[jjo] add DIND support to contrib/ (#3468)

* [jjo] add DIND support to contrib/

- add contrib/dind with ansible playbook to
  create "node" containers, and setup them to mimic
  host nodes as much as possible (using Ubuntu images),
  see contrib/dind/README.md

- nodes' /etc/hosts editing via `blockinfile` and
  `lineinfile` need `unsafe_writes: yes` because /etc/hosts
  are mounted by docker, and thus can't be handled atomically
  (modify copy + rename)

* dind-host role: set node container hostname on creation

* add "Resulting deployment" section with some CLI outputs

* typo

* selectable node_distro: debian, ubuntu

* some fixes for node_distro: ubuntu

* cpu optimization: add early `pkill -STOP agetty`

* typo

* add centos dind support ;)

* add kubespray-dind.yaml, support fedora

- add kubespray-dind.yaml (former custom.yaml at README.md)
- rework README.md as per above
- use some YAML power to share distros' commonality
- add fedora support

* create unique /etc/machine-id and other updates

- create unique /etc/machine-id in each docker node,
  used as seed for e.g. weave mac addresses

- with above, now netchecker 100% passes WoHooOO!
  🎉 🎉 🎉

- updated README.md output from (1.12.1, verified
  netcheck)

* minor typos

* fix centos node creation, needs earlier udevadm removal to avoid flaky facts, also verified netcheck Ok \o/

* add Q&D test-distros.sh, back to manual /etc/machine-id hack

* run-test-distros.sh cosmetics and minor fixes

* run-test-distros.sh: $rc fix and minor formatting changes

* run-test-distros.sh output cosmetics
This commit is contained in:
JuanJo Ciarlante 2018-10-15 04:44:02 -03:00 committed by Antoine Legrand
parent fac8aaa44e
commit 4077934519
12 changed files with 451 additions and 0 deletions

