nterminal 1.2.15 → 1.2.17

package/scripts/nterminalctl

@@ -68,6 +68,7 @@ load_env_file
 
 DEFAULT_STATE_PATH="$DEFAULT_RUNTIME_DIR/state.json"
 DEFAULT_PID_PATH="$DEFAULT_RUNTIME_DIR/nterminal.pid"
+DEFAULT_MONITOR_PID_PATH="$DEFAULT_RUNTIME_DIR/nterminal.monitor.pid"
 DEFAULT_LOG_PATH="$DEFAULT_RUNTIME_DIR/nterminal.log"
 
 HOST="${NTERMINAL_HOST:-127.0.0.1}"
@@ -75,6 +76,7 @@ PORT="${NTERMINAL_PORT:-3107}"
 NODE_BIN="${NTERMINAL_NODE_BIN:-node}"
 NPM_BIN="${NTERMINAL_NPM_BIN:-npm}"
 PID_PATH="$(resolve_path "${NTERMINAL_PID_PATH:-${NTERMINAL_RUNTIME_PID_PATH:-$DEFAULT_PID_PATH}}")"
+MONITOR_PID_PATH="$(resolve_path "${NTERMINAL_MONITOR_PID_PATH:-${NTERMINAL_RUNTIME_MONITOR_PID_PATH:-$DEFAULT_MONITOR_PID_PATH}}")"
 LOG_PATH="$(resolve_path "${NTERMINAL_LOG_PATH:-${NTERMINAL_RUNTIME_LOG_PATH:-$DEFAULT_LOG_PATH}}")"
 STATE_PATH="$(resolve_path "${NTERMINAL_STATE_PATH:-${NTERMINAL_RUNTIME_STATE_PATH:-$DEFAULT_STATE_PATH}}")"
 HEALTH_TIMEOUT_SECONDS="${NTERMINAL_HEALTH_TIMEOUT_SECONDS:-15}"
@@ -100,6 +102,7 @@ export NTERMINAL_APP_DIR="$APP_DIR"
 export NTERMINAL_ENV_FILE="$ENV_FILE"
 export NTERMINAL_STATE_PATH="$STATE_PATH"
 export NTERMINAL_PID_PATH="$PID_PATH"
+export NTERMINAL_MONITOR_PID_PATH="$MONITOR_PID_PATH"
 export NTERMINAL_LOG_PATH="$LOG_PATH"
 
 health_url() {
@@ -130,7 +133,7 @@ ensure_runtime() {
 }
 
 ensure_directories() {
-  mkdir -p "$(dirname "$PID_PATH")" "$(dirname "$LOG_PATH")"
+  mkdir -p "$(dirname "$PID_PATH")" "$(dirname "$MONITOR_PID_PATH")" "$(dirname "$LOG_PATH")"
   : >> "$LOG_PATH"
   chmod 600 "$LOG_PATH" 2>/dev/null || true
 }
@@ -152,6 +155,11 @@ read_pid() {
   read_pid_file "$PID_PATH"
 }
 
+read_monitor_pid() {
+  [[ -f "$MONITOR_PID_PATH" ]] || return 1
+  read_pid_file "$MONITOR_PID_PATH"
+}
+
 read_pid_file() {
   local pid_path="$1"
   [[ -f "$pid_path" ]] || return 1
@@ -166,6 +174,19 @@ is_running() {
   kill -0 "$pid" 2>/dev/null
 }
 
+parent_pid() {
+  local pid="$1"
+  if [[ -r "/proc/$pid/status" ]]; then
+    awk '/^PPid:/ { print $2; exit }' "/proc/$pid/status" 2>/dev/null
+    return
+  fi
+  if command -v ps >/dev/null 2>&1; then
+    ps -o ppid= -p "$pid" 2>/dev/null | tr -d '[:space:]'
+    return
+  fi
+  return 1
+}
+
 pid_controls_app() {
   local pid="$1"
   is_running "$pid" || return 1
@@ -176,7 +197,6 @@ pid_controls_app() {
     local cwd=""
     cwd="$(readlink "/proc/$pid/cwd" 2>/dev/null || true)"
     if [[ "$cmdline" == *"$RUNTIME_ENTRY"* ]]; then
-      [[ -z "$cwd" || "$cwd" == "$APP_DIR" ]] || return 1
       return 0
     fi
     if [[ "$cwd" == "$APP_DIR" && "$cmdline" == *"dist/server/index.js"* ]]; then
@@ -194,6 +214,26 @@ pid_controls_app() {
   return 1
 }
 
