深度剖析Spark分布式执行原理

import java.io.Serializable;

public abstract class Task implements Serializable {
    public void run() {
        System.out.println("run task!");
    }
}

public class SimpleTask extends Task {
    @Override
    public void run() {
        System.out.println("run simple task!");
    }
}

import java.io.{FileInputStream, FileOutputStream, ObjectInputStream, ObjectOutputStream}

object FileSerializer {

  def writeObjectToFile(obj: Object, file: String) = {
    val fileStream = new FileOutputStream(file)
    val oos = new ObjectOutputStream(fileStream)
    oos.writeObject(obj)
    oos.close()
  }

  def readObjectFromFile(file: String): Object = {
    val fileStream = new FileInputStream(file)
    val ois = new ObjectInputStream(fileStream)
    val obj = ois.readObject()
    ois.close()
    obj
  }
}

val task = new SimpleTask()
FileSerializer.writeObjectToFile(task, "task.ser")

val task = FileSerializer.readObjectFromFile("task.ser").asInstanceOf[Task]
task.run()

Exception in thread "main" java.lang.ClassNotFoundException: site.stanzhai.serialization.SimpleTask
    at java.net.URLClassLoader.findClass(URLClassLoader.java:381)
    at java.lang.ClassLoader.loadClass(ClassLoader.java:424)
    at sun.misc.Launcher$AppClassLoader.loadClass(Launcher.java:331)
    at java.lang.ClassLoader.loadClass(ClassLoader.java:357)
    at java.lang.Class.forName0(Native Method)
    at java.lang.Class.forName(Class.java:348)
    at java.io.ObjectInputStream.resolveClass(ObjectInputStream.java:628)
    at java.io.ObjectInputStream.readNonProxyDesc(ObjectInputStream.java:1620)
    at java.io.ObjectInputStream.readClassDesc(ObjectInputStream.java:1521)
    at java.io.ObjectInputStream.readOrdinaryObject(ObjectInputStream.java:1781)
    at java.io.ObjectInputStream.readObject0(ObjectInputStream.java:1353)
    at java.io.ObjectInputStream.readObject(ObjectInputStream.java:373)
    at site.stanzhai.serialization.FileSerializer$.readObjectFromFile(FileSerializer.scala:20)

object ClassManipulator {
  def saveClassFile(obj: AnyRef): Unit = {
    val classLoader = obj.getClass.getClassLoader
    val className = obj.getClass.getName
    val classFile = className.replace(‘.‘, ‘/‘) + ".class"
    val stream = classLoader.getResourceAsStream(classFile)

    // just use the class simple name as the file name
    val outputFile = className.split(‘.‘).last + ".class"
    val fileStream = new FileOutputStream(outputFile)
    var data = stream.read()
    while (data != -1) {
      fileStream.write(data)
      data = stream.read()
    }
    fileStream.flush()
    fileStream.close()
  }
}

val task = new SimpleTask()
FileSerializer.writeObjectToFile(task, "task.ser")
ClassManipulator.saveClassFile(task)

class FileClassLoader() extends ClassLoader {
  override def findClass(fullClassName: String): Class[_] = {
    val file = fullClassName.split(‘.‘).last + ".class"
    val in = new FileInputStream(file)
    val bos = new ByteArrayOutputStream
    val bytes = new Array[Byte](4096)
    var done = false
    while (!done) {
      val num = in.read(bytes)
      if (num >= 0) {
        bos.write(bytes, 0, num)
      } else {
        done = true
      }
    }
    val data = bos.toByteArray
    defineClass(fullClassName, data, 0, data.length)
  }
}

protected Class<?> resolveClass(ObjectStreamClass desc)
    throws IOException, ClassNotFoundException
{
    String name = desc.getName();
    try {
        return Class.forName(name, false, latestUserDefinedLoader());
    } catch (ClassNotFoundException ex) {
        Class<?> cl = primClasses.get(name);
        if (cl != null) {
            return cl;
        } else {
            throw ex;
        }
    }
}

def readObjectFromFile(file: String, classLoader: ClassLoader): Object = {
  val fileStream = new FileInputStream(file)
  val ois = new ObjectInputStream(fileStream) {
    override def resolveClass(desc: ObjectStreamClass): Class[_] =
      Class.forName(desc.getName, false, classLoader)
  }
  val obj = ois.readObject()
  ois.close()
  obj
}

val fileClassLoader = new FileClassLoader()
val task = FileSerializer.readObjectFromFile("task.ser", fileClassLoader).asInstanceOf[Task]
task.run()

scala> val n = 2
n: Int = 2

scala> val f = (x: Int) => x * n
f: Int => Int = <function1>

scala> Seq.range(0, 5).map(f)
res0: Seq[Int] = List(0, 2, 4, 6, 8)

scala> f.getClass
res0: Class[_ <: Int => Int] = class $anonfun$1

scala> f.isInstanceOf[Function1[Int, Int]]
res1: Boolean = true

scala> f.isInstanceOf[Serializable]
res2: Boolean = true

val spark = SparkSession.builder().appName("demo").master("local").getOrCreate()
val sc = spark.sparkContext
val data = Array(1, 2, 3, 4, 5)
val distData = sc.parallelize(data)
val sum = distData.map(x => x * 2).sum()
println(sum)  // 30.0

class Operation {
  val n = 2
  def multiply = (x: Int) => x * n
}
...
val sum = distData.map(new Operation().multiply).sum()
...

