266 lines
9.0 KiB
Markdown
266 lines
9.0 KiB
Markdown
---
|
||
tags: [MySQL, 子查询, EXISTS, IN, 派生表]
|
||
create time: 2026-05-16 00:00
|
||
---
|
||
|
||
# 子查询与派生表
|
||
|
||
## 概述
|
||
|
||
子查询是嵌套在另一个查询中的 SELECT 语句。它可以在 WHERE、FROM、SELECT 等多个位置出现，每种位置的语义和执行方式不同。
|
||
|
||
## 分类
|
||
|
||
```mermaid
|
||
graph BT
|
||
subgraph "标量子查询"
|
||
S1["返回一行一列<br/>用在 SELECT / WHERE"]
|
||
end
|
||
subgraph "行子查询"
|
||
S2["返回一行多列<br/>用在 ROW() 比较"]
|
||
end
|
||
subgraph "列子查询"
|
||
S3["返回多行一列<br/>用在 IN / ANY / ALL"]
|
||
end
|
||
subgraph "表子查询（派生表）"
|
||
S4["返回多行多列<br/>用在 FROM 子句"]
|
||
end
|
||
subgraph "EXISTS 子查询"
|
||
S5["返回布尔值<br/>用于 EXISTS / NOT EXISTS"]
|
||
end
|
||
```
|
||
|
||
## 标量子查询
|
||
|
||
```sql
|
||
-- 用法 1：在 SELECT 中调用
|
||
SELECT
|
||
username,
|
||
(SELECT COUNT(*) FROM orders WHERE user_id = users.id) AS order_count
|
||
FROM users;
|
||
|
||
-- 用法 2：在 WHERE 中等价于常量
|
||
SELECT * FROM products
|
||
WHERE price > (SELECT AVG(price) FROM products);
|
||
|
||
-- 用法 3：在 INSERT 中赋值
|
||
INSERT INTO reports (month, order_total)
|
||
VALUES ('2026-05', (SELECT SUM(amount) FROM orders WHERE MONTH(created_at) = 5));
|
||
```
|
||
|
||
> [!WARNING] 标量子查询的性能隐患
|
||
> MySQL 8.0.21 之前，标量子查询**无法被物化**，会导致每次执行都重新计算（N+1 问题）。
|
||
> ```sql
|
||
> -- ❌ 慢：每个用户都要查一次 orders 表
|
||
> SELECT u.username,
|
||
> (SELECT COUNT(*) FROM orders o WHERE o.user_id = u.id) AS cnt
|
||
> FROM users u;
|
||
> -- 如果有 10 万用户，就要查 10 万次 orders 表
|
||
>
|
||
> -- ✅ 换成 JOIN 或窗口函数
|
||
> SELECT u.username, COALESCE(SUB.cnt, 0) AS cnt
|
||
> FROM users u
|
||
> LEFT JOIN (
|
||
> SELECT user_id, COUNT(*) AS cnt
|
||
> FROM orders GROUP BY user_id
|
||
> ) SUB ON u.id = SUB.user_id;
|
||
> ```
|
||
|
||
## IN vs EXISTS
|
||
|
||
这是面试经典题，也是实际中最纠结的选择。
|
||
|
||
```sql
|
||
-- IN 子查询
|
||
SELECT * FROM users
|
||
WHERE id IN (SELECT user_id FROM orders);
|
||
|
||
-- EXISTS 子查询
|
||
SELECT * FROM users u
|
||
WHERE EXISTS (SELECT 1 FROM orders o WHERE o.user_id = u.id);
|
||
```
|
||
|
||
```mermaid
|
||
flowchart TD
|
||
A["Optimizer 选择策略"] --> B{"哪张表更小？"}
|
||
|
||
B -->|"子查询结果小"<| C["转换 IN → EXISTS<br/>先跑子查询，外层仅匹配结果"]
|
||
B -->|"子查询结果大"<| D["转换 EXISTS → IN<br/>外层先行过滤，减少子查询次数"]
|
||
|
||
C --> E["NOT IN 永远转不成 EXISTS!<br/>NULL 值会导致语义错误"]
|
||
D --> F["NOT EXISTS 是最安全的反查写法"]
|
||
|
||
style E fill:#EE5A24,color:#fff
|
||
style F fill:#00D866,color:#fff
|
||
```
|
||
|
||
> [!QUESTION] NOT IN 和 NOT EXISTS 有什么区别？
|
||
> ```sql
|
||
> -- ⚠️ 致命陷阱
|
||
> SELECT * FROM users WHERE id NOT IN (SELECT user_id FROM orders);
|
||
>
|
||
> -- 如果子查询返回任何 NULL 值，整个 NOT IN 结果为 UNKNOWN
