[SPARK-17365][CORE] Remove/Kill multiple executors together to reduce…

… RPC call time. ## What changes were proposed in this pull request? We are killing multiple executors together instead of iterating over expensive RPC calls to kill single executor. ## How was this patch tested? Executed sample spark job to observe executors being killed/removed with dynamic allocation enabled. Author: Dhruve Ashar <[email protected]> Author: Dhruve Ashar <[email protected]> Closes apache#15152 from dhruve/impr/SPARK-17365.
ammills01 · Sep 22, 2016 · 17b72d3 · 17b72d3
1 parent 8a02410
commit 17b72d3
Show file tree

Hide file tree

Showing 11 changed files with 239 additions and 55 deletions.
diff --git a/core/src/main/scala/org/apache/spark/ExecutorAllocationClient.scala b/core/src/main/scala/org/apache/spark/ExecutorAllocationClient.scala
@@ -54,13 +54,16 @@ private[spark] trait ExecutorAllocationClient {
 
   /**
    * Request that the cluster manager kill the specified executors.
-   * @return whether the request is acknowledged by the cluster manager.
+   * @return the ids of the executors acknowledged by the cluster manager to be removed.
    */
-  def killExecutors(executorIds: Seq[String]): Boolean
+  def killExecutors(executorIds: Seq[String]): Seq[String]
 
   /**
    * Request that the cluster manager kill the specified executor.
    * @return whether the request is acknowledged by the cluster manager.
    */
-  def killExecutor(executorId: String): Boolean = killExecutors(Seq(executorId))
+  def killExecutor(executorId: String): Boolean = {
+    val killedExecutors = killExecutors(Seq(executorId))
+    killedExecutors.nonEmpty && killedExecutors(0).equals(executorId)
+  }
 }
diff --git a/core/src/main/scala/org/apache/spark/ExecutorAllocationManager.scala b/core/src/main/scala/org/apache/spark/ExecutorAllocationManager.scala
@@ -20,6 +20,7 @@ package org.apache.spark
 import java.util.concurrent.TimeUnit
 
 import scala.collection.mutable
+import scala.collection.mutable.ArrayBuffer
 import scala.util.control.ControlThrowable
 
 import com.codahale.metrics.{Gauge, MetricRegistry}
@@ -279,14 +280,18 @@ private[spark] class ExecutorAllocationManager(
 
     updateAndSyncNumExecutorsTarget(now)
 
+    val executorIdsToBeRemoved = ArrayBuffer[String]()
     removeTimes.retain { case (executorId, expireTime) =>
       val expired = now >= expireTime
       if (expired) {
         initializing = false
-        removeExecutor(executorId)
+        executorIdsToBeRemoved += executorId
       }
       !expired
     }
+    if (executorIdsToBeRemoved.nonEmpty) {
+      removeExecutors(executorIdsToBeRemoved)
+    }
   }
 
   /**
@@ -391,11 +396,67 @@ private[spark] class ExecutorAllocationManager(
     }
   }
 
+  /**
+   * Request the cluster manager to remove the given executors.
+   * Returns the list of executors which are removed.
+   */
+  private def removeExecutors(executors: Seq[String]): Seq[String] = synchronized {
+    val executorIdsToBeRemoved = new ArrayBuffer[String]
+
+    logInfo("Request to remove executorIds: " + executors.mkString(", "))
+    val numExistingExecutors = allocationManager.executorIds.size - executorsPendingToRemove.size
+
+    var newExecutorTotal = numExistingExecutors
+    executors.foreach { executorIdToBeRemoved =>
+      if (newExecutorTotal - 1 < minNumExecutors) {
+        logDebug(s"Not removing idle executor $executorIdToBeRemoved because there are only " +
+          s"$newExecutorTotal executor(s) left (limit $minNumExecutors)")
+      } else if (canBeKilled(executorIdToBeRemoved)) {
+        executorIdsToBeRemoved += executorIdToBeRemoved
+        newExecutorTotal -= 1
+      }
+    }
+
+    if (executorIdsToBeRemoved.isEmpty) {
+      return Seq.empty[String]
+    }
+
+    // Send a request to the backend to kill this executor(s)
+    val executorsRemoved = if (testing) {
+      executorIdsToBeRemoved
+    } else {
+      client.killExecutors(executorIdsToBeRemoved)
+    }
+    // reset the newExecutorTotal to the existing number of executors
+    newExecutorTotal = numExistingExecutors
+    if (testing || executorsRemoved.nonEmpty) {
+      executorsRemoved.foreach { removedExecutorId =>
+        newExecutorTotal -= 1
+        logInfo(s"Removing executor $removedExecutorId because it has been idle for " +
+          s"$executorIdleTimeoutS seconds (new desired total will be $newExecutorTotal)")
+        executorsPendingToRemove.add(removedExecutorId)
+      }
+      executorsRemoved
+    } else {
+      logWarning(s"Unable to reach the cluster manager to kill executor/s " +
+        "executorIdsToBeRemoved.mkString(\",\") or no executor eligible to kill!")
+      Seq.empty[String]
+    }
+  }
+
   /**
    * Request the cluster manager to remove the given executor.
-   * Return whether the request is received.
+   * Return whether the request is acknowledged.
    */
   private def removeExecutor(executorId: String): Boolean = synchronized {
+    val executorsRemoved = removeExecutors(Seq(executorId))
+    executorsRemoved.nonEmpty && executorsRemoved(0) == executorId
+  }
+
+  /**
+   * Determine if the given executor can be killed.
+   */
+  private def canBeKilled(executorId: String): Boolean = synchronized {
     // Do not kill the executor if we are not aware of it (should never happen)
     if (!executorIds.contains(executorId)) {
       logWarning(s"Attempted to remove unknown executor $executorId!")
@@ -409,26 +470,7 @@ private[spark] class ExecutorAllocationManager(
       return false
     }
 
