SQL中的半连接在MySQL和Oracle还是存在一些差距,从测试的情况来看,Oracle的处理要更加全面。
首先我们来看看在MySQL中怎么测试,对于MySQL方面的测试也参考了不少海翔兄的博客文章,自己也完整的按照他的测试思路练习了一遍。
首先创建下面的表:
create table users(
userid int(11) unsigned not null,
user_name varchar(64) default null,
primary key(userid)
)engine=innodb default charset=UTF8;
如果要插入数据,可以使用存储过程的方式。比如先插入20000条定制数据。
delimiter $$
drop procedure if exists proc_auto_insertdata$$
create procedure proc_auto_insertdata()
begin
declare
init_data integer default 1;
while init_data<=20000 do
insert into users values(init_data,concat('user' ,init_data));
set init_data=init_data+1;
end while;
end$$
delimiter ;
call proc_auto_insertdata();
初始化的过程会很快,最后一步即插入数据花费了近6秒的时间。
[test]>source insert_proc.sql
Query OK, 0 rows affected (0.12 sec)
Query OK, 0 rows affected (0.00 sec)
Query OK, 0 rows affected (0.00 sec)
Query OK, 1 row affected (5.63 sec)
然后我们使用如下的半连接查询数据,实际上执行了6秒左右。
select u.userid,u.user_name from users u where u.user_name in (select t.user_name from users t where t.userid<2000);
1999 rows in set (6.36 sec)
为了简化测试条件和查询结果,我们使用count的方式来完成对比测试。
[test]>select count(u.userid) from users u where u.user_name in (select t.user_name from users t where t.userid<2000);
+-----------------+
| count(u.userid) |
+-----------------+
| 1999 |
+-----------------+
1 row in set (6.38 sec)
然后使用如下的方式来查看,当然看起来这种结构似乎有些多余,因为userid<-1的数据是不存在的。
select count(u.userid) from users u
where (u.user_name in (select t.user_name from users t where t.userid<2000) or u.user_name in (select t.user_name from users t where userid<-1) );
+-----------------+
| count(u.userid) |
+-----------------+
| 1999 |
+-----------------+
1 row in set (0.06 sec)
但是效果却好很多。
当然两种方式的执行计划差别很大。
第一种效率较差的执行计划如下:
[test]>explain select count(u.userid) from users u where u.user_name in (select t.user_name from users t where t.userid<2000);
+----+--------------+-------------+-------+---------------+---------+---------+------+-------+----------------------------------------------------+
| id | select_type | table | type | possible_keys | key | key_len | ref | rows | Extra |
+----+--------------+-------------+-------+---------------+---------+---------+------+-------+----------------------------------------------------+
| 1 | SIMPLE | | ALL | NULL | NULL | NULL | NULL | NULL | NULL |
| 1 | SIMPLE | u | ALL | NULL | NULL | NULL | NULL | 19762 | Using where; Using join buffer (Block Nested Loop) |
| 2 | MATERIALIZED | t | range | PRIMARY | PRIMARY | 4 | NULL | 1998 | Using where |
+----+--------------+-------------+-------+---------------+---------+---------+------+-------+----------------------------------------------------+
3 rows in set (0.02 sec)
第二个执行效率较高的执行计划如下:
[test]>explain select count(u.userid) from users u where (u.user_name in (select t.user_name from users t where t.userid<2000) or u.user_name in (select t.user_name from users t where userid<-1) );
+----+-------------+-------+-------+---------------+---------+---------+------+-------+-----------------------------------------------------+
| id | select_type | table | type | possible_keys | key | key_len | ref | rows | Extra |
+----+-------------+-------+-------+---------------+---------+---------+------+-------+-----------------------------------------------------+
| 1 | PRIMARY | u | ALL | NULL | NULL | NULL | NULL | 19762 | Using where |
| 3 | SUBQUERY | NULL | NULL | NULL | NULL | NULL | NULL | NULL | Impossible WHERE noticed after reading const tables |
| 2 | SUBQUERY | t | range | PRIMARY | PRIMARY | 4 | NULL | 1998 | Using where |
+----+-------------+-------+-------+---------------+---------+---------+------+-------+-----------------------------------------------------+
3 rows in set (0.00 sec)
我们在这个测试中先不解释更多的原理,只是对比说明。
如果想得到更多的执行效率对比情况,可以使用show status 的方式。
首先flush status
[test]>flush status;
Query OK, 0 rows affected (0.02 sec)
然后执行语句如下:
[test]>select count(u.userid) from users u where u.user_name in (select t.user_name from users t where t.userid<2000);
+-----------------+
| count(u.userid) |
+-----------------+
| 1999 |
+-----------------+
1 row in set (6.22 sec)
查看状态信息,关键词是Handler_read.
