Every system has a maximum rate it can process. When arrivals exceed it, work queues up somewhere: in a message queue, in thread pools, in clients' retry loops. An unbounded queue hides the problem until memory runs out or every request waits minutes. Backpressure is the practice of letting overload push back on the sender: bounded queues, explicit rejections and clients that slow down.
Queue growth is simple arithmetic: (arrival rate - service rate) x time. Once a queue is bounded, decide what to reject first: drop low-priority work, return 429 or 503 with a retry time, and protect the requests that matter most. Retries need exponential backoff with jitter, or a brief outage turns into a retry storm that keeps the service down after it recovers.
Bound queues, shed load deliberately and keep retries from amplifying an outage.