【Hadoop代码笔记】Hadoop作业提交之Child启动map任务

一、概要描述

在上篇博文描述了TaskTracker启动一个独立的java进程来执行Map或Reduce任务。在本篇和下篇博文中我们会关注启动的那个入口是org.apache.hadoop.mapred.Child的这个Java进程是如何执行用户定义的map或Reduce任务的。

接上篇文章，TaskRunner线程执行中，会构造一个java –D** Child address port tasked这样第一个java命令，单独启动一个java进程。在Child的main函数中通过TaskUmbilicalProtocol协议，从TaskTracker获得需要执行的Task，并调用Task的run方法来执行，而Task的run方法会通过java反射机制构造Mapper，InputFormat，mapperContext，然后调用构造的mapper的run方法执行mapper操作。

二、 流程描述  

1.Child类根据前面输入的三个参数，即tasktracher的地址、端口、taskid。通过TaskUmbilicalProtocol协议，从TaskTracker获得需要执行的Task，在Child的main函数中调用执行。

2. 在Chilld中，执行Task的run方法。Task 的run方法。是真正执行用户定义的map或者reduce任务的入口，通过TaskUmbilicalProtocol向tasktracker上报执行进度。

3. 在MapTask的run中执行runMapper方法来调用mapper定义的方法。

4. 在runNewMapper方法中构造mapper实例和mapper执行的配置信息。并执行mapper.run方法来调用到用户定义的mapper的方法。

5. mapper的run方法中，从输入数据中逐一取出调用map方法来处理每一条数据

6. mapper的map方法是真正用户定义的处理数据的类。也是用户唯一需要定义的方法。

三、代码详细

1. Child的main方法每个task进程都会被在单独的进程中执行，这个方法就是这些进程的入口方法。观察下载在这个方法中做了哪些事情？

1)从传入的参数中获得tasktracker的地址、从传入的参数中获得tasktracker的地址
2) 根据获取的taskTracker的地址和端口通过RPC方式和tasktracker通信，umbilical是作为tasktracker的代理来执行操作。
3) 根据JvmId从taskTracker查询获取到JvmTask
4) 执行任务

public static void main(String[] args) throws Throwable {
    LOG.debug("Child starting");

    JobConf defaultConf = new JobConf();

    //从传入的参数中获得taskTracker的地址
    String host = args[0];
  //从传入的参数中获得taskTracker的响应请求的端口。
    int port = Integer.parseInt(args[1]);
    InetSocketAddress address = new InetSocketAddress(host, port);
    final TaskAttemptID firstTaskid = TaskAttemptID.forName(args[2]);
    final int SLEEP_LONGER_COUNT = 5;
    int jvmIdInt = Integer.parseInt(args[3]);
    JVMId jvmId = new JVMId(firstTaskid.getJobID(),firstTaskid.isMap(),jvmIdInt);

