Field note
Reading peak traffic without drowning in charts
Peak periods invite more graphs, not better judgment. Before adding another panel, decide which questions the busy window must answer: Did demand arrive where you expected? Did dependencies absorb the load? Did scaling actions finish in time?
Start with a narrow set of signals tied to user outcomes—request rate by critical path, saturation on the bottleneck resource, and error or timeout rates that customers would notice. Secondary metrics belong on a second pass once the primary story is clear.
Compare the current peak against one or two prior windows with similar marketing or calendar drivers. Absolute numbers alone rarely explain behavior; relative change and timing against scale events do. Note when autoscaling lagged demand by more than your recovery budget.
Write a short narrative after the peak while memory is fresh. Record what the workload did, what you changed mid-event, and which monitors would have shortened the scramble. That note becomes the seed for your next readiness review.