pytest-asyncio is an Apache2 licensed library, written in Python, for testing asyncio code with pytest.
asyncio code is usually written in the form of coroutines, which makes it slightly more difficult to test using normal testing tools. pytest-asyncio provides useful fixtures and markers to make testing easier.
@pytest.mark.asyncio
async def test_some_asyncio_code():
res = await library.do_something()
assert b'expected result' == res
pytest-asyncio has been strongly influenced by pytest-tornado.
- fixtures for creating and injecting versions of the asyncio event loop
- fixtures for injecting unused tcp ports
- pytest markers for treating tests as asyncio coroutines
- easy testing with non-default event loops
- support for async def fixtures and async generator fixtures
To install pytest-asyncio, simply:
$ pip install pytest-asyncio
This is enough for pytest to pick up pytest-asyncio.
Creates and injects a new instance of the default asyncio event loop. By
default, the loop will be closed at the end of the test (i.e. the default
fixture scope is function
).
Note that just using the event_loop
fixture won't make your test function
a coroutine. You'll need to interact with the event loop directly, using methods
like event_loop.run_until_complete
. See the pytest.mark.asyncio
marker
for treating test functions like coroutines.
Simply using this fixture will not set the generated event loop as the
default asyncio event loop, or change the asyncio event loop policy in any way.
Use pytest.mark.asyncio
for this purpose.
def test_http_client(event_loop):
url = 'http://httpbin.org/get'
resp = event_loop.run_until_complete(http_client(url))
assert b'HTTP/1.1 200 OK' in resp
This fixture can be easily overridden in any of the standard pytest locations
(e.g. directly in the test file, or in conftest.py
) to use a non-default
event loop. This will take effect even if you're using the
pytest.mark.asyncio
marker and not the event_loop
fixture directly.
@pytest.yield_fixture()
def event_loop():
loop = MyCustomLoop()
yield loop
loop.close()
If the pytest.mark.asyncio
marker is applied, a pytest hook will
ensure the produced loop is set as the default global loop.
Fixtures depending on the event_loop
fixture can expect the policy to be properly modified when they run.
Finds and yields a single unused TCP port on the localhost interface. Useful for binding temporary test servers.
A callable which returns a different unused TCP port each invocation. Useful when several unused TCP ports are required in a test.
def a_test(unused_tcp_port_factory):
port1, port2 = unused_tcp_port_factory(), unused_tcp_port_factory()
...
Asynchronous fixtures are defined just like ordinary pytest fixtures, except they should be coroutines or asynchronous generators.
@pytest.fixture
async def async_gen_fixture():
await asyncio.sleep(0.1)
yield 'a value'
@pytest.fixture(scope='module')
async def async_fixture():
return await asyncio.sleep(0.1)
All scopes are supported, but if you use a non-function scope you will need
to redefine the event_loop
fixture to have the same or broader scope.
Async fixtures need the event loop, and so must have the same or narrower scope
than the event_loop
fixture.
If you want to do this with Python 3.5, the yield
statement must be replaced with await yield_()
and the coroutine
function must be decorated with @async_generator
, like so:
from async_generator import yield_, async_generator
@pytest.fixture
@async_generator
async def async_gen_fixture():
await asyncio.sleep(0.1)
await yield_('a value')
Mark your test coroutine with this marker and pytest will execute it as an
asyncio task using the event loop provided by the event_loop
fixture. See
the introductory section for an example.
The event loop used can be overriden by overriding the event_loop
fixture
(see above).
In order to make your test code a little more concise, the pytest pytestmark
feature can be used to mark entire modules or classes with this marker.
Only test coroutines will be affected (by default, coroutines prefixed by
test_
), so, for example, fixtures are safe to define.
import asyncio
import pytest
# All test coroutines will be treated as marked.
pytestmark = pytest.mark.asyncio
async def test_example(event_loop):
"""No marker!"""
await asyncio.sleep(0, loop=event_loop)
pytest-asyncio
integrates with Hypothesis to support@given
on async test functions usingasyncio
. #102 <pytest-dev#102>- Pytest 4.1 support. #105 <pytest-dev#105>
- Python 3.7 support.
- Remove
event_loop_process_pool
fixture andpytest.mark.asyncio_process_pool
marker (see https://bugs.python.org/issue34075 for deprecation and removal details)
- Improve integration with other packages (like aiohttp) with more careful event loop handling. #64 <pytest-dev#64>
- Python versions pre-3.6 can use the async_generator library for async fixtures. #62 <pytest-dev#62>
- Support for Python versions pre-3.5 has been dropped.
pytestmark
now works on both module and class level.- The
forbid_global_loop
parameter has been removed. - Support for async and async gen fixtures has been added. #45
- The deprecation warning regarding
asyncio.async()
has been fixed. #51
- Introduced a changelog. #31
- The
event_loop
fixture is again responsible for closing itself. This makes the fixture slightly harder to correctly override, but enables other fixtures to depend on it correctly. #30 - Deal with the event loop policy by wrapping a special pytest hook,
pytest_fixture_setup
. This allows setting the policy before fixtures dependent on theevent_loop
fixture run, thus allowing them to take advantage of theforbid_global_loop
parameter. As a consequence of this, we now depend on pytest 3.0. #29
- Fix a bug preventing the propagation of exceptions from the plugin. #25
- Make
event_loop
fixtures simpler to override by closing them in the plugin, instead of directly in the fixture. #21 - Introduce the
forbid_global_loop
parameter. #21
- Support for Python 3.5
async
/await
syntax. #17
unused_tcp_port_factory
fixture. #10
Initial release.
Contributions are very welcome. Tests can be run with tox
, please ensure
the coverage at least stays the same before you submit a pull request.