fix(amber): surface writer-thread failure as FatalError instead of silent hang#4683
Merged
Yicong-Huang merged 3 commits intoMay 6, 2026
Merged
Conversation
Codecov Report❌ Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## main #4683 +/- ##
============================================
+ Coverage 42.49% 42.51% +0.01%
- Complexity 2180 2185 +5
============================================
Files 1005 1005
Lines 37429 37436 +7
Branches 3914 3918 +4
============================================
+ Hits 15907 15915 +8
+ Misses 20558 20557 -1
Partials 964 964
Flags with carried forward coverage won't be shown. Click here to find out more. ☔ View full report in Codecov by Sentry. 🚀 New features to boost your workflow:
|
Contributor
Author
|
Another natural occurrence to fold into the test plan: PR #4648's release backport leg hit this exact path today.
Tracked separately in #4682. |
Yicong-Huang
force-pushed
the
fix/iceberg-write-fatal-fallback
branch
from
97eb43d to
9262054
Compare
kunwp1
approved these changes
May 4, 2026
kunwp1
left a comment
kunwp1
left a comment
Contributor
There was a problem hiding this comment.
LGTM! I have no changes to suggest.
Contributor
|
Please wait for @Xiao-zhen-Liu 's review. |
Xiao-zhen-Liu
approved these changes
May 4, 2026
Xiao-zhen-Liu
left a comment
Xiao-zhen-Liu
left a comment
Contributor
There was a problem hiding this comment.
Nice fix, and I hope this can also solve the flaky and unreproducible #3950 test case issue. Left two minor suggestions before merge.
…lent hang OutputPortResultWriterThread previously let exceptions in close() escape Thread.run(), so the iceberg commit failure was invisible to the worker, the controller still saw normal portCompleted, downstream operators read incomplete data, and the test/user observed only a 1-minute Await timeout. Capture the failure, re-throw on closeOutputStorageWriterIfNeeded, let DPThread's existing MainThreadDelegateMessage path route it to the worker actor, and let the Controller's AllForOneStrategy supervisor emit FatalError to the client immediately. Closes apache#4682. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Add OutputPortResultWriterThreadSpec with 4 tests: - OutputPortResultWriterThread leaves getFailure empty on a clean run. - OutputPortResultWriterThread captures a close() exception in getFailure so the worker can re-throw it. - OutputManager.closeOutputStorageWriterIfNeeded re-throws the writer thread's captured failure (this is the bridge from the writer thread to the DP thread → worker actor → controller supervisor → FatalError to client). - OutputManager.closeOutputStorageWriterIfNeeded is a no-op when the port has no writer thread. Together these pin every link of the fatal-error chain that this PR introduces. The OutputManager test reaches into the private outputPortResultWriterThreads map by reflection rather than going through addPort, which would require a real iceberg URI; the test file otherwise only depends on a 4-method stub BufferedItemWriter. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Address review: a putOne failure mid-loop bypassed close() and leaked the underlying writer's file handles. Move close() into a finally clause; if both legs fail, attach close()'s exception as suppressed on the original. Tests: rework StubWriter to take onPutOne/onClose thunks, add a putOne-failure test (asserts close() still ran) and a both-fail test (asserts the original is captured with the close() failure as suppressed).
Yicong-Huang
force-pushed
the
fix/iceberg-write-fatal-fallback
branch
from
c709d9e to
b215402
Compare
aglinxinyuan
added a commit
that referenced
this pull request
May 7, 2026
Mirror the result-writer cleanup added in #4683: if the state-writer thread captured a failure (e.g. iceberg commit retries exhausted), re-throw it from closeOutputStorageWriterIfNeeded so the DP thread surfaces a FatalError to the controller instead of silently announcing port completion.
yangzhang75
pushed a commit
to yangzhang75/texera
that referenced
this pull request
Jun 22, 2026
…lent hang (apache#4683) ### What changes were proposed in this PR? When `OutputPortResultWriterThread.run()` throws (e.g. iceberg commit-retry budget exhausted), the writer thread dies silently and the worker still reports `portCompleted` to the controller. The user sees a 1-minute completion timeout with no signal pointing at iceberg. Capture the failure on the writer thread, re-throw it from `OutputManager.closeOutputStorageWriterIfNeeded`, and let the existing DP-thread → worker-actor → controller-supervisor path surface it as a `FatalError` to the client. ### Any related issues, documentation, discussions? Closes apache#4682. ### How was this PR tested? `OutputPortResultWriterThreadSpec` (6 tests) covers clean run, putOne failure (close() still runs), close() failure, both-fail (close() suppressed on putOne), and `OutputManager.closeOutputStorageWriterIfNeeded` re-throw + no-op cases. ### Was this PR authored or co-authored using generative AI tooling? Generated-by: Claude Code (Opus 4.7, 1M context) --------- Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
aglinxinyuan
pushed a commit
to aglinxinyuan/texera
that referenced
this pull request
Jun 27, 2026
…apache#5737) ### What changes were proposed in this PR? - `RegionExecutionCoordinator.terminateWorkersWithRetry` retried worker termination with no limit, so a worker that never finished draining (its `EndWorker` kept failing) left the region's termination future unresolved and the workflow never reached COMPLETED, surfacing as an opaque ~1 minute timeout in `DataProcessingSpec` with no indication of the stuck region or workers. - Bounded the retry by a new `maxTerminationAttempts` budget (default 150, about 30s at the existing 200ms delay); on exhaustion the termination future fails with a descriptive error naming the region and the still-unterminated workers, instead of retrying indefinitely. - Made `maxTerminationAttempts` and `killRetryDelay` constructor parameters with production defaults so the behavior is unit-testable without long waits. - Scope note: this is a fail-fast/diagnosability change (it converts a silent hang into a fast, explicit failure, matching the pattern in apache#4683), not a guaranteed elimination of the underlying termination flake. ### Any related issues, documentation, discussions? Closes: apache#5614 ### How was this PR tested? - Run `sbt "WorkflowExecutionService/testOnly *RegionExecutionCoordinatorSpec"` (under JDK 17); expect 3 tests succeeded, 0 failed. - New test `give up with a descriptive error once the EndWorker retry budget is exhausted`: forces `EndWorker` to always fail, then asserts the completion future fails with `IllegalStateException` containing "could not be terminated after 3 attempts", the coordinator is not marked completed, exactly 3 `EndWorker` calls were made, and the worker actor ref is retained. - Existing tests in the same spec (gracefulStop ordering, transient-failure retry-then-succeed) still pass, confirming the normal and transient paths are unchanged. ### Was this PR authored or co-authored using generative AI tooling? Co-authored with Claude Opus 4.8 in compliance with ASF
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
5 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What changes were proposed in this PR?
When
OutputPortResultWriterThread.run()throws (e.g. iceberg commit-retry budget exhausted), the writer thread dies silently and the worker still reportsportCompletedto the controller. The user sees a 1-minute completion timeout with no signal pointing at iceberg.Capture the failure on the writer thread, re-throw it from
OutputManager.closeOutputStorageWriterIfNeeded, and let the existing DP-thread → worker-actor → controller-supervisor path surface it as aFatalErrorto the client.Any related issues, documentation, discussions?
Closes #4682.
How was this PR tested?
OutputPortResultWriterThreadSpec(6 tests) covers clean run, putOne failure (close() still runs), close() failure, both-fail (close() suppressed on putOne), andOutputManager.closeOutputStorageWriterIfNeededre-throw + no-op cases.Was this PR authored or co-authored using generative AI tooling?
Generated-by: Claude Code (Opus 4.7, 1M context)