    //通过RPC方式和tasktracker通信，umbilical是作为tasktracker的代理来执行操作。
    TaskUmbilicalProtocol umbilical =
      (TaskUmbilicalProtocol)RPC.getProxy(TaskUmbilicalProtocol.class,
          TaskUmbilicalProtocol.versionID,
          address,
          defaultConf);
    int numTasksToExecute = -1; //-1 signifies "no limit"
    int numTasksExecuted = 0;
    //for the memory management, a PID file is written and the PID file
    //is written once per JVM. We simply symlink the file on a per task
    //basis later (see below). Long term, we should change the Memory
    //manager to use JVMId instead of TaskAttemptId
    Path srcPidPath = null;
    Path dstPidPath = null;
    int idleLoopCount = 0;
    Task task = null;
    try {
      while (true) {
        taskid = null;
        //根据JvmId从taskTracker查询获取到JvmTask
        JvmTask myTask = umbilical.getTask(jvmId);
        if (myTask.shouldDie()) {
          break;
        } else {
          if (myTask.getTask() == null) {
            taskid = null;
            if (++idleLoopCount >= SLEEP_LONGER_COUNT) {
              //we sleep for a bigger interval when we don't receive
              //tasks for a while
              Thread.sleep(1500);
            } else {
              Thread.sleep(500);
            }
            continue;
          }
        }
        idleLoopCount = 0;
        task = myTask.getTask();
        taskid = task.getTaskID();
        isCleanup = task.isTaskCleanupTask();
        // reset the statistics for the task
        FileSystem.clearStatistics();
        TaskLog.syncLogs(firstTaskid, taskid, isCleanup);
        JobConf job = new JobConf(task.getJobFile());
        if (job.getBoolean("task.memory.mgmt.enabled", false)) {
          if (srcPidPath == null) {
            srcPidPath = new Path(task.getPidFile());
          }
          //since the JVM is running multiple tasks potentially, we need
          //to do symlink stuff only for the subsequent tasks
          if (!taskid.equals(firstTaskid)) {
            dstPidPath = new Path(task.getPidFile());
            FileUtil.symLink(srcPidPath.toUri().getPath(),
                dstPidPath.toUri().getPath());
          }
        }
        //setupWorkDir actually sets up the symlinks for the distributed
        //cache. After a task exits we wipe the workdir clean, and hence
        //the symlinks have to be rebuilt.
        TaskRunner.setupWorkDir(job);

        numTasksToExecute = job.getNumTasksToExecutePerJvm();
        assert(numTasksToExecute != 0);
        TaskLog.cleanup(job.getInt("mapred.userlog.retain.hours", 24));
        task.setConf(job);
        defaultConf.addResource(new Path(task.getJobFile()));
        // use job-specified working directory
        FileSystem.get(job).setWorkingDirectory(job.getWorkingDirectory());
        try {
          //执行任务
          task.run(job, umbilical);             // run the task
        } finally {
          TaskLog.syncLogs(firstTaskid, taskid, isCleanup);
          if (!taskid.equals(firstTaskid) &&
              job.getBoolean("task.memory.mgmt.enabled", false)) {
            // delete the pid-file's symlink
            new File(dstPidPath.toUri().getPath()).delete();
          }
        }
        if (numTasksToExecute > 0 && ++numTasksExecuted == numTasksToExecute) {
          break;
        }
      }
    } catch (FSError e) {
      LOG.fatal("FSError from child", e);
      umbilical.fsError(taskid, e.getMessage());
    } catch (Throwable throwable) {
      LOG.warn("Error running child", throwable);
      try {
        if (task != null) {
          // do cleanup for the task
          task.taskCleanup(umbilical);
        }
      } catch (Throwable th) {
        LOG.info("Error cleaning up" + th);
      }
      // Report back any failures, for diagnostic purposes
      ByteArrayOutputStream baos = new ByteArrayOutputStream();
      throwable.printStackTrace(new PrintStream(baos));
      if (taskid != null) {
        umbilical.reportDiagnosticInfo(taskid, baos.toString());
      }
    } finally {
      RPC.stopProxy(umbilical);

    }
  }

2. TaskTracker 的getTask方法。TaskTracker实现了TaskUmbilicalProtocol接扣。getTask是该接口定义的一个方法。是子进程Child调用的根据jvmId获取task。

 public synchronized JvmTask getTask(JVMId jvmId)
  throws IOException {
  TaskInProgress tip = jvmManager.getTaskForJvm(jvmId);
    if (tip == null) {
      return new JvmTask(null, false);
    }
    if (tasks.get(tip.getTask().getTaskID()) != null) { //is task still present
      LOG.info("JVM with ID: " + jvmId + " given task: " +
          tip.getTask().getTaskID());
      return new JvmTask(tip.getTask(), false);
    } else {
      LOG.info("Killing JVM with ID: " + jvmId + " since scheduled task: " +
          tip.getTask().getTaskID() + " is " + tip.taskStatus.getRunState());
      return new JvmTask(null, true);
}

3.Task 的run方法。因为map和reduce的执行逻辑大不相同，先看下MapTask中该方法的实现。是真正执行用户定义的map或者reduce任务的入口，通过TaskUmbilicalProtocol向tasktracker上报执行进度。开启线程向TaskTracker上报进度，根据task的不同动作要求执行不同的方法，如jobClean，jobsetup，taskCleanup。对于部分的了解可以产看taskTracker获取Task文章中的 JobTracker的 heartbeat方法处的详细解释。

public void run(final JobConf job, final TaskUmbilicalProtocol umbilical)
    throws IOException, ClassNotFoundException, InterruptedException {

    // 开启线程向TaskTracker上报进度
    TaskReporter reporter = new TaskReporter(getProgress(), umbilical);
    reporter.startCommunicationThread();
    boolean useNewApi = job.getUseNewMapper();
    initialize(job, getJobID(), reporter, useNewApi);

    // 根据task的不同动作要求执行不同的方法，如jobClean，jobsetup，taskCleanup
    if (jobCleanup) {
      runJobCleanupTask(umbilical, reporter);
      return;
    }
    if (jobSetup) {
      runJobSetupTask(umbilical, reporter);
      return;
    }
    if (taskCleanup) {
      runTaskCleanupTask(umbilical, reporter);
      return;
    }

    if (useNewApi) {
      runNewMapper(job, split, umbilical, reporter);
    } else {
      runOldMapper(job, split, umbilical, reporter);
    }
    done(umbilical, reporter);
  }

4. TaskReporter的run方法。定时向父进程TaskTracker上报状态和进度。

 public void run() {
      final int MAX_RETRIES = 3;
      int remainingRetries = MAX_RETRIES;
      // get current flag value and reset it as well
      boolean sendProgress = resetProgressFlag();
      while (!taskDone.get()) {
        try {
          boolean taskFound = true; // whether TT knows about this task
          // sleep for a bit
          try {
            Thread.sleep(PROGRESS_INTERVAL);
          }
                    break;
          }

          if (sendProgress) {
            // we need to send progress update
            updateCounters();
            taskStatus.statusUpdate(taskProgress.get(),
                                    taskProgress.toString(),
                                    counters);
            taskFound = umbilical.statusUpdate(taskId, taskStatus);
            taskStatus.clearStatus();
          }
          else {
            // send ping
            taskFound = umbilical.ping(taskId);
          }

          // if Task Tracker is not aware of our task ID (probably because it died and
          // came back up), kill ourselves
          if (!taskFound) {
            LOG.warn("Parent died.  Exiting "+taskId);
            System.exit(66);
          }

          sendProgress = resetProgressFlag();
          remainingRetries = MAX_RETRIES;
        }
        catch (Throwable t) {
                 }
        }
      }
    }

5. Task 的Initialize方法初始化后续要执行的几个重要变量。包括JobContext OutputFormat OutputCommitter等，这些都是后续执行中要用到的属性实例。

 public void initialize(JobConf job, JobID id,
                         Reporter reporter,
                         boolean useNewApi) throws IOException,
                                                   ClassNotFoundException,
                                                   InterruptedException {
    jobContext = new JobContext(job, id, reporter);
    taskContext = new TaskAttemptContext(job, taskId, reporter);
    if (getState() == TaskStatus.State.UNASSIGNED) {
      setState(TaskStatus.State.RUNNING);
    }
    if (useNewApi) {
      LOG.debug("using new api for output committer");
      outputFormat =
        ReflectionUtils.newInstance(taskContext.getOutputFormatClass(), job);
      committer = outputFormat.getOutputCommitter(taskContext);
    } else {
      committer = conf.getOutputCommitter();
    }
    Path outputPath = FileOutputFormat.getOutputPath(conf);
    if (outputPath != null) {
      if ((committer instanceof FileOutputCommitter)) {
        FileOutputFormat.setWorkOutputPath(conf,
          ((FileOutputCommitter)committer).getTempTaskOutputPath(taskContext));
      } else {
        FileOutputFormat.setWorkOutputPath(conf, outputPath);
      }
    }
    committer.setupTask(taskContext);
  }

