/** Return a Task that can be sent to a TaskTracker for execution. */
public Task getTaskToRun(String taskTracker) {
// Create the 'taskid'; do not count the 'killed' tasks against the job!
TaskAttemptID taskid = null;
if (nextTaskId < (MAX_TASK_EXECS + maxTaskAttempts + numKilledTasks)) {
// Make sure that the attempts are unqiue across restarts
int attemptId = job.getNumRestarts() * NUM_ATTEMPTS_PER_RESTART + nextTaskId;
taskid = new TaskAttemptID(id, attemptId);
++nextTaskId;
} else {
LOG.warn(
"Exceeded limit of "
+ (MAX_TASK_EXECS + maxTaskAttempts)
+ " (plus "
+ numKilledTasks
+ " killed)"
+ " attempts for the tip '"
+ getTIPId()
+ "'");
return null;
}
// keep track of the last time we started an attempt at this TIP
// used to calculate the progress rate of this TIP
setDispatchTime(taskid, JobTracker.getClock().getTime());
if (0 == execStartTime) {
// assume task starts running now
execStartTime = JobTracker.getClock().getTime();
}
return addRunningTask(taskid, taskTracker);
}
/** The TIP's been ordered kill()ed. */
public void kill() {
if (isComplete() || failed) {
return;
}
this.failed = true;
killed = true;
this.execFinishTime = JobTracker.getClock().getTime();
recomputeProgress();
}
/** Indicate that one of the taskids in this TaskInProgress has successfully completed! */
public void completed(TaskAttemptID taskid) {
//
// Record that this taskid is complete
//
completedTask(taskid, TaskStatus.State.SUCCEEDED);
// Note the successful taskid
setSuccessfulTaskid(taskid);
//
// Now that the TIP is complete, the other speculative
// subtasks will be closed when the owning tasktracker
// reports in and calls shouldClose() on this object.
//
this.completes++;
this.execFinishTime = JobTracker.getClock().getTime();
recomputeProgress();
}
/** Initialization common to Map and Reduce */
void init(JobID jobId) {
this.startTime = JobTracker.getClock().getTime();
this.id = new TaskID(jobId, isMapTask(), partition);
this.skipping = startSkipping();
long speculativeDuration;
if (isMapTask()) {
this.speculativeLag = conf.getMapSpeculativeLag();
speculativeDuration = conf.getMapSpeculativeDuration();
} else {
this.speculativeLag = conf.getReduceSpeculativeLag();
speculativeDuration = conf.getReduceSpeculativeDuration();
}
// speculate only if 1/(1000 * progress_rate) > speculativeDuration
// ie. :
// speculate only if progress_rate < 1/(1000 * speculativeDuration)
if (speculativeDuration > 0) {
this.maxProgressRateForSpeculation = 1.0 / (1000.0 * speculativeDuration);
} else {
// disable this check for durations <= 0
this.maxProgressRateForSpeculation = -1.0;
}
}
@SuppressWarnings("deprecation")
public SimulatorJobInProgress(
JobID jobid, JobTracker jobtracker, JobConf default_conf, JobStory jobStory) {
super(jobid, jobStory.getJobConf(), jobtracker);
// jobSetupCleanupNeeded set to false in parent cstr, though
// default is true
restartCount = 0;
jobSetupCleanupNeeded = false;
this.memoryPerMap = conf.getMemoryForMapTask();
this.memoryPerReduce = conf.getMemoryForReduceTask();
this.maxTaskFailuresPerTracker = conf.getMaxTaskFailuresPerTracker();
this.jobId = jobid;
String url =
"http://"
+ jobtracker.getJobTrackerMachine()
+ ":"
+ jobtracker.getInfoPort()
+ "/jobdetails.jsp?jobid="
+ jobid;
this.jobtracker = jobtracker;
this.conf = jobStory.getJobConf();
this.priority = conf.getJobPriority();
Path jobDir = jobtracker.getSystemDirectoryForJob(jobid);
this.jobFile = new Path(jobDir, "job.xml");
this.status =
new JobStatus(jobid, 0.0f, 0.0f, 0.0f, 0.0f, JobStatus.PREP, priority, conf.getUser());
