Question

A financial services company builds a document-processing pipeline orchestrated by AWS Step Functions. The workflow: (1) a Lambda extracts text from an uploaded PDF, (2) a Bedrock InvokeModel call classifies the document, and (3) a second Bedrock call generates a structured summary that is written to DynamoDB. During load testing, transient Bedrock ThrottlingExceptions cause some executions to fail at step 2, and manual reruns occasionally create duplicate DynamoDB summary records. The team needs a resilient design that handles throttling gracefully and prevents duplicate writes without redesigning the entire pipeline. What should they implement?

Accepted Answer

Add Retry blocks with exponential backoff and jitter on the ThrottlingException for the Bedrock task states, and make the DynamoDB write idempotent using a deterministic key derived from the document ID with a conditional PutItem.

Answer

Convert the entire state machine to an Express workflow so throttled steps automatically replay, and enable Step Functions at-least-once delivery to guarantee the summary is written.

Answer

Move the Bedrock calls into a single Lambda function with a try/except loop that retries indefinitely until throttling stops, and write to DynamoDB at the end.

Answer

Increase the Bedrock on-demand quota via a service limit increase and add a Catch block that routes all failures to a dead-letter SQS queue for manual reprocessing.