run bot in a fresh interpreter

main.py

@@ -2,10 +2,11 @@ from __future__ import annotations
 
 import os
 import signal
+import subprocess
 import sys
+import time
 from multiprocessing import Process, get_context
 from pathlib import Path
-from collections.abc import Callable
 from typing import BinaryIO
 
 
@@ -54,66 +55,61 @@ def stop_processes(processes: list[Process]) -> None:
             process.join()
 
 
-def run_worker(target: Callable[[], None], lock_fd: int) -> None:
-    # Keep an explicit duplicate alive in the worker. A Linux flock belongs to
-    # the shared open-file description, so it remains held if the parent dies.
-    with os.fdopen(os.dup(lock_fd), "rb"):
-        target()
+def stop_subprocesses(processes: list[subprocess.Popen]) -> None:
+    for process in processes:
+        if process.poll() is None:
+            process.terminate()
+    for process in processes:
+        try:
+            process.wait(SHUTDOWN_TIMEOUT)
+        except subprocess.TimeoutExpired:
+            process.kill()
+            process.wait()
 
 
 def run() -> int:
     # Import only after taking the lock so a rejected duplicate has no bot/event
-    # side effects. Linux workers explicitly retain the lock descriptor: if the
-    # supervisor dies unexpectedly, an orphan cannot poll alongside a replacement.
+    # side effects. Linux workers are fresh interpreters which explicitly inherit
+    # the lock descriptor. If the supervisor dies, an orphan keeps the lock and a
+    # replacement cannot poll alongside it.
     instance_lock = acquire_instance_lock()
-    import bot
-    import event
     import models.logger
 
-    context = get_context("spawn" if sys.platform == "win32" else "fork")
-    if sys.platform == "win32":
-        processes = [
-            context.Process(target=bot.run, name="bot"),
-            context.Process(target=event.run, name="events"),
-        ]
-    else:
-        processes = [
-            context.Process(
-                target=run_worker,
-                args=(bot.run, instance_lock.fileno()),
-                name="bot",
-            ),
-            context.Process(
-                target=run_worker,
-                args=(event.run, instance_lock.fileno()),
-                name="events",
-            ),
-        ]
+    shutdown_requested = False
 
     def request_shutdown(_signum, _frame) -> None:
-        raise KeyboardInterrupt
+        nonlocal shutdown_requested
+        shutdown_requested = True
 
     signal.signal(signal.SIGINT, request_shutdown)
     signal.signal(signal.SIGTERM, request_shutdown)
 
     os.system("cls" if sys.platform == "win32" else "clear")
     models.logger.log("Main", "Startup...")
+
+    if sys.platform == "win32":
+        import bot
+        import event
+
+        context = get_context("spawn")
+        processes: list[Process] = []
+        definitions = [("bot", bot.run), ("events", event.run)]
         try:
-        for process in processes:
+            for name, target in definitions:
+                process = context.Process(target=target, name=name)
                 process.start()
+                processes.append(process)
 
-        while all(process.is_alive() for process in processes):
-            for process in processes:
-                process.join(timeout=0.5)
+            while not shutdown_requested and all(process.is_alive() for process in processes):
+                time.sleep(0.5)
 
+            if not shutdown_requested:
                 stopped = next(process for process in processes if not process.is_alive())
                 models.logger.error(
                     "Main",
                     f"{stopped.name} process exited unexpectedly with code {stopped.exitcode}",
                 )
                 return 1
-    except KeyboardInterrupt:
-        models.logger.error("Main", "Stopping...")
             return 0
         except Exception as error:
             models.logger.error("Main", f"{error}")
@@ -122,6 +118,41 @@ def run() -> int:
             stop_processes(processes)
             instance_lock.close()
 
+    project_dir = Path(__file__).resolve().parent
+    definitions = [
+        ("bot", "import bot; bot.run()"),
+        ("events", "import event; event.run()"),
+    ]
+    processes: list[tuple[str, subprocess.Popen]] = []
+    try:
+        for name, command in definitions:
+            process = subprocess.Popen(
+                [sys.executable, "-c", command],
+                cwd=project_dir,
+                pass_fds=(instance_lock.fileno(),),
+            )
+            processes.append((name, process))
+
+        while not shutdown_requested and all(process.poll() is None for _, process in processes):
+            time.sleep(0.5)
+
+        if not shutdown_requested:
+            name, stopped = next(
+                (name, process) for name, process in processes if process.poll() is not None
+            )
+            models.logger.error(
+                "Main",
+                f"{name} process exited unexpectedly with code {stopped.returncode}",
+            )
+            return 1
+        return 0
+    except Exception as error:
+        models.logger.error("Main", f"{error}")
+        return 1
+    finally:
+        stop_subprocesses([process for _, process in processes])
+        instance_lock.close()
+
 
 if __name__ == "__main__":
     try: