MartijnVisser opened a new pull request, #29038:
URL: https://github.com/apache/flink/pull/29038
## What is the purpose of the change
Fixes FLINK-40505: `SourceCoordinator` never removed a subtask's last
reported watermark from the watermark-alignment aggregation. After a subtask
failure and reset, the stale entry kept constraining `maxAllowedWatermark` for
the entire alignment group until the restarted subtask happened to report again
— and indefinitely if it never did (e.g. after downscale). The retained value
is not a conservative bound either: a restarted attempt can resume from an
earlier checkpoint position than its last report.
## Brief change log
- Added `WatermarkAggregator#remove(key)`, returning the new aggregated
watermark iff the aggregate changed.
- `SourceCoordinator#subtaskReset` now removes the failed subtask's
watermark and, when the per-source aggregate changed, propagates it to the
group-level aggregator in the coordinator store via the same update path used
for reported watermarks (extracted as `updateAggregatedWatermarkOfGroup`).
- Removal is done in `subtaskReset` (per-subtask, fires only when no
execution attempt is alive) rather than per-attempt `executionAttemptFailed`:
watermark alignment is mutually exclusive with concurrent execution attempts
(enforced in the constructor and in `handleReportedWatermark`), and global
failover recreates the coordinator. Subtasks pick up the updated
`maxAllowedWatermark` via the existing periodic `announceCombinedWatermark`.
Known follow-up (out of scope here): the group-level aggregator in the
coordinator store never removes an `operatorName` entry, so a permanently
stopped source in a multi-source alignment group is still never pruned.
## Verifying this change
This change added tests and can be verified as follows:
- New test
`SourceCoordinatorAlignmentTest#testWatermarkAlignmentStatePrunedAfterSubtaskFailureAndReset`
fails without the fix (announced `maxAllowedWatermark` stayed 1100 from the
stale entry; expected 1200 after the constraining subtask fails and resets) and
passes with it.
- New characterization test
`testSubtaskThatNeverReportedReceivesNoAlignmentEvents` documents that only
subtasks that have reported receive alignment events.
- Regression: `SourceCoordinatorAlignmentTest`, `SourceCoordinatorTest`, and
`SourceCoordinatorContextTest` all pass locally.
Both new tests are additive; no existing tests were modified.
## Does this pull request potentially affect one of the following parts:
- Dependencies (does it add or upgrade a dependency): no
- The public API, i.e., is any changed class annotated with
`@Public(Evolving)`: no
- The serializers: no
- The runtime per-record code paths (performance sensitive): no
(coordinator-side, failover-time only)
- Anything that affects deployment or recovery: JobManager (and its
components), Checkpointing, Kubernetes/Yarn, ZooKeeper: yes — changes
JobManager-side `SourceCoordinator` behavior on subtask reset during partial
failover
- The S3 file system connector: no
## Documentation
- Does this pull request introduce a new feature? no
- If yes, how is the feature documented? not applicable
---
##### Was generative AI tooling used to co-author this PR?
- [X] Yes (please specify the tool below)
Generated-by: Claude Code (Fable 5)
--
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]
