deprecated: with Concourse 5.x the concept changed drastically and there should be no need for this implementation any longer.
WAT
Currently when running concourse-workers in non BOSH environments, like we do with docker-compose, see https://github.com/EugenMayer/concourseci-server-boilerplate
when you shutdown the stack for restart or upgrade, the worker becomes to be in a broken, undefined state, leading to a lot of different issues like
- "file not found" when running a task /job in concourse or during a resource trigger
- job / task is stalled / stuck in the "preparing build" state, not doing anything
The main reasons are, that during the shutdown and the not proper persistence, the worker comes back online with a mixed state, see this issue concourse/concourse#1457 (comment) for more details.
What we do here is simply put, trap the worker and concourse retire-worker before we kill / stop the worker-container
And thats about it
This image always is based on the latest concourse/concourse base image release and will not manipulate anything else then that wrapper
Usage
The docker-image can be found under eugenmayer/concourse-worker-solid
instead of using concourse/concourse as your worker image, use eugenmayer/concourse-worker-solid and either omit the command keyword
in docker-compose or use (default) retire-workeror land-worker. The entrypoint is now change to
worker_wrapper
You can change the default strategy to unregister workers from land-worker to retire-worker by simply
command: retire-worker
All parameters you pass to worker_wrapper will be passed to worker except the first one ( strategy ), so use whatever you like
command: land-worker --special-worker-param1=1 --special-worker-param1=
build it yourself
docker build .t <yourname> will do it - but since we concourse, you can also use this pipeline