Consider allowing changes to VM resource's `cloud_config` argument to force recreation

[ddelnano]

This was discussed in Discord.

It is surprising that the cloud_config argument is capable of being changed on a xenorchestra_vm when it likely will have no effect. The aws_instance resource in the AWS terraform provider allows a similar configuration through launch templates. This argument causes the aws_instance resource to be recreated and so we should consider introducing this breaking change to fix this behavior and align it with how other providers treat cloud config / boot time configuration.

[4censord]

That is already possible with the lifecycle meta argument: using something like:

resource "xenorchestra_cloud_config" "user_data" { }
resource "xenorchestra_cloud_config" "network_data" { }

resource "xenorchestra_vm" "vm" {
  lifecycle {
    replace_triggered_by = [
      xenorchestra_cloud_config.user_data,
      xenorchestra_cloud_config.network_data,
    ]
  }
}

Terraform replaces the vm whenever user_data or network_data has any changes.

That looks something like this

# xenorchestra_vm.vm will be replaced due to changes in replace_triggered_by
-/+ resource "xenorchestra_vm" "vm" {
      - blocked_operations   = [] -> null
      ~ cloud_config         = {REDACTED}
      ~ id                   = {REDACTED} -> (known after apply)
      ~ ipv4_addresses       = [
          - {REDACTED},
        ] -> (known after apply)
      ~ ipv6_addresses       = [
          - {REDACTED},
        ] -> (known after apply)
      ~ power_state          = "Running" -> (known after apply)
        tags                 = [
            {REDACTED}
        ]
        # (16 unchanged attributes hidden)
      ~ cdrom {
          ~ id = {REDACTED} -> {REDACTED}
        }
      ~ disk {
          - attached   = true -> null
          ~ position   = "0" -> (known after apply)
          ~ vbd_id     = {REDACTED} -> (known after apply)
          ~ vdi_id     = {REDACTED} -> (known after apply)
            # (3 unchanged attributes hidden)
        }
      ~ network {
          - attached       = true -> null
          ~ device         = "0" -> (known after apply)
          ~ ipv4_addresses = [
              - {REDACTED},
            ] -> (known after apply)
          ~ ipv6_addresses = [
              - {REDACTED},
            ] -> (known after apply)
          ~ mac_address    = {REDACTED} -> (known after apply)
            # (1 unchanged attribute hidden)
        }
    }

[ddelnano]

@4censord thanks for sharing that workaround. I still think it makes sense to unify this against what other similar providers are doing, but it's great that Terraform has this flexibility.

[4censord]

I would also like to have the option for a change on the cloud-init data to force a power cycle of the vm. This does not seem possible with lifecycle configuration only.

[ddelnano]

to force a power cycle of the vm.

@4censord what I described in the issue description would be to cause the VM to be destroyed and recreated. Are you saying that if your case you want the VM to be rebooted?

[4censord]

Yes, I would like to have the option to only reboot the vm I propose to do this via an option that allows the user to choose:

resource "xenorchestra_vm" "vm" {
  on_cloudinit_change = "reboot" # stop the vm, do the change, start the vm
  on_cloudinit_change = "recreate" # destroy and recreate with the change
  on_cloudinit_change = "ignore" # just change it (current behavior, defaul)
  on_cloudinit_change = "fail" # Do not allow changes to the cloudinit config
}