+monitor_controls_app() {
+  local pid="$1"
+  is_running "$pid" || return 1
+
+  local cmdline=""
+  if [[ -r "/proc/$pid/cmdline" ]]; then
+    cmdline="$(tr '\0' ' ' < "/proc/$pid/cmdline" 2>/dev/null || true)"
+    [[ "$cmdline" == *"$SCRIPT_DIR/nterminalctl"* && "$cmdline" == *" supervise"* ]]
+    return
+  fi
+
+  if command -v ps >/dev/null 2>&1; then
+    cmdline="$(ps -p "$pid" -o command= 2>/dev/null || true)"
+    [[ "$cmdline" == *"$SCRIPT_DIR/nterminalctl"* && "$cmdline" == *" supervise"* ]]
+    return
+  fi
+
+  return 1
+}
+
 current_pid() {
   local pid
   pid="$(read_pid || true)"
@@ -201,6 +241,35 @@ current_pid() {
     printf '%s' "$pid"
     return 0
   fi
+
+  # npm can replace the package directory while the old Node process is still
+  # listening. If the pid file is missing, recover by recognizing the port owner
+  # as this package runtime before deciding the app is stopped.
+  pid="$(port_owner_pid || true)"
+  if [[ -n "$pid" && "$(pid_controls_app "$pid" && echo yes || echo no)" == yes ]]; then
+    printf '%s' "$pid"
+    return 0
+  fi
+  return 1
+}
+
+current_monitor_pid() {
+  local pid
+  pid="$(read_monitor_pid || true)"
+  if [[ -n "$pid" && "$(monitor_controls_app "$pid" && echo yes || echo no)" == yes ]]; then
+    printf '%s' "$pid"
+    return 0
+  fi
+
+  local child_pid parent
+  child_pid="$(current_pid || true)"
+  if [[ -n "$child_pid" ]]; then
+    parent="$(parent_pid "$child_pid" || true)"
+    if [[ -n "$parent" && "$(monitor_controls_app "$parent" && echo yes || echo no)" == yes ]]; then
+      printf '%s' "$parent"
+      return 0
+    fi
+  fi
   return 1
 }
 
@@ -259,6 +328,14 @@ remove_stale_pid() {
   fi
 }
 
+remove_stale_monitor_pid() {
+  local pid
+  pid="$(read_monitor_pid || true)"
+  if [[ -n "$pid" && ! "$(monitor_controls_app "$pid" && echo yes || echo no)" == yes ]]; then
+    rm -f "$MONITOR_PID_PATH"
+  fi
+}
+
 node_health_check() {
   "$NODE_BIN" -e '
 const http = require("node:http");
@@ -273,15 +350,90 @@ request.on("error", () => process.exit(1));
 }
 
