Merge branch 'master' of git://git.apache.org/spark into SPARK-2970

Author: sarutak

.travis.yml

@@ -66,10 +66,15 @@ private[spark] class ContextCleaner(sc: SparkContext) extends Logging {
 
   /**
    * Whether the cleaning thread will block on cleanup tasks.
-   * This is set to true only for tests.
+   *
+   * Due to SPARK-3015, this is set to true by default. This is intended to be only a temporary
+   * workaround for the issue, which is ultimately caused by the way the BlockManager actors
+   * issue inter-dependent blocking Akka messages to each other at high frequencies. This happens,
+   * for instance, when the driver performs a GC and cleans up all broadcast blocks that are no
+   * longer in scope.
    */
   private val blockOnCleanupTasks = sc.conf.getBoolean(
-    "spark.cleaner.referenceTracking.blocking", false)
+    "spark.cleaner.referenceTracking.blocking", true)
 
   @volatile private var stopped = false
 
@@ -174,9 +179,6 @@ private[spark] class ContextCleaner(sc: SparkContext) extends Logging {
   private def blockManagerMaster = sc.env.blockManager.master
   private def broadcastManager = sc.env.broadcastManager
   private def mapOutputTrackerMaster = sc.env.mapOutputTracker.asInstanceOf[MapOutputTrackerMaster]
-
-  // Used for testing. These methods explicitly blocks until cleanup is completed
-  // to ensure that more reliable testing.
 }
 
 private object ContextCleaner {

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

@@ -45,7 +45,7 @@ class SparkConf(loadDefaults: Boolean) extends Cloneable with Logging {
   /** Create a SparkConf that loads defaults from system properties and the classpath */
   def this() = this(true)
 
-  private val settings = new HashMap[String, String]()
+  private[spark] val settings = new HashMap[String, String]()
 
   if (loadDefaults) {
     // Load any spark.* system properties
@@ -210,6 +210,12 @@ class SparkConf(loadDefaults: Boolean) extends Cloneable with Logging {
     new SparkConf(false).setAll(settings)
   }
 
+  /**
+   * By using this instead of System.getenv(), environment variables can be mocked
+   * in unit tests.
+   */
+  private[spark] def getenv(name: String): String = System.getenv(name)
+
   /** Checks for illegal or deprecated config settings. Throws an exception for the former. Not
     * idempotent - may mutate this conf object to convert deprecated settings to supported ones. */
   private[spark] def validateSettings() {
@@ -227,7 +233,7 @@ class SparkConf(loadDefaults: Boolean) extends Cloneable with Logging {
     // Validate spark.executor.extraJavaOptions
     settings.get(executorOptsKey).map { javaOpts =>
       if (javaOpts.contains("-Dspark")) {
-        val msg = s"$executorOptsKey is not allowed to set Spark options (was '$javaOpts)'. " +
+        val msg = s"$executorOptsKey is not allowed to set Spark options (was '$javaOpts'). " +
           "Set them directly on a SparkConf or in a properties file when using ./bin/spark-submit."
         throw new Exception(msg)
       }

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

@@ -210,12 +210,22 @@ object SparkEnv extends Logging {
       "MapOutputTracker",
       new MapOutputTrackerMasterActor(mapOutputTracker.asInstanceOf[MapOutputTrackerMaster], conf))
 
+    // Let the user specify short names for shuffle managers
+    val shortShuffleMgrNames = Map(
+      "hash" -> "org.apache.spark.shuffle.hash.HashShuffleManager",
+      "sort" -> "org.apache.spark.shuffle.sort.SortShuffleManager")
+    val shuffleMgrName = conf.get("spark.shuffle.manager", "hash")
+    val shuffleMgrClass = shortShuffleMgrNames.getOrElse(shuffleMgrName.toLowerCase, shuffleMgrName)
+    val shuffleManager = instantiateClass[ShuffleManager](shuffleMgrClass)
+
+    val shuffleMemoryManager = new ShuffleMemoryManager(conf)
+
     val blockManagerMaster = new BlockManagerMaster(registerOrLookup(
       "BlockManagerMaster",
       new BlockManagerMasterActor(isLocal, conf, listenerBus)), conf)
 
     val blockManager = new BlockManager(executorId, actorSystem, blockManagerMaster,
-      serializer, conf, securityManager, mapOutputTracker)
+      serializer, conf, securityManager, mapOutputTracker, shuffleManager)
 
     val connectionManager = blockManager.connectionManager
 
@@ -250,16 +260,6 @@ object SparkEnv extends Logging {
       "."
     }
 
-    // Let the user specify short names for shuffle managers
-    val shortShuffleMgrNames = Map(
-      "hash" -> "org.apache.spark.shuffle.hash.HashShuffleManager",
-      "sort" -> "org.apache.spark.shuffle.sort.SortShuffleManager")
-    val shuffleMgrName = conf.get("spark.shuffle.manager", "hash")
-    val shuffleMgrClass = shortShuffleMgrNames.getOrElse(shuffleMgrName.toLowerCase, shuffleMgrName)
-    val shuffleManager = instantiateClass[ShuffleManager](shuffleMgrClass)
-
-    val shuffleMemoryManager = new ShuffleMemoryManager(conf)
-
     // Warn about deprecated spark.cache.class property
     if (conf.contains("spark.cache.class")) {
       logWarning("The spark.cache.class property is no longer being used! Specify storage " +

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

@@ -62,7 +62,7 @@ private[spark] class PythonRDD(
     val env = SparkEnv.get
     val localdir = env.blockManager.diskBlockManager.localDirs.map(
       f => f.getPath()).mkString(",")
-    envVars += ("SPARK_LOCAL_DIR" -> localdir) // it's also used in monitor thread
+    envVars += ("SPARK_LOCAL_DIRS" -> localdir) // it's also used in monitor thread
     val worker: Socket = env.createPythonWorker(pythonExec, envVars.toMap)
 
     // Start a thread to feed the process input from our parent's iterator
@@ -315,6 +315,14 @@ private[spark] object PythonRDD extends Logging {
     JavaRDD.fromRDD(sc.sc.parallelize(objs, parallelism))
   }
 
+  def readBroadcastFromFile(sc: JavaSparkContext, filename: String): Broadcast[Array[Byte]] = {
+    val file = new DataInputStream(new FileInputStream(filename))
+    val length = file.readInt()
+    val obj = new Array[Byte](length)
+    file.readFully(obj)
+    sc.broadcast(obj)
+  }
+
   def writeIteratorToStream[T](iter: Iterator[T], dataOut: DataOutputStream) {
     // The right way to implement this would be to use TypeTags to get the full
     // type of T.  Since I don't want to introduce breaking changes throughout the

core/src/main/scala/org/apache/spark/api/python/PythonRDD.scala

@@ -32,8 +32,19 @@ import org.apache.spark.annotation.DeveloperApi
  */
 @DeveloperApi
 trait BroadcastFactory {
+
   def initialize(isDriver: Boolean, conf: SparkConf, securityMgr: SecurityManager): Unit
+
+  /**
+   * Creates a new broadcast variable.
+   *
+   * @param value value to broadcast
+   * @param isLocal whether we are in local mode (single JVM process)
+   * @param id unique id representing this broadcast variable
+   */
   def newBroadcast[T: ClassTag](value: T, isLocal: Boolean, id: Long): Broadcast[T]
+
   def unbroadcast(id: Long, removeFromDriver: Boolean, blocking: Boolean): Unit
+
   def stop(): Unit
 }