sj_ec2.submit_job(jd_ec2)
jobs_ec2.append(sj_ec2)
number_started_jobs = number_started_jobs + 1
for i in range(0, NUMBER_JOBS_GRID):
print "Start job no.: " +str(number_started_jobs + 1) + " on Grid"
sj_tg = bigjob.subjob(advert_host)
jd.output = "stdout_tg.txt."+str(number_started_jobs+1)
jd.error = "stderr_tg.txt."+str(number_started_jobs+1)
sj_tg.submit_job(bj_tg.pilot_url, jd)
jobs_tg.append(sj_tg)
number_started_jobs = number_started_jobs + 1
for i in range(0, NUMBER_JOBS_CONDOR):
print "Start job no.: " +str(number_started_jobs + 1) + " on Condor"
sj_condor = bigjob_condor.subjob(bigjob=bj_condor)
jd_condor.output = "stdout_condor.txt."+str(number_started_jobs+1)
jd_condor.error = "stderr_condor.txt."+str(number_started_jobs+1)
sj_condor.submit_job(jd_condor)
jobs_condor.append(sj_condor)
number_started_jobs = number_started_jobs + 1
""" Monitor Progress """
while True:
try:
print "*******************************************************************************************************"
print "Pilot Job/BigJob URL: " + bj_ec2.pilot_url + " State: " + str(bj_ec2.get_state_detail()) + " Time since launch: " + str(time.time()-start)
print "Pilot Job/BigJob URL: " + bj_nimbus.pilot_url + " State: " + str(bj_nimbus.get_state_detail()) + " Time since launch: " + str(time.time()-start)
print "Pilot Job/BigJob URL: " + bj_tg.pilot_url + " State: " + str(bj_tg.get_state_detail()) + " Time since launch: " + str(time.time()-start)
print "Pilot Job/BigJob URL: " + bj_condor.pilot_url + " State: " + str(bj_condor.get_state_detail()) + " Time since launch: " + str(time.time()-start)
jd.arguments = ["NPT.conf"]
jd.working_directory = "/work/lukas/saga/bigjob/data"
jd.output = "namd.$(CLUSTER).$(PROCESS).$(NODE).out"
jd.error = "namd.$(CLUSTER).$(PROCESS).$(NODE).err"
attr = open(CONDOR_BIN + "/condor_attr", "w")
attr.write("universe = parallel\n")
attr.write("machine_count = 2\n") # number of nodes (not cores)
attr.write("+WantParallelSchedulingGroups = True\n") # to avoid running on nodes from different clusters
attr.close()
jobs = []
for i in range (0, NUMBER_JOBS):
print "Start job no.: " + str(i)
print time.ctime()
sj = bigjob_condor.subjob(bigjob=bj)
sj.submit_job(jd)
jobs.append(sj)
# busy wait for completion
sj = jobs.pop()
while 1:
try:
state = str(sj.get_state())
print "state: " + state
if state=="Failed" or state=="Done" or state=="Canceled" or state=="Suspended":
if len(jobs) > 0:
sj = jobs.pop()
continue
else:
break
