其实二者没有什么可比性,但是对于不包含聚集函数的GROUP BY操作来说,和DISTINCT操作是等价的。不过虽然二者的结果是一样的,但是二者的执行计划并不相同。
在Oracle9i中:
SQL> SELECT * FROM V$VERSION;
BANNER
----------------------------------------------------------------
Oracle9i Enterprise Edition Release 9.2.0.4.0 - Production PL/SQL Release 9.2.0.4.0 - Production
CORE 9.2.0.3.0 Production
TNS for Linux: Version 9.2.0.4.0 - Production
NLSRTL Version 9.2.0.4.0 - Production
SQL> CREATE TABLE T AS SELECT ROWNUM ID, A.* FROM DBA_OBJECTS A;
表已创建。
SQL> CREATE INDEX IND_T_CREATED ON T (CREATED);
索引已创建。
SQL> ALTER TABLE T MODIFY CREATED NOT NULL;
表已更改。
SQL> ALTER SESSION SET NLS_DATE_FORMAT = 'YYYY-MM-DD HH24:MI:SS';
会话已更改。
SQL> EXEC DBMS_STATS.GATHER_TABLE_STATS(USER, 'T')
PL/SQL 过程已成功完成。
SQL> SET AUTOT ON EXP
SQL> SELECT COUNT(*) FROM (SELECT DISTINCT CREATED FROM T);
COUNT(*)
----------
4794
执行计划
----------------------------------------------------------
0 SELECT STATEMENT Optimizer=CHOOSE (Cost=65 Card=1)
1 0 SORT (AGGREGATE)
2 1 VIEW (Cost=65 Card=4794)
3 2 SORT (UNIQUE) (Cost=65 Card=4794 Bytes=38352)
4 3 INDEX (FAST FULL SCAN) OF 'IND_T_CREATED' (NON-UNIQUE) (Cost=4 Card=41802 Bytes=334416)
SQL> SELECT COUNT(*) FROM (SELECT CREATED FROM T GROUP BY CREATED);
COUNT(*)
----------
4794
执行计划
----------------------------------------------------------
0 SELECT STATEMENT Optimizer=CHOOSE (Cost=65 Card=1 Bytes=2)
1 0 SORT (AGGREGATE)
2 1 VIEW (Cost=65 Card=4794 Bytes=9588)
3 2 SORT (GROUP BY) (Cost=65 Card=4794 Bytes=38352)
4 3 INDEX (FAST FULL SCAN) OF 'IND_T_CREATED' (NON-UNIQUE) (Cost=4 Card=41802 Bytes=334416)
从执行计划上看,DISTINCT的操作是SORT (UNIQUE),而GROUP BY是SORT (GROUP BY)。DISTINCT操作只需要找出所有不同的值就可以了。而GROUP BY操作还要为其他聚集函数进行准备工作。从这一点上将,GROUP BY操作做的工作应该比DISTINCT所做的工作要多一些。
除了这一点,基本上看不到DISTINCT和GROUP BY(没有聚集函数的情况)有什么区别,而且从执行效率上也看不到明显的差异。
不过从10g开始,二者的差异开始体现出来了。
SQL> CONN YANGTK/YANGTK@YTK已连接。
SQL> SET AUTOT OFF
SQL> SET TIMING OFF
SQL> CREATE TABLE T AS SELECT ROWNUM ID, A.* FROM DBA_OBJECTS A;
表已创建。
SQL> CREATE INDEX IND_T_CREATED ON T (CREATED);
索引已创建。
SQL> ALTER TABLE T MODIFY CREATED NOT NULL;
表已更改。
SQL> ALTER SESSION SET NLS_DATE_FORMAT = 'YYYY-MM-DD HH24:MI:SS';
会话已更改。
SQL> EXEC DBMS_STATS.GATHER_TABLE_STATS(USER, 'T')
PL/SQL 过程已成功完成。
SQL> SET AUTOT ON
SQL> SET TIMING ON
建立好测试环境后,看一看标准分页函数中,两个操作的差异:
SQL> SELECT *
2 FROM
3 (
4 SELECT ROWNUM RN, A.*
5 FROM
6 (
7 SELECT CREATED
8 FROM T
9 GROUP BY CREATED
10 ) A
11 WHERE ROWNUM < 20
12 )
13 WHERE RN >= 10;
RN CREATED
---------- -------------------
10 2005-12-19 17:07:57
11 2005-12-19 17:07:58
12 2005-12-19 17:08:24
13 2005-12-19 17:08:25
14 2005-12-19 17:08:26
15 2005-12-19 17:08:27
16 2005-12-19 17:08:28
17 2005-12-19 17:08:29
18 2005-12-19 17:08:33
19 2005-12-19 17:08:35
已选择10行。
