class TradingNode:
"""
Provides an asynchronous network node for live trading.
"""
def __init__(
self,
strategies: List[TradingStrategy],
config: Dict[str, object],
):
"""
Initialize a new instance of the TradingNode class.
Parameters
----------
strategies : list[TradingStrategy]
The list of strategies to run on the trading node.
config : dict[str, object]
The configuration for the trading node.
Raises
------
ValueError
If strategies is None or empty.
ValueError
If config is None or empty.
"""
PyCondition.not_none(strategies, "strategies")
PyCondition.not_none(config, "config")
PyCondition.not_empty(strategies, "strategies")
PyCondition.not_empty(config, "config")
self._config = config
# Extract configs
config_trader = config.get("trader", {})
config_system = config.get("system", {})
config_log = config.get("logging", {})
config_exec_db = config.get("exec_database", {})
config_risk = config.get("risk", {})
config_strategy = config.get("strategy", {})
# System config
self._connection_timeout = config_system.get("connection_timeout", 5.0)
self._disconnection_timeout = config_system.get(
"disconnection_timeout", 5.0)
self._check_residuals_delay = config_system.get(
"check_residuals_delay", 5.0)
self._load_strategy_state = config_strategy.get("load_state", True)
self._save_strategy_state = config_strategy.get("save_state", True)
# Setup loop
self._loop = asyncio.get_event_loop()
self._executor = concurrent.futures.ThreadPoolExecutor()
self._loop.set_default_executor(self._executor)
self._loop.set_debug(config_system.get("loop_debug", False))
# Components
self._clock = LiveClock(loop=self._loop)
self._uuid_factory = UUIDFactory()
self.system_id = self._uuid_factory.generate()
self.created_time = self._clock.utc_now()
self._is_running = False
# Setup identifiers
self.trader_id = TraderId(
name=config_trader["name"],
tag=config_trader["id_tag"],
)
# Setup logging
level_stdout = LogLevelParser.from_str_py(
config_log.get("level_stdout"))
self._logger = LiveLogger(
loop=self._loop,
clock=self._clock,
trader_id=self.trader_id,
system_id=self.system_id,
level_stdout=level_stdout,
)
self._log = LoggerAdapter(
component=self.__class__.__name__,
logger=self._logger,
)
self._log_header()
self._log.info("Building...")
if platform.system() != "Windows":
# Requires the logger to be initialized
# Windows does not support signal handling
# https://stackoverflow.com/questions/45987985/asyncio-loops-add-signal-handler-in-windows
self._setup_loop()
# Build platform
# ----------------------------------------------------------------------
self.portfolio = Portfolio(
clock=self._clock,
logger=self._logger,
)
self._data_engine = LiveDataEngine(
loop=self._loop,
portfolio=self.portfolio,
clock=self._clock,
logger=self._logger,
config={"qsize": 10000},
)
self.portfolio.register_cache(self._data_engine.cache)
self.analyzer = PerformanceAnalyzer()
if config_exec_db["type"] == "redis":
exec_db = RedisExecutionDatabase(
trader_id=self.trader_id,
logger=self._logger,
command_serializer=MsgPackCommandSerializer(),
event_serializer=MsgPackEventSerializer(),
config={
"host": config_exec_db["host"],
"port": config_exec_db["port"],
},
)
else:
exec_db = BypassExecutionDatabase(
trader_id=self.trader_id,
logger=self._logger,
)
self._exec_engine = LiveExecutionEngine(
loop=self._loop,
database=exec_db,
portfolio=self.portfolio,
clock=self._clock,
logger=self._logger,
config={"qsize": 10000},
)
self._risk_engine = LiveRiskEngine(
loop=self._loop,
exec_engine=self._exec_engine,
portfolio=self.portfolio,
clock=self._clock,
logger=self._logger,
config=config_risk,
)
self._exec_engine.load_cache()
self._exec_engine.register_risk_engine(self._risk_engine)
self.trader = Trader(
trader_id=self.trader_id,
strategies=strategies,
portfolio=self.portfolio,
data_engine=self._data_engine,
exec_engine=self._exec_engine,
risk_engine=self._risk_engine,
clock=self._clock,
logger=self._logger,
)
if self._load_strategy_state:
self.trader.load()
self._builder = TradingNodeBuilder(
data_engine=self._data_engine,
exec_engine=self._exec_engine,
risk_engine=self._risk_engine,
clock=self._clock,
logger=self._logger,
log=self._log,
)
self._log.info("state=INITIALIZED.")
self.time_to_initialize = self._clock.delta(self.created_time)
self._log.info(
f"Initialized in {self.time_to_initialize.total_seconds():.3f}s.")
self._is_built = False
@property
def is_running(self) -> bool:
"""
If the trading node is running.
Returns
-------
bool
True if running, else False.
"""
return self._is_running
@property
def is_built(self) -> bool:
"""
If the trading node clients are built.
Returns
-------
bool
True if built, else False.
"""
return self._is_built
def get_event_loop(self) -> asyncio.AbstractEventLoop:
"""
Return the event loop of the trading node.
Returns
-------
asyncio.AbstractEventLoop
"""
return self._loop
def get_logger(self) -> LiveLogger:
"""
Return the logger for the trading node.
Returns
-------
LiveLogger
"""
return self._logger
def add_data_client_factory(self, name, factory):
"""
Add the given data client factory to the node.
Parameters
----------
name : str
The name of the client factory.
factory : LiveDataClientFactory or LiveExecutionClientFactory
The factory to add.
