langchain - 💡(How to fix) Fix openai: AzureChatOpenAI should reuse cached default httpx clients [2 comments, 2 participants]

Checked other resources

• This is a bug, not a usage question.
• I added a clear and descriptive title that summarizes this issue.
• I used the GitHub search to find similar issues.
• I am sure that this is a bug in LangChain rather than my code.
• The bug is not resolved by updating to the latest stable version of LangChain.
• I posted a self-contained, minimal, reproducible example.

Package (Required)

• langchain-openai

Description

AzureChatOpenAI does not reuse the cached default httpx clients that ChatOpenAI already uses.

In langchain_openai.chat_models.base.BaseChatOpenAI.validate_environment(), the default path uses cached helper functions for sync and async httpx clients.

However, langchain_openai.chat_models.azure.AzureChatOpenAI.validate_environment() currently passes only the explicitly supplied http_client / http_async_client. When those are unset, the OpenAI SDK creates fresh clients internally for each instance.

This creates unnecessary client churn when applications instantiate many AzureChatOpenAI objects.

Minimal Reproducible Example

import httpx
from langchain_openai import AzureChatOpenAI

original_sync_init = httpx.Client.__init__
original_async_init = httpx.AsyncClient.__init__

counter = {"sync": 0, "async": 0}

def counting_sync_init(self, *args, **kwargs):
    counter["sync"] += 1
    return original_sync_init(self, *args, **kwargs)

def counting_async_init(self, *args, **kwargs):
    counter["async"] += 1
    return original_async_init(self, *args, **kwargs)

httpx.Client.__init__ = counting_sync_init
httpx.AsyncClient.__init__ = counting_async_init

for _ in range(5):
    AzureChatOpenAI(
        api_key="test",
        azure_endpoint="https://example.openai.azure.com/",
        azure_deployment="my-deployment",
        openai_api_version="2023-05-15",
    )

print(counter)

Expected Behavior

Repeated AzureChatOpenAI(...) construction should reuse cached default sync/async httpx clients, matching ChatOpenAI behavior.

Actual Behavior

A new underlying client is created per instance when custom http_client / http_async_client are not supplied.

Proposed Fix

Use the same cached default client helpers in azure.py that BaseChatOpenAI already uses for the non-Azure OpenAI chat model.

Additional Context

I already have a local fix and regression test prepared for this.