[Feature] Allow disks to be managed individually

[smrqdt]

Currently, it is not possible to replace a VM while retaining a disk to reuse it with the new VM. It would be very useful, if the provider would allow to manage a VDI as a separate resource which could be referenced by id, similar to the cdrom block.

Motivation

We use Fedora CoreOS as guest OS, which is meant to be managed as immutable infrastructure. To change the configuration, the Ignition file has to be regenerated and the OS has to be redeployed.
To retain state (container volumes) we mount /var to a different disk, while the configuration is discarded and later rewritten by ignition.

Problem

The easiest way to do this would be to trigger a replacement of the VM by terraform. Sadly, currently this means, the disks will also be destroyed and new ones created.

At the moment we resort to leaving the VM as it is and manually triggering a reprovisioning (e.g. rebooting into PXE), which is tedious.

Solution

Therefore it would be useful if:

• an empty VDI could be created by the xenorchestra_vdi resource and
• a VDI could be referenced in a disk block

Example

resource "xenorchestra_vm" "vm" {
  # ....

  disk {
    sr_id      = data.xenorchestra_sr.sr.id
    name_label = "example_root"
    size       = 10 * pow(1024, 3)
  }

  disk {
    id = xenorchestra_vdi.var.id
  }
}

resource "xenorchestra_vdi" "var" {
    name_label = "example_var"
    sr_id = data.xenorchestra_sr.sr.id
    size = 10 * pow(1024, 3)
}

[ddelnano]

Hey @smrqdt, thanks for reaching out and providing a detailed feature request. I need to spend some time investigating this, but at a first glance it seems feasible to do and I agree it would be worthwhile.

[deefdragon]

Just adding another use case, I'm attempting to deploy some nixos machines using terraform and uploading VDIs. Ive been successful in getting VDIs uploaded and booting after manually attaching the disks, but without the ability to specify a disk by VDI ID in a VM, I'm stuck having to perform that step manually.

The direct goal for me is to be able to use terraform to deploy new kubernetes instances on demand, but given I can deploy theoretically ANY nixos config, this would open up handling basically all of my infra via terraform.

Im willing to take a look at doing some of the work as I know go, but I dont know if the API support is there.
If it is, Im willing to give it a crack.

[deefdragon]

I tried playing with the plugin to add VDI ID as an input parameter. While I could add said id easy enough, it appears that they are ignored on VM creation by the API, instead creating a new disk.

[gCyrille]

@deefdragon We are aware that this is a significant limitation of the current provider. It may be the next feature that we work on for this provider :-)