Merge a5c0874a75 into 72228db5c6

litellm/__init__.py

@@ -192,6 +192,7 @@ standard_logging_payload_excluded_fields: Optional[List[str]] = (
     None  # Fields to exclude from StandardLoggingPayload before callbacks receive it
 )
 log_raw_request_response: bool = False
+max_raw_request_body_log_chars: Optional[int] = None
 log_client_error_tracebacks: bool = False
 log_auth_failure_key_identity: bool = False
 request_correlation_in_logs: bool = False

litellm/constants.py

@@ -2199,6 +2199,7 @@ MICROSOFT_USER_LAST_NAME_ATTRIBUTE: Final = str(os.getenv("MICROSOFT_USER_LAST_N
 # Maximum payload size (in bytes) to fully serialize for DEBUG logging.
 # Payloads larger than this are truncated to avoid multi-second json.dumps blocking the response.
 MAX_PAYLOAD_SIZE_FOR_DEBUG_LOG: Final = get_env_int("MAX_PAYLOAD_SIZE_FOR_DEBUG_LOG", 102400)  # 100 KB
+MAX_RAW_REQUEST_BODY_LOG_CHARS: Final = get_env_int("MAX_RAW_REQUEST_BODY_LOG_CHARS", 102400)  # 100 KB
 
 # Policy template enrichment
 MAX_COMPETITOR_NAMES: Final = get_env_int("MAX_COMPETITOR_NAMES", 100)

litellm/litellm_core_utils/litellm_logging.py

@@ -49,6 +49,7 @@ from litellm.constants import (
     INTERNAL_CALLBACK_NAMES,
     KUBERNETES_POD_ROUTING_KEY,
     LITELLM_LOGGING_NO_UPSTREAM_LLM_CALL,
+    MAX_RAW_REQUEST_BODY_LOG_CHARS,
     PROVIDER_REQUEST_ID_HEADERS,
     REALTIME_METERED_USAGE_KEY,
     REDACTED_BY_LITELLM,
@@ -1841,8 +1842,17 @@ class Logging(LiteLLMLoggingBaseClass):
                 else:
                     verbose_logger.debug("\x1b[92m%s\x1b[0m\n", curl_command)
 
+    def _get_max_raw_request_body_log_chars(self) -> int:
+        if getattr(litellm, "max_raw_request_body_log_chars", None) is not None:
+            return cast(int, litellm.max_raw_request_body_log_chars)
+        return MAX_RAW_REQUEST_BODY_LOG_CHARS
+
     def _get_request_body(self, data: dict) -> str:
-        return str(data)
+        body: Final = str(data)
+        max_chars: Final = self._get_max_raw_request_body_log_chars()
+        if len(body) > max_chars:
+            return f"<{len(body)} chars omitted, over MAX_RAW_REQUEST_BODY_LOG_CHARS={max_chars}>"
+        return body
 
     def _get_request_curl_command(self, api_base: str, headers: dict | None, additional_args: dict, data: dict) -> str:
         masked_api_base: Final = self._get_masked_api_base(api_base)
@@ -1858,7 +1868,10 @@ class Logging(LiteLLMLoggingBaseClass):
         if additional_args.get("request_str", None) is not None:
             # print the sagemaker / bedrock client request
             curl_command = "\nRequest Sent from LiteLLM:\n"
-            request_str: Final = additional_args.get("request_str", "")
+            request_str: str = str(additional_args.get("request_str", ""))
+            max_chars: Final = self._get_max_raw_request_body_log_chars()
+            if len(request_str) > max_chars:
+                request_str = f"<{len(request_str)} chars omitted, over MAX_RAW_REQUEST_BODY_LOG_CHARS={max_chars}>"
             curl_command += request_str
         return curl_command
 

tests/unit/litellm_core_utils/test_litellm_logging.py

@@ -8711,6 +8711,77 @@ def test_pre_call_redacts_and_masks_raw_request(logging_obj):
     assert "key=*****" in raw_api_base
 
 
+def test_pre_call_omits_raw_request_body_over_size_limit(logging_obj, monkeypatch):
+    """Redacting an oversized raw request blocks the event loop, so the body is omitted from the curl command."""
+    monkeypatch.setattr("litellm.litellm_core_utils.litellm_logging.MAX_RAW_REQUEST_BODY_LOG_CHARS", 100)
+    metadata = {"user_api_key_alias": "qa-key"}
+    logging_obj.model_call_details["litellm_params"] = {"metadata": metadata}
+    logging_obj.log_raw_request_response = True
+    body = {"messages": [{"role": "user", "content": "LONG-PROMPT-CONTENT " * 20}]}
+
+    logging_obj.pre_call(
+        input="hi",
+        api_key="",
+        additional_args={
+            "api_base": "https://api.openai.com/v1/chat/completions",
+            "headers": {},
+            "complete_input_dict": body,
+        },
+    )
+
+    raw_request = metadata["raw_request"]
+    assert "LONG-PROMPT-CONTENT" not in raw_request
+    assert "chars omitted, over MAX_RAW_REQUEST_BODY_LOG_CHARS=100" in raw_request
+    assert "https://api.openai.com/v1/chat/completions" in raw_request
+    assert logging_obj.model_call_details["raw_request_typed_dict"]["raw_request_body"] == body
+
+
+def test_pre_call_omits_request_str_over_size_limit(logging_obj, monkeypatch):
+    """Oversized request_str (e.g. from Bedrock or SageMaker) is also omitted from raw request logging."""
+    monkeypatch.setattr("litellm.litellm_core_utils.litellm_logging.MAX_RAW_REQUEST_BODY_LOG_CHARS", 100)
+    metadata = {"user_api_key_alias": "qa-key"}
+    logging_obj.model_call_details["litellm_params"] = {"metadata": metadata}
+    logging_obj.log_raw_request_response = True
+    long_request_str = "BEDROCK-RAW-PAYLOAD-" * 20
+
+    logging_obj.pre_call(
+        input="hi",
+        api_key="",
+        additional_args={
+            "api_base": "https://bedrock-runtime.us-east-1.amazonaws.com",
+            "headers": {},
+            "request_str": long_request_str,
+        },
+    )
+
+    raw_request = metadata["raw_request"]
+    assert "BEDROCK-RAW-PAYLOAD-" not in raw_request
+    assert "chars omitted, over MAX_RAW_REQUEST_BODY_LOG_CHARS=100" in raw_request
+
+
+def test_pre_call_includes_request_body_under_size_limit(logging_obj, monkeypatch):
+    """Normal sized payloads under the limit are fully included and redacted."""
+    monkeypatch.setattr("litellm.litellm_core_utils.litellm_logging.MAX_RAW_REQUEST_BODY_LOG_CHARS", 1000)
+    metadata = {"user_api_key_alias": "qa-key"}
+    logging_obj.model_call_details["litellm_params"] = {"metadata": metadata}
+    logging_obj.log_raw_request_response = True
+    body = {"messages": [{"role": "user", "content": "short prompt"}]}
+
+    logging_obj.pre_call(
+        input="hi",
+        api_key="",
+        additional_args={
+            "api_base": "https://api.openai.com/v1/chat/completions",
+            "headers": {},
+            "complete_input_dict": body,
+        },
+    )
+
+    raw_request = metadata["raw_request"]
+    assert "short prompt" in raw_request
+    assert "chars omitted" not in raw_request
+
+
 _PRIVATE_RAW_REQUEST_ARGS: Final = {
     "api_base": "https://api.openai.com/v1/chat/completions",
     "headers": {},