[jira] [Commented] (SPARK-42199) groupByKey creates columns that may conflict with exising columns

["Apache Spark (Jira)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/SPARK-42199?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17681026#comment-17681026 ] 

Apache Spark commented on SPARK-42199:
--------------------------------------

User 'EnricoMi' has created a pull request for this issue:
https://github.com/apache/spark/pull/39754

> groupByKey creates columns that may conflict with exising columns
> -----------------------------------------------------------------
>
>                 Key: SPARK-42199
>                 URL: https://issues.apache.org/jira/browse/SPARK-42199
>             Project: Spark
>          Issue Type: Bug
>          Components: SQL
>    Affects Versions: 3.0.3, 3.1.3, 3.2.3, 3.3.2, 3.4.0, 3.5.0
>            Reporter: Enrico Minack
>            Priority: Major
>
> Calling {{ds.groupByKey(func: V => K)}} creates columns to store the key value. These columns may conflict with columns that already exist in {{ds}}. Function {{Dataset.groupByKey.agg}} accounts for this with a very specific rule, which has some surprising weaknesses:
> {code:scala}
> spark.range(1)
>   // groupByKey adds column 'value'
>   .groupByKey(id => id)
>   // which cannot be referenced, though it is suggested
>   .agg(count("value"))
> {code}
> {code:java}
> org.apache.spark.sql.AnalysisException: Column 'value' does not exist. Did you mean one of the following? [value, id];
> {code}
> An existing 'value' column can be referenced:
> {code:scala}
> // dataset with column 'value'
> spark.range(1).select($"id".as("value")).as[Long]
>   // groupByKey adds another column 'value'
>   .groupByKey(id => id)
>   // agg accounts for the extra column and excludes it when resolving 'value'
>   .agg(count("value"))
>   .show()
> {code}
> {code:java}
> +---+------------+
> |key|count(value)|
> +---+------------+
> |  0|           1|
> +---+------------+
> {code}
> While column suggestion shows both 'value' columns:
> {code:scala}
> spark.range(1).select($"id".as("value")).as[Long]
>   .groupByKey(id => id)
>   .agg(count("unknown"))
> {code}
> {code:java}
> org.apache.spark.sql.AnalysisException: Column 'unknown' does not exist. Did you mean one of the following? [value, value]
> {code}
> However, {{mapValues}} introduces another 'value' column, which should be referencable, but it breaks the exclusion introduced by {{agg}}:
> {code:scala}
> spark.range(1)
>   // groupByKey adds column 'value'
>   .groupByKey(id => id)
>   // adds another 'value' column
>   .mapValues(value => value)
>   // which cannot be referenced in agg
>   .agg(count("value"))
> {code}
> {code:java}
> org.apache.spark.sql.AnalysisException: Reference 'value' is ambiguous, could be: value, value.
> {code}



--
This message was sent by Atlassian Jira
(v8.20.10#820010)

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