Merge pull request 'fix(deploy): nightly auto-upgrade failed on both ATMs, for different reasons' (#101) from fix/autoupgrade-known-hosts-and-wg into dev

Reviewed-on: #101

deploy/nixos/configuration.nix

@@ -159,6 +159,18 @@
   # Auto-updates (optional - disabled by default for stability)
   # system.autoUpgrade.enable = false;
 
+  # Trust the Forgejo host key up front. system.autoUpgrade fetches the flake
+  # over ssh AS ROOT, and a machine whose root has never connected by hand has
+  # no known_hosts entry, so every nightly run dies at
+  # "Host key verification failed" before it reaches authentication. batm3 did
+  # exactly that, silently, from its 2026-08-06 install until 09-22 (#98): it
+  # sat on its install generation for six weeks while reporting a failed unit
+  # nobody was watching. sintra only ever worked because a human had ssh'd as
+  # root once and accepted the key. Declaring it means a freshly flashed ATM
+  # can update from first boot with no manual step.
+  programs.ssh.knownHosts."git.atitlan.io".publicKey =
+    "ssh-ed25519 AAAAC3NzaC1lZDI1NTE5AAAAIMlo3f05o4+bk0+8x2VG91o9GubshOb46HmBPvND9pJx";
+
   # pragma: allowlist secret
   # Ensure WireGuard private key directory exists with correct permissions
   system.activationScripts.wireguard-key = ''
@@ -169,6 +181,20 @@
     fi
   '';
 
+  # The tunnel is operator-provisioned: wg0.key is written per machine after
+  # flashing, and until it is, `wg set … private-key` exits 1 with
+  # "fopen: No such file or directory". One failed unit makes
+  # switch-to-configuration exit 4, which marks the entire nightly
+  # system.autoUpgrade run as failed — so an ATM that simply never had its
+  # tunnel provisioned reports a broken updater for the life of the machine
+  # (sintra, #98). Skip the unit when there is no key instead of failing
+  # activation over an interface that was never set up; a provisioned machine
+  # is unaffected. Guarded on wg0 still being declared so the live image,
+  # which mkForce's the interfaces away, doesn't get a unit with no ExecStart.
+  systemd.services = lib.mkIf (config.networking.wireguard.interfaces ? wg0) {
+    wireguard-wg0.unitConfig.ConditionPathExists = "/var/lib/wireguard/wg0.key";
+  };
+
   # In-place rename migration: lamassu user → bitspire user.
   # Runs after `users` activation so the bitspire user exists with its UID.
   # Idempotent: re-running on an already-migrated system is a chown no-op.