You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@seatunnel.apache.org by "zhangshenghang (via GitHub)" <gi...@apache.org> on 2026/07/15 14:13:44 UTC

[PR] [Fix][Zeta] Reuse fixed slots after master failover [seatunnel]

zhangshenghang opened a new pull request, #11458:
URL: https://github.com/apache/seatunnel/pull/11458

   ## What does this PR do?
   
   Fixes #11437. During active-master failover, a restored JobMaster now reuses the matching active worker slots recorded for the job instead of requesting them again. This lets fixed-slot jobs recover while workers still retain their task groups.
   
   ## Why is this needed?
   
   The replacement master previously treated retained slots as unavailable and left the restored job pending. If only some slots can be reused, cleanup releases only newly allocated slots and preserves the active retained slots for the next retry.
   
   ## Tests
   
   - Added a fixed-slot master-failover regression test that exhausts all free slots before restore.
   - `./mvnw -pl seatunnel-engine/seatunnel-engine-server -DskipITs -Dtest=JobMasterMasterFailoverResourceTest test`
   - `./mvnw -q -DskipTests verify`


-- 
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: commits-unsubscribe@seatunnel.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org


Re: [PR] [Fix][Zeta] Reuse fixed slots after master failover [seatunnel]

Posted by "DanielLeens (via GitHub)" <gi...@apache.org>.
DanielLeens commented on PR #11458:
URL: https://github.com/apache/seatunnel/pull/11458#issuecomment-5555871099

   CI status (maintainer triage, not a review of the change): the Build on 65009e5c6 finished with a single failure, `rocketmq-connector-it (11)`: `testSinkRocketMqMessageTag` on two engine legs hit the RocketMQ broker registering under two identities after the runtime rename in `RocketMqContainer` (fixed by #12115, green, pending merge). Unrelated to this PR. Retriggered CI with an empty commit (4c539e33); no code change.


-- 
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: commits-unsubscribe@seatunnel.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org


Re: [PR] [Fix][Zeta] Reuse fixed slots after master failover [seatunnel]

Posted by "DanielLeens (via GitHub)" <gi...@apache.org>.
DanielLeens commented on PR #11458:
URL: https://github.com/apache/seatunnel/pull/11458#issuecomment-5552467530

   CI retriggered via an empty commit (`214d066`, pushed by davidzollo). The previous `Build` failed only on `doris-connector-it` (Maven Central transfer reset) and `jdbc-connectors-it-part-7` (`MetalakeIT` Gravitino-readiness timeout — source fix in #12108). Neither touches this PR's diff.


-- 
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: commits-unsubscribe@seatunnel.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org


Re: [PR] [Fix][Zeta] Reuse fixed slots after master failover [seatunnel]

Posted by "davidzollo (via GitHub)" <gi...@apache.org>.
davidzollo commented on PR #11458:
URL: https://github.com/apache/seatunnel/pull/11458#issuecomment-5551474008

   Opened #12108 to stabilize the `MetalakeIT` timeout that's been causing the red `jdbc-connectors-it-part-7` job here (3-minute Gravitino-readiness budget bumped to 5 minutes — the download+start sequence for the Gravitino server can occasionally exceed 3 minutes on a loaded runner, producing curl exit code 7 rather than a real functional failure). Test-only change, no production code touched.


-- 
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: commits-unsubscribe@seatunnel.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org


Re: [PR] [Fix][Zeta] Reuse fixed slots after master failover [seatunnel]

Posted by "DanielLeens (via GitHub)" <gi...@apache.org>.
DanielLeens commented on PR #11458:
URL: https://github.com/apache/seatunnel/pull/11458#issuecomment-5571674719

   CI status (maintainer triage): the Build on `6f47af6ed` finished with two failures, both `rocketmq-connector-it` (JDK 8 and 11): `RocketMqIT` - the known broker double-identity race (#12115, unmerged). Zero overlap with this PR's diff (`JobMaster.java`, `AbstractResourceManager.java`, `ResourceManager.java`, `SlotProfile.java`, `DefaultSlotService.java`). No open code-level blockers; standing `APPROVED` holds. Retriggered CI with an empty commit (`60b368dfa`); no code change.
   


-- 
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: commits-unsubscribe@seatunnel.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org


Re: [PR] [Fix][Zeta] Reuse fixed slots after master failover [seatunnel]

Posted by "nzw921rx (via GitHub)" <gi...@apache.org>.
nzw921rx commented on PR #11458:
URL: https://github.com/apache/seatunnel/pull/11458#issuecomment-5602083922

   Follow-up on F5 after testing the current head (`e97540d9a7`) locally.
   
   I added a temporary focused test for the mixed partial-reuse failure path:
   
   - one persisted slot remains active and is reused;
   - capacity permits exactly one fresh allocation, then the remaining requests fail;
   - `preApplyResources()` returns `false`;
   - the reused slot remains active;
   - the successfully allocated fresh slot is released;
   - `masterFailoverRestore` remains `true` for the retry.
   
   The focused test passed (`1` test, `0` failures/errors). I did not run the full `verify` lifecycle.
   
   The current production cleanup behavior is correct, so this is a test-coverage request only. Please add an equivalent regression case to `JobMasterMasterFailoverResourceTest`. Once this coverage is in the PR, I am ready to +1.


-- 
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: commits-unsubscribe@seatunnel.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org


Re: [PR] [Fix][Zeta] Reuse fixed slots after master failover [seatunnel]

Posted by "DanielLeens (via GitHub)" <gi...@apache.org>.
DanielLeens commented on PR #11458:
URL: https://github.com/apache/seatunnel/pull/11458#issuecomment-5575458697

   CI status (maintainer triage): the Build on `60b368dfa` finished with 2 failures, both confirmed unrelated to this PR's diff (`JobMaster.java`, `AbstractResourceManager.java`, `ResourceManager.java`, `SlotProfile.java`, `DefaultSlotService.java`):
   
   - `doris-connector-it (8)`: `DorisSchemaChangeIT.testDorisWithSchemaEvolutionCase` `ConditionTimeout` - the same Doris-container flake now confirmed across three unrelated PRs today (#11503, #11077, and this one), none of which touch Doris schema-change code paths the same way.
   - `rocketmq-connector-it (8)`: the known broker "No topic route info in name server" race (#12115, unmerged).
   
