Bitzu

Associate Cloud Engineer ยท Operating GKE clusters

How do resource requests work in GKE Autopilot?

How Autopilot bills by Pod requests, default and minimum requests, adjusting requests, and why limits are set equal to requests.

Key points

  • Autopilot bills by the requests of your Pods, so right-sizing requests controls cost.
  • Without requests, general-purpose Pods get defaults of 0.5 vCPU and 2 GiB memory; requests below the class minimum or outside the CPU:memory ratio are raised.
  • Setting requests equal to limits gives Guaranteed QoS, and a limit without a request is copied into the request.

Source: https://docs.cloud.google.com/kubernetes-engine/docs/concepts/autopilot-resource-requests

Practice questions

Single choice

In GKE Autopilot, what is the billing for your workloads primarily based on?

  1. The resources requested by your Pods
  2. The number of nodes you provision
  3. The machine type you choose per node pool
  4. The number of Services in the cluster
Show answer

Answer: A. The resources requested by your Pods

Autopilot charges for the resources your Pods request. Node count and VM machine types are managed by Google and are not what you pay by.

Select all that apply

Which statements about how Autopilot treats requests and limits are true?

  1. Billing follows the Pod's effective requests
  2. A limit with no request is copied into the request
  3. A request larger than its limit is lowered to the limit
  4. You are billed per provisioned node regardless of Pods
  5. Ephemeral storage limits may differ freely from requests
Show answer

Answer: A. Billing follows the Pod's effective requests, B. A limit with no request is copied into the request, C. A request larger than its limit is lowered to the limit

A limit without a request is copied into the request, and requests above limits are lowered to the limits. Billing follows requests, not node count, and ephemeral storage limits must equal requests.

True or false

In Autopilot you are billed for the actual CPU a Pod uses, not for what it requests.

Show answer

Answer: False

False. Billing follows the Pod's resource requests, so over-requesting costs money even if the Pod idles.

Practice 8 questions on this topic in Bitzu

Short lessons, instant explanations and spaced review of what you get wrong. Free on iOS and Android.

Download on the App StoreGet it on Google Play

More Associate Cloud Engineer topics in this module