[jira] [Commented] (SPARK-16827) Query with Join produces excessive amount of shuffle data

["Brian Cho (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/SPARK-16827?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15546142#comment-15546142 ] 

Brian Cho commented on SPARK-16827:
-----------------------------------

We found that the "shuffle write" metric was also including writes of spill files, inflating the amount of shuffle writes. This even showed up for final stages when no shuffle writes should take place. I'll upload screenshots on the PR.

> Query with Join produces excessive amount of shuffle data
> ---------------------------------------------------------
>
>                 Key: SPARK-16827
>                 URL: https://issues.apache.org/jira/browse/SPARK-16827
>             Project: Spark
>          Issue Type: Bug
>          Components: Shuffle, Spark Core
>    Affects Versions: 2.0.0
>            Reporter: Sital Kedia
>              Labels: performance
>
> One of our hive job which looks like this -
> {code}
>  SELECT  userid
>      FROM  table1 a
>      JOIN table2 b
>       ON    a.ds = '2016-07-15'
>       AND  b.ds = '2016-07-15'
>       AND  a.source_id = b.id
> {code}
> After upgrade to Spark 2.0 the job is significantly slow.  Digging a little into it, we found out that one of the stages produces excessive amount of shuffle data.  Please note that this is a regression from Spark 1.6. Stage 2 of the job which used to produce 32KB shuffle data with 1.6, now produces more than 400GB with Spark 2.0. We also tried turning off whole stage code generation but that did not help. 
> PS - Even if the intermediate shuffle data size is huge, the job still produces accurate output.



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

---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@spark.apache.org
For additional commands, e-mail: issues-help@spark.apache.org