sionrui/script/deploy/deploy-all.sh

#!/bin/bash
# ============================================
# 芋道 (Yudao) 生产部署脚本 - 运维级
# ============================================
# 特性:
#   - 优雅启停 (SIGTERM → 等待 → SIGKILL 兜底)
#   - 健康检查 + 自动回滚
#   - 版本备份 + 保留最近 5 个版本
#   - 并发锁，防止同时执行
#   - 前置检查 (磁盘/内存/端口/编译产物)
#   - 结构化日志
# ============================================
set -o pipefail

# ==================== 配置 ====================
readonly PROJECT_DIR="/www/wwwroot/sionrui"
readonly BACKEND_DIR="/www/wwwroot/yudao-server"
readonly FRONTEND_DIR="/www/wwwroot/muyetools.cn"
readonly ADMIN_DIR="/www/wwwroot/8.155.172.147"

readonly BACKEND_JAR="sion-rui.jar"
readonly BACKEND_PORT=9900
readonly BACKUP_KEEP=5                            # 保留最近 N 个版本
readonly SHUTDOWN_GRACE_SEC=30                    # 优雅关闭等待上限
readonly HEALTH_CHECK_RETRIES=30                  # 健康检查重试次数 (30 × 2s = 60s)
readonly HEALTH_CHECK_INTERVAL=2                  # 健康检查间隔(秒)

readonly DEPLOY_LOCK="/tmp/yudao_deploy.lock"
readonly BUILD_LOG="/tmp/yudao_build_$(date +%Y%m%d_%H%M%S).log"
readonly DEPLOY_LOG="/tmp/yudao_deploy.log"
readonly BACKUP_DIR="$BACKEND_DIR/backups"

# Java 环境
export JAVA_HOME="/www/server/java/jdk-17.0.8"
export PATH="$JAVA_HOME/bin:$PATH"
export MAVEN_OPTS="-Xms512m -Xmx1024m"
export NODE_OPTIONS="--max-old-space-size=2048"

# Maven
MVN_EXEC=$(which mvn 2>/dev/null || find /www/server -name "mvn" -type f 2>/dev/null | head -1 || echo "mvn")
MAVEN_REPO="$PROJECT_DIR/.m2_repo"

# ==================== 工具函数 ====================
log()     { echo "[$(date '+%H:%M:%S')] $1" | tee -a "$DEPLOY_LOG"; }
success() { echo "[$(date '+%H:%M:%S')] ✅ $1" | tee -a "$DEPLOY_LOG"; }
warn()    { echo "[$(date '+%H:%M:%S')] ⚠️  $1" | tee -a "$DEPLOY_LOG"; }
error()   { echo "[$(date '+%H:%M:%S')] ❌ $1" | tee -a "$DEPLOY_LOG"; }
die()     { error "$1"; release_lock; exit 1; }

# ==================== 并发锁 ====================
acquire_lock() {
    if [ -f "$DEPLOY_LOCK" ]; then
        local pid
        pid=$(cat "$DEPLOY_LOCK" 2>/dev/null)
        if ps -p "$pid" > /dev/null 2>&1; then
            die "部署脚本已在运行中 (PID=$pid)，请等待完成或手动删除 $DEPLOY_LOCK"
        fi
        warn "发现残留锁文件（进程 $pid 已不存在），清理后继续"
        rm -f "$DEPLOY_LOCK"
    fi
    echo $$ > "$DEPLOY_LOCK"
}

release_lock() {
    rm -f "$DEPLOY_LOCK"
}

# 确保退出时释放锁
trap release_lock EXIT INT TERM

# ==================== 前置检查 ====================
pre_flight() {
    log "=== 前置检查 ==="

    # 磁盘空间 (至少 2GB)
    local avail
    avail=$(df -m "$BACKEND_DIR" | awk 'NR==2 {print $4}')
    if [ "${avail:-0}" -lt 2048 ]; then
        die "磁盘空间不足：${BACKEND_DIR} 可用 ${avail}MB，需要至少 2048MB"
    fi
    log "  磁盘可用: ${avail}MB ✓"

    # 内存 (至少 512MB 空闲)
    local mem_free
    mem_free=$(free -m | awk '/Mem:/ {print $7}')
    if [ "${mem_free:-0}" -lt 512 ]; then
        warn "可用内存偏低：${mem_free}MB，构建可能失败"
    else
        log "  可用内存: ${mem_free}MB ✓"
    fi

    # JDK
    if [ ! -x "$JAVA_HOME/bin/java" ]; then
        die "JDK 未找到：$JAVA_HOME/bin/java"
    fi
    log "  JDK: $($JAVA_HOME/bin/java -version 2>&1 | head -1) ✓"

