Memory leak when use lru_cache

[tanaliko]

Bug report

Bug description:

memory leak when running code:

from functools import *

@lru_cache(maxsize=100_000)
def f(n):
    if n==1 : return 3
    if n>1:  return 5*f(n-1)

step=995
for n in range(1,10**12,step):
    t=f(n)
    if (n-1)%(step*10)==0:
        print(n,(n*100)/10**12, f.cache_info())

more cache size more fast leak

CPython versions tested on:

3.14

Operating systems tested on:

Windows

[picnixz]

What do you mean by "memory leak"? where is the leak?

[sergey-miryanov]

I think it is related to #144345

[picnixz]

AFAICT, the other issue is about instances being cached here the problem seems to be for any usage to lru_cache.

[sunmy2019]

It seem irrelevant with lru_cache.

I wrote a plain cache implementation to replace lru_cache, and lru_cache uses almost same memory with the plain implementation.

Note your integer value increases exponentially, the size it uses increases linearly as expected.

Script on linux

# works on linux
import os
import sys
from collections import OrderedDict
from functools import lru_cache

LRU_SIZE = 100_000


@lru_cache(maxsize=LRU_SIZE)
def f(n):
    if n == 1:
        return 3
    if n > 1:
        return 5 * f(n - 1)


cache = OrderedDict()


def f1(n):
    if n in cache:
        return cache[n]

    if n == 1:
        return 3

    if n > 1:
        result = 5 * f1(n - 1)
        cache[n] = result
        if len(cache) > LRU_SIZE:
            cache.popitem(last=False)
        return result


def resident_size_mb():
    with open("/proc/self/statm", "r", encoding="ascii") as statm:
        fields = statm.read().split()
    return int(fields[1]) * os.sysconf("SC_PAGE_SIZE") // 1024 // 1024


step = 995
for n in range(1, 10**12, step):
    t = f(n)
    t1 = f1(n)
    if (n - 1) % (step * 10) == 0:
        total_size_of_cache = sum(
            sys.getsizeof(v) for v in cache.values()
        ) + sys.getsizeof(cache)
        print(
            f"n={n}, len(cache)={len(cache)}, total_size_of_cache={total_size_of_cache // 1024 // 1024} MB, resident_size={resident_size_mb()} MB"
        )

(expect resident size == 2 * total_size_of_cache)
n=378101, len(cache)=100000, total_size_of_cache=9706 MB, resident_size=19486 MB
n=388051, len(cache)=100000, total_size_of_cache=10000 MB, resident_size=20075 MB
n=398001, len(cache)=100000, total_size_of_cache=10294 MB, resident_size=20665 MB