Flink Backports apache/flink
English

Flink Backports

A fix merged on some branches may still be needed on other maintained branches. We found this while releasing: far too many fixes had not been migrated, and both releases and everyday CI keep failing because of it.

Issue 1.201.20.6 2.22.2.2 2.32.3.1 2.42.4.0 2.5master · 2.5.0
FLINK-38442
ProfilingServiceTest.testRollingDeletion is unstable
Bug · Sergey Nuyanzin · origin 2.2
 PR approved Merged Merged Merged Merged
FLINK-38534
Fix flaky LocalRecoveryTest by waiting for tasks to reach RUNNING state
Bug · mukul · origin 2.3
 Not needed Test does not exist on release-1.20: no LocalRecoveryTest.java, and no testStat… PR approved Merged Merged Merged
FLINK-39984
Thread dump RPC can cause heartbeat timeout and unnecessary JM/TM failover
Bug · zbz · origin 2.4
 Merged Merged PR approved Merged Merged
FLINK-40644
ForSt file cache: CachedDataInputStream.close() does not close the remote (original) stream, leaking S3A connections until the pool is exhausted
Bug · sjchoi · origin 2.4
 Not needed code path absent on this branch PR approved PR open Merged Merged
FLINK-39720
SubQueryDecorrelator produces incorrect plans for correlated EXISTS with HAVING on aggregate outputs
Bug · lincoln.86xy · origin 2.4
 Merged PR open Merged Merged Merged
FLINK-40011
CommonExecSinkITCase.testStreamRecordTimestampInserterSinkRuntimeProvider is flaky: test source can finish before NoMoreSplitsEvent is delivered
Bug · martijnvisser · origin 2.4
 Not needed The fix changes the FLIP-27 TestSource reader in CommonExecSinkITCase. On relea… PR open PR open Merged Merged
FLINK-40088
RestClientITCase.testHttpsConnectionWithDefaultCerts is flaky due to GitHub rate limiting (HTTP 429)
Bug · martijnvisser · origin 2.4
 PR open PR open PR open Merged Merged
FLINK-40475
StatusWatermarkValve loses or stalls watermarks when subpartitions realign after idleness
Bug · martijnvisser · origin 2.4
 PR open Merged Merged Merged Merged
FLINK-40812
[Backport][1.20] Backport FLINK-38558 to release-1.20
Bug · yangjunsen · origin 2.2
 PR open Merged Merged Merged Merged
FLINK-40883
Accumulator corruption in NonTimeRowsUnboundedPrecedingFunction
Bug · bvarghese · origin 2.4
 Not needed code path absent on this branch Needs human Cherry-pick of 7a76f129672 fixes the bug on release-2.2 (regression tests pass;… PR open Merged Merged
FLINK-40903
Add source host info in TM disconnect error
Improvement · skywalker0618 · origin 2.4
 Not needed code path absent on this branch Not needed code path absent on this branch PR open Merged Merged
FLINK-40931
ForSt restore with object-store in restore mode fails when scaling in
Bug · morezaei00 · origin 2.4
 Not needed code path absent on this branch PR open Merged Merged Merged
FLINK-34099
CheckpointIntervalDuringBacklogITCase.testNoCheckpointDuringBacklog is unstable on AZP
Bug · roman · origin 2.3
 Needs human Both commits of #27630 (0d3133b, da75518) cherry-pick cleanly and byte-identica… Needs human Both commits of #27630 (0d3133b, da75518) cherry-pick cleanly and byte-identica… Merged Merged Merged
FLINK-40327
ForSt async multiGet leaks a native ReadOptions per read batch
Bug · YordanPavlov · origin 2.4
 Not needed code path absent on this branch Needs human Bug confirmed on this branch and 0edbffc538a cherry-picks cleanly, byte-identic… Needs human Bug confirmed on this branch and 0edbffc538a cherry-picks cleanly, byte-identic… Merged Merged
FLINK-40477
Partial deletes fail sink NOT NULL enforcement
Bug · fhueske · origin 2.4
 Not needed code path absent on this branch Needs human Fix is absent and needed: NotNullConstraint on this branch still checks every R… Needs human Fix is absent and needed: NotNullConstraint on this branch still checks every R… Merged Merged
FLINK-40727
Support comma-separated security.ssl.protocol and fix internal SSL hostname verification regression
Bug · gaborgsomogyi · origin 2.4
 Needs human Not cherry-pickable as-is: #29265 needs Netty 4.2 APIs (SslContextBuilder.endpo… Needs human Not cherry-pickable as-is: #29265 needs Netty 4.2 APIs (SslContextBuilder.endpo… Merged Merged Merged
FLINK-40886
BlobLibraryCacheManagerTest.testApplicationLevelJarSharing fails with FileAlreadyExistsException on the application directory
Bug · robinvm · origin 2.5
 Not needed code path absent on this branch Not needed code path absent on this branch Needs human Maintainer note: "ping author first". The author robinrvm has already opened ht… Needs human Maintainer note: "ping author first". The author robinrvm has already opened ht… Merged
FLINK-40989
MAP equality compares numeric values by reference
Bug · raminqaf · origin 2.4
 Needs human Not backported: AGENTS.md "Ask first" covers hot-path performance changes. Bug … Needs human Not backported: AGENTS.md "Ask first" covers hot-path performance changes. Bug … Needs human Not backported: AGENTS.md "Ask first" covers hot-path performance changes. Bug … Merged Merged
FLINK-38727
