Reference

FinOps and Kubernetes glossary

The terms that come up in cloud and Kubernetes cost conversations, in plain language, with a path to measure each one in your environment.

FinOps

A practice that brings engineering, finance and business together to make cloud spend decisions with data. The focus is making cost visible, owned, and cutting waste without hurting delivery. See: FinOps maturity quiz

Showback and chargeback

Showback shows each team what it spends, without billing it. Chargeback passes that cost to the team's budget. Showback usually comes first and already changes behavior.

Cost allocation

Assigning the spend of a shared resource, such as a cluster, to teams, products or environments, usually by namespace, labels or tags.

Unit economics

Infrastructure cost divided by a business unit, such as a request, customer or transaction. It shows whether cost grows faster than usage.

Rightsizing

Adjusting CPU and memory requests (or instance sizes) to real usage. On Kubernetes, inflated requests reserve capacity nobody uses and are one of the biggest sources of waste. See: rightsizing-check

Requests and limits

A request is the CPU and memory the scheduler reserves for the pod. A limit is the usage ceiling. Cluster cost follows requests, because that is what fills the nodes.

On-Demand

The list price of an instance, with no commitment and no interruption.

Spot

Spare cloud capacity sold at a discount, in exchange for the instance being reclaimable on short notice. It suits workloads that tolerate interruption. See: Spot cost estimator

Savings Plans

A commitment to spend per hour, for 1 or 3 years, in exchange for a discount on On-Demand usage. They do not apply to Spot instances. See: Spot or Savings Plans

Reserved Instances

A reservation of an instance configuration for 1 or 3 years at a discount. It predates Savings Plans and is less flexible.

Graviton

arm64 processors designed by AWS, usually with better price per performance. They need images and dependencies that support arm64. See: graviton-readiness

Karpenter

A node autoscaler that creates instances directly from the requirements of pending pods and consolidates underused nodes. It uses NodePool and EC2NodeClass on AWS. See: Karpenter or Cluster Autoscaler

Cluster Autoscaler

A node autoscaler that grows and shrinks predefined node groups, such as Auto Scaling Groups, when pods are pending or nodes are underused.

Consolidation

Repacking pods onto fewer or smaller nodes to free capacity and cut cost. In Karpenter, it is controlled by policies and disruption budgets.

HPA and VPA

The Horizontal Pod Autoscaler changes the number of replicas with load. The Vertical Pod Autoscaler adjusts each pod's requests. They complement node autoscalers.

PodDisruptionBudget

A limit on how many pods of a service may be unavailable at once during voluntary disruptions, such as node drains or consolidation.

Topology spread

A rule that spreads replicas across zones or nodes. It protects against losing a zone, but can increase cross-zone traffic.

NAT Gateway

An AWS service that gives private subnets outbound internet access. It charges per hour and per GB processed, and becomes a hidden cost when S3 and AWS API traffic goes through it. See: EKS hidden costs calculator

Cross-AZ traffic

Data that crosses availability zones within a Region. AWS charges per GB in both directions, and the cost goes unnoticed until volume grows.

Idle EBS volumes

Unattached volumes, usually from orphaned PVCs or old nodes. They keep being billed per GB-month until deleted.

SLO and error budget

An SLO is a service's reliability target. The error budget is the failure margin it allows, and it decides how much risk, including cost optimization, the team can take.