merge master

Author: mengxr

bin/pyspark

@@ -50,22 +50,47 @@ fi
 
 . "$FWDIR"/bin/load-spark-env.sh
 
-# Figure out which Python executable to use
+# In Spark <= 1.1, setting IPYTHON=1 would cause the driver to be launched using the `ipython`
+# executable, while the worker would still be launched using PYSPARK_PYTHON.
+#
+# In Spark 1.2, we removed the documentation of the IPYTHON and IPYTHON_OPTS variables and added
+# PYSPARK_DRIVER_PYTHON and PYSPARK_DRIVER_PYTHON_OPTS to allow IPython to be used for the driver.
+# Now, users can simply set PYSPARK_DRIVER_PYTHON=ipython to use IPython and set
+# PYSPARK_DRIVER_PYTHON_OPTS to pass options when starting the Python driver
+# (e.g. PYSPARK_DRIVER_PYTHON_OPTS='notebook').  This supports full customization of the IPython
+# and executor Python executables.
+#
+# For backwards-compatibility, we retain the old IPYTHON and IPYTHON_OPTS variables.
+
+# Determine the Python executable to use if PYSPARK_PYTHON or PYSPARK_DRIVER_PYTHON isn't set:
+if hash python2.7 2>/dev/null; then
+  # Attempt to use Python 2.7, if installed:
+  DEFAULT_PYTHON="python2.7"
+else
+  DEFAULT_PYTHON="python"
+fi
+
+# Determine the Python executable to use for the driver:
+if [[ -n "$IPYTHON_OPTS" || "$IPYTHON" == "1" ]]; then
+  # If IPython options are specified, assume user wants to run IPython
+  # (for backwards-compatibility)
+  PYSPARK_DRIVER_PYTHON_OPTS="$PYSPARK_DRIVER_PYTHON_OPTS $IPYTHON_OPTS"
+  PYSPARK_DRIVER_PYTHON="ipython"
+elif [[ -z "$PYSPARK_DRIVER_PYTHON" ]]; then
+  PYSPARK_DRIVER_PYTHON="${PYSPARK_PYTHON:-"$DEFAULT_PYTHON"}"
+fi
+
+# Determine the Python executable to use for the executors:
 if [[ -z "$PYSPARK_PYTHON" ]]; then
-  if [[ "$IPYTHON" = "1" || -n "$IPYTHON_OPTS" ]]; then
-    # for backward compatibility
-    PYSPARK_PYTHON="ipython"
+  if [[ $PYSPARK_DRIVER_PYTHON == *ipython* && $DEFAULT_PYTHON != "python2.7" ]]; then
+    echo "IPython requires Python 2.7+; please install python2.7 or set PYSPARK_PYTHON" 1>&2
+    exit 1
   else
-    PYSPARK_PYTHON="python"
+    PYSPARK_PYTHON="$DEFAULT_PYTHON"
   fi
 fi
 export PYSPARK_PYTHON
 
-if [[ -z "$PYSPARK_PYTHON_OPTS" && -n "$IPYTHON_OPTS" ]]; then
-  # for backward compatibility
-  PYSPARK_PYTHON_OPTS="$IPYTHON_OPTS"
-fi
-
 # Add the PySpark classes to the Python path:
 export PYTHONPATH="$SPARK_HOME/python/:$PYTHONPATH"
 export PYTHONPATH="$SPARK_HOME/python/lib/py4j-0.8.2.1-src.zip:$PYTHONPATH"
@@ -93,9 +118,9 @@ if [[ -n "$SPARK_TESTING" ]]; then
   unset YARN_CONF_DIR
   unset HADOOP_CONF_DIR
   if [[ -n "$PYSPARK_DOC_TEST" ]]; then
-    exec "$PYSPARK_PYTHON" -m doctest $1
+    exec "$PYSPARK_DRIVER_PYTHON" -m doctest $1
   else
-    exec "$PYSPARK_PYTHON" $1
+    exec "$PYSPARK_DRIVER_PYTHON" $1
   fi
   exit
 fi
@@ -111,5 +136,5 @@ if [[ "$1" =~ \.py$ ]]; then
 else
   # PySpark shell requires special handling downstream
   export PYSPARK_SHELL=1
-  exec "$PYSPARK_PYTHON" $PYSPARK_PYTHON_OPTS
+  exec "$PYSPARK_DRIVER_PYTHON" $PYSPARK_DRIVER_PYTHON_OPTS
 fi

