holdenk commented on a change in pull request #24070: [SPARK-23961][PYTHON] Fix
error when toLocalIterator goes out of scope
URL: https://github.com/apache/spark/pull/24070#discussion_r270947475
##########
File path: core/src/main/scala/org/apache/spark/api/python/PythonRDD.scala
##########
@@ -168,7 +168,42 @@ private[spark] object PythonRDD extends Logging {
}
def toLocalIteratorAndServe[T](rdd: RDD[T]): Array[Any] = {
- serveIterator(rdd.toLocalIterator, s"serve toLocalIterator")
+ val (port, secret) = SocketAuthServer.setupOneConnectionServer(
+ authHelper, "serve toLocalIterator") { s =>
+ val out = new DataOutputStream(s.getOutputStream)
+ val in = new DataInputStream(s.getInputStream)
+ Utils.tryWithSafeFinally {
+
+ // Collects a partition on each iteration
+ val collectPartitionIter = rdd.partitions.indices.iterator.map { i =>
+ rdd.sparkContext.runJob(rdd, (iter: Iterator[Any]) => iter.toArray,
Seq(i)).head
+ }
+
+ // Read request for data and send next partition if nonzero
+ var complete = false
+ while (in.readInt() != 0 && !complete) {
+ if (collectPartitionIter.hasNext) {
+
+ // Send response there is a partition to read
+ out.writeInt(1)
+
+ // Write the next object and signal end of data for this iteration
+ val partitionArray = collectPartitionIter.next()
+ writeIteratorToStream(partitionArray.toIterator, out)
Review comment:
Would it make sense to use something which we can interrupt here instead if
we get a message from Python (or if the socket closes) that we don't need to
send more data?
----------------------------------------------------------------
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.
For queries about this service, please contact Infrastructure at:
[email protected]
With regards,
Apache Git Services
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