 start_daemon() {
-  cd "$APP_DIR"
   if command -v setsid >/dev/null 2>&1; then
-    nohup setsid "$NODE_BIN" "$RUNTIME_ENTRY" >> "$LOG_PATH" 2>&1 &
+    nohup setsid bash "$SCRIPT_DIR/nterminalctl" supervise >> "$LOG_PATH" 2>&1 &
   else
-    nohup "$NODE_BIN" "$RUNTIME_ENTRY" >> "$LOG_PATH" 2>&1 &
+    nohup bash "$SCRIPT_DIR/nterminalctl" supervise >> "$LOG_PATH" 2>&1 &
   fi
-  local pid="$!"
-  disown "$pid" 2>/dev/null || true
-  echo "$pid" > "$PID_PATH"
+  local monitor_pid="$!"
+  disown "$monitor_pid" 2>/dev/null || true
+  echo "$monitor_pid" > "$MONITOR_PID_PATH"
+  chmod 600 "$MONITOR_PID_PATH" 2>/dev/null || true
+
+  local pid=""
+  local deadline=$((SECONDS + HEALTH_TIMEOUT_SECONDS))
+  while (( SECONDS <= deadline )); do
+    if ! is_running "$monitor_pid"; then
+      return 1
+    fi
+    pid="$(current_pid || true)"
+    [[ -n "$pid" ]] && return 0
+    sleep 0.25
+  done
+  return 1
+}
+
+stop_supervised_child() {
+  local pid="$1"
+  [[ -n "$pid" ]] || return 0
+  if ! is_running "$pid"; then
+    rm -f "$PID_PATH"
+    return 0
+  fi
+  kill -TERM "$pid" 2>/dev/null || true
+  local deadline=$((SECONDS + STOP_TIMEOUT_SECONDS))
+  while (( SECONDS <= deadline )); do
+    if ! is_running "$pid"; then
+      rm -f "$PID_PATH"
+      return 0
+    fi
+    sleep 0.25
+  done
+  kill -KILL "$pid" 2>/dev/null || true
+  rm -f "$PID_PATH"
+}
+
+cmd_supervise() {
+  ensure_app_dir
+  ensure_env_file
+  ensure_runtime
+  validate_required_env || die "generate real secrets before starting"
+  ensure_directories
+  echo "$$" > "$MONITOR_PID_PATH"
+  chmod 600 "$MONITOR_PID_PATH" 2>/dev/null || true
+
+  local child_pid=""
+  local stop_requested=0
+  supervisor_shutdown() {
+    stop_requested=1
+    trap - TERM INT
+    stop_supervised_child "$child_pid"
+    rm -f "$MONITOR_PID_PATH"
+    exit 0
+  }
+  trap supervisor_shutdown TERM INT
+
+  while true; do
+    cd "$APP_DIR"
+    "$NODE_BIN" "$RUNTIME_ENTRY" >> "$LOG_PATH" 2>&1 &
+    child_pid="$!"
+    echo "$child_pid" > "$PID_PATH"
+    chmod 600 "$PID_PATH" 2>/dev/null || true
+
+    set +e
+    wait "$child_pid"
+    local status="$?"
+    set -e
+
+    rm -f "$PID_PATH"
+    if (( stop_requested == 1 )); then
+      rm -f "$MONITOR_PID_PATH"
+      exit 0
+    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
+  done
 }
 
 start_launchd() {
@@ -321,8 +473,15 @@ wait_for_health() {
 }
 
 port_owner_pid() {
-  command -v lsof >/dev/null 2>&1 || return 1
-  lsof -nP -iTCP:"$PORT" -sTCP:LISTEN -t 2>/dev/null | head -n 1
+  if command -v lsof >/dev/null 2>&1; then
+    lsof -nP -iTCP:"$PORT" -sTCP:LISTEN -t 2>/dev/null | head -n 1
+    return
+  fi
+  if command -v ss >/dev/null 2>&1; then
+    ss -ltnp "sport = :$PORT" 2>/dev/null | sed -n 's/.*pid=\([0-9]\+\).*/\1/p' | head -n 1
+    return
+  fi
+  return 1
 }
 
 assert_port_available() {
@@ -337,12 +496,19 @@ assert_port_available() {
 
 cmd_status() {
   remove_stale_pid
+  remove_stale_monitor_pid
   local pid
   pid="$(current_pid || true)"
   if [[ -n "$pid" ]]; then
     info "NTerminal running pid=$pid url=$(public_url) log=$LOG_PATH"
     return 0
   fi
+  local monitor_pid
+  monitor_pid="$(current_monitor_pid || true)"
+  if [[ -n "$monitor_pid" ]]; then
+    info "NTerminal supervisor running pid=$monitor_pid; server is restarting url=$(public_url) log=$LOG_PATH"
+    return 4
+  fi
   info "NTerminal stopped pid_file=$PID_PATH"
   return 3
 }
@@ -354,6 +520,7 @@ cmd_start() {
   validate_required_env || die "generate real secrets before starting"
   ensure_directories
   remove_stale_pid
+  remove_stale_monitor_pid
 
   local pid
   pid="$(current_pid || true)"
@@ -361,6 +528,12 @@ cmd_start() {
     info "NTerminal already running pid=$pid url=$(public_url)"
     return 0
   fi
+  local monitor_pid
+  monitor_pid="$(current_monitor_pid || true)"
+  if [[ -n "$monitor_pid" ]]; then
+    info "NTerminal supervisor already running pid=$monitor_pid url=$(public_url)"
+    return 0
+  fi
 
   if launchd_controls_app; then
     start_launchd
@@ -368,12 +541,28 @@ cmd_start() {
   fi
 
