[Refactor] [1/N] to simplify the vLLM serving architecture (#28040)

Signed-off-by: chaunceyjiang <[email protected]>

Author: chaunceyjiang

tests/entrypoints/openai/test_basic.py

@@ -232,7 +232,7 @@ def make_long_completion_request():
 @pytest.mark.asyncio
 async def test_health_check_engine_dead_error():
     # Import the health function directly to test it in isolation
-    from vllm.entrypoints.openai.api_server import health
+    from vllm.entrypoints.serve.instrumentator.health import health
 
     # Create a mock request that simulates what FastAPI would provide
     mock_request = Mock(spec=Request)

vllm/entrypoints/api_server.py

@@ -118,6 +118,7 @@ async def init_app(
         )
     )
     app.state.engine_client = engine
+    app.state.args = args
     return app