/**
* JobTracker.submitJob() kicks off a new job.
*
* <p>Create a 'JobInProgress' object, which contains both JobProfile and JobStatus. Those two
* sub-objects are sometimes shipped outside of the JobTracker. But JobInProgress adds info that's
* useful for the JobTracker alone.
*
* <p>We add the JIP to the jobInitQueue, which is processed asynchronously to handle
* split-computation and build up the right TaskTracker/Block mapping.
*/
public synchronized JobStatus submitJob(String jobFile) throws IOException {
totalSubmissions++;
JobInProgress job = new JobInProgress(jobFile, this, this.conf);
synchronized (jobs) {
synchronized (jobsByArrival) {
synchronized (jobInitQueue) {
jobs.put(job.getProfile().getJobId(), job);
jobsByArrival.add(job);
jobInitQueue.add(job);
jobInitQueue.notifyAll();
}
}
}
return job.getStatus();
}
/**
* 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);
}
}
}
/**
* The run method lives for the life of the JobTracker, and removes Jobs that are not still
* running, but which finished a long time ago.
*/
public void run() {
while (shouldRun) {
try {
Thread.sleep(RETIRE_JOB_CHECK_INTERVAL);
} catch (InterruptedException ie) {
}
synchronized (jobs) {
synchronized (jobInitQueue) {
synchronized (jobsByArrival) {
for (Iterator it = jobs.keySet().iterator(); it.hasNext(); ) {
String jobid = (String) it.next();
JobInProgress job = (JobInProgress) jobs.get(jobid);
if (job.getStatus().getRunState() != JobStatus.RUNNING
&& job.getStatus().getRunState() != JobStatus.PREP
&& (job.getFinishTime() + RETIRE_JOB_INTERVAL < System.currentTimeMillis())) {
it.remove();
jobInitQueue.remove(job);
jobsByArrival.remove(job);
}
}
}
}
}
}
}
public synchronized JobStatus getJobStatus(String jobid) {
JobInProgress job = (JobInProgress) jobs.get(jobid);
if (job != null) {
return job.getStatus();
} else {
return null;
}
}
public synchronized JobProfile getJobProfile(String jobid) {
JobInProgress job = (JobInProgress) jobs.get(jobid);
if (job != null) {
return job.getProfile();
} else {
return null;
}
}
private void printFailures(
JspWriter out, JobTracker tracker, JobID jobId, String kind, String cause)
throws IOException {
JobInProgress job = (JobInProgress) tracker.getJob(jobId);
if (job == null) {
out.print("<b>Job " + jobId + " not found.</b><br>\n");
return;
}
boolean includeMap = false;
boolean includeReduce = false;
if (kind == null) {
includeMap = true;
includeReduce = true;
} else if ("map".equals(kind)) {
includeMap = true;
} else if ("reduce".equals(kind)) {
includeReduce = true;
} else if ("all".equals(kind)) {
includeMap = true;
includeReduce = true;
} else {
out.print("<b>Kind " + kind + " not supported.</b><br>\n");
return;
}
TaskStatus.State state = null;
try {
if (cause != null) {
state = TaskStatus.State.valueOf(cause.toUpperCase());
if (state != TaskStatus.State.FAILED && state != TaskStatus.State.KILLED) {
out.print("<b>Cause '" + cause + "' is not an 'unsuccessful' state.</b><br>\n");
return;
}
}
} catch (IllegalArgumentException e) {
out.print("<b>Cause '" + cause + "' not supported.</b><br>\n");
return;
}
out.print("<table border=2 cellpadding=\"5\" cellspacing=\"2\">");
out.print(
"<tr><th>Attempt</th><th>Task</th><th>Machine</th><th>State</th>"
+ "<th>Error</th><th>Logs</th></tr>\n");
if (includeMap) {
TaskInProgress[] tips = job.getTasks(TaskType.MAP);
for (int i = 0; i < tips.length; ++i) {
printFailedAttempts(out, tracker, jobId, tips[i], state);
}
}
if (includeReduce) {
TaskInProgress[] tips = job.getTasks(TaskType.REDUCE);
for (int i = 0; i < tips.length; ++i) {
printFailedAttempts(out, tracker, jobId, tips[i], state);
}
}
out.print("</table>\n");
}
public Vector completedJobs() {
Vector v = new Vector();
for (Iterator it = jobs.values().iterator(); it.hasNext(); ) {
JobInProgress jip = (JobInProgress) it.next();
JobStatus status = jip.getStatus();
if (status.getRunState() == JobStatus.SUCCEEDED) {
v.add(jip);
}
}
return v;
}
public Vector runningJobs() {
Vector v = new Vector();
for (Iterator it = jobs.values().iterator(); it.hasNext(); ) {
JobInProgress jip = (JobInProgress) it.next();
JobStatus status = jip.getStatus();
if (status.getRunState() == JobStatus.RUNNING) {
v.add(jip);
}
}
return v;
}
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()]);
}
}
/**
* 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());
}
