[Feature] Allow setting the RayClusterName in the job specification to have determinitic service head

[mmourafiq]

Search before asking

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

Description

Currently in order to expose the service to the end user, we need to look for the service name which contains a random number.

For instance of the user has already a unique RayJob: job-22f1d24aff9a4c7fbe5ab9707decbcb0 the cluster name would something like this job-22f1d24aff9a4c7fbe5ab9707decbcb0-raycluster-7ct2d and subsequently the service head would be 22ff9a4c7fbe5ab9707decbcb0-raycluster-7ct2d-head-svc.

It would be nice to allow the user to set the value and only generate them if not specified.

Use case

I understand the logic to prevent crashing the creation of the cluster, unfortunately this make using the operator very hard.

I think the specification should allow setting the RayClusterName similar to the JobId to allow building on top of this operator.

Note: I already tried to set the name of the service and head via metadata.name unfortunately some env vars are auto generated and the ray params and using the env var to resolve the service name.

Related issues

No response

Are you willing to submit a PR?

• Yes I am willing to submit a PR!

[kevin85421]

cc @architkulkarni

[architkulkarni]

Hi @mmourafiq , thanks for raising this issue. I think adding a parameter for RayClusterName sounds reasonable. Would you be willing to submit a PR for this? Let me know if you need any pointers to relevant places in the code.

[mmourafiq]

@architkulkarni Happy to make a PR.
I will start with changes that I can test on a local cluster and then we can discuss in case I miss anything after opening a PR draft.

[architkulkarni]

@mmourafiq that sounds great, thanks for the contribution!