Easy ReviewSAA-C03
Design Resilient Architectures10

Application Auto Scaling

Scale supported application capacity with registered targets, metric policies, schedules and ECS demand forecasts.

Este conteúdo de estudo ainda não foi traduzido, então aparece em inglês. Ajude a traduzir

Scalable targets and policies

Application Auto Scaling changes a supported resource’s capacity as demand changes. It scales an application’s resource dimension, such as an ECS service’s task count or DynamoDB provisioned capacity. A scalable dimensionA scalable dimension identifies the specific capacity value to change, such as ECS desired task count. is the particular capacity value the service can change.

How it works

  1. Register a scalable targetA scalable target is a registered resource capacity dimension that Application Auto Scaling can change. using its service namespaceA service namespace identifies which AWS service owns a scalable target., resource ID and scalable dimension. The namespace identifies the service. Set minimum and maximum capacity.
  2. For ECS, use namespace ecs, resource ID service/cluster-name/service-name and dimension ecs:service:DesiredCount.
  3. Choose a scaling mechanism. Target trackingTarget tracking adjusts capacity to keep a utilization or throughput metric near a chosen value. manages its own CloudWatch alarms; step scaling requires alarms you manage. An alarm breachAn alarm breach occurs when a monitored metric crosses the configured threshold. means a metric crossed its configured threshold.
  4. Policies adjust capacity within the target’s bounds. A cooldownA cooldown is a period that gives a previous scaling activity time to take effect. lets a previous change take effect. During a scale-inScaling in removes capacity when demand decreases. cooldown, a required scale-outScaling out adds capacity to handle more demand. can still run.
  5. Use scheduled actions to change capacity bounds before predictable load. For ECS historical demand forecasts, predictive scaling can work alongside dynamic policies.
Requirement Mechanism Control
Maintain chosen average utilization Target tracking Service-managed alarms and adjustments
Different increments by alarm breach size Step scaling Customer-managed alarms and step adjustments
Prepare for a known time window Scheduled scaling Timed minimum or maximum changes
Forecast recurring ECS demand Predictive scaling Historical demand forecasts

Scaling out adds capacity; scaling in removes it. Target tracking needs a metric that reflects capacity utilization, such as average CPU or requests per target. Missing data can leave its alarm in INSUFFICIENT_DATA and prevent scaling. Leave policy-managed alarms under Application Auto Scaling control.

Exam cues

  • If the question says scale ECS tasks, DynamoDB capacity, Aurora readers or Lambda provisioned concurrency, think Application Auto Scaling, because these are supported resource dimensions.
  • If the question says keep utilization near a target with minimal alarm management, think target tracking, because it manages alarms and adjustments.
  • If the question says ECS tasks lack EC2 capacity, think ASG capacity providerAn ASG capacity provider lets ECS manage the EC2 instance capacity available for tasks. managed scaling, because task count and cluster instance capacity are separate.
  • If the question says prepare for known business hours, think scheduled scaling, because it can raise capacity bounds before requests arrive.

Remember

  • A scalable target needs namespace, resource ID, dimension, minimum and maximum.
  • Default target tracking cooldown: 300 seconds for ECS and Aurora; 0 or null for DynamoDB and Lambda provisioned concurrency.
  • With multiple target tracking policies, any can scale out; all policies with scale-in enabled must agree to scale in.
  • Predictive scaling defaults to ForecastOnly. ForecastAndScale enables forecast-driven scaling. Predictive scaling alone does not remove ECS tasks.
  • Policy support varies. Step scaling does not support every Application Auto Scaling target.

Trap: Application Auto Scaling is a separate service from EC2 Auto Scaling. Registering an ECS service target changes task count, not the size of the EC2 Auto Scaling group hosting those tasks.