Raises
------
ValueError
If name is not a valid string.
KeyError
If name has already been added.
"""
self._builder.add_data_client_factory(name, factory)
def add_exec_client_factory(self, name, factory):
"""
Add the given execution client factory to the node.
Parameters
----------
name : str
The name of the client factory.
factory : LiveDataClientFactory or LiveExecutionClientFactory
The factory to add.
Raises
------
ValueError
If name is not a valid string.
KeyError
If name has already been added.
"""
self._builder.add_exec_client_factory(name, factory)
def build(self) -> None:
"""
Build the nodes clients.
"""
if self._is_built:
raise RuntimeError("The trading nodes clients are already built.")
self._builder.build_data_clients(self._config.get("data_clients"))
self._builder.build_exec_clients(self._config.get("exec_clients"))
self._is_built = True
def start(self) -> None:
"""
Start the trading node.
"""
if not self._is_built:
raise RuntimeError(
"The trading nodes clients have not been built. "
"Please run `node.build()` prior to start.")
try:
if self._loop.is_running():
self._loop.create_task(self._run())
else:
self._loop.run_until_complete(self._run())
except RuntimeError as ex:
self._log.exception(ex)
def stop(self) -> None:
"""
Stop the trading node gracefully.
After a specified delay the internal `Trader` residuals will be checked.
If save strategy is specified then strategy states will then be saved.
"""
try:
if self._loop.is_running():
self._loop.create_task(self._stop())
else:
self._loop.run_until_complete(self._stop())
except RuntimeError as ex:
self._log.exception(ex)
def dispose(self) -> None:
"""
Dispose of the trading node.
Gracefully shuts down the executor and event loop.
"""
try:
timeout = self._clock.utc_now() + timedelta(seconds=5)
while self._is_running:
time.sleep(0.1)
if self._clock.utc_now() >= timeout:
self._log.warning(
"Timed out (5s) waiting for node to stop.")
break
self._log.info("state=DISPOSING...")
self._log.debug(f"{self._data_engine.get_run_queue_task()}")
self._log.debug(f"{self._exec_engine.get_run_queue_task()}")
self._log.debug(f"{self._risk_engine.get_run_queue_task()}")
self.trader.dispose()
self._data_engine.dispose()
self._exec_engine.dispose()
self._risk_engine.dispose()
self._log.info("Shutting down executor...")
if sys.version_info >= (3, 9):
# cancel_futures added in Python 3.9
self._executor.shutdown(wait=True, cancel_futures=True)
else:
self._executor.shutdown(wait=True)
self._log.info("Stopping event loop...")
self._cancel_all_tasks()
self._logger.stop()
self._loop.stop()
except RuntimeError as ex:
self._log.exception(ex)
finally:
if self._loop.is_running():
self._log.warning("Cannot close a running event loop.")
else:
self._log.info("Closing event loop...")
self._loop.close()
# Check and log if event loop is running
if self._loop.is_running():
self._log.warning(f"loop.is_running={self._loop.is_running()}")
else:
self._log.info(f"loop.is_running={self._loop.is_running()}")
# Check and log if event loop is closed
if not self._loop.is_closed():
self._log.warning(f"loop.is_closed={self._loop.is_closed()}")
else:
self._log.info(f"loop.is_closed={self._loop.is_closed()}")
self._log.info("state=DISPOSED.")
def _log_header(self) -> None:
nautilus_header(self._log)
self._log.info(f"redis {redis.__version__}")
self._log.info(
f"msgpack {msgpack.version[0]}.{msgpack.version[1]}.{msgpack.version[2]}"
)
if uvloop_version:
self._log.info(f"uvloop {uvloop_version}")
self._log.info(
"================================================================="
)
def _setup_loop(self) -> None:
if self._loop.is_closed():
self._log.error(
"Cannot setup signal handling (event loop was closed).")
return
signal.signal(signal.SIGINT, signal.SIG_DFL)
signals = (signal.SIGTERM, signal.SIGINT, signal.SIGABRT)
for sig in signals:
self._loop.add_signal_handler(sig, self._loop_sig_handler, sig)
self._log.debug(f"Event loop {signals} handling setup.")
def _loop_sig_handler(self, sig: signal.signal) -> None:
self._loop.remove_signal_handler(signal.SIGTERM)
self._loop.add_signal_handler(signal.SIGINT, lambda: None)
self._log.warning(f"Received {sig!s}, shutting down...")
self.stop()
async def _run(self) -> None:
try:
self._log.info("state=STARTING...")
self._is_running = True
self._logger.start()
self._data_engine.start()
self._exec_engine.start()
self._risk_engine.start()
result: bool = await self._await_engines_connected()
if not result:
return
result: bool = await self._exec_engine.reconcile_state()
if not result:
return
self.trader.start()
if self._loop.is_running():
self._log.info("state=RUNNING.")
else:
self._log.warning("Event loop is not running.")
# Continue to run while engines are running...
await self._data_engine.get_run_queue_task()
await self._exec_engine.get_run_queue_task()
await self._risk_engine.get_run_queue_task()
except asyncio.CancelledError as ex:
self._log.error(str(ex))
async def _await_engines_connected(self) -> bool:
self._log.info(f"Waiting for engines to initialize "
f"({self._connection_timeout}s timeout)...")
