Not sure why this is suddenly happening now and only on the NumPy nightly
wheels, but it migbt be relevant that this warning was seen as a bug by
upstream and they made the resource_tracker reentrant-safe:
https://github.com/python/cpython/issues/131788

This is merged in `main`, backported to 3.13 and released in 3.13.7, but
has not yet been backported to 3.14 because that branch is in the RC freeze.

I happened to know about this because it caused some breakage in joblib,
but that’s been fixed for a month or so.

On Mon, Sep 22, 2025 at 8:05 PM Andrew Nelson via NumPy-Discussion <
[email protected]> wrote:

> Hi all,
> in the last few weeks have there been any PRs that have touched objects
> that are followed by the Python ResourceTracker. I'm talking about things
> like Locks, semaphores, mutexes, etc?
>
> I'm seeing some weird behaviour in scipy CI where a test (
> https://github.com/scipy/scipy/blob/main/scipy/optimize/tests/test_optimize.py#L3442)
> that uses an object that wraps `multiprocessing.Pool` has started giving
> test fails. A typical log (
> https://github.com/scipy/scipy/actions/runs/17925659616/job/50971069339#step:12:1229)
> has text like:
>
> "multiprocessing.resource_tracker.ReentrantCallError: Reentrant call into
> the multiprocessing resource tracker"
>  and
> "UserWarning: ResourceTracker called reentrantly for resource cleanup,
> which is unsupported. The semaphore object '/mp-380ciby1' might leak."
>
> In my investigations (https://github.com/andyfaff/scipy/pull/83) it only
> seems to happen for cp312 and the numpy nightly wheel. It doesn't seem to
> trigger for cp312+np2.3.3 or for cp311.
>
> Given that it only seems to happen with the numpy nightly wheel I'm
> wondering if it's therefore a numpy issue. As far as I can tell it's
> started happening in the last five days or so.
>
>
> --
> _____________________________________
> Dr. Andrew Nelson
>
>
> _____________________________________
> _______________________________________________
> NumPy-Discussion mailing list -- [email protected]
> To unsubscribe send an email to [email protected]
> https://mail.python.org/mailman3//lists/numpy-discussion.python.org
> Member address: [email protected]
>
_______________________________________________
NumPy-Discussion mailing list -- [email protected]
To unsubscribe send an email to [email protected]
https://mail.python.org/mailman3//lists/numpy-discussion.python.org
Member address: [email protected]

Reply via email to