   No open code-level blockers; standing DanielLeens `APPROVED` continues to hold. Retriggering once more.
   


-- 
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: commits-unsubscribe@seatunnel.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org


Re: [PR] [Fix][Zeta] Reuse fixed slots after master failover [seatunnel]

Posted by "davidzollo (via GitHub)" <gi...@apache.org>.
davidzollo commented on PR #11458:
URL: https://github.com/apache/seatunnel/pull/11458#issuecomment-5509598218

   Re-reviewed the outstanding open points from SEZ9's 2026-08-31 review and nzw921rx's 2026-08-27 review against the current head (68640a2e4).
   
   Already resolved by earlier commits on this branch, verified against current source:
   - SEZ9 Issue 1 (High/Blocking) - `masterFailoverRestore` not cleared on the subPlan path: fixed by 43cd0c44e; the flag is now cleared inside the shared `enoughResource` success branch that both the whole-job and per-pipeline paths funnel through, so it is unconditional on `isSubPlan`.
   - nzw921rx's [P2][Blocking] same-job slot reassignment: fixed by f7bf00cbc; `DefaultSlotService#requestSlot` now assigns a fresh random UUID as the allocation sequence on every successful assignment instead of reusing the constant worker-level sequence, so a slot reassigned to another task group of the same job no longer satisfies a stale persisted mapping. A dedicated regression test (`testRestoreRejectsSlotReassignedWithinSameJob`) covers this in `JobMasterMasterFailoverResourceTest`.
   
   Verified as non-issues (no code change needed):
   - SEZ9 Issue 3, second half: `ownerJobID` is a primitive `long` on both sides of the `==` comparison in `AbstractResourceManager#slotActiveCheck` (not boxed `Long`), so the comparison is numeric equality, not reference equality - safe as written.
   - SEZ9 Issue 3, first half: `slotActiveCheck` currently has exactly one production caller (`JobMaster#getReusableSlot`), so there is no other existing caller whose contract the added owner-job-ID check could silently tighten.
   
   Fixed in 04a6b4612 (pushed just now):
   - SEZ9 Issue 4 (Medium): switched the `reusedSlotProfiles` guard set in `JobMaster#preApplyResources` from an `IdentityHashMap`-backed set to value-based membership, since `SlotProfile#equals` already keys on worker+slotID+sequence and every allocation now gets a fresh sequence token (per the f7bf00cbc fix above). This removes the dependency on the exact same object instance flowing from pre-apply through cleanup.
   - SEZ9 Issue 8 (Low): `preApplyResourcesForAll` is now `void`; it only mutates the map passed in and its return value was already discarded at the only call site.
   - SEZ9 Issue 6 (Low): completed the `getReusableSlot` Javadoc with `@param`/`@return` tags and documented the owner-job-ID condition enforced by `slotActiveCheck`.
   - SEZ9 Issue 7 (Low): corrected the EN/ZH resource-management docs - slot reuse after an active-master failover does not check `dynamic-slot`; fixed slots are simply the case that benefits most because re-requesting one the Worker still holds would otherwise fail.
   
   Left open for follow-up (not fixed in this pass):
   - SEZ9 Issue 2 (Medium): making slot reuse a worker-side atomic reclaim (validate+re-pin on the Worker's SlotService) instead of a read-only heartbeat-registry check is a protocol-level change beyond the scope of a review-comment fix pass; recommend tracking separately.
   - SEZ9 Issue 5 (Low): no test currently drives the partial-reuse failure-cleanup path (mix of reused and freshly-allocated slots, or an ownership-changed slot, with insufficient free capacity). Left for the PR author/a follow-up, since authoring a new failover scenario test without the ability to run it locally in this pass risked introducing an unverified/flaky case into this financial-scale code path.
   
   No inline review threads exist on this PR (both reviews are top-level review-body comments with no line-level comments), so there is nothing to resolve via the thread API.
   
   CI: a fresh Build run was triggered by the push above; previous state was `pending` before this push.
   


-- 
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: commits-unsubscribe@seatunnel.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org


Re: [PR] [Fix][Zeta] Reuse fixed slots after master failover [seatunnel]

Posted by "DanielLeens (via GitHub)" <gi...@apache.org>.
DanielLeens commented on PR #11458:
URL: https://github.com/apache/seatunnel/pull/11458#issuecomment-5437712751

   Thanks for digging into this, @nzw921rx — this is a real gap and it's on me for not catching it in my August 25 review.
   
   I re-traced the specific path you're describing against the current head (`bbaf984dc`) and can independently confirm it:
   
   - `getCoordinatorVertexList()`/`getPhysicalVertexList()` (`SubPlan.java:609-614`) are plain unfiltered getters — they return every vertex in the pipeline regardless of status. So during a whole-job `preApplyResources(null)` (the only call issued for active-master-failover restore, via `pendingJobSchedule` -> `CoordinatorService.java:422`), `getReusableSlot` is invoked for an already-FINISHED task group's `TaskGroupLocation` exactly the same as for a still-running one.
   - `releaseTaskGroupResource` (`JobMaster.java:941-983`) releases the resource for a finished task group but never removes its entry from `ownedSlotProfilesIMap` — it only records the release in the in-memory `releasedSlotWhenTaskGroupFinished` map, which guards `releasePipelineResource` against double-releasing it, but only within the same `JobMaster` instance's lifetime (`JobMaster.java:993-995`).
   - After an active-master failover, the restored `JobMaster` is a brand-new object with an empty `releasedSlotWhenTaskGroupFinished` (`JobMaster.java:231`), so that guard is gone, while the persisted `ownedSlotProfilesIMap` entry for the finished task group survives (it's only ever fully wiped by `releasePipelineResource` for the whole pipeline, `JobMaster.java:1011`).
   - `slotActiveCheck` (`AbstractResourceManager.java:277-296`) and the Worker-side `DefaultSlotService#releaseSlot` (`DefaultSlotService.java:203-231`) both key off worker address + slot ID + sequence + owner job ID only — none of which encode task-group identity. So if that freed physical slot gets handed to a different task group of the same job before the failover, both checks pass even though the persisted mapping now points at the wrong task group.
   
