End-to-End Scale-Down Flow

End-to-End Scale-Down Flow A workflow diagram generated by Archify. 01 / HPA / ReplicaSet controller 02 / Cloud nodes (EC2, managed by Karpenter) 03 / On-premises nodes (Hybrid Nodes) Traffic Decrease · load drops · HPA / ReplicaSet controller Traffic Decrease load drops HPA Scale Down · replica count reduced · HPA / ReplicaSet controller HPA Scale Down replica count reduced Pod Deletion Order · compare pod-deletion-cost · HPA / ReplicaSet controller Pod Deletion Order compare pod-deletion-cost Cloud Pods Deleted First · deletion-cost=0 · Cloud nodes (EC2, managed by Karpenter) Cloud Pods Deleted First deletion-cost=0 Cloud Nodes Empty · no running pods · Cloud nodes (EC2, managed by Karpenter) Cloud Nodes Empty no running pods Karpenter Removes Empty Nodes · consolidationPolicy · Cloud nodes (EC2, managed by Karpenter) Karpenter Removes Empty Nodes consolidationPolicy On-Prem Pods Retained · deletion-cost=1000 · On-premises nodes (Hybrid Nodes) On-Prem Pods Retained deletion-cost=1000 deletion-cost=0 deletion-cost=1000 Legend HPA / pod workload Deletion order decision Cloud node / Karpenter Traffic change

Karpenter Integration

  • • pod-deletion-cost: prioritizes cloud pod removal during ReplicaSet scale-down
  • • Karpenter WhenEmpty: automatically removes empty cloud nodes
  • • Karpenter WhenUnderutilized: consolidates underutilized cloud nodes