def main():
"""Reports the state of the job tree.
"""
##########################################
#Construct the arguments.
##########################################
parser = getBasicOptionParser("usage: %prog [options] \nThe colours returned indicate the state of the job.\n\
\twhite: job has not been started yet\n\
\tgrey: job is issued to batch system\n\
\tred: job failed\n\
\tblue: job has children currently being processed\n\
\tblack: job has finished and will be processed (transient state)\n\
\tdead: job is totally finished and is awaiting deletion (transient state)", "%prog 0.1")
parser.add_option("--jobTree", dest="jobTree",
help="Directory containing the job tree")
parser.add_option("--verbose", dest="verbose", action="store_true",
help="Print loads of information, particularly all the log files of errors",
default=False)
parser.add_option("--failIfNotComplete", dest="failIfNotComplete", action="store_true",
help="Return exit value of 1 if job tree jobs not all completed",
default=False)
options, args = parseBasicOptions(parser)
logger.info("Parsed arguments")
assert len(args) == 0
if len(sys.argv) == 1:
parser.print_help()
sys.exit(0)
##########################################
#Do some checks.
##########################################
logger.info("Checking if we have files for job tree")
assert options.jobTree != None
assert os.path.isdir(options.jobTree) #The given job dir tree must exist.
assert os.path.isfile(os.path.join(options.jobTree, "config.xml")) #A valid job tree must contain the config gile
assert os.path.isdir(os.path.join(options.jobTree, "jobs")) #A job tree must have a directory of jobs.
assert os.path.isdir(os.path.join(options.jobTree, "tempDirDir")) #A job tree must have a directory of temporary directories (for jobs to make temp files in).
assert os.path.isdir(os.path.join(options.jobTree, "logFileDir")) #A job tree must have a directory of log files.
assert os.path.isdir(os.path.join(options.jobTree, "slaveLogFileDir")) #A job tree must have a directory of slave log files.
##########################################
#Read the total job number
##########################################
config = ET.parse(os.path.join(options.jobTree, "config.xml")).getroot()
##########################################
#Survey the status of the job and report.
##########################################
colours = {}
jobFiles = TempFileTree(config.attrib["job_file_dir"]).listFiles()
if len(jobFiles) > 0:
logger.info("Collating the colours of the job tree")
for absFileName in jobFiles:
job = parseJobFile(absFileName)
if job != None:
if not colours.has_key(job.attrib["colour"]):
colours[job.attrib["colour"]] = 0
colours[job.attrib["colour"]] += 1
else:
logger.info("There are no jobs to collate")
print "There are %i jobs currently in job tree: %s" % \
(len(jobFiles), options.jobTree)
for colour in colours.keys():
print "\tColour: %s, number of jobs: %s" % (colour, colours[colour])
if options.verbose: #Verbose currently means outputting the files that have failed.
for absFileName in jobFiles:
job = parseJobFile(absFileName)
if job != None:
if job.attrib["colour"] == "red":
if os.path.isfile(job.attrib["log_file"]):
def fn(string):
print string
logFile(job.attrib["log_file"], fn)
else:
logger.info("Log file for job %s is not present" % job.attrib["file"])
if len(jobFiles) != 0 and options.failIfNotComplete:
sys.exit(1)
def mainLoop(config, batchSystem):
"""This is the main loop from which jobs are issued and processed.
"""
waitDuration = float(config.attrib["wait_duration"])
assert waitDuration >= 0
rescueJobsFrequency = float(config.attrib["rescue_jobs_frequency"])
maxJobDuration = float(config.attrib["max_job_duration"])
assert maxJobDuration >= 0
logger.info("Got parameters, wait duration %s, rescue jobs frequency: %s max job duration: %s" % \
(waitDuration, rescueJobsFrequency, maxJobDuration))
#Kill any jobs on the batch system queue from the last time.
assert len(batchSystem.getIssuedJobIDs()) == 0 #Batch system must start with no active jobs!
logger.info("Checked batch system has no running jobs and no updated jobs")
jobFiles = config.attrib["job_file_dir"].listFiles()
logger.info("Got a list of job files")
#Repair the job tree using any .old files
fixJobsList(config, jobFiles)
logger.info("Fixed the job files using any .old files")
#Get jobs that were running, or that had failed reset to 'white' status
restartFailedJobs(config, jobFiles)
logger.info("Reworked failed jobs")
updatedJobFiles = set() #Jobs whose status needs updating, either because they have finished, or because they need to be started.
