Python LRU cache with TTL. Decorator.

lru_with_ttl.py

import time
from functools import lru_cache


def lru_with_ttl(*, ttl_seconds, maxsize=128):
    """
    A decorator to apply LRU in-memory cache to a function with defined maximum(!) TTL in seconds.
    Be design an actual TTL may be shorter then the passed value (in rare randomized cases). But it can't be higher.

    :param ttl_seconds: TTL for a cache record in seconds
    :param maxsize: Maximum size of the LRU cache (a functools.lru_cache argument)
    :return: decorated function
    """
    def deco(foo):
        @lru_cache(maxsize=maxsize)
        def cached_with_ttl(*args, ttl_hash, **kwargs):
            return foo(*args, **kwargs)

        def inner(*args, **kwargs):
            return cached_with_ttl(*args, ttl_hash=round(time.time() / ttl_seconds), **kwargs)

        return inner
    return deco

This is an elegant idea, but it doesn't really implement a proper TTL cache. As it's written it divides the time line into windows of length ttl_seconds and then returns a cached result if the function has already been called with the same arguments in the same time window. This means that if you have a pair of calls in quick succession but one happens to land at the end of a time window and the second happens at the start of the next time window then the cached value is not used.