[ML][SPARK-6529] Add Word2Vec transformer

R/pkg/NAMESPACE

@@ -71,6 +71,7 @@ exportMethods(
               "unpersist",
               "value",
               "values",
+              "zipPartitions",
               "zipRDD",
               "zipWithIndex",
               "zipWithUniqueId"

R/pkg/R/RDD.R

@@ -66,6 +66,11 @@ setMethod("initialize", "RDD", function(.Object, jrdd, serializedMode,
   .Object
 })
 
+setMethod("show", "RDD",
+          function(.Object) {
+              cat(paste(callJMethod(.Object@jrdd, "toString"), "\n", sep=""))
+          })
+
 setMethod("initialize", "PipelinedRDD", function(.Object, prev, func, jrdd_val) {
   .Object@env <- new.env()
   .Object@env$isCached <- FALSE
@@ -1590,3 +1595,49 @@ setMethod("intersection",
 
             keys(filterRDD(cogroup(rdd1, rdd2, numPartitions = numPartitions), filterFunction))
           })
+
+#' Zips an RDD's partitions with one (or more) RDD(s).
+#' Same as zipPartitions in Spark.
+#' 
+#' @param ... RDDs to be zipped.
+#' @param func A function to transform zipped partitions.
+#' @return A new RDD by applying a function to the zipped partitions. 
+#'         Assumes that all the RDDs have the *same number of partitions*, but 
+#'         does *not* require them to have the same number of elements in each partition.
+#' @examples
+#'\dontrun{
+#' sc <- sparkR.init()
+#' rdd1 <- parallelize(sc, 1:2, 2L)  # 1, 2
+#' rdd2 <- parallelize(sc, 1:4, 2L)  # 1:2, 3:4
+#' rdd3 <- parallelize(sc, 1:6, 2L)  # 1:3, 4:6
+#' collect(zipPartitions(rdd1, rdd2, rdd3, 
+#'                       func = function(x, y, z) { list(list(x, y, z))} ))
+#' # list(list(1, c(1,2), c(1,2,3)), list(2, c(3,4), c(4,5,6)))
+#'}
+#' @rdname zipRDD
+#' @aliases zipPartitions,RDD
+setMethod("zipPartitions",
+          "RDD",
+          function(..., func) {
+            rrdds <- list(...)
+            if (length(rrdds) == 1) {
+              return(rrdds[[1]])
+            }
+            nPart <- sapply(rrdds, numPartitions)
+            if (length(unique(nPart)) != 1) {
+              stop("Can only zipPartitions RDDs which have the same number of partitions.")
+            }
+
+            rrdds <- lapply(rrdds, function(rdd) {
+              mapPartitionsWithIndex(rdd, function(partIndex, part) {
+                print(length(part))
+                list(list(partIndex, part))
+              })
+            })
+            union.rdd <- Reduce(unionRDD, rrdds)
+            zipped.rdd <- values(groupByKey(union.rdd, numPartitions = nPart[1]))
+            res <- mapPartitions(zipped.rdd, function(plist) {
+              do.call(func, plist[[1]])
+            })
+            res
+          })

R/pkg/R/generics.R

@@ -217,6 +217,11 @@ setGeneric("unpersist", function(x, ...) { standardGeneric("unpersist") })
 #' @export
 setGeneric("zipRDD", function(x, other) { standardGeneric("zipRDD") })
 
+#' @rdname zipRDD
+#' @export
+setGeneric("zipPartitions", function(..., func) { standardGeneric("zipPartitions") }, 
+           signature = "...")
+
 #' @rdname zipWithIndex
 #' @seealso zipWithUniqueId
 #' @export

