Skip to content

Latest commit

 

History

History
114 lines (79 loc) · 4.55 KB

README.md

File metadata and controls

114 lines (79 loc) · 4.55 KB

FairAsyncRLock

Python PyPI

Main Status: Workflow name

Develop Status: Workflow name

This is a well-tested standalone implementation of a fair reentrant lock for conncurrent programming with asyncio. This was built because python decided not to support RLock in asyncio, their argument being that every extra bit of functionality adds to maintenance cost.

Install with

pip install fair-async-rlock

About Fair Reentrant Lock for AsyncIO

A reentrant lock (or recursive lock) is a particular type of lock that can be "locked" multiple times by the same task without causing a deadlock. This is in contrast to a standard lock (sometimes called a "non-reentrant lock"), which could cause a deadlock if a task attempts to lock it multiple times. A "Fair" Reentrant Lock respects the order of lock granting, so that all lock acquirers get acquisition in the order they asked for it.

A reentrant lock can be useful in a number of scenarios:

  • Recursive function calls: If you have a function that can be called recursively, and that function needs to lock a resource, a reentrant lock would be useful. This is because the same task will be able to acquire the lock multiple times, once for each recursive call.

  • Complex operations: If you're performing a complex operation that consists of multiple steps, and each of these steps needs to lock a resource, a reentrant lock might be necessary. This is especially true if the steps are performed by the same task and if the steps can overlap in some way.

  • Improve readability: Reentrant locks can sometimes make code easier to understand and debug. This is because you don't need to worry about a task deadlocking itself by attempting to acquire a lock it already holds.

  • Task ownership: If your code logic depends on which task owns a lock, reentrant locks can be helpful. They inherently track ownership since they are tied to the task that acquired them first.

This code demonstrates the reentrant property:

from fair_async_rlock import FairAsyncRLock


async def reentrant():
    lock = FairAsyncRLock()
    async with lock:
        async with lock:  # This will not block, whereas it would with asyncio.Lock
            assert True

In the context of Python's asyncio, FairAsyncRLock can be beneficial when used with coroutines that might call each other or be invoked recursively. They also help when a single coroutine holds a lock and wants to ensure fairness among other coroutines waiting for the same lock, so that no coroutine is left waiting for the lock indefinitely while others after it acquire the lock, which can happen with standard asyncio Locks.

This code demonstrates fairness:

import asyncio

from fair_async_rlock import FairAsyncRLock


async def fairness():
    lock = FairAsyncRLock()
    num_tasks = 1000
    order = []

    async def worker(n):
        async with lock:
            order.append(n)
            await asyncio.sleep(0)  # emulate work

    # Start several tasks to acquire the lock
    tasks = [asyncio.create_task(worker(i)) for i in range(num_tasks)]

    # Make sure they all start and try to acquire the lock before releasing it
    await asyncio.sleep(0)

    await asyncio.gather(*tasks)

    assert order == list(range(num_tasks))  # The tasks have run in order

Potential downsides

It's also important to note the potential downsides of using a reentrant lock. For one, they can hide bugs in code that isn't properly synchronized, since the same task can acquire a reentrant lock multiple times without issue.

Non-reentrant locks are often simpler and can expose synchronization bugs more easily, so you should only use a reentrant lock if you have a specific need for it.

Performance

The performance is about 50% slower than asyncio.Lock, i.e. overhead of sequential locks is about 3/2 of same with asyncio.Lock.

Change Log

27 Jan, 2024 - 1.0.7 released. Fixed a bug that allowed another task to get the lock before a waiter got its turn on the event loop.