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 "Li Lu (JIRA)" <ji...@apache.org> on 2015/10/29 02:41:27 UTC

[jira] [Commented] (YARN-4313) Race condition in MiniMRYarnCluster when getting history server address

    [ https://issues.apache.org/jira/browse/YARN-4313?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14979640#comment-14979640 ] 

Li Lu commented on YARN-4313:
-----------------------------

Thanks [~jianhe] for the work. I have one quick w.r.t the atomic boolean in the patch. If we're just using atomic read/write for the boolean flag, I think a volatile boolean would suffice. We're not actually using the "fancy" features of atomic boolean in this fix. The other part of the patch LGTM. 

> Race condition in MiniMRYarnCluster when getting history server address
> -----------------------------------------------------------------------
>
>                 Key: YARN-4313
>                 URL: https://issues.apache.org/jira/browse/YARN-4313
>             Project: Hadoop YARN
>          Issue Type: Bug
>            Reporter: Jian He
>            Assignee: Jian He
>         Attachments: YARN-4313.1.patch
>
>
> Problem in this place when waiting for JHS to be started
> {code}
>         new Thread() {
>           public void run() {
>             historyServer.start();
>           };
>         }.start();
>         while (historyServer.getServiceState() == STATE.INITED) {
>           LOG.info("Waiting for HistoryServer to start...");
>           Thread.sleep(1500);
>         }
> {code}
> The service state is updated before the service is actually started. See AbstractServic#start.  So it's possible that when the while loop breaks, the service is not yet started. 



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