Replica of Kubernetes HPA algorithm and auto scale simulation.

K8sHpaSimulation.java

class K8sHpaSimulation {
    public static void main(String[] args) {
        final double tolerance = 0.1; // 0 < x < 1
        final double targetUtilization = 50; // 0 < x < 100
        final double requestedMetricPerPod = 40;
        final double limitMetricPerPod = 80;
        
        final int podScaleUpPolicyCount = 16;
        final int podScaleDownPolicyCount = 2;
        /* ^ Config Values ^ */

        final double scaleUpLimit = targetUtilization + (tolerance * 100) / 2;
        final double scaleDownLimit = targetUtilization - (tolerance * 100) / 2;
        
        int currentReplicas = 2;
        /* ^ Set current replica to simulate ^ */
        
        int desiredReplicas;
        double currentUtilization;
        System.out.print("Scale steps: " + currentReplicas);
        do {
            // assume all pods using 1m cpu but one of them using cpu at its limit.
            double totalMetricUsage = currentReplicas - 1 + limitMetricPerPod; 
            currentUtilization = (totalMetricUsage / (currentReplicas * requestedMetricPerPod)) * 100;
            
            if (currentUtilization > scaleUpLimit || currentUtilization < scaleDownLimit) {
                desiredReplicas = (int) Math.ceil((currentReplicas * currentUtilization) / targetUtilization);
            } else {
                desiredReplicas = currentReplicas;
            }
            
            if (desiredReplicas > currentReplicas) {
                // limit scale up speed
                currentReplicas += Math.min(desiredReplicas - currentReplicas, podScaleUpPolicyCount);
                System.out.print(" -> " + currentReplicas);
            } else if (desiredReplicas < currentReplicas) {
                // limit scale down speed
                currentReplicas -= Math.min(currentReplicas - desiredReplicas, podScaleDownPolicyCount);
                System.out.print(" -> " + currentReplicas);
            } else {
                break;
            }
        } while (targetUtilization < currentUtilization);
        System.out.println();
        
        if (targetUtilization < currentUtilization)
            System.out.println("Reason of scale stop: Tolerated.");
        else
            System.out.println("Reason of scale stop: Balanced.");

        System.out.println("Current Utilization: " + currentUtilization);
        System.out.println("Target Utilization: " + targetUtilization);
        System.out.println("Scale Up Limit: " + (scaleUpLimit >= 100 ? "No limit" : scaleUpLimit));
        System.out.println("Scale Down Limit: " + (scaleDownLimit < 0 ? "No limit" : scaleDownLimit));
        System.out.println("Desired Replicas: " + desiredReplicas);
    }
}