/** Process incoming heartbeat messages from the task trackers. */
public synchronized int emitHeartbeat(TaskTrackerStatus trackerStatus, boolean initialContact) {
String trackerName = trackerStatus.getTrackerName();
trackerStatus.setLastSeen(System.currentTimeMillis());
synchronized (taskTrackers) {
synchronized (trackerExpiryQueue) {
boolean seenBefore = updateTaskTrackerStatus(trackerName, trackerStatus);
if (initialContact) {
// If it's first contact, then clear out any state hanging around
if (seenBefore) {
lostTaskTracker(trackerName);
}
} else {
// If not first contact, there should be some record of the tracker
if (!seenBefore) {
return InterTrackerProtocol.UNKNOWN_TASKTRACKER;
}
}
if (initialContact) {
trackerExpiryQueue.add(trackerStatus);
}
}
}
updateTaskStatuses(trackerStatus);
// LOG.info("Got heartbeat from "+trackerName);
return InterTrackerProtocol.TRACKERS_OK;
}
public int compare(Object o1, Object o2) {
TaskTrackerStatus p1 = (TaskTrackerStatus) o1;
TaskTrackerStatus p2 = (TaskTrackerStatus) o2;
if (p1.getLastSeen() < p2.getLastSeen()) {
return -1;
} else if (p1.getLastSeen() > p2.getLastSeen()) {
return 1;
} else {
return (p1.getTrackerName().compareTo(p2.getTrackerName()));
}
}
/**
* The run method lives for the life of the JobTracker, and removes TaskTrackers that have not
* checked in for some time.
*/
public void run() {
while (shouldRun) {
//
// Thread runs periodically to check whether trackers should be expired.
// The sleep interval must be no more than half the maximum expiry time
// for a task tracker.
//
try {
Thread.sleep(TASKTRACKER_EXPIRY_INTERVAL / 3);
} catch (InterruptedException ie) {
}
//
// Loop through all expired items in the queue
//
synchronized (taskTrackers) {
synchronized (trackerExpiryQueue) {
long now = System.currentTimeMillis();
TaskTrackerStatus leastRecent = null;
while ((trackerExpiryQueue.size() > 0)
&& ((leastRecent = (TaskTrackerStatus) trackerExpiryQueue.first()) != null)
&& (now - leastRecent.getLastSeen() > TASKTRACKER_EXPIRY_INTERVAL)) {
// Remove profile from head of queue
trackerExpiryQueue.remove(leastRecent);
String trackerName = leastRecent.getTrackerName();
// Figure out if last-seen time should be updated, or if tracker is dead
TaskTrackerStatus newProfile =
(TaskTrackerStatus) taskTrackers.get(leastRecent.getTrackerName());
// Items might leave the taskTracker set through other means; the
// status stored in 'taskTrackers' might be null, which means the
// tracker has already been destroyed.
if (newProfile != null) {
if (now - newProfile.getLastSeen() > TASKTRACKER_EXPIRY_INTERVAL) {
// Remove completely
updateTaskTrackerStatus(trackerName, null);
lostTaskTracker(leastRecent.getTrackerName());
} else {
// Update time by inserting latest profile
trackerExpiryQueue.add(newProfile);
}
}
}
}
}
}
}
/**
* 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());
}
}
}
}
@Override
public synchronized List<Task> assignTasks(TaskTracker taskTracker) throws IOException {
TaskTrackerStatus taskTrackerStatus = taskTracker.getStatus();
ClusterStatus clusterStatus = taskTrackerManager.getClusterStatus();
final int numTaskTrackers = clusterStatus.getTaskTrackers();
final int clusterMapCapacity = clusterStatus.getMaxMapTasks();
final int clusterReduceCapacity = clusterStatus.getMaxReduceTasks();
Collection<JobInProgress> jobQueue = jobQueueJobInProgressListener.getJobQueue();
//
// Get map + reduce counts for the current tracker.
//
final int trackerMapCapacity = taskTrackerStatus.getMaxMapSlots();
final int trackerReduceCapacity = taskTrackerStatus.getMaxReduceSlots();
final int trackerRunningMaps = taskTrackerStatus.countMapTasks();
final int trackerRunningReduces = taskTrackerStatus.countReduceTasks();
// Assigned tasks
List<Task> assignedTasks = new ArrayList<Task>();
//
// Compute (running + pending) map and reduce task numbers across pool
//
int remainingReduceLoad = 0;
int remainingMapLoad = 0;
synchronized (jobQueue) {
for (JobInProgress job : jobQueue) {
if (job.getStatus().getRunState() == JobStatus.RUNNING) {
remainingMapLoad += (job.desiredMaps() - job.finishedMaps());
if (job.scheduleReduces()) {
remainingReduceLoad += (job.desiredReduces() - job.finishedReduces());
}
}
}
}
// Compute the 'load factor' for maps and reduces
double mapLoadFactor = 0.0;
if (clusterMapCapacity > 0) {
mapLoadFactor = (double) remainingMapLoad / clusterMapCapacity;
}
double reduceLoadFactor = 0.0;
if (clusterReduceCapacity > 0) {
reduceLoadFactor = (double) remainingReduceLoad / clusterReduceCapacity;
}
//
// In the below steps, we allocate first map tasks (if appropriate),
// and then reduce tasks if appropriate. We go through all jobs
// in order of job arrival; jobs only get serviced if their
// predecessors are serviced, too.
