[SPARK-6602][Core] Replace direct use of Akka with Spark RPC interface - part 1

This PR replaced the following `Actor`s to `RpcEndpoint`:

1. HeartbeatReceiver
1. ExecutorActor
1. BlockManagerMasterActor
1. BlockManagerSlaveActor
1. CoarseGrainedExecutorBackend and subclasses
1. CoarseGrainedSchedulerBackend.DriverActor

This is the first PR. I will split the work of SPARK-6602 to several PRs for code review.

Author: zsxwing <[email protected]>

Closes apache#5268 from zsxwing/rpc-rewrite and squashes the following commits:

287e9f8 [zsxwing] Fix the code style
26c56b7 [zsxwing] Merge branch 'master' into rpc-rewrite
9cc825a [zsxwing] Rmove setupThreadSafeEndpoint and add ThreadSafeRpcEndpoint
30a9036 [zsxwing] Make self return null after stopping RpcEndpointRef; fix docs and error messages
705245d [zsxwing] Fix some bugs after rebasing the changes on the master
003cf80 [zsxwing] Update CoarseGrainedExecutorBackend and CoarseGrainedSchedulerBackend to use RpcEndpoint
7d0e6dc [zsxwing] Update BlockManagerSlaveActor to use RpcEndpoint
f5d6543 [zsxwing] Update BlockManagerMaster to use RpcEndpoint
30e3f9f [zsxwing] Update ExecutorActor to use RpcEndpoint
478b443 [zsxwing] Update HeartbeatReceiver to use RpcEndpoint

Author: zsxwing

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

@@ -17,15 +17,15 @@
 
 package org.apache.spark
 
-import scala.concurrent.duration._
-import scala.collection.mutable
+import java.util.concurrent.{ScheduledFuture, TimeUnit, Executors}
 
-import akka.actor.{Actor, Cancellable}
+import scala.collection.mutable
 
 import org.apache.spark.executor.TaskMetrics
+import org.apache.spark.rpc.{ThreadSafeRpcEndpoint, RpcEnv, RpcCallContext}
 import org.apache.spark.storage.BlockManagerId
 import org.apache.spark.scheduler.{SlaveLost, TaskScheduler}
-import org.apache.spark.util.ActorLogReceive
+import org.apache.spark.util.Utils
 
 /**
  * A heartbeat from executors to the driver. This is a shared message used by several internal
@@ -51,9 +51,11 @@ private[spark] case class HeartbeatResponse(reregisterBlockManager: Boolean)
  * Lives in the driver to receive heartbeats from executors..
  */
 private[spark] class HeartbeatReceiver(sc: SparkContext)
-  extends Actor with ActorLogReceive with Logging {
+  extends ThreadSafeRpcEndpoint with Logging {
+
+  override val rpcEnv: RpcEnv = sc.env.rpcEnv
 
-  private var scheduler: TaskScheduler = null
+  private[spark] var scheduler: TaskScheduler = null
 
   // executor ID -> timestamp of when the last heartbeat from this executor was received
   private val executorLastSeen = new mutable.HashMap[String, Long]
@@ -69,34 +71,44 @@ private[spark] class HeartbeatReceiver(sc: SparkContext)
     sc.conf.getOption("spark.network.timeoutInterval").map(_.toLong * 1000).
       getOrElse(sc.conf.getLong("spark.storage.blockManagerTimeoutIntervalMs", 60000))
 
