Resource Monitoring Flashcards

A VerticalPodAutoscaler (VPA) is set to 'updateMode: Auto'. What does this mean for running pods?

Answer: VPA will evict and recreate pods with updated resource requests when recommendations change

In Auto mode, VPA evicts pods and allows the controller to recreate them with the updated resource requests generated by its recommendation engine.

What is the purpose of 'ephemeral-storage' as a resource type in Kubernetes pod specs?

Answer: To limit the amount of local disk space a container can use for writable layers, logs, and emptyDir volumes

Ephemeral storage requests and limits control how much local node disk space a container may consume through its writable layer, logs, and emptyDir volumes not backed by memory.

Which cAdvisor metric tracks the cumulative number of times a container has been throttled due to CPU limits?

Answer: container_cpu_cfs_throttled_periods_total

The container_cpu_cfs_throttled_periods_total metric counts the total number of CFS scheduling periods during which the container was CPU-throttled.

How does kubectl top differ from Prometheus when monitoring Kubernetes resource usage?

Answer: kubectl top shows only current instantaneous metrics from Metrics Server, while Prometheus stores historical time-series data

kubectl top fetches a single current snapshot from the Metrics Server API with no history, whereas Prometheus scrapes and stores time-series data enabling historical queries and alerting.

A pod specification sets requests.memory=128Mi and limits.memory=256Mi. What QoS class is assigned to this pod?

Answer: Burstable

Because the memory request (128Mi) does not equal the limit (256Mi), the pod is classified as Burstable rather than Guaranteed.

Which Kubernetes feature gate or component must be enabled for the HorizontalPodAutoscaler to scale based on memory utilization?

Answer: Metrics Server must be installed and the HPA must reference resource metric type 'memory'

HPA supports memory as a built-in resource metric type via the Metrics Server, so no special feature gate is needed — only Metrics Server installation and correct HPA spec configuration.

What does the Kubernetes eviction signal 'memory.available' represent on a node?

Answer: The amount of memory remaining on the node calculated as capacity minus working set bytes of all pods and system processes

The memory.available eviction signal represents total node memory minus the working set of all pods and system processes, which is what the kubelet uses to trigger eviction thresholds.