//
//
// We assign tasks to the current taskTracker if the given machine
// has a workload that's less than the maximum load of that kind of
// task.
// However, if the cluster is close to getting loaded i.e. we don't
// have enough _padding_ for speculative executions etc., we only
// schedule the "highest priority" task i.e. the task from the job
// with the highest priority.
//
final int trackerCurrentMapCapacity =
Math.min((int) Math.ceil(mapLoadFactor * trackerMapCapacity), trackerMapCapacity);
int availableMapSlots = trackerCurrentMapCapacity - trackerRunningMaps;
boolean exceededMapPadding = false;
if (availableMapSlots > 0) {
exceededMapPadding = exceededPadding(true, clusterStatus, trackerMapCapacity);
}
int numLocalMaps = 0;
int numNonLocalMaps = 0;
boolean newIterationJob = false;
scheduleMaps:
for (int i = 0; i < availableMapSlots; ++i) {
synchronized (jobQueue) {
for (JobInProgress job : jobQueue) {
if (job.getStatus().getRunState() != JobStatus.RUNNING) {
continue;
}
if (job.getJobConf().isIterative()) {
String iterativeAppID = job.getJobConf().getIterativeAlgorithmID();
if (iterativeAppID.equals("none")) {
throw new IOException("please specify the iteration ID!");
}
String jointype = job.getJobConf().get("mapred.iterative.jointype");
// prepare the iterationid map and jobtask map
if (!this.tracker_mtask_map.containsKey(iterativeAppID)) {
// a new iterative algorithm
Map<String, LinkedList<Integer>> new_tracker_task_map =
new HashMap<String, LinkedList<Integer>>();
this.tracker_mtask_map.put(iterativeAppID, new_tracker_task_map);
Map<String, LinkedList<Integer>> new_tracker_rtask_map =
new HashMap<String, LinkedList<Integer>>();
this.tracker_rtask_map.put(iterativeAppID, new_tracker_rtask_map);
// record the first job of the series of jobs in the iterations
this.first_job_map.put(iterativeAppID, job.getJobID());
// record the list of jobs for a iteration
HashSet<JobID> jobs = new HashSet<JobID>();
jobs.add(job.getJobID());
this.iteration_jobs_map.put(iterativeAppID, jobs);
}
// this is the first job of the series of jobs
if (this.first_job_map.get(iterativeAppID).equals(job.getJobID())
&& job.getJobConf().isIterative()) {
LOG.info(job.getJobID() + " is the first iteration job");
newIterationJob = true;
}
// this is one of the following jobs, and prepare a assignment list for the assignment
if (!newIterationJob) {
LOG.info(job.getJobID() + " is not the first iteration job");
this.iteration_jobs_map.get(iterativeAppID).add(job.getJobID());
if (this.mtask_assign_map.get(job.getJobID()) == null) {
// prepare the map task assignment list
LOG.info("for job " + job.getJobID() + "'s assignment:");
Map<String, LinkedList<Integer>> map_task_assign =
new HashMap<String, LinkedList<Integer>>();
for (Map.Entry<String, LinkedList<Integer>> entry :
this.tracker_mtask_map.get(iterativeAppID).entrySet()) {
String tracker = entry.getKey();
LinkedList<Integer> taskids = entry.getValue();
LinkedList<Integer> copytaskids = new LinkedList<Integer>();
LOG.info("assign on tracker " + tracker);
for (int taskid : taskids) {
copytaskids.add(taskid);
LOG.info("task id " + taskid);
}
map_task_assign.put(tracker, copytaskids);
}
this.mtask_assign_map.put(job.getJobID(), map_task_assign);
// if one2one copy the map assign to reduce assign, the are with the same mapping
if (jointype.equals("one2one")) {
// prepare the reduce task assignment list
Map<String, LinkedList<Integer>> reduce_task_assign =
new HashMap<String, LinkedList<Integer>>();
for (Map.Entry<String, LinkedList<Integer>> entry :
this.tracker_mtask_map.get(iterativeAppID).entrySet()) {
String tracker = entry.getKey();
LinkedList<Integer> taskids = entry.getValue();
LinkedList<Integer> copytaskids = new LinkedList<Integer>();
for (int taskid : taskids) {
copytaskids.add(taskid);
}
reduce_task_assign.put(tracker, copytaskids);
}
this.tracker_rtask_map.put(iterativeAppID, reduce_task_assign);
}
// prepare the reduce task assignment list for all cases
Map<String, LinkedList<Integer>> reduce_task_assign =
new HashMap<String, LinkedList<Integer>>();
for (Map.Entry<String, LinkedList<Integer>> entry :
this.tracker_rtask_map.get(iterativeAppID).entrySet()) {
String tracker = entry.getKey();
LinkedList<Integer> taskids = entry.getValue();
