[
https://issues.apache.org/jira/browse/HBASE-6377?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Andrew Purtell updated HBASE-6377:
----------------------------------
Attachment: 6377-trunk-simple.patch
Attached as '6377-trunk-simple.patch' for your consideration is a minimal set
of changes that put back HRegionInterface level request metrics. The intent of
HBASE-5533 was to have server side request metrics to match up with client side
metrics. So what we are measuring here? Here is my interpretation: Given all of
the batching/dispatch/increasingly async actions taken by the client a 1:1
correspondence with client API ops is not what we are after; instead, a measure
of the server side time required to process a unit of work sent by the client.
This kind of measurement can be done simply and cheaply, especially given the
nice HRI refactor in trunk. There's no need to ask "how to deconstruct this
server side packet of work into client side ops and account for them
individually?" On the other hand, it's a very coarse measurement.
> HBASE-5533 metrics miss all operations submitted via MultiAction
> ----------------------------------------------------------------
>
> Key: HBASE-6377
> URL: https://issues.apache.org/jira/browse/HBASE-6377
> Project: HBase
> Issue Type: Bug
> Components: metrics, regionserver
> Affects Versions: 0.96.0, 0.94.1
> Reporter: Andrew Purtell
> Fix For: 0.94.2
>
> Attachments: 6377-trunk-simple.patch
>
>
> A client application (LoadTestTool) calls put() on HTables. Internally to the
> HBase client those puts are batched into MultiActions. The total number of
> put operations shown in the RegionServer's put metrics histogram never
> increases from 0 even though millions of such operations are made. Needless
> to say the latency for those operations are not measured either. The value of
> HBASE-5533 metrics are suspect given the client will batch put and delete ops
> like this.
> I had a fix in progress but HBASE-6284 messed it up. Before, MultiAction
> processing in HRegionServer would distingush between puts and deletes and
> dispatch them separately. It was easy to account for the time for them. Now
> both puts and deletes are submitted in batch together as mutations.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators:
https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira
