Mailing-List: contact yarn-issues-help@hadoop.apache.org; run by ezmlm
Precedence: bulk
Date: Tue, 29 Aug 2017 19:23:00 +0000 (UTC)
From: "Gour Saha (JIRA)" <jira@apache.org>
To: yarn-issues@hadoop.apache.org
Message-ID: <JIRA.13019281.1478626454000.151038.1504034580175@Atlassian.JIRA>
In-Reply-To: <JIRA.13019281.1478626454000@Atlassian.JIRA>
References: <JIRA.13019281.1478626454000@Atlassian.JIRA> <JIRA.13019281.1478626454737@jira-lw-us.apache.org>
Subject: [jira] [Commented] (YARN-5855) DELETE call sometimes returns
 success when app is not deleted
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: 7bit
archived-at: Tue, 29 Aug 2017 19:23:07 -0000


    [ https://issues.apache.org/jira/browse/YARN-5855?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16145949#comment-16145949 ] 

Gour Saha commented on YARN-5855:
---------------------------------

DELETE being synchronous makes sense to me too. 

Only one point to think on is - since the intent of the app-owner is to destroy the app, he/she probably doesn't care if the app is stopped elegantly or not (as long as log-aggregation is done successfully after the app dies). Now in worst case the API can take up to 10 secs to respond. I think 10 secs is too high. Do you think we should reduce it? My suggestion is 2 secs. 

> DELETE call sometimes returns success when app is not deleted
> -------------------------------------------------------------
>
>                 Key: YARN-5855
>                 URL: https://issues.apache.org/jira/browse/YARN-5855
>             Project: Hadoop YARN
>          Issue Type: Sub-task
>            Reporter: Billie Rinaldi
>            Assignee: Gour Saha
>
> Looking into this issue with [~gsaha], we noticed that multiple things can contribute to an app continuing to run after a DELETE call, which consists of a stop and a destroy operation. One problem is that the stop call is asynchronous unless a force flag is set. Without the force flag, a message is sent to the AM and success is returned, and with the flag yarnClient.killRunningApplication is called. (There is also an option to wait for a fixed amount of time for the app to stop before returning, but DELETE is not setting this option and force is preferable in this case.) The other issue is that the destroy operation is attempted in a loop, but if the number of retries is exceeded the call returns a 204 response.


--
This message was sent by Atlassian JIRA
(v6.4.14#64029)

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