[
https://issues.apache.org/jira/browse/UIMA-6077?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16872708#comment-16872708
]
Lou DeGenaro commented on UIMA-6077:
------------------------------------
When RM receives state from Orchestrator, it uses the publication's DuccWorkMap
to find job/service/reservation instances to reconstruct the collection of
nodes that may be schedulable. The bug is that the RM considers such instances
that have completed, but it should not. Completed instances may have completed
long ago, and the corresponding node may be long gone. The scheduler would
then attempt to put work on the node which is doomed to failure.
> DUCC stop_ducc -ag should cause RM to stop scheduling there
> -----------------------------------------------------------
>
> Key: UIMA-6077
> URL: https://issues.apache.org/jira/browse/UIMA-6077
> Project: UIMA
> Issue Type: Bug
> Components: DUCC
> Reporter: Lou DeGenaro
> Assignee: Lou DeGenaro
> Priority: Major
> Fix For: future-DUCC
>
>
> When agent is quiesced it will no longer accept work, but RM will continue
> trying to schedule there. This results in unnecessary JP init "failures".
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
