def _tcp_rendezvous_handler(url: str,
timeout: timedelta = default_pg_timeout,
**kwargs):
def _error(msg):
return _rendezvous_error("tcp:// rendezvous: " + msg)
result = urlparse(url)
if not result.port:
raise _error("port number missing")
query: Dict[str, Union[int, str]]
# mypy doesn't allow dict() to accept List of values (#257)
query = dict(
pair.split("=") for pair in filter(None, result.query.split(
"&"))) # type: ignore[misc, arg-type]
if "rank" not in query:
raise _error("rank parameter missing")
if "world_size" not in query:
raise _error("world size parameter missing")
rank = int(query["rank"])
world_size = int(query["world_size"])
start_daemon = rank == 0
assert result.hostname is not None
store = TCPStore(result.hostname, result.port, world_size, start_daemon,
timeout)
yield (store, rank, world_size)
# If this configuration is invalidated, there is nothing we can do about it
raise RuntimeError("Unable to perform rerendezvous using tcp:// method")
def _env_rendezvous_handler(url: str,
timeout: timedelta = default_pg_timeout,
**kwargs):
def _error(msg):
return _rendezvous_error("env:// rendezvous: " + msg)
def _env_error(var):
return _error("environment variable %s expected, but not set" % var)
result = urlparse(url)
query: Dict[str, Union[int, str]]
# mypy doesn't allow dict() to accept List of values (#257)
query = dict(
pair.split("=") for pair in filter(None, result.query.split(
"&"))) # type: ignore[misc, arg-type]
rank: Optional[Union[str, int]]
world_size: Optional[Union[str, int]]
master_port: Optional[Union[str, int]]
if "rank" in query:
rank = int(query["rank"])
else:
rank = os.environ.get("RANK", None)
if rank is None:
raise _env_error("RANK")
if "world_size" in query:
world_size = int(query["world_size"])
else:
world_size = os.environ.get("WORLD_SIZE", None)
if world_size is None:
raise _env_error("WORLD_SIZE")
master_addr = os.environ.get("MASTER_ADDR", None)
if master_addr is None:
raise _env_error("MASTER_ADDR")
master_port = os.environ.get("MASTER_PORT", None)
if master_port is None:
raise _env_error("MASTER_PORT")
# Converting before creating the store
rank = int(rank)
world_size = int(world_size)
master_port = int(master_port)
# Now start the TCP store daemon on the rank 0
start_daemon = rank == 0
store = TCPStore(master_addr, master_port, world_size, start_daemon,
timeout)
yield (store, rank, world_size)
# If this configuration is invalidated, there is nothing we can do about it
raise RuntimeError("Unable to perform rerendezvous using env:// method")
def _tcp_rendezvous_handler(url, timeout=default_pg_timeout, **kwargs):
def _error(msg):
return _rendezvous_error("tcp:// rendezvous: " + msg)
result = urlparse(url)
if not result.port:
raise _error("port number missing")
query = dict(pair.split("=") for pair in filter(None, result.query.split("&")))
if "rank" not in query:
raise _error("rank parameter missing")
if "world_size" not in query:
raise _error("world size parameter missing")
rank = int(query["rank"])
world_size = int(query["world_size"])
start_daemon = rank == 0
store = TCPStore(result.hostname, result.port, world_size, start_daemon, timeout)
yield (store, rank, world_size)
# If this configuration is invalidated, there is nothing we can do about it
raise RuntimeError("Unable to perform rerendezvous using tcp:// method")