Martijn Visser created FLINK-39845:
--------------------------------------
Summary: SavepointITCase.testStopWithSavepointFailsOverToSavepoint
fails on AdaptiveScheduler after JUnit5 migration
Key: FLINK-39845
URL: https://issues.apache.org/jira/browse/FLINK-39845
Project: Flink
Issue Type: Bug
Components: Tests
Affects Versions: 2.3.0
Reporter: Martijn Visser
Assignee: Martijn Visser
SavepointITCase.testStopWithSavepointFailsOverToSavepoint fails
deterministically in the test_cron_adaptive_scheduler tests leg on master only.
It has been red every nightly since 2026-03-21, the first nightly after the
JUnit5 migration FLINK-39124 rewrote this file. It passes under the default
scheduler, and passes under the adaptive scheduler on release-2.0/2.1
pre-migration assertion).
{code:java}
java.lang.AssertionError:
Expecting a throwable with cause being an instance of:
org.apache.flink.runtime.scheduler.stopwithsavepoint.StopWithSavepointStoppingException
but was an instance of:
org.apache.flink.util.FlinkException
...
at
org.apache.flink.test.checkpointing.SavepointITCase.testStopWithSavepointFailsOverToSavepoint(SavepointITCase.java:324)
{code}
Root cause: the migration replaced a cause-chain search
(ExceptionUtils.assertThrowable → findThrowable) with a direct-cause check
(assertThatThrownBy(...).hasCauseInstanceOf(StopWithSavepointStoppingException.class)).
Under the AdaptiveScheduler, StopWithSavepoint.onLeave() wraps the expected
StopWithSavepointStoppingException inside a FlinkException ("Stop with
savepoint operation could not be completed."), so it is no longer the direct
cause. The runtime is unchanged (StopWithSavepoint.java is byte-identical).
Fix: restore the chain search via FlinkAssertions.anyCauseMatches, so the test
passes under both schedulers.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
