An EKS cluster hosts an ML application whose model and container images have grown, causing pod startup times of minutes. The engineer must reduce startup times to seconds, including for pods scheduled on newly launched nodes. They created an EventBridge rule invoking an SSM Automation that prefetches images from ECR when new images are pushed and tag the cluster and node groups. What is the next step to meet the requirement?

The correct answer involves creating an IAM role for EventBridge to use Systems Manager (SSM) to run commands on the EKS worker nodes, and then creating an SSM State Manager association that uses node tags to prefetch images. This is because image prefetching needs to occur on the worker nodes where pods are scheduled, not the control plane nodes. The State Manager association, combined with node tags, ensures that newly launched nodes (which would also have the specified tags) automatically receive the prefetching commands, thereby reducing startup times to seconds. Incorrect options: Running commands on EKS control plane nodes is incorrect because image prefetching is a worker node task. Using machine-size attributes for the SSM State Manager association is less flexible and less precise than using node tags, which can be applied consistently across all relevant worker nodes.