/**
* checks if jobs have completed and updates job and file index returns a list of failed files for
* restarting
*/
void checkJobs() throws IOException {
Iterator<Job> jobIter = jobIndex.keySet().iterator();
while (jobIter.hasNext()) {
Job job = jobIter.next();
try {
if (job.isComplete()) {
long slotSeconds =
job.getCounters().findCounter(JobInProgress.Counter.SLOTS_MILLIS_MAPS).getValue()
/ 1000;
RaidNodeMetrics.getInstance().blockFixSlotSeconds.inc(slotSeconds);
long filesSucceeded =
job.getCounters().findCounter(Counter.FILES_SUCCEEDED) != null
? job.getCounters().findCounter(Counter.FILES_SUCCEEDED).getValue()
: 0;
long filesFailed =
job.getCounters().findCounter(Counter.FILES_FAILED) != null
? job.getCounters().findCounter(Counter.FILES_FAILED).getValue()
: 0;
long filesNoAction =
job.getCounters().findCounter(Counter.FILES_NOACTION) != null
? job.getCounters().findCounter(Counter.FILES_NOACTION).getValue()
: 0;
int files = jobIndex.get(job).size();
if (job.isSuccessful()
&& (filesSucceeded + filesFailed + filesNoAction == ((long) files))) {
// job has processed all files
succeedJob(job, filesSucceeded, filesFailed);
} else {
failJob(job);
}
jobIter.remove();
} else {
LOG.info("Job " + job.getID() + "(" + job.getJobName() + " still running");
}
} catch (Exception e) {
LOG.error(StringUtils.stringifyException(e));
failJob(job);
try {
job.killJob();
} catch (Exception ee) {
LOG.error(StringUtils.stringifyException(ee));
}
jobIter.remove();
}
}
purgeFileIndex();
}
