[
https://issues.apache.org/jira/browse/AMQ-6815?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16322192#comment-16322192
]
ASF subversion and git services commented on AMQ-6815:
------------------------------------------------------
Commit d8f8ae9f925b905968c51b386049c5b2d9b6b2d0 in activemq's branch
refs/heads/activemq-5.15.x from [~gtully]
[ https://git-wip-us.apache.org/repos/asf?p=activemq.git;h=d8f8ae9 ]
[AMQ-6815] rework to drop the batch reference from Location such that batches
are free for gc when index pages are agressively cached
(cherry picked from commit ec6fa190999160676cab900038b268b2d40a4d5c)
> KahaDB checkpoint needs to fail fast in the event of journal write failure
> --------------------------------------------------------------------------
>
> Key: AMQ-6815
> URL: https://issues.apache.org/jira/browse/AMQ-6815
> Project: ActiveMQ
> Issue Type: Bug
> Components: KahaDB
> Affects Versions: 5.15.0
> Reporter: Gary Tully
> Assignee: Gary Tully
> Fix For: 5.16.0, 5.15.3
>
>
> KahaDB checkpoint does some async updated to the journal and syncs to the
> index. However the async journal updates can error out and not get reported
> leading to the index metadata being stale. This can lead to reading an
> invalid location on restart which may result in full recovery of the index.
> Any exception from the async store needs to cause the checkpoint to fail.
> In addition, on reading a corrupt location, we can do a sanity check to avoid
> a potential OOM by validating the size before allocation.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
