Question

A media company is building a new photo-sharing platform. Users upload images that must be resized into multiple resolutions, watermarked, and stored. Upload volume is highly unpredictable — bursting from near zero to tens of thousands of images per minute during viral events. The processing of each image is CPU-intensive and takes 5-45 seconds. The company wants to minimize idle costs, avoid managing servers, and ensure no uploaded image is ever lost even if processing components fail temporarily. Which architecture best meets these requirements?

Accepted Answer

Store uploaded images in S3, send an event to an Amazon SQS queue on object creation, and configure Lambda with the SQS queue as an event source; process images in AWS Fargate tasks driven by the queue for CPU-heavy work with a dead-letter queue for failures.

Answer

Upload images directly to an Amazon S3 bucket that triggers a synchronous Lambda function via S3 event notification to process each image immediately.

Answer

Run a fixed fleet of EC2 instances behind an Application Load Balancer that receives uploads and processes images inline, using an Auto Scaling group scaled on CPU utilization.

Answer

Upload images to S3 and use S3 event notifications to invoke a single large Lambda function synchronously, storing failed events in CloudWatch Logs for manual reprocessing.