Check for existing asyncio event loops after forking

[amyreese]

It seems there are some cases where there is already an existing, active event loop after forking, which causes an exception when the child process tries to create a new event loop. It should check for an existing event loop first, and only create one if there's no active loop available.

Target function: https://github.com/jreese/aiomultiprocess/blob/master/aiomultiprocess/core.py#L93

See #4 for context.

[openalmeida]

Hi,

RuntimeError reproduced, using the code of yours from https://gist.githubusercontent.com/jreese/73e05b7e407f42bb748b406f42e3abd6

i've tried Python 3.6.0, 3.6.1, 3.6.2 all of them pure fresh and clean new installed by pyenv, and with no more operations, just clone aiomultiprocess master (ef791e9) to pwd, and run the code copied from url of above.

Results

python 3.6.1, 3.6.2 works fine, but 3.6.0 will raise RuntimeError the same as what issue #4 said.

Details

• OS: Fedora 22
• Python version: see above
• aiomultiprocess version: see above
• Can you repro on master? all on master as above
• Can you repro in a clean virtualenv? yes, but i used pyenv (an virtualenv wrapper).

I've created a toy script following your example code, but without the database parts, and it works just fine. See https://gist.github.com/jreese/73e05b7e407f42bb748b406f42e3abd6.

It sounds like your code, or one of your dependencies, is automatically creating an asyncio event loop after the process is forked, or otherwise is using as asyncio event loop that isn't friendly to forked processes. In most cases after a fork, there shouldn't be an active asyncio event loop, which is why aiomultiprocess creates one when initializing the child process. That said, it's not difficult to see if there's an existing event loop active first, I've created issue #5 to track that.