数据库内核月报

数据库内核月报 - 2017 / 04

MySQL · 特性分析 · common table expression

Author: 张远

common table expression

Common table expression简称CTE,由SQL:1999标准引入, 目前支持CTE的数据库有Teradata, DB2, Firebird, Microsoft SQL Server, Oracle (with recursion since 11g release 2), PostgreSQL (since 8.4), MariaDB (since 10.2), SQLite (since 3.8.3), HyperSQL and H2 (experimental), MySQL8.0.

CTE的语法如下:

WITH [RECURSIVE] with_query [, ...]
SELECT...

with_query:
query_name [ (column_name [,...]) ] AS (SELECT ...)

以下图示来自MariaDB

Non-recursive CTEs
screenshot.png

Recursive CTEs
screenshot.png

CTE的使用

例如以下两个语句表达的是同一语义,使用CTE比未使用CTE的嵌套查询更简洁明了。

1) 使用嵌套子查询

SELECT MAX(txt), MIN(txt)
FROM
(
  SELECT concat(cte2.txt, cte3.txt) as txt
  FROM
  (
    SELECT CONCAT(cte1.txt,'is a ') as txt
    FROM
    (
      SELECT 'This ' as txt
    ) as cte1
  ) as cte2,
  (
    SELECT 'nice query' as txt
    UNION
    SELECT 'query that rocks'
    UNION
    SELECT 'query'
  ) as cte3
) as cte4;

2) 使用CTE

WITH cte1(txt) AS (SELECT "This "),
     cte2(txt) AS (SELECT CONCAT(cte1.txt,"is a ") FROM cte1),
     cte3(txt) AS (SELECT "nice query" UNION
                   SELECT "query that rocks" UNION
                   SELECT "query"),
     cte4(txt) AS (SELECT concat(cte2.txt, cte3.txt) FROM cte2, cte3)
SELECT MAX(txt), MIN(txt) FROM cte4;
create table t1(id int, value char(10), parent_id int);
insert into t1 values(1, 'A', NULL);
insert into t1 values(2, 'B', 1);
insert into t1 values(3, 'C', 1);
insert into t1 values(4, 'D', 1);
insert into t1 values(5, 'E', 2);
insert into t1 values(6, 'F', 2);
insert into t1 values(7, 'G', 4);
insert into t1 values(8, 'H', 6);

1) 层序遍历

with recursive cte as (
  select id, value, 0 as level from t1 where parent_id is null
  union all
  select t1.id, t1.value, cte.level+1 from cte join t1 on t1.parent_id=cte.id)
select * from cte;
+------+-------+-------+
| id   | value | level |
+------+-------+-------+
|    1 | A     |     0 |
|    2 | B     |     1 |
|    3 | C     |     1 |
|    4 | D     |     1 |
|    5 | E     |     2 |
|    6 | F     |     2 |
|    7 | G     |     2 |
|    8 | H     |     3 |
+------+-------+-------+

2) 深度优先遍历

with recursive cte as (
  select id, value, 0 as level, CAST(id AS CHAR(200)) AS path  from t1 where parent_id is null
  union all
  select t1.id, t1.value, cte.level+1, CONCAT(cte.path, ",", t1.id)  from cte join t1 on t1.parent_id=cte.id)
select * from cte order by path;
+------+-------+-------+---------+
| id   | value | level | path    |
+------+-------+-------+---------+
|    1 | A     |     0 | 1       |
|    2 | B     |     1 | 1,2     |
|    5 | E     |     2 | 1,2,5   |
|    6 | F     |     2 | 1,2,6   |
|    8 | H     |     3 | 1,2,6,8 |
|    3 | C     |     1 | 1,3     |
|    4 | D     |     1 | 1,4     |
|    7 | G     |     2 | 1,4,7   |
+------+-------+-------+---------+

Oracle

Oracle从9.2才开始支持CTE, 但只支持non-recursive with, 直到Oracle 11.2才完全支持CTE。但oracle 之前就支持connect by 的树形查询,recursive with 语句可以与connect by语句相互转化。 一些相互转化案例可以参考这里.

Oracle recursive with 语句不需要指定recursive关键字,可以自动识别是否recursive.