   One nuance I'd add to the framing: `releaseTaskGroupResource`, `releasePipelineResource`, and `ownedSlotProfilesIMap`'s lifecycle are all pre-existing code this PR doesn't touch — the staleness itself isn't new. But this PR is what makes the scenario reachable in practice: before this fix, a fixed-slot job under an active-master failover would just stay deadlocked in PENDING (#11437) once slots were exhausted, so the whole-job restore never succeeded, and a pipeline in that state could never progress far enough afterward to reach the corrupting `releasePipelineResource` call. `getReusableSlot` is exactly the mechanism that now lets that restore succeed via a same-job/different-task-group coincidence, so I agree this needs to be closed in this PR rather than filed as a separate pre-existing-defect ticket — it sits squarely inside the normal target scenario (fixed slots + same job + active-master failover), not an obscure corner case.
   
   I'm revising my August 25 "Ready to merge" conclusion: this is a blocker. `getReusableSlot`'s reuse decision needs to validate the exact retained allocation, not just worker/slot liveness and job ownership — a generation/token minted on each successful `requestSlot` and persisted/compared during recovery, or a direct `TaskGroupLocation` check on the Worker side, both sound like reasonable directions. I'd also fold in a regression test for the release-path half of this (the `releasePipelineResource` double-release from your second focused test), not just the reuse-acceptance half, since both need to hold for the fix to be complete.
   
   @zhangshenghang — there's no new commit yet, so I'm not re-posting a full review record for this round, but this is now the primary blocker to resolve. Happy to do a full re-review once a fix is pushed.
   


-- 
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: commits-unsubscribe@seatunnel.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org


Re: [PR] [Fix][Zeta] Reuse fixed slots after master failover [seatunnel]

Posted by "JeremyXin (via GitHub)" <gi...@apache.org>.
JeremyXin commented on PR #11458:
URL: https://github.com/apache/seatunnel/pull/11458#issuecomment-5433426303

   Reviewed the latest head `bbaf984dc`.
   
   The current version fixes the two main correctness risks I was looking for during master failover recovery:
   
     1. reused fixed slots are only accepted when the worker still reports the same slot identity and the same owner job;
     2. partial pre-allocation failure no longer releases retained slots that should survive for the next retry.
   
   ### Merge Conclusion: Approve to merge
   
   A few minor, non-blocking suggestions below — feel free to address in a follow-up if you prefer:
   
   ### Minor Suggestions
   
   #### 1. Add a comment explaining the `IdentityHashMap` choice
   
   **Location:** `JobMaster.java` — `preApplyResources()`, `reusedSlotProfiles` declaration
   
   Since `SlotProfile.equals()` compares `worker`+`slotID`+`sequence` but excludes `ownerJobID`, identity comparison is needed to distinguish reused slots from newly allocated ones. A brief inline comment would prevent someone from accidentally replacing it with a `HashSet` later.
   
   #### 2. Add a JobMaster-level log when a slot is reused
   
   **Location:** `JobMaster.java` — `getReusableSlot()`
   
   Currently only `AbstractResourceManager.slotActiveCheck()` logs at the ResourceManager level. An INFO log in `getReusableSlot()` when returning a non-null slot (with jobId, taskGroupLocation, slotProfile) would improve operational visibility during failover recovery.
   
   #### 3. Extract shared coordinator/task vertex logic
   
   **Location:** `JobMaster.java` — `preApplyResourcesForSubPlan()`
   
   The coordinator and task vertex loops contain near-identical reuse/apply logic (~15 lines each). Could be extracted into a small helper like `applyOrReuseForVertex(...)` to reduce duplication.
   
   #### 4. Clarify `masterFailoverRestore` reset scope
   
   **Location:** `JobMaster.java` — `preApplyResources()`, the `!isSubPlan` success branch
   
   The flag is only reset on the full-plan success path, not the SubPlan path. I believe this is safe since SubPlan-level applies follow after the full-plan apply has already reset it, but a one-line comment noting this is intentional would help future readers.
   
   ---
   
   Overall, LGTM 👍


-- 
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: commits-unsubscribe@seatunnel.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org


Re: [PR] [Fix][Zeta] Reuse fixed slots after master failover [seatunnel]

Posted by "SEZ9 (via GitHub)" <gi...@apache.org>.
SEZ9 commented on PR #11458:
URL: https://github.com/apache/seatunnel/pull/11458#issuecomment-5595222571

   Thanks @DanielLeens for the re-verification pass.
   
   Based on your description, `e97540d9a7` is a merge of the current `dev` tip `438f9ccfbef` into `5730017c82`, and the diff restricted to this PR's changed files is empty, so there is nothing new to review on the merits at this head. I am fine treating it as a no-op checkpoint, with your `5730017c82` review as the reference point.
   
   Two requests:
   
   1. Your comment appears to be cut off mid-sentence in section 1.4 ("Carried-over, non-blocking items from my la…"). Could you repost the rest so the carried-over list is captured here rather than inferred?
   
