[jira] [Commented] (FLINK-2670) Unstable CombineTaskTest

["Stephan Ewen (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/FLINK-2670?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14768901#comment-14768901 ] 

Stephan Ewen commented on FLINK-2670:
-------------------------------------

I cannot see any issue with this test. It looks absolutely valid. There is also no deadlock, the code is in the test's clean-up phase.

There is a concurrent modification during concurrent shutdown/cancel of the job, but that is perfectly valid and tolerated and handled correctly.

I cannot see why this should take longer than 5 minutes (when the test runs for less than one second). Two cases could happen:

1.  The code gets stuck in the native method that fills in the stack trace. Whyever would it do that, I do not know.

2.  It is just a coincidence that the code is there (in cleanup) after 5 minutes and the code simply took forever to execute the test. I find this highly unlikely, as the test has no busy loops, it emits slowly some data (no tight loop, sleeps for a msec regularly) and cancels after 500 msecs.

No idea what to do with this test, it looks almost like a quirk in Travis/Linux/JVM...

> Unstable CombineTaskTest
> ------------------------
>
>                 Key: FLINK-2670
>                 URL: https://issues.apache.org/jira/browse/FLINK-2670
>             Project: Flink
>          Issue Type: Bug
>          Components: Tests
>            Reporter: Matthias J. Sax
>            Assignee: Stephan Ewen
>            Priority: Critical
>              Labels: test-stability
>
> Fails with
> {noformat}
> ==============================================================================
> Maven produced no output for 300 seconds.
> ==============================================================================
> {noformat}
> https://travis-ci.org/apache/flink/jobs/80344487



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)