LinkedList<Integer> copytaskids = new LinkedList<Integer>();
for (int taskid : taskids) {
copytaskids.add(taskid);
}
reduce_task_assign.put(tracker, copytaskids);
}
this.rtask_assign_map.put(job.getJobID(), reduce_task_assign);
}
}
Task t = null;
// the first iteration or following iteration
// if the first iteration: assign taskid by default (exception for the one2mul case,
// where we assign staring from 0,...,n)
// else if the following iterations: assign taskid based on the first iteration
// assignment
if (newIterationJob) {
/**
* the one2mul case should be carefully taken care, we want to assgin map0,map1,map2
* and reduce0 to a tracker, and assign map3,map4,map5 and reduce1 to another tracker
*/
if (jointype.equals("one2mul")
&& !tracker_rtask_map
.get(iterativeAppID)
.containsKey(taskTracker.getTrackerName())) {
// if contain the tracker, that means we have assigned tasks for this tracker
int scala = job.getJobConf().getInt("mapred.iterative.data.scala", 1);
// int mapsEachTracker = job.getJobConf().getNumMapTasks() / numTaskTrackers;
int reducersEachTracker = job.getJobConf().getNumReduceTasks() / numTaskTrackers;
if (job.getJobConf().getNumReduceTasks() % numTaskTrackers != 0)
throw new IOException(
"job.getJobConf().getNumReduceTasks() % numTaskTrackers != 0");
if (!this.tracker_mtask_map
.get(iterativeAppID)
.containsKey(taskTracker.getTrackerName())) {
LinkedList<Integer> tasklist = new LinkedList<Integer>();
this.tracker_mtask_map
.get(iterativeAppID)
.put(taskTracker.getTrackerName(), tasklist);
}
if (!this.tracker_rtask_map
.get(iterativeAppID)
.containsKey(taskTracker.getTrackerName())) {
LinkedList<Integer> tasklist = new LinkedList<Integer>();
this.tracker_rtask_map
.get(iterativeAppID)
.put(taskTracker.getTrackerName(), tasklist);
}
// for debugging
String debugout1 = "maps: ";
String debugout2 = "reduces: ";
int reduceOffsetId =
(tracker_rtask_map.get(iterativeAppID).size() - 1)
* reducersEachTracker; // the start reduce id
for (int count = 0; count < reducersEachTracker; count++) {
int reducepartitionid = reduceOffsetId + count;
debugout2 += reducepartitionid + " ";
tracker_rtask_map
.get(iterativeAppID)
.get(taskTracker.getTrackerName())
.add(reducepartitionid);
for (int count2 = 0; count2 < scala; count2++) {
int mappartitionid = reducepartitionid * scala + count2;
// int mapid = job.splitTaskMap.get(mappartitionid);
debugout1 += mappartitionid + " ";
this.tracker_mtask_map
.get(iterativeAppID)
.get(taskTracker.getTrackerName())
.add(mappartitionid);
}
}
// print out for debug
LOG.info(
"tracker "
+ taskTracker.getTrackerName()
+ " assigned tasks "
+ debugout1
+ " and "
+ debugout2);
// make the assignment list
String tracker = taskTracker.getTrackerName();
LinkedList<Integer> mtaskids =
this.tracker_mtask_map.get(iterativeAppID).get(taskTracker.getTrackerName());
LinkedList<Integer> mcopytaskids = new LinkedList<Integer>();
for (int taskid : mtaskids) {
mcopytaskids.add(taskid);
}
if (!mtask_assign_map.containsKey(job.getJobID())) {
Map<String, LinkedList<Integer>> map_task_assign =
new HashMap<String, LinkedList<Integer>>();
this.mtask_assign_map.put(job.getJobID(), map_task_assign);
}
this.mtask_assign_map.get(job.getJobID()).put(tracker, mcopytaskids);
// prepare the reduce task assignment list
LinkedList<Integer> rtaskids =
this.tracker_rtask_map.get(iterativeAppID).get(taskTracker.getTrackerName());
LinkedList<Integer> rcopytaskids = new LinkedList<Integer>();
for (int taskid : rtaskids) {
rcopytaskids.add(taskid);
}
if (!rtask_assign_map.containsKey(job.getJobID())) {
Map<String, LinkedList<Integer>> reduce_task_assign =
new HashMap<String, LinkedList<Integer>>();
this.rtask_assign_map.put(job.getJobID(), reduce_task_assign);
}
this.rtask_assign_map.get(job.getJobID()).put(tracker, rcopytaskids);
// assign a map task for this tracker
Integer target = null;
try {
target =
this.mtask_assign_map
.get(job.getJobID())
.get(taskTracker.getTrackerName())
.peekFirst();
} catch (Exception e) {
e.printStackTrace();
}
if (target == null) {
// all have been assigned, no more work, maybe it should help others to process
LOG.info(
"all map tasks on tasktracker "
+ taskTracker.getTrackerName()
+ " have been processed");
break;
} else {
t =