Unstable test cases EventTimeWindowCheckpointingITCase.testPreAggregatedSlidingTimeWindow
Bug · shekharrajak · origin 2.3
 Not needed Per MartijnVisser the delay is not the fix; the failures are FLINK-39103 combin… Not needed Rejected by maintainer: #28727 closed by MartijnVisser, failures come from FLIN… Merged Merged Merged
FLINK-38908
ExecutionPartitionLifecycleTest fails intermittently
Bug · mateczagany · origin 2.3
 Not needed Race does not exist on release-1.20 (3f52d32da60). It comes from FLINK-38114 (#… Merged Merged Merged Merged
FLINK-38967
UnalignedCheckpointFailureHandlingITCase fails intermittently
Bug · mateczagany · origin 2.3
 Merged Merged Merged Merged Merged
FLINK-38970
AdaptiveBatchSchedulerTest is unstable
Bug · shekharrajak · origin 2.3
 Not needed release-1.20 Execution#deploy builds the TaskDeploymentDescriptor synchronously… Merged Merged Merged Merged
FLINK-39117
Flaky ChainUnionTest testMigrationAndRestore
Bug · shekharrajak · origin 2.3
 Not needed Not needed: the race fixed by #27664 does not exist here. The Jira stack goes t… Not needed Not needed: the race fixed by #27664 does not exist here. The Jira stack goes t… Merged Merged Merged
FLINK-39134
UnalignedCheckpointFailureHandlingITCase.testCheckpointSuccessAfterFailure is flaky
Bug · roman · origin 2.3
 Not needed Not needed: the flake does not exist on release-1.20. The fix (setMinPauseBetwe… Not needed Not needed: the flake does not exist on release-2.2. The fix (setMinPauseBetwee… Merged Merged Merged
FLINK-39164
Flaky test HistoryServerTest.testRemoveApplicationArchivesBeyondHistorySizeLimit
Bug · yizh · origin 2.3
 Not needed Test-only fix for HistoryServerTest.testRemoveApplicationArchivesBeyondHistoryS… Not needed Test-only fix for HistoryServerTest.testRemoveApplicationArchivesBeyondHistoryS… Merged Merged Merged
FLINK-39182
Fix the flaky test class ExecutionGraphRestartTest
Bug · izeren · origin 2.3
 Not needed Flake does not exist on release-1.20: condition 3 fails. Cause is FLINK-38114 (… Merged Merged Merged Merged
FLINK-39387
Fix flaky scheduler benchmark tests caused by thread assertion failure in async TDD creation
Bug · Jiangang · origin 2.3
 Not needed Failure path does not exist on release-1.20 (checked at 3745565188f). The bug c… Merged Merged Merged Merged
FLINK-39388
Fix Flaky DataGeneratorSourceITCase#testGatedRateLimiter
Bug · averyzhang · origin 2.4
 Merged Merged Merged Merged Merged
FLINK-39641
Fix unstable position of the Rescales Tab
Bug · panyuepeng · origin 2.4
 Not needed The Rescales tab does not exist on release-1.20: no 'rescales' match in flink-r… Not needed The Rescales tab does not exist on release-2.2: no 'rescales' match in flink-ru… Merged Merged Merged
FLINK-39914
Fix flaky TaskDeploymentDescriptorFactoryTest#testHybridVertexFinish caused by async TDD creation
Bug · chshining · origin 2.4
 Not needed Test-only fix for a failure caused by async TDD creation from FLINK-38114 (6b04… Merged Merged Merged Merged
FLINK-39921
Fix the flaky test case ExecutionVertexCancelTest.testSendCancelAndReceiveFail
Bug · chshining · origin 2.4
 Not needed Branch lacks the async TDD creation this flaky test fix targets: no FLINK-38114… Merged Merged Merged Merged
FLINK-39929
Fix the flaky test case ExecutionTimeBasedSlowTaskDetectorTest due to unexepected ComponentMainThreadExecutor setting.
Bug · chshining · origin 2.4
 Not needed Flaky failures come from async TDD creation added by FLINK-38114 (6b0426c7c08),… Merged Merged Merged Merged
FLINK-39931
Fix flaky AdaptiveBatchSchedulerTest due to unexpected ComponentMainThreadExecutor behavior in async execution
Bug · chshining · origin 2.4
 Not needed Jira FLINK-39931 has no fix commit on any branch: PR 28481 belongs to FLINK-400… Not needed Jira FLINK-39931 has no fix commit on any branch: PR 28481 belongs to FLINK-400… Not needed Jira FLINK-39931 has no fix commit on any branch: PR 28481 belongs to FLINK-400… Merged Merged
FLINK-40099
YARNSessionFIFOSecuredITCase.testDetachedMode is flaky: Kerberos keytab log check reads a killed TaskManager's log
Bug · raminqaf · origin 2.4
 Not needed No backport needed. The failure does not exist on this branch: it pins log4j 2.… Not needed No backport needed. The failure does not exist on this branch: it pins log4j 2.… Merged Net fix on release-2.3 as commit 9ce420f91b7 (FLINK-40109, https://github.com/a… Merged Merged
FLINK-40363
MiscSemanticTests#COALESCE_NESTED_ROW_LEFT_JOIN is unstable
Bug · Sergey Nuyanzin · origin 2.4
 Not needed code path absent on this branch Not needed Test absent on release-2.2: CalcTestPrograms has no COALESCE_NESTED_ROW_LEFT_JO… Not needed Test absent on release-2.3: CalcTestPrograms has no COALESCE_NESTED_ROW_LEFT_JO… Merged Merged
FLINK-41247
AbstractServerTest.testPortRangeSuccess sees an open server connection after the servers are closed
Bug · psinha · origin 2.5
 Not needed code path absent on this branch Not needed code path absent on this branch Merged Merged Merged

