[SPARK-1592][streaming] Automatically remove streaming input blocks

The raw input data is stored as blocks in BlockManagers. Earlier they were cleared by cleaner ttl. Now since streaming does not require cleaner TTL to be set, the block would not get cleared. This increases up the Spark's memory usage, which is not even accounted and shown in the Spark storage UI. It may cause the data blocks to spill over to disk, which eventually slows down the receiving of data (persisting to memory become bottlenecked by writing to disk).

The solution in this PR is to automatically remove those blocks. The mechanism to keep track of which BlockRDDs (which has presents the raw data blocks as a RDD) can be safely cleared already exists. Just use it to explicitly remove blocks from BlockRDDs.

Author: Tathagata Das <[email protected]>

Closes apache#512 from tdas/block-rdd-unpersist and squashes the following commits:

d25e610 [Tathagata Das] Merge remote-tracking branch 'apache/master' into block-rdd-unpersist
5f46d69 [Tathagata Das] Merge remote-tracking branch 'apache/master' into block-rdd-unpersist
2c320cd [Tathagata Das] Updated configuration with spark.streaming.unpersist setting.
2d4b2fd [Tathagata Das] Automatically removed input blocks

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

@@ -19,24 +19,30 @@ package org.apache.spark.rdd
 
 import scala.reflect.ClassTag
 
-import org.apache.spark.{Partition, SparkContext, SparkEnv, TaskContext}
+import org.apache.spark._
 import org.apache.spark.storage.{BlockId, BlockManager}
+import scala.Some
 
 private[spark] class BlockRDDPartition(val blockId: BlockId, idx: Int) extends Partition {
   val index = idx
 }
 
 private[spark]
-class BlockRDD[T: ClassTag](sc: SparkContext, @transient blockIds: Array[BlockId])
+class BlockRDD[T: ClassTag](@transient sc: SparkContext, @transient val blockIds: Array[BlockId])
   extends RDD[T](sc, Nil) {
 
   @transient lazy val locations_ = BlockManager.blockIdsToHosts(blockIds, SparkEnv.get)
+  @volatile private var _isValid = true
 
-  override def getPartitions: Array[Partition] = (0 until blockIds.size).map(i => {
-    new BlockRDDPartition(blockIds(i), i).asInstanceOf[Partition]
-  }).toArray
+  override def getPartitions: Array[Partition] = {
+    assertValid()
+    (0 until blockIds.size).map(i => {
+      new BlockRDDPartition(blockIds(i), i).asInstanceOf[Partition]
+    }).toArray
+  }
 
   override def compute(split: Partition, context: TaskContext): Iterator[T] = {
+    assertValid()
     val blockManager = SparkEnv.get.blockManager
     val blockId = split.asInstanceOf[BlockRDDPartition].blockId
     blockManager.get(blockId) match {
@@ -47,7 +53,36 @@ class BlockRDD[T: ClassTag](sc: SparkContext, @transient blockIds: Array[BlockId
   }
 
   override def getPreferredLocations(split: Partition): Seq[String] = {
+    assertValid()
     locations_(split.asInstanceOf[BlockRDDPartition].blockId)
   }
+
+  /**
+   * Remove the data blocks that this BlockRDD is made from. NOTE: This is an
+   * irreversible operation, as the data in the blocks cannot be recovered back
+   * once removed. Use it with caution.
+   */
+  private[spark] def removeBlocks() {
+    blockIds.foreach { blockId =>
+      sc.env.blockManager.master.removeBlock(blockId)
+    }
+    _isValid = false
+  }
+
+  /**
+   * Whether this BlockRDD is actually usable. This will be false if the data blocks have been
+   * removed using `this.removeBlocks`.
+   */
+  private[spark] def isValid: Boolean = {
+    _isValid
+  }
+
+  /** Check if this BlockRDD is valid. If not valid, exception is thrown. */
+  private[spark] def assertValid() {
+    if (!_isValid) {
+      throw new SparkException(
+        "Attempted to use %s after its blocks have been removed!".format(toString))
+    }
+  }
 }
 

