Leadership asks you to cut the cloud bill for your Kubernetes platform by 30 percent without hurting reliability. How do you approach it?

What the interviewer wants to hear

• Get visibility: cost allocation per team and namespace with labels
• Right-size requests using real usage and VPA recommendations
• Improve bin packing and autoscaling with Karpenter or cluster autoscaler
• Use spot for stateless workloads and commitments for the baseline
• Cut waste: idle environments, orphaned volumes, NAT and data transfer, log volume
• Guard reliability with SLOs and make cost part of regular reviews

Model answer

I start with visibility, because you cannot cut what you cannot attribute. Using a tool like OpenCost or Kubecost plus cloud tags, I break spend down by team, namespace and service, and share it. The biggest lever is usually over-requested Pods: compare requests with real usage, apply VPA recommendations, and since the scheduler packs by requests, this frees whole nodes. Then I improve node efficiency with Karpenter or cluster autoscaler, consolidating underused nodes and picking cheaper instance types, including ARM where images support it. Stateless, replicated workloads move to diversified spot capacity, and the steady baseline is covered by savings plans or reserved capacity. Next is waste: dev environments that run overnight, orphaned volumes and snapshots, NAT and cross-zone traffic, and log ingestion. Throughout, I watch SLOs and error budgets so savings never come from reliability, and I make cost a monthly review metric with owners.

More hard DevOps Engineer questions

• You are asked to build an internal platform so 30 product teams can deploy services to Kubernetes on their own. How would you design it?
• How would you secure the software supply chain from source code to a running container in production?
• Your company wants to run a critical service active-active in two cloud regions. What are the main DevOps challenges and how would you handle them?
• Your DORA metrics show a 30% change failure rate and long recovery times. How would you bring them down without slowing teams down?
• How do you plan and carry out a Kubernetes version upgrade for production clusters with minimal risk?
• You are the incident commander for a major outage affecting all customers. Describe how you run the incident from first page to postmortem.
• Design CI/CD for a monorepo with 60 services and 200 engineers where full builds take over an hour.
• Design a disaster recovery plan for a platform whose business requires an RTO of 1 hour and an RPO of 5 minutes for the loss of a whole cloud region.
• Several teams will share one Kubernetes cluster. How would you isolate tenants and secure the cluster?