Stop 04 · Compute ServicesTopics · 4/10

Load Balancing Compute

In short

Once you have more than one server, someone has to decide which one gets each request, and stop sending people to the one that just crashed. That someone is the load balancer.

Applying Phase 2's load balancer concepts specifically to distributing traffic across a fleet of compute instances — health checks, target groups, and routing rules all connect the load balancer to your compute layer.

The blob character dressed as a traffic cop waves cars towards open toll booths.

Think of it like this

The cop at the toll plaza

At a busy toll plaza, a traffic cop waves cars to whichever booth is free. If one booth's barrier is stuck, he stops sending cars there until it's fixed. A load balancer does that for requests.

Watch it happen

Health checks in action

Users

Load balancer

web-1

web-2

GET /products

1/6A request reaches the load balancer's single public address.

The big ideas

What you need to know

Target groups

The load balancer forwards to a group of targets: instances, IPs or containers. Auto scaling adds and removes them.

Health checks

A small /health endpoint that returns 200 when the app is truly ready. Unhealthy targets get no traffic.

Layer 7 routing

An application load balancer can send /api to one group and /images to another, based on path or host.

TLS at the edge

The load balancer can hold the HTTPS certificate, so your servers deal with plain HTTP inside the VPC.

See it in code

Walk through the code

hcl

1resource "aws_lb_target_group" "web" {2  name     = "web-tg"3  port     = 80804  protocol = "HTTP"5  vpc_id   = aws_vpc.main.id6 7  health_check {8    path                = "/health"9    matcher             = "200"10    interval            = 1511    healthy_threshold   = 212    unhealthy_threshold = 313  }14}

1/3Targets listen on port 8080 inside the VPC.

Terraform for a target group. Attach it to an application load balancer listener and an auto scaling group.

Save yourself the pain

Common mistakes

Oops

Pointing the health check at the home page that loads ten database queries.

Do this instead

Use a light /health route that checks the app is up and its key dependencies are reachable.

Oops

Opening app servers to the internet as well as the load balancer.

Do this instead

Let app servers accept traffic only from the load balancer's security group.

Your mission

Break one server on purpose

~45 min

Quick check

A target fails its health checks. What does the load balancer do?

Go deeper

1Attach a load balancer to a set of VM instances or an auto scaling group.
2Configure a health check and verify unhealthy instances stop receiving traffic.
3Learn to test failover by manually stopping one instance and observing traffic reroute.