template_file with count causes instances to be rebuild on increase

[MansM]

resource "template_file" "masters_cloud_init" {
  count = "${var.mastercount}"
  template = "${file("master.yaml.tpl")}"
  vars {
    hostname = "master${count.index+1}"
  }
}


resource "aws_instance" "kubemaster" {
  count = "${var.mastercount}"
  ami = "${lookup(var.amis, var.region)}"
  instance_type = "t2.micro"
  key_name = "mans"
  vpc_security_group_ids = ["${aws_security_group.kubernetes.id}"]
  subnet_id = "${aws_subnet.kubernetes.id}"
  tags = {
      Role = "kubemaster"
      Env = "dev"

  }
  user_data = "${element(template_file.masters_cloud_init.*.rendered, count.index)}"
  provisioner "remote-exec" {
    inline = [
      "ls -l"
    ]
  }
}

when I increase var.mastercount from 2 to 3, all 3 the instances are getting recreated.

-/+ aws_instance.kubemaster.0
user_data:                        "1974e36c46d425037f2cd7e66f793030fc72cfda" => "66f502e5ff78bee2ebd70b6e04be1c6cb5d78a31" (forces new resource)

-/+ aws_instance.kubemaster.1
    user_data:                        "81e28c58224841c9d6c5c6f6d91163ed9158bd80" => "66f502e5ff78bee2ebd70b6e04be1c6cb5d78a31" (forces new resource)

+ aws_instance.kubemaster.2
    user_data:                        "" => "66f502e5ff78bee2ebd70b6e04be1c6cb5d78a31"

Terraform v0.6.13

[Fodoj]

My favorit bug so far. Was reported so may times already in November 2015: #3885 Shame it still gets zero attention from core team

[apparentlymart]

Hi @MansM,

As @Fodoj mentioned this is an implication of the bug being discussed in #3449. This is not being quickly addressed because it represents a tricky architectural challenge for Terraform. To be specific, the diffing code sees that all of the rendered templates are being used in user_data and thus applies the "computed" state to the result, even though as users we know that element will actually ignore all but a single item from that set. To fix this fully-generally, the diffing code needs a special case for the element function.

However, for this particular case where it's multiple template_file resources you're trying to use, it's likely that this would be resolved by a different architectural change #4169, which would recast template_file as a "data source" rather than a resource, and that means its rendered result would be available immediately on the initial plan, rather than having to wait until this pseudo-resource is "created". This latter issue is being worked on over in #4961, and is planned to be included in the 0.7 minor release along with some other trickier new capabilities.

A potential workaround for this in the mean time is to use the -target argument to terraform apply to force Terraform to update your template_file resources in isolation first, and then subsequently run terraform plan again without a -target to update the aws_instance resources based on the newly-rendered templates. I think something like the following should get your new template_file rendered so that Terraform will not treat it as "computed" on subsequent runs:

$ terraform apply -target=template_file.masters_cloud_init

For now I'm going to close this issue to consolidate discussions in #3449 and #4169 respectively. Thanks for taking the time to file this and I'm sorry that this case is so tricky right now.

[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.