Professional-Cloud-DevOps-Engineer Exam Dumps : Google Cloud Certified - Professional Cloud DevOps Engineer Exam

The best option for implementing the rollback actions is to update the Kubernetes Service to point to the previous Kubernetes Deployment. A Kubernetes Service is a resource that defines how to access a set of Pods. A Kubernetes Deployment is a resource that manages the creation and update of Pods. By using the blue/green deployment methodology, you can create two Deployments, one for the current version (blue) and one for the new version (green), and use a Service to switch traffic between them. If you need to rollback, you can update the Service to point to the previous Deployment (blue) and stop sending traffic to the new Deployment (green).

The best option for configuring all current and future Compute Engine instances to collect system logs and ensure that the Ops Agent remains up to date is to use the gcloud CLI to create an Agent Policy. An Agent Policy is a resource that defines how Ops Agents are installed and configured on VM instances that match certain criteria, such as labels or zones. Ops Agents are software agents that collect metrics and logs from VM instances and send them to Cloud Operations products, such as Cloud Monitoring and Cloud Logging. By creating an Agent Policy, you can ensure that all current and future VM instances that match the policy criteria will have the Ops Agent installed and updated automatically. This way, you can collect system logs from all VM instances and use them to generate dashboards in Cloud Operations.

OOMKilled means the containers exceeded their allocated memory. Since the HPA isn't scaling up (stuck at min), this indicates the issue is not scale-related but container memory limits are too low.

"Pods terminated with OOMKilled have exceeded their memory limits. You should adjust the container's resource requests and limits."

— Kubernetes OOMKill Debug Guide

Scaling the number of replicas won't help if each replica crashes. You need to increase memory limits so pods can run reliably before autoscaling can even work.