--创建2个结构完全相同，数据量几乎相同的表，只不过是分区表，一个是普通的heap表，
然后简单的做个访问时的性能对比，从而更好的理解如何合理的使用分区表
SQL> create table t(object_id number,object_name varchar2(30))
2 partition by range(object_id)
3 (
4 partition p1 values less than(2000) tablespace users,
5 partition p2 values less than(4000) tablespace users,
6 partition p3 values less than(6000) tablespace users,
7 partition p4 values less than(8000) tablespace users,
8 partition p5 values less than(maxvalue) tablespace users
9 );
表已创建。
SQL> insert into t select object_id,object_name from dba_objects;
已创建9848行。
SQL> insert into t select * from t;
已创建9848行。
SQL> insert into t select * from t;
已创建19696行。
SQL> insert into t select * from t;
已创建39392行。
SQL> insert into t select * from t;
已创建78784行。
SQL> insert into t select * from t;
已创建157568行。
SQL> insert into t select * from t;
已创建315136行。
SQL> insert into t select * from t;
已创建630272行。
SQL> commit;
提交完成。
SQL> create table t1 tablespace users as select object_id,object_name from dba_o
bjects;
表已创建。
SQL> insert into t1 select *from t1;
已创建9849行。
SQL> insert into t1 select *from t1;
已创建19698行。
SQL> insert into t1 select *from t1;
已创建39396行。
SQL> insert into t1 select *from t1;
已创建78792行。
SQL> insert into t1 select *from t1;
已创建157584行。
SQL> insert into t1 select *from t1;
已创建315168行。
SQL> insert into t1 select *from t1;
已创建630336行。
SQL> commit;
提交完成。
SQL>
--=====================================
1.先来看不垮分区访问时的性能对比，这里我们主要关注逻辑度(consistent gets),
由于第一次访问sql存在解析，因此我们对比时都看第二次访问时的情况，
下面的试验对比都是这样...
注意：目前2个表上都没有index
SQL> select * from t1 where object_id<2000;
已选择249984行。
执行计划
----------------------------------------------------------
Plan hash value: 3617692013
--------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
--------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 270K| 20M| 1069 (4)| 00:00:13 |
|* 1 | TABLE ACCESS FULL| T1 | 270K| 20M| 1069 (4)| 00:00:13 |
--------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
1 - filter("OBJECT_ID"<2000)
Note
-----
- dynamic sampling used for this statement
统计信息
----------------------------------------------------------
222 recursive calls
0 db block gets
21446 consistent gets
4719 physical reads
0 redo size
7439711 bytes sent via SQL*Net to client
183700 bytes received via SQL*Net from client
16667 SQL*Net roundtrips to/from client
2 sorts (memory)
0 sorts (disk)
249984 rows processed
SQL> select * from t1 where object_id<2000;
已选择249984行。
执行计划
----------------------------------------------------------
Plan hash value: 3617692013
--------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
--------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 270K| 20M| 1069 (4)| 00:00:13 |
|* 1 | TABLE ACCESS FULL| T1 | 270K| 20M| 1069 (4)| 00:00:13 |
--------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
1 - filter("OBJECT_ID"<2000)
Note
-----
- dynamic sampling used for this statement
统计信息
----------------------------------------------------------
0 recursive calls
0 db block gets
21341 consistent gets
0 physical reads
0 redo size
7439711 bytes sent via SQL*Net to client
183700 bytes received via SQL*Net from client
16667 SQL*Net roundtrips to/from client
0 sorts (memory)
0 sorts (disk)
249984 rows processed
SQL> select * from t where object_id<2000;
已选择249984行。
执行计划
----------------------------------------------------------
Plan hash value: 2931986080
--------------------------------------------------------------------------------
---------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
Pstart| Pstop |
--------------------------------------------------------------------------------
---------------
| 0 | SELECT STATEMENT | | 241K| 7063K| 238 (19)| 00:00:03 |
| |
| 1 | PARTITION RANGE SINGLE| | 241K| 7063K| 238 (19)| 00:00:03 |
1 | 1 |
| 2 | TABLE ACCESS FULL | T | 241K| 7063K| 238 (19)| 00:00:03 |
1 | 1 |
--------------------------------------------------------------------------------
---------------
Note
-----
- dynamic sampling used for this statement
统计信息
----------------------------------------------------------
1305 recursive calls
0 db block gets
17770 consistent gets
890 physical reads
0 redo size
7439711 bytes sent via SQL*Net to client
183700 bytes received via SQL*Net from client
16667 SQL*Net roundtrips to/from client
12 sorts (memory)
0 sorts (disk)
249984 rows processed
SQL> select * from t where object_id<2000;
已选择249984行。
执行计划
----------------------------------------------------------
Plan hash value: 2931986080
--------------------------------------------------------------------------------
---------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
Pstart| Pstop |
--------------------------------------------------------------------------------
---------------
| 0 | SELECT STATEMENT | | 241K| 7063K| 238 (19)| 00:00:03 |
| |
| 1 | PARTITION RANGE SINGLE| | 241K| 7063K| 238 (19)| 00:00:03 |
1 | 1 |
| 2 | TABLE ACCESS FULL | T | 241K| 7063K| 238 (19)| 00:00:03 |
1 | 1 |
--------------------------------------------------------------------------------
---------------
Note
-----
- dynamic sampling used for this statement
统计信息
----------------------------------------------------------
0 recursive calls
0 db block gets
17498 consistent gets
0 physical reads
0 redo size
7439711 bytes sent via SQL*Net to client
183700 bytes received via SQL*Net from client
16667 SQL*Net roundtrips to/from client
0 sorts (memory)
0 sorts (disk)
249984 rows processed
SQL>
--========================
效果比较明显，很显然是通过访问分区表的性能更好一些，逻辑读21341大于17498，并且都没有发生物理读，
同时我们从执行计划也能看到访问t1时访问的是单个分区(PARTITION RANGE SINGLE),很明显如果我们能把数据
控制在一个分区之内的话那么分区表的性能要好于普通的heap表
--=============================
2.跨分区访问,依然是分区表的性能好，因为数据集中在了2个分区中，还是比访问
t1要少访问数据...