   2. Could you give a one-line status (still open / addressed / withdrawn) for each of the previously raised findings? Since this head does not touch the PR's own code, I assume they are unchanged from `5730017c82`, but please confirm or correct:
      - F1 (HIGH): `masterFailoverRestore` only cleared on the whole-job pre-apply success branch, not on the subPlan/pipeline path. You mention correcting an inaccuracy in your earlier notes about where the flag is cleared — does that change your view on whether F1 is still open?
      - F2 (MEDIUM): check-then-use of heartbeat-eventually-consistent registry state with no worker-side reclaim.
      - F3 (MEDIUM): `slotActiveCheck` ownerJobID tightening changing the contract for existing callers and relying on `==`.
      - F4 (MEDIUM): identity-based `reusedSlotProfiles` set fragility on the failure/cleanup path.
      - F5 (LOW): missing partial-reuse failure cleanup and ownership-changed fallback test coverage.
      - F6 (LOW): `getReusableSlot` Javadoc missing parameter/return tags and the owner-job-ID condition.
      - F7 (LOW): docs describe reuse as fixed-slot-specific while `getReusableSlot` applies regardless of slot mode.
      - F8 (LOW): `preApplyResourcesForAll` returning its mutated input map with the return value ignored at the call site.
   
   F1 in particular I'd like to see either resolved or explicitly argued down before this moves forward; the rest can stay as non-blocking follow-ups if that matches your carried-over list.
   
   <!-- streview-comment:920 -->


-- 
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: commits-unsubscribe@seatunnel.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org


Re: [PR] [Fix][Zeta] Reuse fixed slots after master failover [seatunnel]

Posted by "nzw921rx (via GitHub)" <gi...@apache.org>.
nzw921rx merged PR #11458:
URL: https://github.com/apache/seatunnel/pull/11458


-- 
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: commits-unsubscribe@seatunnel.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org


Re: [PR] [Fix][Zeta] Reuse fixed slots after master failover [seatunnel]

Posted by "DanielLeens (via GitHub)" <gi...@apache.org>.
DanielLeens commented on PR #11458:
URL: https://github.com/apache/seatunnel/pull/11458#issuecomment-5601168454

   Thanks @SEZ9.
   
   **On the "cut off" text**: it wasn't actually truncated in the underlying comment — I re-pulled the raw review body via the API just now and the full sentence is there ("Carried-over, non-blocking items from my last full review: - Issue 2 ... - Issue 3 ..."). That was most likely GitHub's "show more" collapse on a long review body, not a real cut-off. Reposting the full carried-over list here anyway so it's captured in this thread directly, as requested:
   
   - **Issue 2 (= your F2, Medium, tracked, non-blocking)**: check-then-use gap between `slotActiveCheck` and actual deploy — no Worker-side atomic claim/pin step.
   - **Issue 3 (= your F5, Low, non-blocking)**: no test drives a mixed partial-reuse-then-release-of-non-reused-slots outcome in the cleanup branch.
   
   **Status of every F-item**, re-verified just now directly against the current head's real source (`e97540d9a7`, which is byte-identical to `5730017c82` on every file this PR owns — I fetched the PR head into a local worktree and read the files, not relying on memory of prior rounds):
   
   - **F1 (you rated HIGH)** — Not open. This was my own mischaracterization in the 09-02/09-06 review rounds, corrected in the 09-08 review and re-confirmed just now by reading the method body again: `masterFailoverRestore = false;` sits inside the single `if (enoughResource) {...}` block in `JobMaster.preApplyResources`, which both the whole-job and subPlan paths fall into on success — it is not gated by `isSubPlan` at all, so it fires identically either way. There's an inline comment right above it now ("Retained slots are valid only for the first successful allocation after failover..."), and `testSubPlanRestoreClearsFailoverSlotReuseFlag` directly asserts this by calling `preApplyResources(subPlan)` and checking the flag clears. So: closed, false alarm on my part, not a live concern.
   - **F2** — Still open, Medium, non-blocking (carried-over Issue 2 above). Accepted as an out-of-scope structural follow-up (Worker-side atomic reclaim), not something this PR needs to solve to merge.
   - **F3** — Not open, confirmed false alarm against current source. `SlotProfile.ownerJobID` is `private long` (`SlotProfile.java:37`), so the `==` in `slotActiveCheck` is a primitive comparison, not a boxed-reference hazard. And I grepped every call site of `slotActiveCheck` in the current head: `JobMaster.getReusableSlot` is the only production caller; every other caller is test code (`JobMasterMasterFailoverResourceTest`, `ResourceManagerTest`). So there's no pre-existing production caller whose contract got silently tightened.
   - **F4** — Fixed. `reusedSlotProfiles` is now `Set<SlotProfile> reusedSlotProfiles = new HashSet<>()` (plain value equality via `SlotProfile#equals` on worker+slotID+sequence), not the old `IdentityHashMap`-backed set — verified directly in the current head's `JobMaster.java`, with an inline comment explaining why value-based membership is now safe.
   - **F5** — Still open, Low, non-blocking (carried-over Issue 3 above).
   - **F6** — Fixed. `getReusableSlot`'s Javadoc now carries `@param taskGroupLocation`/`@return` and explicitly states the slot is reused only if "the slot's owner job ID still matches this job" — verified in the current head.
   - **F7** — Fixed. Both EN/ZH `resource-management.md` now read "this check does not look at the `dynamic-slot` setting" — verified in the current head.
   - **F8** — Fixed. `preApplyResourcesForAll` is now `private void` (mutates its map argument in place, no return value to ignore) — verified in the current head.
   
   Net: at this head, 2 non-blocking follow-ups remain open (F2 and F5, both previously agreed as out-of-scope-or-fast-follow, not blockers), and everything else on your list is either fixed in a landed commit or a confirmed false alarm (including F1, which is on me). This doesn't change my Ready-to-merge-after-this-head's-CI-is-green conclusion from the 00:46 UTC review — I'm not asserting CI is green as observed fact there, just that nothing on the source side is blocking.


-- 
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: commits-unsubscribe@seatunnel.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org


Re: [PR] [Fix][Zeta] Reuse fixed slots after master failover [seatunnel]

Posted by "davidzollo (via GitHub)" <gi...@apache.org>.
davidzollo commented on PR #11458:
URL: https://github.com/apache/seatunnel/pull/11458#issuecomment-5476826317

   I traced the current-head call sites before changing this behavior. The active-master restore scheduler enters `CoordinatorService.pendingJobSchedule`, which invokes `jobMaster.preApplyResources()` without a `SubPlan`. On successful whole-job pre-apply, the current code stores the futures and clears the `volatile masterFailoverRestore` flag. The only `preApplyResources(this)` caller is `SubPlan`'s later pipeline-recovery path, reached after that initial scheduling step.
   
