tuantran0910 opened a new pull request, #68756:
URL: https://github.com/apache/doris/pull/68756
### What problem does this PR solve?
Issue Number: close #68755
Related PR: #65196, #67661, #67761
Problem Summary:
In Apache Doris 4.1.3, a cloud stream-load transaction can remain PREPARE
after its coordinator BE is restarted or rescheduled. With FQDN addressing, the
registered backend hostname differs from the recorded client IP; after
rescheduling, the pod IP can also change. FE failure detection and meta-service
coordinator cleanup then miss the abandoned transaction, leaving schema changes
and rollups in WAITING_TXN until expiration or manual abort.
This PR targets master and uses the backend ID when both coordinator IDs are
nonzero, falling back to address matching when either ID is unknown. The cloud
FE checks restart/liveness after backend lookup by ID without requiring
hostname/IP equality. Startup-time and transaction-status filters remain in
place, and non-cloud FE address matching is preserved.
Master already contains the schema-change/rollup retry and
committed/final-status safeguards from #65196 and #67661. This PR complements
those changes and preserves the scan/retry behavior from #67761.
Tests added cover cloud restart and lost-heartbeat detection with differing
addresses, protection of the current process/recent heartbeat, non-cloud
compatibility, changed-address lookup on both meta-service scan paths,
startup-time filtering, different known IDs sharing an address, unknown
request-ID fallback, and changed-address abort with current-process protection.
Validation:
- On this master-based patch: Java Checkstyle passed for both changed Java
files; changed C++ files were formatted with clang-format 16.0.6; `git diff
--check` passed.
- Source-extracted compiled checks on this patch passed 33 FE
failure-detection cases and 54 identity cases for each of the two meta-service
matching expressions. These checks are isolated harnesses, not the project's
native unit-test suites.
- Earlier Docker validation for the 4.1.3 fix passed `AlterTxnRecoveryTest`
(4 tests) and `CloudIndexTest`, `SchemaChangeJobV2Test`, `RollupJobV2Test`, and
`GlobalTransactionMgrTest` (37 tests total). These results do not validate this
master PR head.
- The earlier full FE Docker run is still in progress. Two failures have
been observed in `AuthenticationPluginManagerTest` concerning OIDC
factory/plugin loading. The run uses `maven.test.failure.ignore=true` to
collect failures across modules; a successful Maven exit must not be
interpreted as a passing suite.
- Native execution of the new master-based FE and cloud tests, full-suite
validation of this PR head, and a cloud BE-rescheduling end-to-end reproduction
are pending. No successful result is claimed for them.
CI / review follow-up: request CI workflow approval from a Committer, then
trigger `run buildall` for the current PR head according to the project's
review-and-CI guide. Cloud/meta-service Code Owner review is required. Keep
this PR in draft until the pending validation is resolved.
### Release note
Fix cloud transaction cleanup missing restarted or rescheduled BE
coordinators when their address changes or FQDN addressing is used, which could
leave schema changes and rollups waiting for abandoned prepared transactions.
### Check List (For Author)
- Test
- [ ] Regression test
- [x] Unit Test
- [ ] Manual test (add detailed scripts or steps below)
- [ ] No need to test or manual test. Explain why:
- [ ] This is a refactor/code format and no logic has been changed.
- [ ] Previous test can cover this change.
- [ ] No code files have been changed.
- [ ] Other reason
- Behavior changed:
- [ ] No.
- [x] Yes. Cloud cleanup matches known backend IDs despite address
changes; unknown IDs fall back to address matching.
- Does this need documentation?
- [x] No.
- [ ] Yes.
### Check List (For Reviewer who merge this PR)
- [ ] Confirm the release note
- [ ] Confirm test cases
- [ ] Confirm document
- [ ] Add branch pick label
--
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]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
