Search before asking

• I had searched in the issues and found no similar feature requirement.

Description

It would be nice it there was a ttlSeconds field in RayClusterSpec that would allow the KubeRay controller to automatically cleanup old RayCluster instances.

Use case

I would like to have a staging-like environment where a RayCluster can be shared across multiple data processing steps - for faster startup time.

When a new step runs, it would check whether there is already a RayCluster available to run on. If there isn't, it would create one. However, it's channeling to determine whether a step should delete a RayCluster, because it's unclear whether there are going to be other steps targeting the same cluster launched soon.

Therefore, the only reliable cleanup mechanism can be implemented on the controller side (I mean, I could make a custom external schedule, but that's inferior UX), similarly how RayJobSpec has ttlSecondsAfterFinished.

Related issues

No response

Are you willing to submit a PR?

• Yes I am willing to submit a PR!