The "count" value depends on resource attributes that cannot be determined until apply

Terraform Version

1.0.0

main.tf

resource "random_id" "random_label" {
  byte_length = 2
}

module "a" {
  source    = "./child"
  name      = "a"
  //  label = null
  label = random_id.random_label.hex
}


output "main-test-out" {
  value = random_id.random_label.id
}

###child/main.tf

variable "name" {
  type = string
}

variable "label" {
  type = string
  default = null
}

resource "null_resource" "x" {
  count = var.label == null ? 1 : 0
  //depends_on = [var.label,]
}

output "ids" {
  value = null_resource.x.*.id
}

Crash Output

Error: Invalid count argument

on child/main.tf line 11, in resource "null_resource" "x":
11: count = var.label == null ? 1 : 0

The "count" value depends on resource attributes that cannot be determined until apply, so Terraform cannot predict how many instances will be created. To work around this, use the -target argument to first apply only the resources that the count depends on.

Explaination

I built a minimal example to show the problem I have. Is there any way to give a random label to a module, which uses this to decide whether to create a resource or not (using count)?

Expected Behavior

I expect that random_label gets created first and then count is applied.

Actual Behavior

Apparantly, count is applied first. Since random_label is needed to decide whether a resource is to be created or not, an error is thrown.

Steps to Reproduce

Please list the full steps required to reproduce the issue, for example:

1. terraform init
2. terraform apply

References

I believe this is the same issue as #26078

[apparentlymart]

Hi @yohanuta,

What you've seen here is the intended behavior of Terraform. An important purpose of creating a Terraform plan is to see how many objects will be created, and so Terraform has a rule that the count and for_each arguments must be determined only from values defined directly in the configuration, not from values that a provider will generate during the apply phase.

The random provider is a little unusual here in that it doesn't actually interact with a remote system and so in principle it could generate its random values during the plan step, but it seems to be designed to behave similarly to a more typical resource type that represents a remote API, where often the resulting values won't be known until the provider actually makes the call to create the corresponding object, and therefore those results would not be suitable to use with count or for_each.

To make this work will require designing your configuration to derive count and for_each only from values defined statically in the documentation. Since your example here seems to be a contrived one just to illustrate the situation I won't try to show an example here, but if you have a real situation where you'd like to discuss other designs that fit within the language rules then please start a topic in the community forum, where we can discuss your specific situation in more detail.

Since what you described is Terraform's intended behavior, I'm going to close this. Thanks!

[github-actions]

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.