Esempio n. 1
2
  void removeTaskEntry(String taskid) {
    // taskid --> tracker
    String tracker = (String) taskidToTrackerMap.remove(taskid);

    // tracker --> taskid
    TreeSet trackerSet = (TreeSet) trackerToTaskMap.get(tracker);
    if (trackerSet != null) {
      trackerSet.remove(taskid);
    }

    // taskid --> TIP
    taskidToTIPMap.remove(taskid);
  }
Esempio n. 2
1
  /**
   * We lost the task tracker! All task-tracker structures have already been updated. Just process
   * the contained tasks and any jobs that might be affected.
   */
  void lostTaskTracker(String trackerName) {
    LOG.info("Lost tracker '" + trackerName + "'");
    TreeSet lostTasks = (TreeSet) trackerToTaskMap.get(trackerName);
    trackerToTaskMap.remove(trackerName);

    if (lostTasks != null) {
      for (Iterator it = lostTasks.iterator(); it.hasNext(); ) {
        String taskId = (String) it.next();
        TaskInProgress tip = (TaskInProgress) taskidToTIPMap.get(taskId);

        // Tell the job to fail the relevant task
        JobInProgress job = tip.getJob();
        job.failedTask(tip, taskId, trackerName);
      }
    }
  }
Esempio n. 3
0
 public synchronized JobStatus getJobStatus(String jobid) {
   JobInProgress job = (JobInProgress) jobs.get(jobid);
   if (job != null) {
     return job.getStatus();
   } else {
     return null;
   }
 }
Esempio n. 4
0
 public synchronized JobProfile getJobProfile(String jobid) {
   JobInProgress job = (JobInProgress) jobs.get(jobid);
   if (job != null) {
     return job.getProfile();
   } else {
     return null;
   }
 }
Esempio n. 5
0
 /**
  * A tracker wants to know if any of its Tasks have been closed (because the job completed,
  * whether successfully or not)
  */
 public synchronized String pollForTaskWithClosedJob(String taskTracker) {
   TreeSet taskIds = (TreeSet) trackerToTaskMap.get(taskTracker);
   if (taskIds != null) {
     for (Iterator it = taskIds.iterator(); it.hasNext(); ) {
       String taskId = (String) it.next();
       TaskInProgress tip = (TaskInProgress) taskidToTIPMap.get(taskId);
       if (tip.shouldCloseForClosedJob(taskId)) {
         //
         // This is how the JobTracker ends a task at the TaskTracker.
         // It may be successfully completed, or may be killed in
         // mid-execution.
         //
         return taskId;
       }
     }
   }
   return null;
 }
Esempio n. 6
0
  /**
   * A TaskTracker wants to know the physical locations of completed, but not yet closed, tasks.
   * This exists so the reduce task thread can locate map task outputs.
   */
  public synchronized MapOutputLocation[] locateMapOutputs(
      String taskId, String[][] mapTasksNeeded) {
    ArrayList v = new ArrayList();
    for (int i = 0; i < mapTasksNeeded.length; i++) {
      for (int j = 0; j < mapTasksNeeded[i].length; j++) {
        TaskInProgress tip = (TaskInProgress) taskidToTIPMap.get(mapTasksNeeded[i][j]);
        if (tip != null && tip.isComplete(mapTasksNeeded[i][j])) {
          String trackerId = (String) taskidToTrackerMap.get(mapTasksNeeded[i][j]);
          TaskTrackerStatus tracker;
          synchronized (taskTrackers) {
            tracker = (TaskTrackerStatus) taskTrackers.get(trackerId);
          }
          v.add(new MapOutputLocation(mapTasksNeeded[i][j], tracker.getHost(), tracker.getPort()));
          break;
        }
      }
    }
    // randomly shuffle results to load-balance map output requests
    Collections.shuffle(v);

