Preventing numpy from Multithreading

Question

I know the following question has been ask here already multiple times but sadly none of the suggested solutions seem to works for me.

I tried to get numpy (and scipy) to stop multithreading. The most common posted answer for this was to set some environmental variables to "1", i.e.

import os
N_THREADS = '1'
os.environ['OMP_NUM_THREADS'] = N_THREADS
os.environ['OPENBLAS_NUM_THREADS'] = N_THREADS
os.environ['MKL_NUM_THREADS'] = N_THREADS
os.environ['VECLIB_MAXIMUM_THREADS'] = N_THREADS
os.environ['NUMEXPR_NUM_THREADS'] = N_THREADS

import numpy as np


while True:
    np.linalg.inv(np.identity(100)) 

However, this Processes still uses 70 or so threads. :/

I also checked weather the environmental variables are set corretly with

for name, value in os.environ.items():
    print("{0}: {1}".format(name, value))

And indeed the variables show up:

OMP_NUM_THREADS: 1
OPENBLAS_NUM_THREADS: 1
MKL_NUM_THREADS: 1
VECLIB_MAXIMUM_THREADS: 1
NUMEXPR_NUM_THREADS: 1

An alternative i saw was using a function, i.e.

import mkl
mkl.set_num_threads(1)

However, i still get the same issue.

Next i checked which libary numpy is using with

np.__config__.show()

The results are

openblas64__info:
    libraries = ['openblas64_', 'openblas64_']
    library_dirs = ['/usr/local/lib']
    language = c
    define_macros = [('HAVE_CBLAS', None), ('BLAS_SYMBOL_SUFFIX', '64_'), ('HAVE_BLAS_ILP64', None)]
    runtime_library_dirs = ['/usr/local/lib']
blas_ilp64_opt_info:
    libraries = ['openblas64_', 'openblas64_']
    library_dirs = ['/usr/local/lib']
    language = c
    define_macros = [('HAVE_CBLAS', None), ('BLAS_SYMBOL_SUFFIX', '64_'), ('HAVE_BLAS_ILP64', None)]
    runtime_library_dirs = ['/usr/local/lib']
openblas64__lapack_info:
    libraries = ['openblas64_', 'openblas64_']
    library_dirs = ['/usr/local/lib']
    language = c
    define_macros = [('HAVE_CBLAS', None), ('BLAS_SYMBOL_SUFFIX', '64_'), ('HAVE_BLAS_ILP64', None), ('HAVE_LAPACKE', None)]
    runtime_library_dirs = ['/usr/local/lib']
lapack_ilp64_opt_info:
    libraries = ['openblas64_', 'openblas64_']
    library_dirs = ['/usr/local/lib']
    language = c
    define_macros = [('HAVE_CBLAS', None), ('BLAS_SYMBOL_SUFFIX', '64_'), ('HAVE_BLAS_ILP64', None), ('HAVE_LAPACKE', None)]
    runtime_library_dirs = ['/usr/local/lib']
Supported SIMD extensions in this NumPy install:
    baseline = SSE,SSE2,SSE3
    found = SSSE3,SSE41,POPCNT,SSE42,AVX,F16C,FMA3,AVX2
    not found = AVX512F,AVX512CD,AVX512_KNL,AVX512_KNM,AVX512_SKX,AVX512_CLX,AVX512_CNL,AVX512_ICL

So my numpy seems to be using openblas and not mkl. Next it tried using this random code i found in the net:

def disable_openblas_threading():
    """
    A convenience function for turning off openblas threading to avoid costly overhead.

    Just setting the `OPENBLAS_NUM_THREADS` environment variable to `1` would be much simpler, but
    that only works if the user hasn't already imported `numpy`.  This function attempts to use
    `ctypes` to load the OpenBLAS library and access the `openblas_set_num_threads` function, which
    will work even if the user already imported numpy or scipy.
    """
    import numpy as np
    import ctypes
    from ctypes.util import find_library

    try:
        np_lib_dir = np.__config__.__dict__['openblas_info']['library_dirs'][0]
    except KeyError:
        np_lib_dir = None

    try_paths = ['{}/libopenblas.so'.format(np_lib_dir),
                 '{}/libopenblas.dylib'.format(np_lib_dir),
                 '/opt/OpenBLAS/lib/libopenblas.so',
                 '/lib/libopenblas.so',
                 '/usr/lib/libopenblas.so.0',
                 find_library('openblas')]

    openblas_lib = None
    for path in try_paths:
        try:
            openblas_lib = ctypes.cdll.LoadLibrary(path)
        except OSError:
            continue

    try:
        openblas_lib.openblas_set_num_threads(1)
    except AttributeError:
        raise EnvironmentError('Could not locate an OpenBLAS shared library', 2)

It still doesn't work :/.

Now i'm pretty much out of ideas what i can do. Do you have any ideas?

One fact that is maybe relevant for my problem is that i'm working on a pc-cluster but i asked some people responsible for managing the cluster and they said it shouldn't influence my problem.

I never get any error messages but simply my process is using to many threads.

I'm using numpy 1.22.1

Edit: I count the threads by looking at htop before and after starting the script.

Update: Using threadpoolctl solved the Problem :), i.e.

import numpy as np
from threadpoolctl import threadpool_limits

with threadpool_limits(limits=1, user_api='blas'):
    while True:
        np.linalg.inv(np.identity(100)) 

Thanks.