fix(proxy): check pre-call fallback access with the saved model_group_alias, never the one the client sent (#52077)

Co-authored-by: devin-ai-integration[bot] <158243242+devin-ai-integration[bot]@users.noreply.github.com>
This commit is contained in:
Kerry Lu 2026-11-16 14:02:42 -08:00 • committed by GitHub
parent 0ba3d312d4
commit de20250735
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
3 changed files with 71 additions and 12 deletions

View file

@ -3401,14 +3401,7 @@ class ProxyBaseLLMRequestProcessing:
original_model,
fallback_models,
)
raw_router_settings: Final = _DECODED_JSON.validate_python(
_REQUEST_FIELDS.validate_python(rate_limited_data).get("router_settings_override")
)
group_alias: Final = _DECODED_JSON.validate_python(
_REQUEST_FIELDS.validate_python(raw_router_settings).get("model_group_alias")
if isinstance(raw_router_settings, Mapping)
else None
)
group_alias: Final = router_settings.get("model_group_alias") if router_settings is not None else None
await _release_pre_call_hold(
proxy_logging_obj, _REQUEST_FIELDS.validate_python(rate_limited_data), original_exc, user_api_key_dict
)

View file

@ -49,7 +49,7 @@ def rig(tmp_path_factory: pytest.TempPathFactory) -> Iterator[Rig]:
model_entry(second_fallback, second_fallback_wire),
),
fallbacks=({primary: [fallback, second_fallback]},),
general_settings={},
general_settings={"enforce_fallback_model_access": True},
)
owned: Final = stack.enter_context(
owned_proxy_process(
@ -77,3 +77,20 @@ def test_quota_fallback_response_counts_every_candidate_a_limit_rejected(rig: Ri
assert_served(response, rig.second_fallback, "served pmrl-")
assert response.headers.get("x-litellm-attempted-fallbacks") == "2", dict(response.headers)
assert response.headers.get("x-litellm-fallback-reason") == "rate_limit:model_per_key", dict(response.headers)
def test_a_client_sent_model_group_alias_does_not_authorize_a_fallback_the_key_may_not_call(rig: Rig) -> None:
with rig.gateway.scenario() as scenario:
key: Final = scenario.key(models=[rig.primary], model_tpm_limit={rig.primary: 1})
response: Final = chat(
rig.gateway,
key,
rig.primary,
f"pmrl-{uuid.uuid4().hex}",
router_settings_override={
"model_group_alias": {rig.fallback: rig.primary, rig.second_fallback: rig.primary}
},
)
assert response.status_code == 429, (response.headers.get("x-litellm-model-group"), response.text)
assert "x-litellm-attempted-fallbacks" not in response.headers, dict(response.headers)

View file

@ -8804,7 +8804,7 @@ class TestPreCallWithFallbacksOnLocalRateLimit:
proxy_logging_obj=MagicMock(),
user_api_key_dict=ProxyUserAPIKeyAuth(),
version=None,
proxy_config=_proxy_config(),
proxy_config=_proxy_config({"model_group_alias": {"embed-alias": "embed-only"}}),
user_model=None,
user_temperature=None,
user_request_timeout=None,
@ -9020,6 +9020,55 @@ class TestPreCallWithFallbacksOnLocalRateLimit:
assert processor.data["model"] == "claude-3-haiku"
@pytest.mark.asyncio
async def test_a_client_sent_model_group_alias_does_not_authorize_a_fallback_the_key_may_not_call(
self, monkeypatch
):
from litellm.proxy.common_request_processing import ProxyBaseLLMRequestProcessing
from litellm.proxy.common_utils.proxy_rate_limit_error import ProxyRateLimitError
processor = ProxyBaseLLMRequestProcessing(
data={"model": "gpt-4", "router_settings_override": {"model_group_alias": {"gpt-3.5-turbo": "gpt-4"}}}
)
async def mock_pre_call_logic(**kwargs):
if processor.data.get("model") == "gpt-4":
raise ProxyRateLimitError(detail="TPM limit exceeded", headers={"retry-after": "30"})
return processor.data, MagicMock()
async def only_the_requested_model(*, model, valid_token, llm_router):
return model == "gpt-4"
from litellm.proxy import proxy_server
monkeypatch.setitem(proxy_server.general_settings, "enforce_fallback_model_access", True)
monkeypatch.setattr(
"litellm.proxy.auth.fallback_model_access.is_model_authorized_for_token", only_the_requested_model
)
mock_router = route_entry_points_to_stubs(MagicMock())
mock_router.fallbacks = [{"gpt-4": ["gpt-3.5-turbo"]}]
with (
patch.object(processor, "common_processing_pre_call_logic", side_effect=mock_pre_call_logic),
pytest.raises(ProxyRateLimitError),
):
await processor._pre_call_with_fallbacks(
request=MagicMock(),
general_settings={},
proxy_logging_obj=MagicMock(),
user_api_key_dict=MagicMock(router_settings=None, agent_id=None, end_user_id=None),
version=None,
proxy_config=_proxy_config(),
user_model=None,
user_temperature=None,
user_request_timeout=None,
user_max_tokens=None,
user_api_base=None,
model="gpt-4",
route_type="acompletion",
llm_router=mock_router,
)
@pytest.mark.asyncio
async def test_an_agent_owned_key_never_leaves_the_model_it_asked_for(self):
from litellm.proxy.common_request_processing import ProxyBaseLLMRequestProcessing
@ -9345,7 +9394,7 @@ class TestPreCallWithFallbacksOnLocalRateLimit:
proxy_logging_obj=MagicMock(),
user_api_key_dict=ProxyKey(),
version=None,
proxy_config=_proxy_config(),
proxy_config=_proxy_config({"model_group_alias": {"gpt-3.5-turbo": "pricey"}}),
user_model=None,
user_temperature=None,
user_request_timeout=None,
@ -9394,7 +9443,7 @@ class TestPreCallWithFallbacksOnLocalRateLimit:
proxy_logging_obj=MagicMock(),
user_api_key_dict=ProxyKey(),
version=None,
proxy_config=_proxy_config(),
proxy_config=_proxy_config({"model_group_alias": {"gpt-3.5-turbo": "roomy"}}),
user_model=None,
user_temperature=None,
user_request_timeout=None,