Merge pull request apache#401 from andrewor14/master

External sorting - Add number of bytes spilled to Web UI

Additionally, update test suite for external sorting to induce spilling.

core/src/main/scala/org/apache/spark/Aggregator.scala

@@ -34,7 +34,8 @@ case class Aggregator[K, V, C] (
   private val sparkConf = SparkEnv.get.conf
   private val externalSorting = sparkConf.getBoolean("spark.shuffle.externalSorting", true)
 
-  def combineValuesByKey(iter: Iterator[_ <: Product2[K, V]]) : Iterator[(K, C)] = {
+  def combineValuesByKey(iter: Iterator[_ <: Product2[K, V]],
+                         context: TaskContext) : Iterator[(K, C)] = {
     if (!externalSorting) {
       val combiners = new AppendOnlyMap[K,C]
       var kv: Product2[K, V] = null
@@ -47,17 +48,18 @@ case class Aggregator[K, V, C] (
       }
       combiners.iterator
     } else {
-      val combiners =
-        new ExternalAppendOnlyMap[K, V, C](createCombiner, mergeValue, mergeCombiners)
+      val combiners = new ExternalAppendOnlyMap[K, V, C](createCombiner, mergeValue, mergeCombiners)
       while (iter.hasNext) {
         val (k, v) = iter.next()
         combiners.insert(k, v)
       }
+      context.taskMetrics.memoryBytesSpilled = combiners.memoryBytesSpilled
+      context.taskMetrics.diskBytesSpilled = combiners.diskBytesSpilled
       combiners.iterator
     }
   }
 
-  def combineCombinersByKey(iter: Iterator[(K, C)]) : Iterator[(K, C)] = {
+  def combineCombinersByKey(iter: Iterator[(K, C)], context: TaskContext) : Iterator[(K, C)] = {
     if (!externalSorting) {
       val combiners = new AppendOnlyMap[K,C]
       var kc: Product2[K, C] = null
@@ -75,6 +77,8 @@ case class Aggregator[K, V, C] (
         val (k, c) = iter.next()
         combiners.insert(k, c)
       }
+      context.taskMetrics.memoryBytesSpilled = combiners.memoryBytesSpilled
+      context.taskMetrics.diskBytesSpilled = combiners.diskBytesSpilled
       combiners.iterator
     }
   }

core/src/main/scala/org/apache/spark/executor/Executor.scala

@@ -279,7 +279,7 @@ private[spark] class Executor(
           //System.exit(1)
         }
       } finally {
-        // TODO: Unregister shuffle memory only for ShuffleMapTask
+        // TODO: Unregister shuffle memory only for ResultTask
         val shuffleMemoryMap = env.shuffleMemoryMap
         shuffleMemoryMap.synchronized {
           shuffleMemoryMap.remove(Thread.currentThread().getId)

core/src/main/scala/org/apache/spark/executor/TaskMetrics.scala

@@ -48,6 +48,16 @@ class TaskMetrics extends Serializable {
    */
   var resultSerializationTime: Long = _
 
+  /**
+   * The number of in-memory bytes spilled by this task
+   */
+  var memoryBytesSpilled: Long = _
+
+  /**
+   * The number of on-disk bytes spilled by this task
+   */
+  var diskBytesSpilled: Long = _
+
   /**
    * If this task reads from shuffle output, metrics on getting shuffle data will be collected here
    */

core/src/main/scala/org/apache/spark/rdd/CoGroupedRDD.scala

@@ -106,6 +106,7 @@ class CoGroupedRDD[K](@transient var rdds: Seq[RDD[_ <: Product2[K, _]]], part:
   override val partitioner = Some(part)
 
   override def compute(s: Partition, context: TaskContext): Iterator[(K, CoGroupCombiner)] = {
+
     val externalSorting = sparkConf.getBoolean("spark.shuffle.externalSorting", true)
     val split = s.asInstanceOf[CoGroupPartition]
     val numRdds = split.deps.size
@@ -150,6 +151,8 @@ class CoGroupedRDD[K](@transient var rdds: Seq[RDD[_ <: Product2[K, _]]], part:
           map.insert(kv._1, new CoGroupValue(kv._2, depNum))
         }
       }
+      context.taskMetrics.memoryBytesSpilled = map.memoryBytesSpilled
+      context.taskMetrics.diskBytesSpilled = map.diskBytesSpilled
       new InterruptibleIterator(context, map.iterator)
     }
   }

