Event loop shutdown failures on 0.17

[zanieb]

We use the following to define a session scoped event loop (per #68)

@pytest.fixture(scope="session")
def event_loop(request):
    """
    Redefine the event loop to support session/module-scoped fixtures;
    see https://github.com/pytest-dev/pytest-asyncio/issues/68
    """
    policy = asyncio.get_event_loop_policy()
    loop = policy.new_event_loop()

    try:
        yield loop
    finally:
        loop.close()

After upgrading to 0.17, our test suite fails with

_______________________________________________________________________________________ ERROR at teardown of test_XXXX_______________________________________________________________________________________

fixturedef = <FixtureDef argname='event_loop' scope='session' baseid='tests'>, request = <SubRequest 'event_loop' for <Function test_secret_settings_are_not_serialized>>

    @pytest.hookimpl(trylast=True)
    def pytest_fixture_post_finalizer(fixturedef, request):
        """Called after fixture teardown"""
        if fixturedef.argname == "event_loop":
            policy = asyncio.get_event_loop_policy()
            # Clean up existing loop to avoid ResourceWarnings
>           policy.get_event_loop().close()

/opt/homebrew/Caskroom/miniconda/base/envs/orion-dev-38/lib/python3.8/site-packages/pytest_asyncio/plugin.py:163: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 

self = <asyncio.unix_events._UnixDefaultEventLoopPolicy object at 0x7ff8eaa8f8b0>

    def get_event_loop(self):
        """Get the event loop for the current context.
    
        Returns an instance of EventLoop or raises an exception.
        """
        if (self._local._loop is None and
                not self._local._set_called and
                isinstance(threading.current_thread(), threading._MainThread)):
            self.set_event_loop(self.new_event_loop())
    
        if self._local._loop is None:
>           raise RuntimeError('There is no current event loop in thread %r.'
                               % threading.current_thread().name)
E           RuntimeError: There is no current event loop in thread 'MainThread'.

/opt/homebrew/Caskroom/miniconda/base/envs/orion-dev-38/lib/python3.8/asyncio/events.py:639: RuntimeError

Modifying the fixture to set the loop with the policy at the end resolves the error

    try:
        yield loop
    finally:
        loop.close()

    policy.set_event_loop(loop)

Leaving the loop open or setting the loop with the policy before yielding does not resolve the error.

I'm not sure why this is happening now, not sure if it needs a fix here.

We're using asyncio_mode = auto.

[seifertm]

@madkinsz Sorry to hear the new release is causing issues. The offending line was introduced in v0.17.0.

Your fixture setup looks correct to me and I cannot reproduce the issue. What's the output when adding --setup-show to the pytest call? Could you provide a reproducible example for the error you're getting?

[zanieb]

Thanks for the quick reply.

Here's a repository with a MRE: https://github.com/madkinsz/pytest-asyncio-257

It looks like the issue is calling asyncio.run(...) in one of the tests.

e.g.

import asyncio


async def test_an_async_test_is_required():
    """Without an async test, the error will not be raised"""
    pass


def test_run_async_in_new_loop():
    async def foo():
        pass
    asyncio.run(foo())

❯ pytest . --setup-show
============================================================================================================================ test session starts =============================================================================================================================
platform darwin -- Python 3.8.12, pytest-6.2.5, py-1.11.0, pluggy-1.0.0
rootdir: /Users/mz/dev/contrib/pytest-asyncio-257, configfile: setup.cfg
plugins: env-0.6.2, flaky-3.7.0, anyio-3.4.0, xdist-2.4.0, asyncio-0.17.0, forked-1.3.0
collected 2 items                                                                                                                                                                                                                                                            

test_bug.py 
SETUP    S event_loop
        test_bug.py::test_an_async_test_is_required (fixtures used: event_loop).
        test_bug.py::test_run_async_in_new_loop.
TEARDOWN S event_loopE

=================================================================================================================================== ERRORS ===================================================================================================================================
______________________________________________________________________________________________________________ ERROR at teardown of test_run_async_in_new_loop _______________________________________________________________________________________________________________

fixturedef = <FixtureDef argname='event_loop' scope='session' baseid=''>, request = <SubRequest 'event_loop' for <Function test_an_async_test_is_required>>

    @pytest.hookimpl(trylast=True)
    def pytest_fixture_post_finalizer(fixturedef, request):
        """Called after fixture teardown"""
        if fixturedef.argname == "event_loop":
            policy = asyncio.get_event_loop_policy()
            # Clean up existing loop to avoid ResourceWarnings
>           policy.get_event_loop().close()

/opt/homebrew/Caskroom/miniconda/base/envs/orion-dev-38/lib/python3.8/site-packages/pytest_asyncio/plugin.py:163: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 

self = <asyncio.unix_events._UnixDefaultEventLoopPolicy object at 0x7f934806a8b0>

    def get_event_loop(self):
        """Get the event loop for the current context.
    
