def main(argv):
args = parser.parse_args(argv[1:])
assert args.command == 'mode'
cluster = Cluster.new('tmux')
exp = cluster.new_experiment(EXP_NAME)
exp.set_preamble_cmds(PREAMBLE_CMDS)
serv, cli = create_program(exp)
if args.mode == 'localhost':
node = localhost_setup()
localhost_placement(serv, cli, node)
elif args.mode == 'ssh':
nodeloader = NodeLoader(
ConfigDict(argon.to_nested_dicts(args.cluster_config)),
args.filter_regex)
nodes = nodeloader.nodes
if len(nodes) != 1:
raise Exception(
'For this test condition, please specify just a single ssh node.'
)
ssh_node = nodes[0]
ssh_node.setup(res_files=RES_FILES)
local_node = localhost_setup()
ssh_placement(serv, cli, local_node, ssh_node)
elif args.mode == 'slurm':
nodeloader = NodeLoader(
ConfigDict(argon.to_nested_dicts(args.cluster_config)),
args.filter_regex)
nodes = nodeloader.nodes
if len(nodes) != 1:
raise Exception(
'For this test condition, please specify just a single slurm node.'
)
slurm_node = nodes[0]
slurm_node.setup(res_files=RES_FILES)
local_node = localhost_setup()
slurm_placement(serv, cli, local_node, slurm_node)
else:
raise Exception('Unknown mode %s' % args.mode)
try:
cluster.launch(exp)
while True:
time.sleep(100000)
except KeyboardInterrupt:
cluster.delete(experiment_name=EXP_NAME)
def create_cluster(self): # step 1
return Cluster.new('kube')
def create_cluster(self, server_name):
self.cluster = Cluster.new('tmux', server_name=server_name)
def create_cluster(self): # step 1
# Create a Cluster with Docker Compose backend.
return Cluster.new('docker')
def action_create(self, args):
"""
Spin up a multi-node distributed Surreal experiment.
Put any command line args that pass to the config script after "--"
"""
cluster = Cluster.new('subproc',
stdout_mode='print',
stderr_mode='print',
log_dir=None)
experiment_name = args.experiment_name
exp = cluster.new_experiment(experiment_name)
algorithm_args = args.remainder
algorithm_args += [
"--num-agents",
str(args.num_agents),
]
experiment_folder = os.path.join(self.folder, experiment_name)
print('Writing experiment output to {}'.format(experiment_folder))
algorithm_args += ["--experiment-folder", experiment_folder]
algorithm_args += ["--env", args.env]
executable = self._find_executable(args.algorithm)
cmd_gen = CommandGenerator(num_agents=args.num_agents,
num_evals=args.num_evals,
executable=executable,
config_commands=algorithm_args)
learner = exp.new_process('learner',
cmd=cmd_gen.get_command('learner'))
# learner.set_env('DISABLE_MUJOCO_RENDERING', "1")
replay = exp.new_process('replay', cmd=cmd_gen.get_command('replay'))
ps = exp.new_process('ps', cmd=cmd_gen.get_command('ps'))
tensorboard = exp.new_process('tensorboard',
cmd=cmd_gen.get_command('tensorboard'))
tensorplex = exp.new_process('tensorplex',
cmd=cmd_gen.get_command('tensorplex'))
loggerplex = exp.new_process('loggerplex',
cmd=cmd_gen.get_command('loggerplex'))
agents = []
for i in range(args.num_agents):
agent_name = 'agent-{}'.format(i)
agent = exp.new_process(agent_name,
cmd=cmd_gen.get_command(agent_name))
agents.append(agent)
evals = []
for i in range(args.num_evals):
eval_name = 'eval-{}'.format(i)
eval_p = exp.new_process(eval_name,
cmd=cmd_gen.get_command(eval_name))
evals.append(eval_p)
setup_network(agents=agents,
evals=evals,
learner=learner,
replay=replay,
ps=ps,
tensorboard=tensorboard,
tensorplex=tensorplex,
loggerplex=loggerplex)
self._setup_gpu(agents=agents,
evals=evals,
learner=learner,
gpus=args.gpu)
cluster.launch(exp, dry_run=args.dry_run)
import symphony
from symphony.engine import Cluster
cluster = Cluster.new('subproc',
stdout_mode='print', stderr_mode='print', log_dir=None)
exp = cluster.new_experiment('hello-world')
server = exp.new_process('server', cmd='python simple_server.py')
client = exp.new_process('client', cmd='python simple_client.py')
server.binds('example')
client.connects('example')
print('Server and client are running in subprocess mode')
cluster.launch(exp, dry_run=False)