[
https://issues.apache.org/jira/browse/BEAM-3599?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16354237#comment-16354237
]
Kenneth Knowles commented on BEAM-3599:
---------------------------------------
This is now sickbayed
> kinesis.ShardReadersPoolTest.shouldInterruptKinesisReadingAndStopShortly is
> flaky
> ---------------------------------------------------------------------------------
>
> Key: BEAM-3599
> URL: https://issues.apache.org/jira/browse/BEAM-3599
> Project: Beam
> Issue Type: Bug
> Components: sdk-java-extensions
> Reporter: Thomas Groh
> Assignee: Kenneth Knowles
> Priority: Critical
> Labels: flake
>
> This appears to be unavoidable due to the construction of the test - the
> amount of realtime that the thread waits is asserted to be less than one
> second, but as the class doesn't consume a {{Ticker}} (or related way to
> track elapsed time), we have no way of deterministically controlling that
> value within the test.
>
> Example failure:
> [https://builds.apache.org/job/beam_PreCommit_Java_GradleBuild/1738/testReport/junit/org.apache.beam.sdk.io.kinesis/ShardReadersPoolTest/shouldInterruptKinesisReadingAndStopShortly/]
>
> {{java.lang.AssertionError: Expecting: <4169L> to be less than: <1000L> at
> org.apache.beam.sdk.io.kinesis.ShardReadersPoolTest.shouldInterruptKinesisReadingAndStopShortly(ShardReadersPoolTest.java:159)
> at sun.reflect.NativeMethodAccessorImpl.invoke0(Native Method)}}
>
> The test could also do with a more precise error message (it looks like too
> many elements were received based on the error message, rather than the
> actual "took too long" failure.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
