[ 
https://issues.apache.org/jira/browse/SPARK-21479?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16440475#comment-16440475
 ] 

Apache Spark commented on SPARK-21479:
--------------------------------------

User 'cloud-fan' has created a pull request for this issue:
https://github.com/apache/spark/pull/21083

> Outer join filter pushdown in null supplying table when condition is on one 
> of the joined columns
> -------------------------------------------------------------------------------------------------
>
>                 Key: SPARK-21479
>                 URL: https://issues.apache.org/jira/browse/SPARK-21479
>             Project: Spark
>          Issue Type: Bug
>          Components: Optimizer, SQL
>    Affects Versions: 2.1.0, 2.1.1, 2.2.0
>            Reporter: Abhijit Bhole
>            Priority: Major
>
> Here are two different query plans - 
> {code:java}
> df1 = spark.createDataFrame([{ "a": 1, "b" : 2}, { "a": 3, "b" : 4}])
> df2 = spark.createDataFrame([{ "a": 1, "c" : 5}, { "a": 3, "c" : 6}, { "a": 
> 5, "c" : 8}])
> df1.join(df2, ['a'], 'right_outer').where("b = 2").explain()
> == Physical Plan ==
> *Project [a#16299L, b#16295L, c#16300L]
> +- *SortMergeJoin [a#16294L], [a#16299L], Inner
>    :- *Sort [a#16294L ASC NULLS FIRST], false, 0
>    :  +- Exchange hashpartitioning(a#16294L, 4)
>    :     +- *Filter ((isnotnull(b#16295L) && (b#16295L = 2)) && 
> isnotnull(a#16294L))
>    :        +- Scan ExistingRDD[a#16294L,b#16295L]
>    +- *Sort [a#16299L ASC NULLS FIRST], false, 0
>       +- Exchange hashpartitioning(a#16299L, 4)
>          +- *Filter isnotnull(a#16299L)
>             +- Scan ExistingRDD[a#16299L,c#16300L]
> df1 = spark.createDataFrame([{ "a": 1, "b" : 2}, { "a": 3, "b" : 4}])
> df2 = spark.createDataFrame([{ "a": 1, "c" : 5}, { "a": 3, "c" : 6}, { "a": 
> 5, "c" : 8}])
> df1.join(df2, ['a'], 'right_outer').where("a = 1").explain()
> == Physical Plan ==
> *Project [a#16314L, b#16310L, c#16315L]
> +- SortMergeJoin [a#16309L], [a#16314L], RightOuter
>    :- *Sort [a#16309L ASC NULLS FIRST], false, 0
>    :  +- Exchange hashpartitioning(a#16309L, 4)
>    :     +- Scan ExistingRDD[a#16309L,b#16310L]
>    +- *Sort [a#16314L ASC NULLS FIRST], false, 0
>       +- Exchange hashpartitioning(a#16314L, 4)
>          +- *Filter (isnotnull(a#16314L) && (a#16314L = 1))
>             +- Scan ExistingRDD[a#16314L,c#16315L]
> {code}
> If condition on b can be pushed down on df1 then why not condition on a?



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)

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

Reply via email to