Document type

[strugee]

Description of feature

Both the README and https://dokku.com/docs/deployment/zero-downtime-deploys/ state that type may be one of startup, readiness, or liveness, but they don't actually say what this means! It would be nice to have this information available.

[erickedji]

Seems to refer to this, but not sure if applied to local docker outside of the default startup type: https://kubernetes.io/docs/tasks/configure-pod-container/configure-liveness-readiness-startup-probes/

[josegonzalez]

Yeah it refers to that, but the docker-local scheduler only supports the startup type. All are supported for the k3s scheduler.

In the future, the custom proxy plugins might enable some sort of support for the readiness type so that they don't proxy to things that can't be proxied to yet. We might also finally have a proper daemon that does healthchecks for the docker-local scheduler and handles restarts appropriately.

[drbraden]

Is there a way to set the Docker "Test" command for the container when using a non-web process? For example, I have a queue worker that doesn't serve web requests. If I do a docker inspect the "HealthCheck" looks like this:

"Healthcheck": {
     "Test": [
         "CMD",
         "sh",
         "-c",
         "curl --insecure --silent --location --show-error --fail http://localhost:8080$HEALTHCHECK_PATH || exit 1"
     ],
     "Interval": 5000000000,
     "Timeout": 3000000000,
     "Retries": 3
},

So a docker ps will always list the STATUS as something like "Up 8 minutes (unhealthy)".

I just tried adding a "liveness" type to my app.json in the hope that it would affect the container's healthcheck command but it's still set to the curl command test.

[josegonzalez]

This plugin doesn't change any HEALTHCHECK directives in a dockerfile. Also, for docker-local, only the startup type, so adding a liveness or readiness check there wouldn't do anything unless the scheduler allows for it.

[drbraden]

Ah, that was the clue I needed! It's being set in a base image I'm pulling. That explains why none of my other deploys were showing "unhealthy" for non-web processes, as this is my first deploy with this particular base image (the others just rely on the standard buildpacks which don't set healthcheck).

Got it, thank you!