Capacity Planning

---

1 - Introduction

1.1 Définition

Capacity Planning = Process de détermination des ressources nécessaires pour satisfaire la demande future.

1.2 Pourquoi c'est important

Scenario	Conséquence
Sous-provisionné	Outages, latence, perte de clients
Sur-provisionné	Gaspillage de ressources, coûts élevés
Bien planifié	Performance optimale, coûts maîtrisés

---

2 - Métriques de capacité

2.1 Ressources à monitorer

capacity_metrics:
  compute:
    - CPU utilization
    - Memory usage
    - Network bandwidth
    
  storage:
    - Disk space
    - IOPS
    - Throughput
    
  application:
    - Concurrent connections
    - Requests per second
    - Queue depth
    
  database:
    - Connection pool
    - Query throughput
    - Replication lag

2.2 Saturation Points

saturation_thresholds:
  cpu:
    warning: 70%
    critical: 85%
    action: "Scale up or out"
    
  memory:
    warning: 75%
    critical: 90%
    action: "Add memory or instances"
    
  disk:
    warning: 70%
    critical: 85%
    action: "Extend storage"
    
  connections:
    warning: 70%
    critical: 85%
    action: "Increase pool or scale"

---

3 - Demand Forecasting

3.1 Méthodes

forecasting_methods:
  historical_trend:
    description: "Extrapolation des données passées"
    best_for: "Croissance organique stable"
    
  seasonal:
    description: "Patterns récurrents (jour, semaine, année)"
    best_for: "E-commerce, B2C"
    
  event_driven:
    description: "Événements planifiés (promos, launches)"
    best_for: "Marketing campaigns"
    
  regression:
    description: "Modèle statistique"
    best_for: "Prédictions complexes"

3.2 Prometheus Prediction

# Prédiction linéaire: espace disque dans 7 jours
predict_linear(node_filesystem_avail_bytes[7d], 7*24*3600)

# Prédiction: quand le disque sera plein
(
  node_filesystem_avail_bytes
  /
  deriv(node_filesystem_avail_bytes[7d])
) / 3600 / 24  # Jours restants

3.3 Growth Modeling

growth_model:
  current_metrics:
    users: 100000
    rps: 1000
    instances: 10
    
  growth_rate:
    monthly: 10%
    yearly: 214%  # Compound
    
  projection_6_months:
    users: 177000
    rps: 1770
    instances_needed: 18

---

4 - Load Testing

4.1 Types de tests

Type	Objectif	Durée
Smoke	Validation basique	Minutes
Load	Performance normale	Heures
Stress	Trouver les limites	Heures
Spike	Pics soudains	Minutes
Soak	Endurance	Jours

4.2 k6 Load Test

// load-test.js
import http from 'k6/http';
import { check, sleep } from 'k6';

export const options = {
  stages: [
    { duration: '5m', target: 100 },  // Ramp up
    { duration: '30m', target: 100 }, // Steady
    { duration: '5m', target: 200 },  // Peak
    { duration: '5m', target: 0 },    // Ramp down
  ],
  thresholds: {
    http_req_duration: ['p(99)<500'],
    http_req_failed: ['rate<0.01'],
  },
};

export default function () {
  const res = http.get('https://api.example.com/endpoint');
  check(res, {
    'status is 200': (r) => r.status === 200,
    'response time < 500ms': (r) => r.timings.duration < 500,
  });
  sleep(1);
}

4.3 Déterminer les limites

capacity_limits:
  test_results:
    max_rps: 5000
    breaking_point_rps: 6000
    p99_at_max: 450ms
    error_rate_at_max: 0.5%
    
  headroom:
    target: 30%  # Headroom de sécurité
    safe_max_rps: 3500
    
  scaling_trigger:
    cpu: 70%
    rps_per_instance: 350

---

5 - Capacity Planning Process

5.1 Annual Planning

annual_planning:
  q1_review:
    - Review past year growth
    - Analyze seasonal patterns
    - Update growth projections
    
  q2_planning:
    - Define capacity requirements
    - Budget allocation
    - Infrastructure roadmap
    
  q3_implementation:
    - Provision resources
    - Optimize existing systems
    - Load testing
    
  q4_preparation:
    - Holiday/peak preparation
    - Buffer provisioning
    - Incident readiness

5.2 Capacity Model

capacity_model:
  service: api-gateway
  
  resource_per_unit:
    cpu_cores: 0.5
    memory_gb: 1
    rps_capacity: 100
    
  current_state:
    instances: 20
    total_capacity: 2000 rps
    current_usage: 1400 rps
    utilization: 70%
    
  growth_projection:
    expected_growth: 50%
    required_capacity: 2100 rps
    required_instances: 21
    recommended: 25  # With 20% buffer

---

6 - Auto-scaling

6.1 Kubernetes HPA

apiVersion: autoscaling/v2
kind: HorizontalPodAutoscaler
metadata:
  name: api-hpa
spec:
  scaleTargetRef:
    apiVersion: apps/v1
    kind: Deployment
    name: api
  minReplicas: 3
  maxReplicas: 50
  metrics:
    - type: Resource
      resource:
        name: cpu
        target:
          type: Utilization
          averageUtilization: 70
    - type: Resource
      resource:
        name: memory
        target:
          type: Utilization
          averageUtilization: 80
  behavior:
    scaleUp:
      stabilizationWindowSeconds: 60
      policies:
        - type: Percent
          value: 100
          periodSeconds: 60
    scaleDown:
      stabilizationWindowSeconds: 300
      policies:
        - type: Percent
          value: 10
          periodSeconds: 60

6.2 Custom Metrics

apiVersion: autoscaling/v2
kind: HorizontalPodAutoscaler
spec:
  metrics:
    - type: External
      external:
        metric:
          name: queue_depth
          selector:
            matchLabels:
              queue: orders
        target:
          type: AverageValue
          averageValue: "100"

---

7 - Cost Optimization

7.1 Right-sizing

rightsizing:
  analysis:
    - Review actual resource usage
    - Compare to requested
    - Identify over-provisioned
    
  actions:
    - Reduce oversized instances
    - Use appropriate instance types
    - Implement resource quotas

7.2 Reserved vs On-demand

capacity_strategy:
  reserved:
    for: "Baseline capacity"
    percentage: 60%
    savings: "30-60%"
    
  on_demand:
    for: "Variable capacity"
    percentage: 30%
    
  spot:
    for: "Non-critical, interruptible"
    percentage: 10%
    savings: "60-90%"

---

Résumé

Dans ce chapitre, nous avons appris :

• Les métriques de capacité
• Le Demand Forecasting
• Le Load Testing (k6)
• Le Capacity Planning process
• L'Auto-scaling Kubernetes
• L'optimisation des coûts

---

Prochaine étape

Dans le prochain chapitre, nous verrons Release Engineering.

→ Chapitre suivant : Release Engineering

---

← Retour à la table des matières