DOP-C02 Practice Question 79: CloudWatch Anomaly Detection Seasonal Metric
AWS Certified DevOps Engineer – Professional (DOP-C02) · Test 8 · question 9 of 10
An e-commerce order processing service has request latency that is naturally much higher every weekday between 9 AM and 11 AM and every night during a nightly batch job, both entirely expected. A single static-threshold CloudWatch alarm on the p90 latency metric either misses real degradations during quiet periods (threshold set too high) or pages constantly during the known daily peaks (threshold set too low). The team wants one alarm on this one metric that adapts its threshold to the metric's normal daily and weekly pattern and only pages for genuine deviations.
What should they configure?
Show answer and explanation
Answer:
CloudWatch anomaly detection builds a model of a metric's expected behavior from its historical data, explicitly accounting for typical hourly, daily, and weekly cycles, and renders that model as a band. An anomaly detection alarm compares the live metric against the band's upper and lower limits (configurable independently) rather than against one fixed number, so the known morning peak and nightly batch window are already inside the expected band and only a genuine deviation from that learned pattern triggers the alarm.
Option A approximates the same idea by hand with two brittle static thresholds and a scheduler that must be kept in sync with any change to business hours or the batch job's timing; it also still misses degradations that occur close to, but not exactly on, the schedule boundaries.
Option B's composite alarm changes how many notifications fire for correlated conditions, not how the underlying threshold adapts to time-of-day; a static latency alarm inside it still needs its own correct threshold, which is the problem being solved.
Option C narrows the observation window rather than adapting to expected value, and it would miss degradations entirely outside the two hardcoded windows while still using a static, hand-picked threshold inside them.
Reference: Using CloudWatch anomaly detection