Open fabriziosestito opened 2 years ago
@fabriziosestito Could you elaborate more? I don't understand what's the real meaning of the ticket
Kubernetes needs to know if the pod is ready and alive.
Right now we hacked around it by tricking k8s thinking it is alive using /usr/bin/true/
as the probes (since it returns always true the pod is considered to be alive and ready).
This is obviously an hack, for the trento-web we already have this since we provide the health api.
Refer to this for more info: https://kubernetes.io/docs/tasks/configure-pod-container/configure-liveness-readiness-startup-probes/
Actually, Fabrizio, I removed the hack and disabled the probes altogether because it was causing the runner container some problems at startup.
This fixed the probe in the distroless containers, but do you think we still need to implement those two?
I'm not sure how would we make probing meaningful to assess the status of the runner. It's just a background worker process, so I guess for now we just assume that if it's up, it's okay. We could instrument it properly, for sure, but I'm afraid that's a bit of a stretch goal right now!
As a comment, we will remove ARA sooner than later, so I wouldn't invest much time implementing something rock-solid. We want just some solution that makes the current Demo environment usable.
We need to add liveness and readiness probe to the trento-runner, since right now we use
/usr/bin/true
as an hack.