[8.1.0] `_run_async_function` returns unawaited Future when event loop is running

[mahmoud]

Hey folks! Longtime user via pytest-recording, first time issue-filer. Big fan of v8 because of #943 , allowed me to remove a bunch of monkeypatched conftest.py code. Unfortunately today I found myself adding some more monkeypatching back in, due to some interaction with httpx via dspy/litellm.

When using VCR with libraries that use httpx/httpcore internally (like litellm/DSPy), requests fail with RuntimeError: await wasn't used with future when there's already a running event loop.

For me, this happens when:

1. Running tests with pytest (which may have an event loop)
2. Making HTTP requests through a library that uses httpx internally
3. VCR intercepts the request via httpcore stubs

Repro / Traceback

For example:

import dspy

@pytest.mark.vcr
def test_llm_call():
    lm = dspy.LM("gemini/gemini-2.0-flash", api_key="...")
    with dspy.context(lm=lm):
        result = my_dspy_module(input="test")  # Fails

And the stack:

litellm.exceptions.APIConnectionError: litellm.APIConnectionError: await wasn't used with future
Traceback (most recent call last):
  File "/opt/app/venv/lib/python3.11/site-packages/litellm/main.py", line 3113, in completion
    response = vertex_chat_completion.completion(
               ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/app/venv/lib/python3.11/site-packages/litellm/llms/vertex_ai/gemini/vertex_and_google_ai_studio_gemini.py", line 2599, in completion
    response = client.post(url=url, headers=headers, json=data, logging_obj=logging_obj)
               ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/app/venv/lib/python3.11/site-packages/litellm/llms/custom_httpx/http_handler.py", line 960, in post
    response = self.client.send(req, stream=stream)
               ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/app/venv/lib/python3.11/site-packages/httpx/_client.py", line 926, in send
    response = self._send_handling_auth(
               ^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/app/venv/lib/python3.11/site-packages/httpx/_client.py", line 1027, in _send_single_request
    response = transport.handle_request(request)
               ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/app/venv/lib/python3.11/site-packages/httpx/_transports/default.py", line 236, in handle_request
    resp = self._pool.handle_request(req)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/app/venv/lib/python3.11/site-packages/vcr/stubs/httpcore_stubs.py", line 213, in _inner_handle_request
    return _vcr_handle_request(cassette, real_handle_request, self, real_request)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/app/venv/lib/python3.11/site-packages/vcr/stubs/httpcore_stubs.py", line 195, in _vcr_handle_request
    vcr_request, vcr_response = _run_async_function(
    ^^^^^^^^^^^^^^^^^^^^^^^^^
RuntimeError: await wasn't used with future

I think this is because in vcr/stubs/httpcore_stubs.py, the _run_async_function does this:

vcrpy/vcr/stubs/httpcore_stubs.py

Lines 178 to 191 in ca4b1e1

	def _run_async_function(sync_func, *args, **kwargs):
	"""
	Safely run an asynchronous function from a synchronous context.
	Handles both cases:
	- An event loop is already running.
	- No event loop exists yet.
	"""
	try:
	asyncio.get_running_loop()
	except RuntimeError:
	return asyncio.run(sync_func(*args, **kwargs))
	else:
	# If inside a running loop, create a task and wait for it
	return asyncio.ensure_future(sync_func(*args, **kwargs))

The problem is in the else branch: asyncio.ensure_future() returns a Future object, but doesn't wait for it. The caller expects the actual result, not a Future.

Note that I know this worked with v7 + an earlier version of litellm, but can't honestly say this is for sure the root cause.

Workaround

For now I just throw a thread around it (via monkeypatch in conftest.py):

import asyncio
import threading

def _patched_run_async_function(async_func, *args, **kwargs):
    try:
        asyncio.get_running_loop()
    except RuntimeError:
        return asyncio.run(async_func(*args, **kwargs))
    else:
        # Run in separate thread with its own event loop
        result = None
        exception = None
        
        def run_in_thread():
            nonlocal result, exception
            try:
                loop = asyncio.new_event_loop()
                asyncio.set_event_loop(loop)
                try:
                    result = loop.run_until_complete(async_func(*args, **kwargs))
                finally:
                    loop.close()
                    asyncio.set_event_loop(None)
            except Exception as e:
                exception = e
        
        thread = threading.Thread(target=run_in_thread)
        thread.start()
        thread.join()
        
        if exception is not None:
            raise exception
        return result

# Apply patch
import vcr.stubs.httpcore_stubs as httpcore_stubs
httpcore_stubs._run_async_function = _patched_run_async_function

cc @jairhenrique / @seowalex bc #943

Thanks for all your hard work!

[seowalex]

Hi @mahmoud, this has been solved in #965. The bug was introduced in #943, very sorry for that.

[mahmoud]

@seowalex no worries, it happens! Overall the httpcore approach is definitely superior, so once #965 merges, we'll be in a much better place thanks to you :)

[seowalex]

This should now be fixed in https://github.com/kevin1024/vcrpy/releases/tag/v8.1.1.