[
https://issues.apache.org/jira/browse/UIMA-3365?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13806312#comment-13806312
]
Lou DeGenaro commented on UIMA-3365:
------------------------------------
4. Add field for Unassigned work items count to I/DuccProcessWorkItems in
transport
a) don't decrement dispatch count when counting lost (bug!)
5. Orchestrator (OR) StateManager must copy Job Driver (JD) reported unassigned
work items
6. Job Driver must keep track of CASes reaped due to missing callback
a) JD should not recycle reaped CASes
b) JD should "dequeue" lost CASes (bug!)
c) JD should add more debugging capability to work item processing
d) JD should have improved error injection for testing of missing/delayed
callbacks (1=queued,2=assigned)
e) JD should ignore callbacks for reaped (lost) CASes
f) JD should ignore queued callback if assigned callback has already arrived
g) JD should allow assigned callback even if assigned callback is missing (or
late), due to JD threading
7. Webserver (WS) must properly display dispatch and unassigned counts (bug!)
code delivered.
> DUCC job driver (JD) ERROR client.WorkItemListener onBeforeProcessCASHandler
> N/A seqNo:null casId:1602519189 java.lang.NullPointerException
> -----------------------------------------------------------------------------------------------------------------------------------------------
>
> Key: UIMA-3365
> URL: https://issues.apache.org/jira/browse/UIMA-3365
> Project: UIMA
> Issue Type: Bug
> Components: DUCC
> Reporter: Lou DeGenaro
> Assignee: Lou DeGenaro
> Fix For: 1.0-Ducc
>
>
> Job 93735.
> It seems that the first callback expected by JD for work item 659 comes.
> However the second callback does not arrive within 60 seconds, so the work
> items is reaped. Then, 2 seconds later the missing callback finally (and
> unexpectedly) arrives causing the NPE in JD.
> The Job seems to actually run to completion, but is flagged with lost work
> items. This makes sense, since if the JP does get the work item it will
> process it. However, the JD will begin queuing "extra" work items since it
> does not understand that such work items are not really lost. This may
> increasingly aggravate the problem by having the JD put out to queue more
> work items than the present JP capacity can handle.
--
This message was sent by Atlassian JIRA
(v6.1#6144)
