360aff4a57
* rename ansible groups to use _ instead of - k8s-cluster -> k8s_cluster k8s-node -> k8s_node calico-rr -> calico_rr no-floating -> no_floating Note: kube-node,k8s-cluster groups in upgrade CI need clean-up after v2.16 is tagged * ensure old groups are mapped to the new ones |
||
---|---|---|
.. | ||
group_vars/all | ||
roles | ||
README.md | ||
dind-cluster.yaml | ||
hosts | ||
kubespray-dind.yaml | ||
requirements.txt | ||
run-test-distros.sh | ||
test-most_distros-some_CNIs.env | ||
test-some_distros-kube_router_combo.env | ||
test-some_distros-most_CNIs.env |
README.md
Kubespray DIND experimental setup
This ansible playbook creates local docker containers to serve as Kubernetes "nodes", which in turn will run "normal" Kubernetes docker containers, a mode usually called DIND (Docker-IN-Docker).
The playbook has two roles:
- dind-host: creates the "nodes" as containers in localhost, with appropriate settings for DIND (privileged, volume mapping for dind storage, etc).
- dind-cluster: customizes each node container to have required system packages installed, and some utils (swapoff, lsattr) symlinked to /bin/true to ease mimicking a real node.
This playbook has been test with Ubuntu 16.04 as host and ubuntu:16.04 as docker images (note that dind-cluster has specific customization for these images).
The playbook also creates a /tmp/kubespray.dind.inventory_builder.sh
helper (wraps up running contrib/inventory_builder/inventory.py
with
node containers IPs and prefix).
Deploying
See below for a complete successful run:
- Create the node containers
# From the kubespray root dir
cd contrib/dind
pip install -r requirements.txt
ansible-playbook -i hosts dind-cluster.yaml
# Back to kubespray root
cd ../..
NOTE: if the playbook run fails with something like below error
message, you may need to specifically set ansible_python_interpreter
,
see ./hosts
file for an example expanded localhost entry.
failed: [localhost] (item=kube-node1) => {"changed": false, "item": "kube-node1", "msg": "Failed to import docker or docker-py - No module named requests.exceptions. Try `pip install docker` or `pip install docker-py` (Python 2.6)"}
- Customize kubespray-dind.yaml
Note that there's coupling between above created node containers
and kubespray-dind.yaml
settings, in particular regarding selected node_distro
(as set in group_vars/all/all.yaml
), and docker settings.
$EDITOR contrib/dind/kubespray-dind.yaml
- Prepare the inventory and run the playbook
INVENTORY_DIR=inventory/local-dind
mkdir -p ${INVENTORY_DIR}
rm -f ${INVENTORY_DIR}/hosts.ini
CONFIG_FILE=${INVENTORY_DIR}/hosts.ini /tmp/kubespray.dind.inventory_builder.sh
ansible-playbook --become -e ansible_ssh_user=debian -i ${INVENTORY_DIR}/hosts.ini cluster.yml --extra-vars @contrib/dind/kubespray-dind.yaml
NOTE: You could also test other distros without editing files by
passing --extra-vars
as per below commandline,
replacing DISTRO
by either debian
, ubuntu
, centos
, fedora
:
cd contrib/dind
ansible-playbook -i hosts dind-cluster.yaml --extra-vars node_distro=DISTRO
cd ../..
CONFIG_FILE=inventory/local-dind/hosts.ini /tmp/kubespray.dind.inventory_builder.sh
ansible-playbook --become -e ansible_ssh_user=DISTRO -i inventory/local-dind/hosts.ini cluster.yml --extra-vars @contrib/dind/kubespray-dind.yaml --extra-vars bootstrap_os=DISTRO
Resulting deployment
See below to get an idea on how a completed deployment looks like, from the host where you ran kubespray playbooks.
