Source code for betty.concurrent

"""
Provide utilities for concurrent programming.
"""

import asyncio
import multiprocessing
import threading
import time
from abc import ABC, abstractmethod
from asyncio import sleep
from collections.abc import Hashable
from ctypes import c_longdouble
from math import floor
from multiprocessing.managers import SyncManager
from types import TracebackType
from typing import final, MutableMapping, TypeVar, Self

from typing_extensions import override

from betty.typing import processsafe
from betty.warnings import deprecate

_KeyT = TypeVar("_KeyT")
_ValueT = TypeVar("_ValueT")


[docs] class Lock(ABC): """ Provide an asynchronous lock. """ async def __aenter__(self): await self.acquire() async def __aexit__( self, exc_type: type[BaseException] | None, exc_val: BaseException | None, exc_tb: TracebackType | None, ) -> None: await self.release()
[docs] @abstractmethod async def acquire(self, *, wait: bool = True) -> bool: """ Acquire the lock. """ pass
[docs] @abstractmethod async def release(self) -> None: """ Release the lock. """ pass
[docs] async def asynchronize_acquire(lock: threading.Lock, *, wait: bool = True) -> bool: """ Acquire a synchronous lock asynchronously. """ while not lock.acquire(blocking=False): if not wait: return False # Sleeping for zero seconds does not actually sleep, but gives the event # loop a chance to progress other tasks while we wait for another chance # to acquire the lock. await sleep(0) return True
[docs] @final class AsynchronizedLock(Lock): """ Make a synchronous (blocking) lock asynchronous (non-blocking). """ __slots__ = "_lock"
[docs] def __init__(self, lock: threading.Lock): self._lock = lock
@property def lock(self) -> threading.Lock: """ The underlying, synchronous lock. """ return self._lock
[docs] @override async def acquire(self, *, wait: bool = True) -> bool: return await asynchronize_acquire(self._lock, wait=wait)
[docs] @override async def release(self) -> None: self._lock.release()
[docs] @classmethod def threading(cls) -> Self: """ Create a new thread-safe, asynchronous lock. """ return cls(threading.Lock())
[docs] @final @processsafe class RateLimiter: """ Rate-limit operations. This class implements the `Token Bucket algorithm <https://en.wikipedia.org/wiki/Token_bucket>`_. """
[docs] def __init__( self, maximum: int, period: int = 1, *, manager: SyncManager | None = None ): manager = ensure_manager(manager) self._lock = AsynchronizedLock(manager.Lock()) self._maximum = maximum self._period = period self._available = manager.Value(c_longdouble, maximum) # A Token Bucket fills as time passes. However, we want callers to be able to start # using the limiter immediately, so we 'preload' the first's period's tokens, and # set the last added time to the end of the first period. This ensures there is no # needless waiting if the number of tokens consumed in total is less than the limit # per period. self._last_add = manager.Value(c_longdouble, time.monotonic() + self._period)
def _add_tokens(self): now = time.monotonic() elapsed = now - self._last_add.value added = elapsed * self._maximum possibly_available = floor(self._available.value + added) if possibly_available > 0: self._available.value = min(possibly_available, self._maximum) self._last_add.value = now async def __aenter__(self) -> None: await self.wait() async def __aexit__( self, exc_type: type[BaseException] | None, exc_val: BaseException | None, exc_tb: TracebackType | None, ) -> None: return
[docs] async def is_available(self) -> bool: """ Whether an operation may be performed (again). """ async with self._lock: self._add_tokens() return self._available.value != 0
[docs] async def wait(self) -> None: """ Wait until an operation may be performed (again). """ async with self._lock: while self._available.value < 1: self._add_tokens() if self._available.value < 1: await asyncio.sleep(0) self._available.value -= 1
class _Transaction(Lock): def __init__( self, transaction_id: Hashable, ledger_lock: Lock, ledger: MutableMapping[Hashable, bool], ): self._transaction_id = transaction_id self._ledger_lock = ledger_lock self._ledger = ledger @override async def acquire(self, *, wait: bool = True) -> bool: if wait: while True: async with self._ledger_lock: if self._can_acquire(): return self._acquire() await sleep(0) else: async with self._ledger_lock: if self._can_acquire(): return self._acquire() return False def _can_acquire(self) -> bool: try: return not self._ledger[self._transaction_id] except KeyError: self._ledger[self._transaction_id] = False return True def _acquire(self) -> bool: self._ledger[self._transaction_id] = True return True @override async def release(self) -> None: self._ledger[self._transaction_id] = False
[docs] @processsafe class Ledger: """ Lazily create locks by keeping a ledger. The ledger lock is released once a transaction lock is acquired. """
[docs] def __init__(self, ledger_lock: Lock, *, manager: SyncManager | None = None): manager = ensure_manager(manager) self._ledger_lock = ledger_lock self._ledger: MutableMapping[Hashable, bool] = manager.dict()
[docs] def ledger(self, transaction_id: Hashable) -> Lock: """ Ledger a new lock for the given transaction ID. """ return _Transaction(transaction_id, self._ledger_lock, self._ledger)
[docs] def ensure_manager(manager: SyncManager | None, *, stacklevel: int = 1) -> SyncManager: """ Ensure that a value is a multiprocessing manager. """ if manager: return manager deprecate( "Not providing a multiprocessing manager is deprecated as of Betty 0.4.10.", stacklevel=stacklevel, ) return multiprocessing.Manager()