def recover_job_descriptors(self):
root = self.job_pool.journal_root
if root is None:
return
for job_id in os.listdir(root):
try:
job_dir = os.path.join(root, job_id)
result_path = os.path.join(job_dir, 'result')
if os.path.exists(result_path):
with open(result_path, 'r') as result_file:
result = simplejson.load(result_file, object_hook=json_decode_object_hook)
else:
result = None
journal_path = os.path.join(job_dir, 'task_journal')
journal_file = open(journal_path, 'rb')
record_type, root_task_descriptor_length = RECORD_HEADER_STRUCT.unpack(journal_file.read(RECORD_HEADER_STRUCT.size))
root_task_descriptor_string = journal_file.read(root_task_descriptor_length)
assert record_type == 'T'
assert len(root_task_descriptor_string) == root_task_descriptor_length
root_task_descriptor = simplejson.loads(root_task_descriptor_string, object_hook=json_decode_object_hook)
root_task = build_taskpool_task_from_descriptor(root_task_descriptor, None)
# FIXME: Get the job pool to create this job, because it has access to the scheduler queue and task failure investigator.
# FIXME: Store job options somewhere for recovered job.
job = Job(job_id, root_task, job_dir, JOB_RECOVERED, self.job_pool, {}, journal=False)
root_task.job = job
if result is not None:
with job._lock:
job.completed(result)
self.job_pool.add_job(job)
# Adding the job to the job pool should add the root task.
#self.task_pool.add_task(root_task)
if result is None:
self.load_other_tasks_defer(job, journal_file)
ciel.log.error('Recovered job %s' % job_id, 'RECOVERY', logging.INFO, False)
ciel.log.error('Recovered task %s for job %s' % (root_task.task_id, job_id), 'RECOVERY', logging.INFO, False)
else:
journal_file.close()
ciel.log.error('Found information about job %s' % job_id, 'RECOVERY', logging.INFO, False)
except:
# We have lost critical data for the job, so we must fail it.
ciel.log.error('Error recovering job %s' % job_id, 'RECOVERY', logging.ERROR, True)
self.job_pool.add_failed_job(job_id)
def recover_job_descriptors(self):
root = self.job_pool.journal_root
if root is None:
return
for job_id in os.listdir(root):
try:
job_dir = os.path.join(root, job_id)
result_path = os.path.join(job_dir, 'result')
if os.path.exists(result_path):
with open(result_path, 'r') as result_file:
result = simplejson.load(result_file, object_hook=json_decode_object_hook)
else:
result = None
journal_path = os.path.join(job_dir, 'task_journal')
journal_file = open(journal_path, 'rb')
record_type, root_task_descriptor_length = RECORD_HEADER_STRUCT.unpack(journal_file.read(RECORD_HEADER_STRUCT.size))
root_task_descriptor_string = journal_file.read(root_task_descriptor_length)
assert record_type == 'T'
assert len(root_task_descriptor_string) == root_task_descriptor_length
root_task_descriptor = simplejson.loads(root_task_descriptor_string, object_hook=json_decode_object_hook)
root_task = build_taskpool_task_from_descriptor(root_task_descriptor, None)
# FIXME: Get the job pool to create this job, because it has access to the scheduler queue and task failure investigator.
# FIXME: Store job options somewhere for recovered job.
job = Job(job_id, root_task, job_dir, JOB_RECOVERED, self.job_pool, {}, journal=False)
root_task.job = job
if result is not None:
with job._lock:
job.completed(result)
self.job_pool.add_job(job)
# Adding the job to the job pool should add the root task.
#self.task_pool.add_task(root_task)
if result is None:
self.load_other_tasks_defer(job, journal_file)
ciel.log.error('Recovered job %s' % job_id, 'RECOVERY', logging.INFO, False)
ciel.log.error('Recovered task %s for job %s' % (root_task.task_id, job_id), 'RECOVERY', logging.INFO, False)
else:
journal_file.close()
ciel.log.error('Found information about job %s' % job_id, 'RECOVERY', logging.INFO, False)
except:
# We have lost critical data for the job, so we must fail it.
ciel.log.error('Error recovering job %s' % job_id, 'RECOVERY', logging.ERROR, True)
self.job_pool.add_failed_job(job_id)
def load_other_tasks_for_job(self, job, journal_file):
'''
Process a the task journal for a recovered job.
'''
try:
while True:
record_header = journal_file.read(RECORD_HEADER_STRUCT.size)
if len(record_header) != RECORD_HEADER_STRUCT.size:
ciel.log.error('Journal entry truncated for job %s' % job.id, 'RECOVERY', logging.WARNING, False)
# XXX: Need to truncate the journal file.
break
record_type, record_length = RECORD_HEADER_STRUCT.unpack(record_header)
record_string = journal_file.read(record_length)
if len(record_string) != record_length:
ciel.log.error('Journal entry truncated for job %s' % job.id, 'RECOVERY', logging.WARNING, False)
# XXX: Need to truncate the journal file.
break
rec = simplejson.loads(record_string, object_hook=json_decode_object_hook)
if record_type == 'R':
job.task_graph.publish(rec['ref'])
elif record_type == 'T':
task_id = rec['task_id']
parent_task = job.task_graph.get_task(rec['parent'])
task = build_taskpool_task_from_descriptor(rec, parent_task)
task.job = job
task.parent.children.append(task)
ciel.log.error('Recovered task %s for job %s' % (task_id, job.id), 'RECOVERY', logging.INFO, False)
job.task_graph.spawn(task)
else:
ciel.log.error('Got invalid record type in job %s' % job.id, 'RECOVERY', logging.WARNING, False)
except:
ciel.log.error('Error recovering task_journal for job %s' % job.id, 'RECOVERY', logging.WARNING, True)
finally:
journal_file.close()
job.restart_journalling()
if job.state == JOB_ACTIVE:
ciel.log.error('Restarting recovered job %s' % job.id, 'RECOVERY', logging.INFO)
def load_other_tasks_for_job(self, job, journal_file):
'''
Process a the task journal for a recovered job.
'''
try:
while True:
record_header = journal_file.read(RECORD_HEADER_STRUCT.size)
if len(record_header) != RECORD_HEADER_STRUCT.size:
ciel.log.error('Journal entry truncated for job %s' % job.id, 'RECOVERY', logging.WARNING, False)
# XXX: Need to truncate the journal file.
break
record_type, record_length = RECORD_HEADER_STRUCT.unpack(record_header)
record_string = journal_file.read(record_length)
if len(record_string) != record_length:
ciel.log.error('Journal entry truncated for job %s' % job.id, 'RECOVERY', logging.WARNING, False)
# XXX: Need to truncate the journal file.
break
rec = simplejson.loads(record_string, object_hook=json_decode_object_hook)
if record_type == 'R':
job.task_graph.publish(rec['ref'])
elif record_type == 'T':
task_id = rec['task_id']
parent_task = job.task_graph.get_task(rec['parent'])
task = build_taskpool_task_from_descriptor(rec, parent_task)
task.job = job
task.parent.children.append(task)
ciel.log.error('Recovered task %s for job %s' % (task_id, job.id), 'RECOVERY', logging.INFO, False)
job.task_graph.spawn(task)
else:
ciel.log.error('Got invalid record type in job %s' % job.id, 'RECOVERY', logging.WARNING, False)
except:
ciel.log.error('Error recovering task_journal for job %s' % job.id, 'RECOVERY', logging.WARNING, True)
finally:
journal_file.close()
job.restart_journalling()
if job.state == JOB_ACTIVE:
ciel.log.error('Restarting recovered job %s' % job.id, 'RECOVERY', logging.INFO)