Atomic LRU

What is this?

This is a thread-safe and dependency-free in-memory LRU storage Python 3.10+ library with optional Time To Live (TTL).

You can define:

• limits (max_items or size_limit_in_bytes)
• TTL expiration (globally or per item)

to prevent the storage from growing too big.

You will get an automatic LRU eviction of the least recently used items when the limits are reached.

Features

• Thread-Safe
• (optional) TTL expiration (globally or per item)
• (optional) Total size limit (in bytes) ¹
• (optional) Max items limit
• Automatic LRU eviction (when the limits are reached)
• Full-typing support
• High level Cache API with automatic serialization/deserialization ²
• Low level Storage API without serialization/deserialization (store only references to given objects)

Quickstart

Installation

pip install atomic-lru (or equivalent for your package manager)

High level API (with automatic serialization/deserialization)

The main use-case is to use it as a cache for your data. You store any kind of data type which will be automatically serialized to bytes. ²

from atomic_lru import CACHE_MISS, Cache

# Create a Cache object instance (with a size limit of 1MB)
# (this object is thread-safe, so you can use it from multiple threads)
cache = Cache(size_limit_in_bytes=1_000_000, default_ttl=3600)

# Let's store something (a dictionary here) in the cache with a custom TTL
cache.set(key="user:123", value={"name": "Alice", "age": 30}, ttl=60)

# ...

# Let's retrieve it
user = cache.get(key="user:123")

if user is not CACHE_MISS:
    # cache hit
    print(user["name"])

# Always close to stop the background expiration thread
cache.close()

Low level API (without serialization/deserialization)

But you can use it at a lower level to store any kind of data type without serialization. In that case, you will lose the size_limit_in_bytes feature but you still get the max_items feature.

from atomic_lru import CACHE_MISS, Storage


class ExpensiveObject:
    """An expensive object that is not serializable."""

    pass


# Create a Storage object instance to store ExpensiveObject instances
# (this object is thread-safe, so you can use it from multiple threads)
storage = Storage[ExpensiveObject](max_items=100, default_ttl=3600)

# Create and store an ExpensiveObject instance
value = ExpensiveObject()
storage.set("key1", value, ttl=60)

# ...

# Let's retrieve it
obj = storage.get("key1")

if obj is not CACHE_MISS:
    # cache hit
    assert isinstance(obj, ExpensiveObject)
    assert id(obj) == id(value)  # this is the same object instance

# Always close to stop the background expiration thread
storage.close()

Full API reference

Refer to the API reference for the full API.

DEV

This library is managed with uv and a Makefile. Execute:

• uv sync to create the virtual environment
• make lint to lint the code (style, checks, types, architecture) and fix obvious things
• make test to execute unit tests
• make doc to generate the documentation

See https://docs.astral.sh/uv/getting-started/installation/ to install uv if you need to.