Auto Scaling

---

1 - Types de scaling ECS

---

2 - Target Tracking Scaling

Le plus simple et recommandé. Le service ajuste automatiquement le nombre de tasks pour maintenir une métrique cible.

2.1 Configuration

# CloudFormation
ScalableTarget:
  Type: AWS::ApplicationAutoScaling::ScalableTarget
  Properties:
    MaxCapacity: 10
    MinCapacity: 2
    ResourceId: !Sub service/${ECSCluster}/${ECSService.Name}
    RoleARN: !GetAtt AutoScalingRole.Arn
    ScalableDimension: ecs:service:DesiredCount
    ServiceNamespace: ecs

CPUTargetTracking:
  Type: AWS::ApplicationAutoScaling::ScalingPolicy
  Properties:
    PolicyName: CPUTargetTracking
    PolicyType: TargetTrackingScaling
    ScalingTargetId: !Ref ScalableTarget
    TargetTrackingScalingPolicyConfiguration:
      TargetValue: 70.0
      PredefinedMetricSpecification:
        PredefinedMetricType: ECSServiceAverageCPUUtilization
      ScaleInCooldown: 300
      ScaleOutCooldown: 60

2.2 Métriques prédéfinies

Métrique	Description
ECSServiceAverageCPUUtilization	CPU moyen du service
ECSServiceAverageMemoryUtilization	Mémoire moyenne
ALBRequestCountPerTarget	Requêtes par task

2.3 Métrique personnalisée

CustomMetricTargetTracking:
  Type: AWS::ApplicationAutoScaling::ScalingPolicy
  Properties:
    PolicyName: CustomMetric
    PolicyType: TargetTrackingScaling
    ScalingTargetId: !Ref ScalableTarget
    TargetTrackingScalingPolicyConfiguration:
      TargetValue: 100
      CustomizedMetricSpecification:
        MetricName: QueueDepth
        Namespace: MyApp
        Statistic: Average
        Dimensions:
          - Name: QueueName
            Value: orders-queue
      ScaleInCooldown: 300
      ScaleOutCooldown: 60

---

3 - Step Scaling

Plus de contrôle avec des seuils et des actions définies.

# CloudWatch Alarm
HighCPUAlarm:
  Type: AWS::CloudWatch::Alarm
  Properties:
    AlarmName: HighCPU
    MetricName: CPUUtilization
    Namespace: AWS/ECS
    Statistic: Average
    Period: 60
    EvaluationPeriods: 2
    Threshold: 80
    ComparisonOperator: GreaterThanThreshold
    Dimensions:
      - Name: ClusterName
        Value: !Ref ECSCluster
      - Name: ServiceName
        Value: !GetAtt ECSService.Name
    AlarmActions:
      - !Ref ScaleUpPolicy

# Step Scaling Policy
ScaleUpPolicy:
  Type: AWS::ApplicationAutoScaling::ScalingPolicy
  Properties:
    PolicyName: ScaleUp
    PolicyType: StepScaling
    ScalingTargetId: !Ref ScalableTarget
    StepScalingPolicyConfiguration:
      AdjustmentType: ChangeInCapacity
      Cooldown: 60
      MetricAggregationType: Average
      StepAdjustments:
        - MetricIntervalLowerBound: 0
          MetricIntervalUpperBound: 20
          ScalingAdjustment: 1
        - MetricIntervalLowerBound: 20
          ScalingAdjustment: 2

---

4 - Scheduled Scaling

Scaling basé sur un calendrier prévisible.

ScheduledScaleOut:
  Type: AWS::ApplicationAutoScaling::ScalableTarget
  Properties:
    # ...
    ScheduledActions:
      - ScheduledActionName: ScaleOutMorning
        Schedule: "cron(0 8 * * ? *)"  # 8h UTC
        ScalableTargetAction:
          MinCapacity: 5
          MaxCapacity: 20
      
      - ScheduledActionName: ScaleInNight
        Schedule: "cron(0 22 * * ? *)"  # 22h UTC
        ScalableTargetAction:
          MinCapacity: 2
          MaxCapacity: 5

4.1 Via CLI

# Scale out le matin
aws application-autoscaling put-scheduled-action \
    --service-namespace ecs \
    --resource-id service/cluster-name/service-name \
    --scalable-dimension ecs:service:DesiredCount \
    --scheduled-action-name morning-scale-out \
    --schedule "cron(0 8 * * ? *)" \
    --scalable-target-action MinCapacity=5,MaxCapacity=20

---

5 - Multi-métrique scaling

Combiner plusieurs politiques pour un scaling précis.

