Address_book as unordered list

[lukestu]

It seems the global address_book is reconfigured on every terraform plan even if only the order has changed.
Is it possible to have the address_book rendered as an unordered list, e.g. using Typeset rather than Typelist?

example:

     ~ address_set {
          ~ address = [
              - "net.aa01:cc01:1100::/40",
              + "net.bb02:dd01:500::/40",
            ]
          ~ name    = "site.net1" -> "site.net2"
        }
	  ~ address_set {
          ~ address = [
              - "net.bb02:dd01:500::/40",
              + "net.cc03:ee01:900::/40",
            ]
          ~ name    = "site.net2" -> "site.net3"
        }

Thanks again for your great work on this!

[jeremmfr]

Yes, ordering lists in global_address_book is unnecessary for Junos but is supported in its config and diff when run terraform plan is much more convenient.
Example with this config :

network_address {
    name  = "red"
    value = "192.0.2.3/32"
  }
  network_address {
    name  = "green"
    value = "192.0.2.2/32"
  }
  network_address {
    name  = "blue"
    value = "192.0.2.1/32"
  }
  address_set {
    name    = "color2"
    address = ["green", "blue"]
  }
  address_set {
    name    = "color1"
    address = ["green", "red", "blue"]
  }

When add red in color2 with have :

• plan with ordered list :

      ~ address_set {
         ~ address = [
               # (1 unchanged element hidden)
               "blue",
             + "red",
           ]
           name    = "color2"
       }

       # (4 unchanged blocks hidden)

• plan without ordered list :

      + address_set {
         + address = [
             + "green",
             + "blue",
             + "red",
           ]
         + name    = "color2"
       }
     - address_set {
         - address = [
             - "green",
             - "blue",
           ] -> null
         - name    = "color2" -> null
       }
     - address_set {
         - address = [
             - "green",
             - "red",
             - "blue",
           ] -> null
         - name    = "color1" -> null
       }
     + address_set {
         + address = [
             + "green",
             + "red",
             + "blue",
           ]
         + name    = "color1"
       }

       # (3 unchanged blocks hidden)

So I prefer to stay with an ordered list

[lukestu]

Fair enough.

The problem we face is that we have a large address book, 500+ records. If we add one more, then the terraform plan will show hundreds of changes simply for the one address addition. This makes it very hard to properly review the changes that will be made, before we apply them.

[jeremmfr]

Fair enough.

The problem we face is that we have a large address book, 500+ records. If we add one more, then the terraform plan will show hundreds of changes simply for the one address addition. This makes it very hard to properly review the changes that will be made, before we apply them.

You should add the new entry at the end of the resource definition to have a more readable plan

[tagur87]

I investigated this further myself...and there seems to be a pattern that is followed by the AWS provider that might be helpful.
https://github.com/hashicorp/terraform-provider-aws/blob/main/aws/resource_aws_security_group.go#L78

Seems we can use the Set: option to define a schema.SchemaSetFunc type, which would output a hash of name-value combination, to uniquely identify the individual entries.

I tested this a bit, but it does not seem to work as I would hope.

After doing a bit of digging, it seems that this issue still could exist with a TypeSet type. See the issue in the main terraform project for more information. Seems there is work being done to provide more rich type support to the SDK.

hashicorp/terraform#28281

[jeremmfr]

Following my investigation in issue #316, I found a workaround to avoid large update plan output.
So I reopen this issue to take care of making the change to an unordered block.