|
||
> -- 最终返回空结果集！
|
||
>
|
||
> -- ✅ 正确写法
|
||
> SELECT * FROM users
|
||
> WHERE id NOT IN (SELECT user_id FROM orders WHERE user_id IS NOT NULL);
|
||
>
|
||
> -- 或者直接用 NOT EXISTS（更安全）
|
||
> SELECT * FROM users u
|
||
> WHERE NOT EXISTS (SELECT 1 FROM orders o WHERE o.user_id = u.id);
|
||
> ```
|
||
|
||
### 选型决策：IN vs EXISTS vs JOIN
|
||
|
||
实际开发中三者往往能写出等价逻辑，选法参考下表：
|
||
|
||
| 场景 | 推荐写法 | 理由 |
|
||
|------|----------|------|
|
||
| 外大内小（外层百万、子查询几百） | `EXISTS` | 外层每行只需匹配一条即可停止，短路效应明显 |
|
||
| 外小内大 | `IN` | Optimizer 会先物化子查询结果再匹配，效率高 |
|
||
| 子查询含 `NULL` 值可能 | `EXISTS` / `NOT EXISTS` | `NOT IN` 遇到 NULL 直接返回空集（见上文陷阱） |
|
||
| 需要返回外键表的完整列 | `JOIN` + `DISTINCT` | EXISTS 只能判断存在性，无法获取关联表字段 |
|
||
| 只是判断"有没有" | `EXISTS` | 语义最清晰，性能也最优 |
|
||
|
||
> [!TIP] 经验法则
|
||
> **不确定时优先用 EXISTS**——它的语义是"是否存在"而非"值是否匹配"，即使 MySQL Optimizer 最终把 IN 和 EXISTS 优化成执行计划也一样。用 EXISTS 能让读代码的人一眼看懂意图，同时给 Optimizer 留下优化空间。
|
||
|
||
## 派生表（Derived Table / Subquery in FROM）
|
||
|
||
```sql
|
||
-- 派生表：子查询作为一个虚拟表出现在 FROM 位置
|
||
SELECT dept, avg_salary
|
||
FROM (
|
||
SELECT department AS dept, AVG(salary) AS avg_salary
|
||
FROM employees
|
||
WHERE status = 'active'
|
||
GROUP BY department
|
||
) AS dept_stats
|
||
WHERE avg_salary > 15000
|
||
ORDER BY avg_salary DESC;
|
||
```
|
||
|
||
### 物化（Materialization）
|
||
|
||
MySQL 是否将子查询物化为临时表，取决于查询复杂度：
|
||
|
||
```mermaid
|
||
flowchart LR
|
||
A["子查询"] --> B{"能否被推入外层？"}
|
||
B -->|能| C["Flattening<br/>展开为 JOIN，零额外开销 ✅"]
|
||
B -->|不能| D{"是否需要聚合？"}
|
||
D -->|是| E["Materialization<br/>物化为临时表 ⚠️"]
|
||
D -->|否| F["Semi-join 优化<br/>半连接优化 ✅"]
|
||
|
||
style C fill:#00D866,color:#fff
|
||
style F fill:#00B6BC,color:#fff
|
||
style E fill:#FF9F43,color:#000
|
||
```
|
||
|
||
```sql
|
||
-- ✅ 可以被 Flattening 优化（无额外开销）
|
||
SELECT u.id, o.amount
|
||
FROM users u
|
||
JOIN (SELECT user_id, amount FROM orders) AS o ON u.id = o.user_id;
|
||
-- Optimizer 将其转换为普通的 JOIN
|
||
|
||
-- ❌ 必须 Materialization（有临时表开销）
|
||
SELECT u.id, SUB.total
|
||
FROM users u
|
||
JOIN (
|
||
SELECT user_id, SUM(amount) AS total
|
||
FROM orders GROUP BY user_id
|
||
) AS SUB ON u.id = SUB.user_id;
|
||
-- 子查询因为有 GROUP BY，必须先物化成临时表
|
||
```
|
||
|
||
> [!NOTE] 如何判断是否物化？
|
||
> 使用 `EXPLAIN FORMAT=JSON`：
|
||
> ```json
|
||
> {
|
||
> "select_type": "DERIVED",
|
||
> "materialized_from_subquery": { ... }
|
||
> }
|
||
> ```
|
||
> 如果出现 `"using_temporary_table": true`，说明使用了临时表。
|
||
|
||
实战中更常用的方式是 `EXPLAIN` + 观察 type 和 Extra：
|
||
|
||
```sql
|
||
-- ❌ 物化型派生表 → Extra 出现 "Using temporary"
|
||