   Therefore the reported permanent-reuse window does not occur on the current control flow. The cleanup collection is already identity-based, so retained profiles are excluded without depending on value equality. I found no safe source change to make for this comment; the current-head Build is green.


-- 
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: commits-unsubscribe@seatunnel.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org


Re: [PR] [Fix][Zeta] Reuse fixed slots after master failover [seatunnel]

Posted by "DanielLeens (via GitHub)" <gi...@apache.org>.
DanielLeens commented on PR #11458:
URL: https://github.com/apache/seatunnel/pull/11458#issuecomment-5557010563

   Cross-link: the `doris-connector-it (8)` failures on this PR's recent runs (Doris container `Timed out waiting for container port to open` after the 60 s default, and `DorisIT` reading an empty / partial sink table right after the job exited on the Spark legs) are addressed at the source by #12142 (test-only: six-minute startup budget for the all-in-one image, bounded visibility wait for the 2pc sink checks). Nothing to do on this PR; it will be retriggered once its current Build finishes.


-- 
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: commits-unsubscribe@seatunnel.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org


Re: [PR] [Fix][Zeta] Reuse fixed slots after master failover [seatunnel]

Posted by "nielifeng (via GitHub)" <gi...@apache.org>.
nielifeng commented on PR #11458:
URL: https://github.com/apache/seatunnel/pull/11458#issuecomment-5128163765

   Additional baseline evidence for the acceptance matrix: #11437 remains reproducible on upstream `dev` commit `735cf430`, which does not contain this PR, using a 3-Master / 3-Worker cluster with 24 fixed slots and one streaming job occupying 16 slots.
   
   After the Active Master Java process was terminated, the replacement Master was elected and all six Hazelcast members rejoined safely, but the job stayed `PENDING/RESTORE` for 294 seconds. Checkpoints stopped progressing, and cancellation left the same 16 slots assigned; only 8 of 24 slots were free until the Workers were restarted. The logs repeatedly reported `NoEnoughResourceException` and `Pre resource application failed`.
   
   This adds a 3+3, single high-occupancy workload to the existing 3+2 and concurrent-workload reproductions. It is baseline reproduction evidence, not a test of PR head `81f8dce434404cb32c7fe557df69c88d6fd9bef5`. The same topology and workload would be useful as an acceptance case when this PR is updated and ready for validation.


-- 
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: commits-unsubscribe@seatunnel.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org


Re: [PR] [Fix][Zeta] Reuse fixed slots after master failover [seatunnel]

Posted by "DanielLeens (via GitHub)" <gi...@apache.org>.
DanielLeens commented on PR #11458:
URL: https://github.com/apache/seatunnel/pull/11458#issuecomment-5477025558

   Thanks @SEZ9 for continuing to track this, and @davidzollo for the rebuttal — since this is Zeta's master-failover/slot-reuse path, I traced the full call graph myself before taking a side rather than deferring to either comment. No new commit landed since my last re-verification (`ebebc0b085bd`), so this is a reply, not a new review.
   
   **Issue 1 (F1, marked Blocking): I don't think this is reachable, and I've traced the exact reason why.** The claim is that the `isSubPlan` branch of `preApplyResources` never clears `masterFailoverRestore`, so it stays armed for the JobMaster's whole lifetime and can fire on an ordinary later pipeline restart. Structurally that line is correct — `masterFailoverRestore = false;` only sits in the `!isSubPlan` success branch (`JobMaster.java:601`). But the call graph guarantees that branch always runs, and succeeds, before the `isSubPlan` branch can ever be reached for the same job:
   
   ```
   CoordinatorService.pendingJobSchedule (only preApplyResources() caller with subPlan=null)
     -> jobMaster.preApplyResources()                         [isSubPlan = false]
       -> on enoughResource: physicalPlan.setPreApplyResourceFutures(futures)
                             masterFailoverRestore = false      (JobMaster.java:601)
       -> only THEN does jobMaster.run() let pipelines progress CREATED -> SCHEDULED
   SubPlan.stateProcess(), case SCHEDULED
     -> ResourceUtils.applyResourceForPipeline(jobMaster, this)
       -> reads jobMaster.getPhysicalPlan().getPreApplyResourceFutures()   (the SAME map
          populated by the whole-job call above) -> DEPLOYING -> RUNNING
   SubPlan.stateProcess(), case FAILED/CANCELED (later, ordinary pipeline restart)
     -> checkNeedRestore(state) && prepareRestorePipeline()
       -> jobMaster.preApplyResources(this)                    [isSubPlan = true, SubPlan.java:741]
   ```
   
   A pipeline can only reach the `FAILED`/`CANCELED` restart branch that calls `preApplyResources(this)` after it has already been through `DEPLOYING`/`RUNNING` at least once — and that first `DEPLOYING` transition (`case SCHEDULED`) only happens by consuming futures that `applyResourceForPipeline` reads out of `physicalPlan.getPreApplyResourceFutures()`, which is only non-empty once the whole-job `preApplyResources()` call has already succeeded and, in the same success branch, cleared `masterFailoverRestore`. So by the time any `isSubPlan=true` call is reachable, the flag is already `false`, and `getReusableSlot()`'s `if (!masterFailoverRestore) { return null; }` gate has already closed. I don't see a path where a fresh JobMaster's pipeline reaches a restart-eligible state without first going through the whole-job call. @davidzollo's read matches what I traced independently.
   
   I'd still take the low-risk hardening for free, though: clearing the flag in the `isSubPlan` success branch too (or once the job reaches RUNNING) costs nothing and removes the dependency on this ordering guarantee holding forever as the scheduler code evolves. I'd suggest keeping that as a Low-severity, non-blocking follow-up rather than a merge blocker — it's defense-in-depth, not a live bug on this head.
   