Oracle 还支持CTE相关的hint,

WITH dept_count AS (
  SELECT /*+ MATERIALIZE */ deptno, COUNT(*) AS dept_count
  FROM   emp
  GROUP BY deptno)
SELECT ...

WITH dept_count AS (
  SELECT /*+ INLINE */ deptno, COUNT(*) AS dept_count
  FROM   emp
  GROUP BY deptno)
SELECT ...

“MATERIALIZE”告诉优化器产生一个全局的临时表保存结果,多次引用CTE时直接访问临时表即可。而”INLINE”则表示每次需要解析查询CTE。

PostgreSQL

PostgreSQL从8.4开始支持CTE,PostgreSQL还扩展了CTE的功能, CTE的query中支持DML语句,例如

create table t1 (c1 int, c2 char(10));
 insert into t1 values(1,'a'),(2,'b');
 select * from t1;
 c1 | c2
----+----
  1 | a
  2 | b


 WITH cte AS (
     UPDATE t1 SET c1= c1 * 2 where c1=1
     RETURNING *
 )
 SELECT * FROM cte; //返回更新的值
 c1 |     c2
----+------------
  2 | a

 truncate table t1;
 insert into t1 values(1,'a'),(2,'b');
 WITH cte AS (
     UPDATE t1 SET c1= c1 * 2 where c1=1
     RETURNING *
 )
 SELECT * FROM t1;//返回原值
 c1 |     c2
----+------------
  1 | a
  2 | b


 truncate table t1;
 insert into t1 values(1,'a'),(2,'b');
 WITH cte AS (
     DELETE FROM t1
     WHERE c1=1
     RETURNING *
 )
 SELECT * FROM cte;//返回删除的行
 c1 |     c2
----+------------
  1 | a


 truncate table t1;
 insert into t1 values(1,'a'),(2,'b');
 WITH cte AS (
     DELETE FROM t1
     WHERE c1=1
     RETURNING *
 )
 SELECT * FROM t1;//返回原值
 c1 |     c2
----+------------
  1 | a
  2 | b
(2 rows)

MariaDB

MariaDB从10.2开始支持CTE。10.2.1 支持non-recursive CTE, 10.2.2开始支持recursive CTE。 目前的GA的版本是10.1.

MySQL

MySQL从8.0开始支持完整的CTE。MySQL8.0还在development 阶段,RC都没有,GA还需时日。

AliSQL

AliSQL基于mariadb10.2, port了no-recursive CTE的实现,此功能近期会上线。

以下从源码主要相关函数简要介绍其实现,

//解析识别with table引用
find_table_def_in_with_clauses

//检查依赖关系,比如不能重复定义with table名字
With_clause::check_dependencies

// 为每个引用clone一份定义
With_element::clone_parsed_spec

//替换with table指定的列名
With_element::rename_columns_of_derived_unit

此实现对于多次引用CTE,CTE会解析多次,因此此版本CTE有简化SQL的作用,但效率上没有效提高。

select count(*) from t1 where c2 !='z';
+----------+
| count(*) |
+----------+
|    65536 |
+----------+
1 row in set (0.25 sec)

//从执行时间来看是进行了3次全表扫描
 with t as (select count(*) from t1 where c2 !='z')
     select * from t union select * from t union select * from t;
+----------+
| count(*) |
+----------+
|    65536 |
+----------+
1 row in set (0.59 sec)

 select count(*) from t1 where c2 !='z'
     union
     select count(*) from t1 where c2 !='z'
     union
    select count(*) from t1 where c2 !='z';
+----------+
| count(*) |
+----------+
|    65536 |
+----------+
1 row in set (0.57 sec)

