一次关于INDEX SKIP SCAN的SQL优化-CFANZ编程社区

客户发来一个比较长的SQL，说是跑了几小时还没出结果；这是制造业的一个统计型SQL，发来SQL文本一看SQL比较长要翻好几屏，一时半会也梳理不清；按照DBA的思路先尝试查看执行计划，看是否能通过索引或执行计划的改变在不修改SQL(也就不需要了解SQL业务逻辑)情况下优化；从执行计划来看cost 13676,具体的执行步骤在INDEX SKIP SCAN部分消耗比较大；

关于索引跳跃扫描 index skip scan，它是有一定的适用场景的，如需要满足数据库版本9i及以后CBO模式且有统计信息，适用所有类型复合B树索引(包括唯一和非唯一索引)。当在WHERE条件中没有对目标索引的前导列指定查询条件，但又对非前导列指定了查询条件时依然可以使用该索引(实际是ORACLE对这种SQL进行了等价改写，对索引的前导列所有distinct值做了遍历，以此值做查询再合并结果)。

在ORACLE中索引跳跃扫描适用于:目标索引前导列的distinct值数量少，后续非前导列的可选择性好的情况。这个环境中WHERE条件中的WHERE MSGFID =列的选择率不太好(总共150万行数据，distinct值约170个)，当前存在的组合索引是前导列选择率较好的；与用户沟通了解此表的DML操作也较多，也不适用创建位图索引；让用户尝试创建一个单独包含MSGFID 列的索引，效果还是比较好的，修改后的执行计划使用index range scan，70秒可以执行完成，能满足业务需求；

如下实验测试一下索引跳跃扫描 index skip scan的效率：

测试环境linux+oracle 11.2.0.4：SQL> create table test4 as select * from dba_objects; Table created. SQL> exec dbms_stats.gather_table_stats('BYS','TEST4'); PL/SQL procedure successfully completed.

使用dba_objects的三个列，共 87250数据，owner 对应用户约10多个用户，object_type对应类型约10多个；OBJECT_ID是惟一的。

1.首选测试不使用索引时的执行计划，COST=348：

2.测试test4(owner,OBJECT_TYPE)做组合索引时INDEX SKIP SCAN执行计划，COST=143：

3.测试test4(OBJECT_ID,OBJECT_TYPE)做组合索引时INDEX SKIP SCAN执行计划，COST=331：

4.测试test4(OBJECT_TYPE)做单列索引时INDEX SKIP SCAN执行计划，COST=87：

测试结果比较明显；ORACLE中索引跳跃扫描适用于目标索引前导列的distinct值数量少，后续非前导列的可选择性好的情况。 ================== 1.首选测试不使用索引时的执行计划，COST=348：

SQL> select count(object_name) from test4 where object_TYPE='TABLE';

COUNT(OBJECT_NAME)
------------------
              3181
Execution Plan
----------------------------------------------------------
Plan hash value: 2292881717

----------------------------------------------------------------------------
| Id  | Operation          | Name  | Rows  | Bytes | Cost (%CPU)| Time     |
----------------------------------------------------------------------------
|   0 | SELECT STATEMENT   |       |     1 |    34 |   348   (1)| 00:00:05 |
|   1 |  SORT AGGREGATE    |       |     1 |    34 |            |          |
|*  2 |   TABLE ACCESS FULL| TEST4 |  1983 | 67422 |   348   (1)| 00:00:05 |
----------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
   2 - filter("OBJECT_TYPE"='TABLE')

2.测试test4(owner,OBJECT_TYPE)做组合索引时INDEX SKIP SCAN执行计划，，COST=143：

SQL>  create index idx03_test4 on test4(owner,OBJECT_TYPE);
Index created.

SQL> select count(object_name) from test4 where object_TYPE='TABLE';

COUNT(OBJECT_NAME)
------------------
              3181
Execution Plan
----------------------------------------------------------
Plan hash value: 2478250731
--------------------------------------------------------------------------------------------
| Id  | Operation                    | Name        | Rows  | Bytes | Cost (%CPU)| Time     |
--------------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT             |             |     1 |    34 |   143   (0)| 00:00:02 |
|   1 |  SORT AGGREGATE              |             |     1 |    34 |            |          |
|   2 |   TABLE ACCESS BY INDEX ROWID| TEST4       |  1983 | 67422 |   143   (0)| 00:00:02 |
|*  3 |    INDEX SKIP SCAN           | IDX03_TEST4 |  1983 |       |    60   (0)| 00:00:01 |
--------------------------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
   3 - access("OBJECT_TYPE"='TABLE')
       filter("OBJECT_TYPE"='TABLE')

3.测试test4(OBJECT_ID,OBJECT_TYPE)做组合索引时INDEX SKIP SCAN执行计划，COST=331：

SQL>  create index idx01_test4 on test4(OBJECT_ID,OBJECT_TYPE);

Index created.
SQL> select count(object_name) from test4 where object_TYPE='TABLE';

COUNT(OBJECT_NAME)
------------------
              3181
Execution Plan
----------------------------------------------------------
Plan hash value: 1133214586
--------------------------------------------------------------------------------------------
| Id  | Operation                    | Name        | Rows  | Bytes | Cost (%CPU)| Time     |
--------------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT             |             |     1 |    34 |   331   (0)| 00:00:04 |
|   1 |  SORT AGGREGATE              |             |     1 |    34 |            |          |
|   2 |   TABLE ACCESS BY INDEX ROWID| TEST4       |  1983 | 67422 |   331   (0)| 00:00:04 |
|*  3 |    INDEX SKIP SCAN           | IDX01_TEST4 |  1983 |       |   300   (0)| 00:00:04 |
-------------------------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
   3 - access("OBJECT_TYPE"='TABLE')
       filter("OBJECT_TYPE"='TABLE')

4.测试test4(OBJECT_TYPE)做单列索引时INDEX SKIP SCAN执行计划，COST=87：

SQL>  create index idx02_test4 on test4(OBJECT_TYPE);
Index created.
SQL> select count(object_name) from test4 where object_TYPE='TABLE';
COUNT(OBJECT_NAME)
------------------
              3181
Execution Plan
----------------------------------------------------------
Plan hash value: 1992435611
--------------------------------------------------------------------------------------------
| Id  | Operation                    | Name        | Rows  | Bytes | Cost (%CPU)| Time     |
--------------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT             |             |     1 |    34 |    87   (0)| 00:00:02 |
|   1 |  SORT AGGREGATE              |             |     1 |    34 |            |          |
|   2 |   TABLE ACCESS BY INDEX ROWID| TEST4       |  1983 | 67422 |    87   (0)| 00:00:02 |
|*  3 |    INDEX RANGE SCAN          | IDX02_TEST4 |  1983 |       |     6   (0)| 00:00:01 |
--------------------------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
   3 - access("OBJECT_TYPE"='TABLE')