   **Issues 2-8:** no new information in this round beyond what was already in the previous pass — I'll leave those as the Medium/Low non-blocking follow-ups they already were rather than re-litigating each one here, since nothing about them changed with this reply.
   
   With Issue 1 resolved as non-reachable, my merge recommendation from `5063079162` stands: **ready to merge**, CI green, no blockers. As before, my repository permission is comment-only for final sign-off, so a write-capable maintainer should do the actual approve/merge.


-- 
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: commits-unsubscribe@seatunnel.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org


Re: [PR] [Fix][Zeta] Reuse fixed slots after master failover [seatunnel]

Posted by "DanielLeens (via GitHub)" <gi...@apache.org>.
DanielLeens commented on PR #11458:
URL: https://github.com/apache/seatunnel/pull/11458#issuecomment-5558121341

   CI status (maintainer triage, not a review of the change): the Build on 4c539e330 finished with seven failures, none in this PR's diff: `CouchbaseIT` (#12127), `KuduIT` 90-minute hang (#12132), `PostgresCDCIT` committed-offset restore (#11864), `TestFilterRowKindIT` Flink leg (#12116), RocketMQ broker double identity (fixed by #12115), Doris container port timeout + `DorisIT` reading the sink before the 2pc publish (fixed by #12142), and `PaimonSinkCDCIT#testSinkWithIncompatibleSchema` leaving `st-multi-table-sink-writer-*` threads alive after an expectedly failing job (engine-side, first sighting, watching it). Retriggered CI with an empty commit (02839c50); no code change.


-- 
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: commits-unsubscribe@seatunnel.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org


Re: [PR] [Fix][Zeta] Reuse fixed slots after master failover [seatunnel]

Posted by "nielifeng (via GitHub)" <gi...@apache.org>.
nielifeng commented on PR #11458:
URL: https://github.com/apache/seatunnel/pull/11458#issuecomment-4989049507

   Reviewed the diff and traced the call paths on `dev`. The approach looks solid to me:
   
   - `getReusableSlot` doesn't blindly trust the persisted `ownedSlotProfilesIMap` entry — it confirms with `resourceManager.slotActiveCheck(slotProfile)` that the worker still holds the matching slot sequence before reusing. That's the right guard against a stale slot record after failover.
   - The cleanup filter `!reusedSlotProfiles.contains(slotProfile)` correctly preserves reused slots and only releases newly-applied ones. I checked `SlotProfile.equals/hashCode` (slotID + worker + sequence) — identity is stable, so `contains` is reliable here.
   - The regression test genuinely exercises the fix: it exhausts all free fixed slots (`assertEquals(0, getUnassignedSlots)`) before restore, so the old code would leave the job pending and only the reuse path lets it recover. `dynamicSlot=false` + fixed `slotNum` targets the right scenario.
   
   I also double-checked the multi-pipeline concern: master failover recovery goes through `CoordinatorService#pendingJobSchedule` → `preApplyResources()` (no-arg, all pipelines in one call), and `masterFailoverRestore` is only cleared after that whole call succeeds — so a JobMaster-level flag covers all pipelines correctly. The per-SubPlan `preApplyResources(this)` path is normal-operation pipeline restart where reuse is intentionally skipped. So no issue there.
   
   One minor suggestion (non-blocking): `masterFailoverRestore` is a plain `boolean`. It's written in `init()` and in `preApplyResources`, and read from the per-SubPlan path which runs on `executorService` threads. For consistency with the adjacent `volatile boolean needRestore`, it might be worth marking it `volatile` to avoid any visibility gap. Low priority given the coordinator-thread write happens well before pipeline restarts.
   
   Nice fix overall.


-- 
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: commits-unsubscribe@seatunnel.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org


Re: [PR] [Fix][Zeta] Reuse fixed slots after master failover [seatunnel]

Posted by "nielifeng (via GitHub)" <gi...@apache.org>.
nielifeng commented on PR #11458:
URL: https://github.com/apache/seatunnel/pull/11458#issuecomment-4987598580

   I validated this PR commit (`46d2f3909f7e97f8f516c8fba68d12df471d7718`) against the fixed-slot reproduction from #11437.
   
   Environment: 3 Masters, 2 Workers, 16 fixed slots, `WAIT` scheduling. Workload: FakeSource -> Console, parallelism 10, 20,000 rows/split, 11 assigned / 5 free slots. Package SHA-256: `9bd831336fc547aed1b53adda4af636a3ec1771cfdab4c81b9689c70e7d73c91`.
   
   Three consecutive Active Master Java-process failures passed:
   - election: 1 s, 1 s, 2 s
   - job visible as `RUNNING`: 5 s in every round
   - source and sink QPS >= 90% of the 87.9k rows/s pre-fault median: 5 s in every round
   - slots: exactly 11 assigned / 5 free in every post-fault sample
   - checkpoints: 2 before fault injection -> 19 completed after round 3, 0 failed
   
   After all Masters rejoined, cancelling the job returned all resources in 2 seconds: 16/16 slots free, no running/pending jobs, and empty Worker `runningJobIds`. Neither Worker was restarted.
   
   I found no `RESOURCE_NOT_ENOUGH`, `NoEnoughResourceException`, duplicate `TaskGroupLocation`, or stale-slot message. Only transient Hazelcast member-left/connection errors occurred at the injected failures.
   
   For this topology and workload, the PR fixes both the failover restore and cancellation cleanup symptoms reported in #11437.


-- 
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: commits-unsubscribe@seatunnel.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org


Re: [PR] [Fix][Zeta] Reuse fixed slots after master failover [seatunnel]

Posted by "SEZ9 (via GitHub)" <gi...@apache.org>.
SEZ9 commented on PR #11458:
URL: https://github.com/apache/seatunnel/pull/11458#issuecomment-5564294625

   Thanks @DanielLeens for the detailed disclosure on the `02839c50d9b6` head. I agree with your reading: `4c539e330` and `02839c50d` are empty retrigger commits and `git diff 65009e5c60..02839c50d9b6 --stat` is empty, so there is nothing new to re-derive on the source side.
   
