From 32074cf4de50e6c5276e1973bfc7660b2b03182e Mon Sep 17 00:00:00 2001 From: mateo-berri <277851410+mateo-berri@users.noreply.github.com> Date: Thu, 3 Sep 2026 03:04:03 -0700 Subject: [PATCH 1/4] fix(proxy): stop answering with the literal string "None" on the proxy-server, auth, and health routes The error objects these routes return defaulted `type` and `param` to the four-character string "None", so a client's error handler matched no known OpenAI type and fell into its generic branch. Route them through the shared openai_error_payload helpers instead, so `type` comes from the carried type or the HTTP status and `param` serializes as JSON null. --- litellm/proxy/auth/auth_exception_handler.py | 7 +- litellm/proxy/auth/handle_jwt.py | 2 +- .../health_endpoints/_health_endpoints.py | 5 +- litellm/proxy/proxy_server.py | 165 +++++++++--------- .../proxy/auth/test_auth_exception_handler.py | 34 +++- .../proxy/auth/test_handle_jwt.py | 12 ++ .../health_endpoints/test_health_endpoints.py | 12 ++ tests/test_litellm/proxy/test_proxy_server.py | 11 ++ 8 files changed, 161 insertions(+), 87 deletions(-) diff --git a/litellm/proxy/auth/auth_exception_handler.py b/litellm/proxy/auth/auth_exception_handler.py index b36c8a038fc..3cfd8e047de 100644 --- a/litellm/proxy/auth/auth_exception_handler.py +++ b/litellm/proxy/auth/auth_exception_handler.py @@ -24,6 +24,7 @@ from litellm.proxy.auth.auth_utils import ( is_invalid_virtual_key_error, mark_invalid_virtual_key_error, ) +from litellm.proxy.common_utils.openai_error_payload import openai_error_param from litellm.proxy.db.exception_handler import PrismaDBExceptionHandler from litellm.types.services import ServiceTypes @@ -54,7 +55,7 @@ def _as_proxy_exception(e: Exception) -> ProxyException: return ProxyException( message=getattr(e, "detail", f"Authentication Error({e})"), type=ProxyErrorTypes.auth_error, - param=getattr(e, "param", "None"), + param=openai_error_param(e), code=getattr(e, "status_code", status.HTTP_401_UNAUTHORIZED), ) if isinstance(e, ProxyException): @@ -63,13 +64,13 @@ def _as_proxy_exception(e: Exception) -> ProxyException: return ProxyException( message=PrismaDBExceptionHandler.database_unavailable_message(e), type=ProxyErrorTypes.no_db_connection, - param="None", + param=None, code=status.HTTP_503_SERVICE_UNAVAILABLE, ) return ProxyException( message="Authentication Error, " + str(e), type=ProxyErrorTypes.auth_error, - param=getattr(e, "param", "None"), + param=openai_error_param(e), code=status.HTTP_401_UNAUTHORIZED, ) diff --git a/litellm/proxy/auth/handle_jwt.py b/litellm/proxy/auth/handle_jwt.py index 0795cee7409..0eaa816b223 100644 --- a/litellm/proxy/auth/handle_jwt.py +++ b/litellm/proxy/auth/handle_jwt.py @@ -142,7 +142,7 @@ def jwks_unavailable_exception(error: JWKSUnreachableError) -> ProxyException: f"unreachable, so the JWT signature could not be verified. Please retry shortly. Error: {error}" ), type=ProxyErrorTypes.auth_provider_unavailable, - param="None", + param=None, code=status.HTTP_503_SERVICE_UNAVAILABLE, ) diff --git a/litellm/proxy/health_endpoints/_health_endpoints.py b/litellm/proxy/health_endpoints/_health_endpoints.py index 8b57bdca2fe..06e6f281048 100644 --- a/litellm/proxy/health_endpoints/_health_endpoints.py +++ b/litellm/proxy/health_endpoints/_health_endpoints.py @@ -40,6 +40,7 @@ from litellm.proxy.auth.auth_utils import ( _BANNED_REQUEST_BODY_PARAMS, # pyright: ignore[reportPrivateUsage] # one canonical list, shared with the request-body check ) from litellm.proxy.auth.user_api_key_auth import user_api_key_auth +from litellm.proxy.common_utils.openai_error_payload import openai_error_param from litellm.proxy.db.exception_handler import PrismaDBExceptionHandler from litellm.proxy.db.proxy_worker_heartbeat import count_live_proxy_workers from litellm.proxy.health_check import ( @@ -529,7 +530,7 @@ async def health_services_endpoint( raise ProxyException( message=getattr(e, "detail", f"Authentication Error({e})"), type=ProxyErrorTypes.auth_error, - param=getattr(e, "param", "None"), + param=openai_error_param(e), code=getattr(e, "status_code", status.HTTP_500_INTERNAL_SERVER_ERROR), ) elif isinstance(e, ProxyException): @@ -537,7 +538,7 @@ async def health_services_endpoint( raise ProxyException( message="Authentication Error, " + str(e), type=ProxyErrorTypes.auth_error, - param=getattr(e, "param", "None"), + param=openai_error_param(e), code=status.HTTP_500_INTERNAL_SERVER_ERROR, ) diff --git a/litellm/proxy/proxy_server.py b/litellm/proxy/proxy_server.py index 27132c90e05..c4942e5df1b 100644 --- a/litellm/proxy/proxy_server.py +++ b/litellm/proxy/proxy_server.py @@ -112,6 +112,11 @@ from litellm.proxy.common_utils.callback_utils import ( process_callback, strip_callback_config, ) +from litellm.proxy.common_utils.openai_error_payload import ( + error_status_code, + openai_error_param, + openai_error_type, +) from litellm.proxy.common_utils.realtime_utils import _realtime_request_body from litellm.router_utils.add_retry_fallback_headers import ( get_fallback_errors_from_headers, @@ -8160,9 +8165,9 @@ async def async_assistants_data_generator(response, user_api_key_dict: UserAPIKe proxy_exception: Final = ProxyException( message=getattr(e, "message", error_msg), - type=getattr(e, "type", "None"), - param=getattr(e, "param", "None"), - code=getattr(e, "status_code", 500), + type=openai_error_type(e, error_status_code(e, 500)), + param=openai_error_param(e), + code=error_status_code(e, 500), ) error_returned: Final = json.dumps({"error": proxy_exception.to_dict()}) yield f"data: {error_returned}\n\n" @@ -8867,9 +8872,9 @@ async def async_data_generator( proxy_exception: Final = ProxyException( message=getattr(e, "message", error_msg), - type=getattr(e, "type", "None"), - param=getattr(e, "param", "None"), - code=getattr(e, "status_code", 500), + type=openai_error_type(e, error_status_code(e, 500)), + param=openai_error_param(e), + code=error_status_code(e, 500), ) error_returned: Final = json.dumps({"error": proxy_exception.to_dict()}) stream_completed = True @@ -10819,10 +10824,10 @@ async def completion( error_msg: Final = f"{e}" raise