Throughput budgets for launch weeks

A throughput budget is a plain agreement: expected messages per minute, peak multiplier, acceptable lag minutes, and the abort rule if lag crosses that line.

Product forecasts often arrive as daily totals. Streaming capacity needs minute-level peaks. Convert the daily figure with a peak factor drawn from a prior similar launch, then add headroom for retries. If you have no prior launch, use a conservative peak factor and schedule a short load rehearsal.

Consumer concurrency, partition count, and downstream rate limits must all clear the same budget. Raising only partitions while the downstream API remains capped simply moves the queue.

Write the abort rule before launch day. Example: if lag exceeds fifteen minutes for ten consecutive minutes on the checkout stream, pause the new producer path and keep the legacy path. Ambiguous abort language is how teams argue while customers wait.

Capacity planning sessions exist to force that conversation into the open while there is still time to change the plan.

Back to field notes