[jira] [Commented] (UIMA-5057) DUCC Orchestrator (OR) handle down JD node

["Lou DeGenaro (JIRA)" <de...@uima.apache.org>]

    [ https://issues.apache.org/jira/browse/UIMA-5057?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15421561#comment-15421561 ] 

Lou DeGenaro commented on UIMA-5057:
------------------------------------

Whenever Orchestrator does a publication it calls org.apache.uima.ducc.orchestrator.jd.scheduler.JdScheduler.handle(IDuccWorkMap dwm).  Here we consider expanding or contracting the JD share pool.  But we fail to consider that an already allocated JD node has gone down.   We add a new call to monitor() to consider this possibility.

Note that we rely upon database updated by Resource Manager to determine node downness.

> DUCC Orchestrator (OR) handle down JD node
> ------------------------------------------
>
>                 Key: UIMA-5057
>                 URL: https://issues.apache.org/jira/browse/UIMA-5057
>             Project: UIMA
>          Issue Type: Bug
>          Components: DUCC
>            Reporter: Lou DeGenaro
>            Assignee: Lou DeGenaro
>             Fix For: 2.2.0-Ducc
>
>
> If a node that goes down happens to be a JD node then:
> 1. cancel any running jobs whose JD was assigned there, and
> 2. allocate a new JD node, if needed and possible
> Currently when the Agent hosting the JD of running Job is killed the Job hangs because the Agent is not there to carry out the Orchestrator directed clean-up procedures.



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