[
https://issues.apache.org/jira/browse/HDFS-15311?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17103802#comment-17103802
]
xuzq commented on HDFS-15311:
-----------------------------
Thanks [~shv] and i will watch
[HDFS-15291|https://issues.apache.org/jira/browse/HDFS-15291].
In my test, the Observer Handler only slept for a short period of time, and the
ProcessTime of the client dropped significantly.
> [SBN Read] High frequency reQueue cause Reader's performance to degrade
> -----------------------------------------------------------------------
>
> Key: HDFS-15311
> URL: https://issues.apache.org/jira/browse/HDFS-15311
> Project: Hadoop HDFS
> Issue Type: Improvement
> Reporter: xuzq
> Priority: Major
>
> If _autoMsyncPeriodMs_ is 0, will do _msync_ for each read rpc.
> On the observer server side, it will cause high frequency reQueue in Handler.
> As the Queue is BlockingQueue, so it will cause Readers(small number) and
> Handlers(large number) competing for BlockingQueue locks.
> It will cause the throughput decrease.
>
> Maybe we can let the handler sleep a little time to wait the StateId to
> decrease ReQueue.
>
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
