feat: render Talos bootstrap configuration

terraform/.gitignore

@@ -11,7 +11,8 @@ crash.log
 crash.*.log
 
 # Ignore local Terraform configuration files that can contain sensitive data.
-#terraform.tfvars
+terraform.tfvars
+.maidn/
 
 # Ignore override files as they are typically used for local testing.
 # These should not be checked in to source control.

terraform/locals.tf

@@ -11,7 +11,7 @@ locals {
   all_networks_flat = flatten([
     for node in var.nodes : [
       for network in node.networks : {
-        proxmox_node = lookup(node, "proxmox_node", var.proxmox_node)
+        proxmox_node = coalesce(node.proxmox_node, var.proxmox_node)
         vlan_id      = network.vlan_id
       }
     ]
@@ -57,10 +57,10 @@ locals {
       cores                   = node.cores
       memory                  = node.memory
       role                    = node.role
-      proxmox_node            = lookup(node, "proxmox_node", var.proxmox_node)
+      proxmox_node            = coalesce(node.proxmox_node, var.proxmox_node)
       disk_size               = node.disk_size
       disk_storage            = var.disk_storage
-      tags                    = lookup(node, "tags", [node.role])
+      tags                    = coalesce(node.tags, [node.role])
       iso_file                = local.node_iso_files[node.name]
       additional_disk_size    = lookup(node, "additional_disk_size", null)
       additional_disk_storage = lookup(node, "additional_disk_size", null) != null ? var.additional_disk_storage : null

terraform/main.tf

@@ -21,7 +21,7 @@ resource "null_resource" "stage_talos_image" {
 
   triggers = {
     node_name     = each.value.name
-    proxmox_node  = lookup(each.value, "proxmox_node", var.proxmox_node)
+    proxmox_node  = coalesce(each.value.proxmox_node, var.proxmox_node)
     image_url     = local.talos_iso_download_url
     storage       = var.talos_image_storage
     talos_version = var.talos_version
@@ -31,19 +31,26 @@ resource "null_resource" "stage_talos_image" {
 
   provisioner "local-exec" {
     interpreter = ["python", "-c"]
-    command     = "import subprocess; subprocess.run(['python', r'${path.module}/scripts/stage_talos_image.py', '--api-url', '${var.proxmox_api_url}', '--api-token', '${var.proxmox_api_token}', '--node', '${self.triggers.proxmox_node}', '--storage', '${var.talos_image_storage}', '--image-url', '${local.talos_iso_download_url}', '--filename', '${self.triggers.filename}', '--force'], check=True)"
+    environment = {
+      MAIDN_PROXMOX_API_TOKEN = var.proxmox_api_token
+    }
+    command = "import subprocess; subprocess.run(['python', r'${path.module}/scripts/stage_talos_image.py', '--api-url', '${var.proxmox_api_url}', '--node', '${self.triggers.proxmox_node}', '--storage', '${var.talos_image_storage}', '--image-url', '${local.talos_iso_download_url}', '--filename', '${self.triggers.filename}'], check=True)"
   }
 }
 
 # Create all required network bridges
 resource "proxmox_virtual_environment_network_linux_bridge" "cluster_bridge" {
-  for_each = local.unique_bridges
+  for_each = var.manage_network_bridges ? local.unique_bridges : {}
 
   node_name  = each.value.proxmox_node
   name       = each.value.bridge_name
   comment    = "Auto-created bridge for Talos VLAN ${each.value.vlan_id} on ${each.value.proxmox_node}"
   ports      = var.create_vlan_interface ? ["${each.value.phys_iface}.${each.value.vlan_id}"] : [each.value.phys_iface]
   vlan_aware = false
+
+  lifecycle {
+    prevent_destroy = true
+  }
 }
 