SQL> select * from t1 where object_id<4000;
已选择504448行。
执行计划
----------------------------------------------------------
Plan hash value: 3617692013
--------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
--------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 547K| 41M| 1070 (4)| 00:00:13 |
|* 1 | TABLE ACCESS FULL| T1 | 547K| 41M| 1070 (4)| 00:00:13 |
--------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
1 - filter("OBJECT_ID"<4000)
Note
-----
- dynamic sampling used for this statement
统计信息
----------------------------------------------------------
5 recursive calls
0 db block gets
38318 consistent gets
0 physical reads
0 redo size
15092711 bytes sent via SQL*Net to client
370304 bytes received via SQL*Net from client
33631 SQL*Net roundtrips to/from client
0 sorts (memory)
0 sorts (disk)
504448 rows processed
SQL> set timing on
SQL> select * from t1 where object_id<4000;
已选择504448行。
已用时间: 00: 00: 09.79
执行计划
----------------------------------------------------------
Plan hash value: 3617692013
--------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
--------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 547K| 41M| 1070 (4)| 00:00:13 |
|* 1 | TABLE ACCESS FULL| T1 | 547K| 41M| 1070 (4)| 00:00:13 |
--------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
1 - filter("OBJECT_ID"<4000)
Note
-----
- dynamic sampling used for this statement
统计信息
----------------------------------------------------------
0 recursive calls
0 db block gets
38236 consistent gets
0 physical reads
0 redo size
15092711 bytes sent via SQL*Net to client
370304 bytes received via SQL*Net from client
33631 SQL*Net roundtrips to/from client
0 sorts (memory)
0 sorts (disk)
504448 rows processed
SQL> select * from t where object_id<4000;
已选择504448行。
已用时间: 00: 00: 09.85
执行计划
----------------------------------------------------------
Plan hash value: 1571388083
--------------------------------------------------------------------------------
-----------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time
| Pstart| Pstop |
--------------------------------------------------------------------------------
-----------------
| 0 | SELECT STATEMENT | | 459K| 13M| 837 (53)| 00:00:11
| | |
| 1 | PARTITION RANGE ITERATOR| | 459K| 13M| 837 (53)| 00:00:11
| 1 | 2 |
| 2 | TABLE ACCESS FULL | T | 459K| 13M| 837 (53)| 00:00:11
| 1 | 2 |
--------------------------------------------------------------------------------
-----------------
Note
-----
- dynamic sampling used for this statement
统计信息
----------------------------------------------------------
69 recursive calls
1 db block gets
36078 consistent gets
946 physical reads
48432 redo size
15092711 bytes sent via SQL*Net to client
370304 bytes received via SQL*Net from client
33631 SQL*Net roundtrips to/from client
0 sorts (memory)
0 sorts (disk)
504448 rows processed
SQL> select * from t where object_id<4000;
已选择504448行。
已用时间: 00: 00: 10.12
执行计划
----------------------------------------------------------
Plan hash value: 1571388083
--------------------------------------------------------------------------------
-----------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time
| Pstart| Pstop |
--------------------------------------------------------------------------------
-----------------
| 0 | SELECT STATEMENT | | 459K| 13M| 837 (53)| 00:00:11
| | |
| 1 | PARTITION RANGE ITERATOR| | 459K| 13M| 837 (53)| 00:00:11
| 1 | 2 |
| 2 | TABLE ACCESS FULL | T | 459K| 13M| 837 (53)| 00:00:11
| 1 | 2 |
--------------------------------------------------------------------------------
-----------------
Note
-----
- dynamic sampling used for this statement
统计信息
----------------------------------------------------------
0 recursive calls
0 db block gets
35328 consistent gets
0 physical reads
0 redo size
15092711 bytes sent via SQL*Net to client
370304 bytes received via SQL*Net from client
33631 SQL*Net roundtrips to/from client
0 sorts (memory)
0 sorts (disk)
504448 rows processed
SQL>
--======================
3.访问全部数据,我们发现访问heap表要比访问分区表是少了一些逻辑读：
88429<88573,其实这点差别对性能来说是无关紧要的，重要的是说明了一个问题，
那就是尽量让执行的sql少读、少些，这也是sql调整的最终目的，相差的这100
多逻辑读主要发生在查找metadata上...
SQL> select * from t1 ;
已选择1260672行。
已用时间: 00: 00: 26.70
执行计划
----------------------------------------------------------
Plan hash value: 3617692013
--------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
--------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1482K| 111M| 1064 (3)| 00:00:13 |
| 1 | TABLE ACCESS FULL| T1 | 1482K| 111M| 1064 (3)| 00:00:13 |
--------------------------------------------------------------------------
Note
-----
- dynamic sampling used for this statement
统计信息
----------------------------------------------------------
5 recursive calls
0 db block gets
88511 consistent gets
0 physical reads
0 redo size
44940044 bytes sent via SQL*Net to client
924869 bytes received via SQL*Net from client
84046 SQL*Net roundtrips to/from client
0 sorts (memory)
0 sorts (disk)
1260672 rows processed
SQL> select * from t1 ;
已选择1260672行。
已用时间: 00: 00: 24.59
执行计划
----------------------------------------------------------
Plan hash value: 3617692013
--------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
--------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1482K| 111M| 1064 (3)| 00:00:13 |
| 1 | TABLE ACCESS FULL| T1 | 1482K| 111M| 1064 (3)| 00:00:13 |
--------------------------------------------------------------------------
Note
-----
- dynamic sampling used for this statement
统计信息
----------------------------------------------------------
0 recursive calls
0 db block gets
88429 consistent gets
0 physical reads
0 redo size
44940044 bytes sent via SQL*Net to client
924869 bytes received via SQL*Net from client
84046 SQL*Net roundtrips to/from client
0 sorts (memory)
0 sorts (disk)
1260672 rows processed
SQL> select * from t ;
已选择1260544行。
已用时间: 00: 00: 25.71
执行计划
----------------------------------------------------------
Plan hash value: 3557914527
--------------------------------------------------------------------------------
------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time | Pst
art| Pstop |
--------------------------------------------------------------------------------
------------
| 0 | SELECT STATEMENT | | 1317K| 37M| 1094 (3)| 00:00:14 |
| |
| 1 | PARTITION RANGE ALL| | 1317K| 37M| 1094 (3)| 00:00:14 |
1 | 5 |
| 2 | TABLE ACCESS FULL | T | 1317K| 37M| 1094 (3)| 00:00:14 |
1 | 5 |
--------------------------------------------------------------------------------
------------
Note
-----
- dynamic sampling used for this statement
统计信息
----------------------------------------------------------
7 recursive calls
2 db block gets
90275 consistent gets
2133 physical reads
114820 redo size
39895221 bytes sent via SQL*Net to client
924781 bytes received via SQL*Net from client
