def __pollJobTree(self):
childJobFileToParentJob, childCounts = {}, {}
updatedJobFiles, shellJobs = set(), set()
try:
parseJobFiles(getJobFileDirName(self.jobTreePath),
updatedJobFiles, childJobFileToParentJob,
childCounts, shellJobs)
failedJobs = [ job for job in updatedJobFiles | \
set(childCounts.keys()) \
if job.remainingRetryCount == 0 ]
self.curActiveJobs = set()
for job in updatedJobFiles:
self.curActiveJobs.add(job.getJobFileName())
self.failedJobs = max(len(failedJobs), self.failedJobs)
except:
self.curActiveJobs = set()
if len(self.prevActiveJobs) > 0 and len(self.curActiveJobs) > 0 and\
self.curActiveJobs == self.prevActiveJobs:
self.sameJobsTime += self.pollTime
else:
self.sameJobsTime = 0
self.prevActiveJobs = set(self.curActiveJobs)
def jobtree_is_finished(self, jobtree_path):
"""
See if this jobTree has finished before. Code extracted from the jobTree repo.
"""
childJobFileToParentJob, childCounts, updatedJobFiles, shellJobs = {}, {}, set(), set()
parseJobFiles(getJobFileDirName(jobtree_path), updatedJobFiles, childJobFileToParentJob, childCounts, shellJobs)
return len(updatedJobFiles) == 0
def createFirstJob(command, config, memory=None, cpu=None, time=sys.maxint):
"""Adds the first job to to the jobtree.
"""
logger.info("Adding the first job")
if memory == None or memory == sys.maxint:
memory = float(config.attrib["default_memory"])
if cpu == None or cpu == sys.maxint:
cpu = float(config.attrib["default_cpu"])
job = Job(command=command, memory=memory, cpu=cpu,
tryCount=int(config.attrib["try_count"]), jobDir=getJobFileDirName(config.attrib["job_tree"]))
job.write()
logger.info("Added the first job")
def reloadJobTree(jobTree):
"""Load the job tree from a dir.
"""
logger.info("The job tree appears to already exist, so we'll reload it")
assert os.path.isfile(getConfigFileName(jobTree)) #A valid job tree must contain the config file
assert os.path.isfile(getEnvironmentFileName(jobTree)) #A valid job tree must contain a pickle file which encodes the path environment of the job
assert os.path.isdir(getJobFileDirName(jobTree)) #A job tree must have a directory of jobs.
config = ET.parse(getConfigFileName(jobTree)).getroot()
config.attrib["log_level"] = getLogLevelString()
writeConfig(config) #This updates the on disk config file with the new logging setting
batchSystem = loadTheBatchSystem(config)
logger.info("Reloaded the jobtree")
return config, batchSystem
def createJobTree(options):
logger.info("Starting to create the job tree setup for the first time")
options.jobTree = absSymPath(options.jobTree)
os.mkdir(options.jobTree)
os.mkdir(getJobFileDirName(options.jobTree))
config = ET.Element("config")
config.attrib["log_level"] = getLogLevelString()
config.attrib["job_tree"] = options.jobTree
config.attrib["parasol_command"] = options.parasolCommand
config.attrib["try_count"] = str(int(options.retryCount) + 1)
config.attrib["max_job_duration"] = str(float(options.maxJobDuration))
config.attrib["batch_system"] = options.batchSystem
config.attrib["job_time"] = str(float(options.jobTime))
config.attrib["max_log_file_size"] = str(int(options.maxLogFileSize))
config.attrib["default_memory"] = str(int(options.defaultMemory))
config.attrib["default_cpu"] = str(int(options.defaultCpu))
config.attrib["max_cpus"] = str(int(options.maxCpus))
config.attrib["max_memory"] = str(int(options.maxMemory))
config.attrib["max_threads"] = str(int(options.maxThreads))
if options.bigBatchSystem != None:
config.attrib["big_batch_system"] = options.bigBatchSystem
config.attrib["big_memory_threshold"] = str(
int(options.bigMemoryThreshold))
config.attrib["big_cpu_threshold"] = str(int(options.bigCpuThreshold))
config.attrib["big_max_cpus"] = str(int(options.bigMaxCpus))
config.attrib["big_max_memory"] = str(int(options.bigMaxMemory))
if options.stats:
config.attrib["stats"] = ""
#Load the batch system.
