[Bugfix] Return HTTP 422 for unprocessable image URLs instead of 500 (#47165)
Signed-off-by: Alexis Kinsella <alexis.kinsella@gmail.com>
This commit is contained in:
@@ -51,7 +51,7 @@ from vllm.entrypoints.serve.utils.server_utils import (
|
||||
log_response,
|
||||
validation_exception_handler,
|
||||
)
|
||||
from vllm.exceptions import VLLMValidationError
|
||||
from vllm.exceptions import VLLMUnprocessableEntityError, VLLMValidationError
|
||||
from vllm.logger import init_logger
|
||||
from vllm.reasoning import ReasoningParserManager
|
||||
from vllm.renderers.online_derenderer import OnlineDerenderer
|
||||
@@ -245,6 +245,7 @@ def build_app(
|
||||
app.exception_handler(EngineDeadError)(engine_error_handler)
|
||||
app.exception_handler(GenerationError)(generation_error_handler)
|
||||
app.exception_handler(VLLMValidationError)(exception_handler)
|
||||
app.exception_handler(VLLMUnprocessableEntityError)(exception_handler)
|
||||
app.exception_handler(Exception)(exception_handler)
|
||||
|
||||
# Ensure --api-key option from CLI takes precedence over VLLM_API_KEY
|
||||
|
||||
@@ -27,12 +27,20 @@ def create_error_response(
|
||||
"create_error_response called with %s: %s", type(exc).__name__, exc
|
||||
)
|
||||
|
||||
from vllm.exceptions import VLLMNotFoundError, VLLMValidationError
|
||||
from vllm.exceptions import (
|
||||
VLLMNotFoundError,
|
||||
VLLMUnprocessableEntityError,
|
||||
VLLMValidationError,
|
||||
)
|
||||
|
||||
if isinstance(exc, VLLMValidationError):
|
||||
err_type = "BadRequestError"
|
||||
status_code = HTTPStatus.BAD_REQUEST
|
||||
param = exc.parameter
|
||||
elif isinstance(exc, VLLMUnprocessableEntityError):
|
||||
err_type = "UnprocessableEntityError"
|
||||
status_code = HTTPStatus.UNPROCESSABLE_ENTITY
|
||||
param = exc.parameter
|
||||
elif isinstance(exc, VLLMNotFoundError):
|
||||
err_type = "NotFoundError"
|
||||
status_code = HTTPStatus.NOT_FOUND
|
||||
|
||||
@@ -64,3 +64,37 @@ class LoRAAdapterNotFoundError(VLLMNotFoundError):
|
||||
|
||||
def __str__(self):
|
||||
return self.message
|
||||
|
||||
|
||||
class VLLMUnprocessableEntityError(ValueError):
|
||||
"""vLLM-specific error for unprocessable entity requests.
|
||||
|
||||
This exception is raised when the request content is invalid or cannot be
|
||||
processed, such as when an image URL points to a non-existent or inaccessible
|
||||
resource (404, 403, DNS failure, etc.).
|
||||
|
||||
Args:
|
||||
message: The error message describing the unprocessable entity.
|
||||
parameter: Optional parameter name that failed validation.
|
||||
value: Optional value that was rejected during validation.
|
||||
"""
|
||||
|
||||
def __init__(
|
||||
self,
|
||||
message: str,
|
||||
*,
|
||||
parameter: str | None = None,
|
||||
value: Any = None,
|
||||
) -> None:
|
||||
super().__init__(message)
|
||||
self.parameter = parameter
|
||||
self.value = value
|
||||
|
||||
def __str__(self):
|
||||
base = super().__str__()
|
||||
extras = []
|
||||
if self.parameter is not None:
|
||||
extras.append(f"parameter={self.parameter}")
|
||||
if self.value is not None:
|
||||
extras.append(f"value={self.value}")
|
||||
return f"{base} ({', '.join(extras)})" if extras else base
|
||||
|
||||
@@ -13,14 +13,17 @@ from pathlib import Path
|
||||
from typing import Any, TypeVar
|
||||
from urllib.request import url2pathname
|
||||
|
||||
import aiohttp
|
||||
import numpy as np
|
||||
import numpy.typing as npt
|
||||
import requests
|
||||
import torch
|
||||
from PIL import Image, UnidentifiedImageError
|
||||
from urllib3.util import Url, parse_url
|
||||
|
||||
import vllm.envs as envs
|
||||
from vllm.connections import HTTPConnection, global_http_connection
|
||||
from vllm.exceptions import VLLMUnprocessableEntityError
|
||||
from vllm.logger import init_logger
|
||||
from vllm.multimodal.video import get_video_loader_backend_for_processor
|
||||
from vllm.utils.registry import ExtensionManager
|
||||
@@ -48,6 +51,65 @@ MODALITY_IO_MAP: dict[str, type[MediaIO]] = {
|
||||
}
|
||||
|
||||
|
||||
def _wrap_media_fetch_error(
|
||||
url: str, exc: Exception
|
||||
) -> VLLMUnprocessableEntityError | Exception:
|
||||
"""Convert media fetch exceptions to VLLMUnprocessableEntityError.
