Ensure metrics are logged regardless of requests

vllm/engine/async_llm_engine.py

@@ -506,3 +506,9 @@ def from_engine_args(cls,
                      max_log_len=engine_args.max_log_len,
                      start_engine_loop=start_engine_loop)
         return engine
+
+    async def do_log_stats(self) -> None:
+        if self.engine_use_ray:
+            await self.engine.do_log_stats.remote()
+        else:
+            self.engine.do_log_stats()

vllm/engine/llm_engine.py

@@ -641,6 +641,9 @@ def step(self) -> List[RequestOutput]:
 
         return self._process_model_outputs(output, scheduler_outputs)
 
+    def do_log_stats(self) -> None:
+        self._log_system_stats(False, 0)
+
     def _log_system_stats(
         self,
         prompt_run: bool,

vllm/entrypoints/openai/api_server.py

@@ -6,6 +6,7 @@
 import codecs
 import json
 import time
+from contextlib import asynccontextmanager
 from http import HTTPStatus
 from typing import AsyncGenerator, Dict, List, Optional, Tuple, Union
 
@@ -38,11 +39,28 @@
 
 logger = init_logger(__name__)
 served_model = None
-app = fastapi.FastAPI()
+engine_args = None
 engine = None
 response_role = None
 
 
+@asynccontextmanager
+async def lifespan(app: fastapi.FastAPI):
+
+    async def _force_log():
+        while True:
+            await asyncio.sleep(10)
+            await engine.do_log_stats()
+
+    if not engine_args.disable_log_stats:
+        asyncio.create_task(_force_log())
+
+    yield
+
+
+app = fastapi.FastAPI(lifespan=lifespan)
+
+
 def parse_args():
     parser = argparse.ArgumentParser(
         description="vLLM OpenAI-Compatible RESTful API server.")