84038 SQL*Net roundtrips to/from client
0 sorts (memory)
0 sorts (disk)
1260544 rows processed
SQL> select * from t ;
已选择1260544行。
已用时间: 00: 00: 24.90
执行计划
----------------------------------------------------------
Plan hash value: 3557914527
--------------------------------------------------------------------------------
------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time | Pst
art| Pstop |
--------------------------------------------------------------------------------
------------
| 0 | SELECT STATEMENT | | 1317K| 37M| 1094 (3)| 00:00:14 |
| |
| 1 | PARTITION RANGE ALL| | 1317K| 37M| 1094 (3)| 00:00:14 |
1 | 5 |
| 2 | TABLE ACCESS FULL | T | 1317K| 37M| 1094 (3)| 00:00:14 |
1 | 5 |
--------------------------------------------------------------------------------
------------
Note
-----
- dynamic sampling used for this statement
统计信息
----------------------------------------------------------
0 recursive calls
0 db block gets
88573 consistent gets
0 physical reads
0 redo size
39895221 bytes sent via SQL*Net to client
924781 bytes received via SQL*Net from client
84038 SQL*Net roundtrips to/from client
0 sorts (memory)
0 sorts (disk)
1260544 rows processed
SQL>
--==========================
有index的情况,创建global index：
有index其实也是一样的，首先通过index找到rowid，之后最终通过rowid
再去访问表，如果访问的数据尽可能的可以集中在一些分区内，那么访问
分区表的性能肯定要优于heap表
SQL> create index idx_t on t(object_id) tablespace users;
索引已创建。
已用时间: 00: 00: 10.21
SQL> create index idx_t1 on t1(object_id) tablespace users;
索引已创建。
已用时间: 00: 00: 06.89
SQL>
SQL> exec dbms_stats.gather_table_stats('SYS','T',CASCADE=>TRUE);
PL/SQL 过程已成功完成。
已用时间: 00: 00: 06.00
SQL> exec dbms_stats.gather_table_stats('SYS','T1',CASCADE=>TRUE);
PL/SQL 过程已成功完成。
已用时间: 00: 00: 03.21
SQL>
--有global index跨分区访问
SQL> select /*+ index(t idx_t) */ * from t where object_id<3000;
已选择377984行。
已用时间: 00: 00: 07.75
执行计划
----------------------------------------------------------
Plan hash value: 3317799281
--------------------------------------------------------------------------------
----------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)
| Time | Pstart| Pstop |
--------------------------------------------------------------------------------
----------------------------
| 0 | SELECT STATEMENT | | 368K| 7922K| 363K (1)
| 01:12:41 | | |
| 1 | TABLE ACCESS BY GLOBAL INDEX ROWID| T | 368K| 7922K| 363K (1)
| 01:12:41 | ROWID | ROWID |
|* 2 | INDEX RANGE SCAN | IDX_T | 368K| | 986 (1)
| 00:00:12 | | |
--------------------------------------------------------------------------------
----------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
2 - access("OBJECT_ID"<3000)
统计信息
----------------------------------------------------------
1 recursive calls
0 db block gets
398707 consistent gets
0 physical reads
0 redo size
12757072 bytes sent via SQL*Net to client
277563 bytes received via SQL*Net from client
25200 SQL*Net roundtrips to/from client
0 sorts (memory)
0 sorts (disk)
377984 rows processed
SQL> select /*+ index(t idx_t) */ * from t where object_id<3000;
已选择377984行。
已用时间: 00: 00: 09.43
执行计划
----------------------------------------------------------
Plan hash value: 3317799281
--------------------------------------------------------------------------------
----------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)
| Time | Pstart| Pstop |
--------------------------------------------------------------------------------
----------------------------
| 0 | SELECT STATEMENT | | 368K| 7922K| 363K (1)
| 01:12:41 | | |
| 1 | TABLE ACCESS BY GLOBAL INDEX ROWID| T | 368K| 7922K| 363K (1)
| 01:12:41 | ROWID | ROWID |
|* 2 | INDEX RANGE SCAN | IDX_T | 368K| | 986 (1)
| 00:00:12 | | |
--------------------------------------------------------------------------------
----------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
2 - access("OBJECT_ID"<3000)
统计信息
----------------------------------------------------------
0 recursive calls
0 db block gets
398707 consistent gets
0 physical reads
0 redo size
12757072 bytes sent via SQL*Net to client
277563 bytes received via SQL*Net from client
25200 SQL*Net roundtrips to/from client
0 sorts (memory)
0 sorts (disk)
377984 rows processed
SQL>
--======================
--有global index不垮分区访问：
SQL> select /*+ index(t1 idx_t1) */ * from t1 where object_id<2000;
已选择249984行。
已用时间: 00: 00: 05.57
执行计划
----------------------------------------------------------
Plan hash value: 50753647
--------------------------------------------------------------------------------
------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time
|
--------------------------------------------------------------------------------
------
| 0 | SELECT STATEMENT | | 245K| 5264K| 245K (1)| 00:4
9:12 |
| 1 | TABLE ACCESS BY INDEX ROWID| T1 | 245K| 5264K| 245K (1)| 00:4
9:12 |
|* 2 | INDEX RANGE SCAN | IDX_T1 | 245K| | 522 (2)| 00:0
0:07 |
--------------------------------------------------------------------------------
------
Predicate Information (identified by operation id):
---------------------------------------------------
2 - access("OBJECT_ID"<2000)
统计信息
----------------------------------------------------------
1 recursive calls
0 db block gets
266985 consistent gets
0 physical reads
0 redo size
8439590 bytes sent via SQL*Net to client
183700 bytes received via SQL*Net from client
16667 SQL*Net roundtrips to/from client
0 sorts (memory)
0 sorts (disk)
249984 rows processed
SQL> select /*+ index(t1 idx_t1) */ * from t1 where object_id<2000;
已选择249984行。
已用时间: 00: 00: 05.42
执行计划
----------------------------------------------------------
Plan hash value: 50753647
--------------------------------------------------------------------------------
------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time
|
--------------------------------------------------------------------------------
------
| 0 | SELECT STATEMENT | | 245K| 5264K| 245K (1)| 00:4
9:12 |
| 1 | TABLE ACCESS BY INDEX ROWID| T1 | 245K| 5264K| 245K (1)| 00:4
9:12 |
|* 2 | INDEX RANGE SCAN | IDX_T1 | 245K| | 522 (2)| 00:0
0:07 |
--------------------------------------------------------------------------------
------
Predicate Information (identified by operation id):
---------------------------------------------------
2 - access("OBJECT_ID"<2000)
统计信息
----------------------------------------------------------
0 recursive calls
0 db block gets
266985 consistent gets
0 physical reads
0 redo size
8439590 bytes sent via SQL*Net to client
183700 bytes received via SQL*Net from client
16667 SQL*Net roundtrips to/from client
0 sorts (memory)
0 sorts (disk)
249984 rows processed
SQL> select /*+ index(t idx_t) */ * from t where object_id<2000;
已选择249984行。
已用时间: 00: 00: 04.93
执行计划
----------------------------------------------------------
Plan hash value: 2063514567
