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 "Steve Loughran (JIRA)" <ji...@apache.org> on 2016/08/27 11:11:20 UTC
[jira] [Commented] (YARN-5570) NodeManager blocks kill attempts
during startup if ResourceManager is down
[ https://issues.apache.org/jira/browse/YARN-5570?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15441279#comment-15441279 ]
Steve Loughran commented on YARN-5570:
--------------------------------------
YARN-679 tries to be a single entry point for anything that implements Service/subclasses AbstractService & adds extra support for an interface to take command line arguments.
It does double-exit escalation: first control-C triggers System.exit() and a shutdown with the hooks running; a second SIGINT triggers escalation and a hard System.halt. This addresses the issue that when IPC clients block in the retry/attempt cycle, they can deadlock on the shutdown code
> NodeManager blocks kill attempts during startup if ResourceManager is down
> --------------------------------------------------------------------------
>
> Key: YARN-5570
> URL: https://issues.apache.org/jira/browse/YARN-5570
> Project: Hadoop YARN
> Issue Type: Bug
> Affects Versions: 2.7.3
> Reporter: Andrew Wang
>
> Found while testing a 2.7.3 RC:
> {quote}
> * If you start the NodeManager and not the RM, the NM has a handler for SIGTERM and SIGINT that blocked my Ctrl-C and kill attempts during startup. I had to kill -9 it.
> {quote}
> [~stevel@apache.org] tells me that this is potentially addressed by YARN-679, though that's a large patch.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: yarn-issues-unsubscribe@hadoop.apache.org
For additional commands, e-mail: yarn-issues-help@hadoop.apache.org