tianhanhu-db commented on code in PR #42618:
URL: https://github.com/apache/spark/pull/42618#discussion_r1308051615
##########
connector/avro/src/main/scala/org/apache/spark/sql/avro/SchemaConverters.scala:
##########
@@ -142,18 +143,30 @@ object SchemaConverters {
if (avroSchema.getTypes.asScala.exists(_.getType == NULL)) {
// In case of a union with null, eliminate it and make a recursive
call
val remainingUnionTypes = AvroUtils.nonNullUnionBranches(avroSchema)
- if (remainingUnionTypes.size == 1) {
- toSqlTypeHelper(remainingUnionTypes.head, existingRecordNames,
avroOptions)
- .copy(nullable = true)
- } else {
- toSqlTypeHelper(
- Schema.createUnion(remainingUnionTypes.asJava),
- existingRecordNames,
- avroOptions).copy(nullable = true)
- }
+ toSqlTypeHelper(
+ Schema.createUnion(remainingUnionTypes.asJava),
+ existingRecordNames,
+ avroOptions).copy(nullable = true)
} else avroSchema.getTypes.asScala.map(_.getType).toSeq match {
case Seq(t1) =>
- toSqlTypeHelper(avroSchema.getTypes.get(0), existingRecordNames,
avroOptions)
+ // If spark.sql.avro.alwaysConvertUnionToStructType is set to
false (default),
+ // we convert Avro union with a single primitive type into a
primitive Spark type
+ // instead of a StructType.
+ if (!SQLConf.get.avroAlwaysConvertUnionToStruct) {
+ toSqlTypeHelper(avroSchema.getTypes.get(0), existingRecordNames,
avroOptions)
Review Comment:
I don't think the two approaches are equivalent.
```scala
SchemaConverters.toSqlType(avroSchema, options).dataType match {
case t: StructType => Some(t)
case _ => throw new RuntimeException(
s"""Avro schema cannot be converted to a Spark SQL StructType:
|
|${avroSchema.toString(true)}
|""".stripMargin)
}
```
is for converting Avro top-level records into Spark StructType IIUC. In
deserialization, we use `getRecordWriter` to wrap around Avro schema that are
non-records into records, and in this method we get it back.
##########
connector/avro/src/main/scala/org/apache/spark/sql/avro/SchemaConverters.scala:
##########
@@ -142,18 +143,30 @@ object SchemaConverters {
if (avroSchema.getTypes.asScala.exists(_.getType == NULL)) {
// In case of a union with null, eliminate it and make a recursive
call
val remainingUnionTypes = AvroUtils.nonNullUnionBranches(avroSchema)
- if (remainingUnionTypes.size == 1) {
- toSqlTypeHelper(remainingUnionTypes.head, existingRecordNames,
avroOptions)
- .copy(nullable = true)
- } else {
- toSqlTypeHelper(
- Schema.createUnion(remainingUnionTypes.asJava),
- existingRecordNames,
- avroOptions).copy(nullable = true)
- }
+ toSqlTypeHelper(
+ Schema.createUnion(remainingUnionTypes.asJava),
+ existingRecordNames,
+ avroOptions).copy(nullable = true)
} else avroSchema.getTypes.asScala.map(_.getType).toSeq match {
case Seq(t1) =>
- toSqlTypeHelper(avroSchema.getTypes.get(0), existingRecordNames,
avroOptions)
+ // If spark.sql.avro.alwaysConvertUnionToStructType is set to
false (default),
+ // we convert Avro union with a single primitive type into a
primitive Spark type
+ // instead of a StructType.
+ if (!SQLConf.get.avroAlwaysConvertUnionToStruct) {
+ toSqlTypeHelper(avroSchema.getTypes.get(0), existingRecordNames,
avroOptions)
Review Comment:
When we call this method, we expect the datatype to be `StructType` already,
and that is why all other matches would throw exceptions.
--
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]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
