[quinnmil]

Problem

CI has been failing on master since #866 (and on #868) across every Python test job. Two unpinned test dependencies released new versions between the last green run and now:

1. pytest 9.1.1 → 9.2.0

pytest aborts with INTERNALERROR (exit code 3) before collecting any tests:

pytest.PytestDeprecationWarning: The configuration option "asyncio_default_fixture_loop_scope" is unset.

With pytest 9.2.0, pytest-asyncio's configure-time deprecation warning is turned into an error by our filterwarnings = error in pytest.ini.

2. rq 2.12.0 → 3.0.0

Once tests could run, test_rq.py::test_hello and test_fail failed in every env (assert 1 == 2 on span count). rq < 3.0 called PERSIST on the job key inside Job.perform, so it was captured as a Redis/PERSIST span within Scout's job span. rq 3.0 moved that call into the worker's pipeline. The instrumentation itself is unaffected — the Job span, tags, and error tagging are unchanged; the tests had hard-coded an rq internal.

Fix

• Set asyncio_default_fixture_loop_scope = function in pytest.ini (the scope pytest-asyncio says will become its default).
• Only expect the Redis/PERSIST span in the RQ tests on rq < 3.0.

Testing

• With the repo's pytest.ini and pytest-asyncio 1.4.1, a trivial test crashes on pytest 9.2.0 without the setting and passes with it; still passes on 9.1.1.
• tests/integration/test_rq.py passes (3/3) against a local Redis on both rq 3.0.0 and rq 2.12.0.

🤖 Generated with Claude Code

https://claude.ai/code/session_01NgEia2KuXhw1rAKi99X1ow

[jrothrock]

Confirmed the repo has no async fixtures so function changes no test behavior, and every tox env (py310+) pulls a pytest-asyncio that knows the key — clean fix. 👍