deploy_pickup/start.sh

#!/bin/bash
# ========================================
# 拾音器异响检测系统启动脚本
# ========================================
#
# 使用方法：
#   ./start.sh              # 前台运行
#   ./start.sh -d           # 后台运行
#   ./start.sh --daemon     # 后台运行
#   ./start.sh stop         # 停止服务
#   ./start.sh restart      # 重启服务
#   ./start.sh status       # 查看状态
#
# 日志文件：
#   前台运行：直接输出到控制台
#   后台运行：logs/system.log（RotatingFileHandler 自动轮转）
#

# 切换到脚本所在目录
cd "$(dirname "$0")"

# PID文件路径
PID_FILE="logs/pid.txt"
STARTUP_TIMEOUT=5
HEALTH_CHECK_INTERVAL=1

# ========================================
# 函数：按PID精确清理PID文件
# ========================================
cleanup_pid_file_if_matches() {
    local expected_pid="$1"
    if [ ! -f "$PID_FILE" ]; then
        return 0
    fi

    local current_pid
    current_pid=$(cat "$PID_FILE" 2>/dev/null)
    if [ -z "$expected_pid" ] || [ "$current_pid" = "$expected_pid" ]; then
        rm -f "$PID_FILE"
    fi
}

# ========================================
# 函数：激活conda环境
# ========================================
activate_conda() {
    if command -v conda &> /dev/null; then
        # 激活 conda 环境
        source $(conda info --base)/etc/profile.d/conda.sh
        conda activate pump_asd
        echo "已激活 conda 环境: pump_asd"
    fi
}

# ========================================
# 函数：检查PID是否为当前服务进程
# ========================================
is_expected_process() {
    local pid="$1"
    if [ -z "$pid" ]; then
        return 1
    fi

    if ! ps -p "$pid" > /dev/null 2>&1; then
        return 1
    fi

    local command
    command=$(ps -p "$pid" -o command= 2>/dev/null)
    case "$command" in
        *"run_with_auto_training.py"*)
            return 0
            ;;
        *)
            return 1
            ;;
    esac
}

# ========================================
# 函数：检查进程是否运行
# ========================================
is_running() {
    if [ -f "$PID_FILE" ]; then
        PID=$(cat "$PID_FILE")
        # 不仅检查PID是否存在，还要确认是本服务进程，避免PID复用误判
        if is_expected_process "$PID"; then
            return 0  # 运行中
        fi
        cleanup_pid_file_if_matches "$PID"
    fi
    return 1  # 未运行
}

# ========================================
# 函数：获取当前PID
# ========================================
get_pid() {
    if [ -f "$PID_FILE" ]; then
        cat "$PID_FILE"
    else
        echo ""
    fi
}

# ========================================
# 函数：等待服务稳定启动
# ========================================
wait_for_service_ready() {
    local pid="$1"
    local elapsed=0

    while [ "$elapsed" -lt "$STARTUP_TIMEOUT" ]; do
        if ! is_expected_process "$pid"; then
            return 1
        fi
        sleep "$HEALTH_CHECK_INTERVAL"
        elapsed=$((elapsed + HEALTH_CHECK_INTERVAL))
    done

    return 0
}

# ========================================
# 函数：后台监控PID，进程退出后自动清理PID文件
# ========================================
spawn_pid_watcher() {
    local watched_pid="$1"
    nohup bash -c '
        watched_pid="$1"
        pid_file="$2"

        while ps -p "$watched_pid" > /dev/null 2>&1; do
            sleep 2
        done

        if [ -f "$pid_file" ] && [ "$(cat "$pid_file" 2>/dev/null)" = "$watched_pid" ]; then
            rm -f "$pid_file"
        fi
    ' _ "$watched_pid" "$PID_FILE" > /dev/null 2>&1 &
}

# ========================================
# 函数：启动服务
# ========================================
start_service() {
    # 检查是否已经运行
    if is_running; then
        echo "服务已在运行中, PID: $(get_pid)"
        echo "如需重启，请使用: ./start.sh restart"
        return 1
    fi
    
    # 激活conda环境
    activate_conda
    
    # 检查必要文件
    if [ ! -f "run_with_auto_training.py" ]; then
        echo "错误: run_with_auto_training.py 不存在"
        exit 1
    fi

