Modified version of python's cpu_count that takes into account system constraints to calculate the number of available CPUs

The Python standard library offers an implementation of cpu_count that returns the real number of CPUs even when they are not actually available to be used by the python process (due to constraints such as CPU affinity or CPU scheduler configurations). This is the preferred behaviour for most applications. However, when the interest is the amount of CPUs available for data processing, that approach could be misleading. Especially when it is used behind the scenes such as by the concurrent.futures.Executor when defining its defaults.

The purpose of this module is to provide this functionality in an API equal to the standard implementation. By taking into account the described constraints, this implementation attempts to return the amount of usable CPUs that are available. If no constraint is identified the result will be the same as the standard implementation.

pip install cpu_count

This is the standard way. Just import and call cpu_count

from cpu_count import cpu_count

print(cpu_count())
# $> 8

This an alternative way that replaces python's standard cpu_count with the one from this module (Affected internal modules are posix, os and multiprocessing). The advantage of this approach is not needing to port any code, just import and call setup_monkey_patch ate the begin of your application and everything will just work™.

Note: This will also impact the behaviour of standard libraries that use this function

import os
from cpu_count.monkey_patch import setup_monkey_patch

print(os.cpu_count())
# $> 12

setup_monkey_patch()

print(os.cpu_count())
# $> 8

This approach has one limitation: it can't replace previous code that imported the standard implementation using from os import cpu_count.

from os import cpu_count
from cpu_count.monkey_patch import setup_monkey_patch

print(cpu_count())
# $> 12

setup_monkey_patch()

print(cpu_count())
# $> 12

This approach also replaces python's standard cpu_count. However instead of calling this module's setup_monkey_patch in your application code, it will be called at python startup. For this to work you need to create a file called cpu_count_monkey_patch.pth at your python's global or local site-package's folder with the following content:

import cpu_count; cpu_count.monkey_patch.setup_monkey_patch()

NOTE: This approach is specially useful when creating container images of python applications. An example of using this on a Dockerfile can be found here

• Add logic for Realtime Scheduler constraint
• Create unit tests

All contributions are very welcome.

Code styles is defined in the Editorconfig file. Besides that I use black and isort for auto-format, their configurations are defined in the Editorconfig and the black.toml files respectively. Also I use mypy for statical type check, it's configurations are in the .mypy file.

Thanks to @tomMoral for his loky project, from which I took the base code for this implementation.

Thanks to @Lothiraldan for his in-depth guide on how to monkey patch python: python-production-monkey-patching, which helped me a lot when constructing this module.

BSD 3-Clause “New” or “Revised” License

See LICENSE