[test]>show status like 'Handler_read%';
+-----------------------+-------+
| Variable_name | Value |
+-----------------------+-------+
| Handler_read_first | 2 |
| Handler_read_key | 2 |
| Handler_read_last | 0 |
| Handler_read_next | 1999 |
| Handler_read_prev | 0 |
| Handler_read_rnd | 0 |
| Handler_read_rnd_next | 22001 |
+-----------------------+-------+
7 rows in set (0.04 sec
Handler_read_key这个参数的解释是根据键读一行的请求数。如果较高,说明查询和表的索引正确。
Handler_read_next这个参数的解释是按照键顺序读下一行的请求数。如果用范围约束或如果执行索引扫描来查询索引列,该值增加。
Handler_read_rnd_next这个参数的解释是在数据文件中读下一行的请求数。如果正进行大量的表扫描,该值较高。通常说明表索引不正确或写入的查询没有利用索引。
这是一个count的操作,所以Handler_read_rnd_next的指标较高,这是一个范围查询,所以Handler_read_next 的值也是一个范围值。
然后运行另外一个子查询,可以看到show status的结果如下:
[test]>show status like 'Handler_read%';
+-----------------------+-------+
| Variable_name | Value |
+-----------------------+-------+
| Handler_read_first | 2 |
| Handler_read_key | 20002 |
| Handler_read_last | 0 |
| Handler_read_next | 1999 |
| Handler_read_prev | 0 |
| Handler_read_rnd | 0 |
| Handler_read_rnd_next | 20001 |
+-----------------------+-------+
7 rows in set (0.00 sec)
可以和明显看到Handler_read_key这个值很高,根据参数的解释,说明查询和表的索引使用正确。也就意味着这种方式想必于第一种方案要好很多。
而对于此,MySQL其实也有一些方式方法可以得到更细节的信息。
一种就是explain extended的方式。
[test]>explain extended select count(u.userid) from users u where u.user_name in (select t.user_name from users t where t.userid<2000);
。。。。
3 rows in set, 1 warning (0.00 sec)
然后show warnings就会看到详细的信息。
[test]>show warnings;
| Note | 1003 | /* select#1 */ select count(`test`.`u`.`userid`) AS `count(u.userid)` from `test`.`users` `u` semi join (`test`.`users` `t`) where ((`test`.`u`.`user_name` = ``.`user_name`) and (`test`.`t`.`userid` < 2000)) |
1 row in set (0.00 sec)
第二个语句的情况如下:
[test]>explain extended select count(u.userid) from users u where (u.user_name in (select t.user_name from users t where t.userid<2000) or u.user_name in (select t.user_name from users t where userid<-1) );
3 rows in set, 1 warning (0.00 sec)
[test]>show warnings;
| Note | 1003 | /* select#1 */ select count(`test`.`u`.`userid`) AS `count(u.userid)` from `test`.`users` `u` where ((`test`.`u`.`user_name`,`test`.`u`.`user_name` in ( (/* select#2 */ select `test`.`t`.`user_name` from `test`.`users` `t` where (`test`.`t`.`userid` < 2000) ), (`test`.`u`.`user_name` in on where ((`test`.`u`.`user_name` = `materialized-subquery`.`user_name`))))) or (`test`.`u`.`user_name`,`test`.`u`.`user_name` in ( (/* select#3 */ select `test`.`t`.`user_name` from `test`.`users` `t` where 0 ), (`test`.`u`.`user_name` in on where ((`test`.`u`.`user_name` = `materialized-subquery`.`user_name`)))))) |
1 row in set (0.00 sec)
还有一种方式就是使用 optimizer_trace,在5.6可用
set optimizer_trace="enabled=on";
运行语句后,然后通过下面的查询得到trace信息。
select *from information_schema.optimizer_trace\G
当然可以看出半连接的表现其实还不够好,能不能选择性的关闭呢,有一个参数可以控制,即是optimizer_switch,其实我们也可以看看这个参数的情况。
| optimizer_switch | index_merge=on,index_merge_union=on,index_merge_sort_union=on,index_merge_intersection=on,engine_condition_pushdown=on,index_condition_pushdown=on,mrr=on,mrr_cost_based=on,block_nested_loop=on,batched_key_access=off,materialization=on,semijoin=on,loosescan=on,firstmatch=on,subquery_materialization_cost_based=on,use_index_extensions=on |
关闭半连接的设置
>set optimizer_switch="semijoin=off";
Query OK, 0 rows affected (0.00 sec)
再次运行原本执行时间近6秒的SQL,执行时间大大降低。
[test]> select count(u.userid) from users u where u.user_name in (select t.user_name from users t where t.userid<2000);
+-----------------+
| count(u.userid) |
+-----------------+
| 1999 |
+-----------------+
1 row in set (0.05 sec)
执行第二个语句,情况如下:
