数据库
首页 > 数据库> > mysql 8.0.28 查询语句执行顺序实测结果

mysql 8.0.28 查询语句执行顺序实测结果

作者:互联网

TL;NRs

根据实测结果,MySQL8.0.28 中 SQL 语句的执行顺序为:

(8)     SELECT 
(5)     DISTINCT <select_list>
(1)     FROM <left_table>
(3)     <join_type> JOIN <right_table>
(4)     ON <join_condition>
(2)     WHERE <where_condition>
(6)     GROUP BY <group_by_list>
(7)     HAVING <having_condition>
(9)     ORDER BY <order_by_condition>
(10)    LIMIT <limit_number>

引言

关于 SQL 语句的执行顺序,常见的是以下版本。然而该版本却与实测结果不符。

(7)     SELECT 
(8)     DISTINCT <select_list>
(1)     FROM <left_table>
(3)     <join_type> JOIN <right_table>
(2)     ON <join_condition>
(4)     WHERE <where_condition>
(5)     GROUP BY <group_by_list>
(6)     HAVING <having_condition>
(9)     ORDER BY <order_by_condition>
(10)    LIMIT <limit_number>

MySQL 可以通过 EXPLAIN ANALYZE sql_statement 显示真实的执行过程。那么可以通过一个复杂的语句完成测试。

准备数据

准备三个表 t1, t2, t3, 其中数据分别为:

t1

t2

t3

测试

执行以下语句

EXPLAIN ANALYZE
SELECT
DISTINCT COUNT(p.id) AS cnt, COUNT(e.id) AS nn
FROM t1 p
LEFT JOIN t2 q ON p.id > q.id
INNER JOIN t2 w ON q.id < w.id
RIGHT JOIN t3 e ON w.id = e.id
WHERE p.id < 10
GROUP BY p.id
HAVING cnt > 3
ORDER BY cnt DESC, nn DESC
LIMIT 1;

结果为:

-> Limit: 10 row(s)  (actual time=0.394..0.395 rows=1 loops=1)
    -> Sort with duplicate removal: cnt DESC, nn DESC  (actual time=0.393..0.394 rows=1 loops=1)
        -> Filter: (cnt > 3)  (actual time=0.372..0.374 rows=5 loops=1)
            -> Table scan on <temporary>  (actual time=0.001..0.001 rows=6 loops=1)
                -> Aggregate using temporary table  (actual time=0.370..0.372 rows=6 loops=1)
                    -> Inner hash join (e.id = w.id)  (cost=4.73 rows=3) (actual time=0.314..0.324 rows=32 loops=1)
                        -> Table scan on e  (cost=0.13 rows=5) (actual time=0.008..0.016 rows=5 loops=1)
                        -> Hash
                            -> Filter: (q.id < w.id)  (cost=3.15 rows=3) (actual time=0.265..0.282 rows=32 loops=1)
                                -> Inner hash join (no condition)  (cost=3.15 rows=3) (actual time=0.259..0.271 rows=72 loops=1)
                                    -> Covering index scan on w using PRIMARY  (cost=0.13 rows=3) (actual time=0.007..0.010 rows=4 loops=1)
                                    -> Hash
                                        -> Nested loop inner join  (cost=2.10 rows=3) (actual time=0.084..0.232 rows=18 loops=1)
                                            -> Filter: (p.id < 10)  (cost=1.05 rows=3) (actual time=0.036..0.051 rows=7 loops=1)
                                                -> Table scan on p  (cost=1.05 rows=8) (actual time=0.034..0.046 rows=8 loops=1)
                                            -> Filter: (p.id > q.id)  (cost=0.13 rows=1) (actual time=0.021..0.025 rows=3 loops=7)
                                                -> Covering index range scan on q (re-planned for each iteration)  (cost=0.13 rows=3) (actual time=0.021..0.024 rows=3 loops=7)

结果分析

这是一个调用栈,还原其执行过程为:

筛选 LIMIT 10 {
    排序 ORDER BY cnt DESC, nn DESC {
        调用 HAVING cnt > 3 过滤器 {
            读取临时聚合表 {
                聚合 {
                    第三次联结 RIGHT JOIN t3 e ON w.id = e.id {
                        扫描表 e ;
                        第二次联结 INNER JOIN t2 w ON q.id < w.id {
                            扫描表 w {
                                使用主键扫描
                                得到 4 行
                            }
                            第一次联结 t1 p LEFT JOIN t2 q ON p.id > q.id {
                                扫描表 p {
                                    使用 WHERE p.id < 10 过滤器
                                    共 8 行,返回 7 行
                                }
                                循环扫描表 q {
                                    7 次循环 {
                                        使用过滤器 ON  p.id > q.id 
                                    }
                                }
                                执行哈希,共 21 行,返回 18 行
                            }
                            执行全连接,获得 4 * 18 = 72 行
                            执行 ON q.id < w.id 过滤器,剩余 32 行
                        }
                        执行相等联结 e.id = w.id, 返回 32 行
                    }
                    完成所有的联结,获得 32 行
                    进行聚合 GROUP BY p.id 获得 6 行
                }
                读取临时聚合表,获得 6 行 
            }
            执行过滤,剩余 5 行 
        }
        去重,剩余 2 行 
        排序
		返回 1 行
    }
    输出前 1 项
}

可以看到:

结论

根据实测结果,MySQL8.0.28 中 SQL 语句的执行顺序为:

(8)     SELECT 
(5)     DISTINCT <select_list>
(1)     FROM <left_table>
(3)     <join_type> JOIN <right_table>
(4)     ON <join_condition>
(2)     WHERE <where_condition>
(6)     GROUP BY <group_by_list>
(7)     HAVING <having_condition>
(9)     ORDER BY <order_by_condition>
(10)    LIMIT <limit_number>

标签:8.0,rows,actual,..,time,28,mysql,loops,id
来源: https://www.cnblogs.com/CounterX/p/16583776.html