Application Auto Scaling
Application Auto Scaling extends auto scaling to cover multiple services. They use CloudWatch alarms to determine when to scale out or back in.
Concepts
- Scaling Plans define a set of resources and a strategy for scaling them in and out.
- Scaling Policies define scaling strategies.
Supported services
- ECS clusters
- EC2 Spot Fleets
- EMR
- AppStream 2.0
- DynamoDB
- Aurora replicas
- SageMaker
- Custom resources
Finding scalable resources
Resources which support scaling policies can be identified through a range of means:
- CloudFormation stack.
- Tags, though this doesn't currently work for EC2 Spot Fleets.
- EC2 Auto Scaling Groups.
Scaling policies
Multiple scaling policies are supported.
Target tracking
In this configuration, resources are scaled based on a target value for a CloudWatch metric.
Step-based
Step scaling allows resources to be scaled proportionally to a breach of a CloudWatch alarm.
Scheduled
Scheduled scaling policies allow scaling based on the date and time, and can be useful for fairly predictable workloads where demand is understood.