[20180317]12c TABLE ACCESS BY INDEX ROWID BAT…
2018-06-17 20:05:35来源:未知 阅读 ()
[20180317]12c TABLE ACCESS BY INDEX ROWID BATCHED2.txt
--//简单探究12c TABLE ACCESS BY INDEX ROWID BATCHED特性.
--//当使用12c时,执行计划出现TABLE ACCESS BY INDEX ROWID BATCHED,做一些探究.
--//本文主要探究如何使用提示或者隐含参数控制这种特性.
1.环境:
SCOTT@test01p> @ ver1
PORT_STRING VERSION BANNER CON_ID
------------------------------ -------------- -------------------------------------------------------------------------------- ----------
IBMPC/WIN_NT64-9.1.0 12.1.0.1.0 Oracle Database 12c Enterprise Edition Release 12.1.0.1.0 - 64bit Production 0
SCOTT@test01p> show array
arraysize 200
SCOTT@test01p> create table t as select * from all_objects order by DBMS_RANDOM.random;
Table created.
SCOTT@test01p> create index i_t_object_id on t(object_id);
Index created.
--//分析.
execute sys.dbms_stats.gather_table_stats ( OwnName => nvl('',user),TabName => 't',Estimate_Percent => NULL,Method_Opt => 'FOR ALL COLUMNS SIZE 1 ',Cascade => True ,No_Invalidate => false)
SCOTT@test01p> select rowid ,owner from t where object_id between 1 and 10;
ROWID OWNER
------------------ --------------------
AAAaKPAAJAAAAY7AAE SYS
AAAaKPAAJAAAAaWAAP SYS
AAAaKPAAJAAAARtAAd SYS
AAAaKPAAJAAAAK5AAw SYS
AAAaKPAAJAAAAbGAAV SYS
AAAaKPAAJAAAAIFAAK SYS
AAAaKPAAJAAAASIAAB SYS
AAAaKPAAJAAAAMIAAQ SYS
AAAaKPAAJAAAAOMAAp SYS
9 rows selected.
SCOTT@test01p> select data_object_id,object_id from dba_objects where owner=user and object_name='T';
DATA_OBJECT_ID OBJECT_ID
-------------- ----------
107151 107151
--//看看那些参数可以控制取消TABLE ACCESS BY INDEX ROWID BATCHED特性.
2.测试:
--//通过隐含参数_optimizer_batch_table_access_by_rowid可以改变执行计划.
SYS@test> @ hide _optimizer_batch_table_access_by_rowid
NAME DESCRIPTION DEFAULT_VALUE SESSION_VALUE SYSTEM_VALUE
---------------------------------------- ---------------------------------------- ------------- ------------- ------------
_optimizer_batch_table_access_by_rowid enable table access by ROWID IO batching TRUE TRUE TRUE
SCOTT@test01p> alter session set statistics_level=all;
Session altered.
SCOTT@test01p> select rowid ,owner from t where object_id between 1 and 10;
ROWID OWNER
------------------ -----
AAAaKPAAJAAAAY7AAE SYS
AAAaKPAAJAAAAaWAAP SYS
AAAaKPAAJAAAARtAAd SYS
AAAaKPAAJAAAAK5AAw SYS
AAAaKPAAJAAAAbGAAV SYS
AAAaKPAAJAAAAIFAAK SYS
AAAaKPAAJAAAASIAAB SYS
AAAaKPAAJAAAAMIAAQ SYS
AAAaKPAAJAAAAOMAAp SYS
9 rows selected.
SCOTT@test01p> @ dpc '' ''
PLAN_TABLE_OUTPUT
-------------------------------------
SQL_ID anscphj6zbgpn, child number 0
-------------------------------------
select rowid ,owner from t where object_id between 1 and 10
Plan hash value: 2683697726
---------------------------------------------------------------------------------------------------------------------------------------
| Id | Operation | Name | Starts | E-Rows |E-Bytes| Cost (%CPU)| E-Time | A-Rows | A-Time | Buffers |
---------------------------------------------------------------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1 | | | 10 (100)| | 9 |00:00:00.01 | 12 |
| 1 | TABLE ACCESS BY INDEX ROWID| T | 1 | 8 | 184 | 10 (0)| 00:00:01 | 9 |00:00:00.01 | 12 |
|* 2 | INDEX RANGE SCAN | I_T_OBJECT_ID | 1 | 8 | | 2 (0)| 00:00:01 | 9 |00:00:00.01 | 3 |
---------------------------------------------------------------------------------------------------------------------------------------
Query Block Name / Object Alias (identified by operation id):
-------------------------------------------------------------
1 - SEL$1 / T@SEL$1
2 - SEL$1 / T@SEL$1
Predicate Information (identified by operation id):
---------------------------------------------------
2 - access("OBJECT_ID">=1 AND "OBJECT_ID"<=10)
--//现在执行使用TABLE ACCESS BY INDEX ROWID.顺便测试10046事件的跟踪情况.
SCOTT@test01p> alter system flush buffer_cache ;
System altered.
