DanielLeens commented on issue #11437: URL: https://github.com/apache/seatunnel/issues/11437#issuecomment-4986944413
Thanks for the very complete failover report. I checked the fixed-slot restore path around resource pre-allocation, and this looks like a real Zeta bug rather than a usage or config issue. After active-master failover, the replacement master is still trying to pre-apply a fresh set of slots for the restored task groups, while the workers are still holding the original assigned slots from the running job. In fixed-slot `WAIT` mode, that leaves restore stuck with 5 free slots visible, 11 retained slots still occupied, and no way to move the job back to `RUNNING`. That also matches the second symptom you reported: cancel removes the job from the queue, but the worker-side slot ownership is still left behind. So this is different from #10568. That earlier path was about duplicate task-group deployment; this one is happening earlier, at slot reuse and resource recovery during restore. There is already an in-flight fix in PR #11458 to reuse the retained worker slots during master failover instead of requesting them again. Once that PR lands in `dev`, the right verification is the same sequence you described: kill only the active Master, confirm the job returns from `PENDING/RESTORE` to `RUNNING`, and confirm cancel returns all fixed slots without restarting Workers. Keeping this open as a valid bug. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