--------------------------------------------------------------------------------
----------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)
| Time | Pstart| Pstop |
--------------------------------------------------------------------------------
----------------------------
| 0 | SELECT STATEMENT | | 259K| 5073K| 1241K (1)
| 04:08:20 | | |
| 1 | TABLE ACCESS BY GLOBAL INDEX ROWID| T | 259K| 5073K| 1241K (1)
| 04:08:20 | 1 | 1 |
|* 2 | INDEX RANGE SCAN | IDX_T | 1260K| | 3366 (1)
| 00:00:41 | | |
--------------------------------------------------------------------------------
----------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
2 - access("OBJECT_ID"<2000)
统计信息
----------------------------------------------------------
44 recursive calls
0 db block gets
262733 consistent gets
0 physical reads
0 redo size
8439590 bytes sent via SQL*Net to client
183700 bytes received via SQL*Net from client
16667 SQL*Net roundtrips to/from client
0 sorts (memory)
0 sorts (disk)
249984 rows processed
SQL>
SQL> select /*+ index(t idx_t) */ * from t where object_id<2000;
已选择249984行。
已用时间: 00: 00: 04.62
执行计划
----------------------------------------------------------
Plan hash value: 2063514567
--------------------------------------------------------------------------------
----------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)
| Time | Pstart| Pstop |
--------------------------------------------------------------------------------
----------------------------
| 0 | SELECT STATEMENT | | 259K| 5073K| 1241K (1)
| 04:08:20 | | |
| 1 | TABLE ACCESS BY GLOBAL INDEX ROWID| T | 259K| 5073K| 1241K (1)
| 04:08:20 | 1 | 1 |
|* 2 | INDEX RANGE SCAN | IDX_T | 1260K| | 3366 (1)
| 00:00:41 | | |
--------------------------------------------------------------------------------
----------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
2 - access("OBJECT_ID"<2000)
统计信息
----------------------------------------------------------
0 recursive calls
0 db block gets
262727 consistent gets
0 physical reads
0 redo size
8439590 bytes sent via SQL*Net to client
183700 bytes received via SQL*Net from client
16667 SQL*Net roundtrips to/from client
0 sorts (memory)
0 sorts (disk)
249984 rows processed
SQL>
--======================
SQL> exec dbms_stats.gather_table_stats('SYS','T',CASCADE=>TRUE);
PL/SQL 过程已成功完成。
已用时间: 00: 00: 05.82
SQL> exec dbms_stats.gather_table_stats('SYS','T1',CASCADE=>TRUE);
PL/SQL 过程已成功完成。
已用时间: 00: 00: 03.17
SQL>
--local index:
SQL> select /*+ index(t1 idx_t1) */ * from t1 where object_id<3000;
已选择377984行。
已用时间: 00: 00: 09.46
执行计划
----------------------------------------------------------
Plan hash value: 50753647
--------------------------------------------------------------------------------
------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time
|
--------------------------------------------------------------------------------
------
| 0 | SELECT STATEMENT | | 370K| 7963K| 371K (1)| 01:1
4:16 |
| 1 | TABLE ACCESS BY INDEX ROWID| T1 | 370K| 7963K| 371K (1)| 01:1
4:16 |
|* 2 | INDEX RANGE SCAN | IDX_T1 | 370K| | 786 (2)| 00:0
0:10 |
--------------------------------------------------------------------------------
------
Predicate Information (identified by operation id):
---------------------------------------------------
2 - access("OBJECT_ID"<3000)
统计信息
----------------------------------------------------------
150 recursive calls
0 db block gets
403726 consistent gets
0 physical reads
0 redo size
12757072 bytes sent via SQL*Net to client
277563 bytes received via SQL*Net from client
25200 SQL*Net roundtrips to/from client
3 sorts (memory)
0 sorts (disk)
377984 rows processed
SQL> select /*+ index(t1 idx_t1) */ * from t1 where object_id<3000;
已选择377984行。
已用时间: 00: 00: 08.28
执行计划
----------------------------------------------------------
Plan hash value: 50753647
--------------------------------------------------------------------------------
------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time
|
--------------------------------------------------------------------------------
------
| 0 | SELECT STATEMENT | | 370K| 7963K| 371K (1)| 01:1
4:16 |
| 1 | TABLE ACCESS BY INDEX ROWID| T1 | 370K| 7963K| 371K (1)| 01:1
4:16 |
|* 2 | INDEX RANGE SCAN | IDX_T1 | 370K| | 786 (2)| 00:0
0:10 |
--------------------------------------------------------------------------------
------
Predicate Information (identified by operation id):
---------------------------------------------------
2 - access("OBJECT_ID"<3000)
统计信息
----------------------------------------------------------
0 recursive calls
0 db block gets
403706 consistent gets
0 physical reads
0 redo size
12757072 bytes sent via SQL*Net to client
277563 bytes received via SQL*Net from client
25200 SQL*Net roundtrips to/from client
0 sorts (memory)
0 sorts (disk)
377984 rows processed
SQL> select /*+ index(t idx_t) */ * from t where object_id<3000;
已选择377984行。
已用时间: 00: 00: 07.17
执行计划
----------------------------------------------------------
Plan hash value: 3038871768
--------------------------------------------------------------------------------
----------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)
| Time | Pstart| Pstop |
--------------------------------------------------------------------------------
----------------------------
| 0 | SELECT STATEMENT | | 368K| 7566K| 363K (1)
| 01:12:40 | | |
| 1 | PARTITION RANGE ITERATOR | | 368K| 7566K| 363K (1)
| 01:12:40 | 1 | 2 |
| 2 | TABLE ACCESS BY LOCAL INDEX ROWID| T | 368K| 7566K| 363K (1)
| 01:12:40 | 1 | 2 |
|* 3 | INDEX RANGE SCAN | IDX_T | 368K| | 783 (2)
| 00:00:10 | 1 | 2 |
--------------------------------------------------------------------------------
----------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
3 - access("OBJECT_ID"<3000)
统计信息
----------------------------------------------------------
195 recursive calls
0 db block gets
398680 consistent gets
0 physical reads
0 redo size
5145251 bytes sent via SQL*Net to client
277563 bytes received via SQL*Net from client
25200 SQL*Net roundtrips to/from client
6 sorts (memory)
0 sorts (disk)
377984 rows processed
SQL> select /*+ index(t idx_t) */ * from t where object_id<3000;
已选择377984行。
已用时间: 00: 00: 07.42
执行计划
----------------------------------------------------------
Plan hash value: 3038871768
--------------------------------------------------------------------------------
----------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)
| Time | Pstart| Pstop |
--------------------------------------------------------------------------------
----------------------------
| 0 | SELECT STATEMENT | | 368K| 7566K| 363K (1)
| 01:12:40 | | |
| 1 | PARTITION RANGE ITERATOR | | 368K| 7566K| 363K (1)
| 01:12:40 | 1 | 2 |
| 2 | TABLE ACCESS BY LOCAL INDEX ROWID| T | 368K| 7566K| 363K (1)
| 01:12:40 | 1 | 2 |
