diff --git a/litellm/proxy/auth/auth_checks.py b/litellm/proxy/auth/auth_checks.py index 00f6d44e25a..c2e874864db 100644 --- a/litellm/proxy/auth/auth_checks.py +++ b/litellm/proxy/auth/auth_checks.py @@ -1088,7 +1088,7 @@ async def _check_end_user_budget( max_budget=end_user_budget, fallback_authoritative=True, ) - if end_user_spend > end_user_budget: + if end_user_spend >= end_user_budget: raise litellm.BudgetExceededError( current_cost=end_user_spend, max_budget=end_user_budget, @@ -3900,7 +3900,7 @@ async def _team_max_budget_check( max_budget=team_object.max_budget, ) - if math.isfinite(team_object.max_budget) and spend > team_object.max_budget: + if math.isfinite(team_object.max_budget) and spend >= team_object.max_budget: if valid_token: call_info = CallInfo( token=valid_token.token, @@ -4320,7 +4320,7 @@ async def _tag_max_budget_check( max_budget=tag_object.litellm_budget_table.max_budget, fallback_authoritative=True, ) - if tag_spend <= tag_object.litellm_budget_table.max_budget: + if tag_spend < tag_object.litellm_budget_table.max_budget: continue raise litellm.BudgetExceededError( current_cost=tag_spend, diff --git a/litellm/proxy/hooks/model_max_budget_limiter.py b/litellm/proxy/hooks/model_max_budget_limiter.py index 803fe64c193..d07144e8e4f 100644 --- a/litellm/proxy/hooks/model_max_budget_limiter.py +++ b/litellm/proxy/hooks/model_max_budget_limiter.py @@ -70,7 +70,7 @@ class _PROXY_VirtualKeyModelMaxBudgetLimiter(RouterBudgetLimiting): if ( _current_spend is not None and _current_model_budget_info.max_budget is not None - and _current_spend > _current_model_budget_info.max_budget + and _current_spend >= _current_model_budget_info.max_budget ): raise litellm.BudgetExceededError( message=f"LiteLLM Virtual Key: {user_api_key_dict.token}, key_alias: {user_api_key_dict.key_alias}, exceeded budget for model={model}", @@ -134,7 +134,7 @@ class _PROXY_VirtualKeyModelMaxBudgetLimiter(RouterBudgetLimiting): if ( _current_spend is not None and _current_model_budget_info.max_budget is not None - and _current_spend > _current_model_budget_info.max_budget + and _current_spend >= _current_model_budget_info.max_budget ): raise litellm.BudgetExceededError( message=f"LiteLLM End User: {end_user_id}, exceeded budget for model={model}", diff --git a/tests/proxy_unit_tests/test_unit_test_max_model_budget_limiter.py b/tests/proxy_unit_tests/test_unit_test_max_model_budget_limiter.py index 55459721906..f9860194256 100644 --- a/tests/proxy_unit_tests/test_unit_test_max_model_budget_limiter.py +++ b/tests/proxy_unit_tests/test_unit_test_max_model_budget_limiter.py @@ -87,6 +87,13 @@ async def test_is_key_within_model_budget(budget_limiter): with pytest.raises(litellm.BudgetExceededError): await budget_limiter.is_key_within_model_budget(user_api_key, "gpt-4") + # Spend equal to the limit must block (>=, not >) + with patch.object( + budget_limiter, "_get_virtual_key_spend_for_model", return_value=100.0 + ): + with pytest.raises(litellm.BudgetExceededError): + await budget_limiter.is_key_within_model_budget(user_api_key, "gpt-4") + # Test model not in budget config assert ( await budget_limiter.is_key_within_model_budget(user_api_key, "non-existent") @@ -187,6 +194,17 @@ async def test_is_end_user_within_model_budget(budget_limiter): "gpt-4", ) + # Spend equal to the limit must block (>=, not >) + with patch.object( + budget_limiter, "_get_end_user_spend_for_model", return_value=100.0 + ): + with pytest.raises(litellm.BudgetExceededError): + await budget_limiter.is_end_user_within_model_budget( + "test-user", + {"gpt-4": {"budget_limit": 100.0, "time_period": "1d"}}, + "gpt-4", + ) + # Test model not in budget config assert ( await budget_limiter.is_end_user_within_model_budget( diff --git a/tests/test_litellm/proxy/auth/test_auth_checks.py b/tests/test_litellm/proxy/auth/test_auth_checks.py index 27f43c4948f..bfbc0801ccd 100644 --- a/tests/test_litellm/proxy/auth/test_auth_checks.py +++ b/tests/test_litellm/proxy/auth/test_auth_checks.py @@ -2928,6 +2928,104 @@ async def test_tag_budget_check_reads_from_spend_counter(): assert exc_info.value.max_budget == 1.0 +@pytest.mark.asyncio +async def test_team_budget_check_blocks_at_exact_limit(): + """Team hard budget must block when spend equals max_budget (>=, not >).""" + from litellm.proxy.utils import ProxyLogging + + team_object = LiteLLM_TeamTable( + team_id="test-team", + spend=0.0, + max_budget=10.0, + ) + valid_token = UserAPIKeyAuth(token="test-token", team_id="test-team") + + proxy_logging_obj = ProxyLogging(user_api_key_cache=None) + proxy_logging_obj.budget_alerts = AsyncMock() + + async def mock_get_current_spend( + counter_key, fallback_spend, max_budget=None, **kwargs + ): + if counter_key == "spend:team:test-team": + return 10.0 + return fallback_spend + + with patch("litellm.proxy.proxy_server.get_current_spend", mock_get_current_spend): + with pytest.raises(litellm.BudgetExceededError) as exc_info: + await _team_max_budget_check( + team_object=team_object, + valid_token=valid_token, + proxy_logging_obj=proxy_logging_obj, + ) + assert exc_info.value.current_cost == 10.0 + assert exc_info.value.max_budget == 10.0 + + +@pytest.mark.asyncio +async def test_end_user_budget_check_blocks_at_exact_limit(): + """End-user hard budget must block when spend equals max_budget (>=, not >).""" + end_user_object = LiteLLM_EndUserTable( + user_id="customer-1", + blocked=False, + spend=0.0, + litellm_budget_table=LiteLLM_BudgetTable(max_budget=10.0), + ) + + async def mock_get_current_spend( + counter_key, fallback_spend, max_budget=None, **kwargs + ): + if counter_key == "spend:end_user:customer-1": + return 10.0 + return fallback_spend + + with patch("litellm.proxy.proxy_server.get_current_spend", mock_get_current_spend): + with pytest.raises(litellm.BudgetExceededError) as exc_info: + await _check_end_user_budget( + end_user_obj=end_user_object, + route="/chat/completions", + ) + assert exc_info.value.current_cost == 10.0 + assert exc_info.value.max_budget == 10.0 + + +@pytest.mark.asyncio +async def test_tag_budget_check_blocks_at_exact_limit(): + """Tag hard budget must block when spend equals max_budget (>=, not >).""" + from litellm.proxy.utils import ProxyLogging + + tag_object = LiteLLM_TagTable( + tag_name="paid-tag", + spend=0.0, + litellm_budget_table=LiteLLM_BudgetTable(max_budget=10.0), + ) + + async def mock_get_current_spend( + counter_key, fallback_spend, max_budget=None, **kwargs + ): + if counter_key == "spend:tag:paid-tag": + return 10.0 + return fallback_spend + + with ( + patch("litellm.proxy.proxy_server.get_current_spend", mock_get_current_spend), + patch( + "litellm.proxy.auth.auth_checks.get_tag_objects_batch", + new_callable=AsyncMock, + return_value={"paid-tag": tag_object}, + ), + ): + with pytest.raises(litellm.BudgetExceededError) as exc_info: + await _tag_max_budget_check( + request_body={"metadata": {"tags": ["paid-tag"]}}, + prisma_client=MagicMock(), + user_api_key_cache=MagicMock(), + proxy_logging_obj=ProxyLogging(user_api_key_cache=None), + valid_token=UserAPIKeyAuth(token="test-token"), + ) + assert exc_info.value.current_cost == 10.0 + assert exc_info.value.max_budget == 10.0 + + @pytest.mark.asyncio async def test_team_member_budget_check_reads_from_spend_counter(): """Team member budget check should use get_current_spend when counter exists."""