Improve production.py robustness with orphan detection and kill command

production.py

@@ -52,22 +52,65 @@ def _process_alive(pid: int) -> bool:
         return False
 
 
+def _find_all_pids() -> list[int]:
+    """Search 'ps' for all Gunicorn/Bastebin processes associated with this directory."""
+    pids = []
+    try:
+        # Use 'ps' to find processes related to this Gunicorn instance
+        out = subprocess.check_output(['ps', 'wax', '-o', 'pid,command'], text=True)
+        for line in out.splitlines():
+            # Match gunicorn running from this directory or with this config
+            if 'gunicorn' in line and (BASE_DIR in line or CONF_FILE in line):
+                if 'production.py' in line: continue
+                parts = line.strip().split()
+                if parts:
+                    try:
+                        pids.append(int(parts[0]))
+                    except ValueError:
+                        continue
+    except (subprocess.SubprocessError, FileNotFoundError):
+        pass
+    return sorted(list(set(pids)))
+
+
+def _is_port_in_use(port: int) -> bool:
+    """Check if the given port is already occupied."""
+    import socket
+    with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as s:
+        s.settimeout(0.5)
+        return s.connect_ex(('127.0.0.1', port)) == 0
+
+
 def cmd_status() -> None:
-    pid = _read_pid()
-    if pid and _process_alive(pid):
-        print(f'Running  (PID {pid})')
+    pid_file_val = _read_pid()
+    all_pids = _find_all_pids()
+    
+    if pid_file_val and _process_alive(pid_file_val):
+        print(f'Running  (Master PID {pid_file_val})')
+        if len(all_pids) > 1:
+            print(f'Workers  ({len(all_pids) - 1} processes)')
+    elif all_pids:
+        print(f'Warning: Found {len(all_pids)} orphaned processes not tracked in {PID_FILE}:')
+        print(f'  PIDs: {", ".join(map(str, all_pids))}')
+        print('  Action: Use "python production.py kill" to clean these up.')
     else:
-        if pid:
-            os.remove(PID_FILE)
+        if pid_file_val:
+            try: os.remove(PID_FILE)
+            except OSError: pass
         print('Stopped')
 
 
 def cmd_start(host: str, port: int, workers: int) -> None:
     pid = _read_pid()
-    if pid and _process_alive(pid):
-        print(f'Already running (PID {pid}). Use "restart" to reload.')
+    all_pids = _find_all_pids()
+    
+    if (pid and _process_alive(pid)) or all_pids:
+        print(f'Already running (Master PID {pid or all_pids[0]}). Use "restart" to reload.')
         return
 
+    if _is_port_in_use(port):
+        sys.exit(f'Error: Port {port} is already in use by another program (possibly an old zombie instance).')
+
     gunicorn = _gunicorn_bin()
 
     log = open(LOG_FILE, 'a')
@@ -101,22 +144,36 @@ def cmd_start(host: str, port: int, workers: int) -> None:
     print(f'Logs: {LOG_FILE}')
 
 
-def cmd_stop(graceful: bool = True) -> bool:
+def cmd_stop(graceful: bool = True, force: bool = False) -> bool:
     pid = _read_pid()
+    all_pids = _find_all_pids()
+
     if not pid or not _process_alive(pid):
-        print('Not running.')
+        if all_pids:
+            if force:
+                print(f"Cleaning up {len(all_pids)} orphaned processes...")
+                return cmd_kill()
+            print(f'Error: PID file is missing, but {len(all_pids)} orphaned processes were found.')
+            print('Use "python production.py kill" to force stop.')
+        else:
+            print('Not running.')
         if pid:
-            os.remove(PID_FILE)
+            try: os.remove(PID_FILE)
+            except OSError: pass
         return False
 
+    # Standard shutdown
     sig = signal.SIGTERM if graceful else signal.SIGKILL
     os.kill(pid, sig)
 
     # Wait up to 10 s for a clean shutdown.
+    print(f'Stopping PID {pid}...', end='', flush=True)
     for _ in range(20):
         time.sleep(0.5)
+        print('.', end='', flush=True)
         if not _process_alive(pid):
             break
+    print(' Done.')
 
     if _process_alive(pid):
         # Force-kill if still alive after graceful period.
@@ -124,9 +181,39 @@ def cmd_stop(graceful: bool = True) -> bool:
         time.sleep(0.5)
 
     if os.path.exists(PID_FILE):
-        os.remove(PID_FILE)
+        try: os.remove(PID_FILE)
+        except OSError: pass
 
-    print(f'Stopped  (was PID {pid})')
+    # Double check for ANY remaining processes in this directory
+    remaining = _find_all_pids()
+    if remaining and force:
+        for p in remaining:
+            try: os.kill(p, signal.SIGKILL)
+            except OSError: pass
+    
+    return True
+
+
+def cmd_kill() -> bool:
+    """Nuke all Bastebin-related processes."""
+    pids = _find_all_pids()
+    if not pids:
+        print("Nothing to kill.")
+        if os.path.exists(PID_FILE):
+            os.remove(PID_FILE)
+        return False
+    
+    print(f"Killing {len(pids)} processes...")
+    for pid in pids:
+        try:
+            print(f"  - {pid}")
+            os.kill(pid, signal.SIGKILL)
+        except OSError:
+            pass
+    
+    if os.path.exists(PID_FILE):
+        os.remove(PID_FILE)
+    print("Clean slate achieved.")
     return True
 
 
@@ -159,9 +246,11 @@ def _build_parser() -> argparse.ArgumentParser:
 
     p_start   = sub.add_parser('start',   help='Start Gunicorn in the background')
     p_restart = sub.add_parser('restart', help='Gracefully reload (SIGHUP) or start if stopped')
-    sub.add_parser('stop',   help='Stop Gunicorn gracefully')
-    sub.add_parser('status', help='Show whether Gunicorn is running')
+    p_stop    = sub.add_parser('stop',    help='Stop Gunicorn gracefully')
+    sub.add_parser('kill',    help='Forcefully kill all Bastebin processes')
+    sub.add_parser('status',  help='Show whether Gunicorn is running')
 
+    p_stop.add_argument('--force', action='store_true', help='Kill all orphaned processes')
     _add_server_args(p_start)
     _add_server_args(p_restart)
 
@@ -209,7 +298,9 @@ def main() -> None:
     if args.command == 'start':
         cmd_start(host, port, workers)
     elif args.command == 'stop':
-        cmd_stop()
+        cmd_stop(force=args.force)
+    elif args.command == 'kill':
+        cmd_kill()
     elif args.command == 'restart':
         cmd_restart(host, port, workers)
     elif args.command == 'status':