Question

A developer maintains a REST API backed by Lambda and DynamoDB. Support tickets report that the API feels slow for some users, but the CloudWatch alarm monitoring the Average of the API Gateway Latency metric never triggers because the mean stays low. The team wants an alarm that reliably catches the intermittent slowness affecting a minority of requests. What is the most effective change?

Accepted Answer

Create the alarm on the p99 percentile statistic of the Latency metric instead of the Average.

Answer

Increase the alarm's evaluation period so the Average is calculated over a longer window.

Answer

Switch the alarm to use the Sum statistic of the Latency metric.

Answer

Lower the alarm threshold on the Average statistic to a smaller millisecond value.