scriptoriumadmin/litellm
1
0
Fork 0
mirror of https://github.com/BerriAI/litellm.git synced 2026-10-29 06:33:28 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 267

fix(exception_mapping): preserve streamed errors and explicit rate limits

Browse source 
Assisted-by: OpenAI Codex
This commit is contained in:
dlowzzxx 2026-10-14 11:04:14 +02:00
parent f7115367c3
commit 49cea41cdf
2 changed files with 74 additions and 51 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

114
litellm/litellm_core_utils/exception_mapping_utils.py
View file

@ -57,6 +57,7 @@ _STRUCTURED_ERROR_SIGNAL_KEYS: Final = (
"error_type",
"__type",
)
_ERROR_PAYLOAD_ADAPTER: Final = TypeAdapter(Mapping[str, object])
def _normalise_structured_marker(value: object) -> str:
@ -75,23 +76,42 @@ def _parse_structured_error_payload(error_text: str) -> Mapping[str, object] | N
for candidate in candidates:
try:
payload: Final = json.loads(candidate)
except (TypeError, ValueError):
return _ERROR_PAYLOAD_ADAPTER.validate_json(candidate)
except ValidationError:
continue
if isinstance(payload, dict):
return payload
return None
def _response_json_payload(response: httpx.Response | None) -> object | None:
def _response_json_payload(response: httpx.Response | None) -> Mapping[str, object] | None:
if response is None:
return None
try:
return response.json()
except (TypeError, ValueError):
return _ERROR_PAYLOAD_ADAPTER.validate_json(response.content)
except (ValidationError, httpx.ResponseNotRead):
return None
def _error_payload_mapping(payload: object) -> Mapping[str, object] | None:
try:
return _ERROR_PAYLOAD_ADAPTER.validate_python(payload)
except ValidationError:
return None
def _error_object_has_validation_marker(error_object: Mapping[str, object]) -> bool:
return any(
_normalise_structured_marker(error_object.get(key)) in _VALIDATION_STRUCTURED_MARKERS
for key in _STRUCTURED_ERROR_SIGNAL_KEYS
)
def _payload_has_validation_marker(payload: Mapping[str, object]) -> bool:
nested_error: Final = _error_payload_mapping(payload.get("error"))
return _error_object_has_validation_marker(payload) or (
nested_error is not None and _error_object_has_validation_marker(nested_error)
)
def _structured_validation_signal(
*,
error_str: str,
@ -102,29 +122,14 @@ def _structured_validation_signal(
payloads: Final = tuple(
payload
for payload in (
error_body,
_error_payload_mapping(error_body),
_response_json_payload(response),
_parse_structured_error_payload(error_str),
)
if isinstance(payload, dict)
if payload is not None
)
for payload in payloads:
nested_error: Final = payload.get("error")
error_objects: Final = (payload, nested_error) if isinstance(nested_error, dict) else (payload,)
for error_object in error_objects:
for key in _STRUCTURED_ERROR_SIGNAL_KEYS:
value: Final = error_object.get(key)
if value is None:
continue
marker: Final = _normalise_structured_marker(value)
if not marker:
continue
if marker in _VALIDATION_STRUCTURED_MARKERS:
return True
return False
return any(_payload_has_validation_marker(payload) for payload in payloads)
class ExceptionCheckers:
@ -685,8 +690,8 @@ def _map_anthropic_exception(
extra_information: str,
) -> None:
error_payload: Final = _parse_structured_error_payload(error_str)
structured_error: Final = error_payload.get("error") if error_payload is not None else None
structured_message: Final = structured_error.get("message") if isinstance(structured_error, Mapping) else None
structured_error: Final = _error_payload_mapping(error_payload.get("error")) if error_payload is not None else None
structured_message: Final = structured_error.get("message") if structured_error is not None else None
if (
"prompt is too long" in error_str
or "prompt: length" in error_str
@ -703,14 +708,17 @@ def _map_anthropic_exception(
model=model,
llm_provider="anthropic",
)
elif (
isinstance(structured_message, str)
and structured_message.startswith("You have reached your specified API usage limits.")
) or ExceptionCheckers.is_error_str_rate_limit(
error_str,
status_code=getattr(original_exception, "status_code", None),
