Python-level deadlock in debusine-server

The daphne debusine-server is occasionally "stuck" and no longer responds to anything.

According to strace all Python-level threads are calling futex, i.e. waiting for some kind of lock. The postgresql connections look idle.

Running austin reveals three kinds of threads:

1. Some are worker threads of a concurrent.futures.ThreadPoolExecutor waiting for work items. It is pretty normal for an executor to not have any work, so these probably are harmless.
2. asgiref has a CurrentThreadExecutor which also seems like a work pooling mechanism that has run out of work to do.
3. Another class of threads is trying to join other threads from asgiref's ThreadSensitiveContext.__aexit__. None of these stands out as an obvious cause.

Normally, the asyncio thread resides in EpollScheduler. When locked up, it belongs to the last class.