search

DataDog / dd-trace-py

Datadog Python APM Client
https://ddtrace.readthedocs.io/
Other
532 stars 408 forks source link

Async requests with AzureOpenAI vision returns a 431 from openai in FastAPI #10458

Open RogerThomas opened 2 weeks ago

RogerThomas commented 2 weeks ago

Summary of problem

Async requests with vision returns a 431 from openai

I have confirmed this only happens when using datadog as when I replace this

ENTRYPOINT ["ddtrace-run"]
CMD ["uvicorn", "app.main:app" , "--host", "0.0.0.0", "--port", "80"]

with

ENTRYPOINT []
CMD ["uvicorn", "app.main:app" , "--host", "0.0.0.0", "--port", "80"]

It doesn't happen.

When I run the get_descriptions from a fastapi route

#!/usr/bin/env python
import asyncio
import base64
import io
import time
from typing import Literal, TypedDict

from langchain.prompts import ChatPromptTemplate, HumanMessagePromptTemplate
from langchain.schema import SystemMessage
from langchain_community.callbacks import get_openai_callback
from langchain_core.output_parsers import StrOutputParser
from langchain_openai import AzureChatOpenAI
from PIL import Image

ImageFormat = Literal["JPEG", "PNG"]
ImageURL = TypedDict("ImageURL", {"url": str, "detail": str})
ImageContentObj = TypedDict("ImageContentObj", {"type": str, "image_url": ImageURL})

def _convert_image_to_base64_string(image: Image.Image, fmt: ImageFormat = "JPEG") -> str:
    """Convert image to a base64 utf-8 encoded string"""
    bytes_io = io.BytesIO()
    image.save(bytes_io, format=fmt)
    img_str = base64.b64encode(bytes_io.getvalue()).decode("utf-8")
    return img_str

def _create_image(width: int, height: int, color: tuple[int, int, int]):
    image = Image.new("RGB", (width, height), color)
    return image

async def get_descriptions(llm: AzureChatOpenAI):
    # llm = get_llm(model_type="4o-mini")
    image0 = _create_image(50, 50, (255, 0, 0))
    image1 = _create_image(50, 50, (0, 255, 0))
    image2 = _create_image(50, 50, (0, 0, 255))

    system_message = SystemMessage("Descripe this image")
    human_message_prompt_template = HumanMessagePromptTemplate.from_template(
        template=[
            {  # pyright: ignore[reportArgumentType]
                "type": "image_url",
                "image_url": {"url": "data:image/jpeg;base64,{image_base64}", "detail": "low"},
            }
        ]
    )
    prompt_template = ChatPromptTemplate(
        input_variables=["image_base64"],
        messages=[system_message, human_message_prompt_template],
    )
    chain = prompt_template | llm | StrOutputParser()

    t1 = time.time()
    with get_openai_callback() as cb:
        descriptions = await asyncio.gather(
            *(
                chain.ainvoke({"image_base64": _convert_image_to_base64_string(image)})
                for image in (image0, image1, image2)
            )
        )
    took = time.time() - t1
    for description in descriptions:
        print(description)
    print(
        f"Took {took:,.2f}s to get descriptions for 3 images, "
        f"token consumption (Total | Prompt | Completion): {cb.total_tokens} | "
        f"{cb.prompt_tokens} | {cb.completion_tokens}, cost: ${cb.total_cost:.5f}"
    )

if __name__ == "__main__":
    llm = AzureChatOpenAI()
    asyncio.run(get_descriptions(llm))

