You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@uima.apache.org by "Jerry Cwiklik (JIRA)" <de...@uima.apache.org> on 2017/09/28 20:01:07 UTC

[jira] [Commented] (UIMA-5587) DUCC Process Manager (PM) takes OOM on Apache DUCC demo

    [ https://issues.apache.org/jira/browse/UIMA-5587?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16184751#comment-16184751 ] 

Jerry Cwiklik commented on UIMA-5587:
-------------------------------------

I've checked Camel and AMQ docs and verified that by default Camel uses a single thread in a JMS based route. This is great EXCEPT, the default prefetch in AMQ is 1000. So this explains OOM in the PM when there is a burst of msgs from the OR. Prefetch controls how many messages AMQ code (in the PM) will fetch from a broker. Even though there is a single thread to process msgs, the large prefetch brings into memory many messages and they just sit there waiting in the process heap. 
Am working on a fix for this. 

> DUCC Process Manager (PM) takes OOM on Apache DUCC demo
> -------------------------------------------------------
>
>                 Key: UIMA-5587
>                 URL: https://issues.apache.org/jira/browse/UIMA-5587
>             Project: UIMA
>          Issue Type: Bug
>          Components: DUCC
>            Reporter: Lou DeGenaro
>            Assignee: Jerry Cwiklik
>            Priority: Minor
>             Fix For: future-DUCC
>
>
> Apache DUCC demo has PM that runs with reduced memory -Xmx200M
> On at least 2 occasions PM has taken OOM.  Core dumps have resulted that indicate that PM is flooded with multiple OR publications that is supposed to be handled by "BlastGuard"?
> Investigate what storage is being used for and if multiple OR messages are being processed in parallel.  Perhaps Camel should be configured to process OR message sequentially if it isn't already.



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)