控的metrics。 当应用的Pod漂移了之后怎么办？因为要考虑到Pod的生命周期比虚拟机和物理机短的多，如何持续监控应用的状态？ 更多的监控项，kubernetes本身、容器、应用等。 监控指标的来源，是通过heapster收集后汇聚还是直接从每台主机的docker上取？ 容器的命名规则 首先我们需要清楚使用cAdvisor收集的数据的格式和字段信息。 当我们通过cAdvisor获取到了容器的信息后，例如访问${NODE_IP}:4194/api/v1.3/docker获取的json结果中的某个容器包含如下字段： \"labels\": { \"annotation.io.kubernetes.container.hash\": \"f47f0602\", \"annotation.io.kubernetes.container.ports\": \"[{\\\"containerPort\\\":80,\\\"protocol\\\":\\\"TCP\\\"}]\", \"annotation.io.kubernetes.container.restartCount\": \"0\", \"annotation.io.kubernetes.container.terminationMessagePath\": \"/dev/termination-log\", \"annotation.io.kubernetes.container.terminationMessagePolicy\": \"File\", \"annotation.io.kubernetes.pod.terminationGracePeriod\": \"30\", \"io.kubernetes.container.logpath\": \"/var/log/pods/d8a2e995-3617-11e7-a4b0-ecf4bbe5d414/php-redis_0.log\", \"io.kubernetes.container.name\": \"php-redis\", \"io.kubernetes.docker.type\": \"container\", \"io.kubernetes.pod.name\": \"frontend-2337258262-771lz\", \"io.kubernetes.pod.namespace\": \"default\", \"io.kubernetes.pod.uid\": \"d8a2e995-3617-11e7-a4b0-ecf4bbe5d414\", \"io.kubernetes.sandbox.id\": \"843a0f018c0cef2a5451434713ea3f409f0debc2101d2264227e814ca0745677\" }, 这些信息其实都是kubernetes创建容器时给docker container打的Labels，使用docker inspect $conainer_name命令同样可以看到上述信息。 你是否想过这些label跟容器的名字有什么关系？当你在node节点上执行docker ps看到的容器名字又对应哪个应用的Pod呢？ 在kubernetes代码中pkg/kubelet/dockertools/docker.go中的BuildDockerName方法定义了容器的名称规范。 这段容器名称定义代码如下： // Creates a name which can be reversed to identify both full pod name and container name. // This function returns stable name, unique name and a unique id. // Although rand.Uint32() is not really unique, but it's enough for us because error will // only occur when instances of the same container in the same pod have the same UID. The // chance is really slim. func BuildDockerName(dockerName KubeletContainerName, container *v1.Container) (string, string, string) { containerName := dockerName.ContainerName + \".\" + strconv.FormatUint(kubecontainer.HashContainerLegacy(container), 16) stableName := fmt.Sprintf(\"%s_%s_%s_%s\", containerNamePrefix, containerName, dockerName.PodFullName, dockerName.PodUID) UID := fmt.Sprintf(\"%08x\", rand.Uint32()) return stableName, fmt.Sprintf(\"%s_%s\", stableName, UID), UID } // Unpacks a container name, returning the pod full name and container name we would have used to // construct the docker name. If we are unable to parse the name, an error is returned. func ParseDockerName(name string) (dockerName *KubeletContainerName, hash uint64, err error) { // For some reason docker appears to be appending '/' to names. // If it's there, strip it. name = strings.TrimPrefix(name, \"/\") parts := strings.Split(name, \"_\") if len(parts) == 0 || parts[0] != containerNamePrefix { err = fmt.Errorf(\"failed to parse Docker container name %q into parts\", name) return nil, 0, err } if len(parts) 1 { hash, err = strconv.ParseUint(nameParts[1], 16, 32) if err != nil { glog.Warningf(\"invalid container hash %q in container %q\", nameParts[1], name) } } podFullName := parts[2] + \"_\" + parts[3] podUID := types.UID(parts[4]) return &KubeletContainerName{podFullName, podUID, containerName}, hash, nil } 我们可以看到容器名称中包含如下几个字段，中间用下划线隔开，至少有6个字段，未来可能添加更多字段。 下面的是四个基本字段。 containerNamePrefix_containerName_PodFullName_PodUID 所有kubernetes启动的容器的containerNamePrefix都是k8s。 Kubernetes启动的docker容器的容器名称规范，下面以官方示例guestbook为例，Deployment 名为 frontend中启动的名为php-redis的docker容器的副本书为3。 Deployment frontend的配置如下： apiVersion: extensions/v1beta1 kind: Deployment metadata: name: frontend spec: template: metadata: labels: app: guestbook tier: frontend spec: containers: - name: php-redis image: bj-xg-oam-docker-hub-001.tendcloud.com/library/gb-frontend:v4 resources: requests: cpu: 100m memory: 100Mi env: - name: GET_HOSTS_FROM value: dns ports: - containerPort: 80 我们选取三个实例中的一个运行php-redis的docker容器。 k8s_php-redis_frontend-2337258262-154p7_default_d8a2e2dd-3617-11e7-a4b0-ecf4bbe5d414_0 containerNamePrefix：k8s containerName：php-redis podFullName：frontend-2337258262-154p7 computeHash：154p7 deploymentName：frontend replicaSetName：frontend-2337258262 namespace：default podUID：d8a2e2dd-3617-11e7-a4b0-ecf4bbe5d414 kubernetes容器命名规则解析，见下图所示。 图片 - kubernetes的容器命名规则示意图 使用Heapster进行集群监控 Heapster是kubernetes官方提供的监控方案，我们在前面的章节中已经讲解了如何部署和使用heapster，见安装Heapster插件。 但是Grafana显示的指标只根据Namespace和Pod两层来分类，实在有些单薄，我们希望通过应用的label增加service这一层分类。架构图如下： 图片 - Heapster架构图（改进版） 在不改变原有架构的基础上，通过应用的label来区分不同应用的pod。 应用监控 Kubernetes中应用的监控架构如图： 图片 - 应用监控架构图 这种方式有以下几个要点： 访问kubernetes API获取应用Pod的IP和端口 Pod labels作为监控metric的tag 直接访问应用的Pod的IP和端口获取应用监控数据 metrics发送到OWL中存储和展示 应用拓扑状态图 对于复杂的应用编排和依赖关系，我们希望能够有清晰的图标一览应用状态和拓扑关系，因此我们用到了Weaveworks开源的scope。 安装scope 我们在kubernetes集群上使用standalone方式安装，详情参考Installing Weave Scope。 使用scope.yaml文件安装scope，该服务安装在kube-system namespace下。 $ kubectl apply -f scope.yaml 创建一个新的Ingress：kube-system.yaml，配置如下： apiVersion: extensions/v1beta1 kind: Ingress metadata: name: traefik-ingress namespace: kube-system spec: rules: - host: scope.weave.io http: paths: - path: / backend: serviceName: weave-scope-app servicePort: 80 执行kubectl apply -f kube-system.yaml后在你的主机上的/etc/hosts文件中添加一条记录： 172.20.0.119 scope.weave.io 在浏览器中访问scope.weave.io就可以访问到scope了，详见边缘节点配置。 图片 - 应用拓扑图 如上图所示，scope可以监控kubernetes集群中的一系列资源的状态、资源使用情况、应用拓扑、scale、还可以直接通过浏览器进入容器内部调试等。 参考 Monitoring in the Kubernetes Era Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-10-11 23:41:27 "},"practice/data-persistence-problem.html":{"url":"practice/data-persistence-problem.html","title":"数据持久化问题","keywords":"","body":"数据落盘问题的由来 这本质上是数据持久化问题，对于有些应用依赖持久化数据，比如应用自身产生的日志需要持久化存储的情况，需要保证容器里的数据不丢失，在Pod挂掉后，其他应用依然可以访问到这些数据，因此我们需要将数据持久化存储起来。 数据落盘问题解决方案 下面以一个应用的日志收集为例，该日志需要持久化收集到ElasticSearch集群中，如果不考虑数据丢失的情形，可以直接使用前面提到的应用日志收集一节中的方法，但考虑到Pod挂掉时logstash（或filebeat）并没有收集完该pod内日志的情形，我们想到了如下这种解决方案，示意图如下： 图片 - 日志持久化收集解决方案示意图 首先需要给数据落盘的应用划分node，即这些应用只调用到若干台主机上 给这若干台主机增加label 使用deamonset方式在这若干台主机上启动logstash的Pod（使用nodeSelector来限定在这几台主机上，我们在边缘节点启动的treafik也是这种模式） 将应用的数据通过volume挂载到宿主机上 Logstash（或者filebeat）收集宿主机上的数据，数据持久化不会丢失 Side-effect 首先kubernetes本身就提供了数据持久化的解决方案statefulset，不过需要用到公有云的存储货其他分布式存储，这一点在我们的私有云环境里被否定了。 需要管理主机的label，增加运维复杂度，但是具体问题具体对待 必须保证应用启动顺序，需要先启动logstash 为主机打label使用nodeSelector的方式限制了资源调度的范围 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-08-21 18:23:34 "},"practice/manage-compute-resources-container.html":{"url":"practice/manage-compute-resources-container.html","title":"管理容器的计算资源","keywords":"","body":"管理容器的计算资源 当您定义 Pod 的时候可以选择为每个容器指定需要的 CPU 和内存（RAM）大小。当为容器指定了资源请求后，调度器就能够更好的判断出将容器调度到哪个节点上。如果您还为容器指定了资源限制，节点上的资源就可以按照指定的方式做竞争。关于资源请求和限制的不同点和更多资料请参考 Resource QoS。 资源类型 CPU 和 memory 都是 资源类型。资源类型具有基本单位。CPU 的单位是 core，memory 的单位是 byte。 CPU和内存统称为计算资源，也可以称为资源。计算资源的数量是可以被请求、分配和消耗的可测量的。它们与 API 资源 不同。 API 资源（如 Pod 和 Service）是可通过 Kubernetes API server 读取和修改的对象。 Pod 和 容器的资源请求和限制 Pod 中的每个容器都可以指定以下的一个或者多个值： spec.containers[].resources.limits.cpu spec.containers[].resources.limits.memory spec.containers[].resources.requests.cpu spec.containers[].resources.requests.memory 尽管只能在个别容器上指定请求和限制，但是我们可以方便地计算出 Pod 资源请求和限制。特定资源类型的Pod 资源请求/限制是 Pod 中每个容器的该类型的资源请求/限制的总和。 CPU 的含义 CPU 资源的限制和请求以 cpu 为单位。 Kubernetes 中的一个 cpu 等于： 1 AWS vCPU 1 GCP Core 1 Azure vCore 1 Hyperthread 在带有超线程的裸机 Intel 处理器上 允许浮点数请求。具有 spec.containers[].resources.requests.cpu 为 0.5 的容器保证了一半 CPU 要求 1 CPU的一半。表达式 0.1 等价于表达式 100m，可以看作 “100 millicpu”。有些人说成是“一百毫 cpu”，其实说的是同样的事情。具有小数点（如 0.1）的请求由 API 转换为100m，精度不超过 1m。因此，可能会优先选择 100m 的形式。 CPU 总是要用绝对数量，不可以使用相对数量；0.1 的 CPU 在单核、双核、48核的机器中的意义是一样的。 内存的含义 内存的限制和请求以字节为单位。您可以使用以下后缀之一作为平均整数或定点整数表示内存：E，P，T，G，M，K。您还可以使用两个字母的等效的幂数：Ei，Pi，Ti ，Gi，Mi，Ki。例如，以下代表大致相同的值： 128974848, 129e6, 129M, 123Mi 下面是个例子。 以下 Pod 有两个容器。每个容器的请求为 0.25 cpu 和 64MiB（226 字节）内存，每个容器的限制为 0.5 cpu 和 128MiB 内存。您可以说该 Pod 请求 0.5 cpu 和 128 MiB 的内存，限制为 1 cpu 和 256MiB 的内存。 apiVersion: v1 kind: Pod metadata: name: frontend spec: containers: - name: db image: mysql resources: requests: memory: \"64Mi\" cpu: \"250m\" limits: memory: \"128Mi\" cpu: \"500m\" - name: wp image: wordpress resources: requests: memory: \"64Mi\" cpu: \"250m\" limits: memory: \"128Mi\" cpu: \"500m\" 具有资源请求的 Pod 如何调度 当您创建一个 Pod 时，Kubernetes 调度程序将为 Pod 选择一个节点。每个节点具有每种资源类型的最大容量：可为 Pod 提供的 CPU 和内存量。调度程序确保对于每种资源类型，调度的容器的资源请求的总和小于节点的容量。请注意，尽管节点上的实际内存或 CPU 资源使用量非常低，但如果容量检查失败，则调度程序仍然拒绝在该节点上放置 Pod。当资源使用量稍后增加时，例如在请求率的每日峰值期间，这可以防止节点上的资源短缺。 具有资源限制的 Pod 如何运行 当 kubelet 启动一个 Pod 的容器时，它会将 CPU 和内存限制传递到容器运行时。 当使用 Docker 时： spec.containers[].resources.requests.cpu 的值将转换成 millicore 值，这是个浮点数，并乘以1024，这个数字中的较大者或2用作 docker run 命令中的 --cpu-shares 标志的值。 spec.containers[].resources.limits.cpu 被转换成 millicore 值。被乘以 100000 然后 除以 1000。这个数字用作 docker run 命令中的 --cpu-quota 标志的值。[--cpu-quota ] 标志被设置成了 100000，表示测量配额使用的默认100ms 周期。如果 [--cpu-cfs-quota] 标志设置为 true，则 kubelet 会强制执行 cpu 限制。从 Kubernetes 1.2 版本起，此标志默认为 true。 spec.containers[].resources.limits.memory 被转换为整型，作为 docker run 命令中的 --memory 标志的值。 如果容器超过其内存限制，则可能会被终止。如果可重新启动，则与所有其他类型的运行时故障一样，kubelet 将重新启动它。 如果一个容器超过其内存请求，那么当节点内存不足时，它的 Pod 可能被逐出。 容器可能被允许也可能不被允许超过其 CPU 限制时间。但是，由于 CPU 使用率过高，不会被杀死。 要确定容器是否由于资源限制而无法安排或被杀死，请参阅 疑难解答 部分。 监控计算资源使用 Pod 的资源使用情况被报告为 Pod 状态的一部分。 如果为集群配置了 可选监控，则可以从监控系统检索 Pod 资源的使用情况。 疑难解答 我的 Pod 处于 pending 状态且事件信息显示 failedScheduling 如果调度器找不到任何该 Pod 可以匹配的节点，则该 Pod 将保持不可调度状态，直到找到一个可以被调度到的位置。每当调度器找不到 Pod 可以调度的地方时，会产生一个事件，如下所示： $ kubectl describe pod frontend | grep -A 3 Events Events: FirstSeen LastSeen Count From Subobject PathReason Message 36s 5s 6 {scheduler } FailedScheduling Failed for reason PodExceedsFreeCPU and possibly others 在上述示例中，由于节点上的 CPU 资源不足，名为 “frontend” 的 Pod 将无法调度。由于内存不足（PodExceedsFreeMemory），类似的错误消息也可能会导致失败。一般来说，如果有这种类型的消息而处于 pending 状态，您可以尝试如下几件事情： $ kubectl describe nodes e2e-test-minion-group-4lw4 Name: e2e-test-minion-group-4lw4 [ ... lines removed for clarity ...] Capacity: alpha.kubernetes.io/nvidia-gpu: 0 cpu: 2 memory: 7679792Ki pods: 110 Allocatable: alpha.kubernetes.io/nvidia-gpu: 0 cpu: 1800m memory: 7474992Ki pods: 110 [ ... lines removed for clarity ...] Non-terminated Pods: (5 in total) Namespace Name CPU Requests CPU Limits Memory Requests Memory Limits --------- ---- ------------ ---------- --------------- ------------- kube-system fluentd-gcp-v1.38-28bv1 100m (5%) 0 (0%) 200Mi (2%) 200Mi (2%) kube-system kube-dns-3297075139-61lj3 260m (13%) 0 (0%) 100Mi (1%) 170Mi (2%) kube-system kube-proxy-e2e-test-... 100m (5%) 0 (0%) 0 (0%) 0 (0%) kube-system monitoring-influxdb-grafana-v4-z1m12 200m (10%) 200m (10%) 600Mi (8%) 600Mi (8%) kube-system node-problem-detector-v0.1-fj7m3 20m (1%) 200m (10%) 20Mi (0%) 100Mi (1%) Allocated resources: (Total limits may be over 100 percent, i.e., overcommitted.) CPU Requests CPU Limits Memory Requests Memory Limits ------------ ---------- --------------- ------------- 680m (34%) 400m (20%) 920Mi (12%) 1070Mi (14%) 我的容器被终结了 您的容器可能因为资源枯竭而被终结了。要查看容器是否因为遇到资源限制而被杀死，请在相关的 Pod 上调用 kubectl describe pod： [12:54:41] $ kubectl describe pod simmemleak-hra99 Name: simmemleak-hra99 Namespace: default Image(s): saadali/simmemleak Node: kubernetes-node-tf0f/10.240.216.66 Labels: name=simmemleak Status: Running Reason: Message: IP: 10.244.2.75 Replication Controllers: simmemleak (1/1 replicas created) Containers: simmemleak: Image: saadali/simmemleak Limits: cpu: 100m memory: 50Mi State: Running Started: Tue, 07 Jul 2015 12:54:41 -0700 Last Termination State: Terminated Exit Code: 1 Started: Fri, 07 Jul 2015 12:54:30 -0700 Finished: Fri, 07 Jul 2015 12:54:33 -0700 Ready: False Restart Count: 5 Conditions: Type Status Ready False Events: FirstSeen LastSeen Count From SubobjectPath Reason Message Tue, 07 Jul 2015 12:53:51 -0700 Tue, 07 Jul 2015 12:53:51 -0700 1 {scheduler } scheduled Successfully assigned simmemleak-hra99 to kubernetes-node-tf0f Tue, 07 Jul 2015 12:53:51 -0700 Tue, 07 Jul 2015 12:53:51 -0700 1 {kubelet kubernetes-node-tf0f} implicitly required container POD pulled Pod container image \"gcr.io/google_containers/pause:0.8.0\" already present on machine Tue, 07 Jul 2015 12:53:51 -0700 Tue, 07 Jul 2015 12:53:51 -0700 1 {kubelet kubernetes-node-tf0f} implicitly required container POD created Created with docker id 6a41280f516d Tue, 07 Jul 2015 12:53:51 -0700 Tue, 07 Jul 2015 12:53:51 -0700 1 {kubelet kubernetes-node-tf0f} implicitly required container POD started Started with docker id 6a41280f516d Tue, 07 Jul 2015 12:53:51 -0700 Tue, 07 Jul 2015 12:53:51 -0700 1 {kubelet kubernetes-node-tf0f} spec.containers{simmemleak} created Created with docker id 87348f12526a 在上面的例子中，Restart Count: 5 意味着 Pod 中的 simmemleak 容器被终止并重启了五次。 您可以使用 kubectl get pod 命令加上 -o go-template=... 选项来获取之前终止容器的状态。 [13:59:01] $ kubectl get pod -o go-template='{{range.status.containerStatuses}}{{\"Container Name: \"}}{{.name}}{{\"\\r\\nLastState: \"}}{{.lastState}}{{end}}' simmemleak-60xbc Container Name: simmemleak LastState: map[terminated:map[exitCode:137 reason:OOM Killed startedAt:2015-07-07T20:58:43Z finishedAt:2015-07-07T20:58:43Z containerID:docker://0e4095bba1feccdfe7ef9fb6ebffe972b4b14285d5acdec6f0d3ae8a22fad8b2]] 您可以看到容器因为 reason:OOM killed 被终止，OOM 表示 Out Of Memory。 不透明整型资源（Alpha功能） Kubernetes 1.5 版本中引入不透明整型资源。不透明的整数资源允许集群运维人员发布新的节点级资源，否则系统将不了解这些资源。 用户可以在 Pod 的 spec 中消费这些资源，就像 CPU 和内存一样。调度器负责资源计量，以便在不超过可用量的同时分配给 Pod。 注意： 不透明整型资源在 kubernetes 1.5 中还是 Alpha 版本。只实现了资源计量，节点级别的隔离还处于积极的开发阶段。 不透明整型资源是以 pod.alpha.kubernetes.io/opaque-int-resource- 为前缀的资源。API server 将限制这些资源的数量为整数。有效 数量的例子有 3、3000m 和 3Ki。无效数量的例子有 0.5 和 1500m。 申请使用不透明整型资源需要两步。首先，集群运维人员必须在一个或多个节点上通告每个节点不透明的资源。然后，用户必须在 Pod 中请求不透明资源。 要发布新的不透明整型资源，集群运维人员应向 API server 提交 PATCH HTTP请求，以指定集群中节点的status.capacity 的可用数量。在此操作之后，节点的 status.capacity 将包括一个新的资源。 status.allocatable 字段由 kubelet 异步地使用新资源自动更新。请注意，由于调度器在评估 Pod 适应度时使用节点 status.allocatable 值，所以在使用新资源修补节点容量和请求在该节点上调度资源的第一个 pod 之间可能会有短暂的延迟。 示例 这是一个 HTTP 请求，master 节点是 k8s-master，在 k8s-node-1 节点上通告 5 个 “foo” 资源。 PATCH /api/v1/nodes/k8s-node-1/status HTTP/1.1 Accept: application/json Content-Type: application/json-patch+json Host: k8s-master:8080 [ { \"op\": \"add\", \"path\": \"/status/capacity/pod.alpha.kubernetes.io~1opaque-int-resource-foo\", \"value\": \"5\" } ] curl --header \"Content-Type: application/json-patch+json\" \\ --request PATCH \\ --data '[{\"op\": \"add\", \"path\": \"/status/capacity/pod.alpha.kubernetes.io~1opaque-int-resource-foo\", \"value\": \"5\"}]' \\ http://k8s-master:8080/api/v1/nodes/k8s-node-1/status 注意： 在前面的请求中，~1 是 patch 路径中 / 字符的编码。JSON-Patch 中的操作路径值被解释为 JSON-Pointer。更多详细信息请参阅 IETF RFC 6901, section 3。 apiVersion: v1 kind: Pod metadata: name: my-pod spec: containers: - name: my-container image: myimage resources: requests: cpu: 2 pod.alpha.kubernetes.io/opaque-int-resource-foo: 1 计划改进 在 kubernetes 1.5 版本中仅允许在容器上指定资源量。计划改进对所有容器在 Pod 中共享资源的计量，如 emptyDir volume。 在 kubernetes 1.5 版本中仅支持容器对 CPU 和内存的申请和限制。计划增加新的资源类型，包括节点磁盘空间资源和一个可支持自定义 资源类型 的框架。 Kubernetes 通过支持通过多级别的 服务质量 来支持资源的过度使用。 在 kubernetes 1.5 版本中，一个 CPU 单位在不同的云提供商和同一云提供商的不同机器类型中的意味都不同。例如，在 AWS 上，节点的容量报告为 ECU，而在 GCE 中报告为逻辑内核。我们计划修改 cpu 资源的定义，以便在不同的提供商和平台之间保持一致。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-09-28 17:36:25 "},"practice/storage.html":{"url":"practice/storage.html","title":"存储管理","keywords":"","body":"存储管理 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-08-21 18:23:35 "},"practice/glusterfs.html":{"url":"practice/glusterfs.html","title":"GlusterFS","keywords":"","body":"GlusterFS GlusterFS是Scale-Out存储解决方案Gluster的核心，它是一个开源的分布式文件系统，具有强大的横向扩展能力，通过扩展能够支持数PB存储容量和处理数千客户端。GlusterFS借助TCP/IP或InfiniBand RDMA网络将物理分布的存储资源聚集在一起，使用单一全局命名空间来管理数据。GlusterFS基于可堆叠的用户空间设计，可为各种不同的数据负载提供优异的性能。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-08-21 18:23:34 "},"practice/using-glusterfs-for-persistent-storage.html":{"url":"practice/using-glusterfs-for-persistent-storage.html","title":"使用GlusterFS做持久化存储","keywords":"","body":"使用glusterfs做持久化存储 我们复用kubernetes的三台主机做glusterfs存储。 以下步骤参考自：https://www.xf80.com/2017/04/21/kubernetes-glusterfs/（该网站已无法访问） 安装glusterfs 我们直接在物理机上使用yum安装，如果你选择在kubernetes上安装，请参考：https://github.com/gluster/gluster-kubernetes/blob/master/docs/setup-guide.md # 先安装 gluster 源 $ yum install centos-release-gluster -y # 安装 glusterfs 组件 $ yum install -y glusterfs glusterfs-server glusterfs-fuse glusterfs-rdma glusterfs-geo-replication glusterfs-devel ## 创建 glusterfs 目录 $ mkdir /opt/glusterd ## 修改 glusterd 目录 $ sed -i 's/var\\/lib/opt/g' /etc/glusterfs/glusterd.vol # 启动 glusterfs $ systemctl start glusterd.service # 设置开机启动 $ systemctl enable glusterd.service #查看状态 $ systemctl status glusterd.service 配置 glusterfs # 配置 hosts $ vi /etc/hosts 172.20.0.113 sz-pg-oam-docker-test-001.tendcloud.com 172.20.0.114 sz-pg-oam-docker-test-002.tendcloud.com 172.20.0.115 sz-pg-oam-docker-test-003.tendcloud.com # 开放端口 $ iptables -I INPUT -p tcp --dport 24007 -j ACCEPT # 创建存储目录 $ mkdir /opt/gfs_data # 添加节点到 集群 # 执行操作的本机不需要probe 本机 [root@sz-pg-oam-docker-test-001 ~]# gluster peer probe sz-pg-oam-docker-test-002.tendcloud.com gluster peer probe sz-pg-oam-docker-test-003.tendcloud.com # 查看集群状态 $ gluster peer status Number of Peers: 2 Hostname: sz-pg-oam-docker-test-002.tendcloud.com Uuid: f25546cc-2011-457d-ba24-342554b51317 State: Peer in Cluster (Connected) Hostname: sz-pg-oam-docker-test-003.tendcloud.com Uuid: 42b6cad1-aa01-46d0-bbba-f7ec6821d66d State: Peer in Cluster (Connected) 配置 volume GlusterFS中的volume的模式有很多中，包括以下几种： 分布卷（默认模式）：即DHT, 也叫 分布卷: 将文件已hash算法随机分布到 一台服务器节点中存储。 复制模式：即AFR, 创建volume 时带 replica x 数量: 将文件复制到 replica x 个节点中。 条带模式：即Striped, 创建volume 时带 stripe x 数量： 将文件切割成数据块，分别存储到 stripe x 个节点中 ( 类似raid 0 )。 分布式条带模式：最少需要4台服务器才能创建。 创建volume 时 stripe 2 server = 4 个节点： 是DHT 与 Striped 的组合型。 分布式复制模式：最少需要4台服务器才能创建。 创建volume 时 replica 2 server = 4 个节点：是DHT 与 AFR 的组合型。 条带复制卷模式：最少需要4台服务器才能创建。 创建volume 时 stripe 2 replica 2 server = 4 个节点： 是 Striped 与 AFR 的组合型。 三种模式混合： 至少需要8台 服务器才能创建。 stripe 2 replica 2 , 每4个节点 组成一个 组。 这几种模式的示例图参考：CentOS7安装GlusterFS。 因为我们只有三台主机，在此我们使用默认的分布卷模式。请勿在生产环境上使用该模式，容易导致数据丢失。 # 创建分布卷 $ gluster volume create k8s-volume transport tcp sz-pg-oam-docker-test-001.tendcloud.com:/opt/gfs_data sz-pg-oam-docker-test-002.tendcloud.com:/opt/gfs_data sz-pg-oam-docker-test-003.tendcloud.com:/opt/gfs_data force # 查看volume状态 $ gluster volume info Volume Name: k8s-volume Type: Distribute Volume ID: 9a3b0710-4565-4eb7-abae-1d5c8ed625ac Status: Created Snapshot Count: 0 Number of Bricks: 3 Transport-type: tcp Bricks: Brick1: sz-pg-oam-docker-test-001.tendcloud.com:/opt/gfs_data Brick2: sz-pg-oam-docker-test-002.tendcloud.com:/opt/gfs_data Brick3: sz-pg-oam-docker-test-003.tendcloud.com:/opt/gfs_data Options Reconfigured: transport.address-family: inet nfs.disable: on # 启动 分布卷 $ gluster volume start k8s-volume Glusterfs调优 # 开启 指定 volume 的配额 $ gluster volume quota k8s-volume enable # 限制 指定 volume 的配额 $ gluster volume quota k8s-volume limit-usage / 1TB # 设置 cache 大小, 默认32MB $ gluster volume set k8s-volume performance.cache-size 4GB # 设置 io 线程, 太大会导致进程崩溃 $ gluster volume set k8s-volume performance.io-thread-count 16 # 设置 网络检测时间, 默认42s $ gluster volume set k8s-volume network.ping-timeout 10 # 设置 写缓冲区的大小, 默认1M $ gluster volume set k8s-volume performance.write-behind-window-size 1024MB Kubernetes中配置glusterfs 官方的文档见：https://github.com/kubernetes/kubernetes/tree/master/examples/volumes/glusterfs 以下用到的所有yaml和json配置文件可以在../manifests/glusterfs中找到。注意替换其中私有镜像地址为你自己的镜像地址。 kubernetes安装客户端 # 在所有 k8s node 中安装 glusterfs 客户端 $ yum install -y glusterfs glusterfs-fuse # 配置 hosts $ vi /etc/hosts 172.20.0.113 sz-pg-oam-docker-test-001.tendcloud.com 172.20.0.114 sz-pg-oam-docker-test-002.tendcloud.com 172.20.0.115 sz-pg-oam-docker-test-003.tendcloud.com 因为我们glusterfs跟kubernetes集群复用主机，因为此这一步可以省去。 配置 endpoints $ curl -O https://raw.githubusercontent.com/kubernetes/kubernetes/master/examples/volumes/glusterfs/glusterfs-endpoints.json # 修改 endpoints.json ，配置 glusters 集群节点ip # 每一个 addresses 为一个 ip 组 { \"addresses\": [ { \"ip\": \"172.22.0.113\" } ], \"ports\": [ { \"port\": 1990 } ] }, # 导入 glusterfs-endpoints.json $ kubectl apply -f glusterfs-endpoints.json # 查看 endpoints 信息 $ kubectl get ep 配置 service $ curl -O https://raw.githubusercontent.com/kubernetes/kubernetes/master/examples/volumes/glusterfs/glusterfs-service.json # service.json 里面查找的是 enpointes 的名称与端口，端口默认配置为 1，我改成了1990 # 导入 glusterfs-service.json $ kubectl apply -f glusterfs-service.json # 查看 service 信息 $ kubectl get svc 创建测试 pod $ curl -O https://raw.githubusercontent.com/kubernetes/kubernetes/master/examples/volumes/glusterfs/glusterfs-pod.json # 编辑 glusterfs-pod.json # 修改 volumes 下的 path 为上面创建的 volume 名称 \"path\": \"k8s-volume\" # 导入 glusterfs-pod.json $ kubectl apply -f glusterfs-pod.json # 查看 pods 状态 $ kubectl get pods NAME READY STATUS RESTARTS AGE glusterfs 1/1 Running 0 1m # 查看 pods 所在 node $ kubectl describe pods/glusterfs # 登陆 node 物理机，使用 df 可查看挂载目录 $ df -h 172.20.0.113:k8s-volume 1073741824 0 1073741824 0% 172.20.0.113:k8s-volume 1.0T 0 1.0T 0% /var/lib/kubelet/pods/3de9fc69-30b7-11e7-bfbd-8af1e3a7c5bd/volumes/kubernetes.io~glusterfs/glusterfsvol 配置PersistentVolume PersistentVolume（PV）和 PersistentVolumeClaim（PVC）是kubernetes提供的两种API资源，用于抽象存储细节。管理员关注于如何通过pv提供存储功能而无需关注用户如何使用，同样的用户只需要挂载PVC到容器中而不需要关注存储卷采用何种技术实现。 PVC和PV的关系跟pod和node关系类似，前者消耗后者的资源。PVC可以向PV申请指定大小的存储资源并设置访问模式。 PV属性 storage容量 读写属性：分别为ReadWriteOnce：单个节点读写； ReadOnlyMany：多节点只读 ； ReadWriteMany：多节点读写 $ cat glusterfs-pv.yaml apiVersion: v1 kind: PersistentVolume metadata: name: gluster-dev-volume spec: capacity: storage: 8Gi accessModes: - ReadWriteMany glusterfs: endpoints: \"glusterfs-cluster\" path: \"k8s-volume\" readOnly: false # 导入PV $ kubectl apply -f glusterfs-pv.yaml # 查看 pv $ kubectl get pv NAME CAPACITY ACCESSMODES RECLAIMPOLICY STATUS CLAIM STORAGECLASS REASON AGE gluster-dev-volume 8Gi RWX Retain Available 3s PVC属性 访问属性与PV相同 容量：向PV申请的容量 配置PVC $ cat glusterfs-pvc.yaml kind: PersistentVolumeClaim apiVersion: v1 metadata: name: glusterfs-nginx spec: accessModes: - ReadWriteMany resources: requests: storage: 8Gi # 导入 pvc $ kubectl apply -f glusterfs-pvc.yaml # 查看 pvc $ kubectl get pv NAME STATUS VOLUME CAPACITY ACCESSMODES STORAGECLASS AGE glusterfs-nginx Bound gluster-dev-volume 8Gi RWX 4s 创建 nginx deployment 挂载 volume $ vi nginx-deployment.yaml apiVersion: extensions/v1beta1 kind: Deployment metadata: name: nginx-dm spec: replicas: 2 template: metadata: labels: name: nginx spec: containers: - name: nginx image: nginx:alpine imagePullPolicy: IfNotPresent ports: - containerPort: 80 volumeMounts: - name: gluster-dev-volume mountPath: \"/usr/share/nginx/html\" volumes: - name: gluster-dev-volume persistentVolumeClaim: claimName: glusterfs-nginx # 导入 deployment $ kubectl apply -f nginx-deployment.yaml # 查看 deployment $ kubectl get pods |grep nginx-dm nginx-dm-3698525684-g0mvt 1/1 Running 0 6s nginx-dm-3698525684-hbzq1 1/1 Running 0 6s # 查看 挂载 $ kubectl exec -it nginx-dm-3698525684-g0mvt -- df -h|grep k8s-volume 172.20.0.113:k8s-volume 1.0T 0 1.0T 0% /usr/share/nginx/html # 创建文件 测试 $ kubectl exec -it nginx-dm-3698525684-g0mvt -- touch /usr/share/nginx/html/index.html $ kubectl exec -it nginx-dm-3698525684-g0mvt -- ls -lt /usr/share/nginx/html/index.html -rw-r--r-- 1 root root 0 May 4 11:36 /usr/share/nginx/html/index.html # 验证 glusterfs # 因为我们使用分布卷，所以可以看到某个节点中有文件 [root@sz-pg-oam-docker-test-001 ~] ls /opt/gfs_data/ [root@sz-pg-oam-docker-test-002 ~] ls /opt/gfs_data/ index.html [root@sz-pg-oam-docker-test-003 ~] ls /opt/gfs_data/ 参考 CentOS 7 安装 GlusterFS GlusterFS with kubernetes Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-10-19 18:27:14 "},"practice/storage-for-containers-using-glusterfs-with-openshift.html":{"url":"practice/storage-for-containers-using-glusterfs-with-openshift.html","title":"在OpenShift中使用GlusterFS做持久化存储","keywords":"","body":"在OpenShift中使用GlusterFS做持久化存储 概述 本文由Daniel Messer（Technical Marketing Manager Storage @RedHat）和Keith Tenzer（Solutions Architect @RedHat）共同撰写。 Storage for Containers Overview – Part I Storage for Containers using Gluster – Part II Storage for Containers using Container Native Storage – Part III Storage for Containers using Ceph – Part IV Storage for Containers using NetApp ONTAP NAS – Part V Storage for Containers using NetApp SolidFire – Part VI Gluster作为Container-Ready Storage(CRS) 在本文中，我们将介绍容器存储的首选以及如何部署它。 Kusternet和OpenShift支持GlusterFS已经有一段时间了。 GlusterFS的适用性很好，可用于所有的部署场景：裸机、虚拟机、内部部署和公共云。 在容器中运行GlusterFS的新特性将在本系列后面讨论。 GlusterFS是一个分布式文件系统，内置了原生协议（GlusterFS）和各种其他协议（NFS，SMB，...）。 为了与OpenShift集成，节点将通过FUSE使用原生协议，将GlusterFS卷挂在到节点本身上，然后将它们绑定到目标容器中。 OpenShift / Kubernetes具有实现请求、释放和挂载、卸载GlusterFS卷的原生程序。 CRS概述 在存储方面，根据OpenShift / Kubernetes的要求，还有一个额外的组件管理集群，称为“heketi”。 这实际上是一个用于GlusterFS的REST API，它还提供CLI版本。 在以下步骤中，我们将在3个GlusterFS节点中部署heketi，使用它来部署GlusterFS存储池，将其连接到OpenShift，并使用它来通过PersistentVolumeClaims为容器配置存储。 我们将总共部署4台虚拟机。 一个用于OpenShift（实验室设置），另一个用于GlusterFS。 注意：您的系统应至少需要有四核CPU，16GB RAM和20 GB可用磁盘空间。 部署OpenShift 首先你需要先部署OpenShift。最有效率的方式是直接在虚拟机中部署一个All-in-One环境，部署指南见 the “OpenShift Enterprise 3.4 all-in-one Lab Environment” article.。 确保你的OpenShift虚拟机可以解析外部域名。编辑/etc/dnsmasq.conf文件，增加下面的Google DNS： server=8.8.8.8 重启： # systemctl restart dnsmasq # ping -c1 google.com 部署Gluster GlusterFS至少需要有以下配置的3台虚拟机： RHEL 7.3 2 CPUs 2 GB内存 30 GB磁盘存储给操作系统 10 GB磁盘存储给GlusterFS bricks 修改/etc/hosts文件，定义三台虚拟机的主机名。 例如（主机名可以根据你自己的环境自由调整） # cat /etc/hosts 127.0.0.1 localhost localhost.localdomain localhost4 localhost4.localdomain4 ::1 localhost localhost.localdomain localhost6 localhost6.localdomain6 172.16.99.144 ocp-master.lab ocp-master 172.16.128.7 crs-node1.lab crs-node1 172.16.128.8 crs-node2.lab crs-node2 172.16.128.9 crs-node3.lab crs-node3 在3台GlusterFS虚拟机上都执行以下步骤： # subscription-manager repos --disable=\"*\" # subscription-manager repos --enable=rhel-7-server-rpms 如果你已经订阅了GlusterFS那么可以直接使用，开启rh-gluster-3-for-rhel-7-server-rpms的yum源。 如果你没有的话，那么可以通过EPEL使用非官方支持的GlusterFS的社区源。 # yum -y install http://dl.fedoraproject.org/pub/epel/epel-release-latest-7.noarch.rpm # rpm --import http://dl.fedoraproject.org/pub/epel/RPM-GPG-KEY-EPEL-7 在/etc/yum.repos.d/目录下创建glusterfs-3.10.repo文件： [glusterfs-3.10] name=glusterfs-3.10 description=\"GlusterFS 3.10 Community Version\" baseurl=https://buildlogs.centos.org/centos/7/storage/x86_64/gluster-3.10/ gpgcheck=0 enabled=1 验证源已经被激活。 # yum repolist 现在可以开始安装GlusterFS了。 # yum -y install glusterfs-server 需要为GlusterFS peers打开几个基本TCP端口，以便与OpenShift进行通信并提供存储： # firewall-cmd --add-port=24007-24008/tcp --add-port=49152-49664/tcp --add-port=2222/tcp # firewall-cmd --runtime-to-permanent 现在我们可以启动GlusterFS的daemon进程了： # systemctl enable glusterd # systemctl start glusterd 完成。GlusterFS已经启动并正在运行。其他配置将通过heketi完成。 在GlusterFS的一台虚拟机上安装heketi [root@crs-node1 ~]# yum -y install heketi heketi-client 更新EPEL 如果你没有Red Hat Gluster Storage订阅的话，你可以从EPEL中获取heketi。 在撰写本文时，2016年10月那时候还是3.0.0-1.el7版本，它不适用于OpenShift 3.4。 你将需要更新到更新的版本： [root@crs-node1 ~]# yum -y install wget [root@crs-node1 ~]# wget https://github.com/heketi/heketi/releases/download/v4.0.0/heketi-v4.0.0.linux.amd64.tar.gz [root@crs-node1 ~]# tar -xzf heketi-v4.0.0.linux.amd64.tar.gz [root@crs-node1 ~]# systemctl stop heketi [root@crs-node1 ~]# cp heketi/heketi* /usr/bin/ [root@crs-node1 ~]# chown heketi:heketi /usr/bin/heketi* 在/etc/systemd/system/heketi.service中创建v4版本的heketi二进制文件的更新语法文件： [Unit] Description=Heketi Server [Service] Type=simple WorkingDirectory=/var/lib/heketi EnvironmentFile=-/etc/heketi/heketi.json User=heketi ExecStart=/usr/bin/heketi --config=/etc/heketi/heketi.json Restart=on-failure StandardOutput=syslog StandardError=syslog [Install] WantedBy=multi-user.target [root@crs-node1 ~]# systemctl daemon-reload [root@crs-node1 ~]# systemctl start heketi Heketi使用SSH来配置GlusterFS的所有节点。创建SSH密钥对，将公钥拷贝到所有3个节点上（包括你登陆的第一个节点）： [root@crs-node1 ~]# ssh-keygen -f /etc/heketi/heketi_key -t rsa -N '' [root@crs-node1 ~]# ssh-copy-id -i /etc/heketi/heketi_key.pub root@crs-node1.lab [root@crs-node1 ~]# ssh-copy-id -i /etc/heketi/heketi_key.pub root@crs-node2.lab [root@crs-node1 ~]# ssh-copy-id -i /etc/heketi/heketi_key.pub root@crs-node3.lab [root@crs-node1 ~]# chown heketi:heketi /etc/heketi/heketi_key* 剩下唯一要做的事情就是配置heketi来使用SSH。 编辑/etc/heketi/heketi.json文件使它看起来像下面这个样子（改变的部分突出显示下划线）： { \"_port_comment\":\"Heketi Server Port Number\", \"port\":\"8080\", \"_use_auth\":\"Enable JWT authorization. Please enable for deployment\", \"use_auth\":false, \"_jwt\":\"Private keys for access\", \"jwt\":{ \"_admin\":\"Admin has access to all APIs\", \"admin\":{ \"key\":\"My Secret\" }, \"_user\":\"User only has access to /volumes endpoint\", \"user\":{ \"key\":\"My Secret\" } }, \"_glusterfs_comment\":\"GlusterFS Configuration\", \"glusterfs\":{ \"_executor_comment\":[ \"Execute plugin. Possible choices: mock, ssh\", \"mock: This setting is used for testing and development.\", \" It will not send commands to any node.\", \"ssh: This setting will notify Heketi to ssh to the nodes.\", \" It will need the values in sshexec to be configured.\", \"kubernetes: Communicate with GlusterFS containers over\", \" Kubernetes exec api.\" ], \"executor\":\"ssh\", \"_sshexec_comment\":\"SSH username and private key file information\", \"sshexec\":{ \"keyfile\":\"/etc/heketi/heketi_key\", \"user\":\"root\", \"port\":\"22\", \"fstab\":\"/etc/fstab\" }, \"_kubeexec_comment\":\"Kubernetes configuration\", \"kubeexec\":{ \"host\":\"https://kubernetes.host:8443\", \"cert\":\"/path/to/crt.file\", \"insecure\":false, \"user\":\"kubernetes username\", \"password\":\"password for kubernetes user\", \"namespace\":\"OpenShift project or Kubernetes namespace\", \"fstab\":\"Optional: Specify fstab file on node. Default is /etc/fstab\" }, \"_db_comment\":\"Database file name\", \"db\":\"/var/lib/heketi/heketi.db\", \"_loglevel_comment\":[ \"Set log level. Choices are:\", \" none, critical, error, warning, info, debug\", \"Default is warning\" ], \"loglevel\":\"debug\" } } 完成。heketi将监听8080端口，我们来确认下防火墙规则允许它监听该端口： # firewall-cmd --add-port=8080/tcp # firewall-cmd --runtime-to-permanent 重启heketi： # systemctl enable heketi # systemctl restart heketi 测试它是否在运行： # curl http://crs-node1.lab:8080/hello Hello from Heketi 很好。heketi上场的时候到了。 我们将使用它来配置我们的GlusterFS存储池。 该软件已经在我们所有的虚拟机上运行，但并未被配置。 要将其改造为满足我们需求的存储系统，需要在拓扑文件中描述我们所需的GlusterFS存储池，如下所示： # vi topology.json { \"clusters\": [ { \"nodes\": [ { \"node\": { \"hostnames\": { \"manage\": [ \"crs-node1.lab\" ], \"storage\": [ \"172.16.128.7\" ] }, \"zone\": 1 }, \"devices\": [ \"/dev/sdb\" ] }, { \"node\": { \"hostnames\": { \"manage\": [ \"crs-node2.lab\" ], \"storage\": [ \"172.16.128.8\" ] }, \"zone\": 1 }, \"devices\": [ \"/dev/sdb\" ] }, { \"node\": { \"hostnames\": { \"manage\": [ \"crs-node3.lab\" ], \"storage\": [ \"172.16.128.9\" ] }, \"zone\": 1 }, \"devices\": [ \"/dev/sdb\" ] } ] } ] } 该文件格式比较简单，基本上是告诉heketi要创建一个3节点的集群，其中每个节点包含的配置有FQDN，IP地址以及至少一个将用作GlusterFS块的备用块设备。 现在将该文件发送给heketi： # export HEKETI_CLI_SERVER=http://crs-node1.lab:8080 # heketi-cli topology load --json=topology.json Creating cluster ... ID: 78cdb57aa362f5284bc95b2549bc7e7d Creating node crs-node1.lab ... ID: ffd7671c0083d88aeda9fd1cb40b339b Adding device /dev/sdb ... OK Creating node crs-node2.lab ... ID: 8220975c0a4479792e684584153050a9 Adding device /dev/sdb ... OK Creating node crs-node3.lab ... ID: b94f14c4dbd8850f6ac589ac3b39cc8e Adding device /dev/sdb ... OK 现在heketi已经配置了3个节点的GlusterFS存储池。很简单！你现在可以看到3个虚拟机都已经成功构成了GlusterFS中的可信存储池（Trusted Stroage Pool）。 [root@crs-node1 ~]# gluster peer status Number of Peers: 2 Hostname: crs-node2.lab Uuid: 93b34946-9571-46a8-983c-c9f128557c0e State: Peer in Cluster (Connected) Other names: crs-node2.lab Hostname: 172.16.128.9 Uuid: e3c1f9b0-be97-42e5-beda-f70fc05f47ea State: Peer in Cluster (Connected) 现在回到OpenShift！ 将Gluster与OpenShift集成 为了集成OpenShift，需要两样东西：一个动态的Kubernetes Storage Provisioner和一个StorageClass。 Provisioner在OpenShift中开箱即用。 实际上关键的是如何将存储挂载到容器上。 StorageClass是OpenShift中的用户可以用来实现的PersistentVolumeClaims的实体，它反过来能够触发一个Provisioner实现实际的配置，并将结果表示为Kubernetes PersistentVolume（PV）。 就像OpenShift中的其他组件一样，StorageClass也简单的用YAML文件定义： # cat crs-storageclass.yaml kind: StorageClass apiVersion: storage.k8s.io/v1beta1 metadata: name: container-ready-storage annotations: storageclass.beta.kubernetes.io/is-default-class: \"true\" provisioner: kubernetes.io/glusterfs parameters: resturl: \"http://crs-node1.lab:8080\" restauthenabled: \"false\" 我们的provisioner是kubernetes.io/glusterfs，将它指向我们的heketi实例。 我们将类命名为“container-ready-storage”，同时使其成为所有没有显示指定StorageClass的PersistentVolumeClaim的默认StorageClass。 为你的GlusterFS池创建StorageClass： # oc create -f crs-storageclass.yaml 在OpenShift中使用Gluster 我们来看下如何在OpenShift中使用GlusterFS。首先在OpenShift虚拟机中创建一个测试项目。 # oc new-project crs-storage --display-name=\"Container-Ready Storage\" 这会向Kubernetes/OpenShift发出storage请求，请求一个PersistentVolumeClaim（PVC）。 这是一个简单的对象，它描述最少需要多少容量和应该提供哪种访问模式（非共享，共享，只读）。 它通常是应用程序模板的一部分，但我们只需创建一个独立的PVC： # cat crs-claim.yaml apiVersion: v1 kind: PersistentVolumeClaim metadata: name: my-crs-storage namespace: crs-storage spec: accessModes: - ReadWriteOnce resources: requests: storage: 1Gi 发送该请求： # oc create -f crs-claim.yaml 观察在OpenShfit中，PVC正在以动态创建volume的方式实现： # oc get pvc NAME STATUS VOLUME CAPACITY ACCESSMODES AGE my-crs-storage Bound pvc-41ad5adb-107c-11e7-afae-000c2949cce7 1Gi RWO 58s 太棒了！ 你现在可以在OpenShift中使用存储容量，而不需要直接与存储系统进行任何交互。 我们来看看创建的volume： # oc get pv/pvc-41ad5adb-107c-11e7-afae-000c2949cce7 Name: pvc-41ad5adb-107c-11e7-afae-000c2949cce7 Labels: StorageClass: container-ready-storage Status: Bound Claim: crs-storage/my-crs-storage Reclaim Policy: Delete Access Modes: RWO Capacity: 1Gi Message: Source: Type: Glusterfs (a Glusterfs mount on the host that shares a pod's lifetime) EndpointsName: gluster-dynamic-my-crs-storage Path: vol_85e444ee3bc154de084976a9aef16025 ReadOnly: false What happened in the background was that when the PVC reached the system, our default StorageClass reached out to the GlusterFS Provisioner with the volume specs from the PVC. The provisioner in turn communicates with our heketi instance which facilitates the creation of the GlusterFS volume, which we can trace in it’s log messages: 该volume是根据PVC中的定义特别创建的。 在PVC中，我们没有明确指定要使用哪个StorageClass，因为heketi的GlusterFS StorageClass已经被定义为系统范围的默认值。 在后台发生的情况是，当PVC到达系统时，默认的StorageClass请求具有该PVC中volume声明规格的GlusterFS Provisioner。 Provisioner又与我们的heketi实例通信，这有助于创建GlusterFS volume，我们可以在其日志消息中追踪： [root@crs-node1 ~]# journalctl -l -u heketi.service ... Mar 24 11:25:52 crs-node1.lab heketi[2598]: [heketi] DEBUG 2017/03/24 11:25:52 /src/github.com/heketi/heketi/apps/glusterfs/volume_entry.go:298: Volume to be created on cluster e Mar 24 11:25:52 crs-node1.lab heketi[2598]: [heketi] INFO 2017/03/24 11:25:52 Creating brick 9e791b1daa12af783c9195941fe63103 Mar 24 11:25:52 crs-node1.lab heketi[2598]: [heketi] INFO 2017/03/24 11:25:52 Creating brick 3e06af2f855bef521a95ada91680d14b Mar 24 11:25:52 crs-node1.lab heketi[2598]: [heketi] INFO 2017/03/24 11:25:52 Creating brick e4daa240f1359071e3f7ea22618cfbab ... Mar 24 11:25:52 crs-node1.lab heketi[2598]: [sshexec] INFO 2017/03/24 11:25:52 Creating volume vol_85e444ee3bc154de084976a9aef16025 replica 3 ... Mar 24 11:25:53 crs-node1.lab heketi[2598]: Result: volume create: vol_85e444ee3bc154de084976a9aef16025: success: please start the volume to access data ... Mar 24 11:25:55 crs-node1.lab heketi[2598]: Result: volume start: vol_85e444ee3bc154de084976a9aef16025: success ... Mar 24 11:25:55 crs-node1.lab heketi[2598]: [asynchttp] INFO 2017/03/24 11:25:55 Completed job c3d6c4f9fc74796f4a5262647dc790fe in 3.176522702s ... 成功！ 大约用了3秒钟，GlusterFS池就配置完成了，并配置了一个volume。 默认值是replica 3，这意味着数据将被复制到3个不同节点的3个块上（用GlusterFS作为后端存储）。 该过程是通过Heketi在OpenShift进行编排的。 你也可以从GlusterFS的角度看到有关volume的信息： [root@crs-node1 ~]# gluster volume list vol_85e444ee3bc154de084976a9aef16025 [root@crs-node1 ~]# gluster volume info vol_85e444ee3bc154de084976a9aef16025 Volume Name: vol_85e444ee3bc154de084976a9aef16025 Type: Replicate Volume ID: a32168c8-858e-472a-b145-08c20192082b Status: Started Snapshot Count: 0 Number of Bricks: 1 x 3 = 3 Transport-type: tcp Bricks: Brick1: 172.16.128.8:/var/lib/heketi/mounts/vg_147b43f6f6903be8b23209903b7172ae/brick_9e791b1daa12af783c9195941fe63103/brick Brick2: 172.16.128.9:/var/lib/heketi/mounts/vg_72c0f520b0c57d807be21e9c90312f85/brick_3e06af2f855bef521a95ada91680d14b/brick Brick3: 172.16.128.7:/var/lib/heketi/mounts/vg_67314f879686de975f9b8936ae43c5c5/brick_e4daa240f1359071e3f7ea22618cfbab/brick Options Reconfigured: transport.address-family: inet nfs.disable: on 请注意，GlusterFS中的卷名称如何对应于OpenShift中Kubernetes Persistent Volume的“路径”。 或者，你也可以使用OpenShift UI来配置存储，这样可以很方便地在系统中的所有已知的StorageClasses中进行选择： 图片 - 创建存储 图片 - Screen Shot 2017-03-24 at 11.09.34.png 让我们做点更有趣的事情，在OpenShift中运行工作负载。 在仍运行着crs-storage项目的OpenShift虚拟机中执行： # oc get templates -n openshift 你应该可以看到一个应用程序和数据库模板列表，这个列表将方便你更轻松的使用OpenShift来部署你的应用程序项目。 我们将使用MySQL来演示如何在OpenShift上部署具有持久化和弹性存储的有状态应用程序。 Mysql-persistent模板包含一个用于MySQL数据库目录的1G空间的PVC。 为了演示目的，可以直接使用默认值。 # oc process mysql-persistent -n openshift | oc create -f - 等待部署完成。你可以通过UI或者命令行观察部署进度： # oc get pods NAME READY STATUS RESTARTS AGE mysql-1-h4afb 1/1 Running 0 2m 好了。我们已经使用这个模板创建了一个service，secrets、PVC和pod。我们来使用它（你的pod名字将跟我的不同）： # oc rsh mysql-1-h4afb 你已经成功的将它挂载到MySQL的pod上。我们连接一下数据库试试： sh-4.2$ mysql -u $MYSQL_USER -p$MYSQL_PASSWORD -h $HOSTNAME $MYSQL_DATABASE 这点很方便，所有重要的配置，如MySQL凭据，数据库名称等都是pod模板中的环境变量的一部分，因此可以在pod中作为shell的环境变量。 我们来创建一些数据： mysql> show databases; +--------------------+ | Database | +--------------------+ | information_schema | | sampledb | +--------------------+ 2 rows in set (0.02 sec) mysql> \\u sampledb Database changed mysql> CREATE TABLE IF NOT EXISTS equipment ( -> equip_id int(5) NOT NULL AUTO_INCREMENT, -> type varchar(50) DEFAULT NULL, -> install_date DATE DEFAULT NULL, -> color varchar(20) DEFAULT NULL, -> working bool DEFAULT NULL, -> location varchar(250) DEFAULT NULL, -> PRIMARY KEY(equip_id) -> ); Query OK, 0 rows affected (0.13 sec) mysql> INSERT INTO equipment (type, install_date, color, working, location) -> VALUES -> (\"Slide\", Now(), \"blue\", 1, \"Southwest Corner\"); Query OK, 1 row affected, 1 warning (0.01 sec) mysql> SELECT * FROM equipment; +----------+-------+--------------+-------+---------+------------------+ | equip_id | type | install_date | color | working | location | +----------+-------+--------------+-------+---------+------------------+ | 1 | Slide | 2017-03-24 | blue | 1 | Southwest Corner | +----------+-------+--------------+-------+---------+------------------+ 1 row in set (0.00 sec) 很好，数据库运行正常。 你想看下数据存储在哪里吗？很简单！