在 FROM 子句中选择与子查询的速度差异

Question

我有一个 IMDB 数据库，我正在尝试计算每年制作的平均演员人数。问题是关于从子查询中选择的速度差异。

我的查询是：

SELECT AVG(sub.num) 
FROM 
   (SELECT 
        COUNT(production_cast.person_id) AS num, 
        production.production_year AS pyear 
    FROM production_cast 
    INNER JOIN production ON production.id = production_cast.production_id
    GROUP BY production.id) sub
GROUP BY(sub.pyear)

但是为了简单起见，以下是问题所涉及的两个查询：

带子查询

SELECT sub.num 
FROM 
    (SELECT 
         COUNT(production_cast.person_id) AS num, 
         production.production_year AS pyear  
     FROM production_cast 
     INNER JOIN production ON production.id = production_cast.production_id
     GROUP BY production.id) sub 

没有子查询：

SELECT 
    COUNT(production_cast.person_id) AS num, 
    production.production_year AS pyear  
FROM production_cast 
INNER JOIN production ON production.id = production_cast.production_id
GROUP BY production.id

第二个的持续时间不到一秒，第一个永远不会完成。 -超过5分钟-。

带有子查询的解释

+-------------+------------------+-------+-----------------------------------+-------------+
| select_type | table            | type  | key                               | Extra       |
+-------------+------------------+-------+-----------------------------------+-------------+
| PRIMARY     |  <derived2>      | ALL   | NULL                              | NULL        |
| DERIVED     |  production      | index | idx_Production_id_production_year | Using index |
| DERIVED     |  production_cast | ref   |  production_id                    | NULL        |
+-------------+------------------+-------+-----------------------------------+-------------+

没有子查询的解释：

+-------------+-----------------+-----------------------------------+------------+
| select_type | table           | key                               | Extra      |
+-------------+-----------------+-----------------------------------+------------+
| SIMPLE      | production      | idx_Production_id_production_year | Usingindex |
| SIMPLE      | production_cast | production_id                     | NULL       |
+-------------+-----------------+-----------------------------------+------------+

造成这种性能差异的原因是什么？有什么办法可以预防？

Answer 1

派生：在临时表中扑通一声，没有索引的机会

子查询：像瘟疫一样避免

来自https://www.percona.com/blog/2006/08/31/derived-tables-and-views-performance/

“要注意的是，即使执行 EXPLAIN 语句，派生表也会被物化。因此，如果您在 from 子句中选择错误，即忘记了连接条件，您可能会让 EXPLAIN 永远运行。”