Dead-letter on crash

[bergie]

Currently if the Node.js process segfaults, the connection is not closed. Once AMQP decides that it is dead, the message that caused a segfault is requeued. In many cases this will cause another worker to crash. Then another, until all are dead.

Would be better to dead-letter in this case.

[jonnor]

It looks like our best bet is to use the redelivered tag, and then NACK (requeue=false) the message (or route it differently).
http://stackoverflow.com/questions/17654475/consuming-not-acknowledge-messages-from-rabbitmq

However, redelivery can happen because of several reason. A channel close due to being scaled down (which is expected and happens a lot with guv) will requeue just like a crash will.

So deadlettering is basically mandatory then. So maybe we should only enable this behavior for --deadletter queues.

[bergie]

@jonnor would be good to bump this in priority