def master(group_name: str, is_immediate: bool = False):
"""
The main master logic includes initialize proxy and allocate jobs to workers.
Args:
group_name (str): Identifier for the group of all communication components,
is_immediate (bool): If True, it will be an async mode; otherwise, it will be an sync mode.
Async Mode: The proxy only returns the session id for sending messages. Based on the local task priority,
you can do something with high priority before receiving replied messages from peers.
Sync Mode: It will block until the proxy returns all the replied messages.
"""
proxy = Proxy(group_name=group_name,
component_type="master",
expected_peers={"worker": 1})
random_integer_list = np.random.randint(0, 100, 5)
print(f"generate random integer list: {random_integer_list}.")
for peer in proxy.peers_name["worker"]:
message = SessionMessage(tag="sum",
source=proxy.name,
destination=peer,
payload=random_integer_list,
session_type=SessionType.TASK)
if is_immediate:
session_id = proxy.isend(message)
# Do some tasks with higher priority here.
replied_msgs = proxy.receive_by_id(session_id, timeout=-1)
else:
replied_msgs = proxy.send(message, timeout=-1)
for msg in replied_msgs:
print(
f"{proxy.name} receive {msg.source}, replied payload is {msg.payload}."
)
def as_worker(self, group: str, proxy_options=None, log_dir: str = getcwd()):
"""Executes an event loop where roll-outs are performed on demand from a remote learner.
Args:
group (str): Identifier of the group to which the actor belongs. It must be the same group name
assigned to the learner (and decision clients, if any).
proxy_options (dict): Keyword parameters for the internal ``Proxy`` instance. See ``Proxy`` class
for details. Defaults to None.
"""
if proxy_options is None:
proxy_options = {}
proxy = Proxy(group, "actor", {"learner": 1}, **proxy_options)
logger = Logger(proxy.name, dump_folder=log_dir)
for msg in proxy.receive():
if msg.tag == MessageTag.EXIT:
logger.info("Exiting...")
proxy.close()
sys.exit(0)
elif msg.tag == MessageTag.ROLLOUT:
ep = msg.payload[PayloadKey.ROLLOUT_INDEX]
logger.info(f"Rolling out ({ep})...")
metrics, rollout_data = self.roll_out(
ep,
training=msg.payload[PayloadKey.TRAINING],
model_by_agent=msg.payload[PayloadKey.MODEL],
exploration_params=msg.payload[PayloadKey.EXPLORATION_PARAMS]
)
if rollout_data is None:
logger.info(f"Roll-out {ep} aborted")
else:
logger.info(f"Roll-out {ep} finished")
rollout_finish_msg = Message(
MessageTag.FINISHED,
proxy.name,
proxy.peers_name["learner"][0],
payload={
PayloadKey.ROLLOUT_INDEX: ep,
PayloadKey.METRICS: metrics,
PayloadKey.DETAILS: rollout_data
}
)
proxy.isend(rollout_finish_msg)
self.env.reset()
