Merge pull request 'refactor(provision): Phase 3 — remove legacy host-keys/ fallback' (#59) from worktree-phase0-provision-ordering-fix into main

Merge PR #59: Phase 3 — remove legacy host-keys/ fallback

scripts/proxmox/create-proxmox-resource.sh

@@ -8,9 +8,12 @@
 # script -- there's no multi-gigabyte image to transfer afterward. The first
 # time a node doesn't have that repo path yet, it's bootstrapped: cloned from
 # this checkout's own `origin` remote, then scripts/codex-setup.sh installs
-# the build tooling (Nix, etc.). Every run after that just `git pull`s it and
+# the build tooling (Nix, etc.). Every run after that just `git pull`s it.
-# copies over the locally-managed host-keys/ (gitignored, so a git pull
+# SSH host keys are stored as clan vars (vars/per-machine/<target>/openssh/,
-# alone wouldn't carry it) before building.
+# committed and sops-encrypted) -- the script decrypts them locally and
+# copies only the two files for this target to the node's host-keys/ before
+# building. A target with no clan var is an error (generate one first with
+# scripts/secrets/sync-host-keys.sh <target>).
 #
 # --node (default: $PROXMOX_HOST, see scripts/env.sh) picks which of the two
 # LAN Proxmox nodes this runs against: production, pve1.sweet.home
@@ -497,27 +500,27 @@ sync_args=("$flake_target")
 [[ "$dry_run" -eq 1 ]] && sync_args+=(--dry-run)
 bash "$sync_keys" "${sync_args[@]}"
 
-# If sync-host-keys.sh changed .sops.yaml or secrets/, those changes must be
+# If sync-host-keys.sh changed .sops.yaml, secrets/, or vars/per-machine/,
-# committed and pushed before the remote `git pull` below picks them up --
+# those changes must be committed and pushed before the remote `git pull`
-# the PVE node builds from whatever HEAD is checked out there, not the local
+# below picks them up -- the PVE node builds from whatever HEAD is checked
-# working tree. Detect uncommitted changes and block until the operator
+# out there, not the local working tree. Uncommitted clan vars or sops
-# confirms they've pushed, so the build never runs against a stale flake.
+# recipients mean the image builds fine but the host cannot decrypt its
+# secrets on first boot. Block until the operator confirms they've pushed.
 if [[ "$dry_run" -eq 0 ]]; then
-  _sops_dirty="$(git -C "$repo_root" status --porcelain -- .sops.yaml secrets/ 2>/dev/null || true)"
+  _dirty="$(git -C "$repo_root" status --porcelain -- .sops.yaml secrets/ vars/per-machine/ 2>/dev/null || true)"
-  if [[ -n "$_sops_dirty" ]]; then
+  if [[ -n "$_dirty" ]]; then
     echo
     echo "==> COMMIT + PUSH REQUIRED before the remote build can succeed:"
-    echo "    sync-host-keys.sh modified .sops.yaml / secrets/ to register the"
+    echo "    Uncommitted changes in .sops.yaml, secrets/, or vars/per-machine/."
-    echo "    new host's sops recipient. The PVE node builds from the git-tracked"
+    echo "    The PVE node builds from the git-tracked flake, so these changes"
-    echo "    flake, so these changes must be committed and pushed first -- otherwise"
+    echo "    must be committed and pushed first -- otherwise the image build will"
-    echo "    the image build will succeed but the host cannot decrypt secrets on"
+    echo "    succeed but the host cannot decrypt its secrets on first boot."
-    echo "    first boot (its age key isn't in the encrypted secrets files yet)."
     echo
-    git -C "$repo_root" status --short -- .sops.yaml secrets/ || true
+    git -C "$repo_root" status --short -- .sops.yaml secrets/ vars/per-machine/ || true
     echo
     read -rp "    Commit and push those changes, then press Enter to continue (Ctrl-C to abort): "
   fi
-  unset _sops_dirty
+  unset _dirty
 fi
 
 # --- VMID: pick one, and refuse to touch anything that already exists ---
@@ -643,36 +646,34 @@ ensure_remote_repo() {
 }
 
 # --- sync host key to the node ---------------------------------------------
-# Clan-managed keys live in vars/per-machine/ (committed, sops-encrypted),
+# SSH host keys are stored as clan vars (vars/per-machine/<target>/openssh/).
-# so they arrive on the node via `git pull`. But the build scripts expect a
+# Decrypt locally and scp just the two files for this target to the node's
-# plaintext key file in host-keys/ (NIXOS_HOST_KEYS_DIR for LXC, or
+# host-keys/ directory, where the remote build script picks them up via
-# --pre-format-files for VM). For clan keys, decrypt locally and scp just the
+# NIXOS_HOST_KEYS_DIR (LXC) or --pre-format-files (VM). A target with no
-# two files for this target; for legacy host-keys/ entries, scp the whole dir.
+# clan var is an error -- generate one first with sync-host-keys.sh.
 sync_remote_host_keys() {
   echo
   echo "==> Syncing host key for ${flake_target} to ${node}..."
   if [[ "$dry_run" -eq 1 ]]; then
-    if clan_ssh_key_exists "$flake_target" "$repo_root"; then
     echo "[dry-run] would decrypt clan SSH key for ${flake_target} and copy to ${ssh_target}:${remote_repo_dir}/host-keys/"
-    else
-      echo "[dry-run] would copy ${repo_root}/host-keys/ to ${ssh_target}:${remote_repo_dir}/host-keys/"
-    fi
     return
   fi
-  ssh "$ssh_target" "mkdir -p '${remote_repo_dir}/host-keys'"
+  if ! clan_ssh_key_exists "$flake_target" "$repo_root"; then
-  if clan_ssh_key_exists "$flake_target" "$repo_root"; then
+    echo "ERROR: no clan SSH key found for ${flake_target}" >&2
+    echo "  (expected: ${repo_root}/vars/per-machine/${flake_target}/openssh/ssh_host_ed25519_key/secret)" >&2
+    echo "  Generate one first: bash scripts/secrets/sync-host-keys.sh ${flake_target}" >&2
+    exit 1
+  fi
   local tmpdir
   tmpdir="$(mktemp -d)"
   # shellcheck disable=SC2064
   trap "rm -rf '${tmpdir}'" RETURN
   echo "  Decrypting clan SSH key for ${flake_target}..."
   clan_decrypt_ssh_key "$flake_target" "$repo_root" "$tmpdir"
+  ssh "$ssh_target" "mkdir -p '${remote_repo_dir}/host-keys'"
   scp -p "${tmpdir}/${flake_target}_ssh_host_ed25519_key" \
          "${tmpdir}/${flake_target}_ssh_host_ed25519_key.pub" \
          "${ssh_target}:${remote_repo_dir}/host-keys/"
-  else
-    scp -pr "${repo_root}/host-keys/." "${ssh_target}:${remote_repo_dir}/host-keys/"
-  fi
 }
 
 # --- build (or reuse an image already on the node) ------------------------