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

@@ -21,6 +21,7 @@ import scala.language.implicitConversions
 
 import java.io._
 import java.net.URI
+import java.util.Arrays
 import java.util.concurrent.atomic.AtomicInteger
 import java.util.{Properties, UUID}
 import java.util.UUID.randomUUID
@@ -1429,7 +1430,10 @@ object SparkContext extends Logging {
     simpleWritableConverter[Boolean, BooleanWritable](_.get)
 
   implicit def bytesWritableConverter(): WritableConverter[Array[Byte]] = {
-    simpleWritableConverter[Array[Byte], BytesWritable](_.getBytes)
+    simpleWritableConverter[Array[Byte], BytesWritable](bw =>
+      // getBytes method returns array which is longer then data to be returned
+      Arrays.copyOfRange(bw.getBytes, 0, bw.getLength)
+    )
   }
 
   implicit def stringWritableConverter(): WritableConverter[String] =

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

@@ -26,6 +26,8 @@ import scala.collection.JavaConversions._
 import javax.tools.{JavaFileObject, SimpleJavaFileObject, ToolProvider}
 import com.google.common.io.Files
 
+import org.apache.spark.util.Utils
+
 /**
  * Utilities for tests. Included in main codebase since it's used by multiple
  * projects.
@@ -42,8 +44,7 @@ private[spark] object TestUtils {
    * in order to avoid interference between tests.
    */
   def createJarWithClasses(classNames: Seq[String], value: String = ""): URL = {
-    val tempDir = Files.createTempDir()
-    tempDir.deleteOnExit()
+    val tempDir = Utils.createTempDir()
     val files = for (name <- classNames) yield createCompiledClass(name, tempDir, value)
     val jarFile = new File(tempDir, "testJar-%s.jar".format(System.currentTimeMillis()))
     createJar(files, jarFile)

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

@@ -25,8 +25,6 @@ import java.util.{List => JList, ArrayList => JArrayList, Map => JMap, Collectio
 import scala.collection.JavaConversions._
 import scala.collection.mutable
 import scala.language.existentials
-import scala.reflect.ClassTag
-import scala.util.{Try, Success, Failure}
 
 import net.razorvine.pickle.{Pickler, Unpickler}
 
@@ -42,7 +40,7 @@ import org.apache.spark.rdd.RDD
 import org.apache.spark.util.Utils
 
 private[spark] class PythonRDD(
-    parent: RDD[_],
+    @transient parent: RDD[_],
     command: Array[Byte],
     envVars: JMap[String, String],
     pythonIncludes: JList[String],
@@ -55,9 +53,9 @@ private[spark] class PythonRDD(
   val bufferSize = conf.getInt("spark.buffer.size", 65536)
   val reuse_worker = conf.getBoolean("spark.python.worker.reuse", true)
 
-  override def getPartitions = parent.partitions
+  override def getPartitions = firstParent.partitions
 
-  override val partitioner = if (preservePartitoning) parent.partitioner else None
+  override val partitioner = if (preservePartitoning) firstParent.partitioner else None
 
   override def compute(split: Partition, context: TaskContext): Iterator[Array[Byte]] = {
     val startTime = System.currentTimeMillis
@@ -234,7 +232,7 @@ private[spark] class PythonRDD(
         dataOut.writeInt(command.length)
         dataOut.write(command)
         // Data values
-        PythonRDD.writeIteratorToStream(parent.iterator(split, context), dataOut)
+        PythonRDD.writeIteratorToStream(firstParent.iterator(split, context), dataOut)
         dataOut.writeInt(SpecialLengths.END_OF_DATA_SECTION)
         dataOut.flush()
       } catch {

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

@@ -108,10 +108,12 @@ private[spark] class PythonWorkerFactory(pythonExec: String, envVars: Map[String
       serverSocket = new ServerSocket(0, 1, InetAddress.getByAddress(Array(127, 0, 0, 1)))
 
       // Create and start the worker
-      val pb = new ProcessBuilder(Seq(pythonExec, "-u", "-m", "pyspark.worker"))
+      val pb = new ProcessBuilder(Seq(pythonExec, "-m", "pyspark.worker"))
       val workerEnv = pb.environment()
       workerEnv.putAll(envVars)
       workerEnv.put("PYTHONPATH", pythonPath)
+      // This is equivalent to setting the -u flag; we use it because ipython doesn't support -u:
+      workerEnv.put("PYTHONUNBUFFERED", "YES")
       val worker = pb.start()
 
