[jira] [Created] (YARN-12001) estTimelineReaderHBaseDown leaks TimelineReaderServer storage monitors; serviceStop() skips child services if the web server fails to stop

["Jose Luis López (Jira)" <ji...@apache.org>]

Jose Luis López created YARN-12001:
--------------------------------------

             Summary: estTimelineReaderHBaseDown leaks TimelineReaderServer storage monitors; serviceStop() skips child services if the web server fails to stop
                 Key: YARN-12001
                 URL: https://issues.apache.org/jira/browse/YARN-12001
             Project: Hadoop YARN
          Issue Type: Bug
          Components: test, timelineserver
            Reporter: Jose Luis López


TestTimelineReaderHBaseDown starts a TimelineReaderServer in all five of its tests and never stops one. Each run leaks the TimelineStorageMonitor the server schedules: non-daemon threads that keep polling, every TIMELINE_SERVICE_READER_STORAGE_MONITOR_INTERVAL_MS, an HBase minicluster the test has already shut down. hadoop-yarn-server-timelineservice-hbase-tests runs with forkCount 0, so these threads pile up in Maven's JVM across all eleven test classes in the module.

Stopping the server isn't enough on its own. TimelineReaderServer.serviceStop() calls readerWebServer.stop() before super.serviceStop(). If the web server fails to stop, the exception skips super.serviceStop(), so the child services (the reader and the storage monitor's executor) are never stopped.

Fix:
* TimelineReaderServer.serviceStop(): call super.serviceStop() in a finally. Move the web-app stop into stopTimelineReaderWebApp() (@VisibleForTesting).
* TestTimelineReaderHBaseDown: stop the server in a finally with ServiceOperations.stopQuietly() in all five tests, as every other test in the module already does. testTimelineReaderRecoversAfterHBaseReturns also waits until no regions are in transition before shutting down. configure() replaces hbase.regionserver.abort.timeout.task with a no-op TimerTask, because HBase's default task halts the JVM, and with forkCount 0 that JVM is Maven's.
* TestTimelineReaderServer: new test testChildServicesStoppedWhenWebAppStopFails.

Split out of HADOOP-19979. Reviewed earlier as part of https://github.com/apache/hadoop/pull/8717.



--
This message was sent by Atlassian Jira
(v8.20.10#820010)

---------------------------------------------------------------------
To unsubscribe, e-mail: yarn-dev-unsubscribe@hadoop.apache.org
For additional commands, e-mail: yarn-dev-help@hadoop.apache.org