-    // Do not kill the executor if we have already reached the lower bound
-    val numExistingExecutors = executorIds.size - executorsPendingToRemove.size
-    if (numExistingExecutors - 1 < minNumExecutors) {
-      logDebug(s"Not removing idle executor $executorId because there are only " +
-        s"$numExistingExecutors executor(s) left (limit $minNumExecutors)")
-      return false
-    }
-
-    // Send a request to the backend to kill this executor
-    val removeRequestAcknowledged = testing || client.killExecutor(executorId)
-    if (removeRequestAcknowledged) {
-      logInfo(s"Removing executor $executorId because it has been idle for " +
-        s"$executorIdleTimeoutS seconds (new desired total will be ${numExistingExecutors - 1})")
-      executorsPendingToRemove.add(executorId)
-      true
-    } else {
-      logWarning(s"Unable to reach the cluster manager to kill executor $executorId," +
-        s"or no executor eligible to kill!")
-      false
-    }
+    true
   }
 
   /**

diff --git a/core/src/main/scala/org/apache/spark/SparkContext.scala b/core/src/main/scala/org/apache/spark/SparkContext.scala
@@ -73,7 +73,7 @@ import org.apache.spark.util._
  * @param config a Spark Config object describing the application configuration. Any settings in
  *   this config overrides the default configs as well as system properties.
  */
-class SparkContext(config: SparkConf) extends Logging with ExecutorAllocationClient {
+class SparkContext(config: SparkConf) extends Logging {
 
   // The call site where this SparkContext was constructed.
   private val creationSite: CallSite = Utils.getCallSite()
@@ -534,7 +534,13 @@ class SparkContext(config: SparkConf) extends Logging with ExecutorAllocationCli
     val dynamicAllocationEnabled = Utils.isDynamicAllocationEnabled(_conf)
     _executorAllocationManager =
       if (dynamicAllocationEnabled) {
-        Some(new ExecutorAllocationManager(this, listenerBus, _conf))
+        schedulerBackend match {
+          case b: ExecutorAllocationClient =>
+            Some(new ExecutorAllocationManager(
+              schedulerBackend.asInstanceOf[ExecutorAllocationClient], listenerBus, _conf))
+          case _ =>
+            None
+        }
       } else {
         None
       }
@@ -1473,7 +1479,7 @@ class SparkContext(config: SparkConf) extends Logging with ExecutorAllocationCli
     listenerBus.addListener(listener)
   }
 
