✅ add test for multiprocessing flag

tests/entrypoints/openai/test_basic.py

@@ -1,9 +1,11 @@
+import socket
 from http import HTTPStatus
 
 import openai
 import pytest
 import requests
 
+from vllm import envs
 from vllm.version import __version__ as VLLM_VERSION
 
 from ...utils import RemoteOpenAIServer
@@ -59,3 +61,20 @@ async def test_log_metrics(client: openai.AsyncOpenAI):
     response = requests.get(base_url + "/metrics")
 
     assert response.status_code == HTTPStatus.OK
+
+
+@pytest.mark.asyncio
+async def test_fronted_multiprocessing_flag():
+    # Build server without the flag to disable multiprocessing
+    with RemoteOpenAIServer("facebook/opt-125m", []), \
+        socket.socket(socket.AF_INET, socket.SOCK_STREAM) as s, \
+        pytest.raises(OSError, match="Address already in use"):
+        # Ensure we see the backend port in use
+        s.bind(("localhost", envs.VLLM_RPC_PORT))
+
+    # Build server with the flag to disable multiprocessing
+    with RemoteOpenAIServer("facebook/opt-125m",
+                            ["--disable-frontend-multiprocessing"]), \
+        socket.socket(socket.AF_INET, socket.SOCK_STREAM) as s:
+        # Ensure the backend port is free -> no multiprocessing is happening
+        s.bind(("localhost", envs.VLLM_RPC_PORT))