fix(proxy): recompute budget models after a user fallback rewrite in custom auth checks

Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
This commit is contained in:
jesus 2026-09-18 22:16:51 +00:00
parent 93dc85b15f
commit fe100bbbac
2 changed files with 84 additions and 10 deletions

View file

@ -3754,6 +3754,17 @@ async def _run_post_custom_auth_checks(
llm_router=llm_router,
)
def _budget_models() -> list[str]:
return _get_model_names_for_budget_checks(
model=_get_model_from_request_context(
request_data=request_data,
route=route,
request=request,
llm_router=llm_router,
team_id=valid_token.team_id,
)
)
current_model = _get_model_from_request_context(
request_data=request_data,
route=route,
@ -3794,16 +3805,7 @@ async def _run_post_custom_auth_checks(
llm_router=llm_router,
)
# Recompute after a potential budget-fallback rewrite so
# the end-user check below validates the final model
current_model = _get_model_from_request_context(
request_data=request_data,
route=route,
request=request,
llm_router=llm_router,
team_id=valid_token.team_id,
)
current_models = _get_model_names_for_budget_checks(model=current_model)
current_models = _budget_models() # rebind-ok: refresh after a budget-fallback rewrite
# 3b. Attach and check the internal user's model_max_budget.
# Custom auth builds its own token, so unlike the main path nothing has
@ -3834,6 +3836,7 @@ async def _run_post_custom_auth_checks(
llm_model_list=llm_model_list,
llm_router=llm_router,
)
current_models = _budget_models() # rebind-ok: refresh after a budget-fallback rewrite
# 4. Check end-user model_max_budget
end_user_mmb: Final = valid_token.end_user_model_max_budget

View file

@ -491,6 +491,77 @@ async def test_custom_auth_honors_key_level_model_access_restriction_denied_with
assert exc.value.type == ProxyErrorTypes.key_model_access_denied
@pytest.mark.asyncio
async def test_custom_auth_end_user_check_uses_model_after_user_budget_fallback():
"""A user-level budget fallback rewrites request_data["model"]; the
end-user model budget check must run against the rewritten model, not
the pre-rewrite one."""
valid_token = UserAPIKeyAuth(
token="test_token",
user_id="user-1",
end_user_id="eu-1",
end_user_model_max_budget={"model-a": {"max_budget": 1}},
)
request_data = {"model": "model-a"}
request = MagicMock()
request.scope = {}
end_user_calls: list[str] = []
async def is_user_within_model_budget(user_id=None, user_model_max_budget=None, model=""):
if model == "model-b":
return True
raise litellm.BudgetExceededError(current_cost=10, max_budget=1)
async def is_end_user_within_model_budget(end_user_id=None, end_user_model_max_budget=None, model=""):
end_user_calls.append(model)
if model == "model-b":
return True
raise litellm.BudgetExceededError(current_cost=10, max_budget=1)
limiter = MagicMock()
limiter.is_user_within_model_budget = is_user_within_model_budget
limiter.is_end_user_within_model_budget = is_end_user_within_model_budget
limiter.is_key_within_model_budget = AsyncMock(return_value=True)
limiter.is_team_within_model_budget = AsyncMock(return_value=True)
user_row = MagicMock()
user_row.model_max_budget = {"model-a": {"max_budget": 1}}
user_row.budget_fallbacks = {"model-a": ["model-b"]}
with (
patch( # test-quality-ok: the fallback check reads the limiter seam from module scope; no injection point
"litellm.proxy.auth.user_api_key_auth.can_key_call_model",
new=AsyncMock(return_value=True),
),
patch( # test-quality-ok: the function reads the user row seam from module scope; no injection point
"litellm.proxy.auth.user_api_key_auth._read_user_budget_row",
new=AsyncMock(return_value=user_row),
),
patch( # test-quality-ok: the function reads the end-user lookup seam from module scope; no injection point
"litellm.proxy.auth.user_api_key_auth._lookup_end_user_and_apply_budget",
new=AsyncMock(side_effect=lambda **kwargs: (kwargs["valid_token"], None)),
),
patch("litellm.proxy.proxy_server.general_settings", {}), # test-quality-ok: function reads proxy_server globals internally; no injection point
patch("litellm.proxy.proxy_server.prisma_client", MagicMock()), # test-quality-ok: function reads proxy_server globals internally; no injection point
patch("litellm.proxy.proxy_server.user_api_key_cache", MagicMock()), # test-quality-ok: function reads proxy_server globals internally; no injection point
patch("litellm.proxy.proxy_server.proxy_logging_obj", MagicMock()), # test-quality-ok: function reads proxy_server globals internally; no injection point
patch("litellm.proxy.proxy_server.llm_model_list", []), # test-quality-ok: function reads proxy_server globals internally; no injection point
patch("litellm.proxy.proxy_server.llm_router", None), # test-quality-ok: function reads proxy_server globals internally; no injection point
patch("litellm.proxy.proxy_server.model_max_budget_limiter", limiter), # test-quality-ok: function reads proxy_server globals internally; no injection point
):
await _run_post_custom_auth_checks(
valid_token=valid_token,
request=request,
request_data=request_data,
route="/v1/chat/completions",
parent_otel_span=None,
)
assert request_data["model"] == "model-b"
assert end_user_calls and set(end_user_calls) == {"model-b"}
def _proxy_server_attrs_for_custom_auth(*, user_custom_auth):
"""
Build the minimal set of proxy_server module attributes that