this.profile =
new JobProfile(
jobStory.getUser(),
jobid,
this.jobFile.toString(),
url,
jobStory.getName(),
conf.getQueueName());
this.startTime = JobTracker.getClock().getTime();
status.setStartTime(startTime);
this.resourceEstimator = new ResourceEstimator(this);
this.numMapTasks = jobStory.getNumberMaps();
this.numReduceTasks = jobStory.getNumberReduces();
this.taskCompletionEvents =
new ArrayList<TaskCompletionEvent>(numMapTasks + numReduceTasks + 10);
this.mapFailuresPercent = conf.getMaxMapTaskFailuresPercent();
this.reduceFailuresPercent = conf.getMaxReduceTaskFailuresPercent();
MetricsContext metricsContext = MetricsUtil.getContext("mapred");
this.jobMetrics = MetricsUtil.createRecord(metricsContext, "job");
this.jobMetrics.setTag("user", conf.getUser());
this.jobMetrics.setTag("sessionId", conf.getSessionId());
this.jobMetrics.setTag("jobName", conf.getJobName());
this.jobMetrics.setTag("jobId", jobid.toString());
this.maxLevel = jobtracker.getNumTaskCacheLevels();
this.anyCacheLevel = this.maxLevel + 1;
this.nonLocalMaps = new LinkedList<TaskInProgress>();
this.nonLocalRunningMaps = new LinkedHashSet<TaskInProgress>();
this.runningMapCache = new IdentityHashMap<Node, Set<TaskInProgress>>();
this.nonRunningReduces = new LinkedList<TaskInProgress>();
this.runningReduces = new LinkedHashSet<TaskInProgress>();
this.slowTaskThreshold =
Math.max(0.0f, conf.getFloat("mapred.speculative.execution.slowTaskThreshold", 1.0f));
this.speculativeCap = conf.getFloat("mapred.speculative.execution.speculativeCap", 0.1f);
this.slowNodeThreshold = conf.getFloat("mapred.speculative.execution.slowNodeThreshold", 1.0f);
this.jobStory = jobStory;
// this.jobHistory = this.jobtracker.getJobHistory();
}
// for initTasks, update information from JobStory object
@Override
public synchronized void initTasks() throws IOException {
boolean loggingEnabled = LOG.isDebugEnabled();
if (loggingEnabled) {
LOG.debug("(initTasks@SJIP) Starting Initialization for " + jobId);
}
numMapTasks = jobStory.getNumberMaps();
numReduceTasks = jobStory.getNumberReduces();
JobHistory.JobInfo.logSubmitted(
getJobID(), conf, jobFile.toString(), this.startTime, hasRestarted());
if (loggingEnabled) {
LOG.debug("(initTasks@SJIP) Logged to job history for " + jobId);
}
// checkTaskLimits();
if (loggingEnabled) {
LOG.debug("(initTasks@SJIP) Checked task limits for " + jobId);
}
final String jobFile = "default";
splits = getRawSplits(jobStory.getInputSplits());
if (loggingEnabled) {
LOG.debug(
"(initTasks@SJIP) Created splits for job = "
+ jobId
+ " number of splits = "
+ splits.length);
}
// createMapTasks(jobFile, splits);
numMapTasks = splits.length;
maps = new TaskInProgress[numMapTasks];
for (int i = 0; i < numMapTasks; ++i) {
inputLength += splits[i].getDataLength();
maps[i] = new TaskInProgress(jobId, jobFile, splits[i], conf, this, i, numSlotsPerMap);
}
if (numMapTasks > 0) {
nonRunningMapCache = createCache(splits, maxLevel);
if (loggingEnabled) {
LOG.debug(
"initTasks:numMaps="
+ numMapTasks
+ " Size of nonRunningMapCache="
+ nonRunningMapCache.size()
+ " for "
+ jobId);
}
}
// set the launch time
this.launchTime = JobTracker.getClock().getTime();
// createReduceTasks(jobFile);
//
// Create reduce tasks
//
this.reduces = new TaskInProgress[numReduceTasks];
for (int i = 0; i < numReduceTasks; i++) {
reduces[i] =
new TaskInProgress(jobId, jobFile, numMapTasks, i, conf, this, numSlotsPerReduce);
nonRunningReduces.add(reduces[i]);
}
// Calculate the minimum number of maps to be complete before