job.obtainNewNodeOrRackLocalMapTask(
taskTrackerStatus,
numTaskTrackers,
taskTrackerManager.getNumberOfUniqueHosts(),
target);
}
} else {
t =
job.obtainNewNodeOrRackLocalMapTask(
taskTrackerStatus,
numTaskTrackers,
taskTrackerManager.getNumberOfUniqueHosts());
}
} else {
Integer target = null;
try {
target =
this.mtask_assign_map
.get(job.getJobID())
.get(taskTracker.getTrackerName())
.peekFirst();
} catch (Exception e) {
e.printStackTrace();
}
if (target == null) {
// all have been assigned, no more work, maybe it should help others to process
LOG.info(
"all map tasks on tasktracker "
+ taskTracker.getTrackerName()
+ " have been processed");
break;
} else {
t =
job.obtainNewNodeOrRackLocalMapTask(
taskTrackerStatus,
numTaskTrackers,
taskTrackerManager.getNumberOfUniqueHosts(),
target);
}
}
if (t != null) {
assignedTasks.add(t);
++numLocalMaps;
// new iteration job and the first task for a tasktracker
// for one2mul case, we don't need to record the assignment, since we already made the
// assignment list beforehand
if (!newIterationJob || jointype.equals("one2mul")) {
// poll, remove
this.mtask_assign_map
.get(job.getJobID())
.get(taskTracker.getTrackerName())
.pollFirst();
LOG.info("assigning task " + t.getTaskID() + " on " + taskTracker.getTrackerName());
} else {
// record the assignment list for map tasks
if (!this.tracker_mtask_map
.get(iterativeAppID)
.containsKey(taskTracker.getTrackerName())) {
LinkedList<Integer> tasklist = new LinkedList<Integer>();
this.tracker_mtask_map
.get(iterativeAppID)
.put(taskTracker.getTrackerName(), tasklist);
}
this.tracker_mtask_map
.get(iterativeAppID)
.get(taskTracker.getTrackerName())
.add(t.getTaskID().getTaskID().getId());
// prepare the reduce assignment, for mapping with reduce
if (jointype.equals("one2one")) {
// prepare the reduce assignment, for mapping with reduce
if (!first_job_reduces_map.containsKey(iterativeAppID)) {
Map<String, LinkedList<Integer>> tracker_reduce_map =
new HashMap<String, LinkedList<Integer>>();
first_job_reduces_map.put(iterativeAppID, tracker_reduce_map);
}
if (!first_job_reduces_map
.get(iterativeAppID)
.containsKey(taskTracker.getTrackerName())) {
LinkedList<Integer> reduces = new LinkedList<Integer>();
first_job_reduces_map
.get(iterativeAppID)
.put(taskTracker.getTrackerName(), reduces);
}
first_job_reduces_map
.get(iterativeAppID)
.get(taskTracker.getTrackerName())
.add(t.getTaskID().getTaskID().getId());
}
LOG.info("assigning task " + t.getTaskID() + " on " + taskTracker.getTrackerName());
}
// Don't assign map tasks to the hilt!
// Leave some free slots in the cluster for future task-failures,
// speculative tasks etc. beyond the highest priority job
if (exceededMapPadding) {
break scheduleMaps;
}
// Try all jobs again for the next Map task
break;
}
LOG.error("New Node Or Rack Local Map Task failed!");
if (newIterationJob) {
// Try to schedule a node-local or rack-local Map task
t =
job.obtainNewNonLocalMapTask(
taskTrackerStatus,
numTaskTrackers,
taskTrackerManager.getNumberOfUniqueHosts());
} else {
Integer target =
this.mtask_assign_map
.get(job.getJobID())
.get(taskTracker.getTrackerName())
.peekFirst();
if (target == null) {
// all have been assigned, no more work, maybe it should help others to process
LOG.info(
"all map tasks on tasktracker "
+ taskTracker.getTrackerName()
+ " have been processed");
break;
} else {
t =
job.obtainNewNonLocalMapTask(
taskTrackerStatus,
numTaskTrackers,
taskTrackerManager.getNumberOfUniqueHosts(),
target);
}
}
if (t != null) {
assignedTasks.add(t);
++numNonLocalMaps;
// new iteration job and the first task for a tasktracker
if (newIterationJob) {
if (!this.tracker_mtask_map
.get(iterativeAppID)
.containsKey(taskTracker.getTrackerName())) {
LinkedList<Integer> tasklist = new LinkedList<Integer>();
this.tracker_mtask_map
.get(iterativeAppID)
.put(taskTracker.getTrackerName(), tasklist);
}
this.tracker_mtask_map
.get(iterativeAppID)
.get(taskTracker.getTrackerName())
.add(t.getTaskID().getTaskID().getId());
} else {
// poll, remove
this.mtask_assign_map
.get(job.getJobID())
.get(taskTracker.getTrackerName())
.pollFirst();
}
// We assign at most 1 off-switch or speculative task
// This is to prevent TaskTrackers from stealing local-tasks
// from other TaskTrackers.
