[Question]: Replacing servers one-by-one when using `count`

[johnhamelink]

How might I use create_before_destroy to rotate out my servers one-by-one without losing them all in one go while I wait for the new servers to provision?

When I do this, I of course get a uniqueness error, because the new servers have the same name:

resource "hcloud_server" "app-nodejs-server" {
  name               = "${var.environment}-${var.app}-${var.service}-${count.index + 1}"
  count              = var.app-replicas

  lifecycle {
    create_before_destroy = true
  }
}

I'm using opentofu - could I somehow use for_each to do this? I'd appreciate some advice on this, thanks!

[jooola]

Hey,
I believe the easiest way for you to replace one node after the other is run your replace one by one by hand or using a script:

resource "hcloud_server" "nodes" {
  count = 3

  name        = "host${count.index + 1}"
  server_type = "cx22"
  image       = "debian-13"
  location    = "nbg1"
}

$ tofu state list                                                                                    
hcloud_server.nodes[0]
hcloud_server.nodes[1]
hcloud_server.nodes[2]

Replaces all nodes one by one:

$ tofu state list hcloud_server.nodes | xargs -i tofu apply -replace='{}' -auto-approve

Replace all nodes one by one interactively:

$ for address in $(tofu state list hcloud_server.nodes); do tofu apply -replace="$address"; done

create_before_destroy will likely not work as we require a unique resource name per project.

[johnhamelink]

Thank you, that's super useful. I'm going to implement this and see how I get on!