152
contrib/dind/README.md Normal file
View File

@ -0,0 +1,152 @@
# 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
appropiate 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:
1. 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)"}
~~~
2. 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
~~~
3. 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}
~~~

View File

@ -0,0 +1,9 @@
---
- hosts: localhost
gather_facts: False
roles:
- { role: dind-host }
- hosts: containers
roles:
- { role: dind-cluster }

View File

@ -0,0 +1,2 @@
# See distro.yaml for supported node_distro images
node_distro: debian

View File

@ -0,0 +1,40 @@
distro_settings:
debian: &DEBIAN
image: "debian:9.5"
user: "debian"
pid1_exe: /lib/systemd/systemd
init: |
sh -c "apt-get -qy update && apt-get -qy install systemd-sysv dbus && exec /sbin/init"
raw_setup: apt-get -qy update && apt-get -qy install dbus python sudo iproute2
raw_setup_done: test -x /usr/bin/sudo
agetty_svc: getty@*
ssh_service: ssh
extra_packages: []
ubuntu:
<<: *DEBIAN
image: "ubuntu:16.04"
user: "ubuntu"
init: |
/sbin/init
centos: &CENTOS
image: "centos:7"
user: "centos"
pid1_exe: /usr/lib/systemd/systemd
init: |
/sbin/init
raw_setup: yum -qy install policycoreutils dbus python sudo iproute iptables
raw_setup_done: test -x /usr/bin/sudo
agetty_svc: getty@* serial-getty@*
ssh_service: sshd
extra_packages: []
fedora:
<<: *CENTOS
image: "fedora:latest"
user: "fedora"
raw_setup: yum -qy install policycoreutils dbus python sudo iproute iptables; mkdir -p /etc/modules-load.d
extra_packages:
- hostname
- procps
- findutils
- kmod
- iputils

15
contrib/dind/hosts Normal file
View File

@ -0,0 +1,15 @@
[local]
# If you created a virtualenv for ansible, you may need to specify running the
# python binary from there instead:
#localhost ansible_connection=local ansible_python_interpreter=/home/user/kubespray/.venv/bin/python
localhost ansible_connection=local
[containers]
kube-node1
kube-node2
kube-node3
kube-node4
kube-node5
[containers:vars]
ansible_connection=docker

View File

@ -0,0 +1,24 @@
# kubespray-dind.yaml: minimal kubespray ansible playbook usable for DIND
# See contrib/dind/README.md
kube_api_anonymous_auth: true
kubeadm_enabled: true
kubelet_fail_swap_on: false
# Docker nodes need to have been created with same "node_distro: debian"
# at contrib/dind/group_vars/all/all.yaml
bootstrap_os: debian
docker_version: latest
docker_storage_options: -s overlay2 --storage-opt overlay2.override_kernel_check=true -g /dind/docker
dns_mode: coredns
kube_network_plugin: weave
deploy_netchecker: True
netcheck_agent_image_repo: quay.io/l23network/k8s-netchecker-agent
netcheck_server_image_repo: quay.io/l23network/k8s-netchecker-server
netcheck_agent_image_tag: v1.0
netcheck_server_image_tag: v1.0

View File

@ -0,0 +1 @@
docker

View File

@ -0,0 +1,70 @@
- name: set_fact distro_setup
set_fact:
distro_setup: "{{ distro_settings[node_distro] }}"
- name: set_fact other distro settings
set_fact:
distro_user: "{{ distro_setup['user'] }}"
distro_ssh_service: "{{ distro_setup['ssh_service'] }}"
distro_extra_packages: "{{ distro_setup['extra_packages'] }}"
- name: Null-ify some linux tools to ease DIND
file:
src: "/bin/true"
dest: "{{item}}"
state: link
force: yes
with_items:
# DIND box may have swap enable, don't bother
- /sbin/swapoff
# /etc/hosts handling would fail on trying to copy file attributes on edit,
# void it by successfully returning nil output
- /usr/bin/lsattr
# disable selinux-isms, sp needed if running on non-Selinux host
- /usr/sbin/semodule
- name: Void installing dpkg docs and man pages on Debian based distros
copy:
content: |
# Delete locales
path-exclude=/usr/share/locale/*
# Delete man pages
path-exclude=/usr/share/man/*
# Delete docs
path-exclude=/usr/share/doc/*
path-include=/usr/share/doc/*/copyright
dest: /etc/dpkg/dpkg.cfg.d/01_nodoc
when:
- ansible_os_family == 'Debian'
- name: Install system packages to better match a full-fledge node
package:
name: "{{ item }}"
state: present
with_items: "{{ distro_extra_packages }} + [ 'rsyslog', 'openssh-server' ]"
- name: Start needed services
service:
name: "{{ item }}"
state: started
with_items:
- rsyslog
- "{{ distro_ssh_service }}"
- name: Create distro user "{{distro_user}}"
user:
name: "{{ distro_user }}"
uid: 1000
#groups: sudo
append: yes
- name: Allow password-less sudo to "{{ distro_user }}"
copy:
content: "{{ distro_user }} ALL=(ALL) NOPASSWD:ALL"
dest: "/etc/sudoers.d/{{ distro_user }}"
- name: Add my pubkey to "{{ distro_user }}" user authorized keys
authorized_key:
user: "{{ distro_user }}"
state: present
key: "{{ lookup('file', lookup('env','HOME') + '/.ssh/id_rsa.pub') }}"

View File

@ -0,0 +1,86 @@
- name: set_fact distro_setup
set_fact:
distro_setup: "{{ distro_settings[node_distro] }}"
- name: set_fact other distro settings
set_fact:
distro_image: "{{ distro_setup['image'] }}"
distro_init: "{{ distro_setup['init'] }}"
