[jira] [Reopened] (SPARK-39070) Pivoting a dataframe after two explodes raises an org.codehaus.commons.compiler.CompileException

["Hyukjin Kwon (Jira)" <ji...@apache.org>]

     [ https://issues.apache.org/jira/browse/SPARK-39070?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Hyukjin Kwon reopened SPARK-39070:
----------------------------------

> Pivoting a dataframe after two explodes raises an org.codehaus.commons.compiler.CompileException
> ------------------------------------------------------------------------------------------------
>
>                 Key: SPARK-39070
>                 URL: https://issues.apache.org/jira/browse/SPARK-39070
>             Project: Spark
>          Issue Type: Bug
>          Components: PySpark, SQL
>    Affects Versions: 3.2.0, 3.2.1
>         Environment: Tested on:
> MacOS 12.3.1, python versions 3.8.10, 3.9.12
> Ubuntu 20.04, python version 3.8.10
> Tested spark versions 3.1.2, 3.1.3, 3.2.1 and 3.2.0 installed through conda and pip
>            Reporter: Felix Altenberg
>            Priority: Major
>         Attachments: spark_pivot_error.txt
>
>
> The following Code raises an exception starting in spark version 3.2.0
> {code:java}
> import pyspark.sql.functions as sf
> from pyspark.sql import Row
> data = [
>     Row(
>         other_value=10,
>         ships=[
>             Row(
>                 fields=[
>                     Row(name="field1", value=1),
>                     Row(name="field2", value=2),
>                 ]
>             ),
>             Row(
>                 fields=[
>                     Row(name="field1", value=3),
>                     Row(name="field3", value=4),
>                 ]
>             ),
>         ],
>     )
> ]
> df = spark.createDataFrame(data)
> df = df.withColumn("ships", sf.explode("ships")).select(
>     sf.col("other_value"), sf.col("ships.*")
> )
> df = df.withColumn("fields", sf.explode("fields")).select("other_value", "fields.*")
> df = df.groupBy("other_value").pivot("name") #ERROR THROWN HERE {code}
> The "pivot("name")" is what causes the error to be thrown.
> Having even one less explode (by manually transforming the input dataframe to what it would look like after the first explode) causes this error not to appear.
> Adding a .cache() before the groupBy and pivot also causes this error to dissapear.
> Strangely, this error does not appear when running the above code in Databricks using Databricks Runtime 10.4 which is also using spark version 3.2.1
> I have attached the full stacktrace of the error below.
> I will try to investigate more after work today and will add anything else I find.
> The exception raised is
> {noformat}
> org.codehaus.commons.compiler.CompileException: File 'generated.java', Line 61, Column 78: A method named "numElements" is not declared in any enclosing class nor any supertype, nor through a static import{noformat}
> but seems to be caused by this exception
> {noformat}
> Caused by: java.lang.ClassCastException: org.apache.spark.sql.catalyst.expressions.UnsafeArrayData cannot be cast to org.apache.spark.sql.catalyst.InternalRow{noformat}
> The following stackoverflow question seems to pertain to a very similar error
> [https://stackoverflow.com/questions/70356375/pyspark-pivot-unsafearraydata-cannot-be-cast-to-internalrow]



--
This message was sent by Atlassian Jira
(v8.20.7#820007)

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