[
https://issues.apache.org/jira/browse/HBASE-15163?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15116613#comment-15116613
]
Yu Li commented on HBASE-15163:
-------------------------------
Thanks for the comment [~eclark]
bq. We already have those numbers with the histogram counting
Yes we do have the histogram for put/get/delete/scanNext. However, most of them
count in the actions in multi request (in
{{doNonAtomicRegionMutation}}/{{doBatchOp}}), so these are not from the *rpc*
perspective. We think it's valuable to know how many rpc requests of each type
are assigned to a single RS and help judge whether the workload is balanced.
Makes sense? (maybe I should update the JIRA title to make it more clear)
bq. Additionally this gets us back to the point were instantaneous rate is
controlled by how long inbetween metrics snapshots
Agreed, but we still have metrics like {{readRequestCount/writeRequestCount}}.
I think it's still worthwhile comparing the advantages/disadvantages. :-)
> Add sampling code and metrics for get/scan/multi/mutate count separately
> ------------------------------------------------------------------------
>
> Key: HBASE-15163
> URL: https://issues.apache.org/jira/browse/HBASE-15163
> Project: HBase
> Issue Type: Sub-task
> Reporter: Yu Li
> Assignee: Yu Li
> Attachments: DifferentRequestQPS.png, HBASE-15163.patch
>
>
> This way we could see QPS of different kinds of requests, to better analyze
> what's causing hot spot in system
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
