How do we optimize Google Kubernetes Engine (GKE) for cost and performance in a DevOps setup?

All answers to this question.

To slash costs on GKE, you should first look at Spot VMs. For non-critical workloads or dev environments, Spot instances can be up to 60-91% cheaper than on-demand instances. Regarding autoscaling, the best practice is to use HPA for scaling the number of pods based on CPU/Memory and Cluster Autoscaler to manage the underlying nodes. VPA is great for finding the "right" resource requests, but avoid running it on the same metrics as HPA to prevent "thrashing." If you want to offload the headache of node management entirely, move to GKE Autopilot. It manages the infrastructure for you and charges you only for the pods you run, which naturally eliminates the "wasted node space" issue.

   Answered 2024-03-03 by Michelle Garcia

Are you using the GKE Cost Estimator and "Cost Allocation" labels to see exactly which team or project is responsible for the bulk of your cloud spend?

   Answered 2024-03-05 by David Scott

• We just started doing this, David. By tagging our namespaces (e.g., team:marketing, env:prod), we realized that our staging environment was actually costing more than production because nobody was shutting down experimental clusters! We’ve now automated a "nightly shutdown" script for all dev and staging namespaces using Google Cloud Functions. This alone is projected to save us about $2,500 every month without touching our production performance.

     Commented 2024-03-06 by Thomas Anderson

Check out the "GKE Cost Optimization Insights" in the console. It literally tells you which pods are over-provisioned and gives you a one-click fix to resize them.

   Answered 2024-03-07 by Jennifer Lopez

• I agree with Jennifer. Those native insights are surprisingly accurate. It’s the easiest way to find "low-hanging fruit" for immediate cost reductions.

     Commented 2024-03-08 by Patricia Lewis