Backpressure and admission control

Protect a service by making its capacity and overload behavior explicit. The goal is not to accept every request: it is to preserve useful work and bounded latency when offered load exceeds sustainable capacity.

Map the work before choosing a mechanism

Trace work from request admission through waiting, execution, downstream calls, and response or durable acknowledgement. Identify every place work can wait, including runtime mailboxes, async task creation, batch buffers, stream buffers, client pools, and broker consumers. An unbounded queue hidden behind a bounded queue is still unbounded work.

Distinguish: