def dispatch(filename):
# wait until FS says that the file exists
while not FileSystem.file_exists_new_shell(filename):
time.sleep(1)
job = Serialization.deserialize_object(filename)
job.compute()
def dispatch(filename):
# wait until FS says that the file exists
while not FileSystem.file_exists_new_shell(filename):
time.sleep(1)
job = Serialization.deserialize_object(filename)
job.compute()
def test_serialize_and_deserialize(self):
filename = "temp.bin"
obj = [1, 2, 3]
Serialization.serialize_object_overwrite(obj, filename)
obj2 = Serialization.deserialize_object(filename)
self.assertEqual(obj, obj2)
def test_serialize_and_deserialize(self):
filename = "temp.bin"
obj = [1, 2, 3]
Serialization.serialize_object_overwrite(obj, filename)
obj2 = Serialization.deserialize_object(filename)
self.assertEqual(obj, obj2)
def _get_max_wait_time_exceed_jobs(self):
names = []
current_time = time.time()
for job_name, job_time, _ in self.submitted_jobs:
# load job ressources
job_filename = self.get_job_filename(job_name)
job = Serialization.deserialize_object(job_filename)
if abs(current_time - job_time) > job.walltime:
names += [job_name]
return names
def _get_max_wait_time_exceed_jobs(self):
names = []
current_time = time.time()
for job_name, job_time, _ in self.submitted_jobs:
# load job ressources
job_filename = self.get_job_filename(job_name)
job = Serialization.deserialize_object(job_filename)
if abs(current_time - job_time) > job.walltime:
names += [job_name]
return names
def _wait_until_n_unfinished(self, desired_num_unfinished):
"""
Iteratively checks all non-finished jobs and updates whether they are
finished. Blocks until there are less or exactly desired_num_unfinished
unfinished jobs in the queue. Messages a "waiting for" info message
for the oldest job in the queue.
"""
# save all job list to file for reconstructing results later
self.save_all_job_list()
last_printed = self._get_oldest_job_in_queue()
logger.info("Waiting for %s and %d other jobs" %
(last_printed, self._get_num_unfinished_jobs() - 1))
while self._get_num_unfinished_jobs() > desired_num_unfinished:
oldest = self._get_oldest_job_in_queue()
if oldest != last_printed:
last_printed = oldest
logger.info(
"Waiting for %s and %d other jobs" %
(last_printed, self._get_num_unfinished_jobs() - 1))
# delete all finished jobs from internal list
i = 0
while i < len(self.submitted_jobs):
job_name = self.submitted_jobs[i][0]
if self._check_job_done(job_name):
del self.submitted_jobs[i]
# dont change i as it is now the index of the next element
else:
i += 1
# check for re-submissions
if self.batch_parameters.resubmit_on_timeout:
for job_name in self._get_max_wait_time_exceed_jobs():
# load job ressources
job_filename = self.get_job_filename(job_name)
job = Serialization.deserialize_object(job_filename)
logger.info("%s exceeded maximum waiting time of %dh" %
(job_name, job.walltime))
self._resubmit(job_name)
time.sleep(self.check_interval)
def _wait_until_n_unfinished(self, desired_num_unfinished):
"""
Iteratively checks all non-finished jobs and updates whether they are
finished. Blocks until there are less or exactly desired_num_unfinished
unfinished jobs in the queue. Messages a "waiting for" info message
for the oldest job in the queue.
"""
# save all job list to file for reconstructing results later
self.save_all_job_list()
last_printed = self._get_oldest_job_in_queue()
logger.info("Waiting for %s and %d other jobs" % (last_printed,
self._get_num_unfinished_jobs() - 1))
while self._get_num_unfinished_jobs() > desired_num_unfinished:
oldest = self._get_oldest_job_in_queue()
if oldest != last_printed:
last_printed = oldest
logger.info("Waiting for %s and %d other jobs" % (last_printed,
self._get_num_unfinished_jobs() - 1))
# delete all finished jobs from internal list
i = 0
while i < len(self.submitted_jobs):
job_name = self.submitted_jobs[i][0]
if self._check_job_done(job_name):
del self.submitted_jobs[i]
# dont change i as it is now the index of the next element
else:
i += 1
# check for re-submissions
if self.batch_parameters.resubmit_on_timeout:
for job_name in self._get_max_wait_time_exceed_jobs():
# load job ressources
job_filename = self.get_job_filename(job_name)
job = Serialization.deserialize_object(job_filename)
logger.info("%s exceeded maximum waiting time of %dh"
% (job_name, job.walltime))
self._resubmit(job_name)
time.sleep(self.check_interval)
def _resubmit(self, job_name):
new_job_name = self.create_job_name()
logger.info("Re-submitting under name %s" % new_job_name)
# remove from unfinished jobs list
for i in range(len(self.submitted_jobs)):
if self.submitted_jobs[i][0] == job_name:
del self.submitted_jobs[i]
break
# remove from all jobs list
for i in range(len(self.all_jobs)):
if self.all_jobs[i] == job_name:
del self.all_jobs[i]
break
# load job from disc and re-submit under new name
job_filename = self.get_job_filename(job_name)
wrapped_job = Serialization.deserialize_object(job_filename)
self.submit_wrapped_pbs_job(wrapped_job, new_job_name)
def _resubmit(self, job_name):
new_job_name = self.create_job_name()
logger.info("Re-submitting under name %s" % new_job_name)
# remove from unfinished jobs list
for i in range(len(self.submitted_jobs)):
if self.submitted_jobs[i][0] == job_name:
del self.submitted_jobs[i]
break
# remove from all jobs list
for i in range(len(self.all)):
if self.all_jobs[i] == job_name:
del self.all_jobs[i]
break
# load job from disc and re-submit under new name
job_filename = self.get_job_filename(job_name)
wrapped_job = Serialization.deserialize_object(job_filename)
self.submit_wrapped_pbs_job(wrapped_job, new_job_name)