-  private var timeoutCheckingTask: Cancellable = null
-  
-  override def preStart(): Unit = {
-    import context.dispatcher
-    timeoutCheckingTask = context.system.scheduler.schedule(0.seconds,
-      checkTimeoutIntervalMs.milliseconds, self, ExpireDeadHosts)
-    super.preStart()
+  private var timeoutCheckingTask: ScheduledFuture[_] = null
+
+  private val timeoutCheckingThread = Executors.newSingleThreadScheduledExecutor(
+    Utils.namedThreadFactory("heartbeat-timeout-checking-thread"))
+
+  private val killExecutorThread = Executors.newSingleThreadExecutor(
+    Utils.namedThreadFactory("kill-executor-thread"))
+
+  override def onStart(): Unit = {
+    timeoutCheckingTask = timeoutCheckingThread.scheduleAtFixedRate(new Runnable {
+      override def run(): Unit = Utils.tryLogNonFatalError {
+        Option(self).foreach(_.send(ExpireDeadHosts))
+      }
+    }, 0, checkTimeoutIntervalMs, TimeUnit.MILLISECONDS)
   }
-  
-  override def receiveWithLogging: PartialFunction[Any, Unit] = {
+
+  override def receive: PartialFunction[Any, Unit] = {
+    case ExpireDeadHosts =>
+      expireDeadHosts()
     case TaskSchedulerIsSet =>
       scheduler = sc.taskScheduler
+  }
+
+  override def receiveAndReply(context: RpcCallContext): PartialFunction[Any, Unit] = {
     case heartbeat @ Heartbeat(executorId, taskMetrics, blockManagerId) =>
       if (scheduler != null) {
         val unknownExecutor = !scheduler.executorHeartbeatReceived(
           executorId, taskMetrics, blockManagerId)
         val response = HeartbeatResponse(reregisterBlockManager = unknownExecutor)
         executorLastSeen(executorId) = System.currentTimeMillis()
-        sender ! response
+        context.reply(response)
       } else {
         // Because Executor will sleep several seconds before sending the first "Heartbeat", this
         // case rarely happens. However, if it really happens, log it and ask the executor to
         // register itself again.
         logWarning(s"Dropping $heartbeat because TaskScheduler is not ready yet")
-        sender ! HeartbeatResponse(reregisterBlockManager = true)
+        context.reply(HeartbeatResponse(reregisterBlockManager = true))
       }
-    case ExpireDeadHosts =>
-      expireDeadHosts()
   }
 
   private def expireDeadHosts(): Unit = {
@@ -109,17 +121,25 @@ private[spark] class HeartbeatReceiver(sc: SparkContext)
         scheduler.executorLost(executorId, SlaveLost("Executor heartbeat " +
           s"timed out after ${now - lastSeenMs} ms"))
         if (sc.supportDynamicAllocation) {
-          sc.killExecutor(executorId)
+          // Asynchronously kill the executor to avoid blocking the current thread
+          killExecutorThread.submit(new Runnable {
+            override def run(): Unit = sc.killExecutor(executorId)
+          })
         }
         executorLastSeen.remove(executorId)
       }
     }
   }
 
-  override def postStop(): Unit = {
+  override def onStop(): Unit = {
     if (timeoutCheckingTask != null) {
-      timeoutCheckingTask.cancel()
+      timeoutCheckingTask.cancel(true)
     }
-    super.postStop()
+    timeoutCheckingThread.shutdownNow()
+    killExecutorThread.shutdownNow()
   }
 }
+
+object HeartbeatReceiver {
+  val ENDPOINT_NAME = "HeartbeatReceiver"
+}

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

@@ -32,8 +32,6 @@ import scala.collection.generic.Growable
 import scala.collection.mutable.HashMap
 import scala.reflect.{ClassTag, classTag}
 
