def cancel(self):
darelogger.debug("Terminate Pilot Compute/Data Service")
self.compute_data_service.cancel()
try:
self.pilot_data_service.cancel()
self.pilot_compute_service.cancel()
except:
pass
def quit(self, message=None):
if message:
darelogger.debug(message)
darelogger.debug("Terminating steps")
for step, thread in list(self.step_threads.items()):
darelogger.debug("Stoppping step %s" % step)
thread._Thread__stop()
darelogger.debug("Terminating Pilot Compute/Data Service")
try:
self.compute_data_service.cancel()
self.pilot_data_service.cancel()
self.pilot_compute_service.cancel()
except:
pass
def quit(self, message=None):
if message:
darelogger.debug(message)
darelogger.debug("Terminating steps")
for step, thread in self.step_threads.items():
darelogger.debug("Stoppping step %s" % step)
thread._Thread__stop()
darelogger.debug("Terminating Pilot Compute/Data Service")
try:
self.compute_data_service.cancel()
self.pilot_data_service.cancel()
self.pilot_compute_service.cancel()
except:
pass
def main():
# parse conf files
#parser = optparse.OptionParser()
#parser.add_option("-c", "--conf_job", dest="conf_job", help="job configuration file")
#(options, args) = parser.parse_args()
#confjob = options.conf_job
if (len(sys.argv)> 1):
conffile = sys.argv[1]
else:
raise Exception, "missing dare configurtion file"
darelogger.debug("starting DARE")
try:
dare = DareManager(conffile)
#dare.start()
except KeyboardInterrupt:
dare.cancel()
darelogger.debug("DARE Exec Done")
def run_step(self, step_id):
#self.step_run_lock.acquire()
#job started update status
this_su = self.workflow.step_units_repo[step_id].UnitInfo
self.updater.update_status(
this_su['dare_web_id'],
"%s in step %s" % ('Running', this_su['name']))
darelogger.info(" Started running %s " % step_id)
jobs = []
job_start_times = {}
job_states = {}
NUMBER_JOBS = len(
self.workflow.step_units_repo[step_id].UnitInfo['compute_units'])
for cu_id in self.workflow.step_units_repo[step_id].UnitInfo[
'compute_units']:
compute_unit_desc = self.workflow.compute_units_repo[cu_id]
input_dus = compute_unit_desc.pop('input_data_units')
output_dus = compute_unit_desc.pop('output_data_units')
input_data_units = []
for du_id in input_dus:
input_data_units.append(
self.compute_data_service.submit_data_unit(
self.workflow.data_units_repo[du_id]))
output_data_units = []
for du_id in output_dus:
output_data_units.append(
self.compute_data_service.submit_data_unit(
self.workflow.data_units_repo[du_id]))
compute_unit_desc["input_data"] = [
du.get_url() for du in input_data_units
]
compute_unit_desc["output_data"] = [{
du.get_url(): ['std*']
} for du in output_data_units]
compute_unit = self.compute_data_service.submit_compute_unit(
compute_unit_desc)
darelogger.info("Compute Unit: Description: \n%s" %
(str(self.workflow.compute_units_repo[cu_id])))
jobs.append(compute_unit)
job_start_times[compute_unit] = time.time()
job_states[compute_unit] = compute_unit.get_state()
darelogger.debug(
"************************ All Jobs submitted ************************"
)
while 1:
finish_counter = 0
result_map = {}
for i in range(0, NUMBER_JOBS):
old_state = job_states[jobs[i]]
state = jobs[i].get_state()
if state in result_map == False:
result_map[state] = 0
result_map[state] = result_map.get(state, 0) + 1
#print "counter: " + str(i) + " job: " + str(jobs[i]) + " state: " + state
if old_state != state:
darelogger.debug("Job " + str(jobs[i]) +
" changed from: " + old_state + " to " +
state)
if old_state != state and self.has_finished(state) == True:
darelogger.info("%s step Job: " %
(self.workflow.step_units_repo[step_id].
UnitInfo['name']) + str(jobs[i]) +
" Runtime: " +
str(time.time() -
job_start_times[jobs[i]]) + " s.")
if self.has_finished(state) == True:
finish_counter = finish_counter + 1
job_states[jobs[i]] = state
darelogger.debug("Current states: " + str(result_map))
time.sleep(5)
if finish_counter == NUMBER_JOBS:
break
self.workflow.step_units_repo[step_id].set_status(StepUnitStates.Done)
#self.compute_data_service.wait()
darelogger.debug(" Compute jobs for step %s complete" % step_id)
#runtime = time.time()-starttime
#all jobs done update status
self.updater.update_status(this_su['dare_web_id'],
"%s is Done" % this_su['name'])
def run_step(self, step_id):
#self.step_run_lock.acquire()
#job started update status
this_su = self.workflow.step_units_repo[step_id].UnitInfo
self.updater.update_status(this_su['dare_web_id'], "%s in step %s" % ('Running', this_su['name']))
darelogger.info(" Started running %s " % step_id)
jobs = []
job_start_times = {}
job_states = {}
NUMBER_JOBS = len(self.workflow.step_units_repo[step_id].UnitInfo['compute_units'])
for cu_id in self.workflow.step_units_repo[step_id].UnitInfo['compute_units']:
compute_unit_desc = self.workflow.compute_units_repo[cu_id]
input_dus = compute_unit_desc.pop('input_data_units')
output_dus = compute_unit_desc.pop('output_data_units')
input_data_units = []
for du_id in input_dus:
input_data_units.append(self.compute_data_service.submit_data_unit(self.workflow.data_units_repo[du_id]))
output_data_units = []
for du_id in output_dus:
output_data_units.append(self.compute_data_service.submit_data_unit(self.workflow.data_units_repo[du_id]))
compute_unit_desc["input_data"] = [du.get_url() for du in input_data_units]
compute_unit_desc["output_data"] = [{du.get_url(): ['std*']} for du in output_data_units]
compute_unit = self.compute_data_service.submit_compute_unit(compute_unit_desc)
darelogger.info("Compute Unit: Description: \n%s" % (str(self.workflow.compute_units_repo[cu_id])))
jobs.append(compute_unit)
job_start_times[compute_unit] = time.time()
job_states[compute_unit] = compute_unit.get_state()
darelogger.debug("************************ All Jobs submitted ************************")
while 1:
finish_counter = 0
result_map = {}
for i in range(0, NUMBER_JOBS):
old_state = job_states[jobs[i]]
state = jobs[i].get_state()
if state in result_map == False:
result_map[state] = 0
result_map[state] = result_map.get(state, 0) + 1
#print "counter: " + str(i) + " job: " + str(jobs[i]) + " state: " + state
if old_state != state:
darelogger.debug("Job " + str(jobs[i]) + " changed from: " + old_state + " to " + state)
if old_state != state and self.has_finished(state) == True:
darelogger.info("%s step Job: " % (self.workflow.step_units_repo[step_id].UnitInfo['name']) + str(jobs[i]) + " Runtime: " + str(time.time() - job_start_times[jobs[i]]) + " s.")
if self.has_finished(state) == True:
finish_counter = finish_counter + 1
job_states[jobs[i]] = state
darelogger.debug("Current states: " + str(result_map))
time.sleep(5)
if finish_counter == NUMBER_JOBS:
break
self.workflow.step_units_repo[step_id].set_status(StepUnitStates.Done)
#self.compute_data_service.wait()
darelogger.debug(" Compute jobs for step %s complete" % step_id)
#runtime = time.time()-starttime
#all jobs done update status
self.updater.update_status(this_su['dare_web_id'], "%s is Done" % this_su['name'])
