You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@flink.apache.org by "Martijn Visser (Jira)" <ji...@apache.org> on 2026/06/12 11:33:00 UTC
[jira] [Created] (FLINK-39917) JobMasterTriggerSavepointITCase.testDoNotCancelJobIfSavepointFails: "Disconnect job manager" log assertion races the async JM->RM disconnect
Martijn Visser created FLINK-39917:
--------------------------------------
Summary: JobMasterTriggerSavepointITCase.testDoNotCancelJobIfSavepointFails: "Disconnect job manager" log assertion races the async JM->RM disconnect
Key: FLINK-39917
URL: https://issues.apache.org/jira/browse/FLINK-39917
Project: Flink
Issue Type: Bug
Components: Runtime / Coordination, Tests
Reporter: Martijn Visser
Assignee: Martijn Visser
https://dev.azure.com/apache-flink/apache-flink/_build/results?buildId=75865&view=results (leg: test_cron_azure tests)
{code}
06:17:51.991 [ERROR] org.apache.flink.runtime.jobmaster.JobMasterTriggerSavepointITCase.testDoNotCancelJobIfSavepointFails(ClusterClient) -- Time elapsed: 0.394 s <<< FAILURE!
java.lang.AssertionError:
[not all expected events logged by org.apache.flink.runtime.resourcemanager.StandaloneResourceManager, logged:
[... Message=Registering job manager ..., ... Message=Registered job manager ...]]
Expecting empty but was: [Disconnect job manager .*]
at org.apache.flink.util.JobIDLoggingUtil.assertKeyPresent(JobIDLoggingUtil.java:98)
at org.apache.flink.runtime.jobmaster.JobMasterTriggerSavepointITCase.verifyJobIdIsLogged(JobMasterTriggerSavepointITCase.java:280)
{code}
Root cause: {{waitForDisconnect}} cancels the job and waits for the client-visible {{CANCELED}} status, then {{verifyJobIdIsLogged}} asserts that {{StandaloneResourceManager}} logged "Disconnect job manager ...". The JobMaster disconnects from the ResourceManager asynchronously during shutdown, *after* the job reports CANCELED. The run logs confirm the window: job CANCELED at 06:17:51,115, JobMaster began stopping at 06:17:51,136, and the assertion ran in between, capturing only the "Registering/Registered job manager" events.
Not the same failure as FLINK-37821 (closed), which addressed a different signal in this test.
Proposed fix: in {{waitForDisconnect}}, after the CANCELED wait, additionally wait until the RM has actually logged the disconnect event before returning. No assertion change.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)