AsyncSingleThreadContext calls a blocking function in async context

[helmutg]

Hi,

I am observing daphne hanging in a production deployment using an older version of asgiref. In debugging it, I also looked at asgiref and suggest there being a bug (which may be different from the one causing my hang).

AsyncSingleThreadContext.aexit is an asynchronous function and therefore should not be calling any blocking code. As it happens, the executor variable happens to become an instance of ThreadPoolExecutor and its shutdown method happens to have a wait parameter defaulting to True. Therefore executor.shutdown() theoretically may block.

Indeed, that is exactly what I am observing by hooking up austin at daphne. The root cause may be something else and #492 is a plausible root cause for what I see. And while that change may make the problem at hand unobservable, the fact that blocking code is called from async context looks like a separate bug to me.

[carltongibson]

Hi @helmutg — Interesting. AsyncSingleThreadContext is very new. Can I ask what you're using it for?

I suspect the code here was just lifted from ThreadSensitiveContext, which does the same thing. (And has for years now. Happy to look at a change there, but we'd need to think very carefully about it.)

/cc @Arfey

[helmutg]

I'm sorry for having pointed at the wrong code. Your hinting at ThreadSensitiveContext is spot on. I looked for the code pattern and did not spot it was duplicated from there. Indeed it is ThreadSensitiveContext where we are observing the problem.

The fundamental problem of calling a blocking function in an async context remains the same and affects both classes in the very same way.

Unfortunately, I cannot tell much about how to get into that situation. I have not yet found a reliable reproducer.

[carltongibson]

• It's difficult to say whether we can just pass the wait=False argument without being able to see what's at stake.