6. Task的 runJobCleanupTask方法。即如果在Task是jobCleanup，则调用OutputCommitter删除输出文件

protected void runJobCleanupTask(TaskUmbilicalProtocol umbilical,
                               TaskReporter reporter
                              ) throws IOException, InterruptedException {
    // set phase for this task
    setPhase(TaskStatus.Phase.CLEANUP);
    getProgress().setStatus("cleanup");
    statusUpdate(umbilical);
    // do the cleanup
    committer.cleanupJob(jobContext);
    done(umbilical, reporter);
  }

7.Task的runJobSetupTask。如果Task是setupTask，则调用OutputCommitter，如创建Task要执行的根目录。

protected void runJobSetupTask(TaskUmbilicalProtocol umbilical,
                             TaskReporter reporter
                             ) throws IOException, InterruptedException {
    // do the setup
    getProgress().setStatus("setup");
    committer.setupJob(jobContext);
    done(umbilical, reporter);
  }

8. Task的runTaskCleanupTask。如果Task是taskCleanup，则调用taskCleanup 方法。最终OutputCommitter方法删除task的工作目录。

  protected void runTaskCleanupTask(TaskUmbilicalProtocol umbilical,
                                TaskReporter reporter)
  throws IOException, InterruptedException {
    taskCleanup(umbilical);
    done(umbilical, reporter);
  }

9.MapTask的runNewMapper方法是我们要重点关注的方法，是真正执行用户定义的map的方法。

1)根据传入的jobconf构造一个context，包含了job相关的所有配置信息，如后面用到的mapper、inputformat等。
2)根据配置的mapper类创建一个Mapper实例
3)根据配置的inputformat创建一个InputFormat实例。
4)重新够构建InputSplit
5)创建RecordReader，其实使用的是适配器模式适配了inputFormat的Reader。
6)构造输出RecordWriter。当没有Reducer时，output是配置的outputFormat的RecordWriter，即直接写输出。如果ruducer数量不为0，则构造一个NewOutputCollector
7)构造Mapper.Context，封装了刚才配置的所有信息，在map执行时候时候使用。
8)调用mapper的run方法来执行map动作。

 @SuppressWarnings("unchecked")
  private <INKEY,INVALUE,OUTKEY,OUTVALUE>
  void runNewMapper(final JobConf job,
                    final BytesWritable rawSplit,
                    final TaskUmbilicalProtocol umbilical,
                    TaskReporter reporter
                    ) throws IOException, ClassNotFoundException,
                             InterruptedException {
    // 1. 根据传入的jobconf构造一个context，包含了job相关的所有配置信息，如后面用到的mapper、inputformat等。
    org.apache.hadoop.mapreduce.TaskAttemptContext taskContext =
      new org.apache.hadoop.mapreduce.TaskAttemptContext(job, getTaskID());
    // 2. 根据配置的mapper类创建一个Mapper实例
    org.apache.hadoop.mapreduce.Mapper<INKEY,INVALUE,OUTKEY,OUTVALUE> mapper =
      (org.apache.hadoop.mapreduce.Mapper<INKEY,INVALUE,OUTKEY,OUTVALUE>)
        ReflectionUtils.newInstance(taskContext.getMapperClass(), job);
    // 根据配置的input format创建一个InputFormat实例。
    org.apache.hadoop.mapreduce.InputFormat<INKEY,INVALUE> inputFormat =
      (org.apache.hadoop.mapreduce.InputFormat<INKEY,INVALUE>)
        ReflectionUtils.newInstance(taskContext.getInputFormatClass(), job);
    // 4.重新够构建InputSplit
    org.apache.hadoop.mapreduce.InputSplit split = null;
    DataInputBuffer splitBuffer = new DataInputBuffer();
    splitBuffer.reset(rawSplit.getBytes(), 0, rawSplit.getLength());
    SerializationFactory factory = new SerializationFactory(job);
    Deserializer<? extends org.apache.hadoop.mapreduce.InputSplit>
      deserializer =
        (Deserializer<? extends org.apache.hadoop.mapreduce.InputSplit>)
        factory.getDeserializer(job.getClassByName(splitClass));
    deserializer.open(splitBuffer);
    split = deserializer.deserialize(null);