response=getattr(original_exception, "response", None),
error_body=getattr(original_exception, "body", None),
elif getattr(original_exception, "status_code", None) != 429 and (
(
isinstance(structured_message, str)
and structured_message.startswith("You have reached your specified API usage limits.")
)
or ExceptionCheckers.is_error_str_rate_limit(
error_str,
status_code=getattr(original_exception, "status_code", None),
response=getattr(original_exception, "response", None),
error_body=getattr(original_exception, "body", None),
)
):
raise RateLimitError(
message=f"AnthropicException - {error_str}",
@ -1087,7 +1095,7 @@ def _map_bedrock_exception(
llm_provider="bedrock",
response=getattr(original_exception, "response", None),
)
elif ExceptionCheckers.is_error_str_rate_limit(
elif getattr(original_exception, "status_code", None) != 429 and ExceptionCheckers.is_error_str_rate_limit(
error_str,
status_code=getattr(original_exception, "status_code", None),
response=getattr(original_exception, "response", None),
@ -1425,22 +1433,26 @@ def _map_vertex_exception(
),
),
)
elif not _structured_validation_signal(
error_str=error_str,
response=getattr(original_exception, "response", None),
error_body=getattr(original_exception, "body", None),
) and (
"429 Quota exceeded" in error_str
or "Quota exceeded for" in error_str
or "Resource exhausted" in error_str
or "IndexError: list index out of range" in error_str
or "429 Unable to submit request because the service is temporarily out of capacity." in error_str
or ExceptionCheckers.is_error_str_rate_limit(
error_str,
status_code=getattr(original_exception, "status_code", None),
elif (
getattr(original_exception, "status_code", None) != 429
and not _structured_validation_signal(
error_str=error_str,
response=getattr(original_exception, "response", None),
error_body=getattr(original_exception, "body", None),
)
and (
"429 Quota exceeded" in error_str
or "Quota exceeded for" in error_str
or "Resource exhausted" in error_str
or "IndexError: list index out of range" in error_str
or "429 Unable to submit request because the service is temporarily out of capacity." in error_str
or ExceptionCheckers.is_error_str_rate_limit(
error_str,
status_code=getattr(original_exception, "status_code", None),
response=getattr(original_exception, "response", None),
error_body=getattr(original_exception, "body", None),
)
)
):
raise RateLimitError(
message=f"litellm.RateLimitError: {custom_llm_provider}Exception - {error_str}",
@ -1962,7 +1974,7 @@ def _map_together_ai_exception(
model=model,
llm_provider="together_ai",
)
elif ExceptionCheckers.is_error_str_rate_limit(
elif getattr(original_exception, "status_code", None) != 429 and ExceptionCheckers.is_error_str_rate_limit(
error_str,
status_code=getattr(original_exception, "status_code", None),
response=getattr(original_exception, "response", None),
@ -2240,7 +2252,7 @@ def _map_azure_exception(
response=getattr(original_exception, "response", None),
body=getattr(original_exception, "body", None),
)
elif ExceptionCheckers.is_error_str_rate_limit(
elif getattr(original_exception, "status_code", None) != 429 and ExceptionCheckers.is_error_str_rate_limit(
error_str,
status_code=getattr(original_exception, "status_code", None),
response=getattr(original_exception, "response", None),

11
tests/unit/litellm_core_utils/test_exception_mapping_utils.py
View file

@ -224,6 +224,17 @@ class TestExceptionCheckers:
is True
)
def test_rate_limit_classifier_leaves_unread_stream_available(self):
response: Final = httpx.Response(
status_code=400,
stream=httpx.ByteStream(b'{"error":{"type":"invalid_request_error"}}'),
)
assert (
ExceptionCheckers.is_error_str_rate_limit("too many requests", status_code=400, response=response) is True
)
assert response.is_stream_consumed is False
assert response.read() == b'{"error":{"type":"invalid_request_error"}}'
def test_rate_limit_provider_specific_phrases(self):
"""Test that common vendor rate-limit and quota exhaustion phrases are properly recognized."""
vendor_phrases: Final = (