Question

A media company runs a nightly job that generates SEO metadata for roughly 400,000 product listings using an Amazon Bedrock foundation model. The job must complete within a 6-hour maintenance window every night, but during business hours the same model is used only for occasional, latency-sensitive interactive requests from an internal editorial tool. The team currently uses on-demand invocation for everything and frequently hits ThrottlingExceptions during the nightly run, causing the job to overrun its window. They want the most cost-effective way to reliably complete the nightly bulk workload while keeping the daytime interactive path fast. What should they do?

Accepted Answer

Submit the nightly listings as a Bedrock batch inference job and keep on-demand invocation for the daytime interactive editorial requests.

Answer

Purchase a 1-month Provisioned Throughput commitment sized for the peak nightly volume and route all traffic through it.

Answer

Keep on-demand for everything and implement exponential backoff with jitter to eliminate the throttling during the nightly run.

Answer

Request a large service quota increase for on-demand requests-per-minute so the nightly job can burst through the throttling limits.