GKE-hosted stateless web frontend requires high availability; cluster auto-upgrades will drain nodes. Ensure the application retains serving capacity equal to 10% of Pods before a drain. What do you do?

The correct answer ensures that even during a node drain, there are enough Pods to maintain 10% of the serving capacity. If the cluster auto-upgrades drain nodes, Pods on those nodes will be evicted. By setting the replica count 10% higher than needed for normal operation, you create a buffer. When a drain occurs, these extra Pods can absorb the impact, allowing the application to continue serving traffic at the desired minimum capacity. Incorrect options: VPA adjusts resource requests/limits, not the number of running Pods, so it won't help maintain serving capacity during a drain. A PDB with minAvailable: 10% would ensure that at least 10% of the Pods remain available during voluntary disruptions, but it doesn't proactively add extra Pods to handle the drain while maintaining the original serving capacity. It only guarantees a minimum remaining percentage. Increasing HPA's maxReplicas only sets an upper limit for scaling; it doesn't guarantee a specific number of running Pods or add extra Pods to handle drains.