ProfilingServiceTest.testRollingDeletion is unstable

Type
Bug
Priority
Major
Jira status
Resolved
Owner
Sergey Nuyanzin
Origin version
2.2
Origin PR
#27047
Fix merged
2025-10-02 07:37 UTC
Jira
FLINK-38442

  1. 1.20 1.20.6 PR approved

    • #29756Platform 1996fanrui · Open

  2. 2.2 2.2.2 Merged

  3. 2.3 2.3.1 Merged

  4. 2.4 2.4.0 Merged

  5. 2.5 master · 2.5.0 Merged

Fix flaky LocalRecoveryTest by waiting for tasks to reach RUNNING state

Type
Bug
Priority
Critical
Jira status
Reopened
Owner
mukul
Origin version
2.3
Origin PR
#27767
Fix merged
2026-03-25 06:14 UTC
Jira
FLINK-38534

  1. 1.20 1.20.6 Not needed

    Test does not exist on release-1.20: no LocalRecoveryTest.java, and no testStateSizeIsConsideredForLocalRecovery or waitForCheckpointInProgress anywhere on the branch (git grep at b43c38ae0de). The affected test path is absent, so there is nothing to fix. Previous note: low priority.

  2. 2.2 2.2.2 PR approved

    • #29751Platform 1996fanrui · Open

  3. 2.3 2.3.1 Merged

  4. 2.4 2.4.0 Merged

  5. 2.5 master · 2.5.0 Merged

Thread dump RPC can cause heartbeat timeout and unnecessary JM/TM failover

Type
Bug
Priority
Critical
Jira status
Resolved
Owner
zbz
Origin version
2.4
Origin PR
#28778
Fix merged
2026-08-01 16:13 UTC
Jira
FLINK-39984

  1. 1.20 1.20.6 Merged

  2. 2.2 2.2.2 Merged

  3. 2.3 2.3.1 PR approved

    • #29738Platform 1996fanrui · Open

  4. 2.4 2.4.0 Merged

  5. 2.5 master · 2.5.0 Merged

ForSt file cache: CachedDataInputStream.close() does not close the remote (original) stream, leaking S3A connections until the pool is exhausted

Type
Bug
Priority
Critical
Jira status
Resolved
Owner
sjchoi
Origin version
2.4
Origin PR
#29168
Fix merged
2026-10-06 16:52 UTC
Jira
FLINK-40644

  1. 1.20 1.20.6 Not needed

    code path absent on this branch

  2. 2.2 2.2.2 PR approved

    • #29753Platform 1996fanrui · Open

  3. 2.3 2.3.1 PR open

    • #29752Platform 1996fanrui · Open

  4. 2.4 2.4.0 Merged

  5. 2.5 master · 2.5.0 Merged

SubQueryDecorrelator produces incorrect plans for correlated EXISTS with HAVING on aggregate outputs

Type
Bug
Priority
Critical
Jira status
Closed
Owner
lincoln.86xy
Origin version
2.4
Origin PR
#28217
Fix merged
2026-05-25 06:40 UTC
Jira
FLINK-39720

  1. 1.20 1.20.6 Merged

  2. 2.2 2.2.2 PR open

    • #29757Platform 1996fanrui · Open

  3. 2.3 2.3.1 Merged

  4. 2.4 2.4.0 Merged

  5. 2.5 master · 2.5.0 Merged

CommonExecSinkITCase.testStreamRecordTimestampInserterSinkRuntimeProvider is flaky: test source can finish before NoMoreSplitsEvent is delivered

Type
Bug
Priority
Major
Jira status
Resolved
Owner
martijnvisser
Origin version
2.4
Origin PR
#28572
Fix merged
2026-06-29 20:08 UTC
Jira
FLINK-40011

  1. 1.20 1.20.6 Not needed

    The fix changes the FLIP-27 TestSource reader in CommonExecSinkITCase. On release-1.20 (1dce10347e9) that test uses the legacy SourceFunctionProvider with TestSourceFunction: no SplitEnumerator, no SourceReader, no notifyNoMoreSplits, so no NoMoreSplitsEvent race exists.

  2. 2.2 2.2.2 PR open

    • #29734Platform 1996fanrui · Open

  3. 2.3 2.3.1 PR open

    • #29733Platform 1996fanrui · Open

  4. 2.4 2.4.0 Merged

  5. 2.5 master · 2.5.0 Merged

RestClientITCase.testHttpsConnectionWithDefaultCerts is flaky due to GitHub rate limiting (HTTP 429)

Type
Bug
Priority
Major
Jira status
Resolved
Owner
martijnvisser
Origin version
2.4
Origin PR
#28676
Fix merged
2026-07-15 09:20 UTC
Jira
FLINK-40088

  1. 1.20 1.20.6 PR open

    • #29737Platform 1996fanrui · Open

  2. 2.2 2.2.2 PR open

    • #29736Platform 1996fanrui · Open

  3. 2.3 2.3.1 PR open

    • #29735Platform 1996fanrui · Open

  4. 2.4 2.4.0 Merged

  5. 2.5 master · 2.5.0 Merged

StatusWatermarkValve loses or stalls watermarks when subpartitions realign after idleness

Type
Bug
Priority
Critical
Jira status
Resolved
Owner
martijnvisser
Origin version
2.4
Origin PR
#29024
Fix merged
2026-08-28 12:15 UTC
Jira
FLINK-40475

  1. 1.20 1.20.6 PR open

    • #29758Platform 1996fanrui · Open

  2. 2.2 2.2.2 Merged

  3. 2.3 2.3.1 Merged

  4. 2.4 2.4.0 Merged

  5. 2.5 master · 2.5.0 Merged

[Backport][1.20] Backport FLINK-38558 to release-1.20

Type
Bug
Priority
Major
Jira status
Open
Owner
yangjunsen
Origin version
2.2
Origin PR
#27160
Fix merged
2025-10-31 12:44 UTC
Jira
FLINK-40812

  1. 1.20 1.20.6 PR open

  2. 2.2 2.2.2 Merged

  3. 2.3 2.3.1 Merged

  4. 2.4 2.4.0 Merged

  5. 2.5 master · 2.5.0 Merged

Accumulator corruption in NonTimeRowsUnboundedPrecedingFunction

Type
Bug
Priority
Major
Jira status
Open
Owner
bvarghese
Origin version
2.4
Origin PR
#29373
Fix merged
2026-10-07 20:08 UTC
Jira
FLINK-40883

  1. 1.20 1.20.6 Not needed

    code path absent on this branch

  2. 2.2 2.2.2 Needs human

    Cherry-pick of 7a76f129672 fixes the bug on release-2.2 (regression tests pass; they fail with the main change reverted), but the fix parameterizes the test classes over RocksDB and testInsertAndRetractWithStateTTLExpiration (Rows and Range) then fails on RocksDB at mapState.isEmpty(), with or without the main change. release-2.2 lacks FLINK-38992 (timer-based cleanup, fd9a1ba39ac); TtlMapState.isEmpty() delegates to the backend, which still holds expired entries. Decision needed: backport FLINK-38992 first as a prerequisite, or accept restricting this TTL test to HEAP on release-2.2 (disabling a test variant is a maintainer call). Test conflicts from missing FLINK-39004 are mechanical: take the fix side (@TestTemplate).

