kubeasz/docs/op/ChangeVIP.md

108 lines
5.8 KiB
Markdown
Raw Blame History

This file contains ambiguous Unicode characters!

This file contains ambiguous Unicode characters that may be confused with others in your current locale. If your use case is intentional and legitimate, you can safely ignore this warning. Use the Escape button to highlight these characters.

# 更改高可用 `Master IP`
**WARNING:** 更改集群的 `Master VIP`操作有风险,不建议在生产环境直接操作,此文档实践一个修改的操作流程,帮助理解整个集群运行架构和 `kubeasz`的部署逻辑,请在测试环境操作练手。
**BUG:** 目前该操作只适用于集群网络选用`calico`,如果使用`flannel`操作变更后会出现POD地址分配错误的BUG。
首先分析大概操作思路:
- 修改`/etc/ansible/hosts`里面的配置项`MASTER_IP` `KUBE_APISERVER`
- 修改LB节点的keepalive的配置重启keepalived服务
- 修改kubectl/kube-proxy的配置文件使用新VIP地址更新api-server地址
- 重新生成master证书hosts字段包含新VIP地址
- 修改kubelet的配置文件kubelet的配置文件和证书是由bootstrap机制自动生成的
- 删除kubelet.kubeconfig
- 删除集群所有node 节点
- 所有节点重新bootstrap
## 变更前状态验证
``` bash
$ kubectl get cs,node,pod -o wide
NAME STATUS MESSAGE ERROR
controller-manager Healthy ok
scheduler Healthy ok
etcd-2 Healthy {"health":"true"}
etcd-0 Healthy {"health":"true"}
etcd-1 Healthy {"health":"true"}
NAME STATUS ROLES AGE VERSION EXTERNAL-IP OS-IMAGE KERNEL-VERSION CONTAINER-RUNTIME
192.168.1.41 Ready,SchedulingDisabled <none> 2h v1.10.0 <none> Ubuntu 16.04.3 LTS 4.4.0-97-generic docker://18.3.0
192.168.1.42 Ready,SchedulingDisabled <none> 2h v1.10.0 <none> Ubuntu 16.04.3 LTS 4.4.0-97-generic docker://18.3.0
192.168.1.43 Ready <none> 2h v1.10.0 <none> Ubuntu 16.04.3 LTS 4.4.0-97-generic docker://18.3.0
192.168.1.44 Ready <none> 2h v1.10.0 <none> Ubuntu 16.04.3 LTS 4.4.0-98-generic docker://18.3.0
192.168.1.45 Ready <none> 2h v1.10.0 <none> Ubuntu 16.04.3 LTS 4.4.0-98-generic docker://18.3.0
NAME READY STATUS RESTARTS AGE IP NODE
busy-5d6b6b5d4b-8wxkp 1/1 Running 0 17h 172.20.135.133 192.168.1.41
busy-5d6b6b5d4b-fcmkp 1/1 Running 0 17h 172.20.135.128 192.168.1.41
busy-5d6b6b5d4b-ptvd7 1/1 Running 0 17h 172.20.135.136 192.168.1.41
nginx-768979984b-ncqbp 1/1 Running 0 17h 172.20.135.137 192.168.1.41
# 查看待变更集群 Master VIP
$ kubectl cluster-info
Kubernetes master is running at https://192.168.1.39:8443
```
## 变更操作
- `ansible playbook`可以使用tags来控制只允许部分任务执行这里为简化操作没有细化在deploy节点具体操作如下
``` bash
# 1.修改/etc/ansible/hosts 配置项MASTER_IPKUBE_APISERVER
# 2.删除集群所有node节点等待重新bootstrap
$ kubectl get node |grep Ready|awk '{print $1}' |xargs kubectl delete node
# 3.重置keepalived 和修改kubectl/kube-proxy/bootstrap配置
$ ansible-playbook 01.prepare.yml
# 4.删除旧master证书
$ ansible kube-master -m file -a 'path=/etc/kubernetes/ssl/kubernetes.pem state=absent'
# 5.删除旧kubelet配置文件
$ ansible all -m file -a 'path=/etc/kubernetes/kubelet.kubeconfig state=absent'
# 6.重新配置启动master节点
$ ansible-playbook 04.kube-master.yml
# 7.重新配置启动node节点
$ ansible-playbook 05.kube-node.yml
```
## 变更后验证
``` bash
$ kubectl get cs,node,pod -o wide
NAME STATUS MESSAGE ERROR
scheduler Healthy ok
controller-manager Healthy ok
etcd-2 Healthy {"health":"true"}
etcd-1 Healthy {"health":"true"}
etcd-0 Healthy {"health":"true"}
NAME STATUS ROLES AGE VERSION EXTERNAL-IP OS-IMAGE KERNEL-VERSION CONTAINER-RUNTIME
192.168.1.41 Ready,SchedulingDisabled <none> 4m v1.10.0 <none> Ubuntu 16.04.3 LTS 4.4.0-97-generic docker://18.3.0
192.168.1.42 Ready,SchedulingDisabled <none> 4m v1.10.0 <none> Ubuntu 16.04.3 LTS 4.4.0-97-generic docker://18.3.0
192.168.1.43 Ready <none> 3m v1.10.0 <none> Ubuntu 16.04.3 LTS 4.4.0-97-generic docker://18.3.0
192.168.1.44 Ready <none> 3m v1.10.0 <none> Ubuntu 16.04.3 LTS 4.4.0-98-generic docker://18.3.0
192.168.1.45 Ready <none> 3m v1.10.0 <none> Ubuntu 16.04.3 LTS 4.4.0-98-generic docker://18.3.0
NAME READY STATUS RESTARTS AGE IP NODE
busy-5d6b6b5d4b-25hfr 1/1 Running 0 5m 172.20.237.64 192.168.1.43
busy-5d6b6b5d4b-cdzb5 1/1 Running 0 5m 172.20.145.192 192.168.1.44
busy-5d6b6b5d4b-m2rf7 1/1 Running 0 5m 172.20.26.131 192.168.1.45
nginx-768979984b-2ngww 1/1 Running 0 5m 172.20.145.193 192.168.1.44
# 查看集群master VIP已经变更
$ kubectl cluster-info
Kubernetes master is running at https://192.168.1.40:8443
```
## 小结
本示例操作演示了多主多节点k8s集群变更`Master VIP`的操作,有助于理解整个集群组件架构和`kubeasz`的安装逻辑,小结如下:
- 变更操作不影响集群已运行的业务POD但是操作过程中业务会中断
- 已运行POD会重新调度到各node节点如果业务POD量很大短时间内会对集群造成压力
- 不建议在生成环境直接操作,本示例演示说明为主