Github user JoshRosen commented on a diff in the pull request:
https://github.com/apache/spark/pull/1338#discussion_r15601428
--- Diff: core/src/main/scala/org/apache/spark/api/python/SerDeUtil.scala
---
@@ -65,20 +66,49 @@ private[python] object SerDeUtil extends Logging {
* by PySpark. By default, if serialization fails, toString is called
and the string
* representation is serialized
*/
- def rddToPython(rdd: RDD[(Any, Any)]): RDD[Array[Byte]] = {
+ def pairRDDToPython(rdd: RDD[(Any, Any)], batchSize: Int):
RDD[Array[Byte]] = {
val (keyFailed, valueFailed) = checkPickle(rdd.first())
rdd.mapPartitions { iter =>
val pickle = new Pickler
- iter.map { case (k, v) =>
- if (keyFailed && valueFailed) {
- pickle.dumps(Array(k.toString, v.toString))
- } else if (keyFailed) {
- pickle.dumps(Array(k.toString, v))
- } else if (!keyFailed && valueFailed) {
- pickle.dumps(Array(k, v.toString))
- } else {
- pickle.dumps(Array(k, v))
+ val cleaned = iter.map { case (k, v) =>
+ val key = if (keyFailed) k.toString else k
+ val value = if (valueFailed) v.toString else v
+ Array[Any](key, value)
+ }
+ if (batchSize > 1) {
+ cleaned.grouped(batchSize).map(batched =>
pickle.dumps(seqAsJavaList(batched)))
+ } else {
+ cleaned.map(pickle.dumps(_))
+ }
+ }
+ }
+
+ /**
+ * Convert an RDD of serialized Python tuple (K, V) to RDD[(K, V)].
+ */
+ def pythonToPairRDD[K, V](pyRDD: RDD[Array[Byte]]): RDD[(K, V)] = {
+ def isPair(obj: Any): Boolean = {
+
Option(obj.getClass.getComponentType).map(!_.isPrimitive).getOrElse(false) &&
+ obj.asInstanceOf[Array[_]].length == 2
+ }
+ pyRDD.mapPartitions { iter =>
+ val unpickle = new Unpickler
+ iter.flatMap { row =>
+ unpickle.loads(row) match {
+ // batch serialized Python RDDs
+ case objs: java.util.List[_] => objs
--- End diff --
This also affects `PythonRDD.pythonToJavaMap(); good catch.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
