[
https://issues.apache.org/jira/browse/SPARK-36967?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Wan Kun updated SPARK-36967:
----------------------------
Description:
Now map task will report accurate shuffle block size if the block size is
greater than "spark.shuffle.accurateBlockThreshold"( 100M by default ). But if
there are a large number of map tasks and the shuffle block sizes of these
tasks are smaller than "spark.shuffle.accurateBlockThreshold", there may be
unrecognized data skew.
For example, there are 10000 map task and 10000 reduce task, and each map task
create 50M shuffle blocks for reduce 0, and 10K shuffle blocks for the left
reduce tasks, reduce 0 is data skew, but the stat of this plan do not have this
information.
!map_status2.png!
I think we need to judge if a shuffle block is huge and need to be accurate
reported while running.
was:
Now map task will report accurate shuffle block size if the block size is
greater than "spark.shuffle.accurateBlockThreshold"( 100M by default ). But if
there are many map tasks and shuffle block sizes of these tasks are less than
"spark.shuffle.accurateBlockThreshold", there may be data skew, but not
recognized.
For example, there are 10000 map task and 10000 reduce task, and each task has
50M for reduce 0, and 10K for the left reduce tasks, reduce 0 is data skew, but
the stat of this plan do not have this message.
!map_status.png!
I think we need to judge if a shuffle block is huge and need to be accurate
reported while running.
> Update accurate block size threshold per reduce task
> ----------------------------------------------------
>
> Key: SPARK-36967
> URL: https://issues.apache.org/jira/browse/SPARK-36967
> Project: Spark
> Issue Type: Improvement
> Components: Spark Core
> Affects Versions: 3.3.0
> Reporter: Wan Kun
> Priority: Major
> Attachments: map_status.png, map_status2.png
>
>
> Now map task will report accurate shuffle block size if the block size is
> greater than "spark.shuffle.accurateBlockThreshold"( 100M by default ). But
> if there are a large number of map tasks and the shuffle block sizes of these
> tasks are smaller than "spark.shuffle.accurateBlockThreshold", there may be
> unrecognized data skew.
> For example, there are 10000 map task and 10000 reduce task, and each map
> task create 50M shuffle blocks for reduce 0, and 10K shuffle blocks for the
> left reduce tasks, reduce 0 is data skew, but the stat of this plan do not
> have this information.
> !map_status2.png!
> I think we need to judge if a shuffle block is huge and need to be accurate
> reported while running.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
