[
https://issues.apache.org/jira/browse/HBASE-15315?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15232496#comment-15232496
]
Yu Li commented on HBASE-15315:
-------------------------------
Thanks for the feedback [~busbey] and [~eclark], good to know your thoughts. :-)
I came across this issue during YCSB testing against 1.2.1RC1, and in my case
YCSB clients and hbase servers all launched by user hadoop (never thought this
might cause any problem). In the test I found all requests easily occupied all
{{Priority.RpcServer.handler}}s leaving all {{B.default.RpcServer.handler}}s
idle. Backporting the patch here made the hbase.regionserver.handler.count
setting effective again (creating another user and re-deploy YCSB clients is
annoying I'd say...). Just to share my experience here in case anyone run into
the same problem and get some incorrect perf number against 1.2.x.
> Remove always set super user call as high priority
> --------------------------------------------------
>
> Key: HBASE-15315
> URL: https://issues.apache.org/jira/browse/HBASE-15315
> Project: HBase
> Issue Type: Improvement
> Reporter: Yong Zhang
> Assignee: Yong Zhang
> Fix For: 2.0.0, 1.3.0, 1.4.0
>
> Attachments: HBASE-15315.001.patch
>
>
> Current implementation set superuser call as ADMIN_QOS, but we have many
> customers use superuser to do normal table operation such as put/get data and
> so on. If client put much data during region assignment, RPC from HMaster may
> timeout because of no handle. so it is better to remove always set super user
> call as high priority.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