    # 项目目录
    if [ ! -d "$PROJECT_DIR/.git" ]; then
        die "项目目录不存在或不是 Git 仓库：$PROJECT_DIR"
    fi
    log "  项目目录: $PROJECT_DIR ✓"

    # 端口 (是否被非本应用的进程占用)
    local port_pid
    port_pid=$(ss -tlnp 2>/dev/null | grep ":$BACKEND_PORT " | grep -oP 'pid=\K\d+' || true)
    if [ -n "$port_pid" ]; then
        local port_cmd
        port_cmd=$(ps -p "$port_pid" -o comm= 2>/dev/null || echo "unknown")
        if ! echo "$port_cmd" | grep -q "java"; then
            die "端口 $BACKEND_PORT 被非 Java 进程占用 (PID=$port_pid, cmd=$port_cmd)"
        fi
        log "  端口 $BACKEND_PORT: 被当前后端占用 (PID=$port_pid)，将在停止阶段处理"
    else
        log "  端口 $BACKEND_PORT: 空闲 ✓"
    fi

    success "前置检查通过"
}

# ==================== 停止旧服务 ====================
stop_service() {
    log "=== 停止旧服务 ==="

    local old_pid
    old_pid=$(ps -ef | grep "$BACKEND_JAR" | grep -v grep | awk '{print $2}')

    if [ -z "$old_pid" ]; then
        log "  未发现运行中的旧进程"
        return 0
    fi

    # 记录旧进程信息用于回滚
    local old_start
    old_start=$(ps -p "$old_pid" -o lstart= 2>/dev/null || echo "unknown")
    log "  旧进程: PID=$old_pid, 启动时间=$old_start"

    # Step 1: SIGTERM 优雅关闭
    log "  发送 SIGTERM ..."
    kill -15 "$old_pid" 2>/dev/null || true

    local waited=0
    while [ $waited -lt $SHUTDOWN_GRACE_SEC ]; do
        if ! ps -p "$old_pid" > /dev/null 2>&1; then
            success "旧进程优雅退出 (PID=$old_pid, 耗时 ${waited}s)"
            break
        fi
        sleep 1
        waited=$((waited + 1))
        # 每 10 秒汇报一次状态
        if [ $((waited % 10)) -eq 0 ] && [ $waited -gt 0 ]; then
            log "  等待中... (${waited}s/${SHUTDOWN_GRACE_SEC}s)"
        fi
    done

    # Step 2: 仍未退出 → SIGKILL
    if ps -p "$old_pid" > /dev/null 2>&1; then
        warn "优雅关闭超时 (${SHUTDOWN_GRACE_SEC}s)，执行 kill -9"
        kill -9 "$old_pid" 2>/dev/null || true
        sleep 3
        if ps -p "$old_pid" > /dev/null 2>&1; then
            die "无法杀死进程 PID=$old_pid，请手动处理"
        fi
        log "  进程已强制终止"
    fi

    # Step 3: 确认端口释放
    local retries=10
    while [ $retries -gt 0 ]; do
        if ! ss -tlnp 2>/dev/null | grep -q ":$BACKEND_PORT "; then
            log "  端口 $BACKEND_PORT 已释放 ✓"
            return 0
        fi
        sleep 2
        retries=$((retries - 1))
    done

    # 最后手段：fuser 释放端口
    warn "端口 $BACKEND_PORT 未自动释放，使用 fuser 强制释放"
    fuser -k "${BACKEND_PORT}/tcp" 2>/dev/null || true
    sleep 2
}

# ==================== 版本备份 ====================
backup_current() {
    if [ -f "$BACKEND_DIR/$BACKEND_JAR" ]; then
        mkdir -p "$BACKUP_DIR"
        local backup_name="${BACKEND_JAR%.jar}_$(date +%Y%m%d_%H%M%S).jar"
        cp "$BACKEND_DIR/$BACKEND_JAR" "$BACKUP_DIR/$backpack_name"
        log "  已备份: $backup_name"

        # 清理旧备份，只保留最近 N 个
        local count
        count=$(ls -1t "$BACKUP_DIR"/*.jar 2>/dev/null | wc -l)
        if [ "$count" -gt "$BACKUP_KEEP" ]; then
            ls -1t "$BACKUP_DIR"/*.jar | tail -n +$((BACKUP_KEEP + 1)) | xargs rm -f
            log "  清理旧备份，保留最近 $BACKUP_KEEP 个"
        fi
    fi
}

