Tests of multi-threaded workloads return random results

[pedrocamargo]

Description

We have an Open-Source transportation modeling library for Python that has multiple multi-threaded workloads (Cython code where we release the GIL).
When testing one of the multi-threaded features, tests pass normally in Ubuntu and in all Python versions locally but fail somewhat randomly on Windows in Github Actions (Mostly Python 3.7 and 3.11).

When setting the workload to use a single core , tests started passing in all platforms and Python versions.

Platforms affected

• Azure DevOps
• GitHub Actions - Standard Runners
• GitHub Actions - Larger Runners

Runner images affected

• Ubuntu 20.04
• Ubuntu 22.04
• macOS 11
• macOS 12
• macOS 13
• Windows Server 2019
• Windows Server 2022

Image version and build link

Runner version: '2.306.0'

https://github.com/AequilibraE/aequilibrae/actions/runs/5643403056/job/15285191180?pr=437

Is it regression?

yes

Expected behavior

Tests passing in all platformson

Actual behavior

Tests randomly failing on Windows only

Repro steps

We have left the PR where the issue can be seen here:

AequilibraE/aequilibrae#437

The last commit, c4acf7e9618667029d3aec13ff4651fcfa07f4e3, passes all tests.

The immediately previous commit, c547f61bdf45e2bb323e6ce4fcf939641842c959, fails the multi-threaded workflow.

[v-sergei-pyshnoi]

Hello @pedrocamargo , we will take a look.

[mikhailkoliada]

@dmitry-shibanov @marko-zivic-93 @MaksimZhukov transferring to setup-python, as it looks purely python-specific and not the image specific, if you need help, let us know

[dmitry-shibanov]

Hello @pedrocamargo. Thank you for your report. I think it is not related to setup-python because the actions uses official python binaries for Windows. We do not modify or build them from the source.

[dmitry-shibanov]

Hello @pedrocamargo. For now I'm going to close the issue because it is not related to the setup action but to the set up of environment or project.