Question

A retail company runs an order-processing application on Amazon EC2 in an Auto Scaling group. During flash sales, the front-end web tier receives sudden traffic spikes that overwhelm the back-end processing tier, causing dropped orders and inconsistent throughput. The back-end must process every order exactly once, and orders for the same customer must be handled in the sequence they were received. The solution architect must redesign the workload to absorb spikes, prevent data loss, and scale the processing tier based on backlog. Which design best meets these requirements?

Accepted Answer

Place an Amazon SQS FIFO queue with message group IDs per customer between the tiers, and scale the back-end Auto Scaling group using a target tracking policy on the ApproximateNumberOfMessagesVisible CloudWatch metric.

Answer

Publish orders to an Amazon SNS standard topic that fans out to the back-end instances, and scale the Auto Scaling group on average CPU utilization.

Answer

Place an Amazon SQS standard queue between the tiers and scale the back-end Auto Scaling group on the queue depth, relying on visibility timeouts to prevent duplicates.

Answer

Stream orders through Amazon Kinesis Data Streams with a single shard and scale the back-end on the IteratorAge metric to preserve ordering.