Kuberneteskubernetesautoscalinghpascalingproduction-ready

Horizontal Pod Autoscaler

Kubernetes HPA configuration with CPU/memory metrics, custom metrics, scaling behavior tuning, and stabilization windows.


title: "Horizontal Pod Autoscaler" description: "Kubernetes HPA configuration with CPU/memory metrics, custom metrics, scaling behavior tuning, and stabilization windows." category: "Kubernetes" language: "yaml" tags: ["kubernetes", "autoscaling", "hpa", "scaling", "production-ready"] createdAt: "2025-01-28" updatedAt: "2025-02-18"

Overview

Kubernetes Horizontal Pod Autoscaler with multiple scaling metrics (CPU, memory, custom metrics, external metrics) and fine-tuned scaling behavior. Includes stabilization windows to prevent thrashing and asymmetric scale-up/scale-down policies.

Code

behavior:
  scaleUp:
    stabilizationWindowSeconds: 30
    policies:
      - type: Percent
        value: 100
        periodSeconds: 30
  scaleDown:
    stabilizationWindowSeconds: 300

Notes

  • stabilizationWindowSeconds prevents rapid oscillation by requiring metrics to be sustained
  • Scale-up is aggressive (double in 30s) while scale-down is conservative (10% per minute)
  • selectPolicy: Max for scale-up picks the policy that adds the most pods
  • selectPolicy: Min for scale-down picks the policy that removes the fewest pods
  • Resource requests on pods are required for CPU/memory-based HPA to calculate utilization
  • Custom metrics require a metrics adapter (Prometheus adapter, Datadog, etc.)