core/src/main/scala/org/apache/spark/rdd/PairRDDFunctions.scala

@@ -88,20 +88,22 @@ class PairRDDFunctions[K: ClassTag, V: ClassTag](self: RDD[(K, V)])
     val aggregator = new Aggregator[K, V, C](createCombiner, mergeValue, mergeCombiners)
     if (self.partitioner == Some(partitioner)) {
       self.mapPartitionsWithContext((context, iter) => {
-        new InterruptibleIterator(context, aggregator.combineValuesByKey(iter))
+        new InterruptibleIterator(context, aggregator.combineValuesByKey(iter, context))
       }, preservesPartitioning = true)
     } else if (mapSideCombine) {
-      val combined = self.mapPartitions(aggregator.combineValuesByKey, preservesPartitioning = true)
+      val combined = self.mapPartitionsWithContext((context, iter) => {
+        aggregator.combineValuesByKey(iter, context)
+      }, preservesPartitioning = true)
       val partitioned = new ShuffledRDD[K, C, (K, C)](combined, partitioner)
         .setSerializer(serializerClass)
       partitioned.mapPartitionsWithContext((context, iter) => {
-        new InterruptibleIterator(context, aggregator.combineCombinersByKey(iter))
+        new InterruptibleIterator(context, aggregator.combineCombinersByKey(iter, context))
       }, preservesPartitioning = true)
     } else {
       // Don't apply map-side combiner.
       val values = new ShuffledRDD[K, V, (K, V)](self, partitioner).setSerializer(serializerClass)
       values.mapPartitionsWithContext((context, iter) => {
-        new InterruptibleIterator(context, aggregator.combineValuesByKey(iter))
+        new InterruptibleIterator(context, aggregator.combineValuesByKey(iter, context))
       }, preservesPartitioning = true)
     }
   }

core/src/main/scala/org/apache/spark/ui/jobs/ExecutorSummary.scala

@@ -24,4 +24,6 @@ private[spark] class ExecutorSummary {
   var succeededTasks : Int = 0
   var shuffleRead : Long = 0
   var shuffleWrite : Long = 0
+  var memoryBytesSpilled : Long = 0
+  var diskBytesSpilled : Long = 0
 }

core/src/main/scala/org/apache/spark/ui/jobs/ExecutorTable.scala

@@ -48,6 +48,8 @@ private[spark] class ExecutorTable(val parent: JobProgressUI, val stageId: Int)
         <th>Succeeded Tasks</th>
         <th>Shuffle Read</th>
         <th>Shuffle Write</th>
+        <th>Shuffle Spill (Memory)</th>
+        <th>Shuffle Spill (Disk)</th>
       </thead>
       <tbody>
         {createExecutorTable()}
@@ -80,6 +82,8 @@ private[spark] class ExecutorTable(val parent: JobProgressUI, val stageId: Int)
               <td>{v.succeededTasks}</td>
               <td>{Utils.bytesToString(v.shuffleRead)}</td>
               <td>{Utils.bytesToString(v.shuffleWrite)}</td>
+              <td>{Utils.bytesToString(v.memoryBytesSpilled)}</td>
+              <td>{Utils.bytesToString(v.diskBytesSpilled)}</td>
             </tr>
           }
         }

core/src/main/scala/org/apache/spark/ui/jobs/JobProgressListener.scala

@@ -52,6 +52,8 @@ private[spark] class JobProgressListener(val sc: SparkContext) extends SparkList
   val stageIdToTime = HashMap[Int, Long]()
   val stageIdToShuffleRead = HashMap[Int, Long]()
   val stageIdToShuffleWrite = HashMap[Int, Long]()
+  val stageIdToMemoryBytesSpilled = HashMap[Int, Long]()
+  val stageIdToDiskBytesSpilled = HashMap[Int, Long]()
   val stageIdToTasksActive = HashMap[Int, HashSet[TaskInfo]]()
   val stageIdToTasksComplete = HashMap[Int, Int]()
   val stageIdToTasksFailed = HashMap[Int, Int]()
@@ -78,6 +80,8 @@ private[spark] class JobProgressListener(val sc: SparkContext) extends SparkList
         stageIdToTime.remove(s.stageId)
         stageIdToShuffleRead.remove(s.stageId)
         stageIdToShuffleWrite.remove(s.stageId)
+        stageIdToMemoryBytesSpilled.remove(s.stageId)
+        stageIdToDiskBytesSpilled.remove(s.stageId)
         stageIdToTasksActive.remove(s.stageId)
         stageIdToTasksComplete.remove(s.stageId)
         stageIdToTasksFailed.remove(s.stageId)
@@ -149,6 +153,8 @@ private[spark] class JobProgressListener(val sc: SparkContext) extends SparkList
         Option(taskEnd.taskMetrics).foreach { taskMetrics =>
           taskMetrics.shuffleReadMetrics.foreach { y.shuffleRead += _.remoteBytesRead }
           taskMetrics.shuffleWriteMetrics.foreach { y.shuffleWrite += _.shuffleBytesWritten }
+          y.memoryBytesSpilled += taskMetrics.memoryBytesSpilled
+          y.diskBytesSpilled += taskMetrics.diskBytesSpilled
         }
       }
       case _ => {}
@@ -184,6 +190,14 @@ private[spark] class JobProgressListener(val sc: SparkContext) extends SparkList
     stageIdToShuffleWrite(sid) += shuffleWrite
     totalShuffleWrite += shuffleWrite
 
