[Frontend][V1] Online serving performance improvements (#12287)

vllm/entrypoints/openai/api_server.py

@@ -1,5 +1,6 @@
 import asyncio
 import atexit
+import gc
 import importlib
 import inspect
 import multiprocessing
@@ -104,6 +105,11 @@ async def lifespan(app: FastAPI):
             task.add_done_callback(_running_tasks.remove)
         else:
             task = None
+
+        # Mark the startup heap as static so that it's ignored by GC.
+        # Reduces pause times of oldest generation collections.
+        gc.collect()
+        gc.freeze()
         try:
             yield
         finally: