[SPARK-47525] Support subquery correlation joining on map attributes

• Type: Improvement
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 3.5.0
• Fix Version/s: 4.0.0
• Component/s: SQL
• Labels: pull-request-available
• Assignee: Jack Chen
• Reporter: Jack Chen
• Created: 22/Mar/24 19:07
• Updated: 29/Mar/24 02:52
• Resolved: 29/Mar/24 02:52

Description

Currently, when a subquery is correlated on a condition like `outer_map[1] = inner_map[1]`, DecorrelateInnerQuery generates a join on the map itself,
which is unsupported, so the query cannot run - for example:
 

scala> Seq(Map(0 -> 0)).toDF.createOrReplaceTempView("v")scala> sql("select v1.value[0] from v v1 where v1.value[0] > (select avg(v2.value[0]) from v v2 where v1.value[1] = v2.value[1])").explain
org.apache.spark.sql.AnalysisException: [UNSUPPORTED_SUBQUERY_EXPRESSION_CATEGORY.UNSUPPORTED_CORRELATED_REFERENCE_DATA_TYPE] Unsupported subquery expression: Correlated column reference 'v1.value' cannot be map type. SQLSTATE: 0A000; line 1 pos 49
at org.apache.spark.sql.errors.QueryCompilationErrors$.unsupportedCorrelatedReferenceDataTypeError(QueryCompilationErrors.scala:2463)
... 

However, if we rewrite the query to pull out the map access `outer_map[1]` into the outer plan, it succeeds:
 

scala> sql("""with tmp as (
select value[0] as value0, value[1] as value1 from v
)
select v1.value0 from tmp v1 where v1.value0 > (select avg(v2.value0) from tmp v2 where v1.value1 = v2.value1)""").explain

Another point that can be improved is that, even if the data type supports join, we still don’t need to join on the full attribute, and we can get a better plan by doing the same rewrite to pull out the extract expression.

Issue Links

• links to: GitHub Pull Request #45673