distro_pid1_exe: "{{ distro_setup['pid1_exe'] }}"
distro_raw_setup: "{{ distro_setup['raw_setup'] }}"
distro_raw_setup_done: "{{ distro_setup['raw_setup_done'] }}"
distro_agetty_svc: "{{ distro_setup['agetty_svc'] }}"
- name: Create dind node containers from "containers" inventory section
docker_container:
image: "{{ distro_image }}"
name: "{{ item }}"
state: started
hostname: "{{ item }}"
command: "{{ distro_init }}"
#recreate: yes
privileged: true
tmpfs:
- /sys/module/nf_conntrack/parameters
volumes:
- /boot:/boot
- /lib/modules:/lib/modules
- "{{ item }}:/dind/docker"
register: containers
with_items: "{{groups.containers}}"
tags:
- addresses
- name: Gather list of containers IPs
set_fact:
addresses: "{{ containers.results | map(attribute='ansible_facts') | map(attribute='docker_container') | map(attribute='NetworkSettings') | map(attribute='IPAddress') | list }}"
tags:
- addresses
- name: Create inventory_builder helper already set with the list of node containers' IPs
template:
src: inventory_builder.sh.j2
dest: /tmp/kubespray.dind.inventory_builder.sh
mode: 0755
tags:
- addresses
- name: Install needed packages into node containers via raw, need to wait for possible systemd packages to finish installing
raw: |
# agetty processes churn a lot of cpu time failing on inexistent ttys, early STOP them, to rip them in below task
pkill -STOP agetty || true
{{ distro_raw_setup_done }} && echo SKIPPED && exit 0
until [ "$(readlink /proc/1/exe)" = "{{ distro_pid1_exe }}" ] ; do sleep 1; done
{{ distro_raw_setup }}
delegate_to: "{{ item._ansible_item_label }}"
with_items: "{{ containers.results }}"
register: result
changed_when: result.stdout.find("SKIPPED") < 0
- name: Remove gettys from node containers
raw: |
until test -S /var/run/dbus/system_bus_socket; do sleep 1; done
systemctl disable {{ distro_agetty_svc }}
systemctl stop {{ distro_agetty_svc }}
delegate_to: "{{ item._ansible_item_label }}"
with_items: "{{ containers.results }}"
changed_when: false
# Running systemd-machine-id-setup doesn't create a unique id for each node container on Debian,
# handle manually
- name: Re-create unique machine-id (as we may just get what comes in the docker image), needed by some CNIs for mac address seeding (notably weave)
raw: |
echo {{ item | hash('sha1') }} > /etc/machine-id.new
mv -b /etc/machine-id.new /etc/machine-id
cmp /etc/machine-id /etc/machine-id~ || true
systemctl daemon-reload
delegate_to: "{{ item._ansible_item_label }}"
with_items: "{{ containers.results }}"
- name: Early hack image install to adapt for DIND
raw: |
rm -fv /usr/bin/udevadm /usr/sbin/udevadm
delegate_to: "{{ item._ansible_item_label }}"
with_items: "{{ containers.results }}"
register: result
changed_when: result.stdout.find("removed") >= 0

View File

@ -0,0 +1,3 @@
#!/bin/bash
# NOTE: if you change HOST_PREFIX, you also need to edit ./hosts [containers] section
HOST_PREFIX=kube-node python3 contrib/inventory_builder/inventory.py {% for ip in addresses %} {{ ip }} {% endfor %}

View File

@ -0,0 +1,46 @@
#!/bin/bash
# Q&D test'em all: creates full DIND kubespray deploys
# for each distro, verifying it via netchecker.
pass_or_fail() {
local rc="$?"
local msg="$*"
local date="$(date -Isec)"
[ $rc -eq 0 ] && echo "PASS: [$date] $msg" || echo "FAIL: [$date] $msg"
}
test_distro() {
local distro=${1:?}
ansible-playbook -i hosts dind-cluster.yaml --extra-vars node_distro=$distro
pass_or_fail "$distro: dind-nodes"
(cd ../..
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=$distro -i \
${INVENTORY_DIR}/hosts.ini cluster.yml \
--extra-vars @contrib/dind/kubespray-dind.yaml --extra-vars bootstrap_os=$distro
pass_or_fail "$distro: kubespray"
)
docker exec kube-node1 kubectl get pod --all-namespaces
pass_or_fail "$distro: kube-api"
let n=60
while ((n--)); do
docker exec kube-node1 curl -s http://localhost:31081/api/v1/connectivity_check | grep successfully && break
sleep 2
done
[ $n -ge 0 ]
pass_or_fail "$distro: netcheck"
}
# Get all DISTROS from distro.yaml if $* unset (shame no yaml parsing, but nuff anyway)
DISTROS="${*:-$(egrep -o '^ \w+' group_vars/all/distro.yaml|paste -s)}"
NODES="$(egrep ^kube-node hosts|paste -s)"
echo "DISTROS=${DISTROS}"
for distro in ${DISTROS}; do
docker rm -f ${NODES}
{ time test_distro ${distro} ;} |& tee test-${distro}.out
# sleeping for the sake of the human to verify if they want
sleep 2m
done
egrep '^(PASS:|FAIL:|real)' $(ls -tr test-*out)

View File

@ -8,6 +8,7 @@
state: present
create: yes
backup: yes
unsafe_writes: yes
marker: "# Ansible inventory hosts {mark}"
when: populate_inventory_to_hosts_file
@ -18,6 +19,7 @@
line: "{{ loadbalancer_apiserver.address }} {{ apiserver_loadbalancer_domain_name }}"
state: present
backup: yes
unsafe_writes: yes
when:
- loadbalancer_apiserver is defined
- loadbalancer_apiserver.address is defined
@ -51,4 +53,5 @@
regexp: "^{{ item.key }}.*$"
state: present
backup: yes
unsafe_writes: yes
with_dict: "{{ etc_hosts_localhosts_dict_target }}"