Yes, but how do you manage fault tolerance at the machine level? If the machine is rebooted for some reason, the pending messages will be lost, and will never b

Yes, but how do you manage fault tolerance at the machine level? If the machine is rebooted for some reason, the pending messages will be lost, and will never be processed. A persistent queue is necessary for this reason. This is the reason why people use tools like Celery or RabbitMQ (which is written in Erlang by the way). I don't see how replacing Python with Erlang changes anything in this regard.