scripts/openclaw-gateway-ctl

#!/usr/bin/env bash
set -euo pipefail

MANAGER_PID_FILE="/var/run/openclaw/manager.pid"
LOG_FILE="${OPENCLAW_GATEWAY_CTL_LOG_FILE:-/var/log/openclaw/gateway-ctl.log}"

ts() { date -u +"%Y-%m-%dT%H:%M:%SZ"; }
log_info() { 
    local msg="[$(ts)] [INFO] openclaw-gateway-ctl: $*"
    printf '%s\n' "$msg"
    mkdir -p "$(dirname "$LOG_FILE")"
    printf '%s\n' "$msg" >> "$LOG_FILE"
}
log_error() { 
    local msg="[$(ts)] [ERROR] openclaw-gateway-ctl: $*"
    printf '%s\n' "$msg" >&2
    mkdir -p "$(dirname "$LOG_FILE")"
    printf '%s\n' "$msg" >> "$LOG_FILE"
}

get_manager_pid() {
    if [[ -f "$MANAGER_PID_FILE" ]]; then
        local pid
        pid=$(cat "$MANAGER_PID_FILE" 2>/dev/null | tr -d '[:space:]')
        if [[ -n "$pid" ]] && kill -0 "$pid" 2>/dev/null; then
            echo "$pid"
            return 0
        fi
    fi
    return 1
}

get_gateway_pid() {
    local port="${OPENCLAW_GATEWAY_PORT:-18789}"
    local pid
    pid=$(lsof -ti :"$port" -sTCP:LISTEN 2>/dev/null | head -1 || true)
    if [[ -n "$pid" ]] && kill -0 "$pid" 2>/dev/null; then
        local comm
        comm=$(ps -p "$pid" -o comm= 2>/dev/null || true)
        if [[ "$comm" == *"openclaw"* ]]; then
            echo "$pid"
            return 0
        fi
    fi
    return 1
}

do_status() {
    local manager_pid gateway_pid
    manager_pid=$(get_manager_pid 2>/dev/null) || true
    gateway_pid=$(get_gateway_pid 2>/dev/null) || true

    if [[ -n "$manager_pid" ]]; then
        log_info "manager is running (PID: $manager_pid)"
    else
        log_info "manager is not running"
    fi

    if [[ -n "$gateway_pid" ]]; then
        log_info "gateway is running (PID: $gateway_pid)"
    else
        log_info "gateway is not running"
    fi

    if [[ -n "$manager_pid" && -n "$gateway_pid" ]]; then
        return 0
    else
        return 1
    fi
}

do_start() {
    local manager_pid
    manager_pid=$(get_manager_pid) || true
    if [[ -z "$manager_pid" ]]; then
        log_error "manager is not running, cannot start gateway"
        return 1
    fi

    log_info "Sending SIGUSR1 to manager (PID: $manager_pid) to start gateway..."
    kill -USR1 "$manager_pid" 2>/dev/null || true

    local waited=0
    local max_wait=60
    while [[ $waited -lt $max_wait ]]; do
        sleep 2
        waited=$((waited + 2))
        local gateway_pid
        gateway_pid=$(get_gateway_pid 2>/dev/null) || true
        if [[ -n "$gateway_pid" ]]; then
            log_info "Gateway started successfully (PID: $gateway_pid, waited ${waited}s)"
            return 0
        fi
    done

    log_error "Gateway failed to start after ${max_wait}s"
    if [[ -f /var/log/openclaw/gateway.stderr.log ]]; then
        log_error "Last 5 lines from gateway stderr:"
        tail -5 /var/log/openclaw/gateway.stderr.log | while IFS= read -r line; do
            log_error "  $line"
        done
    fi
    return 1
}

do_stop() {
    local gateway_pid
    gateway_pid=$(get_gateway_pid) || true
    if [[ -z "$gateway_pid" ]]; then
        log_error "gateway is not running"
        return 1
    fi

    log_info "Stopping gateway process (PID: $gateway_pid)..."
    kill -TERM "$gateway_pid" 2>/dev/null || true

    log_info "Waiting for gateway to stop (max 30s)..."
    local waited=0
    while [[ $waited -lt 30 ]]; do
        if ! kill -0 "$gateway_pid" 2>/dev/null; then
            log_info "Gateway stopped gracefully"
            return 0
        fi
        sleep 1
        waited=$((waited + 1))
    done

    log_error "Gateway did not stop gracefully, force killing..."
    kill -9 "$gateway_pid" 2>/dev/null || true
    log_info "Gateway killed"
}

do_restart() {
    local manager_pid gateway_pid
    manager_pid=$(get_manager_pid) || true
    gateway_pid=$(get_gateway_pid) || true

    if [[ -z "$manager_pid" ]]; then
        log_error "manager is not running, cannot restart gateway"
        return 1
    fi

    if [[ -n "$gateway_pid" ]]; then
        log_info "Stopping gateway process (PID: $gateway_pid)..."
        kill -TERM "$gateway_pid" 2>/dev/null || true
        sleep 1
    fi

    log_info "Sending SIGUSR1 to manager (PID: $manager_pid) to restart gateway..."
    kill -USR1 "$manager_pid" 2>/dev/null || true

    local waited=0
    local max_wait=60
    while [[ $waited -lt $max_wait ]]; do
        sleep 2
        waited=$((waited + 2))
        gateway_pid=$(get_gateway_pid 2>/dev/null) || true
        if [[ -n "$gateway_pid" ]]; then
            log_info "Gateway restarted successfully (PID: $gateway_pid, waited ${waited}s)"
            return 0
        fi
    done

    log_error "Gateway failed to restart after ${max_wait}s"
    if [[ -f /var/log/openclaw/gateway.stderr.log ]]; then
        log_error "Last 5 lines from gateway stderr:"
        tail -5 /var/log/openclaw/gateway.stderr.log | while IFS= read -r line; do
            log_error "  $line"
        done
    fi
    return 1
}

do_reload() {
    local gateway_pid
    gateway_pid=$(get_gateway_pid) || true
    if [[ -z "$gateway_pid" ]]; then
        log_error "gateway is not running"
        return 1
    fi
    log_info "Sending SIGHUP to gateway (PID: $gateway_pid)..."
    kill -HUP "$gateway_pid" 2>/dev/null || true
}

ACTION="${1:-}"

case "$ACTION" in
    start|START)
        do_start
        ;;
    stop|STOP)
        do_stop
        ;;
    restart|RESTART)
        do_restart
        ;;
    reload|RELOAD)
        do_reload
        ;;
    status|STATUS|"")
        do_status
        ;;
    *)
        echo "Usage: openclaw-gateway-ctl {start|stop|restart|reload|status}"
        echo ""
        echo "  start   - Start gateway (via manager)"
        echo "  stop    - Stop only gateway process (manager keeps running)"
        echo "  restart - Stop gateway then start via manager (no bootstrap)"
        echo "  reload  - Send SIGHUP to gateway for config reload"
        echo "  status  - Show running status"
        exit 1
        ;;
esac