Backpressure is a product decision

2026-11-14

Every queue has a limit. When you reach it you can drop, block, or shed by priority. Engineers tend to treat this as a tuning problem; it is usually a product problem, because each option is visible to someone.

Blocking is honest but moves the cost upstream. Dropping is cheap but silently loses work. Priority shedding is predictable only if the priority is real.

The useful exercise is to write down, per queue, what the caller experiences at 100% depth. If nobody can answer, the queue is a hidden promise.

← all notes