class __MPIInterfaceCoreImpl:
""" Implementation of the MPIInterfaceCore singleton interface """
def __init__(self):
self.__command_client = MPICommandClient()
self.__monitor_client = MPIMonitorClient()
def start_cluster(self, cl_file=None):
self.__command_client.set_log_mode(MPIInterface._log_mode)
self.__command_client.start_services()
def stop_cluster(self):
self.__command_client.stop_services()
def get_engines(self):
return MPIEnvironment.mpi_server_rank_list()
def get_nodes(self):
# Get list of all server Ids
servers_list = MPIEnvironment.mpi_server_rank_list()
# Get list of hostnames for each server Id
hostnames = []
for server in servers_list:
hostnames.append(
self.__monitor_client.get_server_status_keyword(
server, 'processor'))
# Remove duplicates to get the list of host used in general
hostnames_not_repeated = list(set(hostnames))
return hostnames_not_repeated
def pgc(self, commands, block=True):
"""This method has two modes:
- When the input command is a dictionary of commands execute
each command taking the dictionary key as target node
(Equivalent to various calls to odo/execute)
- When commands is a single command execute it in all engines
"""
# Get list of jobs and commands
ret = None
if isinstance(commands, dict):
# Spawn jobs in non-blocking mode
jobId_list = []
for server in commands:
cmd = commands[server]
jobId = self.__command_client.push_command_request(
cmd, block=False, target_server=server)
jobId_list.append(jobId[0])
# If user requests blocking mode wait until execution is completed
ret = self.__command_client.get_command_response(jobId_list,
block=block,
verbose=True)
else:
cmd = commands
# Get list of all servers
all_servers_list = MPIEnvironment.mpi_server_rank_list()
# Execute command in all servers
ret = self.__command_client.push_command_request(
cmd, block=block, target_server=all_servers_list)
# Return result
return ret
def odo(self, job, nodes):
"""Execute a job on a subset of engines in non-blocking mode"""
cmd_ids = self.__command_client.push_command_request(
job, block=False, target_server=nodes)
return cmd_ids
def execute(self, job, nodes):
"""Execute a job on a subset of engines in blocking mode"""
result_list = self.__command_client.push_command_request(
job, block=True, target_server=nodes)
return result_list
def push(self, variables, targets=None):
"""Set variables in a sub-set of engines"""
# Determine target servers
target_server = []
if targets is None or targets == 'all':
target_server = MPIEnvironment.mpi_server_rank_list()
else:
target_server = list(targets)
# Push variables
ret = self.__command_client.push_command_request(
"push",
block=True,
target_server=target_server,
parameters=dict(variables))
# Return request result for further processing
return ret
def pull(self, varname="", targets=None):
"""Retrieve a variable from a sub-set of engines"""
# Pull variable from all servers
result_list = []
if targets is None:
result_list = self.pgc(varname)
else:
result_list = self.execute(varname, nodes=targets)
# Extract result vars and corresponding server Ids
res = {}
for result in result_list:
res[result['server']] = result['ret']
# Return result
return res
def check_job(self, jobId, verbose=True):
"""Check the status of a non-blocking job"""
jobId_list = list(jobId)
command_response_list = self.__command_client.get_command_response(
jobId_list, block=False, verbose=verbose)
# Aggregate exceptions and completed jobIds
error_msg = ''
completed_jobs = []
for command_response in command_response_list:
if not command_response['successful']:
if len(error_msg) > 0:
error_msg += "\n"
error_msg += "Exception executing command in server %s: %s" % (
command_response['server'],
command_response['traceback'])
else:
completed_jobs.append(command_response['id'])
# Re-throw aggregated exception
if len(error_msg) > 0:
casalog.post(error_msg, "SEVERE",
"MPIInterfaceCore::check_job")
raise Exception(error_msg)
# Check that all jobs have been completed
completed = True
for jobId in jobId_list:
if jobId not in completed_jobs:
completed = False
break
# Return completion status
return completed
def get_server_status(self):
return self.__command_client.get_server_status()
def get_command_request_list(self):
return self.__command_client.get_command_request_list()
def get_command_response_list(self):
return self.__command_client.get_command_response_list()
