default=myip)
parser.add_option(
"-e",
"--eip",
dest="eip",
type="string",
help=
"external ip address to publish. if Not set will default to the same as -i option",
default=None)
parser.add_option(
"-o",
"--notifyport",
dest="nport",
type="int",
help="port to bind to for notifications. default: random",
default=None)
parser.add_option("-f",
"--infoport",
dest="fport",
type="int",
help="port to bind to for information. default: random",
default=None)
options, args = nwperf.parseServerOptions()
if not options.cluster:
parser.error("No Cluster Specified")
main(options.nameserver, options.cluster, options.ip, options.statefile,
options.fport, options.nport, options.eip)
if request[0] == "jobinfo":
info = status.jobInfo(request[1])
sock.send_json(info)
if request[0] == "running":
info = status.runningJobs()
sock.send_json(info)
if request[0] == "completed":
info = status.completedJobs()
sock.send_json(info)
except:
sock.send_json({})
print "done"
else:
ns.updateServices()
status.checkUpdate()
if __name__ == "__main__":
myip = socket.gethostbyname(socket.gethostname())
parser = nwperf.defaultServerOptionParser()
parser.add_option("-S","--name-server",dest="nameserver",type="string",help="The ZMQ URL of the nameserver to register with",default="tcp://nwperf-ns:6967")
parser.add_option("-s","--state-file", help="File to store the current slurm state", dest="statefile", default="/tmp/slurmjobs.state")
parser.add_option("-c","--cluster",dest="cluster",type="string",help="The cluster prefix to publish as",default=None)
parser.add_option("-i", "--ip", dest="ip", type="string", help="ip address to bind to. default: %s" % myip, default=myip)
options,args = nwperf.parseServerOptions()
if not options.cluster:
parser.error("No Cluster Specified")
main(options.nameserver,options.cluster,options.ip,options.statefile)
def main():
parser = nwperf.defaultServerOptionParser()
parser.add_option( "-c", "--cluster", action="store", type="string", dest="cluster",
help="name of cluster to generate graphs for")
parser.add_option( "-S", "--name-server", action="store", type="string", dest="nameserver",
help="The ZMQ URL of the nameserver")
parser.add_option( "-j", "--job-service", action="store", type="string", dest="jobservice",
help="The service name that provides job information")
parser.add_option( "-P", "--point-service", action="store", type="string", dest="pointservice",
help="The service name that provides point data")
(options, args) = nwperf.parseServerOptions()
if not options.nameserver:
parser.error("No name server specified")
if not options.pointservice:
parser.error("No point service specified")
if not options.jobservice:
parser.error("No job service specified")
if not options.cluster:
parser.error("No cluster specified")
pointStore = MongoPointStore.MongoPointStore()
jobStore = MongoJobStore.MongoJobStore()
ns = nnslib.NameServer(options.nameserver)
ctx = zmq.Context()
q = multiprocessing.Queue()
#if not options.generategraphs:
children = [JobArchiveGenerator(pointStore, q, jobStore, extraFields={"cluster": options.cluster}) for i in range(8)]
for child in children:
child.start()
psp = PointStoreProcess(pointStore, options.nameserver, options.pointservice)
psp.start()
sock = ctx.socket(zmq.SUB)
sock.setsockopt(zmq.SUBSCRIBE, "JobEnd")
ns.connectService(sock, options.jobservice)
while True:
poll = zmq.core.poll.Poller()
poll.register(sock, zmq.POLLIN)
res = poll.poll(1000)
if res:
job = sock.recv_multipart()[1]
job = json.loads(job)
job["Nodes"] = hostlist.expand_hostlist(job["NodeList"])
q.put(job)
for i in range(len(children)):
if not children[i].is_alive():
children[i].join()
del(children[i])
newchild = JobArchiveGenerator(pointStore, q, jobStore, extraFields={"cluster": options.cluster})
newchild.start()
children.append(newchild)
try:
ns.updateSocket(sock)
except:
sock.close()
sock = ctx.socket(zmq.SUB)
sock.setsockopt(zmq.SUBSCRIBE, "JobEnd")
connected = False
while not connected:
try:
ns.connectService(sock, options.jobservice)
except:
print "Error connecting to %s. Sleeping" % options.jobservice
time.sleep(1)
finally:
connected = True
poll.register(sock, zmq.POLLIN)
if q.qsize() > 0:
print "Queue size: %d" % q.qsize()
if not psp.is_alive():
psp.join()
psp = PointStoreProcess(pointStore, options.nameserver, options.pointservice)