查看刚使用模板创建的mysql volume： # oc get pvc/mysql NAME STATUS VOLUME CAPACITY ACCESSMODES AGE mysql Bound pvc-a678b583-1082-11e7-afae-000c2949cce7 1Gi RWO 11m # oc describe pv/pvc-a678b583-1082-11e7-afae-000c2949cce7 Name: pvc-a678b583-1082-11e7-afae-000c2949cce7 Labels: StorageClass: container-ready-storage Status: Bound Claim: crs-storage/mysql Reclaim Policy: Delete Access Modes: RWO Capacity: 1Gi Message: Source: Type: Glusterfs (a Glusterfs mount on the host that shares a pod's lifetime) EndpointsName: gluster-dynamic-mysql Path: vol_6299fc74eee513119dafd43f8a438db1 ReadOnly: false GlusterFS的volume名字是vol_6299fc74eee513119dafd43f8a438db1。回到你的GlusterFS虚拟机中，输入： # gluster volume info vol_6299fc74eee513119dafd43f8a438db Volume Name: vol_6299fc74eee513119dafd43f8a438db1 Type: Replicate Volume ID: 4115918f-28f7-4d4a-b3f5-4b9afe5b391f Status: Started Snapshot Count: 0 Number of Bricks: 1 x 3 = 3 Transport-type: tcp Bricks: Brick1: 172.16.128.7:/var/lib/heketi/mounts/vg_67314f879686de975f9b8936ae43c5c5/brick_f264a47aa32be5d595f83477572becf8/brick Brick2: 172.16.128.8:/var/lib/heketi/mounts/vg_147b43f6f6903be8b23209903b7172ae/brick_f5731fe7175cbe6e6567e013c2591343/brick Brick3: 172.16.128.9:/var/lib/heketi/mounts/vg_72c0f520b0c57d807be21e9c90312f85/brick_ac6add804a6a467cd81cd1404841bbf1/brick Options Reconfigured: transport.address-family: inet nfs.disable: on 你可以看到数据是如何被复制到3个GlusterFS块的。我们从中挑一个（最好挑选你刚登陆的那台虚拟机并查看目录）： # ll /var/lib/heketi/mounts/vg_67314f879686de975f9b8936ae43c5c5/brick_f264a47aa32be5d595f83477572becf8/brick total 180300 -rw-r-----. 2 1000070000 2001 56 Mar 24 12:11 auto.cnf -rw-------. 2 1000070000 2001 1676 Mar 24 12:11 ca-key.pem -rw-r--r--. 2 1000070000 2001 1075 Mar 24 12:11 ca.pem -rw-r--r--. 2 1000070000 2001 1079 Mar 24 12:12 client-cert.pem -rw-------. 2 1000070000 2001 1680 Mar 24 12:12 client-key.pem -rw-r-----. 2 1000070000 2001 352 Mar 24 12:12 ib_buffer_pool -rw-r-----. 2 1000070000 2001 12582912 Mar 24 12:20 ibdata1 -rw-r-----. 2 1000070000 2001 79691776 Mar 24 12:20 ib_logfile0 -rw-r-----. 2 1000070000 2001 79691776 Mar 24 12:11 ib_logfile1 -rw-r-----. 2 1000070000 2001 12582912 Mar 24 12:12 ibtmp1 drwxr-s---. 2 1000070000 2001 8192 Mar 24 12:12 mysql -rw-r-----. 2 1000070000 2001 2 Mar 24 12:12 mysql-1-h4afb.pid drwxr-s---. 2 1000070000 2001 8192 Mar 24 12:12 performance_schema -rw-------. 2 1000070000 2001 1676 Mar 24 12:12 private_key.pem -rw-r--r--. 2 1000070000 2001 452 Mar 24 12:12 public_key.pem drwxr-s---. 2 1000070000 2001 62 Mar 24 12:20 sampledb -rw-r--r--. 2 1000070000 2001 1079 Mar 24 12:11 server-cert.pem -rw-------. 2 1000070000 2001 1676 Mar 24 12:11 server-key.pem drwxr-s---. 2 1000070000 2001 8192 Mar 24 12:12 sys 你可以在这里看到MySQL数据库目录。 它使用GlusterFS作为后端存储，并作为绑定挂载给MySQL容器使用。 如果你检查OpenShift VM上的mount表，你将会看到GlusterFS的mount。 总结 在这里我们是在OpenShift之外创建了一个简单但功能强大的GlusterFS存储池。 该池可以独立于应用程序扩展和收缩。 该池的整个生命周期由一个简单的称为heketi的前端管理，你只需要在部署增长时进行手动干预。 对于日常配置操作，使用它的API与OpenShifts动态配置器交互，无需开发人员直接与基础架构团队进行交互。 o这就是我们如何将存储带入DevOps世界 - 无痛苦，并在OpenShift PaaS系统的开发人员工具中直接提供。 GlusterFS和OpenShift可跨越所有环境：裸机，虚拟机，私有和公共云（Azure，Google Cloud，AWS ...），确保应用程序可移植性，并避免云供应商锁定。 祝你愉快在容器中使用GlusterFS！ (c) 2017 Keith Tenzer 原文链接：https://keithtenzer.com/2017/03/24/storage-for-containers-using-gluster-part-ii/ Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-15 21:43:07 "},"practice/cephfs.html":{"url":"practice/cephfs.html","title":"CephFS","keywords":"","body":"CephFS Cephfs 是一个基于 ceph 集群且兼容POSIX标准的文件系统。创建 cephfs 文件系统时需要在 ceph 集群中添加 mds 服务，该服务负责处理 POSIX 文件系统中的 metadata 部分，实际的数据部分交由 ceph 集群中的 OSDs 处理。cephfs 支持以内核模块方式加载也支持 fuse 方式加载。无论是内核模式还是 fuse 模式，都是通过调用 libcephfs 库来实现 cephfs 文件系统的加载，而 libcephfs 库又调用 librados 库与 ceph 集群进行通信，从而实现 cephfs 的加载。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-09-01 17:25:47 "},"practice/using-ceph-for-persistent-storage.html":{"url":"practice/using-ceph-for-persistent-storage.html","title":"使用Ceph做持久化存储","keywords":"","body":"使用Ceph做持久化存储创建MySQL集群 本文中用到的 yaml 文件可以在 ../manifests/mariadb-cluster 目录下找到。 下面我们以部署一个高可用的 MySQL 集群为例，讲解如何使用 Ceph 做数据持久化，其中使用 StorageClass 动态创建 PV，Ceph 集群我们使用 kubernetes 集群外部的已有的集群，我们没有必要重新部署了。 在 1.4 以后，kubernetes 提供了一种更加方便的动态创建 PV 的方式；也就是说使用 StoragaClass 时无需预先创建固定大小的 PV，等待使用者创建 PVC 来使用；而是直接创建 PVC 即可分配使用。 使用 kubernetes 集群外部的 Ceph 存储 在部署 kubernetes 之前我们就已经有了 Ceph 集群，因此我们可以直接拿来用。但是 kubernetes 的所有节点（尤其是 master 节点）上依然需要安装 ceph 客户端。 yum install -y ceph-common 还需要将 ceph 的配置文件 ceph.conf 放在所有节点的 /etc/ceph 目录下。 Kubernetes 使用 ceph 存储需要用到如下配置： Monitors: Ceph montors 列表 Path：作为挂载的根路径，默认是 / User：RADOS用户名，默认是 admin secretFile：keyring 文件路径，默认是 /etc/ceph/user.secret，我们 Ceph 集群提供的文件是 ceph.client.admin.keyring，将在下面用到 secretRef：Ceph 认证 secret 的引用，如果配置了将会覆盖 secretFile。 readOnly：该文件系统是否只读。 Galera Cluster介绍 Galera是一个MySQL(也支持MariaDB，Percona)的同步多主集群软件。 从用户视角看，一组Galera集群可以看作一个具有多入口的MySQL库，用户可以同时从多个IP读写这个库。目前Galera已经得到广泛应用，例如Openstack中，在集群规模不大的情况下，稳定性已经得到了实践考验。真正的multi-master，即所有节点可以同时读写数据库。 详细步骤 以下步骤包括创建 Ceph 的配置 和 MySQL 的配置两部分。 配置 Ceph 关于 Ceph 的 yaml 文件可以在 ../manifest/mariadb-cluster 目录下找到。 1. 生成 Ceph secret 使用 Ceph 管理员提供给你的 ceph.client.admin.keyring 文件，我们将它放在了 /etc/ceph 目录下，用来生成 secret。 grep key /etc/ceph/ceph.client.admin.keyring |awk '{printf \"%s\", $NF}'|base64 将获得加密后的 key：QVFDWDA2aFo5TG5TQnhBQVl1b0lUL2V3YlRSaEtwVEhPWkxvUlE9PQ==，我们将在后面用到。 2. 创建租户namespace 创建 galera-namespace.yaml 文件内容为： apiVersion: v1 kind: Namespace metadata: name: galera 3. 创建 Ceph secret 创建 ceph-secret.yaml 文件内容为： apiVersion: v1 kind: Secret metadata: name: ceph-secret namespace: galera type: \"kubernetes.io/rbd\" data: key: QVFDWDA2aFo5TG5TQnhBQVl1b0lUL2V3YlRSaEtwVEhPWkxvUlE9PQ== 4. 创建 StorageClass 创建 ceph-class.yaml 文件内容为： apiVersion: storage.k8s.io/v1 kind: StorageClass metadata: name: ceph-web provisioner: kubernetes.io/rbd parameters: monitors: 172.28.7.98,172.28.7.99,172.28.7.100 adminId: admin adminSecretName: ceph-secret adminSecretNamespace: galera pool: rbd #此处默认是rbd池，生产上建议自己创建存储池隔离 userId: admin userSecretName: ceph-secret 此配置请参考 kubernetes 官方文档：https://kubernetes.io/docs/concepts/storage/persistent-volumes/#ceph-rbd 配置 MySQL 1. 创建 MySQL 配置文件 创建 mysql-config.yaml 文件内容为： apiVersion: v1 kind: ConfigMap metadata: name: mysql-config-vol namespace: galera labels: app: mysql data: mariadb.cnf: | [client] default-character-set = utf8 [mysqld] character-set-server = utf8 collation-server = utf8_general_ci # InnoDB optimizations innodb_log_file_size = 64M galera.cnf: | [galera] user = mysql bind-address = 0.0.0.0 # Optimizations innodb_flush_log_at_trx_commit = 0 sync_binlog = 0 expire_logs_days = 7 # Required settings default_storage_engine = InnoDB binlog_format = ROW innodb_autoinc_lock_mode = 2 query_cache_size = 0 query_cache_type = 0 # MariaDB Galera settings #wsrep_debug=ON wsrep_on=ON wsrep_provider=/usr/lib/galera/libgalera_smm.so wsrep_sst_method=rsync # Cluster settings (automatically updated) wsrep_cluster_address=gcomm:// wsrep_cluster_name=galera wsrep_node_address=127.0.0.1 2. 创建 MySQL root 用户和密码 创建加密密码 $ echo -n jimmysong|base64 amltbXlzb25n 注意：一定要用-n 去掉换行符，不然会报错。 创建 root 用户 $ echo -n root |base64 cm9vdA== 创建 MySQL secret 创建 mysql-secret.yaml 文件内容为： apiVersion: v1 kind: Secret metadata: name: mysql-secrets namespace: galera labels: app: mysql data: # Root password: changeit run echo -n jimmysong|base64 root-password: amltbXlzb25n # Root user: root root-user: cm9vdA== 3. 创建 yaml 配置文件 创建 MySQL 的 yaml 文件 galera-mariadb.yaml 内容为： apiVersion: v1 kind: Service metadata: annotations: service.alpha.kubernetes.io/tolerate-unready-endpoints: \"true\" name: mysql namespace: galera labels: app: mysql tier: data spec: ports: - port: 3306 name: mysql clusterIP: None selector: app: mysql --- apiVersion: apps/v1beta1 kind: StatefulSet metadata: name: mysql namespace: galera spec: serviceName: \"mysql\" replicas: 3 template: metadata: labels: app: mysql tier: data annotations: pod.beta.kubernetes.io/init-containers: '[ { \"name\": \"galera-init\", \"image\": \"sz-pg-oam-docker-hub-001.tendcloud.com/library/k8s-galera-init:latest\", \"args\": [\"-service=mysql\"], \"env\": [ { \"name\": \"POD_NAMESPACE\", \"valueFrom\": { \"fieldRef\": { \"apiVersion\": \"v1\", \"fieldPath\": \"metadata.namespace\" } } }, { \"name\": \"SAFE_TO_BOOTSTRAP\", \"value\": \"1\" }, { \"name\": \"DEBUG\", \"value\": \"1\" } ], \"volumeMounts\": [ { \"name\": \"config\", \"mountPath\": \"/etc/mysql/conf.d\" }, { \"name\": \"data\", \"mountPath\": \"/var/lib/mysql\" } ] } ]' spec: terminationGracePeriodSeconds: 10 containers: - name: mysql image: sz-pg-oam-docker-hub-001.tendcloud.com/library/mariadb:10.1 imagePullPolicy: IfNotPresent ports: - containerPort: 3306 name: mysql - containerPort: 4444 name: sst - containerPort: 4567 name: replication - containerPort: 4568 name: ist env: - name: MYSQL_ROOT_PASSWORD valueFrom: secretKeyRef: name: mysql-secrets key: root-password - name: MYSQL_ROOT_USER valueFrom: secretKeyRef: name: mysql-secrets key: root-user - name: MYSQL_INITDB_SKIP_TZINFO value: \"yes\" livenessProbe: exec: command: [\"sh\", \"-c\", \"mysql -u\\\"${MYSQL_ROOT_USER:-root}\\\" -p\\\"${MYSQL_ROOT_PASSWORD}\\\" -e 'show databases;'\"] initialDelaySeconds: 60 timeoutSeconds: 5 readinessProbe: exec: command: [\"sh\", \"-c\", \"mysql -u\\\"${MYSQL_ROOT_USER:-root}\\\" -p\\\"${MYSQL_ROOT_PASSWORD}\\\" -e 'show databases;'\"] initialDelaySeconds: 20 timeoutSeconds: 5 volumeMounts: - name: config mountPath: /etc/mysql/conf.d - name: data mountPath: /var/lib/mysql volumes: - name: config configMap: name: mysql-config-vol imagePullSecrets: - name: \"registrykey\" volumeClaimTemplates: - metadata: name: data annotations: volume.beta.kubernetes.io/storage-class: \"ceph-web\" #引用ceph class 的类 spec: accessModes: [ \"ReadWriteOnce\" ] resources: requests: storage: 3Gi 部署 MySQL 集群 在 /etc/mariadb-cluster 目录下执行： kubectl create -f . 验证 存在 issue，参考 Error creating rbd image: executable file not found in $PATH#38923 问题记录 如果没有安装 ceph-common 的话，kubernetes 在创建 PVC 的时候会有如下报错信息： Events: FirstSeen LastSeen Count From SubObjectPath Type Reason Message --------- -------- ----- ---- ------------- -------- ------ ------- 1h 12s 441 {persistentvolume-controller } Warning ProvisioningFailed Failed to provision volume with StorageClass \"ceph-web\": failed to create rbd image: executable file not found in $PATH, command output: 检查 kube-controller-manager 的日志将看到如下错误信息： journalctl -xe -u kube-controller-manager ... rbd_util.go:364] failed to create rbd image, output ... rbd.go:317] rbd: create volume failed, err: failed to create rbd image: executable file not found in $PATH, command output: 这是因为 kube-controller-manager 主机上没有安装 ceph-common 的缘故。 但是安装了 ceph-common 之后依然有问题： Sep 4 15:25:36 bj-xg-oam-kubernetes-001 kube-controller-manager: W0904 15:25:36.032128 13211 rbd_util.go:364] failed to create rbd image, output Sep 4 15:25:36 bj-xg-oam-kubernetes-001 kube-controller-manager: W0904 15:25:36.032201 13211 rbd_util.go:364] failed to create rbd image, output Sep 4 15:25:36 bj-xg-oam-kubernetes-001 kube-controller-manager: W0904 15:25:36.032252 13211 rbd_util.go:364] failed to create rbd image, output Sep 4 15:25:36 bj-xg-oam-kubernetes-001 kube-controller-manager: E0904 15:25:36.032276 13211 rbd.go:317] rbd: create volume failed, err: failed to create rbd image: fork/exec /usr/bin/rbd: invalid argument, command output: 该问题尚未解决，参考 Error creating rbd image: executable file not found in $PATH#38923 从日志记录来看追查到 pkg/volume/rbd/rbd.go 的 func (r *rbdVolumeProvisioner) Provision() (*v1.PersistentVolume, error) { 方法对 ceph-class.yaml 中的参数进行了验证和处理后调用了 pkg/volume/rbd/rdb_utils.go 文件第 344 行 CreateImage 方法（kubernetes v1.6.1版本）： func (util *RBDUtil) CreateImage(p *rbdVolumeProvisioner) (r *v1.RBDVolumeSource, size int, err error) { var output []byte capacity := p.options.PVC.Spec.Resources.Requests[v1.ResourceName(v1.ResourceStorage)] volSizeBytes := capacity.Value() // convert to MB that rbd defaults on sz := int(volume.RoundUpSize(volSizeBytes, 1024*1024)) volSz := fmt.Sprintf(\"%d\", sz) // rbd create l := len(p.rbdMounter.Mon) // pick a mon randomly start := rand.Int() % l // iterate all monitors until create succeeds. for i := start; i 该方法调用失败。 参考 https://github.com/kubernetes/examples/blob/master/staging/volumes/cephfs/README.md k8s-ceph-statefulsets-storageclass-nfs 动态卷有状态应用实践 Kubernetes persistent storage with Ceph https://kubernetes.io/docs/concepts/storage/persistent-volumes/#ceph-rbd Error creating rbd image: executable file not found in $PATH#38923 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-07 10:26:02 "},"practice/monitoring.html":{"url":"practice/monitoring.html","title":"集群与应用监控","keywords":"","body":"监控 Kubernetes 使得管理复杂环境变得更简单，但是对 kubernetes 本身的各种组件还有运行在 kubernetes 集群上的各种应用程序做到很好的洞察就很难了。Kubernetes 本身对应用程序的做了很多抽象，在生产环境下对这些不同的抽象组件的健康就是迫在眉睫的事情。 我们在安装 kubernetes 集群的时候，默认安装了 kubernetes 官方提供的 heapster 插件，可以对 kubernetes 集群上的应用进行简单的监控，获取 pod 级别的内存、CPU和网络监控信息，同时还能够通过 API 监控 kubernetes 中的基本资源监控指标。 然而，Prometheus 的出现让人眼前一亮，与 kubernetes 一样同样为 CNCF 中的项目，而且是第一个加入到 CNCF 中的项目。 Prometheus 是由 SoundCloud 开源监控告警解决方案，从 2012 年开始编写代码，再到 2015 年 GitHub 上开源以来，已经吸引了 9k+ 关注，以及很多大公司的使用；2016 年 Prometheus 成为继 k8s 后，第二名 CNCF(Cloud Native Computing Foundation) 成员。 作为新一代开源解决方案，很多理念与 Google SRE 运维之道不谋而合。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-09 10:53:54 "},"practice/heapster.html":{"url":"practice/heapster.html","title":"Heapster","keywords":"","body":"Heapster Heapster作为kubernetes安装过程中默认安装的一个插件，见安装heapster插件。这对于集群监控十分有用，同时在Horizontal Pod Autoscaling中也用到了，HPA将Heapster作为Resource Metrics API，向其获取metric，做法是在kube-controller-manager 中配置--api-server指向kube-aggregator，也可以使用heapster来实现，通过在启动heapster的时候指定--api-server=true。 Heapster可以收集Node节点上的cAdvisor数据，还可以按照kubernetes的资源类型来集合资源，比如Pod、Namespace域，可以分别获取它们的CPU、内存、网络和磁盘的metric。默认的metric数据聚合时间间隔是1分钟。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-09 10:47:20 "},"practice/using-heapster-to-get-object-metrics.html":{"url":"practice/using-heapster-to-get-object-metrics.html","title":"使用Heapster获取集群和对象的metric数据","keywords":"","body":"使用Heapster获取集群对象的metric数据 Heapster作为kubernetes安装过程中默认安装的一个插件，见安装heapster插件。这对于集群监控十分有用，同时在Horizontal Pod Autoscaling中也用到了，HPA将Heapster作为Resource Metrics API，向其获取metric，做法是在kube-controller-manager 中配置--api-server指向kube-aggregator，也可以使用heapster来实现，通过在启动heapster的时候指定--api-server=true。 Heapster可以收集Node节点上的cAdvisor数据，还可以按照kubernetes的资源类型来集合资源，比如Pod、Namespace域，可以分别获取它们的CPU、内存、网络和磁盘的metric。默认的metric数据聚合时间间隔是1分钟。 架构 下面是Heapster架构图： 图片 - Heapster架构图 Heapser是用Go语言开发Kubernetes集群计算资源使用情况的数据采集工具，编译后可以直接以一个二进制文件运行，通过向heapster传递的参数来指定数据采集行为，这些数据可以选择多种sink方式，例如Graphite、influxDB、OpenTSDB、ElasticSearch、Kafka等。 使用案例 Heapster使用起来很简单，本身就是二进制文件，直接使用命令行启动，也可以放在容器里运行，在作为kubernetes插件运行时，我们是直接放在容器中的，见安装heapster插件。 运行 下面是heapster的启动参数： Flag Description --allowed-users string comma-separated list of allowed users --alsologtostderr log to standard error as well as files --api-server Enable API server for the Metrics API. If set, the Metrics API will be served on --insecure-port (internally) and --secure-port (externally). --authentication-kubeconfig string kubeconfig file pointing at the 'core' kubernetes server with enough rights to create tokenaccessreviews.authentication.k8s.io. --authentication-token-webhook-cache-ttl duration The duration to cache responses from the webhook token authenticator. (default 10s) --authorization-kubeconfig string kubeconfig file pointing at the 'core' kubernetes server with enough rights to create subjectaccessreviews.authorization.k8s.io. --authorization-webhook-cache-authorized-ttl duration The duration to cache 'authorized' responses from the webhook authorizer. (default 10s) --authorization-webhook-cache-unauthorized-ttl duration The duration to cache 'unauthorized' responses from the webhook authorizer. (default 10s) --bind-address ip The IP address on which to listen for the --secure-port port. The associated interface(s) must be reachable by the rest of the cluster, and by CLI/web clients. If blank, all interfaces will be used (0.0.0.0). (default 0.0.0.0) --cert-dir string The directory where the TLS certs are located (by default /var/run/kubernetes). If --tls-cert-file and --tls-private-key-file are provided, this flag will be ignored. (default \"/var/run/kubernetes\") --client-ca-file string If set, any request presenting a client certificate signed by one of the authorities in the client-ca-file is authenticated with an identity corresponding to the CommonName of the client certificate. --contention-profiling Enable contention profiling. Requires --profiling to be set to work. --disable-export Disable exporting metrics in api/v1/metric-export --enable-swagger-ui Enables swagger ui on the apiserver at /swagger-ui --heapster-port int port used by the Heapster-specific APIs (default 8082) --historical-source string which source type to use for the historical API (should be exactly the same as one of the sink URIs), or empty to disable the historical API --label-seperator string seperator used for joining labels (default \",\") --listen-ip string IP to listen on, defaults to all IPs --log-backtrace-at traceLocation when logging hits line file:N, emit a stack trace (default :0) --log-dir string If non-empty, write log files in this directory --log-flush-frequency duration Maximum number of seconds between log flushes (default 5s) --logtostderr log to standard error instead of files (default true) --max-procs int max number of CPUs that can be used simultaneously. Less than 1 for default (number of cores) --metric-resolution duration The resolution at which heapster will retain metrics. (default 1m0s) --profiling Enable profiling via web interface host:port/debug/pprof/ (default true) --requestheader-allowed-names stringSlice List of client certificate common names to allow to provide usernames in headers specified by --requestheader-username-headers. If empty, any client certificate validated by the authorities in --requestheader-client-ca-file is allowed. --requestheader-client-ca-file string Root certificate bundle to use to verify client certificates on incoming requests before trusting usernames in headers specified by --requestheader-username-headers --requestheader-extra-headers-prefix stringSlice List of request header prefixes to inspect. X-Remote-Extra- is suggested. (default [x-remote-extra-]) --requestheader-group-headers stringSlice List of request headers to inspect for groups. X-Remote-Group is suggested. (default [x-remote-group]) --requestheader-username-headers stringSlice List of request headers to inspect for usernames. X-Remote-User is common. (default [x-remote-user]) --secure-port int The port on which to serve HTTPS with authentication and authorization. If 0, don't serve HTTPS at all. (default 6443) --sink *flags.Uris external sink(s) that receive data (default []) --source *flags.Uris source(s) to watch (default []) --stderrthreshold severity logs at or above this threshold go to stderr (default 2) --tls-ca-file string If set, this certificate authority will used for secure access from Admission Controllers. This must be a valid PEM-encoded CA bundle. Altneratively, the certificate authority can be appended to the certificate provided by --tls-cert-file. --tls-cert string file containing TLS certificate --tls-cert-file string File containing the default x509 Certificate for HTTPS. (CA cert, if any, concatenated after server cert). If HTTPS serving is enabled, and --tls-cert-file and --tls-private-key-file are not provided, a self-signed certificate and key are generated for the public address and saved to /var/run/kubernetes. --tls-client-ca string file containing TLS client CA for client cert validation --tls-key string file containing TLS key --tls-private-key-file string File containing the default x509 private key matching --tls-cert-file. --tls-sni-cert-key namedCertKey A pair of x509 certificate and private key file paths, optionally suffixed with a list of domain patterns which are fully qualified domain names, possibly with prefixed wildcard segments. If no domain patterns are provided, the names of the certificate are extracted. Non-wildcard matches trump over wildcard matches, explicit domain patterns trump over extracted names. For multiple key/certificate pairs, use the --tls-sni-cert-key multiple times. Examples: \"example.key,example.crt\" or \"*.foo.com,foo.com:foo.key,foo.crt\". (default []) --v Level log level for V logs --version print version info and exit --vmodule moduleSpec comma-separated list of pattern=N settings for file-filtered logging Version version: v1.4.0 commit: 546ab66f API使用 Heapster提供RESTful API接口，下面以获取spark-cluster namespace的memory usage为例讲解Heapster API的使用。 构造URL地址 https://172.20.0.113:6443/api/v1/proxy/namespaces/kube-system/services/heapster/api/v1/model/namespaces/spark-cluster/metrics/memory/usage?start=2017-10-16T09:14:00Z&end=2017-10-16T09:16:00Z 结果 访问该地址获取的结果是这样的： { \"metrics\": [ { \"timestamp\": \"2017-10-16T09:14:00Z\", \"value\": 322592768 }, { \"timestamp\": \"2017-10-16T09:15:00Z\", \"value\": 322592768 }, { \"timestamp\": \"2017-10-16T09:16:00Z\", \"value\": 322592768 } ], \"latestTimestamp\": \"2017-10-16T09:16:00Z\" } 注意：Heapster中查询的所有值都是以最小单位为单位，比如CPU为1milicore，内存为B。 第一部分：Heapster API地址 https://172.20.0.113:6443/api/v1/proxy/namespaces/kube-system/services/heapster/ 可以使用下面的命令获取: $ kubectl cluster-info Heapster is running at https://172.20.0.113:6443/api/v1/proxy/namespaces/kube-system/services/heapster ... 第二部分：Heapster API参数 /api/v1/model/namespaces/spark-cluster/metrics/memory/usage 表示查询的是spark-cluster namespace中的memory/usage的metrics。 第三部分：时间片 ?start=2017-10-16T09:14:00Z&end=2017-10-16T09:16:00Z 查询参数为时间片：包括start和end。使用RFC-3339时间格式，在Linux系统中可以这样获取： $ date --rfc-3339=\"seconds\" 2017-10-16 17:23:20+08:00 该时间中的空格替换成T，最后的+08:00替换成Z代表时区。可以只指定start时间，end时间自动设置为当前时间。 参考 kubernetes metrics Heapster metric model Heapster storage schema Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-10-16 17:41:04 "},"practice/prometheus.html":{"url":"practice/prometheus.html","title":"Prometheus","keywords":"","body":"Prometheus Prometheus 是由 SoundCloud 开源监控告警解决方案，从 2012 年开始编写代码，再到 2015 年 github 上开源以来，已经吸引了 9k+ 关注，以及很多大公司的使用；2016 年 Prometheus 成为继 k8s 后，第二名 CNCF(Cloud Native Computing Foundation) 成员。 作为新一代开源解决方案，很多理念与 Google SRE 运维之道不谋而合。 主要功能 多维 数据模型（时序由 metric 名字和 k/v 的 labels 构成）。 灵活的查询语句（PromQL）。 无依赖存储，支持 local 和 remote 不同模型。 采用 http 协议，使用 pull 模式，拉取数据，简单易懂。 监控目标，可以采用服务发现或静态配置的方式。 支持多种统计数据模型，图形化友好。 核心组件 Prometheus Server， 主要用于抓取数据和存储时序数据，另外还提供查询和 Alert Rule 配置管理。 client libraries，用于对接 Prometheus Server, 可以查询和上报数据。 push gateway ，用于批量，短期的监控数据的汇总节点，主要用于业务数据汇报等。 各种汇报数据的 exporters ，例如汇报机器数据的 node_exporter, 汇报 MongoDB 信息的 MongoDB exporter 等等。 用于告警通知管理的 alertmanager 。 基础架构 一图胜千言，先来张官方的架构图： 图片 - Prometheus 架构图 从这个架构图，也可以看出 Prometheus 的主要模块包含， Server, Exporters, Pushgateway, PromQL, Alertmanager, WebUI 等。 它大致使用逻辑是这样： Prometheus server 定期从静态配置的 targets 或者服务发现的 targets 拉取数据。 当新拉取的数据大于配置内存缓存区的时候，Prometheus 会将数据持久化到磁盘（如果使用 remote storage 将持久化到云端）。 Prometheus 可以配置 rules，然后定时查询数据，当条件触发的时候，会将 alert 推送到配置的 Alertmanager。 Alertmanager 收到警告的时候，可以根据配置，聚合，去重，降噪，最后发送警告。 可以使用 API， Prometheus Console 或者 Grafana 查询和聚合数据。 注意 Prometheus 的数据是基于时序的 float64 的值，如果你的数据值有更多类型，无法满足。 Prometheus 不适合做审计计费，因为它的数据是按一定时间采集的，关注的更多是系统的运行瞬时状态以及趋势，即使有少量数据没有采集也能容忍，但是审计计费需要记录每个请求，并且数据长期存储，这个和 Prometheus 无法满足，可能需要采用专门的审计系统。 以上介绍来自 https://github.com/songjiayang/prometheus_practice/ Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-11 10:12:28 "},"practice/using-prometheus-to-monitor-kuberentes-cluster.html":{"url":"practice/using-prometheus-to-monitor-kuberentes-cluster.html","title":"使用Prometheus监控kubernetes集群","keywords":"","body":"使用Prometheus监控kubernetes集群 我们使用 Giantswarm 开源的 kubernetes-promethues 来监控 kubernetes 集群，所有的 YAML 文件可以在 ../manifests/prometheus 目录下找到。 需要用到的镜像有： sz-pg-oam-docker-hub-001.tendcloud.com/library/prometheus-alertmanager:v0.7.1 sz-pg-oam-docker-hub-001.tendcloud.com/library/grafana:4.2.0 sz-pg-oam-docker-hub-001.tendcloud.com/library/giantswarm-tiny-tools:latest sz-pg-oam-docker-hub-001.tendcloud.com/library/prom-prometheus:v1.7.0 sz-pg-oam-docker-hub-001.tendcloud.com/library/kube-state-metrics:v1.0.1 sz-pg-oam-docker-hub-001.tendcloud.com/library/dockermuenster-caddy:0.9.3 sz-pg-oam-docker-hub-001.tendcloud.com/library/prom-node-exporter:v0.14.0 同时备份到时速云： index.tenxcloud.com/jimmy/prometheus-alertmanager:v0.7.1 index.tenxcloud.com/jimmy/grafana:4.2.0 index.tenxcloud.com/jimmy/giantswarm-tiny-tools:latest index.tenxcloud.com/jimmy/prom-prometheus:v1.7.0 index.tenxcloud.com/jimmy/kube-state-metrics:v1.0.1 index.tenxcloud.com/jimmy/dockermuenster-caddy:0.9.3 index.tenxcloud.com/jimmy/prom-node-exporter:v0.14.0 注：所有镜像都是从官方镜像仓库下载下。 部署 我将部署时需要用的的配置文件分成了 namespace、serviceaccount、configmaps、clusterrolebinding 和最后的部署 prometheus、grafana 的过程。 ## 创建 monitoring namespaece kubectl create -f prometheus-monitoring-ns.yaml ## 创建 serviceaccount kubectl create -f prometheus-monitoring-serviceaccount.yaml ## 创建 configmaps kubectl create -f prometheus-configmaps.yaml ## 创建 clusterrolebinding kubectl create clusterrolebinding kube-state-metrics --clusterrole=cluster-admin --serviceaccount=monitoring:kube-state-metrics kubectl create clusterrolebinding prometheus --clusterrole=cluster-admin --serviceaccount=monitoring:prometheus ## 部署 Prometheus kubectl create -f prometheus-monitoring.yaml 访问 kubernetes 任何一个 node 上的 Grafana service 的 nodeport： 图片 - Grafana页面 该图中的数据显示明显有问题，还需要修正。 prometheus-monitoring.yaml 文件中有一个 Job 就是用来导入 grafana dashboard 配置信息的，如果该 Job 执行失败，可以单独在在 monitoring 的 namespace 中启动一个容器，将 manifests/prometheus 目录下的 json 文件复制到容器中，然后进入容器 json 文件的目录下执行： for file in *-datasource.json ; do if [ -e \"$file\" ] ; then echo \"importing $file\" && curl --silent --fail --show-error \\ --request POST http://admin:admin@grafana:3000/api/datasources \\ --header \"Content-Type: application/json\" \\ --data-binary \"@$file\" ; echo \"\" ; fi done ; for file in *-dashboard.json ; do if [ -e \"$file\" ] ; then echo \"importing $file\" && ( echo '{\"dashboard\":'; \\ cat \"$file\"; \\ echo ',\"overwrite\":true,\"inputs\":[{\"name\":\"DS_PROMETHEUS\",\"type\":\"datasource\",\"pluginId\":\"prometheus\",\"value\":\"prometheus\"}]}' ) \\ | jq -c '.' \\ | curl --silent --fail --show-error \\ --request POST http://admin:admin@grafana:3000/api/dashboards/import \\ --header \"Content-Type: application/json\" \\ --data-binary \"@-\" ; echo \"\" ; fi done 这样也可以向 grafana 中导入 dashboard。 存在的问题 该项目的代码中存在几个问题。 1. RBAC 角色授权问题 需要用到两个 clusterrolebinding： kube-state-metrics，对应的serviceaccount是kube-state-metrics prometheus，对应的 serviceaccount是 prometheus-k8s 在部署 Prometheus 之前应该先创建 serviceaccount、clusterrole、clusterrolebinding 等对象，否则在安装过程中可能因为权限问题而导致各种错误，所以这些配置应该写在一个单独的文件中，而不应该跟其他部署写在一起，即使要写在一个文件中，也应该写在文件的最前面，因为使用 kubectl 部署的时候，kubectl 不会判断 YAML 文件中的资源依赖关系，只是简单的从头部开始执行部署，因此写在文件前面的对象会先部署。 解决方法 也可以绕过复杂的 RBAC 设置，直接使用下面的命令将对应的 serviceaccount 设置成 admin 权限，如下： kubectl create clusterrolebinding kube-state-metrics --clusterrole=cluster-admin --serviceaccount=monitoring:kube-state-metrics kubectl create clusterrolebinding prometheus --clusterrole=cluster-admin --serviceaccount=monitoring:prometheus 参考 RBAC——基于角色的访问控制 2. API 兼容问题 从 kube-state-metrics 日志中可以看出用户 kube-state-metrics 没有权限访问如下资源类型： *v1.Job *v1.PersistentVolumeClaim *v1beta1.StatefulSet *v2alpha1.CronJob 而在我们使用的 kubernetes 1.6.0 版本的集群中 API 路径跟 kube-state-metrics 中不同，无法 list 以上三种资源对象的资源。详情见：https://github.com/giantswarm/kubernetes-prometheus/issues/77 3. Job 中的权限认证问题 在 grafana-import-dashboards 这个 job 中有个 init-containers 其中指定的 command 执行错误，应该使用 curl -sX GET -H \"Authorization:bearer `cat /var/run/secrets/kubernetes.io/serviceaccount/token`\" -k https://kubernetes.default/api/v1/namespaces/monitoring/endpoints/grafana 不需要指定 csr 文件，只需要 token 即可。 参考 wait-for-endpoints init-containers fails to load with k8s 1.6.0 #56 参考 Kubernetes Setup for Prometheus and Grafana RBAC——基于角色的访问控制 wait-for-endpoints init-containers fails to load with k8s 1.6.0 #56 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-10-11 23:43:48 "},"practice/services-management-tool.html":{"url":"practice/services-management-tool.html","title":"服务编排管理","keywords":"","body":"服务编排管理 Kubernetes虽然提供了多种容器编排对象，例如Deployment、StatefulSet、DeamonSet、Job等，还有多种基础资源封装例如ConfigMap、Secret、Serivce等，但是一个应用往往有多个服务，有的可能还要依赖持久化存储，当这些服务之间直接互相依赖，需要有一定的组合的情况下，使用YAML文件的方式配置应用往往十分繁琐还容易出错，这时候就需要服务编排工具。 服务编排管理工具就是构建在kubernetes的基础object之上，统筹各个服务之间的关系和依赖的。目前常用到的工具是 Helm。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-10-19 15:29:43 "},"practice/helm.html":{"url":"practice/helm.html","title":"使用Helm管理kubernetes应用","keywords":"","body":"使用Helm管理kubernetes应用 读完本文后您应该可以自己创建chart，并创建自己的私有chart仓库。 Helm是一个kubernetes应用的包管理工具，用来管理charts——预先配置好的安装包资源，有点类似于Ubuntu的APT和CentOS中的yum。 Helm chart是用来封装kubernetes原生应用程序的yaml文件，可以在你部署应用的时候自定义应用程序的一些metadata，便与应用程序的分发。 Helm和charts的主要作用： 应用程序封装 版本管理 依赖检查 便于应用程序分发 安装Helm 前提要求 Kubernetes1.5以上版本 集群可访问到的镜像仓库 执行helm命令的主机可以访问到kubernetes集群 安装步骤 首先需要安装helm客户端 curl https://raw.githubusercontent.com/kubernetes/helm/master/scripts/get > get_helm.sh chmod 700 get_helm.sh ./get_helm.sh 创建tiller的serviceaccount和clusterrolebinding kubectl create serviceaccount --namespace kube-system tiller kubectl create clusterrolebinding tiller-cluster-rule --clusterrole=cluster-admin --serviceaccount=kube-system:tiller 然后安装helm服务端tiller helm init -i sz-pg-oam-docker-hub-001.tendcloud.com/library/kubernetes-helm-tiller:v2.3.1 我们使用-i指定自己的镜像，因为官方的镜像因为某些原因无法拉取。 为应用程序设置serviceAccount： kubectl patch deploy --namespace kube-system tiller-deploy -p '{\"spec\":{\"template\":{\"spec\":{\"serviceAccount\":\"tiller\"}}}}' 检查是否安装成功： $ kubectl -n kube-system get pods|grep tiller tiller-deploy-2372561459-f6p0z 1/1 Running 0 1h $ helm version Client: &version.Version{SemVer:\"v2.3.1\", GitCommit:\"32562a3040bb5ca690339b9840b6f60f8ce25da4\", GitTreeState:\"clean\"} Server: &version.Version{SemVer:\"v2.3.1\", GitCommit:\"32562a3040bb5ca690339b9840b6f60f8ce25da4\", GitTreeState:\"clean\"} 创建自己的chart 我们创建一个名为mychart的chart，看一看chart的文件结构。 $ helm create mongodb $ tree mongodb mongodb ├── Chart.yaml #Chart本身的版本和配置信息 ├── charts #依赖的chart ├── templates #配置模板目录 │ ├── NOTES.txt #helm提示信息 │ ├── _helpers.tpl #用于修改kubernetes objcet配置的模板 │ ├── deployment.yaml #kubernetes Deployment object │ └── service.yaml #kubernetes Serivce └── values.yaml #kubernetes object configuration 2 directories, 6 files 模板 Templates目录下是yaml文件的模板，遵循Go template语法。使用过Hugo的静态网站生成工具的人应该对此很熟悉。 我们查看下deployment.yaml文件的内容。 apiVersion: extensions/v1beta1 kind: Deployment metadata: name: {{ template \"fullname\" . }} labels: chart: \"{{ .Chart.Name }}-{{ .Chart.Version | replace \"+\" \"_\" }}\" spec: replicas: {{ .Values.replicaCount }} template: metadata: labels: app: {{ template \"fullname\" . }} spec: containers: - name: {{ .Chart.Name }} image: \"{{ .Values.image.repository }}:{{ .Values.image.tag }}\" imagePullPolicy: {{ .Values.image.pullPolicy }} ports: - containerPort: {{ .Values.service.internalPort }} livenessProbe: httpGet: path: / port: {{ .Values.service.internalPort }} readinessProbe: httpGet: path: / port: {{ .Values.service.internalPort }} resources: {{ toYaml .Values.resources | indent 12 }} 这是该应用的Deployment的yaml配置文件，其中的双大括号包扩起来的部分是Go template，其中的Values是在values.yaml文件中定义的： # Default values for mychart. # This is a YAML-formatted file. # Declare variables to be passed into your templates. replicaCount: 1 image: repository: nginx tag: stable pullPolicy: IfNotPresent service: name: nginx type: ClusterIP externalPort: 80 internalPort: 80 resources: limits: cpu: 100m memory: 128Mi requests: cpu: 100m memory: 128Mi 比如在Deployment.yaml中定义的容器镜像image: \"{{ .Values.image.repository }}:{{ .Values.image.tag }}\"其中的： .Values.image.repository就是nginx .Values.image.tag就是stable 以上两个变量值是在create chart的时候自动生成的默认值。 我们将默认的镜像地址和tag改成我们自己的镜像sz-pg-oam-docker-hub-001.tendcloud.com/library/nginx:1.9。 检查配置和模板是否有效 当使用kubernetes部署应用的时候实际上讲templates渲染成最终的kubernetes能够识别的yaml格式。 使用helm install --dry-run --debug 命令来验证chart配置。该输出中包含了模板的变量配置与最终渲染的yaml文件。 $ helm install --dry-run --debug mychart Created tunnel using local port: '58406' SERVER: \"localhost:58406\" CHART PATH: /Users/jimmy/Workspace/github/bitnami/charts/incubator/mean/charts/mychart NAME: filled-seahorse REVISION: 1 RELEASED: Tue Oct 24 18:57:13 2017 CHART: mychart-0.1.0 USER-SUPPLIED VALUES: {} COMPUTED VALUES: image: pullPolicy: IfNotPresent repository: sz-pg-oam-docker-hub-001.tendcloud.com/library/nginx tag: 1.9 replicaCount: 1 resources: limits: cpu: 100m memory: 128Mi requests: cpu: 100m memory: 128Mi service: externalPort: 80 internalPort: 80 name: nginx type: ClusterIP HOOKS: MANIFEST: --- # Source: mychart/templates/service.yaml apiVersion: v1 kind: Service metadata: name: filled-seahorse-mychart labels: chart: \"mychart-0.1.0\" spec: type: ClusterIP ports: - port: 80 targetPort: 80 protocol: TCP name: nginx selector: app: filled-seahorse-mychart --- # Source: mychart/templates/deployment.yaml apiVersion: extensions/v1beta1 kind: Deployment metadata: name: filled-seahorse-mychart labels: chart: \"mychart-0.1.0\" spec: replicas: 1 template: metadata: labels: app: filled-seahorse-mychart spec: containers: - name: mychart image: \"sz-pg-oam-docker-hub-001.tendcloud.com/library/nginx:1.9\" imagePullPolicy: IfNotPresent ports: - containerPort: 80 livenessProbe: httpGet: path: / port: 80 readinessProbe: httpGet: path: / port: 80 resources: limits: cpu: 100m memory: 128Mi requests: cpu: 100m memory: 128Mi 我们可以看到Deployment和Service的名字前半截由两个随机的单词组成，最后才是我们在values.yaml中配置的值。 部署到kubernetes 在mychart目录下执行下面的命令将nginx部署到kubernetes集群上。 helm install . NAME: eating-hound LAST DEPLOYED: Wed Oct 25 14:58:15 2017 NAMESPACE: default STATUS: DEPLOYED RESOURCES: ==> v1/Service NAME CLUSTER-IP EXTERNAL-IP PORT(S) AGE eating-hound-mychart 10.254.135.68 80/TCP 0s ==> extensions/v1beta1/Deployment NAME DESIRED CURRENT UP-TO-DATE AVAILABLE AGE eating-hound-mychart 1 1 1 0 0s NOTES: 1. Get the application URL by running these commands: export POD_NAME=$(kubectl get pods --namespace default -l \"app=eating-hound-mychart\" -o jsonpath=\"{.items[0].metadata.name}\") echo \"Visit http://127.0.0.1:8080 to use your application\" kubectl port-forward $POD_NAME 8080:80 现在nginx已经部署到kubernetes集群上，本地执行提示中的命令在本地主机上访问到nginx实例。 export POD_NAME=$(kubectl get pods --namespace default -l \"app=eating-hound-mychart\" -o jsonpath=\"{.items[0].metadata.name}\") echo \"Visit http://127.0.0.1:8080 to use your application\" kubectl port-forward $POD_NAME 8080:80 在本地访问http://127.0.0.1:8080即可访问到nginx。 查看部署的relaese $ helm list NAME REVISION UPDATED STATUS CHART NAMESPACE eating-hound 1 Wed Oct 25 14:58:15 2017 DEPLOYED mychart-0.1.0 default 删除部署的release $ helm delete eating-hound release \"eating-hound\" deleted 打包分享 我们可以修改Chart.yaml中的helm chart配置信息，然后使用下列命令将chart打包成一个压缩文件。 helm package . 打包出mychart-0.1.0.tgz文件。 依赖 我们可以在requirement.yaml中定义应用所依赖的chart，例如定义对mariadb的依赖： dependencies: - name: mariadb version: 0.6.0 repository: https://kubernetes-charts.storage.googleapis.com 使用helm lint .命令可以检查依赖和模板配置是否正确。 安装源 我们在前面安装chart可以通过HTTP server的方式提供。 $ helm serve Regenerating index. This may take a moment. Now serving you on 127.0.0.1:8879 访问http://localhost:8879可以看到刚刚安装的chart。 图片 - Helm chart源 点击链接即可以下载chart的压缩包。 部署MEAN测试案例 MEAN是用来构建网站和web应用的免费开源的JavaScript软件栈，该软件栈包括MongoDB、Express.js、Angular和Node.js。 下载charts $ git clone https://github.com/bitnami/charts.git $ cd charts/incubator/mean $ helm dep list NAME VERSION REPOSITORY STATUS mongodb 0.4.x https://kubernetes-charts.storage.googleapis.com/ missing 缺少mongodb的依赖，需要更新一下chart。 注：https://kubernetes-charts.storage.googleapis.com/是Google维护的chart库，访问该地址可以看到所有的chart列表。 $ helm dep update Hang tight while we grab the latest from your chart repositories... ...Unable to get an update from the \"local\" chart repository (http://127.0.0.1:8879/charts): Get http://127.0.0.1:8879/charts/index.yaml: dial tcp 127.0.0.1:8879: getsockopt: connection refused ...Successfully got an update from the \"stable\" chart repository Update Complete. ⎈Happy Helming!⎈ Saving 1 charts Downloading mongodb from repo https://kubernetes-charts.storage.googleapis.com/ 所有的image都在 values.yaml 文件中配置。 下载缺失的chart。 $ helm dep build Hang tight while we grab the latest from your chart repositories... ...Unable to get an update from the \"local\" chart repository (http://127.0.0.1:8879/charts): Get http://127.0.0.1:8879/charts/index.yaml: dial tcp 127.0.0.1:8879: getsockopt: connection refused ...Successfully got an update from the \"stable\" chart repository Update Complete. ⎈Happy Helming!⎈ Saving 1 charts Downloading mongodb from repo https://kubernetes-charts.storage.googleapis.com/ 修改mongodb chart配置 将刚才下载的charts/mongodb-0.4.17.tgz给解压后，修改其中的配置： 将persistence下的enabled设置为false 将image修改为我们的私有镜像：sz-pg-oam-docker-hub-001.tendcloud.com/library/bitnami-mongodb:3.4.9-r1 执行helm install --dry-run --debug .确定模板无误。 将修改后的mongodb chart打包，在mongodb的目录下执行： helm package . 现在再访问前面启动的helm server http://localhost:8879将可以在页面上看到mongodb-0.4.17这个chart。 我们对官方chart配置做了如下修改后推送到了自己的chart仓库： requirements.yaml和requirements.lock文件中的repository为http://localhost:8879 将values.yaml中的storageClass设置为null 将values.yaml中的Image都改为私有镜像 repositroy都设置为http://localhost:8879 注：因为我们没有使用PVC所以将所有的关于持久化存储的配置都设置为false了。 部署MEAN 在mean目录下执行： helm install . NAME: orbiting-platypus LAST DEPLOYED: Wed Oct 25 16:21:48 2017 NAMESPACE: default STATUS: DEPLOYED RESOURCES: ==> v1/Secret NAME TYPE DATA AGE orbiting-platypus-mongodb Opaque 2 2s ==> v1/ConfigMap NAME DATA AGE orbiting-platypus-mean 1 2s ==> v1/Service NAME CLUSTER-IP EXTERNAL-IP PORT(S) AGE orbiting-platypus-mongodb 10.254.144.208 27017/TCP 2s orbiting-platypus-mean 10.254.165.23 80/TCP 2s ==> extensions/v1beta1/Deployment NAME DESIRED CURRENT UP-TO-DATE AVAILABLE AGE orbiting-platypus-mean 1 1 1 0 2s orbiting-platypus-mongodb 1 1 1 0 2s NOTES: Get the URL of your Node app by running: export POD_NAME=$(kubectl get pods --namespace default -l \"app=orbiting-platypus-mean\" -o jsonpath=\"{.items[0].metadata.name}\") echo http://127.0.0.1:8080/ kubectl port-forward $POD_NAME 8080:80 这样MEAN软件栈就部署到你的kuberentes集群里面了（默认是在default namespace下）。 验证检查 为了验证MEAN是否安装成功过，可以使用kubectl get pods查看pod是否启动完成，会先启动mongodb的pod，然后启动MEAN中的4步init。 访问Web UI 在Ingress中增加如下配置： - host: mean.jimmysong.io http: paths: - backend: serviceName: orbiting-platypus-mean servicePort: 80 path: / 然后在页面中更新ingress: kubectl repalce -f ingress.yaml 关于Ingress配置请参考：边缘节点配置 然后在本地的/etc/hosts文件中增加一条配置： 172.20.0.119 mean.jimmysong.io 注：172.20.0.119即边缘节点的VIP。 因为该页面需要加载google的angularjs、还有两个css在国内无法访问，可以使用curl测试： curl mean.jimmysong.io 将会返回HTML内容： Node/Angular Todo App html { overflow-y: scroll; } body { padding-top: 50px; } #todo-list { margin-bottom: 30px; } #todo-form { margin-bottom: 50px; } I'm a Todo-aholic {{ todos.length }} {{ todo.text }} Add A demo by Scotch. Read the tutorial. 