Module 1 · Throughput under limits · Lesson 4
Backpressure, load shedding and batching
What to do when demand exceeds capacity for hours, not seconds. Compute backlog and drain time, order work by criticality with aging, admit or shed at the edge, push back on producers, and move deferrable work into batches.
By the end of this lesson you can
- Compute backlog growth, drain time and FIFO waiting time for a load spike from arrival and service rates
- Design priority classes for an agent fleet, with aging to prevent starvation, and quantify their effect on critical work
- Implement admission control that sheds or defers work whose expected wait exceeds its deadline
- Propagate backpressure upstream to webhooks, queues and sweeps instead of buffering without bound
- Decide when to micro-batch items into one call and when to move work to the Message Batches API, including the data-handling trade-off