// we should start scheduling reduces
completedMapsForReduceSlowstart =
(int)
Math.ceil(
(conf.getFloat(
"mapred.reduce.slowstart." + "completed.maps",
DEFAULT_COMPLETED_MAPS_PERCENT_FOR_REDUCE_SLOWSTART)
* numMapTasks));
tasksInited.set(true);
if (loggingEnabled) {
LOG.debug(
"Initializing job, nowstatus = " + JobStatus.getJobRunState(getStatus().getRunState()));
}
setupComplete();
if (loggingEnabled) {
LOG.debug(
"Initializing job, inited-status = "
+ JobStatus.getJobRunState(getStatus().getRunState()));
}
}
/** Indicate that one of the taskids in this TaskInProgress has failed. */
public void incompleteSubTask(TaskAttemptID taskid, JobStatus jobStatus) {
//
// Note the failure and its location
//
TaskStatus status = taskStatuses.get(taskid);
String trackerName;
String trackerHostName = null;
TaskStatus.State taskState = TaskStatus.State.FAILED;
if (status != null) {
trackerName = status.getTaskTracker();
trackerHostName = JobInProgressTraits.convertTrackerNameToHostName(trackerName);
// Check if the user manually KILLED/FAILED this task-attempt...
Boolean shouldFail = tasksToKill.remove(taskid);
if (shouldFail != null) {
if (status.getRunState() == TaskStatus.State.FAILED
|| status.getRunState() == TaskStatus.State.KILLED) {
taskState = (shouldFail) ? TaskStatus.State.FAILED : TaskStatus.State.KILLED;
} else {
taskState =
(shouldFail) ? TaskStatus.State.FAILED_UNCLEAN : TaskStatus.State.KILLED_UNCLEAN;
}
status.setRunState(taskState);
addDiagnosticInfo(taskid, "Task has been " + taskState + " by the user");
}
taskState = status.getRunState();
if (taskState != TaskStatus.State.FAILED
&& taskState != TaskStatus.State.KILLED
&& taskState != TaskStatus.State.FAILED_UNCLEAN
&& taskState != TaskStatus.State.KILLED_UNCLEAN) {
LOG.info(
"Task '"
+ taskid
+ "' running on '"
+ trackerName
+ "' in state: '"
+ taskState
+ "' being failed!");
status.setRunState(TaskStatus.State.FAILED);
taskState = TaskStatus.State.FAILED;
}
// tasktracker went down and failed time was not reported.
if (0 == status.getFinishTime()) {
status.setFinishTime(JobTracker.getClock().getTime());
}
}
this.activeTasks.remove(taskid);
// Since we do not fail completed reduces (whose outputs go to hdfs), we
// should note this failure only for completed maps, only if this taskid;
// completed this map. however if the job is done, there is no need to
// manipulate completed maps
if (this.isMapTask()
&& !jobSetup
&& !jobCleanup
&& isComplete(taskid)
&& jobStatus.getRunState() != JobStatus.SUCCEEDED) {
this.completes--;
// Reset the successfulTaskId since we don't have a SUCCESSFUL task now
resetSuccessfulTaskid();
}
// Note that there can be failures of tasks that are hosted on a machine
// that has not yet registered with restarted jobtracker
// recalculate the counts only if its a genuine failure
if (tasks.contains(taskid)) {
if (taskState == TaskStatus.State.FAILED) {
numTaskFailures++;
machinesWhereFailed.add(trackerHostName);
if (maxSkipRecords > 0) {
// skipping feature enabled
LOG.debug("TaskInProgress adding" + status.getNextRecordRange());
failedRanges.add(status.getNextRecordRange());
skipping = startSkipping();
}
} else if (taskState == TaskStatus.State.KILLED) {
numKilledTasks++;
}
}
if (numTaskFailures >= maxTaskAttempts) {
LOG.info("TaskInProgress " + getTIPId() + " has failed " + numTaskFailures + " times.");
kill();
}
}
/**
* A status message from a client has arrived. It updates the status of a single
* component-thread-task, which might result in an overall TaskInProgress status update.
