diff --git a/litellm/proxy/auth/user_api_key_auth.py b/litellm/proxy/auth/user_api_key_auth.py index a863de8b0f8..dbd5d927653 100644 --- a/litellm/proxy/auth/user_api_key_auth.py +++ b/litellm/proxy/auth/user_api_key_auth.py @@ -1359,6 +1359,8 @@ def update_valid_token_with_end_user_params(valid_token: UserAPIKeyAuth, end_use valid_token.end_user_rpm_limit = end_user_params["end_user_rpm_limit"] if end_user_params.get("end_user_tpd_limit") is not None: valid_token.end_user_tpd_limit = end_user_params["end_user_tpd_limit"] + if cast(dict[str, object], end_user_params).get("end_user_max_budget") is not None: # cast-ok: pin dict param types + valid_token.end_user_max_budget = end_user_params["end_user_max_budget"] # rebind-ok: sibling limit write-back if end_user_params.get("allowed_model_region") is not None: valid_token.allowed_model_region = end_user_params["allowed_model_region"] if end_user_params.get("end_user_model_max_budget") is not None: @@ -4561,6 +4563,12 @@ async def _lookup_end_user_and_apply_budget( budget_info=default_budget, end_user_id=valid_token.end_user_id or "", ) + # This branch has no end-user row, so the budget here is the key's + # ``end_user_budget_id`` (or the proxy-wide default) - a fallback, not + # this end user's own budget. It must only fill a cap the custom auth + # callable left unset, never loosen one it set, so keep it out of the + # write-back and let the fill-only assignment below apply it. + end_user_params.pop("end_user_max_budget", None) valid_token = update_valid_token_with_end_user_params( valid_token=valid_token, end_user_params=end_user_params ) diff --git a/tests/unit/proxy/auth/test_custom_auth_end_user_budget.py b/tests/unit/proxy/auth/test_custom_auth_end_user_budget.py index 0c3487255f6..201f7132377 100644 --- a/tests/unit/proxy/auth/test_custom_auth_end_user_budget.py +++ b/tests/unit/proxy/auth/test_custom_auth_end_user_budget.py @@ -1,15 +1,17 @@ -import pytest from unittest.mock import AsyncMock, patch + +import pytest + import litellm -from litellm.proxy.auth.user_api_key_auth import ( - _run_post_custom_auth_checks, - update_valid_token_with_end_user_params, -) from litellm.proxy._types import ( LiteLLM_BudgetTable, LiteLLM_EndUserTable, UserAPIKeyAuth, ) +from litellm.proxy.auth.user_api_key_auth import ( + _run_post_custom_auth_checks, + update_valid_token_with_end_user_params, +) @pytest.mark.asyncio @@ -21,9 +23,7 @@ async def test_custom_auth_run_post_custom_auth_checks_without_end_user_id(): valid_token = UserAPIKeyAuth(token="test_token") # Default: common_checks should NOT be called inside the helper - with patch( - "litellm.proxy.auth.user_api_key_auth.common_checks", new_callable=AsyncMock - ) as mock_common: + with patch("litellm.proxy.auth.user_api_key_auth.common_checks", new_callable=AsyncMock) as mock_common: mock_common.return_value = True result = await _run_post_custom_auth_checks( valid_token=valid_token, @@ -39,9 +39,7 @@ async def test_custom_auth_run_post_custom_auth_checks_without_end_user_id(): # With opt-in flag: still not from the helper — the centralized gate # in the wrapper handles it. with ( - patch( - "litellm.proxy.auth.user_api_key_auth.common_checks", new_callable=AsyncMock - ) as mock_common, + patch("litellm.proxy.auth.user_api_key_auth.common_checks", new_callable=AsyncMock) as mock_common, patch( "litellm.proxy.proxy_server.general_settings", {"custom_auth_run_common_checks": True}, @@ -64,15 +62,11 @@ async def test_custom_auth_run_post_custom_auth_checks_with_end_user_budget_exce valid_token = UserAPIKeyAuth( token="test_token", end_user_id="test_user", - end_user_model_max_budget={ - "gpt-4": {"budget_limit": 10.0, "time_period": "1d"} - }, + end_user_model_max_budget={"gpt-4": {"budget_limit": 10.0, "time_period": "1d"}}, ) request_data = {"model": "gpt-4"} - with patch( - "litellm.proxy.auth.user_api_key_auth.common_checks", new_callable=AsyncMock - ): + with patch("litellm.proxy.auth.user_api_key_auth.common_checks", new_callable=AsyncMock): with patch( "litellm.proxy.proxy_server.model_max_budget_limiter.is_end_user_within_model_budget", new_callable=AsyncMock, @@ -106,9 +100,7 @@ async def test_custom_auth_enforces_end_user_budget_when_common_checks_skipped() litellm_budget_table=LiteLLM_BudgetTable(max_budget=1.0), ) - async def mock_get_current_spend( - counter_key, fallback_spend, max_budget=None, **kwargs - ): + async def mock_get_current_spend(counter_key, fallback_spend, max_budget=None, **kwargs): if counter_key == "spend:end_user:customer-1": return 5.0 return fallback_spend @@ -403,3 +395,32 @@ def test_end_user_budget_tpd_limit_reaches_the_token(): assert result.end_user_rpm_limit == 5 assert result.end_user_tpd_limit == 750000 + + +def test_end_user_max_budget_reaches_the_token(): + from litellm.proxy.auth.user_api_key_auth import _apply_budget_limits_to_end_user_params + + end_user_params = {"end_user_id": "user_1"} + _apply_budget_limits_to_end_user_params( + end_user_params=end_user_params, + budget_info=LiteLLM_BudgetTable(max_budget=20.0), + end_user_id="user_1", + ) + result = update_valid_token_with_end_user_params(UserAPIKeyAuth(token="test_token"), end_user_params) + + assert result.end_user_max_budget == 20.0 + + +def test_end_user_max_budget_is_not_leaked_from_a_previous_end_user(): + from litellm.proxy.auth.user_api_key_auth import _apply_budget_limits_to_end_user_params + + reused_token = UserAPIKeyAuth(token="test_token", end_user_max_budget=2.0) + end_user_params = {"end_user_id": "user_2"} + _apply_budget_limits_to_end_user_params( + end_user_params=end_user_params, + budget_info=LiteLLM_BudgetTable(max_budget=20.0), + end_user_id="user_2", + ) + result = update_valid_token_with_end_user_params(reused_token, end_user_params) + + assert result.end_user_max_budget == 20.0