Data source external with dependencies is being evaluated early

[Lasering]

Terraform Version

Terraform v0.11.2
+ provider.external v1.0.0
+ provider.null v1.0.0

Terraform Configuration Files

These are defined inside a module called consul.

module "A" {
  source       = "./region"
  region       = "RegionA"
  environment  = "${var.environment}"
  project-name = "${var.project-name}"
  number-of-servers = "${var.number-of-servers}"
}

module "T" {
  source       = "./region"
  region       = "RegionT"
  environment  = "${var.environment}"
  project-name = "${var.project-name}"
  number-of-servers = "${var.number-of-servers}"
}

locals {
  both-regions-server-ips   = "${concat(module.A.server-ips, module.T.server-ips)}"
  both-regions-server-fqdns = "${concat(module.A.server-fqdns, module.T.server-fqdns)}"
}

resource "null_resource" "provision-both-clusters" {
  count = "${var.number-of-servers * 2}" # * 2 Because we have RegionA and RegionT

  connection {
    host = "${element(local.both-regions-server-ips, count.index)}"

    user        = "${var.ssh-username}"
    private_key = "${file(".${local.keypair}.pem")}"
  }

  provisioner "chef" {
    # Omitted for clarity
  }
}

data "external" "vault-tokens" {
  depends_on = ["null_resource.provision-both-clusters"]
  
  program = ["${path.module}/create-vault-tokens.sh"]

  query = {
    consulAddress = "https://${element(module.tagus.server-ips, 0)}:8500"
    masterToken   = "${var.master-token}"
  }
}

output "vault-backend-token" {
  value     = "${data.external.vault-tokens.result.backend-token}"
  sensitive = true
}

output "vault-consul-auth-backend-token" {
  value     = "${data.external.vault-tokens.result.consul-auth-backend-token}"
  sensitive = true
}

Expected Behavior

The output variables values should only be computed after provision-both-clusters in ran.

Actual Behavior

Error: Error running plan: 2 error(s) occurred:

* module.consul.output.vault-consul-auth-backend-token: Resource 'data.external.vault-tokens' does not have attribute 'result.consul-auth-backend-token' for variable 'data.external.vault-tokens.result.consul-auth-backend-token'
* module.consul.output.vault-backend-token: Resource 'data.external.vault-tokens' does not have attribute 'result.backend-token' for variable 'data.external.vault-tokens.result.backend-token'

The same behavior happens even if I define vault-tokens to be:

data "external" "vault-tokens" {
  program = ["${path.module}/create-vault-tokens.sh"]

  query = {
    consulAddress = "https://${element(module.tagus.server-ips, 0)}:8500"
    masterToken   = "${var.master-token}"
    dummy = "${null_resource.provision-both-clusters.count}"
  }
}

Like described in #10603

If I remove the output variables Terraform resolves the graph correctly and only runs the vault-tokens after the null_resource.

If I set a depends_on = ["null_resource.provision-both-clusters"] on both output variables I still get the error.

Steps to Reproduce

1. terraform init
2. terraform plan

[Lasering]

If I change the output variables to be:

output "vault-backend-token" {
  value     = "${data.external.vault-tokens.result["backend-token"]}"
  sensitive = true
}

output "vault-consul-auth-backend-token" {
  value     = "${data.external.vault-tokens.result["consul-auth-backend-token"]}"
  sensitive = true
}

It works.

If I change to:

output "vault-backend-token" {
  value     = "${data.external.vault-tokens.result.backend-token}"
  sensitive = true
}

output "vault-consul-auth-backend-token" {
  value     = "${data.external.vault-tokens.result.consul-auth-backend-token}"
  sensitive = true
}

It does not work, which is unexpected.

[danvaida]

FWIW, I can reproduce this on v0.11.7 while having a bash script as an external data source. Changing to the square brackets notation from dot, does indeed make the error go away.

[Phylu]

I just reproduced this issue with Terraform v0.11.10. Thanks for the hint with the brackets, @Lasering. That worked as a workaround.

