Merge branch 'master' of https://github.com/apache/spark

Author: nishkamravi2

.gitignore

@@ -7,7 +7,7 @@
 sbt/*.jar
 .settings
 .cache
-.generated-mima-excludes
+.generated-mima*
 /build/
 work/
 out/

bagel/src/test/scala/org/apache/spark/bagel/BagelSuite.scala

@@ -80,7 +80,7 @@ class BagelSuite extends FunSuite with Assertions with BeforeAndAfter with Timeo
   test("large number of iterations") {
     // This tests whether jobs with a large number of iterations finish in a reasonable time,
     // because non-memoized recursion in RDD or DAGScheduler used to cause them to hang
-    failAfter(10 seconds) {
+    failAfter(30 seconds) {
       sc = new SparkContext("local", "test")
       val verts = sc.parallelize((1 to 4).map(id => (id.toString, new TestVertex(true, 0))))
       val msgs = sc.parallelize(Array[(String, TestMessage)]())
@@ -101,7 +101,7 @@ class BagelSuite extends FunSuite with Assertions with BeforeAndAfter with Timeo
       sc = new SparkContext("local", "test")
       val verts = sc.parallelize((1 to 4).map(id => (id.toString, new TestVertex(true, 0))))
       val msgs = sc.parallelize(Array[(String, TestMessage)]())
-      val numSupersteps = 50
+      val numSupersteps = 20
       val result =
         Bagel.run(sc, verts, msgs, sc.defaultParallelism, StorageLevel.DISK_ONLY) {
           (self: TestVertex, msgs: Option[Array[TestMessage]], superstep: Int) =>

bin/compute-classpath.sh

@@ -38,8 +38,10 @@ else
   JAR_CMD="jar"
 fi
 
-# First check if we have a dependencies jar. If so, include binary classes with the deps jar
-if [ -f "$ASSEMBLY_DIR"/spark-assembly*hadoop*-deps.jar ]; then
+# A developer option to prepend more recently compiled Spark classes
+if [ -n "$SPARK_PREPEND_CLASSES" ]; then
+  echo "NOTE: SPARK_PREPEND_CLASSES is set, placing locally compiled Spark"\
+    "classes ahead of assembly." >&2
   CLASSPATH="$CLASSPATH:$FWDIR/core/target/scala-$SCALA_VERSION/classes"
   CLASSPATH="$CLASSPATH:$FWDIR/repl/target/scala-$SCALA_VERSION/classes"
   CLASSPATH="$CLASSPATH:$FWDIR/mllib/target/scala-$SCALA_VERSION/classes"
@@ -51,17 +53,31 @@ if [ -f "$ASSEMBLY_DIR"/spark-assembly*hadoop*-deps.jar ]; then
   CLASSPATH="$CLASSPATH:$FWDIR/sql/core/target/scala-$SCALA_VERSION/classes"
   CLASSPATH="$CLASSPATH:$FWDIR/sql/hive/target/scala-$SCALA_VERSION/classes"
   CLASSPATH="$CLASSPATH:$FWDIR/yarn/stable/target/scala-$SCALA_VERSION/classes"
+fi
 
-  ASSEMBLY_JAR=$(ls "$ASSEMBLY_DIR"/spark-assembly*hadoop*-deps.jar 2>/dev/null)
+# Use spark-assembly jar from either RELEASE or assembly directory
+if [ -f "$FWDIR/RELEASE" ]; then
+  assembly_folder="$FWDIR"/lib
 else
-  # Else use spark-assembly jar from either RELEASE or assembly directory
-  if [ -f "$FWDIR/RELEASE" ]; then
-    ASSEMBLY_JAR=$(ls "$FWDIR"/lib/spark-assembly*hadoop*.jar 2>/dev/null)
-  else
-    ASSEMBLY_JAR=$(ls "$ASSEMBLY_DIR"/spark-assembly*hadoop*.jar 2>/dev/null)
-  fi
+  assembly_folder="$ASSEMBLY_DIR"
 fi
 
