CODEWITHSUNDEEP
amazon-s3terraformterraform-provider-aws
Riley Hun
Riley Hun

Reputation: 2775

Terraform: How to loop through list of elements and create resources based on each element?

I need to setup aws s3 access points for each data_uri passed in the inference variable; this is to provide a cross-account uri(s). I need to parse out just the bucket name from each of these data_uri and then create resources for each. How would I go about doing that?

Here is what I have so far:

resource "aws_s3_access_point" "s3_access_point" {
  count    = var.create ? 1 : 0
  for_each = var.inference

  bucket = split("/", replace(each.value.image_uri, "s3://", ""))[0]
  name   = format("%s-%s", split("/", replace(each.value.image_uri, "s3://", ""))[0], "-access-point")
}

The variable would look like this:

{
"inference": [
  {
  "data_uri": "s3://my_bucket/model.tar.gz"
  },
  {
  "data_uri": "s3://my_bucket_2/model.tar.gz"
  },
  {
  "data_uri": "s3://my_bucket_3/model.tar.gz"
  }
]
}

Upvotes: 1

Views: 1095

Answers (2)

theherk
theherk

Reputation: 7546

I would recommend using split if the naming is this consistent. Also, you cannot mix count and for_each; use one or the other as the case requires.

note: This answer has been reworked based on the comments below

locals {
  inference = [
    { "data_uri" : "s3://my_bucket/model.tar.gz" },
    { "data_uri" : "s3://my_bucket_2/model.tar.gz" },
    { "data_uri" : "s3://my_bucket_3/model.tar.gz" }
  ]

  uri_bucket_map = {
    for x in local.inference : x.data_uri =>
    split("/", split("//", x.data_uri)[1])[0]
  }
}

resource "aws_s3_access_point" "s3_access_point" {
  for_each = local.uri_bucket_map

  bucket = each.value
  name   = var.s3_access_point_name
}

output "s3_access_point" {
  value = { for uri, ap in aws_s3_access_point.s3_access_point : uri => ap.arn }
}

In this iteration, we generate a map from data_uri to bucket_name so that you can access both when needed. You could of course still just use the toset version of the uri's and calculate the bucket within the resources, but since you preferred this method so you can access the bucket names elsewhere, I went with the map.

After the resources are created, we then generate an output that is a map from data_uri to access point arn.

Upvotes: 3

Martin Atkins
Martin Atkins

Reputation: 74084

The documentation for Terraform's regex function includes an example containing a pattern for matching the scheme and authority parts of URLs, which is a simplification of the pattern given in RFC 3986 Appendix B:

> regex("^(?:(?P<scheme>[^:/?#]+):)?(?://(?P<authority>[^/?#]*))?", "https://terraform.io/docs/")
{
  "authority" = "terraform.io"
  "scheme" = "https"
}

These S3 URLs seem to follow the usual URL productions and so one way to parse these would be to run them through that same regex pattern:

> regex("^(?:(?P<scheme>[^:/?#]+):)?(?://(?P<authority>[^/?#]*))?", "s3://my_bucket/model.tar.gz")
{
  "authority" = "my_bucket"
  "scheme" = "s3"
}

To do this I'd first derive a new value from local.inference which expands out those URLs:

locals {
  data_uris = [
    for o in local.inference : merge(
      {uri = o.data_uri},
      regex("^(?:(?P<scheme>[^:/?#]+):)?(?://(?P<authority>[^/?#]*))?", o.data_uri),
    )
  ]
}

The merge in the above is to combine the regex results with the original input URIs to get a single simple object for each of the inputs:

[
  {
    uri       = "s3://my_bucket/model.tar.gz",
    scheme    = "s3"
    authority = "my_bucket"
  },
  {
    uri       = "s3://my_bucket_2/model.tar.gz",
    scheme    = "s3"
    authority = "my_bucket_2"
  },
  {
    uri       = "s3://my_bucket_3/model.tar.gz",
    scheme    = "s3"
    authority = "my_bucket_3"
  },
]

We can then do a little more filtering/projection inside for_each to transform this into a set of bucket names, to filter them all out if var.create isn't set, and to skip any that aren't S3 URLs:

resource "aws_s3_access_point" "s3_access_point" {
  for_each = toset([
    for ap in local.data_uris : ap.authority
    if var.enabled && ap.scheme == "s3"
  ])

  bucket = each.key
  name   = "${each.key}-access-point"
}

The use of regex here is admittedly not ideal because it's pretty unclear what that regex pattern is doing. If you move forward with this then I would suggest including source code comments noting that this regex is parsing the URI and possibly linking to RFC 3986 Appendix B for its explanation of the pattern.

Upvotes: 1

Related Questions