[
https://issues.apache.org/jira/browse/HBASE-19285?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16257830#comment-16257830
]
Josh Elser commented on HBASE-19285:
------------------------------------
Well, color me surprised. This seems to be "OK".
* As we'd expect, aggregating up to the table level (instead of regions) helps
keep memory use under wraps, regardless the size of table. Confirmed this by
JFR object count: 95% is in the write path (bytes, ConcurrentSkipListMap/Nodes,
KeyValue, etc).
* With a sampling rate of 10ms, didn't get a single hit in a FastLongHistogram
or the corresponding metrics2 Histogram wrappers. Nice confirmation.
* Using {{jmap -histo:live}} to compare the current branch-1.3 with my working
patch, the amount of extra heap with this new per-table histograms (again, for
a single table) appears to be a few KB -- nominal.
I need to clean up how the metrics are being shown on the JMX (each table is
getting its own "sub" which is gross), but I think I'm very close to a first
patch. [~apurtell] do you (and/or [~enis] given the prior-art) have any other
testing/data you'd like me to collect while I have my harness set up?
I'm also wondering about some value in adding a configuration property which
can disable these histograms. I think I could even be swayed to have them be
default=false to remove all possibility for unexpected perf impact :)
> Add per-table latency histograms
> --------------------------------
>
> Key: HBASE-19285
> URL: https://issues.apache.org/jira/browse/HBASE-19285
> Project: HBase
> Issue Type: Bug
> Components: metrics
> Reporter: Clay B.
> Assignee: Josh Elser
> Priority: Critical
> Fix For: 2.0.0, 1.4.0, 1.3.3
>
>
> HBASE-17017 removed the per-region latency histograms (e.g. Get, Put, Scan at
> p75, p85, etc)
> HBASE-15518 added some per-table metrics, but not the latency histograms.
> Given the previous conversations, it seems like it these per-table
> aggregations weren't intentionally omitted, just never re-implemented after
> the per-region removal. They're some really nice out-of-the-box metrics we
> can provide to our users/admins as long as it's not detrimental.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