+num_jars=$(ls "$assembly_folder" | grep "spark-assembly.*hadoop.*\.jar" | wc -l)
+if [ "$num_jars" -eq "0" ]; then
+  echo "Failed to find Spark assembly in $assembly_folder"
+  echo "You need to build Spark before running this program."
+  exit 1
+fi
+if [ "$num_jars" -gt "1" ]; then
+  jars_list=$(ls "$assembly_folder" | grep "spark-assembly.*hadoop.*.jar")
+  echo "Found multiple Spark assembly jars in $assembly_folder:"
+  echo "$jars_list"
+  echo "Please remove all but one jar."
+  exit 1
+fi
+
+ASSEMBLY_JAR=$(ls "$assembly_folder"/spark-assembly*hadoop*.jar 2>/dev/null)
+
 # Verify that versions of java used to build the jars and run Spark are compatible
 jar_error_check=$("$JAR_CMD" -tf "$ASSEMBLY_JAR" nonexistent/class/path 2>&1)
 if [[ "$jar_error_check" =~ "invalid CEN header" ]]; then

bin/pyspark

@@ -45,7 +45,7 @@ fi
 . $FWDIR/bin/load-spark-env.sh
 
 # Figure out which Python executable to use
-if [ -z "$PYSPARK_PYTHON" ] ; then
+if [[ -z "$PYSPARK_PYTHON" ]]; then
   PYSPARK_PYTHON="python"
 fi
 export PYSPARK_PYTHON
@@ -59,7 +59,7 @@ export OLD_PYTHONSTARTUP=$PYTHONSTARTUP
 export PYTHONSTARTUP=$FWDIR/python/pyspark/shell.py
 
 # If IPython options are specified, assume user wants to run IPython
-if [ -n "$IPYTHON_OPTS" ]; then
+if [[ -n "$IPYTHON_OPTS" ]]; then
   IPYTHON=1
 fi
 
@@ -76,6 +76,16 @@ for i in "$@"; do
 done
 export PYSPARK_SUBMIT_ARGS
 
+# For pyspark tests
+if [[ -n "$SPARK_TESTING" ]]; then
+  if [[ -n "$PYSPARK_DOC_TEST" ]]; then
+    exec "$PYSPARK_PYTHON" -m doctest $1
+  else
+    exec "$PYSPARK_PYTHON" $1
+  fi
+  exit
+fi
+
 # If a python file is provided, directly run spark-submit.
 if [[ "$1" =~ \.py$ ]]; then
   echo -e "\nWARNING: Running python applications through ./bin/pyspark is deprecated as of Spark 1.0." 1>&2

bin/spark-class

@@ -108,23 +108,6 @@ fi
 export JAVA_OPTS
 # Attention: when changing the way the JAVA_OPTS are assembled, the change must be reflected in CommandUtils.scala!
 
-if [ ! -f "$FWDIR/RELEASE" ]; then
-  # Exit if the user hasn't compiled Spark
-  num_jars=$(ls "$FWDIR"/assembly/target/scala-$SCALA_VERSION/ | grep "spark-assembly.*hadoop.*.jar" | wc -l)
-  jars_list=$(ls "$FWDIR"/assembly/target/scala-$SCALA_VERSION/ | grep "spark-assembly.*hadoop.*.jar")
-  if [ "$num_jars" -eq "0" ]; then
-    echo "Failed to find Spark assembly in $FWDIR/assembly/target/scala-$SCALA_VERSION/" >&2
-    echo "You need to build Spark before running this program." >&2
-    exit 1
-  fi
-  if [ "$num_jars" -gt "1" ]; then
-    echo "Found multiple Spark assembly jars in $FWDIR/assembly/target/scala-$SCALA_VERSION:" >&2
-    echo "$jars_list"
-    echo "Please remove all but one jar."
-    exit 1
-  fi
-fi
-
 TOOLS_DIR="$FWDIR"/tools
 SPARK_TOOLS_JAR=""
 if [ -e "$TOOLS_DIR"/target/scala-$SCALA_VERSION/*assembly*[0-9Tg].jar ]; then

core/pom.xml

@@ -67,6 +67,12 @@
       <groupId>org.apache.commons</groupId>
       <artifactId>commons-lang3</artifactId>
     </dependency>
+    <dependency>
+      <groupId>org.apache.commons</groupId>
+      <artifactId>commons-math3</artifactId>
+      <version>3.3</version>
+      <scope>test</scope>
+    </dependency>
     <dependency>
       <groupId>com.google.code.findbugs</groupId>
       <artifactId>jsr305</artifactId>

