[
https://issues.apache.org/jira/browse/HBASE-3033?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12914267#action_12914267
]
Jean-Daniel Cryans commented on HBASE-3033:
-------------------------------------------
At the same time, instead of batching consecutive rows, we can do even better
by creating a List<Put> per table and do a single call on HTable, leveraging
MultiPut even more. I just rolled that out in production here I got 10x
insertion speed (made a big enough difference that a region server that was
piling up HLogs to replicate, was at about 450 each of 128MB, is now able to
return back to live replication under the same write load).
> [replication] ReplicationSink.replicateEntries doesn't need to be sync'ed
> -------------------------------------------------------------------------
>
> Key: HBASE-3033
> URL: https://issues.apache.org/jira/browse/HBASE-3033
> Project: HBase
> Issue Type: Improvement
> Reporter: Jean-Daniel Cryans
> Assignee: Jean-Daniel Cryans
> Fix For: 0.90.0
>
>
> Once upon a time it was necessary to synchronize the access to
> ReplicationSink.replicateEntries but it's not required anymore. At the same
> time do the cleanup in the put/delete methods to de-dup the call
> this.pool.putTable(table);
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
