[
https://issues.apache.org/jira/browse/AMQ-4789?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13789624#comment-13789624
]
Graham Leggett commented on AMQ-4789:
-------------------------------------
Obviously if we could reproduce it easily it would help, but we're seeing this
manifest itself on a live system after running for many hours, so no we don't
have a test case.
To explain the losing state, the leak eventually causes an out of memory
condition, which causes garbage collection to take forever, in turn triggering
the "kill -9" behaviour of the java service wrapper, in turn causing the loss
of data.
I have a 2.6GB heap file and can run various reports on it to show the objects
being leaked.
> Slow leak: org.apache.activemq.broker.jmx.ManagedTransportConnection
> --------------------------------------------------------------------
>
> Key: AMQ-4789
> URL: https://issues.apache.org/jira/browse/AMQ-4789
> Project: ActiveMQ
> Issue Type: Bug
> Components: JMX
> Affects Versions: 5.8.0
> Environment: INFO | jvm 1 | 2013/10/07 10:17:29 | Java Runtime:
> Sun Microsystems Inc. 1.6.0_24
> /usr/lib/jvm/java-1.6.0-openjdk-1.6.0.0.x86_64/jre
> INFO | jvm 1 | 2013/10/07 10:17:29 | Heap sizes: current=83008k
> free=78872k max=4190080k
> Reporter: Graham Leggett
> Labels: leak
> Attachments: activemq-leak-detail.tiff, activemq-leak.tiff
>
>
> When activemq is configured with an amqp transport and SSL, and where clients
> make short lived polled connections to the queue, a slow leak exists that
> kills the server within about 12 hours or so (for us) with the following:
> java.lang.OutOfMemoryError: GC overhead limit exceeded
> Obtaining a heap dump in this state reveals that about 3/4 of our heap is
> used by the org.apache.activemq.broker.jmx.ManagedTransportConnection as per
> the attached screenshots.
--
This message was sent by Atlassian JIRA
(v6.1#6144)