访问 http://mean.jimmysong.io 可以看到如下界面，我在其中添加几条todo： 图片 - TODO应用的Web页面 注：Todo中的文字来自What does the fox say？ 测试完成后可以使用下面的命令将mean chart推送的本地chart仓库中。 在mean目录下执行： helm package . 再次刷新http://localhost:8879将可以看到如下三个chart： mean mean-0.1.3 mongodb mongodb-0.4.17 mychart mychart-0.1.0 参考 Deploy, Scale And Upgrade An Application On Kubernetes With Helm Helm charts Go template Helm docs How To Create Your First Helm Chart Speed deployment on Kubernetes with Helm Chart – Quick YAML example from scratch Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-01 23:39:54 "},"practice/create-private-charts-repo.html":{"url":"practice/create-private-charts-repo.html","title":"构建私有Chart仓库","keywords":"","body":"构建私有Chart仓库 使用Chart便于封装和管理kubernetes中的应用，因此当企业内部的应用多了以后，互相依赖、部署环境复杂之后，原先的直接使用yaml文件的管理方式已经不再适应生产的需要，因此我们有必要构建自己的chart仓库。本文中我们将使用 GitHub Pages 来构建我们自己的 chart 仓库。 目的 我们需要构建一个GitHub pages存储所有chart的压缩文件，最好还要有一个前端来展示和搜索chart。 什么是Chart Chart是helm管理的应用的打包格式。它包括如下特征： Chart中包括一系列的yaml格式的描述文件。 一个Chart只用来部署单个的应用的，不应该过于复杂，不应该包含多个依赖，相当于一个微服务。 Chart有特定的目录结构，可以打包起来进行版本控制。 Chart的组成结构 我们以nginx的chart为例，讲解chart的组成结构。 nginx/ Chart.yaml # 必须：一个包含chart的名称、版本和启用条件信息等的YAML文件 LICENSE # 可选: chart的许可证 README.md # 可选: 使用说明 requirements.yaml # 可选: 该chart的依赖配置 values.yaml # 必须：该chart的默认配置值 charts/ # 可选: 包含该chart依赖的chart templates/ # 可选：kubernetes manifest文件模板，用于生成kubernetes yaml文件 templates/NOTES.txt # 可选: 该chart的使用说明和提示信息文本文件，作为helm install后的提示信息 Chart的安装方式 安装chart主要分为安装本地定义的chart和远程chart仓库中的chart两种方式。 安装本地chart 指定本地chart目录：helm install . 指定本地chart压缩包：helm install nginx-1.2.3.tgz 安装chart仓库中的chart 使用默认的远程仓库：helm install stable/nginx 使用指定的仓库：helm install localhost:8879/nginx-1.2.3.tgz 实际上可以将chart打包后作为静态文件托管到web服务器上，例如GitHub pages作为chart仓库也可以。 依赖管理 有两种方式来管理chart的依赖。 直接在本的chart的charts目录下定义 通过在requirements.yaml文件中定义依赖的chart 在每个chart的charts目录下可以定义依赖的子chart。子chart有如下特点： 无法访问父chart中的配置 父chart可以覆盖子chart中的配置 Chart仓库 Chart 仓库（repository）是一个用来托管index.yaml文件和打包好的chart文件的web服务器。当前chart仓库本身没有设置身份和权限验证，查看此链接获取该问题的最新进展。 因为chart仓库只是一个HTTP服务，通过HTTP GET获取YAML文件和chart的压缩包，所以可以将这些文件存储在web服务器中，例如GCS、Amazon S3、GitHub Pages等。 关于chart仓库的更多信息请参考Helm chart文档。 使用GitHub pages托管charts 我们在上文中说到，chart可以使用GitHub pages做存储，接下来我们将会把之前够够构建的chart上传到GitHub pages并在helm中新增一个repo。 构建Monocular UI 参考 Monocular UI 构建UI。 克隆项目到本地 git clone https://github.com/kubernetes-helm/monocular.git 依赖环境 Angular 2 angular/cli Typescript Sass Webpack Bootstrap 在monoclar/scr/ui目录下执行以下命令安装依赖： yarn install npm install -g @angular/cli npm install -g typescript npm install -g webpack 运行 使用docker-compose 最简单的运行方式使用docker-compose： docker-comopse up 该命令需要用到如下镜像： bitnami/mongodb:3 bitnami/node:8 quay.io/deis/go-dev:v1.5.0 会有一个很长的build过程，构建失败。 使用helm 首先需要已在本地安装了helm，并在kubernetes集群中安装了tiller，见使用helm管理kubernetes应用。 # 需要安装nginx ingress $ helm install stable/nginx-ingress $ helm repo add monocular https://kubernetes-helm.github.io/monocular $ helm install monocular/monocular 图片 - Helm monocular界面 因为nginx ingress配置问题，官方的chart中api与ui使用的是同样的domain name，我使用的是traefik ingress，api访问不到，所以加载不了chart。 参考 Monocular UI Helm Chart - GitHub 简化Kubernetes应用部署工具-Helm之应用部署 Speed deployment on Kubernetes with Helm Chart – Quick YAML example from scratch Using a private github repo as helm chart repo (https access) Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-15 18:47:40 "},"practice/ci-cd.html":{"url":"practice/ci-cd.html","title":"持续集成与发布","keywords":"","body":"持续集成与发布 持续集成与发布，简称CI/CD，是微服务构建的重要环节，也是DevOps中推崇的方法论。如何在kubernetes中使用持续构建与发布工具？可以既可以与企业内部原有的持续构建集成，例如Jenkins，也可以在kubernetes中部署一套新的持续构建与发布工具，例如Drone。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-10-27 11:40:14 "},"practice/jenkins-ci-cd.html":{"url":"practice/jenkins-ci-cd.html","title":"使用Jenkins进行持续集成与发布","keywords":"","body":"使用Jenkins进行持续集成与发布 我们基于Jenkins的CI/CD流程如下所示。 图片 - 基于Jenkins的持续集成与发布 流程说明 应用构建和发布流程说明。 用户向Gitlab提交代码，代码中必须包含Dockerfile 将代码提交到远程仓库 用户在发布应用时需要填写git仓库地址和分支、服务类型、服务名称、资源数量、实例个数，确定后触发Jenkins自动构建 Jenkins的CI流水线自动编译代码并打包成docker镜像推送到Harbor镜像仓库 Jenkins的CI流水线中包括了自定义脚本，根据我们已准备好的kubernetes的YAML模板，将其中的变量替换成用户输入的选项 生成应用的kubernetes YAML配置文件 更新Ingress的配置，根据新部署的应用的名称，在ingress的配置文件中增加一条路由信息 更新PowerDNS，向其中插入一条DNS记录，IP地址是边缘节点的IP地址。关于边缘节点，请查看边缘节点配置 Jenkins调用kubernetes的API，部署应用 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-08-21 18:23:35 "},"practice/drone-ci-cd.html":{"url":"practice/drone-ci-cd.html","title":"使用Drone进行持续集成与发布","keywords":"","body":"使用Drone进行持续构建与发布 Drone是一个用Go语言开发的基于容器运行的持续集成软件。 配置GitHub 使用Drone对GitHub上的代码进行持续构建与发布，需要首先在GitHub上设置一个OAuth，如下： 1. 在Github上创建一个新的OAtuh应用 访问https://github.com/settings/applications/new，创建新的OAuth应用。 图片 - OAuth注册 填写应用程序的地址，因为是在本地与行，所以我们都填http://localhost。 2. 获取OAtuh Client ID和Client Secret 在注册完成后就可以获得如下图所示的OAuth Client ID和Client Secret，保存下来，我们后面要用到。 图片 - OAuth key 使用docker-compose单机运行 我们在本地环境，使用docker-compose，按照Drone官方安装文档安装配置Drone。 我们将代码托管在Github上，需要Drone可以持续集成和发布Github的代码，因此需要修改docker-compoe.yaml文件中的GitHub配置。 version: '2' services: drone-server: image: drone/drone:0.7 ports: - 80:8000 volumes: - /var/lib/drone:/var/lib/drone/ restart: always environment: - DRONE_OPEN=true - DRONE_ADMIN=jimmy - DRONE_HOST=localhost - DRONE_GITHUB=true - DRONE_GITHUB_CLIENT=${DRONE_GITHUB_CLIENT} - DRONE_GITHUB_SECRET=${DRONE_GITHUB_SECRET} - DRONE_SECRET=admin drone-agent: image: drone/drone:0.7 command: agent restart: always depends_on: - drone-server volumes: - /var/run/docker.sock:/var/run/docker.sock environment: - DRONE_SERVER=ws://drone-server:8000/ws/broker - DRONE_SECRET=admin /var/lib/drone是在本地挂载的目录，请确保该目录已存在，且可以被docker访问到，Mac下可以在docker的共享目录中配置。 DRONE_SECRET可以是一个随机的字符串，要确保drone-server与drone-client的DRONE_SECRET相同。 DRONE_GITHUB_CLIENT和DRONE_GITHUB_SECRET即在前面申请的OAuth的Client ID和Client Secret。 启动Drone 使用下面的命令在本地启动drone： docker-compose up 这样是在前台启动，加上-d参数就可以在后台启动。 访问http://localhost可以看到登陆画面。 图片 - Drone登陆界面 授权后可以看到GitHub repo设置。 图片 - Github启用repo设置 图片 - Github单个repo设置 参考 Drone Installation Github - Drone Drone 搭配 Kubernetes 升級應用程式版本 - blog.wu-boy.com Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-03 19:08:34 "},"practice/update-and-upgrade.html":{"url":"practice/update-and-upgrade.html","title":"更新与升级","keywords":"","body":"更新与升级 Kubernetes到目前为止基本保持三个月发行一个新版本的节奏，更新节奏可以说非常快，这一部分将主要跟踪kubernetes及其相关组件的更新与升级。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-02 16:29:33 "},"practice/manually-upgrade.html":{"url":"practice/manually-upgrade.html","title":"手动升级kubernetes集群","keywords":"","body":"手动升级kubernetes集群 在我最开始写作本书的时候，kubernetes刚发布1.6.0版本，而kubernetes基本按照每三个月发布一个大版本的速度迭代，为了使用新特性和只支持新版本kubernetes的配套软件，升级kubernetes就迫在眉睫，在此我们使用替换kubernets的旧的二进制文件这种暴力的方式来升级测试集群，若升级生产集群还望三思。 另外，自kubernetes1.6版本之后发布的1.7和1.8版本又增加了一些新特性，参考： Kubernetes1.7更新日志 Kubernetes1.8更新日志 目前kubernetes的官方文档上并没有详细的手动安装的集群如何升级的参考资料，只有两篇关于kubernetes集群升级的文档。 在ubuntu上如何使用juju升级：https://kubernetes.io/docs/getting-started-guides/ubuntu/upgrades/ 使用kubeadm升级：https://kubernetes.io/docs/tasks/administer-cluster/kubeadm-upgrade-1-7/ 手动升级的还没有详细的方案，大多是基于管理工具部署和升级，比如juju、kubeadm、kops、kubespray等。 manual upgrade/downgrade testing for Kubernetes 1.6 - google group，在这个Google group中讨论了kubernetes手动升级的问题，并给出了参考建议。 升级步骤 注意：该升级步骤是实验性的，建议在测试集群上使用，无法保证线上服务不中断，实际升级完成后无需对线上服务做任何操作。 大体上的升级步骤是，先升级master节点，然后再一次升级每台node节点。 准备 备份kubernetes原先的二进制文件和配置文件。 下载最新版本的kubernetes二进制包，如1.8.5版本，查看changelog，下载二进制包，我们使用的是kubernetes-server-linux-amd64.tar.gz，分发到集群的每个节点上。 升级master节点 停止master节点的进程 systemctl stop kube-apiserver systemctl stop kube-scheduler systemctl stop kube-controller-manager systemctl stop kube-proxy systemctl stop kubelet 使用新版本的kubernetes二进制文件替换原来老版本的文件，然后启动master节点上的进程： systemctl start kube-apiserver systemctl start kube-scheduler systemctl start kube-controller-manager 因为我们的master节点同时也作为node节点，所有还要执行下面的”升级node节点“中的步骤。 升级node节点 关闭swap # 临时关闭 swapoff -a # 永久关闭，注释掉swap分区即可 vim /etc/fstab #UUID=65c9f92d-4828-4d46-bf19-fb78a38d2fd1 swap swap defaults 0 0 修改kubelet的配置文件 将kubelet的配置文件/etc/kubernetes/kublet配置文件中的KUBELET_API_SERVER=\"--api-servers=http://172.20.0.113:8080\"行注释掉。 注意：：kubernetes1.7及以上版本已经没有该配置了，API server的地址写在了kubeconfig文件中。 停止node节点上的kubernetes进程： systemctl stop kubelet systemctl stop kube-proxy 使用新版本的kubernetes二进制文件替换原来老版本的文件，然后启动node节点上的进程： systemctl start kubelet systemctl start kube-proxy 启动新版本的kube-proxy报错找不到conntrack命令，使用yum install -y conntrack-tools命令安装后重启kube-proxy即可。 检查 到此升级完成，在master节点上检查节点状态： NAME STATUS ROLES AGE VERSION 172.20.0.113 Ready 244d v1.8.5 172.20.0.114 Ready 244d v1.8.5 172.20.0.115 Ready 244d v1.8.5 所有节点的状态都正常，再检查下原先的运行在kubernetes之上的服务是否正常，如果服务正常的话说明这次升级无误。 参考 Cluster Upgrade #2524 Upgrading self-hosted Kubernetes Upgrading Kubernetes - kops Upgrading kubeadm clusters from 1.6 to 1.7 How to Upgrade a Kubernetes Cluster With No Downtime manual upgrade/downgrade testing for Kubernetes 1.6 - google group Notes/Instructions for Manual Upgrade Testing1.5 -> 1.6 Upgrading Kubernetes in Kubespray Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-13 13:02:14 "},"practice/dashboard-upgrade.html":{"url":"practice/dashboard-upgrade.html","title":"升级dashboard","keywords":"","body":"升级Dashboard 我们在kubernetes1.6的时候同时安装了dashboard插件，该插件也是基于kubernetes1.6版本开发的。如今kubernetes1.8版本业已发布，如何升级dashboard以获取新版中功能呢？ Dashboard的升级比较简单，因为它仅仅是一个前端应用，用来展现集群信息和与后端API交互，理论上只需要更新原先dashboard的yaml配置文件中的镜像就可以了，但是为了使用dashboard1.7版本中的用户登陆功能，还需要做一些额外的操作。 dashboard的更新日志请见release note，当前的最新版本为v1.7.1，下面将介绍将dashboard从v1.6.3升级到v1.7.1并开启用户登陆认证的详细步骤。 升级步骤 删除原来的版本 首先删除原来的dashboard资源： kubectl delete -f dashboard/ 将dashboard目录下的所有yaml文件中的资源全部删除，包括Deployment、service和角色绑定等。 部署新版本 我们使用官方的配置文件来安装，首先下载官方配置： wget https://raw.githubusercontent.com/kubernetes/dashboard/master/src/deploy/recommended/kubernetes-dashboard.yaml 修改其中的两个镜像地址为我们的私有地址。 gcr.io/google_containers/kubernetes-dashboard-init-amd64:v1.0.1 gcr.io/google_containers/kubernetes-dashboard-amd64:v1.7.1 这个两个镜像可以同时从时速云上获取： index.tenxcloud.com/jimmy/kubernetes-dashboard-amd64:v1.7.1 index.tenxcloud.com/jimmy/kubernetes-dashboard-init-amd64:v1.0.1 将service type设置为NodePort，修改后的yaml文件见kubernetes-dashboard.yaml，然后就可以部署新版本的dashboard了。 kubectl create -f kubernetes-dashboard.yaml 获取dashboard的外网访问端口： kubectl -n kube-system get svc kubernetes-dashboard NAME CLUSTER-IP EXTERNAL-IP PORT(S) AGE kubernetes-dashboard 10.254.177.181 443:32324/TCP 49m 访问集群中的任何一个节点，即可打开dashboard登陆页面，如https://172.20.0.113:32324/（请使用https访问），支持使用`kubeconfig`和`token`两种的认证方式： 图片 - 登陆界面 选择本地的kubeconfig文件以登陆集群，kubeconfig文件中包括登陆的用户名、证书和token信息。 登陆之后首先看到的界面是这样的： 图片 - 首页 这是因为该用户没有对default命名空间的访问权限。 修改URL地址中的namespace字段为该用户有权限访问的命名空间如brand：https://172.20.0.113:32324/#!/overview?namespace=brand： 图片 - 用户空间 设置界面的语言 我们看到现在 dashboard 的页面都已经被汉化了，当前支持英文、中文简体、中文繁体、日语，根据浏览器的语言自动切换的。如果想要强制设置 dashboard 中显示的语言，需要在 dahsboard 的 Deployment yaml 配置中增加如下配置： env: - name: ACCEPT_LANGUAGE value: english 关于 i18n 的设计文档请参考：https://github.com/kubernetes/dashboard/blob/master/docs/design/i18n.md 更简单的方式是，如果您使用的Chrome浏览器，则在浏览器中的配置中设置语言的顺序后刷新网页，dashboard将以您在Chrome中配置的首选语言显示。 身份认证 登陆 dashboard 的时候支持 kubeconfig 和 token 两种认证方式，kubeconfig 中也依赖 token 字段，所以生成 token 这一步是必不可少的。 下文分两块来讲解两种登陆认证方式： 为 brand 命名空间下的 brand 用户创建 kubeconfig 文件 为集群的管理员（拥有所有命名空间的 amdin 权限）创建 token 使用 kubeconfig 登陆dashboard的时候可以指定kubeconfig文件来认证用户权限，如何生成登陆dashboard时指定的kubeconfig文件请参考创建用户认证授权的kubeconfig文件。 注意我们生成的 kubeconfig 文件中没有 token 字段，需要手动添加该字段。 比如我们为 brand namespace 下的 brand 用户生成了名为 brand.kubeconfig 的 kubeconfig 文件，还要再该文件中追加一行 token 的配置（如何生成 token 将在下文介绍），如下所示： 图片 - kubeconfig文件 这样就可以使用brand.kubeconfig文件来登陆dashboard了，而且只能访问和操作brand命名空间下的对象。 生成 token 需要创建一个admin用户并授予admin角色绑定，使用下面的yaml文件创建admin用户并赋予他管理员权限，然后可以通过token登陆dashbaord，该文件见admin-role.yaml。这种认证方式本质上是通过 Service Account 的身份认证加上 Bearer token 请求 API server 的方式实现，参考 Kubernetes 中的认证。 kind: ClusterRoleBinding apiVersion: rbac.authorization.k8s.io/v1beta1 metadata: name: admin annotations: rbac.authorization.kubernetes.io/autoupdate: \"true\" roleRef: kind: ClusterRole name: cluster-admin apiGroup: rbac.authorization.k8s.io subjects: - kind: ServiceAccount name: admin namespace: kube-system --- apiVersion: v1 kind: ServiceAccount metadata: name: admin namespace: kube-system labels: kubernetes.io/cluster-service: \"true\" addonmanager.kubernetes.io/mode: Reconcile 然后执行下面的命令创建 serviceaccount 和角色绑定，对于其他命名空间的其他用户只要修改上述 yaml 中的 name 和 namespace 字段即可： kubectl create -f admin-role.yaml 创建完成后获取secret和token的值。 # 获取admin-token的secret名字 $ kubectl -n kube-system get secret|grep admin-token admin-token-nwphb kubernetes.io/service-account-token 3 6m # 获取token的值 $ kubectl -n kube-system describe secret admin-token-nwphb Name: admin-token-nwphb Namespace: kube-system Labels: Annotations: kubernetes.io/service-account.name=admin kubernetes.io/service-account.uid=f37bd044-bfb3-11e7-87c0-f4e9d49f8ed0 Type: kubernetes.io/service-account-token Data ==== namespace: 11 bytes token: 非常长的字符串 ca.crt: 1310 bytes 在 dashboard 登录页面上使用上面输出中的那个非常长的字符串进行 base64 解码后作为 token 登录，即可以拥有管理员权限操作整个kubernetes集群中的对象。当然您也可以将这串 token 进行 base64 解码后，加到 admin 用户的kubeconfig文件中，继续使用kubeconfig登录，两种认证方式任您选择。 注意：一定要将 kubectl 的输出中的 token 值进行 base64 解码，在线解码工具 base64decode，Linux 和 Mac 有自带的 base64 命令也可以直接使用，输入 base64 是进行编码，Linux 中base64 -d 表示解码，Mac 中使用 base64 -D。 也可以使用 jsonpath 的方式直接获取 token 的值，如： kubectl -n kube-system get secret admin-token-nwphb -o jsonpath={.data.token}|base64 -d 注意我们使用了 base64 对其重新解码，因为 secret 都是经过 base64 编码的，如果直接使用 kubectl 中查看到的 token 值会认证失败，详见 secret 配置。关于 JSONPath 的使用请参考 JSONPath 手册。 参考 Dashboard log in mechanism #2093 Accessing Dashboard 1.7.X and above Kubernetes dashboard UX for Role-Based Access Control How to sign in kubernetes dashboard? - StackOverflow JSONPath 手册 Kubernetes 中的认证 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-13 16:28:52 "},"usecases/":{"url":"usecases/","title":"领域应用概览","keywords":"","body":"领域应用 Kubernetes 和云原生应用在各个领域中的实践。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-15 20:05:51 "},"usecases/microservices.html":{"url":"usecases/microservices.html","title":"微服务架构","keywords":"","body":"微服务架构 Kubernetes 设计之初就是按照 Cloud Native 的理念设计的，Cloud Native 中有个重要概念就是微服务的架构设计，当将单体应用拆分微服务后， 随着服务数量的增多，如何微服务进行管理以保证服务的 SLA 呢？为了从架构层面上解决这个问题，解放程序员的创造性，避免繁琐的服务发现、监控、分布式追踪等事务，Service mesh 应运而生。 微服务 下图是Bilgin Ibryam给出的微服务中应该关心的主题，图片来自RedHat Developers。 图片 - 微服务关注的部分 当前最成熟最完整的微服务框架可以说非Spring莫属，而Spring又仅限于Java语言开发，其架构本身又跟Kubernetes存在很多重合的部分，如何探索将Kubernetes作为微服务架构平台就成为一个热点话题。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-20 16:43:03 "},"usecases/service-discovery-in-microservices.html":{"url":"usecases/service-discovery-in-microservices.html","title":"微服务中的服务发现","keywords":"","body":"微服务中的服务发现 在单体架构时，因为服务不会经常和动态迁移，所有服务地址可以直接在配置文件中配置，所以也不会有服务发现的问题。但是对于微服务来说，应用的拆分，服务之间的解耦，和服务动态扩展带来的服务迁移，服务发现就成了微服务中的一个关键问题。 服务发现分为客户端服务发现和服务端服务发现两种，架构如下图所示。 图片 - 微服务中的服务发现 这两种架构都各有利弊，我们拿客户端服务发现软件Eureka和服务端服务发现架构Kubernetes/SkyDNS+Ingress LB+Traefik+PowerDNS为例说明。 服务发现方案 Pros Cons Eureka 使用简单，适用于java语言开发的项目，比服务端服务发现少一次网络跳转 对非Java语言的支持不够好，Consumer需要内置特定的服务发现客户端和发现逻辑 Kubernetes Consumer无需关注服务发现具体细节，只需知道服务的DNS域名即可，支持异构语言开发 需要基础设施支撑，多了一次网络跳转，可能有性能损失 Eureka 也不是单独使用的，一般会配合 ribbon 一起使用，ribbon 作为路由和负载均衡。 Ribbon提供一组丰富的功能集： 多种内建的负载均衡规则： Round-robin 轮询负载均衡 平均加权响应时间负载均衡 随机负载均衡 可用性过滤负载均衡（避免跳闸线路和高并发链接数） 自定义负载均衡插件系统 与服务发现解决方案的可拔插集成（包括Eureka） 云原生智能，例如可用区亲和性和不健康区规避 内建的故障恢复能力 参考 谈服务发现的背景、架构以及落地方案 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-09-16 16:38:32 "},"usecases/service-mesh.html":{"url":"usecases/service-mesh.html","title":"Service Mesh 服务网格","keywords":"","body":"Service Mesh 服务网格 Service mesh 又译作 ”服务网格“，作为服务间通信的基础设施层。Buoyant 公司的 CEO Willian Morgan 在他的这篇文章 WHAT’S A SERVICE MESH? AND WHY DO I NEED ONE? 中解释了什么是 Service Mesh，为什么云原生应用需要 Service Mesh。 A service mesh is a dedicated infrastructure layer for handling service-to-service communication. It’s responsible for the reliable delivery of requests through the complex topology of services that comprise a modern, cloud native application. In practice, the service mesh is typically implemented as an array of lightweight network proxies that are deployed alongside application code, without the application needing to be aware. 今年来以 Istio 和 Linkderd 为代表的 Service Mesh 蓬勃发展，大有成为下一代语言异构微服务架构的王者之范，今天又碰巧看到了 Red Hat 的 Burr Sutter 提出了8 Steps to Becoming Awesome with Kubernetes，整个PPT一共60多页，很有建设性，点此跳转到我的GitHub上下载，我将其归档到cloud-native-slides-share中了。 图片 - 下一代异构微服务架构 自我6月份初接触Istio依赖就发觉service mesh很好的解决了异构语言中的很多问题，而且是kuberentes service 上层不可或缺的服务间代理。关于istio的更多内容请参考 istio中文文档。 什么是 service mesh？ Service mesh 有如下几个特点： 应用程序间通讯的中间层 轻量级网络代理 应用程序无感知 解耦应用程序的重试/超时、监控、追踪和服务发现 目前两款流行的 service mesh 开源软件 Istio 和 Linkerd 都可以直接在 kubernetes 中集成，其中 Linkerd 已经成为 CNCF 成员。 理解 Service Mesh 如果用一句话来解释什么是 Service Mesh，可以将它比作是应用程序或者说微服务间的 TCP/IP，负责服务之间的网络调用、限流、熔断和监控。对于编写应用程序来说一般无须关心 TCP/IP 这一层（比如通过 HTTP 协议的 RESTful 应用），同样使用 Service Mesh 也就无须关系服务之间的那些原来是通过应用程序或者其他框架实现的事情，比如 Spring Cloud、OSS，现在只要交给 Service Mesh 就可以了。 Phil Calçado 在他的这篇博客 Pattern: Service Mesh 中详细解释了 Service Mesh 的来龙去脉： 从最原始的主机之间直接使用网线相连 网络层的出现 集成到应用程序内部的控制流 分解到应用程序外部的控制流 应用程序的中集成服务发现和断路器 出现了专门用于服务发现和断路器的软件包/库，如 Twitter 的 Finagle 和 Facebook 的 Proxygen，这时候还是集成在应用程序内部 出现了专门用于服务发现和断路器的开源软件，如 Netflix OSS、Airbnb 的 synapse 和 nerve 最后作为微服务的中间层 service mesh 出现 Service mesh 的架构如下图所示： 图片 - Service Mesh 架构图 图片来自：Pattern: Service Mesh Service mesh 作为 sidecar 运行，对应用程序来说是透明，所有应用程序间的流量都会通过它，所以对应用程序流量的控制都可以在 serivce mesh 中实现。 Service mesh如何工作？ 下面以 Linkerd 为例讲解 service mesh 如何工作，Istio 作为 service mesh 的另一种实现原理与 linkerd 基本类似，后续文章将会详解 Istio 和 Linkerd 如何在 kubernetes 中工作。 Linkerd 将服务请求路由到目的地址，根据中的参数判断是到生产环境、测试环境还是 staging 环境中的服务（服务可能同时部署在这三个环境中），是路由到本地环境还是公有云环境？所有的这些路由信息可以动态配置，可以是全局配置也可以为某些服务单独配置。 当 Linkerd 确认了目的地址后，将流量发送到相应服务发现端点，在 kubernetes 中是 service，然后 service 会将服务转发给后端的实例。 Linkerd 根据它观测到最近请求的延迟时间，选择出所有应用程序的实例中响应最快的实例。 Linkerd 将请求发送给该实例，同时记录响应类型和延迟数据。 如果该实例挂了、不响应了或者进程不工作了，Linkerd 将把请求发送到其他实例上重试。 如果该实例持续返回 error，Linkerd 会将该实例从负载均衡池中移除，稍后再周期性得重试。 如果请求的截止时间已过，Linkerd 主动失败该请求，而不是再次尝试添加负载。 Linkerd 以 metric 和分布式追踪的形式捕获上述行为的各个方面，这些追踪信息将发送到集中 metric 系统。 为何使用 service mesh？ Service mesh 并没有给我们带来新功能，它是用于解决其他工具已经解决过的问题，只不过这次是在 Cloud Native 的 kubernetes 环境下的实现。 在传统的 MVC 三层 Web 应用程序架构下，服务之间的通讯并不复杂，在应用程序内部自己管理即可，但是在现今的复杂的大型网站情况下，单体应用被分解为众多的微服务，服务之间的依赖和通讯十分复杂，出现了 twitter 开发的 Finagle、Netflix 开发的 Hystrix 和 Google 的 Stubby 这样的 ”胖客户端“ 库，这些就是早期的 service mesh，但是它们都近适用于特定的环境和特定的开发语言，并不能作为平台级的 service mesh 支持。 在 Cloud Native 架构下，容器的使用给予了异构应用程序的更多可行性，kubernetes 增强的应用的横向扩容能力，用户可以快速的编排出复杂环境、复杂依赖关系的应用程序，同时开发者又无须过分关心应用程序的监控、扩展性、服务发现和分布式追踪这些繁琐的事情而专注于程序开发，赋予开发者更多的创造性。 Istio VS Linkerd 当前的Service Mesh实现主要有两大阵营，要给是Linkerd（也是最初提出该概念的），另一个是Istio，当然还有很多其他号称也是Service Mesh，比如Nginx出品的Nginmesh。 Feature Istio Linkerd 部署架构 Envoy/Sidecar DaemonSets 易用性 复杂 简单 支持平台 kuberentes kubernetes/mesos/Istio/local 当前版本 0.3.0 1.3.3 是否已有生产部署 否 是 下图是Istio和Linkerd架构的不同，Istio是使用Sidecar模式，将Envoy植入到Pod中，而Linkerd则是在每台node上都以DaemonSet的方式运行。 图片 - Istio vs linkerd 关于Istio和Linkerd的详细信息请参考 安装并试用Istio service mesh 与 Linkerd 使用指南。 另外出品Linkerd的公司buoyant又推出了conduit，这是一种更轻量级的Service Mesh。 参考 WHAT’S A SERVICE MESH? AND WHY DO I NEED ONE? So what even is a Service Mesh? Hot take on Istio and Linkerd linkerd: A service mesh for AWS ECS Introducing Istio: A robust service mesh for microservices Application Network Functions With ESBs, API Management, and Now.. Service Mesh? Pattern: Service Mesh Istio官方文档中文版 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-07 20:09:18 "},"usecases/istio.html":{"url":"usecases/istio.html","title":"Istio","keywords":"","body":"Istio简介 前言 Istio是由Google、IBM和Lyft开源的微服务管理、保护和监控框架。Istio为希腊语，意思是”起航“。 简介 使用istio可以很简单的创建具有负载均衡、服务间认证、监控等功能的服务网络，而不需要对服务的代码进行任何修改。你只需要在部署环境中，例如Kubernetes的pod里注入一个特别的sidecar proxy来增加对istio的支持，用来截获微服务之间的网络流量。 目前版本的istio只支持kubernetes，未来计划支持其他其他环境。 另外，Istio的前身是IBM开源的Amalgam8，追本溯源，我们来看下它的特性。 Amalgam8 Amalgam8的网站上说，它是一个Content-based Routing Fabric for Polyglot Microservices，简单、强大且开源。 Amalgam8是一款基于内容和版本的路由布局，用于集成多语言异构体微服务。 其control plane API可用于动态编程规则，用于在正在运行的应用程序中跨微服务进行路由和操作请求。 以内容/版本感知方式路由请求的能力简化了DevOps任务，如金丝雀和红/黑发布，A/B Test和系统地测试弹性微服务。 可以使用Amalgam8平台与受欢迎的容器运行时（如Docker，Kubernetes，Marathon / Mesos）或其他云计算提供商（如IBM Bluemix，Google Cloud Platform或Amazon AWS）。 特性 使用istio的进行微服务管理有如下特性： 流量管理：控制服务间的流量和API调用流，使调用更可靠，增强不同环境下的网络鲁棒性。 可观测性：了解服务之间的依赖关系和它们之间的性质和流量，提供快速识别定位问题的能力。 策略实施：通过配置mesh而不是以改变代码的方式来控制服务之间的访问策略。 服务识别和安全：提供在mesh里的服务可识别性和安全性保护。 未来将支持多种平台，不论是kubernetes、Mesos、还是云。同时可以集成已有的ACL、日志、监控、配额、审计等。 架构 图片 - Istio架构图 Istio架构分为控制层和数据层。 数据层：由一组智能代理（Envoy）作为sidecar部署，协调和控制所有microservices之间的网络通信。 控制层：负责管理和配置代理路由流量，以及在运行时执行的政策。 Envoy Istio使用Envoy代理的扩展版本，该代理是以C++开发的高性能代理，用于调解service mesh中所有服务的所有入站和出站流量。 Istio利用了Envoy的许多内置功能，例如动态服务发现，负载平衡，TLS终止，HTTP/2＆gRPC代理，断路器，运行状况检查，基于百分比的流量拆分分阶段上线，故障注入和丰富指标。 Envoy在kubernetes中作为pod的sidecar来部署。 这允许Istio将大量关于流量行为的信号作为属性提取出来，这些属性又可以在Mixer中用于执行策略决策，并发送给监控系统以提供有关整个mesh的行为的信息。 Sidecar代理模型还允许你将Istio功能添加到现有部署中，无需重新构建或重写代码。 更多信息参见设计目标。 Mixer Mixer负责在service mesh上执行访问控制和使用策略，并收集Envoy代理和其他服务的遥测数据。代理提取请求级属性，发送到mixer进行评估。有关此属性提取和策略评估的更多信息，请参见Mixer配置。 混音器包括一个灵活的插件模型，使其能够与各种主机环境和基础架构后端进行接口，从这些细节中抽象出Envoy代理和Istio管理的服务。 Istio Manager Istio-Manager用作用户和Istio之间的接口，收集和验证配置，并将其传播到各种Istio组件。它从Mixer和Envoy中抽取环境特定的实现细节，为他们提供独立于底层平台的用户服务的抽象表示。 此外，流量管理规则（即通用4层规则和七层HTTP/gRPC路由规则）可以在运行时通过Istio-Manager进行编程。 Istio-auth Istio-Auth提供强大的服务间和最终用户认证，使用相互TLS，内置身份和凭据管理。它可用于升级service mesh中的未加密流量，并为运营商提供基于服务身份而不是网络控制的策略的能力。 Istio的未来版本将增加细粒度的访问控制和审计，以使用各种访问控制机制（包括属性和基于角色的访问控制以及授权hook）来控制和监控访问你服务、API或资源的人员。 参考 Istio开源平台发布，Google、IBM和Lyft分别承担什么角色？ Istio：用于微服务的服务啮合层 Istio Overview Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-08-21 18:23:35 "},"usecases/istio-installation.html":{"url":"usecases/istio-installation.html","title":"安装并试用Istio service mesh","keywords":"","body":"安装并试用Istio service mesh 本文根据官网的文档整理而成，步骤包括安装istio 0.1.5并创建一个bookinfo的微服务来测试istio的功能。 文中使用的yaml文件可以在kubernetes-handbook的manifests/istio目录中找到，所有的镜像都换成了我的私有镜像仓库地址，请根据官网的镜像自行修改。 安装环境 CentOS 7.3.1611 Docker 1.12.6 Kubernetes 1.6.0 安装 1.下载安装包 下载地址：https://github.com/istio/istio/releases 下载Linux版本的当前最新版安装包 wget https://github.com/istio/istio/releases/download/0.1.5/istio-0.1.5-linux.tar.gz 2.解压 解压后，得到的目录结构如下： . ├── bin │ └── istioctl ├── install │ └── kubernetes │ ├── addons │ │ ├── grafana.yaml │ │ ├── prometheus.yaml │ │ ├── servicegraph.yaml │ │ └── zipkin.yaml │ ├── istio-auth.yaml │ ├── istio-rbac-alpha.yaml │ ├── istio-rbac-beta.yaml │ ├── istio.yaml │ ├── README.md │ └── templates │ ├── istio-auth │ │ ├── istio-auth-with-cluster-ca.yaml │ │ ├── istio-cluster-ca.yaml │ │ ├── istio-egress-auth.yaml │ │ ├── istio-ingress-auth.yaml │ │ └── istio-namespace-ca.yaml │ ├── istio-egress.yaml │ ├── istio-ingress.yaml │ ├── istio-manager.yaml │ └── istio-mixer.yaml ├── istio.VERSION ├── LICENSE └── samples ├── apps │ ├── bookinfo │ │ ├── bookinfo.yaml │ │ ├── cleanup.sh │ │ ├── destination-ratings-test-delay.yaml │ │ ├── loadbalancing-policy-reviews.yaml │ │ ├── mixer-rule-additional-telemetry.yaml │ │ ├── mixer-rule-empty-rule.yaml │ │ ├── mixer-rule-ratings-denial.yaml │ │ ├── mixer-rule-ratings-ratelimit.yaml │ │ ├── README.md │ │ ├── route-rule-all-v1.yaml │ │ ├── route-rule-delay.yaml │ │ ├── route-rule-reviews-50-v3.yaml │ │ ├── route-rule-reviews-test-v2.yaml │ │ ├── route-rule-reviews-v2-v3.yaml │ │ └── route-rule-reviews-v3.yaml │ ├── httpbin │ │ ├── httpbin.yaml │ │ └── README.md │ └── sleep │ ├── README.md │ └── sleep.yaml └── README.md 11 directories, 41 files 从文件里表中可以看到，安装包中包括了kubernetes的yaml文件，示例应用和安装模板。 3.安装istioctl 将./bin/istioctl拷贝到你的$PATH目录下。 4.检查RBAC 因为我们安装的kuberentes版本是1.6.0默认支持RBAC，这一步可以跳过。如果你使用的其他版本的kubernetes，请参考官方文档操作。 执行以下命令，正确的输出是这样的： $ kubectl api-versions | grep rbac rbac.authorization.k8s.io/v1alpha1 rbac.authorization.k8s.io/v1beta1 5.创建角色绑定 $ kubectl create -f install/kubernetes/istio-rbac-beta.yaml clusterrole \"istio-manager\" created clusterrole \"istio-ca\" created clusterrole \"istio-sidecar\" created clusterrolebinding \"istio-manager-admin-role-binding\" created clusterrolebinding \"istio-ca-role-binding\" created clusterrolebinding \"istio-ingress-admin-role-binding\" created clusterrolebinding \"istio-sidecar-role-binding\" created 注意：官网的安装包中的该文件中存在RoleBinding错误，应该是集群级别的clusterrolebinding，而release里的代码只是普通的rolebinding，查看该Issue Istio manager cannot list of create k8s TPR when RBAC enabled #327。 6.安装istio核心组件 用到的镜像有： docker.io/istio/mixer:0.1.5 docker.io/istio/manager:0.1.5 docker.io/istio/proxy_debug:0.1.5 我们暂时不开启Istio Auth。 本文中用到的所有yaml文件中的type: LoadBalancer去掉，使用默认的ClusterIP，然后配置Traefik ingress，就可以在集群外部访问。请参考安装Traefik ingress。 kubectl apply -f install/kubernetes/istio.yaml 7.安装监控插件 用到的镜像有： docker.io/istio/grafana:0.1.5 quay.io/coreos/prometheus:v1.1.1 gcr.io/istio-testing/servicegraph:latest docker.io/openzipkin/zipkin:latest 为了方便下载，其中两个镜像我备份到了时速云： index.tenxcloud.com/jimmy/prometheus:v1.1.1 index.tenxcloud.com/jimmy/servicegraph:latest 安装插件 kubectl apply -f install/kubernetes/addons/prometheus.yaml kubectl apply -f install/kubernetes/addons/grafana.yaml kubectl apply -f install/kubernetes/addons/servicegraph.yaml kubectl apply -f install/kubernetes/addons/zipkin.yaml 在traefik ingress中增加增加以上几个服务的配置，同时增加istio-ingress配置。 - host: grafana.istio.io http: paths: - path: / backend: serviceName: grafana servicePort: 3000 - host: servicegraph.istio.io http: paths: - path: / backend: serviceName: servicegraph servicePort: 8088 - host: prometheus.istio.io http: paths: - path: / backend: serviceName: prometheus servicePort: 9090 - host: zipkin.istio.io http: paths: - path: / backend: serviceName: zipkin servicePort: 9411 - host: ingress.istio.io http: paths: - path: / backend: serviceName: istio-ingress servicePort: 80 测试 我们使用Istio提供的测试应用bookinfo微服务来进行测试。 该微服务用到的镜像有： istio/examples-bookinfo-details-v1 istio/examples-bookinfo-ratings-v1 istio/examples-bookinfo-reviews-v1 istio/examples-bookinfo-reviews-v2 istio/examples-bookinfo-reviews-v3 istio/examples-bookinfo-productpage-v1 该应用架构图如下： 图片 - BookInfo Sample应用架构图 部署应用 kubectl create -f Istio kube-inject命令会在bookinfo.yaml文件中增加Envoy sidecar信息。参考：https://istio.io/docs/reference/commands/istioctl.html#istioctl-kube-inject 在本机的/etc/hosts下增加VIP节点和ingress.istio.io的对应信息。具体步骤参考：边缘节点配置 在浏览器中访问http://ingress.istio.io/productpage 图片 - BookInfo Sample页面 多次刷新页面，你会发现有的页面上的评论里有星级打分，有的页面就没有，这是因为我们部署了三个版本的应用，有的应用里包含了评分，有的没有。Istio根据默认策略随机将流量分配到三个版本的应用上。 查看部署的bookinfo应用中的productpage-v1 service和deployment，查看productpage-v1的pod的详细json信息可以看到这样的结构： $ kubectl get productpage-v1-944450470-bd530 -o json 见productpage-v1-istio.json文件。从详细输出中可以看到这个Pod中实际有两个容器，这里面包括了initContainer，作为istio植入到kubernetes deployment中的sidecar。 \"initContainers\": [ { \"args\": [ \"-p\", \"15001\", \"-u\", \"1337\" ], \"image\": \"docker.io/istio/init:0.1\", \"imagePullPolicy\": \"Always\", \"name\": \"init\", \"resources\": {}, \"securityContext\": { \"capabilities\": { \"add\": [ \"NET_ADMIN\" ] } }, \"terminationMessagePath\": \"/dev/termination-log\", \"terminationMessagePolicy\": \"File\", \"volumeMounts\": [ { \"mountPath\": \"/var/run/secrets/kubernetes.io/serviceaccount\", \"name\": \"default-token-3l9f0\", \"readOnly\": true } ] }, { \"args\": [ \"-c\", \"sysctl -w kernel.core_pattern=/tmp/core.%e.%p.%t \\u0026\\u0026 ulimit -c unlimited\" ], \"command\": [ \"/bin/sh\" ], \"image\": \"alpine\", \"imagePullPolicy\": \"Always\", \"name\": \"enable-core-dump\", \"resources\": {}, \"securityContext\": { \"privileged\": true }, \"terminationMessagePath\": \"/dev/termination-log\", \"terminationMessagePolicy\": \"File\", \"volumeMounts\": [ { \"mountPath\": \"/var/run/secrets/kubernetes.io/serviceaccount\", \"name\": \"default-token-3l9f0\", \"readOnly\": true } ] } ], 监控 不断刷新productpage页面，将可以在以下几个监控中看到如下界面。 Grafana页面 http://grafana.istio.io 图片 - Istio Grafana界面 Prometheus页面 http://prometheus.istio.io 图片 - Prometheus页面 Zipkin页面 http://zipkin.istio.io 图片 - Zipkin页面 ServiceGraph页面 http://servicegraph.istio.io/dotviz 可以用来查看服务间的依赖关系。 访问 http://servicegraph.istio.io/graph 可以获得json格式的返回结果。 图片 - ServiceGraph页面 更进一步 BookInfo示例中有三个版本的reviews，可以使用istio来配置路由请求，将流量分发到不同版本的应用上。参考Configuring Request Routing。 还有一些更高级的功能，我们后续将进一步探索。 参考 Installing Istio BookInfo sample Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-21 21:18:30 "},"usecases/configuring-request-routing.html":{"url":"usecases/configuring-request-routing.html","title":"配置请求的路由规则","keywords":"","body":"配置请求的路由规则 在上一节安装istio中我们创建BookInfo的示例，熟悉了Istio的基本功能，现在我们再来看一下istio的高级特性——配置请求的路由规则。 使用istio我们可以根据权重和HTTP headers来动态配置请求路由。 基于内容的路由 因为BookInfo示例部署了3个版本的评论微服务，我们需要设置一个默认路由。 否则，当你多次访问应用程序时，会注意到有时输出包含星级，有时候又没有。 这是因为没有明确的默认版本集，Istio将以随机方式将请求路由到服务的所有可用版本。 注意：假定您尚未设置任何路由。如果您已经为示例创建了冲突的路由规则，则需要在以下命令中使用replace而不是create。 下面这个例子能够根据网站的不同登陆用户，将流量划分到服务的不同版本和实例。跟kubernetes中的应用一样，所有的路由规则都是通过声明式的yaml配置。关于reviews:v1和reviews:v2的唯一区别是，v1没有调用评分服务，productpage页面上不会显示评分星标。 将微服务的默认版本设置成v1。 istioctl create -f samples/apps/bookinfo/route-rule-all-v1.yaml 使用以下命令查看定义的路由规则。 istioctl get route-rules -o yaml type: route-rule name: details-default namespace: default spec: destination: details.default.svc.cluster.local precedence: 1 route: - tags: version: v1 --- type: route-rule name: productpage-default namespace: default spec: destination: productpage.default.svc.cluster.local precedence: 1 route: - tags: version: v1 --- type: route-rule name: reviews-default namespace: default spec: destination: reviews.default.svc.cluster.local precedence: 1 route: - tags: version: v1 --- type: route-rule name: ratings-default namespace: default spec: destination: ratings.default.svc.cluster.local precedence: 1 route: - tags: version: v1 --- 由于对代理的规则传播是异步的，因此在尝试访问应用程序之前，需要等待几秒钟才能将规则传播到所有pod。 在浏览器中打开BookInfo URL（http://$GATEWAY_URL/productpage ，我们在上一节中使用的是 http://ingress.istio.io/productpage ）您应该会看到BookInfo应用程序的产品页面显示。 注意，产品页面上没有评分星，因为reviews:v1不访问评级服务。 将特定用户路由到reviews:v2。 为测试用户jason启用评分服务，将productpage的流量路由到reviews:v2实例上。 istioctl create -f samples/apps/bookinfo/route-rule-reviews-test-v2.yaml 确认规则生效： istioctl get route-rule reviews-test-v2 destination: reviews.default.svc.cluster.local match: httpHeaders: cookie: regex: ^(.*?;)?(user=jason)(;.*)?$ precedence: 2 route: - tags: version: v2 使用jason用户登陆productpage页面。 你可以看到每个刷新页面时，页面上都有一个1到5颗星的评级。如果你使用其他用户登陆的话，将因继续使用reviews:v1而看不到星标评分。 内部实现 在这个例子中，一开始使用istio将100%的流量发送到BookInfo服务的reviews:v1的实例上。然后又根据请求的header（例如用户的cookie）将流量选择性的发送到reviews:v2实例上。 验证了v2实例的功能后，就可以将全部用户的流量发送到v2实例上，或者逐步的迁移流量，如10%、20%直到100%。 如果你看了故障注入这一节，你会发现v2版本中有个bug，而在v3版本中修复了，你想将流量迁移到reviews:v1迁移到reviews:v3版本上，只需要运行如下命令： 将50%的流量从reviews:v1转移到reviews:v3上。 istioctl replace -f samples/apps/bookinfo/route-rule-reviews-50-v3.yaml 注意这次使用的是replace命令，而不是create，因为该rule已经在前面创建过了。 登出jason用户，或者删除测试规则，可以看到新的规则已经生效。 删除测试规则。 istioctl delete route-rule reviews-test-v2 istioctl delete route-rule ratings-test-delay 现在的规则就是刷新productpage页面，50%的概率看到红色星标的评论，50%的概率看不到星标。 注意：因为使用Envoy sidecar的实现，你需要刷新页面很多次才能看到接近规则配置的概率分布，你可以将v3的概率修改为90%，这样刷新页面时，看到红色星标的概率更高。 当v3版本的微服务稳定以后，就可以将100%的流量分摊到reviews:v3上了。 istioctl replace -f samples/apps/bookinfo/route-rule-reviews-v3.yaml 现在不论你使用什么用户登陆productpage页面，你都可以看到带红色星标评分的评论了。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-22 17:00:47 "},"usecases/install-and-expand-istio-mesh.html":{"url":"usecases/install-and-expand-istio-mesh.html","title":"安装和拓展Istio service mesh","keywords":"","body":"安装和拓展 Istio mesh 前置条件 下面的操作说明需要您可以访问 kubernetes 1.7.3 后更高版本 的集群，并且启用了 RBAC (基于角色的访问控制)。您需要安装了 1.7.3 或更高版本 的 kubectl 命令。如果您希望启用 自动注入 sidecar，您需要启用 kubernetes 集群的 alpha 功能。 注意：如果您安装了 Istio 0.1.x，在安装新版本前请先 卸载 它们（包括已启用 Istio 应用程序 Pod 中的 sidecar）。 取决于您的 kubernetes 提供商： 本地安装 Istio，安装最新版本的 Minikube (version 0.22.1 或者更高)。 Google Container Engine 使用 kubectl 获取证书 （使用您自己的集群的名字替换 ，使用集群实际所在的位置替换 ）： gcloud container clusters get-credentials --zone --project 将集群管理员权限授予当前用户（需要管理员权限才能为Istio创建必要的RBAC规则）： kubectl create clusterrolebinding cluster-admin-binding --clusterrole=cluster-admin --user=$(gcloud config get-value core/account) IBM Bluemix Container Service 使用 kubectl 获取证书 （使用您自己的集群的名字替换）： $(bx cs cluster-config |grep \"export KUBECONFIG\") Openshift Origin 3.7 或者以上版本： 默认情况下，Openshift 不允许以 UID 0运行容器。为 Istio 的入口（ingress）和出口（egress）service account 启用使用UID 0运行的容器： oc adm policy add-scc-to-user anyuid -z istio-ingress-service-account -n istio-system oc adm policy add-scc-to-user anyuid -z istio-egress-service-account -n istio-system oc adm policy add-sc-to-user anyuid -z default -n istio-system 运行应用程序 Pod 的 service account 需要特权安全性上下文限制，以此作为 sidecar 注入的一部分: oc adm policy add-scc-to-user privileged -z default -n 安装或升级 Kubernetes 命令行工具 kubectl 以匹配您的集群版本（1.7或以上版本）。 安装步骤 不论对于哪个 Istio 发行版，都安装到 istio-system namespace 下，即可以管理所有其它 namespace 下的微服务。 到 Istio release 页面上，根据您的操作系统下载对应的发行版。如果您使用的是 MacOS 或者 Linux 系统，可以使用下面的额命令自动下载和解压最新的发行版： curl -L https://git.io/getLatestIstio | sh - 解压安装文件，切换到文件所在目录。安装文件目录下包含： install/ 目录下是 kubernetes 使用的 .yaml 安装文件 samples/ 目录下是示例程序 istioctl 客户端二进制文件在 bin 目录下。istioctl 文件用户手动注入 Envoy sidecar 代理、创建路由和策略等。 istio.VERSION 配置文件 切换到 istio 包的解压目录。例如 istio-0.2.7： cd istio-0.2.7 将 istioctl 客户端二进制文件加到 PATH 中。 例如，在 MacOS 或 Linux 系统上执行下面的命令： export PATH=$PWD/bin:$PATH 安装 Istio 的核心部分。选择面两个 互斥 选项中的之一： a) 安装 Istio 的时候不启用 sidecar 之间的 TLS 双向认证： 为具有现在应用程序的集群选择该选项，使用 Istio sidecar 的服务需要能够与非 Istio Kubernetes 服务以及使用 liveliness 和 readiness 探针、headless service 和 StatefulSet 的应用程序通信。 kubectl apply -f install/kubernetes/istio.yaml 或者 b) 安装 Istio 的时候启用 sidecar 之间的 TLS 双向认证： kubectl apply -f install/kubernetes/istio-auth.yaml 这两个选项都会创建 istio-system 命名空间以及所需的 RBAC 权限，并部署 Istio-Pilot、Istio-Mixer、Istio-Ingress、Istio-Egress 和 Istio-CA（证书颁发机构）。 可选的：如果您的 kubernetes 集群开启了 alpha 功能，并想要启用 自动注入 sidecar，需要安装 Istio-Initializer： kubectl apply -f install/kubernetes/istio-initializer.yaml 验证安装 确认系列 kubernetes 服务已经部署了： istio-pilot、 istio-mixer、istio-ingress、 istio-egress： kubectl get svc -n istio-system NAME CLUSTER-IP EXTERNAL-IP PORT(S) AGE istio-egress 10.83.247.89 80/TCP 5h istio-ingress 10.83.245.171 35.184.245.62 80:32730/TCP,443:30574/TCP 5h istio-pilot 10.83.251.173 8080/TCP,8081/TCP 5h istio-mixer 10.83.244.253 9091/TCP,9094/TCP,42422/TCP 5h 注意：如果您运行的集群不支持外部负载均衡器（如 minikube）， istio-ingress 服务的 EXTERNAL-IP 显示。你必须改为使用 NodePort service 或者 端口转发方式来访问应用程序。 确认对应的 Kubernetes pod 已部署并且所有的容器都启动并运行： istio-pilot-*、 istio-mixer-*、 istio-ingress-*、 istio-egress-*、istio-ca-*， istio-initializer-* 是可以选的。 kubectl get pods -n istio-system istio-ca-3657790228-j21b9 1/1 Running 0 5h istio-egress-1684034556-fhw89 1/1 Running 0 5h istio-ingress-1842462111-j3vcs 1/1 Running 0 5h istio-initializer-184129454-zdgf5 1/1 Running 0 5h istio-pilot-2275554717-93c43 1/1 Running 0 5h istio-mixer-2104784889-20rm8 2/2 Running 0 5h 部署应用 您可以部署自己的应用或者示例应用程序如 BookInfo。 注意：应用程序必须使用 HTTP/1.1 或 HTTP/2.0 协议来传输 HTTP 流量，因为 HTTP/1.0 已经不再支持。 如果您启动了 Istio-Initializer，如上所示，您可以使用 kubectl create 直接部署应用。Istio-Initializer 会向应用程序的 pod 中自动注入 Envoy 容器： kubectl create -f .yaml 如果您没有安装 Istio-initializer 的话，您必须使用 istioctl kube-inject 命令在部署应用之前向应用程序的 pod 中手动注入 Envoy 容器： kubectl create -f .yaml) 卸载 卸载 Istio initializer: 如果您安装 Isto 的时候启用了 initializer，请卸载它： kubectl delete -f install/kubernetes/istio-initializer.yaml 卸载 Istio 核心组件。对于某一 Istio 版本，删除 RBAC 权限，istio-system namespace，和该命名空间的下的各层级资源。 不必理会在层级删除过程中的各种报错，因为这些资源可能已经被删除的。 a) 如果您在安装 Istio 的时候关闭了 TLS 双向认证： kubectl delete -f install/kubernetes/istio.yaml 或者 b) 如果您在安装 Istio 的时候启用到了 TLS 双向认证： kubectl delete -f install/kubernetes/istio-auth.yaml 安装 Istio Sidecar Pod Spec 需满足的条件 为了成为 Service Mesh 中的一部分，kubernetes 集群中的每个 Pod 都必须满足如下条件： Service 注解：每个 pod 都必须只属于某一个 Kubernetes Service （当前不支持一个 pod 同时属于多个 service）。 命名的端口：Service 的端口必须命名。端口的名字必须遵循如下格式 [-]，可以是http、http2、 grpc、 mongo、 或者 redis 作为 ，这样才能使用 Istio 的路由功能。例如name: http2-foo 和 name: http 都是有效的端口名称，而 name: http2foo 不是。如果端口的名称是不可识别的前缀或者未命名，那么该端口上的流量就会作为普通的 TCP 流量（除非使用 Protocol: UDP 明确声明使用 UDP 端口）。 带有 app label 的 Deployment：我们建议 kubernetes 的Deploymenet 资源的配置文件中为 Pod 明确指定 app label。每个Deployment 的配置中都需要有个不同的有意义的 app 标签。app label 用于在分布式坠重中添加上下文信息。 Mesh 中的每个 pod 里都有一个 Sidecar： 最后，Mesh 中的每个 pod 都必须运行与 Istio 兼容的sidecar。下面部分介绍了将 sidecar 注入到 pod 中的两种方法：使用istioctl 命令行工具手动注入，或者使用 istio initializer 自动注入。注意 sidecar 不涉及到容器间的流量，因为它们都在同一个 pod 中。 手动注入 sidecar istioctl 命令行中有一个称为 kube-inject 的便利工具，使用它可以将 Istio 的 sidecar 规范添加到 kubernetes 工作负载的规范配置中。与 Initializer 程序不同，kube-inject 只是将 YAML 规范转换成包含 Istio sidecar 的规范。您需要使用标准的工具如 kubectl 来部署修改后的 YAML。例如，以下命令将 sidecar 添加到 sleep.yaml 文件中指定的 pod 中，并将修改后的规范提交给 kubernetes： kubectl apply -f 示例 我们来试一试将 Istio sidecar 注入到 sleep 服务中去。 kubectl apply -f Kube-inject 子命令将 Istio sidecar 和 init 容器注入到 deployment 配置中，转换后的输出如下所示： ... 略过 ... --- apiVersion: extensions/v1beta1 kind: Deployment metadata: annotations: sidecar.istio.io/status: injected-version-root@69916ebba0fc-0.2.6-081ffece00c82cb9de33cd5617682999aee5298d name: sleep spec: replicas: 1 template: metadata: annotations: sidecar.istio.io/status: injected-version-root@69916ebba0fc-0.2.6-081ffece00c82cb9de33cd5617682999aee5298d labels: app: sleep spec: containers: - name: sleep image: tutum/curl command: [\"/bin/sleep\",\"infinity\"] imagePullPolicy: IfNotPresent - name: istio-proxy image: docker.io/istio/proxy_debug:0.2.6 args: ... 略过 ... initContainers: - name: istio-init image: docker.io/istio/proxy_init:0.2.6 imagePullPolicy: IfNotPresent args: ... 略过 ... --- 注入 sidecar 的关键在于 initContainers 和 istio-proxy 容器。为了简洁起见，上述输出有所省略。 验证 sleep deployment 中包含 sidecar。injected-version 对应于注入的 sidecar 镜像的版本和镜像的 TAG。在您的设置的可能会有所不同。 echo $(kubectl get deployment sleep -o jsonpath='{.metadata.annotations.sidecar\\.istio\\.io\\/status}') injected-version-9c7c291eab0a522f8033decd0f5b031f5ed0e126 你可以查看包含注入的容器和挂载的 volume 的完整 deployment 信息。 kubectl get deployment sleep -o yaml 自动注入 sidecar Istio sidecar 可以在部署之前使用 Kubernetes 中一个名为 Initializer 的 Alpha 功能自动注入到 Pod 中。 注意：Kubernetes InitializerConfiguration没有命名空间，适用于整个集群的工作负载。不要在共享测试环境中启用此功能。 前置条件 Initializer 需要在集群设置期间显示启用，如 此处 所述。 假设集群中已启用RBAC，则可以在不同环境中启用初始化程序，如下所示： GKE gcloud container clusters create NAME \\ --enable-kubernetes-alpha \\ --machine-type=n1-standard-2 \\ --num-nodes=4 \\ --no-enable-legacy-authorization \\ --zone=ZONE IBM Bluemix kubernetes v1.7.4 或更高版本的集群已默认启用 initializer。 Minikube Minikube v0.22.1 或更高版本需要为 GenericAdmissionWebhook 功能配置适当的证书。获取最新版本： https://github.com/kubernetes/minikube/releases. minikube start \\ --extra-config=apiserver.Admission.PluginNames=\"Initializers,NamespaceLifecycle,LimitRanger,ServiceAccount,DefaultStorageClass,GenericAdmissionWebhook,ResourceQuota\" \\ --kubernetes-version=v1.7.5 安装 您现在可以从 Istio 安装根目录设置 Istio Initializer。 kubectl apply -f install/kubernetes/istio-initializer.yaml 将会创建下列资源： istio-sidecar InitializerConfiguration 资源，指定 Istio sidecar 注入的资源。默认情况下 Istio sidecar 将被注入到 deployment、 statefulset、 job 和 daemonset中。 