R/pkg/inst/tests/test_binary_function.R

@@ -66,3 +66,36 @@ test_that("cogroup on two RDDs", {
   expect_equal(sortKeyValueList(actual),
                sortKeyValueList(expected))
 })
+
+test_that("zipPartitions() on RDDs", {
+  rdd1 <- parallelize(sc, 1:2, 2L)  # 1, 2
+  rdd2 <- parallelize(sc, 1:4, 2L)  # 1:2, 3:4
+  rdd3 <- parallelize(sc, 1:6, 2L)  # 1:3, 4:6
+  actual <- collect(zipPartitions(rdd1, rdd2, rdd3, 
+                                  func = function(x, y, z) { list(list(x, y, z))} ))
+  expect_equal(actual,
+               list(list(1, c(1,2), c(1,2,3)), list(2, c(3,4), c(4,5,6))))
+
+  mockFile = c("Spark is pretty.", "Spark is awesome.")
+  fileName <- tempfile(pattern="spark-test", fileext=".tmp")
+  writeLines(mockFile, fileName)
+
+  rdd <- textFile(sc, fileName, 1)
+  actual <- collect(zipPartitions(rdd, rdd, 
+                                  func = function(x, y) { list(paste(x, y, sep = "\n")) }))
+  expected <- list(paste(mockFile, mockFile, sep = "\n"))
+  expect_equal(actual, expected)
+
+  rdd1 <- parallelize(sc, 0:1, 1)
+  actual <- collect(zipPartitions(rdd1, rdd, 
+                                  func = function(x, y) { list(x + nchar(y)) }))
+  expected <- list(0:1 + nchar(mockFile))
+  expect_equal(actual, expected)
+
+  rdd <- map(rdd, function(x) { x })
+  actual <- collect(zipPartitions(rdd, rdd1, 
+                                  func = function(x, y) { list(y + nchar(x)) }))
+  expect_equal(actual, expected)
+
+  unlink(fileName)
+})

R/pkg/inst/tests/test_rdd.R

@@ -759,6 +759,11 @@ test_that("collectAsMap() on a pairwise RDD", {
   expect_equal(vals, list(`1` = "a", `2` = "b"))
 })
 
+test_that("show()", {
+  rdd <- parallelize(sc, list(1:10))    
+  expect_output(show(rdd), "ParallelCollectionRDD\\[\\d+\\] at parallelize at RRDD\\.scala:\\d+")
+})
+
 test_that("sampleByKey() on pairwise RDDs", {
   rdd <- parallelize(sc, 1:2000)
   pairsRDD <- lapply(rdd, function(x) { if (x %% 2 == 0) list("a", x) else list("b", x) })

assembly/pom.xml

@@ -194,7 +194,6 @@
           <plugin>
             <groupId>org.apache.maven.plugins</groupId>
             <artifactId>maven-assembly-plugin</artifactId>
-            <version>2.4</version>
             <executions>
               <execution>
                 <id>dist</id>

bin/spark-class2.cmd

@@ -61,7 +61,10 @@ if not "x%JAVA_HOME%"=="x" set RUNNER=%JAVA_HOME%\bin\java
 
 rem The launcher library prints the command to be executed in a single line suitable for being
 rem executed by the batch interpreter. So read all the output of the launcher into a variable.
-for /f "tokens=*" %%i in ('cmd /C ""%RUNNER%" -cp %LAUNCH_CLASSPATH% org.apache.spark.launcher.Main %*"') do (
+set LAUNCHER_OUTPUT=%temp%\spark-class-launcher-output-%RANDOM%.txt
+"%RUNNER%" -cp %LAUNCH_CLASSPATH% org.apache.spark.launcher.Main %* > %LAUNCHER_OUTPUT%
+for /f "tokens=*" %%i in (%LAUNCHER_OUTPUT%) do (
   set SPARK_CMD=%%i
 )
+del %LAUNCHER_OUTPUT%
 %SPARK_CMD%

conf/spark-env.sh.template

@@ -3,7 +3,7 @@
 # This file is sourced when running various Spark programs.
 # Copy it as spark-env.sh and edit that to configure Spark for your site.
 
-# Options read when launching programs locally with 
+# Options read when launching programs locally with
 # ./bin/run-example or ./bin/spark-submit
 # - HADOOP_CONF_DIR, to point Spark towards Hadoop configuration files
 # - SPARK_LOCAL_IP, to set the IP address Spark binds to on this node
@@ -39,6 +39,7 @@
 # - SPARK_WORKER_DIR, to set the working directory of worker processes
 # - SPARK_WORKER_OPTS, to set config properties only for the worker (e.g. "-Dx=y")
 # - SPARK_HISTORY_OPTS, to set config properties only for the history server (e.g. "-Dx=y")
+# - SPARK_SHUFFLE_OPTS, to set config properties only for the external shuffle service (e.g. "-Dx=y")
 # - SPARK_DAEMON_JAVA_OPTS, to set config properties for all daemons (e.g. "-Dx=y")
 # - SPARK_PUBLIC_DNS, to set the public dns name of the master or workers
 

core/pom.xml

@@ -95,6 +95,11 @@
       <artifactId>spark-network-shuffle_${scala.binary.version}</artifactId>
       <version>${project.version}</version>
     </dependency>
+    <dependency>
+      <groupId>org.apache.spark</groupId>
+      <artifactId>spark-unsafe_${scala.binary.version}</artifactId>
+      <version>${project.version}</version>
+    </dependency>
     <dependency>
       <groupId>net.java.dev.jets3t</groupId>
       <artifactId>jets3t</artifactId>
@@ -478,7 +483,6 @@
           <plugin>
             <groupId>org.codehaus.mojo</groupId>
             <artifactId>exec-maven-plugin</artifactId>
-            <version>1.3.2</version>
             <executions>
               <execution>
                 <id>sparkr-pkg</id>

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

@@ -76,13 +76,15 @@ private[spark] class HeartbeatReceiver(sc: SparkContext)
 
   private var timeoutCheckingTask: ScheduledFuture[_] = null
 
-  private val timeoutCheckingThread =
-    ThreadUtils.newDaemonSingleThreadScheduledExecutor("heartbeat-timeout-checking-thread")
+  // "eventLoopThread" is used to run some pretty fast actions. The actions running in it should not
+  // block the thread for a long time.
+  private val eventLoopThread =
+    ThreadUtils.newDaemonSingleThreadScheduledExecutor("heartbeat-receiver-event-loop-thread")
 
   private val killExecutorThread = ThreadUtils.newDaemonSingleThreadExecutor("kill-executor-thread")
 
   override def onStart(): Unit = {
-    timeoutCheckingTask = timeoutCheckingThread.scheduleAtFixedRate(new Runnable {
+    timeoutCheckingTask = eventLoopThread.scheduleAtFixedRate(new Runnable {
       override def run(): Unit = Utils.tryLogNonFatalError {
         Option(self).foreach(_.send(ExpireDeadHosts))
       }
@@ -99,11 +101,15 @@ private[spark] class HeartbeatReceiver(sc: SparkContext)
   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()
-        context.reply(response)
+        eventLoopThread.submit(new Runnable {
+          override def run(): Unit = Utils.tryLogNonFatalError {
+            val unknownExecutor = !scheduler.executorHeartbeatReceived(
+              executorId, taskMetrics, blockManagerId)
+            val response = HeartbeatResponse(reregisterBlockManager = unknownExecutor)
+            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
@@ -125,7 +131,9 @@ private[spark] class HeartbeatReceiver(sc: SparkContext)
         if (sc.supportDynamicAllocation) {
           // Asynchronously kill the executor to avoid blocking the current thread
           killExecutorThread.submit(new Runnable {
-            override def run(): Unit = sc.killExecutor(executorId)
+            override def run(): Unit = Utils.tryLogNonFatalError {
+              sc.killExecutor(executorId)
+            }
           })
         }
         executorLastSeen.remove(executorId)
@@ -137,7 +145,7 @@ private[spark] class HeartbeatReceiver(sc: SparkContext)
     if (timeoutCheckingTask != null) {
       timeoutCheckingTask.cancel(true)
     }
-    timeoutCheckingThread.shutdownNow()
+    eventLoopThread.shutdownNow()
     killExecutorThread.shutdownNow()
   }
 }

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

@@ -106,7 +106,7 @@ private[spark] abstract class MapOutputTracker(conf: SparkConf) extends Logging
    */
   protected def askTracker[T: ClassTag](message: Any): T = {
     try {
-      trackerEndpoint.askWithReply[T](message)
+      trackerEndpoint.askWithRetry[T](message)
     } catch {
       case e: Exception =>
         logError("Error communicating with MapOutputTracker", e)

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

@@ -211,7 +211,74 @@ class SparkConf(loadDefaults: Boolean) extends Cloneable with Logging {
     Utils.timeStringAsMs(get(key, defaultValue))
   }
 
