Jim Challenger created UIMA-3378:
------------------------------------
Summary: DUCC RM Race if job canceled quickly
Key: UIMA-3378
URL: https://issues.apache.org/jira/browse/UIMA-3378
Project: UIMA
Issue Type: Bug
Components: DUCC
Affects Versions: 1.0-Ducc
Reporter: Jim Challenger
Assignee: Jim Challenger
Priority: Minor
Fix For: 1.0-Ducc
If a job is canceled immediately after it is submitted it can arrive at RM in
state WaitingForResources AND in the same scheduling epoch, a message indicated
the job is canceled can arrive. WaitingForResources usually causes a job to be
entered into the RM scheduling lists, but Completed causes it to be deleted and
overrides the WaitingForResources. The bug is that Completed does not check to
see if the job ever did actually get entered, and bam! NPE. The fix is simple,
in Scheduler. processCompletion, if the indicated job isn't found in the
scheduling list, just ignore and return.
--
This message was sent by Atlassian JIRA
(v6.1#6144)