istio-inject ConfigMap，initializer 的默认注入策略，一组初始化 namespace，以及注入时使用的模版参数。这些配置的详细说明请参考 配置选项。 istio-initializer Deployment，运行 initializer 控制器。 istio-initializer-service-account ServiceAccount，用于 istio-initializer deployment。ClusterRole 和 ClusterRoleBinding 在 install/kubernetes/istio.yaml 中定义。注意所有的资源类型都需要有 initialize 和 patch 。正式处于这个原因，initializer 要作为 deployment 的一部分来运行而不是嵌入到其它控制器中，例如 istio-pilot。 验证 为了验证 sidecar 是否成功注入，为上面的 sleep 服务创建 deployment 和 service。 kubectl apply -f samples/sleep/sleep.yaml 验证 sleep deployment 中包含 sidecar。injected-version 对应于注入的 sidecar 镜像的版本和镜像的 TAG。在您的设置的可能会有所不同。 $ echo $(kubectl get deployment sleep -o jsonpath='{.metadata.annotations.sidecar\\.istio\\.io\\/status}') injected-version-9c7c291eab0a522f8033decd0f5b031f5ed0e126 你可以查看包含注入的容器和挂载的 volume 的完整 deployment 信息。 kubectl get deployment sleep -o yaml 了解发生了什么 以下是将工作负载提交给 Kubernetes 后发生的情况： 1) kubernetes 将 sidecar.initializer.istio.io 添加到工作负载的 pending initializer 列表中。 2) istio-initializer 控制器观察到有一个新的未初始化的工作负载被创建了。pending initializer 列表中的第一个个将作为 sidecar.initializer.istio.io 的名称。 3) istio-initializer 检查它是否负责初始化 namespace 中的工作负载。如果没有为该 namespace 配置 initializer，则不需要做进一步的工作，而且 initializer 会忽略工作负载。默认情况下，initializer 负责所有的 namespace（参考 配置选项）。 4) istio-initializer 将自己从 pending initializer 中移除。如果 pending initializer 列表非空，则 Kubernetes 不回结束工作负载的创建。错误配置的 initializer 意味着破损的集群。 5) istio-initializer 检查 mesh 的默认注入策略，并检查所有单个工作负载的策略负载值，以确定是否需要注入 sidecar。 6) istio-initializer 向工作负载中注入 sidecar 模板，然后通过 PATCH 向 kubernetes 提交。 7) kubernetes 正常的完成了工作负载的创建，并且工作负载中已经包含了注入的 sidecar。 配置选项 istio-initializer 具有用于注入的全局默认策略以及每个工作负载覆盖配置。全局策略由 istio-inject ConfigMap 配置（请参见下面的示例）。Initializer pod 必须重新启动以采用新的配置更改。 apiVersion: v1 kind: ConfigMap metadata: name: istio-inject namespace: istio-system data: config: |- policy: \"enabled\" namespaces: [\"\"] # everything, aka v1.NamepsaceAll, aka cluster-wide # excludeNamespaces: [\"ns1\", \"ns2\"] initializerName: \"sidecar.initializer.istio.io\" params: initImage: docker.io/istio/proxy_init:0.2.6 proxyImage: docker.io/istio/proxy:0.2.6 verbosity: 2 version: 0.2.6 meshConfigMapName: istio imagePullPolicy: IfNotPresent 下面是配置中的关键参数： policy off - 禁用 initializer 修改资源。pending 的 sidecar.initializer.istio.io initializer 将被删除以避免创建阻塞资源。 disable - initializer 不会注入 sidecar 到 watch 的所有 namespace 的资源中。启用 sidecar 注入请将 sidecar.istio.io/inject 注解的值设置为 true。 enable - initializer 将会注入 sidecar 到 watch 的所有 namespace 的资源中。禁用 sidecar 注入请将 sidecar.istio.io/inject 注解的值设置为 false。 namespaces 要 watch 和初始化的 namespace 列表。特殊的 \"\" namespace 对应于 v1.NamespaceAll 并配置初始化程序以初始化所有 namespace。kube-system、kube-publice 和 istio-system 被免除初始化。 excludeNamespaces 从 Istio initializer 中排除的 namespace 列表。不可以定义为 v1.NamespaceAll 或者与 namespaces 一起定义。 initializerName 这必须与 InitializerConfiguration 中 initializer 设定项的名称相匹配。Initializer 只处理匹配其配置名称的工作负载。 params 这些参数允许您对注入的 sidecar 进行有限的更改。更改这些值不会影响已部署的工作负载。 重写自动注入 单个工作负载可以通过使用 sidecar.istio.io/inject 注解重写全局策略。如果注解被省略，则使用全局策略。 如果注解的值是 true，则不管全局策略如何，sidecar 都将被注入。 如果注解的值是 false，则不管全局策略如何，sidecar 都不会被注入。 下表显示全局策略和每个工作负载覆盖的组合。 policy workload annotation injected off N/A no disabled omitted no disabled false no disabled true yes enabled omitted yes enabled false no enabled true yes 例如，即使全局策略是 disable，下面的 deployment 也会被注入sidecar。 apiVersion: extensions/v1beta1 kind: Deployment metadata: name: myapp annotations: sidecar.istio.io/inject: \"true\" spec: replicas: 1 template: ... 这是在包含 Istio 和非 Istio 服务的混合群集中使用自动注入的好方法。 卸载 Initializer 运行下面的命令，删除 Istio initializer： kubectl delete -f install/kubernetes/istio-initializer.yaml 注意上述命令并不会删除已注入到 Pod 中的 sidecar。要想删除这些 sidecar，需要在不使用 initializer 的情况下重新部署这些 pod。 拓展 Istio Mesh 将虚拟机或裸机集成到部署在 kubernetes 集群上的 Istio mesh 中的说明。 前置条件 按照 安装指南 在 kubernetes 集群上安装 Istio service mesh。 机器必须具有到 mesh 端点的 IP 地址连接。这通常需要一个 VPC 或者 VPN，以及一个向端点提供直接路由（没有 NAT 或者防火墙拒绝）的容器网络。及其不需要访问有 Kubernetes 分配的 cluster IP。 虚拟机必须可以访问到 Istio 控制平面服务（如Pilot、Mixer、CA）和 Kubernetes DNS 服务器。通常使用 内部负载均衡器 来实现。 您也可以使用 NodePort，在虚拟机上运行 Istio 的组件，或者使用自定义网络配置，有几个单独的文档会涵盖这些高级配置。 安装步骤 安装过程包括准备用于拓展的 mesh 和安装和配置虚拟机。 install/tools/setupMeshEx.sh ：这是一个帮助大家设置 kubernetes 环境的示例脚本。检查脚本内容和支持的环境变量（如 GCP_OPTS）。 install/tools/setupIstioVM.sh：这是一个用于配置主机环境的示例脚本。 您应该根据您的配置工具和DNS要求对其进行自定义。 准备要拓展的 Kubernetes 集群： 为 Kube DNS、Pilot、Mixer 和 CA 安装内部负载均衡器（ILB）。每个云供应商的配置都有所不同，根据具体情况修改注解。 0.2.7 版本的 YAML 文件的 DNS ILB 的 namespace 配置不正确。 使用 这一个 替代。 setupMeshEx.sh 中也有错误。使用上面链接中的最新文件或者从 GitHub.com/istio/istio 克隆。 kubectl apply -f install/kubernetes/mesh-expansion.yaml 生成要部署到虚拟机上的 Istio cluster.env 配置。该文件中包含要拦截的 cluster IP 地址范围。 export GCP_OPTS=\"--zone MY_ZONE --project MY_PROJECT\" install/tools/setupMeshEx.sh generateClusterEnv MY_CLUSTER_NAME 该示例生成的文件： cat cluster.env ISTIO_SERVICE_CIDR=10.63.240.0/20 产生虚拟机使用的 DNS 配置文件。这样可以让虚拟机上的应用程序解析到集群中的服务名称，这些名称将被 sidecar 拦截和转发。 # Make sure your kubectl context is set to your cluster install/tools/setupMeshEx.sh generateDnsmasq 该示例生成的文件： cat kubedns server=/svc.cluster.local/10.150.0.7 address=/istio-mixer/10.150.0.8 address=/istio-pilot/10.150.0.6 address=/istio-ca/10.150.0.9 address=/istio-mixer.istio-system/10.150.0.8 address=/istio-pilot.istio-system/10.150.0.6 address=/istio-ca.istio-system/10.150.0.9 设置机器 例如，您可以使用下面的“一条龙”脚本复制和安装配置： # 检查该脚本看看它是否满足您的需求 # 在 Mac 上，使用 brew install base64 或者 set BASE64_DECODE=\"/usr/bin/base64 -D\" export GCP_OPTS=\"--zone MY_ZONE --project MY_PROJECT\" install/tools/setupMeshEx.sh machineSetup VM_NAME 或者等效得手动安装步骤如下： ------ 手动安装步骤开始 ------ 将配置文件和 Istio 的 Debian 文件复制到要加入到集群的每台机器上。重命名为 /etc/dnsmasq.d/kubedns 和/var/lib/istio/envoy/cluster.env。 配置和验证 DNS 配置。需要安装 dnsmasq 或者直接将其添加到 /etc/resolv.conf 中，或者通过 DHCP 脚本。验证配置是否有效，检查虚拟机是否可以解析和连接到 pilot，例如： 在虚拟机或外部主机上： host istio-pilot.istio-system 产生的消息示例： # Verify you get the same address as shown as \"EXTERNAL-IP\" in 'kubectl get svc -n istio-system istio-pilot-ilb' istio-pilot.istio-system has address 10.150.0.6 检查是否可以解析 cluster IP。实际地址取决您的 deployment： host istio-pilot.istio-system.svc.cluster.local. 该示例产生的消息： istio-pilot.istio-system.svc.cluster.local has address 10.63.247.248 同样检查 istio-ingress： host istio-ingress.istio-system.svc.cluster.local. 该示例产生的消息： istio-ingress.istio-system.svc.cluster.local has address 10.63.243.30 验证连接性，检查迅即是否可以连接到 Pilot 的端点： curl 'http://istio-pilot.istio-system:8080/v1/registration/istio-pilot.istio-system.svc.cluster.local|http-discovery' { \"hosts\": [ { \"ip_address\": \"10.60.1.4\", \"port\": 8080 } ] } # 在虚拟机上使用上面的地址。将直接连接到运行 istio-pilot 的 pod。 curl 'http://10.60.1.4:8080/v1/registration/istio-pilot.istio-system.svc.cluster.local|http-discovery' 提取出实话 Istio 认证的 secret 并将它复制到机器上。Istio 的默认安装中包括 CA，即使是禁用了自动 mTLS 设置（她为每个 service account 创建 secret，secret 命名为 istio.）也会生成 Istio secret。建议您执行此步骤，以便日后启用 mTLS，并升级到默认启用 mTLS 的未来版本。 # ACCOUNT 默认是 'default'，SERVICE_ACCOUNT 是环境变量 # NAMESPACE 默认为当前 namespace，SERVICE_NAMESPACE 是环境变量 # （这一步由 machineSetup 完成） # 在 Mac 上执行 brew install base64 或者 set BASE64_DECODE=\"/usr/bin/base64 -D\" install/tools/setupMeshEx.sh machineCerts ACCOUNT NAMESPACE 生成的文件（key.pem, root-cert.pem, cert-chain.pem）必须拷贝到每台主机的 /etc/certs 目录，并且让 istio-proxy 可读。 安装 Istio Debian 文件，启动 istio 和 istio-auth-node-agent 服务。 从 github releases 获取 Debian 安装包： # 注意：在软件源配置好后，下面的额命令可以使用 'apt-get' 命令替代。 source istio.VERSION # defines version and URLs env var curl -L ${PILOT_DEBIAN_URL}/istio-agent.deb > ${ISTIO_STAGING}/istio-agent.deb curl -L ${AUTH_DEBIAN_URL}/istio-auth-node-agent.deb > ${ISTIO_STAGING}/istio-auth-node-agent.deb curl -L ${PROXY_DEBIAN_URL}/istio-proxy.deb > ${ISTIO_STAGING}/istio-proxy.deb dpkg -i istio-proxy-envoy.deb dpkg -i istio-agent.deb dpkg -i istio-auth-node-agent.deb systemctl start istio systemctl start istio-auth-node-agent ------ 手动安装步骤结束 ------ 安装完成后，机器就能访问运行在 Kubernetes 集群上的服务或者其他的 mesh 拓展的机器。 # 假设您在 'bookinfo' namespace 下安装的 bookinfo curl productpage.bookinfo.svc.cluster.local:9080 ... html content ... 检查进程是否正在运行： ps aux |grep istio root 6941 0.0 0.2 75392 16820 ? Ssl 21:32 0:00 /usr/local/istio/bin/node_agent --logtostderr root 6955 0.0 0.0 49344 3048 ? Ss 21:32 0:00 su -s /bin/bash -c INSTANCE_IP=10.150.0.5 POD_NAME=demo-vm-1 POD_NAMESPACE=default exec /usr/local/bin/pilot-agent proxy > /var/log/istio/istio.log istio-proxy istio-p+ 7016 0.0 0.1 215172 12096 ? Ssl 21:32 0:00 /usr/local/bin/pilot-agent proxy istio-p+ 7094 4.0 0.3 69540 24800 ? Sl 21:32 0:37 /usr/local/bin/envoy -c /etc/istio/proxy/envoy-rev1.json --restart-epoch 1 --drain-time-s 2 --parent-shutdown-time-s 3 --service-cluster istio-proxy --service-node sidecar~10.150.0.5~demo-vm-1.default~default.svc.cluster.local 检查 Istio auth-node-agent 是否健康： sudo systemctl status istio-auth-node-agent ● istio-auth-node-agent.service - istio-auth-node-agent: The Istio auth node agent Loaded: loaded (/lib/systemd/system/istio-auth-node-agent.service; disabled; vendor preset: enabled) Active: active (running) since Fri 2017-10-13 21:32:29 UTC; 9s ago Docs: http://istio.io/ Main PID: 6941 (node_agent) Tasks: 5 Memory: 5.9M CPU: 92ms CGroup: /system.slice/istio-auth-node-agent.service └─6941 /usr/local/istio/bin/node_agent --logtostderr Oct 13 21:32:29 demo-vm-1 systemd[1]: Started istio-auth-node-agent: The Istio auth node agent. Oct 13 21:32:29 demo-vm-1 node_agent[6941]: I1013 21:32:29.469314 6941 main.go:66] Starting Node Agent Oct 13 21:32:29 demo-vm-1 node_agent[6941]: I1013 21:32:29.469365 6941 nodeagent.go:96] Node Agent starts successfully. Oct 13 21:32:29 demo-vm-1 node_agent[6941]: I1013 21:32:29.483324 6941 nodeagent.go:112] Sending CSR (retrial #0) ... Oct 13 21:32:29 demo-vm-1 node_agent[6941]: I1013 21:32:29.862575 6941 nodeagent.go:128] CSR is approved successfully. Will renew cert in 29m59.137732603s 在拓展的 mesh 中的机器上运行服务 配置 sidecar 拦截端口。在 /var/lib/istio/envoy/sidecar.env 中通过 ISTIO_INBOUND_PORTS 环境变量配置。 例如（运行服务的虚拟机）： echo \"ISTIO_INBOUND_PORTS=27017,3306,8080\" > /var/lib/istio/envoy/sidecar.env systemctl restart istio 手动配置 selector-less 的 service 和 endpoint。“selector-less” service 用于那些不依托 Kubernetes pod 的 service。 例如，在有权限的机器上修改 Kubernetes 中的 service： # istioctl register servicename machine-ip portname:port istioctl -n onprem register mysql 1.2.3.4 3306 istioctl -n onprem register svc1 1.2.3.4 http:7000 安装完成后，Kubernetes pod 和其它 mesh 扩展将能够访问集群上运行的服务。 整合到一起 请参阅 拓展 BookInfo Mesh 指南。 部署 bookinfo 示例应用 该示例部署由四个单独的微服务组成的简单应用程序，用于演示Istio服务网格的各种功能。 概况 在本示例中，我们将部署一个简单的应用程序，显示书籍的信息，类似于网上书店的书籍条目。在页面上有书籍的描述、详细信息（ISBN、页数等）和书评。 BookInfo 应用程序包括四个独立的微服务： productpage：productpage(产品页面)微服务，调用 details 和 reviews 微服务来填充页面。 details：details 微服务包含书籍的详细信息。 reviews：reviews 微服务包含书籍的点评。它也调用 ratings 微服务。 ratings：ratings 微服务包含随书评一起出现的评分信息。 有3个版本的 reviews 微服务： 版本v1不调用 ratings 服务。 版本v2调用 ratings ，并将每个评级显示为1到5个黑色星。 版本v3调用 ratings ，并将每个评级显示为1到5个红色星。 应用程序的端到端架构如下所示。 图片 - BookInfo 该应用程序是多语言构建的，即这些微服务是用不同的语言编写的。值得注意的是，这些服务与 Istio 没有任何依赖关系，单这是个有趣的 Service Mesh 示例，特别是因为评论服务和众多的语言和版本。 开始之前 如果您还没有这样做，请按照与您的平台 安装指南 对应的说明安装Istio。 部署应用程序 使用 Istio 运行应用程序示例不需要修改应用程序本身。相反，我们只需要在支持 Istio 的环境中配置和运行服务， Envoy sidecar 将会注入到每个服务中。所需的命令和配置根据运行时环境的不同而有所不同，但在所有情况下，生成的部署将如下所示： 图片 - BookInfo 所有的微服务都将与一个 Envoy sidecar 一起打包，拦截这些服务的入站和出站的调用请求，提供通过 Istio 控制平面从外部控制整个应用的路由，遥测收集和策略执行所需的 hook。 要启动该应用程序，请按照以下对应于您的 Istio 运行时环境的说明进行操作。 在 Kubernetes 中运行 注意：如果您使用 GKE，清确保您的集群至少有 4 个标准的 GKE 节点。如果您使用 Minikube，请确保您至少有 4GB 内存。 将目录更改为 Istio 安装目录的根目录。 构建应用程序容器： 如果您使用 自动注入 sidecar 的方式部署的集群，那么只需要使用 kubectl 命令部署服务： kubectl apply -f samples/bookinfo/kube/bookinfo.yaml 如果您使用 手动注入 sidecar 的方式部署的集群，清使用下面的命令： kubectl apply -f 请注意，该 istioctl kube-inject 命令用于在创建部署之前修改 bookinfo.yaml 文件。这将把 Envoy 注入到 Kubernetes 资源。 上述命令启动四个微服务并创建网关入口资源，如下图所示。3 个版本的评论的服务 v1、v2、v3 都已启动。 请注意在实际部署中，随着时间的推移部署新版本的微服务，而不是同时部署所有版本。 确认所有服务和 pod 已正确定义并运行： kubectl get services 这将产生以下输出： NAME CLUSTER-IP EXTERNAL-IP PORT(S) AGE details 10.0.0.31 9080/TCP 6m istio-ingress 10.0.0.122 80:31565/TCP 8m istio-pilot 10.0.0.189 8080/TCP 8m istio-mixer 10.0.0.132 9091/TCP,42422/TCP 8m kubernetes 10.0.0.1 443/TCP 14d productpage 10.0.0.120 9080/TCP 6m ratings 10.0.0.15 9080/TCP 6m reviews 10.0.0.170 9080/TCP 6m 而且 kubectl get pods 将产生: NAME READY STATUS RESTARTS AGE details-v1-1520924117-48z17 2/2 Running 0 6m istio-ingress-3181829929-xrrk5 1/1 Running 0 8m istio-pilot-175173354-d6jm7 2/2 Running 0 8m istio-mixer-3883863574-jt09j 2/2 Running 0 8m productpage-v1-560495357-jk1lz 2/2 Running 0 6m ratings-v1-734492171-rnr5l 2/2 Running 0 6m reviews-v1-874083890-f0qf0 2/2 Running 0 6m reviews-v2-1343845940-b34q5 2/2 Running 0 6m reviews-v3-1813607990-8ch52 2/2 Running 0 6m 确定 ingress IP 和端口 如果您的 kubernetes 集群环境支持外部负载均衡器的话，可以使用下面的命令获取 ingress 的IP地址： kubectl get ingress -o wide 输出如下所示： NAME HOSTS ADDRESS PORTS AGE gateway * 130.211.10.121 80 1d Ingress 服务的地址是： export GATEWAY_URL=130.211.10.121:80 GKE：如果服务无法获取外部 IP，kubectl get ingress -o wide 会显示工作节点的列表。在这种情况下，您可以使用任何地址以及 NodePort 访问入口。但是，如果集群具有防火墙，则还需要创建防火墙规则以允许TCP流量到NodePort，您可以使用以下命令创建防火墙规则： export GATEWAY_URL=:$(kubectl get svc istio-ingress -n istio-system -o jsonpath='{.spec.ports[0].nodePort}') gcloud compute firewall-rules create allow-book --allow tcp:$(kubectl get svc istio-ingress -n istio-system -o jsonpath='{.spec.ports[0].nodePort}') IBM Bluemix Free Tier：在免费版的 Bluemix 的 kubernetes 集群中不支持外部负载均衡器。您可以使用工作节点的公共 IP，并通过 NodePort 来访问 ingress。工作节点的公共 IP可以通过如下命令获取： bx cs workers export GATEWAY_URL=:$(kubectl get svc istio-ingress -n istio-system -o jsonpath='{.spec.ports[0].nodePort}') Minikube：Minikube 不支持外部负载均衡器。您可以使用 ingress 服务的主机 IP 和 NodePort 来访问 ingress： export GATEWAY_URL=$(kubectl get po -l istio=ingress -o 'jsonpath={.items[0].status.hostIP}'):$(kubectl get svc istio-ingress -o 'jsonpath={.spec.ports[0].nodePort}') 在 Consul 或 Eureka 环境下使用 Docker 运行 切换到 Istio 的安装根目录下。 启动应用程序容器。 执行下面的命令测试 Consul： docker-compose -f samples/bookinfo/consul/bookinfo.yaml up -d 执行下面的命令测试 Eureka： docker-compose -f samples/bookinfo/eureka/bookinfo.yaml up -d 确认所有容器都在运行： docker ps -a 如果 Istio Pilot 容器终止了，重新执行上面的命令重新运行。 设置 GATEWAY_URL: export GATEWAY_URL=localhost:9081 下一步 使用以下 curl 命令确认 BookInfo 应用程序正在运行: curl -o /dev/null -s -w \"%{http_code}\\n\" http://${GATEWAY_URL}/productpage 200 你也可以通过在浏览器中打开 http://$GATEWAY_URL/productpage 页面访问 Bookinfo 网页。如果您多次刷新浏览器将在 productpage 中看到评论的不同的版本，它们会按照 round robin（红星、黑星、没有星星）的方式展现，因为我们还没有使用 Istio 来控制版本的路由。 现在，您可以使用此示例来尝试 Istio 的流量路由、故障注入、速率限制等功能。要继续的话，请参阅 Istio 指南，具体取决于您的兴趣。智能路由 是初学者入门的好方式。 清理 在完成 BookInfo 示例后，您可以卸载它，如下所示： 卸载 Kubernetes 环境 删除路由规则，终止应用程序 pod samples/bookinfo/kube/cleanup.sh 确认关闭 istioctl get routerules #-- there should be no more routing rules kubectl get pods #-- the BookInfo pods should be deleted 卸载 docker 环境 删除路由规则和应用程序容器 若使用 Consul 环境安装，执行下面的命令： samples/bookinfo/consul/cleanup.sh 若使用 Eureka 环境安装，执行下面的命令： samples/bookinfo/eureka/cleanup.sh 确认清理完成： istioctl get routerules #-- there should be no more routing rules docker ps -a #-- the BookInfo containers should be delete Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-21 21:18:30 "},"usecases/integrating-vms.html":{"url":"usecases/integrating-vms.html","title":"集成虚拟机","keywords":"","body":"集成虚拟机 该示例跨越 Kubernetes 集群和一组虚拟机上部署 Bookinfo 服务，描述如何使用 Istio service mesh 将此基础架构以单一 mesh 的方式操控。 注意：本文档还在建设中，并且只在 Google Cloud Platform 上进行过测试。 在 IBM Bluemix 或其它平台上，pod 的 overlay 网络跟虚拟机的网络是隔离的。 即使使用 Istio，虚拟机也不能直接与 Kubernetes Pod 进行通信。 概览 图片 - Bookinfo应用的拓展Mesh 开始之前 按照 安装指南 上的步骤部署 Istio。 部署 BookInfo 示例应用程序（在 bookinfo namespace 下）。 在 Istio 集群相同的项目下创建名为 vm-1 的虚拟机，并 加入到 Mesh。 在虚拟机上运行 mysql 我们将首先在虚拟机上安装 mysql，将其配置成评分服务的后台存储。 在虚拟机上执行： sudo apt-get update && sudo apt-get install -y mariadb-server sudo mysql # 授权 root 用户访问 GRANT ALL PRIVILEGES ON *.* TO 'root'@'localhost' IDENTIFIED BY 'password' WITH GRANT OPTION; quit; sudo systemctl restart mysql 关于 Mysql 的详细配置请见： Mysql。 在虚拟机上执行下面的命令，向 mysql 中添加评分数据库。 # 向 mysql 数据库中添加评分数据库 curl -q https://raw.githubusercontent.com/istio/istio/master/samples/bookinfo/src/mysql/mysqldb-init.sql | mysql -u root -ppassword 为了便于直观地查看bookinfo应用程序输出的差异，可以更改使用以下命令生成的评分： # 查看评分 mysql -u root -ppassword test -e \"select * from ratings;\" +----------+--------+ | ReviewID | Rating | +----------+--------+ | 1 | 5 | | 2 | 4 | +----------+--------+ # 修改评分 mysql -u root -ppassword test -e \"update ratings set rating=1 where reviewid=1;select * from ratings;\" +----------+--------+ | ReviewID | Rating | +----------+--------+ | 1 | 1 | | 2 | 4 | +----------+--------+ 找出将添加到 mesh 中的虚拟机的 IP 地址 在虚拟机上执行： hostname -I 将 mysql 服务注册到 mesh 中 在一台可以访问 istioctl 命令的主机上，注册该虚拟机和 mysql db service： istioctl register -n vm mysqldb 3306 # 示例输出 $ istioctl register mysqldb 192.168.56.112 3306 I1015 22:24:33.846492 15465 register.go:44] Registering for service 'mysqldb' ip '192.168.56.112', ports list [{3306 mysql}] I1015 22:24:33.846550 15465 register.go:49] 0 labels ([]) and 1 annotations ([alpha.istio.io/kubernetes-serviceaccounts=default]) W1015 22:24:33.866410 15465 register.go:123] Got 'services \"mysqldb\" not found' looking up svc 'mysqldb' in namespace 'default', attempting to create it W1015 22:24:33.904162 15465 register.go:139] Got 'endpoints \"mysqldb\" not found' looking up endpoints for 'mysqldb' in namespace 'default', attempting to create them I1015 22:24:33.910707 15465 register.go:180] No pre existing exact matching ports list found, created new subset {[{192.168.56.112 nil}] [] [{mysql 3306 }]} I1015 22:24:33.921195 15465 register.go:191] Successfully updated mysqldb, now with 1 endpoints 集群管理 如果你之前在 kubernetes 上运行过 mysql，您需要将 kubernetes 的 mysql service 移除： kubectl delete service mysql 执行 istioctl 来配置 service（在您的 admin 机器上）： istioctl register mysql IP mysql:PORT 注意：mysqldb 虚拟机不需要也不应该有特别的 kubernetes 权限。 使用 mysql 服务 bookinfo 中的评分服务将使用机器上的数据库。要验证它是否有效，请创建使用虚拟机上的 mysql db 的评分服务的 v2 版本。然后指定强制评论服务使用评分 v2 版本的路由规则。 # 创建使用 mysql 后端的评分服务版本 istioctl kube-inject -n bookinfo -f samples/bookinfo/kube/bookinfo-ratings-v2-mysql-vm.yaml | kubectl apply -n bookinfo -f - # 强制 bookinfo 使用评分后端的路有规则 istioctl create -n bookinfo -f samples/bookinfo/kube/route-rule-ratings-mysql-vm.yaml 您可以验证 bookinfo 应用程序的输出结果，显示来自 Reviewer1 的 1 星级和来自 Reviewer2 的 4 星级，或者更改虚拟机上的评分并查看结果。 同时，您还可以在 RawVM MySQL 的文档中找到一些故障排查和其它信息。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-07 23:54:19 "},"usecases/linkerd.html":{"url":"usecases/linkerd.html","title":"Linkerd","keywords":"","body":"Linkerd简介 Linkerd是一个用于云原生应用的开源、可扩展的service mesh（一般翻译成服务网格，还有一种说法叫”服务啮合层“，见Istio：用于微服务的服务啮合层）。 Linkerd是什么 Linkerd的出现是为了解决像twitter、google这类超大规模生产系统的复杂性问题。Linkerd不是通过控制服务之间的通信机制来解决这个问题，而是通过在服务实例之上添加一个抽象层来解决的。 图片 - source https://linkerd.io Linkerd负责跨服务通信中最困难、易出错的部分，包括延迟感知、负载平衡、连接池、TLS、仪表盘、请求路由等——这些都会影响应用程序伸缩性、性能和弹性。 如何运行 Linkerd作为独立代理运行，无需特定的语言和库支持。应用程序通常会在已知位置运行linkerd实例，然后通过这些实例代理服务调用——即不是直接连接到目标服务，服务连接到它们对应的linkerd实例，并将它们视为目标服务。 在该层上，linkerd应用路由规则，与现有服务发现机制通信，对目标实例做负载均衡——与此同时调整通信并报告指标。 通过延迟调用linkerd的机制，应用程序代码与以下内容解耦： 生产拓扑 服务发现机制 负载均衡和连接管理逻辑 应用程序也将从一致的全局流量控制系统中受益。这对于多语言应用程序尤其重要，因为通过库来实现这种一致性是非常困难的。 Linkerd实例可以作为sidecar（既为每个应用实体或每个主机部署一个实例）来运行。 由于linkerd实例是无状态和独立的，因此它们可以轻松适应现有的部署拓扑。它们可以与各种配置的应用程序代码一起部署，并且基本不需要去协调它们。 参考 Buoyant发布服务网格Linkerd的1.0版本 Linkerd documentation Istio：用于微服务的服务啮合层 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-10-22 22:32:51 "},"usecases/linkerd-user-guide.html":{"url":"usecases/linkerd-user-guide.html","title":"Linkerd 使用指南","keywords":"","body":"Linkerd 使用指南 前言 Linkerd 作为一款 service mesh 与kubernetes 结合后主要有以下几种用法： 作为服务网关，可以监控 kubernetes 中的服务和实例 使用 TLS 加密服务 通过流量转移到持续交付 开发测试环境（Eat your own dog food）、Ingress 和边缘路由 给微服务做 staging 分布式 tracing 作为 Ingress controller 使用 gRPC 更方便 以下我们着重讲解在 kubernetes 中如何使用 linkerd 作为 kubernetes 的 Ingress controller，并作为边缘节点代替 Traefik 的功能，详见 边缘节点的配置。 准备 安装测试时需要用到的镜像有： buoyantio/helloworld:0.1.4 buoyantio/jenkins-plus:2.60.1 buoyantio/kubectl:v1.4.0 buoyantio/linkerd:1.1.2 buoyantio/namerd:1.1.2 buoyantio/nginx:1.10.2 linkerd/namerctl:0.8.6 openzipkin/zipkin:1.20 tutum/dnsutils:latest 这些镜像可以直接通过 Docker Hub 获取，我将它们下载下来并上传到了自己的私有镜像仓库 sz-pg-oam-docker-hub-001.tendcloud.com 中，下文中用到的镜像皆来自我的私有镜像仓库，yaml 配置见 linkerd 目录，并在使用时将配置中的镜像地址修改为你自己的。 部署 首先需要先创建 RBAC，因为使用 namerd 和 ingress 时需要用到。 $ kubectl create -f linkerd-rbac-beta.yml Linkerd 提供了 Jenkins 示例，在部署的时候使用以下命令： $ kubectl create -f jenkins-rbac-beta.yml $ kubectl create -f jenkins.yml 访问 http://jenkins.jimmysong.io 图片 - Jenkins pipeline 图片 - Jenkins config 注意：要访问 Jenkins 需要在 Ingress 中增加配置，下文会提到。 在 kubernetes 中使用 Jenkins 的时候需要注意 Pipeline 中的配置： def currentVersion = getCurrentVersion() def newVersion = getNextVersion(currentVersion) def frontendIp = kubectl(\"get svc l5d -o jsonpath=\\\"{.status.loadBalancer.ingress[0].*}\\\"\").trim() def originalDst = getDst(getDtab()) frontendIP 的地址要配置成 service 的 Cluster IP ，因为我们没有用到LoadBalancer。 需要安装 namerd，namerd 负责 dtab 信息的存储，当然也可以存储在 etcd、consul中。dtab 保存的是路由规则信息，支持递归解析，详见 dtab。 流量切换主要是通过 dtab 来实现的，通过在 HTTP 请求的 header 中增加 l5d-dtab 和 Host 信息可以对流量分离到 kubernetes 中的不同 service 上。 遇到的问题 Failed with the following error(s) Error signal dtab is already marked as being deployed! 因为该 dtab entry 已经存在，需要删除后再运行。 访问 http://namerd.jimmysong.io 图片 - namerd dtab 保存在 namerd 中，该页面中的更改不会生效，需要使用命令行来操作。 使用 namerctl 来操作。 $ namerctl --base-url http://namerd-backend.jimmysong.io dtab update internal file 注意：update 时需要将更新文本先写入文件中。 部署 Linkerd 直接使用 yaml 文件部署，注意修改镜像仓库地址。 # 创建 namerd $ kubectl create -f namerd.yaml # 创建 ingress $ kubectl create -f linkerd-ingress.yml # 创建测试服务 hello-world $ kubectl create -f hello-world.yml # 创建 API 服务 $ kubectl create -f api.yml # 创建测试服务 world-v2 $ kubectl create -f world-v2.yml 为了在本地调试 linkerd，我们将 linkerd 的 service 加入到 ingress 中，详见 边缘节点配置。 在 Ingress 中增加如下内容： - host: linkerd.jimmysong.io http: paths: - path: / backend: serviceName: l5d servicePort: 9990 - host: linkerd-viz.jimmysong.io http: paths: - path: / backend: serviceName: linkerd-viz servicePort: 80 - host: l5d.jimmysong.io http: paths: - path: / backend: serviceName: l5d servicePort: 4141 - host: jenkins.jimmysong.io http: paths: - path: / backend: serviceName: jenkins servicePort: 80 在本地/etc/hosts中添加如下内容： 172.20.0.119 linkerd.jimmysong.io 172.20.0.119 linkerd-viz.jimmysong.io 172.20.0.119 l5d.jimmysong.io 测试路由功能 使用 curl 简单测试。 单条测试 $ curl -s -H \"Host: www.hello.world\" 172.20.0.120:4141 Hello (172.30.60.14) world (172.30.71.19)!!% 请注意请求返回的结果，表示访问的是 world-v1 service。 $ for i in $(seq 0 10000);do echo $i;curl -s -H \"Host: www.hello.world\" 172.20.0.120:4141;done 使用 ab test。 $ ab -c 4 -n 10000 -H \"Host: www.hello.world\" http://172.20.0.120:4141/ This is ApacheBench, Version 2.3 Copyright 1996 Adam Twiss, Zeus Technology Ltd, http://www.zeustech.net/ Licensed to The Apache Software Foundation, http://www.apache.org/ Benchmarking 172.20.0.120 (be patient) Completed 1000 requests Completed 2000 requests Completed 3000 requests Completed 4000 requests Completed 5000 requests Completed 6000 requests Completed 7000 requests Completed 8000 requests Completed 9000 requests Completed 10000 requests Finished 10000 requests Server Software: Server Hostname: 172.20.0.120 Server Port: 4141 Document Path: / Document Length: 43 bytes Concurrency Level: 4 Time taken for tests: 262.505 seconds Complete requests: 10000 Failed requests: 0 Total transferred: 2210000 bytes HTML transferred: 430000 bytes Requests per second: 38.09 [#/sec] (mean) Time per request: 105.002 [ms] (mean) Time per request: 26.250 [ms] (mean, across all concurrent requests) Transfer rate: 8.22 [Kbytes/sec] received Connection Times (ms) min mean[+/-sd] median max Connect: 36 51 91.1 39 2122 Processing: 39 54 29.3 46 585 Waiting: 39 52 20.3 46 362 Total: 76 105 96.3 88 2216 Percentage of the requests served within a certain time (ms) 50% 88 66% 93 75% 99 80% 103 90% 119 95% 146 98% 253 99% 397 100% 2216 (longest request) 监控 kubernets 中的服务与实例 访问 http://linkerd.jimmysong.io 查看流量情况 Outcoming 图片 - linkerd监控 Incoming 图片 - linkerd监控 访问 http://linkerd-viz.jimmysong.io 查看应用 metric 监控 图片 - linkerd性能监控 测试路由 测试在 http header 中增加 dtab 规则。 $ curl -H \"Host: www.hello.world\" -H \"l5d-dtab:/host/world => /srv/world-v2;\" 172.20.0.120:4141 Hello (172.30.60.14) earth (172.30.94.40)!! 请注意调用返回的结果，表示调用的是 world-v2 的 service。 另外再对比 ab test 的结果与 linkerd-viz 页面上的结果，可以看到结果一致。 但是我们可能不想把该功能暴露给所有人，所以可以在前端部署一个 nginx 来过滤 header 中的 l5d-dtab 打头的字段，并通过设置 cookie 的方式来替代 header 里的 l5d-dtab 字段。 $ http_proxy=http://172.20.0.120:4141 curl -s http:/hello Hello (172.30.60.14) world (172.30.71.19)!! 将 Linkerd 作为 Ingress controller 将 Linkerd 作为 kubernetes ingress controller 的方式跟将 Treafik 作为 ingress controller 的过程过程完全一样，可以直接参考 边缘节点配置。 架构如下图所示。 图片 - Linkerd ingress controller (图片来自 A Service Mesh for Kubernetes - Buoyant.io) 当然可以绕过 kubernetes ingress controller 直接使用 linkerd 作为边界路由，通过 dtab 和 linkerd 前面的 nginx 来路由流量。 参考 https://github.com/linkerd/linkerd-examples/ A Service Mesh for Kubernetes dtab Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-08-21 18:23:35 "},"usecases/conduit.html":{"url":"usecases/conduit.html","title":"Conduit","keywords":"","body":"Conduit - 基于Kubernetes的轻量级Service Mesh 2017年12月在得克萨斯州的Asdin，KubeCon和CloudNativeCon上，创造了Service Mesh这个词汇并开源了Linkerd的公司Buoyant，又开源了一款针对Kubernetes的超轻量Service Sesh——Conduit。它可以透明得管理服务运行时之间的通信，使得在Kubernetes上运行服务更加安全和可靠；它还具有不用修改任何应用程序代码即可改进应用程序的可观测性、可靠性及安全性等方面的特性。 Condiut与Linkerd的设计方式不同，它跟Istio一样使用的是Sidecar模式，但架构又没Istio那么复杂。Conduit只支持Kubernetes，且只支持HTTP2（包括gRPC）协议。 Conduit使用Rust和Go语言开发，GitHub地址https://github.com/runconduit/conduit 安装Conduit必须使用Kubernetes1.8以上版本。 参考 Conduit GitHub：https://github.com/runconduit/conduit 关于Conduit的更多资源请参考官方网站：https://conduit.io/ 另外，我们翻译Conduit的官方文档见：https://github.com/doczhcn/conduit 关于Service Mesh的更多内容请访问Service Mesh中国：http://www.servicemesh.cn/ Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-10 21:53:24 "},"usecases/conduit-overview.html":{"url":"usecases/conduit-overview.html","title":"Condiut概览","keywords":"","body":"Conduit 概览 Conduit是一款针对Kubernetes的超轻量级的service mesh。它可以透明得管理服务运行时之间的通信，使得在Kubernetes上运行服务更加安全和可靠；它还具有不用修改任何应用程序代码即可改进应用程序的可观测性、可靠性及安全性等方面的特性。 本文档将从一个较高层次介绍Conduit及其是如何工作的。如果不熟悉service mesh模型，或许可以先阅读William Morgan的概览文章 什么是service mesh？为什么需要它？ Conduit架构 Conduit service mesh部署到Kubernetes集群时有两个基本组件：数据平面和控制平面。数据平面承载服务实例间实际的应用请求流量，而控制平面则驱动数据平面并修改其行为（及访问聚合指标）提供API。Conduit CLI和Web UI使用此API并为人类提供符合人体工程学的控制。 让我们依次认识这些组件。 Conduit的数据平面由轻量级的代理组成，这些代理作为sidecar容器与每个服务代码的实例部署在一起。如果将服务“添加”到Conduit servie mesh中，必须重新部署该服务的pod，以便在每个pod中包含一个数据平面代理。（conduit inject 命令可以完成这个任务，以及透明地从每个实例通过代理汇集流量所需的配置工作。） 这些代理透明地拦截进出每个pod的通信，并增加诸如重试和超时、检测及加密（TLS）等特性，并根据相关策略来允许和拒绝请求。 这些代理并未设计成通过手动方式配置；相反，它们的行为是由控制平面驱动的。 Conduit控制平面是一组运行在专用Kubernetes名称空间（默认情况下为conduit）的服务。这些服务完成各种事情 - 聚合遥测数据，提供面向用户的API，向数据平面代理提供控制数据等。它们一起驱动数据平面的行为。 使用Conduit 为了支持Conduit的人机交互，可以使用Conduit CLI及web UI（也可以通过相关工具比如 kubectl）。CLI 和 web UI通过API驱动控制平面，而控制平面相应地驱动数据平面的行为。 控制平面API设计得足够通用，以便能基于此构建其他工具。比如，你可能希望从另外一个CI/CD系统来驱动API。 运行 conduit --help 可查看关于CLI功能的简短概述。 Conduit 与 Kubernetes Conduit设计用于无缝地融入现有的Kubernetes系统。该设计有几个重要特征。 第一，Conduit CLI（conduit）设计成尽可能地与 kubectl 一起使用。比如，conduit install 和 conduit inject 生成的Kubernetes配置，被设计成直接送入kubectl。这是为了在service mesh和编排系统之间提供一个明确的分工，并且使得Conduit适配现有的Kubernetes工作流程。 第二，Kubernetes中Conduit的核心词是Deployment，而不是Service。举个例子，conduit inject 增加一个Deployment，Conduit web UI会显示这些Deployment，并且每个Deployment都会给出聚合的性能指标。这是因为单个pod可以是任意数量Service的一部分，而这会导致流量与pod之间出现复杂的映射。相比之下，Deployment要求单个pod只能属于一个Deployment的一部分。通过基于Deployment而不是Service来构建，流量与pod间的映射就总是清晰的。 这两个设计特性能很好地组合。比如，conduit inject可用于一个运行的Deployment，因为当它更新Deployment时， Kubernetes会回滚pod以包括数据平面代理。 扩展Conduit的行为 Conduit控制平面还为构建自定义功能提供了一个便捷的入口。Conduit最初发布时并不支持这一点，在不远的将来，通过编写gRPC插件来作为控制平面的一部分运行，将能扩展Conduit的功能，而无需重新编译Conduit。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-11 14:19:23 "},"usecases/conduit-installation.html":{"url":"usecases/conduit-installation.html","title":"安装Conduit","keywords":"","body":"安装Conduit 本文档指导您如何在kubernetes上安装Conduit service mesh。 前提条件 kubernetes版本为1.8或以上 用到的镜像如下： buoyantio/kubectl:v1.6.2 gcr.io/runconduit/controller:v0.1.0 gcr.io/runconduit/web:v0.1.0 prom/prometheus:v1.8.1 其中位于gcr.io的镜像我备份到了DockerHub： jimmysong/runconduit-web:v0.1.0 jimmysong/runconduit-controller:v0.1.0 另外两个镜像本身就可以从DockerHub上下载。 部署 到release页面上下载conduit的二进制文件。 使用conduit install命令生成了用于部署到kubernetes中yaml文件，然后修改文件中的镜像仓库地址为你自己的镜像地址。 conduit install>conduit-0.1.0.yaml # 修改完镜像地址执行 kubectl apply -f conduit-0.1.0.yaml 修改后的yaml文件见：conduit-0.1.0.yaml。 注意：Conduit官方给出的yaml文件中不包括RBAC授权，我重新修改了，增加了RBAC和ServiceAccount。 使用kubectl proxy来开放外网访问conduit dashboard： kubectl proxy --address='172.20.0.113' --port=8001 --accept-hosts='^*$' 在浏览器中访问http://172.20.0.113:8001/api/v1/namespaces/conduit/services/web:http/proxy/servicemesh将看到如下页面： 图片 - Conduit dashboard Conduit inject Conduit注入的时候需要用到如下两个镜像： gcr.io/runconduit/proxy:v0.1.0 gcr.io/runconduit/proxy-init:v0.1.0 我将其备份到了DockerHub： jimmysong/runconduit-proxy:v0.1.0 jimmysong/runconduit-proxy-init:v0.1.0 查看conduit向yaml文件中注入了哪些配置，我们使用my-nginx.yaml为例： conduit inject --init-image sz-pg-oam-docker-hub-001.tendcloud.com/library/runconduit-proxy-init --proxy-image sz-pg-oam-docker-hub-001.tendcloud.com/library/runconduit-proxy my-nginx.yaml|kubectl apply -f - 注意：只需要指定镜像名称即可，tag与使用的conduit server版本相同，会自动注入。 my-nginx.yaml的内容如下： apiVersion: extensions/v1beta1 kind: Deployment metadata: name: my-nginx spec: replicas: 2 template: metadata: labels: run: my-nginx spec: containers: - name: my-nginx image: sz-pg-oam-docker-hub-001.tendcloud.com/library/nginx:1.9 ports: - containerPort: 80 --- apiVersion: v1 kind: Service metadata: name: my-nginx labels: app: my-nginx spec: ports: - port: 80 protocol: TCP name: http selector: run: my-nginx Conduit自动注入后生成的新的yaml文件内容如下： apiVersion: extensions/v1beta1 kind: Deployment metadata: creationTimestamp: null name: my-nginx spec: replicas: 2 strategy: {} template: metadata: annotations: conduit.io/created-by: conduit/cli v0.1.0 conduit.io/proxy-version: v0.1.0 creationTimestamp: null labels: conduit.io/controller: conduit conduit.io/plane: data run: my-nginx spec: containers: - image: sz-pg-oam-docker-hub-001.tendcloud.com/library/nginx:1.9 name: my-nginx ports: - containerPort: 80 resources: {} - env: - name: CONDUIT_PROXY_LOG value: trace,h2=debug,mio=info,tokio_core=info - name: CONDUIT_PROXY_CONTROL_URL value: tcp://proxy-api.conduit.svc.cluster.local:8086 - name: CONDUIT_PROXY_CONTROL_LISTENER value: tcp://0.0.0.0:4190 - name: CONDUIT_PROXY_PRIVATE_LISTENER value: tcp://127.0.0.1:4140 - name: CONDUIT_PROXY_PUBLIC_LISTENER value: tcp://0.0.0.0:4143 - name: CONDUIT_PROXY_NODE_NAME valueFrom: fieldRef: fieldPath: spec.nodeName - name: CONDUIT_PROXY_POD_NAME valueFrom: fieldRef: fieldPath: metadata.name - name: CONDUIT_PROXY_POD_NAMESPACE valueFrom: fieldRef: fieldPath: metadata.namespace image: sz-pg-oam-docker-hub-001.tendcloud.com/library/runconduit-proxy:v0.1.0 imagePullPolicy: IfNotPresent name: conduit-proxy ports: - containerPort: 4143 name: conduit-proxy resources: {} securityContext: runAsUser: 2102 initContainers: - args: - -p - \"4143\" - -o - \"4140\" - -i - \"4190\" - -u - \"2102\" image: sz-pg-oam-docker-hub-001.tendcloud.com/library/runconduit-proxy-init:v0.1.0 imagePullPolicy: IfNotPresent name: conduit-init resources: {} securityContext: capabilities: add: - NET_ADMIN privileged: false status: {} --- apiVersion: v1 kind: Service metadata: name: my-nginx labels: app: my-nginx spec: ports: - port: 80 protocol: TCP name: http selector: run: my-nginx --- 部署示例应用 使用下面的命令部署官方提供的示例应用： curl https://raw.githubusercontent.com/rootsongjc/kubernetes-handbook/master/manifests/conduit/emojivoto.yml | conduit inject --init-image sz-pg-oam-docker-hub-001.tendcloud.com/library/runconduit-proxy-init --proxy-image sz-pg-oam-docker-hub-001.tendcloud.com/library/runconduit-proxy - --skip-inbound-ports=80 | kubectl apply -f - 注意：其中使用的镜像地址已经改为我的私有镜像仓库地址，大家使用时请注意修改。 参考 Getting started - conduit.io Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-13 21:33:43 "},"usecases/big-data.html":{"url":"usecases/big-data.html","title":"大数据","keywords":"","body":"大数据 Kubernetes community中已经有了一个Big data SIG，大家可以通过这个SIG了解kubernetes结合大数据的应用。 在Swarm、Mesos、kubernetes这三种流行的容器编排调度架构中，Mesos对于大数据应用支持是最好的，spark原生就是运行在mesos上的，当然也可以容器化运行在kubernetes上。当前在kubernetes上运行大数据应用主要是spark应用。 Spark on Kubernetes Spark原生支持standalone、mesos和YARN的调度方式，当前kubernetes社区正在支持kubernetes的原生调度来运行spark - 。 当然您也可以在kubernetes直接部署spark on yarn或者spark standalone模式，仍然沿用已有的 Spark Standalone 使用spark standalone模式在kubernetes上运行，kubernetes不负责spark任务的调度。参考：Spark standalone on Kubernetes 这种模式中使用的spark本身负责任务调度，kubernetes只是作为一个spark的部署平台。 Spark on Yarn 使用StatefulSet和Headless serverless来实现，请参考 Spark on Yarn 这种模式中kubernetes依然不负责spark应用的调度，而只是将Yarn换了一个部署环境而已。 下面是架构图： 图片 - Spark on yarn with kubernetes Spark on Kubernetes Spark on kubernetes，使用kubernetes作为调度引擎，spark的任务直接调度到node节点上。参考：运行支持kubernetes原生调度的Spark程序 调度方式总结 下图显示的是三种调度方式中单个kubernetes node节点上运行的spark相关容器的调度情况。 图片 - 在kubernetes上使用多种调度方式 毫无疑问，使用kubernetes原生调度的spark任务才是最节省资源的。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-06 14:45:35 "},"usecases/spark-standalone-on-kubernetes.html":{"url":"usecases/spark-standalone-on-kubernetes.html","title":"Spark standalone on Kubernetes","keywords":"","body":"Spark standalone on Kubernetes 该项目是基于 Spark standalone 模式，对资源的分配调度还有作业状态查询的功能实在有限，对于让 spark 使用真正原生的 kubernetes 资源调度推荐大家尝试 https://github.com/apache-spark-on-k8s/ 本文中使用的镜像我已编译好上传到了时速云上，大家可以直接下载。 index.tenxcloud.com/jimmy/spark:1.5.2_v1 index.tenxcloud.com/jimmy/zeppelin:0.7.1 代码和使用文档见Github地址：https://github.com/rootsongjc/spark-on-kubernetes 本文中用到的 yaml 文件可以在 ../manifests/spark-standalone 目录下找到，也可以在上面的 https://github.com/rootsongjc/spark-on-kubernetes/ 项目的 manifests 目录下找到。 注意：时速云上本来已经提供的镜像 index.tenxcloud.com/google_containers/spark:1.5.2_v1 ，但是该镜像似乎有问题，下载总是失败。 在Kubernetes上启动spark 创建名为spark-cluster的namespace，所有操作都在该namespace中进行。 所有yaml文件都在manifests目录下。 $ kubectl create -f manifests/ 将会启动一个拥有三个worker的spark集群和zeppelin。 同时在该namespace中增加ingress配置，将spark的UI和zeppelin页面都暴露出来，可以在集群外部访问。 该ingress后端使用traefik。 访问spark 通过上面对ingress的配置暴露服务，需要修改本机的/etc/hosts文件，增加以下配置，使其能够解析到上述service。 172.20.0.119 zeppelin.traefik.io 172.20.0.119 spark.traefik.io 172.20.0.119是我设置的VIP地址，VIP的设置和traefik的配置请查看kubernetes-handbook。 spark ui 访问http://spark.traefik.io 图片 - spark master ui zeppelin ui 访问http://zepellin.treafik.io Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-08-30 14:16:22 "},"usecases/running-spark-with-kubernetes-native-scheduler.html":{"url":"usecases/running-spark-with-kubernetes-native-scheduler.html","title":"运行支持kubernetes原生调度的Spark程序","keywords":"","body":"运行支持kubernetes原生调度的Spark程序 TL;DR 这个主题比较大，该开源项目也还在不断进行中，我单独做了一个 web 用来记录 spark on kubernetes 的研究和最新进展见: https://jimmysong.io/spark-on-k8s 我们之前就在 kubernetes 中运行过 standalone 方式的 spark 集群，见 Spark standalone on kubernetes。 目前运行支持 kubernetes 原生调度的 spark 程序由 Google 主导，目前运行支持 kubernetes 原生调度的 spark 程序由 Google 主导，fork 自 spark 的官方代码库，见https://github.com/apache-spark-on-k8s/spark/ ，属于Big Data SIG。 参与到该项目的公司有： Bloomberg Google Haiwen Hyperpilot Intel Palantir Pepperdata Red Hat 为何使用 spark on kubernetes 使用kubernetes原生调度的spark on kubernetes是对现有的spark on yarn/mesos的资源使用方式的革命性的改进，主要表现在以下几点： Kubernetes原生调度：不再需要二层调度，直接使用kubernetes的资源调度功能，跟其他应用共用整个kubernetes管理的资源池； 资源隔离，粒度更细：原先yarn中的queue在spark on kubernetes中已不存在，取而代之的是kubernetes中原生的namespace，可以为每个用户分别指定一个namespace，限制用户的资源quota； 细粒度的资源分配：可以给每个spark任务指定资源限制，实际指定多少资源就使用多少资源，因为没有了像yarn那样的二层调度（圈地式的），所以可以更高效和细粒度的使用资源； 监控的变革：因为做到了细粒度的资源分配，所以可以对用户提交的每一个任务做到资源使用的监控，从而判断用户的资源使用情况，所有的metric都记录在数据库中，甚至可以为每个用户的每次任务提交计量； 日志的变革：用户不再通过yarn的web页面来查看任务状态，而是通过pod的log来查看，可将所有的kuberentes中的应用的日志等同看待收集起来，然后可以根据标签查看对应应用的日志； 所有这些变革都可以让我们更高效的获取资源、更有效率的获取资源！ Spark 概念说明 Apache Spark 是一个围绕速度、易用性和复杂分析构建的大数据处理框架。最初在2009年由加州大学伯克利分校的AMPLab开发，并于2010年成为Apache的开源项目之一。 在 Spark 中包括如下组件或概念： Application：Spark Application 的概念和 Hadoop 中的 MapReduce 类似，指的是用户编写的 Spark 应用程序，包含了一个 Driver 功能的代码和分布在集群中多个节点上运行的 Executor 代码； Driver：Spark 中的 Driver 即运行上述 Application 的 main() 函数并且创建 SparkContext，其中创建 SparkContext 的目的是为了准备Spark应用程序的运行环境。在 Spark 中由 SparkContext 负责和 ClusterManager 通信，进行资源的申请、任务的分配和监控等；当 Executor 部分运行完毕后，Driver负责将SparkContext 关闭。通常用 SparkContext 代表 Driver； Executor：Application运行在Worker 节点上的一个进程，该进程负责运行Task，并且负责将数据存在内存或者磁盘上，每个Application都有各自独立的一批Executor。在Spark on Yarn模式下，其进程名称为CoarseGrainedExecutorBackend，类似于 Hadoop MapReduce 中的 YarnChild。一个 CoarseGrainedExecutorBackend 进程有且仅有一个 executor 对象，它负责将 Task 包装成 taskRunner，并从线程池中抽取出一个空闲线程运行 Task。每个 CoarseGrainedExecutorBackend 能并行运行 Task 的数量就取决于分配给它的 CPU 的个数了； Cluster Manager：指的是在集群上获取资源的外部服务，目前有： Standalone：Spark原生的资源管理，由Master负责资源的分配； Hadoop Yarn：由YARN中的ResourceManager负责资源的分配； Worker：集群中任何可以运行Application代码的节点，类似于YARN中的NodeManager节点。在Standalone模式中指的就是通过Slave文件配置的Worker节点，在Spark on Yarn模式中指的就是NodeManager节点； 作业（Job）：包含多个Task组成的并行计算，往往由Spark Action催生，一个JOB包含多个RDD及作用于相应RDD上的各种Operation； 阶段（Stage）：每个Job会被拆分很多组 Task，每组任务被称为Stage，也可称TaskSet，一个作业分为多个阶段，每一个stage的分割点是action。比如一个job是：（transformation1 -> transformation1 -> action1 -> transformation3 -> action2），这个job就会被分为两个stage，分割点是action1和action2。 任务（Task）： 被送到某个Executor上的工作任务； Context：启动spark application的时候创建，作为Spark 运行时环境。 Dynamic Allocation（动态资源分配）：一个配置选项，可以将其打开。从Spark1.2之后，对于On Yarn模式，已经支持动态资源分配（Dynamic Resource Allocation），这样，就可以根据Application的负载（Task情况），动态的增加和减少executors，这种策略非常适合在YARN上使用spark-sql做数据开发和分析，以及将spark-sql作为长服务来使用的场景。Executor 的动态分配需要在 cluster mode 下启用 \"external shuffle service\"。 动态资源分配策略：开启动态分配策略后，application会在task因没有足够资源被挂起的时候去动态申请资源，这意味着该application现有的executor无法满足所有task并行运行。spark一轮一轮的申请资源，当有task挂起或等待 spark.dynamicAllocation.schedulerBacklogTimeout (默认1s)时间的时候，会开始动态资源分配；之后会每隔 spark.dynamicAllocation.sustainedSchedulerBacklogTimeout (默认1s)时间申请一次，直到申请到足够的资源。每次申请的资源量是指数增长的，即1,2,4,8等。之所以采用指数增长，出于两方面考虑：其一，开始申请的少是考虑到可能application会马上得到满足；其次要成倍增加，是为了防止application需要很多资源，而该方式可以在很少次数的申请之后得到满足。 架构设计 关于 spark standalone 的局限性与 kubernetes native spark 架构之间的区别请参考 Anirudh Ramanathan 在 2016年10月8日提交的 issue Support Spark natively in Kubernetes #34377。 简而言之，spark standalone on kubernetes 有如下几个缺点： 无法对于多租户做隔离，每个用户都想给 pod 申请 node 节点可用的最大的资源。 Spark 的 master／worker 本来不是设计成使用 kubernetes 的资源调度，这样会存在两层的资源调度问题，不利于与 kuberentes 集成。 而 kubernetes native spark 集群中，spark 可以调用 kubernetes API 获取集群资源和调度。要实现 kubernetes native spark 需要为 spark 提供一个集群外部的 manager 可以用来跟 kubernetes API 交互。 调度器后台 使用 kubernetes 原生调度的 spark 的基本设计思路是将 spark 的 driver 和 executor 都放在 kubernetes 的 pod 中运行，另外还有两个附加的组件：ResourceStagingServer 和 KubernetesExternalShuffleService。 Spark driver 其实可以运行在 kubernetes 集群内部（cluster mode）可以运行在外部（client mode），executor 只能运行在集群内部，当有 spark 作业提交到 kubernetes 集群上时，调度器后台将会为 executor pod 设置如下属性： 使用我们预先编译好的包含 kubernetes 支持的 spark 镜像，然后调用 CoarseGrainedExecutorBackend main class 启动 JVM。 调度器后台为 executor pod 的运行时注入环境变量，例如各种 JVM 参数，包括用户在 spark-submit 时指定的那些参数。 Executor 的 CPU、内存限制根据这些注入的环境变量保存到应用程序的 SparkConf 中。 可以在配置中指定 spark 运行在指定的 namespace 中。 