batchSystem = loadTheBatchSystem(config, options)
logger.info("Loaded the batch system %s" % batchSystem)
#Set the parameters determining the polling frequency of the system.
config.attrib["rescue_jobs_frequency"] = str(
float(batchSystem.getRescueJobFrequency()))
if options.rescueJobsFrequency != None:
config.attrib["rescue_jobs_frequency"] = str(
float(options.rescueJobsFrequency))
writeConfig(config)
logger.info("Finished the job tree setup")
return config, batchSystem
def createJobTree(options):
logger.info("Starting to create the job tree setup for the first time")
options.jobTree = absSymPath(options.jobTree)
os.mkdir(options.jobTree)
os.mkdir(getJobFileDirName(options.jobTree))
config = ET.Element("config")
config.attrib["log_level"] = getLogLevelString()
config.attrib["job_tree"] = options.jobTree
config.attrib["parasol_command"] = options.parasolCommand
config.attrib["try_count"] = str(int(options.retryCount) + 1)
config.attrib["max_job_duration"] = str(float(options.maxJobDuration))
config.attrib["batch_system"] = options.batchSystem
config.attrib["job_time"] = str(float(options.jobTime))
config.attrib["max_log_file_size"] = str(int(options.maxLogFileSize))
config.attrib["default_memory"] = str(int(options.defaultMemory))
config.attrib["default_cpu"] = str(int(options.defaultCpu))
config.attrib["max_cpus"] = str(int(options.maxCpus))
config.attrib["max_memory"] = str(int(options.maxMemory))
config.attrib["max_threads"] = str(int(options.maxThreads))
if options.bigBatchSystem != None:
config.attrib["big_batch_system"] = options.bigBatchSystem
config.attrib["big_memory_threshold"] = str(int(options.bigMemoryThreshold))
config.attrib["big_cpu_threshold"] = str(int(options.bigCpuThreshold))
config.attrib["big_max_cpus"] = str(int(options.bigMaxCpus))
config.attrib["big_max_memory"] = str(int(options.bigMaxMemory))
if options.stats:
config.attrib["stats"] = ""
#Load the batch system.
batchSystem = loadTheBatchSystem(config)
#Set the parameters determining the polling frequency of the system.
config.attrib["rescue_jobs_frequency"] = str(float(batchSystem.getRescueJobFrequency()))
if options.rescueJobsFrequency != None:
config.attrib["rescue_jobs_frequency"] = str(float(options.rescueJobsFrequency))
writeConfig(config)
logger.info("Finished the job tree setup")
return config, batchSystem
def __pollJobTree(self):
childJobFileToParentJob, childCounts = {}, {}
updatedJobFiles, shellJobs = set(), set()
try:
parseJobFiles(getJobFileDirName(self.jobTreePath), updatedJobFiles,
childJobFileToParentJob, childCounts, shellJobs)
failedJobs = [ job for job in updatedJobFiles | \
set(childCounts.keys()) \
if job.remainingRetryCount == 0 ]
self.curActiveJobs = set()
for job in updatedJobFiles:
self.curActiveJobs.add(job.getJobFileName())
self.failedJobs = max(len(failedJobs), self.failedJobs)
except:
self.curActiveJobs = set()
if len(self.prevActiveJobs) > 0 and len(self.curActiveJobs) > 0 and\
self.curActiveJobs == self.prevActiveJobs:
self.sameJobsTime += self.pollTime
else:
self.sameJobsTime = 0
self.prevActiveJobs = set(self.curActiveJobs)
def main():
"""Reports the state of the job tree.
"""
##########################################
#Construct the arguments.