ProxyException( message=getattr(e, "message", error_msg), - type=getattr(e, "type", "None"), - param=getattr(e, "param", "None"), + type=openai_error_type(e, error_status_code(e, 500)), + param=openai_error_param(e), openai_code=getattr(e, "code", None), - code=getattr(e, "status_code", 500), + code=error_status_code(e, 500), ) @@ -11052,17 +11057,17 @@ async def moderations( if isinstance(e, HTTPException): raise ProxyException( message=getattr(e, "message", str(e)), - type=getattr(e, "type", "None"), - param=getattr(e, "param", "None"), - code=getattr(e, "status_code", status.HTTP_400_BAD_REQUEST), + type=openai_error_type(e, error_status_code(e, status.HTTP_400_BAD_REQUEST)), + param=openai_error_param(e), + code=error_status_code(e, status.HTTP_400_BAD_REQUEST), ) else: error_msg: Final = f"{e}" raise ProxyException( message=getattr(e, "message", error_msg), - type=getattr(e, "type", "None"), - param=getattr(e, "param", "None"), - code=getattr(e, "status_code", 500), + type=openai_error_type(e, error_status_code(e, 500)), + param=openai_error_param(e), + code=error_status_code(e, 500), ) @@ -11199,10 +11204,10 @@ async def audio_speech( raise e raise ProxyException( message=getattr(e, "message", f"{e}"), - type=getattr(e, "type", "None"), - param=getattr(e, "param", "None"), + type=openai_error_type(e, error_status_code(e, 500)), + param=openai_error_param(e), openai_code=getattr(e, "code", None), - code=getattr(e, "status_code", 500), + code=error_status_code(e, 500), ) @@ -11347,18 +11352,18 @@ async def audio_transcriptions( if isinstance(e, HTTPException): raise ProxyException( message=getattr(e, "message", str(e.detail)), - type=getattr(e, "type", "None"), - param=getattr(e, "param", "None"), - code=getattr(e, "status_code", status.HTTP_400_BAD_REQUEST), + type=openai_error_type(e, error_status_code(e, status.HTTP_400_BAD_REQUEST)), + param=openai_error_param(e), + code=error_status_code(e, status.HTTP_400_BAD_REQUEST), ) else: error_msg: Final = f"{e}" raise ProxyException( message=getattr(e, "message", error_msg), - type=getattr(e, "type", "None"), - param=getattr(e, "param", "None"), + type=openai_error_type(e, error_status_code(e, 500)), + param=openai_error_param(e), openai_code=getattr(e, "code", None), - code=getattr(e, "status_code", 500), + code=error_status_code(e, 500), ) @@ -11645,18 +11650,18 @@ async def get_assistants( if isinstance(e, HTTPException): raise ProxyException( message=getattr(e, "message", str(e.detail)), - type=getattr(e, "type", "None"), - param=getattr(e, "param", "None"), - code=getattr(e, "status_code", status.HTTP_400_BAD_REQUEST), + type=openai_error_type(e, error_status_code(e, status.HTTP_400_BAD_REQUEST)), + param=openai_error_param(e), + code=error_status_code(e, status.HTTP_400_BAD_REQUEST), ) else: error_msg: Final = f"{e}" raise ProxyException( message=getattr(e, "message", error_msg), - type=getattr(e, "type", "None"), - param=getattr(e, "param", "None"), + type=openai_error_type(e, error_status_code(e, 500)), + param=openai_error_param(e), openai_code=getattr(e, "code", None), - code=getattr(e, "status_code", 500), + code=error_status_code(e, 500), ) @@ -11736,16 +11741,16 @@ async def create_assistant( if isinstance(e, HTTPException): raise ProxyException( message=getattr(e, "message", str(e.detail)), - type=getattr(e, "type", "None"), - param=getattr(e, "param", "None"), - code=getattr(e, "status_code", status.HTTP_400_BAD_REQUEST), + type=openai_error_type(e, error_status_code(e, status.HTTP_400_BAD_REQUEST)), + param=openai_error_param(e), + code=error_status_code(e, status.HTTP_400_BAD_REQUEST), ) else: error_msg: Final = f"{e}" raise ProxyException( message=getattr(e, "message", error_msg), - type=getattr(e, "type", "None"), - param=getattr(e, "param", "None"), + type=openai_error_type(e, error_status_code(e, 500)), + param=openai_error_param(e), code=getattr(e, "code", getattr(e, "status_code", 500)), ) @@ -11825,16 +11830,16 @@ async def delete_assistant( if isinstance(e, HTTPException): raise ProxyException( message=getattr(e, "message", str(e.detail)), - type=getattr(e, "type", "None"), - param=getattr(e, "param", "None"), - code=getattr(e, "status_code", status.HTTP_400_BAD_REQUEST), + type=openai_error_type(e, error_status_code(e, status.HTTP_400_BAD_REQUEST)), + param=openai_error_param(e), + code=error_status_code(e, status.HTTP_400_BAD_REQUEST), ) else: error_msg: Final = f"{e}" raise ProxyException( message=getattr(e, "message", error_msg), - type=getattr(e, "type", "None"), - param=getattr(e, "param", "None"), + type=openai_error_type(e, error_status_code(e, 500)), + param=openai_error_param(e), code=getattr(e, "code", getattr(e, "status_code", 500)), ) @@ -11914,16 +11919,16 @@ async def create_threads( if isinstance(e, HTTPException): raise ProxyException( message=getattr(e, "message", str(e.detail)), - type=getattr(e, "type", "None"), - param=getattr(e, "param", "None"), - code=getattr(e, "status_code", status.HTTP_400_BAD_REQUEST), + type=openai_error_type(e, error_status_code(e, status.HTTP_400_BAD_REQUEST)), + param=openai_error_param(e), + code=error_status_code(e, status.HTTP_400_BAD_REQUEST), ) else: error_msg: Final = f"{e}" raise ProxyException( message=getattr(e, "message", error_msg), - type=getattr(e, "type", "None"), - param=getattr(e, "param", "None"), + type=openai_error_type(e, error_status_code(e, 500)), + param=openai_error_param(e), code=getattr(e, "code", getattr(e, "status_code", 500)), ) @@ -12001,16 +12006,16 @@ async def get_thread( if isinstance(e, HTTPException): raise ProxyException( message=getattr(e, "message", str(e.detail)), - type=getattr(e, "type", "None"), - param=getattr(e, "param", "None"), - code=getattr(e, "status_code", status.HTTP_400_BAD_REQUEST), + type=openai_error_type(e, error_status_code(e, status.HTTP_400_BAD_REQUEST)), + param=openai_error_param(e), + code=error_status_code(e, status.HTTP_400_BAD_REQUEST), ) else: error_msg: Final = f"{e}" raise ProxyException( message=getattr(e, "message", error_msg), - type=getattr(e, "type", "None"), - param=getattr(e, "param", "None"), + type=openai_error_type(e, error_status_code(e, 500)), + param=openai_error_param(e), code=getattr(e, "code", getattr(e, "status_code", 