  3. 2.3 2.3.1 PR open

    • #29755Platform 1996fanrui · Open

  4. 2.4 2.4.0 Merged

  5. 2.5 master · 2.5.0 Merged

Add source host info in TM disconnect error

Type
Improvement
Priority
Minor
Jira status
Closed
Owner
skywalker0618
Origin version
2.4
Origin PR
#29391
Fix merged
2026-10-12 15:23 UTC
Jira
FLINK-40903

  1. 1.20 1.20.6 Not needed

    code path absent on this branch

  2. 2.2 2.2.2 Not needed

    code path absent on this branch

  3. 2.3 2.3.1 PR open

    • #29759Platform 1996fanrui · Open

  4. 2.4 2.4.0 Merged

  5. 2.5 master · 2.5.0 Merged

ForSt restore with object-store in restore mode fails when scaling in

Type
Bug
Priority
Critical
Jira status
Resolved
Owner
morezaei00
Origin version
2.4
Origin PR
#29475
Fix merged
2026-10-17 11:50 UTC
Jira
FLINK-40931

  1. 1.20 1.20.6 Not needed

    code path absent on this branch

  2. 2.2 2.2.2 PR open

    • #29529 mohsenrezaeithe · Open

  3. 2.3 2.3.1 Merged

  4. 2.4 2.4.0 Merged

  5. 2.5 master · 2.5.0 Merged

CheckpointIntervalDuringBacklogITCase.testNoCheckpointDuringBacklog is unstable on AZP

Type
Bug
Priority
Major
Jira status
Closed
Owner
roman
Origin version
2.3
Origin PR
#27630
Fix merged
2026-02-18 22:53 UTC
Jira
FLINK-34099

  1. 1.20 1.20.6 Needs human

    Both commits of #27630 (0d3133b, da75518) cherry-pick cleanly and byte-identical; 5 runs: Tests run 4, Failures 0, Skipped 1; spotless/checkstyle pass. Not opened: the guide requires no skipped tests, but the fix itself adds @Ignore to testNoCheckpointDuringBacklog, so the gate cannot be met. Un-ignored, the hardened test passed 5 of 5 runs locally on JDK 11, while Jira comments report release-1.20 nightly failures on 2026-09-22 and 2026-10-04 (not checked in CI). Decision needed: is disabling this test on 1.20 wanted despite the gate and the local passes, or wait for FLINK-39108 (#29110, open on master)?

  2. 2.2 2.2.2 Needs human

    Both commits of #27630 (0d3133b, da75518) cherry-pick cleanly and byte-identical; 5 runs: Tests run 4, Failures 0, Skipped 1; spotless/checkstyle pass. Not opened: the guide requires no skipped tests, but the fix itself adds @Ignore to testNoCheckpointDuringBacklog, so the gate cannot be met. Un-ignored, the hardened test failed 4 of 5 runs (expected 0, was 1), matching open bug FLINK-39108. Decision needed: accept a backport that disables the test despite the none-skipped gate, or wait for FLINK-39108 (#29110, open on master).

  3. 2.3 2.3.1 Merged

  4. 2.4 2.4.0 Merged

  5. 2.5 master · 2.5.0 Merged

ForSt async multiGet leaks a native ReadOptions per read batch

Type
Bug
Priority
Critical
Jira status
Resolved
Owner
YordanPavlov
Origin version
2.4
Origin PR
#28985
Fix merged
2026-10-08 13:38 UTC
Jira
FLINK-40327

  1. 1.20 1.20.6 Not needed

    code path absent on this branch

  2. 2.2 2.2.2 Needs human

    Bug confirmed on this branch and 0edbffc538a cherry-picks cleanly, byte-identical. Stopped because Flink AGENTS.md "Ask first" lists hot-path changes (state access), and #28985 itself answers "runtime per-record code paths: yes" (one extra native close per read batch, which it calls negligible). Decision needed: does a maintainer accept this for the patch branch? If yes, rerun; the tests were not run and no PR was opened.

  3. 2.3 2.3.1 Needs human

    Bug confirmed on this branch and 0edbffc538a cherry-picks cleanly, byte-identical. Stopped because Flink AGENTS.md "Ask first" lists hot-path changes (state access), and #28985 itself answers "runtime per-record code paths: yes" (one extra native close per read batch, which it calls negligible). Decision needed: does a maintainer accept this for the patch branch? If yes, rerun; the tests were not run and no PR was opened.

  4. 2.4 2.4.0 Merged

  5. 2.5 master · 2.5.0 Merged

Partial deletes fail sink NOT NULL enforcement

Type
Bug
Priority
Critical
Jira status
Resolved
Owner
fhueske
Origin version
2.4
Origin PR
#29067
Fix merged
2026-09-03 10:52 UTC
Jira
FLINK-40477

  1. 1.20 1.20.6 Not needed

    code path absent on this branch

  2. 2.2 2.2.2 Needs human

    Fix is absent and needed: NotNullConstraint on this branch still checks every RowKind, and ChangelogMode#keyOnlyDeletes exists. Stopped before cherry-picking d554162df59: it changes NotNullConstraint#enforce, a per-record path (#29067 answers "runtime per-record code paths: yes"), and Flink AGENTS.md lists hot-path changes under "Ask first". Decision needed: does a maintainer accept this per-record change on the patch branch? If yes, reset to needs_backport.

  3. 2.3 2.3.1 Needs human

    Fix is absent and needed: NotNullConstraint on this branch still checks every RowKind, and ChangelogMode#keyOnlyDeletes exists. Stopped before cherry-picking d554162df59: it changes NotNullConstraint#enforce, a per-record path (#29067 answers "runtime per-record code paths: yes"), and Flink AGENTS.md lists hot-path changes under "Ask first". Decision needed: does a maintainer accept this per-record change on the patch branch? If yes, reset to needs_backport.

  4. 2.4 2.4.0 Merged

  5. 2.5 master · 2.5.0 Merged

Support comma-separated security.ssl.protocol and fix internal SSL hostname verification regression

Type
Bug
Priority
Blocker
Jira status
Closed
Owner
gaborgsomogyi
Origin version
2.4
Origin PR
#29265
Fix merged
2026-09-23 15:25 UTC
Jira
FLINK-40727

  1. 1.20 1.20.6 Needs human

    Not cherry-pickable as-is: #29265 needs Netty 4.2 APIs (SslContextBuilder.endpointIdentificationAlgorithm, MultiThreadIoEventLoopGroup, NioIoHandler), but this branch has Netty 4.1.91 (javap on the shaded jar). The hostname regression does not exist here: FLINK-38953 and Netty 4.2 are absent. The Pekko bug is present (unquoted protocol in PekkoUtils, no engine override in CustomSSLEngineProvider), though SecurityOptions documents comma lists as unsupported. Decision needed: backport only the comma-separated security.ssl.protocol part (drop the SSLUtils change and hostname test, rewrite tests for Netty 4.1, change the documented option contract on a patch branch), or mark not_needed. Jira fixVersions are 2.3.1 and 2.4.0 only, no maintainer ruling.

