Question

An ML team runs a SageMaker Model Monitor data-quality monitoring schedule against a production endpoint. The monitoring reports are written to Amazon S3, but the operations team complains that they only discover constraint violations hours later when someone manually reviews the JSON reports. The team wants near-real-time alerting when the number of feature violations exceeds a threshold, using their existing CloudWatch dashboards and alarms with the least custom development. What should they do?

Accepted Answer

Enable emission of Model Monitor CloudWatch metrics, then create a CloudWatch alarm on the violation metric that notifies an SNS topic.

Answer

Write a Lambda function triggered by S3 PutObject events to parse each Model Monitor JSON report and publish custom metrics to CloudWatch.

Answer

Configure an S3 event notification to trigger an EventBridge rule that emails the operations team the raw JSON report on each run.

Answer

Increase the monitoring schedule frequency to hourly so violations are detected sooner in the S3 reports.