Can I add healthcheck to the services dependencies?

[unkcpz]

In our docker image, we start the main service (oneshot) based on the rabbitmq (longrun) (https://github.com/aiidateam/aiida-core/blob/fa28ebfc1a6dbd25a86de2ab2d38ea36794362db/.docker/aiida-core-with-services/s6-assets/s6-rc.d/aiida-prepare/dependencies.d/rabbitmq)

We start the RMQ service (type longrun) with:

#!/command/execlineb -P

with-contenv

foreground { s6-echo "start RMQ server" }
rabbitmq-server

But we realised the RabbitMQ service start to fast that it is not ready for our service to detect it. We workaround it by adding a sleep in the startup script.

Is s6-overlay support to add a healthcheck to the dependencies services only do the following services launching after healthcheck is passing as docker compose?

We also have a docker-compose file to start the rabbitmq as another independent container, so we can do healthcheck and then start the main container. It 'll be great to have similar thing with s6-overlay I assume?

[skarnet]

Your problem is readiness notification. Here is how s6 handles it. Please read this page before anything else.

If RabbitMQ can support direct readiness notification via a command-line option, then that's what you should use.

If it cannot, then you should put a call to your healthchecker in a script named check in a data subdirectory of your service directory (remember to make it executable!) and add a call to s6-notifyoncheck right before rabbitmq-server in your run script. If the service takes a particularly long time to start, you may want to adjust various options to s6-notifyoncheck so that it doesn't report failure before the service has had the chance to become ready.

[unkcpz]

Thanks! Sounds it is the tool I need. I'll give it a try. So I'll close the issue for the moment, will reopen if more questions coming up.

[unkcpz]

Hi @skarnet, I did apply a pool healthchecker and tested works on Linux with both docker and podman but now with docker on windows with WSL. The problem in WSL docker looks like the pool healthcheck is running but block the following service to start. Do you have any clue on what could be wrong?

[skarnet]

I answered directly on the linked issue. If it's specific to a given environment, especially a weird one such as WSL, then I don't think the problem is related to s6-overlay; it's likely something in the environment that makes the readiness reporting fail. We'll see.