:info: Simple lru cache for asyncio

.. image:: https://github.com/aio-libs/async-lru/actions/workflows/ci-cd.yml/badge.svg?event=push :target: https://github.com/aio-libs/async-lru/actions/workflows/ci-cd.yml?query=event:push :alt: GitHub Actions CI/CD workflows status

.. image:: https://img.shields.io/pypi/v/async-lru.svg?logo=Python&logoColor=white :target: https://pypi.org/project/async-lru :alt: async-lru @ PyPI

.. image:: https://codecov.io/gh/aio-libs/async-lru/branch/master/graph/badge.svg :target: https://codecov.io/gh/aio-libs/async-lru

.. image:: https://img.shields.io/matrix/aio-libs:matrix.org?label=Discuss%20on%20Matrix%20at%20%23aio-libs%3Amatrix.org&logo=matrix&server_fqdn=matrix.org&style=flat :target: https://matrix.to/#/%23aio-libs:matrix.org :alt: Matrix Room — #aio-libs:matrix.org

.. image:: https://img.shields.io/matrix/aio-libs-space:matrix.org?label=Discuss%20on%20Matrix%20at%20%23aio-libs-space%3Amatrix.org&logo=matrix&server_fqdn=matrix.org&style=flat :target: https://matrix.to/#/%23aio-libs-space:matrix.org :alt: Matrix Space — #aio-libs-space:matrix.org

.. code-block:: shell

pip install async-lru

This package is a port of Python's built-in functools.lru_cache <https://docs.python.org/3/library/functools.html#functools.lru_cache>_ function for asyncio <https://docs.python.org/3/library/asyncio.html>_. To better handle async behaviour, it also ensures multiple concurrent calls will only result in 1 call to the wrapped function, with all await\s receiving the result of that call when it completes.

.. code-block:: python

import asyncio

import aiohttp
from async_lru import alru_cache


@alru_cache(maxsize=32)
async def get_pep(num):
    resource = 'http://www.python.org/dev/peps/pep-%04d/' % num
    async with aiohttp.ClientSession() as session:
        try:
            async with session.get(resource) as s:
                return await s.read()
        except aiohttp.ClientError:
            return 'Not Found'


async def main():
    for n in 8, 290, 308, 320, 8, 218, 320, 279, 289, 320, 9991:
        pep = await get_pep(n)
        print(n, len(pep))

    print(get_pep.cache_info())
    # CacheInfo(hits=3, misses=8, maxsize=32, currsize=8)

    # closing is optional, but highly recommended
    await get_pep.cache_close()


asyncio.run(main())

TTL (time-to-live in seconds, expiration on timeout) is supported by accepting ttl configuration parameter (off by default):

.. code-block:: python

@alru_cache(ttl=5)
async def func(arg):
    return arg * 2

To prevent thundering herd issues when many cache entries expire simultaneously, you can add jitter to randomize the TTL for each entry:

.. code-block:: python

@alru_cache(ttl=3600, jitter=1800)
async def func(arg):
    return arg * 2

With ttl=3600, jitter=1800, each cache entry will have a random TTL between 3600 and 5400 seconds, spreading out invalidations over time.

The library supports explicit invalidation for specific function call by cache_invalidate():

.. code-block:: python

@alru_cache(ttl=5)
async def func(arg1, arg2):
    return arg1 + arg2

func.cache_invalidate(1, arg2=2)

The method returns True if corresponding arguments set was cached already, False otherwise.

To check whether a specific set of arguments is present in the cache without affecting hit/miss counters or LRU ordering, use cache_contains():

.. code-block:: python

@alru_cache(maxsize=32)
async def func(arg1, arg2):
    return arg1 + arg2

await func(1, arg2=2)

func.cache_contains(1, arg2=2)  # True
func.cache_contains(3, arg2=4)  # False

The method returns True if the result for the given arguments is cached, False otherwise.

Event Loop Affinity: alru_cache enforces that a cache instance is used with only one event loop. If you attempt to use a cached function from a different event loop than where it was first called, a RuntimeError will be raised:

.. code-block:: text

RuntimeError: alru_cache is not safe to use across event loops: this cache
instance was first used with a different event loop.
Use separate cache instances per event loop.

For typical asyncio applications using a single event loop, this is automatic and requires no configuration. If your application uses multiple event loops, create separate cache instances per loop:

.. code-block:: python

import threading

_local = threading.local()

def get_cached_fetcher():
    if not hasattr(_local, 'fetcher'):
        @alru_cache(maxsize=100)
        async def fetch_data(key):
            ...
        _local.fetcher = fetch_data
    return _local.fetcher

You can also reuse the logic of an already decorated function in a new loop by accessing __wrapped__:

.. code-block:: python

@alru_cache(maxsize=32)
async def my_task(x):
    ...

# In Loop 1:
# my_task() uses the default global cache instance

# In Loop 2 (or a new thread):
# Create a fresh cache instance for the same logic
cached_task_loop2 = alru_cache(maxsize=32)(my_task.__wrapped__)
await cached_task_loop2(x)

async-lru uses CodSpeed <https://codspeed.io/>_ for performance regression testing.

To run the benchmarks locally:

.. code-block:: shell

pip install -r requirements-dev.txt
pytest --codspeed benchmark.py

The benchmark suite covers both bounded (with maxsize) and unbounded (no maxsize) cache configurations. Scenarios include:

• Cache hit
• Cache miss
• Cache fill/eviction (cycling through more keys than maxsize)
• Cache clear
• TTL expiry
• Cache invalidation
• Cache info retrieval
• Concurrent cache hits
• Baseline (uncached async function)

On CI, benchmarks are run automatically via GitHub Actions on Python 3.13, and results are uploaded to CodSpeed (if a CODSPEED_TOKEN is configured). You can view performance history and detect regressions on the CodSpeed dashboard.

The library was donated by Ocean S.A. <https://ocean.io/>_

Thanks to the company for contribution.