def register( self ):
self.set_node_timeout( 500 )
msgid = remap_utils.unique_id()
logger.info( "Registering with node" )
self.forward( "node._hello.%d"%(self.pid), {"msgid":msgid,"pid":self.pid,"priority":self.priority} )
# The while loop will terminate as soon as node stops sending messages,
# so this should be safe to do.
while True:
try:
msg = self.sub.recv()
msgprefix, data = remap_utils.unpack_msg( msg )
recipientid,msgtype,senderid = remap_utils.split_prefix(msgprefix)
if msgtype != "_hey":
continue
replymsgid = remap_utils.safe_get(data, "msgid")
if replymsgid == msgid:
# this is us
self.coreid = remap_utils.safe_get(data, "coreid" )
self.sub.set_string_option( nn.SUB, nn.SUB_UNSUBSCRIBE, "" )
self.sub.set_string_option( nn.SUB, nn.SUB_SUBSCRIBE, "global" )
self.sub.set_string_option( nn.SUB, nn.SUB_SUBSCRIBE, "local" )
self.sub.set_string_option( nn.SUB, nn.SUB_SUBSCRIBE, "notlocal" )
self.sub.set_string_option( nn.SUB, nn.SUB_SUBSCRIBE, self.coreid )
logger.info( "Received coreid %s."%( self.coreid ))
return True
except nn.NanoMsgAPIError as e:
logger.error( "Node is currently not available." )
break
logger.error( "Registration failed" )
return False
def start_job(self, jobdata):
if self.job_status != "waiting":
raise RemapException(
"A job is currently in progress on this monitor")
if "type" not in jobdata:
raise RemapException("Must have job type specified")
if "priority" not in jobdata:
raise RemapException("Must have priority specified")
if "parallellism" not in jobdata:
raise RemapException("Must have parallellism specified")
self.job_status = "preparing"
self.prepare_start = time.time()
self.jobtype = jobdata["type"]
self.priority = jobdata["priority"]
self.parallellism = jobdata["parallellism"]
plugin = self.load_plugin(self.jobtype)
self.rejectedtasks = {}
self.completedtasks = {}
if self.jobid != None:
self.bsub.set_string_option(nn.SUB, nn.SUB_UNSUBSCRIBE, self.jobid)
if "jobid" in jobdata:
self.jobid = jobdata["jobid"]
del jobdata["jobid"]
else:
self.jobid = remap_utils.unique_id()
self.bsub.set_string_option(nn.SUB, nn.SUB_SUBSCRIBE, self.jobid)
if "app" not in jobdata:
raise RemapException("The name of the app must be provided")
if jobdata["app"] not in self.list_apps():
raise RemapException("No such application: %s" % (jobdata["app"]))
config = {"jobid": self.jobid, "remaproot": self.remaproot}
logger.info("Started a new job: %s" % (self.jobid))
self.manager = plugin.create_manager(jobdata, config)
if ((time.time() - self.refreshed) > 60):
# Not refreshed > 60s
self.refresh_nodes(self.priority)
# Wait for a bunch of nodes to advertise themselves
r = Timer(1.0, self.resume, ())
r.start()
else:
self.resume()
return {"jobid": self.jobid}
def start_job(self, jobdata):
if self.job_status != "waiting":
raise RemapException("A job is currently in progress on this monitor")
if "type" not in jobdata:
raise RemapException("Must have job type specified")
if "priority" not in jobdata:
raise RemapException("Must have priority specified")
if "parallellism" not in jobdata:
raise RemapException("Must have parallellism specified")
self.job_status = "preparing"
self.prepare_start = time.time()
self.jobtype = jobdata["type"]
self.priority = jobdata["priority"]
self.parallellism = jobdata["parallellism"]
plugin = self.load_plugin(self.jobtype)
self.rejectedtasks = {}
self.completedtasks = {}
if self.jobid != None:
self.bsub.set_string_option(nn.SUB, nn.SUB_UNSUBSCRIBE, self.jobid)
if "jobid" in jobdata:
self.jobid = jobdata["jobid"]
del jobdata["jobid"]
else:
self.jobid = remap_utils.unique_id()
self.bsub.set_string_option(nn.SUB, nn.SUB_SUBSCRIBE, self.jobid)
if "app" not in jobdata:
raise RemapException("The name of the app must be provided")
if jobdata["app"] not in self.list_apps():
raise RemapException("No such application: %s" % (jobdata["app"]))
config = {"jobid": self.jobid, "remaproot": self.remaproot}
logger.info("Started a new job: %s" % (self.jobid))
self.manager = plugin.create_manager(jobdata, config)
if (time.time() - self.refreshed) > 60:
# Not refreshed > 60s
self.refresh_nodes(self.priority)
# Wait for a bunch of nodes to advertise themselves
r = Timer(1.0, self.resume, ())
r.start()
else:
self.resume()
return {"jobid": self.jobid}