[
https://issues.apache.org/jira/browse/AMQ-6940?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16417567#comment-16417567
]
Johno Crawford commented on AMQ-6940:
-------------------------------------
I created a canary build of the broker and removed the dispatched array from
org.apache.activemq.broker.region.TopicSubscription , the heap has more or less
dropped to the old baseline but I will follow up later as it still may be too
early to tell. The dispatched array is cleaned via
org.apache.activemq.broker.region.TopicSubscription#updateStatsOnAck(org.apache.activemq.command.MessageAck)
which is invoked every x thousand messages, the problem here is that in our
cluster we have roughly 7300 consumers and 900 producers which is clearly not
going to scale..
> Significant memory footprint increase from 5.8.0 to 5.15.3
> ----------------------------------------------------------
>
> Key: AMQ-6940
> URL: https://issues.apache.org/jira/browse/AMQ-6940
> Project: ActiveMQ
> Issue Type: Bug
> Components: Broker
> Affects Versions: 5.15.3
> Environment: Java jdk1.8.0_144
> OS Linux 4.4.19-29.55.amzn1.x86_64
> Reporter: Johno Crawford
> Priority: Major
> Attachments: activemq.xml, amq_heap2.png, messageid_analysis.png
>
>
> Recently we upgraded from 5.8.0 to 5.15.3 on one of our environments and
> found the broker running out of memory with the old broker configurations.
> I suspect this may be related to the new inflight message size calculation
> but haven't been able to create a reproducer just yet.
> Broker JVM running with opts
> {code:java}
> -Xss256k -Xms600m -Xmx600m -Xmn128m -Dfile.encoding=ISO-8859-1
> -Dlog4j.watchDelay=60000 -XX:+PrintGCDetails -XX:+PrintGCTimeStamps
> -XX:+PrintGCDateStamps -XX:+PrintPromotionFailure -XX:PrintFLSStatistics=1
> -XX:+UseConcMarkSweepGC -XX:+UseParNewGC -XX:+PrintClassHistogram -verbose:gc
> -Dsun.rmi.dgc.client.gcInterval=7200000
> -Dsun.rmi.dgc.server.gcInterval=7200000
> {code}
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
