Github user icexelloss commented on a diff in the pull request:
https://github.com/apache/spark/pull/18732#discussion_r141886413
--- Diff:
sql/core/src/main/scala/org/apache/spark/sql/execution/python/ArrowEvalPythonExec.scala
---
@@ -44,14 +44,17 @@ case class ArrowEvalPythonExec(udfs: Seq[PythonUDF],
output: Seq[Attribute], chi
val schemaOut =
StructType.fromAttributes(output.drop(child.output.length).zipWithIndex
.map { case (attr, i) => attr.withName(s"_$i") })
+ val batchedIter: Iterator[Iterator[InternalRow]] =
+ iter.grouped(conf.arrowMaxRecordsPerBatch).map(_.iterator)
+
--- End diff --
I actually find this code doesn't work now. I will fix it.
@ueshin is right, this is to reuse `ArrowEvalPython` for both the current
pandas udf and `apply()`. I basically want to lift the batching logic out of
`ArrowEvalPython` so the called and decide how they want rows to be batched
into `RecordBatch`.
In the current pandas udf case, it batches it by
`conf.arrowMaxRecordsPerBatch` and in `apply` it batches by one group per batch.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org
For additional commands, e-mail: reviews-h...@spark.apache.org
