We have a small limitation right now due to the way health-checks are performed. You need to configure on which port the container is listening (it can only be internal, doesn’t need to be exposed on the internet), the container will be restarted if the health-check fails.
For workers, this is annoying as they usually don't listen on any port.
The current workaround is to spawn a dummy web server or a
nc
process which will allow the health-checks to succeed.