How to recreate EC2 instances of an autoscaling group with terraform?

问题

Scenario: I am running an AWS autoscaling group (ASG), and I have changed the associated launch configuration during terraform apply. The ASG stays unaffected.

How do I recreate now the instances in that ASG (i.e., replace them one-by-one to do a rolling replace), which then is based on the changed/new launch configuration?

What I've tried: With terraform taint one can mark resources to be destroyed and recreated during the next apply. However, I don't want to taint the autoscaling group (which is a resource, and single instances are not in this case), but single instances in it. Is there a way to taint single instances or am I thinking in the wrong direction?

回答1:

The normal thing to do here is to use Terraform's lifecycle management to force it to create new resources before destroying the old ones.

In this case you might set your launch configuration and autoscaling group up something like this:

resource "aws_launch_configuration" "as_conf" {
    name_prefix = "terraform-lc-example-"
    image_id = "${var.ami_id}"
    instance_type = "t1.micro"

    lifecycle {
      create_before_destroy = true
    }
}

resource "aws_autoscaling_group" "bar" {
    name = "terraform-asg-example-${aws_launch_configuration.as_conf.name}"
    launch_configuration = "${aws_launch_configuration.as_conf.name}"

    lifecycle {
      create_before_destroy = true
    }
}

Then if you change the ami_id variable to use another AMI Terraform will realise it has to change the launch configuration and so create a new one before destroying the old one. The new name generated by the new LC is then interpolated in the ASG name forcing a new ASG to be rebuilt.

As you are using create_before_destroy Terraform will create the new LC and ASG and wait for the new ASG to reach the desired capacity (which can be configured with health checks) before destroying the old ASG and then the old LC.

This will flip all the instances in the ASG at once. So if you had a minimum capacity of 2 in the ASG then this will create 2 more instances and as soon as both of those pass health checks then the 2 older instances will be destroyed. In the event you are using an ELB with the ASG then it will join the 2 new instances to the ELB so, temporarily, you will have all 4 instances in service before then destroying the older 2.

回答2:

You cannot solve this with terraform alone. Please see this AWS document (http://docs.aws.amazon.com/autoscaling/latest/userguide/LaunchConfiguration.html):

When you change the launch configuration for your Auto Scaling group, any new instances are launched using the new configuration parameters, but existing instances are not affected.

回答3:

This isn't an answer per se, but I struggled with this as well. In one context it was general, in another, Kubernetes.

I ended up writing an open-source service to watch and auto-rolling update. It should solve this, but it is new, more than happy to get feedback/issues/PRs on it.

https://github.com/deitch/aws-asg-roller

来源：https://stackoverflow.com/questions/39345609/how-to-recreate-ec2-instances-of-an-autoscaling-group-with-terraform