|* 3 | INDEX RANGE SCAN | IDX_T | 368K| | 783 (2)
| 00:00:10 | 1 | 2 |
--------------------------------------------------------------------------------
----------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
3 - access("OBJECT_ID"<3000)
统计信息
----------------------------------------------------------
0 recursive calls
0 db block gets
398633 consistent gets
0 physical reads
0 redo size
5145251 bytes sent via SQL*Net to client
277563 bytes received via SQL*Net from client
25200 SQL*Net roundtrips to/from client
0 sorts (memory)
0 sorts (disk)
377984 rows processed
SQL>
--===============================
有local index不垮分区访问：
SQL> select /*+ index(t idx_t) */ * from t where object_id<2000;
已选择249984行。
已用时间: 00: 00: 04.93
执行计划
----------------------------------------------------------
Plan hash value: 3965956311
--------------------------------------------------------------------------------
----------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)
| Time | Pstart| Pstop |
--------------------------------------------------------------------------------
----------------------------
| 0 | SELECT STATEMENT | | 254K| 4962K| 245K (1)
| 00:49:11 | | |
| 1 | PARTITION RANGE SINGLE | | 254K| 4962K| 245K (1)
| 00:49:11 | 1 | 1 |
| 2 | TABLE ACCESS BY LOCAL INDEX ROWID| T | 254K| 4962K| 245K (1)
| 00:49:11 | 1 | 1 |
|* 3 | INDEX RANGE SCAN | IDX_T | 249K| | 528 (2)
| 00:00:07 | 1 | 1 |
--------------------------------------------------------------------------------
----------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
3 - access("OBJECT_ID"<2000)
统计信息
----------------------------------------------------------
24 recursive calls
0 db block gets
262679 consistent gets
0 physical reads
0 redo size
3406055 bytes sent via SQL*Net to client
183700 bytes received via SQL*Net from client
16667 SQL*Net roundtrips to/from client
0 sorts (memory)
0 sorts (disk)
249984 rows processed
SQL> select /*+ index(t idx_t) */ * from t where object_id<2000;
已选择249984行。
已用时间: 00: 00: 03.48
执行计划
----------------------------------------------------------
Plan hash value: 3965956311
--------------------------------------------------------------------------------
----------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)
| Time | Pstart| Pstop |
--------------------------------------------------------------------------------
----------------------------
| 0 | SELECT STATEMENT | | 254K| 4962K| 245K (1)
| 00:49:11 | | |
| 1 | PARTITION RANGE SINGLE | | 254K| 4962K| 245K (1)
| 00:49:11 | 1 | 1 |
| 2 | TABLE ACCESS BY LOCAL INDEX ROWID| T | 254K| 4962K| 245K (1)
| 00:49:11 | 1 | 1 |
|* 3 | INDEX RANGE SCAN | IDX_T | 249K| | 528 (2)
| 00:00:07 | 1 | 1 |
--------------------------------------------------------------------------------
----------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
3 - access("OBJECT_ID"<2000)
统计信息
----------------------------------------------------------
0 recursive calls
0 db block gets
262676 consistent gets
0 physical reads
0 redo size
3406055 bytes sent via SQL*Net to client
183700 bytes received via SQL*Net from client
16667 SQL*Net roundtrips to/from client
0 sorts (memory)
0 sorts (disk)
249984 rows processed
SQL> select /*+ index(t1 idx_t1) */ * from t1 where object_id<2000;
已选择249984行。
已用时间: 00: 00: 05.40
执行计划
----------------------------------------------------------
Plan hash value: 50753647
--------------------------------------------------------------------------------
------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time
|
--------------------------------------------------------------------------------
------
| 0 | SELECT STATEMENT | | 247K| 5306K| 247K (1)| 00:4
9:29 |
| 1 | TABLE ACCESS BY INDEX ROWID| T1 | 247K| 5306K| 247K (1)| 00:4
9:29 |
|* 2 | INDEX RANGE SCAN | IDX_T1 | 247K| | 525 (2)| 00:0
0:07 |
--------------------------------------------------------------------------------
------
Predicate Information (identified by operation id):
---------------------------------------------------
2 - access("OBJECT_ID"<2000)
统计信息
----------------------------------------------------------
0 recursive calls
0 db block gets
266985 consistent gets
0 physical reads
0 redo size
8439590 bytes sent via SQL*Net to client
183700 bytes received via SQL*Net from client
16667 SQL*Net roundtrips to/from client
0 sorts (memory)
0 sorts (disk)
249984 rows processed
SQL> select /*+ index(t1 idx_t1) */ * from t1 where object_id<2000;
已选择249984行。
已用时间: 00: 00: 05.70
执行计划
----------------------------------------------------------
Plan hash value: 50753647
--------------------------------------------------------------------------------
------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time
|
--------------------------------------------------------------------------------
------
| 0 | SELECT STATEMENT | | 247K| 5306K| 247K (1)| 00:4
9:29 |
| 1 | TABLE ACCESS BY INDEX ROWID| T1 | 247K| 5306K| 247K (1)| 00:4
9:29 |
|* 2 | INDEX RANGE SCAN | IDX_T1 | 247K| | 525 (2)| 00:0
0:07 |
--------------------------------------------------------------------------------
------
Predicate Information (identified by operation id):
---------------------------------------------------
2 - access("OBJECT_ID"<2000)
统计信息
----------------------------------------------------------
0 recursive calls
0 db block gets
266985 consistent gets
0 physical reads
0 redo size
8439590 bytes sent via SQL*Net to client
183700 bytes received via SQL*Net from client
16667 SQL*Net roundtrips to/from client
0 sorts (memory)
0 sorts (disk)
249984 rows processed
SQL>
--=========================
SQL> create index idx_t_name on t(object_name) tablespace users;
索引已创建。
SQL>
--=========================
SQL> select index_name,status from dba_indexes where index_name='IDX_T_NAME';
INDEX_NAME STATUS
------------------------------ --------
IDX_T_NAME VALID
SQL> alter table t truncate partition p4;
表被截断。
--在分区表上创建的global index(注意非分区)，即使我们truncate partition
其实这个global index是不会自动维护的。
SQL> select index_name,status from dba_indexes where index_name='IDX_T_NAME';
INDEX_NAME STATUS
------------------------------ --------
IDX_T_NAME UNUSABLE
SQL> COL OBJECT_NAME FORMAT A20
SQL> select object_name,status from dba_objects where object_name='IDX_T_NAME';
OBJECT_NAME STATUS
-------------------- -------
IDX_T_NAME VALID
SQL>
SQL> select bytes/1024/1024 m from dba_segments where segment_name='IDX_T_NAME';
M
----------
47
SQL> alter index idx_t_name rebuild;
索引已更改。
SQL> select bytes/1024/1024 m from dba_segments where segment_name='IDX_T_NAME';
M
----------
37
SQL> select index_name,status from dba_indexes where index_name='IDX_T_NAME';
INDEX_NAME STATUS
------------------------------ --------
IDX_T_NAME VALID
SQL>
--==========================
验证一下dml对global index的影响,dml操作oracle会自动update global index(注意:并非global partition index)，这个
比较好理解，也应该这样做...