500)), ) @@ -12092,16 +12097,16 @@ async def add_messages( if isinstance(e, HTTPException): raise ProxyException( message=getattr(e, "message", str(e.detail)), - type=getattr(e, "type", "None"), - param=getattr(e, "param", "None"), - code=getattr(e, "status_code", status.HTTP_400_BAD_REQUEST), + type=openai_error_type(e, error_status_code(e, status.HTTP_400_BAD_REQUEST)), + param=openai_error_param(e), + code=error_status_code(e, status.HTTP_400_BAD_REQUEST), ) else: error_msg: Final = f"{e}" raise ProxyException( message=getattr(e, "message", error_msg), - type=getattr(e, "type", "None"), - param=getattr(e, "param", "None"), + type=openai_error_type(e, error_status_code(e, 500)), + param=openai_error_param(e), code=getattr(e, "code", getattr(e, "status_code", 500)), ) @@ -12179,16 +12184,16 @@ async def get_messages( if isinstance(e, HTTPException): raise ProxyException( message=getattr(e, "message", str(e.detail)), - type=getattr(e, "type", "None"), - param=getattr(e, "param", "None"), - code=getattr(e, "status_code", status.HTTP_400_BAD_REQUEST), + type=openai_error_type(e, error_status_code(e, status.HTTP_400_BAD_REQUEST)), + param=openai_error_param(e), + code=error_status_code(e, status.HTTP_400_BAD_REQUEST), ) else: error_msg: Final = f"{e}" raise ProxyException( message=getattr(e, "message", error_msg), - type=getattr(e, "type", "None"), - param=getattr(e, "param", "None"), + type=openai_error_type(e, error_status_code(e, 500)), + param=openai_error_param(e), code=getattr(e, "code", getattr(e, "status_code", 500)), ) @@ -12301,16 +12306,16 @@ async def run_thread( if isinstance(e, HTTPException): raise ProxyException( message=getattr(e, "message", str(e.detail)), - type=getattr(e, "type", "None"), - param=getattr(e, "param", "None"), - code=getattr(e, "status_code", status.HTTP_400_BAD_REQUEST), + type=openai_error_type(e, error_status_code(e, status.HTTP_400_BAD_REQUEST)), + param=openai_error_param(e), + code=error_status_code(e, status.HTTP_400_BAD_REQUEST), ) else: error_msg: Final = f"{e}" raise ProxyException( message=getattr(e, "message", error_msg), - type=getattr(e, "type", "None"), - param=getattr(e, "param", "None"), + type=openai_error_type(e, error_status_code(e, 500)), + param=openai_error_param(e), code=getattr(e, "code", getattr(e, "status_code", 500)), ) @@ -13980,7 +13985,7 @@ async def model_streaming_metrics( raise ProxyException( message=CommonProxyErrors.db_not_connected_error.value, type="internal_error", - param="None", + param=None, code=status.HTTP_500_INTERNAL_SERVER_ERROR, ) @@ -14114,7 +14119,7 @@ async def model_metrics( raise ProxyException( message="Prisma Client is not initialized", type="internal_error", - param="None", + param=None, code=status.HTTP_500_INTERNAL_SERVER_ERROR, ) startTime = startTime or datetime.now() - timedelta(days=DAYS_IN_A_MONTH) @@ -14229,7 +14234,7 @@ async def model_metrics_slow_responses( raise ProxyException( message="Prisma Client is not initialized", type="internal_error", - param="None", + param=None, code=status.HTTP_500_INTERNAL_SERVER_ERROR, ) if api_key is None or api_key == "undefined": @@ -14318,7 +14323,7 @@ async def model_metrics_exceptions( raise ProxyException( message="Prisma Client is not initialized", type="internal_error", - param="None", + param=None, code=status.HTTP_500_INTERNAL_SERVER_ERROR, ) @@ -15247,7 +15252,7 @@ async def async_queue_request( raise ProxyException( message=getattr(e, "detail", f"Authentication Error({e})"), type=ProxyErrorTypes.auth_error, - param=getattr(e, "param", "None"), + param=openai_error_param(e), code=getattr(e, "status_code", status.HTTP_400_BAD_REQUEST), ) elif isinstance(e, ProxyException): @@ -15255,7 +15260,7 @@ async def async_queue_request( raise ProxyException( message="Authentication Error, " + str(e), type=ProxyErrorTypes.auth_error, - param=getattr(e, "param", "None"), + param=openai_error_param(e), code=status.HTTP_400_BAD_REQUEST, ) @@ -15419,7 +15424,7 @@ async def login_v2(request: Request): raise ProxyException( message=getattr(e, "detail", str(e)), type=ProxyErrorTypes.auth_error, - param=getattr(e, "param", "None"), + param=openai_error_param(e), code=getattr(e, "status_code", status.HTTP_500_INTERNAL_SERVER_ERROR), ) else: @@ -15427,7 +15432,7 @@ async def login_v2(request: Request): raise ProxyException( message=error_msg, type=ProxyErrorTypes.auth_error, - param="None", + param=None, code=status.HTTP_500_INTERNAL_SERVER_ERROR, ) @@ -15494,7 +15499,7 @@ async def login_v3(request: Request): raise ProxyException( message=getattr(e, "detail", str(e)), type=ProxyErrorTypes.auth_error, - param=getattr(e, "param", "None"), + param=openai_error_param(e), code=getattr(e, "status_code", status.HTTP_500_INTERNAL_SERVER_ERROR), ) else: @@ -15502,7 +15507,7 @@ async def login_v3(request: Request): raise ProxyException( message=error_msg, type=ProxyErrorTypes.auth_error, - param="None", + param=None, code=status.HTTP_500_INTERNAL_SERVER_ERROR, ) @@ -15566,7 +15571,7 @@ async def login_v3_exchange(request: Request): raise ProxyException( message=str(e), type=ProxyErrorTypes.auth_error, - param="None", + param=None, code=status.HTTP_500_INTERNAL_SERVER_ERROR, ) @@ -16465,7 +16470,7 @@ async def update_config( raise ProxyException( message=getattr(e, "detail", f"Authentication Error({e})"), type=ProxyErrorTypes.auth_error, - param=getattr(e, "param", "None"), + param=openai_error_param(e), code=getattr(e, "status_code", status.HTTP_400_BAD_REQUEST), ) elif isinstance(e, ProxyException): @@ -16473,7 +16478,7 @@ async def update_config( raise ProxyException( message="Authentication Error, " + str(e), type=ProxyErrorTypes.auth_error, - param=getattr(e, "param", "None"), + param=openai_error_param(e), code=status.HTTP_400_BAD_REQUEST, ) @@ -17464,7 +17469,7 @@ async def get_config( raise ProxyException( message=getattr(e, "detail", f"Authentication Error({e})"), type=ProxyErrorTypes.auth_error, - param=getattr(e, "param", "None"), + param=openai_error_param(e), code=getattr(e, "status_code", status.HTTP_400_BAD_REQUEST), ) elif isinstance(e, ProxyException): @@ -17472,7 +17477,7 @@ async def