##########################################
parser = getBasicOptionParser(
"usage: %prog [--jobTree] JOB_TREE_DIR [options]", "%prog 0.1")
parser.add_option(
"--jobTree",
dest="jobTree",
help=
"Directory containing the job tree. The jobTree location can also be specified as the argument to the script. default=%default",
default='./jobTree')
parser.add_option(
"--verbose",
dest="verbose",
action="store_true",
help=
"Print loads of information, particularly all the log files of jobs that failed. default=%default",
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=%default",
default=False)
options, args = parseBasicOptions(parser)
logger.info("Parsed arguments")
if len(sys.argv) == 1:
parser.print_help()
sys.exit(0)
assert len(args) <= 1 #Only jobtree may be specified as argument
if len(args) == 1: #Allow jobTree directory as arg
options.jobTree = args[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(getConfigFileName(
options.jobTree)) #A valid job tree must contain the config gile
assert os.path.isdir(getJobFileDirName(
options.jobTree)) #A job tree must have a directory of jobs.
##########################################
#Survey the status of the job and report.
##########################################
childJobFileToParentJob, childCounts, updatedJobFiles, shellJobs = {}, {}, set(
), set()
parseJobFiles(getJobFileDirName(options.jobTree), updatedJobFiles,
childJobFileToParentJob, childCounts, shellJobs)
failedJobs = [
job for job in updatedJobFiles | set(childCounts.keys())
if job.remainingRetryCount == 0
]
print "There are %i active jobs, %i parent jobs with children, %i totally failed jobs and %i empty jobs (i.e. finished but not cleaned up) currently in job tree: %s" % \
(len(updatedJobFiles), len(childCounts), len(failedJobs), len(shellJobs), options.jobTree)
if options.verbose: #Verbose currently means outputting the files that have failed.
for job in failedJobs:
if os.path.isfile(job.getLogFileName()):
print "Log file of failed job: %s" % job.getLogFileName()
logFile(job.getLogFileName(), logger.critical)
else:
print "Log file for job %s is not present" % job.getJobFileName(
)
if len(failedJobs) == 0:
print "There are no failed jobs to report"
if (len(updatedJobFiles) +
len(childCounts)) != 0 and options.failIfNotComplete:
sys.exit(1)
def main():
"""Reports the state of the job tree.
"""
##########################################
#Construct the arguments.
##########################################
parser = getBasicOptionParser("usage: %prog [--jobTree] JOB_TREE_DIR [options]", "%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=%default",
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=%default",
default=False)
options, args = parseBasicOptions(parser)
logger.info("Parsed arguments")
if len(sys.argv) == 1:
parser.print_help()
sys.exit(0)
assert len(args) <= 1 #Only jobtree may be specified as argument
if len(args) == 1: #Allow jobTree directory as arg
options.jobTree = args[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(getConfigFileName(options.jobTree)) #A valid job tree must contain the config gile
assert os.path.isdir(getJobFileDirName(options.jobTree)) #A job tree must have a directory of jobs.
##########################################
#Survey the status of the job and report.
##########################################
childJobFileToParentJob, childCounts, updatedJobFiles, shellJobs = {}, {}, set(), set()
parseJobFiles(getJobFileDirName(options.jobTree), updatedJobFiles, childJobFileToParentJob, childCounts, shellJobs)
failedJobs = [ job for job in updatedJobFiles | set(childCounts.keys()) if job.remainingRetryCount == 0 ]
print "There are %i active jobs, %i parent jobs with children, %i totally failed jobs and %i empty jobs (i.e. finished but not cleaned up) currently in job tree: %s" % \
(len(updatedJobFiles), len(childCounts), len(failedJobs), len(shellJobs), options.jobTree)
if options.verbose: #Verbose currently means outputting the files that have failed.
for job in failedJobs:
if os.path.isfile(job.getLogFileName()):
print "Log file of failed job: %s" % job.getLogFileName()
logFile(job.getLogFileName(), logger.critical)
else:
print "Log file for job %s is not present" % job.getJobFileName()
if len(failedJobs) == 0:
print "There are no failed jobs to report"
if (len(updatedJobFiles) + len(childCounts)) != 0 and options.failIfNotComplete:
sys.exit(1)