SQL> select name,del_lf_rows,del_lf_rows_len from index_stats;
NAME DEL_LF_ROWS DEL_LF_ROWS_LEN
------------------------------ ----------- ---------------
IDX_T_NAME 256 10112
SQL>
SQL> delete from t where object_id=5002;
已删除128行。
SQL> commit;
提交完成。
SQL> analyze index idx_t_name validate structure;
索引已分析
SQL> select name,del_lf_rows,del_lf_rows_len from index_stats;
NAME DEL_LF_ROWS DEL_LF_ROWS_LEN
------------------------------ ----------- ---------------
IDX_T_NAME 384 15616
SQL>
SQL> select index_name,status from dba_indexes where index_name='IDX_T_NAME';
INDEX_NAME STATUS
------------------------------ --------
IDX_T_NAME VALID
--=========================
验证在分区表上创建主键和唯一索引的情况：
SQL> truncate table t ;
表被截断。
SQL> alter table t add pk number;
表已更改。
SQL> insert into t(pk,object_id,object_name) select rownum,object_id,object_name
from t1;
已创建1260672行。
SQL> commit;
提交完成。
SQL> alter table t add constraint pk_t primary key (pk);
表已更改。
SQL> alter table t drop constraint pk_t;
表已更改。
SQL> create unique index idx_t_pk on t(pk) tablespace users;
索引已创建。
SQL> create unique index idx_t_pk on t(pk) local tablespace users;
create unique index idx_t_pk on t(pk) local tablespace users
*
第 1 行出现错误:
ORA-14039: 分区列必须构成 UNIQUE 索引的关键字列子集
SQL> create unique index idx_t_pk on t(pk,object_id) local tablespace users;
索引已创建。
SQL> drop index idx_t_pk;
索引已删除。
SQL> alter table t add constraint pk_t primary key (pk) using index local;
alter table t add constraint pk_t primary key (pk) using index local
*
第 1 行出现错误:
ORA-14039: 分区列必须构成 UNIQUE 索引的关键字列子集
SQL> alter table t add constraint pk_t primary key (pk,object_id) using index l
ocal;
表已更改。
SQL> alter table t drop constraint pk_t;
表已更改。
SQL> create index idx_t_pk on t(pk) local tablespace users;
索引已创建。
SQL> drop index idx_t_pk;
索引已删除。
SQL>
结论:不管我们在创建unique index还是添加primary key时，只要我们
想把index创建成local分区index，那么就应该至少包含分区列...而不是象
网上有些人说的在分区表上创建primary key或者nuique index时必须包含分区列。
--=======================
ddl对local partition index的影响,oracle会自动维护ddl对local partition index的影响：
SQL> select name,del_lf_rows,del_lf_rows_len from index_stats;
NAME DEL_LF_ROWS DEL_LF_ROWS_LEN
------------------------------ ----------- ---------------
IDX_T 0 0
SQL> alter table t truncate partition p4;
表被截断。
SQL> analyze index idx_t validate structure;
索引已分析
SQL> select name,del_lf_rows,del_lf_rows_len from index_stats;
NAME DEL_LF_ROWS DEL_LF_ROWS_LEN
------------------------------ ----------- ---------------
IDX_T 0 0
--看来通过analyze index idx_t validate structure;没法验证ddl对local partition index的维护，因为
在truncate partition前后index_stats里的数据没有变化
SQL> select index_name,partition_name from dba_ind_partitions where index_name i
n ('IDX_T','IDX_T1','IDX_T_NAME');
INDEX_NAME PARTITION_NAME
------------------------------ ------------------------------
IDX_T P5
IDX_T P4
IDX_T P3
IDX_T P2
IDX_T P1
SQL> alter table t drop partition p4;
表已更改。
SQL> select index_name,partition_name from dba_ind_partitions where index_name i
n ('IDX_T','IDX_T1','IDX_T_NAME');
INDEX_NAME PARTITION_NAME
------------------------------ ------------------------------
IDX_T P5
IDX_T P3
IDX_T P2
IDX_T P1
--drop partition之后和该分区相关的local partition index也被drop了...
SQL> select object_name,subobject_name,last_ddl_time from dba_objects where obje
ct_name='IDX_T';
OBJECT_NAM SUBOBJECT_NAME LAST_DDL_TIME
---------- ------------------------------ -------------------
IDX_T P1 2010/11/02 12:55:38
IDX_T P2 2010/11/02 12:55:38
IDX_T P3 2010/11/02 12:55:38
IDX_T P5 2010/11/02 12:55:38
IDX_T 2010/11/02 13:42:46
SQL> alter table t truncate partition p3;
表被截断。
SQL> select object_name,subobject_name,last_ddl_time from dba_objects where obje
ct_name='IDX_T';
OBJECT_NAM SUBOBJECT_NAME LAST_DDL_TIME
---------- ------------------------------ -------------------
IDX_T P1 2010/11/02 12:55:38
IDX_T P2 2010/11/02 12:55:38
IDX_T P3 2010/11/02 13:50:29
IDX_T P5 2010/11/02 12:55:38
IDX_T 2010/11/02 13:50:29
--通过ddl里的时间我们可以清楚的发现在truncate partition时oracle维护了
其对应的local index，因为p3对应的ddl时间由原来的12:55:38变成了13:50:29
SQL> select segment_name,partition_name,bytes/1024/1024 m from dba_segments whe
re segment_name='IDX_T';
SEGMENT_NAME PARTITION_NAME M
-------------------- ------------------------------ ----------
IDX_T P1 5
IDX_T P2 5
IDX_T P3 .0625
IDX_T P5 6
SQL> alter table t truncate partition p2;
表被截断。
SQL> select segment_name,partition_name,bytes/1024/1024 m from dba_segments whe
re segment_name='IDX_T';
SEGMENT_NAME PARTITION_NAME M
-------------------- ------------------------------ ----------
IDX_T P1 5
IDX_T P2 .0625
IDX_T P3 .0625
IDX_T P5 6
SQL>
--当然通过local index的大小我们也可以清楚的观查到，local partition index
p2的大小在truncate p2之前是5m，之后变成了0.625m，可见oracle维护了local index
--======================
dml对global partition index的影响：
SQL> select name,del_lf_rows,del_lf_rows_len from index_stats;
NAME DEL_LF_ROWS DEL_LF_ROWS_LEN
------------------------------ ----------- ---------------
IDX_T 0 0
SQL> delete from t where object_id=100;
已删除128行。
SQL> commit;
提交完成。
SQL> analyze index idx_t_g validate structure;
索引已分析
SQL> select name,del_lf_rows,del_lf_rows_len from index_stats;
NAME DEL_LF_ROWS DEL_LF_ROWS_LEN
------------------------------ ----------- ---------------
IDX_T_G 127 2565
SQL>
--很显然oracle会自动维护dml对global partition index的维护，这个其实
不用验证也没有问题，如果不能自动维护dml，那么怎么使用这种类型的index
--=========================
dll对global partition index的影响：
SQL> select object_name,subobject_name,last_ddl_time from dba_objects where obje
ct_name='IDX_T_G';
OBJECT_NAM SUBOBJECT_NAME LAST_DDL_TIME
---------- ------------------------------ -------------------
IDX_T_G GP1 2010/11/02 14:17:49
