[
https://issues.apache.org/jira/browse/HBASE-3632?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13006572#comment-13006572
]
Gary Helmling commented on HBASE-3632:
--------------------------------------
This definitely makes sense to me, as a moving average seems a much more useful
metric. So seems like we'd want this for both "blockCacheHitRatio" and
"blockCacheHitCachingRatio".
As an aside, we should probably not change the way that the related "count"
metrics are reported though. If a metrics collection setup is configured to
see those as incrementing values, the displayed values freak out if they start
reseting (I've seen this with cacti at least).
> ability to record the block cache hit ratio for the last few minutes
> --------------------------------------------------------------------
>
> Key: HBASE-3632
> URL: https://issues.apache.org/jira/browse/HBASE-3632
> Project: HBase
> Issue Type: Improvement
> Components: regionserver
> Reporter: dhruba borthakur
> Assignee: dhruba borthakur
>
> In the curent code, the block cache hit ratio is calculated by using the
> total number of cache hits since the regin server was rebooted. This means
> that this metric does not reflect the short term changes that occur in the
> workload; the longer the region reserver is alive slower is the rate of
> change of this metric.
> I propose that this metric reflect the cache hit ratio for the work processed
> in the most recent 10 minutes, is that reasonable?
--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira
