[GitHub] [spark] kazuyukitanimura commented on a diff in pull request #33721: [SPARK-32210][CORE] Fix NegativeArraySizeException in MapOutputTracker with large spark.default.parallelism

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

kazuyukitanimura commented on code in PR #33721:
URL: https://github.com/apache/spark/pull/33721#discussion_r844126152


##########
core/src/test/scala/org/apache/spark/MapOutputTrackerSuite.scala:
##########
@@ -664,4 +664,74 @@ class MapOutputTrackerSuite extends SparkFunSuite with LocalSparkContext {
     tracker.stop()
     rpcEnv.shutdown()
   }
+
+  test("SPARK-32210: serialize mapStatuses to a nested Array and deserialize them") {
+    val newConf = new SparkConf
+
+    // needs TorrentBroadcast so need a SparkContext
+    withSpark(new SparkContext("local", "MapOutputTrackerSuite", newConf)) { sc =>
+      val tracker = sc.env.mapOutputTracker.asInstanceOf[MapOutputTrackerMaster]
+      val rpcEnv = sc.env.rpcEnv
+      val masterEndpoint = new MapOutputTrackerMasterEndpoint(rpcEnv, tracker, sc.getConf)
+      rpcEnv.stop(tracker.trackerEndpoint)
+      rpcEnv.setupEndpoint(MapOutputTracker.ENDPOINT_NAME, masterEndpoint)
+      val shuffleId = 20
+      val numMaps = 1000
+
+      tracker.registerShuffle(shuffleId, numMaps, MergeStatus.SHUFFLE_PUSH_DUMMY_NUM_REDUCES)
+      val r = new scala.util.Random(912)
+      (0 until numMaps).foreach { i =>
+        tracker.registerMapOutput(shuffleId, i, HighlyCompressedMapStatus(
+          BlockManagerId(s"node$i", s"node$i.spark.apache.org", 1000),
+          Array.fill[Long](1000)((r.nextDouble() * 1024 * 1024 * 1024).toLong), i))
+      }
+
+      val shuffleStatus = tracker.shuffleStatuses.get(shuffleId).head
+      val (serializedMapStatus, serializedBroadcast) = MapOutputTracker.serializeOutputStatuses(
+        shuffleStatus.mapStatuses, tracker.broadcastManager, tracker.isLocal, 0, sc.getConf)
+      assert(serializedBroadcast.value.length > 1)
+      assert(serializedBroadcast.value.dropRight(1).forall(_.length == 1024 * 1024))
+
+      val result = MapOutputTracker.deserializeOutputStatuses(serializedMapStatus, sc.getConf)
+      assert(result.length == numMaps)
+
+      tracker.unregisterShuffle(shuffleId)
+      tracker.stop()
+    }
+  }
+
+  ignore("SPARK-32210: serialize and deserialize over 2GB compressed mapStatuses") {
+    // This test requires 8GB heap memory settings

Review Comment:
   @pan3793 `project/SparkBuild.scala` is hard-coding the `Xmx` sizes unfortunately. You may need to manually change them, something like https://github.com/apache/spark/pull/34214/files
   If you are using 8G should work. If you are using JDK 11 or 17, you may need more than 8G



-- 
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