    //5. 创建RecordReader，其实使用的是适配器模式适配了inputFormat的Reader。
    org.apache.hadoop.mapreduce.RecordReader<INKEY,INVALUE> input =
      new NewTrackingRecordReader<INKEY,INVALUE>
          (inputFormat.createRecordReader(split, taskContext), reporter);

    job.setBoolean("mapred.skip.on", isSkipping());
    org.apache.hadoop.mapreduce.RecordWriter output = null;
    org.apache.hadoop.mapreduce.Mapper<INKEY,INVALUE,OUTKEY,OUTVALUE>.Context
         mapperContext = null;
    try {
      Constructor<org.apache.hadoop.mapreduce.Mapper.Context> contextConstructor =
        org.apache.hadoop.mapreduce.Mapper.Context.class.getConstructor
        (new Class[]{org.apache.hadoop.mapreduce.Mapper.class,
                     Configuration.class,
                     org.apache.hadoop.mapreduce.TaskAttemptID.class,
                     org.apache.hadoop.mapreduce.RecordReader.class,
                     org.apache.hadoop.mapreduce.RecordWriter.class,
                     org.apache.hadoop.mapreduce.OutputCommitter.class,
                     org.apache.hadoop.mapreduce.StatusReporter.class,
                     org.apache.hadoop.mapreduce.InputSplit.class});

      //6. 构造输出RecordWriter。当没有Reducer时，output是配置的outputFormat的RecordWriter，即直接写输出。如果ruducer数量不为0，则构造一个NewOutputCollector
      if (job.getNumReduceTasks() == 0) {
        output = outputFormat.getRecordWriter(taskContext);
      } else {
        output = new NewOutputCollector(job, umbilical, reporter);
      }

      //7.构造Mapper.Context，封装了刚才配置的所有信息，在map执行时候时候使用。
      mapperContext = contextConstructor.newInstance(mapper, job, getTaskID(),
                                                     input, output, committer,
                                                     reporter, split);

      input.initialize(split, mapperContext);
      //8. 调用mapper的run方法来执行map动作。
      mapper.run(mapperContext);
      input.close();
      output.close(mapperContext);
    } catch (NoSuchMethodException e) {
      throw new IOException("Can't find Context constructor", e);
    } catch (InstantiationException e) {
      throw new IOException("Can't create Context", e);
    } catch (InvocationTargetException e) {
      throw new IOException("Can't invoke Context constructor", e);
    } catch (IllegalAccessException e) {
      throw new IOException("Can't invoke Context constructor", e);
    }
  }

10.Mapper的run方法。即对每一个输入的记录执行map方法。一般不会改变，就是拿出输入记录逐条执行map方法。除非要改变记录的执行方式，（如MultithreadedMapper需要多线程来执行），一般该方法不用override。

public void run(Context context) throws IOException, InterruptedException {
    setup(context);
    while (context.nextKeyValue()) {
      map(context.getCurrentKey(), context.getCurrentValue(), context);
    }
    cleanup(context);
  }
}

11.Mapper的map方法。即对每一个输入的记录执行map方法。这个只是默然的map执行方法，把输入不变的输出即可。用户定义的mapper就是override这个方法来按照自己定义的逻辑来处理数据。

protected void map(KEYIN key, VALUEIN value,
                     Context context) throws IOException, InterruptedException {
    context.write((KEYOUT) key, (VALUEOUT) value);
  }

完。

为了转载内容的一致性、可追溯性和保证及时更新纠错，转载时请注明来自：http://www.cnblogs.com/douba/p/hadoop_mapreduce_tasktracker_child_map.html。谢谢！