def process_compute_mode(self, numtasks):
# Send superstep
self.surveyor.send("%d" % (self.superstep))
halt = True
respondents = 0
try:
while True:
msg = remap_utils.decode(self.surveyor.recv())
if msg != "H":
halt = False
respondents = respondents + 1
if respondents == numtasks:
# all replied
logger.info("All respondents replied")
self.mode = MODE_PROCESS
break
except nn.NanoMsgAPIError as nme:
logger.error("No vertex nodes connected")
print(nme)
self.mode = MODE_RECOVERY
return True
if halt:
self.mode = MODE_HALT
else:
self.superstep = self.superstep + 1
return True
def process_compute_mode( self, numtasks ):
# Send superstep
self.surveyor.send( "%d"%(self.superstep) )
halt = True
respondents = 0
try:
while( True ):
msg = remap_utils.decode( self.surveyor.recv() )
if msg != "H":
halt = False
respondents = respondents + 1
if respondents == numtasks:
# all replied
logger.info("All respondents replied")
self.mode = MODE_PROCESS
break
except nn.NanoMsgAPIError as nme:
logger.error("No vertex nodes connected")
print(nme)
self.mode = MODE_RECOVERY
return True
if halt:
self.mode = MODE_HALT
else:
self.superstep = self.superstep + 1
return True
def process_halt_mode(self, numtasks):
self.surveyor.send("H")
respondents = 0
try:
while True:
msg = remap_utils.decode(self.surveyor.recv())
respondents = respondents + 1
if respondents == numtasks:
logger.info("All respondents replied")
break
except nn.NanoMsgAPIError as nme:
print(nme)
return False
def process_halt_mode( self, numtasks ):
self.surveyor.send( "H" )
respondents = 0
try:
while( True ):
msg = remap_utils.decode( self.surveyor.recv() )
respondents = respondents + 1
if respondents == numtasks:
logger.info("All respondents replied")
break
except nn.NanoMsgAPIError as nme:
print(nme)
return False
def process_process_mode(self, numtasks):
self.surveyor.send("P")
respondents = 0
try:
while True:
msg = remap_utils.decode(self.surveyor.recv())
respondents = respondents + 1
if respondents == numtasks:
logger.info("All respondents replied")
self.mode = MODE_SHIFT
break
except nn.NanoMsgAPIError as nme:
logger.error("No vertex nodes replied")
print(nme)
self.mode = MODE_RECOVERY
return True
def process_shift_mode( self, numtasks ):
self.surveyor.send( "S" )
respondents = 0
try:
while( True ):
msg = remap_utils.decode( self.surveyor.recv() )
respondents = respondents + 1
if respondents == numtasks:
logger.info("All respondents replied")
self.mode = MODE_COMPUTE
break
except nn.NanoMsgAPIError as nme:
logger.error("No vertex nodes replied")
print(nme)
self.mode = MODE_RECOVERY
return True
def work( self ):
if not self.ready:
return True
surveyormsg = None
try:
surveyormsg = remap_utils.decode( self.surveyor.recv() )
except nn.NanoMsgAPIError as e:
return True
if surveyormsg[0] == 'S':
# Shift messages
if self.mode != MODE_MSGS:
self.mode = MODE_MSGS
# We haven't done this in a previous step. Due to recovery, it might be
# used by the initiator to get others up to speed.
for key, (vertex,messages,messagesNext) in self.vertices.items():
self.vertices[ key ] = ( vertex, messagesNext, [] )
self.surveyor.send( "D" )
return True
if surveyormsg[0] == 'H':
self.mode = MODE_HALT
logger.info("Halting core.")
self.surveyor.close()
return False
if surveyormsg[0] == 'P':
if self.mode != MODE_PROCESS:
# First time in this state, we need to grab all messages and
# allocate them to vertex queue
self.mode = MODE_PROCESS
logger.info("Processing messages 1")
while True:
try:
msg = self.vsub.recv()
prefix, data = remap_utils.unpack_msg( msg )
if prefix in self.vertices:
# This vertex is indeed on this host. Add the message to its new msg list for next iteration
vertex, messages, messagesNext = self.vertices[ prefix ]
messagesNext.append( data )
except nn.NanoMsgAPIError as e:
logger.error( "No more messages available." )
break
else:
logger.info("Processing messages 2")
# doing things twice does not make a difference. Second time around, just throw away all messages
while True:
try:
msg = self.vsub.recv()
print("Received and thrown away: ", msg)
except nn.NanoMsgAPIError as e:
logger.error( "No more messages available." )
break
self.surveyor.send( "D" )
return True
self.mode = MODE_RUN
self.superstep = int(surveyormsg)
mainHalt = True
for key, (vertex,messages,messagesNext) in self.vertices.items():
vertex, halt = self.app.compute( self.forward, self.subscribe, self.unsubscribe, self.superstep, vertex, messages )
if vertex != None:
# Store the new vertex object in its place, maintaining the messagesNext list as we know it
self.vertices[ key ] = (vertex,[],messagesNext)
if not halt:
mainHalt = False
if mainHalt:
self.surveyor.send( "H" )
else:
self.surveyor.send( "D" )
return True