shared_memory object results in incorrect memory statistics

[GYHHAHA]

Test code usage: Launch 4 processes and make a sum operation on the shared numpy array with shared_memory in each process.

from multiprocessing import shared_memory
import time
from functools import partial
import numpy as np
from multiprocessing import Pool

def f(shape, dtype, name, n):
    my_sm = shared_memory.SharedMemory(name=name)
    arr = np.ndarray(shape=shape, dtype=dtype, buffer=my_sm.buf)
    time.sleep(n)
    arr.sum()
    time.sleep(n)
if __name__ == "__main__":
    p = Pool(4)
    arr = np.random.rand(int(5e8))
    shm = shared_memory.SharedMemory(create=True, size=arr.nbytes)
    shm_arr = np.ndarray(shape=arr.shape, dtype=arr.dtype, buffer=shm.buf)
    shm_arr[:] = arr[:]
    del arr
    f_ = partial(f, shm_arr.shape, shm_arr.dtype, shm.name)
    p.map(f_, [10, 10, 10, 10])

Issue: The windows system memory monitor's result is different from memory profiler, and I believe the former one is correct.

Thus some overcounts happen here. Thanks for your attention on this.

[shaleenanuj]

Is this issue solved. I am even stuck with the same problem. Is there a flag that I can switch off to count shared variable only once?