Skip to content

[P2] Detect silent watch stalls during control-plane upgrades #13

Open

Description

@brendandburns

Upstream: kubernetes-client#2462
Priority: P2 (reliability)

A long-running job watch with timeout_seconds=0 can stop delivering events during a control-plane upgrade without raising an exception. Controllers may stop making progress without noticing.

Investigate current watch behavior under dropped or stalled connections. Provide a documented recovery path or explicit failure rather than an indefinitely silent stall. Preserve normal idle watches, explicit stop behavior, and server-side timeouts; do not treat lack of events alone as proof of failure.

Acceptance criteria:

  • Reproduce the failure on the current client with a controlled connection/transport test.
  • Add focused regression coverage for recovery/failure signaling and intentional shutdown.
  • Document any new timeout or recovery behavior.

Related upstream work: kubernetes-client#2527. Check its current status and reuse relevant work before implementing.

Activity

  1. linked a pull request that will close this issueDocument recovery from silent watch stalls #32on Nov 19, 2026
  2. brendandburns commented on Nov 19, 2026

    @brendandburns
    OwnerAuthor

    Clarification to the original assignment: this task's designated agent/base branch is agent/p2-issue-13. Raise/target the PR against that branch, not master; the existing draft PR has been retargeted accordingly. Keep the requested commit order: first a test-only commit demonstrating the issue, then a separate fix commit only if reproduced. If it does not reproduce, leave the evidence/test commit and explain why no fix is appropriate.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Labels

bugSomething isn't working

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions