The GitHub Actions job "Flink Kubernetes Operator CI" on 
flink-kubernetes-operator.git/fix/FLINK-41074-bluegreen-readiness has succeeded.
Run started by GitHub user luca-p-castelli (triggered by gyfora).

Head commit for run:
921a76c95c800284706d4bfd912105a752ecd6e5 / Luca Castelli 
<[email protected]>
[FLINK-41074][operator] Fence blue/green readiness to the candidate revision

Editing the parent FlinkBlueGreenDeployment while the incoming (candidate)
child is in its deletion delay patches that child in place. The parent then
kept using readiness evidence from the candidate's previous revision:

1. isFlinkDeploymentReady() ignored status.observedGeneration, so a
   STABLE/RUNNING status reported for generation N authorized retiring the
   source while generation N+1 was still restarting.
2. deploymentReadyTimestamp was never reset when the candidate was patched,
   so the new revision inherited the remainder of the previous delay, or
   none at all if it had already expired.

Fix:
- Add isCandidateReady(): the candidate must not be terminating and must
  have metadata.generation == status.observedGeneration on top of the
  existing STABLE/RUNNING check. The source keeps the unchanged
  isFlinkDeploymentReady() check.
- Reset deploymentReadyTimestamp when the candidate is patched during a
  transition and whenever it is observed not ready, so an edited or
  recovered candidate only retires the source after a full deletion delay
  measured from when its current revision is observed ready.

No CRD or API changes.

Generated-by: OpenAI gpt-6.1-sol via Pi 1.2.0; Claude claude-fable-5-1 via Pi 
1.3.0

Report URL: 
https://github.com/apache/flink-kubernetes-operator/actions/runs/39781035042

With regards,
GitHub Actions via GitBox

Reply via email to