-import akka.actor.Props
-
 import org.apache.hadoop.conf.Configuration
 import org.apache.hadoop.fs.Path
 import org.apache.hadoop.io.{ArrayWritable, BooleanWritable, BytesWritable, DoubleWritable,
@@ -48,12 +46,13 @@ import org.apache.mesos.MesosNativeLibrary
 import org.apache.spark.annotation.{DeveloperApi, Experimental}
 import org.apache.spark.broadcast.Broadcast
 import org.apache.spark.deploy.{LocalSparkCluster, SparkHadoopUtil}
-import org.apache.spark.executor.TriggerThreadDump
+import org.apache.spark.executor.{ExecutorEndpoint, TriggerThreadDump}
 import org.apache.spark.input.{StreamInputFormat, PortableDataStream, WholeTextFileInputFormat,
   FixedLengthBinaryInputFormat}
 import org.apache.spark.io.CompressionCodec
 import org.apache.spark.partial.{ApproximateEvaluator, PartialResult}
 import org.apache.spark.rdd._
+import org.apache.spark.rpc.RpcAddress
 import org.apache.spark.scheduler._
 import org.apache.spark.scheduler.cluster.{CoarseGrainedSchedulerBackend,
   SparkDeploySchedulerBackend, SimrSchedulerBackend}
@@ -360,14 +359,14 @@ class SparkContext(config: SparkConf) extends Logging with ExecutorAllocationCli
 
   // We need to register "HeartbeatReceiver" before "createTaskScheduler" because Executor will
   // retrieve "HeartbeatReceiver" in the constructor. (SPARK-6640)
-  private val heartbeatReceiver = env.actorSystem.actorOf(
-    Props(new HeartbeatReceiver(this)), "HeartbeatReceiver")
+  private val heartbeatReceiver = env.rpcEnv.setupEndpoint(
+    HeartbeatReceiver.ENDPOINT_NAME, new HeartbeatReceiver(this))
 
   // Create and start the scheduler
   private[spark] var (schedulerBackend, taskScheduler) =
     SparkContext.createTaskScheduler(this, master)
 
-  heartbeatReceiver ! TaskSchedulerIsSet
+  heartbeatReceiver.send(TaskSchedulerIsSet)
 
   @volatile private[spark] var dagScheduler: DAGScheduler = _
   try {
@@ -455,10 +454,12 @@ class SparkContext(config: SparkConf) extends Logging with ExecutorAllocationCli
       if (executorId == SparkContext.DRIVER_IDENTIFIER) {
         Some(Utils.getThreadDump())
       } else {
-        val (host, port) = env.blockManager.master.getActorSystemHostPortForExecutor(executorId).get
-        val actorRef = AkkaUtils.makeExecutorRef("ExecutorActor", conf, host, port, env.actorSystem)
-        Some(AkkaUtils.askWithReply[Array[ThreadStackTrace]](TriggerThreadDump, actorRef,
-          AkkaUtils.numRetries(conf), AkkaUtils.retryWaitMs(conf), AkkaUtils.askTimeout(conf)))
+        val (host, port) = env.blockManager.master.getRpcHostPortForExecutor(executorId).get
+        val endpointRef = env.rpcEnv.setupEndpointRef(
+          SparkEnv.executorActorSystemName,
+          RpcAddress(host, port),
+          ExecutorEndpoint.EXECUTOR_ENDPOINT_NAME)
+        Some(endpointRef.askWithReply[Array[ThreadStackTrace]](TriggerThreadDump))
       }
     } catch {
       case e: Exception =>
@@ -1418,7 +1419,7 @@ class SparkContext(config: SparkConf) extends Logging with ExecutorAllocationCli
     dagScheduler = null
     listenerBus.stop()
     eventLogger.foreach(_.stop())
-    env.actorSystem.stop(heartbeatReceiver)
+    env.rpcEnv.stop(heartbeatReceiver)
     progressBar.foreach(_.stop())
     taskScheduler = null
     // TODO: Cache.stop()?

