HPA, VPA, KEDA — K8s Autoscaling Tam Rehber

"3 farklı autoscaler, 3 farklı niche. HPA: pod sayısı CPU'ya göre. VPA: pod resource'u kullanıma göre. KEDA: pod sayısı event'e göre (queue, cron, custom metric). Hangisini ne zaman?"

Bu rehber Kubernetes'in 3 ana autoscaler'ını karşılaştırır, hangi senaryoda hangisinin tercih edileceğini, ve birlikte çalışma pattern'lerini anlatır.

---

⚖️ 3 Autoscaler

Tool	Ne	Yöntem
HPA	Horizontal Pod Autoscaler	Pod sayısı ↑↓ (CPU/Memory/custom metric)
VPA	Vertical Pod Autoscaler	Pod resource (CPU/Memory request) ↑↓
KEDA	Kubernetes Event-Driven Autoscaler	Pod sayısı, dış event'lere göre

---

🎯 HPA — Horizontal Pod Autoscaler

Pod sayısını yatay ölçekler (örn: 3 → 10 → 3).

Basit (CPU)

apiVersion: autoscaling/v2
kind: HorizontalPodAutoscaler
metadata:
  name: payments-hpa
spec:
  scaleTargetRef:
    apiVersion: apps/v1
    kind: Deployment
    name: payments
  minReplicas: 3
  maxReplicas: 20
  metrics:
    - type: Resource
      resource:
        name: cpu
        target:
          type: Utilization
          averageUtilization: 70   # CPU %70 hedef

Multi-metric (CPU + Memory + Custom)

metrics:
  - type: Resource
    resource:
      name: cpu
      target: {type: Utilization, averageUtilization: 70}
  - type: Resource
    resource:
      name: memory
      target: {type: Utilization, averageUtilization: 80}
  - type: Pods
    pods:
      metric:
        name: http_requests_per_second
      target:
        type: AverageValue
        averageValue: 100

Behavior tuning (scale up/down hızı)

behavior:
  scaleUp:
    stabilizationWindowSeconds: 0   # hızlı scale up
    policies:
      - type: Percent
        value: 100   # max 2x kat
        periodSeconds: 60
      - type: Pods
        value: 4     # max 4 pod ekle
        periodSeconds: 60
    selectPolicy: Max
  scaleDown:
    stabilizationWindowSeconds: 300   # 5 dk bekle
    policies:
      - type: Percent
        value: 25
        periodSeconds: 60

🔑 Scale up hızlı, scale down yavaş → flap engelleme.

Custom metric (Prometheus)

# prometheus-adapter
- seriesQuery: 'http_requests_total{namespace!="",pod!=""}'
  resources:
    overrides:
      namespace: {resource: "namespace"}
      pod: {resource: "pod"}
  name: {matches: "^(.*)_total$", as: "${1}_per_second"}
  metricsQuery: sum(rate(<<.Series>>{<<.LabelMatchers>>}[2m])) by (<<.GroupBy>>)

# HPA custom metric kullanır
- type: Pods
  pods:
    metric: {name: http_requests_per_second}
    target: {type: AverageValue, averageValue: 100}

---

📦 VPA — Vertical Pod Autoscaler

Pod resource (CPU/Memory) request/limit ayarlar.

Modes

Mode	Davranış
Off	Sadece öneri, resource değişmez (recommendation mode)
Initial	Pod oluşturulurken resource set, sonra değişmez
Auto	Pod yeniden başlatarak resource günceller
Recreate	Eski pod kill, yeni resource'la yarat

Manifest

apiVersion: autoscaling.k8s.io/v1
kind: VerticalPodAutoscaler
metadata:
  name: payments-vpa
spec:
  targetRef:
    apiVersion: apps/v1
    kind: Deployment
    name: payments
  updatePolicy:
    updateMode: "Auto"
  resourcePolicy:
    containerPolicies:
      - containerName: payments
        minAllowed: {cpu: 100m, memory: 128Mi}
        maxAllowed: {cpu: 2000m, memory: 4Gi}
        controlledResources: [cpu, memory]

Off mode (recommendation only)

spec:
  updatePolicy:
    updateMode: "Off"

# Önerileri gör
kubectl describe vpa payments-vpa
# Recommendation: cpu=234m memory=512Mi

→ İnsan review eder, deployment'a manuel uygular.

---

⚠️ HPA + VPA Birlikte

HPA: pod sayısı (yatay)
VPA: pod resource (dikey)
   ↑
   └── Aynı resource (CPU) için ikisini birlikte kullanma!
       İhtilaf riski.

✅ İYİ: HPA CPU + VPA Memory
✅ İYİ: VPA Off mode + HPA prod'da
❌ KÖTÜ: HPA Auto + VPA Auto aynı resource

🔑 Production önerisi: HPA aktif (pod sayısı), VPA Off mode (öneri only) → manuel review + deployment'a uygula.

---

⚡ KEDA — Event-Driven Autoscaler

HPA'yı "event source"larla genişletir. Queue depth, cron, AWS SQS, Kafka, Prometheus, vb.