 # Create the Virtual Machines

terraform/scripts/stage_talos_image.py

@@ -1,5 +1,6 @@
 import argparse
 import json
+import os
 import ssl
 import subprocess
 import sys
@@ -82,13 +83,15 @@ def wait_for_storage_content(base_url, token, node, storage, filename):
 def main():
     parser = argparse.ArgumentParser()
     parser.add_argument("--api-url", required=True)
-    parser.add_argument("--api-token", required=True)
+    parser.add_argument("--api-token", default=os.environ.get("MAIDN_PROXMOX_API_TOKEN"))
     parser.add_argument("--node", required=True)
     parser.add_argument("--storage", required=True)
     parser.add_argument("--image-url", required=True)
     parser.add_argument("--filename", required=True)
     parser.add_argument("--force", action="store_true")
     args = parser.parse_args()
+    if not args.api_token:
+        parser.error("--api-token or MAIDN_PROXMOX_API_TOKEN is required")
 
     try:
         filename, downloaded = ensure_image(args.api_url, args.api_token, args.node, args.storage, args.image_url, args.filename, args.force)

terraform/talconfig_generator.tf

@@ -7,6 +7,8 @@ resource "local_file" "talconfig" {
     cluster_endpoint           = local.cluster_endpoint
     cluster_domain             = var.cluster_domain
     talos_version              = var.talos_version
+    kubernetes_version         = var.kubernetes_version
+    architecture               = var.talos_image_architecture
     talos_factory_schematic_id = var.talos_factory_schematic_id
     control_plane_vip          = var.control_plane_vip
 
@@ -20,34 +22,34 @@ resource "local_file" "talconfig" {
       name         = node.name,
       role         = node.role,
       networks     = node.networks,
-      tags         = node.tags,
+      tags         = coalesce(node.tags, []),
-      proxmox_node = node.proxmox_node
+      proxmox_node = coalesce(node.proxmox_node, var.proxmox_node)
     } if node.role == "controlplane"]
 
     compute_workers = [for node in var.nodes : {
       name         = node.name,
       role         = node.role,
       networks     = node.networks,
-      tags         = node.tags,
+      tags         = coalesce(node.tags, []),
-      proxmox_node = node.proxmox_node
+      proxmox_node = coalesce(node.proxmox_node, var.proxmox_node)
-    } if node.role == "worker" && contains(lookup(node, "tags", []), "compute")]
+    } if node.role == "worker" && !contains(coalesce(node.tags, []), "storage") && !contains(coalesce(node.tags, []), "gpu")]
 
     storage_workers = [for node in var.nodes : {
       name                 = node.name,
       role                 = node.role,
       networks             = node.networks,
-      tags                 = node.tags,
+      tags                 = coalesce(node.tags, []),
-      proxmox_node         = node.proxmox_node,
+      proxmox_node         = coalesce(node.proxmox_node, var.proxmox_node),
       additional_disk_size = lookup(node, "additional_disk_size", "")
-    } if node.role == "worker" && contains(lookup(node, "tags", []), "storage")]
+    } if node.role == "worker" && contains(coalesce(node.tags, []), "storage")]
 
     gpu_workers = [for node in var.nodes : {
       name         = node.name,
       role         = node.role,
       networks     = node.networks,
-      tags         = node.tags,
+      tags         = coalesce(node.tags, []),
-      proxmox_node = node.proxmox_node
+      proxmox_node = coalesce(node.proxmox_node, var.proxmox_node)
-    } if node.role == "worker" && contains(lookup(node, "tags", []), "gpu")]
+    } if node.role == "worker" && contains(coalesce(node.tags, []), "gpu")]
 
