[ 
https://issues.apache.org/jira/browse/YARN-4846?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15251230#comment-15251230
 ] 

Sunil G commented on YARN-4846:
-------------------------------

Thanks [~bibinchundatt] and [~leftnoteasy].
As {{editpolicy}} was called twice at same time, problem maker was below code 
{{context.getQueueByPartition(queueName, RMNodeLabelsManager.NO_LABEL)}}. This 
was returning null. I think in my earlier comment, I meant the return value of 
this method will be null rather queueName becoming null. In 
{{containerBasedPreemptOrKill}}, {{queueToPartitions}} will be cleared in 
second editPolicy call.

Root cause seems correct to me, two calls to {{editpolicy}} is the real issue. 
Good analysis :). And new patch from [~leftnoteasy] seems fixing the Assert 
issue correctly. For fixing editPolicy call, adding synchronized looks good. 
Along with a big preemption interval timeout, this will solve the test failures.
[~bibinchundatt], may be can  you update a new patch based on [~leftnoteasy]'s 
updated patch.

> Random failures for 
> TestCapacitySchedulerPreemption#testPreemptionPolicyShouldRespectAlreadyMarkedKillableContainers
> --------------------------------------------------------------------------------------------------------------------
>
>                 Key: YARN-4846
>                 URL: https://issues.apache.org/jira/browse/YARN-4846
>             Project: Hadoop YARN
>          Issue Type: Sub-task
>            Reporter: Bibin A Chundatt
>            Assignee: Bibin A Chundatt
>         Attachments: 0001-YARN-4846.patch, 0002-YARN-4846.patch, 
> YARN-4846-update-PCPP.patch
>
>
> {noformat}
> java.lang.AssertionError: expected:<1> but was:<0>
>       at org.junit.Assert.fail(Assert.java:88)
>       at org.junit.Assert.failNotEquals(Assert.java:743)
>       at org.junit.Assert.assertEquals(Assert.java:118)
>       at org.junit.Assert.assertEquals(Assert.java:555)
>       at org.junit.Assert.assertEquals(Assert.java:542)
>       at 
> org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.TestCapacitySchedulerPreemption.testPreemptionPolicyShouldRespectAlreadyMarkedKillableContainers(TestCapacitySchedulerPreemption.java:473)
> {noformat}
> https://builds.apache.org/job/PreCommit-YARN-Build/10826/testReport/org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity/TestCapacitySchedulerPreemption/testPreemptionPolicyShouldRespectAlreadyMarkedKillableContainers/



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

Reply via email to