AMQP is so nice from an engineer's perspective. If shit goes south somewhere in the system, I get a pagerduty alerting me that a deferred queue has exceeded so

AMQP is so nice from an engineer's perspective. If shit goes south somewhere in the system, I get a pagerduty alerting me that a deferred queue has exceeded some depth and I can start working to resolve the issue (if a message fails to be processed it is sent to the deferred queue). Once the issue is resolved, all the messages are still waiting in the deferred queue and will get processed as soon as they are requeued, and I don't have to replay nginx logs. Anything that can be asynchronous, make asynchronous and save yourself the headache of dealing with 500s when a component of your distributed system fails.