[ 
https://issues.apache.org/jira/browse/BEAM-10542?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17162316#comment-17162316
 ] 

Valentyn Tymofieiev commented on BEAM-10542:
--------------------------------------------

There appears to be a regression in Nexmark Query 1 between 2.22.0 and 2.23.0 
that is reproducible on Flink runner. Identified via `git bisect` and the 
following command: 

gradlew clean; ./gradlew :sdks:java:testing:nexmark:assemble  ; ./gradlew 
:sdks:java:testing:nexmark:run -Pnexmark.runner=":runners:flink:1.10" 
-Pnexmark.args=" --runner=FlinkRunner --suite=SMOKE --streamTimeout=60 
--streaming=false --manageResources=false --monitorJobs=true 
--flinkMaster=[local] --query=1 --numEvents=1000000"

According to the git bisect:

{noformat}
6dc08a4fb361b19a3353e9bd93563bc1120723df is the first bad commit
commit 6dc08a4fb361b19a3353e9bd93563bc1120723df
Author: David Moravek <[email protected]>
Date:   Fri May 29 17:41:32 2020 +0200

    [BEAM-8608] Chain DoFns in Flink batch runner when possible.
{noformat}

Here are some datapoints with numElements increased 100x (to 1000000) on my 
machine: 
Before:
{noformat}
Performance:                                                                    
               
  Conf  Runtime(sec)    (Baseline)  Events(/sec)    (Baseline)       Results    
(Baseline)     
  0000           3.5                    285551.1                      920000    
               
==========================================================================================
     
{noformat}
After:
{noformat}
Performance:
  Conf  Runtime(sec)    (Baseline)  Events(/sec)    (Baseline)       Results    
(Baseline)
  0000           4.7                    214868.9                      920000    
          
==========================================================================================
{noformat}
The timing of the merge  of https://github.com/apache/beam/pull/11858 (Jun 1) 
does not quite align with the jump visible on the dashboard (Jun 16), so 
perhaps something else is also at play.


> Investigate a possible Nexmark performance regression around 06/16  
> --------------------------------------------------------------------
>
>                 Key: BEAM-10542
>                 URL: https://issues.apache.org/jira/browse/BEAM-10542
>             Project: Beam
>          Issue Type: Bug
>          Components: sdk-java-core
>            Reporter: Valentyn Tymofieiev
>            Priority: P0
>             Fix For: 2.23.0
>
>
> There is a jump in benchmark metric visible on the Dashboard:
> http://104.154.241.245/d/ahuaA_zGz/nexmark?orgId=1&from=now-90d&to=now



--
This message was sent by Atlassian Jira
(v8.3.4#803005)

Reply via email to