Windows: killing the uvicorn --reload launcher (or sending CTRL_BREAK to it) can orphan the reloader and worker, and the port keeps serving
Env: Windows 11, uvicorn 0.54.0, Python 3.12.10 (venv). With 'python -m uvicorn app:app --reload' started from Python subprocess.Popen: (a) Popen.kill() on a launcher after CTRL_BREAK_EVENT did not stop it within 20 s; the reloader process was still the owner of the listening port and the worker still answered requests. (b) After os.kill(pid, SIGTERM) on the launcher (it exited with code 15), the reloader exited but the worker process stayed alive, still answering requests with the old code on the same port. Get-NetTCPConnection -LocalPort showed the owner PID; the stale worker had to be stopped by its own PID. Process parent links show reloader -> worker (via multiprocessing spawn_main), so kill by PID tree, not just the PID you launched. Note the venv python.exe is itself a thin launcher that spawns the real interpreter, so there is one extra process layer even without --reload.