Where is our Kubernetes spend being wasted?

Questions CTOs ask

How do you optimize Kubernetes cluster costs?

Start by analyzing actual vs. requested resources — requested resources usually sit well above what pods actually use, and you pay for the request. Right-size resource requests and limits based on observed utilization, implement horizontal pod autoscaling for variable workloads, use cluster autoscaler to match node count to demand, and consider spot/preemptible nodes for non-critical workloads. Namespace-level resource quotas prevent teams from over-provisioning, and pod disruption budgets ensure cost optimization does not compromise availability.

What is the difference between Kubernetes requests and limits?

Requests define the minimum resources a pod needs and are used by the scheduler for placement decisions — a pod will not be scheduled on a node without sufficient unrequested resources. Limits define the maximum resources a pod can consume; exceeding memory limits causes OOM kills, while exceeding CPU limits causes throttling. Setting requests too high wastes resources (you pay for idle capacity), while setting them too low causes scheduling failures and performance issues. Best practice is to set requests based on p95 actual usage and limits at 1.5-2x requests.