fix: bring talos static links up

terraform/main.tf

@@ -26,56 +26,15 @@ resource "null_resource" "stage_talos_image" {
     storage       = var.talos_image_storage
     talos_version = var.talos_version
     filename      = local.node_iso_filenames[each.key]
+    stager        = filesha256("${path.module}/scripts/stage_talos_image.py")
   }
 
   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', '${self.triggers.filename}'], check=True)"
+    command     = "import subprocess; subprocess.run(['python', r'${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)"
   }
 }
 
-resource "proxmox_virtual_environment_file" "nocloud_network_config" {
-  for_each = { for node in var.nodes : node.name => node }
-
-  node_name    = lookup(each.value, "proxmox_node", var.proxmox_node)
-  datastore_id = var.talos_image_storage
-  content_type = "snippets"
-
-  source_raw {
-    data      = local_file.metal_network_config[each.key].content
-    file_name = "${each.key}-network-config.yaml"
-  }
-
-  depends_on = [local_file.metal_network_config]
-}
-
-resource "proxmox_virtual_environment_file" "nocloud_metadata" {
-  for_each = { for node in var.nodes : node.name => node }
-
-  node_name    = lookup(each.value, "proxmox_node", var.proxmox_node)
-  datastore_id = var.talos_image_storage
-  content_type = "snippets"
-
-  source_raw {
-    data      = "instance-id: ${each.key}\nlocal-hostname: ${each.key}\n"
-    file_name = "${each.key}-meta-data.yaml"
-  }
-}
-
-resource "proxmox_virtual_environment_file" "nocloud_user_data" {
-  for_each = { for node in var.nodes : node.name => node }
-
-  node_name    = lookup(each.value, "proxmox_node", var.proxmox_node)
-  datastore_id = var.talos_image_storage
-  content_type = "snippets"
-
-  source_raw {
-    data      = "#cloud-config\n"
-    file_name = "${each.key}-user-data.yaml"
-  }
-}
-
-
 # Create all required network bridges
 resource "proxmox_virtual_environment_network_linux_bridge" "cluster_bridge" {
   for_each = local.unique_bridges
@@ -162,21 +121,29 @@ resource "proxmox_virtual_environment_vm" "vm" {
 
   # Talos reads this Proxmox-generated cidata disk through its NoCloud platform.
   initialization {
-    datastore_id         = var.talos_image_storage
-    interface            = "ide3"
-    type                 = "nocloud"
-    meta_data_file_id    = proxmox_virtual_environment_file.nocloud_metadata[each.key].id
-    network_data_file_id = proxmox_virtual_environment_file.nocloud_network_config[each.key].id
-    user_data_file_id    = proxmox_virtual_environment_file.nocloud_user_data[each.key].id
+    datastore_id = var.disk_storage
+    interface    = "ide3"
+    type         = "nocloud"
+
+    dynamic "ip_config" {
+      for_each = each.value.ip != null ? [each.value] : []
+      content {
+        ipv4 {
+          address = "${ip_config.value.ip}/${split("/", ip_config.value.network_cidr)[1]}"
+          gateway = ip_config.value.gateway
+        }
+      }
+    }
+
+    dns {
+      servers = var.dns_servers
+    }
   }
 
   # Ensure image staging and bridge setup complete before VMs
   depends_on = [
     proxmox_virtual_environment_network_linux_bridge.cluster_bridge,
     null_resource.stage_talos_image,
-    proxmox_virtual_environment_file.nocloud_network_config,
-    proxmox_virtual_environment_file.nocloud_metadata,
-    proxmox_virtual_environment_file.nocloud_user_data,
   ]
 }
 

terraform/scripts/stage_talos_image.py

@@ -21,13 +21,18 @@ def api_request(base_url, token, method, path, data=None):
         return json.loads(response.read().decode())
 
 
-def ensure_image(base_url, token, node, storage, image_url, filename):
+def ensure_image(base_url, token, node, storage, image_url, filename, force=False):
     content_path = f"/api2/json/nodes/{node}/storage/{storage}/content"
     response = api_request(base_url, token, "GET", content_path)
 
     for item in response.get("data", []):
         if item.get("volid", "").endswith(filename):
-            return filename, False
+            if not force:
+                return filename, False
+            volume_path = urllib.parse.quote(f"iso/{filename}", safe="")
+            task = api_request(base_url, token, "DELETE", f"{content_path}/{volume_path}")
+            wait_for_task(base_url, token, node, task.get("data"))
+            break
 
     download_path = f"/api2/json/nodes/{node}/storage/{storage}/download-url"
     task = api_request(base_url, token, "POST", download_path, {
@@ -82,10 +87,11 @@ def main():
     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()
 