  2. 2.2 2.2.2 Needs human

    Not cherry-pickable as-is: #29265 needs Netty 4.2 APIs (SslContextBuilder.endpointIdentificationAlgorithm, MultiThreadIoEventLoopGroup, NioIoHandler), but this branch has Netty 4.1.100 (javap on the shaded jar). The hostname regression does not exist here: FLINK-38953 and Netty 4.2 are absent. The Pekko bug is present (unquoted protocol in PekkoUtils, no engine override in CustomSSLEngineProvider), though SecurityOptions documents comma lists as unsupported. Decision needed: backport only the comma-separated security.ssl.protocol part (drop the SSLUtils change and hostname test, rewrite tests for Netty 4.1, change the documented option contract on a patch branch), or mark not_needed. Jira fixVersions are 2.3.1 and 2.4.0 only, no maintainer ruling.

  3. 2.3 2.3.1 Merged

  4. 2.4 2.4.0 Merged

  5. 2.5 master · 2.5.0 Merged

BlobLibraryCacheManagerTest.testApplicationLevelJarSharing fails with FileAlreadyExistsException on the application directory

Type
Bug
Priority
Major
Jira status
Open
Owner
robinvm
Origin version
2.5
Origin PR
#29647
Fix merged
2026-10-30 17:30 UTC
Jira
FLINK-40886

  1. 1.20 1.20.6 Not needed

    code path absent on this branch

  2. 2.2 2.2.2 Not needed

    code path absent on this branch

  3. 2.3 2.3.1 Needs human

    Maintainer note: "ping author first". The author robinrvm has already opened https://github.com/apache/flink/pull/29742 for release-2.3, so no platform backport was started. Decision needed: confirm the note is satisfied by that PR, then set pr_open with pr_source external.

  4. 2.4 2.4.0 Needs human

    Maintainer note: "ping author first". The author robinrvm has already opened https://github.com/apache/flink/pull/29741 for release-2.4, so no platform backport was started. Decision needed: confirm the note is satisfied by that PR, then set pr_open with pr_source external.

  5. 2.5 master · 2.5.0 Merged

MAP equality compares numeric values by reference

Type
Bug
Priority
Critical
Jira status
Resolved
Owner
raminqaf
Origin version
2.4
Origin PR
#29481
Fix merged
2026-10-17 19:44 UTC
Jira
FLINK-40989

  1. 1.20 1.20.6 Needs human

    Not backported: AGENTS.md "Ask first" covers hot-path performance changes. Bug is present here (generateMapComparison unchanged), but fix bbf37d099464 replaces HashMap lookups with O(n^2) key scans in generated per-record code, reached through RecordEqualiser in deduplication and group aggregation (PR 29481 declares per-record paths: yes). It also makes those operators drop updates they used to emit and changes key matching to SQL =. Decision needed: does a maintainer accept this performance and changelog behavior change in a patch release on this branch? If yes, reset to needs_backport.

  2. 2.2 2.2.2 Needs human

    Not backported: AGENTS.md "Ask first" covers hot-path performance changes. Bug is present here (generateMapComparison unchanged), but fix bbf37d099464 replaces HashMap lookups with O(n^2) key scans in generated per-record code, reached through RecordEqualiser in deduplication and group aggregation (PR 29481 declares per-record paths: yes). It also makes those operators drop updates they used to emit and changes key matching to SQL =. Decision needed: does a maintainer accept this performance and changelog behavior change in a patch release on this branch? If yes, reset to needs_backport.

  3. 2.3 2.3.1 Needs human

    Not backported: AGENTS.md "Ask first" covers hot-path performance changes. Bug is present here (generateMapComparison unchanged), but fix bbf37d099464 replaces HashMap lookups with O(n^2) key scans in generated per-record code, reached through RecordEqualiser in deduplication and group aggregation (PR 29481 declares per-record paths: yes). It also makes those operators drop updates they used to emit and changes key matching to SQL =. Decision needed: does a maintainer accept this performance and changelog behavior change in a patch release on this branch? If yes, reset to needs_backport.

  4. 2.4 2.4.0 Merged

  5. 2.5 master · 2.5.0 Merged

Unstable test cases EventTimeWindowCheckpointingITCase.testPreAggregatedSlidingTimeWindow

Type
Bug
Priority
Critical
Jira status
Closed
Owner
shekharrajak
Origin version
2.3
Origin PR
#27659
Fix merged
2026-04-11 03:16 UTC
Jira
FLINK-38727

  1. 1.20 1.20.6 Not needed

    Per MartijnVisser the delay is not the fix; the failures are FLINK-39103 combined with FLINK-38180. release-1.20 lacks FLINK-38180: Task.java has no postFailureCleanUpRegistry and restoreAndInvoke still wraps cleanUp in try/addSuppressed. Same reason he closed the 2.1 backport #28728 as not affected.

  2. 2.2 2.2.2 Not needed

    Rejected by maintainer: #28727 closed by MartijnVisser, failures come from FLINK-39103, which a delay cannot fix. FLINK-39103 merged to release-2.2 as 1f90dad29b3 (#29338). FLINK-40846 (#29589) removes the delay from master.

  3. 2.3 2.3.1 Merged

  4. 2.4 2.4.0 Merged

  5. 2.5 master · 2.5.0 Merged

ExecutionPartitionLifecycleTest fails intermittently

Type
Bug
Priority
Major
Jira status
Resolved
Owner
mateczagany
Origin version
2.3
Origin PR
#27406
Fix merged
2026-01-28 11:50 UTC
Jira
FLINK-38908

  1. 1.20 1.20.6 Not needed

    Race does not exist on release-1.20 (3f52d32da60). It comes from FLINK-38114 (#26821, 6b0426c7c08), absent from this branch (git log --grep empty): Execution#deploy here creates the TaskDeploymentDescriptor synchronously, and the async submitTask completion only calls notifyCompletedDeployment on ack, so no partition release path is reachable. The fix also relies on ExecutionUtils and TestingUtils.jmMainThreadExecutorExtension, both missing here. Decided from code; test not run.