    return (MapOutputLocation[]) v.toArray(new MapOutputLocation[v.size()]);
  }
Esempio n. 7
0
 /**
  * Update the last recorded status for the given task tracker. It assumes that the taskTrackers
  * are locked on entry.
  *
  * @author Owen O'Malley
  * @param trackerName The name of the tracker
  * @param status The new status for the task tracker
  * @return Was an old status found?
  */
 private boolean updateTaskTrackerStatus(String trackerName, TaskTrackerStatus status) {
   TaskTrackerStatus oldStatus = (TaskTrackerStatus) taskTrackers.get(trackerName);
   if (oldStatus != null) {
     totalMaps -= oldStatus.countMapTasks();
     totalReduces -= oldStatus.countReduceTasks();
     if (status == null) {
       taskTrackers.remove(trackerName);
     }
   }
   if (status != null) {
     totalMaps += status.countMapTasks();
     totalReduces += status.countReduceTasks();
     taskTrackers.put(trackerName, status);
   }
   return oldStatus != null;
 }
Esempio n. 8
0
 public synchronized TaskReport[] getReduceTaskReports(String jobid) {
   JobInProgress job = (JobInProgress) jobs.get(jobid);
   if (job == null) {
     return new TaskReport[0];
   } else {
     Vector reports = new Vector();
     Vector completeReduceTasks = job.reportTasksInProgress(false, true);
     for (Iterator it = completeReduceTasks.iterator(); it.hasNext(); ) {
       TaskInProgress tip = (TaskInProgress) it.next();
       reports.add(tip.generateSingleReport());
     }
     Vector incompleteReduceTasks = job.reportTasksInProgress(false, false);
     for (Iterator it = incompleteReduceTasks.iterator(); it.hasNext(); ) {
       TaskInProgress tip = (TaskInProgress) it.next();
       reports.add(tip.generateSingleReport());
     }
     return (TaskReport[]) reports.toArray(new TaskReport[reports.size()]);
   }
 }
Esempio n. 9
0
  /**
   * Accept and process a new TaskTracker profile. We might have known about the TaskTracker
   * previously, or it might be brand-new. All task-tracker structures have already been updated.
   * Just process the contained tasks and any jobs that might be affected.
   */
  void updateTaskStatuses(TaskTrackerStatus status) {
    for (Iterator it = status.taskReports(); it.hasNext(); ) {
      TaskStatus report = (TaskStatus) it.next();
      TaskInProgress tip = (TaskInProgress) taskidToTIPMap.get(report.getTaskId());
      if (tip == null) {
        LOG.info(
            "Serious problem.  While updating status, cannot find taskid " + report.getTaskId());
      } else {
        JobInProgress job = tip.getJob();
        job.updateTaskStatus(tip, report);

        if (report.getRunState() == TaskStatus.SUCCEEDED) {
          job.completedTask(tip, report.getTaskId());
        } else if (report.getRunState() == TaskStatus.FAILED) {
          // Tell the job to fail the relevant task
          job.failedTask(tip, report.getTaskId(), status.getTrackerName());
        }
      }
    }
  }
Esempio n. 10
0
  ///////////////////////////////////////////////////////
  // Maintain lookup tables; called by JobInProgress
  // and TaskInProgress
  ///////////////////////////////////////////////////////
  void createTaskEntry(String taskid, String taskTracker, TaskInProgress tip) {
    LOG.info(
        "Adding task '"
            + taskid
            + "' to tip "
            + tip.getTIPId()
            + ", for tracker '"
            + taskTracker
            + "'");

    // taskid --> tracker
    taskidToTrackerMap.put(taskid, taskTracker);

    // tracker --> taskid
    TreeSet taskset = (TreeSet) trackerToTaskMap.get(taskTracker);
    if (taskset == null) {
      taskset = new TreeSet();
      trackerToTaskMap.put(taskTracker, taskset);
    }
    taskset.add(taskid);

