[
https://issues.apache.org/jira/browse/SPARK-39302?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17554015#comment-17554015
]
Sean R. Owen commented on SPARK-39302:
--------------------------------------
The error seems correct. You group the result of the join by country_id, but in
the join that definitely resolves to country_id 910, as there is no other
country_id. The projection happens after, which creates the second country_id.
Group the result of the join after the projection.
> Spark SQL - wrong field selection in group by
> ---------------------------------------------
>
> Key: SPARK-39302
> URL: https://issues.apache.org/jira/browse/SPARK-39302
> Project: Spark
> Issue Type: Bug
> Components: SQL
> Affects Versions: 3.1.2
> Reporter: shaharl...@gmail.com
> Priority: Minor
>
> SQL parser select wrong field to group by.
> See following example.
> Sample data:
>
> {code:java}
> spark.sql("""
> select "US" as pv_countryCode
> union all
> select "IL" as pv_countryCode
> """).createOrReplaceTempView("my_test_data")
> spark.sql("""
> select -3 as id, "US" as countryCode
> """).createOrReplaceTempView("country_codes_sample")
> {code}
> code:
> {code:java}
>
> spark.sql("""
> WITH BASE
> (SELECT
> pv_countryCode AS country_id
> FROM my_test_data)
> SELECT
> COALESCE(cct.id, -3) as country_id,
> count(1) as count
> FROM BASE
> LEFT JOIN country_codes_sample cct ON cct.countryCode =
> country_id
> GROUP BY
> country_id
> """).createOrReplaceTempView("campaign_country_bug") {code}
> Error:
> {code:java}
> org.apache.spark.sql.AnalysisException: expression 'cct.`id`' is neither
> present in the group by, nor is it an aggregate function. Add to group by or
> wrap in first() (or first_value) if you don't care which value you get.;
> Aggregate [country_id#910], [coalesce(id#886, -3) AS country_id#908, count(1)
> AS count#909L]
> +- Join LeftOuter, (countryCode#887 = country_id#910)
> :- SubqueryAlias BASE
> : +- Project [pv_countryCode#883 AS country_id#910]
> : +- SubqueryAlias my_test_data
> : +- Union false, false
> : :- Project [US AS pv_countryCode#883]
> : : +- OneRowRelation
> : +- Project [IL AS pv_countryCode#884]
> : +- OneRowRelation
> +- SubqueryAlias cct
> +- SubqueryAlias country_codes_sample
> +- Project [-3 AS id#886, US AS countryCode#887]
> +- OneRowRelation {code}
> I expected Spark to choose the selected country_id(country_id#908) instead of
> country_id#910
> Or at least throw ambiguous exception when grouping by `country_id`.
> This lead developers to add `cct.id` into group by which results in
> unexpected results.
> (In case country_id has both null and -3 values)
>
--
This message was sent by Atlassian Jira
(v8.20.7#820007)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
