For_each on resources causing recreation when item removed from list

Terraform 0.12.6

Code snippet

data "aws_ec2_transit_gateway_vpn_attachment" "vpn_propagations" {
  for_each = { for v in var.vpn_propagations : v => v }
  transit_gateway_id = var.transit_gateway_id
  vpn_connection_id  = v.value

resource "aws_ec2_transit_gateway_route_table_propagation" "vpn_propagations" {
  for_each                       = { for v in var.vpn_propagations : v => v }
  transit_gateway_attachment_id  = data.aws_ec2_transit_gateway_vpn_attachment.vpn_propagations[v.key].id
  transit_gateway_route_table_id =

Terraform plan - after an item is removed from vpn_propagations[ ]

Not sure why terraform is trying to replace vpn_propagations[2]. Shouldn’t it destroy vpn_propagations[2] and move vpn_propagations[3] to [2] in state file?

(Or am I using for_each the wrong way?)

1 Like

Hi @arunamat!

It seems that something strange is going on here, because a resource using for_each should never have numeric instance keys like [0], [1], etc. Instead, it should have string keys like ["foo"] where foo is a key from your map.

Is it possible that the resources being planned here are from a different module than the one whose code you shared here? Perhaps it’s an older version of the module that’s still using count? :thinking:

Hi @apparentlymart - They are from the same module.

State file snippet -

I believe the above state format is for for_each^^

When for_each is used, the "index_key" in the state appears as a string, so it still seems that something strange is happening here.

Is this the first time running terraform apply or terraform plan after switching from count to for_each?

No @apparentlymart.

Also I did do a terraform destroy and terraform apply with for_each and TF 0.12.16 - just to make sure that it was not something to do with the state migration.

1 Like

Hi again @arunamat!

I’m still not sure what exactly is going on here. Would you mind opening a bug report about this behavior? The various bits of information requested in that issue template should hopefully give the engineers on the Terraform team more information to understand better what exactly is going on here and fix it.

If you do open one, please post a link to it here for future readers of this thread to see, and also link to this thread from the issue so that the engineering team can see the context from the discussion we already had here.


Thanks @apparentlymart. After some debugging we figured out that our root module was pointing to the wrong code base using count . After switching to the right module version everything is working as expected. I’m really sorry for reporting a false issue.