A 311 intake fleet on Auto Scaling may terminate instances during scale-in. In-flight citizen reports must not be lost when a worker dies. Which design best prevents lost work?
Select an answer to reveal the explanation.
Short Explanation
Don’t keep the only copy of a 311 report in a worker’s short-term memory. Park it on a durable queue, hide it while working, and make handlers safe to retry.
Full Explanation
When a county must process sensitive citizen images with a managed AI service but policy requires data not be used to train the provider's foundational models, architects should select AWS AI services and account configurations that support data protection commitments and opt-out or no-training postures as documented for that service. Sending images to arbitrary public internet AI endpoints without DPAs or training opt-out violates the policy intent. Storing sensitive images on public S3 websites for the model to scrape expands exposure. Disabling all encryption because the AI service is managed ignores defense in depth. Exam caveat: service capabilities and opt-out controls change—verify the current AWS documentation and BAAs for the specific AI API used. Operational check: review the service's data usage configuration for the account, process a lab image through the private pattern, confirm CloudTrail shows the approved endpoint, and document the residual training and retention posture for counsel. In municipal cloud governance, record the owners, success metrics, and rollback or restore contacts beside this technical choice so the next incident, budget review, or audit does not rediscover the decision from scratch under time pressure.