已用时间: 00: 00: 00.06
执行计划
----------------------------------------------------------
Plan hash value: 3639065582
-------------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)|
-------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 19 | 418 | 1 (0)|
|* 1 | VIEW | | 19 | 418 | 1 (0)|
|* 2 | COUNT STOPKEY | | | | |
| 3 | VIEW | | 969 | 8721 | 1 (0)|
|* 4 | SORT GROUP BY STOPKEY| | 969 | 7752 | 1 (0)|
| 5 | INDEX FULL SCAN | IND_T_CREATED | 969 | 7752 | 1 (0)|
-------------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
1 - filter("RN">=10)
2 - filter(ROWNUM<20)
4 - filter(ROWNUM<20)
统计信息
----------------------------------------------------------
1 recursive calls
0 db block gets
67 consistent gets
0 physical reads
0 redo size
642 bytes sent via SQL*Net to client
385 bytes received via SQL*Net from client
2 SQL*Net roundtrips to/from client
1 sorts (memory)
0 sorts (disk)
10 rows processed
SQL> SELECT *
2 FROM
3 (
4 SELECT ROWNUM RN, A.*
5 FROM
6 (
7 SELECT DISTINCT CREATED
8 FROM T
9 ) A
10 WHERE ROWNUM < 20
11 )
12 WHERE RN >= 10;
RN CREATED
---------- -------------------
10 2005-12-19 17:07:57
11 2005-12-19 17:07:58
12 2005-12-19 17:08:24
13 2005-12-19 17:08:25
14 2005-12-19 17:08:26
15 2005-12-19 17:08:27
16 2005-12-19 17:08:28
17 2005-12-19 17:08:29
18 2005-12-19 17:08:33
19 2005-12-19 17:08:35
已选择10行。
已用时间: 00: 00: 00.03
执行计划
----------------------------------------------------------
Plan hash value: 1650124153
-------------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)|
-------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 19 | 418 | 14 (36)|
|* 1 | VIEW | | 19 | 418 | 14 (36)|
|* 2 | COUNT STOPKEY | | | | |
| 3 | VIEW | | 987 | 8883 | 14 (36)|
|* 4 | SORT GROUP BY STOPKEY| | 987 | 7896 | 14 (36)|
| 5 | INDEX FAST FULL SCAN| IND_T_CREATED | 50333 | 393K| 10 (10)|
-------------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
1 - filter("RN">=10)
2 - filter(ROWNUM<20)
4 - filter(ROWNUM<20)
统计信息
----------------------------------------------------------
1 recursive calls
0 db block gets
73 consistent gets
0 physical reads
0 redo size
642 bytes sent via SQL*Net to client
385 bytes received via SQL*Net from client
2 SQL*Net roundtrips to/from client
1 sorts (memory)
0 sorts (disk)
10 rows processed
出乎意料的是,GROUP BY操作的COST更低,而且逻辑读也小,这似乎与二者的工作量成反比。仔细观察执行计划发现,问题的根源来自于GROUP BY使用INDEX FULL SCAN,而DISTINCT使用了INDEX FAST FULL SCAN。也许有人会感到奇怪,索引的快速全扫描不是要比索引全扫描效率更高吗?对于读取所有数据的情况下,确实是索引快速全扫效率更高。但是由于这里采用了分页,只取前20条数据,而且Oracle的10g增加了GROUP BY STOPKEY这种新的执行路径,因此在这里GROUP BY操作的效率更高。
观察执行计划中的处理行数可以发现,索引全扫描由于是按照索引的顺序扫描,因此利用了STOPKEY,仅仅处理了969条记录就停了下来。而对于DISTINCT操作的快速索引全速而言,显然没有使用STOPKEY,读取了所有的50333条记录。这就是GROUP BY和DISTINCT的性能差异原因。