README.md

This template demostrates the use of Cluster Proportional Autoscaler (CPA) in EKS

What is expected?

• An EKS Cluster
• Two set of resources that demostrates CPA Linear and CPA Ladder
  • cpa-ladder.tf:
    • kubernetes_config_map.nginx-ladder-autoscaler: ConfigMap for CPA Ladder
    • kubernetes_deployment.nginx-ladder-autoscaler: Deployment for CPA
    • kubernetes_deployment.nginx-ladder: Deployment for the application
  • cpa-linear.tf:
    • kubernetes_config_map.nginx-linear-autoscaler: ConfigMap for CPA Linear
    • kubernetes_deployment.nginx-linear-autoscaler: Deployment for CPA
    • kubernetes_deployment.nginx-linear: Deployment for the application

Test the autoscaling

The EKS Node Group will deploy 2 Worker Nodes with m5.large (2 vCPU and 8GB Memory) instance type, there are total 2 x 2 = 4 cores available.

Ladder

{
  "coresToReplicas":
  [
    [ 1,1 ],
    [ 3,3 ],
    [ 5,5 ],
    [ 7,7 ]
  ],
  "nodesToReplicas":
  [
    [ 1,1 ],
    [ 2,2 ]
  ]
}

Available Nodes	Available Cores	Deployment Replicas	Reason
2	4	3	[3,3] < 4 < [5,5] -> choose [3,3] = 3 replicas
3	6	5	[5,5] < 6 < [7,7] -> choose [5,5] = 5 replicas
4	8	7	[7,7] < 8 < [7,7] -> choose [7,7] = 7 replicas

Linear

{
  "coresPerReplica": 2,
  "nodesPerReplica": 1,
  "preventSinglePointFailure": true
}

Formula

replicas = max( ceil( cores * 1/coresPerReplica ) , ceil( nodes * 1/nodesPerReplica ) )
replicas = min(replicas, max)
replicas = max(replicas, min)

Available Nodes	Available Cores	Deployment Replicas	Reason
2	4	2	max ( ceil( 4 * 1/2 ) , ceil( 2 * 1/1 ) ) = max(2,2) = 2
3	6	3	max ( ceil( 6 * 1/2 ) , ceil( 3 * 1/1 ) ) = max(3,3) = 3
4	8	4	max ( ceil( 8 * 1/2 ) , ceil( 4 * 1/1 ) ) = max(4,4) = 4

Reference

https://github.com/kubernetes-sigs/cluster-proportional-autoscaler