goutamadwant commented on PR #12454:
URL: https://github.com/apache/seatunnel/pull/12454#issuecomment-5841703965
I stress-tested the three flaky-test parts of this PR against the PRs they
overlap with. Each PR's diff was applied to `dev` at `deb16a3c3`, one fresh JVM
per run, on a loaded 10-core macOS host. JDK 11 runs used
`-XX:ActiveProcessorCount=2`.
- **`CheckpointCoordinatorFailoverIT` trigger-dispatch test:**
- `dev` 10 / 78 failed (`Checkpoint notify complete failed`, mostly JDK 11)
- this PR 0 / 78
- #12449 0 / 78
- #12288 8 / 78
This PR and #12449 use the same injection point (`pendingCounter == 0`
under the coordinator lock). #12449 does it in the test with no production
accessor, and is green.
- **`PayPalClientTest#closeWakesRetryWait`:**
- `dev` 5 / 50 on JDK 11 at `:416`
- this PR 0 / 50
- #12444 0 / 50
With a forced `IOException` on the body write, `dev` failed 20 / 20; this
PR and #12444 failed 0 / 20. Both work.
- **`TestFilterRowKindIT#testFilterRowKindMultiTable`, Flink 1.15/1.18
only:**
- `dev` 8 / 14 runs failed
- this PR's `parallelism = 1` 0 / 6 (`Writer (1/1)` confirmed)
- #12216 0 / 13
Since #12449 and #12444 each cover one of these on their own and are small,
it might be easier for everyone if this PR dropped the
`runWhenNoCheckpointPending` / `CheckpointCoordinatorFailoverIT` part and the
PayPal part. That would leave the TaskExecutionService/PhysicalVertex recovery
fix and the PostgreSQL CDC LSN fix, which I did not test here. That would also
make this one smaller to review and avoid three-way conflicts on the same files.
The FilterRowKind conf change is fine as a stopgap. #12216 covers the same
Assert-sink race for every multi-table conf.
--
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]