for jobFile in jobFiles:
job = ET.parse(jobFile).getroot()
if job.attrib["colour"] not in ("grey", "blue"):
updatedJobFiles.add(jobFile)
logger.info("Got the active (non grey/blue) job files")
totalJobFiles = len(jobFiles) #Total number of job files we have.
jobIDsToJobsHash = {} #A hash of the currently running jobs ids, made by the batch system.
idealJobTime = float(config.attrib["job_time"])
assert idealJobTime > 0.0
maxIssuedJobs = int(config.attrib["max_jobs"]) #The maximum number of jobs to issue to the batch system
assert maxIssuedJobs >= 1
stats = config.attrib.has_key("stats")
if stats:
startTime = time.time()
startClock = getTotalCpuTime()
logger.info("Starting the main loop")
timeSinceJobsLastRescued = time.time() - rescueJobsFrequency + 100 #We hack it so that we rescue jobs after the first 100 seconds to get around an apparent parasol bug
while True:
if len(updatedJobFiles) > 0:
logger.debug("Built the jobs list, currently have %i job files, %i jobs to update and %i jobs currently issued" % (totalJobFiles, len(updatedJobFiles), len(jobIDsToJobsHash)))
jobsToIssue = []
for jobFile in list(updatedJobFiles):
job = ET.parse(jobFile).getroot()
assert job.attrib["colour"] not in ("grey", "blue")
if job.attrib["colour"] == "white": #Get ready to start the job
if len(jobIDsToJobsHash) < maxIssuedJobs:
logger.debug("Job: %s is being started" % job.attrib["file"])
updatedJobFiles.remove(job.attrib["file"])
#Reset the log files for the job.
open(job.attrib["slave_log_file"], 'w').close()
open(job.attrib["log_file"], 'w').close()
job.attrib["colour"] = "grey"
#writeJobs([ job ]) #Check point, do this before issuing job, so state is not read until issued
#issueJobs([ job ], jobIDsToJobsHash, batchSystem)
jobsToIssue.append(job)
else:
logger.debug("Job: %s is not being issued yet because we have %i jobs issued" % (job.attrib["file"], len(jobIDsToJobsHash)))
elif job.attrib["colour"] == "black": #Job has finished okay
logger.debug("Job: %s has finished okay" % job.attrib["file"])
#Deal with stats
if stats:
system("cat %s >> %s" % (job.attrib["stats"], config.attrib["stats"]))
open(job.attrib["stats"], 'w').close() #Reset the stats file
childCount = int(job.attrib["child_count"])
blackChildCount = int(job.attrib["black_child_count"])
assert childCount == blackChildCount #Has no currently running child jobs
#Launch any unborn children
unbornChildren = job.find("children")
unbornChild = unbornChildren.find("child")
if unbornChild != None: #We must give birth to the unborn children
logger.debug("Job: %s has children to schedule" % job.attrib["file"])
newChildren = []
while unbornChild != None:
cummulativeChildTime = float(unbornChild.attrib["time"])
newJob = createJob(unbornChild.attrib.copy(), job.attrib["file"], config)
totalJobFiles += 1
updatedJobFiles.add(newJob.attrib["file"])
newChildren.append(newJob)
unbornChildren.remove(unbornChild)
unbornChild = unbornChildren.find("child")
#This was code to aggregate groups of children into one job, but we don't do this now
#while cummulativeChildTime < idealJobTime and unbornChild != None: #Cummulate a series of children into each job as a stack of jobs (to balance cost of parellelism with cost of running stuff in serially).
# cummulativeChildTime += float(unbornChild.attrib["time"])
# ET.SubElement(newJob.find("followOns"), "followOn", unbornChild.attrib.copy())
# unbornChildren.remove(unbornChild)
# unbornChild = unbornChildren.find("child")
newJob.attrib["total_time"] = str(cummulativeChildTime)
updatedJobFiles.remove(job.attrib["file"])
job.attrib["child_count"] = str(childCount + len(newChildren))
job.attrib["colour"] = "blue" #Blue - has children running.
