Auto Scaling Group Basics

Desired capacity and self-healing

An Auto Scaling group (ASG) maintains capacity between minimum and maximum, aiming at desired capacity. If an instance terminates or fails the configured health check, the group launches a replacement to restore desired capacity. With a target group attached, new instances register automatically.

flowchart LR
  LT["Launch template v3<br/>AMI, type, SG"] --> G["ASG web-asg<br/>min 2, desired 2, max 6"]
  G --> A["Instance<br/>AZ 1a"]
  G --> B["Instance<br/>AZ 1b"]
  A --> TG["Target group<br/>auto-registered"]
  B --> TG
  B -. "fails health check" .-> R["Terminated<br/>and replaced"]
  %% step LT: The launch template says what each new instance looks like.
  %% step G: The group keeps the number of running instances at desired, never below min or above max.
  %% step A: Instances are spread evenly across the subnets and AZs you choose.
  %% step B: If one AZ has problems, the group rebalances when it can.
  %% step TG: New instances join the target group, so the load balancer starts sending them traffic.
  %% step R: An unhealthy instance is terminated and a fresh one launched from the template.

Launch templates

Launch Templates have versions and are the current choice; Launch Configurations are legacy. A template holds the AMI, instance type, key pair, Security Groups, IAM instance profile, user data, and tags. Change an AMI with a new template version and run an instance refresh (rolling replacement with a minimum healthy percentage) rather than editing running instances and expecting automatic rollout.

Health checks

• EC2 health check (default) — replaces instances whose EC2 status checks fail.
• ELB health check — also replaces instances the load balancer marks unhealthy, which catches a crashed application on a healthy VM. Enable it whenever the ASG serves a load balancer, with a grace period long enough for boot and app start.

Spreading across AZs and instance types

Subnets determine AZs and network paths; pick subnets in at least two AZs for HA. Private instances need NAT or VPC endpoints to download packages; avoid public IPs as a workaround for missing egress.

A mixed instances policy lets one group use several instance types and combine On-Demand with Spot, for example "first 2 On-Demand, the rest 70% Spot across c7g.large, c6g.large, m7g.large". More types and AZs mean fewer Spot interruptions.

Lifecycle hooks and warm pools

Lifecycle hooks pause an instance in Pending:Wait or Terminating:Wait so a script can finish setup or copy logs before it goes into or out of service. Warm pools keep pre-initialised, stopped instances ready so scale-out is faster for apps with slow startup.

The ASG itself is free; you pay for the instances it runs.

Traps: updating the template and expecting running instances to change; using only EC2 health checks behind a load balancer; a grace period shorter than the app's start time.