zookeeper-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Andor Molnar (JIRA)" <j...@apache.org>
Subject [jira] [Resolved] (ZOOKEEPER-2641) AvgRequestLatency metric improves to be more accurate
Date Mon, 07 Jan 2019 13:40:00 GMT

     [ https://issues.apache.org/jira/browse/ZOOKEEPER-2641?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel

Andor Molnar resolved ZOOKEEPER-2641.
       Resolution: Fixed
    Fix Version/s: 3.6.0

Issue resolved by pull request 748

> AvgRequestLatency metric improves to be more accurate
> -----------------------------------------------------
>                 Key: ZOOKEEPER-2641
>                 URL: https://issues.apache.org/jira/browse/ZOOKEEPER-2641
>             Project: ZooKeeper
>          Issue Type: Improvement
>    Affects Versions: 3.4.7, 3.4.9
>            Reporter: Allen Chan
>            Assignee: maoling
>            Priority: Minor
>              Labels: pull-request-available
>             Fix For: 3.6.0
>          Time Spent: 3h 50m
>  Remaining Estimate: 0h
> I posted a thread on the mailing list about finding AvgRequestLatency metric to be 0
all the time. I believe this is a valuable metric because it is useful to baseline performance
of ZK and know when something is going wrong. 
> Another user (Arshad Mohammad) wrote up these notes. I am not a developer so i do not
have ability to patch this. Filing this so hopefully someone with developer abilities can
add this improvement. 
> "I find two reason whys AvgRequestLatency is almost all the time 0
> 1) Ping requests are counted the most:
> AvgRequestLatency is calculated as
> AvgRequestLatency=totalLatency/count
> Ping requests come very often and complete very fast, these request add
> nothing to totalLatency but add one to count.
> 2) Wrong data type is chosen to store AvgRequestLatency:
> AvgRequestLatency is calculated and store as the long value instead of
> double vlaue.
> In my opinion ZooKeeper code should be modified to improve this metrics
> i) Ping request should be ignored while recording the statistics or at
> least should be configurable whether to ignore or not. If ping request is
> not counted even other metrics will be more meaningful.
> ii)  AvgRequestLatency should be of double type"

This message was sent by Atlassian JIRA

View raw message