[test]>select count(u.userid) from users u where (u.user_name in (select t.user_name from users t where t.userid<2000) or u.user_name in (select t.user_name from users t where userid<-1) );
+-----------------+
| count(u.userid) |
+-----------------+
| 1999 |
+-----------------+
1 row in set (0.07 sec)
参考内容如下:
http://dbaplus.cn/news-11-133-1.html
http://blog.chinaunix.net/uid-16909016-id-214888.html
而在Oracle中表现如何呢。
创建测试表
create table users(
userid number not null,
user_name varchar2(64) default null,
primary key(userid)
);
初始化数据,其实一句SQL就可以搞定。递归查询可以换种方式来用,效果杠杠的。
insert into users select level,'user'||level from dual connect by level<=20000;
收集一下统计信息
exec dbms_stats.gather_table_stats(ownname=>'CYDBA',tabname=>'USERS',cascade=>true);
然后执行和MySQL中同样的语句。
我们使用trace的方式来查看,我们仅列出trace的情况。
SQL> set autot trace exp stat
SQL> select u.userid,u.user_name from users u where u.user_name in (select t.user_name from users t where t.userid<2000);
1999 rows selected.
Execution Plan
----------------------------------------------------------
Plan hash value: 771105466
---------------------------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
---------------------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 2003 | 52078 | 21 (5)| 00:00:01 |
|* 1 | HASH JOIN RIGHT SEMI | | 2003 | 52078 | 21 (5)| 00:00:01 |
| 2 | TABLE ACCESS BY INDEX ROWID| USERS | 1999 | 25987 | 3 (0)| 00:00:01 |
|* 3 | INDEX RANGE SCAN | SYS_C0042448 | 1999 | | 2 (0)| 00:00:01 |
| 4 | TABLE ACCESS FULL | USERS | 20000 | 253K| 17 (0)| 00:00:01 |
---------------------------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
1 - access("U"."USER_NAME"="T"."USER_NAME")
3 - access("T"."USERID"<2000)
Statistics
----------------------------------------------------------
0 recursive calls
0 db block gets
205 consistent gets
0 physical reads
0 redo size
52196 bytes sent via SQL*Net to client
1983 bytes received via SQL*Net from client
135 SQL*Net roundtrips to/from client
0 sorts (memory)
0 sorts (disk)
1999 rows processed
SQL> select u.userid,u.user_name from users u where (u.user_name in (select t.user_name from users t where t.userid<2000) or u.user_name in (select t.user_name from users t where userid<-1) );
1999 rows selected.
Execution Plan
----------------------------------------------------------
Plan hash value: 1012235795
------------------------------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
------------------------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 2004 | 94188 | 22 (5)| 00:00:01 |
|* 1 | HASH JOIN | | 2004 | 94188 | 22 (5)| 00:00:01 |
| 2 | VIEW | VW_NSO_1 | 2000 | 68000 | 4 (0)| 00:00:01 |
| 3 | HASH UNIQUE | | 2000 | 26000 | 4 (25)| 00:00:01 |
| 4 | UNION-ALL | | | | | |
| 5 | TABLE ACCESS BY INDEX ROWID| USERS | 1 | 13 | 1 (0)| 00:00:01 |
|* 6 | INDEX RANGE SCAN | SYS_C0042448 | 1 | | 1 (0)| 00:00:01 |
| 7 | TABLE ACCESS BY INDEX ROWID| USERS | 1999 | 25987 | 3 (0)| 00:00:01 |
|* 8 | INDEX RANGE SCAN | SYS_C0042448 | 1999 | | 2 (0)| 00:00:01 |
| 9 | TABLE ACCESS FULL | USERS | 20000 | 253K| 17 (0)| 00:00:01 |
------------------------------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
1 - access("U"."USER_NAME"="USER_NAME")
6 - access("USERID"<(-1))
8 - access("T"."USERID"<2000)
Statistics
----------------------------------------------------------
0 recursive calls
0 db block gets
207 consistent gets
0 physical reads
0 redo size
52196 bytes sent via SQL*Net to client
1983 bytes received via SQL*Net from client
135 SQL*Net roundtrips to/from client
0 sorts (memory)
0 sorts (disk)
1999 rows processed
从Oracle的表现来看,支持的力度要全面很多。当然半连接的玩法还有很多,比如exists,这些限于篇幅暂没有展开。而且对于对比测试中的更多知识点分析,我们后期也会逐步补充。