参考：Scheduler backend 文档 安装指南 我们可以直接使用官方已编译好的 docker 镜像来部署，下面是官方发布的镜像： 组件 镜像 Spark Driver Image kubespark/spark-driver:v2.1.0-kubernetes-0.3.1 Spark Executor Image kubespark/spark-executor:v2.1.0-kubernetes-0.3.1 Spark Initialization Image kubespark/spark-init:v2.1.0-kubernetes-0.3.1 Spark Staging Server Image kubespark/spark-resource-staging-server:v2.1.0-kubernetes-0.3.1 PySpark Driver Image kubespark/driver-py:v2.1.0-kubernetes-0.3.1 PySpark Executor Image kubespark/executor-py:v2.1.0-kubernetes-0.3.1 我将这些镜像放到了我的私有镜像仓库中了。 还需要安装支持 kubernetes 的 spark 客户端，在这里下载：https://github.com/apache-spark-on-k8s/spark/releases 根据使用的镜像版本，我下载的是 v2.1.0-kubernetes-0.3.1 运行 SparkPi 测试 我们将任务运行在 spark-cluster 的 namespace 中，启动 5 个 executor 实例。 ./bin/spark-submit \\ --deploy-mode cluster \\ --class org.apache.spark.examples.SparkPi \\ --master k8s://https://172.20.0.113:6443 \\ --kubernetes-namespace spark-cluster \\ --conf spark.executor.instances=5 \\ --conf spark.app.name=spark-pi \\ --conf spark.kubernetes.driver.docker.image=sz-pg-oam-docker-hub-001.tendcloud.com/library/kubespark-spark-driver:v2.1.0-kubernetes-0.3.1 \\ --conf spark.kubernetes.executor.docker.image=sz-pg-oam-docker-hub-001.tendcloud.com/library/kubespark-spark-executor:v2.1.0-kubernetes-0.3.1 \\ --conf spark.kubernetes.initcontainer.docker.image=sz-pg-oam-docker-hub-001.tendcloud.com/library/kubespark-spark-init:v2.1.0-kubernetes-0.3.1 \\ local:///opt/spark/examples/jars/spark-examples_2.11-2.1.0-k8s-0.3.1-SNAPSHOT.jar 关于该命令参数的介绍请参考：https://apache-spark-on-k8s.github.io/userdocs/running-on-kubernetes.html 注意： 该 jar 包实际上是 spark.kubernetes.executor.docker.image 镜像中的。 这时候提交任务运行还是失败，报错信息中可以看到两个问题： Executor 无法找到 driver pod 用户 system:serviceaccount:spark-cluster:defaul 没有权限获取 spark-cluster 中的 pod 信息。 提了个 issue Failed to run the sample spark-pi test using spark-submit on the doc #478 需要为 spark 集群创建一个 serviceaccount 和 clusterrolebinding： kubectl create serviceaccount spark --namespace spark-cluster kubectl create rolebinding spark-edit --clusterrole=edit --serviceaccount=spark-cluster:spark --namespace=spark-cluster 该 Bug 将在新版本中修复。 用户指南 编译 Fork 并克隆项目到本地： git clone https://github.com/rootsongjc/spark.git 编译前请确保你的环境中已经安装 Java8 和 Maven3。 ## 第一次编译前需要安装依赖 build/mvn install -Pkubernetes -pl resource-managers/kubernetes/core -am -DskipTests ## 编译 spark on kubernetes build/mvn compile -Pkubernetes -pl resource-managers/kubernetes/core -am -DskipTests ## 发布 dev/make-distribution.sh --tgz -Phadoop-2.7 -Pkubernetes 第一次编译和发布的过程耗时可能会比较长，请耐心等待，如果有依赖下载不下来，请自备梯子。 详细的开发指南请见：https://github.com/apache-spark-on-k8s/spark/blob/branch-2.2-kubernetes/resource-managers/kubernetes/README.md 构建镜像 使用该脚本来自动构建容器镜像：https://github.com/apache-spark-on-k8s/spark/pull/488 将该脚本放在 dist 目录下，执行： ./build-push-docker-images.sh -r sz-pg-oam-docker-hub-001.tendcloud.com/library -t v2.1.0-kubernetes-0.3.1-1 build ./build-push-docker-images.sh -r sz-pg-oam-docker-hub-001.tendcloud.com/library -t v2.1.0-kubernetes-0.3.1-1 push 注意：如果你使用的 MacOS，bash 的版本可能太低，执行改脚本将出错，请检查你的 bash 版本： bash --version GNU bash, version 3.2.57(1)-release (x86_64-apple-darwin16) Copyright (C) 2007 Free Software Foundation, Inc. 上面我在升级 bash 之前获取的版本信息，使用下面的命令升级 bash： brew install bash 升级后的 bash 版本为 4.4.12(1)-release (x86_64-apple-darwin16.3.0)。 编译并上传镜像到我的私有镜像仓库，将会构建出如下几个镜像： sz-pg-oam-docker-hub-001.tendcloud.com/library/spark-driver:v2.1.0-kubernetes-0.3.1-1 sz-pg-oam-docker-hub-001.tendcloud.com/library/spark-resource-staging-server:v2.1.0-kubernetes-0.3.1-1 sz-pg-oam-docker-hub-001.tendcloud.com/library/spark-init:v2.1.0-kubernetes-0.3.1-1 sz-pg-oam-docker-hub-001.tendcloud.com/library/spark-shuffle:v2.1.0-kubernetes-0.3.1-1 sz-pg-oam-docker-hub-001.tendcloud.com/library/spark-executor:v2.1.0-kubernetes-0.3.1-1 sz-pg-oam-docker-hub-001.tendcloud.com/library/spark-executor-py:v2.1.0-kubernetes-0.3.1-1 sz-pg-oam-docker-hub-001.tendcloud.com/library/spark-driver-py:v2.1.0-kubernetes-0.3.1-1 运行测试 在 dist/bin 目录下执行 spark-pi 测试： ./spark-submit \\ --deploy-mode cluster \\ --class org.apache.spark.examples.SparkPi \\ --master k8s://https://172.20.0.113:6443 \\ --kubernetes-namespace spark-cluster \\ --conf spark.kubernetes.authenticate.driver.serviceAccountName=spark \\ --conf spark.executor.instances=5 \\ --conf spark.app.name=spark-pi \\ --conf spark.kubernetes.driver.docker.image=sz-pg-oam-docker-hub-001.tendcloud.com/library/spark-driver:v2.1.0-kubernetes-0.3.1-1 \\ --conf spark.kubernetes.executor.docker.image=sz-pg-oam-docker-hub-001.tendcloud.com/library/spark-executor:v2.1.0-kubernetes-0.3.1-1 \\ --conf spark.kubernetes.initcontainer.docker.image=sz-pg-oam-docker-hub-001.tendcloud.com/library/spark-init:v2.1.0-kubernetes-0.3.1-1 \\ local:///opt/spark/examples/jars/spark-examples_2.11-2.2.0-k8s-0.4.0-SNAPSHOT.jar 详细的参数说明见：https://apache-spark-on-k8s.github.io/userdocs/running-on-kubernetes.html 注意：local:///opt/spark/examples/jars/spark-examples_2.11-2.2.0-k8s-0.4.0-SNAPSHOT.jar 文件是在 spark-driver 和 spark-executor 镜像里的，在上一步构建镜像时已经构建并上传到了镜像仓库中。 执行日志显示： 2017-09-14 14:59:01 INFO Client:54 - Waiting for application spark-pi to finish... 2017-09-14 14:59:01 INFO LoggingPodStatusWatcherImpl:54 - State changed, new state: pod name: spark-pi-1505372339796-driver namespace: spark-cluster labels: spark-app-selector -> spark-f4d3a5d3ad964a05a51feb6191d50357, spark-role -> driver pod uid: 304cf440-991a-11e7-970c-f4e9d49f8ed0 creation time: 2017-09-14T06:59:01Z service account name: spark volumes: spark-token-zr8wv node name: N/A start time: N/A container images: N/A phase: Pending status: [] 2017-09-14 14:59:01 INFO LoggingPodStatusWatcherImpl:54 - State changed, new state: pod name: spark-pi-1505372339796-driver namespace: spark-cluster labels: spark-app-selector -> spark-f4d3a5d3ad964a05a51feb6191d50357, spark-role -> driver pod uid: 304cf440-991a-11e7-970c-f4e9d49f8ed0 creation time: 2017-09-14T06:59:01Z service account name: spark volumes: spark-token-zr8wv node name: 172.20.0.114 start time: N/A container images: N/A phase: Pending status: [] 2017-09-14 14:59:01 INFO LoggingPodStatusWatcherImpl:54 - State changed, new state: pod name: spark-pi-1505372339796-driver namespace: spark-cluster labels: spark-app-selector -> spark-f4d3a5d3ad964a05a51feb6191d50357, spark-role -> driver pod uid: 304cf440-991a-11e7-970c-f4e9d49f8ed0 creation time: 2017-09-14T06:59:01Z service account name: spark volumes: spark-token-zr8wv node name: 172.20.0.114 start time: 2017-09-14T06:59:01Z container images: sz-pg-oam-docker-hub-001.tendcloud.com/library/spark-driver:v2.1.0-kubernetes-0.3.1-1 phase: Pending status: [ContainerStatus(containerID=null, image=sz-pg-oam-docker-hub-001.tendcloud.com/library/spark-driver:v2.1.0-kubernetes-0.3.1-1, imageID=, lastState=ContainerState(running=null, terminated=null, waiting=null, additionalProperties={}), name=spark-kubernetes-driver, ready=false, restartCount=0, state=ContainerState(running=null, terminated=null, waiting=ContainerStateWaiting(message=null, reason=ContainerCreating, additionalProperties={}), additionalProperties={}), additionalProperties={})] 2017-09-14 14:59:03 INFO LoggingPodStatusWatcherImpl:54 - State changed, new state: pod name: spark-pi-1505372339796-driver namespace: spark-cluster labels: spark-app-selector -> spark-f4d3a5d3ad964a05a51feb6191d50357, spark-role -> driver pod uid: 304cf440-991a-11e7-970c-f4e9d49f8ed0 creation time: 2017-09-14T06:59:01Z service account name: spark volumes: spark-token-zr8wv node name: 172.20.0.114 start time: 2017-09-14T06:59:01Z container images: sz-pg-oam-docker-hub-001.tendcloud.com/library/spark-driver:v2.1.0-kubernetes-0.3.1-1 phase: Running status: [ContainerStatus(containerID=docker://5c5c821c482a1e35552adccb567020532b79244392374f25754f0050e6cd4c62, image=sz-pg-oam-docker-hub-001.tendcloud.com/library/spark-driver:v2.1.0-kubernetes-0.3.1-1, imageID=docker-pullable://sz-pg-oam-docker-hub-001.tendcloud.com/library/spark-driver@sha256:beb92a3e3f178e286d9e5baebdead88b5ba76d651f347ad2864bb6f8eda26f94, lastState=ContainerState(running=null, terminated=null, waiting=null, additionalProperties={}), name=spark-kubernetes-driver, ready=true, restartCount=0, state=ContainerState(running=ContainerStateRunning(startedAt=2017-09-14T06:59:02Z, additionalProperties={}), terminated=null, waiting=null, additionalProperties={}), additionalProperties={})] 2017-09-14 14:59:12 INFO LoggingPodStatusWatcherImpl:54 - State changed, new state: pod name: spark-pi-1505372339796-driver namespace: spark-cluster labels: spark-app-selector -> spark-f4d3a5d3ad964a05a51feb6191d50357, spark-role -> driver pod uid: 304cf440-991a-11e7-970c-f4e9d49f8ed0 creation time: 2017-09-14T06:59:01Z service account name: spark volumes: spark-token-zr8wv node name: 172.20.0.114 start time: 2017-09-14T06:59:01Z container images: sz-pg-oam-docker-hub-001.tendcloud.com/library/spark-driver:v2.1.0-kubernetes-0.3.1-1 phase: Succeeded status: [ContainerStatus(containerID=docker://5c5c821c482a1e35552adccb567020532b79244392374f25754f0050e6cd4c62, image=sz-pg-oam-docker-hub-001.tendcloud.com/library/spark-driver:v2.1.0-kubernetes-0.3.1-1, imageID=docker-pullable://sz-pg-oam-docker-hub-001.tendcloud.com/library/spark-driver@sha256:beb92a3e3f178e286d9e5baebdead88b5ba76d651f347ad2864bb6f8eda26f94, lastState=ContainerState(running=null, terminated=null, waiting=null, additionalProperties={}), name=spark-kubernetes-driver, ready=false, restartCount=0, state=ContainerState(running=null, terminated=ContainerStateTerminated(containerID=docker://5c5c821c482a1e35552adccb567020532b79244392374f25754f0050e6cd4c62, exitCode=0, finishedAt=2017-09-14T06:59:11Z, message=null, reason=Completed, signal=null, startedAt=null, additionalProperties={}), waiting=null, additionalProperties={}), additionalProperties={})] 2017-09-14 14:59:12 INFO LoggingPodStatusWatcherImpl:54 - Container final statuses: Container name: spark-kubernetes-driver Container image: sz-pg-oam-docker-hub-001.tendcloud.com/library/spark-driver:v2.1.0-kubernetes-0.3.1-1 Container state: Terminated Exit code: 0 2017-09-14 14:59:12 INFO Client:54 - Application spark-pi finished. 从日志中可以看到任务运行的状态信息。 使用下面的命令可以看到 kubernetes 启动的 Pod 信息： kubectl --namespace spark-cluster get pods -w 将会看到 spark-driver 和 spark-exec 的 Pod 信息。 依赖管理 上文中我们在运行测试程序时，命令行中指定的 jar 文件已包含在 docker 镜像中，是不是说我们每次提交任务都需要重新创建一个镜像呢？非也！如果真是这样也太麻烦了。 创建 resource staging server 为了方便用户提交任务，不需要每次提交任务的时候都创建一个镜像，我们使用了 resource staging server 。 kubectl create -f conf/kubernetes-resource-staging-server.yaml 我们同样将其部署在 spark-cluster namespace 下，该 yaml 文件见 kubernetes-handbook 的 manifests/spark-with-kubernetes-native-scheduler 目录。 优化 其中有一点需要优化，在使用下面的命令提交任务时，使用 --conf spark.kubernetes.resourceStagingServer.uri 参数指定 resource staging server 地址，用户不应该关注 resource staging server 究竟运行在哪台宿主机上，可以使用下面两种方式实现： 使用 nodeSelector 将 resource staging server 固定调度到某一台机器上，该地址依然使用宿主机的 IP 地址 改变 spark-resource-staging-service service 的 type 为 ClusterIP， 然后使用 Ingress 将其暴露到集群外部，然后加入的内网 DNS 里，用户使用 DNS 名称指定 resource staging server 的地址。 然后可以执行下面的命令来提交本地的 jar 到 kubernetes 上运行。 ./spark-submit \\ --deploy-mode cluster \\ --class org.apache.spark.examples.SparkPi \\ --master k8s://https://172.20.0.113:6443 \\ --kubernetes-namespace spark-cluster \\ --conf spark.kubernetes.authenticate.driver.serviceAccountName=spark \\ --conf spark.executor.instances=5 \\ --conf spark.app.name=spark-pi \\ --conf spark.kubernetes.driver.docker.image=sz-pg-oam-docker-hub-001.tendcloud.com/library/spark-driver:v2.1.0-kubernetes-0.3.1-1 \\ --conf spark.kubernetes.executor.docker.image=sz-pg-oam-docker-hub-001.tendcloud.com/library/spark-executor:v2.1.0-kubernetes-0.3.1-1 \\ --conf spark.kubernetes.initcontainer.docker.image=sz-pg-oam-docker-hub-001.tendcloud.com/library/spark-init:v2.1.0-kubernetes-0.3.1-1 \\ --conf spark.kubernetes.resourceStagingServer.uri=http://172.20.0.114:31000 \\ ../examples/jars/spark-examples_2.11-2.2.0-k8s-0.4.0-SNAPSHOT.jar 该命令将提交本地的 ../examples/jars/spark-examples_2.11-2.2.0-k8s-0.4.0-SNAPSHOT.jar 文件到 resource staging server，executor 将从该 server 上获取 jar 包并运行，这样用户就不需要每次提交任务都编译一个镜像了。 详见：https://apache-spark-on-k8s.github.io/userdocs/running-on-kubernetes.html#dependency-management 设置 HDFS 用户 如果 Hadoop 集群没有设置 kerbros 安全认证的话，在指定 spark-submit 的时候可以通过指定如下四个环境变量， 设置 Spark 与 HDFS 通信使用的用户： --conf spark.kubernetes.driverEnv.SPARK_USER=hadoop --conf spark.kubernetes.driverEnv.HADOOP_USER_NAME=hadoop --conf spark.executorEnv.HADOOP_USER_NAME=hadoop --conf spark.executorEnv.SPARK_USER=hadoop 使用 hadoop 用户提交本地 jar 包的命令示例： ./spark-submit \\ --deploy-mode cluster \\ --class com.talkingdata.alluxio.hadooptest \\ --master k8s://https://172.20.0.113:6443 \\ --kubernetes-namespace spark-cluster \\ --conf spark.kubernetes.driverEnv.SPARK_USER=hadoop \\ --conf spark.kubernetes.driverEnv.HADOOP_USER_NAME=hadoop \\ --conf spark.executorEnv.HADOOP_USER_NAME=hadoop \\ --conf spark.executorEnv.SPARK_USER=hadoop \\ --conf spark.kubernetes.authenticate.driver.serviceAccountName=spark \\ --conf spark.executor.instances=5 \\ --conf spark.app.name=spark-pi \\ --conf spark.kubernetes.driver.docker.image=sz-pg-oam-docker-hub-001.tendcloud.com/library/spark-driver:v2.1.0-kubernetes-0.3.1-1 \\ --conf spark.kubernetes.executor.docker.image=sz-pg-oam-docker-hub-001.tendcloud.com/library/spark-executor:v2.1.0-kubernetes-0.3.1-1 \\ --conf spark.kubernetes.initcontainer.docker.image=sz-pg-oam-docker-hub-001.tendcloud.com/library/spark-init:v2.1.0-kubernetes-0.3.1-1 \\ --conf spark.kubernetes.resourceStagingServer.uri=http://172.20.0.114:31000 \\ ~/Downloads/tendcloud_2.10-1.0.jar 详见：https://github.com/apache-spark-on-k8s/spark/issues/408 限制 Driver 和 Executor 的资源使用 在执行 spark-submit 时使用如下参数设置内存和 CPU 资源限制： --conf spark.driver.memory=3G --conf spark.executor.memory=3G --conf spark.driver.cores=2 --conf spark.executor.cores=10 这几个参数中值如何传递到 Pod 的资源设置中的呢？ 比如我们设置在执行 spark-submit 的时候传递了这样的两个参数：--conf spark.driver.cores=2 和 --conf spark.driver.memory=100G 那么查看 driver pod 的 yaml 输出结果将会看到这样的资源设置： resources: limits: memory: 110Gi requests: cpu: \"2\" memory: 100Gi 以上参数是对 request 值的设置，那么 limit 的资源设置的值又是从何而来？ 可以使用 spark.kubernetes.driver.limit.cores 和 spark.kubernetes.executor.limit.cores 来设置 CPU的 hard limit。 memory limit 的值是根据 memory request 的值加上 spark.kubernetes.executor.memoryOverhead 的值计算而来的，该配置项用于设置分配给每个 executor 的超过 heap 内存的值（可以使用k、m、g单位）。该值用于虚拟机的开销、其他本地服务开销。根据 executor 的大小设置（通常是 6%到10%）。 我们可以这样来提交一个任务，同时设置 driver 和 executor 的 CPU、内存的资源 request 和 limit 值（driver 的内存 limit 值为 request 值的 110%）。 ./spark-submit \\ --deploy-mode cluster \\ --class org.apache.spark.examples.SparkPi \\ --master k8s://https://172.20.0.113:6443 \\ --kubernetes-namespace spark-cluster \\ --conf spark.kubernetes.authenticate.driver.serviceAccountName=spark \\ --conf spark.driver.memory=100G \\ --conf spark.executor.memory=10G \\ --conf spark.driver.cores=30 \\ --conf spark.executor.cores=2 \\ --conf spark.driver.maxResultSize=10240m \\ --conf spark.kubernetes.driver.limit.cores=32 \\ --conf spark.kubernetes.executor.limit.cores=3 \\ --conf spark.kubernetes.executor.memoryOverhead=2g \\ --conf spark.executor.instances=5 \\ --conf spark.app.name=spark-pi \\ --conf spark.kubernetes.driver.docker.image=sz-pg-oam-docker-hub-001.tendcloud.com/library/spark-driver:v2.1.0-kubernetes-0.3.1-1 \\ --conf spark.kubernetes.executor.docker.image=sz-pg-oam-docker-hub-001.tendcloud.com/library/spark-executor:v2.1.0-kubernetes-0.3.1-1 \\ --conf spark.kubernetes.initcontainer.docker.image=sz-pg-oam-docker-hub-001.tendcloud.com/library/spark-init:v2.1.0-kubernetes-0.3.1-1 \\ local:///opt/spark/examples/jars/spark-examples_2.11-2.2.0-k8s-0.4.0-SNAPSHOT.jar 10000000 这将启动一个包含一千万个 task 的计算 pi 的 spark 任务，任务运行过程中，drvier 的 CPU 实际消耗大约为 3 核，内存 40G，每个 executor 的 CPU 实际消耗大约不到 1 核，内存不到 4G，我们可以根据实际资源消耗不断优化资源的 request 值。 SPARK_DRIVER_MEMORY 和 SPARK_EXECUTOR_MEMORY 和分别作为 Driver 容器和 Executor 容器启动的环境变量，比如下面这个 Driver 启动的 CMD 中： CMD SPARK_CLASSPATH=\"${SPARK_HOME}/jars/*\" && \\ env | grep SPARK_JAVA_OPT_ | sed 's/[^=]*=\\(.*\\)/\\1/g' > /tmp/java_opts.txt && \\ readarray -t SPARK_DRIVER_JAVA_OPTS 我们可以看到对 SPARK_DRIVER_MEMORY 环境变量的引用。Executor 的设置与 driver 类似。 而我们可以使用这样的参数来传递环境变量的值 spark.executorEnv.[EnvironmentVariableName]，只要将 EnvironmentVariableName 替换为环境变量名称即可。 参考 Spark动态资源分配-Dynamic Resource Allocation Running Spark on Kubernetes Apache Spark Jira Issue - 18278 - SPIP: Support native submission of spark jobs to a kubernetes cluster Kubernetes Github Issue - 34377 Support Spark natively in Kubernetes Kubernetes example spark https://github.com/rootsongjc/spark-on-kubernetes Scheduler backend Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-09-27 21:03:00 "},"usecases/serverless.html":{"url":"usecases/serverless.html","title":"Serverless架构","keywords":"","body":"Serverless架构 Serverless（无服务器架构）指的是由开发者实现的服务端逻辑运行在无状态的计算容器中，它由事件触发， 完全被第三方管理，其业务层面的状态则被开发者使用的数据库和存储资源所记录。 下图来自谷歌云平台官网，是对云计算的一个很好的分层概括，其中 serverless 就是构建在虚拟机和容器之上的一层，与应用本身的关系更加密切。 图片 - 从物理机到函数计算 Serverless架构的优点 今天大多数公司在开发应用程序并将其部署在服务器上的时候，无论是选择公有云还是私有的数据中心，都需要提前了解究竟需要多少台服务器、多大容量的存储和数据库的功能等。并需要部署运行应用程序和依赖的软件到基础设施之上。假设我们不想在这些细节上花费精力，是否有一种简单的架构模型能够满足我们这种想法？这个答案已经存在，这就是今天软件架构世界中新鲜但是很热门的一个话题——Serverless（无服务器）架构。 ——AWS 费良宏 降低运营成本： Serverless是非常简单的外包解决方案。它可以让您委托服务提供商管理服务器、数据库和应用程序甚至逻辑，否则您就不得不自己来维护。由于这个服务使用者的数量会非常庞大，于是就会产生规模经济效应。在降低成本上包含了两个方面，即基础设施的成本和人员（运营/开发）的成本。 降低开发成本： IaaS和PaaS存在的前提是，服务器和操作系统管理可以商品化。Serverless作为另一种服务的结果是整个应用程序组件被商品化。 扩展能力： Serverless架构一个显而易见的优点即“横向扩展是完全自动的、有弹性的、且由服务提供者所管理”。从基本的基础设施方面受益最大的好处是，您只需支付您所需要的计算能力。 更简单的管理： Serverless架构明显比其他架构更简单。更少的组件，就意味着您的管理开销会更少。 “绿色”的计算： 按照《福布斯》杂志的统计，在商业和企业数据中心的典型服务器仅提供5%～15%的平均最大处理能力的输出。这无疑是一种资源的巨大浪费。随着Serverless架构的出现，让服务提供商提供我们的计算能力最大限度满足实时需求。这将使我们更有效地利用计算资源。 Kubernetes上的serverless 架构 目前已经有一批优秀的基于 kubernetes 的 serverless 架构（FaaS）开源项目如下： faas - 🐳 Functions as a Service - a serverless framework for Docker & Kubernetes https://blog.alexellis.io/introducing… faas-netes - Enable Kubernetes as a backend for Functions as a Service (OpenFaaS) https://github.com/alexellis/faas fn - The container native, cloud agnostic serverless platform. http://fnproject.io funktion - a CLI tool for working with funktion https://funktion.fabric8.io/ fx - Poor man's serverless framework based on Docker, Function as a Service with painless. IronFunctions - IronFunctions - the serverless microservices platform. http://iron.io kubeless - Kubernetes Native Serverless Framework http://kubeless.io OpenWhisk - Apache OpenWhisk (Incubating) is a serverless, open source cloud platform that executes functions in response to events at any scale. 以上项目收录于 awsome-cloud-native FaaS Function-as-a-Service 全景图（图片来自https://github.com/amyers1793/FunctionasaServiceLandscape) 图片 - FaaS Landscape 参考 Serverless Architectures - Martin Fowler Serverless架构综述 2017年会是Serverless爆发之年吗？ 从IaaS到FaaS—— Serverless架构的前世今生 Introducing Redpoint's FaaS Landscape Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-21 18:11:54 "},"usecases/understanding-serverless.html":{"url":"usecases/understanding-serverless.html","title":"理解Serverless","keywords":"","body":"理解Serverless No silver bullet. - The Mythical Man-Month 许多年前，我们开发的软件还是C/S（客户端/服务器）和MVC（模型-试图-控制器）的形式，再后来有了SOA，最近几年又出现了微服务架构，更新一点的有Cloud Native（云原生）应用，企业应用从单体架构，到服务化，再到更细粒度的微服务化，应用开发之初就是为了应对互联网的特有的高并发、不间断的特性，需要很高的性能和可扩展性，人们对软件开发的追求孜孜不倦，希望力求在软件开发的复杂度和效率之间达到一个平衡。但可惜的是，NO SILVER BULLET！几十年前（1975年）Fred Brooks就在The Mythical Man-Month中就写到了这句话。那么Serverlss会是那颗银弹吗？ 云改变了我们对操作系统的认知，原来一个系统的计算资源、存储和网络是可以分离配置的，而且还可以弹性扩展，但是长久以来，我们在开发应用时始终没有摆脱的服务器的束缚（或者说认知），应用必须运行在不论是实体还是虚拟的服务器上，必须经过部署、配置、初始化才可以运行，还需要对服务器和应用进行监控和管理，还需要保证数据的安全性，这些云能够帮我们简化吗？让我们只要关注自己代码的逻辑就好了，其它的东西让云帮我实现就好了。 Serverless介绍 Serverless架构是云的自然延伸，为了理解serverless，我们有必要回顾一下云计算的发展。 IaaS 2006年AWS推出EC2（Elastic Compute Cloud），作为第一代IaaS（Infrastructure as a Service），用户可以通过AWS快速的申请到计算资源，并在上面部署自己的互联网服务。IaaS从本质上讲是服务器租赁并提供基础设施外包服务。就比如我们用的水和电一样，我们不会自己去引入自来水和发电，而是直接从自来水公司和电网公司购入，并根据实际使用付费。 EC2真正对IT的改变是硬件的虚拟化（更细粒度的虚拟化），而EC2给用户带来了以下五个好处： 降低劳动力成本：减少了企业本身雇佣IT人员的成本 降低风险：不用再像自己运维物理机那样，担心各种意外风险，EC2有主机损坏，再申请一个就好了。 降低基础设施成本：可以按小时、周、月或者年为周期租用EC2。 扩展性：不必过早的预期基础设施采购，因为通过云厂商可以很快的获取。 节约时间成本：快速的获取资源开展业务实验。 以上说了是IaaS或者说基础设施外包的好处，当然其中也有弊端，我们将在后面讨论。 以上是AWS为代表的公有云IaaS，还有使用OpenStack构建的私有云也能够提供IaaS能力。 PaaS PaaS（Platform as a Service）是构建在IaaS之上的一种平台服务，提供操作系统安装、监控和服务发现等功能，用户只需要部署自己的应用即可，最早的一代是Heroku。Heroko是商业的PaaS，还有一个开源的PaaS——Cloud Foundry，用户可以基于它来构建私有PaaS，如果同时使用公有云和私有云，如果能在两者之间构建一个统一的PaaS，那就是“混合云”了。 在PaaS上最广泛使用的技术就要数docker了，因为使用容器可以很清晰的描述应用程序，并保证环境一致性。管理云上的容器，可以称为是CaaS（Container as a Service），如GCE（Google Container Engine）。也可以基于Kubernetes、Mesos这类开源软件构件自己的CaaS，不论是直接在IaaS构建还是基于PaaS。 PaaS是对软件的一个更高的抽象层次，已经接触到应用程序的运行环境本身，可以由开发者自定义，而不必接触更底层的操作系统。 Serverless的定义 Serverless不如IaaS和PaaS那么好理解，因为它通常包含了两个领域BaaS（Backend as a Service）和FaaS（Function as a Service）。 BaaS BaaS（Backend as a Service）后端即服务，一般是一个个的API调用后端或别人已经实现好的程序逻辑，比如身份验证服务Auth0，这些BaaS通常会用来管理数据，还有很多公有云上提供的我们常用的开源软件的商用服务，比如亚马逊的RDS可以替代我们自己部署的MySQL，还有各种其它数据库和存储服务。 FaaS FaaS（Functions as a Service）函数即服务，FaaS是无服务器计算的一种形式，当前使用最广泛的是AWS的Lambada。 现在当大家讨论Serverless的时候首先想到的就是FaaS，有点甚嚣尘上了。FaaS本质上是一种事件驱动的由消息触发的服务，FaaS供应商一般会集成各种同步和异步的事件源，通过订阅这些事件源，可以突发或者定期的触发函数运行。 图片 - 服务端软件的运行环境 传统的服务器端软件不同是经应用程序部署到拥有操作系统的虚拟机或者容器中，一般需要长时间驻留在操作系统中运行，而FaaS是直接将程序部署上到平台上即可，当有事件到来时触发执行，执行完了就可以卸载掉。 图片 - FaaS应用架构 总结 两者都为我们的计算资源提供了弹性的保障，BaaS其实依然是服务外包，而FaaS使我们更加关注应用程序的逻辑，两者使我们不需要关注应用程序所在的服务器，但实际上服务器依然是客观存在的。 当我们将应用程序迁移到容器和虚拟机中时，其实对于应用程序本身的体系结构并没有多少改变，只不过有些流程和规定需要遵守，比如12因素应用守则，但是serverlss对应用程序的体系结构来说就是一次颠覆了，通常我们需要考虑事件驱动模型，更加细化的不熟形式，以及在FaaS组件之外保持状态的需求。 Serverless应用 我们以一个游戏应用为例，来说明什么是serverless应用。 一款移动端游戏至少包含如下几个特性： 移动端友好的用户体验 用户管理和权限认证 关卡、升级等游戏逻辑，游戏排行，玩家的等级、任务等信息 传统的应用程序架构可能是这样的： 图片 - 传统应用程序架构 一个app前端，iOS后者安卓 用Java写的后端，使用JBoss或者Tomcat做server运行 使用关系型数据库存储用户数据，如MySQL 这样的架构可以让前端十分轻便，不需要做什么应用逻辑，只是负责渲染用户界面，将请求通过HTTP发送给后端，而所有的数据操作都是有由后端的Java程序来完成的。 这样的架构开发起来比较容易，但是维护起来确十分复杂，前端开发、后端的开发都需要十分专业的人员、环境的配置，还要有人专门维护数据库、应用的更新和升级。 图片 - Serverless架构 而在serverless架构中，我们不再需要在服务器端代码中存储任何会话状态，而是直接将它们存储在NoSQL中，这样将使应用程序无状态，有助于弹性扩展。前端可以直接利用BaaS而减少后端的编码需求，这样架构的本质上是减少了应用程序开发的人力成本，降低了自己维护基础设施的风险，而且利用云的能力更便于扩展和快速迭代。 Serverless的优势 在最前面我们提到了使用IaaS给我们带来了五点好处，FaaS当然也包括了这些好处，但是它给我们带来的最大的好处就是多快好省。减少从概念原型到实施的等待时间，比自己维护服务更省钱。 降低人力成本 不需要再自己维护服务器，操心服务器的各种性能指标和资源利用率，而是关心应用程序本身的状态和逻辑。而且serverless应用本身的部署也十分容易，我们只要上传基本的代码但愿，例如Javascript或Python的源代码的zip文件，以及基于JVM的语言的纯JAR文件。不需使用Puppet、Chef、Ansible或Docker来进行配置管理，降低了运维成本。同时，对于运维来说，也不再需要监控那些更底层的如磁盘使用量、CPU使用率等底层和长期的指标信息，而是监控应用程序本身的度量，这将更加直观和有效。 在此看来有人可能会提出“NoOps”的说法，其实这是不存在的，只要有应用存在的一天就会有Ops，只是人员的角色会有所转变，部署将变得更加自动化，监控将更加面向应用程序本身，更底层的运维依然需要专业的人员去做。 降低风险 对于组件越多越复杂的系统，出故障的风险就越大。我们使用BaaS或FaaS将它们外包出去，让专业人员来处理这些故障，有时候比我们自己来修复更可靠，利用专业人员的知识来降低停机的风险，缩短故障修复的时间，让我们的系统稳定性更高。 减少资源开销 我们在申请主机资源一般会评估一个峰值最大开销来申请资源，往往导致过度的配置，这意味着即使在主机闲置的状态下也要始终支付峰值容量的开销。对于某些应用来说这是不得已的做法，比如数据库这种很难扩展的应用，而对于普通应用这就显得不太合理了，虽然我们都觉得即使浪费了资源也比当峰值到来时应用程序因为资源不足而挂掉好。 解决这个问题最好的办法就是，不计划到底需要使用多少资源，而是根据实际需要来请求资源，当然前提必须是整个资源池是充足的（公有云显然更适合）。根据使用时间来付费，根据每次申请的计算资源来付费，让计费的粒度更小，将更有利于降低资源的开销。这是对应用程序本身的优化，例如让每次请求耗时更短，让每次消耗的资源更少将能够显著节省成本。 增加缩放的灵活性 以AWS Lamba为例，当平台接收到第一个触发函数的事件时，它将启动一个容器来运行你的代码。如果此时收到了新的事件，而第一个容器仍在处理上一个事件，平台将启动第二个代码实例来处理第二个事件。AWS lambad的这种自动的零管理水平缩放，将持续到有足够的代码实例来处理所有的工作负载。 但是，AWS仍然只会向您收取代码的执行时间，无论它需要启动多少个容器实例要满足你的负载请求。例如，假设所有事件的总执行时间是相同的，在一个容器中按顺序调用Lambda 100次与在100个不同容器中同时调用100次Lambda的成本是 一样的。当然AWS Lambada也不会无限制的扩展实例个数，如果有人对你发起了DDos攻击怎么办，那么不就会产生高昂的成本吗？AWS是有默认限制的，默认执行Lambada函数最大并发数是1000。 缩短创新周期 小团队的开发人员正可以在几天之内从头开始开发应用程序并部署到生产。使用短而简单的函数和事件来粘合强大的驱动数据存储和服务的API。完成的应用程序具有高度可用性和可扩展性，利用率高，成本低，部署速度快。 以docker为代表的容器技术仅仅是缩短了应用程序的迭代周期，而serverless技术是直接缩短了创新周期，从概念到最小可行性部署的时间，让初级开发人员也能在很短的时间内完成以前通常要经验丰富的工程师才能完成的项目。 Serverless的劣势 我们知道没有十全十美的技术，在说了serverless的那么多优势之后，我们再来探讨以下serverless的劣势，或者说局限性和适用场景。 状态管理 要想实现自由的缩放，无状态是必须的，而对于有状态的服务，使用serverless这就丧失了灵活性，有状态服务需要与存储交互就不可避免的增加了延迟和复杂性。 延迟 应用程序中不同组件的访问延迟是一个大问题，我们可以通过使用专有的网络协议、RPC调用、数据格式来优化，或者是将实例放在同一个机架内或同一个主机实例上来优化以减少延迟。 而serverless应用程序是高度分布式、低耦合的，这就意味着延迟将始终是一个问题，单纯使用serverless的应用程序是不太现实的。 本地测试 Serverless应用的本地测试困难是一个很棘手的问题。虽然可以在测试环境下使用各种数据库和消息队列来模拟生产环境，但是对于无服务应用的集成或者端到端测试尤其困难，很难在本地模拟应用程序的各种连接，并与性能和缩放的特性结合起来测试，并且serverless应用本身也是分布式的，简单的将无数的FaaS和BaaS组件粘合起来也是有挑战性的。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-19 20:06:23 "},"usecases/faas.html":{"url":"usecases/faas.html","title":"FaaS-函数即服务","keywords":"","body":"FaaS-函数即服务 FaaS（Functions as a Service）函数即服务，FaaS是无服务器计算的一种形式，当前使用最广泛的是AWS的Lambada。 现在当大家讨论Serverless的时候首先想到的就是FaaS，有点甚嚣尘上了。FaaS本质上是一种事件驱动的由消息触发的服务，FaaS供应商一般会集成各种同步和异步的事件源，通过订阅这些事件源，可以突发或者定期的触发函数运行。 当前开源的FaaS框架大部分都是基于Kuberentes来实现的，例如： faas-netes - Enable Kubernetes as a backend for Functions as a Service (OpenFaaS) https://github.com/alexellis/faas fn - The container native, cloud agnostic serverless platform. http://fnproject.io funktion - a CLI tool for working with funktion https://funktion.fabric8.io/ fx - Poor man's serverless framework based on Docker, Function as a Service with painless. IronFunctions - IronFunctions - the serverless microservices platform. http://iron.io kubeless - Kubernetes Native Serverless Framework http://kubeless.io OpenFaaS - OpenFaaS - Serverless Functions Made Simple for Docker & Kubernetes https://blog.alexellis.io/introducing-functions-as-a-service/ OpenWhisk - Apache OpenWhisk (Incubating) is a serverless, open source cloud platform that executes functions in response to events at any scale. 关于整个Cloud Native开源生态，请参考awesome-cloud-native。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-19 20:06:08 "},"usecases/openfaas.html":{"url":"usecases/openfaas.html","title":"OpenFaaS","keywords":"","body":"OpenFaaS OpenFaaS一款高人气的开源的faas框架，可以在 在Kubernetes上部署OpenFaaS十分简单，用到的镜像如下： functions/faas-netesd:0.3.4 functions/gateway:0.6.14 functions/prometheus:latest-k8s functions/alertmanager:latest-k8s 这些镜像都存储在DockerHub上。 OpenFaaS的架构如下图： 图片 - OpenFaaS架构 部署 参考Deployment guide for Kubernetes部署OpenFaaS。 如果您的Kuberentes集群可以访问DockerHub那么直接使用官方提供的YAML文件即可。 YAML文件见官方仓库：https://github.com/openfaas/faas-netes 部署同步请求 一共用到了三个YAML文件： faas.yml monitoring.yml rbac.yml OpenFaaS安装好后会启动一个Prometheus。 图片 - OpenFaaS Prometheus 在这里可以看到函数的运行情况。 同时OpenFaaS还提供了Dashboard，需要我们自己向Grafana中配置Prometheus数据源后导入，JSON配置见：https://grafana.com/dashboards/3526，可以下载后直接导入到Grafana中。 图片 - OpenFaas Grafana监控 OpenFaaS的使用 OpenFaaS提供了便捷的UI，在部署完成后就可以通过NodePort方式访问。 图片 - OpenFaas Dashboard 其中已经内置了一些函数应用可供我们试用，还可以创建自己的函数。 比如内置的NodeInfo应用，可以获取该应用所部署到的主机的信息，如下图： 图片 - NodeInfo执行结果 注意：有一些js和css文件需要翻墙才能访问，否则页面将出现格式错误。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-19 22:37:24 "},"usecases/edge-computing.html":{"url":"usecases/edge-computing.html","title":"边缘计算","keywords":"","body":"边缘计算 TBD 参考 The Birth of an Edge Orchestrator – Cloudify Meets Edge Computing K8s(Kubernetes) and SDN for Multi-access Edge Computing deployment Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-10-27 19:19:45 "},"develop/":{"url":"develop/","title":"开发指南概览","keywords":"","body":"开发指南说明 讲解如何在原生 Kubernetes 的基础上做定制开发。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-15 19:44:17 "},"develop/sigs-and-working-group.html":{"url":"develop/sigs-and-working-group.html","title":"SIG和工作组","keywords":"","body":"SIG和工作组 Kubernetes的社区是以SIG（Special Interest Group特别兴趣小组）和工作组的形式组织起来的，每个工作组都会定期召开视频会议。 所有的SIG和工作组都使用slack和邮件列表沟通。 图片 - Kubernetes SIG 主要SIG列表 api-machinery：所有API级别的功能，包括了API server、API注册和发现、通用的API CRUD语义，准入控制，编码/解码，转换，默认值，持久化层（etcd），OpenAPI，第三方资源，垃圾手机和客户端库的方方面面。 aws：如何在AWS上支持和使用kubernetes。 apps：在kubernetes上部署和运维应用程序。关注开发者和DevOps在kubernetes上运行应用程序的体验。 architecture：维持kubernetes在架构设计上的一致性和原则。 auth：kubernetes的认证授权、权限管理和安全性策略。 autoscaling：集群的自动缩放，pod的水平和垂直自动缩放，pod的资源初始化，pod监控和指标收集等主题。 azure：如何在Azure上支持和使用kubernetes。 big-data：在kubernetes上部署和运行大数据应用，如Spark、Kafka、Hadoop、Flink、Storm等。 CLI：kubectl和相关工具。 cluster-lifecycle：部署和升级kubernetes集群。 cluster-ops：促进kubernetes集群本身的可操作性和集群间的互操作性，使不同的运营商之间协调一致。 contributor-experience：维持良好的开发者社区。 docs：文档，流程和出版物。 GCP：在Google Cloud Platform上使用kubernetes。 instrumentation：集群可观测性的最佳实践，包括指标设置、日志收集、事件等。 multicluster：多kubernetes集群的用例和工具。 network：kubernetes集群的网络。 node：node节点、kubelet。 onprem：在非云供应商的环境下运行kubernetes，例如on premise、裸机等环境。 openstack：协调跨OpenStack和Kubernetes社区的努力。 product-management：侧重于产品管理方面。 release：发布、PR和bug提交等。 scalability：负责回答可伸缩性相关的问题。 scheduling：资源调度。 service-catalog：为CNCF service broker和Kubernetes broker实现开发API。 storage：存储和volume插件。 testing：测试。 ui：与UI相关的话题。 windows：在kubernets上运行Windows Server Container。 工作组列表 App Def：改进API中的声明性原语、客户端库、工具的用户体验。 Cloud Provider：云供应商工作组。 Cluster API：定义一个代表Kubernetes集群的可移植API。 API将包含控制平面及其配置和底层基础设施（节点，节点池等）。 Container Identity：确保容器能够获得安全的身份认证并与外部连通的解决方案。 Kubeadm Adoption：提高kubeadm工具的采用率。 Resource ManagementJ：资源隔离和提高资源利用率。 详细信息请参考 https://github.com/kubernetes/community/blob/master/sig-list.md Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-20 18:50:16 "},"develop/developing-environment.html":{"url":"develop/developing-environment.html","title":"开发环境搭建","keywords":"","body":"配置Kubernetes开发环境 我们将在Mac上使用docker环境编译kuberentes。 安装依赖 brew install gnu-tar Docker环境，至少需要给容器分配4G内存，在低于3G内存的时候可能会编译失败。 执行编译 切换目录到kuberentes源码的根目录下执行： ./build/run.sh make可以在docker中执行跨平台编译出二进制文件。 需要用的的docker镜像： gcr.io/google_containers/kube-cross:v1.7.5-2 我将该镜像备份到时速云上了，可供大家使用： index.tenxcloud.com/jimmy/kube-cross:v1.7.5-2 该镜像基于Ubuntu构建，大小2.15G，编译环境中包含以下软件： Go1.7.5 etcd protobuf g++ 其他golang依赖包 在我自己的电脑上的整个编译过程大概要半个小时。 编译完成的二进制文件在/_output/local/go/bin/目录下。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-08-21 18:23:34 "},"develop/testing.html":{"url":"develop/testing.html","title":"单元测试和集成测试","keywords":"","body":"Kubernetes测试 单元测试 单元测试仅依赖于源代码，是测试代码逻辑是否符合预期的最简单方法。 运行所有的单元测试 make test 仅测试指定的package # 单个package make test WHAT=./pkg/api # 多个packages make test WHAT=./pkg/{api,kubelet} 或者，也可以直接用go test go test -v k8s.io/kubernetes/pkg/kubelet 仅测试指定package的某个测试case # Runs TestValidatePod in pkg/api/validation with the verbose flag set make test WHAT=./pkg/api/validation KUBE_GOFLAGS=\"-v\" KUBE_TEST_ARGS='-run ^TestValidatePod$' # Runs tests that match the regex ValidatePod|ValidateConfigMap in pkg/api/validation make test WHAT=./pkg/api/validation KUBE_GOFLAGS=\"-v\" KUBE_TEST_ARGS=\"-run ValidatePod\\|ValidateConfigMap$\" 或者直接用go test go test -v k8s.io/kubernetes/pkg/api/validation -run ^TestValidatePod$ 并行测试 并行测试是root out flakes的一种有效方法： # Have 2 workers run all tests 5 times each (10 total iterations). make test PARALLEL=2 ITERATION=5 生成测试报告 make test KUBE_COVER=y Benchmark测试 go test ./pkg/apiserver -benchmem -run=XXX -bench=BenchmarkWatch 集成测试 Kubernetes集成测试需要安装etcd（只要按照即可，不需要启动），比如 hack/install-etcd.sh # Installs in ./third_party/etcd echo export PATH=\"\\$PATH:$(pwd)/third_party/etcd\" >> ~/.profile # Add to PATH 集成测试会在需要的时候自动启动etcd和kubernetes服务，并运行test/integration里面的测试。 运行所有集成测试 make test-integration # Run all integration tests. 指定集成测试用例 # Run integration test TestPodUpdateActiveDeadlineSeconds with the verbose flag set. make test-integration KUBE_GOFLAGS=\"-v\" KUBE_TEST_ARGS=\"-run ^TestPodUpdateActiveDeadlineSeconds$\" End to end (e2e)测试 End to end (e2e) 测试模拟用户行为操作Kubernetes，用来保证Kubernetes服务或集群的行为完全符合设计预期。 在开启e2e测试之前，需要先编译测试文件，并设置KUBERNETES_PROVIDER（默认为gce）： make WHAT='test/e2e/e2e.test' make ginkgo export KUBERNETES_PROVIDER=local 启动cluster，测试，最后停止cluster # build Kubernetes, up a cluster, run tests, and tear everything down go run hack/e2e.go -- -v --build --up --test --down 仅测试指定的用例 go run hack/e2e.go -v -test --test_args='--ginkgo.focus=Kubectl\\sclient\\s\\[k8s\\.io\\]\\sKubectl\\srolling\\-update\\sshould\\ssupport\\srolling\\-update\\sto\\ssame\\simage\\s\\[Conformance\\]$' 略过测试用例 go run hack/e2e.go -- -v --test --test_args=\"--ginkgo.skip=Pods.*env 并行测试 # Run tests in parallel, skip any that must be run serially GINKGO_PARALLEL=y go run hack/e2e.go --v --test --test_args=\"--ginkgo.skip=\\[Serial\\]\" # Run tests in parallel, skip any that must be run serially and keep the test namespace if test failed GINKGO_PARALLEL=y go run hack/e2e.go --v --test --test_args=\"--ginkgo.skip=\\[Serial\\] --delete-namespace-on-failure=false\" 清理测试 go run hack/e2e.go -- -v --down 有用的-ctl # -ctl can be used to quickly call kubectl against your e2e cluster. Useful for # cleaning up after a failed test or viewing logs. Use -v to avoid suppressing # kubectl output. go run hack/e2e.go -- -v -ctl='get events' go run hack/e2e.go -- -v -ctl='delete pod foobar' Fedaration e2e测试 export FEDERATION=true export E2E_ZONES=\"us-central1-a us-central1-b us-central1-f\" # or export FEDERATION_PUSH_REPO_BASE=\"quay.io/colin_hom\" export FEDERATION_PUSH_REPO_BASE=\"gcr.io/${GCE_PROJECT_NAME}\" # build container images KUBE_RELEASE_RUN_TESTS=n KUBE_FASTBUILD=true go run hack/e2e.go -- -v -build # push the federation container images build/push-federation-images.sh # Deploy federation control plane go run hack/e2e.go -- -v --up # Finally, run the tests go run hack/e2e.go -- -v --test --test_args=\"--ginkgo.focus=\\[Feature:Federation\\]\" # Don't forget to teardown everything down go run hack/e2e.go -- -v --down 可以用cluster/log-dump.sh 方便的下载相关日志，帮助排查测试中碰到的问题。 Node e2e测试 Node e2e仅测试Kubelet的相关功能，可以在本地或者集群中测试 export KUBERNETES_PROVIDER=local make test-e2e-node FOCUS=\"InitContainer\" make test_e2e_node TEST_ARGS=\"--experimental-cgroups-per-qos=true\" 补充说明 借助kubectl的模版可以方便获取想要的数据，比如查询某个container的镜像的方法为 kubectl get pods nginx-4263166205-ggst4 -o template '--template={{if (exists . \"status\" \"containerStatuses\")}}{{range .status.containerStatuses}}{{if eq .name \"nginx\"}}{{.image}}{{end}}{{end}}{{end}}' 参考文档 Kubernetes testing End-to-End Testing Node e2e test How to write e2e test Coding Conventions Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-08-21 18:23:34 "},"develop/client-go-sample.html":{"url":"develop/client-go-sample.html","title":"client-go示例","keywords":"","body":"client-go示例 访问kubernetes集群有几下几种方式： 方式 特点 支持者 Kubernetes dashboard 直接通过Web UI进行操作，简单直接，可定制化程度低 官方支持 kubectl 命令行操作，功能最全，但是比较复杂，适合对其进行进一步的分装，定制功能，版本适配最好 官方支持 client-go 从kubernetes的代码中抽离出来的客户端包，简单易用，但需要小心区分kubernetes的API版本 官方支持 client-python python客户端，kubernetes-incubator 官方支持 Java client fabric8中的一部分，kubernetes的java客户端 redhat 下面，我们基于client-go，对Deployment升级镜像的步骤进行了定制，通过命令行传递一个Deployment的名字、应用容器名和新image名字的方式来升级。代码和使用方式见 https://github.com/rootsongjc/kubernetes-client-go-sample 。 kubernetes-client-go-sample 代码如下： package main import ( \"flag\" \"fmt\" \"os\" \"path/filepath\" \"k8s.io/apimachinery/pkg/api/errors\" metav1 \"k8s.io/apimachinery/pkg/apis/meta/v1\" \"k8s.io/client-go/kubernetes\" \"k8s.io/client-go/tools/clientcmd\" ) func main() { var kubeconfig *string if home := homeDir(); home != \"\" { kubeconfig = flag.String(\"kubeconfig\", filepath.Join(home, \".kube\", \"config\"), \"(optional) absolute path to the kubeconfig file\") } else { kubeconfig = flag.String(\"kubeconfig\", \"\", \"absolute path to the kubeconfig file\") } deploymentName := flag.String(\"deployment\", \"\", \"deployment name\") imageName := flag.String(\"image\", \"\", \"new image name\") appName := flag.String(\"app\", \"app\", \"application name\") flag.Parse() if *deploymentName == \"\" { fmt.Println(\"You must specify the deployment name.\") os.Exit(0) } if *imageName == \"\" { fmt.Println(\"You must specify the new image name.\") os.Exit(0) } // use the current context in kubeconfig config, err := clientcmd.BuildConfigFromFlags(\"\", *kubeconfig) if err != nil { panic(err.Error()) } // create the clientset clientset, err := kubernetes.NewForConfig(config) if err != nil { panic(err.Error()) } deployment, err := clientset.AppsV1beta1().Deployments(\"default\").Get(*deploymentName, metav1.GetOptions{}) if err != nil { panic(err.Error()) } if errors.IsNotFound(err) { fmt.Printf(\"Deployment not found\\n\") } else if statusError, isStatus := err.(*errors.StatusError); isStatus { fmt.Printf(\"Error getting deployment%v\\n\", statusError.ErrStatus.Message) } else if err != nil { panic(err.Error()) } else { fmt.Printf(\"Found deployment\\n\") name := deployment.GetName() fmt.Println(\"name ->\", name) containers := &deployment.Spec.Template.Spec.Containers found := false for i := range *containers { c := *containers if c[i].Name == *appName { found = true fmt.Println(\"Old image ->\", c[i].Image) fmt.Println(\"New image ->\", *imageName) c[i].Image = *imageName } } if found == false { fmt.Println(\"The application container not exist in the deployment pods.\") os.Exit(0) } _, err := clientset.AppsV1beta1().Deployments(\"default\").Update(deployment) if err != nil { panic(err.Error()) } } } func homeDir() string { if h := os.Getenv(\"HOME\"); h != \"\" { return h } return os.Getenv(\"USERPROFILE\") // windows } 我们使用kubeconfig文件认证连接kubernetes集群，该文件默认的位置是$HOME/.kube/config。 该代码编译后可以直接在kubernetes集群之外，任何一个可以连接到API server的机器上运行。 编译运行 $ go get github.com/rootsongjc/kubernetes-client-go-sample $ cd $GOPATH/src/github.com/rootsongjc/kubernetes-client-go-sample $ make $ ./update-deployment-image -h Usage of ./update-deployment-image: -alsologtostderr log to standard error as well as files -app string application name (default \"app\") -deployment string deployment name -image string new image name -kubeconfig string (optional) absolute path to the kubeconfig file (default \"/Users/jimmy/.kube/config\") -log_backtrace_at value when logging hits line file:N, emit a stack trace -log_dir string If non-empty, write log files in this directory -logtostderr log to standard error instead of files -stderrthreshold value logs at or above this threshold go to stderr -v value log level for V logs -vmodule value comma-separated list of pattern=N settings for file-filtered logging 使用不存在的image更新 $ ./update-deployment-image -deployment filebeat-test -image sz-pg-oam-docker-hub-001.tendcloud.com/library/analytics-docker-test:Build_9 Found deployment name -> filebeat-test Old image -> sz-pg-oam-docker-hub-001.tendcloud.com/library/analytics-docker-test:Build_8 New image -> sz-pg-oam-docker-hub-001.tendcloud.com/library/analytics-docker-test:Build_9 查看Deployment的event。 $ kubectl describe deployment filebeat-test Name: filebeat-test Namespace: default CreationTimestamp: Fri, 19 May 2017 15:12:28 +0800 Labels: k8s-app=filebeat-test Selector: k8s-app=filebeat-test Replicas: 2 updated | 3 total | 2 available | 2 unavailable StrategyType: RollingUpdate MinReadySeconds: 0 RollingUpdateStrategy: 1 max unavailable, 1 max surge Conditions: Type Status Reason ---- ------ ------ Available True MinimumReplicasAvailable Progressing True ReplicaSetUpdated OldReplicaSets: filebeat-test-2365467882 (2/2 replicas created) NewReplicaSet: filebeat-test-2470325483 (2/2 replicas created) Events: FirstSeen LastSeen Count From SubObjectPath Type ReasoMessage --------- -------- ----- ---- ------------- -------- ------------ 2h 1m 3 {deployment-controller } Normal ScalingReplicaSet Scaled down replica set filebeat-test-2365467882 to 2 1m 1m 1 {deployment-controller } Normal ScalingReplicaSet Scaled up replica set filebeat-test-2470325483 to 1 1m 1m 1 {deployment-controller } Normal ScalingReplicaSet Scaled up replica set filebeat-test-2470325483 to 2 可以看到老的ReplicaSet从3个replica减少到了2个，有2个使用新配置的replica不可用，目前可用的replica是2个。 