Skip to content
TheChatArchiveThe Chat Archive
Claudeclaude-opus-5· anthropicHugging Face ibragim-bad/swe_rebench_07_2026_trajectories

ISSUE DESCRIPTION Admitted workloads are requeued too early when a worker disconnects An…

  1. User

    ISSUE DESCRIPTION

    Admitted workloads are requeued too early when a worker disconnects

    An admitted workload must remain ready during a temporary worker disconnect and be retried only after the configured worker-loss timeout. Reconnection with the same configuration must restore processing without affecting other connected workers.

    Interface

    No new interfaces are introduced.

    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 using git reset --hard command. Additionally, you can use git stash and git stash pop to temporarily save your changes and restore them later. Finally, you can see the changes that you've made (which will be submitted when you call submit) using git status or git diff commands. However you don't need to use git add or git commit before submitting your changes. In fact, submit only submits currently unstaged changes because it uses git diff to 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-$

  2. Claude Assistant

    Now let me implement the fix.

  3. Claude Assistant

    Now let me add the helper function after reconcileGroup.

  4. Claude Assistant

    Now update the Reconcile path for when no worker cluster is connected at all:

  5. Claude Assistant

    I've completed my analysis and fix. Let me do a final review of the change:

  6. Claude Assistant

    The fix is complete. Here's my checklist review:

    1. If an error was reported in the issue description, I have successfully reproduced it. Yes. The issue is behavioral (no stack trace). I traced the root cause in pkg/controller/admissionchecks/multikueue/workload.go: the "keep the workload Ready while the reserving worker is lost" window was measured from acs.LastTransitionTime, which is the time the MultiKueue admission check transitioned to Ready (admission time) — workload.SetAdmissionCheckState only refreshes that timestamp on a state change, so for any workload admitted longer than workerLostTimeout the window is already expired. I wrote a temporary unit test (built on the existing fake-client harness) that admits a workload on worker2, then disconnects worker2 (client reconnecting / client removed + MultiKueueCluster Active=False one minute ago) and asserts the check stays Ready. Against the unmodified code it failed with want "Ready" got "Retry" (message "Reserving remote lost") for every temporary-disconnect case — i.e., the workload was requeued immediately.

    2. If an error was reported in the issue description, I have confirmed that my fix resolves the error. Yes. With the fix, the same temporary test showed: disconnect 1 min ago → stays Ready, requeued after 4m (5m timeout); reconnecting worker with cluster still Active → stays Ready, requeue after 5m; disconnected longer than the timeout → Retry; all workers disconnected (previously a permanent reconcile error) → Ready while recent, Retry after the timeout; deleted MultiKueueCluster → Retry. I also verified with a temporary test that reconnecting a worker with the same configuration recreates its client, sets Active=True again, and leaves the other connected worker untouched. Both temporary test files were removed before submitting.

    3. I have thought about edge cases and made sure my fix handles them as well. Yes: unknown reserving cluster (status.clusterName empty) and connected-but-missing-remote-workload keep the previous LastTransitionTime semantics (so a genuinely lost remote workload is still retried); the disconnect instant is taken from the persisted MultiKueueCluster Active=False transition time so it survives a manager restart and repeated failed reconnects (the timestamp is not reset by identical condition updates); the short race window between detecting a disconnect and the cluster status update is treated as "just lost" so no premature requeue; a deleted MultiKueueCluster (no client, no object) falls back to the old behavior so quota is eventually released; the new "no connected clusters" path is guarded by !isDeleted, Ready state, quota reservation and not-finished so deletion/cleanup paths keep their error-retry behavior; the clock is now taken from w.clock for testability.

    4. I have run existing tests in the repository that might have been affected by the change I made and confirmed that they pass. Yes: go test ./pkg/... (including pkg/controller/admissionchecks/multikueue and pkg/workload) and go test ./apis/... all pass; go build ./..., go vet ./pkg/controller/admissionchecks/... and go vet ./test/integration/multikueue/... are clean; gofmt reports no issues and no added line exceeds the repo's 200-char golines limit. Integration/e2e suites cannot run here (no envtest/kind assets), but they compile. I did not modify or add any test files.

Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.