# The data engine clients will be set as connected when all
# instruments are received and updated with the data engine.
# The execution engine clients will be set as connected when all
# accounts are updated and the current order and position status is
# reconciled. Thus any delay here will be due to blocking network IO.
seconds = self._connection_timeout
timeout: timedelta = self._clock.utc_now() + timedelta(seconds=seconds)
while True:
await asyncio.sleep(0)
if self._clock.utc_now() >= timeout:
self._log.error(
f"Timed out ({seconds}s) waiting for engines to connect.")
return False
if not self._data_engine.check_connected():
continue
if not self._exec_engine.check_connected():
continue
break
return True # Engines connected
async def _stop(self) -> None:
self._is_stopping = True
self._log.info("state=STOPPING...")
if self.trader.state == ComponentState.RUNNING:
self.trader.stop()
self._log.info(
f"Awaiting residual state ({self._check_residuals_delay}s delay)..."
)
await asyncio.sleep(self._check_residuals_delay)
self.trader.check_residuals()
if self._save_strategy_state:
self.trader.save()
if self._data_engine.state == ComponentState.RUNNING:
self._data_engine.stop()
if self._exec_engine.state == ComponentState.RUNNING:
self._exec_engine.stop()
if self._risk_engine.state == ComponentState.RUNNING:
self._risk_engine.stop()
await self._await_engines_disconnected()
# Clean up remaining timers
timer_names = self._clock.timer_names()
self._clock.cancel_timers()
for name in timer_names:
self._log.info(f"Cancelled Timer(name={name}).")
self._log.info("state=STOPPED.")
self._is_running = False
async def _await_engines_disconnected(self) -> None:
self._log.info(f"Waiting for engines to disconnect "
f"({self._disconnection_timeout}s timeout)...")
seconds = self._disconnection_timeout
timeout: timedelta = self._clock.utc_now() + timedelta(seconds=seconds)
while True:
await asyncio.sleep(0)
if self._clock.utc_now() >= timeout:
self._log.error(
f"Timed out ({seconds}s) waiting for engines to disconnect."
)
break
if not self._data_engine.check_disconnected():
continue
if not self._exec_engine.check_disconnected():
continue
break
def _cancel_all_tasks(self) -> None:
to_cancel = asyncio.tasks.all_tasks(self._loop)
if not to_cancel:
self._log.info("All tasks finished.")
return
for task in to_cancel:
self._log.warning(f"Cancelling pending task {task}")
task.cancel()
if self._loop.is_running():
self._log.warning(
"Event loop still running during `cancel_all_tasks`.")
return
finish_all_tasks: asyncio.Future = asyncio.tasks.gather(
*to_cancel,
loop=self._loop,
return_exceptions=True,
)
self._loop.run_until_complete(finish_all_tasks)
self._log.debug(f"{finish_all_tasks}")
for task in to_cancel:
if task.cancelled():
continue
if task.exception() is not None:
self._loop.call_exception_handler({
"message":
"unhandled exception during asyncio.run() shutdown",
"exception": task.exception(),
"task": task,
})
class TradingNode:
"""
Provides an asynchronous network node for live trading.
"""
def __init__(
self,
strategies: List[TradingStrategy],
config: Dict[str, object],
):
"""
Initialize a new instance of the TradingNode class.
Parameters
----------
strategies : list[TradingStrategy]
The list of strategies to run on the trading node.
config : dict[str, object]
The configuration for the trading node.
Raises
------
ValueError
If strategies is None or empty.
ValueError
If config is None or empty.
"""
PyCondition.not_none(strategies, "strategies")
PyCondition.not_none(config, "config")
PyCondition.not_empty(strategies, "strategies")
PyCondition.not_empty(config, "config")
# Extract configs
config_trader = config.get("trader", {})
config_log = config.get("logging", {})
config_exec_db = config.get("exec_database", {})
config_strategy = config.get("strategy", {})
config_adapters = config.get("adapters", {})
self._uuid_factory = UUIDFactory()
self._loop = asyncio.get_event_loop()
self._executor = concurrent.futures.ThreadPoolExecutor()
self._loop.set_default_executor(self._executor)
self._clock = LiveClock(loop=self._loop)
self.created_time = self._clock.utc_now()
self._is_running = False
# Uncomment for debugging
# self._loop.set_debug(True)
# Setup identifiers
self.trader_id = TraderId(
name=config_trader["name"],
tag=config_trader["id_tag"],
)
# Setup logging
self._logger = LiveLogger(
clock=self._clock,
name=self.trader_id.value,
level_console=LogLevelParser.from_str_py(config_log.get("log_level_console")),
level_file=LogLevelParser.from_str_py(config_log.get("log_level_file")),
level_store=LogLevelParser.from_str_py(config_log.get("log_level_store")),
run_in_process=config_log.get("run_in_process", True), # Run logger in a separate process
log_thread=config_log.get("log_thread_id", False),
log_to_file=config_log.get("log_to_file", False),
log_file_path=config_log.get("log_file_path", ""),
)
self._log = LoggerAdapter(component_name=self.__class__.__name__, logger=self._logger)
self._log_header()
self._log.info("Building...")
