hadoop-yarn-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Tsuyoshi OZAWA (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (YARN-659) RMStateStore's removeApplication APIs should just take an applicationId
Date Mon, 14 Oct 2013 10:19:43 GMT

    [ https://issues.apache.org/jira/browse/YARN-659?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13794028#comment-13794028
] 

Tsuyoshi OZAWA commented on YARN-659:
-------------------------------------

[~maysamyabandeh], in the FileSystemRMStateStore case, we can avoid using ApplicationAttemptId
by deleting the directory recursively. The trunk code do this.
One problem is that ZKRMStateStore cannot remove znode recursively. If removeApplication is
only passed ApplicationId, we cannot know applicationAttemptIds to be deleted. We need to
change znode structure. My proposal is as follows:
{code}
RM_APP_ROOT
 |---/(#appId)
 |---/attemptIds_(#appId)/(#applicationAttemptIds)
{code}

Note that #appId and #applicationAttemptIds are evaluated as String value, and attemptIds_
is a just prefix. To do this, we need to change not only removeApplicationState and storeApplicationAttemptState
but also loadRMAppState.

> RMStateStore's removeApplication APIs should just take an applicationId
> -----------------------------------------------------------------------
>
>                 Key: YARN-659
>                 URL: https://issues.apache.org/jira/browse/YARN-659
>             Project: Hadoop YARN
>          Issue Type: Sub-task
>          Components: resourcemanager
>            Reporter: Vinod Kumar Vavilapalli
>            Assignee: Tsuyoshi OZAWA
>
> There is no need to give in the whole state for removal - just an ID should be enough
when an app finishes.



--
This message was sent by Atlassian JIRA
(v6.1#6144)

Mime
View raw message