[BugFix] Prevent the task of _force_log from being garbage collected (#4567)

Author: Atry

vllm/entrypoints/openai/api_server.py

@@ -4,6 +4,7 @@
 import re
 from contextlib import asynccontextmanager
 from http import HTTPStatus
+from typing import Any, Set
 
 import fastapi
 import uvicorn
@@ -33,6 +34,8 @@
 openai_serving_completion: OpenAIServingCompletion
 logger = init_logger(__name__)
 
+_running_tasks: Set[asyncio.Task[Any]] = set()
+
 
 @asynccontextmanager
 async def lifespan(app: fastapi.FastAPI):
@@ -43,7 +46,9 @@ async def _force_log():
             await engine.do_log_stats()
 
     if not engine_args.disable_log_stats:
-        asyncio.create_task(_force_log())
+        task = asyncio.create_task(_force_log())
+        _running_tasks.add(task)
+        task.add_done_callback(_running_tasks.remove)
 
     yield