docs/configuration.md

@@ -469,10 +469,13 @@ Apart from these, the following properties are also available, and may be useful
 </tr>
 <tr>
   <td>spark.streaming.unpersist</td>
-  <td>false</td>
+  <td>true</td>
   <td>
     Force RDDs generated and persisted by Spark Streaming to be automatically unpersisted from
-    Spark's memory. Setting this to true is likely to reduce Spark's RDD memory usage.
+    Spark's memory. The raw input data received by Spark Streaming is also automatically cleared.
+    Setting this to false will allow the raw data and persisted RDDs to be accessible outside the
+    streaming application as they will not be cleared automatically. But it comes at the cost of
+    higher memory usage in Spark.
   </td>
 </tr>
 <tr>

streaming/src/main/scala/org/apache/spark/streaming/Time.scala

@@ -68,5 +68,5 @@ case class Time(private val millis: Long) {
 }
 
 object Time {
-  val ordering = Ordering.by((time: Time) => time.millis)
+  implicit val ordering = Ordering.by((time: Time) => time.millis)
 }

streaming/src/main/scala/org/apache/spark/streaming/dstream/DStream.scala

@@ -25,7 +25,7 @@ import scala.reflect.ClassTag
 import java.io.{IOException, ObjectInputStream, ObjectOutputStream}
 
 import org.apache.spark.Logging
-import org.apache.spark.rdd.RDD
+import org.apache.spark.rdd.{BlockRDD, RDD}
 import org.apache.spark.storage.StorageLevel
 import org.apache.spark.util.MetadataCleaner
 import org.apache.spark.streaming._
@@ -340,13 +340,23 @@ abstract class DStream[T: ClassTag] (
    * this to clear their own metadata along with the generated RDDs.
    */
   private[streaming] def clearMetadata(time: Time) {
+    val unpersistData = ssc.conf.getBoolean("spark.streaming.unpersist", true)
     val oldRDDs = generatedRDDs.filter(_._1 <= (time - rememberDuration))
     logDebug("Clearing references to old RDDs: [" +
       oldRDDs.map(x => s"${x._1} -> ${x._2.id}").mkString(", ") + "]")
     generatedRDDs --= oldRDDs.keys
-    if (ssc.conf.getBoolean("spark.streaming.unpersist", false)) {
+    if (unpersistData) {
       logDebug("Unpersisting old RDDs: " + oldRDDs.values.map(_.id).mkString(", "))
-      oldRDDs.values.foreach(_.unpersist(false))
+      oldRDDs.values.foreach { rdd =>
+        rdd.unpersist(false)
+        // Explicitly remove blocks of BlockRDD
+        rdd match {
+          case b: BlockRDD[_] =>
+            logInfo("Removing blocks of RDD " + b + " of time " + time)
+            b.removeBlocks()
+          case _ =>
+        }
+      }
     }
     logDebug("Cleared " + oldRDDs.size + " RDDs that were older than " +
       (time - rememberDuration) + ": " + oldRDDs.keys.mkString(", "))

streaming/src/test/scala/org/apache/spark/streaming/BasicOperationsSuite.scala

@@ -19,14 +19,16 @@ package org.apache.spark.streaming
 
 import org.apache.spark.streaming.StreamingContext._
 
-import org.apache.spark.rdd.RDD
+import org.apache.spark.rdd.{BlockRDD, RDD}
 import org.apache.spark.SparkContext._
 
 import util.ManualClock
 import org.apache.spark.{SparkContext, SparkConf}
 import org.apache.spark.streaming.dstream.{WindowedDStream, DStream}
 import scala.collection.mutable.{SynchronizedBuffer, ArrayBuffer}
 import scala.reflect.ClassTag
+import org.apache.spark.storage.StorageLevel
+import scala.collection.mutable
 
 class BasicOperationsSuite extends TestSuiteBase {
   test("map") {
@@ -450,6 +452,78 @@ class BasicOperationsSuite extends TestSuiteBase {
     assert(!stateStream.generatedRDDs.contains(Time(4000)))
   }
 
