mirror of
https://github.com/headroomlabs-ai/headroom.git
synced 2026-11-22 13:29:19 +08:00
fix(backends/litellm): keep the upstream 4xx status on send_message errors (#3944)
## Description Fixes #3943. The LiteLLM backend turned any upstream error it did not recognise by message text into HTTP 500, so an upstream 400 reached the client as a 500 and its retry-on-400 fallback never ran. The Anthropic and OpenAI error paths now use the exception's own 4xx `status_code` (400, 401, 403, 404, 413, 422, 429) with a matching error type, and keep the old message matching otherwise. ## Type of Change - [x] Bug fix (non-breaking change that fixes an issue) - [ ] New feature (non-breaking change that adds functionality) - [ ] Breaking change (fix or feature that would cause existing functionality to change) - [ ] Documentation update - [ ] Performance improvement - [ ] Code refactoring (no functional changes) ## Changes Made - Map 4xx status; add tests. ## Testing - [x] Unit tests pass (`pytest`) (new file only) - [x] Linting passes (`ruff check .`) (changed files) - [ ] Type checking passes (`mypy headroom`) (not run) - [x] New tests added for new functionality - [ ] Manual testing performed ### Test Output ```text # Paste relevant command output or artifact links here tests/test_litellm_transport_errors.py: 22 passed, 4 skipped ruff check (changed files): All checks passed! ``` ## Real Behavior Proof - Environment: local venv, mocked upstream - Exact command / steps: the pytest run above - Observed result: BadRequestError gives 400, was 500 - Not tested: real Bedrock ## Runtime Rollout Safety - Rollout-managed feature(s): none - Minimum rollout channel: n/a - Stable/default behavior changed: upstream 4xx keep their status - Kill switch / disable path: none - Unsafe override required: no - Qualification impact: none - Rollback path: revert the commit ## Review Readiness - [x] I have performed a self-review - [x] This PR is ready for human review ## Checklist - [ ] My code follows the project's style guidelines - [x] I have performed a self-review of my code - [ ] I have commented my code, particularly in hard-to-understand areas - [ ] I have made corresponding changes to the documentation - [ ] My changes generate no new warnings - [x] I have added tests that prove my fix is effective or that my feature works - [ ] New and existing unit tests pass locally with my changes - [x] I did **not** edit `CHANGELOG.md` — it is generated by release-please from my Conventional Commit PR title (a CI guard enforces this) ## Screenshots (if applicable) N/A ## Additional Notes N/A <!-- devin-review-badge-begin --> --- <a href="https://app.devin.ai/review/headroomlabs-ai/headroom/pull/3944" target="_blank"><picture><source media="(prefers-color-scheme: dark)" srcset="https://static.devin.ai/assets/gh-devin-review-dark.svg?v=4"><img src="https://static.devin.ai/assets/gh-devin-review-light.svg?v=4" alt="Devin Review"></picture></a> <!-- devin-review-badge-end --> --------- Co-authored-by: Claude Sonnet 5.5 <noreply@anthropic.com> Co-authored-by: JD Davis <mxjerrett@gmail.com>
This commit is contained in:
co-authored by
Claude Sonnet 5.5
JD Davis
parent
d0e9c4ca48
commit
e0e41cd2f5
@@ -557,6 +557,44 @@ def get_provider_config(provider: str) -> ProviderConfig:
|
||||
)
|
||||
|
||||
|
||||
_ANTHROPIC_ERROR_TYPES = {
|
||||
400: "invalid_request_error",
|
||||
401: "authentication_error",
|
||||
403: "permission_error",
|
||||
404: "not_found_error",
|
||||
413: "request_too_large",
|
||||
422: "invalid_request_error",
|
||||
429: "rate_limit_error",
|
||||
}
|
||||
|
||||
_OPENAI_ERROR_TYPES = {
|
||||
400: "invalid_request_error",
|
||||
401: "invalid_api_key",
|
||||
403: "permission_error",
|
||||
404: "model_not_found",
|
||||
413: "invalid_request_error",
|
||||
422: "invalid_request_error",
|
||||
429: "rate_limit_exceeded",
|
||||
}
|
||||
|
||||
|
||||
def _upstream_client_error(exc: Exception, error_types: dict[int, str]) -> tuple[int, str] | None:
|
||||
"""Map a LiteLLM exception's own 4xx ``status_code`` to ``(status, error type)``.
|
||||
|
||||
A 4xx status outside ``error_types`` keeps its status with the generic
|
||||
``invalid_request_error`` type. Returns None when the exception carries no
|
||||
4xx status, so the caller falls back to matching on the message.