writeJobs([ job ] + newChildren ) #Check point
elif len(job.find("followOns").findall("followOn")) != 0: #Has another job
logger.debug("Job: %s has a new command that we can now issue" % job.attrib["file"])
##Reset the job run info
job.attrib["remaining_retry_count"] = config.attrib["retry_count"]
job.attrib["colour"] = "white"
##End resetting the job
writeJobs([ job ])
else: #Job has finished, so we can defer to any parent
logger.debug("Job: %s is now dead" % job.attrib["file"])
job.attrib["colour"] = "dead"
if job.attrib.has_key("parent"):
parent = ET.parse(job.attrib["parent"]).getroot()
assert parent.attrib["colour"] == "blue"
assert int(parent.attrib["black_child_count"]) < int(parent.attrib["child_count"])
parent.attrib["black_child_count"] = str(int(parent.attrib["black_child_count"]) + 1)
if int(parent.attrib["child_count"]) == int(parent.attrib["black_child_count"]):
parent.attrib["colour"] = "black"
assert parent.attrib["file"] not in updatedJobFiles
updatedJobFiles.add(parent.attrib["file"])
writeJobs([ job, parent ]) #Check point
else:
writeJobs([ job ])
elif job.attrib["colour"] == "red": #Job failed
logger.critical("Job: %s failed" % job.attrib["file"])
logger.critical("The log file of the failed job")
logFile(job.attrib["log_file"], logger.critical)
logger.critical("The log file of the slave for the failed job")
logFile(job.attrib["slave_log_file"], logger.critical) #We log the job log file in the main loop
#Checks
assert len(job.find("children").findall("child")) == 0
assert int(job.attrib["child_count"]) == int(job.attrib["black_child_count"])
remainingRetyCount = int(job.attrib["remaining_retry_count"])
if remainingRetyCount > 0: #Give it another try, maybe there is a bad node somewhere
job.attrib["remaining_retry_count"] = str(remainingRetyCount-1)
job.attrib["colour"] = "white"
logger.critical("Job: %s will be restarted, it has %s goes left" % (job.attrib["file"], job.attrib["remaining_retry_count"]))
writeJobs([ job ]) #Check point
else:
assert remainingRetyCount == 0
updatedJobFiles.remove(job.attrib["file"])
logger.critical("Job: %s is completely failed" % job.attrib["file"])
else:
logger.debug("Job: %s is already dead, we'll get rid of it" % job.attrib["file"])
assert job.attrib["colour"] == "dead"
updatedJobFiles.remove(job.attrib["file"])
totalJobFiles -= 1
deleteJob(job, config) #This could be done earlier, but I like it this way.
###End of for loop
writeJobs(jobsToIssue) #Check point, do this before issuing job, so state is not read until issued
issueJobs(jobsToIssue, jobIDsToJobsHash, batchSystem)
if len(jobIDsToJobsHash) == 0 and len(updatedJobFiles) == 0:
logger.info("Only failed jobs and their dependents (%i total) are remaining, so exiting." % totalJobFiles)
break
if len(updatedJobFiles) > 0:
updatedJobs = batchSystem.getUpdatedJobs() #Asks the batch system what jobs have been completed.
else:
updatedJobs = pauseForUpdatedJobs(batchSystem.getUpdatedJobs) #Asks the batch system what jobs have been completed.
for jobID in updatedJobs.keys(): #Runs through a map of updated jobs and there status,
result = updatedJobs[jobID]
if jobIDsToJobsHash.has_key(jobID):
if result == 0:
logger.debug("Batch system is reporting that the job %s ended sucessfully" % jobIDsToJobsHash[jobID])
else:
logger.critical("Batch system is reporting that the job %s failed with exit value %i" % (jobIDsToJobsHash[jobID], result))
processFinishedJob(jobID, result, updatedJobFiles, jobIDsToJobsHash)
else:
logger.info("A result seems to already have been processed: %i" % jobID) #T
if time.time() - timeSinceJobsLastRescued >= rescueJobsFrequency: #We only rescue jobs every N seconds
reissueOverLongJobs(updatedJobFiles, jobIDsToJobsHash, config, batchSystem)
logger.info("Reissued any over long jobs")
reissueMissingJobs(updatedJobFiles, jobIDsToJobsHash, batchSystem)
logger.info("Rescued any (long) missing jobs")
timeSinceJobsLastRescued = time.time()
#Going to sleep to let the job system catch up.
time.sleep(waitDuration)
if stats:
fileHandle = open(config.attrib["stats"], 'a')
fileHandle.write("<total_time time='%s' clock='%s'/></stats>" % (str(time.time() - startTime), str(getTotalCpuTime() - startClock)))
fileHandle.close()
logger.info("Finished the main loop")
return totalJobFiles #Returns number of failed jobs