# CPU Scaling
CPUScaling:
  Type: AWS::ApplicationAutoScaling::ScalingPolicy
  Properties:
    PolicyName: CPUScaling
    PolicyType: TargetTrackingScaling
    ScalingTargetId: !Ref ScalableTarget
    TargetTrackingScalingPolicyConfiguration:
      TargetValue: 70
      PredefinedMetricSpecification:
        PredefinedMetricType: ECSServiceAverageCPUUtilization

# Request Count Scaling
RequestScaling:
  Type: AWS::ApplicationAutoScaling::ScalingPolicy
  Properties:
    PolicyName: RequestScaling
    PolicyType: TargetTrackingScaling
    ScalingTargetId: !Ref ScalableTarget
    TargetTrackingScalingPolicyConfiguration:
      TargetValue: 1000
      PredefinedMetricSpecification:
        PredefinedMetricType: ALBRequestCountPerTarget
        ResourceLabel: !Sub 
          - "${ALBFullName}/${TargetGroupFullName}"
          - ALBFullName: !GetAtt ALB.LoadBalancerFullName
            TargetGroupFullName: !GetAtt TargetGroup.TargetGroupFullName

---

6 - Capacity Providers Auto Scaling

Pour les clusters EC2, scaling automatique des instances.

ECSCluster:
  Type: AWS::ECS::Cluster
  Properties:
    CapacityProviders:
      - !Ref EC2CapacityProvider
    DefaultCapacityProviderStrategy:
      - CapacityProvider: !Ref EC2CapacityProvider
        Weight: 1

EC2CapacityProvider:
  Type: AWS::ECS::CapacityProvider
  Properties:
    AutoScalingGroupProvider:
      AutoScalingGroupArn: !Ref ASG
      ManagedScaling:
        Status: ENABLED
        TargetCapacity: 80
        MinimumScalingStepSize: 1
        MaximumScalingStepSize: 10
      ManagedTerminationProtection: ENABLED

---

7 - IAM Role pour Auto Scaling

AutoScalingRole:
  Type: AWS::IAM::Role
  Properties:
    AssumeRolePolicyDocument:
      Version: '2012-10-17'
      Statement:
        - Effect: Allow
          Principal:
            Service: application-autoscaling.amazonaws.com
          Action: sts:AssumeRole
    Policies:
      - PolicyName: AutoScalingPolicy
        PolicyDocument:
          Version: '2012-10-17'
          Statement:
            - Effect: Allow
              Action:
                - ecs:UpdateService
                - ecs:DescribeServices
                - cloudwatch:DescribeAlarms
                - cloudwatch:PutMetricAlarm
                - cloudwatch:DeleteAlarms
              Resource: '*'

---

8 - Bonnes pratiques

8.1 Recommandations

Aspect	Recommandation
Cooldown	Scale-out: 60s, Scale-in: 300s
Min capacity	Au moins 2 pour HA
Métriques	Combiner CPU et Request count
Monitoring	Alertes sur scaling events

8.2 Éviter les problèmes

# Éviter le thrashing (scale up/down constant)
TargetTrackingScalingPolicyConfiguration:
  TargetValue: 70
  ScaleInCooldown: 300   # 5 minutes avant scale-in
  ScaleOutCooldown: 60   # 1 minute avant scale-out
  DisableScaleIn: false

8.3 Dashboard CloudWatch

ScalingDashboard:
  Type: AWS::CloudWatch::Dashboard
  Properties:
    DashboardBody: !Sub |
      {
        "widgets": [
          {
            "type": "metric",
            "properties": {
              "title": "ECS Service Scaling",
              "metrics": [
                ["AWS/ECS", "CPUUtilization", "ServiceName", "${ECSService.Name}", "ClusterName", "${ECSCluster}"],
                [".", "DesiredCount", ".", ".", ".", "."],
                [".", "RunningCount", ".", ".", ".", "."]
              ]
            }
          }
        ]
      }

---

Résumé

Dans ce chapitre, nous avons appris :

• Le Target Tracking scaling (recommandé)
• Le Step Scaling pour plus de contrôle
• Le Scheduled Scaling pour les patterns prévisibles
• Le multi-métrique scaling
• Les Capacity Providers pour EC2
• Les bonnes pratiques de scaling

---

Prochaine étape

Dans le prochain chapitre, nous verrons le Monitoring et Logging.

→ Chapitre suivant : Monitoring et Logging

---

← Retour à la table des matières