  2. 2.2 2.2.2 Merged

  3. 2.3 2.3.1 Merged

  4. 2.4 2.4.0 Merged

  5. 2.5 master · 2.5.0 Merged

UnalignedCheckpointFailureHandlingITCase fails intermittently

Type
Bug
Priority
Major
Jira status
Closed
Owner
mateczagany
Origin version
2.3
Origin PR
#27466
Fix merged
2026-01-28 11:54 UTC
Jira
FLINK-38967

  1. 1.20 1.20.6 Merged

  2. 2.2 2.2.2 Merged

  3. 2.3 2.3.1 Merged

  4. 2.4 2.4.0 Merged

  5. 2.5 master · 2.5.0 Merged

AdaptiveBatchSchedulerTest is unstable

Type
Bug
Priority
Major
Jira status
Resolved
Owner
shekharrajak
Origin version
2.3
Origin PR
#27660
Fix merged
2026-03-19 16:16 UTC
Jira
FLINK-38970

  1. 1.20 1.20.6 Not needed

    release-1.20 Execution#deploy builds the TaskDeploymentDescriptor synchronously on the main thread, without the async thenComposeAsync path behind the failure. Unfixed test: 0 failures in 60000 repeats (JDK 11 and 17); release-2.2 unfixed: 2 failures in 300.

  2. 2.2 2.2.2 Merged

  3. 2.3 2.3.1 Merged

  4. 2.4 2.4.0 Merged

  5. 2.5 master · 2.5.0 Merged

Flaky ChainUnionTest testMigrationAndRestore

Type
Bug
Priority
Major
Jira status
Resolved
Owner
shekharrajak
Origin version
2.3
Origin PR
#27664
Fix merged
2026-04-01 01:51 UTC
Jira
FLINK-39117

  1. 1.20 1.20.6 Not needed

    Not needed: the race fixed by #27664 does not exist here. The Jira stack goes through ExecutionTrackingCheckpointCoordinatorDeActivator.onStateUpdate -> startCheckpointScheduler -> stopCheckpointScheduler, and that class came with 62566c4b445 [FLINK-38940] (2.3), which is absent on this branch. Here CheckpointCoordinatorDeActivator starts the scheduler only on job status RUNNING, before the test can trigger a savepoint; the task-RUNNING start in DefaultExecutionGraph.updateStateInternal is guarded by !isPeriodicCheckpointingStarted, and the failure-handler start in triggerSavepoint finishes before the retry is sent.

  2. 2.2 2.2.2 Not needed

    Not needed: the race fixed by #27664 does not exist here. The Jira stack goes through ExecutionTrackingCheckpointCoordinatorDeActivator.onStateUpdate -> startCheckpointScheduler -> stopCheckpointScheduler, and that class came with 62566c4b445 [FLINK-38940] (2.3), which is absent on this branch. Here CheckpointCoordinatorDeActivator starts the scheduler only on job status RUNNING, before the test can trigger a savepoint; the task-RUNNING start in DefaultExecutionGraph.updateStateInternal is guarded by !isPeriodicCheckpointingStarted, and the failure-handler start in triggerSavepoint finishes before the retry is sent.

  3. 2.3 2.3.1 Merged

  4. 2.4 2.4.0 Merged

  5. 2.5 master · 2.5.0 Merged

UnalignedCheckpointFailureHandlingITCase.testCheckpointSuccessAfterFailure is flaky

Type
Bug
Priority
Major
Jira status
Closed
Owner
roman
Origin version
2.3
Origin PR
#27693
Fix merged
2026-02-27 09:22 UTC
Jira
FLINK-39134

  1. 1.20 1.20.6 Not needed

    Not needed: the flake does not exist on release-1.20. The fix (setMinPauseBetweenCheckpoints in UnalignedCheckpointFailureHandlingITCase) only delays periodic checkpoints that FLINK-38939 (7c90b9e4eb6, fix version 2.3.0) enabled in this test by changing the interval from Long.MAX_VALUE to 100000. release-1.20 lacks that commit: the test still calls enableCheckpointing(Long.MAX_VALUE), which equals CheckpointCoordinatorConfiguration.DISABLED_CHECKPOINT_INTERVAL, so isPeriodicCheckpointingConfigured() is false and the scheduler never starts. Min pause applies only to periodic requests (CheckpointRequestDecider), so the fix would be a no-op here.

  2. 2.2 2.2.2 Not needed

    Not needed: the flake does not exist on release-2.2. The fix (setMinPauseBetweenCheckpoints in UnalignedCheckpointFailureHandlingITCase) only delays periodic checkpoints that FLINK-38939 (7c90b9e4eb6, fix version 2.3.0) enabled in this test by changing the interval from Long.MAX_VALUE to 100000. release-2.2 lacks that commit: the test still calls enableCheckpointing(Long.MAX_VALUE), which equals CheckpointCoordinatorConfiguration.DISABLED_CHECKPOINT_INTERVAL, so isPeriodicCheckpointingConfigured() is false and the scheduler never starts. Min pause applies only to periodic requests (CheckpointRequestDecider), so the fix would be a no-op here.

  3. 2.3 2.3.1 Merged

  4. 2.4 2.4.0 Merged

  5. 2.5 master · 2.5.0 Merged

Flaky test HistoryServerTest.testRemoveApplicationArchivesBeyondHistorySizeLimit

Type
Bug
Priority
Major
Jira status
Closed
Owner
yizh
Origin version
2.3
Origin PR
#27672
Fix merged
2026-02-28 02:47 UTC
Jira
FLINK-39164

  1. 1.20 1.20.6 Not needed

    Test-only fix for HistoryServerTest.testRemoveApplicationArchivesBeyondHistorySizeLimit (3cf73f3d809). That test does not exist on release-1.20: it was added by FLINK-38777 (e3adbdce5a0, application archives), which is not an ancestor of the branch, and HISTORY_SERVER_RETAINED_APPLICATIONS is absent too. The remaining sibling test testRemoveOldestModifiedArchivesBeyondHistorySizeLimit sets explicit modification times via setLastModified one minute apart, so it lacks this identical-timestamp race.