self._setup_loop() # Requires the logger to be initialized
self.portfolio = Portfolio(
clock=self._clock,
logger=self._logger,
)
self._data_engine = LiveDataEngine(
loop=self._loop,
portfolio=self.portfolio,
clock=self._clock,
logger=self._logger,
config={"qsize": 10000},
)
self.portfolio.register_cache(self._data_engine.cache)
self.analyzer = PerformanceAnalyzer()
if config_exec_db["type"] == "redis":
exec_db = RedisExecutionDatabase(
trader_id=self.trader_id,
logger=self._logger,
command_serializer=MsgPackCommandSerializer(),
event_serializer=MsgPackEventSerializer(),
config={
"host": config_exec_db["host"],
"port": config_exec_db["port"],
}
)
else:
exec_db = BypassExecutionDatabase(
trader_id=self.trader_id,
logger=self._logger,
)
self._exec_engine = LiveExecutionEngine(
loop=self._loop,
database=exec_db,
portfolio=self.portfolio,
clock=self._clock,
logger=self._logger,
config={"qsize": 10000},
)
self._exec_engine.load_cache()
self._setup_adapters(config_adapters, self._logger)
self.trader = Trader(
trader_id=self.trader_id,
strategies=strategies,
portfolio=self.portfolio,
data_engine=self._data_engine,
exec_engine=self._exec_engine,
clock=self._clock,
logger=self._logger,
)
self._check_residuals_delay = config_trader.get("check_residuals_delay", 5.0)
self._load_strategy_state = config_strategy.get("load_state", True)
self._save_strategy_state = config_strategy.get("save_state", True)
if self._load_strategy_state:
self.trader.load()
self._log.info("state=INITIALIZED.")
self.time_to_initialize = self._clock.delta(self.created_time)
self._log.info(f"Initialized in {self.time_to_initialize.total_seconds():.3f}s.")
@property
def is_running(self) -> bool:
"""
If the trading node is running.
Returns
-------
bool
True if running, else False.
"""
return self._is_running
def get_event_loop(self) -> asyncio.AbstractEventLoop:
"""
Return the event loop of the trading node.
Returns
-------
asyncio.AbstractEventLoop
"""
return self._loop
def get_logger(self) -> LiveLogger:
"""
Return the logger for the trading node.
Returns
-------
LiveLogger
"""
return self._logger
def start(self) -> None:
"""
Start the trading node.
"""
try:
if self._loop.is_running():
self._loop.create_task(self._run())
else:
self._loop.run_until_complete(self._run())
except RuntimeError as ex:
self._log.exception(ex)
def stop(self) -> None:
"""
Stop the trading node gracefully.
After a specified delay the internal `Trader` residuals will be checked.
If save strategy is specified then strategy states will then be saved.
"""
try:
if self._loop.is_running():
self._loop.create_task(self._stop())
else:
self._loop.run_until_complete(self._stop())
except RuntimeError as ex:
self._log.exception(ex)
def dispose(self) -> None:
"""
Dispose of the trading node.
Gracefully shuts down the executor and event loop.
"""
try:
timeout = self._clock.utc_now() + timedelta(seconds=5)
while self._is_running:
time.sleep(0.1)
if self._clock.utc_now() >= timeout:
self._log.warning("Timed out (5s) waiting for node to stop.")
break
self._log.info("state=DISPOSING...")
self._log.debug(f"{self._data_engine.get_run_queue_task()}")
self._log.debug(f"{self._exec_engine.get_run_queue_task()}")
self.trader.dispose()
self._data_engine.dispose()
self._exec_engine.dispose()
self._log.info("Shutting down executor...")
if sys.version_info >= (3, 9):
# cancel_futures added in Python 3.9
self._executor.shutdown(wait=True, cancel_futures=True)
else:
self._executor.shutdown(wait=True)
self._log.info("Stopping event loop...")
self._loop.stop()
self._cancel_all_tasks()
except RuntimeError as ex:
self._log.error("Shutdown coro issues will be fixed soon...") # TODO: Remove when fixed
self._log.exception(ex)
finally:
if self._loop.is_running():
self._log.warning("Cannot close a running event loop.")
else:
self._log.info("Closing event loop...")
self._loop.close()
# Check and log if event loop is running
if self._loop.is_running():
self._log.warning(f"loop.is_running={self._loop.is_running()}")
else:
self._log.info(f"loop.is_running={self._loop.is_running()}")
# Check and log if event loop is closed
if not self._loop.is_closed():
self._log.warning(f"loop.is_closed={self._loop.is_closed()}")
else:
self._log.info(f"loop.is_closed={self._loop.is_closed()}")
self._log.info("state=DISPOSED.")
self._logger.stop() # Ensure process is stopped
time.sleep(0.1) # Ensure final log messages
def _log_header(self) -> None:
nautilus_header(self._log)
self._log.info(f"redis {redis.__version__}")
self._log.info(f"msgpack {msgpack.version[0]}.{msgpack.version[1]}.{msgpack.version[2]}")
if uvloop_version:
self._log.info(f"uvloop {uvloop_version}")
self._log.info("=================================================================")
def _setup_loop(self) -> None:
if self._loop.is_closed():
self._log.error("Cannot setup signal handling (event loop was closed).")
return
signal.signal(signal.SIGINT, signal.SIG_DFL)
signals = (signal.SIGTERM, signal.SIGINT, signal.SIGHUP, signal.SIGABRT)
for sig in signals:
self._loop.add_signal_handler(sig, self._loop_sig_handler, sig)
self._log.debug(f"Event loop {signals} handling setup.")
def _loop_sig_handler(self, sig: signal.signal) -> None:
self._loop.remove_signal_handler(signal.SIGTERM)
self._loop.add_signal_handler(signal.SIGINT, lambda: None)
self._log.warning(f"Received {sig!s}, shutting down...")
self.stop()
def _setup_adapters(self, config: Dict[str, object], logger: LiveLogger) -> None:
# Setup each data client
for name, config in config.items():
if name.startswith("ccxt-"):
try:
import ccxtpro # TODO: Find a better way of doing this
except ImportError:
raise ImportError("ccxtpro is not installed, "
"installation instructions can be found at https://ccxt.pro")
client_cls = getattr(ccxtpro, name.partition('-')[2].lower())
data_client, exec_client = CCXTClientsFactory.create(
client_cls=client_cls,
config=config,
data_engine=self._data_engine,
exec_engine=self._exec_engine,
clock=self._clock,
logger=logger,
)
elif name == "oanda":
data_client = OandaDataClientFactory.create(
config=config,
data_engine=self._data_engine,
clock=self._clock,
logger=logger,
)
exec_client = None # TODO: Implement
else:
self._log.error(f"No adapter available for `{name}`.")
continue
if data_client is not None:
self._data_engine.register_client(data_client)
if exec_client is not None:
self._exec_engine.register_client(exec_client)
async def _run(self) -> None:
try:
self._log.info("state=STARTING...")
self._is_running = True
self._data_engine.start()
self._exec_engine.start()
result: bool = await self._await_engines_connected()
if not result:
return
result: bool = await self._exec_engine.resolve_state()
if not result:
return
self.trader.start()
if self._loop.is_running():
self._log.info("state=RUNNING.")
else:
self._log.warning("Event loop is not running.")
# Continue to run while engines are running...
await self._data_engine.get_run_queue_task()
await self._exec_engine.get_run_queue_task()
except asyncio.CancelledError as ex:
self._log.error(str(ex))
async def _await_engines_connected(self) -> bool:
self._log.info("Waiting for engines to initialize...")
# The data engine clients will be set as connected when all
# instruments are received and updated with the data engine.
# The execution engine clients will be set as connected when all
# accounts are updated and the current order and position status is
# confirmed. Thus any delay here will be due to blocking network IO.
seconds = 5 # Hard coded for now
timeout: timedelta = self._clock.utc_now() + timedelta(seconds=seconds)
while True:
await asyncio.sleep(0.1)
if self._clock.utc_now() >= timeout:
self._log.error(f"Timed out ({seconds}s) waiting for "
f"engines to initialize.")
return False
if not self._data_engine.check_connected():
continue
if not self._exec_engine.check_connected():
continue
break
return True # Engines initialized
async def _stop(self) -> None:
self._is_stopping = True
self._log.info("state=STOPPING...")
if self.trader.state == ComponentState.RUNNING:
self.trader.stop()
self._log.info(f"Awaiting residual state ({self._check_residuals_delay}s delay)...")
await asyncio.sleep(self._check_residuals_delay)
self.trader.check_residuals()
if self._save_strategy_state:
self.trader.save()
if self._data_engine.state == ComponentState.RUNNING:
self._data_engine.stop()
if self._exec_engine.state == ComponentState.RUNNING:
self._exec_engine.stop()
await self._await_engines_disconnected()
# Clean up remaining timers
timer_names = self._clock.timer_names()
self._clock.cancel_timers()
for name in timer_names:
self._log.info(f"Cancelled Timer(name={name}).")
self._log.info("state=STOPPED.")
self._is_running = False
async def _await_engines_disconnected(self) -> None:
self._log.info("Waiting for engines to disconnect...")
seconds = 5 # Hard coded for now
timeout: timedelta = self._clock.utc_now() + timedelta(seconds=seconds)
while True:
await asyncio.sleep(0.1)
if self._clock.utc_now() >= timeout:
self._log.warning(f"Timed out ({seconds}s) waiting for engines to disconnect.")
break
if not self._data_engine.check_disconnected():
continue
if not self._exec_engine.check_disconnected():
continue
break # Engines initialized
def _cancel_all_tasks(self) -> None:
to_cancel = asyncio.tasks.all_tasks(self._loop)
if not to_cancel:
self._log.info("All tasks finished.")
return
for task in to_cancel:
self._log.warning(f"Cancelling pending task {task}")
task.cancel()
if self._loop.is_running():
self._log.warning("Event loop still running during `cancel_all_tasks`.")
return
finish_all_tasks: asyncio.Future = asyncio.tasks.gather(
*to_cancel,
loop=self._loop,
return_exceptions=True,
)
self._loop.run_until_complete(finish_all_tasks)
self._log.debug(f"{finish_all_tasks}")
for task in to_cancel:
if task.cancelled():
continue
if task.exception() is not None:
self._loop.call_exception_handler({
'message': 'unhandled exception during asyncio.run() shutdown',
'exception': task.exception(),
'task': task,
})
class TradingNode:
"""
Provides an asynchronous network node for live trading.
Parameters
----------
config : TradingNodeConfig, optional
The configuration for the instance.
Raises
------
TypeError
If `config` is not of type `TradingNodeConfig`.
"""
def __init__(self, config: Optional[TradingNodeConfig] = None):
if config is None:
config = TradingNodeConfig()
PyCondition.not_none(config, "config")
PyCondition.type(config, TradingNodeConfig, "config")
# Configuration
self._config = config
# Setup loop
self._loop = asyncio.get_event_loop()
self._executor = concurrent.futures.ThreadPoolExecutor()
self._loop.set_default_executor(self._executor)
self._loop.set_debug(config.loop_debug)
# Components
self._clock = LiveClock(loop=self._loop)
self._uuid_factory = UUIDFactory()
self.created_time = self._clock.utc_now()
self._is_running = False
# Identifiers
self.trader_id = TraderId(config.trader_id)
self.machine_id = socket.gethostname()
self.instance_id = self._uuid_factory.generate()
# Setup logging
self._logger = LiveLogger(
loop=self._loop,
clock=self._clock,
trader_id=self.trader_id,
machine_id=self.machine_id,
instance_id=self.instance_id,
level_stdout=LogLevelParser.from_str_py(config.log_level.upper()),
)
self._log = LoggerAdapter(
component_name=type(self).__name__,
logger=self._logger,
)
self._log_header()
self._log.info("Building...")
if platform.system() != "Windows":
# Windows does not support signal handling
# https://stackoverflow.com/questions/45987985/asyncio-loops-add-signal-handler-in-windows
self._setup_loop()
########################################################################
# Build platform
########################################################################
if config.cache_database is None or config.cache_database.type == "in-memory":
cache_db = None
elif config.cache_database.type == "redis":
cache_db = RedisCacheDatabase(
trader_id=self.trader_id,
logger=self._logger,
serializer=MsgPackSerializer(timestamps_as_str=True),
config=config.cache_database,
)
else: # pragma: no cover (design-time error)
raise ValueError(
"The cache_db_type in the configuration is unrecognized, "
"can one of {{'in-memory', 'redis'}}.",
)
self._msgbus = MessageBus(
trader_id=self.trader_id,
clock=self._clock,
logger=self._logger,
)
self._cache = Cache(
database=cache_db,
logger=self._logger,
config=config.cache,
)
self.portfolio = Portfolio(
msgbus=self._msgbus,
cache=self._cache,
clock=self._clock,
logger=self._logger,
)
self._data_engine = LiveDataEngine(
loop=self._loop,
msgbus=self._msgbus,
cache=self._cache,
clock=self._clock,
logger=self._logger,
config=config.data_engine,
)
self._exec_engine = LiveExecutionEngine(
loop=self._loop,
msgbus=self._msgbus,
cache=self._cache,
clock=self._clock,
logger=self._logger,
config=config.exec_engine,
)
self._exec_engine.load_cache()
self._risk_engine = LiveRiskEngine(
loop=self._loop,
portfolio=self.portfolio,
msgbus=self._msgbus,
cache=self._cache,
clock=self._clock,
logger=self._logger,
config=config.risk_engine,
)
self.trader = Trader(
trader_id=self.trader_id,
msgbus=self._msgbus,
cache=self._cache,
portfolio=self.portfolio,
data_engine=self._data_engine,
risk_engine=self._risk_engine,
exec_engine=self._exec_engine,
clock=self._clock,
logger=self._logger,
)
if config.load_strategy_state:
self.trader.load()
# Setup persistence (requires trader)
self.persistence_writers: List[Any] = []
if config.persistence:
self._setup_persistence(config=config.persistence)
self._builder = TradingNodeBuilder(
loop=self._loop,
data_engine=self._data_engine,
exec_engine=self._exec_engine,
msgbus=self._msgbus,
cache=self._cache,
clock=self._clock,
logger=self._logger,
log=self._log,
)
self._log.info("INITIALIZED.")
self.time_to_initialize = self._clock.delta(self.created_time)
self._log.info(f"Initialized in {int(self.time_to_initialize.total_seconds() * 1000)}ms.")
self._is_built = False
@property
def is_running(self) -> bool:
"""
If the trading node is running.
Returns
-------
bool
"""
return self._is_running
@property
def is_built(self) -> bool:
"""
If the trading node clients are built.
Returns
-------
bool
"""
return self._is_built
def get_event_loop(self) -> asyncio.AbstractEventLoop:
"""
Return the event loop of the trading node.
Returns
-------
asyncio.AbstractEventLoop
"""
return self._loop
def get_logger(self) -> LiveLogger:
"""
Return the logger for the trading node.
Returns
-------
LiveLogger
"""
return self._logger
def add_log_sink(self, handler: Callable[[Dict], None]):
"""
Register the given sink handler with the nodes logger.
Parameters
----------
handler : Callable[[Dict], None]
The sink handler to register.
Raises
------
KeyError
If `handler` already registered.
"""
self._logger.register_sink(handler=handler)
def add_data_client_factory(self, name: str, factory):
"""
Add the given data client factory to the node.
Parameters
----------
name : str
The name of the client factory.
factory : LiveDataClientFactory or LiveExecutionClientFactory
The factory to add.
Raises
------
ValueError
If `name` is not a valid string.
KeyError
If `name` has already been added.
"""
self._builder.add_data_client_factory(name, factory)
def add_exec_client_factory(self, name: str, factory):
"""
Add the given execution client factory to the node.
Parameters
----------
name : str
The name of the client factory.
factory : LiveDataClientFactory or LiveExecutionClientFactory
The factory to add.
Raises
------
ValueError
If `name` is not a valid string.
KeyError
If `name` has already been added.
"""
self._builder.add_exec_client_factory(name, factory)
def build(self) -> None:
"""
Build the nodes clients.
"""
if self._is_built:
raise RuntimeError("the trading nodes clients are already built.")
self._builder.build_data_clients(self._config.data_clients)
self._builder.build_exec_clients(self._config.exec_clients)
self._is_built = True
def start(self) -> Optional[asyncio.Task]:
"""
Start the trading node.
"""
if not self._is_built:
raise RuntimeError(
"The trading nodes clients have not been built. "
"Please run `node.build()` prior to start."
