issue_comments

Well I'm not proud of it, but that did seem to do the trick!

Trying this fix: diff diff --git a/datasette/app.py b/datasette/app.py index f2a6763a5..b17888a07 100644 --- a/datasette/app.py +++ b/datasette/app.py @@ -231,7 +231,13 @@ def __init__( self.inspect_data = inspect_data self.immutables = set(immutables or []) self.databases = collections.OrderedDict() - self._refresh_schemas_lock = asyncio.Lock() + try: + self._refresh_schemas_lock = asyncio.Lock() + except RuntimeError as rex: + if "There is no current event loop in thread" in str(rex): + loop = asyncio.new_event_loop() + asyncio.set_event_loop(loop) + self._refresh_schemas_lock = asyncio.Lock() self.crossdb = crossdb self.nolock = nolock if memory or crossdb or not self.files: This is mainly superstition at this point! I found that recipe on https://techoverflow.net/2020/10/01/how-to-fix-python-asyncio-runtimeerror-there-is-no-current-event-loop-in-thread/

The code that I control that shows up in the trace is this:

https://github.com/simonw/datasette/blob/fd1086c6867f3e3582b1eca456e4ea95f6cecf8b/datasette/app.py#L234

Removing the ensure_eventloop() hack produced the exact same test failure: https://github.com/simonw/datasette/runs/8197570109?check_suite_focus=true

I'm suspicious that this hack here may be hurting me:

https://github.com/simonw/datasette/blob/b91e17280c05bbb9cf97432081bdcea8665879f9/tests/test_cli.py#L25-L29

I added that two years ago in https://github.com/simonw/datasette/commit/3159263f05ac4baf968929d59384d9223a539071

https://docs.python.org/3/library/asyncio-task.html#asyncio.run notes that asyncio.run() was added in Python 3.7. I dropped support for 3.6 a while ago.

https://stackoverflow.com/a/56446830/6083 suggests that switching to asyncio.run() has helped other people dealing with the RuntimeError: There is no current event loop in thread error in the past.

That didn't help. This time it was 3.9 that failed first: https://github.com/simonw/datasette/runs/8197403744?check_suite_focus=true - same error though.

Tried again, this time it failed with 3.8: https://github.com/simonw/datasette/runs/8197344556?check_suite_focus=true

I'm going to try ditching the parallel test execution.

That passed too:

``` ========================================================= test session starts ========================================================== platform linux -- Python 3.7.13, pytest-7.1.3, pluggy-1.0.0 SQLite: 3.27.2 rootdir: /mnt, configfile: pytest.ini plugins: asyncio-0.19.0, anyio-3.6.1, timeout-2.1.0, xdist-2.5.0, forked-1.4.0 asyncio: mode=strict gw0 [1024] / gw1 [1024] / gw2 [1024] / gw3 [1024] / gw4 [1024] / gw5 [1024] ................................................................................................................................ [ 12%] ................................................................................................................................ [ 25%] ...............................................................................................................ss............... [ 37%] ................................................................................................................................ [ 50%] ................................................................................................................................ [ 62%] .................................................s..........................s................................................... [ 75%] ................................................................................................................................ [ 87%] ................................................................................................................................ [100%] ============================================= 1020 passed, 4 skipped in 294.43s (0:04:54) ============================================== ========================================================= test session starts ========================================================== platform linux -- Python 3.7.13, pytest-7.1.3, pluggy-1.0.0 SQLite: 3.27.2 rootdir: /mnt, configfile: pytest.ini plugins: asyncio-0.19.0, anyio-3.6.1, timeout-2.1.0, xdist-2.5.0, forked-1.4.0 asyncio: mode=strict collected 1054 items / 1024 deselected / 30 selected

tests/test_package.py . [ 3%] tests/test_cli_serve_server.py ... [ 13%] tests/test_publish_cloudrun.py ...................... [ 86%] tests/test_publish_heroku.py .... [100%]

================================================= 30 passed, 1024 deselected in 7.62s ================================================== ```

Trying this instead: docker run -it -v `pwd`:/mnt \ python:3.7-buster \ bash -c "cd /mnt && pip install -e '.[test]' && pytest -n auto -m 'not serial' && pytest -m 'serial'"

Here's how GitHub Actions runs them: https://github.com/simonw/datasette/blob/294ecd45f7801971dbeef383d0c5456ee95ab839/.github/workflows/test.yml#L27-L37

Annoyingly those tests pass just fine on my machine using that Docker recipe.

Turned that into a TIL: https://til.simonwillison.net/docker/pytest-docker

Some previous issues where I've run Python in Docker:

• https://github.com/simonw/datasette/issues/1344#issuecomment-849805778
• https://github.com/simonw/datasette/issues/1647#issuecomment-1059819628

For this one I'm going to try this image: python:3.7-buster - from https://hub.docker.com/layers/library/python/3.7-buster/images/sha256-8a79c5ddabd9e3bceb7d274e515d85a3f5898bd983bb5761ec94186bc54ae968?context=explore

docker run -it -v `pwd`:/mnt \ python:3.7-buster \ bash -c "cd /mnt && pip install -e '.[test]' && pytest"

Step one: replicate this failure on my local machine. I'll probably use Docker for this.