[SPARK-32618] ORC writer doesn't support colon in column names

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Invalid
• Affects Version/s: 2.3.0
• Component/s: Input/Output
• Assignee: Unassigned
• Reporter: Pierre Gramme
• Created: 14/Aug/20 15:10
• Updated: 17/Aug/20 14:54
• Resolved: 16/Aug/20 17:02

Description

Hi,

I'm getting an IllegalArgumentException: Can't parse category at 'struct<a:b^:int>' when exporting to ORC a dataframe whose column names contain colon (:). Reproducible as hereunder. Same problem also occurs if the name with colon appears nested as member of a struct.

Seems related with SPARK-21791(which was solved in 2.3.0).

In my real-life case, the column was actually xsi:type, coming from some parsed xml. Thus other users may be affected too.

Has it been fixed after Spark 2.3.0? (sorry, can't test easily)

Any workaround? Would be acceptable for me to find and replace all colons with underscore in column names, but not easy to do in a big set of nested struct columns...

Thanks

 

 

 spark.conf.set("spark.sql.orc.impl", "native")

 val dfColon = Seq(1).toDF("a:b")
 dfColon.printSchema()
 dfColon.show()
 dfColon.write.orc("test_colon")
 // Fails with IllegalArgumentException: Can't parse category at 'struct<a:b^:int>'
 
 import org.apache.spark.sql.functions.struct
 val dfColonStruct = dfColon.withColumn("x", struct($"a:b")).drop("a:b")
 dfColonStruct.printSchema()
 dfColonStruct.show()
 dfColon.write.orc("test_colon_struct")
 // Fails with IllegalArgumentException: Can't parse category at 'struct<x:struct<a:b^:int>>'