Github user cloud-fan commented on a diff in the pull request:
https://github.com/apache/spark/pull/22009#discussion_r208439490
--- Diff:
sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/v2/DataSourceRDD.scala
---
@@ -51,18 +58,19 @@ class DataSourceRDD[T: ClassTag](
valuePrepared
}
- override def next(): T = {
+ override def next(): Any = {
if (!hasNext) {
throw new java.util.NoSuchElementException("End of stream")
}
valuePrepared = false
reader.get()
}
}
- new InterruptibleIterator(context, iter)
+ // TODO: get rid of this type hack.
+ new InterruptibleIterator(context,
iter.asInstanceOf[Iterator[InternalRow]])
--- End diff --
The problem is that, we don't really have a batch API in Spark SQL. We rely
on type erasure and codegen hack to implement columnar scan. It's hardcoded in
the engine: `SparkPlan#execute` returns `RDD[InternalRow]`.
if we have a RDD iterate over the rows in the batch, then whole stage
codegen will break, as it iterates the input RDD and cast the record to
`ColumnarBatch`.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