        Returns an instance of EventLoop or raises an exception.
        """
        if (self._local._loop is None and
                not self._local._set_called and
                isinstance(threading.current_thread(), threading._MainThread)):
            self.set_event_loop(self.new_event_loop())
    
        if self._local._loop is None:
>           raise RuntimeError('There is no current event loop in thread %r.'
                               % threading.current_thread().name)
E           RuntimeError: There is no current event loop in thread 'MainThread'.

/opt/homebrew/Caskroom/miniconda/base/envs/orion-dev-38/lib/python3.8/asyncio/events.py:639: RuntimeError
========================================================================================================================== short test summary info ===========================================================================================================================
ERROR test_bug.py::test_run_async_in_new_loop - RuntimeError: There is no current event loop in thread 'MainThread'.
========================================================================================================================= 2 passed, 1 error in 0.10s =========================================================================================================================
 

[seifertm]

Great example, thanks! It seems that asyncio.run replaces the event loop from the fixture. When asyncio.run finishes, it sets the current event loop to None. This causes the fixture teardown to fail.

[asvetlov]

Correct. Session-scoped event_loop fixture doesn't work well with asyncio.run(). I can imagine other use-cases when session-scoped loop can cause problems.
Should pytest-asyncio solve them? This is the question.
Dark grey zone is dangerous by the definition, dragons live here.

[zanieb]

Of course asyncio clobbers the event loop 😮‍💨

Interestingly though, this worked fine in the previous version and all of our tests are working as intended. Perhaps if there's not an event loop pytest-asyncio should just pass during teardown?

[zanieb]

In https://github.com/madkinsz/pytest-asyncio-257/blob/main/test_event_loop.py I show that the session scoped loop is not actually clobbered during the test run and async tests share the same loop correctly. I'm definitely not familiar with the implementation details of pytest, but it seems like the behavior is correct and the only issue is that the loop is not present when fixture teardown occurs.

[eirikur-grid]

I am seeing something similar with 0.17.2. This happens after the last test has been run.

____________________________________________________ ERROR at teardown of test_get_mimetype_from_filename_raises_error[empty string] _____________________________________________________
.venv/lib/python3.9/site-packages/pytest_asyncio/plugin.py:362: in finalizer
    loop.run_until_complete(async_finalizer())
../../.pyenv/versions/3.9.1/lib/python3.9/asyncio/base_events.py:617: in run_until_complete
    self._check_closed()
../../.pyenv/versions/3.9.1/lib/python3.9/asyncio/base_events.py:510: in _check_closed
    raise RuntimeError('Event loop is closed')
E   RuntimeError: Event loop is closed

We also override the event_loop fixture like so:

# Override the pytest-asyncio event_loop fixture to make it session scoped. This is required in order to enable
# async test fixtures with a session scope. More info: https://github.com/pytest-dev/pytest-asyncio/issues/68
@pytest.fixture(scope="session")
def event_loop(request):
    loop = asyncio.get_event_loop_policy().new_event_loop()
    yield loop
    loop.close()

[eirikur-grid]

OK, so I found out how to fix this. A our folder structure for the tests is as follows

/tests
  /integration
    conftest.py
    ...
  /unit
    conftest.py
    ...
  conftest.py

Our def event_loop fixture was located under /tests/integration/conftest.py. Moving it to /tests/conftest.py solved the issue.

[MetRonnie]

I have a similar issue - getting RuntimeError: There is no current event loop in thread 'MainThread'. But in my case it seems to be caused by a conflict with pytest-tornasync. Not sure why considering the project I'm running the tests in doesn't use it. But it happened after I removed @pytest.mark.asyncio decorators when using asyncio_mode = auto

eukaryote/pytest-tornasync#13

[seifertm]

@MetRonnie After a brief look at pytest-tornasync it seems the plugin has a similar behaviour to pytest-asyncio with async_mode=auto: It tries to run all tests that are coroutines in a tornado.IoLoop. When both plugins try to do the same thing, unexpected things happen.

If your project is using both pytest-asyncio and pytest-tornasync, you should stick with asyncio_mode=strict.