node_distro: debian
Running from an Ubuntu Xenial host:
$ uname -a
Linux ip-xx-xx-xx-xx 4.4.0-1069-aws #79-Ubuntu SMP Mon Sep 24
15:01:41 UTC 2018 x86_64 x86_64 x86_64 GNU/Linux
$ docker ps
CONTAINER ID IMAGE COMMAND CREATED STATUS PORTS NAMES
1835dd183b75 debian:9.5 "sh -c 'apt-get -qy …" 43 minutes ago Up 43 minutes kube-node5
30b0af8d2924 debian:9.5 "sh -c 'apt-get -qy …" 43 minutes ago Up 43 minutes kube-node4
3e0d1510c62f debian:9.5 "sh -c 'apt-get -qy …" 43 minutes ago Up 43 minutes kube-node3
738993566f94 debian:9.5 "sh -c 'apt-get -qy …" 44 minutes ago Up 44 minutes kube-node2
c581ef662ed2 debian:9.5 "sh -c 'apt-get -qy …" 44 minutes ago Up 44 minutes kube-node1
$ docker exec kube-node1 kubectl get node
NAME STATUS ROLES AGE VERSION
kube-node1 Ready master,node 18m v1.12.1
kube-node2 Ready master,node 17m v1.12.1
kube-node3 Ready node 17m v1.12.1
kube-node4 Ready node 17m v1.12.1
kube-node5 Ready node 17m v1.12.1
$ docker exec kube-node1 kubectl get pod --all-namespaces
NAMESPACE NAME READY STATUS RESTARTS AGE
default netchecker-agent-67489 1/1 Running 0 2m51s
default netchecker-agent-6qq6s 1/1 Running 0 2m51s
default netchecker-agent-fsw92 1/1 Running 0 2m51s
default netchecker-agent-fw6tl 1/1 Running 0 2m51s
default netchecker-agent-hostnet-8f2zb 1/1 Running 0 3m
default netchecker-agent-hostnet-gq7ml 1/1 Running 0 3m
default netchecker-agent-hostnet-jfkgv 1/1 Running 0 3m
default netchecker-agent-hostnet-kwfwx 1/1 Running 0 3m
default netchecker-agent-hostnet-r46nm 1/1 Running 0 3m
default netchecker-agent-lxdrn 1/1 Running 0 2m51s
default netchecker-server-864bd4c897-9vstl 1/1 Running 0 2m40s
default sh-68fcc6db45-qf55h 1/1 Running 1 12m
kube-system coredns-7598f59475-6vknq 1/1 Running 0 14m
kube-system coredns-7598f59475-l5q5x 1/1 Running 0 14m
kube-system kube-apiserver-kube-node1 1/1 Running 0 17m
kube-system kube-apiserver-kube-node2 1/1 Running 0 18m
kube-system kube-controller-manager-kube-node1 1/1 Running 0 18m
kube-system kube-controller-manager-kube-node2 1/1 Running 0 18m
kube-system kube-proxy-5xx9d 1/1 Running 0 17m
kube-system kube-proxy-cdqq4 1/1 Running 0 17m
kube-system kube-proxy-n64ls 1/1 Running 0 17m
kube-system kube-proxy-pswmj 1/1 Running 0 18m
kube-system kube-proxy-x89qw 1/1 Running 0 18m
kube-system kube-scheduler-kube-node1 1/1 Running 4 17m
kube-system kube-scheduler-kube-node2 1/1 Running 4 18m
kube-system kubernetes-dashboard-5db4d9f45f-548rl 1/1 Running 0 14m
kube-system nginx-proxy-kube-node3 1/1 Running 4 17m
kube-system nginx-proxy-kube-node4 1/1 Running 4 17m
kube-system nginx-proxy-kube-node5 1/1 Running 4 17m
kube-system weave-net-42bfr 2/2 Running 0 16m
kube-system weave-net-6gt8m 2/2 Running 0 16m
kube-system weave-net-88nnc 2/2 Running 0 16m
kube-system weave-net-shckr 2/2 Running 0 16m
kube-system weave-net-xr46t 2/2 Running 0 16m
$ docker exec kube-node1 curl -s http://localhost:31081/api/v1/connectivity_check
{"Message":"All 10 pods successfully reported back to the server","Absent":null,"Outdated":null}
Using ./run-test-distros.sh
You can use ./run-test-distros.sh
to run a set of tests via DIND,
and excerpt from this script, to get an idea:
# The SPEC file(s) must have two arrays as e.g.
# DISTROS=(debian centos)
# EXTRAS=(
# 'kube_network_plugin=calico'
# 'kube_network_plugin=flannel'
# 'kube_network_plugin=weave'
# )
# that will be tested in a "combinatory" way (e.g. from above there'll be
# be 6 test runs), creating a sequenced <spec_filename>-nn.out with each output.
#
# Each $EXTRAS element will be whitespace split, and passed as --extra-vars
# to main kubespray ansible-playbook run.
See e.g. test-some_distros-most_CNIs.env
and
test-some_distros-kube_router_combo.env
in particular for a richer
set of CNI specific --extra-vars
combo.