-  private[spark] override def getExecutorIds(): Seq[String] = {
+  private[spark] def getExecutorIds(): Seq[String] = {
     schedulerBackend match {
       case b: CoarseGrainedSchedulerBackend =>
         b.getExecutorIds()
@@ -1498,7 +1504,7 @@ class SparkContext(config: SparkConf) extends Logging with ExecutorAllocationCli
    * @return whether the request is acknowledged by the cluster manager.
    */
   @DeveloperApi
-  override def requestTotalExecutors(
+  def requestTotalExecutors(
       numExecutors: Int,
       localityAwareTasks: Int,
       hostToLocalTaskCount: scala.collection.immutable.Map[String, Int]
@@ -1518,7 +1524,7 @@ class SparkContext(config: SparkConf) extends Logging with ExecutorAllocationCli
    * @return whether the request is received.
    */
   @DeveloperApi
-  override def requestExecutors(numAdditionalExecutors: Int): Boolean = {
+  def requestExecutors(numAdditionalExecutors: Int): Boolean = {
     schedulerBackend match {
       case b: CoarseGrainedSchedulerBackend =>
         b.requestExecutors(numAdditionalExecutors)
@@ -1540,10 +1546,10 @@ class SparkContext(config: SparkConf) extends Logging with ExecutorAllocationCli
    * @return whether the request is received.
    */
   @DeveloperApi
-  override def killExecutors(executorIds: Seq[String]): Boolean = {
+  def killExecutors(executorIds: Seq[String]): Boolean = {
     schedulerBackend match {
       case b: CoarseGrainedSchedulerBackend =>
-        b.killExecutors(executorIds, replace = false, force = true)
+        b.killExecutors(executorIds, replace = false, force = true).nonEmpty
       case _ =>
         logWarning("Killing executors is only supported in coarse-grained mode")
         false
@@ -1562,7 +1568,7 @@ class SparkContext(config: SparkConf) extends Logging with ExecutorAllocationCli
    * @return whether the request is received.
    */
   @DeveloperApi
-  override def killExecutor(executorId: String): Boolean = super.killExecutor(executorId)
+  def killExecutor(executorId: String): Boolean = killExecutors(Seq(executorId))
 
   /**
    * Request that the cluster manager kill the specified executor without adjusting the
@@ -1581,7 +1587,7 @@ class SparkContext(config: SparkConf) extends Logging with ExecutorAllocationCli
   private[spark] def killAndReplaceExecutor(executorId: String): Boolean = {
     schedulerBackend match {
       case b: CoarseGrainedSchedulerBackend =>
-        b.killExecutors(Seq(executorId), replace = true, force = true)
+        b.killExecutors(Seq(executorId), replace = true, force = true).nonEmpty
       case _ =>
         logWarning("Killing executors is only supported in coarse-grained mode")
         false

diff --git a/core/src/main/scala/org/apache/spark/scheduler/cluster/CoarseGrainedSchedulerBackend.scala b/core/src/main/scala/org/apache/spark/scheduler/cluster/CoarseGrainedSchedulerBackend.scala
@@ -528,7 +528,7 @@ class CoarseGrainedSchedulerBackend(scheduler: TaskSchedulerImpl, val rpcEnv: Rp
    * @return whether the kill request is acknowledged. If list to kill is empty, it will return
    *         false.
    */
-  final override def killExecutors(executorIds: Seq[String]): Boolean = {
+  final override def killExecutors(executorIds: Seq[String]): Seq[String] = {
     killExecutors(executorIds, replace = false, force = false)
   }
 
@@ -548,7 +548,7 @@ class CoarseGrainedSchedulerBackend(scheduler: TaskSchedulerImpl, val rpcEnv: Rp
   final def killExecutors(
       executorIds: Seq[String],
       replace: Boolean,
-      force: Boolean): Boolean = {
+      force: Boolean): Seq[String] = {
     logInfo(s"Requesting to kill executor(s) ${executorIds.mkString(", ")}")
 
     val response = synchronized {
@@ -564,6 +564,8 @@ class CoarseGrainedSchedulerBackend(scheduler: TaskSchedulerImpl, val rpcEnv: Rp
         .filter { id => force || !scheduler.isExecutorBusy(id) }
       executorsToKill.foreach { id => executorsPendingToRemove(id) = !replace }
 
+      logInfo(s"Actual list of executor(s) to be killed is ${executorsToKill.mkString(", ")}")
+
       // If we do not wish to replace the executors we kill, sync the target number of executors
       // with the cluster manager to avoid allocating new ones. When computing the new target,
       // take into account executors that are pending to be added or removed.
@@ -583,7 +585,11 @@ class CoarseGrainedSchedulerBackend(scheduler: TaskSchedulerImpl, val rpcEnv: Rp
           _ => Future.successful(false)
         }
 
-      adjustTotalExecutors.flatMap(killExecutors)(ThreadUtils.sameThread)
+      val killResponse = adjustTotalExecutors.flatMap(killExecutors)(ThreadUtils.sameThread)
+
+      killResponse.flatMap(killSuccessful =>
+        Future.successful (if (killSuccessful) executorsToKill else Seq.empty[String])
+      )(ThreadUtils.sameThread)
     }
 
     defaultAskTimeout.awaitResult(response)

diff --git a/core/src/test/resources/META-INF/services/org.apache.spark.scheduler.ExternalClusterManager b/core/src/test/resources/META-INF/services/org.apache.spark.scheduler.ExternalClusterManager
@@ -1,2 +1,3 @@
 org.apache.spark.scheduler.DummyExternalClusterManager
-org.apache.spark.scheduler.MockExternalClusterManager
+org.apache.spark.scheduler.MockExternalClusterManager
+org.apache.spark.DummyLocalExternalClusterManager