break scheduleMaps;
}
} else {
// not an iterative algorithm, normal schedule
Task t = null;
// Try to schedule a node-local or rack-local Map task
t =
job.obtainNewNodeOrRackLocalMapTask(
taskTrackerStatus,
numTaskTrackers,
taskTrackerManager.getNumberOfUniqueHosts());
if (t != null) {
assignedTasks.add(t);
++numLocalMaps;
// Don't assign map tasks to the hilt!
// Leave some free slots in the cluster for future task-failures,
// speculative tasks etc. beyond the highest priority job
if (exceededMapPadding) {
break scheduleMaps;
}
// Try all jobs again for the next Map task
break;
}
// Try to schedule a node-local or rack-local Map task
t =
job.obtainNewNonLocalMapTask(
taskTrackerStatus,
numTaskTrackers,
taskTrackerManager.getNumberOfUniqueHosts());
if (t != null) {
assignedTasks.add(t);
++numNonLocalMaps;
// We assign at most 1 off-switch or speculative task
// This is to prevent TaskTrackers from stealing local-tasks
// from other TaskTrackers.
break scheduleMaps;
}
}
}
}
}
int assignedMaps = assignedTasks.size();
//
// Same thing, but for reduce tasks
// However we _never_ assign more than 1 reduce task per heartbeat
//
/** should maintain the reduce task location for the termination check */
final int trackerCurrentReduceCapacity =
Math.min((int) Math.ceil(reduceLoadFactor * trackerReduceCapacity), trackerReduceCapacity);
final int availableReduceSlots =
Math.min((trackerCurrentReduceCapacity - trackerRunningReduces), 1);
boolean exceededReducePadding = false;
// LOG.info("availableReduceSlots " + availableReduceSlots);
if (availableReduceSlots > 0) {
exceededReducePadding = exceededPadding(false, clusterStatus, trackerReduceCapacity);
synchronized (jobQueue) {
for (JobInProgress job : jobQueue) {
LOG.info("job " + job.getJobID());
if (job.getStatus().getRunState() != JobStatus.RUNNING || job.numReduceTasks == 0) {
LOG.info("have to continue " + job.getStatus().getRunState());
continue;
}
Task t = null;
if (job.getJobConf().isIterative()) {
String iterativeAppID = job.getJobConf().getIterativeAlgorithmID();
if (iterativeAppID.equals("none")) {
throw new IOException("please specify the iteration ID!");
}
String jointype = job.getJobConf().get("mapred.iterative.jointype");
if (jointype.equals("one2one")) {
// one-to-one or one-to-mul jobs
if (this.first_job_map.get(iterativeAppID).equals(job.getJobID())
&& job.getJobConf().isIterative()) {
LOG.info(job.getJobID() + " is the first iteration job for reduce");
newIterationJob = true;
}
Integer target = null;
if (newIterationJob) {
if (first_job_reduces_map.get(iterativeAppID) == null) {
throw new IOException(
"I think something is wrong since the tasktracker never receive "
+ "a map task with iterativeapp id "
+ iterativeAppID);
}
if (first_job_reduces_map.get(iterativeAppID).get(taskTracker.getTrackerName())
== null) {
throw new IOException(
"I think something is wrong since the tasktracker never receive "
+ "a map task with iterativeapp id "
+ iterativeAppID
+ " from "
+ taskTracker.getTrackerName());
}
target =
this.first_job_reduces_map
.get(iterativeAppID)
.get(taskTracker.getTrackerName())
.pollFirst();
} else {
// the task assignment has already been processed during the map task assignment, so
// never use tracker_rtask_map
target =
this.rtask_assign_map
.get(job.getJobID())
.get(taskTracker.getTrackerName())
.pollFirst();
}
if (target == null) {
// all have been assigned, no more work, maybe it should help others to process
LOG.info(
"all reduce tasks on tasktracker "
+ taskTracker.getTrackerName()
+ " have been processed");
break;
} else {
t =
job.obtainNewReduceTask(
taskTrackerStatus,
numTaskTrackers,
taskTrackerManager.getNumberOfUniqueHosts(),
target);
}
} else if (jointype.equals("one2mul")) {
Integer target =
this.rtask_assign_map
.get(job.getJobID())
.get(taskTracker.getTrackerName())
.pollFirst();
if (target == null) {
// all have been assigned, no more work, maybe it should help others to process
LOG.info(
"all reduce tasks on tasktracker "
+ taskTracker.getTrackerName()
+ " have been processed");
break;
} else {
t =
job.obtainNewReduceTask(
taskTrackerStatus,
numTaskTrackers,
taskTrackerManager.getNumberOfUniqueHosts(),
target);
}
} else {
// one-to-all case, assign tasks in the first iteration job, and remember this mapping
