A company migrated its application to an Amazon EKS cluster on EC2 and enabled autoscaling based on CPU. Under heavy load the application shows memory errors and does not scale out sufficiently. The company needs to collect and analyze memory metrics over time. Which combination of steps will achieve this? (Choose three.)

To collect memory metrics from EKS nodes and pods, the CloudWatch agent needs appropriate permissions and deployment. Attaching the CloudWatchAgentServerPolicy to the EC2 instance profile grants the necessary permissions for the agent running on the nodes to send metrics to CloudWatch. Deploying the unified CloudWatch agent to existing EC2 instances and baking it into the AMI for new nodes ensures comprehensive collection of performance metrics, including memory. Finally, analyzing the podmemoryutilization metric in the ContainerInsights namespace with the Service dimension allows for granular monitoring of memory usage at the application (pod) level, which is critical for identifying why the application is experiencing memory errors under load. Attaching the policy to a service account role is incorrect because the CloudWatch agent runs on the EC2 instances, not within a specific pod managed by a service account. Deploying ADOT for performance logs is incorrect because the problem specifies collecting and analyzing memory metrics, not logs. Analyzing nodememoryutilization is less effective for diagnosing application-specific memory errors, as it provides an aggregate view for the node rather than individual pod or service performance.