diff --git a/litellm/proxy/common_utils/openai_error_payload.py b/litellm/proxy/common_utils/openai_error_payload.py index 2f01dbe11f7..7a60069659b 100644 --- a/litellm/proxy/common_utils/openai_error_payload.py +++ b/litellm/proxy/common_utils/openai_error_payload.py @@ -17,6 +17,7 @@ _OPENAI_ERROR_TYPE_BY_STATUS: Final[Mapping[int, str]] = MappingProxyType( { status.HTTP_401_UNAUTHORIZED: "authentication_error", status.HTTP_403_FORBIDDEN: "permission_error", + status.HTTP_408_REQUEST_TIMEOUT: "timeout_error", status.HTTP_429_TOO_MANY_REQUESTS: "rate_limit_error", } ) diff --git a/tests/integration/authorization/test_guardrail_submissions.py b/tests/integration/authorization/test_guardrail_submissions.py index f120028f46d..16b9cbf31dc 100644 --- a/tests/integration/authorization/test_guardrail_submissions.py +++ b/tests/integration/authorization/test_guardrail_submissions.py @@ -310,7 +310,7 @@ def test_non_admin_keys_scoped_to_the_submission_routes_get_403_from_the_admin_c f"Your user_id={_mask_user_id(user_id)}" ), type="auth_error", - param="None", + param=None, code="401", ) ) diff --git a/tests/integration/authorization/test_jwt_key_mapping_lifecycle.py b/tests/integration/authorization/test_jwt_key_mapping_lifecycle.py index 715f3731175..da830293145 100644 --- a/tests/integration/authorization/test_jwt_key_mapping_lifecycle.py +++ b/tests/integration/authorization/test_jwt_key_mapping_lifecycle.py @@ -114,7 +114,7 @@ class _ChatRequest(BaseModel): class _ProxyError(BaseModel): message: str type: str - param: str + param: str | None code: str @@ -398,7 +398,7 @@ def _assert_unmapped(response: httpx.Response, client_id: str) -> None: assert response.status_code == 403, response.text error: Final = _ErrorResponse.model_validate_json(response.content) expected: Final = f"JWT Key Mapping: No registered mapping for {_CLIENT_CLAIM}='{client_id}'. Access denied." - assert error.error == _ProxyError(message=expected, type="auth_error", param="None", code="403"), response.text + assert error.error == _ProxyError(message=expected, type="auth_error", param=None, code="403"), response.text def _mapping_info(gateway: Gateway, identity: str) -> _JWTMappingResponse: diff --git a/tests/integration/authorization/test_unknown_key_generic_401.py b/tests/integration/authorization/test_unknown_key_generic_401.py index 8e6ff3fa3ab..025f11d3973 100644 --- a/tests/integration/authorization/test_unknown_key_generic_401.py +++ b/tests/integration/authorization/test_unknown_key_generic_401.py @@ -40,7 +40,7 @@ class _ProxyError(BaseModel): message: str type: str - param: str + param: str | None code: str @@ -260,12 +260,12 @@ def _control_error(response: httpx.Response) -> _ProxyError: def test_requests_without_an_sk_key_keep_their_own_401(gateway: Gateway, peer: Gateway, side: Side) -> None: client: Final = _side(gateway, peer, side).client assert _control_error(client.get("/v1/models")) == _ProxyError( - message="Authentication Error, No api key passed in.", type="auth_error", param="None", code="401" + message="Authentication Error, No api key passed in.", type="auth_error", param=None, code="401" ) assert _control_error(client.get("/v1/models", headers={"Authorization": "Bearer not-a-real-key"})) == _ProxyError( message="LiteLLM Virtual Key expected. Received=not-****-key, expected to start with 'sk-'.", type="auth_error", - param="None", + param=None, code="401", ) @@ -277,7 +277,7 @@ def test_a_blocked_key_keeps_its_own_401_on_both_proxies(gateway: Gateway, peer: expected: Final = _ProxyError( message="Authentication Error, Key is blocked. Update via `/key/unblock` if you're an admin.", type="auth_error", - param="None", + param=None, code="401", ) for candidate in (gateway, peer): diff --git a/tests/unit/proxy/common_utils/test_openai_error_payload.py b/tests/unit/proxy/common_utils/test_openai_error_payload.py index 40bb84ff538..9b85a630347 100644 --- a/tests/unit/proxy/common_utils/test_openai_error_payload.py +++ b/tests/unit/proxy/common_utils/test_openai_error_payload.py @@ -20,7 +20,7 @@ from litellm.proxy.common_utils.openai_error_payload import ( (401, "authentication_error"), (403, "permission_error"), (404, "invalid_request_error"), - (408, "invalid_request_error"), + (408, "timeout_error"), (422, "invalid_request_error"), (429, "rate_limit_error"), (499, "invalid_request_error"), @@ -161,6 +161,16 @@ def test_an_upstream_5xx_body_does_not_relabel_the_internal_server_error(): assert openai_error_type(carried, error_status_code(carried, 400)) == "internal_server_error" +def test_an_upstream_timeout_is_not_typed_as_the_clients_mistake(): + """A provider timeout carries status 408 and no type, and calling it an + invalid_request_error tells a client branching on type not to retry a request that can succeed.""" + from litellm.exceptions import Timeout + + timeout = Timeout(message="Request timed out", model="gpt-5.4-mini", llm_provider="openai") + + assert openai_error_type(timeout, error_status_code(timeout, 500)) == "timeout_error" + + def test_a_stringified_none_type_or_param_is_treated_as_absent(): from litellm.exceptions import BadRequestError