)
try:
if self._loop.is_running():
return self._loop.create_task(self._run())
else:
self._loop.run_until_complete(self._run())
return None
except RuntimeError as ex:
self._log.exception("Error on run", ex)
return None
def stop(self) -> None:
"""
Stop the trading node gracefully.
After a specified delay the internal `Trader` residuals will be checked.
If save strategy is specified then strategy states will then be saved.
"""
try:
if self._loop.is_running():
self._loop.create_task(self._stop())
else:
self._loop.run_until_complete(self._stop())
except RuntimeError as ex:
self._log.exception("Error on stop", ex)
def dispose(self) -> None:
"""
Dispose of the trading node.
Gracefully shuts down the executor and event loop.
"""
try:
timeout = self._clock.utc_now() + timedelta(seconds=self._config.timeout_disconnection)
while self._is_running:
time.sleep(0.1)
if self._clock.utc_now() >= timeout:
self._log.warning(
f"Timed out ({self._config.timeout_disconnection}s) waiting for node to stop."
f"\nStatus"
f"\n------"
f"\nDataEngine.check_disconnected() == {self._data_engine.check_disconnected()}"
f"\nExecEngine.check_disconnected() == {self._exec_engine.check_disconnected()}"
)
break
self._log.info("DISPOSING...")
self._log.debug(f"{self._data_engine.get_run_queue_task()}")
self._log.debug(f"{self._exec_engine.get_run_queue_task()}")
self._log.debug(f"{self._risk_engine.get_run_queue_task()}")
self.trader.dispose()
self._data_engine.dispose()
self._exec_engine.dispose()
self._risk_engine.dispose()
self._log.info("Shutting down executor...")
if sys.version_info >= (3, 9):
# cancel_futures added in Python 3.9
self._executor.shutdown(wait=True, cancel_futures=True)
else:
self._executor.shutdown(wait=True)
self._log.info("Stopping event loop...")
self._cancel_all_tasks()
self._loop.stop()
except RuntimeError as ex:
self._log.exception("Error on dispose", ex)
finally:
if self._loop.is_running():
self._log.warning("Cannot close a running event loop.")
else:
self._log.info("Closing event loop...")
self._loop.close()
# Check and log if event loop is running
if self._loop.is_running():
self._log.warning(f"loop.is_running={self._loop.is_running()}")
else:
self._log.info(f"loop.is_running={self._loop.is_running()}")
# Check and log if event loop is closed
if not self._loop.is_closed():
self._log.warning(f"loop.is_closed={self._loop.is_closed()}")
else:
self._log.info(f"loop.is_closed={self._loop.is_closed()}")
self._log.info("DISPOSED.")
def _log_header(self) -> None:
nautilus_header(self._log)
self._log.info(f"redis {redis.__version__}") # type: ignore
self._log.info(f"msgpack {msgpack.version[0]}.{msgpack.version[1]}.{msgpack.version[2]}")
if uvloop_version:
self._log.info(f"uvloop {uvloop_version}")
self._log.info("\033[36m=================================================================")
def _setup_loop(self) -> None:
if self._loop.is_closed():
self._log.error("Cannot setup signal handling (event loop was closed).")
return
signal.signal(signal.SIGINT, signal.SIG_DFL)
signals = (signal.SIGTERM, signal.SIGINT, signal.SIGABRT)
for sig in signals:
self._loop.add_signal_handler(sig, self._loop_sig_handler, sig)
self._log.debug(f"Event loop signal handling setup for {signals}.")
def _setup_persistence(self, config: PersistenceConfig) -> None:
# Setup persistence
path = f"{config.catalog_path}/live/{self.instance_id}.feather"
writer = FeatherWriter(
path=path,
fs_protocol=config.fs_protocol,
flush_interval=config.flush_interval,
)
self.persistence_writers.append(writer)
self.trader.subscribe("*", writer.write)
self._log.info(f"Persisting data & events to {path=}")
# Setup logging
if config.persist_logs:
def sink(record, f):
f.write(orjson.dumps(record) + b"\n")
path = f"{config.catalog_path}/logs/{self.instance_id}.log"
log_sink = open(path, "wb")
self.persistence_writers.append(log_sink)
self._logger.register_sink(partial(sink, f=log_sink))
self._log.info(f"Persisting logs to {path=}")
def _loop_sig_handler(self, sig) -> None:
self._loop.remove_signal_handler(signal.SIGTERM)
self._loop.add_signal_handler(signal.SIGINT, lambda: None)
self._log.warning(f"Received {sig!s}, shutting down...")
self.stop()
async def _run(self) -> None:
try:
self._log.info("STARTING...")
self._is_running = True
# Start system
self._logger.start()
self._data_engine.start()
self._exec_engine.start()
self._risk_engine.start()
# Connect all clients
self._data_engine.connect()
self._exec_engine.connect()
# Await engine connection and initialization
self._log.info(
f"Waiting for engines to connect and initialize "
f"({self._config.timeout_connection}s timeout)...",
color=LogColor.BLUE,
)
if not await self._await_engines_connected():
self._log.warning(
f"Timed out ({self._config.timeout_connection}s) waiting for engines to connect and initialize."
