Mailing-List: contact yarn-issues-help@hadoop.apache.org; run by ezmlm
Precedence: bulk
Reply-To: yarn-issues@hadoop.apache.org
Date: Tue, 24 Feb 2015 06:04:12 +0000 (UTC)
From: "Xuan Gong (JIRA)" <jira@apache.org>
To: yarn-issues@hadoop.apache.org
Message-ID: <JIRA.12777056.1424756014000.119194.1424757851989@Atlassian.JIRA>
In-Reply-To: <JIRA.12777056.1424756014000@Atlassian.JIRA>
References: <JIRA.12777056.1424756014000@Atlassian.JIRA>
 <JIRA.12777056.1424756014781@arcas>
Subject: [jira] [Commented] (YARN-3246) Add a CLEANUP state in RMApp to
 track the status of clean-up container.
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: 7bit


    [ https://issues.apache.org/jira/browse/YARN-3246?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14334458#comment-14334458 ] 

Xuan Gong commented on YARN-3246:
---------------------------------

More details:
We will add a cleanup state for RMApp. If the clean-up container is enabled, RMApp will transits into cleanup state from Finished/Failed/Killed state. In the clean-up state, we will keep track on how long this clean-up container runs (kill it if it is out of time-out period), how many times the container has retried, whether it is finished, how to launch this container.

A couple of cover case that I can think:
* NM non-work persevering restart: We will kill all the containers in that NM, the clean-up container resource will not be released immediately, and will be re-use to launch the clean-up container again if it is not out of maxRetry.
* NM work persevering restart: It is fine. The clean-up container will be re-launched in NM.
* RM non-work persevering restart/RM work persevering restart: this will be tricky.
** We could ignore it
** OR, we could add a separate clean-up status : disabled, finished, failed, killed, running. Based on the different the clean up status, we could decide whether the clean-up container will be recovered/relaunched or not.

> Add a CLEANUP state in RMApp to track the status of clean-up container.
> -----------------------------------------------------------------------
>
>                 Key: YARN-3246
>                 URL: https://issues.apache.org/jira/browse/YARN-3246
>             Project: Hadoop YARN
>          Issue Type: Sub-task
>          Components: resourcemanager
>            Reporter: Xuan Gong
>            Assignee: Xuan Gong
>
> The clean-up container will be launched by RM, and we need to keep track the status for it, such as when to kill the container because of time-out, whether need to relaunch the clean-up container, etc


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