import logging import pickle from collections import deque from queue import Queue, Empty from datetime import datetime from threading import RLock, Thread from .tl import types as tl __log__ = logging.getLogger(__name__) class UpdateState: """Used to hold the current state of processed updates. To retrieve an update, .poll() should be called. """ 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 = [] self.handlers = [] self._updates_lock = RLock() self._updates = Queue() # 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 not self._updates.empty() 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. """ try: update = self._updates.get(timeout=timeout) except Empty: return 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: while self._updates: self._updates.get() 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.put(StopIteration()) 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: # We don't want to crash a worker thread due to any reason __log__.exception('Unhandled exception on worker %d', wid) 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 with self._updates_lock: if isinstance(update, tl.updates.State): __log__.debug('Saved new updates state') self._state = update return # Nothing else to be done if hasattr(update, 'pts'): self._state.pts = update.pts # After running the script for over an hour and receiving over # 1000 updates, the only duplicates received were users going # online or offline. We can trust the server until new reports. if isinstance(update, tl.UpdateShort): self._updates.put(update.update) # 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. elif isinstance(update, (tl.Updates, tl.UpdatesCombined)): for u in update.updates: self._updates.put(u) # TODO Handle "tl.UpdatesTooLong" else: self._updates.put(update)