fix(proxy): mirror chat completions routing rule in usage ai chat model gate

Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
This commit is contained in:
Devin AI 2026-09-18 22:07:49 +00:00
parent 6a916c2f57
commit 2bd80e89ba
2 changed files with 45 additions and 5 deletions

View file

@ -447,7 +447,11 @@ def _sse(event: SSEEvent) -> str:
def _router_knows_model(router: "Router", model: str) -> bool:
return model in router.get_model_names() or router.pattern_router.route(request=model) is not None # pyright: ignore[reportUnknownMemberType] # PatternMatchDeployments.route returns untyped dicts
return (
model in router.get_model_names()
or router.default_deployment is not None
or len(router.pattern_router.patterns) > 0 # pyright: ignore[reportUnknownMemberType, reportUnknownArgumentType] # PatternMatchRouter.patterns values are bare `list`
)
async def _acompletion(

View file

@ -544,7 +544,8 @@ class TestUsageAiChatCompletionRouting:
async def test_stream_usage_ai_chat_falls_back_to_litellm_for_unknown_model(self):
mock_router = MagicMock()
mock_router.get_model_names.return_value = ["other"]
mock_router.pattern_router.route.return_value = None
mock_router.default_deployment = None
mock_router.pattern_router.patterns = {}
mock_router.acompletion = AsyncMock(return_value=self._plain_content_response())
mock_acompletion = AsyncMock(return_value=self._plain_content_response())
@ -574,10 +575,46 @@ class TestUsageAiChatCompletionRouting:
assert "error" not in event_types
@pytest.mark.asyncio
async def test_stream_usage_ai_chat_uses_router_for_wildcard_matched_model(self):
async def test_stream_usage_ai_chat_uses_router_when_wildcard_pattern_exists(self):
mock_router = MagicMock()
mock_router.get_model_names.return_value = []
mock_router.pattern_router.route.return_value = [{"model_name": "openai/*"}]
mock_router.default_deployment = None
mock_router.pattern_router.patterns = {"openai/.*": [{"model_name": "openai/*"}]}
mock_router.acompletion = AsyncMock(return_value=self._plain_content_response())
mock_acompletion = AsyncMock(return_value=self._plain_content_response())
with (
patch( # test-quality-ok: ai_usage_chat reads the proxy_server.llm_router module global; no injection seam
"litellm.proxy.proxy_server.llm_router", mock_router
),
patch( # test-quality-ok: the stream calls the module-level litellm.acompletion directly; no injection seam
"litellm.proxy.management_endpoints.usage_endpoints.ai_usage_chat.litellm.acompletion",
new=mock_acompletion,
),
):
events = []
async for event in stream_usage_ai_chat(
messages=[{"role": "user", "content": "hi"}],
model="gpt-5.6-terra",
is_admin=True,
):
events.append(json.loads(event.replace("data: ", "").strip()))
mock_router.acompletion.assert_awaited_once()
assert mock_router.acompletion.await_args.kwargs["model"] == "gpt-5.6-terra"
assert mock_router.acompletion.await_args.kwargs["drop_params"] is True
mock_acompletion.assert_not_called()
event_types = [e["type"] for e in events]
assert "chunk" in event_types
assert "error" not in event_types
@pytest.mark.asyncio
async def test_stream_usage_ai_chat_uses_router_when_default_deployment_exists(self):
mock_router = MagicMock()
mock_router.get_model_names.return_value = []
mock_router.default_deployment = {"model_name": "default", "litellm_params": {}}
mock_router.pattern_router.patterns = {}
mock_router.acompletion = AsyncMock(return_value=self._plain_content_response())
mock_acompletion = AsyncMock(return_value=self._plain_content_response())
@ -598,7 +635,6 @@ class TestUsageAiChatCompletionRouting:
):
events.append(json.loads(event.replace("data: ", "").strip()))
mock_router.pattern_router.route.assert_called_once_with(request="gpt-5.6-terra")
mock_router.acompletion.assert_awaited_once()
assert mock_router.acompletion.await_args.kwargs["model"] == "gpt-5.6-terra"
assert mock_router.acompletion.await_args.kwargs["drop_params"] is True