A retail ordering application currently uses a load-balanced EC2 fleet for web hosting, database API services, and business logic. The company wants to refactor to a decoupled, scalable architecture that retains failed orders for later processing while minimizing operational overhead. Which architecture best meets these goals?

The correct architecture leverages Amazon S3 for static website hosting, which is highly scalable and cost-effective. AWS AppSync provides a managed GraphQL API for database interactions, simplifying data access. Amazon SQS offers a fully managed message queue for order processing, ensuring decoupling and scalability. AWS Lambda executes business logic in a serverless manner, minimizing operational overhead. Crucially, an Amazon SQS dead-letter queue (DLQ) automatically retains failed orders, allowing for later processing without additional custom code or infrastructure. The other options are less optimal: Using Amazon ECS for business logic with SQS long polling for failed orders is less serverless than Lambda and requires more operational management. AWS Elastic Beanstalk and Amazon MQ introduce more operational overhead than S3/Lambda/SQS. Storing failed orders in S3 Glacier Deep Archive is for long-term archival, not immediate reprocessing. Amazon Lightsail is a simpler, less scalable option than S3 for web hosting, and Amazon SES is an email service, not an order queuing service. Amazon OpenSearch Service is for search and analytics, not primary retention of failed orders.