fix: bake talos bootstrap network into iso

terraform/locals.tf

@@ -39,7 +39,15 @@ locals {
 
   talos_iso_download_url = "https://factory.talos.dev/image/${var.talos_factory_schematic_id}/${var.talos_version}/nocloud-${var.talos_image_architecture}.iso"
 
-  image_target_nodes = toset(distinct([for node in var.nodes : lookup(node, "proxmox_node", var.proxmox_node)]))
+  image_target_nodes = toset([for node in var.nodes : node.name])
+
+  node_iso_filenames = {
+    for node in var.nodes : node.name => "talos-${trimprefix(var.talos_version, "v")}-${substr(var.talos_factory_schematic_id, 0, 12)}-${node.name}-${var.talos_image_architecture}.iso"
+  }
+
+  node_iso_files = {
+    for node in var.nodes : node.name => format("%s:iso/%s", var.talos_image_storage, local.node_iso_filenames[node.name])
+  }
 
   # --- VM Resource Transformation (No changes needed below this line) ---
   all_nodes_transformed = {
@@ -53,7 +61,7 @@ locals {
       disk_size               = node.disk_size
       disk_storage            = var.disk_storage
       tags                    = lookup(node, "tags", [node.role])
-      iso_file                = var.talos_iso_file
+      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

@@ -16,24 +16,28 @@ provider "proxmox" {
   api_token = var.proxmox_api_token
   insecure  = true
 }
-
 resource "null_resource" "stage_talos_image" {
-  for_each = var.talos_image_update_mode == "download" ? local.image_target_nodes : []
+  for_each = var.talos_image_update_mode == "download" ? { for node in var.nodes : node.name => node } : {}
 
   triggers = {
-    node_name     = each.value
-    image_url     = local.talos_iso_download_url
-    storage       = var.talos_image_storage
-    talos_version = var.talos_version
-    filename      = "talos-${trimprefix(var.talos_version, "v")}-${substr(var.talos_factory_schematic_id, 0, 12)}-${var.talos_image_architecture}.iso"
+    node_name      = each.value.name
+    proxmox_node   = lookup(each.value, "proxmox_node", var.proxmox_node)
+    image_url      = local.talos_iso_download_url
+    storage        = var.talos_image_storage
+    talos_version  = var.talos_version
+    filename       = local.node_iso_filenames[each.key]
+    network_config = sha256(local_file.metal_network_config[each.key].content)
   }
 
   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', '${var.proxmox_api_token}', '--node', '${each.value}', '--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}', '--network-config', r'${local_file.metal_network_config[each.key].filename}'], check=True)"
   }
+
+  depends_on = [local_file.metal_network_config]
 }
 
+
 # Create all required network bridges
 resource "proxmox_virtual_environment_network_linux_bridge" "cluster_bridge" {
   for_each = local.unique_bridges

terraform/scripts/stage_talos_image.py

@@ -1,12 +1,18 @@
 import argparse
+import base64
+import gzip
 import json
 import os
+import shutil
+import subprocess
 import sys
+import tempfile
 import ssl
 import time
 import urllib.error
 import urllib.parse
 import urllib.request
+from pathlib import Path
 
 
 def api_request(base_url, token, method, path, data=None):
@@ -23,20 +29,30 @@ 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, network_config_path=None):
     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
 
-    download_path = f"/api2/json/nodes/{node}/storage/{storage}/download-url"
-    task = api_request(base_url, token, "POST", download_path, {
-        "content": "iso",
-        "filename": filename,
-        "url": image_url,
-    })
-    wait_for_task(base_url, token, node, task.get("data"))
+    source_filename = filename
+    if network_config_path:
+        source_filename = f"base-{filename}"
+
+    response = api_request(base_url, token, "GET", content_path)
+    source_present = any(item.get("volid", "").endswith(source_filename) for item in response.get("data", []))
+    if not source_present:
+        download_path = f"/api2/json/nodes/{node}/storage/{storage}/download-url"
+        task = api_request(base_url, token, "POST", download_path, {
+            "content": "iso",
+            "filename": source_filename,
+            "url": image_url,
+        })
+        wait_for_task(base_url, token, node, task.get("data"))
+
+    if network_config_path:
+        build_custom_iso(base_url, token, node, storage, source_filename, filename, network_config_path)
 
     response = api_request(base_url, token, "GET", content_path)
     for item in response.get("data", []):
@@ -63,6 +79,42 @@ def wait_for_task(base_url, token, node, upid):
     raise RuntimeError("timed out waiting for ISO download task to finish")
 
