Auto Scaling Group

Scaling Policies

• Simple / step scaling: increase or decrease based on CloudWatch alarms.

• Target tracking: select a metric and a target value, ASG will smartly adjust it.

  • keep average CPU at 40%

  • keep request count per target at 1000

• To scale based on RAM, must use a custom CloudWatch metric.

Feature

• Spot Fleet (mix on Spot and On-demand instances) supports 3 types of ASG

• Launch configuration once created cannot be changed

• To upgrade an AMI, must re-associate the launch configuration / template.

  • must terminate instances manually.

  • CloudFormation can help with the step.

• Enable CloudWatch detailed monitoring which is disabled in creation

  • ​Create a new Launch Config with detail monitoring enabled

  • Associate it to the Auto Scaling Group

• Scheduled scaling actions:

  • Modify the ASG settings (min / max / desired) at pre-defined time

  • Helpful when patterns are known in advance

• Lifecycle Hooks:

  • Perform actions before an instance is in service, or before it is terminated.

  • Ex. cleanup, log extraction, special health checks.

• AZ Rebalance

  • Be mindful of unfinished tasks

Scaling Processes

• Launch: Add a new EC2 to the group, increasing the capacity.

• Terminate: Removes an EC2 instance from the group, decreasing its capacity.

• HealthCheck: Checks the health of the instances.

• ReplaceUnhealthty: Terminate unhealthy instances and re-create them.

• AZRebalance: Balance the number of EC2 instances across AZs.

• AlarmNotification: Accepts notification from CloudWatch.

• ScheduledActions: Performs scheduled actions that you create.

• AddToLoadBalancer: Adds instances to the load balancer or target group.

Health Checks

• Types:

  • EC2 status checks (default)

    • The health of EC2 instance doesn't equal to the health of ELB. Change health check targets to terminate instances correctly.

  • ELB health checks (HTTP)

• ASG will launch a new instance after terminating an unhealthy one.

  • With Grace Period: 300s to wait for the health of new instance

  • With Cooldown Period: 300s to wait for another scaling / termination

Comparison of Auto Scaling update stratigies

• Use the same ASG with new EC2 Launch Template

  • No chance to test the service in new version.

  • requests are handled by different version of service.

• Create another ASG, EC2 Launch Template but use the same ALB.

  • Can send small traffic from ALB to new Scaling Group to verify the service in new version.

• Create another ALB, ASG, EC2 Launch Template, and use Route53 to change the routing.

  • Relying customers on good behavior about DNS queries and respecting TTL.

  • Can test service in new version with a different ALB before update the Route53.