ISSUE DESCRIPTION Standalone elastic RayCluster self-updates raycluster-generation…
- User
ISSUE DESCRIPTION
Standalone elastic RayCluster self-updates raycluster-generation annotation What happened:
A standalone Kueue-managed elastic RayCluster is continuously updated by Kueue.
The
kueue.x-k8s.io/raycluster-generationannotation tracks the RayCluster's ownmetadata.generation, but remains one value behind. For example:metadata.generation: 273390 kueue.x-k8s.io/raycluster-generation: "273389"A later read showed the same pattern:
metadata.generation: 273462 kueue.x-k8s.io/raycluster-generation: "273461"Kueue logs showed repeated RayCluster reconciles followed by RayCluster update webhook calls from the Kueue controller manager.
This is a self-triggered loop in the RayCluster custom annotation path. For standalone RayClusters,
GetCustomAnnotationscallsGetWorkloadslicingRayClusterCustomAnnotations(ctx, c, j.Object(), podSets, j.Name). That helper fetches the same RayCluster and returns the fetched RayCluster generation as an annotation to write back onto the RayCluster. The metadata patch then advances the RayCluster generation, making the annotation stale again.What you expected to happen:
Kueue should not continuously patch a standalone elastic RayCluster when its replica counts have not changed.
The
raycluster-generationannotation is useful for RayJob and RayService, where the top-level object needs to account for an underlying RayCluster generation. A standalone RayCluster should not write its own generation back onto itself.How to reproduce it (as minimally and precisely as possible):
- Install Kueue v0.16.5 with
ElasticJobsViaWorkloadSlices=true. - Enable the RayCluster integration.
- Create a standalone RayCluster with:
metadata: annotations: kueue.x-k8s.io/elastic-job: "true" labels: kueue.x-k8s.io/queue-name: <local-queue>- Inspect the RayCluster generation and annotation repeatedly:
kubectl get raycluster.ray.io <name> -n <namespace> \ -o jsonpath='{.metadata.generation}{"\t"}{.metadata.annotations.kueue\.x-k8s\.io/raycluster-generation}{"\n"}'Anything else we need to know?:
This may be related to #10160, which added generation-based elastic workload names on the release-0.16 branch.
A fix is to omit
kueue.x-k8s.io/raycluster-generationwhen thejobObjectis the same standalone RayCluster asrayClusterName.Environment:
- Kubernetes version (use
kubectl version):- Client Version: v1.33.9
- Server Version: v1.33.10-gke.1115000
- Kueue version (use
git describe --tags --dirty --always):- Deployed controller image:
kueue:v0.16.5 - Helm chart:
kueue-0.16.5
- Deployed controller image:
- Cloud provider or hardware configuration: GKE
- OS (e.g:
cat /etc/os-release): not collected - Kernel (e.g.
uname -a): not collected - Install tools: Helm
- Others:
ElasticJobsViaWorkloadSlices=trueTopologyAwareScheduling=false
ADDITIONAL ADVICE
Since you are given a git repository, you can use git commands to simplify your work. For example, if you made a mistake and want to revert your changes, you can use
git checkout <file>to restore the file to its original state. You can also reset all changes in the repository usinggit reset --hardcommand. Additionally, you can usegit stashandgit stash popto temporarily save your changes and restore them later. Finally, you can see the changes that you've made (which will be submitted when you callsubmit) usinggit statusorgit diffcommands. However you don't need to usegit addorgit commitbefore submitting your changes. In fact,submitonly submits currently unstaged changes because it usesgit diffto get the changes that need to be submitted.CHECKLIST
Before submitting your solution, please go over the following checklist and make sure you've done everything:
- If an error was reported in the issue description, I have successfully reproduced it.
- If an error was reported in the issue description, I have confirmed that my fix resolves the error.
- I have thought about edge cases and made sure my fix handles them as well.
- I have run existing tests in the repository that might have been affected by the change I made and confirmed that they pass. I want you to list every bullet from this checklist and write your reasoning for why you think you did it or didn't need to.