+    stageIdToMemoryBytesSpilled.getOrElseUpdate(sid, 0L)
+    val memoryBytesSpilled = metrics.map(m => m.memoryBytesSpilled).getOrElse(0L)
+    stageIdToMemoryBytesSpilled(sid) += memoryBytesSpilled
+
+    stageIdToDiskBytesSpilled.getOrElseUpdate(sid, 0L)
+    val diskBytesSpilled = metrics.map(m => m.diskBytesSpilled).getOrElse(0L)
+    stageIdToDiskBytesSpilled(sid) += diskBytesSpilled
+
     val taskList = stageIdToTaskInfos.getOrElse(
       sid, HashSet[(TaskInfo, Option[TaskMetrics], Option[ExceptionFailure])]())
     taskList -= ((taskEnd.taskInfo, None, None))

core/src/main/scala/org/apache/spark/ui/jobs/StagePage.scala

@@ -56,6 +56,9 @@ private[spark] class StagePage(parent: JobProgressUI) {
       val hasShuffleRead = shuffleReadBytes > 0
       val shuffleWriteBytes = listener.stageIdToShuffleWrite.getOrElse(stageId, 0L)
       val hasShuffleWrite = shuffleWriteBytes > 0
+      val memoryBytesSpilled = listener.stageIdToMemoryBytesSpilled.getOrElse(stageId, 0L)
+      val diskBytesSpilled = listener.stageIdToDiskBytesSpilled.getOrElse(stageId, 0L)
+      val hasBytesSpilled = (memoryBytesSpilled > 0 && diskBytesSpilled > 0)
 
       var activeTime = 0L
       listener.stageIdToTasksActive(stageId).foreach(activeTime += _.timeRunning(now))
@@ -81,6 +84,16 @@ private[spark] class StagePage(parent: JobProgressUI) {
                 {Utils.bytesToString(shuffleWriteBytes)}
               </li>
             }
+            {if (hasBytesSpilled)
+            <li>
+              <strong>Shuffle spill (memory): </strong>
+              {Utils.bytesToString(memoryBytesSpilled)}
+            </li>
+            <li>
+              <strong>Shuffle spill (disk): </strong>
+              {Utils.bytesToString(diskBytesSpilled)}
+            </li>
+            }
           </ul>
         </div>
 
@@ -89,9 +102,10 @@ private[spark] class StagePage(parent: JobProgressUI) {
         Seq("Duration", "GC Time", "Result Ser Time") ++
         {if (hasShuffleRead) Seq("Shuffle Read")  else Nil} ++
         {if (hasShuffleWrite) Seq("Write Time", "Shuffle Write") else Nil} ++
+        {if (hasBytesSpilled) Seq("Shuffle Spill (Memory)", "Shuffle Spill (Disk)") else Nil} ++
         Seq("Errors")
 
-      val taskTable = listingTable(taskHeaders, taskRow(hasShuffleRead, hasShuffleWrite), tasks)
+      val taskTable = listingTable(taskHeaders, taskRow(hasShuffleRead, hasShuffleWrite, hasBytesSpilled), tasks)
 
       // Excludes tasks which failed and have incomplete metrics
       val validTasks = tasks.filter(t => t._1.status == "SUCCESS" && (t._2.isDefined))
@@ -153,13 +167,29 @@ private[spark] class StagePage(parent: JobProgressUI) {
           }
           val shuffleWriteQuantiles = "Shuffle Write" +: getQuantileCols(shuffleWriteSizes)
 
+          val memoryBytesSpilledSizes = validTasks.map {
+            case(info, metrics, exception) =>
+              metrics.get.memoryBytesSpilled.toDouble
+          }
+          val memoryBytesSpilledQuantiles = "Shuffle spill (memory)" +:
+            getQuantileCols(memoryBytesSpilledSizes)
+
+          val diskBytesSpilledSizes = validTasks.map {
+            case(info, metrics, exception) =>
+              metrics.get.diskBytesSpilled.toDouble
+          }
+          val diskBytesSpilledQuantiles = "Shuffle spill (disk)" +:
+            getQuantileCols(diskBytesSpilledSizes)
+
           val listings: Seq[Seq[String]] = Seq(
             serializationQuantiles,
             serviceQuantiles,
             gettingResultQuantiles,
             schedulerDelayQuantiles,
             if (hasShuffleRead) shuffleReadQuantiles else Nil,
-            if (hasShuffleWrite) shuffleWriteQuantiles else Nil)
+            if (hasShuffleWrite) shuffleWriteQuantiles else Nil,
+            if (hasBytesSpilled) memoryBytesSpilledQuantiles else Nil,
+            if (hasBytesSpilled) diskBytesSpilledQuantiles else Nil)
 
           val quantileHeaders = Seq("Metric", "Min", "25th percentile",
             "Median", "75th percentile", "Max")
@@ -178,8 +208,7 @@ private[spark] class StagePage(parent: JobProgressUI) {
     }
   }
 
