MySQL窗口函数实战：排名与聚合应用详解

1. MySQL窗口函数深度解析：排名与聚合实战指南

在数据分析与报表开发中，我们经常需要对数据进行分组排序、计算移动平均值或生成累计统计值。传统SQL通过GROUP BY实现的聚合查询存在明显局限——它会压缩多行数据为单行结果，丢失原始记录的细节信息。而MySQL 8.0引入的窗口函数（Window Functions）完美解决了这一痛点，它能在保留原始行数据的同时，进行复杂的跨行计算。

窗口函数的核心价值在于其"透视"能力：就像透过一个滑动的窗口观察数据子集，这个窗口可以定义为当前行的前N行、后N行或整个分区。我曾在电商促销分析中，仅用一行窗口函数就替代了原先需要多次自连接才能实现的"同品类商品价格排名对比"需求，查询性能提升了20倍。本文将深入剖析排名函数（RANK/DENSE_RANK/ROW_NUMBER）和聚合类窗口函数的使用技巧，包含大量来自生产环境的实战案例。

2. 窗口函数基础与执行原理

2.1 窗口函数语法结构解析

标准窗口函数语法包含四个关键部分：

sql复制function_name([arguments]) 
OVER (
    [PARTITION BY partition_expression_list]
    [ORDER BY order_expression_list]
    [frame_clause]
)

其中PARTITION BY类似GROUP BY的分组，但不会合并行；ORDER BY决定分区内数据的排序方式；frame_clause定义计算范围，比如：

• ROWS BETWEEN 3 PRECEDING AND CURRENT ROW 包含当前行及前3行
• RANGE BETWEEN INTERVAL 7 DAY PRECEDING AND CURRENT ROW 包含最近7天的数据

重要提示：MySQL中窗口函数的执行发生在WHERE、GROUP BY、HAVING之后，但在ORDER BY之前。这意味着不能在WHERE条件中直接引用窗口函数结果，需要通过派生表或CTE实现。

2.2 与GROUP BY聚合的本质区别

通过一个简单例子说明差异。假设有销售表sales_data：

sql复制-- 传统GROUP BY聚合
SELECT product_id, SUM(amount) as total_amount
FROM sales_data
GROUP BY product_id;  -- 丢失了销售日期等明细

-- 窗口函数聚合
SELECT 
    product_id,
    sale_date,
    amount,
    SUM(amount) OVER(PARTITION BY product_id) as product_total
FROM sales_data;  -- 保留所有原始列

窗口函数不会减少结果行数，反而会新增计算列，这是其核心特征。在用户行为分析场景中，这种特性尤其宝贵——我们既能计算用户累计访问次数，又能同时看到每次访问的具体参数。

3. 排名函数实战技巧

3.1 RANK vs DENSE_RANK vs ROW_NUMBER

这三种排名函数的区别常被混淆，通过具体案例说明：

sql复制SELECT 
    student_id,
    exam_score,
    RANK() OVER(ORDER BY exam_score DESC) as rank_score,
    DENSE_RANK() OVER(ORDER BY exam_score DESC) as dense_rank_score,
    ROW_NUMBER() OVER(ORDER BY exam_score DESC) as row_num
FROM exam_results;

当存在并列分数时：

• RANK()：会产生间隔的排名（如1,2,2,4）
• DENSE_RANK()：连续排名（1,2,2,3）
• ROW_NUMBER()：强制连续编号（1,2,3,4）

在电商价格竞争力分析中，我常用DENSE_RANK计算商品在同类中的价格排名，因为间隔的RANK值会导致后续百分比计算出现偏差。

3.2 分区排名典型应用

案例：部门薪资排名

sql复制SELECT 
    emp_name,
    department,
    salary,
    RANK() OVER(PARTITION BY department ORDER BY salary DESC) as dept_rank
FROM employees;

案例：用户连续登录天数排名

sql复制WITH login_streaks AS (
    SELECT 
        user_id,
        COUNT(*) as streak_days
    FROM (
        SELECT 
            user_id,
            login_date,
            DATE_SUB(login_date, INTERVAL ROW_NUMBER() OVER(PARTITION BY user_id ORDER BY login_date) DAY) as grp
        FROM user_logins
    ) t
    GROUP BY user_id, grp
)
SELECT 
    user_id,
    streak_days,
    RANK() OVER(ORDER BY streak_days DESC) as streak_rank
FROM login_streaks
WHERE streak_days >= 7;