    # 检查配置文件（YAML 或 DB 至少存在一个）
    if [ ! -f "config/pickup_config.db" ] && [ ! -f "config/rtsp_config.yaml" ]; then
        echo "错误: 找不到配置文件"
        echo "需要 config/pickup_config.db 或 config/rtsp_config.yaml 之一"
        exit 1
    fi

    # 创建日志目录
    mkdir -p logs

    # 启动服务
    echo "后台运行模式..."
    # stdout/stderr 丢弃，所有日志由 RotatingFileHandler 写入 logs/system.log
    nohup python run_with_auto_training.py > /dev/null 2>&1 &
    PID=$!
    echo $PID > "$PID_FILE"
    
    # 等待一段观察窗口，避免“刚启动1秒就退出”仍被误判为成功
    if wait_for_service_ready "$PID"; then
        spawn_pid_watcher "$PID"
        echo "服务启动成功, PID: $PID"
        echo "日志文件: logs/system.log"

        # 自动启动辅助进程（根据配置开关决定）
        start_auxiliary_workers

        echo ""
        echo "查看日志: tail -f logs/system.log"
        echo "停止服务: ./start.sh stop"
        echo "重启服务: ./start.sh restart"
    else
        echo "服务启动失败，请检查日志: logs/system.log"
        cleanup_pid_file_if_matches "$PID"
        return 1
    fi
}

# ========================================
# 函数：启动辅助进程（上传Worker + 模型同步）
# ========================================
UPLOAD_PID_FILE="logs/upload_worker.pid"
MODEL_SYNC_PID_FILE="logs/model_sync.pid"

start_auxiliary_workers() {
    # 读取 YAML 配置中的开关，决定是否启动辅助进程
    local yaml_file="config/rtsp_config.yaml"
    if [ ! -f "$yaml_file" ]; then
        return 0
    fi

    # 检查 cloud_sync.enabled
    if grep -A1 'cloud_sync:' "$yaml_file" | grep -q 'enabled: [Tt]rue'; then
        if [ -f "run_upload_worker.py" ]; then
            nohup nice -n 19 python run_upload_worker.py > /dev/null 2>&1 &
            echo $! > "$UPLOAD_PID_FILE"
            echo "上传Worker已启动, PID: $!"
        fi
    fi

    # 检查 model_sync.enabled
    if grep -A1 'model_sync:' "$yaml_file" | grep -q 'enabled: [Tt]rue'; then
        if [ -f "run_model_sync.py" ]; then
            nohup nice -n 19 python run_model_sync.py > /dev/null 2>&1 &
            echo $! > "$MODEL_SYNC_PID_FILE"
            echo "模型同步已启动, PID: $!"
        fi
    fi
}

stop_auxiliary_workers() {
    # 停止上传Worker
    if [ -f "$UPLOAD_PID_FILE" ]; then
        local upid
        upid=$(cat "$UPLOAD_PID_FILE" 2>/dev/null)
        if [ -n "$upid" ] && ps -p "$upid" > /dev/null 2>&1; then
            kill "$upid" 2>/dev/null
            echo "上传Worker已停止, PID: $upid"
        fi
        rm -f "$UPLOAD_PID_FILE"
    fi

    # 停止模型同步
    if [ -f "$MODEL_SYNC_PID_FILE" ]; then
        local mpid
        mpid=$(cat "$MODEL_SYNC_PID_FILE" 2>/dev/null)
        if [ -n "$mpid" ] && ps -p "$mpid" > /dev/null 2>&1; then
            kill "$mpid" 2>/dev/null
            echo "模型同步已停止, PID: $mpid"
        fi
        rm -f "$MODEL_SYNC_PID_FILE"
    fi
}

# ========================================
# 函数：停止服务
# ========================================
stop_service() {
    # 先停辅助进程
    stop_auxiliary_workers

    if ! is_running; then
        echo "主服务未运行"
        cleanup_pid_file_if_matches ""
        return 0
    fi
    
    PID=$(get_pid)
    echo "正在停止主服务, PID: $PID"
    
