diff --git a/litellm/proxy/health_endpoints/_health_endpoints.py b/litellm/proxy/health_endpoints/_health_endpoints.py index 333687b07cd..86e454d903b 100644 --- a/litellm/proxy/health_endpoints/_health_endpoints.py +++ b/litellm/proxy/health_endpoints/_health_endpoints.py @@ -102,7 +102,7 @@ from litellm.router_api import deployment_usable_by_team from litellm.secret_managers.main import get_secret_bool from litellm.types.proxy.litellm_pre_call_utils import RedactedDict, SecretFields from litellm.types.router import reject_server_owned_wif_fields -from litellm.types.utils import AdminApiKeySecretReference, admin_api_key_secret_reference +from litellm.types.utils import AdminApiKeySecretReference, CustomPricingLiteLLMParams, admin_api_key_secret_reference #### Health ENDPOINTS #### @@ -155,6 +155,15 @@ _CONFIG_CONNECTION_FIELDS: Final[frozenset[str]] = frozenset( ) ) +# The banned request-body params that actually describe a CONNECTION — the +# banned list minus the custom-pricing fields. Pricing fields are banned from a +# request body because they poison the shared model-cost registry, not because +# they retarget or re-authenticate the outbound call, so the full list would +# treat `input_cost_per_token` as a credential. +_CONNECTION_OVERRIDE_REQUEST_PARAMS: Final[tuple[str, ...]] = tuple( + param for param in _BANNED_REQUEST_BODY_PARAMS if param not in CustomPricingLiteLLMParams.model_fields +) + def _request_inherits_config_credentials( config_params: Mapping[str, object], @@ -173,6 +182,14 @@ def _request_inherits_config_credentials( model name. A blank name or key is no name or key: ``load_credentials_from_list`` resolves nothing from it, so it must not cost the request the credentials it would otherwise be probed with. + + The trigger is ``_CONNECTION_OVERRIDE_REQUEST_PARAMS``, not the full banned + list: the custom-pricing fields are banned from a request body for a + different reason (they poison the shared model-cost registry) and say + nothing about which connection a test describes. Treating them as a + connection override empties the configuration under a request that only + named a model and its price, which reports a healthy deployment as + "Missing credentials". """ requested_credential: Final = request_params.get("litellm_credential_name") if requested_credential and requested_credential != config_params.get("litellm_credential_name"): @@ -181,7 +198,7 @@ def _request_inherits_config_credentials( return False if allow_client_side_credentials: return True - return not any(param in request_params for param in _BANNED_REQUEST_BODY_PARAMS) + return not any(param in request_params for param in _CONNECTION_OVERRIDE_REQUEST_PARAMS) def _admin_api_key_from_request( @@ -2714,12 +2731,20 @@ async def test_model_connection( or _string_mode_or_bad_request(raw_params_mode) or resolve_health_check_mode(probe_model_info, _OBJECT_MAPPING.validate_python(litellm_params)) ) + # A connection test needs no prices, and ``completion`` registers any + # request pricing under the shared ``{provider}/{model}`` cost-map key + # (a probe carries no router deployment id), which would re-price every + # sibling deployment of the same backend model. Probe without them, + # whether they came from the request or from the configuration. + probe_params: Final = CustomPricingLiteLLMParams.strip_custom_pricing_fields( + dict(_OBJECT_MAPPING.validate_python(litellm_params)) + ) raw_headers: Final[dict[str, str]] = TypeAdapter(dict[str, str]).validate_python( http_parsing_utils.safe_get_request_headers(request) ) health_check_params: Final = { - **_OBJECT_MAPPING.validate_python(litellm_params), + **probe_params, **_admin_api_key_from_request( _OBJECT_MAPPING.validate_python(request_litellm_params), _OBJECT_MAPPING.validate_python(litellm_params), diff --git a/tests/unit/proxy/health_endpoints/test_health_endpoints.py b/tests/unit/proxy/health_endpoints/test_health_endpoints.py index 2258eb95fb8..20ddd6231fd 100644 --- a/tests/unit/proxy/health_endpoints/test_health_endpoints.py +++ b/tests/unit/proxy/health_endpoints/test_health_endpoints.py @@ -5320,6 +5320,51 @@ class TestConfigBaseForHealthCheck: assert "litellm_credential_name" not in base assert "api_key" not in base + @pytest.mark.parametrize( + "pricing_field,value", + [ + ("input_cost_per_token", 1e-9), + ("output_cost_per_token", 2e-9), + ("cache_read_input_token_cost", 5e-10), + ], + ) + def test_pricing_field_is_not_a_connection_override(self, pricing_field, value): + """Pricing fields are banned from a request body because they poison the + shared model-cost registry, not because they describe a connection. A + connection test that carries one still gets the configured credentials — + otherwise a healthy deployment reports "Missing credentials".""" + base = self._base(self.CONFIG, {"model": "openai/gpt-4o", pricing_field: value}, selected_by_id=False) + assert base["api_key"] == "sk-configured" + assert base["api_base"] == "https://configured.example/v1" + + def test_pricing_field_alongside_a_real_override_still_drops_credentials(self): + """The pricing field is neutral, so the api_base beside it still decides.""" + base = self._base( + self.CONFIG, + {"api_base": "https://caller.example/v1", "input_cost_per_token": 1e-9}, + selected_by_id=False, + ) + assert "api_key" not in base + assert "sk-configured" not in str(base) + + def test_every_custom_pricing_field_is_excluded_from_the_connection_list(self): + from litellm.proxy.health_endpoints._health_endpoints import _CONNECTION_OVERRIDE_REQUEST_PARAMS + from litellm.types.utils import CustomPricingLiteLLMParams + + connection_params = set(_CONNECTION_OVERRIDE_REQUEST_PARAMS) + for field in CustomPricingLiteLLMParams.model_fields: + assert field not in connection_params, ( + f"CustomPricingLiteLLMParams.{field} is treated as a connection override, " + "so a connection test that sets it loses the configured credentials." + ) + + def test_connection_list_keeps_the_real_credential_and_endpoint_fields(self): + from litellm.proxy.health_endpoints._health_endpoints import _CONNECTION_OVERRIDE_REQUEST_PARAMS + + connection_params = set(_CONNECTION_OVERRIDE_REQUEST_PARAMS) + for field in ("api_base", "base_url", "azure_ad_token", "vertex_credentials", "user_config"): + assert field in connection_params + def test_stored_credential_reference_kept_when_request_sets_no_connection(self): """The Admin UI tests a configured model by naming it plus its stored credential and nothing else; that keeps working.""" @@ -5761,6 +5806,78 @@ class TestTestConnectionUsesTheNamedCredential: assert probe.headers["authorization"] == f"Bearer {self.CREDENTIAL_KEY}" +class TestTestConnectionLeavesSharedPricingAlone: + """A connection test must not write prices into ``litellm.model_cost``: a probe + carries no router deployment id, so ``completion`` would register them under + the shared ``{provider}/{model}`` key that sibling deployments price from.""" + + MODEL = "openai/test-connection-pricing-probe" + SHARED_KEY = MODEL + CONFIG_KEY = "sk-configured" + COMPLETION = TestTestConnectionUsesTheNamedCredential.COMPLETION + + def _probe(self, monkeypatch, config_litellm_params: dict, request_litellm_params: dict) -> httpx.Request: + monkeypatch.setattr(litellm, "disable_aiohttp_transport", True) + monkeypatch.setattr(litellm, "model_cost", dict(litellm.model_cost)) + litellm.in_memory_llm_clients_cache.flush_cache() + + app = FastAPI() + app.include_router(_health_endpoints_module.router) + app.dependency_overrides[user_api_key_auth] = lambda: UserAPIKeyAuth(user_role=LitellmUserRoles.PROXY_ADMIN) + + router = MagicMock() + router.get_deployment.return_value = None + router.get_model_list.return_value = [ + {"model_name": "pricing-probe", "litellm_params": config_litellm_params, "model_info": {}} + ] + + with ( + patch( # test-quality-ok: the endpoint reads the proxy-global DB client and 500s when it is None; it has no injection seam + "litellm.proxy.proxy_server.prisma_client", MagicMock() + ), + patch( # test-quality-ok: the deployment the probe is matched against is a proxy global; it has no injection seam + "litellm.proxy.proxy_server.llm_router", router + ), + respx.mock(assert_all_called=True) as respx_mock, + ): + respx_mock.post(path__regex=r".*/chat/completions").respond(json=self.COMPLETION) + response = TestClient(app).post( + "/health/test_connection", + json={"mode": "chat", "litellm_params": request_litellm_params, "model_info": {"mode": "chat"}}, + ) + probe = respx_mock.calls.last.request + + assert response.status_code == 200, response.text + assert response.json()["status"] == "success", response.text + return probe + + def test_request_pricing_is_not_registered_under_the_shared_key(self, monkeypatch): + probe = self._probe( + monkeypatch, + {"model": self.MODEL, "api_key": self.CONFIG_KEY, "api_base": "https://configured.example/v1"}, + {"model": self.MODEL, "input_cost_per_token": 1.0, "output_cost_per_token": 2.0}, + ) + + assert probe.headers["authorization"] == f"Bearer {self.CONFIG_KEY}" + assert self.SHARED_KEY not in litellm.model_cost + + def test_configured_pricing_is_not_registered_under_the_shared_key(self, monkeypatch): + probe = self._probe( + monkeypatch, + { + "model": self.MODEL, + "api_key": self.CONFIG_KEY, + "api_base": "https://configured.example/v1", + "input_cost_per_token": 3.0, + "output_cost_per_token": 4.0, + }, + {"model": self.MODEL}, + ) + + assert probe.url.host == "configured.example" + assert self.SHARED_KEY not in litellm.model_cost + + class TestNoRedisWarning: """`show_no_redis_warning` drives the Admin UI's default-on "no Redis" banner."""