A Java application runs on ECS Fargate (port 8080); JVM metrics are exposed on port 9404. You want to scale the service based on the JVM thread count with minimal operational overhead. Which two approaches meet this requirement?

The first correct option leverages the CloudWatch agent as a sidecar to scrape custom metrics directly from the application's exposed port. These metrics are then sent to CloudWatch, where alarms can be set to trigger a step-scaling policy for the Fargate service. This is a direct and efficient way to use existing AWS services for custom metric-based scaling. The second correct option utilizes Amazon Managed Service for Prometheus (AMP) with AWS Distro for OpenTelemetry (ADOT) as a sidecar. ADOT collects the JVM metrics and pushes them to AMP. Fargate's target-tracking scaling policies can directly consume metrics from AMP, providing a robust and scalable solution for monitoring and scaling based on Prometheus metrics. The incorrect option involving a metric filter from logs is less efficient and potentially less accurate for real-time thread count monitoring compared to direct metric scraping. The other incorrect option, using Prometheus rules with a step-scaling policy, is not how Fargate's scaling policies directly integrate with Prometheus metrics for scaling. Fargate scaling policies typically consume metrics directly, not rules.