Kif jaħdem il-Horizontal Pod Autoscaler, u kif jirrelata mal-Cluster Autoscaler?

Question

IT Interview Practice · Accepted Answer

Il-**Horizontal Pod Autoscaler (HPA)** iżid jew ineħħi **Pod replicas** ibbażat fuq metrics osservati (CPU, memorja, jew custom). Il-**Cluster Autoscaler (CA)** iżid jew ineħħi **nodes** meta l-Pods ma jistgħux jiġu skedjati jew in-nodes ma jintużawx biżżejjed. Joperaw f'layers differenti u jaħdmu flimkien.

## Żewġ autoscalers, żewġ layers

```text
HPA  → scales PODS  (more/fewer replicas of a Deployment)     — reacts to LOAD
CA   → scales NODES (grows/shrinks the cluster's machines)     — reacts to PENDING Pods
VPA  → (bonus) adjusts a Pod's requests/limits                 — reacts to right-sizing

Typical flow: load rises → HPA adds Pods → Pods can't fit → CA adds a node →
              Pods schedule. Load drops → HPA removes Pods → node idle → CA removes it.
```

## Kif jaħdem il-control loop tal-HPA

```text
Every ~15s the HPA controller:
  1. reads current metric (e.g. avg CPU across Pods) from metrics-server / custom API
  2. computes:  desiredReplicas = ceil(currentReplicas * currentMetric / targetMetric)
  3. scales the Deployment toward that (bounded by minReplicas/maxReplicas)
Example: 4 Pods at 80% CPU, target 50% → 4 * 80/50 = 6.4 → 7 replicas.
Stabilization windows dampen flapping (esp. on scale-down).
```

## Manifest ta' HPA

```yaml
apiVersion: autoscaling/v2
kind: HorizontalPodAutoscaler
metadata:
  name: web-hpa
spec:
  scaleTargetRef:
    apiVersion: apps/v1
    kind: Deployment
    name: web
  minReplicas: 3
  maxReplicas: 20
  metrics:
    - type: Resource
      resource:
        name: cpu
        target:
          type: Utilization
          averageUtilization: 60      # żomm is-CPU medja madwar 60% tar-REQUEST
  behavior:                           # fakultattiv: irranġa l-veloċità tal-iscale up/down
    scaleDown:
      stabilizationWindowSeconds: 300
```

```bash
kubectl get hpa                       # metric attwali/mira, għadd ta' replicas
kubectl top pods                      # jeħtieġ metrics-server
```

## Dipendenzi u twissijiet importanti

```text
• HPA on CPU/memory REQUIRES metrics-server installed; utilization is relative to REQUESTS
  → if requests are unset/wrong, HPA math is meaningless. Set requests correctly.
• Custom/external metrics (RPS, queue depth) need an adapter (Prometheus Adapter, KEDA).
• HPA and manual `replicas` conflict — let the HPA own the replica count.
• Cluster Autoscaler needs correct requests too: it looks at Pending Pods' REQUESTS to
  decide node scaling. Node scale-up takes minutes (provisioning), so keep headroom.
```

## Għaliex huwa importanti

L-autoscaling huwa dak li jħalli lil Kubernetes iqabbel il-kapaċità mad-domanda awtomatikament, u li tifhem iż-żewġ layers huwa għarfien tal-livell senior li jifred clusters effiċjenti fl-ispiża u reżiljenti minn dawk over- jew under-provisioned. Il-mudell mentali ewlieni huwa li l-**HPA jiskala l-Pods** b'risposta għall-load (jaġġusta l-għadd ta' replicas mis-CPU, memorja, jew metrics custom osservati) filwaqt li l-**Cluster Autoscaler jiskala n-nodes** b'risposta għall-pressjoni tal-iscceduling (iżid magni meta l-Pods huma Pending, ineħħihom meta n-nodes huma idle) — joperaw f'layers differenti u jikkooperaw: l-HPA jista' jżid biss Pods jekk ikun hemm spazju, u s-CA jipprovdi dak l-ispazju. Li tifhem il-control loop tal-HPA — il-formula `desiredReplicas = ceil(currentReplicas * currentMetric / targetMetric)`, ir-rikonċiljazzjoni perjodika, u l-istabilization windows li jipprevjenu l-flapping — iħallik tikkonfigurah u tiddibaggjah minflok ma tittrattah bħala maġija. It-twissijiet huma fejn jidher il-ġudizzju senior: l-HPA fuq CPU/memorja **jeħtieġ metrics-server** u jikkalkula l-utilizzazzjoni **relattiva għar-requests**, għalhekk requests ħżiena jew neqsin jagħmlu l-matematika tiegħu bla sens; metrics custom (requests-per-second, queue depth) jeħtieġu adapter bħal Prometheus Adapter jew KEDA; u l-iscale-up tan-nodes jieħu minuti, għalhekk il-headroom jgħodd. Peress li li tqassam il-kapaċità sew jaffettwa direttament l-affidabbiltà taħt spikes tat-traffiku u n-nefqa fil-cloud waqt perjodi kwieti, u peress li ż-żewġ autoscalers jiddependu fuq resource requests issettjati korrettament, li tqabbad l-HPA u l-Cluster Autoscaler huwa fundamentali biex topera Kubernetes tal-produzzjoni b'mod effiċjenti u affidabbli.

Żewġ autoscalers, żewġ layers

Kif jaħdem il-control loop tal-HPA

Manifest ta' HPA

Dipendenzi u twissijiet importanti

Għaliex huwa importanti

Mistoqsijiet relatati