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

@@ -295,7 +295,9 @@ object SparkEnv extends Logging {
       }
     }
 
-    def registerOrLookupEndpoint(name: String, endpointCreator: => RpcEndpoint): RpcEndpointRef = {
+    def registerOrLookupEndpoint(
+        name: String, endpointCreator: => RpcEndpoint):
+      RpcEndpointRef = {
       if (isDriver) {
         logInfo("Registering " + name)
         rpcEnv.setupEndpoint(name, endpointCreator)
@@ -334,12 +336,13 @@ object SparkEnv extends Logging {
           new NioBlockTransferService(conf, securityManager)
       }
 
-    val blockManagerMaster = new BlockManagerMaster(registerOrLookup(
-      "BlockManagerMaster",
-      new BlockManagerMasterActor(isLocal, conf, listenerBus)), conf, isDriver)
+    val blockManagerMaster = new BlockManagerMaster(registerOrLookupEndpoint(
+      BlockManagerMaster.DRIVER_ENDPOINT_NAME,
+      new BlockManagerMasterEndpoint(rpcEnv, isLocal, conf, listenerBus)),
+      conf, isDriver)
 
     // NB: blockManager is not valid until initialize() is called later.
-    val blockManager = new BlockManager(executorId, actorSystem, blockManagerMaster,
+    val blockManager = new BlockManager(executorId, rpcEnv, blockManagerMaster,
       serializer, conf, mapOutputTracker, shuffleManager, blockTransferService, securityManager,
       numUsableCores)
 

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

@@ -21,39 +21,45 @@ import java.net.URL
 import java.nio.ByteBuffer
 
 import scala.collection.mutable
-import scala.concurrent.Await
+import scala.util.{Failure, Success}
 
-import akka.actor.{Actor, ActorSelection, Props}
-import akka.pattern.Patterns
-import akka.remote.{RemotingLifecycleEvent, DisassociatedEvent}
-
-import org.apache.spark.{Logging, SecurityManager, SparkConf, SparkEnv}
+import org.apache.spark.rpc._
+import org.apache.spark._
 import org.apache.spark.TaskState.TaskState
 import org.apache.spark.deploy.SparkHadoopUtil
 import org.apache.spark.deploy.worker.WorkerWatcher
 import org.apache.spark.scheduler.TaskDescription
 import org.apache.spark.scheduler.cluster.CoarseGrainedClusterMessages._
-import org.apache.spark.util.{ActorLogReceive, AkkaUtils, SignalLogger, Utils}
+import org.apache.spark.util.{SignalLogger, Utils}
 
 private[spark] class CoarseGrainedExecutorBackend(
+    override val rpcEnv: RpcEnv,
     driverUrl: String,
     executorId: String,
     hostPort: String,
     cores: Int,
     userClassPath: Seq[URL],
     env: SparkEnv)
-  extends Actor with ActorLogReceive with ExecutorBackend with Logging {
+  extends ThreadSafeRpcEndpoint with ExecutorBackend with Logging {
 
   Utils.checkHostPort(hostPort, "Expected hostport")
 
   var executor: Executor = null
-  var driver: ActorSelection = null
+  @volatile var driver: Option[RpcEndpointRef] = None
 
-  override def preStart() {
+  override def onStart() {
+    import scala.concurrent.ExecutionContext.Implicits.global
     logInfo("Connecting to driver: " + driverUrl)
-    driver = context.actorSelection(driverUrl)
-    driver ! RegisterExecutor(executorId, hostPort, cores, extractLogUrls)
-    context.system.eventStream.subscribe(self, classOf[RemotingLifecycleEvent])
+    rpcEnv.asyncSetupEndpointRefByURI(driverUrl).flatMap { ref =>
+      driver = Some(ref)
+      ref.sendWithReply[RegisteredExecutor.type](
+        RegisterExecutor(executorId, self, hostPort, cores, extractLogUrls))
+    } onComplete {
+      case Success(msg) => Utils.tryLogNonFatalError {
+        Option(self).foreach(_.send(msg)) // msg must be RegisteredExecutor
+      }
+      case Failure(e) => logError(s"Cannot register with driver: $driverUrl", e)
+    }
   }
 
   def extractLogUrls: Map[String, String] = {
@@ -62,7 +68,7 @@ private[spark] class CoarseGrainedExecutorBackend(
       .map(e => (e._1.substring(prefix.length).toLowerCase, e._2))
   }
 
