[
https://issues.apache.org/jira/browse/HBASE-12748?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14256559#comment-14256559
]
Andrew Purtell commented on HBASE-12748:
----------------------------------------
On second thought let me take that back.
I was tinkering around with this.
This patch 'coprocessorList', a simple ArrayList that can be used for iterating
the coprocessor list without creating an Iterator. However 'coprocessorList'
replaces use of SortedCopyOnWriteSet for iteration, which is safe for iteration
concurrent with adding or removing coprocessors, without providing a similar
guarantee. A CopyOnWriteArrayList would make a fresh copy of the underlying
array at each {{add()}}. Synchronization on 'coprocessorList' would undo why we
have SortedCopyOnWriteSet here in the first place. Maybe making
'coprocessorList' volatile, building up a list, and replacing the reference
with the completed list would work, or use an AtomicReference (with lazySet)?
> RegionCoprocessorHost.execOperation creates too many iterator objects
> ---------------------------------------------------------------------
>
> Key: HBASE-12748
> URL: https://issues.apache.org/jira/browse/HBASE-12748
> Project: HBase
> Issue Type: Sub-task
> Affects Versions: 0.94.25, 0.98.9
> Reporter: Vladimir Rodionov
> Assignee: Vladimir Rodionov
> Fix For: 1.0.0, 2.0.0, 0.94.26, 0.98.10
>
> Attachments: HBase-12748.patch
>
>
> This is typical pattern of enhanced for ... loop usage in a hot code path.
> For every HBase operation it instantiates iterator for coprocessor list
> twice.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