}
}
}
public void run() {
while (shouldRun) {
JobInProgress job = null;
synchronized (jobInitQueue) {
if (jobInitQueue.size() > 0) {
job = (JobInProgress) jobInitQueue.elementAt(0);
jobInitQueue.remove(job);
} else {
try {
jobInitQueue.wait(JOBINIT_SLEEP_INTERVAL);
} catch (InterruptedException iex) {
}
}
}
try {
if (job != null) {
job.initTasks();
}
} catch (Exception e) {
LOG.log(Level.WARNING, "job init failed", e);
job.kill();
}
}
}
public void _jspService(HttpServletRequest request, HttpServletResponse response)
throws java.io.IOException, ServletException {
PageContext pageContext = null;
HttpSession session = null;
ServletContext application = null;
ServletConfig config = null;
JspWriter out = null;
Object page = this;
JspWriter _jspx_out = null;
PageContext _jspx_page_context = null;
try {
response.setContentType("text/html; charset=UTF-8");
pageContext = _jspxFactory.getPageContext(this, request, response, null, true, 8192, true);
_jspx_page_context = pageContext;
application = pageContext.getServletContext();
config = pageContext.getServletConfig();
session = pageContext.getSession();
out = pageContext.getOut();
_jspx_out = out;
_jspx_resourceInjector =
(org.apache.jasper.runtime.ResourceInjector)
application.getAttribute("com.sun.appserv.jsp.resource.injector");
out.write('\n');
JobTracker tracker = (JobTracker) application.getAttribute("job.tracker");
ClusterStatus status = tracker.getClusterStatus();
String trackerName = StringUtils.simpleHostname(tracker.getJobTrackerMachine());
out.write("\n<html>\n<head>\n<title>");
out.print(trackerName);
out.write(
" Hadoop Locality Statistics</title>\n<link rel=\"stylesheet\" type=\"text/css\" href=\"/static/hadoop.css\">\n</head>\n<body>\n<h1>");
out.print(trackerName);
out.write(" Hadoop Locality Statistics</h1>\n\n<b>State:</b> ");
out.print(status.getJobTrackerState());
out.write("<br>\n<b>Started:</b> ");
out.print(new Date(tracker.getStartTime()));
out.write("<br>\n<b>Version:</b> ");
out.print(VersionInfo.getVersion());
out.write(",\n r");
out.print(VersionInfo.getRevision());
out.write("<br>\n<b>Compiled:</b> ");
out.print(VersionInfo.getDate());
out.write(" by\n ");
out.print(VersionInfo.getUser());
out.write("<br>\n<b>Identifier:</b> ");
out.print(tracker.getTrackerIdentifier());
out.write("<br>\n\n<hr>\n\n");
Collection<JobInProgress> jobs = new ArrayList<JobInProgress>();
jobs.addAll(tracker.completedJobs());
jobs.addAll(tracker.runningJobs());
jobs.addAll(tracker.failedJobs());
int dataLocalMaps = 0;
int rackLocalMaps = 0;
int totalMaps = 0;
int totalReduces = 0;
for (JobInProgress job : jobs) {
Counters counters = job.getCounters();
dataLocalMaps += counters.getCounter(JobInProgress.Counter.DATA_LOCAL_MAPS);
rackLocalMaps += counters.getCounter(JobInProgress.Counter.RACK_LOCAL_MAPS);
totalMaps += counters.getCounter(JobInProgress.Counter.TOTAL_LAUNCHED_MAPS);
totalReduces += counters.getCounter(JobInProgress.Counter.TOTAL_LAUNCHED_REDUCES);
}
int dataLocalMapPct = totalMaps == 0 ? 0 : (100 * dataLocalMaps) / totalMaps;
int rackLocalMapPct = totalMaps == 0 ? 0 : (100 * rackLocalMaps) / totalMaps;
int dataRackLocalMapPct =
totalMaps == 0 ? 0 : (100 * (dataLocalMaps + rackLocalMaps)) / totalMaps;
out.write("\n<p>\n<b>Data Local Maps:</b> ");
out.print(dataLocalMaps);
out.write(' ');
out.write('(');
out.print(dataLocalMapPct);
out.write("%) <br>\n<b>Rack Local Maps:</b> ");
out.print(rackLocalMaps);
out.write(' ');
out.write('(');
out.print(rackLocalMapPct);
out.write("%) <br>\n<b>Data or Rack Local:</b> ");
out.print(dataLocalMaps + rackLocalMaps);
out.write(' ');
out.write('(');
out.print(dataRackLocalMapPct);
out.write("%) <br>\n<b>Total Maps:</b> ");
out.print(totalMaps);
out.write(" <br>\n<b>Total Reduces:</b> ");
out.print(totalReduces);
out.write(" <br>\n</p>\n\n");
out.println(ServletUtil.htmlFooter());
out.write('\n');
} catch (Throwable t) {
if (!(t instanceof SkipPageException)) {
out = _jspx_out;
if (out != null && out.getBufferSize() != 0) out.clearBuffer();
if (_jspx_page_context != null) _jspx_page_context.handlePageException(t);
}
} finally {
_jspxFactory.releasePageContext(_jspx_page_context);
}
}
public synchronized void killJob(String jobid) {
JobInProgress job = (JobInProgress) jobs.get(jobid);
job.kill();
}
/**
* 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;
}
