pyamqp stops reading from TCP socket receive buffer

[fgebala]

pyamqp stops reading from TCP socket receive buffer

[fgebala]

Closing this issue, but leaving a comment in case our findings help anyone. We eventually solved this by moving from gevent to threads for our celery workers. For context, it seems gevent works correctly for most of our workers the majority of the time. The particular issue noted in this comment was very intermittent and very hard to reproduce consistently in our testing environment.

I was not able to track down exactly what happened here, but my best guess is that there's some type of gevent contention issue, due to the fact that each celery worker uses one TCP connection to RabbitMQ, regardless of the size of the worker pool.