     try:
-        filename, downloaded = ensure_image(args.api_url, args.api_token, args.node, args.storage, args.image_url, args.filename)
+        filename, downloaded = ensure_image(args.api_url, args.api_token, args.node, args.storage, args.image_url, args.filename, args.force)
     except urllib.error.HTTPError as exc:
         sys.stderr.write(exc.read().decode())
         raise

terraform/talconfig_generator.tf

@@ -58,16 +58,6 @@ resource "local_file" "talconfig" {
   })
 }
 
-resource "local_file" "metal_network_config" {
-  for_each = { for node in var.nodes : node.name => node }
-
-  filename = "${path.module}/../generated/metal-network/${each.key}.yaml"
-  content = templatefile("${path.module}/templates/talos-metal-network.yaml.tpl", {
-    node        = each.value
-    dns_servers = var.dns_servers
-  })
-}
-
 # (The output block is already correct from the last fix)
 output "node_organization" {
   description = "How your nodes are organized for Talos configuration"
@@ -85,4 +75,4 @@ output "node_organization" {
       name = node.name, ip = node.networks[0].ip, tags = lookup(node, "tags", [])
     } if node.role == "worker" && contains(lookup(node, "tags", []), "gpu")]
   }
-}
+}

terraform/templates/talconfig.yaml.tpl

@@ -25,7 +25,13 @@ nodes:
     machineSpec:
       mode: metal
       arch: amd64
-    talosImageURL: "factory.talos.dev/installer/${talos_factory_schematic_id}/${talos_version}"
+    talosImageURL: "factory.talos.dev/installer/${talos_factory_schematic_id}"
+    patches:
+      - |-
+        apiVersion: v1alpha1
+        kind: LinkConfig
+        name: eth0
+        up: true
     nodeLabels:
       node-role.kubernetes.io/control-plane: ""
       cluster.local/node-pool: "control-plane"
@@ -59,7 +65,13 @@ nodes:
     machineSpec:
       mode: metal
       arch: amd64
-    talosImageURL: "factory.talos.dev/installer/${talos_factory_schematic_id}/${talos_version}"
+    talosImageURL: "factory.talos.dev/installer/${talos_factory_schematic_id}"
+    patches:
+      - |-
+        apiVersion: v1alpha1
+        kind: LinkConfig
+        name: eth0
+        up: true
     nodeLabels:
       node-role.kubernetes.io/worker: ""
       cluster.local/node-pool: "compute"
@@ -92,7 +104,13 @@ nodes:
     machineSpec:
       mode: metal
       arch: amd64
-    talosImageURL: "factory.talos.dev/installer/${talos_factory_schematic_id}/${talos_version}"
+    talosImageURL: "factory.talos.dev/installer/${talos_factory_schematic_id}"
+    patches:
+      - |-
+        apiVersion: v1alpha1
+        kind: LinkConfig
+        name: eth0
+        up: true
     nodeLabels:
       node-role.kubernetes.io/worker: ""
       cluster.local/node-pool: "storage"
@@ -125,7 +143,13 @@ nodes:
     machineSpec:
       mode: metal
       arch: amd64
-    talosImageURL: "factory.talos.dev/installer/${talos_factory_schematic_id}/${talos_version}"
+    talosImageURL: "factory.talos.dev/installer/${talos_factory_schematic_id}"
+    patches:
+      - |-
+        apiVersion: v1alpha1
+        kind: LinkConfig
+        name: eth0
+        up: true
     nodeLabels:
       node-role.kubernetes.io/worker: ""
       cluster.local/node-pool: "gpu"
@@ -195,4 +219,4 @@ controlPlane:
             - ${network_cidr}
         proxy: {}
         discovery:
-          enabled: true
+          enabled: true

terraform/templates/talos-metal-network.yaml.tpl

@@ -1,32 +0,0 @@
-version: 2
-ethernets:
-%{ for i, net in node.networks ~}
-  eth${i}:
-    match:
-      macaddress: "${net.mac_address}"
-    dhcp4: false
-%{ if i == 0 && net.vlan_id == 0 ~}
-    addresses:
-      - ${net.ip}/${split("/", net.cidr)[1]}
-    gateway4: ${net.gateway}
-    nameservers:
-      addresses:
-%{ for dns in dns_servers ~}
-        - ${dns}
-%{ endfor ~}
-%{ endif ~}
-%{ endfor ~}
-%{ if node.networks[0].vlan_id > 0 ~}
-vlans:
-  eth0.${node.networks[0].vlan_id}:
-    id: ${node.networks[0].vlan_id}
-    link: eth0
-    addresses:
-      - ${node.networks[0].ip}/${split("/", node.networks[0].cidr)[1]}
-    gateway4: ${node.networks[0].gateway}
-    nameservers:
-      addresses:
-%{ for dns in dns_servers ~}
-        - ${dns}
-%{ endfor ~}
-%{ endif ~}