[GitHub] [spark] mridulm commented on a diff in pull request #37728: [SPARK-40276][CORE] Reduce the result size of RDD.takeOrdered

[GitBox <gi...@apache.org>]

mridulm commented on code in PR #37728:
URL: https://github.com/apache/spark/pull/37728#discussion_r958815540


##########
core/src/main/scala/org/apache/spark/rdd/RDD.scala:
##########
@@ -1523,22 +1523,21 @@ abstract class RDD[T: ClassTag](
    * @return an array of top elements
    */
   def takeOrdered(num: Int)(implicit ord: Ordering[T]): Array[T] = withScope {
-    if (num == 0) {
+    if (num == 0 || this.getNumPartitions == 0) {
       Array.empty
     } else {
-      val mapRDDs = mapPartitions { items =>
-        // Priority keeps the largest elements, so let's reverse the ordering.
-        val queue = new BoundedPriorityQueue[T](num)(ord.reverse)
-        queue ++= collectionUtils.takeOrdered(items, num)(ord)
-        Iterator.single(queue)
-      }
-      if (mapRDDs.partitions.length == 0) {
-        Array.empty
-      } else {
-        mapRDDs.reduce { (queue1, queue2) =>
-          queue1 ++= queue2
-          queue1
-        }.toArray.sorted(ord)
+      this.mapPartitionsWithIndex { case (pid, iter) =>
+        if (iter.nonEmpty) {
+          // Priority keeps the largest elements, so let's reverse the ordering.
+          Iterator.single(collectionUtils.takeOrdered(iter, num)(ord).toArray)
+        } else if (pid == 0) {
+          // make sure partition 0 always returns an array to avoid reduce on empty RDD
+          Iterator.single(Array.empty[T])
+        } else {
+          Iterator.empty
+        }
+      }.reduce { (array1, array2) =>
+        collectionUtils.mergeOrdered[T](Seq(array1, array2))(ord).take(num).toArray

Review Comment:
   Nit: you can statically create the array upfront (`min(num, array1.length + array2.length)` ) and copy into it, instead of `toArray`



##########
core/src/main/scala/org/apache/spark/util/collection/Utils.scala:
##########
@@ -37,6 +37,23 @@ private[spark] object Utils {
     ordering.leastOf(input.asJava, num).iterator.asScala
   }
 
+  /**
+   * Returns an iterator over the merged contents of all given iterators,
+   * traversing every element of the input iterators.
+   * Equivalent entries will not be de-duplicated.
+   *
+   * Callers must ensure that the source iterators are already sorted by
+   * the same ordering `ord`, otherwise the result is likely to be incorrect.
+   */
+  def mergeOrdered[T](iterators: Iterable[TraversableOnce[T]])(

Review Comment:
   nit: Rename `iterators` ?



-- 
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: reviews-unsubscribe@spark.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org


---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For additional commands, e-mail: reviews-help@spark.apache.org