Ubuntu autoinstall emitted invalid YAML for every role -- and now asks for IPv6

Two separate things, found because adding IPv6 to the provisioning templates
meant parsing the output for the first time.

PRE-EXISTING, AND THE BIGGER NEWS: renderUbuntuAutoinstall produced a document
that does not parse, for vanilla, worker and infra alike. There was no test on
this template, and `toContain` assertions would never have caught it. Three
faults:

  * late-commands were serialised as `- "${c}"` with no escaping at all.
    Embedded double quotes ended the scalar early -- `echo "tmpfs /tmp ..." >>
    /etc/fstab` gives "expected <block end>, but found '<scalar>'".
  * the same naive quoting put REAL newlines inside a double-quoted scalar, and
    YAML folds those into spaces. So even where it parsed, each heredoc reached
    the target as one long line and wrote a file with no line breaks -- the k3s
    modules-load and sysctl files among them.
  * the longhorn/rancher LVM entries were indented 8 spaces where their siblings
    in storage.config sit at 6, so the storage list became a nested map.

Fixed by serialising with JSON.stringify (JSON is a subset of YAML, so escaping
comes for free and \n survives the round trip) and dedenting the LVM blocks.
Verified by parsing the rendered document: the modules heredoc now arrives with
its four lines intact and the fstab command keeps its quotes.

Guarded by a new test that parses the output with a real YAML parser for every
role, mirroring how kickstart.test.ts shells out to ksvalidator. python3's yaml
rather than a new npm dependency.

DUAL-STACK: both templates now request an IPv6 lease. install.ks.ts uses
--ipv6=auto rather than =dhcp so the RA's managed-flag steers the node to DHCPv6
while a VLAN with no DHCPv6 yet still installs -- failing an OS install because
IPv6 was not ready is the worse trade. ubuntu-autoinstall.ts gets a netplan
block with dhcp4+dhcp6 and optional: true, for the same reason. The address
comes from a kea DHCPv6 reservation keyed on MAC, the same source of truth as
the v4 address, so neither template needs to know it.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DMVzWZgiKW2wquf5z8S1yH

bastion/src/bastion/src/templates/install.ks.ts

@@ -134,7 +134,13 @@ lang ${locale}
 keyboard uk
 timezone ${timezone} --utc
 
-network --bootproto=dhcp --activate --hostname=${fqdn}
+# --ipv6=auto, not =dhcp: "auto" follows the router advertisement, so the RA's
+# managed-flag is what steers the node to DHCPv6, and a VLAN with no DHCPv6 yet
+# still installs instead of blocking on a lease that will never come. Failing an
+# OS install because IPv6 was not ready would be a worse trade than a node that
+# briefly has no v6. The address itself comes from a kea DHCPv6 reservation
+# keyed on MAC -- the same source of truth as the v4 address.
+network --bootproto=dhcp --ipv6=auto --activate --hostname=${fqdn}
 
 ${auth}
 ${userDirective}

bastion/src/bastion/src/templates/ubuntu-autoinstall.ts

@@ -40,6 +40,11 @@ export function renderUbuntuAutoinstall(params: UbuntuAutoinstallParams): string
   // Build the LVM layout to match Fedora kickstart sizes
   const extraLvs: string[] = [];
   if (hasLonghorn) {
+    // 6 spaces for the list item, 8 for its keys -- these are siblings of the
+    // lv-home/lv-srv entries in storage.config, which sit at 6. At 8 the
+    // rendered document is not valid YAML at all ("expected <block end>, but
+    // found '-'"), so an Ubuntu node with the longhorn role could never have
+    // installed. Ubuntu + longhorn is exactly the worker shape.
     extraLvs.push(`      - id: lv-longhorn
         name: longhorn
         type: lvm_partition
@@ -121,7 +126,20 @@ export function renderUbuntuAutoinstall(params: UbuntuAutoinstallParams): string
     `curtin in-target -- bash -c 'IP_ADDR=$(ip -4 addr show | awk "/inet / && !/127.0.0/ {split(\\$2,a,\\"/\\"); print a[1]; exit}"); curl -sf -X POST "http://${serverIp}:${httpPort}/api/progress" -H "Content-Type: application/json" -d "{\\"mac\\":\\"$(ip link show | awk "/ether/ && !/00:00:00:00/ {print \\$2; exit}")\\",\\"stage\\":\\"complete\\",\\"detail\\":\\"ready at $IP_ADDR\\"}" || true'`,
   );
 
-  const lateCommandsYaml = lateCommands.map((c) => `        - "${c}"`).join("\n");
+  // JSON.stringify, not `"${c}"`. JSON is a subset of YAML, so this produces a
+  // correctly escaped double-quoted scalar for free -- and the naive version
+  // was broken in two ways at once, for every role:
+  //
+  //   * embedded double quotes ended the scalar early
+  //     (`echo "tmpfs /tmp ..." >> /etc/fstab` -> "expected <block end>")
+  //   * the heredocs contain REAL newlines, and YAML folds newlines inside a
+  //     double-quoted scalar into spaces -- so even where it parsed, the
+  //     heredoc arrived at the target as one long line and wrote a file with
+  //     no line breaks.
+  //
+  // JSON escaping turns the newlines into \n, which YAML unescapes back to
+  // real newlines on parse, so the heredoc survives intact.
+  const lateCommandsYaml = lateCommands.map((c) => `        - ${JSON.stringify(c)}`).join("\n");
 