*
* @return has the task changed its state noticeably?
*/
synchronized boolean updateStatus(TaskStatus status) {
TaskAttemptID taskid = status.getTaskID();
String taskTracker = status.getTaskTracker();
String diagInfo = status.getDiagnosticInfo();
TaskStatus oldStatus = taskStatuses.get(taskid);
boolean changed = true;
if (diagInfo != null && diagInfo.length() > 0) {
long runTime = status.getRunTime();
LOG.info(
"Error from "
+ taskid
+ " on "
+ taskTracker
+ " runTime(msec) "
+ runTime
+ ": "
+ diagInfo);
addDiagnosticInfo(taskid, diagInfo);
}
if (skipping) {
failedRanges.updateState(status);
}
if (oldStatus != null) {
TaskStatus.State oldState = oldStatus.getRunState();
TaskStatus.State newState = status.getRunState();
// We should never recieve a duplicate success/failure/killed
// status update for the same taskid! This is a safety check,
// and is addressed better at the TaskTracker to ensure this.
// @see {@link TaskTracker.transmitHeartbeat()}
if ((newState != TaskStatus.State.RUNNING
&& newState != TaskStatus.State.COMMIT_PENDING
&& newState != TaskStatus.State.FAILED_UNCLEAN
&& newState != TaskStatus.State.KILLED_UNCLEAN
&& newState != TaskStatus.State.UNASSIGNED)
&& (oldState == newState)) {
LOG.warn(
"Recieved duplicate status update of '"
+ newState
+ "' for '"
+ taskid
+ "' of TIP '"
+ getTIPId()
+ "'"
+ "oldTT="
+ oldStatus.getTaskTracker()
+ " while newTT="
+ status.getTaskTracker());
return false;
}
// The task is not allowed to move from completed back to running.
// We have seen out of order status messagesmoving tasks from complete
// to running. This is a spot fix, but it should be addressed more
// globally.
if ((newState == TaskStatus.State.RUNNING || newState == TaskStatus.State.UNASSIGNED)
&& (oldState == TaskStatus.State.FAILED
|| oldState == TaskStatus.State.KILLED
|| oldState == TaskStatus.State.FAILED_UNCLEAN
|| oldState == TaskStatus.State.KILLED_UNCLEAN
|| oldState == TaskStatus.State.SUCCEEDED
|| oldState == TaskStatus.State.COMMIT_PENDING)) {
return false;
}
// Do not accept any status once the task is marked FAILED/KILLED
// This is to handle the case of the JobTracker timing out a task
// due to launch delay, but the TT comes back with any state or
// TT got expired
if (oldState == TaskStatus.State.FAILED || oldState == TaskStatus.State.KILLED) {
tasksToKill.put(taskid, true);
return false;
}
changed = oldState != newState;
}
// if task is a cleanup attempt, do not replace the complete status,
// update only specific fields.
// For example, startTime should not be updated,
// but finishTime has to be updated.
if (!isCleanupAttempt(taskid)) {
taskStatuses.put(taskid, status);
// we don't want to include setup tasks in the task execution stats
if (!isJobSetupTask()
&& !isJobCleanupTask()
&& ((isMapTask() && job.hasSpeculativeMaps())
|| (!isMapTask() && job.hasSpeculativeReduces()))) {
updateProgressRate(JobTracker.getClock().getTime());
}
} else {
taskStatuses
.get(taskid)
.statusUpdate(
status.getRunState(),
status.getProgress(),
status.getStateString(),
status.getPhase(),
status.getFinishTime());
}
// Recompute progress
recomputeProgress();
return changed;
}