Repository has been uploaded and your shell is currently at the repository root. Time to solve the issue!
(Current directory: /kueue, current file: none) bash-$
- Install Kueue v0.16.5 with
- Assistant
I’ll locate the Ray integration and annotation helper, inspect existing tests and call sites, then reproduce the problematic standalone behavior at unit-test level before making the smallest targeted source change. Afterward I’ll run the relevant test packages and review the diff.
- Assistant
rgis unavailable in the environment, so I’ll use recursivegrepto find the implementation and tests. - Assistant
The shared helper is used by RayJob and RayService; I’ll inspect its identity/generation logic and the standalone RayCluster adapter, plus the current table tests, to determine whether the guard belongs centrally or in the standalone call site.
- Assistant
The helper currently always fetches and returns the generation. I’ll inspect the standalone controller’s
GetCustomAnnotationsimplementation and the full test table setup so the fix preserves RayJob/RayService behavior and handles exact object identity correctly. - Assistant
On this checkout, standalone
RayClusterdoes not yet implementJobWithCustomAnnotations, while the issue describes that call path. I’ll inspect interface usage and recent history/branches around elastic workload naming to understand whether the intended fix should be a helper-level same-object guard (which also future-proofs standalone callers) rather than removing a call that is absent here. - Assistant
The safest condition is not name equality alone: a RayJob can share a name with its underlying cluster. The helper should recognize a canonical
*rayv1.RayClusterwhose namespace/name matchesrayClusterName, return only replica-size metadata, and avoid fetching its own generation. Existing ConfigMap-based tests already guard against accidentally treating all same-named objects as standalone clusters.Before editing, I’ll run a temporary external reproduction program (outside the repository, so no tests are added or modified) that invokes the helper with a standalone RayCluster and confirms the current unwanted generation annotation.
- Assistant
The first reproduction attempt used the wrong feature helper import path. I’ll read the source imports and correct the temporary program.
- Assistant
I’ll correct the import and enable the feature gate in the reproduction, matching the reported configuration.
- Assistant
Go requires an explicit cache path in this stripped environment; I’ll rerun with a temporary cache.
- Assistant
The reproduction confirms generation
42is returned for the object itself. I’ll inspect merge semantics before editing: if omission merely leaves an already-present stale annotation, that still stops the loop, but it determines whether explicit deletion is needed or consistent with the requested “omit” behavior. - Assistant
The first excerpt missed the function body; I’ll open the exact lines.
- Assistant
Omitted keys are not deleted by
mergeAnnotations; importantly, they also cannot cause further patches. Thus the targeted behavior is to return the replica-size annotation but not the self-generation key. I’ll add the standalone identity guard before the clientGet, preserving generation tracking and error behavior for RayJob/RayService and avoiding the self-read entirely. - Assistant
The same reproduction now returns only replica sizes, confirming the self-generation annotation is omitted. Next I’ll run the RayCluster package tests, then RayJob and RayService tests because they rely on the helper and must continue receiving underlying-cluster generation annotations.
- Assistant
All directly affected packages pass. Edge-case coverage is also preserved by existing tests: a non-RayCluster top-level object with the same name still receives generation metadata, missing underlying clusters still return the established empty generation value, and client errors still propagate. I’ll now run formatting/whitespace checks and inspect the final diff and repository status for unintended changes.
Other models, same or similar prompt
Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.
Identical first prompt
Qwen/Qwen3.6-27BCompare side by side →
Qwen/Qwen3.6-27BCompare side by side →
Qwen/Qwen3.6-27BCompare side by side →
Qwen/Qwen3.6-27BCompare side by side →
Qwen/Qwen3.6-27BCompare side by side →
- xiaomi/mimo-v2.5-proCompare side by side →
- xiaomi/mimo-v2.5-proCompare side by side →
- xiaomi/mimo-v2.5-proCompare side by side →
- xiaomi/mimo-v2.5-proCompare side by side →
- xiaomi/mimo-v2.5-proCompare side by side →
- gpt-5.6-solCompare side by side →