[
https://issues.apache.org/jira/browse/SPARK-960?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Josh Rosen updated SPARK-960:
-----------------------------
Assignee: Sean Owen
> JobCancellationSuite "two jobs sharing the same stage" is broken
> ----------------------------------------------------------------
>
> Key: SPARK-960
> URL: https://issues.apache.org/jira/browse/SPARK-960
> Project: Spark
> Issue Type: Bug
> Components: Spark Core
> Affects Versions: 0.8.1, 0.9.0
> Reporter: Mark Hamstra
> Assignee: Sean Owen
> Fix For: 1.3.0
>
>
> This test doesn't work as it appears to be intended since the map tasks can
> never acquire sem2. The simplest way to demonstrate this is to comment out
> f1.cancel() in the future. I believe the intention is that f1 and f2 would
> then complete normally; but they won't. Instead, both jobs block, waiting on
> sem2. It doesn't look like closing over Semaphores works even in a Local
> context, since sem2.hashCode() is different in each of f1, f2 and in the
> future containing f1.cancel, so the map jobs never see the sem2.release(10)
> in the future.
> Instead, the test only completes because all of the stages (the two final
> stages and the common dependent stage) get cancelled and aborted. When job
> <--> stage dependencies are fully accounted for and job cancellation changed
> so that f1.cancel does not abort the common stage, then this test can never
> finish since it then becomes hung waiting on sem2.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
