nterminal 1.2.23 → 1.2.25

package/scripts/nterminalctl

@@ -81,6 +81,7 @@ LOG_PATH="$(resolve_path "${NTERMINAL_LOG_PATH:-${NTERMINAL_RUNTIME_LOG_PATH:-$D
 STATE_PATH="$(resolve_path "${NTERMINAL_STATE_PATH:-${NTERMINAL_RUNTIME_STATE_PATH:-$DEFAULT_STATE_PATH}}")"
 HEALTH_TIMEOUT_SECONDS="${NTERMINAL_HEALTH_TIMEOUT_SECONDS:-15}"
 STOP_TIMEOUT_SECONDS="${NTERMINAL_STOP_TIMEOUT_SECONDS:-20}"
+FATAL_STARTUP_EXIT_CODE="${NTERMINAL_FATAL_STARTUP_EXIT_CODE:-78}"
 RUNTIME_ENTRY="$APP_DIR/dist/server/index.js"
 LAUNCHD_LABEL="${NTERMINAL_LAUNCHD_LABEL:-com.nterminal.local}"
 UPDATE_LOCK_PATH="${NTERMINAL_UPDATE_LOCK_PATH:-}"
@@ -147,9 +148,41 @@ validate_required_env() {
       missing=1
     fi
   done
+  if [[ "${NTERMINAL_ROLE:-}" != "main" && "${NTERMINAL_ROLE:-}" != "secondary" ]]; then
+    echo "missing or invalid env: NTERMINAL_ROLE (expected main or secondary)"
+    missing=1
+  fi
+  if [[ "${NTERMINAL_ROLE:-}" == "secondary" ]]; then
+    local agent_token="${NTERMINAL_AGENT_TOKEN:-}"
+    if (( ${#agent_token} < 32 )); then
+      echo "missing or too-short env: NTERMINAL_AGENT_TOKEN"
+      missing=1
+    fi
+  fi
   return "$missing"
 }
 
+doctor_runtime_state() {
+  local role="${NTERMINAL_ROLE:-}"
+  if [[ "$role" != "main" && "$role" != "secondary" ]]; then
+    return 1
+  fi
+  if [[ ! -f "$STATE_PATH" ]]; then
+    echo "fail state: state file not found at $STATE_PATH"
+    return 1
+  fi
+  if [[ "$role" == "secondary" ]]; then
+    echo "ok state: $STATE_PATH"
+    return 0
+  fi
+  if "$NODE_BIN" -e 'const fs = require("node:fs"); const state = JSON.parse(fs.readFileSync(process.argv[1], "utf8")); process.exit(state && typeof state.passwordCredential === "object" && state.passwordCredential ? 0 : 2);' "$STATE_PATH" >/dev/null 2>&1; then
+    echo "ok state: main password credential present"
+    return 0
+  fi
+  echo "fail state: main state has no passwordCredential at $STATE_PATH"
+  return 1
+}
+
 prepare_start() {
   ensure_app_dir
   ensure_env_file
@@ -438,6 +471,11 @@ cmd_supervise() {
       rm -f "$MONITOR_PID_PATH"
       exit 0
     fi
+    if (( status == FATAL_STARTUP_EXIT_CODE )); then
+      printf '[%s] NTerminal process exited with fatal startup status=%s; supervisor will not restart it\n' "$(date -u '+%Y-%m-%dT%H:%M:%SZ')" "$status" >> "$LOG_PATH"
+      rm -f "$MONITOR_PID_PATH"
+      exit "$status"
+    fi
 
     printf '[%s] NTerminal process exited status=%s; restarting in 2s\n' "$(date -u '+%Y-%m-%dT%H:%M:%SZ')" "$status" >> "$LOG_PATH"
     sleep 2
@@ -480,30 +518,6 @@ wait_for_health() {
   return 1
 }
 
-restart_supervised_child() {
-  local monitor_pid="$1"
-  local old_pid="${2:-}"
-
-  if [[ -n "$old_pid" ]]; then
-    stop_supervised_child "$old_pid"
-  fi
-
-  local pid=""
-  local deadline=$((SECONDS + HEALTH_TIMEOUT_SECONDS))
-  while (( SECONDS <= deadline )); do
-    if ! is_running "$monitor_pid"; then
-      return 1
-    fi
-    pid="$(current_pid || true)"
-    if [[ -n "$pid" && "$pid" != "$old_pid" ]] && wait_for_health "$pid"; then
-      info "NTerminal restarted pid=$pid supervisor=$monitor_pid url=$(public_url) log=$LOG_PATH"
-      return 0
-    fi
-    sleep 0.25
-  done
-  return 1
-}
-
 port_owner_pid() {
   if command -v lsof >/dev/null 2>&1; then
     lsof -nP -iTCP:"$PORT" -sTCP:LISTEN -t 2>/dev/null | head -n 1
@@ -669,26 +683,27 @@ cmd_restart() {
     start_launchd
   else
     prepare_start
-    remove_stale_pid
-    remove_stale_monitor_pid
-    local monitor_pid
-    monitor_pid="$(current_monitor_pid || true)"
-    if [[ -n "$monitor_pid" ]]; then
-      local pid
-      pid="$(current_pid || true)"
-      if restart_supervised_child "$monitor_pid" "$pid"; then
-        return 0
-      fi
-      echo "NTerminal failed to restart under supervisor. Last log lines:" >&2
-      tail -n 40 "$LOG_PATH" >&2 || true
-      return 1
-    fi
-    cmd_stop >/dev/null
+    # Replace the supervisor instead of only restarting its child. The
+    # supervisor keeps its original environment, so reusing it after onboarding
+    # or npm replacement can restart the server with stale .env values or old
+    # package code.
+    cmd_stop >/dev/null || true
     cmd_start
   fi
   info "NTerminal restarted"
 }
 
+cmd_restart_after_update() {
+  prepare_start
+  if launchd_controls_app; then
+    start_launchd
+    info "NTerminal restarted after update"
+    return 0
+  fi
+  cmd_stop >/dev/null || true
+  cmd_start
+}
+
 cmd_build() {
   ensure_app_dir
   ensure_runtime
@@ -842,6 +857,7 @@ cmd_doctor() {
     echo "fail env: .env not found at $ENV_FILE"
     failed=1
   fi
+  doctor_runtime_state || failed=1
 
   if [[ -f "$RUNTIME_ENTRY" ]]; then
     echo "ok runtime: dist/server/index.js"
@@ -880,6 +896,8 @@ Commands:
   start          Start the built server in the background.
   stop           Stop the managed server and supervisor with SIGTERM, then SIGKILL after timeout.
   restart        Stop and start the server.
+  restart-after-update
+                 Stop the supervisor and start again after an npm package update.
   build          No-op for package installs; verifies the bundled runtime exists.
   deploy         Restart the installed package using the bundled runtime.
   clean --force  Stop server, kill NTerminal tmux sessions, and remove local state/log/pid files.
@@ -913,6 +931,7 @@ case "$COMMAND" in
   supervise) cmd_supervise "$@" ;;
   stop) cmd_stop "$@" ;;
   restart|reload) cmd_restart "$@" ;;
+  restart-after-update) cmd_restart_after_update "$@" ;;
   build) cmd_build "$@" ;;
   deploy) cmd_deploy "$@" ;;
   clean|purge) cmd_clean "$@" ;;