# ==================== 回滚 ====================
rollback() {
    error "=== 健康检查失败，开始回滚 ==="

    # 停止新进程
    local new_pid="$1"
    if [ -n "$new_pid" ] && ps -p "$new_pid" > /dev/null 2>&1; then
        log "  停止新进程 PID=$new_pid"
        kill -9 "$new_pid" 2>/dev/null || true
        sleep 2
    fi

    # 恢复最近的备份
    local latest_backup
    latest_backup=$(ls -1t "$BACKUP_DIR"/*.jar 2>/dev/null | head -1)
    if [ -n "$latest_backup" ]; then
        log "  恢复备份: $latest_backup"
        cp -f "$latest_backup" "$BACKEND_DIR/$BACKEND_JAR"
    else
        error "没有可用的备份文件！"
        return 1
    fi

    # 启动旧版本
    log "  启动旧版本..."
    cd "$BACKEND_DIR"
    nohup "$JAVA_HOME/bin/java" -Xms512m -Xmx1024m \
        -jar "$BACKEND_JAR" --server.port="$BACKEND_PORT" \
        > ./logs/console.log 2>&1 &

    local rollback_pid=$!
    log "  回滚进程 PID=$rollback_pid"

    # 等待旧版本启动
    for i in $(seq 1 $HEALTH_CHECK_RETRIES); do
        if curl -s -o /dev/null -w "%{http_code}" \
            "http://127.0.0.1:${BACKEND_PORT}/actuator/health" 2>/dev/null | grep -q "200"; then
            success "回滚成功 (PID=$rollback_pid)"
            return 0
        fi
        if ! ps -p "$rollback_pid" > /dev/null 2>&1; then
            break
        fi
        sleep $HEALTH_CHECK_INTERVAL
    done

    error "回滚失败！请手动检查！"
    return 1
}

# ==================== 启动新服务 ====================
start_service() {
    log "=== 启动新服务 ==="

    cd "$BACKEND_DIR"

    # 确保日志目录
    mkdir -p "$BACKEND_DIR/logs"

    nohup "$JAVA_HOME/bin/java" \
        -Xms512m -Xmx1024m \
        -jar "$BACKEND_JAR" --server.port="$BACKEND_PORT" \
        > ./logs/console.log 2>&1 &

    local new_pid=$!
    log "  新进程 PID=$new_pid"

    # 健康检查
    log "  等待健康检查..."
    local started=false
    for i in $(seq 1 $HEALTH_CHECK_RETRIES); do
        if ! ps -p "$new_pid" > /dev/null 2>&1; then
            error "新进程已退出！最近日志："
            tail -30 "$BACKEND_DIR/logs/console.log" | while read -r line; do
                error "  $line"
            done
            rollback "$new_pid"
            die "启动失败，已回滚到上一版本"
        fi

        if curl -s -o /dev/null -w "%{http_code}" \
            "http://127.0.0.1:${BACKEND_PORT}/actuator/health" 2>/dev/null | grep -q "200"; then
            success "后端启动成功 (PID=$new_pid, 耗时 $((i * HEALTH_CHECK_INTERVAL))s)"
            started=true
            break
        fi

        # 每 5 次汇报进度
        if [ $((i % 5)) -eq 0 ]; then
            log "  等待中... ($((i * HEALTH_CHECK_INTERVAL))s / $((HEALTH_CHECK_RETRIES * HEALTH_CHECK_INTERVAL))s)"
        fi
        sleep $HEALTH_CHECK_INTERVAL
    done

    if [ "$started" != "true" ]; then
        error "健康检查超时"
        rollback "$new_pid"
        die "启动超时，已回滚到上一版本"
    fi
}

# ==================== 后端部署 ====================
deploy_backend() {
    log ""
    log "╔══════════════════════════════════════════╗"
    log "║         后端部署                          ║"
    log "╚══════════════════════════════════════════╝"

    # 0. 前置检查
    pre_flight

    # 1. 拉取代码 & 构建
    log "=== 构建 ==="
    cd "$PROJECT_DIR"

    log "  拉取最新代码..."
    git fetch origin 2>&1 | tee -a "$BUILD_LOG"
    local old_commit new_commit
    old_commit=$(git rev-parse --short HEAD)
    git reset --hard origin/main 2>&1 | tee -a "$BUILD_LOG"
    new_commit=$(git rev-parse --short HEAD)

    if [ "$old_commit" != "$new_commit" ]; then
        log "  代码更新: ${old_commit} → ${new_commit}"
    else
        log "  代码无变化 (${old_commit})"
    fi

    log "  Maven 构建..."
    mkdir -p "$MAVEN_REPO" && chmod -R 777 "$MAVEN_REPO" 2>/dev/null || true
    if ! $MVN_EXEC clean package -Dmaven.test.skip=true -pl yudao-server -am -T 1C \
        -Dmaven.repo.local="$MAVEN_REPO" >> "$BUILD_LOG" 2>&1; then
        error "Maven 构建失败！查看：tail -100 $BUILD_LOG"
        die "构建失败"
    fi
    success "Maven 构建完成"

