[
https://issues.apache.org/jira/browse/AMQ-7126?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16736533#comment-16736533
]
ASF subversion and git services commented on AMQ-7126:
------------------------------------------------------
Commit 9845853b6860619588e50b7a56a2f80af024d28e in activemq's branch
refs/heads/activemq-5.15.x from Jamie Mark Goodyear
[ https://git-wip-us.apache.org/repos/asf?p=activemq.git;h=9845853 ]
[AMQ-7126] Improvement to perf of 5266Test
> Prevent OOM when recovering KahaDB and memory space is insufficient to load
> full page
> -------------------------------------------------------------------------------------
>
> Key: AMQ-7126
> URL: https://issues.apache.org/jira/browse/AMQ-7126
> Project: ActiveMQ
> Issue Type: Bug
> Affects Versions: 5.15.8
> Reporter: Heath Kesler
> Assignee: Jamie goodyear
> Priority: Critical
> Fix For: 5.16.0, 5.15.9
>
> Attachments: OOM.patch, amq7126.patch
>
>
> We have some brokers that are having OOM issues (on reboot) after having
> accepted messages without OOM. We investigated the problem and diagnosed that
> this happens when we attempt to load a full page during recovery.
> If we update KahaDBStore recoverNextMessage to use the following:
> "if (counter >= maxReturned || !listener.hasSpace())"
> To check if we have enough space to load the next message, than our OOM is
> averted.
> AMQ-4495 had the above change that was backed out due to performance
> concerns, our testing does not show us a degradation - perhaps we could
> provide an optional parameter to use this code?
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