I get

descriptions = await asyncio.gather(
                    ^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.12/site-packages/ddtrace/contrib/langchain/patch.py", line 879, in traced_lcel_runnable_sequence_async
    final_output = await func(*args, **kwargs)
                    ^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.12/site-packages/langchain_core/runnables/base.py", line 2920, in ainvoke
    input = await asyncio.create_task(part(), context=context)  # type: ignore
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.12/site-packages/langchain_core/language_models/chat_models.py", line 298, in ainvoke
    llm_result = await self.agenerate_prompt(
                  ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.12/site-packages/langchain_core/language_models/chat_models.py", line 787, in agenerate_prompt
    return await self.agenerate(
            ^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.12/site-packages/ddtrace/contrib/langchain/patch.py", line 523, in traced_chat_model_agenerate
    chat_completions = await func(*args, **kwargs)
                        ^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.12/site-packages/langchain_core/language_models/chat_models.py", line 747, in agenerate
    raise exceptions[0]
  File "/usr/local/lib/python3.12/site-packages/langchain_core/language_models/chat_models.py", line 923, in _agenerate_with_cache
    result = await self._agenerate(
              ^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.12/site-packages/langchain_openai/chat_models/base.py", line 752, in _agenerate
    response = await self.async_client.create(**payload)
                ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.12/site-packages/ddtrace/contrib/openai/patch.py", line 290, in patched_endpoint
    resp = await func(*args, **kwargs)
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.12/site-packages/openai/resources/chat/completions.py", line 1295, in create
    return await self._post(
            ^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.12/site-packages/openai/_base_client.py", line 1826, in post
    return await self.request(cast_to, opts, stream=stream, stream_cls=stream_cls)
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.12/site-packages/openai/_base_client.py", line 1519, in request
    return await self._request(
            ^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.12/site-packages/openai/_base_client.py", line 1620, in _request
    raise self._make_status_error_from_response(err.response) from None
openai.APIStatusError: Error code: 431

This doesn't seem to happen with the non azure chat, i.e. ChatOpenAI doesn't seem to have this issue

Which version of dd-trace-py are you using?

ddtrace==2.11.1

Which version of pip are you using?

Not using pip, using uv

uv --version uv 0.3.4 (Homebrew 2024-08-26)

Which libraries and their versions are you using?

