[
https://issues.apache.org/jira/browse/BEAM-1140?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15794073#comment-15794073
]
Thomas Weise commented on BEAM-1140:
------------------------------------
I was able to reproduce on a VM.
{code}
Failed tests:
ParDoBoundTranslatorTest.testMultiOutputParDoWithSideInputs:307
expected:<[processing: -42: [11, 222], processing: 3: [11, 222], processing:
666: [11, 222]]> but was:<[processing: -42: [11, 222], processing: 3: [11,
222]]>
{code}
When the test fails, input 666 was read, but never processed by the
MultiOutputParDo, because the topology gets terminated before. I cannot see any
issue in the runner code, so need to look at this further at the Apex level. I
suspect it is the same issue that also manifests in other flaky tests.
> Apex ParDoBoundTranslatorTest.testMultiOutputParDoWithSideInputs is flaky
> -------------------------------------------------------------------------
>
> Key: BEAM-1140
> URL: https://issues.apache.org/jira/browse/BEAM-1140
> Project: Beam
> Issue Type: Bug
> Components: runner-apex
> Reporter: Eugene Kirpichov
> Assignee: Thomas Weise
>
> The test fails about one in 10-20 times or so when run in "Run until failure"
> mode in IntelliJ (I did it 3 times; once it failed on the 3rd run, once on
> the 15th, once on the 40th).
> It's making precommits unstable (e.g. currently
> https://github.com/apache/incubator-beam/pull/1565 failed with this test
> error).
> Unfortunately I don't understand enough about the Apex runner to understand
> what's going on and fix this myself...
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