[ervinb]

To work around this, you can also use lookup():

output "vault-backend-token" {
  value     = "${lookup(data.external.vault-tokens.result, "backend_token")}"
  sensitive = true
}

hashicorp/terraform-provider-external#4

[akumadare]

Hi,
I have an issue along these lines where I am using an external provider (a bash script) to create an SSL certificate with the common name based on the dns_name attribute of an aws_alb:

data "external" "ssl_webcert_generator" {
  program = ["bash", "${path.module}/externalproviders/generatessl.sh"]
  query = {
    cn = "${aws_alb.ecs-load-balancer.dns_name}"
    pass = "${var.sslpass}"
  }
}
resource "aws_iam_server_certificate" "lb_cert" {
  depends_on = ["data.external.ssl_webcert_generator"]
  name             = "lb_cert"
  certificate_body = "${data.external.ssl_webcert_generator.result.public_cert_contents}"
  private_key      = "${data.external.ssl_webcert_generator.result.private_cert_contents}"
}

This gives an error on terraform plan:

aws_iam_server_certificate.lb_cert: Resource 'data.external.ssl_webcert_generator' does not have attribute 'result.public_cert_contents' for variable 'data.external.ssl_webcert_generator.result.public_cert_contents'

If i comment the aws_iam_server_certificate resource out, the stack can be launched.

Once the stack is launched the aws_iam_server_certificate resource can be uncommented, the config can be reapplied and the stack updates with this resource

Any ideas on how i can the stack to launch in one step without this workaround?

[Phylu]

It should work like this:

resource "aws_iam_server_certificate" "lb_cert" {
  depends_on = ["data.external.ssl_webcert_generator"]
  name             = "lb_cert"
  certificate_body = "${data.external.ssl_webcert_generator.result["public_cert_contents"]}"
  private_key      = "${data.external.ssl_webcert_generator.result["private_cert_contents"]}"
}

[Lasering]

@akumadare why not use https://www.terraform.io/docs/providers/tls/r/locally_signed_cert.html?

[akumadare]

Thanks both. I hadnt seen the support for locally signed certs in the tls provider - it would make much more sense to use this!

[Vlaaaaaaad]

As an FYI, what @Lasering said in #17173 (comment) happens for EKS auth too when using modules.

Until hashicorp/terraform-provider-aws#4904 is merged a workaround proposed in hashicorp/terraform-provider-kubernetes#161 needs to be used to use the kubernetes provider with an AWS EKS cluster.

I initially had everything in a single folder, without any modules:

data "external" "aws_iam_authenticator" {
  program = ["sh", "-c", "aws-iam-authenticator token -i ${var.cluster_name} | jq -r -c .status"]
}

provider "kubernetes" {
  host = "${var.cluster_endpoint}"
  cluster_ca_certificate = "${base64decode(var.cluster_certificate_authority_data)}"
  token  = "${data.external.aws_iam_authenticator.result.token}"
  load_config_file = false
}

As soon as I moved to a module, I had to switch to ${data.external.aws_iam_authenticator.result["token"]} instead of ${data.external.aws_iam_authenticator.result.token} otherwise I would get an Resource 'data.external.aws_iam_authenticator' does not have attribute 'result.token' for variable 'data.external.aws_iam_authenticator.result.token' error:

data "external" "aws_iam_authenticator" {
  program = ["sh", "-c", "aws-iam-authenticator token -i ${var.cluster_name} | jq -r -c .status"]
}

data "aws_region" "current" {}

provider "kubernetes" {
  host                   = "${var.cluster_endpoint}"
  cluster_ca_certificate = "${base64decode(var.cluster_certificate_authority_data)}"

  # This can actually be
  #   token  = "${data.external.aws_iam_authenticator.result.token}"
  #   but that fails due to... nobody knows why
  #  see https://github.com/hashicorp/terraform/issues/17173
  token = "${data.external.aws_iam_authenticator.result["token"]}"

  load_config_file = false
}

This is... weird.

[queglay]

I am getting a strikingly similar issue to this. I have multiple null resources that run as a linear stream of dependencies, but they all seem to ignore the dependencies and just run at once. v0.11.13

[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.