Running Cluster Autoscaler and Karpenter side by side on ROSA with HCP

This article demonstrates how to run Cluster Autoscaler and Karpenter node autoscalers simultaneously on ROSA with HCP, using node labels to route workloads to their preferred provisioning strategy.

• Autoscaling operates at two tiers: pod-tier (HPA, KEDA) and node-tier (Cluster Autoscaler, Karpenter)
• Node autoscalers coexist using separate "lanes" labeled with demo-scaler values identifying their provisioner
• Cluster Autoscaler manages predefined pools with fixed instance types and warm minimums
• Karpenter provisions right-sized nodes on-demand and consolidates to zero when idle
• Workloads select their autoscaler via nodeSelector matching the lane label
• ResourceQuota and NodePool CPU limits prevent runaway scaling and control costs
• Both strategies can run simultaneously, each optimized for different workload patterns

By combining pod and node autoscaling tiers with dual node provisioners, teams can optimize for both uniform and heterogeneous workloads while maintaining cost controls.