get_config( raise ProxyException( message="Authentication Error, " + str(e), type=ProxyErrorTypes.auth_error, - param=getattr(e, "param", "None"), + param=openai_error_param(e), code=status.HTTP_400_BAD_REQUEST, ) diff --git a/tests/test_litellm/proxy/auth/test_auth_exception_handler.py b/tests/test_litellm/proxy/auth/test_auth_exception_handler.py index 21e0b83791f..b82072086e6 100644 --- a/tests/test_litellm/proxy/auth/test_auth_exception_handler.py +++ b/tests/test_litellm/proxy/auth/test_auth_exception_handler.py @@ -30,7 +30,10 @@ from litellm._logging import verbose_proxy_logger from litellm.constants import INVALID_VIRTUAL_KEY_ERROR_MARKER from litellm.exceptions import BudgetExceededError from litellm.proxy._types import ProxyErrorTypes, ProxyException, UserAPIKeyAuth -from litellm.proxy.auth.auth_exception_handler import UserAPIKeyAuthExceptionHandler +from litellm.proxy.auth.auth_exception_handler import ( + UserAPIKeyAuthExceptionHandler, + _as_proxy_exception, +) class _EngineHttp500: @@ -871,3 +874,32 @@ async def test_handle_authentication_error_traceback_only_for_unexpected_errors( assert records[0].levelname == expect_level expected_logger_name = "LiteLLM Proxy.stdout" if expect_level == "WARNING" else "LiteLLM Proxy" assert records[0].name == expected_logger_name + + +@pytest.mark.parametrize( + "auth_failure,expected_type", + [ + pytest.param( + HTTPException(status_code=401, detail="Authentication Error, invalid key"), + ProxyErrorTypes.auth_error.value, + id="http_exception", + ), + pytest.param( + httpx.ConnectError("All connection attempts failed"), + ProxyErrorTypes.no_db_connection.value, + id="database_unavailable", + ), + pytest.param( + Exception("Invalid proxy server token passed"), + ProxyErrorTypes.auth_error.value, + id="bare_exception", + ), + ], +) +def test_auth_failure_error_body_is_openai_shaped(auth_failure, expected_type): + """Every auth rejection must serialize as an OpenAI error object: a real + `type` string and a JSON null `param`, never the literal string "None".""" + body = json.loads(json.dumps({"error": _as_proxy_exception(auth_failure).to_dict()})) + + assert body["error"]["type"] == expected_type + assert body["error"]["param"] is None diff --git a/tests/test_litellm/proxy/auth/test_handle_jwt.py b/tests/test_litellm/proxy/auth/test_handle_jwt.py index 99a0a4c0a8b..abfa16a74df 100644 --- a/tests/test_litellm/proxy/auth/test_handle_jwt.py +++ b/tests/test_litellm/proxy/auth/test_handle_jwt.py @@ -1,4 +1,5 @@ import asyncio +import json import re import time from collections.abc import Mapping, Sequence @@ -30,6 +31,7 @@ from litellm.proxy.auth.handle_jwt import ( JWTAuthManager, JWTHandler, NoMatchingJWTPublicKeyError, + jwks_unavailable_exception, ) @@ -6734,3 +6736,13 @@ async def test_sync_user_role_and_teams_singular_claim_only_recognized_under_fla } assert mock_patch.call_args.kwargs["teams_ids_to_add_user_to"] == [] assert user.teams == [] + + +def test_jwks_unavailable_exception_is_openai_shaped(): + """An unreachable JWKS endpoint must answer with a real `type` string and a + JSON null `param`, never the literal string "None".""" + body = json.loads(json.dumps({"error": jwks_unavailable_exception(JWKSUnreachableError("jwks down")).to_dict()})) + + assert body["error"]["type"] == ProxyErrorTypes.auth_provider_unavailable.value + assert body["error"]["param"] is None + assert body["error"]["code"] == "503" diff --git a/tests/test_litellm/proxy/health_endpoints/test_health_endpoints.py b/tests/test_litellm/proxy/health_endpoints/test_health_endpoints.py index e3f71692c78..3d14294eff4 100644 --- a/tests/test_litellm/proxy/health_endpoints/test_health_endpoints.py +++ b/tests/test_litellm/proxy/health_endpoints/test_health_endpoints.py @@ -1089,6 +1089,18 @@ async def test_health_services_endpoint_rejects_unknown_service(): await health_services_endpoint(service="totally_unknown_service_xyz") +@pytest.mark.asyncio +async def test_health_services_endpoint_rejection_is_openai_shaped(): + """A caller mistake here must serialize as an OpenAI error object: a real + `type` string and a JSON null `param`, never the literal string "None".""" + with pytest.raises(ProxyException) as exc_info: + await health_services_endpoint(service="totally_unknown_service_xyz") + + body = json.loads(json.dumps({"error": exc_info.value.to_dict()})) + assert body["error"]["type"] == "auth_error" + assert body["error"]["param"] is None + + @pytest.mark.asyncio @pytest.mark.parametrize( "role", diff --git a/tests/test_litellm/proxy/test_proxy_server.py b/tests/test_litellm/proxy/test_proxy_server.py index 4ed6a468371..1972245764f 100644 --- a/tests/test_litellm/proxy/test_proxy_server.py +++ b/tests/test_litellm/proxy/test_proxy_server.py @@ -12507,3 +12507,14 @@ def test_disabling_docs_does_not_disable_other_routes(monkeypatch): assert client.get("/redoc").status_code == 404 assert client.get("/health/liveliness").status_code == 200 + + +def test_assistants_error_body_is_openai_shaped(client_no_auth): + """A failed /v1/assistants call must answer with an OpenAI error object: a + real `type` string and a JSON null `param`, never the literal string "None".""" + response = client_no_auth.get("/v1/assistants") + + assert response.status_code >= 400 + error = response.json()["error"] + assert error["type"] == "internal_server_error" + assert error["param"] is None From f9051a16bb69ea05527611b04d98e0230fd6c1c0 Mon Sep 17 00:00:00 2001 From: mateo-berri <277851410+mateo-berri@users.noreply.github.com> Date: Thu, 3 Sep 2026 03:29:59 -0700 Subject: [PATCH 2/4] fix(proxy): stop answering with the literal string "None" on the spend and management routes The error objects these routes return defaulted `type` and `param` to the four-character string "None", so a client's error handler matched no known OpenAI type and fell into its generic branch. Route them through the shared openai_error_payload helpers instead, so `type` comes from the carried type or the HTTP status and `param` serializes as JSON null. --- .../internal_user_endpoints.py | 5 +- .../key_management_endpoints.py | 15 +++--- .../model_management_endpoints.py | 13 ++--- .../organization_endpoints.py | 5 +- .../team_callback_endpoints.py | 11 ++-- .../management_endpoints/team_endpoints.py | 5 +- .../spend_management_endpoints.py | 53 ++++++++++--------- .../test_key_management_endpoints.py | 20 +++++++ .../test_spend_management_endpoints.py | 22 ++++++++ 9 files changed, 101 insertions(+), 48 deletions(-) diff --git a/litellm/proxy/management_endpoints/internal_user_endpoints.py b/litellm/proxy/management_endpoints/internal_user_endpoints.py index 5326074ad3c..dbe45867010 100644 --- a/litellm/proxy/management_endpoints/internal_user_endpoints.py +++ b/litellm/proxy/management_endpoints/internal_user_endpoints.py @@ -29,6 +29,7 @@ from litellm.proxy._types import * from litellm.proxy.auth.auth_checks import get_team_object, get_user_object from litellm.proxy.auth.password_policy import validate_password_policy from litellm.proxy.auth.user_api_key_auth import user_api_key_auth +from litellm.proxy.common_utils.openai_error_payload import openai_error_param from litellm.proxy.common_utils.user_api_key_cache import ( object_permission_cache_key, user_object_permission_id_cache_key, @@ -1649,7 +1650,7 @@ async def user_update( raise ProxyException( message=getattr(e, "detail", f"Authentication Error({e})"), type=ProxyErrorTypes.auth_error, - param=getattr(e, "param", "None"), + param=openai_error_param(e), code=getattr(e, "status_code", status.HTTP_400_BAD_REQUEST), ) elif isinstance(e, ProxyException): @@ -1657,7 +1658,7 @@ async def user_update( raise ProxyException( message="Authentication Error, " + str(e), type=ProxyErrorTypes.auth_error, - param=getattr(e, "param", "None"), + param=openai_error_param(e), code=status.HTTP_400_BAD_REQUEST, ) diff --git a/litellm/proxy/management_endpoints/key_management_endpoints.py b/litellm/proxy/management_endpoints/key_management_endpoints.py index d7d20d168b5..668301724f5 100644 --- a/litellm/proxy/management_endpoints/key_management_endpoints.py +++ b/litellm/proxy/management_endpoints/key_management_endpoints.py @@ -76,6 +76,7 @@ from litellm.proxy.common_utils.config_sync_pubsub import ( coordination_redis_cache, publish_config_change, ) +from litellm.proxy.common_utils.openai_error_payload import openai_error_param from litellm.proxy.common_utils.rbac_utils import check_org_admin_can_generate_keys from litellm.proxy.common_utils.timezone_utils import get_budget_reset_time from litellm.proxy.common_utils.user_api_key_cache import UserApiKeyCache @@ -3040,7 +3041,7 @@ async def update_key_fn( raise ProxyException( message=getattr(e, "detail", f"Authentication Error({e})"), type=ProxyErrorTypes.auth_error, - param=getattr(e, "param", "None"), + param=openai_error_param(e), code=getattr(e, "status_code", status.HTTP_400_BAD_REQUEST), ) elif isinstance(e, ProxyException): @@ -3048,7 +3049,7 @@ async def update_key_fn( raise ProxyException( message="Authentication Error, " + str(e), type=ProxyErrorTypes.auth_error, - param=getattr(e, "param", "None"), + param=openai_error_param(e), code=status.HTTP_400_BAD_REQUEST, ) @@ -5955,7 +5956,7 @@ async def list_keys( raise ProxyException( message=getattr(e, "detail", f"error({e})"), type=ProxyErrorTypes.internal_server_error, - param=getattr(e, "param", "None"), + param=openai_error_param(e), code=getattr(e, "status_code", fastapi.status.HTTP_500_INTERNAL_SERVER_ERROR), ) elif isinstance(e, ProxyException): @@ -5963,7 +5964,7 @@ async def list_keys( raise ProxyException( message="Authentication Error, " + str(e), type=ProxyErrorTypes.internal_server_error, - param=getattr(e, "param", "None"), + param=openai_error_param(e), code=fastapi.status.HTTP_500_INTERNAL_SERVER_ERROR, ) @@ -6111,7 +6112,7 @@ async def key_aliases( raise ProxyException( message=getattr(e, "detail", f"error({e})"), type=ProxyErrorTypes.internal_server_error, - param=getattr(e, "param", "None"), + param=openai_error_param(e), code=getattr(e, "status_code", status.HTTP_500_INTERNAL_SERVER_ERROR), ) elif isinstance(e, ProxyException): @@ -6119,7 +6120,7 @@ async def key_aliases( raise ProxyException( message="Authentication Error, " + str(e), type=ProxyErrorTypes.internal_server_error, - param=getattr(e, "param", "None"), + param=openai_error_param(e), code=status.HTTP_500_INTERNAL_SERVER_ERROR, ) @@ -6851,7 +6852,7 @@ async def key_health( raise ProxyException( message=f"Key health check failed: {e}", type=ProxyErrorTypes.internal_server_error, - param=getattr(e, "param", "None"), + param=openai_error_param(e), code=status.HTTP_500_INTERNAL_SERVER_ERROR, ) diff --git a/litellm/proxy/management_endpoints/model_management_endpoints.py b/litellm/proxy/management_endpoints/model_management_endpoints.py index ca66640bf46..f2ca8e1f541 100644 --- a/litellm/proxy/management_endpoints/model_management_endpoints.py +++ b/litellm/proxy/management_endpoints/model_management_endpoints.py @@ -58,6 +58,7 @@ from litellm.proxy.common_utils.encrypt_decrypt_utils import ( decrypt_value_helper, encrypt_value_helper, ) +from litellm.proxy.common_utils.openai_error_payload import openai_error_param from litellm.proxy.common_utils.user_api_key_cache import UserApiKeyCache from litellm.proxy.management_endpoints.common_utils import _is_user_team_admin from litellm.proxy.management_endpoints.team_endpoints import ( @@ -1700,7 +1701,7 @@ async def delete_model( raise ProxyException( message=getattr(e, "detail", f"Authentication Error({e})"), type=ProxyErrorTypes.auth_error, - param=getattr(e, "param", "None"), + param=openai_error_param(e), code=getattr(e, "status_code", status.HTTP_400_BAD_REQUEST), ) elif isinstance(e, ProxyException): @@ -1708,7 +1709,7 @@ async def delete_model( raise ProxyException( message="Authentication Error, " + str(e), type=ProxyErrorTypes.auth_error, - param=getattr(e, "param", "None"), + param=openai_error_param(e), code=status.HTTP_400_BAD_REQUEST, ) @@ -1924,7 +1925,7 @@ async def add_new_model( raise ProxyException( message=getattr(e, "detail", f"Authentication Error({e})"), type=ProxyErrorTypes.auth_error, - param=getattr(e, "param", "None"), + param=openai_error_param(e), code=getattr(e, "status_code", status.HTTP_400_BAD_REQUEST), ) elif isinstance(e, ProxyException): @@ -1932,7 +1933,7 @@ async def add_new_model( raise ProxyException( message="Authentication Error, " + str(e), type=ProxyErrorTypes.auth_error, - param=getattr(e, "param", "None"), + param=openai_error_param(e), code=status.HTTP_400_BAD_REQUEST, ) @@ -2087,7 +2088,7 @@ async def update_model( raise ProxyException( message=getattr(e, "detail", f"Authentication Error({e})"), type=ProxyErrorTypes.auth_error, - param=getattr(e, "param", "None"), + param=openai_error_param(e), code=getattr(e, "status_code", status.HTTP_400_BAD_REQUEST), ) elif isinstance(e, ProxyException): @@ -2095,7 +2096,7 @@ async def update_model( raise ProxyException( message="Authentication Error, " + str(e), type=ProxyErrorTypes.auth_error, - param=getattr(e, "param", "None"), + param=openai_error_param(e), code=status.HTTP_400_BAD_REQUEST, ) diff --git a/litellm/proxy/management_endpoints/organization_endpoints.py b/litellm/proxy/management_endpoints/organization_endpoints.py index 5e38a016099..19147885582 100644 --- a/litellm/proxy/management_endpoints/organization_endpoints.py +++ b/litellm/proxy/management_endpoints/organization_endpoints.py @@ -32,6 +32,7 @@ from litellm._uuid import uuid from litellm.proxy._types import * from litellm.proxy.auth.auth_checks import can_user_call_model, get_user_object from litellm.proxy.auth.user_api_key_auth import user_api_key_auth +from litellm.proxy.common_utils.openai_error_payload import openai_error_param from litellm.proxy.common_utils.timezone_utils import get_budget_reset_time from litellm.proxy.management_endpoints.budget_management_endpoints import ( new_budget, @@ -1272,7 +1273,7 @@ async def organization_member_add( raise ProxyException( message=getattr(e, "detail", f"Authentication Error({e})"), type=ProxyErrorTypes.auth_error, - param=getattr(e, "param", "None"), + param=openai_error_param(e), code=getattr(e, "status_code", status.HTTP_500_INTERNAL_SERVER_ERROR), ) elif isinstance(e, ProxyException): @@ -1280,7 +1281,7 @@ async def organization_member_add( raise ProxyException( message="Authentication Error, " + str(e), type=ProxyErrorTypes.auth_error, - param=getattr(e, "param", "None"), + param=openai_error_param(e), code=status.HTTP_500_INTERNAL_SERVER_ERROR, ) diff --git a/litellm/proxy/management_endpoints/team_callback_endpoints.py b/litellm/proxy/management_endpoints/team_callback_endpoints.py index c2f5dbb4032..61841eaba32 100644 --- a/litellm/proxy/management_endpoints/team_callback_endpoints.py +++ b/litellm/proxy/management_endpoints/team_callback_endpoints.py @@ -35,6 +35,7 @@ from litellm.proxy.common_utils.callback_utils import ( encrypt_callback_vars, is_sensitive_callback_key, ) +from litellm.proxy.common_utils.openai_error_payload import openai_error_param from litellm.proxy.litellm_pre_call_utils import ( _get_validated_callback_metadata, convert_key_logging_metadata_to_callback, @@ -387,7 +388,7 @@ async def add_team_callbacks( raise ProxyException( message="Internal Server Error, " + str(e), type=ProxyErrorTypes.internal_server_error.value, - param=getattr(e, "param", "None"), + param=openai_error_param(e), code=status.HTTP_500_INTERNAL_SERVER_ERROR, ) @@ -530,7 +531,7 @@ async def delete_team_callback( raise ProxyException( message="Internal Server Error, " + str(e), type=ProxyErrorTypes.internal_server_error.value, - param=getattr(e, "param", "None"), + param=openai_error_param(e), code=status.HTTP_500_INTERNAL_SERVER_ERROR, ) else: @@ -671,7 +672,7 @@ async def disable_team_logging( raise ProxyException( message="Internal Server Error, " + str(e), type=ProxyErrorTypes.internal_server_error.value, - param=getattr(e, "param", "None"), + param=openai_error_param(e), code=status.HTTP_500_INTERNAL_SERVER_ERROR, ) @@ -765,7 +766,7 @@ async def get_team_callbacks( raise ProxyException( message=getattr(e, "detail", f"Internal Server Error({e})"), type=ProxyErrorTypes.internal_server_error.value, - param=getattr(e, "param", "None"), + param=openai_error_param(e), code=getattr(e, "status_code", status.HTTP_500_INTERNAL_SERVER_ERROR), ) elif isinstance(e, ProxyException): @@ -773,6 +774,6 @@ async def get_team_callbacks( raise ProxyException( message="Internal Server Error, " + str(e), type=ProxyErrorTypes.internal_server_error.value, - param=getattr(e, "param", "None"), + param=openai_error_param(e), code=status.HTTP_500_INTERNAL_SERVER_ERROR, ) diff --git a/litellm/proxy/management_endpoints/team_endpoints.py b/litellm/proxy/management_endpoints/team_endpoints.py index 714cf252e69..cf1a20b0177 100644 --- a/litellm/proxy/management_endpoints/team_endpoints.py +++ b/litellm/proxy/management_endpoints/team_endpoints.py @@ -92,6 +92,7 @@ from litellm.proxy.auth.auth_utils import ( from litellm.proxy.auth.user_api_key_auth import user_api_key_auth from litellm.proxy.common_utils.callback_utils import encrypt_callback_vars from litellm.proxy.common_utils.json_merge_patch import apply_json_merge_patch +from litellm.proxy.common_utils.openai_error_payload import openai_error_param from litellm.proxy.common_utils.user_api_key_cache import UserApiKeyCache from litellm.proxy.management_endpoints.common_daily_activity import ( get_daily_activity_aggregated, @@ -4461,7 +4462,7 @@ async def team_info( raise ProxyException( message=getattr(e, "detail", f"Authentication Error({e})"), type=ProxyErrorTypes.auth_error, - param=getattr(e, "param", "None"), + param=openai_error_param(e), code=getattr(e, "status_code", status.HTTP_400_BAD_REQUEST), ) elif isinstance(e, ProxyException): @@ -4469,7 +4470,7 @@ async def team_info( raise ProxyException( message="Authentication Error, " + str(e), type=ProxyErrorTypes.auth_error, - param=getattr(e, "param", "None"), + param=openai_error_param(e), code=status.HTTP_400_BAD_REQUEST, ) diff --git a/litellm/proxy/spend_tracking/spend_management_endpoints.py b/litellm/proxy/spend_tracking/spend_management_endpoints.py index bb7dfafb297..276985debb8 100644 --- a/litellm/proxy/spend_tracking/spend_management_endpoints.py +++ b/litellm/proxy/spend_tracking/spend_management_endpoints.py @@ -27,6 +27,11 @@ from litellm.constants import LITTELM_INTERNAL_HEALTH_SERVICE_ACCOUNT_NAME from litellm.proxy._types import * from