Exception in thread "main" org.apache.spark.SparkException: Task not serializable
    at org.apache.spark.util.ClosureCleaner$.ensureSerializable(ClosureCleaner.scala:298)
    ...
Caused by: java.io.NotSerializableException: Operation

class Operation {
  def multiply = (x: Int) => x * 2
}
...
val sum = distData.map(new Operation().multiply).sum()
...

private[spark] class JavaSerializationStream(
    out: OutputStream, counterReset: Int, extraDebugInfo: Boolean)
  extends SerializationStream {
  private val objOut = new ObjectOutputStream(out)
  ...
}

private[spark] class JavaDeserializationStream(in: InputStream, loader: ClassLoader)
  extends DeserializationStream {

  private val objIn = new ObjectInputStream(in) {
    override def resolveClass(desc: ObjectStreamClass): Class[_] =
      try {
        Class.forName(desc.getName, false, loader)
      } catch {
        case e: ClassNotFoundException =>
          JavaDeserializationStream.primitiveMappings.getOrElse(desc.getName, throw e)
      }
  }
  ...
}

./bin/spark-submit   --class org.apache.spark.examples.SparkPi   ...   --jars /path/to/dep-libs.jar   /path/to/examples.jar \

/**
 * Adds a JAR dependency for all tasks to be executed on this SparkContext in the future.
 * The `path` passed can be either a local file, a file in HDFS (or other Hadoop-supported
 * filesystems), an HTTP, HTTPS or FTP URI, or local:/path for a file on every worker node.
 */
def addJar(path: String) {
  if (path == null) {
    logWarning("null specified as parameter to addJar")
  } else {
    var key = ""
    if (path.contains("\\")) {
      // For local paths with backslashes on Windows, URI throws an exception
      key = env.rpcEnv.fileServer.addJar(new File(path))
    } else {
      val uri = new URI(path)
      // SPARK-17650: Make sure this is a valid URL before adding it to the list of dependencies
      Utils.validateURL(uri)
      key = uri.getScheme match {
        // A JAR file which exists only on the driver node
        case null | "file" =>
          try {
            env.rpcEnv.fileServer.addJar(new File(uri.getPath))
          } catch {
            case exc: FileNotFoundException =>
              logError(s"Jar not found at $path")
              null
          }
        // A JAR file which exists locally on every worker node
        case "local" =>
          "file:" + uri.getPath
        case _ =>
          path
      }
    }
    if (key != null) {
      val timestamp = System.currentTimeMillis
      if (addedJars.putIfAbsent(key, timestamp).isEmpty) {
        logInfo(s"Added JAR $path at $key with timestamp $timestamp")
        postEnvironmentUpdate()
      }
    }
  }
}

override def run(): Unit = {
  ...

  try {
    val (taskFiles, taskJars, taskProps, taskBytes) =
      Task.deserializeWithDependencies(serializedTask)

    // Must be set before updateDependencies() is called, in case fetching dependencies
    // requires access to properties contained within (e.g. for access control).
    Executor.taskDeserializationProps.set(taskProps)

    updateDependencies(taskFiles, taskJars)
    task = ser.deserialize[Task[Any]](taskBytes, Thread.currentThread.getContextClassLoader)
    ...
  } finally {
    runningTasks.remove(taskId)
  }
}

scala> val f = (x: Int) => x + 1
f: Int => Int = <function1>

scala> val data = Array(1, 2, 3, 4, 5)
data: Array[Int] = Array(1, 2, 3, 4, 5)

scala> val distData = sc.parallelize(data)
distData: org.apache.spark.rdd.RDD[Int] = ParallelCollectionRDD[0] at parallelize at <console>:26

scala> distData.map(f).sum()
res0: Double = 20.0

object Main extends Logging {
  ...
  val rootDir = conf.getOption("spark.repl.classdir").getOrElse(Utils.getLocalDir(conf))
  val outputDir = Utils.createTempDir(root = rootDir, namePrefix = "repl")

  def main(args: Array[String]) {
    doMain(args, new SparkILoop)
  }

  // Visible for testing
  private[repl] def doMain(args: Array[String], _interp: SparkILoop): Unit = {
    interp = _interp
    val jars = Utils.getUserJars(conf, isShell = true).mkString(File.pathSeparator)
    val interpArguments = List(
      "-Yrepl-class-based",
      "-Yrepl-outdir", s"${outputDir.getAbsolutePath}",
      "-classpath", jars
    ) ++ args.toList

    val settings = new GenericRunnerSettings(scalaOptionError)
    settings.processArguments(interpArguments, true)

    if (!hasErrors) {
      interp.process(settings) // Repl starts and goes in loop of R.E.P.L
      Option(sparkContext).map(_.stop)
    }
  }
  ...
}

private val urlClassLoader = createClassLoader()
private val replClassLoader = addReplClassLoaderIfNeeded(urlClassLoader)

/**
 * If the REPL is in use, add another ClassLoader that will read
 * new classes defined by the REPL as the user types code
 */
private def addReplClassLoaderIfNeeded(parent: ClassLoader): ClassLoader = {
  val classUri = conf.get("spark.repl.class.uri", null)
  if (classUri != null) {
    logInfo("Using REPL class URI: " + classUri)
    try {
      val _userClassPathFirst: java.lang.Boolean = userClassPathFirst
      val klass = Utils.classForName("org.apache.spark.repl.ExecutorClassLoader")
        .asInstanceOf[Class[_ <: ClassLoader]]
      val constructor = klass.getConstructor(classOf[SparkConf], classOf[SparkEnv],
        classOf[String], classOf[ClassLoader], classOf[Boolean])
      constructor.newInstance(conf, env, classUri, parent, _userClassPathFirst)
    } catch {
      case _: ClassNotFoundException =>
        logError("Could not find org.apache.spark.repl.ExecutorClassLoader on classpath!")
        System.exit(1)
        null
    }
  } else {
    parent
  }
}

override def run(): Unit = {
  ...
  Thread.currentThread.setContextClassLoader(replClassLoader)
  val ser = env.closureSerializer.newInstance()
  ...
}