f"\nStatus"
f"\n------"
f"\nDataEngine.check_connected() == {self._data_engine.check_connected()}"
f"\nExecEngine.check_connected() == {self._exec_engine.check_connected()}"
)
return
self._log.info("Engines connected.", color=LogColor.GREEN)
# Await execution state reconciliation
self._log.info(
f"Waiting for execution state to reconcile "
f"({self._config.timeout_reconciliation}s timeout)...",
color=LogColor.BLUE,
)
if not await self._exec_engine.reconcile_state(
timeout_secs=self._config.timeout_reconciliation,
):
self._log.error("Execution state could not be reconciled.")
return
self._log.info("State reconciled.", color=LogColor.GREEN)
# Initialize portfolio
self.portfolio.initialize_orders()
self.portfolio.initialize_positions()
# Await portfolio initialization
self._log.info(
"Waiting for portfolio to initialize "
f"({self._config.timeout_portfolio}s timeout)...",
color=LogColor.BLUE,
)
if not await self._await_portfolio_initialized():
self._log.warning(
f"Timed out ({self._config.timeout_portfolio}s) waiting for portfolio to initialize."
f"\nStatus"
f"\n------"
f"\nPortfolio.initialized == {self.portfolio.initialized}"
)
return
self._log.info("Portfolio initialized.", color=LogColor.GREEN)
# Start trader and strategies
self.trader.start()
if self._loop.is_running():
self._log.info("RUNNING.")
else:
self._log.warning("Event loop is not running.")
# Continue to run while engines are running...
await self._data_engine.get_run_queue_task()
await self._exec_engine.get_run_queue_task()
await self._risk_engine.get_run_queue_task()
except asyncio.CancelledError as ex:
self._log.error(str(ex))
async def _await_engines_connected(self) -> bool:
# - The data engine clients will be set connected when all
# instruments are received and updated with the data engine.
# - The execution engine clients will be set connected when all
# accounts are updated and the current order and position status is
# reconciled.
# Thus any delay here will be due to blocking network I/O.
seconds = self._config.timeout_connection
timeout: timedelta = self._clock.utc_now() + timedelta(seconds=seconds)
while True:
await asyncio.sleep(0)
if self._clock.utc_now() >= timeout:
return False
if not self._data_engine.check_connected():
continue
if not self._exec_engine.check_connected():
continue
break
return True # Engines connected
async def _await_portfolio_initialized(self) -> bool:
# - The portfolio will be set initialized when all margin and unrealized
# PnL calculations are completed (maybe waiting on first quotes).
# Thus any delay here will be due to blocking network I/O.
seconds = self._config.timeout_portfolio
timeout: timedelta = self._clock.utc_now() + timedelta(seconds=seconds)
while True:
await asyncio.sleep(0)
if self._clock.utc_now() >= timeout:
return False
if not self.portfolio.initialized:
continue
break
return True # Portfolio initialized
async def _stop(self) -> None:
self._is_stopping = True
self._log.info("STOPPING...")
if self.trader.is_running:
self.trader.stop()
self._log.info(
f"Awaiting residual state ({self._config.check_residuals_delay}s delay)...",
color=LogColor.BLUE,
)
await asyncio.sleep(self._config.check_residuals_delay)
self.trader.check_residuals()
if self._config.save_strategy_state:
self.trader.save()
# Disconnect all clients
self._data_engine.disconnect()
self._exec_engine.disconnect()
if self._data_engine.is_running:
self._data_engine.stop()
if self._exec_engine.is_running:
self._exec_engine.stop()
if self._risk_engine.is_running:
self._risk_engine.stop()
self._log.info(
f"Waiting for engines to disconnect "
f"({self._config.timeout_disconnection}s timeout)...",
color=LogColor.BLUE,
)
if not await self._await_engines_disconnected():
self._log.error(
f"Timed out ({self._config.timeout_disconnection}s) waiting for engines to disconnect."
f"\nStatus"
f"\n------"
f"\nDataEngine.check_disconnected() == {self._data_engine.check_disconnected()}"
f"\nExecEngine.check_disconnected() == {self._exec_engine.check_disconnected()}"
)
# Clean up remaining timers
timer_names = self._clock.timer_names()
self._clock.cancel_timers()
for name in timer_names:
self._log.info(f"Cancelled Timer(name={name}).")
# Clean up persistence
for writer in self.persistence_writers:
writer.close()
self._log.info("STOPPED.")
self._logger.stop()
self._is_running = False
async def _await_engines_disconnected(self) -> bool:
seconds = self._config.timeout_disconnection
timeout: timedelta = self._clock.utc_now() + timedelta(seconds=seconds)
while True:
await asyncio.sleep(0)
if self._clock.utc_now() >= timeout:
return False
if not self._data_engine.check_disconnected():
continue
if not self._exec_engine.check_disconnected():
continue
break
return True # Engines disconnected
def _cancel_all_tasks(self) -> None:
to_cancel = asyncio.tasks.all_tasks(self._loop)
if not to_cancel:
self._log.info("All tasks canceled.")
return
for task in to_cancel:
self._log.warning(f"Canceling pending task {task}")
task.cancel()
if self._loop.is_running():
self._log.warning("Event loop still running during `cancel_all_tasks`.")
return
finish_all_tasks: asyncio.Future = asyncio.tasks.gather( # type: ignore
*to_cancel,
loop=self._loop,
return_exceptions=True,
)
self._loop.run_until_complete(finish_all_tasks)
self._log.debug(f"{finish_all_tasks}")
for task in to_cancel: # pragma: no cover
if task.cancelled():
continue
if task.exception() is not None:
self._loop.call_exception_handler(
{
"message": "unhandled exception during asyncio.run() shutdown",
"exception": task.exception(),
"task": task,
}
)
