[
https://issues.apache.org/jira/browse/HBASE-8772?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13695004#comment-13695004
]
Sameer Vaishampayan commented on HBASE-8772:
--------------------------------------------
In my testing I have seen replication slow down significantly on several
unrelated things (to replication). Observed itfor compactions, major
compactions, gc in regionservers. More nuanced is that the replication
balancing of files is based on the regionserver's regions. Maybe I am not
putting it correctly, but heres the scenario I see,
in a cluster of 16 data nodes, couple of them have active region servers and
are seeing pileup of to-replicate-files in zkdump. 14 other nodes are sitting
idle saying nothing to replicate. surely the cluster as a whole has much
greater capability to replicate. Maybe this replication rebalancing is a
separate feature but it surely could be separated out from balancing regions.
As after the pileup has occured even balancing regions does little good, right
away. If only I could reassign/move the to-replicate files to idle machines.
> Separate Replication from HBase RegionServer process
> ----------------------------------------------------
>
> Key: HBASE-8772
> URL: https://issues.apache.org/jira/browse/HBASE-8772
> Project: HBase
> Issue Type: New Feature
> Components: regionserver, Replication
> Reporter: Sameer Vaishampayan
> Labels: performance
>
> Replication is a separate functionality than managing regions and should be
> able to be managed separately as a service rather than rolled into
> RegionServer. Load on RegionServer, gc etc shouldn't affect the replication
> service.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