IDX_T_G GP2 2010/11/02 14:17:49
IDX_T_G GP3 2010/11/02 14:17:49
IDX_T_G GP4 2010/11/02 14:17:49
IDX_T_G GP5 2010/11/02 14:17:49
IDX_T_G GP6 2010/11/02 14:17:49
IDX_T_G GP_MAX 2010/11/02 14:17:49
IDX_T_G 2010/11/02 14:17:49
已选择8行。
SQL> select segment_name,partition_name,bytes/1024/1024 m from dba_segments whe
re segment_name='IDX_T_G';
SEGMENT_NAME PARTITION_NAME M
-------------------- ------------------------------ ----------
IDX_T_G GP1 .0625
IDX_T_G GP2 .0625
IDX_T_G GP3 .0625
IDX_T_G GP4 .0625
IDX_T_G GP5 .0625
IDX_T_G GP6 .0625
IDX_T_G GP_MAX 12
已选择7行。
SQL> alter table t truncate partition p5;
表被截断。
SQL> select object_name,subobject_name,last_ddl_time from dba_objects where obje
ct_name='IDX_T_G';
OBJECT_NAM SUBOBJECT_NAME LAST_DDL_TIME
---------- ------------------------------ -------------------
IDX_T_G GP1 2010/11/02 14:17:49
IDX_T_G GP2 2010/11/02 14:17:49
IDX_T_G GP3 2010/11/02 14:17:49
IDX_T_G GP4 2010/11/02 14:17:49
IDX_T_G GP5 2010/11/02 14:17:49
IDX_T_G GP6 2010/11/02 14:17:49
IDX_T_G GP_MAX 2010/11/02 14:17:49
IDX_T_G 2010/11/02 14:17:49
已选择8行。
SQL> select segment_name,partition_name,bytes/1024/1024 m from dba_segments whe
re segment_name='IDX_T_G';
SEGMENT_NAME PARTITION_NAME M
-------------------- ------------------------------ ----------
IDX_T_G GP1 .0625
IDX_T_G GP2 .0625
IDX_T_G GP3 .0625
IDX_T_G GP4 .0625
IDX_T_G GP5 .0625
IDX_T_G GP6 .0625
IDX_T_G GP_MAX 12
已选择7行。
SQL>
--global index partition GP_MAX记录的数据>9000,而表分区p5的数据>8000,
在truncate p5之后对比GP_MAX前后的ddl修改时间和段大小，发现都没有任何变化,
我就认为ddl对global index partition不自动维护吧，实在没有太好的验证办法，当然
可以dump index的结构来看，感兴趣的自己测试一下吧。
其实从下面查询index的状态就可以看出来，此时的index状态全部变成了UNUSABLE：
SQL> select index_name,partition_name,status from dba_ind_partitions where index
_name in ('IDX_T_G');
INDEX_NAME PARTITION_NAME STATUS
------------------------------ ------------------------------ --------
IDX_T_G GP1 UNUSABLE
IDX_T_G GP2 UNUSABLE
IDX_T_G GP3 UNUSABLE
IDX_T_G GP4 UNUSABLE
IDX_T_G GP5 UNUSABLE
IDX_T_G GP6 UNUSABLE
IDX_T_G GP_MAX UNUSABLE
已选择7行。
SQL>
尝试重建这个global partition index ：
SQL> alter index idx_t_g rebuild;
alter index idx_t_g rebuild
*
第 1 行出现错误:
ORA-14086: 不能将分区索引作为整体重建
SQL>
--看来只能一个一个重建了：
SQL> alter index idx_t_g rebuild partition gp1 ;
索引已更改。
SQL> alter index idx_t_g rebuild partition gp2 ;
索引已更改。
SQL> alter index idx_t_g rebuild partition gp3 ;
索引已更改。
SQL> alter index idx_t_g rebuild partition gp4 ;
索引已更改。
SQL> alter index idx_t_g rebuild partition gp5 ;
索引已更改。
SQL> alter index idx_t_g rebuild partition gp6 ;
索引已更改。
SQL> alter index idx_t_g rebuild partition gp_max ;
索引已更改。
SQL> select index_name,partition_name,status from dba_ind_partitions where index
_name in ('IDX_T_G');
INDEX_NAME PARTITION_NAME STATUS
------------------------------ ------------------------------ --------
IDX_T_G GP1 USABLE
IDX_T_G GP2 USABLE
IDX_T_G GP3 USABLE
IDX_T_G GP4 USABLE
IDX_T_G GP5 USABLE
IDX_T_G GP6 USABLE
IDX_T_G GP_MAX USABLE
已选择7行。
SQL>
--重建之后状态变成了USABLE，注意只要我们truncate一个分区，那么global
partition index的分区index的状态全部变成UNUSABLE
如何自动维护global partition index：
SQL> select index_name,partition_name,status from dba_ind_partitions where index
_name in ('IDX_T_G');
INDEX_NAME PARTITION_NAME STATUS
------------------------------ ------------------------------ --------
IDX_T_G GP1 USABLE
IDX_T_G GP2 USABLE
IDX_T_G GP3 USABLE
IDX_T_G GP4 USABLE
IDX_T_G GP5 USABLE
IDX_T_G GP6 USABLE
IDX_T_G GP_MAX USABLE
已选择7行。
SQL> alter table t truncate partition p1 update global indexes;
表被截断。
SQL> select index_name,partition_name,status from dba_ind_partitions where index
_name in ('IDX_T_G');
INDEX_NAME PARTITION_NAME STATUS
------------------------------ ------------------------------ --------
IDX_T_G GP1 USABLE
IDX_T_G GP2 USABLE
IDX_T_G GP3 USABLE
IDX_T_G GP4 USABLE
IDX_T_G GP5 USABLE
IDX_T_G GP6 USABLE
IDX_T_G GP_MAX USABLE
已选择7行。
SQL>
显然在ddl语句truncate后面加上update global indexes全局分区index的状态都变成了USABLE
，很显然oracle自动为了index，不过如果分区表和分区index比较大的话，那么update global indexes
的处理时间可能会很长，通过锁定表的时间也会很长，所以使用update global indexes子句要慎重。
如果全局分区index的状态原来就是UNUSABLE，那么即使加上update global indexes子句，oracle也不会
自动维护index：
SQL> select index_name,partition_name,status from dba_ind_partitions where index
_name in ('IDX_T_G');
INDEX_NAME PARTITION_NAME STATUS
------------------------------ ------------------------------ --------
IDX_T_G GP1 UNUSABLE
IDX_T_G GP2 UNUSABLE
IDX_T_G GP3 UNUSABLE
IDX_T_G GP4 UNUSABLE
IDX_T_G GP5 UNUSABLE
IDX_T_G GP6 UNUSABLE
IDX_T_G GP_MAX UNUSABLE
已选择7行。
SQL> alter table t truncate partition p1 update global indexes;
表被截断。
SQL>
SQL> select index_name,partition_name,status from dba_ind_partitions where index
_name in ('IDX_T_G');
INDEX_NAME PARTITION_NAME STATUS
------------------------------ ------------------------------ --------
IDX_T_G GP1 UNUSABLE
IDX_T_G GP2 UNUSABLE
IDX_T_G GP3 UNUSABLE
IDX_T_G GP4 UNUSABLE
IDX_T_G GP5 UNUSABLE
IDX_T_G GP6 UNUSABLE
IDX_T_G GP_MAX UNUSABLE
已选择7行。
--==========================
下面简单对比一下存在global partition index的分区表和heap表的性能：
SQL> desc t1
名称 是否为空? 类型
----------------------------------------- -------- ----------------------------
OBJECT_ID NUMBER
OBJECT_NAME VARCHAR2(128)
SQL> desc t
名称 是否为空? 类型
----------------------------------------- -------- ----------------------------
OBJECT_ID NUMBER
OBJECT_NAME VARCHAR2(30)
PK NUMBER
SQL> insert into t select t1.*,rownum from t1;
已创建1260672行。
SQL> commit;
提交完成。
SQL> truncate table t1;
表被截断。
SQL> alter table t1 add pk number;
表已更改。
SQL> insert into t1 select * from t;
已创建1260672行。
SQL> commit;
提交完成。
SQL> select index_name,partition_name,status from dba_ind_partitions where index
