AutoScaling Groups

Autoscaling aims to maintain available resource inline with demands on those resources. This is horizontal scaling (Number of instances) not vertical (more tin/ram) and is is either scale out (increasing) scaling in (reduction) or to maintaining a resource level (launching new instances if failures occur).

There are now two ways to configure an Autoscaling Group in AWS.

Launch configuration

This is the older way, create aws_launch_configuration.web.tf as below:

resource "aws_launch_configuration" "web" {
  name          = "web_config"
  image_id      = data.aws_ami.xenial.id
  instance_type = "t2.micro"
}

If the properties in a launch configuration look familiar, its because they are used replace the aws_instance resoure, when combined with an Autoscaling group aws_autoscaling_group.web.tf.

resource "aws_autoscaling_group" "web" {
  availability_zones   = ["eu-west-1a"]
  desired_capacity     = 1
  max_size             = 1
  min_size             = 1
  launch_configuration = aws_launch_configuration.web.id
}

Replacing one resource for two might not seem like an obvious gain, but functional it's big, now should you lose the instance a new one will be started automatically. A launch configuration can be used as a foundational component of a blue-green deployment strategy.

Typically this can be achieved by updating the AMI of the Launch config, and then Scaling out and then In, this action removes the oldest EC2 instance.

Launch template

Launch templates are the new way, being almost identical to Launch configuration but with more support for parameters and support for multiple versions.

For our terraform example, swap out the launch configuration and add aws_launch_template.web.tf:

resource "aws_launch_template" "web" {
  name_prefix   = "web"
  image_id      = data.aws_ami.xenial.id
  instance_type = "t2.micro"
}

Together with a small change to the scaling group.

resource "aws_autoscaling_group" "web" {
  availability_zones = ["eu-west-1a"]
  desired_capacity   = 1
  max_size           = 1
  min_size           = 1

  launch_template {
    id      = aws_launch_template.web.id
    version = "$$Latest"
  }
}

TODO:extract numbers to tfvars modify and reapply

Autoscaling Policy

resource "aws_autoscaling_policy" "defaultmetrics" {
  name                   = "web-default-scaling"
  scaling_adjustment     = 4
  adjustment_type        = "ChangeInCapacity"
  cooldown               = 300
  autoscaling_group_name = aws_autoscaling_group.web.name
}

There's nothing very "smart" or reactive here yet. I'd like increase or decrease the capacity of the web tier in response to application performance metric.

TODO:scaling no point scaling down paid for instances if not reserved? EC2 instance custom Metrics - cpu utilisation from hypervisor, io bound?

Autoscaling groups for other objects

Links

https://docs.aws.amazon.com/autoscaling/ec2/userguide/AutoScalingGroup.html https://www.terraform.io/docs/providers/aws/r/autoscaling_policy.html