[
https://issues.apache.org/jira/browse/HBASE-29276?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Wellington Chevreuil resolved HBASE-29276.
------------------------------------------
Release Note:
This proposes two additional properties:
1) "hbase.blockcache.stats.periods" which allows for defining a multiple window
period;
2) "hbase.blockcache.stats.period.minute" which defines the length of each of
these periods (in minutes);
If "hbase.blockcache.stats.periods" is defined and is greater than one, it
creates a scheduled executor that rolls the metrics calculation at
"hbase.blockcache.stats.period.minute" rate.
This will calculate hit ratio for each of the last periods (as defined by
hbase.blockcache.stats.periods), accounting for only the hits and requests
occurred during the interval of the given period (as defined by
hbase.blockcache.stats.period.minute).
For example, assuming default values for these two properties (12 and 5
minutes, respectively), the RS UI would display the hit ratio for each of the
last twelve, five minutes periods, in the L2 Cache stats tab. It also displays
the aggregated last hour hit ratio (by counting the 12*5mins periods).
Resolution: Fixed
Merged into master, branch-3, branch-2 and branch-2.6. Thanks for reviewing
this, [~taklwu] !
> Compute and display hit ratio by configurable, granular time periods
> --------------------------------------------------------------------
>
> Key: HBASE-29276
> URL: https://issues.apache.org/jira/browse/HBASE-29276
> Project: HBase
> Issue Type: Improvement
> Affects Versions: 3.0.0-beta-1, 4.0.0-alpha-1, 2.7.0, 2.6.2
> Reporter: Wellington Chevreuil
> Assignee: Wellington Chevreuil
> Priority: Major
> Labels: pull-request-available
> Fix For: 3.0.0, 4.0.0-alpha-1, 2.7.0, 2.6.3
>
> Attachments: Screenshot 2025-04-28 at 18.30.37.png
>
>
> Currently, we only compute and display hit ratio stats for the all time
> running period of the given region servers. For long running setups where
> most of the accessed data is cached, small recent periods of low hits would
> go undetected, since the larger number of long term hits would drive the hit
> average up.
> This proposes two additional properties:
> 1) "hbase.blockcache.stats.periods" which allows for defining a multiple
> window period;
> 2) "hbase.blockcache.stats.period.minute" which defines the length of each
> of these periods;
> If "hbase.blockcache.stats.periods" is defined and is greater than one, it
> creates a scheduled executor that rolls the metrics calculation at
> "hbase.blockcache.stats.period.minute" rate.
> BucketCache can then set a default window of 12 periods each 5 minutes long,
> providing metrics over the last hour, in addition to the all time hit ratio.
>
> !Screenshot 2025-04-28 at 18.30.37.png|width=1045,height=336!
>
>
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
