[
https://issues.apache.org/jira/browse/HBASE-1354?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12720525#action_12720525
]
stack commented on HBASE-1354:
------------------------------
This patch looks good.
In src/java/org/apache/hadoop/hbase/regionserver/HRegionServer.java, why remove
the try/catch that converts an IOE to a LOG.WARN?
Do all tests pass Evgeny?
Looks like the data structure StoreSize is no longer needed? Why not remove it?
I'd say this will become annoying:
+ LOG.debug(this.storeNameStr + ": no need in compaction");
Remove it I'd say.
Why not have compact return max size rather than have Region call it? In
compact we are getting size anyways, aren't we (I see a getReader().length()
call)? Doing size of a file is expensive operation -- any call out to hdfs is
expensive. In other words, this patch would seem at first to be saving on hdfs
stat calls but is it really?
Thanks Evgeny
> Separate check split from compaction method
> -------------------------------------------
>
> Key: HBASE-1354
> URL: https://issues.apache.org/jira/browse/HBASE-1354
> Project: Hadoop HBase
> Issue Type: Sub-task
> Components: regionserver
> Affects Versions: 0.19.0
> Reporter: Evgeny Ryabitskiy
> Assignee: Evgeny Ryabitskiy
> Priority: Minor
> Fix For: 0.21.0
>
> Attachments: HBASE-1354.patch
>
>
> Now check for split is running from compaction method.
> This check if called after each store compaction. But results of this check
> used only for one Store. All other calling method are for nothing.
> So check split runs much more times that it should.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
