[ 
https://issues.apache.org/jira/browse/CALCITE-642?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15118333#comment-15118333
 ] 

Josh Elser commented on CALCITE-642:
------------------------------------

Teaser:

{noformat}
16/01/26 19:16:56 INFO PhoenixQueryServer.Metrics: type=TIMER, 
name=AvaticaProtobufHandler.Handler.RequestTimings, count=28400, 
min=0.33899999999999997, max=177.731, mean=1.5453050391570862, 
stddev=12.124410598888346, median=0.41, p75=0.428, p95=0.602, p98=0.819, 
p99=4.987, p999=177.731, mean_rate=35.50429362298856, m1=82.95682251176358, 
m5=55.21720112587369, m15=25.688667272597616, rate_unit=events/second, 
duration_unit=milliseconds
16/01/26 19:16:56 INFO PhoenixQueryServer.Metrics: type=TIMER, 
name=ProtobufHandler.Handler.Serialization, count=56800, 
min=0.041999999999999996, max=3.8779999999999997, mean=0.11241678160721483, 
stddev=0.18166755227525663, median=0.13899999999999998, p75=0.147, p95=0.181, 
p98=0.22399999999999998, p99=0.238, p999=3.501, mean_rate=71.00847681444093, 
m1=165.91451214958184, m5=110.43436338590747, m15=51.37730672715273, 
rate_unit=events/second, duration_unit=milliseconds
{noformat}

What I have here is granular timings at the Jetty handler level 
(Handler.RequestTImings) and serialization timings around the protobuf 
serde'ing (Handler.Serialization).

I'm not sure how useful the rates are, but the stats are great. See can easily 
see things like the longest request and how good/bad the tail is as we approach 
that max. I hope to get a patch up this week.

> Avatica server metrics
> ----------------------
>
>                 Key: CALCITE-642
>                 URL: https://issues.apache.org/jira/browse/CALCITE-642
>             Project: Calcite
>          Issue Type: Improvement
>          Components: avatica
>            Reporter: Nick Dimiduk
>            Assignee: Josh Elser
>              Labels: avatica
>             Fix For: next
>
>
> Along the same lines as CALCITE-641, avatica server should track some basic 
> metrics. We can expose them through a "standard" API (such as Hadoop metrics, 
> or Dropwizard metrics), and include them on some interval in the service log.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Reply via email to