| msg355214 - (view) |
Author: Carlton Gibson (carltongibson) |
Date: 2019-10-23 12:13 |
| 6ea29c5e90dde6c240bd8e0815614b52ac307ea1 for https://bugs.python.org/issue34687 "Make asynico use ProactorEventLoop by default" This introducesd a regression in asgiref, and hence Django 3.0.x. https://github.com/django/asgiref/issues/132 https://code.djangoproject.com/ticket/30900 The error from the asgiref issue is clearest: ``` asgiref\sync.py:97: in __call__ loop_future.result() ..\..\..\AppData\Local\Programs\Python\Python38\lib\concurrent\futures\_base.py:432: in result return self.__get_result() ..\..\..\AppData\Local\Programs\Python\Python38\lib\concurrent\futures\_base.py:388: in __get_result raise self._exception ..\..\..\AppData\Local\Programs\Python\Python38\lib\concurrent\futures\thread.py:57: in run result = self.fn(*self.args, **self.kwargs) asgiref\sync.py:130: in _run_event_loop loop.close() _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ self = def close(self): if self.is_running(): raise RuntimeError("Cannot close a running event loop") if self.is_closed(): return > signal.set_wakeup_fd(-1) E ValueError: set_wakeup_fd only works in main thread ``` Off the main thread, we use a ThreadPoolExecutor and wait for the future to finish. It looks as if `ProactorEventLoop` needs to check the current thread before triggering the signal in close(). |
|
|
| msg355216 - (view) |
Author: Carlton Gibson (carltongibson) |
Date: 2019-10-23 12:22 |
| Reproduce steps, on Windows with Python 3.8.0, and a clean venv. git clone https://github.com/django/asgiref.git cd asgiref pip install -e .[tests] py.test . |
|
|
| msg355217 - (view) |
Author: Andrew Svetlov (asvetlov) *  |
Date: 2019-10-23 13:12 |
| Pull Request is welcome! |
|
|
| msg355220 - (view) |
Author: STINNER Victor (vstinner) * |
Date: 2019-10-23 14:28 |
| That's basically the same issue than bpo-34679 but on close() rather than on creating the event loop. |
|
|
| msg355222 - (view) |
Author: Carlton Gibson (carltongibson) |
Date: 2019-10-23 14:43 |
| Thanks for the speedy patch Victor. Great effort! |
|
|
| msg355224 - (view) |
Author: STINNER Victor (vstinner) * |
Date: 2019-10-23 14:46 |
| > Thanks for the speedy patch Victor. Great effort! Can you try to manually test PR 16901? |
|
|
| msg355225 - (view) |
Author: Carlton Gibson (carltongibson) |
Date: 2019-10-23 14:50 |
| OK, that seems fair :) Will do that tomorrow AM. |
|
|
| msg355231 - (view) |
Author: STINNER Victor (vstinner) * |
Date: 2019-10-23 15:25 |
| New changeset 1b53a24fb4417c764dd5933bce505f5c94249ca6 by Victor Stinner in branch 'master': bpo-34679: ProactorEventLoop only uses set_wakeup_fd() in main thread (GH-16901) https://github.com/python/cpython/commit/1b53a24fb4417c764dd5933bce505f5c94249ca6 |
|
|
| msg355234 - (view) |
Author: miss-islington (miss-islington) |
Date: 2019-10-23 15:44 |
| New changeset cbf474c98e702d12c97cd16a1e44ede10ea52b5b by Miss Skeleton (bot) in branch '3.8': bpo-34679: ProactorEventLoop only uses set_wakeup_fd() in main thread (GH-16901) https://github.com/python/cpython/commit/cbf474c98e702d12c97cd16a1e44ede10ea52b5b |
|
|
| msg355235 - (view) |
Author: STINNER Victor (vstinner) * |
Date: 2019-10-23 15:46 |
| Yury and Andrew approved my PR 16901, so I merged it. Thanks for the bug report Carlton Gibson. |
|
|
| msg355240 - (view) |
Author: Carlton Gibson (carltongibson) |
Date: 2019-10-23 16:32 |
| Super turn-around. Thank you all. |
|
|