[K8s] HPA / VPA / KEDA: 자동 확장

HPA, Horizontal Pod Autoscaler, VPA, Vertical Pod Autoscaler, KEDA, Cluster Autoscaler, Karpenter

3가지 스케일링

flowchart TB
    Q[스케일링 종류]
    Q --> HPA[HPA<br/>Pod 수 ↑↓]
    Q --> VPA[VPA<br/>Pod resource ↑↓]
    Q --> CA[Cluster Autoscaler<br/>Karpenter<br/>Node 수 ↑↓]

HPA (Horizontal Pod Autoscaler)

CPU / Memory / custom metric 기반 pod 수 조정.

apiVersion: autoscaling/v2
kind: HorizontalPodAutoscaler
metadata: { name: web }
spec:
  scaleTargetRef:
    apiVersion: apps/v1
    kind: Deployment
    name: web
  minReplicas: 3
  maxReplicas: 50
  metrics:
    - type: Resource
      resource:
        name: cpu
        target: { type: Utilization, averageUtilization: 70 }
    - type: Pods
      pods:
        metric: { name: http_requests_per_second }
        target: { type: AverageValue, averageValue: "1000" }
  behavior:
    scaleUp:
      stabilizationWindowSeconds: 0
      policies:
        - type: Percent
          value: 100
          periodSeconds: 15
    scaleDown:
      stabilizationWindowSeconds: 300
      policies:
        - type: Percent
          value: 10
          periodSeconds: 60

HPA 알고리즘

desired = ceil(currentReplicas × (currentMetric / targetMetric))

예: 현재 4 pod, CPU 80%, 목표 50%:

• desired = ceil(4 × (80/50)) = 7 pod

VPA (Vertical Pod Autoscaler)

apiVersion: autoscaling.k8s.io/v1
kind: VerticalPodAutoscaler
metadata: { name: web }
spec:
  targetRef:
    apiVersion: apps/v1
    kind: Deployment
    name: web
  updatePolicy:
    updateMode: "Auto"   # Off (recommendation only) / Initial / Recreate / Auto
  resourcePolicy:
    containerPolicies:
      - containerName: '*'
        minAllowed: { cpu: 50m, memory: 64Mi }
        maxAllowed: { cpu: 2, memory: 2Gi }

KEDA (Kubernetes Event-Driven Autoscaling)

flowchart LR
    Trigger[외부 트리거<br/>Kafka lag, SQS depth, ...] --> KEDA
    KEDA --> HPA[표준 HPA 생성]
    HPA --> Deploy[Deployment scale]

apiVersion: keda.sh/v1alpha1
kind: ScaledObject
metadata: { name: kafka-consumer }
spec:
  scaleTargetRef:
    name: consumer
  minReplicaCount: 0
  maxReplicaCount: 30
  triggers:
    - type: kafka
      metadata:
        bootstrapServers: kafka:9092
        topic: events
        consumerGroup: my-group
        lagThreshold: '1000'

Cluster Autoscaler / Karpenter

flowchart LR
    HPA --> NeedMore[새 pod 들이 Pending]
    NeedMore --> CA[Cluster Autoscaler]
    CA -->|EC2 / GCE| AddNode[새 node 추가]
    AddNode --> Schedule[pending pod 스케줄]

	Cluster Autoscaler	Karpenter
종류	표준 K8s 도구	AWS 발 (이제 CNCF)
Node 선택	미리 정의된 node group	동적 instance type 선택
속도	분 단위	수십 초
비용 최적화	제한적	우수 (spot, right-sizing)
클라우드	다양	AWS, Azure

흔한 함정

관련 위키

• k8s-deployment
• k8s-pod
• prometheus (custom metric)
• kafka-consumer-group (lag-based scaling)