  2. 2.2 2.2.2 Not needed

    Test-only fix for HistoryServerTest.testRemoveApplicationArchivesBeyondHistorySizeLimit (3cf73f3d809). That test does not exist on release-2.2: it was added by FLINK-38777 (e3adbdce5a0, application archives), which is not an ancestor of the branch, and HISTORY_SERVER_RETAINED_APPLICATIONS is absent too. The remaining sibling test testRemoveOldestModifiedArchivesBeyondHistorySizeLimit sets explicit modification times via setLastModified one minute apart, so it lacks this identical-timestamp race.

  3. 2.3 2.3.1 Merged

  4. 2.4 2.4.0 Merged

  5. 2.5 master · 2.5.0 Merged

Fix the flaky test class ExecutionGraphRestartTest

Type
Bug
Priority
Minor
Jira status
Closed
Owner
izeren
Origin version
2.3
Origin PR
#27740
Fix merged
2026-03-05 16:14 UTC
Jira
FLINK-39182

  1. 1.20 1.20.6 Not needed

    Flake does not exist on release-1.20: condition 3 fails. Cause is FLINK-38114 (commit 6b0426c7c08, on release-2.2+ only, absent from release-1.20), which moved TDD creation behind thenComposeAsync(.., mainThreadExecutor) from the I/O executor, so forMainThread() is hit off-thread and fails deployment. On 1.20 Execution.deploy creates the TDD synchronously on the main thread; the only cross-thread callback handles the ack and marks nothing failed on success. Probe with @RepeatedTest(300) on all 7 tests: release-1.20 3f52d32da60 (JDK 11) 2100 runs, 0 failures; pre-fix release-2.2 77849c6f877 (JDK 17) failed with expected RUNNING/FINISHED/50, the original symptoms.

  2. 2.2 2.2.2 Merged

  3. 2.3 2.3.1 Merged

  4. 2.4 2.4.0 Merged

  5. 2.5 master · 2.5.0 Merged

Fix flaky scheduler benchmark tests caused by thread assertion failure in async TDD creation

Type
Bug
Priority
Major
Jira status
Resolved
Owner
Jiangang
Origin version
2.3
Origin PR
#27882
Fix merged
2026-04-10 02:25 UTC
Jira
FLINK-39387

  1. 1.20 1.20.6 Not needed

    Failure path does not exist on release-1.20 (checked at 3745565188f). The bug comes from FLINK-38114 (6b0426c7c08), which is not on the branch: no initOffloadedTaskRestoreRef or tryGetTaskDeploymentDescriptorForSlot in Execution.java, the TDD is created synchronously in deploy(). There is no intermediate main-executor stage whose failure can reach markFailed; the terminal whenCompleteAsync dispatch only drops the success callback (notifyCompletedDeployment), so no failover occurs even though SchedulerBenchmarkUtils still uses forMainThread().

  2. 2.2 2.2.2 Merged

  3. 2.3 2.3.1 Merged

  4. 2.4 2.4.0 Merged

  5. 2.5 master · 2.5.0 Merged

Fix Flaky DataGeneratorSourceITCase#testGatedRateLimiter

Type
Bug
Priority
Major
Jira status
Resolved
Owner
averyzhang
Origin version
2.4
Origin PR
#27883
Fix merged
2026-04-18 06:09 UTC
Jira
FLINK-39388

  1. 1.20 1.20.6 Merged

  2. 2.2 2.2.2 Merged

  3. 2.3 2.3.1 Merged

  4. 2.4 2.4.0 Merged

  5. 2.5 master · 2.5.0 Merged

Fix unstable position of the Rescales Tab

Type
Bug
Priority
Minor
Jira status
Resolved
Owner
panyuepeng
Origin version
2.4
Origin PR
#28131
Fix merged
2026-05-09 11:18 UTC
Jira
FLINK-39641

  1. 1.20 1.20.6 Not needed

    The Rescales tab does not exist on release-1.20: no 'rescales' match in flink-runtime-web/web-dashboard/src, and job-status.component.ts has no rescalesIndexOfNav. The tab is added by later work on release-2.3 and newer (FLINK-38900 series), so the bug cannot occur here.

  2. 2.2 2.2.2 Not needed

    The Rescales tab does not exist on release-2.2: no 'rescales' match in flink-runtime-web/web-dashboard/src, and job-status.component.ts has no rescalesIndexOfNav. The tab is added by later work on release-2.3 and newer (FLINK-38900 series), so the bug cannot occur here.

  3. 2.3 2.3.1 Merged

  4. 2.4 2.4.0 Merged

  5. 2.5 master · 2.5.0 Merged

Fix flaky TaskDeploymentDescriptorFactoryTest#testHybridVertexFinish caused by async TDD creation

Type
Bug
Priority
Minor
Jira status
Resolved
Owner
chshining
Origin version
2.4
Origin PR
#28398
Fix merged
2026-06-13 19:10 UTC
Jira
FLINK-39914

  1. 1.20 1.20.6 Not needed

    Test-only fix for a failure caused by async TDD creation from FLINK-38114 (6b0426c7c08), which release-1.20 lacks: Execution.deploy() there creates the TDD synchronously on the main thread (no tddCreatedDuringDeployFuture), and executiongraph/utils/ExecutionUtils.java does not exist.

  2. 2.2 2.2.2 Merged

  3. 2.3 2.3.1 Merged

  4. 2.4 2.4.0 Merged

  5. 2.5 master · 2.5.0 Merged

Fix the flaky test case ExecutionVertexCancelTest.testSendCancelAndReceiveFail

Type
Bug
Priority
Major
Jira status
Resolved
Owner
chshining
Origin version
2.4
Origin PR
#28449
Fix merged
2026-06-20 03:21 UTC
Jira
FLINK-39921

  1. 1.20 1.20.6 Not needed

    Branch lacks the async TDD creation this flaky test fix targets: no FLINK-38114 commit (6b0426c7c08) on release-1.20; Execution.deploy() builds the TDD synchronously (Execution.java:625); tddCreatedDuringDeployFuture and executiongraph/utils/ExecutionUtils.java are absent.