       // Redirect worker stdout and stderr
@@ -149,10 +151,12 @@ private[spark] class PythonWorkerFactory(pythonExec: String, envVars: Map[String
 
       try {
         // Create and start the daemon
-        val pb = new ProcessBuilder(Seq(pythonExec, "-u", "-m", "pyspark.daemon"))
+        val pb = new ProcessBuilder(Seq(pythonExec, "-m", "pyspark.daemon"))
         val workerEnv = pb.environment()
         workerEnv.putAll(envVars)
         workerEnv.put("PYTHONPATH", pythonPath)
+        // This is equivalent to setting the -u flag; we use it because ipython doesn't support -u:
+        workerEnv.put("PYTHONUNBUFFERED", "YES")
         daemon = pb.start()
 
         val in = new DataInputStream(daemon.getInputStream)

core/src/main/scala/org/apache/spark/deploy/Client.scala

@@ -130,7 +130,7 @@ private class ClientActor(driverArgs: ClientArguments, conf: SparkConf)
       println(s"Error connecting to master ${driverArgs.master} ($remoteAddress), exiting.")
       System.exit(-1)
 
-    case AssociationErrorEvent(cause, _, remoteAddress, _) =>
+    case AssociationErrorEvent(cause, _, remoteAddress, _, _) =>
       println(s"Error connecting to master ${driverArgs.master} ($remoteAddress), exiting.")
       println(s"Cause was: $cause")
       System.exit(-1)

core/src/main/scala/org/apache/spark/deploy/PythonRunner.scala

@@ -34,7 +34,8 @@ object PythonRunner {
     val pythonFile = args(0)
     val pyFiles = args(1)
     val otherArgs = args.slice(2, args.length)
-    val pythonExec = sys.env.get("PYSPARK_PYTHON").getOrElse("python") // TODO: get this from conf
+    val pythonExec =
+      sys.env.getOrElse("PYSPARK_DRIVER_PYTHON", sys.env.getOrElse("PYSPARK_PYTHON", "python"))
 
     // Format python file paths before adding them to the PYTHONPATH
     val formattedPythonFile = formatPath(pythonFile)
@@ -57,6 +58,7 @@ object PythonRunner {
     val builder = new ProcessBuilder(Seq(pythonExec, formattedPythonFile) ++ otherArgs)
     val env = builder.environment()
     env.put("PYTHONPATH", pythonPath)
+    // This is equivalent to setting the -u flag; we use it because ipython doesn't support -u:
     env.put("PYTHONUNBUFFERED", "YES") // value is needed to be set to a non-empty string
     env.put("PYSPARK_GATEWAY_PORT", "" + gatewayServer.getListeningPort)
     builder.redirectErrorStream(true) // Ugly but needed for stdout and stderr to synchronize

core/src/main/scala/org/apache/spark/deploy/client/AppClient.scala

@@ -154,7 +154,7 @@ private[spark] class AppClient(
         logWarning(s"Connection to $address failed; waiting for master to reconnect...")
         markDisconnected()
 
-      case AssociationErrorEvent(cause, _, address, _) if isPossibleMaster(address) =>
+      case AssociationErrorEvent(cause, _, address, _, _) if isPossibleMaster(address) =>
         logWarning(s"Could not connect to $address: $cause")
 
       case StopAppClient =>

core/src/main/scala/org/apache/spark/deploy/worker/WorkerWatcher.scala

@@ -54,7 +54,7 @@ private[spark] class WorkerWatcher(workerUrl: String)
     case AssociatedEvent(localAddress, remoteAddress, inbound) if isWorker(remoteAddress) =>
       logInfo(s"Successfully connected to $workerUrl")
 
-    case AssociationErrorEvent(cause, localAddress, remoteAddress, inbound)
+    case AssociationErrorEvent(cause, localAddress, remoteAddress, inbound, _)
         if isWorker(remoteAddress) =>
       // These logs may not be seen if the worker (and associated pipe) has died
       logError(s"Could not initialize connection to worker $workerUrl. Exiting.")

core/src/main/scala/org/apache/spark/network/ManagedBuffer.scala

@@ -66,13 +66,27 @@ sealed abstract class ManagedBuffer {
 final class FileSegmentManagedBuffer(val file: File, val offset: Long, val length: Long)
   extends ManagedBuffer {
 
+  /**
+   * Memory mapping is expensive and can destabilize the JVM (SPARK-1145, SPARK-3889).
+   * Avoid unless there's a good reason not to.
+   */
+  private val MIN_MEMORY_MAP_BYTES = 2 * 1024 * 1024;
+
   override def size: Long = length
 
   override def nioByteBuffer(): ByteBuffer = {
     var channel: FileChannel = null
     try {
       channel = new RandomAccessFile(file, "r").getChannel
-      channel.map(MapMode.READ_ONLY, offset, length)
+      // Just copy the buffer if it's sufficiently small, as memory mapping has a high overhead.
+      if (length < MIN_MEMORY_MAP_BYTES) {
+        val buf = ByteBuffer.allocate(length.toInt)
+        channel.read(buf, offset)
+        buf.flip()
+        buf
+      } else {
+        channel.map(MapMode.READ_ONLY, offset, length)
+      }
     } catch {
       case e: IOException =>
         Try(channel.size).toOption match {