How to translate carrier throttling signals into queue-safe system behavior.
Retry-After is not just a retry timer. It is feedback that your system must slow down now. Honor the carrier's guidance, reduce worker throughput, and propagate backpressure to the code paths that enqueue new work. Without that propagation, the queue becomes the real outage even if the carrier is recovering.
Use proactive throttles such as token buckets for steady-state control, then use exponential backoff only for exceptional failures. Dead-letter queues belong at the end of that flow, not in place of it.