这是因为我们指定的镜像不存在，查看Deployment的pod的状态。 $ kubectl get pods -l k8s-app=filebeat-test NAME READY STATUS RESTARTS AGE filebeat-test-2365467882-4zwx8 2/2 Running 0 33d filebeat-test-2365467882-rqskl 2/2 Running 0 33d filebeat-test-2470325483-6vjbw 1/2 ImagePullBackOff 0 4m filebeat-test-2470325483-gc14k 1/2 ImagePullBackOff 0 4m 我们可以看到有两个pod正在拉取image。 还原为原先的镜像 将image设置为原来的镜像。 $ ./update-deployment-image -deployment filebeat-test -image sz-pg-oam-docker-hub-001.tendcloud.com/library/analytics-docker-test:Build_8 Found deployment name -> filebeat-test Old image -> sz-pg-oam-docker-hub-001.tendcloud.com/library/analytics-docker-test:Build_9 New image -> sz-pg-oam-docker-hub-001.tendcloud.com/library/analytics-docker-test:Build_8 现在再查看Deployment的状态。 $ kubectl describe deployment filebeat-test Name: filebeat-test Namespace: default CreationTimestamp: Fri, 19 May 2017 15:12:28 +0800 Labels: k8s-app=filebeat-test Selector: k8s-app=filebeat-test Replicas: 3 updated | 3 total | 3 available | 0 unavailable StrategyType: RollingUpdate MinReadySeconds: 0 RollingUpdateStrategy: 1 max unavailable, 1 max surge Conditions: Type Status Reason ---- ------ ------ Available True MinimumReplicasAvailable Progressing True NewReplicaSetAvailable OldReplicaSets: NewReplicaSet: filebeat-test-2365467882 (3/3 replicas created) Events: FirstSeen LastSeen Count From SubObjectPath Type ReasoMessage --------- -------- ----- ---- ------------- -------- ------------ 2h 8m 3 {deployment-controller } Normal ScalingReplicaSet Scaled down replica set filebeat-test-2365467882 to 2 8m 8m 1 {deployment-controller } Normal ScalingReplicaSet Scaled up replica set filebeat-test-2470325483 to 1 8m 8m 1 {deployment-controller } Normal ScalingReplicaSet Scaled up replica set filebeat-test-2470325483 to 2 2h 1m 3 {deployment-controller } Normal ScalingReplicaSet Scaled up replica set filebeat-test-2365467882 to 3 1m 1m 1 {deployment-controller } Normal ScalingReplicaSet Scaled down replica set filebeat-test-2470325483 to 0 可以看到available的replica个数恢复成3了。 其实在使用该命令的过程中，通过kubernetes dashboard的页面上查看Deployment的状态更直观，更加方便故障排查。 图片 - 使用kubernetes dashboard进行故障排查 这也是dashboard最大的优势，简单、直接、高效。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-08-21 18:23:34 "},"develop/contribute.html":{"url":"develop/contribute.html","title":"社区贡献","keywords":"","body":"Kubernetes社区贡献 Contributing guidelines Kubernetes Developer Guide Special Interest Groups Feature Tracking and Backlog Community Expectations Kubernetes 官方文档汉化项目 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-09-03 13:18:12 "},"develop/minikube.html":{"url":"develop/minikube.html","title":"Minikube","keywords":"","body":"Minikube Minikube用于在本地运行kubernetes环境，用来开发和测试。 在Mac上安装xhyve-driver docker-machine-driver-xhyve # docker-machine-driver-xhyve need root owner and uid $ sudo chown root:wheel $(brew --prefix)/opt/docker-machine-driver-xhyve/bin/docker-machine-driver-xhyve $ sudo chmod u+s $(brew --prefix)/opt/docker-machine-driver-xhyve/bin/docker-machine-driver-xhyve 到 https://github.com/kubernetes/minikube/releases 下载 minikube，我安装的是minikube v0.22.3 下载完成后修改文件名为minikube，然后chmod +x minikube，移动到$PATH目录下： mv ~/Download/minikube-darwin-adm64 /usr/loal/bin/ chmod +x /usr/local/bin/minikube 安装kubectl 参考Install and Set Up kubectl，直接使用二进制文件安装即可。 curl -LO https://storage.googleapis.com/kubernetes-release/release/`curl -s https://storage.googleapis.com/kubernetes-release/release/stable.txt`/bin/darwin/amd64/kubectl 启动Minikube minikube start Starting local Kubernetes v1.7.5 cluster... Starting VM... Downloading Minikube ISO 139.09 MB / 139.09 MB [============================================] 100.00% 0s Getting VM IP address... Moving files into cluster... Setting up certs... Connecting to cluster... Setting up kubeconfig... Starting cluster components... Kubectl is now configured to use the cluster. 这将生成默认的~/.kube/config文件，自动指向minikube。 停止Minikube minikube stop 参考 Running Kubernetes Locally via Minikube Install minikube Driver plugin installation - xhyve-driver Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-10-23 19:06:48 "},"appendix/":{"url":"appendix/","title":"附录说明","keywords":"","body":"附录说明 参考文档以及一些实用的资源链接。 Kubernetes documentation Awesome Kubernetes Kubernetes the hard way Kubernetes Bootcamp Design patterns for container-based distributed systems Awesome Cloud Native Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-15 19:43:33 "},"appendix/debug-kubernetes-services.html":{"url":"appendix/debug-kubernetes-services.html","title":"Kubernetes中的应用故障排查","keywords":"","body":"Kubernetes service中的故障排查 查看某个资源的定义和用法 kubectl explain 查看Pod的状态 kubectl get pods kubectl describe pods my-pod 监控Pod状态的变化 kubectl get pod -w 可以看到一个 namespace 中所有的 pod 的 phase 变化，请参考 Pod 的生命周期。 查看 Pod 的日志 kubectl logs my-pod kubectl logs my-pod -c my-container kubectl logs -f my-pod kubectl logs -f my-pod -c my-container -f 参数可以 follow 日志输出。 交互式 debug kubectl exec my-pod -it /bin/bash kubectl top pod POD_NAME --containers Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-09-21 14:58:20 "},"appendix/material-share.html":{"url":"appendix/material-share.html","title":"Kubernetes相关资讯和情报链接","keywords":"","body":"Kubernetes相关资讯和情报链接 授人以鱼，不如授人以渔。下面的资料将有助于大家了解kubernetes生态圈当前发展状况和发展趋势，我特此整理相关资料如下。 生态环境 包括kubernetes和cloud native相关的开源软件、工具和全景图。 awesome-kubernetes - A curated list for awesome kubernetes sources 🚢🎉 https://ramitsurana.github.io/awesome… awesome-cloud-native - A curated list for awesome cloud native architectures https://jimmysong.io/awesome-cloud-native/ cloud native landscape - Cloud Native Landscape https://cncf.io 开源书籍和教程 arun-gupta/kubernetes-aws-workshop arun-gupta/kubernetes-java-sample feiskyer/kubernetes-handbook kelseyhightower/kubernetes-the-hard-way ks - A series of Kubernetes walk-throughs opsnull/follow-me-install-kubernetes-cluster rootsongjc/kubernetes-handbook 幻灯片、图书和情报资料分享 cloud-native-slides-share - Cloud Native 相关meetup、会议PPT、图书资料分享 博客与网站 Kubernetes和Cloud Native相关网站、专栏、博客等。 网站与专栏 thenewstack.io k8sport.org giantswarm blog k8smeetup.com dockone.io Cloud Native知乎专栏 kubernetes.org.cn 博客 apcera aporeto applatix apprenda bitnami buoyant cisco cncf codeship containership coreos coscale deis fabric8 grafana gravitational heptio istio jimmysong kubernetes moby openshift pivotal platform9 prometheus rancher sysdig spinnaker supergiant thecodeteam twistlock vamp weave wercker Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-21 18:51:19 "},"appendix/docker-best-practice.html":{"url":"appendix/docker-best-practice.html","title":"Docker最佳实践","keywords":"","body":"Docker最佳实践 本文档旨在实验Docker1.13新特性和帮助大家了解docker集群的管理和使用。 环境配置 Docker1.13环境配置 docker源码编译 网络管理 网络配置和管理是容器使用中的的一个重点和难点，对比我们之前使用的docker版本是1.11.1，docker1.13中网络模式跟之前的变动比较大，我们会花大力气讲解。 如何创建docker network Rancher网络探讨和扁平网络实现 swarm mode的路由网络 docker扁平化网络插件Shrike（基于docker1.11） 存储管理 Docker存储插件 infinit 被docker公司收购的法国团队开发 convoy rancher开发的docker volume plugin torus 已废弃 flocker ClusterHQ开发 日志管理 Docker提供了一系列log drivers，如fluentd、journald、syslog等。 需要配置docker engine的启动参数。 创建应用 官方文档：Docker swarm sample app overview 基于docker1.13手把手教你创建swarm app swarm集群应用管理 使用docker-compose创建应用 集群管理 我们使用docker内置的swarm来管理docker集群。 swarm mode介绍 我们推荐使用开源的docker集群管理配置方案： Crane：由数人云开源的基于swarmkit的容器管理软件，可以作为docker和go语言开发的一个不错入门项目 Rancher:Rancher是一个企业级的容器管理平台，可以使用Kubernetes、swarm和rancher自研的cattle来管理集群。 Crane的部署和使用 Rancher的部署和使用 资源限制 内存资源限制 CPU资源限制 IO资源限制 服务发现 下面罗列一些列常见的服务发现工具。 Etcd:服务发现/全局分布式键值对存储。这是CoreOS的创建者提供的工具，面向容器和宿主机提供服务发现和全局配置存储功能。它在每个宿主机有基于http协议的API和命令行的客户端。https://github.com/docker/etcd Cousul：服务发现/全局分布式键值对存储。这个服务发现平台有很多高级的特性，使得它能够脱颖而出，例如：配置健康检查、ACL功能、HAProxy配置等等。 Zookeeper：诞生于Hadoop生态系统里的组件，Apache的开源项目。服务发现/全局分布式键值对存储。这个工具较上面两个都比较老，提供一个更加成熟的平台和一些新特性。 Crypt：加密etcd条目的项目。Crypt允许组建通过采用公钥加密的方式来保护它们的信息。需要读取数据的组件或被分配密钥，而其他组件则不能读取数据。 Confd：观测键值对存储变更和新值的触发器重新配置服务。Confd项目旨在基于服务发现的变化，而动态重新配置任意应用程序。该系统包含了一个工具来检测节点中的变化、一个模版系统能够重新加载受影响的应用。 Vulcand：vulcand在组件中作为负载均衡使用。它使用etcd作为后端，并基于检测变更来调整它的配置。 Marathon：虽然marathon主要是调度器，它也实现了一个基本的重新加载HAProxy的功能，当发现变更时，它来协调可用的服务。 Frontrunner：这个项目嵌入在marathon中对HAproxy的更新提供一个稳定的解决方案。 Synapse：由Airbnb出品的，Ruby语言开发，这个项目引入嵌入式的HAProxy组件，它能够发送流量给各个组件。http://bruth.github.io/synapse/docs/ Nerve：它被用来与synapse结合在一起来为各个组件提供健康检查，如果组件不可用，nerve将更新synapse并将该组件移除出去。 插件开发 插件开发示例-sshfs 我的docker插件开发文章 Docker17.03-CE插件开发举例 网络插件 Contiv 思科出的Docker网络插件，趟坑全记录，目前还无法上生产，1.0正式版还没出，密切关注中。 Calico 产品化做的不错，已经有人用在生产上了。 存储插件 业界使用案例 京东从OpenStack切换到Kubernetes的经验之谈 美团点评容器平台介绍 阿里超大规模docker化之路 TalkingData-容器技术在大数据场景下的应用Yarn on Docker 乐视云基于Kubernetes的PaaS平台建设 资源编排 建议使用kuberentes，虽然比较复杂，但是专业的工具做专业的事情，将编排这么重要的生产特性绑定到docker上的风险还是很大的，我已经转投到kubernetes怀抱了，那么你呢？ 我的kubernetes探险之旅 相关资源 容器技术工具与资源 容器技术2016年总结 关于 Author: Jimmy Song rootsongjc@gmail.com 更多关于Docker、MicroServices、Big Data、DevOps、Deep Learning的内容请关注Jimmy Song's Blog，将不定期更新。 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-10 18:13:03 "},"appendix/tricks.html":{"url":"appendix/tricks.html","title":"使用技巧","keywords":"","body":"1. 在容器中获取 Pod 的IP 通过环境变量来实现，该环境变量直接引用 resource 的状态字段，示例如下： apiVersion: v1 kind: ReplicationController metadata: name: world-v2 spec: replicas: 3 selector: app: world-v2 template: metadata: labels: app: world-v2 spec: containers: - name: service image: test env: - name: POD_IP valueFrom: fieldRef: fieldPath: status.podIP ports: - name: service containerPort: 777 容器中可以直接使用 POD_IP 环境变量获取容器的 IP。 2. 指定容器的启动参数 我们可以在 Pod 中为容器使用 command 为容器指定启动参数： command: [\"/bin/bash\",\"-c\",\"bootstrap.sh\"] 看似很简单，使用数组的方式定义，所有命令使用跟 Dockerfile 中的 CMD 配置是一样的，但是有一点不同的是，bootsttap.sh 必须具有可执行权限，否则容器启动时会出错。 3. 让Pod调用宿主机的docker能力 我们可以想象一下这样的场景，让 Pod 来调用宿主机的 docker 能力，只需要将宿主机的 docker 命令和 docker.sock 文件挂载到 Pod 里面即可，如下： apiVersion: v1 kind: Pod metadata: name: busybox-cloudbomb spec: containers: - image: busybox command: - /bin/sh - \"-c\" - \"while true; \\ do \\ docker run -d --name BOOM_$(cat /dev/urandom | tr -cd 'a-f0-9' | head -c 6) nginx ; \\ done\" name: cloudbomb volumeMounts: - mountPath: /var/run/docker.sock name: docker-socket - mountPath: /bin/docker name: docker-binary volumes: - name: docker-socket hostPath: path: /var/run/docker.sock - name: docker-binary hostPath: path: /bin/docker 参考：Architecture Patterns for Microservices in Kubernetes 4. 使用Init container初始化应用配置 Init container可以在应用程序的容器启动前先按顺序执行一批初始化容器，只有所有Init容器都启动成功后，Pod才算启动成功。看下下面这个例子（来源：kubernetes: mounting volume from within init container - Stack Overflow）： apiVersion: v1 kind: Pod metadata: name: init labels: app: init annotations: pod.beta.kubernetes.io/init-containers: '[ { \"name\": \"download\", \"image\": \"axeclbr/git\", \"command\": [ \"git\", \"clone\", \"https://github.com/mdn/beginner-html-site-scripted\", \"/var/lib/data\" ], \"volumeMounts\": [ { \"mountPath\": \"/var/lib/data\", \"name\": \"git\" } ] } ]' spec: containers: - name: run image: docker.io/centos/httpd ports: - containerPort: 80 volumeMounts: - mountPath: /var/www/html name: git volumes: - emptyDir: {} name: git 这个例子就是用来再应用程序启动前首先从GitHub中拉取代码并存储到共享目录下。 关于Init容器的更详细说明请参考 init容器。 5. 使容器内时间与宿主机同步 我们下载的很多容器内的时区都是格林尼治时间，与北京时间差8小时，这将导致容器内的日志和文件创建时间与实际时区不符，有两种方式解决这个问题： 修改镜像中的时区配置文件 将宿主机的时区配置文件/etc/localtime使用volume方式挂载到容器中 第二种方式比较简单，不需要重做镜像，只要在应用的yaml文件中增加如下配置： volumeMounts: - name: host-time mountPath: /etc/localtime readOnly: true volumes: - name: host-time hostPath: path: /etc/localtime 6. 在Pod中获取宿主机的主机名、namespace等 这条技巧补充了第一条获取 podIP 的内容，方法都是一样的，只不过列出了更多的引用字段。 参考下面的 pod 定义，每个 pod 里都有一个 {.spec.nodeName} 字段，通过 fieldRef 和环境变量，就可以在Pod中获取宿主机的主机名（访问环境变量MY_NODE_NAME）。 apiVersion: v1 kind: Pod metadata: name: dapi-test-pod spec: containers: - name: test-container image: busybox command: [ \"/bin/sh\", \"-c\", \"env\" ] env: - name: MY_NODE_NAME valueFrom: fieldRef: fieldPath: spec.nodeName - name: MY_POD_NAME valueFrom: fieldRef: fieldPath: metadata.name - name: MY_POD_NAMESPACE valueFrom: fieldRef: fieldPath: metadata.namespace - name: MY_POD_IP valueFrom: fieldRef: fieldPath: status.podIP - name: MY_POD_SERVICE_ACCOUNT valueFrom: fieldRef: fieldPath: spec.serviceAccountName restartPolicy: Never 7. 配置Pod使用外部DNS 修改kube-dns的使用的ConfigMap。 apiVersion: v1 kind: ConfigMap metadata: name: kube-dns namespace: kube-system data: stubDomains: | {\"k8s.com\": [\"192.168.10.10\"]} upstreamNameservers: | [\"8.8.8.8\", \"8.8.4.4\"] upstreamNameservers 即使用的外部DNS，参考：Configuring Private DNS Zones and Upstream Nameservers in Kubernetes Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-11-21 16:44:45 "},"appendix/issues.html":{"url":"appendix/issues.html","title":"问题记录","keywords":"","body":"问题记录 安装、使用kubernetes的过程中遇到的所有问题的记录。 推荐直接在Kubernetes的GitHub上提issue，在此记录所提交的issue。 1.Failed to start ContainerManager failed to initialise top level QOS containers #43856 重启kubelet时报错，目前的解决方法是： 1.在docker.service配置中增加的--exec-opt native.cgroupdriver=systemd配置。 2.手动删除slice（貌似不管用） 3.重启主机，这招最管用😄 for i in $(systemctl list-unit-files —no-legend —no-pager -l | grep —color=never -o .*.slice | grep kubepod);do systemctl stop $i;done 上面的几种方法在该bug修复前只有重启主机管用，该bug已于2017年4月27日修复，merge到了master分支，见https://github.com/kubernetes/kubernetes/pull/44940 2.High Availability of Kube-apiserver #19816 API server的HA如何实现？或者说这个master节点上的服务api-server、scheduler、controller 如何实现HA？目前的解决方案是什么？ 目前的解决方案是api-server是无状态的可以启动多个，然后在前端再加一个nginx或者ha-proxy。而scheduler和controller都是直接用容器的方式启动的。 3.Kubelet启动时Failed to start ContainerManager systemd version does not support ability to start a slice as transient unit CentOS系统版本7.2.1511 kubelet启动时报错systemd版本不支持start a slice as transient unit。 尝试升级CentOS版本到7.3，看看是否可以修复该问题。 与kubeadm init waiting for the control plane to become ready on CentOS 7.2 with kubeadm 1.6.1 #228类似。 另外有一个使用systemd管理kubelet的proposal。 4.kube-proxy报错kube-proxy[2241]: E0502 15:55:13.889842 2241 conntrack.go:42] conntrack returned error: error looking for path of conntrack: exec: \"conntrack\": executable file not found in $PATH 导致的现象 kubedns启动成功，运行正常，但是service之间无法解析，kubernetes中的DNS解析异常 解决方法 CentOS中安装conntrack-tools包后重启kubernetes集群即可。 5. Pod stucks in terminating if it has a privileged container but has been scheduled to a node which doesn't allow privilege issue#42568 当pod被调度到无法权限不足的node上时，pod一直处于pending状态，且无法删除pod，删除时一直处于terminating状态。 kubelet中的报错信息 Error validating pod kube-keepalived-vip-1p62d_default(5d79ccc0-3173-11e7-bfbd-8af1e3a7c5bd) from api, ignoring: spec.containers[0].securityContext.privileged: Forbidden: disallowed by cluster policy 6.PVC中对Storage的容量设置不生效 使用glusterfs做持久化存储文档中我们构建了PV和PVC，当时给glusterfs-nginx的PVC设置了8G的存储限额，nginx-dm这个Deployment使用了该PVC，进入该Deployment中的Pod执行测试： dd if=/dev/zero of=test bs=1G count=10 图片 - pvc-storage-limit 从截图中可以看到创建了9个size为1G的block后无法继续创建了，已经超出了8G的限额。 7. 使用 Headless service 的时候 kubedns 解析不生效 kubelet 的配置文件 /etc/kubernetes/kubelet 中的配置中将集群 DNS 的 domain name 配置成了 ––cluster-domain=cluster.local. ，虽然对于 service 的名字能够正常的完成 DNS 解析，但是对于 headless service 中的 pod 名字解析不了，查看 pod 的 /etc/resolv.conf 文件可以看到以下内容： nameserver 10.0.254.2 search default.svc.cluster.local. svc.cluster.local. cluster.local. tendcloud.com options ndots:5 修改 /etc/kubernetes/kubelet 文件中的 ––cluster-domain=cluster.local. 将 local 后面的点去掉后重启所有的 kubelet，这样新创建的 pod 中的 /etc/resolv.conf文件的 DNS 配置和解析就正常了。 8. kubernetes 集成 ceph 存储 rbd 命令组装问题 kubernetes 使用 ceph 创建 PVC 的时候会有如下报错信息： Events: FirstSeen LastSeen Count From SubObjectPath Type Reason Message --------- -------- ----- ---- ------------- -------- ------ ------- 1h 12s 441 {persistentvolume-controller } Warning ProvisioningFailed Failed to provision volume with StorageClass \"ceph-web\": failed to create rbd image: executable file not found in $PATH, command output: 检查 kube-controller-manager 的日志将看到如下错误信息： Sep 4 15:25:36 bj-xg-oam-kubernetes-001 kube-controller-manager: W0904 15:25:36.032128 13211 rbd_util.go:364] failed to create rbd image, output Sep 4 15:25:36 bj-xg-oam-kubernetes-001 kube-controller-manager: W0904 15:25:36.032201 13211 rbd_util.go:364] failed to create rbd image, output Sep 4 15:25:36 bj-xg-oam-kubernetes-001 kube-controller-manager: W0904 15:25:36.032252 13211 rbd_util.go:364] failed to create rbd image, output Sep 4 15:25:36 bj-xg-oam-kubernetes-001 kube-controller-manager: E0904 15:25:36.032276 13211 rbd.go:317] rbd: create volume failed, err: failed to create rbd image: fork/exec /usr/bin/rbd: invalid argument, command output: 该问题尚未解决，参考 Error creating rbd image: executable file not found in $PATH#38923 9. Helm: Error: no available release name found 在开启了RBAC的kubernetes集群中，当使用helm部署应用，执行helm install的时候，会报着个错误： Error: no available release name found Error: the server does not allow access to the requested resource (get configmaps) 这是因为我们使用的2.3.1版本的helm init的时候没有为tiller创建serviceaccount和clusterrolebiding的缘故导致的。 kubectl create serviceaccount --namespace kube-system tiller kubectl create clusterrolebinding tiller-cluster-rule --clusterrole=cluster-admin --serviceaccount=kube-system:tiller # helm init -i sz-pg-oam-docker-hub-001.tendcloud.com/library/kubernetes-helm-tiller:v2.3.1 kubectl patch deploy --namespace kube-system tiller-deploy -p '{\"spec\":{\"template\":{\"spec\":{\"serviceAccount\":\"tiller\"}}}}' 参考 Helm: Error: no available release name found - StackOverflow Helm 2.2.3 not working properly with kubeadm 1.6.1 default RBAC rules #2224 参考 Persistent Volume Resource Design Proposals Helm: Error: no available release name found Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-10-24 19:04:57 "},"appendix/kubernetes-changelog.html":{"url":"appendix/kubernetes-changelog.html","title":"kubernetes版本更新日志","keywords":"","body":"Kubernetes版本更新日志 刚开始写作本书时kubernetes1.6刚刚发布，随后kubernetes基本以每3个月发布一个版本的速度不断迭代，为了追踪不同版本的新特性，我们有必要在此记录一下。 每个kubernetes版本的详细更新日志请参考：https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG.md 发布记录 2017年6月29日 Kubernetes1.7发布 2017年9月28日 Kubernetes1.8发布 2017年12月15日 Kubernetes1.9发布 Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-16 10:11:41 "},"appendix/kubernetes-1.7-changelog.html":{"url":"appendix/kubernetes-1.7-changelog.html","title":"Kubernetes1.7更新日志","keywords":"","body":"Kubernetes1.7更新日志 2017年6月29日，kuberentes1.7发布。该版本的kubernetes在安全性、存储和可扩展性方面有了很大的提升。改版本的详细更新文档请查看Changelog。 这些新特性中包含了安全性更高的加密的secret、pod间通讯的网络策略，限制kubelet访问的节点授权程序以及客户端/服务器TLS证书轮换。 对于那些在Kubernetes上运行横向扩展数据库的人来说，这个版本有一个主要的特性，可以为StatefulSet添加自动更新并增强DaemonSet的更新。我们还宣布了对本地存储的Alpha支持，以及用于更快地缩放StatefulSets的突发模式。 此外，对于高级用户，此发行版中的API聚合允许使用用于自定义的API与API server同时运行。其他亮点包括支持可扩展的准入控制器，可插拔云供应商程序和容器运行时接口（CRI）增强功能。 新功能 安全 Network Policy API 提升为稳定版本。用户可以通过使用网络插件实现的网络策略来控制哪些Pod之间能够互相通信。 节点授权和准入控制插件是新增加的功能，可以用于限制kubelet可以访问的secret、pod和其它基于节点的对象。 加密的Secret和etcd中的其它资源，现在是alpha版本。 Kubelet TLS bootstrapping现在支持客户端和服务器端的证书轮换。 由API server存储的审计日志现在更具可定制性和可扩展性，支持事件过滤和webhook。它们还为系统审计提供更丰富的数据。 有状态负载 StatefulSet更新是1.7版本的beta功能，它允许使用包括滚动更新在内的一系列更新策略自动更新诸如Kafka，Zookeeper和etcd等有状态应用程序。 StatefulSets现在还支持对不需要通过Pod管理策略进行排序的应用程序进行快速扩展和启动。这可以是主要的性能改进。 本地存储（alpha）是有状态应用程序最常用的功能之一。用户现在可以通过标准的PVC/PV接口和StatefulSet中的StorageClass访问本地存储卷。 DaemonSet——为每个节点创建一个Pod，现在有了更新功能，在1.7中增加了智能回滚和历史记录功能。 新的StorageOS Volume插件可以使用本地或附加节点存储中以提供高可用的集群范围的持久卷。 可扩展性 运行时的API聚合是此版本中最强大的扩展功能，允许高级用户将Kubernetes风格的预先构建的第三方或用户创建的API添加到其集群中。 容器运行时接口（CRI）已经增强，可以使用新的RPC调用从运行时检索容器度量。 CRI的验证测试已经发布，与containerd进行了Alpha集成，现在支持基本的生命周期和镜像管理。参考深入介绍CRI的文章。 其它功能 引入了对外部准入控制器的Alpha支持，提供了两个选项，用于向API server添加自定义业务逻辑，以便在创建对象和验证策略时对其进行修改。 基于策略的联合资源布局提供Alpha版本，用于根据自定义需求（如法规、定价或性能）为联合（federated）集群提供布局策略。 弃用 第三方资源（TPR）已被自定义资源定义（Custom Resource Definitions，CRD）取代，后者提供了一个更清晰的API，并解决了TPR测试期间引发的问题和案例。如果您使用TPR测试版功能，则建议您迁移，因为它将在Kubernetes 1.8中被移除。 以上是Kubernetes1.7中的主要新特性，详细更新文档请查看Changelog。 参考 Kuberentes 1.7: Security Hardening, Stateful Application Updates and Extensibility Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-16 10:10:15 "},"appendix/kubernetes-1.8-changelog.html":{"url":"appendix/kubernetes-1.8-changelog.html","title":"Kubernetes1.8更新日志","keywords":"","body":"Kubernetes1.8更新日志 2017年9月28日，kubernetes1.8版本发布。该版本中包括了一些功能改进和增强，并增加了项目的成熟度，将强了kubernetes的治理模式，这些都将有利于kubernetes项目的持续发展。 聚焦安全性 Kubernetes1.8的基于角色的访问控制（RBAC）成为stable支持。RBAC允许集群管理员动态定义角色对于Kubernetes API的访问策略。通过网络策略筛选出站流量的Beta支持，增强了对入站流量进行过滤的现有支持。 RBAC和网络策略是强化Kubernetes内组织和监管安全要求的两个强大工具。 Kubelet的传输层安全性（TLS）证书轮换成为beta版。自动证书轮换减轻了集群安全性运维的负担。 聚焦工作负载支持 Kubernetes 1.8通过apps/v1beta2组和版本推动核心工作负载API的beta版本。Beta版本包含当前版本的Deployment、DaemonSet、ReplicaSet和StatefulSet。 工作负载API是将现有工作负载迁移到Kubernetes以及开发基于Kubernetes的云原生应用程序提供了基石。 对于那些考虑在Kubernetes上运行大数据任务的，现在的工作负载API支持运行kubernetes原生支持的Apache Spark。 批量工作负载，比如夜间ETL工作，将从CronJobs的beta版本中受益。 自定义资源定义（CRD）在Kubernetes 1.8中依然为测试版。CRD提供了一个强大的机制来扩展Kubernetes和用户定义的API对象。 CRD的一个用例是通过Operator Pattern自动执行复杂的有状态应用，例如键值存储、数据库和存储引擎。随着稳定性的继续推进，预计将继续加强对CRD的验证。 更多 Volume快照、PV调整大小、自动taint、pod优先级、kubectl插件等！ 除了稳定现有的功能，Kubernetes 1.8还提供了许多预览新功能的alpha版本。 社区中的每个特别兴趣小组（SIG）都在继续为所在领域的用户提供更多的功能。有关完整列表，请访问发行说明。 参考 Kubernetes 1.8: Security, Workloads and Feature Depth Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-12 13:44:48 "},"appendix/kubernetes-1.9-changelog.html":{"url":"appendix/kubernetes-1.9-changelog.html","title":"Kubernetes1.9更新日志","keywords":"","body":"Kubernetes1.9更新日志 2017年12月15日，kubernetes1.9版本发布。Kubernetes依然按照每三个月一个大版本发布的速度稳定迭代，这是今年发布的第四个版本，也是今年的最后一个版本，该版本最大的改进是Apps Workloads API成为稳定版本，这消除了很多潜在用户对于该功能稳定性的担忧。还有一个重大更新，就是测试支持了Windows了，这打开了在kubernetes中运行Windows工作负载的大门。 Workloads API GA apps/v1 Workloads API成为GA（General Availability），且默认启用。 Apps Workloads API将DaemonSet、Deployment、ReplicaSet和StatefulSet API组合在一起，作为Kubernetes中长时间运行的无状态和有状态工作负载的基础。 Deployment和ReplicaSet是Kubernetes中最常用的两个对象，经过一年多的实际使用和反馈后，现在已经趋于稳定。SIG apps同时将这些经验应用到另外的两个对象上，使得DaemonSet和StatefulSet也能顺利毕业走向成熟。v1（GA）意味着已经生产可用，并保证长期的向后兼容。 Windows支持（beta） Kubernetes最初是为Linux系统开发的，但是用户逐渐意识到容器编排的好处，我们看到有人需要在Kubernetes上运行Windows工作负载。在12个月前，我们开始认真考虑在Kubernetes上支持Windows Server的工作。 SIG-Windows现在已经将这个功能推广到beta版本，这意味着我们可以评估它的使用情况。 增强存储 kubernetes从第一个版本开始就支持多种持久化数据存储，包括常用的NFS或iSCSI，以及对主要公共云和私有云提供商的存储解决方案的原生支持。随着项目和生态系统的发展，Kubernetes的存储选择越来越多。然而，为新的存储系统添加volume插件一直是一个挑战。 容器存储接口（CSI）是一个跨行业标准计划，旨在降低云原生存储开发的障碍并确保兼容性。 SIG-Storage和CSI社区正在合作提供一个单一接口，用于配置、附着和挂载与Kubernetes兼容的存储。 Kubernetes 1.9引入了容器存储接口（CSI）的alpha实现，这将使挂载新的volume插件就像部署一个pod一样简单，并且第三方存储提供商在开发他们的解决方案时也无需修改kubernetes的核心代码。 由于该功能在1.9版本中为alpha，因此必须明确启用该功能，不建议用于生产使用，但它为更具扩展性和基于标准的Kubernetes存储生态系统提供了清晰的路线图。 其它功能 自定义资源定义（CRD）校验，现在已经成为beta，默认情况下已启用，可以用来帮助CRD作者对于无效对象定义给出清晰和即时的反馈。 SIG Node硬件加速器转向alpha，启用GPU，从而实现机器学习和其他高性能工作负载。 CoreDNS alpha可以使用标准工具来安装CoreDNS。 kube-proxy的IPVS模式进入beta版，为大型集群提供更好的可扩展性和性能。 社区中的每个特别兴趣小组（SIG）继续提供其所在领域的用户最迫切需要的功能。有关完整列表，请访问发行说明。 获取 Kubernetes1.9已经可以通过GitHub下载。 参考 Kubernetes 1.9: Apps Workloads GA and Expanded Ecosystem Copyright © jimmysong.io 2017 all right reserved，powered by GitbookUpdated: 2017-12-16 10:47:44 "}} \ No newline at end of file diff --git a/usecases/big-data.html b/usecases/big-data.html index 861d8e135..da99ca6b7 100644 --- a/usecases/big-data.html +++ b/usecases/big-data.html @@ -2797,7 +2797,7 @@ diff --git a/usecases/conduit-installation.html b/usecases/conduit-installation.html index 3254dc08c..a8d0513bb 100644 --- a/usecases/conduit-installation.html +++ b/usecases/conduit-installation.html @@ -2958,7 +2958,7 @@ kubectl apply -f conduit-0.1.0.yaml diff --git a/usecases/conduit-overview.html b/usecases/conduit-overview.html index a5702ac6c..140e32269 100644 --- a/usecases/conduit-overview.html +++ b/usecases/conduit-overview.html @@ -2798,7 +2798,7 @@ diff --git a/usecases/conduit.html b/usecases/conduit.html index 928e72b98..ca48be213 100644 --- a/usecases/conduit.html +++ b/usecases/conduit.html @@ -2787,7 +2787,7 @@ diff --git a/usecases/configuring-request-routing.html b/usecases/configuring-request-routing.html index 2f2148cea..ff30898a0 100644 --- a/usecases/configuring-request-routing.html +++ b/usecases/configuring-request-routing.html @@ -2875,7 +2875,7 @@ istioctl delete route-rule ratings-test-delay diff --git a/usecases/edge-computing.html b/usecases/edge-computing.html index 362c485df..30dfa261c 100644 --- a/usecases/edge-computing.html +++ b/usecases/edge-computing.html @@ -2782,7 +2782,7 @@ diff --git a/usecases/faas.html b/usecases/faas.html index fcc35038d..9830cf688 100644 --- a/usecases/faas.html +++ b/usecases/faas.html @@ -2792,7 +2792,7 @@ diff --git a/usecases/index.html b/usecases/index.html index 4b70cc2c2..a81db7757 100644 --- a/usecases/index.html +++ b/usecases/index.html @@ -2779,7 +2779,7 @@ diff --git a/usecases/install-and-expand-istio-mesh.html b/usecases/install-and-expand-istio-mesh.html index 8846c25c9..efc9876f8 100644 --- a/usecases/install-and-expand-istio-mesh.html +++ b/usecases/install-and-expand-istio-mesh.html @@ -3523,7 +3523,7 @@ docker ps -a var gitbook = gitbook || []; gitbook.push(function() { - gitbook.page.hasChanged({"page":{"title":"安装和拓展Istio service mesh","level":"6.3.1.3","depth":3,"next":{"title":"集成虚拟机","level":"6.3.1.4","depth":3,"path":"usecases/integrating-vms.md","ref":"usecases/integrating-vms.md","articles":[]},"previous":{"title":"配置请求的路由规则","level":"6.3.1.2","depth":3,"path":"usecases/configuring-request-routing.md","ref":"usecases/configuring-request-routing.md","articles":[]},"dir":"ltr"},"config":{"plugins":["github","codesnippet","splitter","page-toc-button","image-captions","editlink","back-to-top-button","-lunr","-search","search-plus","github-buttons@2.1.0","favicon@^0.0.2","tbfed-pagefooter@^0.0.1","3-ba","theme-default"],"styles":{"website":"styles/website.css","pdf":"styles/pdf.css","epub":"styles/epub.css","mobi":"styles/mobi.css","ebook":"styles/ebook.css","print":"styles/print.css"},"pluginsConfig":{"tbfed-pagefooter":{"copyright":"Copyright © jimmysong.io 2017","modify_label":"Updated:","modify_format":"YYYY-MM-DD HH:mm:ss"},"github":{"url":"https://github.com/rootsongjc/kubernetes-handbook"},"editlink":{"label":"编辑本页","multilingual":false,"base":"https://github.com/rootsongjc/kubernetes-handbook/blob/master/"},"splitter":{},"codesnippet":{},"fontsettings":{"theme":"white","family":"sans","size":2},"highlight":{},"favicon":{"shortcut":"favicon.ico","bookmark":"favicon.ico"},"page-toc-button":{},"back-to-top-button":{},"github-buttons":{"repo":"rootsongjc/kubernetes-handbook","types":["star"],"size":"small"},"3-ba":{"configuration":"auto","token":"11f7d254cfa4e0ca44b175c66d379ecc"},"sharing":{"facebook":true,"twitter":true,"google":false,"weibo":false,"instapaper":false,"vk":false,"all":["facebook","google","twitter","weibo","instapaper"]},"theme-default":{"styles":{"website":"styles/website.css","pdf":"styles/pdf.css","epub":"styles/epub.css","mobi":"styles/mobi.css","ebook":"styles/ebook.css","print":"styles/print.css"},"showLevel":true},"search-plus":{},"image-captions":{"caption":"图片 - _CAPTION_","variable_name":"_pictures"}},"theme":"default","author":"Jimmy Song","pdf":{"pageNumbers":true,"fontSize":12,"fontFamily":"Arial","paperSize":"a4","chapterMark":"pagebreak","pageBreaksBefore":"/","margin":{"right":62,"left":62,"top":56,"bottom":56}},"structure":{"langs":"LANGS.md","readme":"README.md","glossary":"GLOSSARY.md","summary":"SUMMARY.md"},"variables":{"_pictures":[{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.1","level":"2.1","list_caption":"Figure: 云计算演进历程","alt":"云计算演进历程","nro":1,"url":"../images/cloud-computing-evolution-road.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"云计算演进历程","attributes":{},"skip":false,"key":"2.1.1"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.2","level":"2.1","list_caption":"Figure: 来自Twitter @MarcWilczek","alt":"来自Twitter @MarcWilczek","nro":2,"url":"../images/cloud-native-comes-of-age.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"来自Twitter @MarcWilczek","attributes":{},"skip":false,"key":"2.1.2"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.3","level":"2.1","list_caption":"Figure: Cloud native思维导图","alt":"Cloud native思维导图","nro":3,"url":"../images/cloud-native-architecutre-mindnode.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"Cloud native思维导图","attributes":{},"skip":false,"key":"2.1.3"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.4","level":"2.1","list_caption":"Figure: 十二因素应用","alt":"十二因素应用","nro":4,"url":"../images/12-factor-app.png","index":4,"caption_template":"图片 - _CAPTION_","label":"十二因素应用","attributes":{},"skip":false,"key":"2.1.4"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.5","level":"2.1","list_caption":"Figure: 容器生态","alt":"容器生态","nro":5,"url":"../images/container-ecosystem.png","index":5,"caption_template":"图片 - _CAPTION_","label":"容器生态","attributes":{},"skip":false,"key":"2.1.5"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.6","level":"2.1","list_caption":"Figure: 使用Jenkins进行持续集成与发布流程图","alt":"使用Jenkins进行持续集成与发布流程图","nro":6,"url":"../images/kubernetes-jenkins-ci-cd.png","index":6,"caption_template":"图片 - _CAPTION_","label":"使用Jenkins进行持续集成与发布流程图","attributes":{},"skip":false,"key":"2.1.6"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.7","level":"2.1","list_caption":"Figure: filebeat日志收集架构图","alt":"filebeat日志收集架构图","nro":7,"url":"../images/filebeat-log-collector-arch.png","index":7,"caption_template":"图片 - _CAPTION_","label":"filebeat日志收集架构图","attributes":{},"skip":false,"key":"2.1.7"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.8","level":"2.1","list_caption":"Figure: API文档","alt":"API文档","nro":8,"url":"../images/k8s-app-monitor-test-api-doc.jpg","index":8,"caption_template":"图片 - _CAPTION_","label":"API文档","attributes":{},"skip":false,"key":"2.1.8"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.9","level":"2.1","list_caption":"Figure: 迁移步骤示意图","alt":"迁移步骤示意图","nro":9,"url":"../images/migrating-hadoop-yarn-to-kubernetes.png","index":9,"caption_template":"图片 - _CAPTION_","label":"迁移步骤示意图","attributes":{},"skip":false,"key":"2.1.9"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.10","level":"2.1","list_caption":"Figure: service mesh架构图","alt":"service mesh架构图","nro":10,"url":"../images/serivce-mesh-control-plane.png","index":10,"caption_template":"图片 - _CAPTION_","label":"service mesh架构图","attributes":{},"skip":false,"key":"2.1.10"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.11","level":"2.1","list_caption":"Figure: kibana界面","alt":"kibana界面","nro":11,"url":"../images/filebeat-docker-test.jpg","index":11,"caption_template":"图片 - _CAPTION_","label":"kibana界面","attributes":{},"skip":false,"key":"2.1.11"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.12","level":"2.1","list_caption":"Figure: Grafana界面示意图1","alt":"Grafana界面示意图1","nro":12,"url":"../images/kubernetes-devops-example-grafana-1.png","index":12,"caption_template":"图片 - _CAPTION_","label":"Grafana界面示意图1","attributes":{},"skip":false,"key":"2.1.12"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.13","level":"2.1","list_caption":"Figure: Grafana界面示意图2","alt":"Grafana界面示意图2","nro":13,"url":"../images/kubernetes-devops-example-grafana-2.png","index":13,"caption_template":"图片 - _CAPTION_","label":"Grafana界面示意图2","attributes":{},"skip":false,"key":"2.1.13"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.14","level":"2.1","list_caption":"Figure: Grafana界面示意图3","alt":"Grafana界面示意图3","nro":14,"url":"../images/kubernetes-devops-example-grafana-3.png","index":14,"caption_template":"图片 - _CAPTION_","label":"Grafana界面示意图3","attributes":{},"skip":false,"key":"2.1.14"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.15","level":"2.1","list_caption":"Figure: dashboard","alt":"dashboard","nro":15,"url":"../images/spark-job-on-kubernetes-example-1.jpg","index":15,"caption_template":"图片 - _CAPTION_","label":"dashboard","attributes":{},"skip":false,"key":"2.1.15"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.16","level":"2.1","list_caption":"Figure: Grafana","alt":"Grafana","nro":16,"url":"../images/spark-job-on-kubernetes-example-2.jpg","index":16,"caption_template":"图片 - _CAPTION_","label":"Grafana","attributes":{},"skip":false,"key":"2.1.16"},{"backlink":"concepts/index.html#fig3.1.1","level":"3.1","list_caption":"Figure: Borg架构","alt":"Borg架构","nro":17,"url":"../images/borg.png","index":1,"caption_template":"图片 - _CAPTION_","label":"Borg架构","attributes":{},"skip":false,"key":"3.1.1"},{"backlink":"concepts/index.html#fig3.1.2","level":"3.1","list_caption":"Figure: Kubernetes架构","alt":"Kubernetes架构","nro":18,"url":"../images/architecture.png","index":2,"caption_template":"图片 - _CAPTION_","label":"Kubernetes架构","attributes":{},"skip":false,"key":"3.1.2"},{"backlink":"concepts/index.html#fig3.1.3","level":"3.1","list_caption":"Figure: Kuberentes架构（图片来自于网络）","alt":"Kuberentes架构（图片来自于网络）","nro":19,"url":"../images/kubernetes-high-level-component-archtecture.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"Kuberentes架构（图片来自于网络）","attributes":{},"skip":false,"key":"3.1.3"},{"backlink":"concepts/index.html#fig3.1.4","level":"3.1","list_caption":"Figure: kubernetes整体架构示意图","alt":"kubernetes整体架构示意图","nro":20,"url":"../images/kubernetes-whole-arch.png","index":4,"caption_template":"图片 - _CAPTION_","label":"kubernetes整体架构示意图","attributes":{},"skip":false,"key":"3.1.4"},{"backlink":"concepts/index.html#fig3.1.5","level":"3.1","list_caption":"Figure: Kubernetes master架构示意图","alt":"Kubernetes master架构示意图","nro":21,"url":"../images/kubernetes-master-arch.png","index":5,"caption_template":"图片 - _CAPTION_","label":"Kubernetes master架构示意图","attributes":{},"skip":false,"key":"3.1.5"},{"backlink":"concepts/index.html#fig3.1.6","level":"3.1","list_caption":"Figure: kubernetes node架构示意图","alt":"kubernetes node架构示意图","nro":22,"url":"../images/kubernetes-node-arch.png","index":6,"caption_template":"图片 - _CAPTION_","label":"kubernetes node架构示意图","attributes":{},"skip":false,"key":"3.1.6"},{"backlink":"concepts/index.html#fig3.1.7","level":"3.1","list_caption":"Figure: Kubernetes分层架构示意图","alt":"Kubernetes分层架构示意图","nro":23,"url":"../images/kubernetes-layers-arch.jpg","index":7,"caption_template":"图片 - _CAPTION_","label":"Kubernetes分层架构示意图","attributes":{},"skip":false,"key":"3.1.7"},{"backlink":"concepts/concepts.html#fig3.2.1","level":"3.2","list_caption":"Figure: 分层架构示意图","alt":"分层架构示意图","nro":24,"url":"../images/kubernetes-layers-arch.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"分层架构示意图","attributes":{},"skip":false,"key":"3.2.1"},{"backlink":"concepts/pod-overview.html#fig3.3.1.1","level":"3.3.1","list_caption":"Figure: pod diagram","alt":"pod diagram","nro":25,"url":"../images/pod-overview.png","index":1,"caption_template":"图片 - _CAPTION_","label":"pod diagram","attributes":{},"skip":false,"key":"3.3.1.1"},{"backlink":"concepts/pod.html#fig3.3.1.1.1","level":"3.3.1.1","list_caption":"Figure: Pod示意图","alt":"Pod示意图","nro":26,"url":"../images/pod-overview.png","index":1,"caption_template":"图片 - _CAPTION_","label":"Pod示意图","attributes":{},"skip":false,"key":"3.3.1.1.1"},{"backlink":"concepts/pod.html#fig3.3.1.1.2","level":"3.3.1.1","list_caption":"Figure: Pod Cheatsheet","alt":"Pod Cheatsheet","nro":27,"url":"../images/kubernetes-pod-cheatsheet.png","index":2,"caption_template":"图片 - _CAPTION_","label":"Pod Cheatsheet","attributes":{},"skip":false,"key":"3.3.1.1.2"},{"backlink":"concepts/service.html#fig3.3.4.1","level":"3.3.4","list_caption":"Figure: userspace代理模式下Service概览图","alt":"userspace代理模式下Service概览图","nro":28,"url":"../images/services-userspace-overview.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"userspace代理模式下Service概览图","attributes":{},"skip":false,"key":"3.3.4.1"},{"backlink":"concepts/service.html#fig3.3.4.2","level":"3.3.4","list_caption":"Figure: iptables代理模式下Service概览图","alt":"iptables代理模式下Service概览图","nro":29,"url":"../images/services-iptables-overview.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"iptables代理模式下Service概览图","attributes":{},"skip":false,"key":"3.3.4.2"},{"backlink":"concepts/deployment.html#fig3.3.6.1","level":"3.3.6","list_caption":"Figure: kubernetes deployment cheatsheet","alt":"kubernetes deployment cheatsheet","nro":30,"url":"../images/deployment-cheatsheet.png","index":1,"caption_template":"图片 - _CAPTION_","label":"kubernetes deployment cheatsheet","attributes":{},"skip":false,"key":"3.3.6.1"},{"backlink":"concepts/horizontal-pod-autoscaling.html#fig3.3.16.1","level":"3.3.16","list_caption":"Figure: horizontal-pod-autoscaler","alt":"horizontal-pod-autoscaler","nro":31,"url":"../images/horizontal-pod-autoscaler.png","index":1,"caption_template":"图片 - _CAPTION_","label":"horizontal-pod-autoscaler","attributes":{},"skip":false,"key":"3.3.16.1"},{"backlink":"concepts/label.html#fig3.3.17.1","level":"3.3.17","list_caption":"Figure: label示意图","alt":"label示意图","nro":32,"url":"../images/labels.png","index":1,"caption_template":"图片 - _CAPTION_","label":"label示意图","attributes":{},"skip":false,"key":"3.3.17.1"},{"backlink":"guide/using-kubectl.html#fig4.3.2.1","level":"4.3.2","list_caption":"Figure: kubectl cheatsheet","alt":"kubectl cheatsheet","nro":33,"url":"../images/kubernetes-kubectl-cheatsheet.png","index":1,"caption_template":"图片 - _CAPTION_","label":"kubectl cheatsheet","attributes":{},"skip":false,"key":"4.3.2.1"},{"backlink":"guide/using-kubectl.html#fig4.3.2.2","level":"4.3.2","list_caption":"Figure: kube-shell页面","alt":"kube-shell页面","nro":34,"url":"../images/kube-shell.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"kube-shell页面","attributes":{},"skip":false,"key":"4.3.2.2"},{"backlink":"guide/ip-masq-agent.html#fig4.4.6.1","level":"4.4.6","list_caption":"Figure: IP伪装代理示意图","alt":"IP伪装代理示意图","nro":35,"url":"../images/ip-masq.png","index":1,"caption_template":"图片 - _CAPTION_","label":"IP伪装代理示意图","attributes":{},"skip":false,"key":"4.4.6.1"},{"backlink":"guide/auth-with-kubeconfig-or-token.html#fig4.4.7.1","level":"4.4.7","list_caption":"Figure: kubeconfig文件","alt":"kubeconfig文件","nro":36,"url":"../images/brand-kubeconfig-yaml.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"kubeconfig文件","attributes":{},"skip":false,"key":"4.4.7.1"},{"backlink":"guide/authentication.html#fig4.4.8.1","level":"4.4.8","list_caption":"Figure: Kubernetes OpenID Connect Flow","alt":"Kubernetes OpenID Connect Flow","nro":37,"url":"../images/kubernetes-oidc-login.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"Kubernetes OpenID Connect Flow","attributes":{},"skip":false,"key":"4.4.8.1"},{"backlink":"guide/carbin-mobile-dashboard-for-kubernetes.html#fig4.5.6.1","level":"4.5.6","list_caption":"Figure: App Store","alt":"App Store","nro":38,"url":"../images/carbin-kubernetes-mobile-dashboard-1.