Update handlers works; it also seems stable

This commit is contained in:
Andrey Egorov
2017-10-22 15:06:36 +03:00
parent 917665852d
commit 780e0ceddf
9 changed files with 300 additions and 265 deletions

View File

@@ -1,8 +1,8 @@
import logging
import pickle
import asyncio
from collections import deque
from datetime import datetime
from threading import RLock, Event, Thread
from .tl import types as tl
@@ -13,177 +13,72 @@ class UpdateState:
"""
WORKER_POLL_TIMEOUT = 5.0 # Avoid waiting forever on the workers
def __init__(self, workers=None):
"""
:param workers: This integer parameter has three possible cases:
workers is None: Updates will *not* be stored on self.
workers = 0: Another thread is responsible for calling self.poll()
workers > 0: 'workers' background threads will be spawned, any
any of them will invoke all the self.handlers.
"""
self._workers = workers
self._worker_threads = []
def __init__(self, loop=None):
self.handlers = []
self._updates_lock = RLock()
self._updates_available = Event()
self._updates = deque()
self._latest_updates = deque(maxlen=10)
self._loop = loop if loop else asyncio.get_event_loop()
self._logger = logging.getLogger(__name__)
# https://core.telegram.org/api/updates
self._state = tl.updates.State(0, 0, datetime.now(), 0, 0)
def can_poll(self):
"""Returns True if a call to .poll() won't lock"""
return self._updates_available.is_set()
def poll(self, timeout=None):
"""Polls an update or blocks until an update object is available.
If 'timeout is not None', it should be a floating point value,
and the method will 'return None' if waiting times out.
"""
if not self._updates_available.wait(timeout=timeout):
return
with self._updates_lock:
if not self._updates_available.is_set():
return
update = self._updates.popleft()
if not self._updates:
self._updates_available.clear()
if isinstance(update, Exception):
raise update # Some error was set through (surely StopIteration)
return update
def get_workers(self):
return self._workers
def set_workers(self, n):
"""Changes the number of workers running.
If 'n is None', clears all pending updates from memory.
"""
self.stop_workers()
self._workers = n
if n is None:
self._updates.clear()
else:
self.setup_workers()
workers = property(fget=get_workers, fset=set_workers)
def stop_workers(self):
"""Raises "StopIterationException" on the worker threads to stop them,
and also clears all of them off the list
"""
if self._workers:
with self._updates_lock:
# Insert at the beginning so the very next poll causes an error
# on all the worker threads
# TODO Should this reset the pts and such?
for _ in range(self._workers):
self._updates.appendleft(StopIteration())
self._updates_available.set()
for t in self._worker_threads:
t.join()
self._worker_threads.clear()
def setup_workers(self):
if self._worker_threads or not self._workers:
# There already are workers, or workers is None or 0. Do nothing.
return
for i in range(self._workers):
thread = Thread(
target=UpdateState._worker_loop,
name='UpdateWorker{}'.format(i),
daemon=True,
args=(self, i)
)
self._worker_threads.append(thread)
thread.start()
def _worker_loop(self, wid):
while True:
try:
update = self.poll(timeout=UpdateState.WORKER_POLL_TIMEOUT)
# TODO Maybe people can add different handlers per update type
if update:
for handler in self.handlers:
handler(update)
except StopIteration:
break
except Exception as e:
# We don't want to crash a worker thread due to any reason
self._logger.debug(
'[ERROR] Unhandled exception on worker {}'.format(wid), e
)
def handle_update(self, update):
for handler in self.handlers:
asyncio.ensure_future(handler(update), loop=self._loop)
def process(self, update):
"""Processes an update object. This method is normally called by
the library itself.
"""
if self._workers is None:
return # No processing needs to be done if nobody's working
if isinstance(update, tl.updates.State):
self._state = update
return # Nothing else to be done
with self._updates_lock:
if isinstance(update, tl.updates.State):
self._state = update
return # Nothing else to be done
pts = getattr(update, 'pts', self._state.pts)
if hasattr(update, 'pts') and pts <= self._state.pts:
return # We already handled this update
pts = getattr(update, 'pts', self._state.pts)
if hasattr(update, 'pts') and pts <= self._state.pts:
return # We already handled this update
self._state.pts = pts
self._state.pts = pts
# TODO There must be a better way to handle updates rather than
# keeping a queue with the latest updates only, and handling
# the 'pts' correctly should be enough. However some updates
# like UpdateUserStatus (even inside UpdateShort) will be called
# repeatedly very often if invoking anything inside an update
# handler. TODO Figure out why.
"""
client = TelegramClient('anon', api_id, api_hash, update_workers=1)
client.connect()
def handle(u):
client.get_me()
client.add_update_handler(handle)
input('Enter to exit.')
"""
data = pickle.dumps(update.to_dict())
if data in self._latest_updates:
return # Duplicated too
# TODO There must be a better way to handle updates rather than
# keeping a queue with the latest updates only, and handling
# the 'pts' correctly should be enough. However some updates
# like UpdateUserStatus (even inside UpdateShort) will be called
# repeatedly very often if invoking anything inside an update
# handler. TODO Figure out why.
"""
client = TelegramClient('anon', api_id, api_hash, update_workers=1)
client.connect()
def handle(u):
client.get_me()
client.add_update_handler(handle)
input('Enter to exit.')
"""
data = pickle.dumps(update.to_dict())
if data in self._latest_updates:
return # Duplicated too
self._latest_updates.append(data)
self._latest_updates.append(data)
if type(update).SUBCLASS_OF_ID == 0x8af52aac: # crc32(b'Updates')
# Expand "Updates" into "Update", and pass these to callbacks.
# Since .users and .chats have already been processed, we
# don't need to care about those either.
if isinstance(update, tl.UpdateShort):
self.handle_update(update.update)
if type(update).SUBCLASS_OF_ID == 0x8af52aac: # crc32(b'Updates')
# Expand "Updates" into "Update", and pass these to callbacks.
# Since .users and .chats have already been processed, we
# don't need to care about those either.
if isinstance(update, tl.UpdateShort):
self._updates.append(update.update)
self._updates_available.set()
elif isinstance(update, (tl.Updates, tl.UpdatesCombined)):
for upd in update.updates:
self.handle_update(upd)
elif isinstance(update, (tl.Updates, tl.UpdatesCombined)):
self._updates.extend(update.updates)
self._updates_available.set()
elif not isinstance(update, tl.UpdatesTooLong):
# TODO Handle "Updates too long"
self.handle_update(update)
elif not isinstance(update, tl.UpdatesTooLong):
# TODO Handle "Updates too long"
self._updates.append(update)
self._updates_available.set()
elif type(update).SUBCLASS_OF_ID == 0x9f89304e: # crc32(b'Update')
self._updates.append(update)
self._updates_available.set()
else:
self._logger.debug('Ignoring "update" of type {}'.format(
type(update).__name__)
)
elif type(update).SUBCLASS_OF_ID == 0x9f89304e: # crc32(b'Update')
self.handle_update(update)
else:
self._logger.debug('Ignoring "update" of type {}'.format(
type(update).__name__)
)