[
https://issues.apache.org/jira/browse/GEODE-1700?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15396029#comment-15396029
]
ASF subversion and git services commented on GEODE-1700:
--------------------------------------------------------
Commit 9389c6fece1adb3e83bc7a8863a2d84f52bfc0fa in incubator-geode's branch
refs/heads/develop from [~dschneider]
[ https://git-wip-us.apache.org/repos/asf?p=incubator-geode.git;h=9389c6f ]
GEODE-1700: fix stat corrupting when async queue full
A common method is now used for writing async ops to disk.
The same method is now used for both flushing an item that
has been taken out of the queue and for when the queue is
full causing a sync write to be done.
Also did a minor refactoring to get rid of the "removed" flag.
This revision fixes a bug introduced in GEODE-93 and should be
included with that fix.
> The DiskRegionStats entriesInVM and entriesOnlyOnDisk for overflow lru
> regions with async disk may be wrong
> -----------------------------------------------------------------------------------------------------------
>
> Key: GEODE-1700
> URL: https://issues.apache.org/jira/browse/GEODE-1700
> Project: Geode
> Issue Type: Bug
> Components: persistence
> Reporter: Darrel Schneider
> Assignee: Darrel Schneider
>
> The DiskRegionStats entriesInVM and entriesOnlyOnDisk for overflow lru
> regions with async disk may be wrong.
> A race condition exists that can cause the async queue to fill. In that case
> additional evictions are written to disk immediately instead of going into
> the queue. The code for this condition (handleFullAsyncQueue) was not updated
> when a recent change was made to how the stats entriesInVM and
> entriesOnlyOnDisk were updated.
> This can also cause partition balancing decisions to be wrong since this size
> of the bucket can be larger than it should be.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