   assert_port_available
-  start_daemon
+  if ! start_daemon; then
+    rm -f "$MONITOR_PID_PATH"
+    echo "NTerminal failed to start supervisor. Last log lines:" >&2
+    tail -n 40 "$LOG_PATH" >&2 || true
+    return 1
+  fi
   chmod 600 "$PID_PATH" 2>/dev/null || true
-  pid="$(read_pid)"
+  pid="$(current_pid || true)"
+  if [[ -z "$pid" ]]; then
+    rm -f "$MONITOR_PID_PATH"
+    echo "NTerminal failed to start supervisor. Last log lines:" >&2
+    tail -n 40 "$LOG_PATH" >&2 || true
+    return 1
+  fi
 
   if wait_for_health "$pid"; then
-    info "NTerminal started pid=$pid url=$(public_url) log=$LOG_PATH"
+    monitor_pid="$(current_monitor_pid || true)"
+    if [[ -n "$monitor_pid" ]]; then
+      info "NTerminal started pid=$pid supervisor=$monitor_pid url=$(public_url) log=$LOG_PATH"
+    else
+      info "NTerminal started pid=$pid url=$(public_url) log=$LOG_PATH"
+    fi
     return 0
   fi
 
@@ -381,7 +570,11 @@ cmd_start() {
   if ! is_running "$pid"; then
     exit_hint="process exited during startup"
   fi
-  rm -f "$PID_PATH"
+  monitor_pid="$(current_monitor_pid || true)"
+  if [[ -n "$monitor_pid" ]]; then
+    kill -TERM "$monitor_pid" 2>/dev/null || true
+  fi
+  rm -f "$PID_PATH" "$MONITOR_PID_PATH"
   echo "NTerminal $exit_hint. Last log lines:" >&2
   tail -n 40 "$LOG_PATH" >&2 || true
   return 1
@@ -389,13 +582,44 @@ cmd_start() {
 
 cmd_stop() {
   remove_stale_pid
+  remove_stale_monitor_pid
   local pid
   pid="$(current_pid || true)"
-  if [[ -z "$pid" ]]; then
+  local monitor_pid
+  monitor_pid="$(current_monitor_pid || true)"
+  if [[ -z "$pid" && -z "$monitor_pid" ]]; then
     info "NTerminal already stopped"
     return 0
   fi
 
+  if [[ -n "$monitor_pid" ]]; then
+    kill -TERM "$monitor_pid" 2>/dev/null || true
+    local monitor_deadline=$((SECONDS + STOP_TIMEOUT_SECONDS))
+    while (( SECONDS <= monitor_deadline )); do
+      if ! is_running "$monitor_pid"; then
+        rm -f "$MONITOR_PID_PATH"
+        break
+      fi
+      sleep 0.25
+    done
+    if is_running "$monitor_pid"; then
+      kill -KILL "$monitor_pid" 2>/dev/null || true
+      rm -f "$MONITOR_PID_PATH"
+    fi
+
+    if [[ -n "$pid" && "$(is_running "$pid" && echo yes || echo no)" == yes ]]; then
+      stop_supervised_child "$pid"
+    fi
+
+    rm -f "$PID_PATH" "$MONITOR_PID_PATH"
+    if [[ -n "$pid" ]]; then
+      info "NTerminal stopped pid=$pid supervisor=$monitor_pid"
+    else
+      info "NTerminal stopped supervisor=$monitor_pid"
+    fi
+    return 0
+  fi
+
   kill -TERM "$pid" 2>/dev/null || true
   local deadline=$((SECONDS + STOP_TIMEOUT_SECONDS))
   while (( SECONDS <= deadline )); do
@@ -453,43 +677,17 @@ cmd_logs() {
   fi
 }
 