     # Other variables
     cluster_pod_nets = var.cluster_pod_nets
@@ -63,16 +65,16 @@ output "node_organization" {
   description = "How your nodes are organized for Talos configuration"
   value = {
     control_plane_nodes = [for node in var.nodes : {
-      name = node.name, ip = node.networks[0].ip, tags = lookup(node, "tags", [])
+      name = node.name, ip = node.networks[0].ip, tags = coalesce(node.tags, [])
     } if node.role == "controlplane"]
     compute_workers = [for node in var.nodes : {
-      name = node.name, ip = node.networks[0].ip, tags = lookup(node, "tags", [])
+      name = node.name, ip = node.networks[0].ip, tags = coalesce(node.tags, [])
-    } if node.role == "worker" && contains(lookup(node, "tags", []), "compute")]
+    } if node.role == "worker" && !contains(coalesce(node.tags, []), "storage") && !contains(coalesce(node.tags, []), "gpu")]
     storage_workers = [for node in var.nodes : {
-      name = node.name, ip = node.networks[0].ip, tags = lookup(node, "tags", []), additional_disk = lookup(node, "additional_disk_size", null)
+      name = node.name, ip = node.networks[0].ip, tags = coalesce(node.tags, []), additional_disk = node.additional_disk_size
-    } if node.role == "worker" && contains(lookup(node, "tags", []), "storage")]
+    } if node.role == "worker" && contains(coalesce(node.tags, []), "storage")]
     gpu_workers = [for node in var.nodes : {
-      name = node.name, ip = node.networks[0].ip, tags = lookup(node, "tags", [])
+      name = node.name, ip = node.networks[0].ip, tags = coalesce(node.tags, [])
-    } if node.role == "worker" && contains(lookup(node, "tags", []), "gpu")]
+    } if node.role == "worker" && contains(coalesce(node.tags, []), "gpu")]
   }
 }

terraform/templates/talconfig.yaml.tpl

@@ -2,6 +2,7 @@ clusterName: ${cluster_name}
 endpoint: ${cluster_endpoint}
 domain: ${cluster_domain}
 talosVersion: ${talos_version}
+kubernetesVersion: "${kubernetes_version}"
 cniConfig:
   name: ${cni_name}
 allowSchedulingOnMasters: ${length(control_plane_nodes) == 1 ? "true" : "false"}
@@ -26,7 +27,7 @@ nodes:
     installDisk: /dev/sda
     machineSpec:
       mode: metal
-      arch: amd64
+      arch: ${architecture}
     talosImageURL: "factory.talos.dev/installer/${talos_factory_schematic_id}"
     patches:
       - |-
@@ -70,7 +71,7 @@ nodes:
     installDisk: /dev/sda
     machineSpec:
       mode: metal
-      arch: amd64
+      arch: ${architecture}
     talosImageURL: "factory.talos.dev/installer/${talos_factory_schematic_id}"
     patches:
       - |-
@@ -111,7 +112,7 @@ nodes:
     installDisk: /dev/sda
     machineSpec:
       mode: metal
-      arch: amd64
+      arch: ${architecture}
     talosImageURL: "factory.talos.dev/installer/${talos_factory_schematic_id}"
     patches:
       - |-
@@ -152,7 +153,7 @@ nodes:
     installDisk: /dev/sda
     machineSpec:
       mode: metal
-      arch: amd64
+      arch: ${architecture}
     talosImageURL: "factory.talos.dev/installer/${talos_factory_schematic_id}"
     patches:
       - |-

terraform/variables.tf

@@ -50,6 +50,12 @@ variable "cni_name" {
   default     = "none"
 }
 
+variable "kubernetes_version" {
+  description = "The Kubernetes version rendered into Talos configuration."
+  type        = string
+  default     = "v1.33.4"
+}
+
 # --- Global Network Configuration ---
 variable "dns_servers" {
   description = "Global DNS servers for the cluster"
@@ -133,7 +139,7 @@ variable "nodes" {
       mac_address = string
       cidr        = string
       vlan_id     = number
-      ip          = optional(string) # Static IP, required for control plane, optional for workers
+      ip          = optional(string) # Static IP required for every rendered Talos node
       gateway     = optional(string) # Should only be defined on the primary interface
     }))
   }))
@@ -144,11 +150,17 @@ variable "nodes" {
   }
 
   validation {
-    condition     = alltrue([for node in var.nodes : node.role == "controlplane" ? node.networks[0].ip != null : true])
+    condition     = alltrue([for node in var.nodes : node.networks[0].ip != null])
-    error_message = "The first network for a 'controlplane' node must have a static IP address."
+    error_message = "The first network for every node must have a static IP address."
   }
 }
 
+variable "manage_network_bridges" {
+  description = "Whether Terraform owns Proxmox host bridges. Leave false for shared or pre-existing bridges."
+  type        = bool
+  default     = false
+}
+
 variable "default_physical_interface" {
   description = "Default physical interface for nodes (fallback)"
   type        = string