RKE has built-in support for (legacy) in-tree cloud providers deployment (AWS, Azure, OpenStack and vSphere), this limits deployment scenarios where a different cloud-provider is needed. Yagan should support out-tree cloud providers listed in kubernetes official repos through HCL-defined resources as well as allowing a bring your own CCM approach.

So defining a variable like cloud_provider = "openstack" along with credentials and additional configuration enables the openstack out-tree provider within RKE.

Currently this module consumes a map of nodes which contains tightly defined roles per node group

nodes = {
  controlplane = [
    {
      name = "node-1"
      ip = "192.168.2.123"
    }
  ]
  etcd = [
    {
      name = "node-2"
      ip = "192.168.2.25"
    }
  ]
  worker = [
    {
      name = "node-3"
      ip = "192.168.2.95"
    }
  ]
}

which is then consumed by

resource rke_cluster cluster {
...
  dynamic nodes {
    for_each = flatten(
      [
        for type, node in var.nodes : [
          for n in node : {
            ip   = n.ip
            name = n.name
            role = type
          }
        ]
      ]
    )

    content {
      address           = nodes.value.ip
      hostname_override = nodes.value.name
      internal_address  = nodes.value.ip
      node_name         = nodes.value.name
      role              = [nodes.value.role]
      ssh_key           = var.private_key
      user              = var.vm_user
    }
  }
...
}

this approach does not offer much flexibility if one needs to define multi-role nodes
(e.g.: one node for controlplane and etcd + one node for worker).

A more flexible way to define nodes and their roles is necessary, as well as setting labels for each one, this way needs
to be standard for yagan infrasctructure modules to come (e.g.: yagan-openstack, yagan-vsphere, etc)