-stop_pid_file() {
-  local pid_path="$1"
-  local pid
-  pid="$(read_pid_file "$pid_path" || true)"
-  if [[ -z "$pid" ]]; then
-    return 0
-  fi
-  if ! is_running "$pid"; then
-    rm -f "$pid_path"
-    return 0
-  fi
-  if ! pid_controls_app "$pid"; then
-    rm -f "$pid_path"
-    return 0
-  fi
-
-  kill -TERM "$pid" 2>/dev/null || true
-  local deadline=$((SECONDS + STOP_TIMEOUT_SECONDS))
-  while (( SECONDS <= deadline )); do
-    if ! is_running "$pid"; then
-      rm -f "$pid_path"
-      return 0
-    fi
-    sleep 0.25
-  done
-
-  kill -KILL "$pid" 2>/dev/null || true
-  rm -f "$pid_path"
-}
-
 clean_state_paths() {
   local state_path="$1"
   local pid_path="$2"
   local log_path="$3"
+  local monitor_pid_path="${4:-}"
   local state_dir
   state_dir="$(dirname "$state_path")"
   rm -f "$state_path" "$pid_path" "$log_path" "$state_dir/update.lock"
+  if [[ -n "$monitor_pid_path" ]]; then
+    rm -f "$monitor_pid_path"
+  fi
   rm -rf "$state_dir/notification-assets"
 }
 
@@ -519,7 +717,7 @@ cmd_clean() {
 
   local state_dir
   state_dir="$(dirname "$STATE_PATH")"
-  clean_state_paths "$STATE_PATH" "$PID_PATH" "$LOG_PATH"
+  clean_state_paths "$STATE_PATH" "$PID_PATH" "$LOG_PATH" "$MONITOR_PID_PATH"
   if (( remove_env == 1 )); then
     rm -f "$ENV_FILE"
   fi
@@ -545,11 +743,11 @@ cmd_uninstall() {
 
   ensure_app_dir
   unload_launchd_if_controlled
-  stop_pid_file "$PID_PATH"
+  cmd_stop >/dev/null
   if (( kill_tmux == 1 )) && command -v tmux >/dev/null 2>&1; then
     tmux -L nterminal kill-server >/dev/null 2>&1 || true
   fi
-  clean_state_paths "$STATE_PATH" "$PID_PATH" "$LOG_PATH"
+  clean_state_paths "$STATE_PATH" "$PID_PATH" "$LOG_PATH" "$MONITOR_PID_PATH"
   rm -f "$ENV_FILE"
   remove_owned_launchd_plist
 
@@ -610,10 +808,10 @@ cmd_doctor() {
   fi
 
   ensure_directories
-  if [[ -w "$(dirname "$PID_PATH")" && -w "$(dirname "$LOG_PATH")" ]]; then
-    echo "ok writable: pid/log directories"
+  if [[ -w "$(dirname "$PID_PATH")" && -w "$(dirname "$MONITOR_PID_PATH")" && -w "$(dirname "$LOG_PATH")" ]]; then
+    echo "ok writable: pid/monitor/log directories"
   else
-    echo "fail writable: pid/log directories"
+    echo "fail writable: pid/monitor/log directories"
     failed=1
   fi
 
@@ -635,9 +833,9 @@ cmd_help() {
 Usage: nterminal <command>
 
 Commands:
-  status         Show process status. Exit 0 when running, 3 when stopped.
+  status         Show process status. Exit 0 when running, 3 when stopped, 4 when the supervisor is restarting it.
   start          Start the built server in the background.
-  stop           Stop the managed server with SIGTERM, then SIGKILL after timeout.
+  stop           Stop the managed server and supervisor with SIGTERM, then SIGKILL after timeout.
   restart        Stop and start the server.
   build          No-op for package installs; verifies the bundled runtime exists.
   deploy         Restart the installed package using the bundled runtime.
@@ -658,6 +856,7 @@ Environment:
   NTERMINAL_ENV_FILE                 Override .env path.
   NTERMINAL_STATE_PATH               Default ~/.nterminal/state.json.
   NTERMINAL_PID_PATH                 Default ~/.nterminal/nterminal.pid.
+  NTERMINAL_MONITOR_PID_PATH         Default ~/.nterminal/nterminal.monitor.pid.
   NTERMINAL_LOG_PATH                 Default ~/.nterminal/nterminal.log.
   NTERMINAL_HEALTH_TIMEOUT_SECONDS   Default 15.
   NTERMINAL_STOP_TIMEOUT_SECONDS     Default 20.
@@ -668,6 +867,7 @@ EOF
 case "$COMMAND" in
   status) cmd_status "$@" ;;
   start) cmd_start "$@" ;;
+  supervise) cmd_supervise "$@" ;;
   stop) cmd_stop "$@" ;;
   restart|reload) cmd_restart "$@" ;;
   build) cmd_build "$@" ;;