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

@@ -32,10 +32,14 @@ private[spark] class CacheManager(blockManager: BlockManager) extends Logging {
   private val loading = new HashSet[RDDBlockId]()
 
   /** Gets or computes an RDD split. Used by RDD.iterator() when an RDD is cached. */
-  def getOrCompute[T](rdd: RDD[T], split: Partition, context: TaskContext,
+  def getOrCompute[T](
+      rdd: RDD[T],
+      split: Partition,
+      context: TaskContext,
       storageLevel: StorageLevel): Iterator[T] = {
+
     val key = RDDBlockId(rdd.id, split.index)
-    logDebug("Looking for partition " + key)
+    logDebug(s"Looking for partition $key")
     blockManager.get(key) match {
       case Some(values) =>
         // Partition is already materialized, so just return its values
@@ -45,7 +49,7 @@ private[spark] class CacheManager(blockManager: BlockManager) extends Logging {
         // Mark the split as loading (unless someone else marks it first)
         loading.synchronized {
           if (loading.contains(key)) {
-            logInfo("Another thread is loading %s, waiting for it to finish...".format(key))
+            logInfo(s"Another thread is loading $key, waiting for it to finish...")
             while (loading.contains(key)) {
               try {
                 loading.wait()
@@ -54,7 +58,7 @@ private[spark] class CacheManager(blockManager: BlockManager) extends Logging {
                   logWarning(s"Got an exception while waiting for another thread to load $key", e)
               }
             }
-            logInfo("Finished waiting for %s".format(key))
+            logInfo(s"Finished waiting for $key")
             /* See whether someone else has successfully loaded it. The main way this would fail
              * is for the RDD-level cache eviction policy if someone else has loaded the same RDD
              * partition but we didn't want to make space for it. However, that case is unlikely
@@ -64,7 +68,7 @@ private[spark] class CacheManager(blockManager: BlockManager) extends Logging {
               case Some(values) =>
                 return new InterruptibleIterator(context, values.asInstanceOf[Iterator[T]])
               case None =>
-                logInfo("Whoever was loading %s failed; we'll try it ourselves".format(key))
+                logInfo(s"Whoever was loading $key failed; we'll try it ourselves")
                 loading.add(key)
             }
           } else {
@@ -73,7 +77,7 @@ private[spark] class CacheManager(blockManager: BlockManager) extends Logging {
         }
         try {
           // If we got here, we have to load the split
-          logInfo("Partition %s not found, computing it".format(key))
+          logInfo(s"Partition $key not found, computing it")
           val computedValues = rdd.computeOrReadCheckpoint(split, context)
 
           // Persist the result, so long as the task is not running locally
@@ -97,8 +101,8 @@ private[spark] class CacheManager(blockManager: BlockManager) extends Logging {
                 case Some(values) =>
                   values.asInstanceOf[Iterator[T]]
                 case None =>
-                  logInfo("Failure to store %s".format(key))
-                  throw new Exception("Block manager failed to return persisted valued")
+                  logInfo(s"Failure to store $key")
+                  throw new SparkException("Block manager failed to return persisted value")
               }
             } else {
               // In this case the RDD is cached to an array buffer. This will save the results

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

@@ -96,7 +96,7 @@ private[spark] class ContextCleaner(sc: SparkContext) extends Logging {
   }
 
   /** Register a ShuffleDependency for cleanup when it is garbage collected. */
-  def registerShuffleForCleanup(shuffleDependency: ShuffleDependency[_, _]) {
+  def registerShuffleForCleanup(shuffleDependency: ShuffleDependency[_, _, _]) {
     registerForCleanup(shuffleDependency, CleanShuffle(shuffleDependency.shuffleId))
   }
 

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

@@ -20,6 +20,7 @@ package org.apache.spark
 import org.apache.spark.annotation.DeveloperApi
 import org.apache.spark.rdd.RDD
 import org.apache.spark.serializer.Serializer
+import org.apache.spark.shuffle.ShuffleHandle
 
