diff --git a/litellm/litellm_core_utils/litellm_logging.py b/litellm/litellm_core_utils/litellm_logging.py index 0ef5bbf807a..28d72702f3e 100644 --- a/litellm/litellm_core_utils/litellm_logging.py +++ b/litellm/litellm_core_utils/litellm_logging.py @@ -20,11 +20,7 @@ from httpx import Response from pydantic import BaseModel, JsonValue import litellm -from litellm import ( - _custom_logger_compatible_callbacks_literal, - json_logs, - turn_off_message_logging, -) +from litellm import _custom_logger_compatible_callbacks_literal from litellm._logging import ( _is_debugging_on, _redact_string, @@ -43,6 +39,7 @@ from litellm.constants import ( DEFAULT_MOCK_RESPONSE_PROMPT_TOKEN_COUNT, EMPTY_MAPPING, PROVIDER_REQUEST_ID_HEADERS, + REDACTED_BY_LITELLM, ) from litellm.cost_calculator import ( RealtimeAPITokenUsageProcessor, @@ -1358,10 +1355,19 @@ class Logging(LiteLLMLoggingBaseClass): _litellm_params: Final = self.model_call_details.get("litellm_params", {}) _metadata: Final = _litellm_params.get("metadata", {}) or {} try: - # [Non-blocking Extra Debug Information in metadata] - if turn_off_message_logging is True: - _metadata["raw_request"] = "redacted by litellm. \ - 'litellm.turn_off_message_logging=True'" + self.model_call_details["raw_request_typed_dict"] = RawRequestTypedDict( + raw_request_api_base=self._get_masked_api_base(str(additional_args.get("api_base") or "")), + raw_request_body=self._get_raw_request_body(additional_args.get("complete_input_dict", {})), + # NOTE: setting ignore_sensitive_headers to True will cause + # the Authorization header to be leaked when calls to the health + # endpoint are made and fail. + raw_request_headers=self._get_masked_headers( + additional_args.get("headers", {}) or {}, + ), + error=None, + ) + if should_redact_message_logging(self.model_call_details): + _metadata["raw_request"] = REDACTED_BY_LITELLM else: curl_command: Final = self._get_request_curl_command( api_base=additional_args.get("api_base", ""), @@ -1369,20 +1375,7 @@ class Logging(LiteLLMLoggingBaseClass): additional_args=additional_args, data=additional_args.get("complete_input_dict", {}), ) - _metadata["raw_request"] = _redact_string(str(curl_command)) - # split up, so it's easier to parse in the UI - self.model_call_details["raw_request_typed_dict"] = RawRequestTypedDict( - raw_request_api_base=self._get_masked_api_base(str(additional_args.get("api_base") or "")), - raw_request_body=self._get_raw_request_body(additional_args.get("complete_input_dict", {})), - # NOTE: setting ignore_sensitive_headers to True will cause - # the Authorization header to be leaked when calls to the health - # endpoint are made and fail. - raw_request_headers=self._get_masked_headers( - additional_args.get("headers", {}) or {}, - ), - error=None, - ) except Exception as e: self.model_call_details["raw_request_typed_dict"] = RawRequestTypedDict( error=str(e), @@ -1476,7 +1469,7 @@ class Logging(LiteLLMLoggingBaseClass): def _print_llm_call_debugging_log( self, api_base: str, - headers: dict, + headers: dict | None, additional_args: dict, ): """ @@ -1485,8 +1478,8 @@ class Logging(LiteLLMLoggingBaseClass): Prints the RAW curl command sent from LiteLLM """ if _is_debugging_on() or self.litellm_request_debug: - if json_logs: - masked_headers: Final = self._get_masked_headers(headers) + if litellm.json_logs: + masked_headers: Final = self._get_masked_headers(headers or {}) masked_api_base: Final = self._get_masked_api_base(str(api_base or "")) if self.litellm_request_debug: verbose_logger.warning( # .warning ensures this shows up in all environments @@ -1563,20 +1556,12 @@ class Logging(LiteLLMLoggingBaseClass): else: attr = "debug" - if json_logs: - callattr = verbose_logger.warning if attr == "warning" else verbose_logger.debug - callattr( - "RAW RESPONSE:\n{}\n\n".format( - self.model_call_details.get("original_response", self.model_call_details) - ), - ) - else: - callattr = verbose_logger.warning if attr == "warning" else verbose_logger.debug - callattr( - "RAW RESPONSE:\n{}\n\n".format( - self.model_call_details.get("original_response", self.model_call_details) - ) + callattr: Final = verbose_logger.warning if attr == "warning" else verbose_logger.debug + callattr( + "RAW RESPONSE:\n{}\n\n".format( + self.model_call_details.get("original_response", self.model_call_details) ) + ) if getattr(self, "logger_fn", None) and callable(self.logger_fn): try: self.logger_fn( diff --git a/tests/test_litellm/litellm_core_utils/test_litellm_logging.py b/tests/test_litellm/litellm_core_utils/test_litellm_logging.py index 265fdb50836..d717718cba2 100644 --- a/tests/test_litellm/litellm_core_utils/test_litellm_logging.py +++ b/tests/test_litellm/litellm_core_utils/test_litellm_logging.py @@ -20,7 +20,7 @@ from openai._legacy_response import HttpxBinaryResponseContent import litellm from litellm._logging import session_id_var, trace_id_var -from litellm.constants import SENTRY_PII_DENYLIST +from litellm.constants import REDACTED_BY_LITELLM, SENTRY_PII_DENYLIST from litellm.cost_calculator import ocr_batch_cost from litellm.integrations.custom_logger import CustomLogger from litellm.litellm_core_utils.litellm_logging import Logging as LitellmLogging @@ -6615,6 +6615,65 @@ def test_pre_call_redacts_and_masks_raw_request(logging_obj): assert "key=*****" in raw_api_base +_PRIVATE_RAW_REQUEST_ARGS: Final = { + "api_base": "https://api.openai.com/v1/chat/completions", + "headers": {}, + "complete_input_dict": {"messages": [{"role": "user", "content": "PRIVATE-PHRASE"}]}, +} + + +def _pre_call_with_raw_request_logging(logging_obj) -> dict: + metadata: Final = {"user_api_key_alias": "qa-key"} + logging_obj.model_call_details["litellm_params"] = {"metadata": metadata} + logging_obj.log_raw_request_response = True + logging_obj.pre_call(input="hi", api_key="", additional_args=_PRIVATE_RAW_REQUEST_ARGS) + return metadata + + +def _assert_raw_request_redacted_for_callbacks_only(logging_obj, metadata: dict) -> None: + assert metadata["raw_request"] == REDACTED_BY_LITELLM + typed_dict: Final = logging_obj.model_call_details["raw_request_typed_dict"] + assert typed_dict["raw_request_body"] == _PRIVATE_RAW_REQUEST_ARGS["complete_input_dict"] + assert typed_dict["error"] is None + + +def test_pre_call_raw_request_honors_turn_off_message_logging_set_after_import(logging_obj, monkeypatch): + monkeypatch.setattr(litellm, "turn_off_message_logging", True) + + metadata = _pre_call_with_raw_request_logging(logging_obj) + + _assert_raw_request_redacted_for_callbacks_only(logging_obj, metadata) + + +def test_pre_call_raw_request_honors_per_request_turn_off_message_logging(logging_obj, monkeypatch): + monkeypatch.setattr(litellm, "turn_off_message_logging", False) + logging_obj.model_call_details["standard_callback_dynamic_params"] = {"turn_off_message_logging": True} + + metadata = _pre_call_with_raw_request_logging(logging_obj) + + _assert_raw_request_redacted_for_callbacks_only(logging_obj, metadata) + + +def test_debugging_log_honors_json_logs_set_after_import(logging_obj, monkeypatch): + monkeypatch.setattr(litellm, "json_logs", True) + logging_obj.litellm_request_debug = True + + with patch("litellm.litellm_core_utils.litellm_logging.verbose_logger.warning") as warning: + logging_obj._print_llm_call_debugging_log(api_base="https://api.openai.com/v1", headers={}, additional_args={}) + + assert "https://api.openai.com/v1" in warning.call_args.kwargs["extra"]["api_base"] + + +def test_debugging_log_with_json_logs_tolerates_missing_headers(logging_obj, monkeypatch): + monkeypatch.setattr(litellm, "json_logs", True) + logging_obj.litellm_request_debug = True + + with patch("litellm.litellm_core_utils.litellm_logging.verbose_logger.warning") as warning: + logging_obj._print_llm_call_debugging_log(api_base="https://api.openai.com/v1", headers=None, additional_args={}) + + assert "https://api.openai.com/v1" in warning.call_args.kwargs["extra"]["api_base"] + + def _streaming_logging_obj_with_callbacks(callbacks: list[CustomLogger]): import datetime diff --git a/tests/unit/test_main.py b/tests/unit/test_main.py index effc038f85b..c06216e4f4e 100644 --- a/tests/unit/test_main.py +++ b/tests/unit/test_main.py @@ -626,6 +626,29 @@ def test_return_raw_request_does_not_call_provider(respx_mock: respx.MockRouter) ] +def test_return_raw_request_ignores_turn_off_message_logging( + respx_mock: respx.MockRouter, monkeypatch: pytest.MonkeyPatch +): + from litellm.types.utils import CallTypes + from litellm.utils import return_raw_request + + model: Final = "gpt-4o" + messages: Final = [{"role": "user", "content": "PRIVATE-PHRASE"}] + route: Final = respx_mock.post("https://api.openai.com/v1/chat/completions").mock( + return_value=_mocked_openai_chat_response(model) + ) + monkeypatch.setattr(litellm, "turn_off_message_logging", True) + + request: Final = return_raw_request( + endpoint=CallTypes.completion, + kwargs={"model": model, "messages": messages}, + ) + + assert route.call_count == 0 + assert request.get("error") is None + assert request["raw_request_body"]["messages"] == messages + + def test_completion_forwards_verbosity_in_raw_request(respx_mock: respx.MockRouter): """Regression test: completion() must forward the verbosity param to the provider request body.""" from litellm.types.utils import CallTypes