def init_lmon(self, argv):
"""Initialize LaunchMON communication.
argv is the set of arguments to initialize with.
"""
self.lmon = LMON_be()
try:
self.lmon.init(len(argv), argv)
self.lmon.regPackForBeToFe(lmon.pack)
self.lmon.regUnpackForFeToBe(lmon.unpack)
self.lmon.handshake(None)
self.lmon.ready(None)
self.lmon_rank = self.lmon.getMyRank()
self.lmon_size = self.lmon.getSize()
self.lmon_master = self.lmon.amIMaster()
self.proctab_size = self.lmon.getMyProctabSize()
self.proctab, unused = self.lmon.getMyProctab(self.proctab_size)
except lmon.LMONException as e:
e.print_lmon_error()
traceback.print_exc()
return False
self._init_mpiranks()
return True
def init_lmon(self, argv):
"""Initialize LaunchMON communication.
argv is the set of arguments to initialize with.
"""
self.lmon = LMON_be()
try:
self.lmon.init(len(argv), argv)
self.lmon.regPackForBeToFe(lmon.pack)
self.lmon.regUnpackForFeToBe(lmon.unpack)
self.lmon.handshake(None)
self.lmon.ready(None)
self.lmon_rank = self.lmon.getMyRank()
self.lmon_size = self.lmon.getSize()
self.lmon_master = self.lmon.amIMaster()
self.proctab_size = self.lmon.getMyProctabSize()
self.proctab, unused = self.lmon.getMyProctab(self.proctab_size)
except lmon.LMONException as e:
e.print_lmon_error()
traceback.print_exc()
return False
self._init_mpiranks()
return True
class CommunicatorBE (Communicator):
"""Communicator for the back-end."""
def __init__(self, locking = False):
Communicator.__init__(self, locking)
def init_lmon(self, argv):
"""Initialize LaunchMON communication.
argv is the set of arguments to initialize with.
"""
self.lmon = LMON_be()
try:
self.lmon.init(len(argv), argv)
self.lmon.regPackForBeToFe(lmon.pack)
self.lmon.regUnpackForFeToBe(lmon.unpack)
self.lmon.handshake(None)
self.lmon.ready(None)
self.lmon_rank = self.lmon.getMyRank()
self.lmon_size = self.lmon.getSize()
self.lmon_master = self.lmon.amIMaster()
self.proctab_size = self.lmon.getMyProctabSize()
self.proctab, unused = self.lmon.getMyProctab(self.proctab_size)
except lmon.LMONException as e:
e.print_lmon_error()
traceback.print_exc()
return False
self._init_mpiranks()
return True
def _wait_for_hello(self):
"""Wait until we receive a HELLO message on MRnet from the front-end."""
msg, stream = self.recv(ret_stream = True)
if msg.msg_type != HELLO_MSG:
print "First message is not hello!"
sys.exit(1)
self.mrnet_frontend_stream = stream
def _init_mrnet_streams(self):
"""Initialize basic MRNet streams."""
self.broadcast_communicator = None
self.mrnet_broadcast_stream = None
self.mrnet_frontend_stream = None # Filled in by back-ends later.
def init_mrnet(self):
"""Initialize MRNet."""
local_node_info = None
try:
if self.lmon_master:
# Receive topology information from front-end.
node_info = self.lmon.recvUsrData(gdbconf.topology_transmit_size)
# Scatter topology information to back-end.
# Presently uses a node info size of 256.
local_node_info = self.lmon.scatter(node_info, 256)
else:
# Receive scattered topology.
local_node_info = self.lmon.scatter(None, 256)
except lmon.LMONException as e:
e.print_lmon_error()
traceback.print_exc()
return False
# Construct MRNet arguments and create network.
argv = [sys.argv[0], # Program name.
str(local_node_info.host), # Comm node host.
str(local_node_info.port), # Comm node port.
str(local_node_info.mrnrank), # Comm node rank.
socket.getfqdn(), # My host.
str(local_node_info.be_rank)] # My rank.
# Initialize.
self.mrnet = MRN.Network.CreateNetworkBE(6, argv)
self._init_shared_mrnet()
self._wait_for_hello()
return True
def shutdown(self):
"""Shut down the communication infrastructure."""
while not self.mrnet_frontend_stream.is_Closed():
time.sleep(0.1)
#del self.mrnet_frontend_stream
self.mrnet.waitfor_ShutDown()
del self.mrnet
self.lmon.finalize()
self.been_shutdown = True
class CommunicatorBE(Communicator):
"""Communicator for the back-end."""
def __init__(self, locking=False):
Communicator.__init__(self, locking)
def init_lmon(self, argv):
"""Initialize LaunchMON communication.
argv is the set of arguments to initialize with.
"""
self.lmon = LMON_be()
try:
self.lmon.init(len(argv), argv)
self.lmon.regPackForBeToFe(lmon.pack)
self.lmon.regUnpackForFeToBe(lmon.unpack)
self.lmon.handshake(None)
self.lmon.ready(None)
self.lmon_rank = self.lmon.getMyRank()
self.lmon_size = self.lmon.getSize()
self.lmon_master = self.lmon.amIMaster()
self.proctab_size = self.lmon.getMyProctabSize()
self.proctab, unused = self.lmon.getMyProctab(self.proctab_size)
except lmon.LMONException as e:
e.print_lmon_error()
traceback.print_exc()
return False
self._init_mpiranks()
return True
def _wait_for_hello(self):
"""Wait until we receive a HELLO message on MRnet from the front-end."""
msg, stream = self.recv(ret_stream=True)
if msg.msg_type != HELLO_MSG:
print "First message is not hello!"
sys.exit(1)
self.mrnet_frontend_stream = stream
def _init_mrnet_streams(self):
"""Initialize basic MRNet streams."""
self.broadcast_communicator = None
self.mrnet_broadcast_stream = None
self.mrnet_frontend_stream = None # Filled in by back-ends later.
def init_mrnet(self):
"""Initialize MRNet."""
local_node_info = None
try:
if self.lmon_master:
# Receive topology information from front-end.
node_info = self.lmon.recvUsrData(
gdbconf.topology_transmit_size)
# Scatter topology information to back-end.
# Presently uses a node info size of 256.
local_node_info = self.lmon.scatter(node_info, 256)
else:
# Receive scattered topology.
local_node_info = self.lmon.scatter(None, 256)
except lmon.LMONException as e:
e.print_lmon_error()
traceback.print_exc()
return False
# Construct MRNet arguments and create network.
argv = [
sys.argv[0], # Program name.
str(local_node_info.host), # Comm node host.
str(local_node_info.port), # Comm node port.
str(local_node_info.mrnrank), # Comm node rank.
socket.getfqdn(), # My host.
str(local_node_info.be_rank)
] # My rank.
# Initialize.
self.mrnet = MRN.Network.CreateNetworkBE(6, argv)
self._init_shared_mrnet()
self._wait_for_hello()
return True
def shutdown(self):
"""Shut down the communication infrastructure."""
while not self.mrnet_frontend_stream.is_Closed():
time.sleep(0.1)
#del self.mrnet_frontend_stream
self.mrnet.waitfor_ShutDown()
del self.mrnet
self.lmon.finalize()
self.been_shutdown = True