class __MPIInterfaceCoreImpl:
""" Implementation of the MPIInterfaceCore singleton interface """
def __init__(self):
self.__command_client = MPICommandClient()
self.__monitor_client = MPIMonitorClient()
def start_cluster(self, cl_file=None):
self.__command_client.set_log_mode(MPIInterface._log_mode)
self.__command_client.start_services()
def stop_cluster(self):
self.__command_client.stop_services()
def get_engines(self):
return MPIEnvironment.mpi_server_rank_list()
def get_nodes(self):
# Get list of all server Ids
servers_list = MPIEnvironment.mpi_server_rank_list()
# Get list of hostnames for each server Id
hostnames = []
for server in servers_list:
hostnames.append(self.__monitor_client.get_server_status_keyword(server,'processor'))
# Remove duplicates to get the list of host used in general
hostnames_not_repeated = list(set(hostnames))
return hostnames_not_repeated
def pgc(self,commands,block=True):
"""This method has two modes:
- When the input command is a dictionary of commands execute
each command taking the dictionary key as target node
(Equivalent to various calls to odo/execute)
- When commands is a single command execute it in all engines
"""
# Get list of jobs and commands
ret = None
if isinstance(commands,dict):
# Spawn jobs in non-blocking mode
jobId_list = []
for server in commands:
cmd = commands[server]
jobId = self.__command_client.push_command_request( cmd,
block=False,
target_server=server)
jobId_list.append(jobId[0])
# If user requests blocking mode wait until execution is completed
ret = self.__command_client.get_command_response(jobId_list,block=block,verbose=True)
else:
cmd = commands
# Get list of all servers
all_servers_list = MPIEnvironment.mpi_server_rank_list()
# Execute command in all servers
ret = self.__command_client.push_command_request( cmd,
block=block,
target_server=all_servers_list)
# Return result
return ret
def odo(self, job, nodes):
"""Execute a job on a subset of engines in non-blocking mode"""
cmd_ids = self.__command_client.push_command_request(job,block=False,target_server=nodes)
return cmd_ids
def execute(self, job, nodes):
"""Execute a job on a subset of engines in blocking mode"""
result_list = self.__command_client.push_command_request(job,block=True,target_server=nodes)
return result_list
def push(self, variables, targets=None):
"""Set variables in a sub-set of engines"""
# Determine target servers
target_server = []
if targets is None or targets == 'all':
target_server = MPIEnvironment.mpi_server_rank_list()
else:
target_server = list(targets)
# Push variables
ret = self.__command_client.push_command_request( "push",
block=True,
target_server=target_server,
parameters=dict(variables))
# Return request result for further processing
return ret
def pull(self, varname="", targets=None):
"""Retrieve a variable from a sub-set of engines"""
# Pull variable from all servers
result_list = []
if targets is None:
result_list = self.pgc(varname)
else:
result_list = self.execute(varname,nodes=targets)
# Extract result vars and corresponding server Ids
res = {}
for result in result_list:
res[result['server']] = result['ret']
# Return result
return res
def check_job(self, jobId, verbose=True):
"""Check the status of a non-blocking job"""
jobId_list = list(jobId)
command_response_list = self.__command_client.get_command_response(jobId_list,block=False,verbose=verbose)
# Aggregate exceptions and completed jobIds
error_msg = ''
completed_jobs = []
for command_response in command_response_list:
if not command_response['successful']:
if len(error_msg) > 0: error_msg += "\n"
error_msg += "Exception executing command in server %s: %s" % (command_response['server'],
command_response['traceback'])
else:
completed_jobs.append(command_response['id'])
# Re-throw aggregated exception
if len(error_msg) > 0:
casalog.post(error_msg,"SEVERE","MPIInterfaceCore::check_job")
raise Exception(error_msg)
# Check that all jobs have been completed
completed = True
for jobId in jobId_list:
if jobId not in completed_jobs:
completed = False
break
# Return completion status
return completed
def get_server_status(self):
return self.__command_client.get_server_status()
def get_command_request_list(self):
return self.__command_client.get_command_request_list()
def get_command_response_list(self):
return self.__command_client.get_command_response_list()
