A multi-tenant SaaS uses shared Amazon DynamoDB tables and AWS Lambda for its application layer. Each request includes a tenant ID. The company wants to bill tenants based on their individual DynamoDB resource consumption and already has an AWS Cost and Usage Report (CUR). Which approach provides a granular per-tenant view of DynamoDB costs with the least operational overhead?

The correct approach leverages detailed per-transaction logging and programmatic cost calculation. By logging the tenant ID and consumed RCUs/WCUs for each DynamoDB interaction, you capture the granular data needed for accurate attribution. A separate Lambda can then process these logs, correlate them with the overall DynamoDB cost obtained from the AWS Cost Explorer API, and calculate per-tenant costs. This method provides high accuracy because it directly measures resource consumption per tenant. An EventBridge rule automates this process, minimizing operational overhead compared to manual analysis. Incorrect options: Tagging DynamoDB tables with tenant IDs is not feasible for shared tables; tags apply to the entire table, not individual items or transactions. Using a new per-tenant partition key for cost attribution is indirect and doesn't directly measure consumed capacity units, making cost allocation less precise. Athena queries on item usage alone won't provide RCU/WCU consumption. Publishing custom metrics to CloudWatch and using CloudWatch Logs Insights for queries is a good monitoring strategy, but combining this with the AWS Pricing Calculator for overall costs is not dynamic or accurate enough for granular, real-time cost allocation based on actual usage. The pricing calculator provides estimates, not actual billing data.