|
||||
"""
|
||||
status = getattr(exc, "status_code", None)
|
||||
if not isinstance(status, int) or isinstance(status, bool):
|
||||
return None
|
||||
if status in error_types:
|
||||
return status, error_types[status]
|
||||
if 400 <= status <= 499:
|
||||
return status, "invalid_request_error"
|
||||
return None
|
||||
|
||||
|
||||
def _anthropic_usage_from_litellm(litellm_usage: Any) -> dict[str, Any]:
|
||||
"""Map LiteLLM usage to Anthropic-shape usage, surfacing cache tokens.
|
||||
|
||||
@@ -1291,7 +1329,10 @@ class LiteLLMBackend(Backend):
|
||||
status_code = 500
|
||||
|
||||
error_str = str(e).lower()
|
||||
if "authentication" in error_str or "credentials" in error_str:
|
||||
upstream = _upstream_client_error(e, _ANTHROPIC_ERROR_TYPES)
|
||||
if upstream is not None:
|
||||
status_code, error_type = upstream
|
||||
elif "authentication" in error_str or "credentials" in error_str:
|
||||
error_type = "authentication_error"
|
||||
status_code = 401
|
||||
elif "rate" in error_str or "limit" in error_str:
|
||||
@@ -1925,7 +1966,10 @@ class LiteLLMBackend(Backend):
|
||||
status_code = 500
|
||||
|
||||
error_str = str(e).lower()
|
||||
if "authentication" in error_str or "credentials" in error_str:
|
||||
upstream = _upstream_client_error(e, _OPENAI_ERROR_TYPES)
|
||||
if upstream is not None:
|
||||
status_code, error_type = upstream
|
||||
elif "authentication" in error_str or "credentials" in error_str:
|
||||
error_type = "invalid_api_key"
|
||||
status_code = 401
|
||||
elif "rate" in error_str or "limit" in error_str:
|
||||
|
||||
@@ -217,3 +217,160 @@ def test_bedrock_stream_names_transport_error_without_message() -> None:
|
||||
assert response.status_code == 200
|
||||
assert _TIMEOUT in response.text
|
||||
assert '"code": "upstream_timeout"' in response.text
|
||||
|
||||
|
||||
def _upstream_error(status_code: int, message: str) -> Exception:
|
||||
"""A LiteLLM-style exception: carries the upstream HTTP status as an attribute."""
|
||||
import litellm
|
||||
|
||||
classes = {
|
||||
400: litellm.BadRequestError,
|
||||
401: litellm.AuthenticationError,
|
||||
403: litellm.PermissionDeniedError,
|
||||
404: litellm.NotFoundError,
|
||||
422: litellm.UnprocessableEntityError,
|
||||
429: litellm.RateLimitError,
|
||||
500: litellm.InternalServerError,
|
||||
}
|
||||
if status_code not in classes:
|
||||
# Unmapped 4xx: litellm bakes the status into each class, so use a bare
|
||||
# exception carrying only the ``status_code`` attribute.
|
||||
err = Exception(message)
|
||||
err.status_code = status_code # type: ignore[attr-defined]
|
||||
return err
|
||||
response = httpx.Response(status_code, request=httpx.Request("POST", "https://upstream.test"))
|
||||
return classes[status_code](
|
||||
message=message, model="bedrock/claude", llm_provider="bedrock", response=response
|
||||
)
|
||||
|
||||
|
||||
_BAD_THINKING = (
|
||||
"BedrockException - thinking.adaptive.display: Input should be 'summarized', 'omitted'"
|
||||
)
|
||||
|
||||
|
||||
@pytest.mark.asyncio
|
||||
@pytest.mark.parametrize(
|
||||
("status", "error_type"),
|
||||
[
|
||||
(400, "invalid_request_error"),
|
||||
(403, "permission_error"),
|
||||
(422, "invalid_request_error"),
|
||||
(401, "authentication_error"),
|
||||
(404, "not_found_error"),
|
||||
(429, "rate_limit_error"),
|
||||
(402, "invalid_request_error"),
|
||||
(408, "invalid_request_error"),
|
||||
(409, "invalid_request_error"),
|
||||
(500, "api_error"),
|
||||
],
|
||||
)
|
||||
async def test_send_message_keeps_upstream_litellm_status(status: int, error_type: str) -> None:
|
||||
with (
|
||||
patch(
|
||||
"headroom.backends.litellm.acompletion",
|
||||
new_callable=AsyncMock,
|
||||
side_effect=_upstream_error(status, _BAD_THINKING),
|
||||
),
|
||||
