Google Professional Cloud DevOps Engineer Question 255
Select 3Google Cloud PlatformYour organization is running a web application on Google Kubernetes Engine (GKE) and has started experiencing intermittent downtime during peak traffic hours. Upon investigation, you notice that some pods are frequently being evicted, and node CPU utilization is consistently high. What steps should you take to address the infrastructure issue and ensure high availability of the application?
- A
Increase the node pool size to add more nodes to the cluster.
- B
Configure Horizontal Pod Autoscaler (HPA) to scale the pods based on CPU utilization.
- C
Increase the CPU and memory requests for the pods in the application deployment.
- D
Switch to a zonal cluster to reduce the complexity of resource management.
- E
Enable preemptible instances for the node pool to lower the cost of operations.
Show answer and explanation
Correct answers: A, B, C
Explanation
The intermittent downtime and pod evictions during peak traffic hours are due to resource constraints in the GKE cluster. To resolve this, you must increase the overall cluster capacity by adding more nodes, dynamically scale pods based on traffic patterns using HPA, and adjust CPU and memory requests to avoid scheduling issues. Choosing a zonal cluster or preemptible instances does not address the core problem and could potentially introduce new issues.
- A. Correct.
Increasing the node pool size adds more nodes to the cluster, providing more resources to handle the increasing demands during peak traffic and preventing pod evictions caused by resource constraints.
- B. Correct.
Configuring Horizontal Pod Autoscaler (HPA) allows the application to automatically scale the number of pods based on CPU utilization, ensuring that the application can handle variable traffic loads effectively.
- C. Correct.
Increasing the CPU and memory requests for the application pods ensures that Kubernetes schedules them on nodes with sufficient resources, reducing the chance of eviction during high utilization.
- D. Incorrect.
Switching to a zonal cluster does not address the resource constraints or pod eviction issues. Zonal clusters are more prone to downtime compared to regional clusters in case of zone failures.
- E. Incorrect.
Enabling preemptible instances could reduce costs, but these instances are short-lived and can be terminated at any time, potentially worsening availability issues during peak traffic hours.