GitHub user JoshRosen opened a pull request:
https://github.com/apache/spark/pull/6772
[SPARK-8317] Do not push sort into shuffle in Exchange operator
In some cases, Spark SQL pushes sorting operations into the shuffle layer
by specifying a key ordering as part of the shuffle dependency. I think that we
should not do this:
- Since we do not delegate aggregation to Spark's shuffle, specifying the
keyOrdering as part of the shuffle has no effect on the shuffle map side.
- By performing the shuffle ourselves (by inserting a sort operator after
the shuffle instead), we can use the Exchange planner to choose specialized
sorting implementations based on the types of rows being sorted.
- We can remove some complexity from SqlSerializer2 by not requiring it to
know about sort orderings, since SQL's own sort operators will already perform
the necessary defensive copying.
This patch removes Exchange's `canSortWithShuffle` path and the associated
code in `SqlSerializer2`. Shuffles that used to go through the
`canSortWithShuffle` path would always wind up using Spark's `ExternalSorter`
(inside of `HashShuffleReader`); to avoid a performance regression as a result
of handling these shuffles ourselves, I've changed the SQLConf defaults so that
external sorting is enabled by default.
You can merge this pull request into a Git repository by running:
$ git pull https://github.com/JoshRosen/spark SPARK-8317
Alternatively you can review and apply these changes as the patch at:
https://github.com/apache/spark/pull/6772.patch
To close this pull request, make a commit to your master/trunk branch
with (at least) the following in the commit message:
This closes #6772
----
commit bf3b4c875555079a43807927fbceb75c5f03d08b
Author: Josh Rosen <[email protected]>
Date: 2015-06-12T00:07:08Z
Enable external sort by default
commit ebf9c0f5c57ebda64aeb676e28424482150e0fab
Author: Josh Rosen <[email protected]>
Date: 2015-06-12T00:07:47Z
Do not push sort into shuffle in Exchange operator
----
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
