fix: req changes from greptile

litellm/router.py

@@ -1501,20 +1501,21 @@ class Router:
             loop = asyncio.new_event_loop()
             asyncio.set_event_loop(loop)
             try:
-                loop.run_until_complete(
-                    self.acompletion(
+
+                async def _run_silent_completion():
+                    await self.acompletion(
                         model=silent_model,
                         messages=cast(List[AllMessageValues], messages),
                         **silent_kwargs,
                     )
-                )
-                # Drain any remaining fire-and-forget tasks (e.g. alerting hooks)
-                # scheduled via asyncio.create_task during the acompletion call.
-                pending = asyncio.all_tasks(loop)
-                if pending:
-                    loop.run_until_complete(
-                        asyncio.gather(*pending, return_exceptions=True)
-                    )
+                    # Drain any fire-and-forget tasks (e.g. alerting hooks)
+                    # scheduled via asyncio.create_task during acompletion.
+                    pending = asyncio.all_tasks()
+                    pending.discard(asyncio.current_task())
+                    if pending:
+                        await asyncio.gather(*pending, return_exceptions=True)
+
+                loop.run_until_complete(_run_silent_completion())
             finally:
                 loop.close()
         except Exception as e:

tests/test_litellm/test_router_silent_experiment.py

@@ -1,4 +1,5 @@
 import asyncio
+import time
 from unittest.mock import AsyncMock, MagicMock, patch
 
 import pytest
@@ -203,15 +204,10 @@ def test_router_silent_experiment_completion():
 
         assert response.choices[0].message.content == "hello"
 
-        # The sync background call uses a thread pool. We might need to wait a bit.
-        import time
-
+        # The sync background call uses a thread pool. We might need to wait.
         time.sleep(2.0)
 
         # Should have 1 acompletion call (the silent background call)
-        # The primary completion call still goes to the real litellm.completion (or we can mock it separately, but here it's testing the background one)
-        # Wait, the primary call in the test is router.completion.
-        # Actually, let's just mock both to avoid real network calls if it's hitting one.
         assert mock_acompletion_mock.call_count == 1
 
         call_args_list = mock_acompletion_mock.call_args_list