The content of the api_key or default_headers of an AsyncOpenAI instance can be modified surreptitiously?

[chenk-gd]

Confirm this is an issue with the Python library and not an underlying OpenAI API

• This is an issue with the Python library

Describe the bug

Deploying a local LLM. Create an instance of AsyncOpenAI called aclient, then call aclient.chat.completions.create and return an asynchronous iterator. When processing the chunks yielded by the iterator, it appears that the api_key in the aclient instance is continuously changing.

To Reproduce

Set the N in user_id to increment from 1 for each request, i.e. 'user_id_1', 'user_id_2', ... 'user_id_30' ... When the number of concurrent requests is greater than a certain number, the id(aclient) changes every time the for loop is entered, and user_id_from_header and user_id_from_api_key change, and the member variables of different AsyncOpenAI instances (aclient) are mixed together: the same aclient instance, e.g. user_id_from_header='user_id_20', user_id_from_api_key='user_id_15',

Code snippets

user_id = <user_id_N>
aclient = AsyncOpenAI(api_key= user_id, default_headers= {'user_id': user_id})
stream =  await aclient.chat.completions.create(...)
async for chunk in stream:
     do_something(...)
     logger.info(f'user_id_from_header={aclient.default_headers['user_id']}, user_id_from_api_key={aclient.api_key}, id={id(aclient)}')

OS

CentOS

Python version

Python 3.11.4

Library version

openai v1.43.0

[RobertCraigie]

Thanks for the report, it looks like your code snippet doesn't fully replicate what you're describing in your comment, would you mind sharing a full runnable example snippet?

[chenk-gd]

sorry，it's a bug in my code.