Flaky Windows CI: subprocess teardown tests fail intermittently #6

Problem

Several tests fail intermittently on Windows CI runners (windows-latest). The failures appear across different Python versions (3.10, 3.14) and are not version-specific — they hit whichever run lands on a slow runner.

This is a pre-existing issue on main (observed in run 23162193291) and is not caused by any recent changes.

Failing tests

CI error details

The ResourceWarning and PytestUnraisableExceptionWarning errors stem from Windows TerminateProcess() not allowing graceful cleanup of _ProactorBasePipeTransport, Popen, and StreamWriter objects.

Suggested fixes

1. Timing-sensitive process cleanup tests (test_stdio.py)

Replace single-check assertions with polling loops that retry with short intervals inside a generous anyio.fail_after() timeout, rather than checking file growth once after a fixed delay.

2. Command execution timeout (test_config.py)

Increase subprocess.run(..., timeout=20) to a more CI-friendly value (e.g., 60 seconds).

3. Unraisable exception warnings (test_memory.py, test_stdio.py)

Extend the existing ResourceWarning filter to also cover PytestUnraisableExceptionWarning on Windows:

@pytest.mark.filterwarnings(
    "ignore::pytest.PytestUnraisableExceptionWarning" if sys.platform == "win32" else "default"
)

Some tests already filter ResourceWarning on Windows but don't filter the PytestUnraisableExceptionWarning that wraps it.

References

• Failing on main: CI run 23162193291 (checks / test (3.10, locked, windows-latest))
• Failing on feature/multi-tenant-managers: CI run 23197887963 (checks / test (3.14, locked, windows-latest))