terraform/website/docs/language/functions/setproduct.html.md

5.9 KiB

layout page_title sidebar_current description
language setproduct - Functions - Configuration Language docs-funcs-collection-setproduct The setproduct function finds all of the possible combinations of elements from all of the given sets by computing the cartesian product.

setproduct Function

The setproduct function finds all of the possible combinations of elements from all of the given sets by computing the Cartesian product.

setproduct(sets...)

This function is particularly useful for finding the exhaustive set of all combinations of members of multiple sets, such as per-application-per-environment resources.

> setproduct(["development", "staging", "production"], ["app1", "app2"])
[
  [
    "development",
    "app1",
  ],
  [
    "development",
    "app2",
  ],
  [
    "staging",
    "app1",
  ],
  [
    "staging",
    "app2",
  ],
  [
    "production",
    "app1",
  ],
  [
    "production",
    "app2",
  ],
]

You must pass at least two arguments to this function.

Although defined primarily for sets, this function can also work with lists. If all of the given arguments are lists then the result is a list, preserving the ordering of the given lists. Otherwise the result is a set. In either case, the result's element type is a list of values corresponding to each given argument in turn.

Examples

There is an example of the common usage of this function above. There are some other situations that are less common when hand-writing but may arise in reusable module situations.

If any of the arguments is empty then the result is always empty itself, similar to how multiplying any number by zero gives zero:

> setproduct(["development", "staging", "production"], [])
[]

Similarly, if all of the arguments have only one element then the result has only one element, which is the first element of each argument:

> setproduct(["a"], ["b"])
[
  [
    "a",
    "b",
  ],
]

Each argument must have a consistent type for all of its elements. If not, Terraform will attempt to convert to the most general type, or produce an error if such a conversion is impossible. For example, mixing both strings and numbers results in the numbers being converted to strings so that the result elements all have a consistent type:

> setproduct(["staging", "production"], ["a", 2])
[
  [
    "staging",
    "a",
  ],
  [
    "staging",
    "2",
  ],
  [
    "production",
    "a",
  ],
  [
    "production",
    "2",
  ],
]

Finding combinations for for_each

The resource for_each and dynamic block language features both require a collection value that has one element for each repetition.

Sometimes your input data comes in separate values that cannot be directly used in a for_each argument, and setproduct can be a useful helper function for the situation where you want to find all unique combinations of elements in a number of different collections.

For example, consider a module that declares variables like the following:

variable "networks" {
  type = map(object({
    base_cidr_block = string
  }))
}

variable "subnets" {
  type = map(object({
    number = number
  }))
}

If the goal is to create each of the defined subnets per each of the defined networks, creating the top-level networks can directly use var.networks because it's already in a form where the resulting instances match one-to-one with map elements:

resource "aws_vpc" "example" {
  for_each = var.networks

  cidr_block = each.value.base_cidr_block
}

However, in order to declare all of the subnets with a single resource block, we must first produce a collection whose elements represent all of the combinations of networks and subnets, so that each element itself represents a subnet:

locals {
  # setproduct works with sets and lists, but our variables are both maps
  # so we'll need to convert them first.
  networks = [
    for key, network in var.networks : {
      key        = key
      cidr_block = network.cidr_block
    }
  ]
  subnets = [
    for key, subnet in var.subnets : {
      key    = key
      number = subnet.number
    }
  ]

  network_subnets = [
    # in pair, element zero is a network and element one is a subnet,
    # in all unique combinations.
    for pair in setproduct(local.networks, local.subnets) : {
      network_key = pair[0].key
      subnet_key  = pair[1].key
      network_id  = aws_vpc.example[pair[0].key].id

      # The cidr_block is derived from the corresponding network. See the
      # cidrsubnet function for more information on how this calculation works.
      cidr_block = cidrsubnet(pair[0].cidr_block, 4, pair[1].number)
    }
  ]
}

resource "aws_subnet" "example" {
  # local.network_subnets is a list, so we must now project it into a map
  # where each key is unique. We'll combine the network and subnet keys to
  # produce a single unique key per instance.
  for_each = {
    for subnet in local.network_subnets : "${subnet.network_key}.${subnet.subnet_key}" => subnet
  }

  vpc_id            = each.value.network_id
  availability_zone = each.value.subnet_key
  cidr_block        = each.value.cidr_block
}

The above results in one subnet instance per combination of network and subnet elements in the input variables.

  • contains tests whether a given list or set contains a given element value.
  • flatten is useful for flattening hierarchical data into a single list, for situations where the relationships between two object types are defined explicitly.
  • setintersection computes the intersection of multiple sets.
  • setsubtract computes the relative complement of two sets
  • setunion computes the union of multiple sets.