def test_get_webui(shutdown_only):
addresses = ray.init(include_webui=True, num_cpus=1)
webui_url = addresses["webui_url"]
assert ray.get_webui_url() == webui_url
assert re.match(r"^(localhost|\d+\.\d+\.\d+\.\d+):\d+$", webui_url)
start_time = time.time()
while True:
try:
node_info = requests.get("http://" + webui_url +
"/api/node_info").json()
break
except requests.exceptions.ConnectionError:
if time.time() > start_time + 30:
error_log = None
out_log = None
with open(
"{}/logs/dashboard.out".format(
addresses["session_dir"]), "r") as f:
out_log = f.read()
with open(
"{}/logs/dashboard.err".format(
addresses["session_dir"]), "r") as f:
error_log = f.read()
raise Exception(
"Timed out while waiting for dashboard to start. "
"Dashboard output log: {}\n"
"Dashboard error log: {}\n".format(out_log, error_log))
assert node_info["error"] is None
assert node_info["result"] is not None
assert isinstance(node_info["timestamp"], float)
def test_get_webui(shutdown_only):
addresses = ray.init(include_webui=True, num_cpus=1)
webui_url = addresses["webui_url"]
assert ray.get_webui_url() == webui_url
assert re.match(r"^http://\d+\.\d+\.\d+\.\d+:8080$", webui_url)
start_time = time.time()
while True:
try:
node_info = requests.get(webui_url + "/api/node_info").json()
break
except requests.exceptions.ConnectionError:
if time.time() > start_time + 30:
raise Exception(
"Timed out while waiting for dashboard to start.")
assert node_info["error"] is None
assert node_info["result"] is not None
assert isinstance(node_info["timestamp"], float)
def test_get_webui(shutdown_only):
addresses = ray.init(include_webui=True, num_cpus=1)
webui_url = addresses["webui_url"]
assert ray.get_webui_url() == webui_url
base, token = webui_url.split("?")
assert token.startswith("token=")
start_time = time.time()
while True:
try:
node_info = requests.get(base + "api/node_info?" + token).json()
break
except requests.exceptions.ConnectionError:
if time.time() > start_time + 30:
raise Exception(
"Timed out while waiting for dashboard to start.")
assert node_info["error"] is None
assert node_info["result"] is not None
assert isinstance(node_info["timestamp"], float)
while True:
current_payoff_matrix_size = size_checker.get_size_of_current_payoff_table(
)
if current_payoff_matrix_size < expected_payoff_matrix_size:
logger.info(
f"waiting for payoff matrix to reach size {expected_payoff_matrix_size} (currently {current_payoff_matrix_size})..."
)
time.sleep(5)
# elif current_payoff_matrix_size > expected_payoff_matrix_size:
# raise ValueError(f"payoff matrix is now larger than expected (expected {expected_payoff_matrix_size}, currently {current_payoff_matrix_size})")
else:
break
expected_payoff_matrix_size = current_payoff_matrix_size + 1
ray.init(address=os.getenv('RAY_HEAD_NODE'), ignore_reinit_error=True)
logger.info("Ray Web UI at {}".format(ray.get_webui_url()))
base_experiment_name = f"{CLOUD_PREFIX}learner_{POKER_GAME_VERSION}_sac_arch1_psro_sequential_explore_coeff_{PSRO_EXPLORATION_COEFF}"
full_experiment_name = f"{base_experiment_name}_{gethostname()}_pid_{os.getpid()}_{datetime_str()}"
experiment_save_dir = os.path.join(DEFAULT_RESULTS_DIR,
full_experiment_name)
def init_static_policy_distribution_after_trainer_init_callback(
trainer):
trainer.storage_client = connect_storage_client()
logger.info("Initializing trainer manager interface")
trainer.manager_interface = LearnerManagerInterface(
server_host=MANAGER_SEVER_HOST,
port=MANAGER_PORT,
worker_id=full_experiment_name,
#### Plot the simulation results ###################################################################
logger.plot()
####################################################################################################
#### Part 2 of 2 of _devp.y: training and testing MARLFlockAviary as an RLlib MultiAgentEnv ########
####################################################################################################
elif PART == 2:
#### WIP notes #####################################################################################
#### use ENV_STATE: github.com/ray-project/ray/blob/master/rllib/examples/env/two_step_game.py #####
#### Initialize Ray Tune ###########################################################################
ray.shutdown()
ray.init(ignore_reinit_error=True)
print("Dashboard URL: http://{}".format(ray.get_webui_url()))
#### Register the environment ######################################################################
register_env(
"marl-flock-aviary-v0",
lambda _: MARLFlockAviary(drone_model=DRONE,
num_drones=NUM_DRONES,
physics=PHYSICS,
freq=SIMULATION_FREQ_HZ))
#### Set up the trainer's config ###################################################################
config = ppo.DEFAULT_CONFIG.copy()
config["num_workers"] = 0
config["env"] = "marl-flock-aviary-v0"
#### Unused env to extract correctly sized action and observation spaces ###########################
unused_env = MARLFlockAviary(num_drones=NUM_DRONES)
