A single EC2 instance is a single point of failure. To make an app highly available and elastic — able to handle variable load and survive failures automatically — AWS gives you two services that work hand in hand: ELB and ASG.

The two pieces

ServiceWhat it doesQuestion it answers
ELBDistributes incoming traffic across multiple EC2s“How do users reach my fleet?”
ASGAdds, removes, and replaces EC2s based on demand or health“How big should my fleet be?”

They are almost always used together.

Mental model

Users ──> ELB ──> [EC2, EC2, EC2, ...]  ← fleet managed by ASG
              spreads traffic         adds/removes/replaces instances
  • ELB is the traffic cop — it routes requests.
  • ASG is the fleet manager — it right-sizes the fleet and heals it.

ELB — Elastic Load Balancer

What it does:

  • Presents a single, stable DNS name in front of a constantly changing fleet.
  • Runs health checks, and only routes to healthy instances.
  • Spreads traffic across Availability Zones.
  • Terminates SSL/TLS, offloading that work from the EC2s.

Four types:

TypeLayerUse
ALB (Application LB)Layer 7 (HTTP/HTTPS)Web apps, path/host-based routing, microservices
NLB (Network LB)Layer 4 (TCP/UDP)Extreme performance, static IPs, non-HTTP traffic
GLB (Gateway LB)Layer 3Routing through third-party appliances such as firewalls
CLB (Classic LB)L4 / L7Legacy — don’t use it for anything new

For web apps, the default choice is ALB.

Key ALB features:

  • Path-based routing → /api/* goes to the API fleet, /img/* goes to the image fleet.
  • Host-based routing → api.app.com and www.app.com route to different fleets.
  • Target groups → groupings of instances, containers, or Lambdas that the load balancer sends traffic to.

ASG — Auto Scaling Group

What it does:

  • Maintains a defined number of EC2s — min, desired, and max.
  • Launches replacements when instances fail their health checks.
  • Scales out and in based on metrics such as CPU, request count, or custom signals.
  • Spreads instances across Availability Zones automatically.

Key settings:

  • Min / Desired / Max — the bounds on fleet size.
  • Launch Template — the AMI, instance type, security group, IAM role, and user data to use for new instances.
  • Health checks — at the EC2 level (is the VM alive?) and/or the ELB level (is the app responding?).
  • Scaling policies:
    • Target tracking → “keep CPU at 50%.” The most common and simplest.
    • Step scaling → “if CPU > 70%, add 2 instances.”
    • Scheduled → “scale up every weekday at 9am.”

How they work together — a running example

Consider a photo app with a typical weekday traffic pattern:

  1. The ASG is configured with min=2, max=10, desired=2.
  2. An ALB sits in front, registered to the ASG’s target group.
  3. Morning: traffic spikes and CPU hits 70% on both EC2s.
  4. The ASG launches EC2 #3 from the launch template.
  5. The new EC2 boots, passes its health check, and the ALB starts sending it traffic.
  6. Evening: traffic drops, CPU falls to 20%, and the ASG terminates the extra EC2s.
  7. Middle of the night: EC2 #2 crashes, fails its health check, and the ASG terminates it and launches a replacement.

The app stayed up the whole time, with no human involved.

Sticky sessions (an ALB feature)

  • Normally each request can go to any EC2 — the app is treated as stateless.
  • Sticky sessions make the same user always land on the same EC2, tracked via a cookie.
  • This is useful when session state lives on the instance, but stateless apps scale better — keep session state in Redis or a database instead.

Cross-zone load balancing

  • Enabled → the load balancer spreads traffic evenly across all instances, regardless of which AZ they are in.
  • Disabled → traffic is spread evenly across AZs first, then across the instances within each AZ.
  • On ALB it is on by default and free. On NLB it is off by default and costs extra.

Key principles

  • An ELB needs at least 2 AZs to be useful — a single AZ means no high availability.
  • An ASG spanning multiple AZs survives the loss of an entire AZ.
  • Health checks are critical — bad health checks send traffic to dead instances.
  • Use launch templates, not the older launch configurations.
  • Stateless EC2s scale best — push state out to S3, RDS, or Redis.
  • An ASG replaces, it does not repair — treat instances as cattle, not pets.

Summary

  • ELB spreads traffic across a fleet; ASG manages the size and health of that fleet.
  • Together they give you an app that is both elastic and self-healing.
  • Use an ALB for web apps, span multiple AZs for real availability, and keep your instances stateless so the fleet can grow, shrink, and heal freely.