hbase-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Bryan Beaudreault (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HBASE-15967) Metric for active ipc Readers and make default fraction of cpu count
Date Thu, 16 Jun 2016 18:28:05 GMT

    [ https://issues.apache.org/jira/browse/HBASE-15967?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15334392#comment-15334392

Bryan Beaudreault commented on HBASE-15967:

Sort of. I'm suggesting we adopt the convention of using {{IdlePercentage}} for measuring
anything of the class of handler and reader usage, instead of the past approach of using point-in-time
counts of active X. The EWMATickTask is an implementation detail.  Typically this is all handled
by yammer metrics' top-level Meter class. But that has a bunch of overhead -- at the very
least it calculates a 1min, 5min, and 15min averages. We wanted to avoid that overhead, so
we used EWMA directly.

The EWMA documentation says it expects {{tick()}} to be called every 5sec, so we created and
registered the EWMATickTask to do so.

So the answer to your question is: both! It'd be great if this new metric you're creating
can use a similar implementation to our {{getIdlePctOfHandlerPool}}. But in order to achieve
that, if you want to use EWMA, you need something akin to the EWMATickTask (or use the heavier
Meter, which would give more averaging options).

If there's a more hbase-standard way than using yammer metrics to do something like this,
we'd be all ears.

> Metric for active ipc Readers and make default fraction of cpu count
> --------------------------------------------------------------------
>                 Key: HBASE-15967
>                 URL: https://issues.apache.org/jira/browse/HBASE-15967
>             Project: HBase
>          Issue Type: Sub-task
>            Reporter: stack
>            Assignee: stack
>         Attachments: HBASE-15967.master.001.patch
> Our ipc Readers are hard coded at 10 regardless since . Running w/ less Readers, we go
faster..(e.g. 12 Readers has us doing 135k with workloadc and 6 readers has us doing 145k)..
.but hard to tell what count of Readers are needed since no metric.
> This issue changes Readers to be 1/4 the installed CPUs or 8, whichever is the minimum,
and then adds a new hbase.regionserver.ipc.runningReaders metric so you have a chance seeing
whats needed.

This message was sent by Atlassian JIRA

View raw message