+  test("rdd cleanup - input blocks and persisted RDDs") {
+    // Actually receive data over through receiver to create BlockRDDs
+
+    // Start the server
+    val testServer = new TestServer()
+    testServer.start()
+
+    // Set up the streaming context and input streams
+    val ssc = new StreamingContext(conf, batchDuration)
+    val networkStream = ssc.socketTextStream("localhost", testServer.port, StorageLevel.MEMORY_AND_DISK)
+    val mappedStream = networkStream.map(_ + ".").persist()
+    val outputBuffer = new ArrayBuffer[Seq[String]] with SynchronizedBuffer[Seq[String]]
+    val outputStream = new TestOutputStream(mappedStream, outputBuffer)
+
+    outputStream.register()
+    ssc.start()
+
+    // Feed data to the server to send to the network receiver
+    val clock = ssc.scheduler.clock.asInstanceOf[ManualClock]
+    val input = Seq(1, 2, 3, 4, 5, 6)
+
+    val blockRdds = new mutable.HashMap[Time, BlockRDD[_]]
+    val persistentRddIds = new mutable.HashMap[Time, Int]
+
+    def collectRddInfo() { // get all RDD info required for verification
+      networkStream.generatedRDDs.foreach { case (time, rdd) =>
+        blockRdds(time) = rdd.asInstanceOf[BlockRDD[_]]
+      }
+      mappedStream.generatedRDDs.foreach { case (time, rdd) =>
+        persistentRddIds(time) = rdd.id
+      }
+    }
+
+    Thread.sleep(200)
+    for (i <- 0 until input.size) {
+      testServer.send(input(i).toString + "\n")
+      Thread.sleep(200)
+      clock.addToTime(batchDuration.milliseconds)
+      collectRddInfo()
+    }
+
+    Thread.sleep(200)
+    collectRddInfo()
+    logInfo("Stopping server")
+    testServer.stop()
+    logInfo("Stopping context")
+
+    // verify data has been received
+    assert(outputBuffer.size > 0)
+    assert(blockRdds.size > 0)
+    assert(persistentRddIds.size > 0)
+
+    import Time._
+
+    val latestPersistedRddId = persistentRddIds(persistentRddIds.keySet.max)
+    val earliestPersistedRddId = persistentRddIds(persistentRddIds.keySet.min)
+    val latestBlockRdd = blockRdds(blockRdds.keySet.max)
+    val earliestBlockRdd = blockRdds(blockRdds.keySet.min)
+    // verify that the latest mapped RDD is persisted but the earliest one has been unpersisted
+    assert(ssc.sparkContext.persistentRdds.contains(latestPersistedRddId))
+    assert(!ssc.sparkContext.persistentRdds.contains(earliestPersistedRddId))
+
+    // verify that the latest input blocks are present but the earliest blocks have been removed
+    assert(latestBlockRdd.isValid)
+    assert(latestBlockRdd.collect != null)
+    assert(!earliestBlockRdd.isValid)
+    earliestBlockRdd.blockIds.foreach { blockId =>
+      assert(!ssc.sparkContext.env.blockManager.master.contains(blockId))
+    }
+    ssc.stop()
+  }
+
   /** Test cleanup of RDDs in DStream metadata */
   def runCleanupTest[T: ClassTag](
       conf2: SparkConf,

streaming/src/test/scala/org/apache/spark/streaming/InputStreamsSuite.scala

@@ -42,8 +42,6 @@ import org.apache.spark.streaming.receiver.{ActorHelper, Receiver}
 
 class InputStreamsSuite extends TestSuiteBase with BeforeAndAfter {
 
-  val testPort = 9999
-
   test("socket input stream") {
     // Start the server
     val testServer = new TestServer()
@@ -288,17 +286,6 @@ class TestServer(portToBind: Int = 0) extends Logging {
   def port = serverSocket.getLocalPort
 }
 
-object TestServer {
-  def main(args: Array[String]) {
-    val s = new TestServer()
-    s.start()
-    while(true) {
-      Thread.sleep(1000)
-      s.send("hello")
-    }
-  }
-}
-
 /** This is an actor for testing actor input stream */
 class TestActor(port: Int) extends Actor with ActorHelper {
 

streaming/src/test/scala/org/apache/spark/streaming/NetworkReceiverSuite.scala

@@ -29,6 +29,7 @@ import org.scalatest.FunSuite
 import org.scalatest.concurrent.Timeouts
 import org.scalatest.concurrent.Eventually._
 import org.scalatest.time.SpanSugar._
+import scala.language.postfixOps
 
 /** Testsuite for testing the network receiver behavior */
 class NetworkReceiverSuite extends FunSuite with Timeouts {