    # 发送 SIGTERM 信号，优雅停止
    kill "$PID" 2>/dev/null
    
    # 等待进程结束（最多等待10秒）
    WAIT_COUNT=0
    while ps -p "$PID" > /dev/null 2>&1; do
        if [ $WAIT_COUNT -ge 10 ]; then
            echo "进程未响应，强制终止..."
            kill -9 "$PID" 2>/dev/null
            break
        fi
        sleep 1
        WAIT_COUNT=$((WAIT_COUNT + 1))
        echo "等待进程结束... ($WAIT_COUNT/10)"
    done
    
    cleanup_pid_file_if_matches "$PID"
    echo "服务已停止"
}

# ========================================
# 函数：重启服务
# ========================================
restart_service() {
    echo "=========================================="
    echo "重启拾音器异响检测服务"
    echo "=========================================="
    
    stop_service
    echo ""
    sleep 2  # 等待2秒确保资源完全释放
    start_service
}

# ========================================
# 函数：查看服务状态
# ========================================
show_status() {
    echo "=========================================="
    echo "拾音器异响检测服务状态"
    echo "=========================================="
    
    if is_running; then
        PID=$(get_pid)
        echo "状态: 运行中"
        echo "PID:  $PID"
        echo ""
        
        # 显示进程信息
        echo "进程详情:"
        ps -p "$PID" -o pid,ppid,user,%cpu,%mem,etime,command | head -2
        echo ""
        
        # 显示最近日志
        echo "最近10行日志:"
        echo "------------------------------------------"
        tail -10 logs/system.log 2>/dev/null || echo "(无日志)"
    else
        echo "主服务状态: 未运行"
        if [ -f "$PID_FILE" ]; then
            echo "注意: PID文件存在但进程已停止，可能是异常退出"
            cleanup_pid_file_if_matches ""
        fi
    fi

    # 辅助进程状态
    echo ""
    echo "辅助进程:"
    if [ -f "$UPLOAD_PID_FILE" ]; then
        local upid
        upid=$(cat "$UPLOAD_PID_FILE" 2>/dev/null)
        if [ -n "$upid" ] && ps -p "$upid" > /dev/null 2>&1; then
            echo "  上传Worker:  运行中 (PID: $upid)"
        else
            echo "  上传Worker:  已停止"
        fi
    else
        echo "  上传Worker:  未启动"
    fi

    if [ -f "$MODEL_SYNC_PID_FILE" ]; then
        local mpid
        mpid=$(cat "$MODEL_SYNC_PID_FILE" 2>/dev/null)
        if [ -n "$mpid" ] && ps -p "$mpid" > /dev/null 2>&1; then
            echo "  模型同步:    运行中 (PID: $mpid)"
        else
            echo "  模型同步:    已停止"
        fi
    else
        echo "  模型同步:    未启动"
    fi
}

# ========================================
# 函数：前台运行
# ========================================
run_foreground() {
    # 检查是否已经运行
    if is_running; then
        echo "服务已在后台运行中, PID: $(get_pid)"
        echo "请先停止: ./start.sh stop"
        return 1
    fi
    
    # 激活conda环境
    activate_conda
    
    # 检查必要文件
    if [ ! -f "run_with_auto_training.py" ]; then
        echo "错误: run_with_auto_training.py 不存在"
        exit 1
    fi

    # 检查配置文件（YAML 或 DB 至少存在一个）
    if [ ! -f "config/pickup_config.db" ] && [ ! -f "config/rtsp_config.yaml" ]; then
        echo "错误: 找不到配置文件"
        echo "需要 config/pickup_config.db 或 config/rtsp_config.yaml 之一"
        exit 1
    fi

    # 创建日志目录
    mkdir -p logs

    echo "前台运行模式..."
    python run_with_auto_training.py
}

# ========================================
# 函数：显示帮助
# ========================================
show_help() {
    echo "拾音器异响检测系统 - 启动脚本"
    echo ""
    echo "用法: ./start.sh [命令]"
    echo ""
    echo "命令:"
    echo "  (无参数)    前台运行"
    echo "  -d, --daemon  后台运行"
    echo "  start       后台启动服务"
    echo "  stop        停止服务"
    echo "  restart     重启服务"
    echo "  status      查看服务状态"
    echo "  help        显示帮助信息"
    echo ""
    echo "示例:"
    echo "  ./start.sh -d        # 后台启动"
    echo "  ./start.sh restart   # 重启服务"
    echo "  ./start.sh status    # 查看状态"
}

# ========================================
# 主逻辑
# ========================================
case "$1" in
    stop)
        stop_service
        ;;
    restart)
        restart_service
        ;;
    status)
        show_status
        ;;
    start|-d|--daemon)
        start_service
        ;;
    help|--help|-h)
        show_help
        ;;
    "")
        run_foreground
        ;;
    *)
        echo "未知命令: $1"
        echo ""
        show_help
        exit 1
        ;;
esac