GCP Associate Cloud Engineer Practice Question

Your team runs a web API on a Google Kubernetes Engine (GKE) Autopilot cluster. The Deployment manifest requests only 100 m CPU and 128 Mi of memory per container and omits any limits. During a sudden load increase, Cloud Monitoring shows some Pods consuming around 400 m CPU and 300 Mi memory, yet they continue to run without throttling and no other workloads are affected. What best explains this observation?

Autopilot disregards the specified requests and always allocates each Pod a fixed 0.5 vCPU and 1 GiB memory, so higher usage is expected.
Because limits were omitted, Autopilot automatically caps resource consumption at the request values, but Cloud Monitoring displays four-times-scaled metrics.
GKE's vertical pod autoscaler immediately increased the Pod's requests to match actual use, so the metrics reflect the new guaranteed values rather than excess consumption.
Autopilot permits Pods without explicit limits to burst beyond their requested CPU and memory when unused node capacity is available; throttling happens only if contention arises.

Report Issue

Answer Description

In GKE Autopilot, a Pod's resource request is the amount of CPU and memory that is guaranteed when contention exists, not a strict ceiling. If the Pod does not set explicit limits, Autopilot allows it to burst above its requested resources whenever the node has spare capacity. The scheduler still reserves only the requested amount when placing the Pod, so other workloads retain their guaranteed resources; if the node later experiences pressure, the bursting Pod will be throttled back to its request. The other choices describe behaviors that Autopilot does not perform: it does not ignore user-defined requests, Cloud Monitoring does not inflate usage figures, and Autopilot does not silently raise requests without user action.

Ask Bash

Bash is our AI bot, trained to help you pass your exam. AI Generated Content may display inaccurate information, always double-check anything important.