You are viewing a plain text version of this content. The canonical link for it is here.
Posted to yarn-issues@hadoop.apache.org by "Zhijie Shen (JIRA)" <ji...@apache.org> on 2013/12/02 19:04:36 UTC

[jira] [Commented] (YARN-1446) Change killing application to wait until state store is done

    [ https://issues.apache.org/jira/browse/YARN-1446?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13836740#comment-13836740 ] 

Zhijie Shen commented on YARN-1446:
-----------------------------------

Hm... I see. The patch is generally good. I've the following comments.

1. It's better to use KillApplicationRequest.newInstance
{code}
+    KillApplicationRequest req =
+        Records.newRecord(KillApplicationRequest.class);
{code}

2. Fix the grammar bellow
{code}
+  // reaches killed state.and also check that attempt state is saved before app
{code}

3. Ago, app can be killed at FINISHING. With the following change in ClientRMSerivce, it seems to be no longer applicable.
{code}
+    if (application.isAppSafeToTerminate()) {
+      return KillApplicationResponse.newInstance(true);
+    }
{code}

4. Instead of logging the killing info every 100ms, how about doing something similar in YarnClientImpl#submitApplication?

5. Do you have an estimation on the number of KILLING requests that is sent before KILLING is succeeded?

6. Does this ticket a bit overlap YARN-261? After the change, it is actually killing the attempt instead of the app, but we doesn't allow retry here.

> Change killing application to wait until state store is done
> ------------------------------------------------------------
>
>                 Key: YARN-1446
>                 URL: https://issues.apache.org/jira/browse/YARN-1446
>             Project: Hadoop YARN
>          Issue Type: Sub-task
>          Components: resourcemanager
>            Reporter: Jian He
>            Assignee: Jian He
>         Attachments: YARN-1446.1.patch, YARN-1446.1.patch, YARN-1446.1.patch
>
>
> When user kills an application, it should wait until the state store is done with saving the killed status of the application. Otherwise, if RM crashes in the middle between user killing the application and writing the status to the store, RM will relaunch this application after it restarts.



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