jonvex commented on code in PR #10956:
URL: https://github.com/apache/hudi/pull/10956#discussion_r1566258283
##########
hudi-spark-datasource/hudi-spark2/src/main/scala/org/apache/spark/sql/execution/datasources/parquet/Spark24ParquetReader.scala:
##########
@@ -156,30 +177,51 @@ class Spark24ParquetReader(enableVectorizedReader:
Boolean,
iter.asInstanceOf[Iterator[InternalRow]]
} else {
// ParquetRecordReader returns UnsafeRow
+ val readSupport = new ParquetReadSupport(convertTz)
val reader = if (pushed.isDefined && enableRecordFilter) {
val parquetFilter = FilterCompat.get(pushed.get, null)
- new ParquetRecordReader[UnsafeRow](new ParquetReadSupport(convertTz),
parquetFilter)
+ new ParquetRecordReader[UnsafeRow](readSupport, parquetFilter)
} else {
- new ParquetRecordReader[UnsafeRow](new ParquetReadSupport(convertTz))
+ new ParquetRecordReader[UnsafeRow](readSupport)
}
val iter = new RecordReaderIterator(reader)
// SPARK-23457 Register a task completion lister before `initialization`.
taskContext.foreach(_.addTaskCompletionListener[Unit](_ => iter.close()))
reader.initialize(split, hadoopAttemptContext)
val fullSchema = requiredSchema.toAttributes ++
partitionSchema.toAttributes
- val joinedRow = new JoinedRow()
- val appendPartitionColumns =
GenerateUnsafeProjection.generate(fullSchema, fullSchema)
+ val unsafeProjection = if (implicitTypeChangeInfos.isEmpty) {
+ GenerateUnsafeProjection.generate(fullSchema, fullSchema)
+ } else {
+ val newFullSchema = new
StructType(requiredSchema.fields.zipWithIndex.map { case (f, i) =>
+ if (implicitTypeChangeInfos.containsKey(i)) {
+ StructField(f.name, implicitTypeChangeInfos.get(i).getRight,
f.nullable, f.metadata)
+ } else f
+ }).toAttributes ++ partitionSchema.toAttributes
+ val castSchema = newFullSchema.zipWithIndex.map { case (attr, i) =>
+ if (implicitTypeChangeInfos.containsKey(i)) {
+ val srcType = implicitTypeChangeInfos.get(i).getRight
+ val dstType = implicitTypeChangeInfos.get(i).getLeft
+ val needTimeZone = Cast.needsTimeZone(srcType, dstType)
+ Cast(attr, dstType, if (needTimeZone) timeZoneId else None)
+ } else attr
+ }
+ GenerateUnsafeProjection.generate(castSchema, newFullSchema)
+ }
Review Comment:
Spark 2 schema evolution diverges, so we just do what's in the legacy
format. We can follow up and discuss with the original authors of schema on read
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
