사용자 지정 지표와 CloudWatch 알람

Lambda publishes built-in metrics like invocations and errors, but your app has its own signals, such as orders processed or items failed. These are custom metrics.

The easiest way to emit metrics is the Embedded Metric Format (EMF): print structured JSON to logs and CloudWatch extracts metrics automatically.

Build the EMF object and print it as JSON. No extra API call or latency is needed.

Alternatively call put_metric_data directly. This is synchronous and adds latency, so EMF is usually preferred for hot paths.

Dimensions are key-value tags that slice a metric, such as by region or customer tier. Keep dimension cardinality low to control cost.

A CloudWatch alarm watches a metric and changes state when it breaches a threshold for a set number of periods, then triggers an action.

This alarm fires when the error count exceeds 5 in a single one-minute period.

Point the alarm action at an SNS topic so it emails, texts, or pages your on-call when triggered.

Alarms have three states: OK, ALARM, and INSUFFICIENT_DATA. The last means the metric had no data points in the window.

Too many noisy alarms train people to ignore them. Alarm on symptoms users feel, like error rate and latency, not every minor metric.

Build a CloudWatch dashboard combining built-in and custom metrics so you see the whole system health at a glance during an incident.

Test your metrics knowledge.

You learned to emit custom metrics via EMF or PutMetricData, add dimensions, create threshold alarms wired to SNS, and avoid alarm fatigue with dashboards.