[ORC-556] ConvertTreeReader can incorrectly be applied on columns of the same primitive type

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 1.6.0, 1.6.1
• Fix Version/s: 1.6.3, 1.7.0
• Assignee: Shardul Mahadik
• Reporter: Ratandeep Ratti
• Created: 03/Oct/19 17:07
• Updated: 27/Apr/20 04:27
• Resolved: 11/Dec/19 18:36

Description

I'm seeing the following exception when reading old ORC data with Iceberg

0.0 in stage 0.0 (TID 0, executor 1): java.lang.IllegalArgumentException: No conversion of type INT to self needed
	at org.apache.iceberg.shaded.org.apache.orc.impl.ConvertTreeReaderFactory.createAnyIntegerConvertTreeReader(ConvertTreeReaderFactory.java:1659)
	at org.apache.iceberg.shaded.org.apache.orc.impl.ConvertTreeReaderFactory.createConvertTreeReader(ConvertTreeReaderFactory.java:2112)
	at org.apache.iceberg.shaded.org.apache.orc.impl.TreeReaderFactory.createTreeReader(TreeReaderFactory.java:2327)
	at org.apache.iceberg.shaded.org.apache.orc.impl.TreeReaderFactory$StructTreeReader.<init>(TreeReaderFactory.java:1957)
	at org.apache.iceberg.shaded.org.apache.orc.impl.TreeReaderFactory.createTreeReader(TreeReaderFactory.java:2367)
	at org.apache.iceberg.shaded.org.apache.orc.impl.TreeReaderFactory$StructTreeReader.<init>(TreeReaderFactory.java:1957)
	at org.apache.iceberg.shaded.org.apache.orc.impl.TreeReaderFactory.createTreeReader(TreeReaderFactory.java:2367)
	at org.apache.iceberg.shaded.org.apache.orc.impl.RecordReaderImpl.<init>(RecordReaderImpl.java:230)
	at org.apache.iceberg.shaded.org.apache.orc.impl.ReaderImpl.rows(ReaderImpl.java:741)
	at org.apache.iceberg.orc.OrcIterable.newOrcIterator(OrcIterable.java:87)
	at org.apache.iceberg.orc.OrcIterable.iterator(OrcIterable.java:72)
	at org.apache.iceberg.spark.source.Reader$TaskDataReader.open(Reader.java:470)
	at org.apache.iceberg.spark.source.Reader$TaskDataReader.open(Reader.java:422)
	at org.apache.iceberg.spark.source.Reader$TaskDataReader.<init>(Reader.java:356)
	at org.apache.iceberg.spark.source.Reader$ReadTask.createPartitionReader(Reader.java:305)
	at org.apache.spark.sql.execution.datasources.v2.DataSourceRDD.compute(DataSourceRDD.scala:42)
	at org.apache.spark.rdd.RDD.computeOrReadCheckpoint(RDD.scala:324)
	at org.apache.spark.rdd.RDD.iterator(RDD.scala:288)
	at org.apache.spark.rdd.MapPartitionsRDD.compute(MapPartitionsRDD.scala:38)
	at org.apache.spark.rdd.RDD.computeOrReadCheckpoint(RDD.scala:324)
	at org.apache.spark.rdd.RDD.iterator(RDD.scala:288)

I think the problem lies in the following snippet in method org.apache.orc.impl.TreeReaderFactory#createTreeReader

if (!fileType.equals(readerType) &&
    ... // elided)) {
      ...
}

We are doing an equals comparison on the TypeDescription class. This equals comparison can now fail for at least 2 reasons

1. Reader schema has annotations [properties] and old file schema does not
2. Reader schema field name does not match in case with the file schema. This, I suspect, is because the old data was written by Hive.

At least 1 can be fixed if we change

fileType.equals(readerType) => fileType.getCategory().equals(readerType.getCategory()) 

I'm currently unsure of the repercussions of this so haven't made this change myself.

Issue Links

• links to: GitHub Pull Request #457