Queues, backpressure and dead letters

Why put a queue between two services instead of calling directly?

To decouple availability and rate. A direct call means the caller is down when the callee is down, and the caller's throughput is capped by the callee's. A queue lets the producer keep accepting work while the consumer is slow or restarting, absorbs spikes rather than shedding them, and lets you scale the consumer independently. What you buy it with is latency, ordering that is now your problem, and a whole second system to operate.

What is backpressure, and what happens without it?

It is the consumer telling the producer to slow down. Without it the queue grows without bound: memory or disk fills, latency through the queue climbs until the results are useless, and the failure arrives all at once instead of gradually. Real backpressure means bounded queues and an explicit decision about what to do when full — block the producer, shed the load, or drop the oldest — and that decision is a product question, not a technical one.

The failure to name: a queue that is always empty is a queue doing its job; a queue that is always growing is an outage that has not happened yet.

What is a dead letter queue for?

Messages that fail repeatedly. Without one, a single poison message is retried forever at the head of the queue and blocks everything behind it, or is dropped and silently lost. After N attempts the message is moved to a separate queue with its failure reason, the consumer moves on, and a human gets to look at the ones that actually broke.

An empty DLQ that nobody alerts on is the same as no DLQ. The alert is the feature.