/**
* Checks if the map-reduce job has completed.
*
* @return true if the job completed, false otherwise.
* @throws IOException
*/
public boolean checkComplete() throws IOException {
JobID jobID = runningJob.getID();
if (runningJob.isComplete()) {
// delete job directory
final String jobdir = jobconf.get(JOB_DIR_LABEL);
if (jobdir != null) {
final Path jobpath = new Path(jobdir);
jobpath.getFileSystem(jobconf).delete(jobpath, true);
}
if (runningJob.isSuccessful()) {
LOG.info("Job Complete(Succeeded): " + jobID);
} else {
LOG.info("Job Complete(Failed): " + jobID);
}
raidPolicyPathPairList.clear();
Counters ctrs = runningJob.getCounters();
if (ctrs != null) {
RaidNodeMetrics metrics = RaidNodeMetrics.getInstance(RaidNodeMetrics.DEFAULT_NAMESPACE_ID);
if (ctrs.findCounter(Counter.FILES_FAILED) != null) {
long filesFailed = ctrs.findCounter(Counter.FILES_FAILED).getValue();
metrics.raidFailures.inc(filesFailed);
}
long slotSeconds =
ctrs.findCounter(JobInProgress.Counter.SLOTS_MILLIS_MAPS).getValue() / 1000;
metrics.raidSlotSeconds.inc(slotSeconds);
}
return true;
} else {
String report =
(" job "
+ jobID
+ " map "
+ StringUtils.formatPercent(runningJob.mapProgress(), 0)
+ " reduce "
+ StringUtils.formatPercent(runningJob.reduceProgress(), 0));
if (!report.equals(lastReport)) {
LOG.info(report);
lastReport = report;
}
TaskCompletionEvent[] events = runningJob.getTaskCompletionEvents(jobEventCounter);
jobEventCounter += events.length;
for (TaskCompletionEvent event : events) {
if (event.getTaskStatus() == TaskCompletionEvent.Status.FAILED) {
LOG.info(" Job " + jobID + " " + event.toString());
}
}
return false;
}
}
private void getTaskInfos() throws IOException, MalformedURLException {
int startIndex = 0;
while (true) {
TaskCompletionEvent[] taskCompletions = rj.getTaskCompletionEvents(startIndex);
if (taskCompletions == null || taskCompletions.length == 0) {
break;
}
boolean more = true;
boolean firstError = true;
for (TaskCompletionEvent t : taskCompletions) {
// For each task completion event, get the associated task id, job id
// and the logs
String taskId = t.getTaskAttemptId().getTaskID().toString();
String jobId = t.getTaskAttemptId().getJobID().toString();
if (firstError) {
console.printError("Examining task ID: " + taskId + " (and more) from job " + jobId);
firstError = false;
}
TaskInfo ti = taskIdToInfo.get(taskId);
if (ti == null) {
ti = new TaskInfo(jobId);
taskIdToInfo.put(taskId, ti);
}
// These tasks should have come from the same job.
assert (ti.getJobId() != null && ti.getJobId().equals(jobId));
String taskAttemptLogUrl =
ShimLoader.getHadoopShims()
.getTaskAttemptLogUrl(conf, t.getTaskTrackerHttp(), t.getTaskId());
if (taskAttemptLogUrl != null) {
ti.getLogUrls().add(taskAttemptLogUrl);
}
// If a task failed, fetch its error code (if available).
// Also keep track of the total number of failures for that
// task (typically, a task gets re-run up to 4 times if it fails.
if (t.getTaskStatus() != TaskCompletionEvent.Status.SUCCEEDED) {
String[] diags = rj.getTaskDiagnostics(t.getTaskAttemptId());
ti.setDiagnosticMesgs(diags);
if (ti.getErrorCode() == 0) {
ti.setErrorCode(extractErrorCode(diags));
}
Integer failAttempts = failures.get(taskId);
if (failAttempts == null) {
failAttempts = Integer.valueOf(0);
}
failAttempts = Integer.valueOf(failAttempts.intValue() + 1);
failures.put(taskId, failAttempts);
} else {
successes.add(taskId);
}
}
if (!more) {
break;
}
startIndex += taskCompletions.length;
}
}