    // taskid --> TIP
    taskidToTIPMap.put(taskid, tip);
  }
Esempio n. 11
0
 ///////////////////////////////////////////////////////////////
 // JobTracker methods
 ///////////////////////////////////////////////////////////////
 public JobInProgress getJob(String jobid) {
   return (JobInProgress) jobs.get(jobid);
 }
Esempio n. 12
0
 public synchronized void killJob(String jobid) {
   JobInProgress job = (JobInProgress) jobs.get(jobid);
   job.kill();
 }
Esempio n. 13
0
  /**
   * A tracker wants to know if there's a Task to run. Returns a task we'd like the TaskTracker to
   * execute right now.
   *
   * <p>Eventually this function should compute load on the various TaskTrackers, and incorporate
   * knowledge of DFS file placement. But for right now, it just grabs a single item out of the
   * pending task list and hands it back.
   */
  public synchronized Task pollForNewTask(String taskTracker) {
    //
    // Compute average map and reduce task numbers across pool
    //
    int avgMaps = 0;
    int avgReduces = 0;
    int numTaskTrackers;
    TaskTrackerStatus tts;
    synchronized (taskTrackers) {
      numTaskTrackers = taskTrackers.size();
      tts = (TaskTrackerStatus) taskTrackers.get(taskTracker);
    }
    if (numTaskTrackers > 0) {
      avgMaps = totalMaps / numTaskTrackers;
      avgReduces = totalReduces / numTaskTrackers;
    }
    int totalCapacity = numTaskTrackers * maxCurrentTasks;
    //
    // Get map + reduce counts for the current tracker.
    //
    if (tts == null) {
      LOG.warning("Unknown task tracker polling; ignoring: " + taskTracker);
      return null;
    }

    int numMaps = tts.countMapTasks();
    int numReduces = tts.countReduceTasks();

    //
    // In the below steps, we allocate first a map task (if appropriate),
    // and then a reduce task if appropriate.  We go through all jobs
    // in order of job arrival; jobs only get serviced if their
    // predecessors are serviced, too.
    //

    //
    // We hand a task to the current taskTracker if the given machine
    // has a workload that's equal to or less than the averageMaps
    // +/- TASK_ALLOC_EPSILON.  (That epsilon is in place in case
    // there is an odd machine that is failing for some reason but
    // has not yet been removed from the pool, making capacity seem
    // larger than it really is.)
    //
    synchronized (jobsByArrival) {
      if ((numMaps < maxCurrentTasks) && (numMaps <= (avgMaps + TASK_ALLOC_EPSILON))) {

        int totalNeededMaps = 0;
        for (Iterator it = jobsByArrival.iterator(); it.hasNext(); ) {
          JobInProgress job = (JobInProgress) it.next();
          if (job.getStatus().getRunState() != JobStatus.RUNNING) {
            continue;
          }

          Task t = job.obtainNewMapTask(taskTracker, tts);
          if (t != null) {
            return t;
          }

          //
          // Beyond the highest-priority task, reserve a little
          // room for failures and speculative executions; don't
          // schedule tasks to the hilt.
          //
          totalNeededMaps += job.desiredMaps();
          double padding = 0;
          if (totalCapacity > MIN_SLOTS_FOR_PADDING) {
            padding = Math.min(maxCurrentTasks, totalNeededMaps * PAD_FRACTION);
          }
          if (totalNeededMaps + padding >= totalCapacity) {
            break;
          }
        }
      }

      //
      // Same thing, but for reduce tasks
      //
      if ((numReduces < maxCurrentTasks) && (numReduces <= (avgReduces + TASK_ALLOC_EPSILON))) {

        int totalNeededReduces = 0;
        for (Iterator it = jobsByArrival.iterator(); it.hasNext(); ) {
          JobInProgress job = (JobInProgress) it.next();
          if (job.getStatus().getRunState() != JobStatus.RUNNING) {
            continue;
          }

          Task t = job.obtainNewReduceTask(taskTracker, tts);
          if (t != null) {
            return t;
          }

          //
          // Beyond the highest-priority task, reserve a little
          // room for failures and speculative executions; don't
          // schedule tasks to the hilt.
          //
          totalNeededReduces += job.desiredReduces();
          double padding = 0;
          if (totalCapacity > MIN_SLOTS_FOR_PADDING) {
            padding = Math.min(maxCurrentTasks, totalNeededReduces * PAD_FRACTION);
          }
          if (totalNeededReduces + padding >= totalCapacity) {
            break;
          }
        }
      }
    }
    return null;
  }
Esempio n. 14
0
 public TaskTrackerStatus getTaskTracker(String trackerID) {
   synchronized (taskTrackers) {
     return (TaskTrackerStatus) taskTrackers.get(trackerID);
   }
 }