fix(talos): upgrading now actually works without destroying the cluster

Use only one pve_download_file resource and key it to <node>_<schematic-id>_<version>

remodel/tofu/kubernetes/main.tf

@@ -6,8 +6,7 @@ module "talos" {
   }
 
   talos_image = {
-    version = "v1.7.5"
-    update_version = "v1.8.0-alpha.1"
+    version        = "v1.7.5"
     schematic = file("${path.module}/config/talos-image-schematic.yaml")
   }
   cluster_config = var.cluster_config
@@ -18,6 +17,7 @@ module "talos" {
 }
 
 module "proxmox_csi_plugin" {
+  depends_on = [module.talos]
   source = "./bootstrap/proxmox-csi-plugin"
 
   providers = {
@@ -29,6 +29,7 @@ module "proxmox_csi_plugin" {
 }
 
 module "sealed_secrets" {
+  depends_on = [module.talos]
   source = "./bootstrap/sealed-secrets"
 
   providers = {
@@ -43,6 +44,7 @@ module "sealed_secrets" {
 }
 
 module "volumes" {
+  depends_on = [module.proxmox_csi_plugin]
   source = "./bootstrap/volumes"
 
   providers = {

remodel/tofu/kubernetes/providers.tf

@@ -6,7 +6,7 @@ terraform {
     }
     proxmox = {
       source  = "bpg/proxmox"
-      version = "0.60.0"
+      version = "0.61.1"
     }
     talos = {
       source  = "siderolabs/talos"

remodel/tofu/kubernetes/talos/image.tf

@@ -1,14 +1,21 @@
 locals {
+  version = var.talos_image.version
+  schematic = var.talos_image.schematic
   schematic_id = jsondecode(data.http.schematic_id.response_body)["id"]
-  update_schematic = coalesce(var.talos_image.update_schematic, var.talos_image.schematic)
+  url = "${var.talos_image.factory_url}/image/${local.schematic_id}/${local.version}/${var.talos_image.platform}-${var.talos_image.arch}.raw.gz"
+  image_id = "${local.schematic_id}_${local.version}"
+
   update_version = coalesce(var.talos_image.update_version, var.talos_image.version)
+  update_schematic = coalesce(var.talos_image.update_schematic, var.talos_image.schematic)
   update_schematic_id = jsondecode(data.http.updated_schematic_id.response_body)["id"]
+  update_url = "${var.talos_image.factory_url}/image/${local.update_schematic_id}/${local.update_version}/${var.talos_image.platform}-${var.talos_image.arch}.raw.gz"
+  update_image_id = "${local.update_schematic_id}_${local.update_version}"
 }
 
 data "http" "schematic_id" {
   url          = "${var.talos_image.factory_url}/schematics"
   method       = "POST"
-  request_body = var.talos_image.schematic
+  request_body = local.schematic
 }
 
 data "http" "updated_schematic_id" {
@@ -18,27 +25,14 @@ data "http" "updated_schematic_id" {
 }
 
 resource "proxmox_virtual_environment_download_file" "talos_image" {
-  for_each = toset(distinct([for k, v in var.cluster_config.nodes : v.host_node]))
+  for_each = toset(distinct([for k, v in var.cluster_config.nodes : "${v.host_node}_${v.update == true ? local.update_image_id : local.image_id}"]))
 
-  node_name    = each.key
+  node_name    = split("_", each.key)[0]
   content_type = "iso"
   datastore_id = var.talos_image.proxmox_datastore
 