Kurulum

helm install keda kedacore/keda \
  -n keda --create-namespace

ScaledObject — SQS queue

apiVersion: keda.sh/v1alpha1
kind: ScaledObject
metadata:
  name: queue-worker-scaler
spec:
  scaleTargetRef:
    name: queue-worker
  minReplicaCount: 0
  maxReplicaCount: 50
  pollingInterval: 30   # 30 saniyede bir queue check
  cooldownPeriod: 300   # 5 dk idle sonrası scale down
  triggers:
    - type: aws-sqs-queue
      metadata:
        queueURL: <SQS_URL>
        queueLength: '10'
        awsRegion: <REGION>
        identityOwner: pod

Kafka consumer

triggers:
  - type: kafka
    metadata:
      bootstrapServers: <KAFKA_BROKER>:9092
      consumerGroup: payments-group
      topic: orders
      lagThreshold: '50'

Cron (predictable scale)

triggers:
  - type: cron
    metadata:
      timezone: Europe/Istanbul
      start: 0 9 * * 1-5     # weekday 09:00 → scale up
      end: 0 18 * * 1-5      # 18:00 → scale down
      desiredReplicas: '20'

Prometheus custom metric

triggers:
  - type: prometheus
    metadata:
      serverAddress: http://prometheus.<NS>.svc:9090
      metricName: http_requests_per_second
      threshold: '100'
      query: |
        sum(rate(http_requests_total{service="payments"}[2m]))

Multi-trigger

triggers:
  - type: prometheus
    metadata: {...}
  - type: cron
    metadata: {...}
  - type: aws-sqs-queue
    metadata: {...}
# → herhangi biri tetiklendiğinde scale

Scale to zero

minReplicaCount: 0   # idle'da hiç pod yok

→ Queue boşken 0 pod (cost tasarruf). Mesaj gelince hızla scale.

---

🌳 Karar Ağacı

START
  │
  ├── Sürekli HTTP traffic (req/sec ile bağlı)?
  │     │
  │     └── HPA (CPU + custom: http_requests_per_second)
  │
  ├── Background queue worker (SQS, Kafka, RabbitMQ)?
  │     │
  │     └── KEDA + queue trigger
  │
  ├── Predictable load (business hours)?
  │     │
  │     └── KEDA + cron
  │
  ├── Idle period uzun (saatler boşta)?
  │     │
  │     └── KEDA + scale-to-zero
  │
  ├── Pod resource yanlış set, sürekli OOM/idle?
  │     │
  │     └── VPA recommendation mode + manuel uygula
  │
  └── Multi-source event (queue + traffic + schedule)?
         │
         └── KEDA multi-trigger

---

🛡️ Cluster Autoscaler / Karpenter

HPA pod ekler ama node yetersizse ne olur?

Cluster Autoscaler

# AWS EKS örneği
apiVersion: autoscaling/v1
kind: ClusterAutoscaler
spec:
  resourceLimits:
    minNodes: 3
    maxNodes: 50
  scaleDown:
    enabled: true
    delayAfterAdd: 10m

Karpenter (önerilen, AWS)

apiVersion: karpenter.sh/v1
kind: NodePool
metadata:
  name: default
spec:
  template:
    spec:
      requirements:
        - key: kubernetes.io/arch
          operator: In
          values: [amd64, arm64]
        - key: karpenter.sh/capacity-type
          operator: In
          values: [spot, on-demand]
      nodeClassRef:
        name: default
  limits:
    cpu: "1000"
  disruption:
    consolidationPolicy: WhenUnderutilized

→ Karpenter otomatik node oluşturur (instance type seçer), idle'da consolidate eder.

---

🚫 Anti-Pattern Tablosu

Anti-pattern	Niye kötü	Doğru
HPA + VPA Auto aynı resource	Race condition	HPA + VPA Off (recommendation)
HPA min=1, max=2	Effective scaling yok	min 3 (HA), max gerçek peak
Scale down çok hızlı	Pod thrashing	stabilizationWindowSeconds 300+
KEDA scale-to-zero ama health check yetersiz	İlk request slow (cold start)	warmup veya prewarm
Cluster Autoscaler yok HPA + büyük scale up	Pod pending	CA / Karpenter
HPA target %95 CPU	Spike'da çökme	%70 hedef (headroom)
Custom metric production'da test'siz	False scaling	Prometheus query test
KEDA queue trigger threshold yanlış	Yanlış scale	Burst pattern analiz
Resource request set değil	HPA % hesaplayamaz	requests zorunlu
HPA üzerinde manuel kubectl scale	HPA geri alır	HPA disable veya değiştir

---

📋 Autoscaling Production Checklist

[ ] Resource requests/limits set (HPA için zorunlu)
[ ] HPA: min 3 replica (HA)
[ ] HPA: target %70 CPU (headroom)
[ ] HPA behavior: scale up fast, scale down slow
[ ] Multi-metric: CPU + Memory veya custom
[ ] Custom metric: Prometheus adapter
[ ] VPA: Off mode (recommendation), manuel uygula
[ ] KEDA: queue / cron / Prometheus trigger
[ ] Scale-to-zero (uygunsa, idle queue)
[ ] Cluster Autoscaler / Karpenter (node scaling)
[ ] PodDisruptionBudget (HPA scale down sırası)
[ ] Quarterly: HPA target review
[ ] Monitoring: HPA events + scaling history
[ ] Load test: HPA gerçek peak'i karşılıyor mu

---

📚 Referanslar

• HPA Docs — kubernetes.io/docs/tasks/run-application/horizontal-pod-autoscale/
• VPA Docs — github.com/kubernetes/autoscaler/tree/master/vertical-pod-autoscaler
• KEDA — keda.sh
• Karpenter — karpenter.sh
• Cluster Autoscaler — github.com/kubernetes/autoscaler
• Production-Checklist.md
• Resource-Limits-Guide.md
• 11-SRE/Capacity-Planning.md
• 14-Sustainability/Carbon-Aware-Computing.md — KEDA carbon scaler

---

"Autoscaling 'pod sayısı oynamak' değil — doğru tool, doğru yer. HPA + VPA + KEDA + Karpenter dörtlüsü doğru kombinasyon ile maliyet %50 azaltır + reliability ↑."