EXPLAIN SELECT u.id, SUB.total
|
||
FROM users u
|
||
JOIN (SELECT user_id, SUM(amount) AS total FROM orders GROUP BY user_id) AS SUB ON u.id = SUB.user_id;
|
||
|
||
-- id | select_type | table | type | Extra
|
||
-- ---|----------------|------------|-------|----------------------------
|
||
-- 1 | PRIMARY | <derived2> | ALL | NULL
|
||
-- 1 | PRIMARY | u | ALL | Using where
|
||
-- 2 | DERIVED | orders | ALL | Using temporary; Using filesort
|
||
|
||
-- ✅ 可 Flattening → 无临时表，Extra 干净
|
||
EXPLAIN SELECT u.id, o.amount
|
||
FROM users u JOIN (SELECT user_id, amount FROM orders) AS o ON u.id = o.user_id;
|
||
|
||
-- id | select_type | table | type | Extra
|
||
-- 1 | SIMPLE | orders| ALL | NULL
|
||
-- 1 | SIMPLE | u | index | Primary key
|
||
```
|
||
|
||
> [!TIP] 性能优化技巧
|
||
> - 当发现派生表产生临时表时，考虑**手动提升为 CTE**（MySQL 8.0+），有时能改变 Optimizer 行为
|
||
> - `optimizer_switch='derived_merge=on'`（默认开启）控制是否允许 Flattening
|
||
> - 对于超大结果集物化，注意 `tmp_table_size` / `max_heap_table_size` 限制
|
||
|
||
## ANY / ALL / SOME
|
||
|
||
```sql
|
||
-- ANY：子查询返回的值中，任意一个满足即可
|
||
SELECT product_name, price
|
||
FROM products
|
||
WHERE price > ANY (SELECT price FROM products WHERE category = 'premium');
|
||
-- 只要比任意一件 premium 产品便宜就行
|
||
|
||
-- ALL：必须大于子查询返回的所有值
|
||
SELECT product_name, price
|
||
FROM products
|
||
WHERE price > ALL (SELECT price FROM products WHERE category = 'premium');
|
||
-- 必须比所有 premium 产品都贵（即最大值之上）
|
||
|
||
-- SOME 等价于 ANY（同义词）
|
||
```
|
||
|
||
```mermaid
|
||
graph TB
|
||
P["Products: 10, 50, 100, 200"]
|
||
|
||
ANY -->|"price > ANY(...)"<| A1["> 10 OR > 50 OR > 100 OR > 200"]
|
||
A1 --> R1["结果: 所有 > 10 的商品"]
|
||
|
||
ALL -->|"price > ALL(...)"<| A2["> 10 AND > 50 AND > 100 AND > 200"]
|
||
A2 --> R2["结果: 只有 > 200 的商品"]
|
||
|
||
style R1 fill:#00B6BC,color:#fff
|
||
style R2 fill:#C44569,color:#fff
|
||
```
|
||
|
||
## 总结：核心要点回顾
|
||
|
||
| 主题 | 一句话 |
|
||
|------|--------|
|
||
| 标量子查询 | MySQL 8.0.21 之前不可物化，百万行数据必踩 N+1 陷阱，优先改写为 JOIN |
|
||
| IN vs EXISTS | 语义上 IN 看值、EXISTS 看存在性；不确定时选 EXISTS，更安全的默认选项 |
|
||
| NOT IN 陷阱 | 子查询出现 NULL 即返回空集，生产中几乎永远该用 `NOT EXISTS` 替代 |
|
||
| 派生表物化 | 有 GROUP BY / LIMIT / UNION 等操作的子查询无法被展开，必然产生临时表开销 |
|
||
| ANY / ALL | 对应 SQL 的 OR / AND 累加，ALL 在空子查询结果时恒返回 TRUE（反直觉，需注意） |
|
||
|
||
> [!TIP] 核心心法
|
||
> **子查询不是万能的——它首先是为了表达清晰，其次才是性能。**
|
||
> 写完后务必跑 `EXPLAIN`，确认没有意料之外的临时表或多表扫描。当数据量上去后，能改写成 JOIN 的子查询就尽量改写，因为 JOIN 的执行路径对 Optimizer 更加透明。
|
||
|
||
## 关联笔记
|
||
|
||
- [[hhs/MySQL/12-DQL SELECT 全解析]] — SELECT 执行顺序与子查询的关系
|
||
- [[hhs/MySQL/13-JOIN 原理与优化]] — EXISTS 子查询常可重写为 JOIN
|