-
-  def taskRow(shuffleRead: Boolean, shuffleWrite: Boolean)
+  def taskRow(shuffleRead: Boolean, shuffleWrite: Boolean, bytesSpilled: Boolean)
              (taskData: (TaskInfo, Option[TaskMetrics], Option[ExceptionFailure])): Seq[Node] = {
     def fmtStackTrace(trace: Seq[StackTraceElement]): Seq[Node] =
       trace.map(e => <span style="display:block;">{e.toString}</span>)
@@ -205,6 +234,14 @@ private[spark] class StagePage(parent: JobProgressUI) {
     val writeTimeReadable = maybeWriteTime.map{ t => t / (1000 * 1000)}.map{ ms =>
       if (ms == 0) "" else parent.formatDuration(ms)}.getOrElse("")
 
+    val maybeMemoryBytesSpilled = metrics.map{m => m.memoryBytesSpilled}
+    val memoryBytesSpilledSortable = maybeMemoryBytesSpilled.map(_.toString).getOrElse("")
+    val memoryBytesSpilledReadable = maybeMemoryBytesSpilled.map{Utils.bytesToString(_)}.getOrElse("")
+
+    val maybeDiskBytesSpilled = metrics.map{m => m.diskBytesSpilled}
+    val diskBytesSpilledSortable = maybeDiskBytesSpilled.map(_.toString).getOrElse("")
+    val diskBytesSpilledReadable = maybeDiskBytesSpilled.map{Utils.bytesToString(_)}.getOrElse("")
+
     <tr>
       <td>{info.index}</td>
       <td>{info.taskId}</td>
@@ -234,6 +271,14 @@ private[spark] class StagePage(parent: JobProgressUI) {
            {shuffleWriteReadable}
          </td>
       }}
+      {if (bytesSpilled) {
+        <td sorttable_customkey={memoryBytesSpilledSortable}>
+          {memoryBytesSpilledReadable}
+        </td>
+        <td sorttable_customkey={diskBytesSpilledSortable}>
+          {diskBytesSpilledReadable}
+        </td>
+      }}
       <td>{exception.map(e =>
         <span>
           {e.className} ({e.description})<br/>

core/src/main/scala/org/apache/spark/util/collection/ExternalAppendOnlyMap.scala

@@ -77,14 +77,18 @@ private[spark] class ExternalAppendOnlyMap[K, V, C](
   }
 
   // Number of pairs in the in-memory map
-  private var numPairsInMemory = 0
+  private var numPairsInMemory = 0L
 
   // Number of in-memory pairs inserted before tracking the map's shuffle memory usage
   private val trackMemoryThreshold = 1000
 
   // How many times we have spilled so far
   private var spillCount = 0
 
+  // Number of bytes spilled in total
+  private var _memoryBytesSpilled = 0L
+  private var _diskBytesSpilled = 0L
+
   private val fileBufferSize = sparkConf.getInt("spark.shuffle.file.buffer.kb", 100) * 1024
   private val syncWrites = sparkConf.getBoolean("spark.shuffle.sync", false)
   private val comparator = new KCComparator[K, C]
@@ -150,6 +154,7 @@ private[spark] class ExternalAppendOnlyMap[K, V, C](
       writer.commit()
     } finally {
       // Partial failures cannot be tolerated; do not revert partial writes
+      _diskBytesSpilled += writer.bytesWritten
       writer.close()
     }
     currentMap = new SizeTrackingAppendOnlyMap[K, C]
@@ -161,8 +166,12 @@ private[spark] class ExternalAppendOnlyMap[K, V, C](
       shuffleMemoryMap(Thread.currentThread().getId) = 0
     }
     numPairsInMemory = 0
+    _memoryBytesSpilled += mapSize
   }
 
+  def memoryBytesSpilled: Long = _memoryBytesSpilled
+  def diskBytesSpilled: Long = _diskBytesSpilled
+
   /**
    * Return an iterator that merges the in-memory map with the spilled maps.
    * If no spill has occurred, simply return the in-memory map's iterator.