 /**
  * :: DeveloperApi ::
@@ -50,19 +51,24 @@ abstract class NarrowDependency[T](rdd: RDD[T]) extends Dependency(rdd) {
  * Represents a dependency on the output of a shuffle stage.
  * @param rdd the parent RDD
  * @param partitioner partitioner used to partition the shuffle output
- * @param serializer [[org.apache.spark.serializer.Serializer Serializer]] to use. If set to null,
+ * @param serializer [[org.apache.spark.serializer.Serializer Serializer]] to use. If set to None,
  *                   the default serializer, as specified by `spark.serializer` config option, will
  *                   be used.
  */
 @DeveloperApi
-class ShuffleDependency[K, V](
+class ShuffleDependency[K, V, C](
     @transient rdd: RDD[_ <: Product2[K, V]],
     val partitioner: Partitioner,
-    val serializer: Serializer = null)
+    val serializer: Option[Serializer] = None,
+    val keyOrdering: Option[Ordering[K]] = None,
+    val aggregator: Option[Aggregator[K, V, C]] = None)
   extends Dependency(rdd.asInstanceOf[RDD[Product2[K, V]]]) {
 
   val shuffleId: Int = rdd.context.newShuffleId()
 
+  val shuffleHandle: ShuffleHandle = rdd.context.env.shuffleManager.registerShuffle(
+    shuffleId, rdd.partitions.size, this)
+
   rdd.sparkContext.cleaner.foreach(_.registerShuffleForCleanup(this))
 }
 

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

@@ -290,14 +290,17 @@ class SparkContext(config: SparkConf) extends Logging {
     value <- Option(System.getenv(envKey)).orElse(Option(System.getProperty(propKey)))} {
     executorEnvs(envKey) = value
   }
+  Option(System.getenv("SPARK_PREPEND_CLASSES")).foreach { v => 
+    executorEnvs("SPARK_PREPEND_CLASSES") = v
+  }
   // The Mesos scheduler backend relies on this environment variable to set executor memory.
   // TODO: Set this only in the Mesos scheduler.
   executorEnvs("SPARK_EXECUTOR_MEMORY") = executorMemory + "m"
   executorEnvs ++= conf.getExecutorEnv
 
   // Set SPARK_USER for user who is running SparkContext.
   val sparkUser = Option {
-    Option(System.getProperty("user.name")).getOrElse(System.getenv("SPARK_USER"))
+    Option(System.getenv("SPARK_USER")).getOrElse(System.getProperty("user.name"))
   }.getOrElse {
     SparkContext.SPARK_UNKNOWN_USER
   }
@@ -431,12 +434,21 @@ class SparkContext(config: SparkConf) extends Logging {
 
   // Methods for creating RDDs
 
-  /** Distribute a local Scala collection to form an RDD. */
+  /** Distribute a local Scala collection to form an RDD.
+   *
+   * @note Parallelize acts lazily. If `seq` is a mutable collection and is
+   * altered after the call to parallelize and before the first action on the
+   * RDD, the resultant RDD will reflect the modified collection. Pass a copy of
+   * the argument to avoid this.
+   */
   def parallelize[T: ClassTag](seq: Seq[T], numSlices: Int = defaultParallelism): RDD[T] = {
     new ParallelCollectionRDD[T](this, seq, numSlices, Map[Int, Seq[String]]())
   }
 
-  /** Distribute a local Scala collection to form an RDD. */
+  /** Distribute a local Scala collection to form an RDD.
+   *
+   * This method is identical to `parallelize`.
+   */
   def makeRDD[T: ClassTag](seq: Seq[T], numSlices: Int = defaultParallelism): RDD[T] = {
     parallelize(seq, numSlices)
   }
@@ -823,9 +835,11 @@ class SparkContext(config: SparkConf) extends Logging {
   }
 
   /**
+   * :: DeveloperApi ::
    * Return information about what RDDs are cached, if they are in mem or on disk, how much space
    * they take, etc.
    */
+  @DeveloperApi
   def getRDDStorageInfo: Array[RDDInfo] = {
     StorageUtils.rddInfoFromStorageStatus(getExecutorStorageStatus, this)
   }
@@ -837,8 +851,10 @@ class SparkContext(config: SparkConf) extends Logging {
   def getPersistentRDDs: Map[Int, RDD[_]] = persistentRdds.toMap
 
   /**
+   * :: DeveloperApi ::
    * Return information about blocks stored in all of the slaves
    */
+  @DeveloperApi
   def getExecutorStorageStatus: Array[StorageStatus] = {
     env.blockManager.master.getStorageStatus
   }