This is a collection of packer templates for a RabbitMQ/Pacemaker VM nodes and docker VM-like containers. It also contains templates to build a Libqb/Corosync/Pacemaker (Linux HA stack) as a lightweight docker apps.
For Packer to Atlas and CI integration examples, see Stefan Scherer's blog post.
The automated build is triggered by a WebHook in GitHub to start a build in CircleCI that triggers a packer build and pushes the produced container image to DockerHub.
See the .circleci/config.yml for details how the glue works.
Builds a docker image for the RabbitMQ RA OCF clustering features testing. The
rabbitmq-cluster-ocf-docker.json builds a Docker VM image either based on a
prebuilt image with Linux HA stack or some other image, with Erlang, RabbitMQ
and some other packages. Setup the base env var as either debian:latest or
optionally use base=bogdando/pcscrm prebuilt=1 to get the required build type,
for example:
$headless=true base=debian:latest packer build -only=docker -color=false \
rabbitmq-cluster-ocf-docker.json
$base=ubuntu:latest prebuilt=yes packer build -only=docker \
rabbitmq-cluster-ocf-docker.json
If you don't have a prebuilt image but want to install only a Corosync/Pacemaker from
packages use pacemaker-cluster-ocf-docker.json.
Builds from trunc the DockerHub images for a Corosync/Pacemaker apps and a runner container containing related HA stack/generic tools like pcs, crm, wget, iptables. Unlike to the templates above creating a "heavies", a VM-like containers, the apps are a lightweight docker containers. The runner is still a "heavy" VM-like though.
Allowed values for the wanted are libqb, corosync, pacemaker, pcscrm.
And for the base: debian:latest (or perhaps anything suchlike), bogdando/libqb,
bogdando/corosync, bogdando/pacemaker, bogdando/pcssrm.
The build chain is supposed to be as the following:
- To build the libqb libs container, use
base=debian:latest wanted=libqb(orbase=bogdando/libqb, if you don't want to go from the clean debian again) - To build the corosync, use
base=bogdando/libqb wanted=corosync - To build the pacemaker container, use
base=bogdando/corosync wanted=pacemaker - And for the pcssrm runner container, use
base=bogdando/pacemaker wanted=pcscrm.
Note, you can try to build the runner from another bases, although it might require additional dependencies to be installed.
F.e. to "rebase" the corosync on top of the changed libqb, run:
$headless=true base=bogdando/libqb wanted=corosync \
repo_path=/home/fuser/gitrepos/ packer build ha-stack-docker.json
And to rebuild the corosync leaving the libqb base as is, use:
$headless=true base=bogdando/corosync wanted=corosync \
repo_path=/home/fuser/gitrepos/ packer build ha-stack-docker.json
Make sure the required source repos (or extracted signed tarballs), which is
ones for a libqb, corosync, pacemaker, crmsh and pcs (named exactly the same way
as listed here) present at a given repo_path.
Also use the rebuild-ha-stack.sh script to rebuild all but the pcscrm.
There are distro base specific shared volumes for docker build templates. For Ubuntu,
those are mounts for /var/cache and /var/lib/apt`. For example, for base=foo,
the volumes ``lib_apt_foo`` and ``cache_foo`` will be used across consequent
packer builds, hopefully making things faster. If something is wrong, just
remove the volumes to be rebuilt from the scratch.
Moved here. It allows to bootstrap a cluster and perform smoke/jepsen tests. See also a generic Vagrantfile for a given Pacemaker OCF RA resource under test.
The Packer template and provision scripts are based on box-cutter/ubuntu-vm. Check out the more up to date version at github.com/boxcutter.
The packer templates and provision scripts in this fork are based on the original repo StefanScherer/packer-atlas-example and on the jakobadam/packer-qemu-templates.