aiobotocore==2.13.2
aiohappyeyeballs==2.3.4
aiohttp==3.10.3
aioitertools==0.11.0
aiosignal==1.3.1
annotated-types==0.7.0
anyio==4.4.0
astroid==3.2.4
attrs==23.2.0
azure-common==1.1.28
azure-core==1.30.2
azure-identity==1.17.1
azure-search==1.0.0b2
azure-search-documents==11.5.1
basedpyright==1.17.0
beautifulsoup4==4.12.3
blinker==1.8.2
boto3==1.34.131
boto3-stubs==1.34.160
botocore==1.34.131
botocore-stubs==1.34.160
bytecode==0.15.1
cattrs==23.2.3
certifi==2024.7.4
cffi==1.16.0
cfgv==3.4.0
charset-normalizer==3.3.2
click==8.1.7
cryptography==43.0.0
dataclasses-json==0.6.7
ddsketch==3.0.1
ddtrace==2.11.1
debugpy==1.8.2
deprecated==1.2.14
dill==0.3.8
distlib==0.3.8
distro==1.9.0
dnspython==2.6.1
docker==7.1.0
docopt==0.6.2
dynaconf==3.2.6
email-validator==2.2.0
envier==0.5.2
et-xmlfile==1.1.0
fastapi==0.111.1
fastapi-cli==0.0.4
filelock==3.15.4
flask==3.0.0
freezegun==1.5.1
frozenlist==1.4.1
git-python==1.0.3
gitdb==4.0.11
gitpython==3.1.43
greenlet==3.0.3
h11==0.14.0
hiredis==2.4.0
httpcore==1.0.5
httptools==0.6.1
httpx==0.27.0
identify==2.6.0
idna==3.7
importlib-metadata==8.0.0
iniconfig==2.0.0
isodate==0.6.1
isort==5.13.2
itsdangerous==2.2.0
jinja2==3.1.4
jmespath==1.0.1
jsonpatch==1.33
jsonpointer==3.0.0
langchain==0.2.15
langchain-community==0.2.10
langchain-core==0.2.36
langchain-openai==0.1.20
langchain-text-splitters==0.2.2
langsmith==0.1.95
language-tags==1.2.0
lxml==5.3.0
markdown-it-py==3.0.0
markupsafe==2.1.5
marshmallow==3.21.3
mccabe==0.7.0
mdurl==0.1.2
more-itertools==10.3.0
moto==5.0.11
msal==1.30.0
msal-extensions==1.2.0
msrest==0.7.1
multidict==6.0.5
mypy-boto3-s3==1.34.158
mypy-extensions==1.0.0
nodeenv==1.9.1
nodejs-wheel-binaries==20.16.0
numpy==1.26.4
oauthlib==3.2.2
openai==1.37.1
openpyxl==3.1.5
opentelemetry-api==1.26.0
orjson==3.10.6
packaging==24.1
pandas==2.2.2
pikepdf==9.1.2
pillow==10.4.0
platformdirs==4.2.2
playwright==1.45.1
pluggy==1.5.0
portalocker==2.10.1
pre-commit==3.8.0
protobuf==5.27.3
pycparser==2.22
pycryptodome==3.20.0
pydantic==2.4.2
pydantic-core==2.10.1
pyee==11.1.0
pygments==2.18.0
pyjwt==2.8.0
pylint==3.2.6
pylint-protobuf==0.22.0
pymupdf==1.24.9
pymupdfb==1.24.9
pynamodb==6.0.1
pynamodb-attributes==0.5.0
pytest==8.3.2
pytest-asyncio==0.23.8
pytest-mock==3.14.0
python-dateutil==2.9.0.post0
python-dotenv==1.0.1
python-json-logger==2.0.7
python-multipart==0.0.9
python-ulid==1.1.0
pytz==2024.1
pyyaml==6.0.1
redis==5.0.8
redis-om==0.2.2
regex==2024.7.24
requests==2.32.3
requests-mock==1.12.1
requests-oauthlib==2.0.0
requests-toolbelt==1.0.0
responses==0.25.3
rich==13.7.1
s3transfer==0.10.2
setuptools==69.5.1
shellingham==1.5.4
six==1.16.0
smmap==5.0.1
sniffio==1.3.1
soupsieve==2.5
sqlalchemy==2.0.31
sse-starlette==2.1.3
sseclient-py==1.8.0
starlette==0.37.2
stomp-py==8.1.2
tenacity==8.5.0
termcolor==2.4.0
tiktoken==0.7.0
tinydb==4.8.0
tomlkit==0.13.0
tqdm==4.66.4
typer==0.12.3
types-aiobotocore==2.13.2.post1
types-aiobotocore-lambda==2.13.2
types-aiobotocore-s3==2.13.3
types-awscrt==0.21.2
types-cffi==1.16.0.20240331
types-pyopenssl==24.1.0.20240722
types-redis==4.6.0.20240726
types-s3transfer==0.10.1
types-setuptools==71.1.0.20240726
typing-extensions==4.12.2
typing-inspect==0.9.0
tzdata==2024.1
urllib3==2.2.2
uvicorn==0.30.4
uvloop==0.19.0
virtualenv==20.26.3
watchfiles==0.22.0
websocket-client==1.8.0
websockets==12.0
werkzeug==3.0.3
wrapt==1.16.0
xmltodict==0.13.0
yarl==1.9.4
zipp==3.19.2

How can we reproduce your problem?

Create a fastapi route and run the above

What is the result that you get?

Error is above

What is the result that you expected?

No error

lievan commented 1 week ago

Hi Roger, thanks for flagging this—I'll work on reproducing your issue.

In the meantime, what is the full error message you get from OpenAI (not just the status code)?

try:
    ...
except openai.APIStatusError as e:
    print("Another non-200-range status code was received")
    print(e.status_code)
    print(e.response)
RogerThomas commented 1 week ago

@lievan I can't really do this as I'd have to change the code in the underyling library

lievan commented 2 days ago

@RogerThomas, thanks for providing the example snippet, but i haven't been able to reproduce your error by

Are you running into this error consistently? Our integration doesn't add data to request headers as the 431 error suggests so we'll need some more details about this error to investigate this further.

It looks like the error is being raised by:

        descriptions = await asyncio.gather(
            *(
                chain.ainvoke({"image_base64": _convert_image_to_base64_string(image)})
                for image in (image0, image1, image2)
            )
        )

Could you not wrap this with:

try:
     descriptions = await asyncio.gather(
            *(
                chain.ainvoke({"image_base64": _convert_image_to_base64_string(image)})
                for image in (image0, image1, image2)
            )
        )
except openai.APIStatusError as e:
    print("Another non-200-range status code was received")
    print(e.status_code)
    print(e.response)