hbase-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Anoop Sam John (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HBASE-6284) Introduce HRegion#doMiniBatchDelete()
Date Thu, 28 Jun 2012 05:53:43 GMT

    [ https://issues.apache.org/jira/browse/HBASE-6284?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13402873#comment-13402873
] 

Anoop Sam John commented on HBASE-6284:
---------------------------------------

HBASE-5941 is doing almost the same thing which I had done in my testing. Batching the deletes
in region level and grabbing as many locks as possible and do that operations in one shot.
The write and sync to WAL will be just one time for this one shot.

Already in 94 and Trunk there is mutateRowsWithLocks() which is also doing almost the same.
The only difference is that this method needs the locks on all the rows in one time and will
do those many ops in one time.[Yes it need to]
Instead our new method will do in mini batches. At least one row lock (batch of one Put /Delete)
is needed.
                
> Introduce HRegion#doMiniBatchDelete()
> -------------------------------------
>
>                 Key: HBASE-6284
>                 URL: https://issues.apache.org/jira/browse/HBASE-6284
>             Project: HBase
>          Issue Type: Bug
>            Reporter: Zhihong Ted Yu
>            Assignee: Anoop Sam John
>
> From Anoop under thread 'Can there be a doMiniBatchDelete in HRegion':
> The HTable#delete(List<Delete>) groups the Deletes for the same RS and make one
n/w call only. But within the RS, there will be N number of delete calls on the region one
by one. This will include N number of HLog write and sync. If this also can be grouped can
we get better performance for the multi row delete.
> I have made the new miniBatchDelete () and made the HTable#delete(List<Delete>)
to call this new batch delete.
> Just tested initially with the one node cluster.  In that itself I am getting a performance
boost which is very much promising.
> Only one CF and qualifier.
> 10K total rows delete with a batch of 100 deletes. Only deletes happening on the table
from one thread.
> With the new way the net time taken is reduced by more than 1/10
> Will test in a 4 node cluster also. I think it will worth doing this change.

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators: https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira

        

Mime
View raw message