A customer-facing image-storage service will receive batches of large images from millions of users, resize them, and keep them in Amazon S3 for up to 6 months. Demand varies widely. The solution must be enterprise-scale, reliable, and allow failed processing jobs to be retried. Which design is the most cost-effective while meeting these requirements?

The correct option uses Amazon SQS for reliable, asynchronous processing of S3 object-created events. SQS decouples the image upload from processing, providing a buffer for varying demand and enabling retries of failed Lambda invocations, which is crucial for enterprise-scale reliability. Storing resized objects in S3 Standard-IA is cost-effective for data accessed less frequently but still requiring rapid access, and transitioning to S3 Glacier Deep Archive after 6 months minimizes long-term storage costs. The other options are less suitable: Using AWS Step Functions or Amazon EventBridge directly to trigger a Lambda that resizes in place and replaces the original is problematic. If the Lambda fails, the original image might be lost or corrupted, and retries are harder to manage without a queue. Resizing in place and replacing the original is generally not a good practice for reliability and auditability. Retaining the original in S3 and transitioning to S3 Standard-IA after 6 months might be more expensive if the original is not needed after resizing, and doesn't address the reliability of processing. The question asks to keep the resized images for 6 months.