// this is the first job of the series of jobs
if (this.first_job_map.get(iterativeAppID).equals(job.getJobID())) {
LOG.info(job.getJobID() + " is the first iteration job for reduce");
newIterationJob = true;
}
/*
//this is one of the following jobs, and prepare a assignment list for the assignment
else{
LOG.info(job.getJobID() + " is not the first iteration job for reduce");
if(this.rtask_assign_map.get(job.getJobID()) == null){
//prepare the map task assignment list
Map<String, LinkedList<Integer>> reduce_task_assign = new HashMap<String, LinkedList<Integer>>();
for(Map.Entry<String, LinkedList<Integer>> entry : this.tracker_rtask_map.get(iterativeAppID).entrySet()){
String tracker = entry.getKey();
LinkedList<Integer> taskids = entry.getValue();
LinkedList<Integer> copytaskids = new LinkedList<Integer>();
for(int taskid : taskids){
copytaskids.add(taskid);
}
reduce_task_assign.put(tracker, copytaskids);
}
this.rtask_assign_map.put(job.getJobID(), reduce_task_assign);
}
}
*/
// the first iteration or following iteration
// if the first iteration: assign taskid by default
// else if the following iterations: assign taskid based on the first iteration
// assignment
if (newIterationJob) {
t =
job.obtainNewReduceTask(
taskTrackerStatus,
numTaskTrackers,
taskTrackerManager.getNumberOfUniqueHosts());
if (t != null) {
if (!this.tracker_rtask_map
.get(iterativeAppID)
.containsKey(taskTracker.getTrackerName())) {
LinkedList<Integer> tasklist = new LinkedList<Integer>();
this.tracker_rtask_map
.get(iterativeAppID)
.put(taskTracker.getTrackerName(), tasklist);
}
this.tracker_rtask_map
.get(iterativeAppID)
.get(taskTracker.getTrackerName())
.add(t.getTaskID().getTaskID().getId());
LOG.info(
"assigning reduce task "
+ t.getTaskID()
+ " on "
+ taskTracker.getTrackerName());
}
} else {
Integer target =
this.rtask_assign_map
.get(job.getJobID())
.get(taskTracker.getTrackerName())
.peekFirst();
if (target == null) {
// all have been assigned, no more work, maybe it should help others to process
LOG.info(
"all map tasks on tasktracker "
+ taskTracker.getTrackerName()
+ " have been processed");
break;
} else {
t =
job.obtainNewReduceTask(
taskTrackerStatus,
numTaskTrackers,
taskTrackerManager.getNumberOfUniqueHosts(),
target);
}
if (t != null) {
// poll, remove
this.rtask_assign_map
.get(job.getJobID())
.get(taskTracker.getTrackerName())
.pollFirst();
LOG.info(
"assigning reduce task "
+ t.getTaskID()
+ " on "
+ taskTracker.getTrackerName());
}
}
}
} else {
t =
job.obtainNewReduceTask(
taskTrackerStatus,
numTaskTrackers,
taskTrackerManager.getNumberOfUniqueHosts());
}
LOG.info("try to assign new task " + t);
if (t != null) {
assignedTasks.add(t);
break;
}
// Don't assign reduce tasks to the hilt!
// Leave some free slots in the cluster for future task-failures,
// speculative tasks etc. beyond the highest priority job
if (exceededReducePadding) {
break;
}
}
}
}
if (LOG.isDebugEnabled()) {
LOG.debug(
"Task assignments for "
+ taskTrackerStatus.getTrackerName()
+ " --> "
+ "["
+ mapLoadFactor
+ ", "
+ trackerMapCapacity
+ ", "
+ trackerCurrentMapCapacity
+ ", "
+ trackerRunningMaps
+ "] -> ["
+ (trackerCurrentMapCapacity - trackerRunningMaps)
+ ", "
+ assignedMaps
+ " ("
+ numLocalMaps
+ ", "
+ numNonLocalMaps
+ ")] ["
+ reduceLoadFactor
+ ", "
+ trackerReduceCapacity
+ ", "
+ trackerCurrentReduceCapacity
+ ","
+ trackerRunningReduces
+ "] -> ["
+ (trackerCurrentReduceCapacity - trackerRunningReduces)
+ ", "
+ (assignedTasks.size() - assignedMaps)
+ "]");
}
return assignedTasks;
}
/*
* TODO:
* For Elf: need to change the major schedule logic, scheduling need
* to be *datacenter-aware*
* */
@Override
public synchronized List<Task> assignTasks(TaskTracker taskTracker) throws IOException {
TaskTrackerStatus taskTrackerStatus = taskTracker.getStatus();
ClusterStatus clusterStatus = taskTrackerManager.getClusterStatus();
final int numTaskTrackers = clusterStatus.getTaskTrackers();
final int clusterMapCapacity = clusterStatus.getMaxMapTasks();
final int clusterReduceCapacity = clusterStatus.getMaxReduceTasks();
Collection<JobInProgress> jobQueue = jobQueueJobInProgressListener.getJobQueue();
//
// Get map + reduce counts for the current tracker.
