[KubeRay] "Gang Scheduling with RayJob and Kueue" tutorial only works without workers

[fg91]

Description

The tutorial Gang Scheduling with RayJob and Kueue shows how to integrate with Kueue and GKE's "dynamic workload scheduling" feature to achieve gang scheduling.

However, the tutorial should mention the constraint that this only works if the RayJob only has a head but no workers. Otherwise one gets this error:

 dmissionChecks:
  - lastTransitionTime: "2025-10-17T13:14:25Z"
    message: |-
      Error creating ProvisioningRequest "rayjob-rayjob-sleep-test-062a5-dws-prov-1": admission webhook "warden-validating.common-webhooks.networking.gke.io" denied the request: GKE Warden rejected the request because it violates one or more constraints.
      Violations details: {"[denied by provisioning-request-cr-validation]":["the queued provisioning feature currently supports only single PodSet per request"]}

Only having a single Pod for the head node alleviates the need for gang scheduling so this is unfortunately not really useful yet.

Link

No response

[Future-Outlier]

HI, @fg91 long time no see!!!
which kueue version are you using?

[fg91]

Hi @Future-Outlier, nice to see you here :)

I'm currently on v0.14.1.

I discovered the following option:

apiVersion: kueue.x-k8s.io/v1beta1
kind: ProvisioningRequestConfig
metadata:
  name: dws-config
spec:
  provisioningClassName: queued-provisioning.gke.io
  podSetMergePolicy: IdenticalWorkloadSchedulingRequirements # Add this
  managedResources:
    - nvidia.com/gpu

When adding the podSetMergePolicy and when using the same resources, affinities, tolerations, ... for both the head and worker pod templates, kueue merges them into a single pod set, and this GKE limitation can be circumvented.

[Future-Outlier]

Do you mind open a doc PR to fix this?
I can help review, thank you!

[fg91]

@Future-Outlier could you please review? #59070
Thank you!