Terraform assumes aws_appautoscaling_policy names are globally unique

[cbroglie]

AWS only requires the policy name to be unique given a resource id, service namespace, and scalable dimension, but Terraform assumes the policy names are globally unique. This makes the behavior of aws_appautoscaling_policy non-deterministic when the same policy name is reused for different resources.

Terraform Version

v0.9.4

Affected Resource(s)

Please list the resources as a list, for example:

• aws_appautoscaling_policy

Terraform Configuration Files

Our configuration leverages a module named ecs_autoscaling, and its contents are here:

variable "service_name" {}
variable "cluster_name" {}
variable "min" {}
variable "max" {}
variable "iam_role" {}

variable "cooldown" {
  default = 300
}

variable "metric_aggregation_type" {
  default = "Average"
}

variable "scale_up_adjustment" {
  default = 1
}

variable "scale_down_adjustment" {
  default = -1
}

resource "aws_appautoscaling_target" "ecs_target" {
  min_capacity = "${var.min}"
  max_capacity = "${var.max}"
  resource_id = "service/${var.cluster_name}/${var.service_name}"
  role_arn = "${var.iam_role}"
  scalable_dimension = "ecs:service:DesiredCount"
  service_namespace  = "ecs"
}

resource "aws_appautoscaling_policy" "ecs_policy_up" {
  depends_on = ["aws_appautoscaling_target.ecs_target"]

  adjustment_type = "ChangeInCapacity"
  cooldown = "${var.cooldown}"
  metric_aggregation_type = "${var.metric_aggregation_type}"
  name = "scale-up"
  resource_id = "service/${var.cluster_name}/${var.service_name}"
  scalable_dimension = "ecs:service:DesiredCount"
  service_namespace = "ecs"

  step_adjustment {
    metric_interval_lower_bound = 0
    scaling_adjustment = "${var.scale_up_adjustment}"
  }
}

resource "aws_appautoscaling_policy" "ecs_policy_down" {
  depends_on = ["aws_appautoscaling_target.ecs_target"]

  adjustment_type = "ChangeInCapacity"
  cooldown = "${var.cooldown}"
  metric_aggregation_type = "${var.metric_aggregation_type}"
  name = "scale-down"
  resource_id = "service/${var.cluster_name}/${var.service_name}"
  scalable_dimension = "ecs:service:DesiredCount"
  service_namespace = "ecs"

  step_adjustment {
    metric_interval_lower_bound = 0
    scaling_adjustment = "${var.scale_down_adjustment}"
  }
}

output "target_resource_id" {
  value = "${aws_appautoscaling_target.ecs_target.resource_id}"
}

output "scale_up_policy_arn" {
  value = "${aws_appautoscaling_policy.ecs_policy_up.arn}"
}

output "scale_down_policy_arn" {
  value = "${aws_appautoscaling_policy.ecs_policy_down.arn}"
}

Then the main configuration file leverages it like this:

module "ecs_autoscaling_example1" {
  source  = "modules/ecs_autoscaling"
  service_name = "example1"
  cluster_name = "${aws_ecs_cluster.cluster.name}"
  min = "${var.min}"
  max = "${var.max}"
  iam_role = "${var.ecs_iam_role}"
}

module "ecs_autoscaling_example2" {
  source  = "modules/ecs_autoscaling"
  service_name = "example2"
  cluster_name = "${aws_ecs_cluster.cluster.name}"
  min = "${var.min}"
  max = "${var.max}"
  iam_role = "${var.ecs_iam_role}"
}

Actual Behavior

Since Terraform assumes the names are globally unique, the behavior depends on the order of results from the call to applicationautoscaling.DescribeScalingPolicies.

Steps to Reproduce

Execute terraform apply multiple times and observe that changes are made continuously.

Important Factoids

The issue is getAwsAppautoscalingPolicy only passes the policy name and service namespace parameters to applicationautoscaling.DescribeScalingPolicies, and uses the first result which matches the policy name. It should also pass the resource id and scalable dimension parameters, and assert that it gets at most 1 result back.

[ghost]

I'm going to lock this issue because it has been closed for 30 days ⏳. This helps our maintainers find and focus on the active issues.

If you have found a problem that seems similar to this, please open a new issue and complete the issue template so we can capture all the details necessary to investigate further.