-  override def receiveWithLogging: PartialFunction[Any, Unit] = {
+  override def receive: PartialFunction[Any, Unit] = {
     case RegisteredExecutor =>
       logInfo("Successfully registered with driver")
       val (hostname, _) = Utils.parseHostPort(hostPort)
@@ -92,23 +98,28 @@ private[spark] class CoarseGrainedExecutorBackend(
         executor.killTask(taskId, interruptThread)
       }
 
-    case x: DisassociatedEvent =>
-      if (x.remoteAddress == driver.anchorPath.address) {
-        logError(s"Driver $x disassociated! Shutting down.")
-        System.exit(1)
-      } else {
-        logWarning(s"Received irrelevant DisassociatedEvent $x")
-      }
-
     case StopExecutor =>
       logInfo("Driver commanded a shutdown")
       executor.stop()
-      context.stop(self)
-      context.system.shutdown()
+      stop()
+      rpcEnv.shutdown()
+  }
+
+  override def onDisconnected(remoteAddress: RpcAddress): Unit = {
+    if (driver.exists(_.address == remoteAddress)) {
+      logError(s"Driver $remoteAddress disassociated! Shutting down.")
+      System.exit(1)
+    } else {
+      logWarning(s"An unknown ($remoteAddress) driver disconnected.")
+    }
   }
 
   override def statusUpdate(taskId: Long, state: TaskState, data: ByteBuffer) {
-    driver ! StatusUpdate(executorId, taskId, state, data)
+    val msg = StatusUpdate(executorId, taskId, state, data)
+    driver match {
+      case Some(driverRef) => driverRef.send(msg)
+      case None => logWarning(s"Drop $msg because has not yet connected to driver")
+    }
   }
 }
 
@@ -132,16 +143,14 @@ private[spark] object CoarseGrainedExecutorBackend extends Logging {
       // Bootstrap to fetch the driver's Spark properties.
       val executorConf = new SparkConf
       val port = executorConf.getInt("spark.executor.port", 0)
-      val (fetcher, _) = AkkaUtils.createActorSystem(
+      val fetcher = RpcEnv.create(
         "driverPropsFetcher",
         hostname,
         port,
         executorConf,
         new SecurityManager(executorConf))
-      val driver = fetcher.actorSelection(driverUrl)
-      val timeout = AkkaUtils.askTimeout(executorConf)
-      val fut = Patterns.ask(driver, RetrieveSparkProps, timeout)
-      val props = Await.result(fut, timeout).asInstanceOf[Seq[(String, String)]] ++
+      val driver = fetcher.setupEndpointRefByURI(driverUrl)
+      val props = driver.askWithReply[Seq[(String, String)]](RetrieveSparkProps) ++
         Seq[(String, String)](("spark.app.id", appId))
       fetcher.shutdown()
 
@@ -162,16 +171,14 @@ private[spark] object CoarseGrainedExecutorBackend extends Logging {
       val boundPort = env.conf.getInt("spark.executor.port", 0)
       assert(boundPort != 0)
 
-      // Start the CoarseGrainedExecutorBackend actor.
+      // Start the CoarseGrainedExecutorBackend endpoint.
       val sparkHostPort = hostname + ":" + boundPort
-      env.actorSystem.actorOf(
-        Props(classOf[CoarseGrainedExecutorBackend],
-          driverUrl, executorId, sparkHostPort, cores, userClassPath, env),
-        name = "Executor")
+      env.rpcEnv.setupEndpoint("Executor", new CoarseGrainedExecutorBackend(
+        env.rpcEnv, driverUrl, executorId, sparkHostPort, cores, userClassPath, env))
       workerUrl.foreach { url =>
         env.rpcEnv.setupEndpoint("WorkerWatcher", new WorkerWatcher(env.rpcEnv, url))
       }
-      env.actorSystem.awaitTermination()
+      env.rpcEnv.awaitTermination()
     }
   }