litellm.proxy._types import ProviderBudgetResponse, ProviderBudgetResponseObject from litellm.proxy.auth.user_api_key_auth import user_api_key_auth +from litellm.proxy.common_utils.openai_error_payload import ( + error_status_code, + openai_error_param, + openai_error_type, +) # NOTE: Avoid module-level import from common_utils: proxy_server imports this # module while common_utils may pull proxy_server during init, which can leave @@ -447,7 +452,7 @@ async def view_spend_tags( raise ProxyException( message=getattr(e, "detail", f"/spend/tags Error({e})"), type="internal_error", - param=getattr(e, "param", "None"), + param=openai_error_param(e), code=getattr(e, "status_code", status.HTTP_500_INTERNAL_SERVER_ERROR), ) elif isinstance(e, ProxyException): @@ -455,7 +460,7 @@ async def view_spend_tags( raise ProxyException( message="/spend/tags Error" + str(e), type="internal_error", - param=getattr(e, "param", "None"), + param=openai_error_param(e), code=status.HTTP_500_INTERNAL_SERVER_ERROR, ) @@ -1875,7 +1880,7 @@ async def global_get_all_tag_names(): raise ProxyException( message=getattr(e, "detail", f"/spend/all_tag_names Error({e})"), type="internal_error", - param=getattr(e, "param", "None"), + param=openai_error_param(e), code=getattr(e, "status_code", status.HTTP_500_INTERNAL_SERVER_ERROR), ) elif isinstance(e, ProxyException): @@ -1883,7 +1888,7 @@ async def global_get_all_tag_names(): raise ProxyException( message="/spend/all_tag_names Error" + str(e), type="internal_error", - param=getattr(e, "param", "None"), + param=openai_error_param(e), code=status.HTTP_500_INTERNAL_SERVER_ERROR, ) @@ -1957,7 +1962,7 @@ async def global_view_spend_tags( raise ProxyException( message=getattr(e, "detail", f"/spend/tags Error({error_str})"), type="internal_error", - param=getattr(e, "param", "None"), + param=openai_error_param(e), code=getattr(e, "status_code", status.HTTP_500_INTERNAL_SERVER_ERROR), ) elif isinstance(e, ProxyException): @@ -1965,7 +1970,7 @@ async def global_view_spend_tags( raise ProxyException( message="/spend/tags Error" + error_str, type="internal_error", - param=getattr(e, "param", "None"), + param=openai_error_param(e), code=status.HTTP_500_INTERNAL_SERVER_ERROR, ) @@ -2175,16 +2180,16 @@ async def calculate_spend(request: SpendCalculateRequest): if isinstance(e, HTTPException): raise ProxyException( message=getattr(e, "detail", str(e)), - type=getattr(e, "type", "None"), - param=getattr(e, "param", "None"), - code=getattr(e, "status_code", status.HTTP_400_BAD_REQUEST), + type=openai_error_type(e, error_status_code(e, status.HTTP_400_BAD_REQUEST)), + param=openai_error_param(e), + code=error_status_code(e, status.HTTP_400_BAD_REQUEST), ) error_msg: Final = f"{e}" raise ProxyException( message=getattr(e, "message", error_msg), - type=getattr(e, "type", "None"), - param=getattr(e, "param", "None"), - code=getattr(e, "status_code", 500), + type=openai_error_type(e, error_status_code(e, 500)), + param=openai_error_param(e), + code=error_status_code(e, 500), ) @@ -2300,7 +2305,7 @@ async def ui_view_spend_logs( raise ProxyException( message="Prisma Client is not initialized", type="internal_error", - param="None", + param=None, code=status.HTTP_401_UNAUTHORIZED, ) @@ -2358,7 +2363,7 @@ async def ui_view_spend_logs( raise ProxyException( message="Start date and end date are required", type="bad_request", - param="None", + param=None, code=status.HTTP_400_BAD_REQUEST, ) formats: Final = ["%Y-%m-%d %H:%M:%S", "%Y-%m-%d"] if is_v2 else ["%Y-%m-%d %H:%M:%S"] @@ -3118,7 +3123,7 @@ async def view_spend_logs( raise ProxyException( message=getattr(e, "detail", f"/spend/logs Error({e})"), type="internal_error", - param=getattr(e, "param", "None"), + param=openai_error_param(e), code=getattr(e, "status_code", status.HTTP_500_INTERNAL_SERVER_ERROR), ) elif isinstance(e, ProxyException): @@ -3126,7 +3131,7 @@ async def view_spend_logs( raise ProxyException( message="/spend/logs Error" + str(e), type="internal_error", - param=getattr(e, "param", "None"), + param=openai_error_param(e), code=status.HTTP_500_INTERNAL_SERVER_ERROR, ) @@ -3153,7 +3158,7 @@ async def global_spend_reset(): raise ProxyException( message="Prisma Client is not initialized", type="internal_error", - param="None", + param=None, code=status.HTTP_401_UNAUTHORIZED, ) @@ -3184,7 +3189,7 @@ async def global_spend_refresh(): raise ProxyException( message="Prisma Client is not initialized", type="internal_error", - param="None", + param=None, code=status.HTTP_401_UNAUTHORIZED, ) @@ -3255,7 +3260,7 @@ async def global_spend_for_internal_user( raise ProxyException( message="Prisma Client is not initialized", type="internal_error", - param="None", + param=None, code=status.HTTP_500_INTERNAL_SERVER_ERROR, ) try: @@ -3316,7 +3321,7 @@ async def global_spend_logs( raise ProxyException( message="Prisma Client is not initialized", type="internal_error", - param="None", + param=None, code=status.HTTP_500_INTERNAL_SERVER_ERROR, ) @@ -3360,7 +3365,7 @@ async def global_spend_logs( raise ProxyException( message=getattr(e, "detail", f"/global/spend/logs Error({error_str})"), type="internal_error", - param=getattr(e, "param", "None"), + param=openai_error_param(e), code=getattr(e, "status_code", status.HTTP_500_INTERNAL_SERVER_ERROR), ) elif isinstance(e, ProxyException): @@ -3368,7 +3373,7 @@ async def global_spend_logs( raise ProxyException( message="/global/spend/logs Error" + error_str, type="internal_error", - param=getattr(e, "param", "None"), + param=openai_error_param(e), code=status.HTTP_500_INTERNAL_SERVER_ERROR, ) @@ -3408,7 +3413,7 @@ async def global_spend(): raise ProxyException( message=getattr(e, "detail", f"/global/spend Error({error_str})"), type="internal_error", - param=getattr(e, "param", "None"), + param=openai_error_param(e), code=getattr(e, "status_code", status.HTTP_500_INTERNAL_SERVER_ERROR), ) elif isinstance(e, ProxyException): @@ -3416,7 +3421,7 @@ async def global_spend(): raise ProxyException( message="/global/spend Error" + error_str, type="internal_error", - param=getattr(e, "param", "None"), + param=openai_error_param(e), code=status.HTTP_500_INTERNAL_SERVER_ERROR, ) diff --git