[Feature] Support `runtimeClassName` in values.yaml

[unclebenel]

Search before asking

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

Description

What problem does this solve? KubeRay clusters that run with containerd + GPU runtimes (e.g. k3s/rke + nvidia) must set spec.template.spec.runtimeClassName: nvidia on the worker Pods.

With the current chart (¹) there is no supported way to pass that field. the template only renders an explicit whitelist of PodSpec attributes, so any additional keys placed under

worker: template: spec:

are silently discarded.

The only workaround today is to patch the RayCluster after Helm deploy: kubectl patch raycluster --type=json
-p='[ {"op":"add", "path":"/spec/workerGroupSpecs/0/template/spec/runtimeClassName", "value":"nvidia"} ]'

Use case

worker: template: spec: runtimeClassName: nvidia

Related issues

No response

Are you willing to submit a PR?

• [x] Yes I am willing to submit a PR!

[Narwhal-fish]

Hi @Future-Outlier, is this issue still valid? I'd like to give it a shot.

[Future-Outlier]

Hi @Future-Outlier, is this issue still valid? I'd like to give it a shot.

go for it

[Narwhal-fish]

Hi @Future-Outlier, is this issue still valid? I'd like to give it a shot.

go for it

Thanks! I'll take a look into it.

[Narwhal-fish]

I’m planning to modify helm-chart/ray-cluster/templates/raycluster-cluster.yaml to include something like: {{- with .Values.worker.runtimeClassName }} runtimeClassName: {{ . }} {{- end }}

However, I noticed there’s also another repository kuberay-helm. Should I submit the PR there, or in this main repository?

[win5923]

Hi @Narwhal-fish,

You just need to open a PR in this repo, the release manager will handle syncing it to the kuberay-helm repo during the release process.

[Narwhal-fish]

Hi @Narwhal-fish,

You just need to open a PR in this repo, the release manager will handle syncing it to the kuberay-helm repo during the release process.

Got it!