def add_Job(system, a_job):
if system.job_set.filter(acct_id=int(a_job.id)):
print "Job %s already exists" % a_job.id
return
owner = get_user(a_job.acct['account'], system)
#nr_bad_hosts = len(filter(lambda h: len(h.times) < 2,
# a_job.hosts.values()))
job_dict = {
'system': system,
'acct_id': a_job.id,
'owner': owner,
'queue': a_job.acct['queue'],
'queue_wait_time': a_job.start_time - a_job.acct['submission_time'],
'begin': a_job.start_time,
'end': a_job.end_time,
#'nr_bad_hots': nr_bad_hosts,
'nr_slots': a_job.acct['slots'],
'pe': a_job.acct['granted_pe'],
'failed': a_job.acct['failed'],
'exit_status': a_job.acct['exit_status'],
}
job_dict.update(job.JobAggregator(a_job).stats)
#newJob.nr_hosts = len(a_job.hosts)
try:
newJob = tsm_Job(**job_dict)
newJob.save()
except DatabaseError:
print "Error on job,", a_job.id
transaction.rollback()
return
hosts = map(lambda node: get_node(system, node), a_job.hosts.keys())
newJob.hosts = hosts
print "Added job:", a_job.id
def updateDatabase(system,a_job):
if system.job_set.filter(acct_id=int(a_job.id)):
print "Job %s already exists" % a_job.id
return
owner = get_user(a_job.acct['account'], system)
#nr_bad_hosts = len(filter(lambda h: len(h.times) < 2,
# a_job.hosts.values()))
job_dict = {
'system': system,
'acct_id': a_job.id,
'owner': owner,
'queue': a_job.acct['queue'],
'queue_wait_time': a_job.start_time - a_job.acct['submission_time'],
'begin': a_job.start_time,
'end': a_job.end_time,
#'nr_bad_hots': nr_bad_hosts,
'nr_slots': a_job.acct['slots'],
'pe': a_job.acct['granted_pe'],
'failed': a_job.acct['failed'],
'exit_status': a_job.acct['exit_status'],
}
job_dict.update(job.JobAggregator(a_job).stats)
#newJob.nr_hosts = len(a_job.hosts)
try:
newJob = tsm_Job(**job_dict)
newJob.save()
except DatabaseError,TypeError:
print "Error on job,", a_job.id
transaction.rollback()
return
def add_Job(system, a_job):
if system.job_set.filter(acct_id=int(a_job.id)):
print "Job %s already exists" % a_job.id
return
owner = get_user(a_job.acct['account'], system)
job_dict = {
'system': system,
'acct_id': a_job.id,
'owner': owner,
'begin': a_job.start_time,
'end': a_job.end_time,
'nr_slots': a_job.acct['slots'],
'pe': a_job.acct['granted_pe'],
'failed': a_job.acct['failed'],
'exit_status': a_job.acct['exit_status'],
}
#newJob.nr_hosts = len(a_job.hosts)
newJob = tsm_Job(**job_dict)
newJob.save()
hosts = map(lambda node: get_node(system, node), a_job.hosts.keys())
newJob.hosts = hosts
print "Added job:", a_job.id
def add_Job(system, a_job):
if system.job_set.filter(acct_id=int(a_job.id)):
print "Job %s already exists" % a_job.id
return
owner = get_user(a_job.acct['owner'], system)
flops = NP.zeros(a_job.times.size)
total_flops = NP.array([])
mem_total_per_host = NP.array([])
for a_host in a_job.hosts:
a_host = a_job.hosts[a_host]
for k,v in a_host.interpret_amd64_pmc_cpu().iteritems():
if 'SSEFLOPS' in v:
flops += v['SSEFLOPS']
index = a_job.schema['mem'].keys['MemUsed'].index
mem_total = NP.zeros(a_job.times.size)
for v in a_host.stats['mem']:
mem_total += a_host.stats['mem'][v][:,2]
mem_total_per_host = NP.append(mem_total, mem_total_per_host)
total_flops = NP.append(total_flops, NP.diff(flops) / NP.diff(a_job.times) / 10 ** 9)
avg_flops = total_flops.mean()
flops_25 = NP.count_nonzero([total_flops < FLOPS_25])*100 /total_flops.size
flops_50 = NP.count_nonzero([total_flops < FLOPS_50])*100 /total_flops.size
flops_75 = NP.count_nonzero([total_flops < FLOPS_75])*100 /total_flops.size
mem_25 = NP.count_nonzero([mem_total_per_host < MEM_25])*100 / mem_total_per_host.size
mem_50 = NP.count_nonzero([mem_total_per_host < MEM_50])*100 / mem_total_per_host.size
mem_75 = NP.count_nonzero([mem_total_per_host < MEM_75])*100 / mem_total_per_host.size
#nr_bad_hosts = len(filter(lambda h: len(h.times) < 2,
# a_job.hosts.values()))
job_dict = {
'system': system,
'acct_id': a_job.id,
'owner': owner,
'queue': a_job.acct['queue'],
'queue_wait_time': a_job.start_time - a_job.acct['submission_time'],
'begin': a_job.start_time,
'end': a_job.end_time,
#'nr_bad_hots': nr_bad_hosts,
'nr_slots': a_job.acct['slots'],
'pe': a_job.acct['granted_pe'],
'failed': a_job.acct['failed'],
'exit_status': a_job.acct['exit_status'],
'avg_flops': avg_flops,
'flops_25': flops_25,
'flops_50': flops_50,
'flops_75': flops_75,
'mem_25': mem_25,
'mem_50': mem_50,
'mem_75': mem_75,
}
job_dict.update(job.JobAggregator(a_job).stats)
#newJob.nr_hosts = len(a_job.hosts)
try:
newJob = tsm_Job(**job_dict)
newJob.save()
except DatabaseError:
print "Error on job,", a_job.id
transaction.rollback()
return
hosts = map(lambda node: get_node(system, node), a_job.hosts.keys())
newJob.hosts = hosts
print "Added job:", a_job.id