a/tests/test_litellm/proxy/management_endpoints/test_key_management_endpoints.py b/tests/test_litellm/proxy/management_endpoints/test_key_management_endpoints.py index 7e2e680743f..11cf2bfe9f1 100644 --- a/tests/test_litellm/proxy/management_endpoints/test_key_management_endpoints.py +++ b/tests/test_litellm/proxy/management_endpoints/test_key_management_endpoints.py @@ -17535,3 +17535,23 @@ def test_key_generation_check_blank_team_id_uses_personal_permissions(monkeypatc ) is True ) + + +def test_key_health_failure_body_is_openai_shaped(): + """A /key/health failure must answer with an OpenAI error object whose `param` + is JSON null, never the literal string "None".""" + import litellm.proxy.proxy_server as ps + + app.dependency_overrides[ps.user_api_key_auth] = lambda: UserAPIKeyAuth( + api_key="hashed-key", metadata={"logging": ["not-a-callback-object"]} + ) + try: + response = client.post("/key/health", headers={"Authorization": "Bearer sk-test"}) + finally: + app.dependency_overrides.pop(ps.user_api_key_auth, None) + + assert response.status_code == 500 + error = response.json()["error"] + assert error["type"] == "internal_server_error" + assert error["param"] is None + assert error["code"] == "500" diff --git a/tests/test_litellm/proxy/spend_tracking/test_spend_management_endpoints.py b/tests/test_litellm/proxy/spend_tracking/test_spend_management_endpoints.py index 30b086bab61..3bc05692344 100644 --- a/tests/test_litellm/proxy/spend_tracking/test_spend_management_endpoints.py +++ b/tests/test_litellm/proxy/spend_tracking/test_spend_management_endpoints.py @@ -5980,3 +5980,25 @@ def test_scoped_spend_report_range_at_max_allowed(client, monkeypatch): mock_prisma.db.query_raw.assert_awaited_once() finally: app.dependency_overrides.pop(ps.user_api_key_auth, None) + + +def test_spend_calculate_rejection_body_is_openai_shaped(client): + """A /spend/calculate rejection must answer with an OpenAI error object: a real + `type` string and a JSON null `param`, never the literal string "None".""" + app.dependency_overrides[ps.user_api_key_auth] = lambda: UserAPIKeyAuth( + user_role=LitellmUserRoles.PROXY_ADMIN, user_id="admin_user", api_key="hashed-k" + ) + try: + response = client.post( + "/spend/calculate", + json={}, + headers={"Authorization": "Bearer sk-test"}, + ) + finally: + app.dependency_overrides.pop(ps.user_api_key_auth, None) + + assert response.status_code == 400 + error = response.json()["error"] + assert error["type"] == "invalid_request_error" + assert error["param"] is None + assert error["code"] == "400" From 17a86d7b2c1ee2b614cfdfa23762964ebaf32ec0 Mon Sep 17 00:00:00 2001 From: mateo-berri <277851410+mateo-berri@users.noreply.github.com> Date: Tue, 20 Oct 2026 22:17:31 -0700 Subject: [PATCH 3/4] test(proxy): pin the mid-stream failure frame and the DB outage body to a JSON null param --- .../proxy_server/test_exception_handlers.py | 2 +- ...test_proxy_server_endpoints_and_startup.py | 33 +++++++++++++++++++ 2 files changed, 34 insertions(+), 1 deletion(-) diff --git a/tests/unit/proxy/proxy_server/test_exception_handlers.py b/tests/unit/proxy/proxy_server/test_exception_handlers.py index f6efdecc884..40a07890f15 100644 --- a/tests/unit/proxy/proxy_server/test_exception_handlers.py +++ b/tests/unit/proxy/proxy_server/test_exception_handlers.py @@ -438,7 +438,7 @@ _DB_OUTAGE_503_BODY: Final = { "error": { "message": "Service Unavailable, the authentication database is temporarily unreachable. Please retry shortly.", "type": "no_db_connection", - "param": "None", + "param": None, "code": "503", } } diff --git a/tests/unit/proxy/test_proxy_server_endpoints_and_startup.py b/tests/unit/proxy/test_proxy_server_endpoints_and_startup.py index 0864477d9b5..facaa6da090 100644 --- a/tests/unit/proxy/test_proxy_server_endpoints_and_startup.py +++ b/tests/unit/proxy/test_proxy_server_endpoints_and_startup.py @@ -7250,6 +7250,39 @@ async def test_async_data_generator_closes_the_guardrail_chain_before_returning_ assert cleanup_ran == [True] +@pytest.mark.asyncio +async def test_async_data_generator_midstream_failure_frame_is_openai_shaped_with_a_json_null_param(): + from litellm.proxy._types import UserAPIKeyAuth + from litellm.proxy.proxy_server import async_data_generator + + async def chain_failing_after_first_chunk(**_kwargs): + yield {"choices": [{"delta": {"content": "Hello"}}]} + raise ValueError("audit hook could not reach its sink") + + proxy_logging = _guarded_chain_logging(chain_failing_after_first_chunk) + proxy_logging.post_call_failure_hook = AsyncMock(return_value=None) + + with patch("litellm.proxy.proxy_server.proxy_logging_obj", proxy_logging): + frames = [ + frame + async for frame in async_data_generator( + MagicMock(), MagicMock(spec=UserAPIKeyAuth), {"model": "gpt-4o-mini"} + ) + ] + + error_bodies = [json.loads(frame.removeprefix("data: ")) for frame in frames if '"error"' in frame] + assert error_bodies == [ + { + "error": { + "message": "audit hook could not reach its sink", + "type": "internal_server_error", + "param": None, + "code": "500", + } + } + ] + + @pytest.mark.asyncio async def test_async_data_generator_closes_the_guardrail_chain_while_a_keepalive_read_is_pending(): from litellm.proxy._types import UserAPIKeyAuth From 32470e4aa341f3598bd08dace09b792e4a090430 Mon Sep 17 00:00:00 2001 From: mateo-berri <277851410+mateo-berri@users.noreply.github.com> Date: Wed, 28 Oct 2026 20:12:18 -0700 Subject: [PATCH 4/4] fix(proxy): type upstream timeouts as timeout_error and update integration asserts for the null param A 408 with no carried type now answers timeout_error instead of invalid_request_error, so a client branching on type keeps retrying it. Three integration suites still compared auth rejections against param="None". --- litellm/proxy/common_utils/openai_error_payload.py | 1 + .../authorization/test_guardrail_submissions.py | 2 +- .../authorization/test_jwt_key_mapping_lifecycle.py | 4 ++-- .../authorization/test_unknown_key_generic_401.py | 8 ++++---- .../proxy/common_utils/test_openai_error_payload.py | 12 +++++++++++- 5 files changed, 19 insertions(+), 8 deletions(-) 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