A company runs several applications in the same AWS account.

Amazon CloudWatch metric streams are designed to deliver metrics in near real time to downstream services like Amazon Data Firehose with minimal configuration and management overhead. By configuring a metric stream to include all metrics, you ensure that any newly created performance or custom metrics are automatically included without needing to update filters or code when namespaces or metrics change.

Option B uses a CloudWatch metric stream → Firehose → Lambda → S3 pattern. Firehose natively supports invoking a Lambda function for in-flight transformation of the metric data before final delivery to S3. This satisfies the requirement to transform metrics before storing them and provides a fully managed, serverless pipeline with very little ongoing operational burden.

Option A mentions including metrics from “the application and the CloudWatch namespace,” which implies more manual selection and maintenance, potentially missing new metrics if not explicitly added. Options C and D work at the log level, not at the metric level, and would require building additional logic to derive metrics from logs. This is more complex and not aligned with “least operational overhead.”

Therefore, Option B is the most efficient and scalable solution.