|
||||
|
||||
This handles HTTP errors that indicate the media resource is invalid
|
||||
(4xx responses except 408/429, malformed URLs) and converts them to a
|
||||
422 Unprocessable Entity error instead of 500.
|
||||
|
||||
Transient errors (5xx, 408, 429, DNS failures, connection errors,
|
||||
timeouts) are returned as-is to allow retry logic to handle them
|
||||
appropriately.
|
||||
|
||||
Returns:
|
||||
VLLMUnprocessableEntityError for permanent client errors (4xx except
|
||||
408/429, invalid URL)
|
||||
Original exception for transient errors (5xx, 408, 429, network blips)
|
||||
or other exceptions
|
||||
"""
|
||||
if isinstance(exc, aiohttp.ClientResponseError):
|
||||
if exc.status in (408, 429):
|
||||
return exc
|
||||
if exc.status < 500:
|
||||
return VLLMUnprocessableEntityError(
|
||||
f"Failed to fetch media from URL: HTTP {exc.status} error",
|
||||
parameter="image_url",
|
||||
value=url,
|
||||
)
|
||||
return exc
|
||||
|
||||
if isinstance(exc, requests.exceptions.HTTPError):
|
||||
if exc.response is not None:
|
||||
status_code = exc.response.status_code
|
||||
if status_code in (408, 429):
|
||||
return exc
|
||||
if status_code < 500:
|
||||
return VLLMUnprocessableEntityError(
|
||||
f"Failed to fetch media from URL: HTTP {status_code} error",
|
||||
parameter="image_url",
|
||||
value=url,
|
||||
)
|
||||
return exc
|
||||
|
||||
if isinstance(exc, requests.exceptions.InvalidURL):
|
||||
return VLLMUnprocessableEntityError(
|
||||
"Failed to fetch media from URL: Invalid URL format",
|
||||
parameter="image_url",
|
||||
value=url,
|
||||
)
|
||||
|
||||
if isinstance(exc, ValueError):
|
||||
return VLLMUnprocessableEntityError(
|
||||
"Failed to fetch media from URL: Invalid URL",
|
||||
parameter="image_url",
|
||||
value=url,
|
||||
)
|
||||
return exc
|
||||
|
||||
|
||||
def merge_media_io_kwargs(
|
||||
defaults: dict[str, dict[str, Any]] | None,
|
||||
overrides: dict[str, dict[str, Any]] | None,
|
||||
@@ -303,11 +365,17 @@ class MediaConnector:
|
||||
return media_io.load_bytes(cached)
|
||||
|
||||
connection = self.connection
|
||||
data = connection.get_bytes(
|
||||
url_spec.url,
|
||||
timeout=fetch_timeout,
|
||||
allow_redirects=envs.VLLM_MEDIA_URL_ALLOW_REDIRECTS,
|
||||
)
|
||||
try:
|
||||
data = connection.get_bytes(
|
||||
url_spec.url,
|
||||
timeout=fetch_timeout,
|
||||
allow_redirects=envs.VLLM_MEDIA_URL_ALLOW_REDIRECTS,
|
||||
)
|
||||
except Exception as e:
|
||||
wrapped = _wrap_media_fetch_error(url, e)
|
||||
if isinstance(wrapped, VLLMUnprocessableEntityError):
|
||||
raise wrapped from e
|
||||
raise
|
||||
|
||||
self._put_cached_bytes(url, data)
|
||||
return media_io.load_bytes(data)
|
||||
@@ -348,11 +416,17 @@ class MediaConnector:
|
||||
return await future
|
||||
|
||||
connection = self.connection
|
||||
data = await connection.async_get_bytes(
|
||||
url_spec.url,
|
||||
timeout=fetch_timeout,
|
||||
allow_redirects=envs.VLLM_MEDIA_URL_ALLOW_REDIRECTS,
|
||||
)
|
||||
try:
|
||||
data = await connection.async_get_bytes(
|
||||
url_spec.url,
|
||||
timeout=fetch_timeout,
|
||||
allow_redirects=envs.VLLM_MEDIA_URL_ALLOW_REDIRECTS,
|
||||
)
|
||||
except Exception as e:
|
||||
wrapped = _wrap_media_fetch_error(url, e)
|
||||
if isinstance(wrapped, VLLMUnprocessableEntityError):
|
||||
raise wrapped from e
|
||||
raise
|
||||
|
||||
await loop.run_in_executor(
|
||||
global_thread_pool, self._put_cached_bytes, url, data
|
||||
|
||||
Reference in New Issue
Block a user