From 55d18c602d679658e61fa1f8bcf95b3de9e786a8 Mon Sep 17 00:00:00 2001 From: mateo-berri <277851410+mateo-berri@users.noreply.github.com> Date: Tue, 20 Oct 2026 14:51:06 -0700 Subject: [PATCH] fix(proxy): route Ask AI deployment names only to deployments the router still has --- .../usage_endpoints/ai_usage_chat.py | 3 ++- .../usage_endpoints/test_ai_usage_chat.py | 16 ++++++++++++++-- 2 files changed, 16 insertions(+), 3 deletions(-) diff --git a/litellm/proxy/management_endpoints/usage_endpoints/ai_usage_chat.py b/litellm/proxy/management_endpoints/usage_endpoints/ai_usage_chat.py index 1cd67ba4770..e2bcfd9380a 100644 --- a/litellm/proxy/management_endpoints/usage_endpoints/ai_usage_chat.py +++ b/litellm/proxy/management_endpoints/usage_endpoints/ai_usage_chat.py @@ -481,7 +481,8 @@ def _router_route(router: "Router", model: str, caller_is_proxy_admin: bool) -> return None if router.default_deployment is not None or router.pattern_router.route(model) is not None: # pyright: ignore[reportUnknownMemberType] # PatternMatchDeployments.route returns untyped dicts return "model_group" - return "deployment" if model in router.deployment_names else None + serves_model: Final = any(deployment["litellm_params"]["model"] == model for deployment in router.model_list) # pyright: ignore[reportUnknownVariableType] # Router.model_list is an untyped list + return "deployment" if serves_model else None class _TaggedMetadata(BaseModel): diff --git a/tests/unit/proxy/management_endpoints/usage_endpoints/test_ai_usage_chat.py b/tests/unit/proxy/management_endpoints/usage_endpoints/test_ai_usage_chat.py index f541875eca9..0af9f403415 100644 --- a/tests/unit/proxy/management_endpoints/usage_endpoints/test_ai_usage_chat.py +++ b/tests/unit/proxy/management_endpoints/usage_endpoints/test_ai_usage_chat.py @@ -929,7 +929,7 @@ class TestUsageAiChatProviderRequest: def _tool_names(sent: dict[str, object]) -> list[object]: return [tool.get("name") or tool["function"]["name"] for tool in sent["tools"]] - async def _ask(self, model: str, monkeypatch: pytest.MonkeyPatch) -> None: + async def _ask(self, model: str, monkeypatch: pytest.MonkeyPatch, router: "Router | None" = None) -> None: import litellm from litellm.proxy._types import LitellmUserRoles, UserAPIKeyAuth from litellm.proxy.management_endpoints.usage_endpoints.ai_usage_chat import ( @@ -939,7 +939,7 @@ class TestUsageAiChatProviderRequest: monkeypatch.setattr(litellm, "disable_aiohttp_transport", True) monkeypatch.setenv("OPENAI_API_KEY", "sk-env") with patch( # test-quality-ok: _acompletion reads the proxy_server.llm_router module global; no injection seam - "litellm.proxy.proxy_server.llm_router", self._router() + "litellm.proxy.proxy_server.llm_router", router or self._router() ): await _acompletion( model=model, @@ -993,6 +993,18 @@ class TestUsageAiChatProviderRequest: assert respx_mock.calls.last.request.headers["Authorization"] == "Bearer sk-gpt5-deployment" + @pytest.mark.asyncio + async def test_deleted_deployment_name_is_called_with_the_environment_key( + self, respx_mock: respx.MockRouter, monkeypatch: pytest.MonkeyPatch + ) -> None: + self._mock_openai(respx_mock) + router: Final = self._router() + router.delete_deployment(id=router.get_model_ids(model_name="gpt-5-prod")[0]) + + await self._ask("openai/gpt-5.6", monkeypatch, router) + + assert respx_mock.calls.last.request.headers["Authorization"] == "Bearer sk-env" + class TestUsageAiChatKeepalive: async def _collect_endpoint_body(self, monkeypatch, interval, delay=0.3) -> tuple[list[bytes], dict]: