def main(args=None):
# Fix a bug introduced by an annoying Google extension
import absl.logging
try:
logging.root.removeHandler(absl.logging._absl_handler)
absl.logging._warn_preinit_stderr = False
except Exception as e:
print("Failed to fix absl logging bug", e)
logging.basicConfig(format="%(levelname)s:%(message)s",
level=logging.INFO)
args = get_args(args)
if args is None:
log.error("Something went wrong while parsing arguments!")
exit(1)
log.info("Registering the DC environment...")
register_env("dc_env", get_env)
# Configure all ray input parameters based on the arguments
config = configure_ray(args)
output_dir = config["env_config"]["output_dir"]
# Check if the output directory exists before running
dc_utils.check_dir(output_dir)
# Dump the configuration
dc_utils.dump_json(path=output_dir, name="ray_config", data=config)
log.info("Starting Ray...")
ts = time.time()
ray.init(ignore_reinit_error=True,
logging_level=logging.INFO,
temp_dir=output_dir,
plasma_store_socket_name="/tmp/plasma_socket%s" % ts,
raylet_socket_name="/tmp/raylet_socket%s" % ts)
log.info("Starting experiment.")
if args.tune:
tune_run(config, args.episodes, args.root_output, args.schedule)
else:
run_ray(config, args.episodes)
# Wait until the topology is torn down completely
# The flaky Mininet stop() call necessitates this
# This is an unfortunate reality and may conflict with other ovs setups
log.info("Waiting for environment to complete...")
wait_for_ovs()
# Take control back from root
dc_utils.change_owner(args.root_output)
# Ray doesn't play nice and prevents proper shutdown sometimes
ray.shutdown()
# time.sleep(1)
# kill_ray()
log.info("Experiment has completed.")
def dump_config(path, pattern):
test_config = {}
test_config["transport"] = TRANSPORT
test_config["episodes"] = EPISODES
test_config["runs"] = RUNS
test_config["topology"] = TOPO
test_config["tcp_algorithms"] = TCP_ALGOS
test_config["rl_algorithms"] = RL_ALGOS
test_config["rates"] = NETWORK_RATES
test_config["pattern"] = pattern
# Get a string formatted time stamp
ts = time.time()
st = datetime.datetime.fromtimestamp(ts).strftime('%Y_%m_%d_%H_%M_%S')
test_config["timestamp"] = st
dc_utils.dump_json(path=path, name="bench_config", data=test_config)
def __init__(self, conf={}):
self.conf = DEFAULT_CONF
self.conf.update(conf)
# Init one-to-one mapped variables
self.net_man = None
self.state_man = None
self.traffic_gen = None
self.bw_ctrl = None
self.sampler = None
self.input_file = None
self.terminated = False
self.reward = RawValue('d', 0)
# set the id of this environment
self.short_id = dc_utils.generate_id()
if self.conf["parallel_envs"]:
self.conf["topo_conf"]["id"] = self.short_id
# initialize the topology
self.topo = TopoFactory.create(self.conf["topo"],
self.conf["topo_conf"])
# Save the configuration we have, id does not matter here
dc_utils.dump_json(path=self.conf["output_dir"],
name="env_config",
data=self.conf)
dc_utils.dump_json(path=self.conf["output_dir"],
name="topo_config",
data=self.topo.conf)
# set the dimensions of the state matrix
self._set_gym_matrices()
# Set the active traffic matrix
self._set_traffic_matrix(self.conf["tf_index"], self.conf["input_dir"],
self.topo)
# each unique id has its own sub folder
if self.conf["parallel_envs"]:
self.conf["output_dir"] += f"/{self.short_id}"
# check if the directory we are going to work with exists
dc_utils.check_dir(self.conf["output_dir"])
# handle unexpected exits scenarios gracefully
atexit.register(self.close)