[Feature] Expose Cluster State via cluster.status and api response

[Jeffwan]

Search before asking

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

Description

Currently, we have not leveraged state field well in our implementation. This is critical because our caller need detail information to let user know the exact stage the cluster is at.

1. Define key stage in cluster CRD and emit events.
2. Reuse those information and expose them to apiserver get/list response

Use case

No response

Related issues

No response

Are you willing to submit a PR?

• Yes I am willing to submit a PR!

[daikeshi]

+1 to this! cc @davidxia

We also need to access cluster status (e.g. ray cluster readiness, head node ip, etc) info in our Ray cluster management lifecycle.

[davidxia]

@Jeffwan happy to contribute something. Just need someone to sketch out an approach or some tips.

[Jeffwan]

/cc @scarlet25151 Can you reach out to @daikeshi and @daikeshi to see whether our internal change meet Spotify's case? I think we forgot to upstream this change.

If there's additional changes to make, we can ask spotify engineers' help.

[daikeshi]

Thanks @Jeffwan! That's awesome. @scarlet25151 please let @davidxia and me know if there's anything we can help with.

[scarlet25151]

Hi @daikeshi and @davidxia , we have designed and implemented some state machine for raycluster lifecycle and expose some status like ports and state. Let's discuss more detail about what we have and what you would like for the raycluster lifecycle and then we can start the work, shall we call a meeting?

[daikeshi]

Thanks @scarlet25151. Look forward to the collaboration! Let's chat more on the slack.

[asm582]

Can we please know the different states that we are planning to expose?

[scarlet25151]

@asm582 sure for now we will implement the state, the design would look like in the doc: kuberay state design