[feature] <RuntimeClass> Ability to specify spec.runtimeClassName inside Pod pipeline

[namirinz]

Feature Area

In my GPU cluster to make a K8s pod able to use GPU is to specify spec.runtimeClassName: nvidia .
(https://kubernetes.io/docs/concepts/containers/runtime-class/#usage)

What feature would you like to see?

Ability to add RuntimeClass Name into the component or task.

Ex.

@container_component
def train_from_csv():
    return dsl.ContainerSpec(
        image="",
        command=["python", "train.py"],
        args=[],
    )

@pipeline(name="complex_pipeline")
def complex_pipeline(kernel: str):
    model = train_from_csv()
    
    ##### HERE ####
    set_runtimeclass_name(task=model, runtimeclass="nvidia")

What is the use case or pain point?

Able to have a component that able to use GPU (ex. Train Model)

Is there a workaround currently?

---

Love this idea? Give it a 👍.

[namirinz]

#8052

[namirinz]

/assign @chensun @rimolive

[aybchan]

We also ran into this issue with Kubeflow Pipelines on k3s. You can implement a mutating admission controller to apply on-the-fly custom modifications (such as adding a runtimeClassName field) to resource creation requests

[github-actions]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

[rimolive]

/lifecycle frozen