import os

import numpy as np
from joblib import Parallel, delayed, parallel_config


def work():
    arr = np.ones((1_000_000,))
    for _ in range(1_000):
        arr.sum()
    return arr.sum()


def start_thread_pool():
    with parallel_config(backend="threading"):
        return Parallel(n_jobs=-1)([delayed(work)() for _ in range(10)])


with parallel_config(backend="threading", n_jobs=os.cpu_count() // 2):
    expected = work()
    for _ in Parallel()(
        [delayed(start_thread_pool)() for _ in range(os.cpu_count() // 2)]
    ):
        pass