  2. 2.2 2.2.2 Merged

  3. 2.3 2.3.1 Merged

  4. 2.4 2.4.0 Merged

  5. 2.5 master · 2.5.0 Merged

Fix the flaky test case ExecutionTimeBasedSlowTaskDetectorTest due to unexepected ComponentMainThreadExecutor setting.

Type
Bug
Priority
Major
Jira status
Resolved
Owner
chshining
Origin version
2.4
Origin PR
#28434
Fix merged
2026-06-24 05:05 UTC
Jira
FLINK-39929

  1. 1.20 1.20.6 Not needed

    Flaky failures come from async TDD creation added by FLINK-38114 (6b0426c7c08), which release-1.20 lacks: Execution.deploy() there creates the TDD synchronously on the main thread (createDeploymentDescriptor), with no thenComposeAsync/tryGetTaskDeploymentDescriptorForSlot/tddCreatedDuringDeployFuture. Fix targets behavior the branch does not have; its helpers ExecutionUtils and top-level NoMainThreadCheckComponentMainThreadExecutor are also absent.

  2. 2.2 2.2.2 Merged

  3. 2.3 2.3.1 Merged

  4. 2.4 2.4.0 Merged

  5. 2.5 master · 2.5.0 Merged

Fix flaky AdaptiveBatchSchedulerTest due to unexpected ComponentMainThreadExecutor behavior in async execution

Type
Bug
Priority
Minor
Jira status
Closed
Owner
chshining
Origin version
2.4
Origin PR
#28481
Fix merged
2026-06-27 07:38 UTC
Jira
FLINK-39931

  1. 1.20 1.20.6 Not needed

    Jira FLINK-39931 has no fix commit on any branch: PR 28481 belongs to FLINK-40000 and is a no-behavior refactor.

  2. 2.2 2.2.2 Not needed

    Jira FLINK-39931 has no fix commit on any branch: PR 28481 belongs to FLINK-40000 and is a no-behavior refactor.

  3. 2.3 2.3.1 Not needed

    Jira FLINK-39931 has no fix commit on any branch: PR 28481 belongs to FLINK-40000 and is a no-behavior refactor.

  4. 2.4 2.4.0 Merged

  5. 2.5 master · 2.5.0 Merged

YARNSessionFIFOSecuredITCase.testDetachedMode is flaky: Kerberos keytab log check reads a killed TaskManager's log

Type
Bug
Priority
Minor
Jira status
Resolved
Owner
raminqaf
Origin version
2.4
Origin PR
#28687
Fix merged
2026-07-09 08:38 UTC
Jira
FLINK-40099

  1. 1.20 1.20.6 Not needed

    No backport needed. The failure does not exist on this branch: it pins log4j 2.24.3 (pom.xml log4j.version) and lacks FLINK-39146 (406568ee12e). Per FLINK-40109 and FLINK-40833 the missing keytab line is always in a rolled <file>.1, and that startup rollover needs log4j >= 2.25.2 (apache/logging-log4j2#4362). Reproducer, 100 JVM starts each: JDK 17 log4j 2.25.3 rolled 12, 2.24.3 rolled 0 (14 runs met the rounding condition); JDK 11 2.25.3 rolled 6, 2.24.3 rolled 0. Also #28687 must not be picked alone: it caused FLINK-40109 and #28694 undid it. FLINK-40109 fix versions are 2.3.1 and 2.4.0 only. Failed misc CI logs still readable (1 on 2.2, 4 on 1.20; 13 expired) show no keytab assertion failure.

  2. 2.2 2.2.2 Not needed

    No backport needed. The failure does not exist on this branch: it pins log4j 2.24.3 (pom.xml log4j.version) and lacks FLINK-39146 (406568ee12e). Per FLINK-40109 and FLINK-40833 the missing keytab line is always in a rolled <file>.1, and that startup rollover needs log4j >= 2.25.2 (apache/logging-log4j2#4362). Reproducer, 100 JVM starts each: JDK 17 log4j 2.25.3 rolled 12, 2.24.3 rolled 0 (14 runs met the rounding condition); JDK 11 2.25.3 rolled 6, 2.24.3 rolled 0. Also #28687 must not be picked alone: it caused FLINK-40109 and #28694 undid it. FLINK-40109 fix versions are 2.3.1 and 2.4.0 only. Failed misc CI logs still readable (1 on 2.2, 4 on 1.20; 13 expired) show no keytab assertion failure.

  3. 2.3 2.3.1 Merged

    Net fix on release-2.3 as commit 9ce420f91b7 (FLINK-40109, https://github.com/apache/flink/pull/29340). That PR states #28687 and #28694 need no backport of their own: #28694 undid #28687 and #29324 replaced its poll. YARNSessionFIFOSecuredITCase matches master.

  4. 2.4 2.4.0 Merged

  5. 2.5 master · 2.5.0 Merged

MiscSemanticTests#COALESCE_NESTED_ROW_LEFT_JOIN is unstable

Type
Bug
Priority
Major
Jira status
Resolved
Owner
Sergey Nuyanzin
Origin version
2.4
Origin PR
#28950
Fix merged
2026-08-10 20:11 UTC
Jira
FLINK-40363

  1. 1.20 1.20.6 Not needed

    code path absent on this branch

  2. 2.2 2.2.2 Not needed

    Test absent on release-2.2: CalcTestPrograms has no COALESCE_NESTED_ROW_LEFT_JOIN; it was added by FLINK-39695 (4f8b4a88b1e), which is master-only.

  3. 2.3 2.3.1 Not needed

    Test absent on release-2.3: CalcTestPrograms has no COALESCE_NESTED_ROW_LEFT_JOIN; it was added by FLINK-39695 (4f8b4a88b1e), which is master-only.

  4. 2.4 2.4.0 Merged

  5. 2.5 master · 2.5.0 Merged

AbstractServerTest.testPortRangeSuccess sees an open server connection after the servers are closed

Type
Bug
Priority
Major
Jira status
Open
Owner
psinha
Origin version
2.5
Origin PR
#29660
Fix merged
2026-10-30 18:06 UTC
Jira
FLINK-41247

  1. 1.20 1.20.6 Not needed

    code path absent on this branch

  2. 2.2 2.2.2 Not needed

    code path absent on this branch

  3. 2.3 2.3.1 Merged

  4. 2.4 2.4.0 Merged

  5. 2.5 master · 2.5.0 Merged