patch("headroom.backends.litellm._fetch_bedrock_inference_profiles", return_value={}),
|
||||
):
|
||||
backend = LiteLLMBackend(provider="bedrock", region="us-east-1")
|
||||
result = await backend.send_message(_BODY, {})
|
||||
|
||||
assert result.status_code == status
|
||||
assert result.body["error"]["type"] == error_type
|
||||
|
||||
|
||||
@pytest.mark.asyncio
|
||||
@pytest.mark.parametrize(
|
||||
("status", "error_type"),
|
||||
[
|
||||
(400, "invalid_request_error"),
|
||||
(403, "permission_error"),
|
||||
(422, "invalid_request_error"),
|
||||
(401, "invalid_api_key"),
|
||||
(404, "model_not_found"),
|
||||
(429, "rate_limit_exceeded"),
|
||||
(402, "invalid_request_error"),
|
||||
(408, "invalid_request_error"),
|
||||
(409, "invalid_request_error"),
|
||||
(500, "api_error"),
|
||||
],
|
||||
)
|
||||
async def test_send_openai_message_keeps_upstream_litellm_status(
|
||||
status: int, error_type: str
|
||||
) -> None:
|
||||
with (
|
||||
patch(
|
||||
"headroom.backends.litellm.acompletion",
|
||||
new_callable=AsyncMock,
|
||||
side_effect=_upstream_error(status, _BAD_THINKING),
|
||||
),
|
||||
patch("headroom.backends.litellm._fetch_bedrock_inference_profiles", return_value={}),
|
||||
):
|
||||
backend = LiteLLMBackend(provider="bedrock", region="us-east-1")
|
||||
result = await backend.send_openai_message(_BODY, {})
|
||||
|
||||
assert result.status_code == status
|
||||
assert result.body["error"]["type"] == error_type
|
||||
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_send_message_substring_fallback_without_status_attribute() -> None:
|
||||
with (
|
||||
patch(
|
||||
"headroom.backends.litellm.acompletion",
|
||||
new_callable=AsyncMock,
|
||||
side_effect=RuntimeError("Authentication failed: bad credentials"),
|
||||
),
|
||||
patch("headroom.backends.litellm._fetch_bedrock_inference_profiles", return_value={}),
|
||||
):
|
||||
backend = LiteLLMBackend(provider="bedrock", region="us-east-1")
|
||||
result = await backend.send_message(_BODY, {})
|
||||
|
||||
assert result.status_code == 401
|
||||
assert result.body["error"]["type"] == "authentication_error"
|
||||
|
||||
|
||||
class _StatusError(Exception):
|
||||
def __init__(self, status_code: object) -> None:
|
||||
super().__init__("payload too large")
|
||||
self.status_code = status_code
|
||||
|
||||
|
||||
@pytest.mark.asyncio
|
||||
@pytest.mark.parametrize(
|
||||
("method", "error_type"),
|
||||
[("send_message", "request_too_large"), ("send_openai_message", "invalid_request_error")],
|
||||
)
|
||||
async def test_413_status_attribute_is_kept(method: str, error_type: str) -> None:
|
||||
with (
|
||||
patch(
|
||||
"headroom.backends.litellm.acompletion",
|
||||
new_callable=AsyncMock,
|
||||
side_effect=_StatusError(413),
|
||||
),
|
||||
patch("headroom.backends.litellm._fetch_bedrock_inference_profiles", return_value={}),
|
||||
):
|
||||
backend = LiteLLMBackend(provider="bedrock", region="us-east-1")
|
||||
result = await getattr(backend, method)(_BODY, {})
|
||||
|
||||
assert result.status_code == 413
|
||||
assert result.body["error"]["type"] == error_type
|
||||
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_non_int_status_attribute_falls_back_to_substring() -> None:
|
||||
with (
|
||||
patch(
|
||||
"headroom.backends.litellm.acompletion",
|
||||
new_callable=AsyncMock,
|
||||
side_effect=_StatusError("429"),
|
||||
),
|
||||
patch("headroom.backends.litellm._fetch_bedrock_inference_profiles", return_value={}),
|
||||
):
|
||||
backend = LiteLLMBackend(provider="bedrock", region="us-east-1")
|
||||
result = await backend.send_message(_BODY, {})
|
||||
|
||||
assert result.status_code == 500
|
||||
assert result.body["error"]["type"] == "api_error"
|
||||
|
||||
Reference in New Issue
Block a user