[
https://issues.apache.org/jira/browse/HBASE-15967?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15316448#comment-15316448
]
Bryan Beaudreault commented on HBASE-15967:
-------------------------------------------
We've found point-in-time gauges to be non-ideal for tracking rapidly changing
values like # of active handlers (or in this case readers). They are less
sensitive to spikes and also makes the metric configuration dependent (a value
of 8 is only interesting when compared to the known max value or cpu count).
One nice thing that kafka does is provide a `NetworkProcessorAvgIdlePercent`
metrics, which is tracked somewhat like so in 0.8.x:
https://apache.googlesource.com/kafka/+/5174df53778cb5cb2d6d86e4cec9f3185a2c85db/core/src/main/scala/kafka/network/SocketServer.scala#58
In 0.9 they refactored a bit and the code is slightly more complicated, but you
can see some of the updated logic here
https://github.com/cloudera/kafka/blob/cdh5-0.9.0_2.0.1/clients/src/main/java/org/apache/kafka/common/network/Selector.java#L630
and search for `selectTime` to see it updated.
They provide similar metrics for both the network threads (similar to ipc
readers) and request handler threads (ipc handlers). We've added similar
metrics to our internal hbase fork for tracking the RPC handler usage, but
figured I'd mention for discussion here since you're adding a net new metric.
Thoughts?
> 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
(v6.3.4#6332)