这个复杂查询先用ROW_NUMBER识别连续登录日期组，再计算每个连续区间的天数，最后全局排名。曾帮助我们发现核心用户的活跃规律。

4. 聚合窗口函数高级应用

4.1 累计聚合与移动平均

销售累计计算

sql复制SELECT 
    sale_date,
    daily_sales,
    SUM(daily_sales) OVER(ORDER BY sale_date) as running_total,
    AVG(daily_sales) OVER(ORDER BY sale_date 
                          ROWS BETWEEN 6 PRECEDING AND CURRENT ROW) as weekly_avg
FROM sales;

库存预警分析

sql复制SELECT 
    product_id,
    date,
    stock_qty,
    AVG(stock_qty) OVER(PARTITION BY product_id 
                        ORDER BY date 
                        RANGE BETWEEN INTERVAL 7 DAY PRECEDING AND CURRENT ROW) as avg_weekly_stock,
    stock_qty - LAG(stock_qty, 1) OVER(PARTITION BY product_id ORDER BY date) as daily_change
FROM inventory
WHERE date >= CURRENT_DATE - INTERVAL 30 DAY;

4.2 首末值获取技巧

FIRST_VALUE和LAST_VALUE配合frame子句特别适合分析用户行为序列：

sql复制SELECT 
    user_id,
    event_time,
    event_type,
    FIRST_VALUE(event_type) OVER(PARTITION BY user_id 
                                 ORDER BY event_time 
                                 ROWS BETWEEN CURRENT ROW AND 5 FOLLOWING) as next_5_events,
    LAST_VALUE(page_url) OVER(PARTITION BY user_id, session_id 
                              ORDER BY event_time
                              RANGE BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW) as last_page_in_session
FROM user_events;

在转化漏斗分析中，这种查询能高效识别用户从浏览到下单的完整路径。

5. 性能优化与常见陷阱

5.1 执行计划解读

使用EXPLAIN分析窗口函数查询时，重点关注：

• "window"：表示窗口函数计算阶段
• "using filesort"：可能需要对大结果集排序
• "using temporary"：可能创建临时表

优化案例：

sql复制-- 优化前：全表排序
EXPLAIN 
SELECT student_id, RANK() OVER(ORDER BY score DESC) 
FROM exam_results;

-- 优化后：利用索引
ALTER TABLE exam_results ADD INDEX idx_score (score DESC);
EXPLAIN 
SELECT student_id, RANK() OVER(ORDER BY score DESC) 
FROM exam_results USE INDEX(idx_score);

5.2 分区设计黄金法则

1. 分区粒度：PARTITION BY列的选择直接影响性能。经验值是每个分区包含500-5000行效率最佳
2. 排序代价：ORDER BY应尽量使用索引列，避免filesort
3. 内存控制：通过SET windowing_use_high_precision=OFF可降低内存使用（但可能损失精度）

5.3 典型错误排查

错误1：忽略NULL值排序

sql复制-- NULL会排在最后导致计算偏差
SELECT RANK() OVER(ORDER BY nullable_column) FROM tbl;

-- 正确做法
SELECT RANK() OVER(ORDER BY COALESCE(nullable_column, 0)) FROM tbl;

错误2：框架范围定义不当

sql复制-- 错误：缺少CURRENT ROW会导致意外结果
SELECT SUM(val) OVER(ROWS 3 PRECEDING) FROM tbl;

-- 正确
SELECT SUM(val) OVER(ROWS BETWEEN 3 PRECEDING AND CURRENT ROW) FROM tbl;

错误3：在WHERE中使用窗口函数

sql复制-- 语法错误
SELECT * FROM tbl WHERE RANK() OVER(...) > 5;

-- 正确方案
WITH ranked_data AS (
    SELECT *, RANK() OVER(...) as rnk 
    FROM tbl
)
SELECT * FROM ranked_data WHERE rnk > 5;

6. 真实业务场景综合案例

6.1 电商场景：竞品价格监控系统

sql复制WITH product_prices AS (
    SELECT 
        product_id,
        competitor,
        price,
        collection_date,
        -- 计算各平台价格排名
        RANK() OVER(PARTITION BY product_id, competitor 
                    ORDER BY collection_date DESC) as latest_flag,
        -- 价格波动分析
        AVG(price) OVER(PARTITION BY product_id 
                        ORDER BY collection_date 
                        RANGE BETWEEN INTERVAL 7 DAY PRECEDING AND CURRENT ROW) as weekly_avg,
        -- 最低价标识
        FIRST_VALUE(competitor) OVER(PARTITION BY product_id 
                                     ORDER BY price 
                                     RANGE BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING) as cheapest_vendor
    FROM competitor_price_tracking
    WHERE collection_date >= CURRENT_DATE - INTERVAL 30 DAY
)
SELECT 
    product_id,
    competitor,
    price,
    weekly_avg,
    cheapest_vendor,
    (price - weekly_avg) / weekly_avg as price_deviation
FROM product_prices
WHERE latest_flag = 1  -- 只取最新记录
ORDER BY product_id, price_deviation DESC;

