fix(terraform): correct NVIDIA extension names to nonfree-kmod-nvidia-production/nvidia-container-toolkit-production, add required nvidia kernel modules to worker config

terraform/talos_schematic.tf

@@ -19,14 +19,17 @@ data "talos_image_factory_urls" "longhorn_installer" {
   platform      = "metal"
 }
 
-# GPU-enabled schematic with NVIDIA drivers and toolkit
+# GPU-enabled schematic with NVIDIA drivers and toolkit (+ Longhorn deps, since
+# worker nodes also run the Longhorn CSI DaemonSet)
 resource "talos_image_factory_schematic" "gpu_enabled" {
   schematic = jsonencode({
     customization = {
       systemExtensions = {
         officialExtensions = [
-          "siderolabs/nvidia-driver",
-          "siderolabs/nvidia-container-toolkit",
+          "siderolabs/iscsi-tools",
+          "siderolabs/util-linux-tools",
+          "siderolabs/nonfree-kmod-nvidia-production",
+          "siderolabs/nvidia-container-toolkit-production",
         ]
       }
     }

terraform/templates/worker.tftpl

@@ -48,6 +48,14 @@ machine:
     image: ${factory_image}
     wipe: true
     grubUseUKICmdline: true
+%{ if gpu_count > 0 ~}
+  kernel:
+    modules:
+      - name: nvidia
+      - name: nvidia_uvm
+      - name: nvidia_drm
+      - name: nvidia_modeset
+%{ endif ~}
   disks:
 %{ for disk in extra_disks ~}
     - device: ${disk.device}