[
https://issues.apache.org/jira/browse/SPARK-16869?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15405669#comment-15405669
]
GUAN Hao commented on SPARK-16869:
----------------------------------
Oh, {{ i = colaesce(b.i, c.i) }} is actually just a comment.
Please try the code at the bottom of the original post.
> Wrong projection when join on columns with the same name which are derived
> from the same dataframe
> --------------------------------------------------------------------------------------------------
>
> Key: SPARK-16869
> URL: https://issues.apache.org/jira/browse/SPARK-16869
> Project: Spark
> Issue Type: Bug
> Components: SQL
> Affects Versions: 2.0.0
> Reporter: GUAN Hao
>
> I have to DataFrames, both contain a column named *i* which are derived from
> a same DataFrame (join).
> {code}
> b
> +---+---+---+---+
> | j| p| i| k|
> +---+---+---+---+
> | 3| 2| 3| 3|
> | 2| 1| 2| 2|
> +---+---+---+---+
> c
> +---+---+---+---+
> | j| k| q| i|
> +---+---+---+---+
> | 1| 1| 0| 1|
> | 2| 2| 1| 2|
> +---+---+---+---+
> {code}
> The result of OUTER join of two DataFrames above is:
> {code}
> i = colaesce(b.i, c.i)
> +----+----+----+---+---+----+----+
> | b_i| c_i| i| j| k| p| q|
> +----+----+----+---+---+----+----+
> | 2| 2| 2| 2| 2| 1| 1|
> |null| 1| 1| 1| 1|null| 0|
> | 3|null| 3| 3| 3| 2|null|
> +----+----+----+---+---+----+----+
> {code}
> However, what I got is:
> {code}
> +----+----+----+---+---+----+----+
> | b_i| c_i| i| j| k| p| q|
> +----+----+----+---+---+----+----+
> | 2| 2| 2| 2| 2| 1| 1|
> |null|null|null| 1| 1|null| 0|
> | 3| 3| 3| 3| 3| 2|null|
> +----+----+----+---+---+----+----+
> {code}
> {code}
> == Physical Plan ==
> *Project [i#0L AS b_i#146L, i#0L AS c_i#147L, coalesce(i#0L, i#0L) AS i#148L,
> coalesce(j#12L, j#21L) AS j#149L, coalesce(k#2L, k#22L) AS k#150L, p#13L,
> q#23L]
> +- SortMergeJoin [i#0L, j#12L, k#2L], [i#113L, j#21L, k#22L], FullOuter
> ....
> {code}
> As shown in the plan, columns {{b.i}} and {{c.i}} are correctly resolved to
> {{i#0L}} and {{i#113L}} correspondingly in the join condition part. However,
> in the projection part, both {{b.i}} and {{c.i}} are resolved to {{i#0L}}.
> Complete code to re-produce:
> {code}
> from pyspark import SparkContext, SQLContext
> from pyspark.sql import Row, functions
> sc = SparkContext()
> sqlContext = SQLContext(sc)
> data_a = sc.parallelize([
> Row(i=1, j=1, k=1),
> Row(i=2, j=2, k=2),
> Row(i=3, j=3, k=3),
> ])
> table_a = sqlContext.createDataFrame(data_a)
> table_a.show()
> data_b = sc.parallelize([
> Row(j=2, p=1),
> Row(j=3, p=2),
> ])
> table_b = sqlContext.createDataFrame(data_b)
> table_b.show()
> data_c = sc.parallelize([
> Row(j=1, k=1, q=0),
> Row(j=2, k=2, q=1),
> ])
> table_c = sqlContext.createDataFrame(data_c)
> table_c.show()
> b = table_b.join(table_a, table_b.j == table_a.j).drop(table_a.j)
> c = table_c.join(table_a, (table_c.j == table_a.j)
> & (table_c.k == table_a.k)) \
> .drop(table_a.j) \
> .drop(table_a.k)
> b.show()
> c.show()
> {code}
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
