A company uses an Amazon Elastic Kubernetes Service (Amazon EKS) cluster to host its machine...

The startup delay occurs because large container images must be pulled from Amazon ECR when pods are scheduled, especially on newly added nodes that do not have cached images. To consistently reduce pod startup time to seconds, container images must be prefetched directly onto the worker nodes that run the pods, not the EKS control plane.

Amazon EKS control plane nodes are fully managed by AWS and cannot be accessed or modified using AWS Systems Manager. Therefore, any solution that attempts to run Systems Manager commands on control plane nodes is invalid. Prefetching must target the EKS worker nodes (EC2 instances in managed node groups) where container images are actually stored and used.

Option C correctly creates an IAM role that allows Amazon EventBridge to invoke AWS Systems Manager to run commands on the EKS worker nodes. By using Systems Manager State Manager associations with node tags, the automation dynamically targets both existing nodes and newly added nodes that inherit the same tags. This ensures that container images are pulled immediately when a new image is pushed to ECR or when new nodes join the cluster.

Option B incorrectly targets nodes based on machine size, which is not reliable for lifecycle automation. Options A and D incorrectly reference control plane nodes, which cannot be managed with Systems Manager.

Therefore, Option C is the correct and AWS-aligned solution to ensure fast pod startup times across all nodes.