Limits argument can never get passed down?

[dmakhervaks]

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

• [X] This is an issue with the Python library

Describe the bug

It seems as though when initializing an openAI client like so:

since you cannot pass in the limits argument as a parameter, it will always default to the DEFAULT value here:

https://github.com/openai/openai-python/blob/aa6818997468b753546d55365d8142e2bb1c6640/src/openai/_base_client.py#L1127

openai.AsyncClient(())

To Reproduce

1. Initialize an openai client like so: openai.AsyncClient(())
2. Try to pass in 'limits' argument or modify the 'http_client' to have custom Limits.
3. This will not actually be propagated

Code snippets

openai.AsyncClient(())

OS

linux

Python version

3.9.12

Library version

1.50.0

[RobertCraigie]

modify the 'http_client' to have custom Limits.

I cannot reproduce this part, passing in a custom HTTP client with different limits does get propagated

import httpx
from openai import AsyncOpenAI

client = AsyncOpenAI(
    http_client=httpx.AsyncClient(
        limits=httpx.Limits(
            max_connections=1,
        )
    )
)
print(client._client._transport._pool._max_connections)

This prints 1 for me. Is it different for you?

---

Separately it is intentional that you can't pass limits directly to OpenAI(), the code you linked to is left-over from the original implementation and we haven't removed that yet.

The intended path for customizing limits is by passing your own HTTP client instance directly.