    # 2. 检查产物
    local source_jar="$PROJECT_DIR/yudao-server/target/$BACKEND_JAR"
    if [ ! -f "$source_jar" ]; then
        die "编译产物不存在：$source_jar"
    fi
    log "  产物大小: $(du -h "$source_jar" | cut -f1)"

    # 3. 备份当前版本
    backup_current

    # 4. 停止旧服务
    stop_service

    # 5. 部署新 JAR
    log "=== 部署 ==="
    rm -f "$BACKEND_DIR/$BACKEND_JAR"
    cp -f "$source_jar" "$BACKEND_DIR/"
    log "  JAR 已复制到 $BACKEND_DIR/"

    # 6. 启动
    start_service

    log ""
    success "后端部署完成 ($old_commit → $new_commit)"
}

# ==================== 前端部署 ====================
deploy_frontend() {
    log ""
    log "╔══════════════════════════════════════════╗"
    log "║         用户端前端部署                     ║"
    log "╚══════════════════════════════════════════╝"

    log "=== 构建 ==="
    cd "$PROJECT_DIR/frontend"

    log "  pnpm install..."
    pnpm install >> "$BUILD_LOG" 2>&1 || die "pnpm install 失败"

    log "  pnpm build:gold..."
    pnpm build:gold >> "$BUILD_LOG" 2>&1 || die "pnpm build:gold 失败"

    log "=== 部署 ==="
    rm -rf "${FRONTEND_DIR:?}"/*
    cp -r "$PROJECT_DIR/frontend/app/web-gold/dist/"* "$FRONTEND_DIR/"
    success "用户端前端部署完成"
}

deploy_admin() {
    log ""
    log "╔══════════════════════════════════════════╗"
    log "║         管理后台前端部署                    ║"
    log "╚══════════════════════════════════════════╝"

    log "=== 构建 ==="
    cd "$PROJECT_DIR/yudao-ui-admin-vue3"

    log "  pnpm install..."
    pnpm install >> "$BUILD_LOG" 2>&1 || die "pnpm install 失败"

    log "  pnpm build:prod..."
    pnpm build:prod >> "$BUILD_LOG" 2>&1 || die "pnpm build:prod 失败"

    log "=== 部署 ==="
    rm -rf "${ADMIN_DIR:?}"/*
    cp -r "$PROJECT_DIR/yudao-ui-admin-vue3/dist-prod/"* "$ADMIN_DIR/"
    success "管理后台部署完成"
}

# ==================== 部署后摘要 ====================
print_summary() {
    log ""
    log "╔══════════════════════════════════════════╗"
    log "║         部署摘要                          ║"
    log "╚══════════════════════════════════════════╝"
    log "  构建日志: $BUILD_LOG"
    log "  部署日志: $DEPLOY_LOG"
    log "  版本备份: $BACKUP_DIR/ ($(ls -1 "$BACKUP_DIR"/*.jar 2>/dev/null | wc -l) 个)"
    log "  控制台日志: $BACKEND_DIR/logs/console.log"

    # 快速验证
    if curl -s -o /dev/null -w "%{http_code}" \
        "http://127.0.0.1:${BACKEND_PORT}/actuator/health" 2>/dev/null | grep -q "200"; then
        success "健康检查: UP"
    else
        warn "健康检查: DOWN (可能仍在启动中)"
    fi

    # 磁盘
    local disk_use
    disk_use=$(df -h "$BACKEND_DIR" | awk 'NR==2 {print $5}')
    log "  磁盘使用: ${disk_use}"

    log ""
}

# ==================== 主流程 ====================
main() {
    acquire_lock

    # 清理残留 Maven 进程
    pkill -f "maven" 2>/dev/null || true

    echo "" >> "$DEPLOY_LOG"
    log "═══════════════════════════════════════════"
    log "  部署开始 (模式: ${1:-full})"
    log "═══════════════════════════════════════════"

    case "${1:-full}" in
        backend-only)
            deploy_backend
            ;;
        frontend-only)
            deploy_frontend
            ;;
        admin-only)
            deploy_admin
            ;;
        backend|full|"")
            deploy_backend
            deploy_frontend
            deploy_admin
            ;;
        *)
            echo "用法: $0 {backend-only|frontend-only|admin-only|full}"
            release_lock
            exit 1
            ;;
    esac

    print_summary
    log "🏁 部署流水线全部执行完毕 ($(date '+%H:%M:%S'))"
}

main "$@"