[Core] cleanup zmq ipc sockets on exit (#11115)

Signed-off-by: Russell Bryant <rbryant@redhat.com>

vllm/entrypoints/openai/api_server.py

@@ -1,4 +1,5 @@
 import asyncio
+import atexit
 import importlib
 import inspect
 import multiprocessing
@@ -196,6 +197,14 @@ async def build_async_engine_client_from_engine_args(
         assert engine_pid is not None, "Engine process failed to start."
         logger.info("Started engine process with PID %d", engine_pid)
 
+        def _cleanup_ipc_path():
+            socket_path = ipc_path.replace("ipc://", "")
+            if os.path.exists(socket_path):
+                os.remove(socket_path)
+
+        # Ensure we clean up the local IPC socket file on exit.
+        atexit.register(_cleanup_ipc_path)
+
         # Build RPCClient, which conforms to EngineClient Protocol.
         engine_config = engine_args.create_engine_config()
         build_client = partial(MQLLMEngineClient, ipc_path, engine_config,