diff --git a/litellm/proxy/db/db_spend_update_writer.py b/litellm/proxy/db/db_spend_update_writer.py index 599ae90bcae..357d69d1371 100644 --- a/litellm/proxy/db/db_spend_update_writer.py +++ b/litellm/proxy/db/db_spend_update_writer.py @@ -299,6 +299,7 @@ class DBSpendUpdateWriter: response_obj=completion_response, start_time=start_time, end_time=end_time, + llm_router=get_llm_router(), ) payload["spend"] = response_cost or 0.0 if isinstance(payload["startTime"], datetime): diff --git a/litellm/proxy/spend_tracking/spend_tracking_utils.py b/litellm/proxy/spend_tracking/spend_tracking_utils.py index 56438fe45bd..3c96a2e6ea7 100644 --- a/litellm/proxy/spend_tracking/spend_tracking_utils.py +++ b/litellm/proxy/spend_tracking/spend_tracking_utils.py @@ -5,7 +5,7 @@ from collections.abc import Mapping, Sequence from datetime import datetime, timezone from datetime import datetime as dt from types import MappingProxyType -from typing import Final, Literal, Protocol, cast, runtime_checkable +from typing import TYPE_CHECKING, Final, Literal, Protocol, cast, runtime_checkable from pydantic import BaseModel @@ -43,6 +43,7 @@ from litellm.proxy._types import SpendLogsMetadata, SpendLogsPayload, SpendLogsR from litellm.proxy.route_llm_request import ProxyModelNotFoundError from litellm.proxy.spend_tracking.spend_log_error_logger import spend_log_error from litellm.proxy.utils import PrismaClient, hash_token +from litellm.types.router import DeploymentTypedDict, LiteLLM_Params from litellm.types.utils import ( PROMPT_CARRYING_GUARDRAIL_FIELDS, CallTypes, @@ -57,6 +58,9 @@ from litellm.types.utils import ( ) from litellm.utils import get_end_user_id_for_cost_tracking +if TYPE_CHECKING: + from litellm.router import Router + def _get_max_string_length_prompt_in_db() -> int: """ @@ -339,12 +343,36 @@ def _sl_attribution_fallback( return standard_logging_payload.get(field) or "" +def _deployment_provider(deployment: DeploymentTypedDict) -> str | None: + litellm_params: Final = LiteLLM_Params.model_validate(deployment["litellm_params"]) + try: + _, provider, _, _ = litellm.get_llm_provider( + model=litellm_params.model, custom_llm_provider=litellm_params.custom_llm_provider + ) + except litellm.exceptions.BadRequestError: + return None + return provider or None + + +def _model_group_provider(model_group: str, llm_router: "Router | None") -> str | None: + if llm_router is None or not model_group: + return None + providers: Final = frozenset( + provider + for deployment in llm_router.get_model_list(model_name=model_group) or () + if (provider := _deployment_provider(deployment)) is not None + ) + return next(iter(providers)) if len(providers) == 1 else None + + def _looks_like_model_name(model: str) -> bool: candidate: Final = model.removeprefix(MCP_SPEND_LOG_MODEL_PREFIX) return len(candidate) <= MAX_SPEND_LOG_MODEL_NAME_LENGTH and not any(char.isspace() for char in candidate) -def get_logging_payload(kwargs, response_obj, start_time, end_time) -> SpendLogsPayload: +def get_logging_payload( + kwargs, response_obj, start_time, end_time, llm_router: "Router | None" = None +) -> SpendLogsPayload: if kwargs is None: kwargs = {} @@ -443,7 +471,7 @@ def get_logging_payload(kwargs, response_obj, start_time, end_time) -> SpendLogs custom_llm_provider: Final = ( kwargs.get("custom_llm_provider") or _sl_attribution_fallback(standard_logging_payload, "custom_llm_provider") - or None + or _model_group_provider(_model_group, llm_router) ) raw_model: Final = cast(str, kwargs.get("model") or "") resolved_model: Final = ( diff --git a/tests/test_litellm/proxy/db/test_db_spend_update_writer.py b/tests/test_litellm/proxy/db/test_db_spend_update_writer.py index 8f72e1d6248..b3f5a60877d 100644 --- a/tests/test_litellm/proxy/db/test_db_spend_update_writer.py +++ b/tests/test_litellm/proxy/db/test_db_spend_update_writer.py @@ -76,6 +76,49 @@ async def test_daily_spend_tracking_with_disabled_spend_logs(): assert call_args["payload"]["custom_llm_provider"] == "openai" +@pytest.mark.asyncio +async def test_update_database_attributes_router_rejected_failure_to_model_group_provider(): + db_writer = DBSpendUpdateWriter() + db_writer._insert_spend_log_to_db = AsyncMock() + db_writer.add_spend_log_transaction_to_daily_user_transaction = AsyncMock() + llm_router: Final = litellm.Router( + model_list=[ + {"model_name": "openai-outage", "litellm_params": {"model": "openai/gpt-4o-mini", "api_key": "sk-a"}}, + {"model_name": "openai-outage", "litellm_params": {"model": "openai/gpt-4o-mini", "api_key": "sk-b"}}, + ] + ) + + with ( + patch("litellm.proxy.proxy_server.disable_spend_logs", True), # test-quality-ok: update_database reads this proxy_server module global at call time; no injection seam + patch("litellm.proxy.proxy_server.prisma_client", MagicMock()), # test-quality-ok: update_database reads this proxy_server module global at call time; no injection seam + patch("litellm.proxy.proxy_server.user_api_key_cache", MagicMock()), # test-quality-ok: update_database reads this proxy_server module global at call time; no injection seam + patch("litellm.proxy.proxy_server.litellm_proxy_budget_name", "test-budget"), # test-quality-ok: update_database reads this proxy_server module global at call time; no injection seam + patch("litellm.proxy.proxy_server.llm_router", llm_router), # test-quality-ok: get_llm_router reads this proxy_server module global at call time; no injection seam + ): + await db_writer.update_database( + token="test-token", + user_id="test-user", + end_user_id=None, + team_id=None, + org_id=None, + kwargs={ + "model": "openai-outage", + "litellm_params": { + "metadata": {"user_api_key": "test-token", "model_group": "openai-outage", "status": "failure"} + }, + }, + completion_response={}, + start_time=datetime.now(timezone.utc), + end_time=datetime.now(timezone.utc), + response_cost=0.0, + ) + await asyncio.sleep(0) + + payload: Final = db_writer.add_spend_log_transaction_to_daily_user_transaction.call_args[1]["payload"] + assert payload["model_group"] == "openai-outage" + assert payload["custom_llm_provider"] == "openai" + + def _tool_call_response(*names: str) -> object: from types import SimpleNamespace diff --git a/tests/test_litellm/proxy/spend_tracking/test_spend_tracking_utils.py b/tests/test_litellm/proxy/spend_tracking/test_spend_tracking_utils.py index 8b105e94d19..5a5fbabe771 100644 --- a/tests/test_litellm/proxy/spend_tracking/test_spend_tracking_utils.py +++ b/tests/test_litellm/proxy/spend_tracking/test_spend_tracking_utils.py @@ -4003,6 +4003,72 @@ def test_get_logging_payload_failed_request_without_standard_logging_payload_lea assert payload["custom_llm_provider"] == "" +def _router_rejected_failure_payload(model_group: str, llm_router: litellm.Router | None) -> SpendLogsPayload: + return get_logging_payload( + kwargs={ + "model": model_group, + "litellm_params": { + "metadata": {"user_api_key": "test-key", "model_group": model_group, "status": "failure"} + }, + }, + response_obj={}, + start_time=datetime.datetime.now(timezone.utc), + end_time=datetime.datetime.now(timezone.utc), + llm_router=llm_router, + ) + + +def _openai_and_anthropic_router() -> litellm.Router: + return litellm.Router( + model_list=[ + {"model_name": "openai-group", "litellm_params": {"model": "openai/gpt-4o-mini", "api_key": "sk-a"}}, + {"model_name": "openai-group", "litellm_params": {"model": "openai/gpt-4o-mini", "api_key": "sk-b"}}, + {"model_name": "mixed-group", "litellm_params": {"model": "openai/gpt-4o-mini", "api_key": "sk-a"}}, + { + "model_name": "mixed-group", + "litellm_params": {"model": "anthropic/claude-haiku-4-5", "api_key": "sk-c"}, + }, + ] + ) + + +@pytest.mark.parametrize( + "model_group,expected_provider", + [("openai-group", "openai"), ("mixed-group", ""), ("not-in-router", "")], +) +def test_get_logging_payload_router_rejected_request_takes_provider_from_model_group( + model_group: str, expected_provider: str +): + payload = _router_rejected_failure_payload(model_group, _openai_and_anthropic_router()) + + assert payload["model_group"] == model_group + assert payload["custom_llm_provider"] == expected_provider + + +def test_get_logging_payload_router_rejected_request_without_router_leaves_provider_empty(): + assert _router_rejected_failure_payload("openai-group", None)["custom_llm_provider"] == "" + + +def test_get_logging_payload_logged_provider_wins_over_model_group_provider(): + payload = get_logging_payload( + kwargs={ + "model": "openai-group", + "litellm_params": {"metadata": {"user_api_key": "test-key", "model_group": "openai-group"}}, + "standard_logging_object": { + **_make_failed_request_standard_logging_payload(), + "model_group": "openai-group", + "custom_llm_provider": "azure", + }, + }, + response_obj={}, + start_time=datetime.datetime.now(timezone.utc), + end_time=datetime.datetime.now(timezone.utc), + llm_router=_openai_and_anthropic_router(), + ) + + assert payload["custom_llm_provider"] == "azure" + + class _ModelRouterSpendLogKwargs(TypedDict): model: ReadOnly[str] litellm_params: ReadOnly[dict[str, dict[str, str]]]