[Bug]: `timeout` is ignored in `BaseLLMAIOHTTPHandler` async path (`ClientSession.post` missing timeout arg)

[dario-fumarola]

What happened?

timeout is accepted and propagated through the async aiohttp handler path, but it is not applied on the actual network request.

In litellm/llms/custom_httpx/aiohttp_handler.py:

• _make_common_async_call(..., timeout: Union[float, httpx.Timeout], ...) receives timeout
• async_completion(...) forwards timeout into _make_common_async_call(...)
• but inside _make_common_async_call, the request is made via:

response = await async_client_session.post(
    url=api_base,
    headers=headers,
    json=data,
    data=form_data,
)

No timeout=... is passed, so caller-provided timeout is silently ignored on this path.

By contrast, sync path _make_common_sync_call(...) does pass timeout=timeout.

Why this matters

This creates async/sync inconsistency and makes timeout ineffective for providers routed through BaseLLMAIOHTTPHandler (for example aiohttp_openai path in litellm/main.py).

Minimal repro idea

Use litellm.acompletion(..., custom_llm_provider="aiohttp_openai", timeout=<very small value>) against a slow endpoint. Expected timeout behavior does not trigger from per-call timeout parameter because it is not forwarded to aiohttp post().

Expected behavior

timeout passed by user should be applied to ClientSession.post(...) in _make_common_async_call.

Suggested fix

In _make_common_async_call, pass timeout explicitly:

response = await async_client_session.post(
    url=api_base,
    headers=headers,
    json=data,
    data=form_data,
    timeout=timeout,
)

Are you a ML Ops Team?

No

What LiteLLM version are you on?

Observed on current main (as of 2026-02-23).

Twitter / LinkedIn details

No response