Support VolumeConfig for EPHEMERAL partition (including RAID devices)

Author: dbackeus

app/controllers/machine_configs_controller.rb

@@ -24,6 +24,7 @@ def create
       :private_ip,
       :already_configured,
       :install_disk,
+      :ephemeral_disk_identifier,
     )
     @machine_config = MachineConfig.new(machine_config_params)
     @server = @machine_config.server

app/models/machine_config.rb

@@ -23,6 +23,7 @@ def initialize(message, output)
   validates_presence_of :private_ip
   validate :validate_private_ip_format
   validates_presence_of :install_disk
+  validate :validate_ephemeral_disk_identifier_format
 
   after_create :set_configured, if: :already_configured
 
@@ -97,7 +98,38 @@ def generate_config(output_type: server.talos_type)
     end
 
     # NOTE: This also gives us consistent 2 space indentation
-    talosconfig.to_yaml.delete_prefix("---\n")
+    config = talosconfig.to_yaml
+
+    # Add a VolumeConfig document if the server has an ephemeral disk identifier
+    if ephemeral_disk_identifier.present?
+      # id_type will be "wwid" for regular disks or "uuid" for raid arrays
+      id_type, id = ephemeral_disk_identifier.split(":", 2)
+
+      # https://www.talos.dev/v1.10/talos-guides/configuration/disk-management/#disk-selector
+      disk_selector_cel =
+        if id_type == "wwid"
+          "disk.wwid == '#{id}'"
+        elsif id_type == "uuid"
+          # Talos takes the UUID hex and puts colons every 8 characters
+          # 1a462672-bd83-888c-df8f-a57e6b38f998 -> 1a462672:bd83888c:df8fa57e:6b38f998
+          uuid_talos_style = id.delete("-").chars.each_slice(8).map(&:join).join(":")
+          "'dev/disk/by-id/md-uuid-#{uuid_talos_style}' in disk.symlinks"
+        end
+
+      config += <<~YAML
+        ---
+        apiVersion: v1alpha1
+        kind: VolumeConfig
+        name: EPHEMERAL
+        provisioning:
+          diskSelector:
+            match: "#{disk_selector_cel}"
+          minSize: 10GB
+          grow: true
+      YAML
+    end
+
+    config
   end
 
   private
@@ -144,6 +176,14 @@ def validate_private_ip_format
     end
   end
 
+  def validate_ephemeral_disk_identifier_format
+    return if ephemeral_disk_identifier.blank?
+
+    unless ephemeral_disk_identifier.match?(/^(wwid|uuid):.+$/)
+      errors.add(:ephemeral_disk_identifier, "must be in the format 'wwid:<wwid>' or 'uuid:<uuid>'")
+    end
+  end
+
   def set_configured
     server.update!(
       last_configured_at: Time.now,

app/views/machine_configs/new.html.erb

@@ -5,9 +5,51 @@
 
   <%= f.text_field :hostname, required: true, placeholder: "worker-x", hint: "Will substitute ${hostname} in your config patches." %>
   <%= f.text_field :private_ip, required: true, placeholder: "10.0.x.x", label: "Private IP", hint: "Will substitute ${private_ip} in your config patches." %>
-  <%= f.text_field :install_disk, required: true, placeholder: "/dev/sda", hint: "Gets passed into the --install-disk argument for talosctl gen config." %>
+  <% if f.object.server.lsblk.present? %>
+    <%= f.text_field :install_disk, required: true, disabled: true, hint: "Gets passed into the --install-disk argument for talosctl gen config. Non modifiable: must use the same disk which was used for bootstrapping." %>
+    <%
+      # - RAID devices are listed as children of disks
+      # - UUID of disks containing RAID devices are actually the UUID of the RAID device
+      # Hence we reduce RAID disks to a single entry per RAID device, using their UUID as value.
+      disks = f.object.server.lsblk.fetch("blockdevices").select { |disk| disk.fetch("type") == "disk" }
+      raids, other = disks.partition { |disk| disk.fetch("children", []).any? { |child| child.fetch("type").start_with?("raid") } }
+      raids = raids
+        .group_by { |disk| disk.fetch("children").find { |child| child.fetch("type").start_with?("raid") }&.fetch("name") }
+        .map do |raid_name, disks|
+          raid = disks.first.fetch("children").find { |child| child.fetch("type").start_with?("raid") }
+          type = raid.fetch("type").upcase
+          label = "/dev/#{raid_name} (#{number_to_human_size(raid.fetch('size'))}) [#{type}]"
+          value = disks.first.fetch("uuid")
+
+          [label, "uuid:#{value}"]
+        end
+      other = other
+        .reject { |disk| disk.fetch("wwn") == f.object.server.bootstrap_disk_wwid }
+        .map do |disk|
+          partitions = disk.fetch("children", []).select { it.fetch("type").start_with?("part") }
+          suffix = partitions.empty? ? "" : "[#{partitions.length} existing partition#{'s' if partitions.length != 1}]"
+          label = "/dev/#{disk['name']} (#{number_to_human_size(disk['size'])}) #{suffix}"
+          value = disk.fetch("wwn")
+
+          [label, "wwid:#{value}"]
+        end
+      ephemeral_disk_options = (raids + other).sort_by(&:first)
+    %>
+    <%=
+      f.select(
+        :ephemeral_disk_identifier,
+        ephemeral_disk_options,
+        { include_blank: "Same as install disk" },
+        required: false,
+        label: "Disk for EPHEMERAL partition (optional)",
+        hint: "If an ephemeral disk is selected, a <a href='https://www.talos.dev/v1.10/reference/configuration/block/volumeconfig/' class='text-blue-500 hover:underline' target='_blank'>VolumeConfig</a> will be added to the server's MachineConfig to use the disk for Talos Linux's EPHEMERAL partition. This is the disk mounted at /var which includes local persistent volumes, so the name ephemeral is misleading. RAID devices will be listed here but note that your Talos Factory Image must contain the <a href='https://github.com/siderolabs/extensions/tree/main/storage/mdadm' target='_blank' class='text-blue-500 hover:underline'>mdadm extension</a> for them to work.",
+      )
+    %>
+  <% else %>
+    <%= f.text_field :install_disk, required: true, placeholder: "/dev/nvme0n1", hint: "Gets passed into the --install-disk argument for talosctl gen config." %>
+  <% end %>
   <%= f.select :config_id, Config.pluck(:name, :id), { include_blank: true }, required: true %>
   <%= f.check_box :already_configured, class: "mr-1", hint: "Check to apply green status to the server immediately" %>
 