+  /**
+   * Get a size parameter as bytes; throws a NoSuchElementException if it's not set. If no
+   * suffix is provided then bytes are assumed.
+   * @throws NoSuchElementException
+   */
+  def getSizeAsBytes(key: String): Long = {
+    Utils.byteStringAsBytes(get(key))
+  }
+
+  /**
+   * Get a size parameter as bytes, falling back to a default if not set. If no
+   * suffix is provided then bytes are assumed.
+   */
+  def getSizeAsBytes(key: String, defaultValue: String): Long = {
+    Utils.byteStringAsBytes(get(key, defaultValue))
+  }
+
+  /**
+   * Get a size parameter as Kibibytes; throws a NoSuchElementException if it's not set. If no
+   * suffix is provided then Kibibytes are assumed.
+   * @throws NoSuchElementException
+   */
+  def getSizeAsKb(key: String): Long = {
+    Utils.byteStringAsKb(get(key))
+  }
+
+  /**
+   * Get a size parameter as Kibibytes, falling back to a default if not set. If no
+   * suffix is provided then Kibibytes are assumed.
+   */
+  def getSizeAsKb(key: String, defaultValue: String): Long = {
+    Utils.byteStringAsKb(get(key, defaultValue))
+  }
+
+  /**
+   * Get a size parameter as Mebibytes; throws a NoSuchElementException if it's not set. If no
+   * suffix is provided then Mebibytes are assumed.
+   * @throws NoSuchElementException
+   */
+  def getSizeAsMb(key: String): Long = {
+    Utils.byteStringAsMb(get(key))
+  }
+
+  /**
+   * Get a size parameter as Mebibytes, falling back to a default if not set. If no
+   * suffix is provided then Mebibytes are assumed.
+   */
+  def getSizeAsMb(key: String, defaultValue: String): Long = {
+    Utils.byteStringAsMb(get(key, defaultValue))
+  }
+
+  /**
+   * Get a size parameter as Gibibytes; throws a NoSuchElementException if it's not set. If no
+   * suffix is provided then Gibibytes are assumed.
+   * @throws NoSuchElementException
+   */
+  def getSizeAsGb(key: String): Long = {
+    Utils.byteStringAsGb(get(key))
+  }
 
+  /**
+   * Get a size parameter as Gibibytes, falling back to a default if not set. If no
+   * suffix is provided then Gibibytes are assumed.
+   */
+  def getSizeAsGb(key: String, defaultValue: String): Long = {
+    Utils.byteStringAsGb(get(key, defaultValue))
+  }
+
   /** Get a parameter as an Option */
   def getOption(key: String): Option[String] = {
     Option(settings.get(key)).orElse(getDeprecatedConfig(key, this))
@@ -407,7 +474,13 @@ private[spark] object SparkConf extends Logging {
         "The spark.cache.class property is no longer being used! Specify storage levels using " +
         "the RDD.persist() method instead."),
       DeprecatedConfig("spark.yarn.user.classpath.first", "1.3",
-        "Please use spark.{driver,executor}.userClassPathFirst instead."))
+        "Please use spark.{driver,executor}.userClassPathFirst instead."),
+      DeprecatedConfig("spark.kryoserializer.buffer.mb", "1.4",
+        "Please use spark.kryoserializer.buffer instead. The default value for " +
+          "spark.kryoserializer.buffer.mb was previously specified as '0.064'. Fractional values " +
+          "are no longer accepted. To specify the equivalent now, one may use '64k'.")
+    )
+
     Map(configs.map { cfg => (cfg.key -> cfg) }:_*)
   }
 
@@ -432,6 +505,21 @@ private[spark] object SparkConf extends Logging {
       AlternateConfig("spark.yarn.applicationMaster.waitTries", "1.3",
         // Translate old value to a duration, with 10s wait time per try.
         translation = s => s"${s.toLong * 10}s")),
+    "spark.reducer.maxSizeInFlight" -> Seq(
+      AlternateConfig("spark.reducer.maxMbInFlight", "1.4")),
+    "spark.kryoserializer.buffer" ->
+        Seq(AlternateConfig("spark.kryoserializer.buffer.mb", "1.4", 
+          translation = s => s"${s.toDouble * 1000}k")),
+    "spark.kryoserializer.buffer.max" -> Seq(
+      AlternateConfig("spark.kryoserializer.buffer.max.mb", "1.4")),
+    "spark.shuffle.file.buffer" -> Seq(
+      AlternateConfig("spark.shuffle.file.buffer.kb", "1.4")),
+    "spark.executor.logs.rolling.maxSize" -> Seq(
+      AlternateConfig("spark.executor.logs.rolling.size.maxBytes", "1.4")),
+    "spark.io.compression.snappy.blockSize" -> Seq(
+      AlternateConfig("spark.io.compression.snappy.block.size", "1.4")),
+    "spark.io.compression.lz4.blockSize" -> Seq(
+      AlternateConfig("spark.io.compression.lz4.block.size", "1.4")),
     "spark.rpc.numRetries" -> Seq(
       AlternateConfig("spark.akka.num.retries", "1.4")),
     "spark.rpc.retry.wait" -> Seq(