//
final int trackerMapCapacity = taskTrackerStatus.getMaxMapSlots();
final int trackerReduceCapacity = taskTrackerStatus.getMaxReduceSlots();
final int trackerRunningMaps = taskTrackerStatus.countMapTasks();
final int trackerRunningReduces = taskTrackerStatus.countReduceTasks();
// Assigned tasks
List<Task> assignedTasks = new ArrayList<Task>();
//
// Compute (running + pending) map and reduce task numbers across pool
//
int remainingReduceLoad = 0;
int remainingMapLoad = 0;
synchronized (jobQueue) {
for (JobInProgress job : jobQueue) {
if (job.getStatus().getRunState() == JobStatus.RUNNING) {
remainingMapLoad += (job.desiredMaps() - job.finishedMaps());
if (job.scheduleReduces()) {
remainingReduceLoad += (job.desiredReduces() - job.finishedReduces());
}
}
}
}
// Compute the 'load factor' for maps and reduces
double mapLoadFactor = 0.0;
if (clusterMapCapacity > 0) {
mapLoadFactor = (double) remainingMapLoad / clusterMapCapacity;
}
double reduceLoadFactor = 0.0;
if (clusterReduceCapacity > 0) {
reduceLoadFactor = (double) remainingReduceLoad / clusterReduceCapacity;
}
//
// In the below steps, we allocate first map tasks (if appropriate),
// and then reduce tasks if appropriate. We go through all jobs
// in order of job arrival; jobs only get serviced if their
// predecessors are serviced, too.
//
//
// We assign tasks to the current taskTracker if the given machine
// has a workload that's less than the maximum load of that kind of
// task.
// However, if the cluster is close to getting loaded i.e. we don't
// have enough _padding_ for speculative executions etc., we only
// schedule the "highest priority" task i.e. the task from the job
// with the highest priority.
//
final int trackerCurrentMapCapacity =
Math.min((int) Math.ceil(mapLoadFactor * trackerMapCapacity), trackerMapCapacity);
int availableMapSlots = trackerCurrentMapCapacity - trackerRunningMaps;
boolean exceededMapPadding = false;
if (availableMapSlots > 0) {
exceededMapPadding = exceededPadding(true, clusterStatus, trackerMapCapacity);
}
int numLocalMaps = 0;
int numNonLocalMaps = 0;
scheduleMaps:
// TODO: for Elf
// The main schedule logic here, outer for loop is for every slot, inner loop is for each job
for (int i = 0; i < availableMapSlots; ++i) {
synchronized (jobQueue) {
for (JobInProgress job : jobQueue) {
if (job.getStatus().getRunState() != JobStatus.RUNNING) {
continue;
}
Task t = null;
// Try to schedule a node-local or rack-local Map task
t =
job.obtainNewLocalMapTask(
taskTrackerStatus, numTaskTrackers, taskTrackerManager.getNumberOfUniqueHosts());
if (t != null) {
assignedTasks.add(t);
++numLocalMaps;
// Don't assign map tasks to the hilt!
// Leave some free slots in the cluster for future task-failures,
// speculative tasks etc. beyond the highest priority job
if (exceededMapPadding) {
break scheduleMaps;
}
// Try all jobs again for the next Map task
// Note: it's FIFO here: next time in the inner for loop the head-of-queue
// will still be chosen
break;
}
// If no locality for this job, try launching non-local
// Try to schedule a node-local or rack-local Map task --> original comments
// FIXME: is the above comment correct? seems should be non-local task
t =
job.obtainNewNonLocalMapTask(
taskTrackerStatus, numTaskTrackers, taskTrackerManager.getNumberOfUniqueHosts());
if (t != null) {
assignedTasks.add(t);
++numNonLocalMaps;
// We assign at most 1 off-switch or speculative task
// This is to prevent TaskTrackers from stealing local-tasks
// from other TaskTrackers.