Target tracking adjusts ECS service task count
12CloudWatch metricScaling policyECS task count
  1. Managed alarm

  2. Adjust within bounds

Da documentação da AWS (18)
  • Application Auto Scaling scales supported resource dimensions including ECS service task count, DynamoDB table and index capacity, Aurora replicas and Lambda provisioned concurrency.
  • EC2 Auto Scaling groups use Amazon EC2 Auto Scaling rather than Application Auto Scaling.
  • ECS cluster auto scaling uses managed scaling on Auto Scaling group capacity providers to adjust the EC2 instances available to tasks.
  • A scalable target is uniquely identified by service namespace, resource ID and scalable dimension.
  • Registering a scalable target requires minimum and maximum capacity; scaling policies operate within these bounds.
  • For ECS service scaling, the namespace is ecs, the resource ID is service/cluster-name/service-name and the dimension is ecs:service:DesiredCount.
  • Target tracking maintains a target utilization or throughput metric by managing CloudWatch alarms and calculating capacity changes.
  • A target tracking metric must represent utilization that changes proportionally with target capacity, such as average CPU or requests per target.
  • Do not manually create, edit or delete CloudWatch alarms managed by a target tracking policy.
  • Missing target tracking metric data can put alarms into INSUFFICIENT_DATA, preventing scaling until new data is available.
  • With multiple target tracking policies, any policy can trigger scale-out, but all policies with scale-in enabled must agree before scale-in.
  • A cooldown gives a previous scaling activity time to take effect; a scale-in cooldown does not prevent an urgent scale-out.
  • Default target tracking cooldown is 300 seconds for ECS and Aurora, and 0 or null for DynamoDB and Lambda provisioned concurrency; null is treated as zero.
  • Step scaling uses customer-managed CloudWatch alarms and capacity adjustments based on the size of the alarm breach.
  • Step scaling supports ECS services and Aurora clusters but does not list DynamoDB or Lambda provisioned concurrency as supported scalable targets.
  • Scheduled actions change minimum or maximum capacity at configured times and can work alongside dynamic scaling policies.
  • ECS predictive scaling uses historical demand to forecast capacity, can work with dynamic scaling and does not scale in tasks by itself.
  • Predictive scaling defaults to ForecastOnly; ForecastAndScale enables capacity changes based on forecasts.

Flashcards

toque para virar
Resposta

·

Todos os 10 flashcards em lista

Which workloads use Application Auto Scaling, and how does ECS cluster capacity scaling differ?

Application Auto Scaling handles ECS service task count, DynamoDB capacity, Aurora replicas and Lambda provisioned concurrency. EC2 groups use EC2 Auto Scaling; ECS cluster auto scaling adjusts EC2 instances through ASG capacity providers.

An Amazon ECS service needs to scale its task count as demand changes. Which AWS scaling service fits?

Application Auto Scaling can adjust the ECS service task count within configured minimum and maximum capacity.

An ECS service should maintain a chosen average CPU utilization without manually maintained alarms. Which Application Auto Scaling policy fits?

Target tracking. It creates and manages CloudWatch alarms and calculates capacity changes to approach the target.

An application needs different capacity increments for small and large alarm breaches. Which Application Auto Scaling policy fits?

Step scaling. The team manages the CloudWatch alarms and defines adjustments by breach size. Check that the resource supports step scaling.

An ECS application expects a known busy window and needs capacity before it starts. Which Application Auto Scaling feature fits?

Scheduled scaling. Raise the minimum capacity before the busy window, then let dynamic policies operate within the scheduled bounds.

An Application Auto Scaling target tracking alarm enters INSUFFICIENT_DATA. Should missing samples be treated as permission to scale in?

No. Target tracking cannot scale while required metric data is missing. Restore metric reporting.

Two Application Auto Scaling target tracking policies disagree about scale-in. What happens?

Scale-in waits until all policies with scale-in enabled agree. Either policy can request scale-out.

Does increasing an ECS service task count through Application Auto Scaling also directly scale its EC2 Auto Scaling group?

No. Service task scaling and cluster instance scaling are separate. Use ECS managed scaling on an ASG capacity provider for the underlying EC2 capacity.

An application has a predictable daily traffic peak. How can Application Auto Scaling prepare capacity?

Use scheduled scaling to change minimum or maximum capacity at configured times. It can work alongside dynamic scaling.

An ECS service has recurring demand patterns and slow task startup. How should the team assess predictive scaling?

Start with ForecastOnly, evaluate forecasts, and use ForecastAndScale to enable forecast-driven capacity changes. Combine with dynamic scaling for scale-in and unexpected changes.

Você concluiu todos os tópicos da SAA-C03!

Seu badge ficou dourado. Próximo passo: agendar a prova na AWS. Depois de passar, adicione-se ao Quem passou.

Compartilhar

Conte que você concluiu esta certificação

Concluí todos os tópicos da AWS Certified Solutions Architect - Associate (SAA-C03) no Easy Review. Venha estudar também!