/**
* 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();
}
private static Vector<String> tokenizeDoc(String cur_doc) {
String[] words = cur_doc.split("\\s+");
Vector<String> tokens = new Vector<String>();
for (int i = 0; i < words.length; i++) {
words[i] = words[i].replaceAll("\\W", "");
if (words[i].length() > 0) {
tokens.add(words[i]);
}
}
return tokens;
}
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()]);
}
}
// Information needed to get a single file:
// BASE_PATH, FILE_ID, TIMESTAMP_START, TIMESTAMP_STOP, SOURCE, FILESYSTEM
private static Vector<Path> getFile(
FileSystem fs, Hashtable<String, String> config, dbutil db_util) throws Exception {
Long latestVersion = latestVersion(config, db_util);
try {
config.put("timestamp_start", config.get("timestamp_start"));
config.put("timestamp_real", latestVersion.toString());
config.put("timestamp_stop", latestVersion.toString());
} catch (Exception E) {
logger.error("Tryign to get file that is impossible to generate: " + getFullPath(config));
return null;
}
if (Integer.parseInt(config.get("timestamp_start"))
> Integer.parseInt(config.get("timestamp_stop"))) {
return null;
}
logger.debug(
"Getting DB for timestamp "
+ config.get("timestamp_start")
+ " to "
+ config.get("timestamp_stop"));
String final_result = getFullPath(config);
String temp_path_base =
config.get("local_temp_path")
+ "_"
+ config.get("task_id")
+ "_"
+ config.get("run_id")
+ "/";
Path newPath = new Path(final_result + "*");
Vector<Path> ret_path = new Vector<Path>();
String lockName = lock(final_result.replaceAll("/", "_"));
if (fs.globStatus(newPath).length != 0) {
ret_path.add(newPath);
unlock(lockName);
config.put("full_file_name", final_result);
return ret_path;
} else {
if (!config.get("source").equals("local")) {
config.put("temp_path_base", temp_path_base);
config.put("timestamp_start", config.get("timestamp_start"));
config.put("timestamp_real", latestVersion.toString());
config.put("timestamp_stop", latestVersion.toString());
Class<?> sourceClass =
Class.forName("org.gestore.plugin.source." + config.get("source") + "Source");
Method process_data = sourceClass.getMethod("process", Hashtable.class, FileSystem.class);
Object processor = sourceClass.newInstance();
Object retVal;
try {
retVal = process_data.invoke(processor, config, fs);
} catch (InvocationTargetException E) {
Throwable exception = E.getTargetException();
logger.error("Unable to call method in child class: " + exception.toString());
exception.printStackTrace(System.out);
unlock(lockName);
return null;
}
FileStatus[] files = (FileStatus[]) retVal;
if (files == null) {
logger.error("Error getting files, no files returned");
return null;
}
for (FileStatus file : files) {
Path cur_file = file.getPath();
Path cur_local_path = new Path(temp_path_base + config.get("file_id"));
String suffix = getSuffix(config.get("file_id"), cur_file.getName());
cur_local_path = cur_local_path.suffix(suffix);
Path res_path = new Path(new String(final_result + suffix));
logger.debug("Moving file" + cur_file.toString() + " to " + res_path.toString());
if (config.get("copy").equals("true")) {
fs.moveFromLocalFile(cur_file, res_path);
} else {
fs.rename(cur_file, res_path);
}
}
config.put("full_file_name", final_result);
}
}
unlock(lockName);
return ret_path;
}
/**
* 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;
}