   <%= f.submit "Configure" %>
-<% end %>
+<% end %>

db/migrate/20250804112912_add_ephemeral_disk_identifier_to_machine_configs.rb

@@ -0,0 +1,5 @@
+class AddEphemeralDiskIdentifierToMachineConfigs < ActiveRecord::Migration[8.0]
+  def change
+    add_column :machine_configs, :ephemeral_disk_identifier, :string
+  end
+end

db/schema.rb

@@ -1,6 +1,6 @@
 default:
   name: default
-  endpoint: http://kubernetes.example.com:6443
+  endpoint: https://kubernetes.example.com:6443
   secrets: |
     cluster:
         id: d4ztP13Cr27K_0eTNQmM0I1_k50EHxci7bT_KROwxrI=

spec/fixtures/clusters.yml

@@ -1,4 +1,22 @@
 RSpec.describe MachineConfig do
+  it "validates format of ephemeral_disk_identifier" do
+    machine_config = MachineConfig.new
+    machine_config.validate
+    expect(machine_config.errors[:ephemeral_disk_identifier]).to be_empty
+
+    machine_config.ephemeral_disk_identifier = "wwid:1234567890abcdef"
+    machine_config.validate
+    expect(machine_config.errors[:ephemeral_disk_identifier]).to be_empty
+
+    machine_config.ephemeral_disk_identifier = "uuid:12345678-1234-5678-1234-567812345678"
+    machine_config.validate
+    expect(machine_config.errors[:ephemeral_disk_identifier]).to be_empty
+
+    machine_config.ephemeral_disk_identifier = "invalid_format"
+    machine_config.validate
+    expect(machine_config.errors[:ephemeral_disk_identifier]).to include("must be in the format 'wwid:<wwid>' or 'uuid:<uuid>'")
+  end
+
   describe "#generate_config" do
     it "raises an error if hostname is blank" do
       server = Server.new(name: "worker-1")
@@ -12,7 +30,7 @@
       expect { machine_config.generate_config }.to raise_error "can't generate config before assigning private_ip"
     end
 
-    it "generates a config" do
+    it "generates a config including substitution variables: bootstrap_disk_wwid, hostname, public_ip, private_ip and vlan" do
       hetzner_vswitch = HetznerVswitch.new(name: "vswitch", vlan: 1337)
       cluster = Cluster.create!(
         name: "cluster",
@@ -69,6 +87,7 @@
         server:,
       )
       expect(machine_config.generate_config).to eq <<~YAML
+        ---
         version: v1alpha1
         debug: false
         persist: true
@@ -147,5 +166,59 @@
               service: {}
       YAML
     end
+
+    context "with an ephemeral_disk_identifier" do
+      it "generates a config including a VolumeConfig for the ephemeral disk with different matchers for uuid and wwid" do
+        server = servers(:cloud_botstrapped)
+
+        config = Config.new(
+          name: "config",
+          install_image: "ghcr.io/siderolabs/installer:v1.10.6",
+          kubernetes_version: "1.33.3",
+          kubespan: true,
+          patch: "",
+        )
+        machine_config = MachineConfig.new(
+          hostname: server.name,
+          private_ip: "10.0.1.2",
+          install_disk: "/dev/nvme0n1",
+          ephemeral_disk_identifier: "wwid:eui.36344630528029720025384500000002",
+          config:,
+          server:,
+        )
+
+        generated_config = machine_config.generate_config
+        expect(YAML.load_stream(generated_config).length).to eq 2 # sanity check that YAML is valid
+
+        volume_config = generated_config.split("---\n").last # expect on raw String to ensure good formatting
+        expect(volume_config).to eq <<~YAML
+          apiVersion: v1alpha1
+          kind: VolumeConfig
+          name: EPHEMERAL
+          provisioning:
+            diskSelector:
+              match: "disk.wwid == 'eui.36344630528029720025384500000002'"
+            minSize: 10GB
+            grow: true
+        YAML
+
+        machine_config.ephemeral_disk_identifier = "uuid:1a462672-bd83-888c-df8f-a57e6b38f998"
+
+        generated_config = machine_config.generate_config
+        expect(YAML.load_stream(generated_config).length).to eq 2 # sanity check that YAML is valid
+
+        volume_config = generated_config.split("---\n").last # expect on raw String to ensure good formatting
+        expect(volume_config).to eq <<~YAML
+          apiVersion: v1alpha1
+          kind: VolumeConfig
+          name: EPHEMERAL
+          provisioning:
+            diskSelector:
+              match: "'dev/disk/by-id/md-uuid-1a462672:bd83888c:df8fa57e:6b38f998' in disk.symlinks"
+            minSize: 10GB
+            grow: true
+        YAML
+      end
+    end
   end
 end