[SPARK-26996] Scalar Subquery not handled properly in Spark 2.4

• Type: Bug
• Status: Closed
• Priority: Critical
• Resolution: Duplicate
• Affects Version/s: 2.4.0
• Component/s: SQL
• Assignee: Unassigned
• Reporter: Ilya Peysakhov
• Created: 26/Feb/19 16:42
• Updated: 27/Feb/19 13:13
• Resolved: 26/Feb/19 18:52

Description

Spark 2.4 reports an error when querying a dataframe that has only 1 row and 1 column (scalar subquery). 

 

Reproducer is below. No other data is needed to reproduce the error.

This will write a table of dates and strings, write another "fact" table of ints and dates, then read both tables as views and filter the "fact" based on the max(date) from the first table. This is done within spark-shell in spark 2.4 vanilla (also reproduced in AWS EMR 5.20.0)

-------------------------

spark.sql("select '2018-01-01' as latest_date, 'source1' as source UNION ALL select '2018-01-02', 'source2' UNION ALL select '2018-01-03' , 'source3' UNION ALL select '2018-01-04' ,'source4' ").write.mode("overwrite").save("/latest_dates")
val mydatetable = spark.read.load("/latest_dates")
mydatetable.createOrReplaceTempView("latest_dates")

spark.sql("select 50 as mysum, '2018-01-01' as date UNION ALL select 100, '2018-01-02' UNION ALL select 300, '2018-01-03' UNION ALL select 3444, '2018-01-01' UNION ALL select 600, '2018-08-30' ").write.mode("overwrite").partitionBy("date").save("/mypartitioneddata")
val source1 = spark.read.load("/mypartitioneddata")
source1.createOrReplaceTempView("source1")

spark.sql("select max(date), 'source1' as category from source1 where date >= (select latest_date from latest_dates where source='source1') ").show

 ----------------------------

 

Error summary

—

java.lang.UnsupportedOperationException: Cannot evaluate expression: scalar-subquery#35 []
at org.apache.spark.sql.catalyst.expressions.Unevaluable$class.eval(Expression.scala:258)
at org.apache.spark.sql.catalyst.expressions.ScalarSubquery.eval(subquery.scala:246)

-------

This reproducer works in previous versions (2.3.2, 2.3.1, etc).

 

Issue Links

• duplicates: SPARK-26709 OptimizeMetadataOnlyQuery does not correctly handle the files with zero record