 
+def build_custom_iso(base_url, token, node, storage, source_filename, output_filename, network_config_path):
+    storage_path = f"/var/lib/vz/template/iso/{source_filename}"
+    output_path = f"/var/lib/vz/template/iso/{output_filename}"
+    network_config = Path(network_config_path).read_text(encoding="utf-8").strip()
+    installer_meta = base64.b64encode(gzip.compress(f"0xa={network_config}".encode("utf-8"), compresslevel=9)).decode("ascii")
+
+    if shutil.which("docker") is None:
+        raise RuntimeError("docker is required to build Talos ISO with embedded META")
+
+    with tempfile.TemporaryDirectory() as temp_dir:
+        temp_path = Path(temp_dir)
+        script_path = temp_path / "build.sh"
+        script_path.write_text(
+            "#!/bin/sh\n"
+            "set -eu\n"
+            "apk add --no-cache xorriso >/dev/null\n"
+            "cp /input/source.iso /work/source.iso\n"
+            "xorriso -indev /work/source.iso -outdev /output/result.iso -boot_image any replay \\\n"
+            f"  -append_partition 2 0x0 /dev/null \\\n"
+            f"  -boot_image any keep \\\n"
+            f"  -map /input/cmdline /cmdline\n",
+            encoding="utf-8",
+        )
+        cmdline_path = temp_path / "cmdline"
+        cmdline_path.write_text(f"talos.environment=INSTALLER_META_BASE64={installer_meta}\n", encoding="utf-8")
+        subprocess.run([
+            "docker", "run", "--rm",
+            "-v", f"{script_path}:/input/build.sh:ro",
+            "-v", f"{cmdline_path}:/input/cmdline:ro",
+            "-v", f"{Path(storage_path).resolve()}:/input/source.iso:ro",
+            "-v", f"{temp_path}:/work",
+            "-v", f"{Path(output_path).resolve().parent}:/output",
+            "alpine:3.20", "sh", "/input/build.sh",
+        ], check=True)
+
+
 def main():
     parser = argparse.ArgumentParser()
     parser.add_argument("--api-url", required=True)
@@ -71,10 +123,11 @@ def main():
     parser.add_argument("--storage", required=True)
     parser.add_argument("--image-url", required=True)
     parser.add_argument("--filename", required=True)
+    parser.add_argument("--network-config")
     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.network_config)
     except urllib.error.HTTPError as exc:
         sys.stderr.write(exc.read().decode())
         raise

terraform/talconfig_generator.tf

@@ -54,11 +54,20 @@ resource "local_file" "talconfig" {
     cluster_svc_nets = var.cluster_svc_nets
     cni_name         = var.cni_name
 
-    # --- Pass the proxy config ---
     image_cache_proxy = var.image_cache_proxy
   })
 }
 
+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"

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

@@ -0,0 +1,39 @@
+links:
+%{ for i, net in node.networks ~}
+  - name: eth${i}
+    up: true
+%{ if net.vlan_id > 0 ~}
+  - name: eth${i}.${net.vlan_id}
+    logical: true
+    up: true
+    kind: vlan
+    type: ether
+    parentName: eth${i}
+    vlan:
+      vlanID: ${net.vlan_id}
+      vlanProtocol: 802.1q
+%{ endif ~}
+%{ endfor ~}
+addresses:
+  - address: ${node.networks[0].ip}/${split("/", node.networks[0].cidr)[1]}
+    linkName: %{ if node.networks[0].vlan_id > 0 }eth0.${node.networks[0].vlan_id}%{ else }eth0%{ endif }
+    family: inet4
+    scope: global
+    flags: permanent
+    layer: platform
+routes:
+  - family: inet4
+    gateway: ${node.networks[0].gateway}
+    outLinkName: %{ if node.networks[0].vlan_id > 0 }eth0.${node.networks[0].vlan_id}%{ else }eth0%{ endif }
+    table: main
+    priority: 1024
+    scope: global
+    type: unicast
+    protocol: static
+    layer: platform
+resolvers:
+  - dnsServers:
+%{ for dns in dns_servers ~}
+      - ${dns}
+%{ endfor ~}
+    layer: platform