 explain with t as (select count(*) from t1 where c2 !='z')
    -> select * from t union select * from t union select * from t;
+------+-----------------+--------------+------+---------------+------+---------+------+-------+-------------+
| id   | select_type     | table        | type | possible_keys | key  | key_len | ref  | rows  | Extra       |
+------+-----------------+--------------+------+---------------+------+---------+------+-------+-------------+
|    1 | PRIMARY         | <derived2>   | ALL  | NULL          | NULL | NULL    | NULL | 65536 |             |
|    2 | SUBQUERY        | t1           | ALL  | NULL          | NULL | NULL    | NULL | 65536 | Using where |
|    3 | RECURSIVE UNION | <derived5>   | ALL  | NULL          | NULL | NULL    | NULL | 65536 |             |
|    5 | SUBQUERY        | t1           | ALL  | NULL          | NULL | NULL    | NULL | 65536 | Using where |
|    4 | RECURSIVE UNION | <derived6>   | ALL  | NULL          | NULL | NULL    | NULL | 65536 |             |
|    6 | SUBQUERY        | t1           | ALL  | NULL          | NULL | NULL    | NULL | 65536 | Using where |
| NULL | UNION RESULT    | <union1,3,4> | ALL  | NULL          | NULL | NULL    | NULL |  NULL |             |
+------+-----------------+--------------+------+---------------+------+---------+------+-------+-------------+
7 rows in set (0.00 sec)

 explain  select count(*) from t1 where c2 !='z'
    union
    select count(*) from t1 where c2 !='z'
    union
    select count(*) from t1 where c2 !='z';
+------+--------------+--------------+------+---------------+------+---------+------+-------+-------------+
| id   | select_type  | table        | type | possible_keys | key  | key_len | ref  | rows  | Extra       |
+------+--------------+--------------+------+---------------+------+---------+------+-------+-------------+
|    1 | PRIMARY      | t1           | ALL  | NULL          | NULL | NULL    | NULL | 65536 | Using where |
|    2 | UNION        | t1           | ALL  | NULL          | NULL | NULL    | NULL | 65536 | Using where |
|    3 | UNION        | t1           | ALL  | NULL          | NULL | NULL    | NULL | 65536 | Using where |
| NULL | UNION RESULT | <union1,2,3> | ALL  | NULL          | NULL | NULL    | NULL |  NULL |             |
+------+--------------+--------------+------+---------------+------+---------+------+-------+-------------+
4 rows in set (0.00 sec)

以下是MySQL8.0 只扫描一次的执行计划

mysql> explain select count(*) from t1 where c2 !='z' union select count(*) from t1 where c2 !='z' union select count(*) from t1 where c2 !='z';
+----+--------------+--------------+------------+------+---------------+------+---------+------+-------+----------+-----------------+
| id | select_type  | table        | partitions | type | possible_keys | key  | key_len | ref  | rows  | filtered | Extra           |
+----+--------------+--------------+------------+------+---------------+------+---------+------+-------+----------+-----------------+
|  1 | PRIMARY      | t1           | NULL       | ALL  | NULL          | NULL | NULL    | NULL | 62836 |    90.00 | Using where     |
|  2 | UNION        | t1           | NULL       | ALL  | NULL          | NULL | NULL    | NULL | 62836 |    90.00 | Using where     |
|  3 | UNION        | t1           | NULL       | ALL  | NULL          | NULL | NULL    | NULL | 62836 |    90.00 | Using where     |
| NULL | UNION RESULT | <union1,2,3> | NULL       | ALL  | NULL          | NULL | NULL    | NULL |  NULL |     NULL | Using temporary |
+----+--------------+--------------+------------+------+---------------+------+---------+------+-------+----------+-----------------+
4 rows in set, 1 warning (0.00 sec)

以下是PostgreSQL9.4 只扫描一次的执行计划

postgres=# explain with t as (select count(*) from t1 where c2 !='z')
postgres-# select * from t union select * from t union select * from t;
 HashAggregate  (cost=391366.28..391366.31 rows=3 width=8)
   Group Key: t.count
   CTE t
     ->  Aggregate  (cost=391366.17..391366.18 rows=1 width=0)
           ->  Seq Scan on t1  (cost=0.00..384392.81 rows=2789345 width=0)
                 Filter: ((c2)::text <> 'z'::text)
   ->  Append  (cost=0.00..0.09 rows=3 width=8)
         ->  CTE Scan on t  (cost=0.00..0.02 rows=1 width=8)
         ->  CTE Scan on t t_1  (cost=0.00..0.02 rows=1 width=8)
         ->  CTE Scan on t t_2  (cost=0.00..0.02 rows=1 width=8)

AliSQL还有待改进。