break scheduleMaps;
}
}
}
}
int assignedMaps = assignedTasks.size();
//
// Same thing, but for reduce tasks
// However we _never_ assign more than 1 reduce task per heartbeat
//
final int trackerCurrentReduceCapacity =
Math.min((int) Math.ceil(reduceLoadFactor * trackerReduceCapacity), trackerReduceCapacity);
final int availableReduceSlots =
Math.min((trackerCurrentReduceCapacity - trackerRunningReduces), 1);
boolean exceededReducePadding = false;
if (availableReduceSlots > 0) {
exceededReducePadding = exceededPadding(false, clusterStatus, trackerReduceCapacity);
synchronized (jobQueue) {
for (JobInProgress job : jobQueue) {
if (job.getStatus().getRunState() != JobStatus.RUNNING || job.numReduceTasks == 0) {
continue;
}
Task t =
job.obtainNewReduceTask(
taskTrackerStatus, numTaskTrackers, taskTrackerManager.getNumberOfUniqueHosts());
if (t != null) {
assignedTasks.add(t);
break;
}
// Don't assign reduce tasks to the hilt!
// Leave some free slots in the cluster for future task-failures,
// speculative tasks etc. beyond the highest priority job
if (exceededReducePadding) {
break;
}
}
}
}
if (LOG.isDebugEnabled()) {
LOG.debug(
"Task assignments for "
+ taskTrackerStatus.getTrackerName()
+ " --> "
+ "["
+ mapLoadFactor
+ ", "
+ trackerMapCapacity
+ ", "
+ trackerCurrentMapCapacity
+ ", "
+ trackerRunningMaps
+ "] -> ["
+ (trackerCurrentMapCapacity - trackerRunningMaps)
+ ", "
+ assignedMaps
+ " ("
+ numLocalMaps
+ ", "
+ numNonLocalMaps
+ ")] ["
+ reduceLoadFactor
+ ", "
+ trackerReduceCapacity
+ ", "
+ trackerCurrentReduceCapacity
+ ","
+ trackerRunningReduces
+ "] -> ["
+ (trackerCurrentReduceCapacity - trackerRunningReduces)
+ ", "
+ (assignedTasks.size() - assignedMaps)
+ "]");
}
return assignedTasks;
}
/** Check refreshNodes for decommissioning blacklisted nodes. */
public void testBlacklistedNodeDecommissioning() throws Exception {
LOG.info("Testing blacklisted node decommissioning");
MiniMRCluster mr = null;
JobTracker jt = null;
try {
// start mini mr
JobConf jtConf = new JobConf();
jtConf.set("mapred.max.tracker.blacklists", "1");
mr = new MiniMRCluster(0, 0, 2, "file:///", 1, null, null, null, jtConf);
jt = mr.getJobTrackerRunner().getJobTracker();
assertEquals("Trackers not up", 2, jt.taskTrackers().size());
// validate the total tracker count
assertEquals(
"Active tracker count mismatch", 2, jt.getClusterStatus(false).getTaskTrackers());
// validate blacklisted count
assertEquals(
"Blacklisted tracker count mismatch",
0,
jt.getClusterStatus(false).getBlacklistedTrackers());
// run a failing job to blacklist the tracker
JobConf jConf = mr.createJobConf();
jConf.set("mapred.max.tracker.failures", "1");
jConf.setJobName("test-job-fail-once");
jConf.setMapperClass(FailOnceMapper.class);
jConf.setReducerClass(IdentityReducer.class);
jConf.setNumMapTasks(1);
jConf.setNumReduceTasks(0);
RunningJob job =
UtilsForTests.runJob(jConf, new Path(TEST_DIR, "in"), new Path(TEST_DIR, "out"));
job.waitForCompletion();
// validate the total tracker count
assertEquals(
"Active tracker count mismatch", 1, jt.getClusterStatus(false).getTaskTrackers());
// validate blacklisted count
assertEquals(
"Blacklisted tracker count mismatch",
1,
jt.getClusterStatus(false).getBlacklistedTrackers());
// find the blacklisted tracker
String trackerName = null;
for (TaskTrackerStatus status : jt.taskTrackers()) {
if (jt.isBlacklisted(status.getTrackerName())) {
trackerName = status.getTrackerName();
break;
}
}
// get the hostname
String hostToDecommission = JobInProgress.convertTrackerNameToHostName(trackerName);
LOG.info("Decommissioning tracker " + hostToDecommission);
// decommission the node
HashSet<String> decom = new HashSet<String>(1);
decom.add(hostToDecommission);
jt.decommissionNodes(decom);
// validate
// check the cluster status and tracker size
assertEquals(
"Tracker is not lost upon host decommissioning",
1,
jt.getClusterStatus(false).getTaskTrackers());
assertEquals(
"Blacklisted tracker count incorrect in cluster status " + "after decommissioning",
0,
jt.getClusterStatus(false).getBlacklistedTrackers());
assertEquals("Tracker is not lost upon host decommissioning", 1, jt.taskTrackers().size());
} finally {
if (mr != null) {
mr.shutdown();
mr = null;
jt = null;
FileUtil.fullyDelete(new File(TEST_DIR.toString()));
}
}
}