[
https://issues.apache.org/jira/browse/HBASE-21186?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16642065#comment-16642065
]
Josh Elser commented on HBASE-21186:
------------------------------------
200 seems really high to me. I don't think the Master would ever get close to
saturating that number of threads. So, while I agree with your assessment that
extra threads isn't the end of the world, I do think it's unnecessarily high.
Did you happen to tweak any properties in the Master so that it would assign
out more regions, and saturate these threads?
I would think something like 50 would be an upper bound, but maybe there's a
back-of-the-envelope formula to be had considering:
# How many regions the Master will dole out for assignment at a time
# How many RegionServers you have
# The avg Regions per RegionServer number.
With this, you could guess at a formula like .. ({{$1 / $2}}) to get the number
of Regions that will be assigned to each RS for a "round" of assignments (the
Master does these in batches). You could also factor in {{$3}} to come up with
how many rounds need to then happen (not sure if that would be helpful..)
> Document hbase.regionserver.executor.openregion.threads in MTTR section
> -----------------------------------------------------------------------
>
> Key: HBASE-21186
> URL: https://issues.apache.org/jira/browse/HBASE-21186
> Project: HBase
> Issue Type: Improvement
> Components: documentation
> Reporter: Sahil Aggarwal
> Assignee: Sahil Aggarwal
> Priority: Minor
> Attachments: HBASE-21186.master.001.patch,
> HBASE-21186.master.002.patch
>
>
> hbase.regionserver.executor.openregion.threads helps in improving MTTR by
> increasing assign rpc processing rate at RS from HMaster but is not
> documented.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
