DanielLeens commented on issue #12985: URL: https://github.com/apache/seatunnel/issues/12985#issuecomment-6415641686
Thanks for separating scheduling policy from #12119's queued-job lifetime problem, and thanks @JeremyXin for offering to coordinate. This is a useful STIP, but I would keep it **Under Review** until two compatibility decisions are resolved. First, the draft says existing configurations remain FIFO-equivalent, yet it also changes an impossible request under today's `WAIT` from indefinitely `PENDING` to terminal `FAILED`. That is observable behavior for existing jobs. Please decide whether prompt `UNSCHEDULABLE` failure is opt-in with the new policy or is an explicitly documented incompatible change, and test the default accordingly. Second, define what makes capacity *permanently* infeasible. Fixed slots are registered per worker, but the eligible worker set or tags can change. A snapshot of current capacity cannot by itself prove that a request can never fit after membership changes. Specify the evaluation epoch, what triggers re-evaluation, and the behavior when no eligible workers are currently registered. Keep the final all-or-nothing allocation attempt authoritative, as the draft already says. For ownership, #12119 should settle the durable acceptance record and normal submission lifecycle first; open #12201 already owns the master-recovery queue-order/lazy-init slice. A bounded first implementation could then address feasibility classification and diagnostics before backfill, without duplicating either lane. Please take these two policy choices and the dependency order to `[email protected]` for community review, then update this STIP before implementation is claimed ready. I added the `improve` and `Zeta` labels; no implementation owner is assigned while the design remains open. -- 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]
