Question

A team runs a high-throughput payment service on AWS Lambda. They need per-customer latency and error metrics for thousands of customers, but calling PutMetricData for each customer on every invocation is causing throttling and increasing invocation duration. They want custom metrics generated asynchronously without adding synchronous API calls in the hot path, while keeping the raw structured log data queryable. Which approach best meets these requirements?

Accepted Answer

Emit logs using the CloudWatch Embedded Metric Format (EMF) to stdout so CloudWatch automatically extracts custom metrics from the structured log entries while preserving the raw logs.

Answer

Batch all customer measurements in memory and call PutMetricData once per invocation with multiple MetricData entries to reduce the number of API calls.

Answer

Write the measurements to a CloudWatch Logs log group and create a metric filter per customer to convert matching log events into custom metrics.

Answer

Send the measurements to Amazon Kinesis Data Firehose and use a Lambda transform to call PutMetricData in batches downstream.