Question

An ML engineering team orchestrates their end-to-end workflow with AWS Step Functions. After a model finishes training and evaluation, company policy requires a compliance officer to manually review the evaluation report and explicitly authorize deployment to the production endpoint before the deploy step runs. The team wants to implement this pause within the existing Step Functions state machine without polling or building a custom scheduling loop. Which approach best meets this requirement?

Accepted Answer

Use a Task state with the .waitForTaskToken (callback) integration pattern that sends an approval request (e.g., via SNS or a Lambda function), then blocks until the compliance officer's action calls SendTaskSuccess or SendTaskFailure with the task token.

Answer

Insert a Wait state configured with a fixed 24-hour delay so the compliance officer has time to review the report before the deploy step automatically proceeds.

Answer

Add a Choice state that reads an S3 flag file, and have the state machine loop back to the Choice state repeatedly until the flag indicates approval.

Answer

Split the workflow into two separate state machines and have the compliance officer manually start the second (deployment) state machine from the console after review.