Bug: Idempotency Issue with Disk Block

[trfore]

I am running into a idempotency issue whenever an ide drive is added, Terraform will attempt to reshuffle all attached disk on all subsequent calls to terraform <plan/apply>.

• So for a baseline, if you add two scsi disks testing for idempotency will pass.
• Once you add an ide drive (cdrom or cloudinit), the shuffling will occur.
• If you only have two ide drives (cdrom), the shuffling does not occur, however, the issue described in New Disk feature block issue #1090 occurs.

This is problematic when you want to add a cloud-init drive to VMs and occurs in the 3.0.1-rc4 release, master and newdisk branches.

Code Example

terraform {
  required_providers {
    proxmox = {
      source  = "Telmate/proxmox"
      version = "3.0.1-rc4"
    }
  }
}

provider "proxmox" {
  pm_api_url          = "REDACTED"
  pm_api_token_id     = "REDACTED"
  pm_api_token_secret = "REDACTED"
}

resource "proxmox_vm_qemu" "vm" {
  target_node = "pve"
  vmid        = "100"
  vm_state    = "stopped"

  disk {
    type    = "disk"
    slot    = "scsi0"
    storage = "local-lvm"
    size    = "8G"
  }

  disk {
    backup  = false
    type    = "cloudinit"
    slot    = "ide2"
    storage = "local-lvm"
  }
}

Terraform Output

Log output after initial apply:

Terraform will perform the following actions:

  # proxmox_vm_qemu.vm will be updated in-place
  ~ resource "proxmox_vm_qemu" "vm" {
        id                     = "pve/qemu/100"
        # (32 unchanged attributes hidden)

      ~ disk {
          ~ backup               = false -> true
          + format               = "raw"
            id                   = 0
          + size                 = "8G"
          ~ slot                 = "ide2" -> "scsi0"
          ~ type                 = "cloudinit" -> "disk"
            # (18 unchanged attributes hidden)
        }
      ~ disk {
          ~ backup               = true -> false
            id                   = 0
          ~ slot                 = "scsi0" -> "ide2"
          ~ type                 = "disk" -> "cloudinit"
            # (20 unchanged attributes hidden)
        }

        # (1 unchanged block hidden)
    }

Plan: 0 to add, 1 to change, 0 to destroy.

[Tinyblargon]

@trfore the shuffling in the Terraform state is due to the complexity of the disk block. Putting the disks in by id in alphanumeric irder seems to mostly fix this. This is due to a limitation in Terraform, and partially the reason why the disks block was introduced.

As far as I'm aware, the old disk block had this same problem but would actually mutate the vm.

FYI, we currently use the ordered list type for disk. I've tried many times to get the unordered list working. The ordered list allows us to somewhat mitigate this with the alphanumeric id ordering, whereas the unordered list makes it completely random.

If someone knows a terraform provider that handles these big objects in an unordered list please point me to it.