这个查询实现了：

1. 识别各平台最新价格
2. 计算7日移动平均价
3. 标记历史最低价供应商
4. 计算当前价格偏离度

6.2 金融场景：交易风险检测

sql复制SELECT 
    account_id,
    transaction_time,
    amount,
    SUM(amount) OVER(PARTITION BY account_id 
                     ORDER BY transaction_time 
                     RANGE BETWEEN INTERVAL 1 HOUR PRECEDING AND CURRENT ROW) as hourly_total,
    COUNT(*) OVER(PARTITION BY account_id 
                  ORDER BY transaction_time 
                  RANGE BETWEEN INTERVAL 1 DAY PRECEDING AND CURRENT ROW) as daily_count,
    AVG(amount) OVER(PARTITION BY merchant_id 
                     ORDER BY transaction_time 
                     RANGE BETWEEN INTERVAL 7 DAY PRECEDING AND CURRENT ROW) as merchant_avg
FROM transactions
WHERE transaction_time >= NOW() - INTERVAL '2' HOUR
HAVING hourly_total > 10000 OR daily_count > 20;

该方案实时监测：

• 单账户1小时内累计交易超限额
• 单日交易频次异常
• 交易金额偏离商户平均水平

7. 进阶技巧与最佳实践

7.1 动态分区策略

对于不确定分区粒度的场景，可以使用表达式动态分区：

sql复制SELECT 
    user_id,
    event_time,
    -- 按小时动态分区计算会话内事件序列
    SUM(CASE WHEN event_type = 'purchase' THEN 1 ELSE 0 END) 
        OVER(PARTITION BY user_id, 
                          DATE_FORMAT(event_time, '%Y%m%d%H')
             ORDER BY event_time) as hourly_purchase_count
FROM user_events;

7.2 窗口函数组合技

多个窗口函数组合能实现复杂分析：

sql复制SELECT 
    product_id,
    month,
    sales,
    -- 计算市场份额
    sales / SUM(sales) OVER(PARTITION BY month) as market_share,
    -- 计算同类产品排名
    RANK() OVER(PARTITION BY month ORDER BY sales DESC) as category_rank,
    -- 计算三个月移动平均
    AVG(sales) OVER(ORDER BY month 
                    ROWS BETWEEN 2 PRECEDING AND CURRENT ROW) as moving_avg,
    -- 计算同比增长
    sales - LAG(sales, 12) OVER(ORDER BY month) as yoy_growth
FROM product_sales;

7.3 与CTE的配合使用

窗口函数与CTE结合能大幅提升复杂查询的可读性：

sql复制WITH monthly_sales AS (
    SELECT 
        salesperson,
        DATE_FORMAT(sale_date, '%Y-%m') as month,
        SUM(amount) as total_sales
    FROM sales
    GROUP BY salesperson, DATE_FORMAT(sale_date, '%Y-%m')
),
ranked_sales AS (
    SELECT 
        salesperson,
        month,
        total_sales,
        RANK() OVER(PARTITION BY month ORDER BY total_sales DESC) as monthly_rank,
        total_sales - LAG(total_sales) OVER(PARTITION BY salesperson ORDER BY month) as mom_growth
    FROM monthly_sales
)
SELECT 
    salesperson,
    month,
    total_sales,
    monthly_rank,
    mom_growth,
    AVG(total_sales) OVER(PARTITION BY salesperson 
                          ORDER BY month 
                          ROWS BETWEEN 3 PRECEDING AND CURRENT ROW) as quarterly_avg
FROM ranked_sales
WHERE monthly_rank <= 5  -- 每月Top5销售
ORDER BY month, monthly_rank;

这个查询实现了：

1. 按月汇总销售数据
2. 计算每月排名
3. 计算环比增长
4. 计算季度移动平均
5. 筛选每月Top5

在实际项目中，窗口函数的使用边界取决于业务需求与数据规模。我曾遇到一个包含15个嵌套窗口函数的报表查询，通过重构为多个CTE并合理设计分区策略，将执行时间从45分钟优化到90秒。关键是要理解每个窗口函数的计算成本，并利用EXPLAIN持续监控性能。