-  file_name               = "talos-${local.schematic_id}-${var.talos_image.version}-${var.talos_image.platform}-${var.talos_image.arch}.img"
-  url                     = "${var.talos_image.factory_url}/image/${local.schematic_id}/${var.talos_image.version}/${var.talos_image.platform}-${var.talos_image.arch}.raw.gz"
-  decompression_algorithm = "gz"
-  overwrite               = false
-}
-
-resource "proxmox_virtual_environment_download_file" "updated_talos_image" {
-  for_each = toset(distinct([for k, v in var.cluster_config.nodes : v.host_node if v.update]))
-
-  node_name    = each.key
-  content_type = "iso"
-  datastore_id = var.talos_image.proxmox_datastore
-
-  file_name               = "talos-update-${local.update_schematic_id}-${local.update_version}-${var.talos_image.platform}-${var.talos_image.arch}.img"
-  url                     = "${var.talos_image.factory_url}/image/${local.update_schematic_id}/${var.talos_image.update_version}/${var.talos_image.platform}-${var.talos_image.arch}.raw.gz"
+  file_name               = "talos-${split("_",each.key)[1]}-${split("_", each.key)[2]}-${var.talos_image.platform}-${var.talos_image.arch}.img"
+  url = "${var.talos_image.factory_url}/image/${split("_", each.key)[1]}/${split("_", each.key)[2]}/${var.talos_image.platform}-${var.talos_image.arch}.raw.gz"
   decompression_algorithm = "gz"
   overwrite               = false
 }

remodel/tofu/kubernetes/talos/talos-config.tf

@@ -47,11 +47,14 @@ resource "talos_machine_configuration_apply" "talos_config_apply" {
 resource "talos_machine_bootstrap" "talos_bootstrap" {
   depends_on = [talos_machine_configuration_apply.talos_config_apply]
   client_configuration = talos_machine_secrets.machine_secrets.client_configuration
-  node                 = [for k, v in var.cluster_config.nodes : v.ip if v.machine_type == "controlplane" && !v.update][0]
+  node                 = [for k, v in var.cluster_config.nodes : v.ip if v.machine_type == "controlplane"][0]
 }
 
 data "talos_cluster_health" "health" {
-  depends_on = [talos_machine_bootstrap.talos_bootstrap]
+  depends_on = [
+    talos_machine_configuration_apply.talos_config_apply,
+    talos_machine_bootstrap.talos_bootstrap
+  ]
   client_configuration = data.talos_client_configuration.talos_config.client_configuration
   control_plane_nodes  = [for k, v in var.cluster_config.nodes : v.ip if v.machine_type == "controlplane"]
   worker_nodes         = [for k, v in var.cluster_config.nodes : v.ip if v.machine_type == "worker"]
@@ -62,10 +65,12 @@ data "talos_cluster_health" "health" {
 }
 
 data "talos_cluster_kubeconfig" "kubeconfig" {
-  #  depends_on = [talos_machine_bootstrap.talos_bootstrap]
-  depends_on = [talos_machine_bootstrap.talos_bootstrap, data.talos_cluster_health.health]
+  depends_on = [
+    talos_machine_bootstrap.talos_bootstrap,
+    data.talos_cluster_health.health
+  ]
   client_configuration = talos_machine_secrets.machine_secrets.client_configuration
-  node                 = [for k, v in var.cluster_config.nodes : v.ip if v.machine_type == "controlplane" && !v.update][0]
+  node                 = [for k, v in var.cluster_config.nodes : v.ip if v.machine_type == "controlplane"][0]
   timeouts = {
     read = "1m"
   }

remodel/tofu/kubernetes/talos/virtual-machines.tf

@@ -40,7 +40,7 @@ resource "proxmox_virtual_environment_vm" "talos_vm" {
     ssd          = true
     file_format  = "raw"
     size         = 20
-    file_id      = each.value.update ? proxmox_virtual_environment_download_file.updated_talos_image[each.value.host_node].id : proxmox_virtual_environment_download_file.talos_image[each.value.host_node].id
+    file_id      = proxmox_virtual_environment_download_file.talos_image["${each.value.host_node}_${each.value.update == true ? local.update_image_id : local.image_id}"].id
   }
 
   boot_order = ["scsi0"]

remodel/tofu/kubernetes/variables.tf

@@ -33,6 +33,12 @@ variable "cluster_config" {
       })
     )
   })
+
+  validation {
+    condition     = length([
+      for n in var.cluster_config.nodes : n if contains(["controlplane", "worker"], n.machine_type)]) == length(var.cluster_config.nodes)
+    error_message = "Node machine_type must be either 'controlplane' or 'worker'."
+  }
 }
 
 variable "volumes" {