Server shutdown conflict with Numba

[Giofraant]

Before submitting the issue

• I have checked for Compatibility issues
• I have searched among the existing issues
• I am using a Python virtual environment

Description of the bug

I am currently trying to do numba accelerated post processing on data, acquired from an .rst file, by using the Ansys DPF core library. When I try to shut down the server to enable numba, it seems that the server still exists, which prevents numba from performing, resulting in the following error:

Traceback (most recent call last):
  File "User\AppData\Roaming\Python\Python311\site-packages\ansys\dpf\core\dpf_operator.py", line 732, in __del__
AttributeError: 'NoneType' object has no attribute 'warn'
Exception ignored in: <function Operator.__del__ at 0x0000027DB55F5C60>

Followed by many identical errors.

I'm unsure how to fix this or if I'm simply on the wrong track and what I have to do to fix this issue.

Steps To Reproduce

# to measure speeds of mockup calculations
import time 

import numpy as np
import ansys.dpf.core as dpf

from numba import njit
from pathlib import Path

SERVERCONFIG = dpf.ServerConfig(protocol=dpf.server.CommunicationProtocols.InProcess, legacy=False)
SERVER = dpf.start_local_server(
    config=SERVERCONFIG,
    as_global=False,
    load_operators=False,
    use_docker_by_default=False,
    use_pypim_by_default=False
    )

my_path = Path(r"example_bracket.rst")
my_source = dpf.DataSources(my_path, server= SERVER)

my_scope = dpf.Model(my_source, server= SERVER)
my_result = my_scope.results.stress

my_array = np.copy(my_result.eval()[0].data)

del(my_source)

# see if server exists
print(dpf.server.available_servers())

# kill it
dpf.server.shutdown_all_session_servers()
dpf.server.shutdown_global_server()

# try again
print(dpf.server.available_servers())

# The following functions are random, computational heavy calculations to see performance results

def arraydicolous(arr: np.array, anumber: int):
    num = np.sum(arr)
    for i in range(anumber+1):
        num += np.sum(arr * i)
    return num

@njit
def anumbadicolous(arr: np.array, anumber: int):
    num = np.sum(arr)
    for i in range(anumber+1):
        num += np.sum(arr * i)
    return num

# measure both functions' time.
test1_t1 = time.time()
print(arraydicolous(my_array, 200000))
test1_t2 = time.time()

print(test1_t2-test1_t1)

test2_t1 = time.time()
print(anumbadicolous(my_array, 200000))
test2_t2 = time.time()

print(test2_t2-test2_t1)

Which Operating System causes the issue?

Windows

Which DPF/Ansys version are you using?

Ansys 2024 R2

Which Python version causes the issue?

3.11

Installed packages

ansys-dpf-core==0.13.3
numba @ file:///C:/b/abs_51yyu3qucu/croot/numba_1720540987599/work
numpy==1.26.4

[MeenaBytes]

@Giofraant How about if you move the PyDPF code inside another function?

import time

import numpy as np

from numba import njit
from pathlib import Path
from ansys.dpf import core as dpf
def function():
    
    rst = r"C:\Users\4565734\.ansys_python_venvs\pydpf_demo_1202\Lib\site-packages\ansys\dpf\core\examples\result_files\ASimpleBar.rst"

    SERVERCONFIG = dpf.ServerConfig(protocol=dpf.server.CommunicationProtocols.InProcess, legacy=False)
    SERVER = dpf.start_local_server(
    config=SERVERCONFIG,
    as_global=False,
    load_operators=False,
    use_docker_by_default=False,
    use_pypim_by_default=False
    )

    my_path = Path(rst)
    my_source = dpf.DataSources(my_path, server= SERVER)
    my_scope = dpf.Model(my_source, server= SERVER)
    my_result = my_scope.results.displacement
    return my_result.eval()[0].data


my_array = function()

def arraydicolous(arr: np.array, anumber: int):
    num = np.sum(arr)
    for i in range(anumber+1):
        num += np.sum(arr * i)
    return num

@njit
def anumbadicolous(arr: np.array, anumber: int):
    num = np.sum(arr)
    for i in range(anumber+1):
        num += np.sum(arr * i)
    return num


test1_t1 = time.time()
print(arraydicolous(my_array, 200000))
test1_t2 = time.time()

print(test1_t2-test1_t1)

test2_t1 = time.time()
print(anumbadicolous(my_array, 200000))
test2_t2 = time.time()

print(test2_t2-test2_t1)

[PProfizi]

Hi @Giofraant , hi @MeenaBytes ,

I've reproduced the posted issue.
I've also found that by removing the njit decorator the script works without any warning at closure.
It looks to me like some kind of conflict with numba as you described.
Even if the array used is a numpy array made from scratch, simply having a DPF server running raises similar warnings.
I will investigate this further.

[PProfizi]

@ansys-akarcher FYI

[Giofraant]

Good day and happy new years to you @MeenaBytes & @PProfizi,

Thank you in advance for investigating this further. If there is anything I can do to assist, please let me know.

For now I will try to experiment with this issue and I will keep you in the loop.

[Giofraant]

Hi @Giofraant , hi @MeenaBytes ,

I've reproduced the posted issue. I've also found that by removing the njit decorator the script works without any warning at closure. It looks to me like some kind of conflict with numba as you described. Even if the array used is a numpy array made from scratch, simply having a DPF server running raises similar warnings. I will investigate this further.

Unfortunately I haven't found any solution yet.. @PProfizi any findings?