hadoop-yarn-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Eric Payne (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (YARN-8774) Memory leak when CapacityScheduler allocates from reserved container with non-default label
Date Wed, 19 Sep 2018 22:05:00 GMT

    [ https://issues.apache.org/jira/browse/YARN-8774?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16621254#comment-16621254

Eric Payne commented on YARN-8774:

Thanks [~Tao Yang] for reporting this issue and providing a patch. The fix looks good to me.
I will finish reviewing the unit test tomorrow.

> Memory leak when CapacityScheduler allocates from reserved container with non-default
> -------------------------------------------------------------------------------------------
>                 Key: YARN-8774
>                 URL: https://issues.apache.org/jira/browse/YARN-8774
>             Project: Hadoop YARN
>          Issue Type: Bug
>          Components: capacityscheduler
>    Affects Versions: 3.2.0, 2.8.5
>            Reporter: Tao Yang
>            Assignee: Tao Yang
>            Priority: Critical
>         Attachments: YARN-8774.001.patch
> The cause is that the RMContainerImpl instance of reserved container lost its node label
expression, when scheduler reserves containers for non-default node-label requests, it will
be wrongly added into LeafQueue#ignorePartitionExclusivityRMContainers and never be removed.
> To reproduce this memory leak:
> (1) create reserved container
> RegularContainerAllocator#doAllocation:  create RMContainerImpl instanceA (nodeLabelExpression="")
> LeafQueue#allocateResource:  RMContainerImpl instanceA is put into  LeafQueue#ignorePartitionExclusivityRMContainers
> (2) allocate from reserved container
> RegularContainerAllocator#doAllocation: create RMContainerImpl instanceB (nodeLabelExpression="test-label")
> (3) From now on, RMContainerImpl instanceA will be left in memory (be kept in LeafQueue#ignorePartitionExclusivityRMContainers)
forever until RM restarted

This message was sent by Atlassian JIRA

To unsubscribe, e-mail: yarn-issues-unsubscribe@hadoop.apache.org
For additional commands, e-mail: yarn-issues-help@hadoop.apache.org

View raw message