Question

A developer needs to publish custom application metrics (such as order value and items-per-order) from a high-throughput Lambda function. The function already writes structured JSON logs to CloudWatch Logs. Calling the PutMetricData API on every invocation has caused throttling and added latency to the function. What is the MOST efficient way to emit these custom metrics while avoiding the API throttling and extra latency?

Accepted Answer

Write log entries in the CloudWatch embedded metric format (EMF) so CloudWatch automatically extracts the values as custom metrics from the logs.

Answer

Batch the metrics in memory and call PutMetricData once per minute using a background thread inside the Lambda function.

Answer

Enable Lambda enhanced monitoring so the order value and items-per-order are captured as built-in metrics.

Answer

Create a CloudWatch metric filter on the log group that increments a metric each time the function logs an order event.