[
https://issues.apache.org/jira/browse/SPARK-21497?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16268033#comment-16268033
]
a283791411 commented on SPARK-21497:
------------------------------------
This feature is very useful. Please review.
> Pull non-deterministic joining keys from Join operator
> ------------------------------------------------------
>
> Key: SPARK-21497
> URL: https://issues.apache.org/jira/browse/SPARK-21497
> Project: Spark
> Issue Type: New Feature
> Components: SQL
> Affects Versions: 2.2.0
> Reporter: Liang-Chi Hsieh
>
> Currently SparkSQL doesn't support non-deterministic joining conditions in
> Join. This kind of joining conditions can be useful in some cases, e.g.,
> http://apache-spark-developers-list.1001551.n3.nabble.com/SQL-Syntax-quot-case-when-quot-doesn-t-be-supported-in-JOIN-tc21953.html#a21973.
> To pull non-deterministic joining conditions from Join operator, there seems
> no standard behavior. Based on the discussion on
> https://github.com/apache/spark/pull/18652#issuecomment-316344905,
> https://github.com/apache/spark/pull/18652#issuecomment-316391759 and
> https://github.com/apache/spark/pull/18652#issuecomment-316665649, Hive
> doesn't have special consideration for non-deterministic join conditions and
> simply pushes down it or uses it as joining keys.
> In this attempt, we initially allow non-deterministic equi join keys in Join
> operators. Because based on SparkSQL's join implementations the equi join
> keys are evaluated once on joining tables, pulling equi join keys from Join
> operators won't change the number of calls on non-deterministic expressions.
> It is more safer than other kinds of joining conditions, e.g. rand(10) > a &&
> rand(20) < b where pulling it and pushing down it will possibly change the
> number of calls of rand().
> We also add a SQL conf to control this new behavior. It is disabled by
> default.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
