Arvid Heise created FLINK-23807:
-----------------------------------
Summary: Use metrics to detect restarts in
MiniClusterTestEnvironment#triggerTaskManagerFailover
Key: FLINK-23807
URL: https://issues.apache.org/jira/browse/FLINK-23807
Project: Flink
Issue Type: Bug
Components: Connectors / Common
Reporter: Arvid Heise
Fix For: 1.14.0
{{MiniClusterTestEnvironment#triggerTaskManagerFailover}} checks the job status
to detect a restart
{noformat}
terminateTaskManager();
CommonTestUtils.waitForJobStatus(
jobClient,
Arrays.asList(JobStatus.FAILING, JobStatus.FAILED,
JobStatus.RESTARTING),
Deadline.fromNow(Duration.ofMinutes(5)));
afterFailAction.run();
startTaskManager();
{noformat}
However, `waitForJobStatus` polls every 100ms while the restart can happen
within 10ms and thus can easily miss the actual restart and wait forever (or
when the next restart happens because slots are missing).
We should rather use the metric `numRestarts`, check before the induced error,
and wait until the counter increased.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
