Concepts

Veneer bridges the gap between Lumina’s real-time cost data and Karpenter’s provisioning decisions. It watches Prometheus for Savings Plan and Reserved Instance utilization metrics, then creates and manages NodeOverlay custom resources that adjust Karpenter’s effective instance pricing. This steers Karpenter toward cost-optimal instance types without replacing its core scheduling and bin-packing logic.

The key mechanism is the NodeOverlay CRD: each overlay targets a set of instance types via label requirements and applies a price adjustment. Karpenter uses these adjusted prices as Priority values in the AWS CreateFleet API, causing AWS to prefer the instances Veneer has identified as cost-effective.

The pages in this section cover:

• Architecture – The end-to-end data flow from Lumina through Veneer to Karpenter, the two reconciliation loops (Metrics Reconciler and NodePool Reconciler), and the overlay lifecycle.
• Instance Selection Deep Dive – How Karpenter translates NodeOverlay price adjustments into AWS CreateFleet Priority values, and how allocation strategies affect the final instance choice.
• Bin-Packing and NodeOverlay – How Karpenter’s bin-packing step can filter out instance types before NodeOverlay price adjustments take effect, and what that means for provisioning outcomes.
• Instance Preferences – How to use veneer.io/preference.N annotations on NodePools to express instance type preferences independent of cost data.