_name in ('IDX_T_G');
INDEX_NAME PARTITION_NAME STATUS
------------------------------ ------------------------------ --------
IDX_T_G GP1 USABLE
IDX_T_G GP2 USABLE
IDX_T_G GP3 USABLE
IDX_T_G GP4 USABLE
IDX_T_G GP5 USABLE
IDX_T_G GP6 USABLE
IDX_T_G GP_MAX USABLE
已选择7行。
SQL> create index idx_t1_pk on t1(pk) tablespace users;
索引已创建。
SQL> set autotrace traceonly
SQL>
--===========================
SQL> select * from t1 where pk<3000;
已选择2999行。
执行计划
----------------------------------------------------------
Plan hash value: 4165294307
--------------------------------------------------------------------------------
---------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| T
ime |
--------------------------------------------------------------------------------
---------
| 0 | SELECT STATEMENT | | 63401 | 1362K| 224 (1)| 0
0:00:03 |
| 1 | TABLE ACCESS BY INDEX ROWID| T1 | 63401 | 1362K| 224 (1)| 0
0:00:03 |
|* 2 | INDEX RANGE SCAN | IDX_T1_PK | 11412 | | 28 (0)| 0
0:00:01 |
--------------------------------------------------------------------------------
---------
Predicate Information (identified by operation id):
---------------------------------------------------
2 - access("PK"<3000)
统计信息
----------------------------------------------------------
1 recursive calls
0 db block gets
423 consistent gets
8 physical reads
0 redo size
113183 bytes sent via SQL*Net to client
2574 bytes received via SQL*Net from client
201 SQL*Net roundtrips to/from client
0 sorts (memory)
0 sorts (disk)
2999 rows processed
SQL> select * from t1 where pk<3000;
已选择2999行。
执行计划
----------------------------------------------------------
Plan hash value: 4165294307
--------------------------------------------------------------------------------
---------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| T
ime |
--------------------------------------------------------------------------------
---------
| 0 | SELECT STATEMENT | | 63401 | 1362K| 224 (1)| 0
0:00:03 |
| 1 | TABLE ACCESS BY INDEX ROWID| T1 | 63401 | 1362K| 224 (1)| 0
0:00:03 |
|* 2 | INDEX RANGE SCAN | IDX_T1_PK | 11412 | | 28 (0)| 0
0:00:01 |
--------------------------------------------------------------------------------
---------
Predicate Information (identified by operation id):
---------------------------------------------------
2 - access("PK"<3000)
统计信息
----------------------------------------------------------
0 recursive calls
0 db block gets
423 consistent gets
0 physical reads
0 redo size
113183 bytes sent via SQL*Net to client
2574 bytes received via SQL*Net from client
201 SQL*Net roundtrips to/from client
0 sorts (memory)
0 sorts (disk)
2999 rows processed
SQL> select * from t where pk<3000;
已选择2999行。
执行计划
----------------------------------------------------------
Plan hash value: 2375091741
--------------------------------------------------------------------------------
-------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%C
PU)| Time | Pstart| Pstop |
--------------------------------------------------------------------------------
-------------------------------
| 0 | SELECT STATEMENT | | 63125 | 1294K| 62
(0)| 00:00:01 | | |
| 1 | PARTITION RANGE ITERATOR | | 63125 | 1294K| 62
(0)| 00:00:01 | 1 | 2 |
| 2 | TABLE ACCESS BY GLOBAL INDEX ROWID| T | 63125 | 1294K| 62
(0)| 00:00:01 | ROWID | ROWID |
|* 3 | INDEX RANGE SCAN | IDX_T_G | 11362 | | 16
(0)| 00:00:01 | 1 | 2 |
--------------------------------------------------------------------------------
-------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
3 - access("PK"<3000)
统计信息
----------------------------------------------------------
1 recursive calls
0 db block gets
430 consistent gets
20 physical reads
668 redo size
101244 bytes sent via SQL*Net to client
2574 bytes received via SQL*Net from client
201 SQL*Net roundtrips to/from client
0 sorts (memory)
0 sorts (disk)
2999 rows processed
SQL> select * from t where pk<3000;
已选择2999行。
执行计划
----------------------------------------------------------
Plan hash value: 2375091741
--------------------------------------------------------------------------------
-------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%C
PU)| Time | Pstart| Pstop |
--------------------------------------------------------------------------------
-------------------------------
| 0 | SELECT STATEMENT | | 63125 | 1294K| 62
(0)| 00:00:01 | | |
| 1 | PARTITION RANGE ITERATOR | | 63125 | 1294K| 62
(0)| 00:00:01 | 1 | 2 |
| 2 | TABLE ACCESS BY GLOBAL INDEX ROWID| T | 63125 | 1294K| 62
(0)| 00:00:01 | ROWID | ROWID |
|* 3 | INDEX RANGE SCAN | IDX_T_G | 11362 | | 16
(0)| 00:00:01 | 1 | 2 |
--------------------------------------------------------------------------------
-------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
3 - access("PK"<3000)
统计信息
----------------------------------------------------------
0 recursive calls
0 db block gets
422 consistent gets
0 physical reads
0 redo size
101244 bytes sent via SQL*Net to client
2574 bytes received via SQL*Net from client
201 SQL*Net roundtrips to/from client
0 sorts (memory)
0 sorts (disk)
2999 rows processed
SQL>
发现逻辑读取相差不大，原因是，跨的分区比较少，其实想一想能差在哪儿呢？
首先都是通过在index里找rowid，其次通过rowid在表里或者分区里找数据，在index里
找到的rowid数量就差不多，因为我们的数据量一样大，而且数据完全一致，差别可能就是
通过访问分区表可能查找metadata需要多消耗几个逻辑读而已
--===========================
通过对比分区表和普通表，简单的测试了性能和对分区表存在index时的维护，最后我们不得不说
使用分区表很多时候不一定能提高性能，主要是维护起来方便，如果我们能把访问的数据集中
在一个或者有限的几个分区里，那么性能肯定比访问普通的全表要好，还是那句话
让执行的sql尽可能的少读、少写，这样才是提高sql性能的关键，就像anlinew提到的公式：
T=S/V,其实少读、少写的目的就是减小S.
和分区表相关的一点总结！
-
410192979
2016-03-22 21:47:41 -
Oracle
- 转载