   return `#cloud-config
 autoinstall:
@@ -139,6 +157,30 @@ autoinstall:
     allow-pw: false
     authorized-keys:
 ${sshKeysYaml}
+  # Both address families. Without dhcp6 the installer's default is IPv4-only,
+  # so a node provisioned into a dual-stack cluster comes up with no IPv6, k3s
+  # has no v6 node-ip to bind, and it joins as an IPv4-only member of a
+  # dual-stack cluster -- which surfaces later as pods on that node being
+  # unreachable over v6 while the node itself reads Ready.
+  #
+  # The address itself comes from a kea DHCPv6 reservation keyed on MAC, the
+  # same source of truth as the v4 address, so nothing here needs to know it.
+  #
+  # optional: true matters -- it lets the install proceed if the v6 lease is
+  # slow or the VLAN has no DHCPv6 yet, rather than blocking on a timeout. The
+  # node still needs the address before k3s starts, but that is a later step's
+  # problem and failing the OS install over it would be worse.
+  # (No backticks in this comment: it lives inside a TS template literal, and a
+  # backtick here ends the literal and breaks the build.)
+  network:
+    version: 2
+    ethernets:
+      primary:
+        match:
+          name: "en*"
+        dhcp4: true
+        dhcp6: true
+        optional: true
   storage:
     config:
       - id: disk0

bastion/src/bastion/tests/ubuntu-autoinstall.test.ts

@@ -0,0 +1,80 @@
+// The Ubuntu autoinstall document must be valid YAML for EVERY role.
+//
+// There was no test here, and the template shipped a document that did not
+// parse: the longhorn/rancher LVM entries were indented 8 spaces while their
+// siblings in storage.config sit at 6, giving "expected <block end>, but found
+// '-'". Every role that gets a longhorn volume -- which is the worker shape --
+// rendered an uninstallable document. A `toContain` assertion would not have
+// caught that; only parsing does.
+//
+// Parsed with python3's yaml rather than a new npm dependency, mirroring how
+// kickstart.test.ts shells out to `ksvalidator`: the point is to check the
+// artefact with a real parser, not to grow the dependency tree.
+
+import { describe, it, expect } from "vitest";
+import { execFileSync } from "node:child_process";
+import { writeFileSync, unlinkSync } from "node:fs";
+import { renderUbuntuAutoinstall } from "../src/templates/ubuntu-autoinstall.js";
+
+const base = {
+  hostname: "n6",
+  disk: "/dev/sda",
+  domain: "ad.itaz.eu",
+  ubuntuVersion: "24.04",
+  timezone: "Europe/London",
+  locale: "en_GB.UTF-8",
+  serverIp: "10.0.0.1",
+  httpPort: 8080,
+  sshKeys: ["ssh-ed25519 AAAAtest test@lab"],
+  adminUser: "root",
+};
+
+/** Parse with python3's yaml and return the document as JSON. */
+function parseYaml(text: string, label: string): Record<string, any> {
+  const tmp = `/tmp/autoinstall-test-${label}.yaml`;
+  writeFileSync(tmp, text);
+  try {
+    const out = execFileSync(
+      "python3",
+      ["-c", "import sys,yaml,json; json.dump(yaml.safe_load(open(sys.argv[1])), sys.stdout)", tmp],
+      { encoding: "utf-8" },
+    );
+    return JSON.parse(out);
+  } catch (err: unknown) {
+    const msg = err instanceof Error ? (err as { stderr?: string }).stderr ?? err.message : String(err);
+    throw new Error(`autoinstall YAML did not parse for ${label}: ${msg}`);
+  } finally {
+    try { unlinkSync(tmp); } catch { /* ignore */ }
+  }
+}
+
+describe("renderUbuntuAutoinstall", () => {
+  for (const role of ["vanilla", "worker", "infra"]) {
+    it(`renders parseable YAML for role=${role}`, () => {
+      const doc = parseYaml(renderUbuntuAutoinstall({ ...base, role }), role);
+      expect(doc.autoinstall).toBeDefined();
+      expect(doc.autoinstall.version).toBe(1);
+      // storage.config must be a flat list; the indentation bug produced a
+      // nested map here, which is how it went unnoticed.
+      expect(Array.isArray(doc.autoinstall.storage.config)).toBe(true);
+    });
+  }
+
+  it("gives the longhorn role its volume as a sibling entry, not a nested map", () => {
+    const doc = parseYaml(renderUbuntuAutoinstall({ ...base, role: "worker" }), "longhorn");
+    const ids = doc.autoinstall.storage.config.map((e: { id: string }) => e.id);
+    expect(ids).toContain("lv-longhorn");
+    expect(ids).toContain("mount-longhorn");
+  });
+
+  it("requests both address families on the primary NIC", () => {
+    const doc = parseYaml(renderUbuntuAutoinstall({ ...base, role: "worker" }), "net");
+    const eth = doc.autoinstall.network.ethernets.primary;
+    expect(eth.dhcp4).toBe(true);
+    // Without this a node provisioned into a dual-stack cluster comes up with
+    // no IPv6 and joins as an IPv4-only member.
+    expect(eth.dhcp6).toBe(true);
+    // The install must not block waiting for a v6 lease that may never come.
+    expect(eth.optional).toBe(true);
+  });
+});