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"App Store","attributes":{},"skip":false,"key":"4.5.6.1"},{"backlink":"guide/carbin-mobile-dashboard-for-kubernetes.html#fig4.5.6.2","level":"4.5.6","list_caption":"Figure: 在手机上操作Kubernetes集群","alt":"在手机上操作Kubernetes集群","nro":39,"url":"../images/carbin-kubernetes-mobile-dashboard-4.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"在手机上操作Kubernetes集群","attributes":{},"skip":false,"key":"4.5.6.2"},{"backlink":"guide/kubernetes-desktop-client.html#fig4.5.7.1","level":"4.5.7","list_caption":"Figure: Kubernetic客户端","alt":"Kubernetic客户端","nro":40,"url":"../images/kubernetic-desktop-ui.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"Kubernetic客户端","attributes":{},"skip":false,"key":"4.5.7.1"},{"backlink":"guide/deploy-applications-in-kubernetes.html#fig4.6.1.1","level":"4.6.1","list_caption":"Figure: API","alt":"API","nro":41,"url":"../images/k8s-app-monitor-test-api-doc.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"API","attributes":{},"skip":false,"key":"4.6.1.1"},{"backlink":"guide/deploy-applications-in-kubernetes.html#fig4.6.1.2","level":"4.6.1","list_caption":"Figure: wercker","alt":"wercker","nro":42,"url":"../images/k8s-app-monitor-agent-wercker.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"wercker","attributes":{},"skip":false,"key":"4.6.1.2"},{"backlink":"guide/deploy-applications-in-kubernetes.html#fig4.6.1.3","level":"4.6.1","list_caption":"Figure: 图表","alt":"图表","nro":43,"url":"../images/k8s-app-monitor-agent.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"图表","attributes":{},"skip":false,"key":"4.6.1.3"},{"backlink":"guide/migrating-hadoop-yarn-to-kubernetes.html#fig4.6.2.1","level":"4.6.2","list_caption":"Figure: spark on yarn with kubernetes","alt":"spark on yarn with kubernetes","nro":44,"url":"../images/spark-on-yarn-with-kubernetes.png","index":1,"caption_template":"图片 - _CAPTION_","label":"spark on yarn with kubernetes","attributes":{},"skip":false,"key":"4.6.2.1"},{"backlink":"guide/migrating-hadoop-yarn-to-kubernetes.html#fig4.6.2.2","level":"4.6.2","list_caption":"Figure: Terms","alt":"Terms","nro":45,"url":"../images/terms-in-kubernetes-app-deployment.png","index":2,"caption_template":"图片 - _CAPTION_","label":"Terms","attributes":{},"skip":false,"key":"4.6.2.2"},{"backlink":"guide/migrating-hadoop-yarn-to-kubernetes.html#fig4.6.2.3","level":"4.6.2","list_caption":"Figure: 分解步骤解析","alt":"分解步骤解析","nro":46,"url":"../images/migrating-hadoop-yarn-to-kubernetes.png","index":3,"caption_template":"图片 - _CAPTION_","label":"分解步骤解析","attributes":{},"skip":false,"key":"4.6.2.3"},{"backlink":"practice/node-installation.html#fig5.2.7.1","level":"5.2.7","list_caption":"Figure: welcome nginx","alt":"welcome nginx","nro":47,"url":"../images/kubernetes-installation-test-nginx.png","index":1,"caption_template":"图片 - _CAPTION_","label":"welcome nginx","attributes":{},"skip":false,"key":"5.2.7.1"},{"backlink":"practice/dashboard-addon-installation.html#fig5.2.9.1","level":"5.2.9","list_caption":"Figure: kubernetes dashboard","alt":"kubernetes dashboard","nro":48,"url":"../images/kubernetes-dashboard-raw.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"kubernetes dashboard","attributes":{},"skip":false,"key":"5.2.9.1"},{"backlink":"practice/dashboard-addon-installation.html#fig5.2.9.2","level":"5.2.9","list_caption":"Figure: V1.6.3版本的dashboard界面","alt":"V1.6.3版本的dashboard界面","nro":49,"url":"../images/dashboard-v163.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"V1.6.3版本的dashboard界面","attributes":{},"skip":false,"key":"5.2.9.2"},{"backlink":"practice/heapster-addon-installation.html#fig5.2.10.1","level":"5.2.10","list_caption":"Figure: dashboard-heapster","alt":"dashboard-heapster","nro":50,"url":"../images/kubernetes-dashboard-with-heapster.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"dashboard-heapster","attributes":{},"skip":false,"key":"5.2.10.1"},{"backlink":"practice/heapster-addon-installation.html#fig5.2.10.2","level":"5.2.10","list_caption":"Figure: grafana","alt":"grafana","nro":51,"url":"../images/kubernetes-heapster-grafana.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"grafana","attributes":{},"skip":false,"key":"5.2.10.2"},{"backlink":"practice/heapster-addon-installation.html#fig5.2.10.3","level":"5.2.10","list_caption":"Figure: kubernetes-influxdb-heapster","alt":"kubernetes-influxdb-heapster","nro":52,"url":"../images/kubernetes-influxdb-heapster.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"kubernetes-influxdb-heapster","attributes":{},"skip":false,"key":"5.2.10.3"},{"backlink":"practice/heapster-addon-installation.html#fig5.2.10.4","level":"5.2.10","list_caption":"Figure: 修改grafana模板","alt":"修改grafana模板","nro":53,"url":"../images/grafana-dashboard-setting.jpg","index":4,"caption_template":"图片 - _CAPTION_","label":"修改grafana模板","attributes":{},"skip":false,"key":"5.2.10.4"},{"backlink":"practice/efk-addon-installation.html#fig5.2.11.1","level":"5.2.11","list_caption":"Figure: es-setting","alt":"es-setting","nro":54,"url":"../images/es-setting.png","index":1,"caption_template":"图片 - _CAPTION_","label":"es-setting","attributes":{},"skip":false,"key":"5.2.11.1"},{"backlink":"practice/efk-addon-installation.html#fig5.2.11.2","level":"5.2.11","list_caption":"Figure: es-home","alt":"es-home","nro":55,"url":"../images/kubernetes-efk-kibana.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"es-home","attributes":{},"skip":false,"key":"5.2.11.2"},{"backlink":"practice/traefik-ingress-installation.html#fig5.3.1.1","level":"5.3.1","list_caption":"Figure: kubernetes-dashboard","alt":"kubernetes-dashboard","nro":56,"url":"../images/traefik-dashboard.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"kubernetes-dashboard","attributes":{},"skip":false,"key":"5.3.1.1"},{"backlink":"practice/traefik-ingress-installation.html#fig5.3.1.2","level":"5.3.1","list_caption":"Figure: traefik-nginx","alt":"traefik-nginx","nro":57,"url":"../images/traefik-nginx.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"traefik-nginx","attributes":{},"skip":false,"key":"5.3.1.2"},{"backlink":"practice/traefik-ingress-installation.html#fig5.3.1.3","level":"5.3.1","list_caption":"Figure: traefik-guestbook","alt":"traefik-guestbook","nro":58,"url":"../images/traefik-guestbook.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"traefik-guestbook","attributes":{},"skip":false,"key":"5.3.1.3"},{"backlink":"practice/distributed-load-test.html#fig5.3.2.1","level":"5.3.2","list_caption":"Figure: 使用dashboard来扩容","alt":"使用dashboard来扩容","nro":59,"url":"../images/dashbaord-scale.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"使用dashboard来扩容","attributes":{},"skip":false,"key":"5.3.2.1"},{"backlink":"practice/distributed-load-test.html#fig5.3.2.2","level":"5.3.2","list_caption":"Figure: Traefik的UI","alt":"Traefik的UI","nro":60,"url":"../images/traefik-dashboard-locust.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"Traefik的UI","attributes":{},"skip":false,"key":"5.3.2.2"},{"backlink":"practice/distributed-load-test.html#fig5.3.2.3","level":"5.3.2","list_caption":"Figure: Locust启动界面","alt":"Locust启动界面","nro":61,"url":"../images/locust-start-swarming.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"Locust启动界面","attributes":{},"skip":false,"key":"5.3.2.3"},{"backlink":"practice/distributed-load-test.html#fig5.3.2.4","level":"5.3.2","list_caption":"Figure: Dashboard查看页面","alt":"Dashboard查看页面","nro":62,"url":"../images/sample-webapp-rc.jpg","index":4,"caption_template":"图片 - _CAPTION_","label":"Dashboard查看页面","attributes":{},"skip":false,"key":"5.3.2.4"},{"backlink":"practice/distributed-load-test.html#fig5.3.2.5","level":"5.3.2","list_caption":"Figure: Locust测试结果页面","alt":"Locust测试结果页面","nro":63,"url":"../images/locust-dashboard.jpg","index":5,"caption_template":"图片 - _CAPTION_","label":"Locust测试结果页面","attributes":{},"skip":false,"key":"5.3.2.5"},{"backlink":"practice/network-and-cluster-perfermance-test.html#fig5.3.3.1","level":"5.3.3","list_caption":"Figure: kubernetes-dashboard","alt":"kubernetes-dashboard","nro":64,"url":"../images/kubenetes-e2e-test.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"kubernetes-dashboard","attributes":{},"skip":false,"key":"5.3.3.1"},{"backlink":"practice/network-and-cluster-perfermance-test.html#fig5.3.3.2","level":"5.3.3","list_caption":"Figure: locust测试页面","alt":"locust测试页面","nro":65,"url":"../images/kubernetes-locust-test.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"locust测试页面","attributes":{},"skip":false,"key":"5.3.3.2"},{"backlink":"practice/edge-node-configuration.html#fig5.3.4.1","level":"5.3.4","list_caption":"Figure: 边缘节点架构","alt":"边缘节点架构","nro":66,"url":"../images/kubernetes-edge-node-architecture.png","index":1,"caption_template":"图片 - _CAPTION_","label":"边缘节点架构","attributes":{},"skip":false,"key":"5.3.4.1"},{"backlink":"practice/app-log-collection.html#fig5.4.2.1","level":"5.4.2","list_caption":"Figure: filebeat日志收集架构图","alt":"filebeat日志收集架构图","nro":67,"url":"../images/filebeat-log-collector.png","index":1,"caption_template":"图片 - _CAPTION_","label":"filebeat日志收集架构图","attributes":{},"skip":false,"key":"5.4.2.1"},{"backlink":"practice/app-log-collection.html#fig5.4.2.2","level":"5.4.2","list_caption":"Figure: Kibana页面","alt":"Kibana页面","nro":68,"url":"../images/filebeat-docker-test.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"Kibana页面","attributes":{},"skip":false,"key":"5.4.2.2"},{"backlink":"practice/app-log-collection.html#fig5.4.2.3","level":"5.4.2","list_caption":"Figure: filebeat收集的日志详细信息","alt":"filebeat收集的日志详细信息","nro":69,"url":"../images/kubernetes-filebeat-detail.png","index":3,"caption_template":"图片 - _CAPTION_","label":"filebeat收集的日志详细信息","attributes":{},"skip":false,"key":"5.4.2.3"},{"backlink":"practice/monitor.html#fig5.4.4.1","level":"5.4.4","list_caption":"Figure: Kubernetes集群中的监控","alt":"Kubernetes集群中的监控","nro":70,"url":"../images/monitoring-in-kubernetes.png","index":1,"caption_template":"图片 - _CAPTION_","label":"Kubernetes集群中的监控","attributes":{},"skip":false,"key":"5.4.4.1"},{"backlink":"practice/monitor.html#fig5.4.4.2","level":"5.4.4","list_caption":"Figure: kubernetes的容器命名规则示意图","alt":"kubernetes的容器命名规则示意图","nro":71,"url":"../images/kubernetes-container-naming-rule.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"kubernetes的容器命名规则示意图","attributes":{},"skip":false,"key":"5.4.4.2"},{"backlink":"practice/monitor.html#fig5.4.4.3","level":"5.4.4","list_caption":"Figure: Heapster架构图（改进版）","alt":"Heapster架构图（改进版）","nro":72,"url":"../images/kubernetes-heapster-monitoring.png","index":3,"caption_template":"图片 - _CAPTION_","label":"Heapster架构图（改进版）","attributes":{},"skip":false,"key":"5.4.4.3"},{"backlink":"practice/monitor.html#fig5.4.4.4","level":"5.4.4","list_caption":"Figure: 应用监控架构图","alt":"应用监控架构图","nro":73,"url":"../images/kubernetes-app-monitoring.png","index":4,"caption_template":"图片 - _CAPTION_","label":"应用监控架构图","attributes":{},"skip":false,"key":"5.4.4.4"},{"backlink":"practice/monitor.html#fig5.4.4.5","level":"5.4.4","list_caption":"Figure: 应用拓扑图","alt":"应用拓扑图","nro":74,"url":"../images/weave-scope-service-topology.jpg","index":5,"caption_template":"图片 - _CAPTION_","label":"应用拓扑图","attributes":{},"skip":false,"key":"5.4.4.5"},{"backlink":"practice/data-persistence-problem.html#fig5.4.5.1","level":"5.4.5","list_caption":"Figure: 日志持久化收集解决方案示意图","alt":"日志持久化收集解决方案示意图","nro":75,"url":"../images/log-persistence-logstash.png","index":1,"caption_template":"图片 - _CAPTION_","label":"日志持久化收集解决方案示意图","attributes":{},"skip":false,"key":"5.4.5.1"},{"backlink":"practice/storage-for-containers-using-glusterfs-with-openshift.html#fig5.5.1.2.1","level":"5.5.1.2","list_caption":"Figure: 创建存储","alt":"创建存储","nro":76,"url":"../images/create-gluster-storage.png","index":1,"caption_template":"图片 - _CAPTION_","label":"创建存储","attributes":{},"skip":false,"key":"5.5.1.2.1"},{"backlink":"practice/storage-for-containers-using-glusterfs-with-openshift.html#fig5.5.1.2.2","level":"5.5.1.2","list_caption":"Figure: Screen Shot 2017-03-24 at 11.09.34.png","alt":"Screen Shot 2017-03-24 at 11.09.34.png","nro":77,"url":"https://keithtenzer.files.wordpress.com/2017/03/screen-shot-2017-03-24-at-11-09-341.png?w=440","index":2,"caption_template":"图片 - _CAPTION_","label":"Screen Shot 2017-03-24 at 11.09.34.png","attributes":{},"skip":false,"key":"5.5.1.2.2"},{"backlink":"practice/using-heapster-to-get-object-metrics.html#fig5.6.1.1.1","level":"5.6.1.1","list_caption":"Figure: Heapster架构图","alt":"Heapster架构图","nro":78,"url":"../images/heapster-architecture.png","index":1,"caption_template":"图片 - _CAPTION_","label":"Heapster架构图","attributes":{},"skip":false,"key":"5.6.1.1.1"},{"backlink":"practice/prometheus.html#fig5.6.2.1","level":"5.6.2","list_caption":"Figure: Prometheus 架构图","alt":"Prometheus 架构图","nro":79,"url":"../images/prometheus-architecture.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"Prometheus 架构图","attributes":{},"skip":false,"key":"5.6.2.1"},{"backlink":"practice/using-prometheus-to-monitor-kuberentes-cluster.html#fig5.6.2.1.1","level":"5.6.2.1","list_caption":"Figure: Grafana页面","alt":"Grafana页面","nro":80,"url":"../images/kubernetes-prometheus-monitoring.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"Grafana页面","attributes":{},"skip":false,"key":"5.6.2.1.1"},{"backlink":"practice/helm.html#fig5.7.1.1","level":"5.7.1","list_caption":"Figure: Helm chart源","alt":"Helm chart源","nro":81,"url":"../images/helm-charts-repository.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"Helm chart源","attributes":{},"skip":false,"key":"5.7.1.1"},{"backlink":"practice/helm.html#fig5.7.1.2","level":"5.7.1","list_caption":"Figure: TODO应用的Web页面","alt":"TODO应用的Web页面","nro":82,"url":"../images/helm-mean-todo-aholic.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"TODO应用的Web页面","attributes":{},"skip":false,"key":"5.7.1.2"},{"backlink":"practice/create-private-charts-repo.html#fig5.7.2.1","level":"5.7.2","list_caption":"Figure: Helm monocular界面","alt":"Helm monocular界面","nro":83,"url":"../images/helm-monocular-jimmysong.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"Helm monocular界面","attributes":{},"skip":false,"key":"5.7.2.1"},{"backlink":"practice/jenkins-ci-cd.html#fig5.8.1.1","level":"5.8.1","list_caption":"Figure: 基于Jenkins的持续集成与发布","alt":"基于Jenkins的持续集成与发布","nro":84,"url":"../images/kubernetes-jenkins-ci-cd.png","index":1,"caption_template":"图片 - _CAPTION_","label":"基于Jenkins的持续集成与发布","attributes":{},"skip":false,"key":"5.8.1.1"},{"backlink":"practice/drone-ci-cd.html#fig5.8.2.1","level":"5.8.2","list_caption":"Figure: OAuth注册","alt":"OAuth注册","nro":85,"url":"../images/github-oauth-register.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"OAuth注册","attributes":{},"skip":false,"key":"5.8.2.1"},{"backlink":"practice/drone-ci-cd.html#fig5.8.2.2","level":"5.8.2","list_caption":"Figure: OAuth key","alt":"OAuth key","nro":86,"url":"../images/github-oauth-drone-key.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"OAuth key","attributes":{},"skip":false,"key":"5.8.2.2"},{"backlink":"practice/drone-ci-cd.html#fig5.8.2.3","level":"5.8.2","list_caption":"Figure: Drone登陆界面","alt":"Drone登陆界面","nro":87,"url":"../images/drone-login-github.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"Drone登陆界面","attributes":{},"skip":false,"key":"5.8.2.3"},{"backlink":"practice/drone-ci-cd.html#fig5.8.2.4","level":"5.8.2","list_caption":"Figure: Github启用repo设置","alt":"Github启用repo设置","nro":88,"url":"../images/drone-github-active.jpg","index":4,"caption_template":"图片 - _CAPTION_","label":"Github启用repo设置","attributes":{},"skip":false,"key":"5.8.2.4"},{"backlink":"practice/drone-ci-cd.html#fig5.8.2.5","level":"5.8.2","list_caption":"Figure: Github单个repo设置","alt":"Github单个repo设置","nro":89,"url":"../images/drone-github-repo-setting.jpg","index":5,"caption_template":"图片 - _CAPTION_","label":"Github单个repo设置","attributes":{},"skip":false,"key":"5.8.2.5"},{"backlink":"practice/dashboard-upgrade.html#fig5.9.2.1","level":"5.9.2","list_caption":"Figure: 登陆界面","alt":"登陆界面","nro":90,"url":"../images/kubernetes-dashboard-1.7.1-login.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"登陆界面","attributes":{},"skip":false,"key":"5.9.2.1"},{"backlink":"practice/dashboard-upgrade.html#fig5.9.2.2","level":"5.9.2","list_caption":"Figure: 首页","alt":"首页","nro":91,"url":"../images/kubernetes-dashboard-1.7.1-default-page.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"首页","attributes":{},"skip":false,"key":"5.9.2.2"},{"backlink":"practice/dashboard-upgrade.html#fig5.9.2.3","level":"5.9.2","list_caption":"Figure: 用户空间","alt":"用户空间","nro":92,"url":"../images/kubernetes-dashboard-1.7.1-brand.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"用户空间","attributes":{},"skip":false,"key":"5.9.2.3"},{"backlink":"practice/dashboard-upgrade.html#fig5.9.2.4","level":"5.9.2","list_caption":"Figure: kubeconfig文件","alt":"kubeconfig文件","nro":93,"url":"../images/brand-kubeconfig-yaml.jpg","index":4,"caption_template":"图片 - _CAPTION_","label":"kubeconfig文件","attributes":{},"skip":false,"key":"5.9.2.4"},{"backlink":"usecases/microservices.html#fig6.2.1","level":"6.2","list_caption":"Figure: 微服务关注的部分","alt":"微服务关注的部分","nro":94,"url":"../images/microservices-concerns.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"微服务关注的部分","attributes":{},"skip":false,"key":"6.2.1"},{"backlink":"usecases/service-discovery-in-microservices.html#fig6.2.1.1","level":"6.2.1","list_caption":"Figure: 微服务中的服务发现","alt":"微服务中的服务发现","nro":95,"url":"../images/service-discovery-in-microservices.png","index":1,"caption_template":"图片 - _CAPTION_","label":"微服务中的服务发现","attributes":{},"skip":false,"key":"6.2.1.1"},{"backlink":"usecases/service-mesh.html#fig6.3.1","level":"6.3","list_caption":"Figure: 下一代异构微服务架构","alt":"下一代异构微服务架构","nro":96,"url":"../images/polyglot-microservices-serivce-mesh.png","index":1,"caption_template":"图片 - _CAPTION_","label":"下一代异构微服务架构","attributes":{},"skip":false,"key":"6.3.1"},{"backlink":"usecases/service-mesh.html#fig6.3.2","level":"6.3","list_caption":"Figure: Service Mesh 架构图","alt":"Service Mesh 架构图","nro":97,"url":"../images/serivce-mesh-control-plane.png","index":2,"caption_template":"图片 - _CAPTION_","label":"Service Mesh 架构图","attributes":{},"skip":false,"key":"6.3.2"},{"backlink":"usecases/service-mesh.html#fig6.3.3","level":"6.3","list_caption":"Figure: Istio vs linkerd","alt":"Istio vs linkerd","nro":98,"url":"../images/istio-vs-linkerd.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"Istio vs linkerd","attributes":{},"skip":false,"key":"6.3.3"},{"backlink":"usecases/istio.html#fig6.3.1.1","level":"6.3.1","list_caption":"Figure: Istio架构图","alt":"Istio架构图","nro":99,"url":"../images/istio-arch.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"Istio架构图","attributes":{},"skip":false,"key":"6.3.1.1"},{"backlink":"usecases/istio-installation.html#fig6.3.1.1.1","level":"6.3.1.1","list_caption":"Figure: BookInfo Sample应用架构图","alt":"BookInfo Sample应用架构图","nro":100,"url":"../images/bookinfo-sample-arch.png","index":1,"caption_template":"图片 - _CAPTION_","label":"BookInfo Sample应用架构图","attributes":{},"skip":false,"key":"6.3.1.1.1"},{"backlink":"usecases/istio-installation.html#fig6.3.1.1.2","level":"6.3.1.1","list_caption":"Figure: BookInfo Sample页面","alt":"BookInfo Sample页面","nro":101,"url":"../images/bookinfo-sample.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"BookInfo Sample页面","attributes":{},"skip":false,"key":"6.3.1.1.2"},{"backlink":"usecases/istio-installation.html#fig6.3.1.1.3","level":"6.3.1.1","list_caption":"Figure: Istio Grafana界面","alt":"Istio Grafana界面","nro":102,"url":"../images/istio-grafana.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"Istio Grafana界面","attributes":{},"skip":false,"key":"6.3.1.1.3"},{"backlink":"usecases/istio-installation.html#fig6.3.1.1.4","level":"6.3.1.1","list_caption":"Figure: Prometheus页面","alt":"Prometheus页面","nro":103,"url":"../images/istio-prometheus.jpg","index":4,"caption_template":"图片 - _CAPTION_","label":"Prometheus页面","attributes":{},"skip":false,"key":"6.3.1.1.4"},{"backlink":"usecases/istio-installation.html#fig6.3.1.1.5","level":"6.3.1.1","list_caption":"Figure: Zipkin页面","alt":"Zipkin页面","nro":104,"url":"../images/istio-zipkin.jpg","index":5,"caption_template":"图片 - _CAPTION_","label":"Zipkin页面","attributes":{},"skip":false,"key":"6.3.1.1.5"},{"backlink":"usecases/istio-installation.html#fig6.3.1.1.6","level":"6.3.1.1","list_caption":"Figure: ServiceGraph页面","alt":"ServiceGraph页面","nro":105,"url":"../images/istio-servicegraph.jpg","index":6,"caption_template":"图片 - _CAPTION_","label":"ServiceGraph页面","attributes":{},"skip":false,"key":"6.3.1.1.6"},{"backlink":"usecases/install-and-expand-istio-mesh.html#fig6.3.1.3.1","level":"6.3.1.3","list_caption":"Figure: BookInfo","alt":"BookInfo","nro":106,"url":"../images/noistio.png","index":1,"caption_template":"图片 - _CAPTION_","label":"BookInfo","attributes":{},"skip":false,"key":"6.3.1.3.1"},{"backlink":"usecases/install-and-expand-istio-mesh.html#fig6.3.1.3.2","level":"6.3.1.3","list_caption":"Figure: BookInfo","alt":"BookInfo","nro":107,"url":"../images/noistio.png","index":2,"caption_template":"图片 - _CAPTION_","label":"BookInfo","attributes":{},"skip":false,"key":"6.3.1.3.2"},{"backlink":"usecases/integrating-vms.html#fig6.3.1.4.1","level":"6.3.1.4","list_caption":"Figure: Bookinfo应用的拓展Mesh","alt":"Bookinfo应用的拓展Mesh","nro":108,"url":"../images/istio-mesh-expansion.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"Bookinfo应用的拓展Mesh","attributes":{},"skip":false,"key":"6.3.1.4.1"},{"backlink":"usecases/linkerd.html#fig6.3.2.1","level":"6.3.2","list_caption":"Figure: source https://linkerd.io","alt":"source https://linkerd.io","nro":109,"url":"../images/diagram-individual-instance.png","index":1,"caption_template":"图片 - _CAPTION_","label":"source https://linkerd.io","attributes":{},"skip":false,"key":"6.3.2.1"},{"backlink":"usecases/linkerd-user-guide.html#fig6.3.2.1.1","level":"6.3.2.1","list_caption":"Figure: Jenkins pipeline","alt":"Jenkins pipeline","nro":110,"url":"../images/linkerd-jenkins-pipeline.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"Jenkins pipeline","attributes":{},"skip":false,"key":"6.3.2.1.1"},{"backlink":"usecases/linkerd-user-guide.html#fig6.3.2.1.2","level":"6.3.2.1","list_caption":"Figure: Jenkins config","alt":"Jenkins config","nro":111,"url":"../images/linkerd-jenkins.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"Jenkins config","attributes":{},"skip":false,"key":"6.3.2.1.2"},{"backlink":"usecases/linkerd-user-guide.html#fig6.3.2.1.3","level":"6.3.2.1","list_caption":"Figure: namerd","alt":"namerd","nro":112,"url":"../images/namerd-internal.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"namerd","attributes":{},"skip":false,"key":"6.3.2.1.3"},{"backlink":"usecases/linkerd-user-guide.html#fig6.3.2.1.4","level":"6.3.2.1","list_caption":"Figure: linkerd监控","alt":"linkerd监控","nro":113,"url":"../images/linkerd-helloworld-outgoing.jpg","index":4,"caption_template":"图片 - _CAPTION_","label":"linkerd监控","attributes":{},"skip":false,"key":"6.3.2.1.4"},{"backlink":"usecases/linkerd-user-guide.html#fig6.3.2.1.5","level":"6.3.2.1","list_caption":"Figure: linkerd监控","alt":"linkerd监控","nro":114,"url":"../images/linkerd-helloworld-incoming.jpg","index":5,"caption_template":"图片 - _CAPTION_","label":"linkerd监控","attributes":{},"skip":false,"key":"6.3.2.1.5"},{"backlink":"usecases/linkerd-user-guide.html#fig6.3.2.1.6","level":"6.3.2.1","list_caption":"Figure: linkerd性能监控","alt":"linkerd性能监控","nro":115,"url":"../images/linkerd-grafana.png","index":6,"caption_template":"图片 - _CAPTION_","label":"linkerd性能监控","attributes":{},"skip":false,"key":"6.3.2.1.6"},{"backlink":"usecases/linkerd-user-guide.html#fig6.3.2.1.7","level":"6.3.2.1","list_caption":"Figure: Linkerd ingress controller","alt":"Linkerd ingress controller","nro":116,"url":"../images/linkerd-ingress-controller.jpg","index":7,"caption_template":"图片 - _CAPTION_","label":"Linkerd ingress controller","attributes":{},"skip":false,"key":"6.3.2.1.7"},{"backlink":"usecases/conduit-installation.html#fig6.3.3.2.1","level":"6.3.3.2","list_caption":"Figure: Conduit dashboard","alt":"Conduit dashboard","nro":117,"url":"../images/conduit-dashboard.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"Conduit dashboard","attributes":{},"skip":false,"key":"6.3.3.2.1"},{"backlink":"usecases/big-data.html#fig6.4.1","level":"6.4","list_caption":"Figure: Spark on yarn with kubernetes","alt":"Spark on yarn with kubernetes","nro":118,"url":"../images/spark-on-yarn-with-kubernetes.png","index":1,"caption_template":"图片 - _CAPTION_","label":"Spark on yarn with kubernetes","attributes":{},"skip":false,"key":"6.4.1"},{"backlink":"usecases/big-data.html#fig6.4.2","level":"6.4","list_caption":"Figure: 在kubernetes上使用多种调度方式","alt":"在kubernetes上使用多种调度方式","nro":119,"url":"../images/spark-on-kubernetes-with-different-schedulers.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"在kubernetes上使用多种调度方式","attributes":{},"skip":false,"key":"6.4.2"},{"backlink":"usecases/spark-standalone-on-kubernetes.html#fig6.4.1.1","level":"6.4.1","list_caption":"Figure: spark master ui","alt":"spark master ui","nro":120,"url":"../images/spark-ui.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"spark master ui","attributes":{},"skip":false,"key":"6.4.1.1"},{"backlink":"usecases/spark-standalone-on-kubernetes.html#fig6.4.1.2","level":"6.4.1","list_caption":"Figure: zeppelin ui","alt":"zeppelin ui","nro":121,"url":"../images/zeppelin-ui.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"zeppelin ui","attributes":{},"skip":false,"key":"6.4.1.2"},{"backlink":"usecases/serverless.html#fig6.5.1","level":"6.5","list_caption":"Figure: 从物理机到函数计算","alt":"从物理机到函数计算","nro":122,"url":"../images/from-bare-metal-to-functions.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"从物理机到函数计算","attributes":{},"skip":false,"key":"6.5.1"},{"backlink":"usecases/serverless.html#fig6.5.2","level":"6.5","list_caption":"Figure: FaaS Landscape","alt":"FaaS Landscape","nro":123,"url":"../images/redpoint-faas-landscape.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"FaaS Landscape","attributes":{},"skip":false,"key":"6.5.2"},{"backlink":"usecases/understanding-serverless.html#fig6.5.1.1","level":"6.5.1","list_caption":"Figure: 服务端软件的运行环境","alt":"服务端软件的运行环境","nro":124,"url":"../images/serverless-server-side-software.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"服务端软件的运行环境","attributes":{},"skip":false,"key":"6.5.1.1"},{"backlink":"usecases/understanding-serverless.html#fig6.5.1.2","level":"6.5.1","list_caption":"Figure: FaaS应用架构","alt":"FaaS应用架构","nro":125,"url":"../images/serverless-faas-platform.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"FaaS应用架构","attributes":{},"skip":false,"key":"6.5.1.2"},{"backlink":"usecases/understanding-serverless.html#fig6.5.1.3","level":"6.5.1","list_caption":"Figure: 传统应用程序架构","alt":"传统应用程序架构","nro":126,"url":"../images/non-serverless-game-arch.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"传统应用程序架构","attributes":{},"skip":false,"key":"6.5.1.3"},{"backlink":"usecases/understanding-serverless.html#fig6.5.1.4","level":"6.5.1","list_caption":"Figure: Serverless架构","alt":"Serverless架构","nro":127,"url":"../images/serverless-game-arch.jpg","index":4,"caption_template":"图片 - _CAPTION_","label":"Serverless架构","attributes":{},"skip":false,"key":"6.5.1.4"},{"backlink":"usecases/openfaas.html#fig6.5.2.1.1","level":"6.5.2.1","list_caption":"Figure: OpenFaaS架构","alt":"OpenFaaS架构","nro":128,"url":"../images/openfaas-arch.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"OpenFaaS架构","attributes":{},"skip":false,"key":"6.5.2.1.1"},{"backlink":"usecases/openfaas.html#fig6.5.2.1.2","level":"6.5.2.1","list_caption":"Figure: OpenFaaS Prometheus","alt":"OpenFaaS Prometheus","nro":129,"url":"../images/openfaas-prometheus.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"OpenFaaS Prometheus","attributes":{},"skip":false,"key":"6.5.2.1.2"},{"backlink":"usecases/openfaas.html#fig6.5.2.1.3","level":"6.5.2.1","list_caption":"Figure: OpenFaas Grafana监控","alt":"OpenFaas Grafana监控","nro":130,"url":"../images/openfaas-grafana.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"OpenFaas Grafana监控","attributes":{},"skip":false,"key":"6.5.2.1.3"},{"backlink":"usecases/openfaas.html#fig6.5.2.1.4","level":"6.5.2.1","list_caption":"Figure: OpenFaas Dashboard","alt":"OpenFaas Dashboard","nro":131,"url":"../images/openfaas-deploy-a-function.jpg","index":4,"caption_template":"图片 - _CAPTION_","label":"OpenFaas Dashboard","attributes":{},"skip":false,"key":"6.5.2.1.4"},{"backlink":"usecases/openfaas.html#fig6.5.2.1.5","level":"6.5.2.1","list_caption":"Figure: NodeInfo执行结果","alt":"NodeInfo执行结果","nro":132,"url":"../images/openfaas-nodeinfo.jpg","index":5,"caption_template":"图片 - _CAPTION_","label":"NodeInfo执行结果","attributes":{},"skip":false,"key":"6.5.2.1.5"},{"backlink":"develop/client-go-sample.html#fig7.5.1","level":"7.5","list_caption":"Figure: 使用kubernetes dashboard进行故障排查","alt":"使用kubernetes dashboard进行故障排查","nro":133,"url":"../images/kubernetes-client-go-sample-update.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"使用kubernetes dashboard进行故障排查","attributes":{},"skip":false,"key":"7.5.1"},{"backlink":"appendix/issues.html#fig8.6.1","level":"8.6","list_caption":"Figure: pvc-storage-limit","alt":"pvc-storage-limit","nro":134,"url":"../images/pvc-storage-limit.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"pvc-storage-limit","attributes":{},"skip":false,"key":"8.6.1"}]},"title":"Kubernetes Handbook - jimmysong.io","language":"zh-hans","links":{"sidebar":{"Jimmy Song":"https://jimmysong.io"}},"gitbook":"*","description":"Kubernetes中文指南/实践手册|kubernetes handbook|jimmysong.io|宋净超"},"file":{"path":"usecases/install-and-expand-istio-mesh.md","mtime":"2017-11-21T13:18:30.606Z","type":"markdown"},"gitbook":{"version":"3.2.2","time":"2017-12-20T09:26:03.762Z"},"basePath":"..","book":{"language":""}}); + gitbook.page.hasChanged({"page":{"title":"安装和拓展Istio service mesh","level":"6.3.1.3","depth":3,"next":{"title":"集成虚拟机","level":"6.3.1.4","depth":3,"path":"usecases/integrating-vms.md","ref":"usecases/integrating-vms.md","articles":[]},"previous":{"title":"配置请求的路由规则","level":"6.3.1.2","depth":3,"path":"usecases/configuring-request-routing.md","ref":"usecases/configuring-request-routing.md","articles":[]},"dir":"ltr"},"config":{"plugins":["github","codesnippet","splitter","page-toc-button","image-captions","editlink","back-to-top-button","-lunr","-search","search-plus","github-buttons@2.1.0","favicon@^0.0.2","tbfed-pagefooter@^0.0.1","3-ba","theme-default"],"styles":{"website":"styles/website.css","pdf":"styles/pdf.css","epub":"styles/epub.css","mobi":"styles/mobi.css","ebook":"styles/ebook.css","print":"styles/print.css"},"pluginsConfig":{"tbfed-pagefooter":{"copyright":"Copyright © jimmysong.io 2017","modify_label":"Updated:","modify_format":"YYYY-MM-DD HH:mm:ss"},"github":{"url":"https://github.com/rootsongjc/kubernetes-handbook"},"editlink":{"label":"编辑本页","multilingual":false,"base":"https://github.com/rootsongjc/kubernetes-handbook/blob/master/"},"splitter":{},"codesnippet":{},"fontsettings":{"theme":"white","family":"sans","size":2},"highlight":{},"favicon":{"shortcut":"favicon.ico","bookmark":"favicon.ico"},"page-toc-button":{},"back-to-top-button":{},"github-buttons":{"repo":"rootsongjc/kubernetes-handbook","types":["star"],"size":"small"},"3-ba":{"configuration":"auto","token":"11f7d254cfa4e0ca44b175c66d379ecc"},"sharing":{"facebook":true,"twitter":true,"google":false,"weibo":false,"instapaper":false,"vk":false,"all":["facebook","google","twitter","weibo","instapaper"]},"theme-default":{"styles":{"website":"styles/website.css","pdf":"styles/pdf.css","epub":"styles/epub.css","mobi":"styles/mobi.css","ebook":"styles/ebook.css","print":"styles/print.css"},"showLevel":true},"search-plus":{},"image-captions":{"caption":"图片 - _CAPTION_","variable_name":"_pictures"}},"theme":"default","author":"Jimmy Song","pdf":{"pageNumbers":true,"fontSize":12,"fontFamily":"Arial","paperSize":"a4","chapterMark":"pagebreak","pageBreaksBefore":"/","margin":{"right":62,"left":62,"top":56,"bottom":56}},"structure":{"langs":"LANGS.md","readme":"README.md","glossary":"GLOSSARY.md","summary":"SUMMARY.md"},"variables":{"_pictures":[{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.1","level":"2.1","list_caption":"Figure: 云计算演进历程","alt":"云计算演进历程","nro":1,"url":"../images/cloud-computing-evolution-road.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"云计算演进历程","attributes":{},"skip":false,"key":"2.1.1"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.2","level":"2.1","list_caption":"Figure: 来自Twitter @MarcWilczek","alt":"来自Twitter @MarcWilczek","nro":2,"url":"../images/cloud-native-comes-of-age.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"来自Twitter @MarcWilczek","attributes":{},"skip":false,"key":"2.1.2"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.3","level":"2.1","list_caption":"Figure: Cloud native思维导图","alt":"Cloud native思维导图","nro":3,"url":"../images/cloud-native-architecutre-mindnode.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"Cloud native思维导图","attributes":{},"skip":false,"key":"2.1.3"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.4","level":"2.1","list_caption":"Figure: 十二因素应用","alt":"十二因素应用","nro":4,"url":"../images/12-factor-app.png","index":4,"caption_template":"图片 - _CAPTION_","label":"十二因素应用","attributes":{},"skip":false,"key":"2.1.4"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.5","level":"2.1","list_caption":"Figure: 容器生态","alt":"容器生态","nro":5,"url":"../images/container-ecosystem.png","index":5,"caption_template":"图片 - _CAPTION_","label":"容器生态","attributes":{},"skip":false,"key":"2.1.5"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.6","level":"2.1","list_caption":"Figure: 使用Jenkins进行持续集成与发布流程图","alt":"使用Jenkins进行持续集成与发布流程图","nro":6,"url":"../images/kubernetes-jenkins-ci-cd.png","index":6,"caption_template":"图片 - _CAPTION_","label":"使用Jenkins进行持续集成与发布流程图","attributes":{},"skip":false,"key":"2.1.6"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.7","level":"2.1","list_caption":"Figure: filebeat日志收集架构图","alt":"filebeat日志收集架构图","nro":7,"url":"../images/filebeat-log-collector-arch.png","index":7,"caption_template":"图片 - _CAPTION_","label":"filebeat日志收集架构图","attributes":{},"skip":false,"key":"2.1.7"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.8","level":"2.1","list_caption":"Figure: API文档","alt":"API文档","nro":8,"url":"../images/k8s-app-monitor-test-api-doc.jpg","index":8,"caption_template":"图片 - _CAPTION_","label":"API文档","attributes":{},"skip":false,"key":"2.1.8"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.9","level":"2.1","list_caption":"Figure: 迁移步骤示意图","alt":"迁移步骤示意图","nro":9,"url":"../images/migrating-hadoop-yarn-to-kubernetes.png","index":9,"caption_template":"图片 - _CAPTION_","label":"迁移步骤示意图","attributes":{},"skip":false,"key":"2.1.9"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.10","level":"2.1","list_caption":"Figure: service mesh架构图","alt":"service mesh架构图","nro":10,"url":"../images/serivce-mesh-control-plane.png","index":10,"caption_template":"图片 - _CAPTION_","label":"service mesh架构图","attributes":{},"skip":false,"key":"2.1.10"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.11","level":"2.1","list_caption":"Figure: kibana界面","alt":"kibana界面","nro":11,"url":"../images/filebeat-docker-test.jpg","index":11,"caption_template":"图片 - _CAPTION_","label":"kibana界面","attributes":{},"skip":false,"key":"2.1.11"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.12","level":"2.1","list_caption":"Figure: Grafana界面示意图1","alt":"Grafana界面示意图1","nro":12,"url":"../images/kubernetes-devops-example-grafana-1.png","index":12,"caption_template":"图片 - _CAPTION_","label":"Grafana界面示意图1","attributes":{},"skip":false,"key":"2.1.12"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.13","level":"2.1","list_caption":"Figure: Grafana界面示意图2","alt":"Grafana界面示意图2","nro":13,"url":"../images/kubernetes-devops-example-grafana-2.png","index":13,"caption_template":"图片 - _CAPTION_","label":"Grafana界面示意图2","attributes":{},"skip":false,"key":"2.1.13"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.14","level":"2.1","list_caption":"Figure: Grafana界面示意图3","alt":"Grafana界面示意图3","nro":14,"url":"../images/kubernetes-devops-example-grafana-3.png","index":14,"caption_template":"图片 - _CAPTION_","label":"Grafana界面示意图3","attributes":{},"skip":false,"key":"2.1.14"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.15","level":"2.1","list_caption":"Figure: dashboard","alt":"dashboard","nro":15,"url":"../images/spark-job-on-kubernetes-example-1.jpg","index":15,"caption_template":"图片 - _CAPTION_","label":"dashboard","attributes":{},"skip":false,"key":"2.1.15"},{"backlink":"cloud-native/kubernetes-and-cloud-native-app-overview.html#fig2.1.16","level":"2.1","list_caption":"Figure: Grafana","alt":"Grafana","nro":16,"url":"../images/spark-job-on-kubernetes-example-2.jpg","index":16,"caption_template":"图片 - _CAPTION_","label":"Grafana","attributes":{},"skip":false,"key":"2.1.16"},{"backlink":"concepts/index.html#fig3.1.1","level":"3.1","list_caption":"Figure: Borg架构","alt":"Borg架构","nro":17,"url":"../images/borg.png","index":1,"caption_template":"图片 - _CAPTION_","label":"Borg架构","attributes":{},"skip":false,"key":"3.1.1"},{"backlink":"concepts/index.html#fig3.1.2","level":"3.1","list_caption":"Figure: Kubernetes架构","alt":"Kubernetes架构","nro":18,"url":"../images/architecture.png","index":2,"caption_template":"图片 - _CAPTION_","label":"Kubernetes架构","attributes":{},"skip":false,"key":"3.1.2"},{"backlink":"concepts/index.html#fig3.1.3","level":"3.1","list_caption":"Figure: Kuberentes架构（图片来自于网络）","alt":"Kuberentes架构（图片来自于网络）","nro":19,"url":"../images/kubernetes-high-level-component-archtecture.