   That said, an unchanged diff also means the previously raised points are still open. Before this can move forward I need an explicit response (fix or rationale) on each:
   
   - **PR11458-F1 (HIGH, Logic)** — `masterFailoverRestore` is only cleared on the whole-job pre-apply success branch in `JobMaster.java`; the subPlan/pipeline path never clears it, so slot reuse can fire on ordinary pipeline restarts long after failover. This is the blocker; please clear the flag on the pipeline path too, or explain why that path is unreachable while the flag is set.
   - **PR11458-F2 (MEDIUM, Robustness)** — slot reuse is a check-then-use of heartbeat-eventually-consistent registry state with no worker-side reclaim. How do you see the window right after failover (slots genuinely retained but not yet reported), and the dead or mid-reassignment worker case?
   - **PR11458-F3 (MEDIUM, Compatibility)** — the added `ownerJobID` equality in `slotActiveCheck` (`AbstractResourceManager.java`) changes the contract for existing callers and relies on `==`. Please use a proper equality comparison and confirm existing callers are unaffected.
   - **PR11458-F4 (MEDIUM, Bug)** — the identity-based `reusedSlotProfiles` set is fragile: if the reused `SlotProfile` reference is replaced between reuse and cleanup, the failure path releases a worker-retained slot and reintroduces #11437. Keying on a stable identifier would remove this.
   - **PR11458-F5 (LOW, Test)** — `JobMasterMasterFailoverResourceTest.java` covers only the happy full-reuse path. Please add the partial-reuse failure cleanup case (reused slots survive a failed pre-apply) and the ownership-changed fallback at JobMaster level.
   - **PR11458-F6 / F7 (LOW, Docs)** — the `getReusableSlot` Javadoc is missing param and return tags and the owner-job-ID condition; `docs/en/architecture/engine/resource-management.md` should not describe reuse as specific to `dynamic-slot: false` when `getReusableSlot` applies to any retained slot.
   - **PR11458-F8 (LOW, Style)** — `preApplyResourcesForAll` returns its mutated input map and the return value is ignored at the only call site; either use it or drop the return value.
   
   A single follow-up commit covering F1, F3 and F4 plus the F5 tests, with the doc/style items folded in, would be ideal. If you disagree with any of these, a short note per finding is fine — I just need the status stated explicitly rather than inferred from an unchanged diff.
   
   <!-- streview-comment:855 -->


-- 
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: commits-unsubscribe@seatunnel.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org


Re: [PR] [Fix][Zeta] Reuse fixed slots after master failover [seatunnel]

Posted by "DanielLeens (via GitHub)" <gi...@apache.org>.
DanielLeens commented on PR #11458:
URL: https://github.com/apache/seatunnel/pull/11458#issuecomment-5548368028

   ## CI follow-up on the current head (`a71c68c426f8`)
   
   My last review (2026-09-04T10:38:12Z) noted 2 jobs still in progress on the fork's Build run (`33854690642`) and said I'd expect to re-affirm `APPROVED` once they completed. They have completed now, but the overall `Build` check is red — not because of those two, though: it's 3 different jobs, and I traced each one to root cause rather than assuming the obvious "flaky E2E" label:
   
   1. **`doris-connector-it (8, ubuntu-latest)` — failed.** Root cause: `Could not transfer artifact com.google.protobuf:protobuf-java:jar:3.19.4 from/to central ... Connection reset`. Pure Maven Central network flake during dependency resolution, before any test even ran. Unrelated to this PR.
   2. **`jdbc-connectors-it-part-7 (8, ubuntu-latest)` and `(11, ubuntu-latest)` — both failed**, same root cause on both JDK versions: `org.apache.seatunnel.connectors.seatunnel.jdbc.MetalakeIT.startUp` — `org.awaitility.core.ConditionTimeoutException: expected: <0> but was: <7> within 3 minutes`. This is a pre-existing readiness-timing issue in the Metalake catalog IT test itself, not something this PR's diff touches.
   3. `kudu-connector-it (8/11)` show as cancelled, which is just the workflow's fail-fast cancellation once the above jobs went red — not an independent failure.
   
   I confirmed none of these three job's underlying modules appear in this PR's file list at all (`git diff --name-only` against this head: `docs/*`, `connector-rocketmq/*` (source + the two E2E test files already discussed in prior rounds), `seatunnel-engine-server/*` main+test). Nothing under `connector-doris-e2e` or `connector-jdbc-e2e-part-7` is touched. I also checked whether `dev` already carries a fix for the `MetalakeIT` timeout that this branch just hasn't synced yet — it doesn't; this looks like a standing flaky/slow test on `dev` itself, unrelated to this PR, and not something I'm going to paper over with a blind timeout bump without someone actually root-causing why the catalog isn't converging in 3 minutes.
   
   **Conclusion: this Build failure is not a blocker for this PR.** Both failing modules are outside this PR's diff and the failures reproduce for reasons (network transfer reset, pre-existing catalog-readiness flakiness) that have nothing to do with fixed-slot failover reuse. My substantive review verdict from Sept 2 (`APPROVED`, mechanism sound) still stands. @zhangshenghang, a CI re-run should clear both — no code change needed on your side for this. The one item still keeping this PR from merge is the same one from every recent round: `mergeStateStatus: BLOCKED` / `reviewDecision: REVIEW_REQUIRED` needs an approval from an account with write access to `apache/seatunnel`, since mine doesn't count toward that gate.
   


-- 
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: commits-unsubscribe@seatunnel.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org


Re: [PR] [Fix][Zeta] Reuse fixed slots after master failover [seatunnel]

Posted by "DanielLeens (via GitHub)" <gi...@apache.org>.
DanielLeens commented on PR #11458:
URL: https://github.com/apache/seatunnel/pull/11458#issuecomment-5129181207

   Thanks @nielifeng, this extra baseline reproduction is very helpful. I read it as additional evidence for the original fixed-slot failover problem, not as validation of the current PR head.
   
