[
https://issues.apache.org/jira/browse/PHOENIX-5291?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16845357#comment-16845357
]
Lars Hofhansl commented on PHOENIX-5291:
----------------------------------------
Case 2: UngroupedAggregateRegionObserver.collectStats(...). In the case of
concurrent UPDATE STATISTICS we do not run the StatsCollectionCallable to close
the passed scanner. Yet, even then we return wrapped scanner that crucially
does not close the inner scanner (since that one might be closed by the
callable later).
I'm sure there're more cases.
> Some cases where Phoenix does close scanners
> --------------------------------------------
>
> Key: PHOENIX-5291
> URL: https://issues.apache.org/jira/browse/PHOENIX-5291
> Project: Phoenix
> Issue Type: Bug
> Reporter: Lars Hofhansl
> Priority: Major
>
> With HBase 1.5 and later this is a disaster, as it causes the wrong reference
> counting of HFiles in HBase, and those subsequently will *never* be removed
> until the region closes and reopens for any reason.
> We found at least two cases... See comments below.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