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"Kuberentes架构（图片来自于网络）","attributes":{},"skip":false,"key":"3.1.3"},{"backlink":"concepts/index.html#fig3.1.4","level":"3.1","list_caption":"Figure: kubernetes整体架构示意图","alt":"kubernetes整体架构示意图","nro":20,"url":"../images/kubernetes-whole-arch.png","index":4,"caption_template":"图片 - _CAPTION_","label":"kubernetes整体架构示意图","attributes":{},"skip":false,"key":"3.1.4"},{"backlink":"concepts/index.html#fig3.1.5","level":"3.1","list_caption":"Figure: Kubernetes master架构示意图","alt":"Kubernetes master架构示意图","nro":21,"url":"../images/kubernetes-master-arch.png","index":5,"caption_template":"图片 - _CAPTION_","label":"Kubernetes master架构示意图","attributes":{},"skip":false,"key":"3.1.5"},{"backlink":"concepts/index.html#fig3.1.6","level":"3.1","list_caption":"Figure: kubernetes node架构示意图","alt":"kubernetes node架构示意图","nro":22,"url":"../images/kubernetes-node-arch.png","index":6,"caption_template":"图片 - _CAPTION_","label":"kubernetes node架构示意图","attributes":{},"skip":false,"key":"3.1.6"},{"backlink":"concepts/index.html#fig3.1.7","level":"3.1","list_caption":"Figure: Kubernetes分层架构示意图","alt":"Kubernetes分层架构示意图","nro":23,"url":"../images/kubernetes-layers-arch.jpg","index":7,"caption_template":"图片 - _CAPTION_","label":"Kubernetes分层架构示意图","attributes":{},"skip":false,"key":"3.1.7"},{"backlink":"concepts/concepts.html#fig3.2.1","level":"3.2","list_caption":"Figure: 分层架构示意图","alt":"分层架构示意图","nro":24,"url":"../images/kubernetes-layers-arch.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"分层架构示意图","attributes":{},"skip":false,"key":"3.2.1"},{"backlink":"concepts/pod-overview.html#fig3.3.1.1","level":"3.3.1","list_caption":"Figure: pod diagram","alt":"pod diagram","nro":25,"url":"../images/pod-overview.png","index":1,"caption_template":"图片 - _CAPTION_","label":"pod diagram","attributes":{},"skip":false,"key":"3.3.1.1"},{"backlink":"concepts/pod.html#fig3.3.1.1.1","level":"3.3.1.1","list_caption":"Figure: Pod示意图","alt":"Pod示意图","nro":26,"url":"../images/pod-overview.png","index":1,"caption_template":"图片 - _CAPTION_","label":"Pod示意图","attributes":{},"skip":false,"key":"3.3.1.1.1"},{"backlink":"concepts/pod.html#fig3.3.1.1.2","level":"3.3.1.1","list_caption":"Figure: Pod Cheatsheet","alt":"Pod Cheatsheet","nro":27,"url":"../images/kubernetes-pod-cheatsheet.png","index":2,"caption_template":"图片 - _CAPTION_","label":"Pod Cheatsheet","attributes":{},"skip":false,"key":"3.3.1.1.2"},{"backlink":"concepts/service.html#fig3.3.4.1","level":"3.3.4","list_caption":"Figure: userspace代理模式下Service概览图","alt":"userspace代理模式下Service概览图","nro":28,"url":"../images/services-userspace-overview.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"userspace代理模式下Service概览图","attributes":{},"skip":false,"key":"3.3.4.1"},{"backlink":"concepts/service.html#fig3.3.4.2","level":"3.3.4","list_caption":"Figure: iptables代理模式下Service概览图","alt":"iptables代理模式下Service概览图","nro":29,"url":"../images/services-iptables-overview.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"iptables代理模式下Service概览图","attributes":{},"skip":false,"key":"3.3.4.2"},{"backlink":"concepts/deployment.html#fig3.3.6.1","level":"3.3.6","list_caption":"Figure: kubernetes deployment cheatsheet","alt":"kubernetes deployment cheatsheet","nro":30,"url":"../images/deployment-cheatsheet.png","index":1,"caption_template":"图片 - _CAPTION_","label":"kubernetes deployment cheatsheet","attributes":{},"skip":false,"key":"3.3.6.1"},{"backlink":"concepts/horizontal-pod-autoscaling.html#fig3.3.16.1","level":"3.3.16","list_caption":"Figure: horizontal-pod-autoscaler","alt":"horizontal-pod-autoscaler","nro":31,"url":"../images/horizontal-pod-autoscaler.png","index":1,"caption_template":"图片 - _CAPTION_","label":"horizontal-pod-autoscaler","attributes":{},"skip":false,"key":"3.3.16.1"},{"backlink":"concepts/label.html#fig3.3.17.1","level":"3.3.17","list_caption":"Figure: label示意图","alt":"label示意图","nro":32,"url":"../images/labels.png","index":1,"caption_template":"图片 - _CAPTION_","label":"label示意图","attributes":{},"skip":false,"key":"3.3.17.1"},{"backlink":"guide/using-kubectl.html#fig4.3.2.1","level":"4.3.2","list_caption":"Figure: kubectl cheatsheet","alt":"kubectl cheatsheet","nro":33,"url":"../images/kubernetes-kubectl-cheatsheet.png","index":1,"caption_template":"图片 - _CAPTION_","label":"kubectl cheatsheet","attributes":{},"skip":false,"key":"4.3.2.1"},{"backlink":"guide/using-kubectl.html#fig4.3.2.2","level":"4.3.2","list_caption":"Figure: kube-shell页面","alt":"kube-shell页面","nro":34,"url":"../images/kube-shell.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"kube-shell页面","attributes":{},"skip":false,"key":"4.3.2.2"},{"backlink":"guide/ip-masq-agent.html#fig4.4.6.1","level":"4.4.6","list_caption":"Figure: IP伪装代理示意图","alt":"IP伪装代理示意图","nro":35,"url":"../images/ip-masq.png","index":1,"caption_template":"图片 - _CAPTION_","label":"IP伪装代理示意图","attributes":{},"skip":false,"key":"4.4.6.1"},{"backlink":"guide/auth-with-kubeconfig-or-token.html#fig4.4.7.1","level":"4.4.7","list_caption":"Figure: kubeconfig文件","alt":"kubeconfig文件","nro":36,"url":"../images/brand-kubeconfig-yaml.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"kubeconfig文件","attributes":{},"skip":false,"key":"4.4.7.1"},{"backlink":"guide/authentication.html#fig4.4.8.1","level":"4.4.8","list_caption":"Figure: Kubernetes OpenID Connect Flow","alt":"Kubernetes OpenID Connect Flow","nro":37,"url":"../images/kubernetes-oidc-login.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"Kubernetes OpenID Connect Flow","attributes":{},"skip":false,"key":"4.4.8.1"},{"backlink":"guide/carbin-mobile-dashboard-for-kubernetes.html#fig4.5.6.1","level":"4.5.6","list_caption":"Figure: App Store","alt":"App Store","nro":38,"url":"../images/carbin-kubernetes-mobile-dashboard-1.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"App Store","attributes":{},"skip":false,"key":"4.5.6.1"},{"backlink":"guide/carbin-mobile-dashboard-for-kubernetes.html#fig4.5.6.2","level":"4.5.6","list_caption":"Figure: 在手机上操作Kubernetes集群","alt":"在手机上操作Kubernetes集群","nro":39,"url":"../images/carbin-kubernetes-mobile-dashboard-4.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"在手机上操作Kubernetes集群","attributes":{},"skip":false,"key":"4.5.6.2"},{"backlink":"guide/kubernetes-desktop-client.html#fig4.5.7.1","level":"4.5.7","list_caption":"Figure: Kubernetic客户端","alt":"Kubernetic客户端","nro":40,"url":"../images/kubernetic-desktop-ui.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"Kubernetic客户端","attributes":{},"skip":false,"key":"4.5.7.1"},{"backlink":"guide/deploy-applications-in-kubernetes.html#fig4.6.1.1","level":"4.6.1","list_caption":"Figure: API","alt":"API","nro":41,"url":"../images/k8s-app-monitor-test-api-doc.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"API","attributes":{},"skip":false,"key":"4.6.1.1"},{"backlink":"guide/deploy-applications-in-kubernetes.html#fig4.6.1.2","level":"4.6.1","list_caption":"Figure: wercker","alt":"wercker","nro":42,"url":"../images/k8s-app-monitor-agent-wercker.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"wercker","attributes":{},"skip":false,"key":"4.6.1.2"},{"backlink":"guide/deploy-applications-in-kubernetes.html#fig4.6.1.3","level":"4.6.1","list_caption":"Figure: 图表","alt":"图表","nro":43,"url":"../images/k8s-app-monitor-agent.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"图表","attributes":{},"skip":false,"key":"4.6.1.3"},{"backlink":"guide/migrating-hadoop-yarn-to-kubernetes.html#fig4.6.2.1","level":"4.6.2","list_caption":"Figure: spark on yarn with kubernetes","alt":"spark on yarn with kubernetes","nro":44,"url":"../images/spark-on-yarn-with-kubernetes.png","index":1,"caption_template":"图片 - _CAPTION_","label":"spark on yarn with kubernetes","attributes":{},"skip":false,"key":"4.6.2.1"},{"backlink":"guide/migrating-hadoop-yarn-to-kubernetes.html#fig4.6.2.2","level":"4.6.2","list_caption":"Figure: Terms","alt":"Terms","nro":45,"url":"../images/terms-in-kubernetes-app-deployment.png","index":2,"caption_template":"图片 - _CAPTION_","label":"Terms","attributes":{},"skip":false,"key":"4.6.2.2"},{"backlink":"guide/migrating-hadoop-yarn-to-kubernetes.html#fig4.6.2.3","level":"4.6.2","list_caption":"Figure: 分解步骤解析","alt":"分解步骤解析","nro":46,"url":"../images/migrating-hadoop-yarn-to-kubernetes.png","index":3,"caption_template":"图片 - _CAPTION_","label":"分解步骤解析","attributes":{},"skip":false,"key":"4.6.2.3"},{"backlink":"practice/node-installation.html#fig5.2.7.1","level":"5.2.7","list_caption":"Figure: welcome nginx","alt":"welcome nginx","nro":47,"url":"../images/kubernetes-installation-test-nginx.png","index":1,"caption_template":"图片 - _CAPTION_","label":"welcome nginx","attributes":{},"skip":false,"key":"5.2.7.1"},{"backlink":"practice/dashboard-addon-installation.html#fig5.2.9.1","level":"5.2.9","list_caption":"Figure: kubernetes dashboard","alt":"kubernetes dashboard","nro":48,"url":"../images/kubernetes-dashboard-raw.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"kubernetes dashboard","attributes":{},"skip":false,"key":"5.2.9.1"},{"backlink":"practice/dashboard-addon-installation.html#fig5.2.9.2","level":"5.2.9","list_caption":"Figure: V1.6.3版本的dashboard界面","alt":"V1.6.3版本的dashboard界面","nro":49,"url":"../images/dashboard-v163.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"V1.6.3版本的dashboard界面","attributes":{},"skip":false,"key":"5.2.9.2"},{"backlink":"practice/heapster-addon-installation.html#fig5.2.10.1","level":"5.2.10","list_caption":"Figure: dashboard-heapster","alt":"dashboard-heapster","nro":50,"url":"../images/kubernetes-dashboard-with-heapster.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"dashboard-heapster","attributes":{},"skip":false,"key":"5.2.10.1"},{"backlink":"practice/heapster-addon-installation.html#fig5.2.10.2","level":"5.2.10","list_caption":"Figure: grafana","alt":"grafana","nro":51,"url":"../images/kubernetes-heapster-grafana.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"grafana","attributes":{},"skip":false,"key":"5.2.10.2"},{"backlink":"practice/heapster-addon-installation.html#fig5.2.10.3","level":"5.2.10","list_caption":"Figure: kubernetes-influxdb-heapster","alt":"kubernetes-influxdb-heapster","nro":52,"url":"../images/kubernetes-influxdb-heapster.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"kubernetes-influxdb-heapster","attributes":{},"skip":false,"key":"5.2.10.3"},{"backlink":"practice/heapster-addon-installation.html#fig5.2.10.4","level":"5.2.10","list_caption":"Figure: 修改grafana模板","alt":"修改grafana模板","nro":53,"url":"../images/grafana-dashboard-setting.jpg","index":4,"caption_template":"图片 - _CAPTION_","label":"修改grafana模板","attributes":{},"skip":false,"key":"5.2.10.4"},{"backlink":"practice/efk-addon-installation.html#fig5.2.11.1","level":"5.2.11","list_caption":"Figure: es-setting","alt":"es-setting","nro":54,"url":"../images/es-setting.png","index":1,"caption_template":"图片 - _CAPTION_","label":"es-setting","attributes":{},"skip":false,"key":"5.2.11.1"},{"backlink":"practice/efk-addon-installation.html#fig5.2.11.2","level":"5.2.11","list_caption":"Figure: es-home","alt":"es-home","nro":55,"url":"../images/kubernetes-efk-kibana.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"es-home","attributes":{},"skip":false,"key":"5.2.11.2"},{"backlink":"practice/traefik-ingress-installation.html#fig5.3.1.1","level":"5.3.1","list_caption":"Figure: kubernetes-dashboard","alt":"kubernetes-dashboard","nro":56,"url":"../images/traefik-dashboard.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"kubernetes-dashboard","attributes":{},"skip":false,"key":"5.3.1.1"},{"backlink":"practice/traefik-ingress-installation.html#fig5.3.1.2","level":"5.3.1","list_caption":"Figure: traefik-nginx","alt":"traefik-nginx","nro":57,"url":"../images/traefik-nginx.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"traefik-nginx","attributes":{},"skip":false,"key":"5.3.1.2"},{"backlink":"practice/traefik-ingress-installation.html#fig5.3.1.3","level":"5.3.1","list_caption":"Figure: traefik-guestbook","alt":"traefik-guestbook","nro":58,"url":"../images/traefik-guestbook.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"traefik-guestbook","attributes":{},"skip":false,"key":"5.3.1.3"},{"backlink":"practice/distributed-load-test.html#fig5.3.2.1","level":"5.3.2","list_caption":"Figure: 使用dashboard来扩容","alt":"使用dashboard来扩容","nro":59,"url":"../images/dashbaord-scale.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"使用dashboard来扩容","attributes":{},"skip":false,"key":"5.3.2.1"},{"backlink":"practice/distributed-load-test.html#fig5.3.2.2","level":"5.3.2","list_caption":"Figure: Traefik的UI","alt":"Traefik的UI","nro":60,"url":"../images/traefik-dashboard-locust.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"Traefik的UI","attributes":{},"skip":false,"key":"5.3.2.2"},{"backlink":"practice/distributed-load-test.html#fig5.3.2.3","level":"5.3.2","list_caption":"Figure: Locust启动界面","alt":"Locust启动界面","nro":61,"url":"../images/locust-start-swarming.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"Locust启动界面","attributes":{},"skip":false,"key":"5.3.2.3"},{"backlink":"practice/distributed-load-test.html#fig5.3.2.4","level":"5.3.2","list_caption":"Figure: Dashboard查看页面","alt":"Dashboard查看页面","nro":62,"url":"../images/sample-webapp-rc.jpg","index":4,"caption_template":"图片 - _CAPTION_","label":"Dashboard查看页面","attributes":{},"skip":false,"key":"5.3.2.4"},{"backlink":"practice/distributed-load-test.html#fig5.3.2.5","level":"5.3.2","list_caption":"Figure: Locust测试结果页面","alt":"Locust测试结果页面","nro":63,"url":"../images/locust-dashboard.jpg","index":5,"caption_template":"图片 - _CAPTION_","label":"Locust测试结果页面","attributes":{},"skip":false,"key":"5.3.2.5"},{"backlink":"practice/network-and-cluster-perfermance-test.html#fig5.3.3.1","level":"5.3.3","list_caption":"Figure: kubernetes-dashboard","alt":"kubernetes-dashboard","nro":64,"url":"../images/kubenetes-e2e-test.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"kubernetes-dashboard","attributes":{},"skip":false,"key":"5.3.3.1"},{"backlink":"practice/network-and-cluster-perfermance-test.html#fig5.3.3.2","level":"5.3.3","list_caption":"Figure: locust测试页面","alt":"locust测试页面","nro":65,"url":"../images/kubernetes-locust-test.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"locust测试页面","attributes":{},"skip":false,"key":"5.3.3.2"},{"backlink":"practice/edge-node-configuration.html#fig5.3.4.1","level":"5.3.4","list_caption":"Figure: 边缘节点架构","alt":"边缘节点架构","nro":66,"url":"../images/kubernetes-edge-node-architecture.png","index":1,"caption_template":"图片 - _CAPTION_","label":"边缘节点架构","attributes":{},"skip":false,"key":"5.3.4.1"},{"backlink":"practice/app-log-collection.html#fig5.4.2.1","level":"5.4.2","list_caption":"Figure: filebeat日志收集架构图","alt":"filebeat日志收集架构图","nro":67,"url":"../images/filebeat-log-collector.png","index":1,"caption_template":"图片 - _CAPTION_","label":"filebeat日志收集架构图","attributes":{},"skip":false,"key":"5.4.2.1"},{"backlink":"practice/app-log-collection.html#fig5.4.2.2","level":"5.4.2","list_caption":"Figure: Kibana页面","alt":"Kibana页面","nro":68,"url":"../images/filebeat-docker-test.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"Kibana页面","attributes":{},"skip":false,"key":"5.4.2.2"},{"backlink":"practice/app-log-collection.html#fig5.4.2.3","level":"5.4.2","list_caption":"Figure: filebeat收集的日志详细信息","alt":"filebeat收集的日志详细信息","nro":69,"url":"../images/kubernetes-filebeat-detail.png","index":3,"caption_template":"图片 - _CAPTION_","label":"filebeat收集的日志详细信息","attributes":{},"skip":false,"key":"5.4.2.3"},{"backlink":"practice/monitor.html#fig5.4.4.1","level":"5.4.4","list_caption":"Figure: Kubernetes集群中的监控","alt":"Kubernetes集群中的监控","nro":70,"url":"../images/monitoring-in-kubernetes.png","index":1,"caption_template":"图片 - _CAPTION_","label":"Kubernetes集群中的监控","attributes":{},"skip":false,"key":"5.4.4.1"},{"backlink":"practice/monitor.html#fig5.4.4.2","level":"5.4.4","list_caption":"Figure: kubernetes的容器命名规则示意图","alt":"kubernetes的容器命名规则示意图","nro":71,"url":"../images/kubernetes-container-naming-rule.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"kubernetes的容器命名规则示意图","attributes":{},"skip":false,"key":"5.4.4.2"},{"backlink":"practice/monitor.html#fig5.4.4.3","level":"5.4.4","list_caption":"Figure: Heapster架构图（改进版）","alt":"Heapster架构图（改进版）","nro":72,"url":"../images/kubernetes-heapster-monitoring.png","index":3,"caption_template":"图片 - _CAPTION_","label":"Heapster架构图（改进版）","attributes":{},"skip":false,"key":"5.4.4.3"},{"backlink":"practice/monitor.html#fig5.4.4.4","level":"5.4.4","list_caption":"Figure: 应用监控架构图","alt":"应用监控架构图","nro":73,"url":"../images/kubernetes-app-monitoring.png","index":4,"caption_template":"图片 - _CAPTION_","label":"应用监控架构图","attributes":{},"skip":false,"key":"5.4.4.4"},{"backlink":"practice/monitor.html#fig5.4.4.5","level":"5.4.4","list_caption":"Figure: 应用拓扑图","alt":"应用拓扑图","nro":74,"url":"../images/weave-scope-service-topology.jpg","index":5,"caption_template":"图片 - _CAPTION_","label":"应用拓扑图","attributes":{},"skip":false,"key":"5.4.4.5"},{"backlink":"practice/data-persistence-problem.html#fig5.4.5.1","level":"5.4.5","list_caption":"Figure: 日志持久化收集解决方案示意图","alt":"日志持久化收集解决方案示意图","nro":75,"url":"../images/log-persistence-logstash.png","index":1,"caption_template":"图片 - _CAPTION_","label":"日志持久化收集解决方案示意图","attributes":{},"skip":false,"key":"5.4.5.1"},{"backlink":"practice/storage-for-containers-using-glusterfs-with-openshift.html#fig5.5.1.2.1","level":"5.5.1.2","list_caption":"Figure: 创建存储","alt":"创建存储","nro":76,"url":"../images/create-gluster-storage.png","index":1,"caption_template":"图片 - _CAPTION_","label":"创建存储","attributes":{},"skip":false,"key":"5.5.1.2.1"},{"backlink":"practice/storage-for-containers-using-glusterfs-with-openshift.html#fig5.5.1.2.2","level":"5.5.1.2","list_caption":"Figure: Screen Shot 2017-03-24 at 11.09.34.png","alt":"Screen Shot 2017-03-24 at 11.09.34.png","nro":77,"url":"https://keithtenzer.files.wordpress.com/2017/03/screen-shot-2017-03-24-at-11-09-341.png?w=440","index":2,"caption_template":"图片 - _CAPTION_","label":"Screen Shot 2017-03-24 at 11.09.34.png","attributes":{},"skip":false,"key":"5.5.1.2.2"},{"backlink":"practice/using-heapster-to-get-object-metrics.html#fig5.6.1.1.1","level":"5.6.1.1","list_caption":"Figure: Heapster架构图","alt":"Heapster架构图","nro":78,"url":"../images/heapster-architecture.png","index":1,"caption_template":"图片 - _CAPTION_","label":"Heapster架构图","attributes":{},"skip":false,"key":"5.6.1.1.1"},{"backlink":"practice/prometheus.html#fig5.6.2.1","level":"5.6.2","list_caption":"Figure: Prometheus 架构图","alt":"Prometheus 架构图","nro":79,"url":"../images/prometheus-architecture.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"Prometheus 架构图","attributes":{},"skip":false,"key":"5.6.2.1"},{"backlink":"practice/using-prometheus-to-monitor-kuberentes-cluster.html#fig5.6.2.1.1","level":"5.6.2.1","list_caption":"Figure: Grafana页面","alt":"Grafana页面","nro":80,"url":"../images/kubernetes-prometheus-monitoring.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"Grafana页面","attributes":{},"skip":false,"key":"5.6.2.1.1"},{"backlink":"practice/helm.html#fig5.7.1.1","level":"5.7.1","list_caption":"Figure: Helm chart源","alt":"Helm chart源","nro":81,"url":"../images/helm-charts-repository.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"Helm chart源","attributes":{},"skip":false,"key":"5.7.1.1"},{"backlink":"practice/helm.html#fig5.7.1.2","level":"5.7.1","list_caption":"Figure: TODO应用的Web页面","alt":"TODO应用的Web页面","nro":82,"url":"../images/helm-mean-todo-aholic.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"TODO应用的Web页面","attributes":{},"skip":false,"key":"5.7.1.2"},{"backlink":"practice/create-private-charts-repo.html#fig5.7.2.1","level":"5.7.2","list_caption":"Figure: Helm monocular界面","alt":"Helm monocular界面","nro":83,"url":"../images/helm-monocular-jimmysong.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"Helm monocular界面","attributes":{},"skip":false,"key":"5.7.2.1"},{"backlink":"practice/jenkins-ci-cd.html#fig5.8.1.1","level":"5.8.1","list_caption":"Figure: 基于Jenkins的持续集成与发布","alt":"基于Jenkins的持续集成与发布","nro":84,"url":"../images/kubernetes-jenkins-ci-cd.png","index":1,"caption_template":"图片 - _CAPTION_","label":"基于Jenkins的持续集成与发布","attributes":{},"skip":false,"key":"5.8.1.1"},{"backlink":"practice/drone-ci-cd.html#fig5.8.2.1","level":"5.8.2","list_caption":"Figure: OAuth注册","alt":"OAuth注册","nro":85,"url":"../images/github-oauth-register.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"OAuth注册","attributes":{},"skip":false,"key":"5.8.2.1"},{"backlink":"practice/drone-ci-cd.html#fig5.8.2.2","level":"5.8.2","list_caption":"Figure: OAuth key","alt":"OAuth key","nro":86,"url":"../images/github-oauth-drone-key.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"OAuth key","attributes":{},"skip":false,"key":"5.8.2.2"},{"backlink":"practice/drone-ci-cd.html#fig5.8.2.3","level":"5.8.2","list_caption":"Figure: Drone登陆界面","alt":"Drone登陆界面","nro":87,"url":"../images/drone-login-github.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"Drone登陆界面","attributes":{},"skip":false,"key":"5.8.2.3"},{"backlink":"practice/drone-ci-cd.html#fig5.8.2.4","level":"5.8.2","list_caption":"Figure: Github启用repo设置","alt":"Github启用repo设置","nro":88,"url":"../images/drone-github-active.jpg","index":4,"caption_template":"图片 - _CAPTION_","label":"Github启用repo设置","attributes":{},"skip":false,"key":"5.8.2.4"},{"backlink":"practice/drone-ci-cd.html#fig5.8.2.5","level":"5.8.2","list_caption":"Figure: Github单个repo设置","alt":"Github单个repo设置","nro":89,"url":"../images/drone-github-repo-setting.jpg","index":5,"caption_template":"图片 - _CAPTION_","label":"Github单个repo设置","attributes":{},"skip":false,"key":"5.8.2.5"},{"backlink":"practice/dashboard-upgrade.html#fig5.9.2.1","level":"5.9.2","list_caption":"Figure: 登陆界面","alt":"登陆界面","nro":90,"url":"../images/kubernetes-dashboard-1.7.1-login.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"登陆界面","attributes":{},"skip":false,"key":"5.9.2.1"},{"backlink":"practice/dashboard-upgrade.html#fig5.9.2.2","level":"5.9.2","list_caption":"Figure: 首页","alt":"首页","nro":91,"url":"../images/kubernetes-dashboard-1.7.1-default-page.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"首页","attributes":{},"skip":false,"key":"5.9.2.2"},{"backlink":"practice/dashboard-upgrade.html#fig5.9.2.3","level":"5.9.2","list_caption":"Figure: 用户空间","alt":"用户空间","nro":92,"url":"../images/kubernetes-dashboard-1.7.1-brand.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"用户空间","attributes":{},"skip":false,"key":"5.9.2.3"},{"backlink":"practice/dashboard-upgrade.html#fig5.9.2.4","level":"5.9.2","list_caption":"Figure: kubeconfig文件","alt":"kubeconfig文件","nro":93,"url":"../images/brand-kubeconfig-yaml.jpg","index":4,"caption_template":"图片 - _CAPTION_","label":"kubeconfig文件","attributes":{},"skip":false,"key":"5.9.2.4"},{"backlink":"usecases/microservices.html#fig6.2.1","level":"6.2","list_caption":"Figure: 微服务关注的部分","alt":"微服务关注的部分","nro":94,"url":"../images/microservices-concerns.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"微服务关注的部分","attributes":{},"skip":false,"key":"6.2.1"},{"backlink":"usecases/service-discovery-in-microservices.html#fig6.2.1.1","level":"6.2.1","list_caption":"Figure: 微服务中的服务发现","alt":"微服务中的服务发现","nro":95,"url":"../images/service-discovery-in-microservices.png","index":1,"caption_template":"图片 - _CAPTION_","label":"微服务中的服务发现","attributes":{},"skip":false,"key":"6.2.1.1"},{"backlink":"usecases/service-mesh.html#fig6.3.1","level":"6.3","list_caption":"Figure: 下一代异构微服务架构","alt":"下一代异构微服务架构","nro":96,"url":"../images/polyglot-microservices-serivce-mesh.png","index":1,"caption_template":"图片 - _CAPTION_","label":"下一代异构微服务架构","attributes":{},"skip":false,"key":"6.3.1"},{"backlink":"usecases/service-mesh.html#fig6.3.2","level":"6.3","list_caption":"Figure: Service Mesh 架构图","alt":"Service Mesh 架构图","nro":97,"url":"../images/serivce-mesh-control-plane.png","index":2,"caption_template":"图片 - _CAPTION_","label":"Service Mesh 架构图","attributes":{},"skip":false,"key":"6.3.2"},{"backlink":"usecases/service-mesh.html#fig6.3.3","level":"6.3","list_caption":"Figure: Istio vs linkerd","alt":"Istio vs linkerd","nro":98,"url":"../images/istio-vs-linkerd.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"Istio vs linkerd","attributes":{},"skip":false,"key":"6.3.3"},{"backlink":"usecases/istio.html#fig6.3.1.1","level":"6.3.1","list_caption":"Figure: Istio架构图","alt":"Istio架构图","nro":99,"url":"../images/istio-arch.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"Istio架构图","attributes":{},"skip":false,"key":"6.3.1.1"},{"backlink":"usecases/istio-installation.html#fig6.3.1.1.1","level":"6.3.1.1","list_caption":"Figure: BookInfo Sample应用架构图","alt":"BookInfo Sample应用架构图","nro":100,"url":"../images/bookinfo-sample-arch.png","index":1,"caption_template":"图片 - _CAPTION_","label":"BookInfo Sample应用架构图","attributes":{},"skip":false,"key":"6.3.1.1.1"},{"backlink":"usecases/istio-installation.html#fig6.3.1.1.2","level":"6.3.1.1","list_caption":"Figure: BookInfo Sample页面","alt":"BookInfo Sample页面","nro":101,"url":"../images/bookinfo-sample.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"BookInfo Sample页面","attributes":{},"skip":false,"key":"6.3.1.1.2"},{"backlink":"usecases/istio-installation.html#fig6.3.1.1.3","level":"6.3.1.1","list_caption":"Figure: Istio Grafana界面","alt":"Istio Grafana界面","nro":102,"url":"../images/istio-grafana.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"Istio Grafana界面","attributes":{},"skip":false,"key":"6.3.1.1.3"},{"backlink":"usecases/istio-installation.html#fig6.3.1.1.4","level":"6.3.1.1","list_caption":"Figure: Prometheus页面","alt":"Prometheus页面","nro":103,"url":"../images/istio-prometheus.jpg","index":4,"caption_template":"图片 - _CAPTION_","label":"Prometheus页面","attributes":{},"skip":false,"key":"6.3.1.1.4"},{"backlink":"usecases/istio-installation.html#fig6.3.1.1.5","level":"6.3.1.1","list_caption":"Figure: Zipkin页面","alt":"Zipkin页面","nro":104,"url":"../images/istio-zipkin.jpg","index":5,"caption_template":"图片 - _CAPTION_","label":"Zipkin页面","attributes":{},"skip":false,"key":"6.3.1.1.5"},{"backlink":"usecases/istio-installation.html#fig6.3.1.1.6","level":"6.3.1.1","list_caption":"Figure: ServiceGraph页面","alt":"ServiceGraph页面","nro":105,"url":"../images/istio-servicegraph.jpg","index":6,"caption_template":"图片 - _CAPTION_","label":"ServiceGraph页面","attributes":{},"skip":false,"key":"6.3.1.1.6"},{"backlink":"usecases/install-and-expand-istio-mesh.html#fig6.3.1.3.1","level":"6.3.1.3","list_caption":"Figure: BookInfo","alt":"BookInfo","nro":106,"url":"../images/noistio.png","index":1,"caption_template":"图片 - _CAPTION_","label":"BookInfo","attributes":{},"skip":false,"key":"6.3.1.3.1"},{"backlink":"usecases/install-and-expand-istio-mesh.html#fig6.3.1.3.2","level":"6.3.1.3","list_caption":"Figure: BookInfo","alt":"BookInfo","nro":107,"url":"../images/noistio.png","index":2,"caption_template":"图片 - _CAPTION_","label":"BookInfo","attributes":{},"skip":false,"key":"6.3.1.3.2"},{"backlink":"usecases/integrating-vms.html#fig6.3.1.4.1","level":"6.3.1.4","list_caption":"Figure: Bookinfo应用的拓展Mesh","alt":"Bookinfo应用的拓展Mesh","nro":108,"url":"../images/istio-mesh-expansion.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"Bookinfo应用的拓展Mesh","attributes":{},"skip":false,"key":"6.3.1.4.1"},{"backlink":"usecases/linkerd.html#fig6.3.2.1","level":"6.3.2","list_caption":"Figure: source https://linkerd.io","alt":"source https://linkerd.io","nro":109,"url":"../images/diagram-individual-instance.png","index":1,"caption_template":"图片 - _CAPTION_","label":"source https://linkerd.io","attributes":{},"skip":false,"key":"6.3.2.1"},{"backlink":"usecases/linkerd-user-guide.html#fig6.3.2.1.1","level":"6.3.2.1","list_caption":"Figure: Jenkins pipeline","alt":"Jenkins pipeline","nro":110,"url":"../images/linkerd-jenkins-pipeline.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"Jenkins pipeline","attributes":{},"skip":false,"key":"6.3.2.1.1"},{"backlink":"usecases/linkerd-user-guide.html#fig6.3.2.1.2","level":"6.3.2.1","list_caption":"Figure: Jenkins config","alt":"Jenkins config","nro":111,"url":"../images/linkerd-jenkins.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"Jenkins config","attributes":{},"skip":false,"key":"6.3.2.1.2"},{"backlink":"usecases/linkerd-user-guide.html#fig6.3.2.1.3","level":"6.3.2.1","list_caption":"Figure: namerd","alt":"namerd","nro":112,"url":"../images/namerd-internal.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"namerd","attributes":{},"skip":false,"key":"6.3.2.1.3"},{"backlink":"usecases/linkerd-user-guide.html#fig6.3.2.1.4","level":"6.3.2.1","list_caption":"Figure: linkerd监控","alt":"linkerd监控","nro":113,"url":"../images/linkerd-helloworld-outgoing.jpg","index":4,"caption_template":"图片 - _CAPTION_","label":"linkerd监控","attributes":{},"skip":false,"key":"6.3.2.1.4"},{"backlink":"usecases/linkerd-user-guide.html#fig6.3.2.1.5","level":"6.3.2.1","list_caption":"Figure: linkerd监控","alt":"linkerd监控","nro":114,"url":"../images/linkerd-helloworld-incoming.jpg","index":5,"caption_template":"图片 - _CAPTION_","label":"linkerd监控","attributes":{},"skip":false,"key":"6.3.2.1.5"},{"backlink":"usecases/linkerd-user-guide.html#fig6.3.2.1.6","level":"6.3.2.1","list_caption":"Figure: linkerd性能监控","alt":"linkerd性能监控","nro":115,"url":"../images/linkerd-grafana.png","index":6,"caption_template":"图片 - _CAPTION_","label":"linkerd性能监控","attributes":{},"skip":false,"key":"6.3.2.1.6"},{"backlink":"usecases/linkerd-user-guide.html#fig6.3.2.1.7","level":"6.3.2.1","list_caption":"Figure: Linkerd ingress controller","alt":"Linkerd ingress controller","nro":116,"url":"../images/linkerd-ingress-controller.jpg","index":7,"caption_template":"图片 - _CAPTION_","label":"Linkerd ingress controller","attributes":{},"skip":false,"key":"6.3.2.1.7"},{"backlink":"usecases/conduit-installation.html#fig6.3.3.2.1","level":"6.3.3.2","list_caption":"Figure: Conduit dashboard","alt":"Conduit dashboard","nro":117,"url":"../images/conduit-dashboard.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"Conduit dashboard","attributes":{},"skip":false,"key":"6.3.3.2.1"},{"backlink":"usecases/big-data.html#fig6.4.1","level":"6.4","list_caption":"Figure: Spark on yarn with kubernetes","alt":"Spark on yarn with kubernetes","nro":118,"url":"../images/spark-on-yarn-with-kubernetes.png","index":1,"caption_template":"图片 - _CAPTION_","label":"Spark on yarn with kubernetes","attributes":{},"skip":false,"key":"6.4.1"},{"backlink":"usecases/big-data.html#fig6.4.2","level":"6.4","list_caption":"Figure: 在kubernetes上使用多种调度方式","alt":"在kubernetes上使用多种调度方式","nro":119,"url":"../images/spark-on-kubernetes-with-different-schedulers.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"在kubernetes上使用多种调度方式","attributes":{},"skip":false,"key":"6.4.2"},{"backlink":"usecases/spark-standalone-on-kubernetes.html#fig6.4.1.1","level":"6.4.1","list_caption":"Figure: spark master ui","alt":"spark master ui","nro":120,"url":"../images/spark-ui.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"spark master ui","attributes":{},"skip":false,"key":"6.4.1.1"},{"backlink":"usecases/spark-standalone-on-kubernetes.html#fig6.4.1.2","level":"6.4.1","list_caption":"Figure: zeppelin ui","alt":"zeppelin ui","nro":121,"url":"../images/zeppelin-ui.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"zeppelin ui","attributes":{},"skip":false,"key":"6.4.1.2"},{"backlink":"usecases/serverless.html#fig6.5.1","level":"6.5","list_caption":"Figure: 从物理机到函数计算","alt":"从物理机到函数计算","nro":122,"url":"../images/from-bare-metal-to-functions.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"从物理机到函数计算","attributes":{},"skip":false,"key":"6.5.1"},{"backlink":"usecases/serverless.html#fig6.5.2","level":"6.5","list_caption":"Figure: FaaS Landscape","alt":"FaaS Landscape","nro":123,"url":"../images/redpoint-faas-landscape.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"FaaS Landscape","attributes":{},"skip":false,"key":"6.5.2"},{"backlink":"usecases/understanding-serverless.html#fig6.5.1.1","level":"6.5.1","list_caption":"Figure: 服务端软件的运行环境","alt":"服务端软件的运行环境","nro":124,"url":"../images/serverless-server-side-software.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"服务端软件的运行环境","attributes":{},"skip":false,"key":"6.5.1.1"},{"backlink":"usecases/understanding-serverless.html#fig6.5.1.2","level":"6.5.1","list_caption":"Figure: FaaS应用架构","alt":"FaaS应用架构","nro":125,"url":"../images/serverless-faas-platform.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"FaaS应用架构","attributes":{},"skip":false,"key":"6.5.1.2"},{"backlink":"usecases/understanding-serverless.html#fig6.5.1.3","level":"6.5.1","list_caption":"Figure: 传统应用程序架构","alt":"传统应用程序架构","nro":126,"url":"../images/non-serverless-game-arch.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"传统应用程序架构","attributes":{},"skip":false,"key":"6.5.1.3"},{"backlink":"usecases/understanding-serverless.html#fig6.5.1.4","level":"6.5.1","list_caption":"Figure: Serverless架构","alt":"Serverless架构","nro":127,"url":"../images/serverless-game-arch.jpg","index":4,"caption_template":"图片 - _CAPTION_","label":"Serverless架构","attributes":{},"skip":false,"key":"6.5.1.4"},{"backlink":"usecases/openfaas.html#fig6.5.2.1.1","level":"6.5.2.1","list_caption":"Figure: OpenFaaS架构","alt":"OpenFaaS架构","nro":128,"url":"../images/openfaas-arch.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"OpenFaaS架构","attributes":{},"skip":false,"key":"6.5.2.1.1"},{"backlink":"usecases/openfaas.html#fig6.5.2.1.2","level":"6.5.2.1","list_caption":"Figure: OpenFaaS Prometheus","alt":"OpenFaaS Prometheus","nro":129,"url":"../images/openfaas-prometheus.jpg","index":2,"caption_template":"图片 - _CAPTION_","label":"OpenFaaS Prometheus","attributes":{},"skip":false,"key":"6.5.2.1.2"},{"backlink":"usecases/openfaas.html#fig6.5.2.1.3","level":"6.5.2.1","list_caption":"Figure: OpenFaas Grafana监控","alt":"OpenFaas Grafana监控","nro":130,"url":"../images/openfaas-grafana.jpg","index":3,"caption_template":"图片 - _CAPTION_","label":"OpenFaas Grafana监控","attributes":{},"skip":false,"key":"6.5.2.1.3"},{"backlink":"usecases/openfaas.html#fig6.5.2.1.4","level":"6.5.2.1","list_caption":"Figure: OpenFaas Dashboard","alt":"OpenFaas Dashboard","nro":131,"url":"../images/openfaas-deploy-a-function.jpg","index":4,"caption_template":"图片 - _CAPTION_","label":"OpenFaas Dashboard","attributes":{},"skip":false,"key":"6.5.2.1.4"},{"backlink":"usecases/openfaas.html#fig6.5.2.1.5","level":"6.5.2.1","list_caption":"Figure: NodeInfo执行结果","alt":"NodeInfo执行结果","nro":132,"url":"../images/openfaas-nodeinfo.jpg","index":5,"caption_template":"图片 - _CAPTION_","label":"NodeInfo执行结果","attributes":{},"skip":false,"key":"6.5.2.1.5"},{"backlink":"develop/sigs-and-working-group.html#fig7.2.1","level":"7.2","list_caption":"Figure: Kubernetes SIG","alt":"Kubernetes SIG","nro":133,"url":"../images/kubernetes-sigs.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"Kubernetes SIG","attributes":{},"skip":false,"key":"7.2.1"},{"backlink":"develop/client-go-sample.html#fig7.5.1","level":"7.5","list_caption":"Figure: 使用kubernetes dashboard进行故障排查","alt":"使用kubernetes dashboard进行故障排查","nro":134,"url":"../images/kubernetes-client-go-sample-update.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"使用kubernetes dashboard进行故障排查","attributes":{},"skip":false,"key":"7.5.1"},{"backlink":"appendix/issues.html#fig8.6.1","level":"8.6","list_caption":"Figure: pvc-storage-limit","alt":"pvc-storage-limit","nro":135,"url":"../images/pvc-storage-limit.jpg","index":1,"caption_template":"图片 - _CAPTION_","label":"pvc-storage-limit","attributes":{},"skip":false,"key":"8.6.1"}]},"title":"Kubernetes Handbook - jimmysong.io","language":"zh-hans","links":{"sidebar":{"Jimmy Song":"https://jimmysong.io"}},"gitbook":"*","description":"Kubernetes中文指南/实践手册|kubernetes handbook|jimmysong.io|宋净超"},"file":{"path":"usecases/install-and-expand-istio-mesh.md","mtime":"2017-11-21T13:18:30.606Z","type":"markdown"},"gitbook":{"version":"3.2.2","time":"2017-12-20T10:50:52.717Z"},"basePath":"..","book":{"language":""}}); }); diff --git a/usecases/integrating-vms.html b/usecases/integrating-vms.html index e550345f7..d79a56684 100644 --- a/usecases/integrating-vms.html +++ b/usecases/integrating-vms.html @@ -2859,7 +2859,7 @@ istioctl create -n bookinfo -f samples/bookinfo/kube diff --git a/usecases/istio-installation.html b/usecases/istio-installation.html index a7c519493..dc802a223 100644 --- a/usecases/istio-installation.html +++ b/usecases/istio-installation.html @@ -3033,7 +3033,7 @@ istio/examples-bookinfo-productpage-v1 diff --git a/usecases/istio.html b/usecases/istio.html index 14f106a82..62bc4abfa 100644 --- a/usecases/istio.html +++ b/usecases/istio.html @@ -2818,7 +2818,7 @@ diff --git a/usecases/linkerd-user-guide.html b/usecases/linkerd-user-guide.html index faeee797a..e88887dac 100644 --- a/usecases/linkerd-user-guide.html +++ b/usecases/linkerd-user-guide.html @@ -2977,7 +2977,7 @@ Hello (172.30.60.14) world (172.30.71.19)!! diff --git a/usecases/linkerd.html b/usecases/linkerd.html index 51f951374..ecc0bc214 100644 --- a/usecases/linkerd.html +++ b/usecases/linkerd.html @@ -2798,7 +2798,7 @@ diff --git a/usecases/microservices.html b/usecases/microservices.html index 6761d366d..bdf98b810 100644 --- a/usecases/microservices.html +++ b/usecases/microservices.html @@ -2783,7 +2783,7 @@ diff --git a/usecases/openfaas.html b/usecases/openfaas.html index 2b1c72e59..969fd2de8 100644 --- a/usecases/openfaas.html +++ b/usecases/openfaas.html @@ -2812,7 +2812,7 @@ diff --git a/usecases/running-spark-with-kubernetes-native-scheduler.html b/usecases/running-spark-with-kubernetes-native-scheduler.html index b8e155cfd..2ca118018 100644 --- a/usecases/running-spark-with-kubernetes-native-scheduler.html +++ b/usecases/running-spark-with-kubernetes-native-scheduler.html @@ -3166,7 +3166,7 @@ sz-pg-oam-docker-hub-001.tendcloud.com/library/spark-driver-py:v2.1.0-kubernetes diff --git a/usecases/serverless.html b/usecases/serverless.html index 2b44a57fe..dc86118bb 100644 --- a/usecases/serverless.html +++ b/usecases/serverless.html @@ -2826,7 +2826,7 @@ diff --git a/usecases/service-discovery-in-microservices.html b/usecases/service-discovery-in-microservices.html index fac9228a2..5633164f8 100644 --- a/usecases/service-discovery-in-microservices.html +++ b/usecases/service-discovery-in-microservices.html @@ -2820,7 +2820,7 @@ diff --git a/usecases/service-mesh.html b/usecases/service-mesh.html index 9ba6a4427..0e43866af 100644 --- a/usecases/service-mesh.html +++ b/usecases/service-mesh.html @@ -2877,7 +2877,7 @@ diff --git a/usecases/spark-standalone-on-kubernetes.html b/usecases/spark-standalone-on-kubernetes.html index 92ee25b84..305e8424a 100644 --- a/usecases/spark-standalone-on-kubernetes.html +++ b/usecases/spark-standalone-on-kubernetes.html @@ -2803,7 +2803,7 @@ index.tenxcloud.com/jimmy/zeppelin:0.7.1 diff --git a/usecases/understanding-serverless.html b/usecases/understanding-serverless.html index 964d13639..c6d98d66c 100644 --- a/usecases/understanding-serverless.html +++ b/usecases/understanding-serverless.html @@ -2856,7 +2856,7 @@