   No new code was pushed after Daniel's previous review, so I am not reopening a full code review in this pass. When the PR is updated, the 3-Master / 3-Worker / 24-slot case you described would be a good acceptance scenario to use alongside the existing coverage.


-- 
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: commits-unsubscribe@seatunnel.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org


Re: [PR] [Fix][Zeta] Reuse fixed slots after master failover [seatunnel]

Posted by "DanielLeens (via GitHub)" <gi...@apache.org>.
DanielLeens commented on PR #11458:
URL: https://github.com/apache/seatunnel/pull/11458#issuecomment-5564511944

   CI status (maintainer triage): the Build on `02839c50d` finished with five failures, none in this PR's diff (`JobMaster.java`, `AbstractResourceManager.java`, `ResourceManager.java`, `SlotProfile.java`, `DefaultSlotService.java`):
   
   - `transform-v2-it-part-1`: `TestFilterRowKindIT` - the `AssertSinkWriter` static-counter issue this branch already carries (dev merge doesn't help; root cause refiled as #12116, still open).
   - `doris-connector-it`: `DorisIT.testDoris` - known, fix exists as #12142, not yet merged.
   - `rocketmq-connector-it`: `RocketMqIT`, 43/87 tests - same broker double-identity race as #12115, not yet merged, wider blast radius than the single-test hit seen elsewhere.
   - `elasticsearch-connector-it`: `ElasticsearchIT.testElasticsearchWithPIT` - new, not previously catalogued; PR touches zero ES files.
   - `connector-file-local-it`: `LocalFileWithMultipleTableIT` (parquet) - Testcontainers image-pull `ContainerFetchException`; PR touches zero local-file files.
   
   No open code-level blockers on this head; standing DanielLeens `APPROVED` reaffirmed through today with zero source delta since 2026-08-29. Retriggered CI with an empty commit (`6f47af6ed`); no code change.
   


-- 
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: commits-unsubscribe@seatunnel.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org


Re: [PR] [Fix][Zeta] Reuse fixed slots after master failover [seatunnel]

Posted by "DanielLeens (via GitHub)" <gi...@apache.org>.
DanielLeens commented on PR #11458:
URL: https://github.com/apache/seatunnel/pull/11458#issuecomment-5554201679

   CI status (maintainer triage, not a review of the change): the Build on 214d06656 finished with four failures, all unrelated to this PR and already tracked upstream:
   
   - `transform-v2-it-part-1 (11)`: `TestFilterRowKindIT#testFilterRowKindMultiTable` on the Flink 1.18 leg, static JVM-wide counters in `AssertSinkWriter` (#12116).
   - `kudu-connector-it (11)`: `KuduIT` hang on the Flink 1.15.3 leg after the JobManager stopped responding; cancelled at the 90-minute limit (issue being filed).
   - `doris-connector-it (11)`: `DorisSchemaChangeIT`, MODIFY COLUMN cancelled the open stream load and the pipeline restart failed with "schema-change-after checkpoint is already completed" (issue being filed).
   - `all-connectors-it-7 (11)`: `PostgresCDCIT` committed-offset restore (#11847 / #11864).
   
   Retriggered CI with an empty commit (65009e5c) so the run reflects the current dev state; no code change.


-- 
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: commits-unsubscribe@seatunnel.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org


Re: [PR] [Fix][Zeta] Reuse fixed slots after master failover [seatunnel]

Posted by "DanielLeens (via GitHub)" <gi...@apache.org>.
DanielLeens commented on PR #11458:
URL: https://github.com/apache/seatunnel/pull/11458#issuecomment-5549703430

   The `rocketmq-connector-it` `MQClientException: CODE: 17 ... No topic route info in name server for the topic: test_topic_source` failure seen on this branch's earlier runs is a known Testcontainers startup-timing race (RocketMQ name-server route metadata not yet propagated when the test starts producing/consuming), unrelated to this PR's own diff.
   
   It's already fixed on `dev` via #12050 ([Fix][Zeta] Stabilize CI test isolation, merged 2026-09-04, commit bbfc188e6), which adds a bounded `waitForTopicRoute()` Awaitility poll in `RocketMqIT` before producing/asserting against `test_topic_source` / `test_topic` / `test_text_topic` / the restore-test topics, plus a related offset-sync timeout bump (30s -> 60s) to match the queue-count increase in the same PR.
   
   This branch's current head (`a71c68c42`) already contains that commit, and the [2026-09-04 CI run](https://github.com/apache/seatunnel/runs/100966104650) confirms both `rocketmq-connector-it` legs (JDK 8 and 11) passing. The `Build` failure on that same run comes from unrelated `doris-connector-it` and `jdbc-connectors-it-part-7` jobs.
   
   No action needed here for the RocketMQ flake specifically — flagging for visibility in case an older/unsynced run shows it again.


-- 
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: commits-unsubscribe@seatunnel.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org


Re: [PR] [Fix][Zeta] Reuse fixed slots after master failover [seatunnel]

Posted by "DanielLeens (via GitHub)" <gi...@apache.org>.
DanielLeens commented on PR #11458:
URL: https://github.com/apache/seatunnel/pull/11458#issuecomment-5552965041

   The recurring `transform-v2-it-part-1` red (`TestFilterRowKindIT.testFilterRowKindMultiTable`, Flink legs only) is now root-caused: `AssertSinkWriter` keeps its row counters in static JVM-wide maps and evaluates MIN_ROW/MAX_ROW in each writer's own `close()`, which is only safe under Zeta's per-job classloader isolation — on Flink the per-table writers/subtasks share one TaskManager JVM, so counts bleed (125/150) or are read early (75/97). Already-merged #11995 doesn't cover it. Filed as #12116 (a `src/main` connector-assert fix, not a test change). This job's failure is unrelated to this PR's diff; treating it as a known recurring flake for retrigger purposes until #12116 is fixed.


-- 
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: commits-unsubscribe@seatunnel.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org