Load Spikes, Slowdowns & Retries

Start load Target QPS Client deadline (ms) Retry None Immediate Exp backoff Backoff + jitter Max retries Spike load 4× Spike latency 4× Clear queue

Shed mode None Fixed max QPS (shared) In-flight quota Max QPS In-flight quota capacity ≈ 80 ops/s

Queue strategy FIFO (oldest first) LIFO (newest first) Queue max size (0 = unbounded) Deadline propagation None (always do work) Drop if past deadline Drop if now + margin > deadline Margin (ms) Work latency Uniform Long-tail Base latency (ms)