@ 10046on 12
select rowid ,owner from t where object_id between 1 and 10;
@ 10046off
--//转储文件:
=====================
PARSING IN CURSOR #180365864 len=61 dep=0 uid=109 oct=3 lid=109 tim=1785413706 hv=1307950772 ad='7ff1292e768' sqlid='anscphj6zbgpn'
select rowid ,owner from t where object_id between 1 and 10
END OF STMT
PARSE #180365864:c=62400,e=231012,p=15,cr=141,cu=0,mis=1,r=0,dep=0,og=1,plh=2683697726,tim=1785413704
EXEC #180365864:c=0,e=45,p=0,cr=0,cu=0,mis=0,r=0,dep=0,og=1,plh=2683697726,tim=1785413932
WAIT #180365864: nam='SQL*Net message to client' ela= 7 driver id=1413697536 #bytes=1 p3=0 obj#=665 tim=1785414043
WAIT #180365864: nam='db file sequential read' ela= 14145 file#=9 block#=363 blocks=1 obj#=107152 tim=1785428269
WAIT #180365864: nam='db file sequential read' ela= 315 file#=9 block#=364 blocks=1 obj#=107152 tim=1785428826
WAIT #180365864: nam='db file sequential read' ela= 5921 file#=9 block#=1595 blocks=1 obj#=107151 tim=1785434811
FETCH #180365864:c=0,e=20917,p=3,cr=3,cu=0,mis=0,r=1,dep=0,og=1,plh=2683697726,tim=1785435009
WAIT #180365864: nam='SQL*Net message from client' ela= 594 driver id=1413697536 #bytes=1 p3=0 obj#=107151 tim=1785435681
WAIT #180365864: nam='db file sequential read' ela= 7635 file#=9 block#=1686 blocks=1 obj#=107151 tim=1785443416
WAIT #180365864: nam='SQL*Net message to client' ela= 5 driver id=1413697536 #bytes=1 p3=0 obj#=107151 tim=1785443628
WAIT #180365864: nam='db file sequential read' ela= 17772 file#=9 block#=1133 blocks=1 obj#=107151 tim=1785461472
WAIT #180365864: nam='db file sequential read' ela= 6360 file#=9 block#=697 blocks=1 obj#=107151 tim=1785468019
WAIT #180365864: nam='db file sequential read' ela= 276 file#=9 block#=1734 blocks=1 obj#=107151 tim=1785468533
WAIT #180365864: nam='db file sequential read' ela= 6200 file#=9 block#=517 blocks=1 obj#=107151 tim=1785474873
WAIT #180365864: nam='db file sequential read' ela= 10777 file#=9 block#=1160 blocks=1 obj#=107151 tim=1785485815
WAIT #180365864: nam='db file sequential read' ela= 11727 file#=9 block#=776 blocks=1 obj#=107151 tim=1785497741
WAIT #180365864: nam='db file sequential read' ela= 11447 file#=9 block#=908 blocks=1 obj#=107151 tim=1785509403
FETCH #180365864:c=0,e=73927,p=8,cr=9,cu=0,mis=0,r=8,dep=0,og=1,plh=2683697726,tim=1785509674
STAT #180365864 id=1 cnt=9 pid=0 pos=1 obj=107151 op='TABLE ACCESS BY INDEX ROWID T (cr=12 pr=11 pw=0 time=94758 us cost=10 size=184 card=8)'
STAT #180365864 id=2 cnt=9 pid=1 pos=1 obj=107152 op='INDEX RANGE SCAN I_T_OBJECT_ID (cr=3 pr=2 pw=0 time=14819 us cost=2 size=0 card=8)'
*** 2018-03-17 19:56:06.875
WAIT #180365864: nam='SQL*Net message from client' ela= 3228967 driver id=1413697536 #bytes=1 p3=0 obj#=107151 tim=1788738943
CLOSE #180365864:c=0,e=11,dep=0,type=0,tim=1788739114
=====================
--//可以看到IO的等待事件全是'db file sequential read'.没有'db file parallel read'.
3.测试二:
--//使用NO_BATCH_TABLE_ACCESS_BY_ROWID提示.
SCOTT@test01p> @ sqlhint BATCH_TABLE_ACCESS_BY_ROWID
NAME SQL_FEATURE CLASS INVERSE TARGET_LEVEL PROPERTY VERSION VERSION_OUTLINE CON_ID
------------------------------ --------------- --------------------------- ------------------------------ ------------ -------- -------- --------------- ------
BATCH_TABLE_ACCESS_BY_ROWID QKSFM_EXECUTION BATCH_TABLE_ACCESS_BY_ROWID NO_BATCH_TABLE_ACCESS_BY_ROWID 4 272 12.1.0.1 12.1.0.1 0
NO_BATCH_TABLE_ACCESS_BY_ROWID QKSFM_EXECUTION BATCH_TABLE_ACCESS_BY_ROWID BATCH_TABLE_ACCESS_BY_ROWID 4 272 12.1.0.1 12.1.0.1 0
---//重新登录,取消参数隐含参数_optimizer_batch_table_access_by_rowid设置,执行如下:
SCOTT@test01p> select /*+ NO_BATCH_TABLE_ACCESS_BY_ROWID */ rowid ,owner from t where object_id between 1 and 10;
SCOTT@test01p> @ dpc '' outline
PLAN_TABLE_OUTPUT
-------------------------------------
SQL_ID 9c0n73wph7xax, child number 0
-------------------------------------
select /*+ NO_BATCH_TABLE_ACCESS_BY_ROWID */ rowid ,owner from t where
object_id between 1 and 10
Plan hash value: 2044526593
-----------------------------------------------------------------------------------------------------------------------------------------------
| Id | Operation | Name | Starts | E-Rows |E-Bytes| Cost (%CPU)| E-Time | A-Rows | A-Time | Buffers |
-----------------------------------------------------------------------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1 | | | 10 (100)| | 9 |00:00:00.01 | 12 |
| 1 | TABLE ACCESS BY INDEX ROWID BATCHED| T | 1 | 8 | 184 | 10 (0)| 00:00:01 | 9 |00:00:00.01 | 12 |
|* 2 | INDEX RANGE SCAN | I_T_OBJECT_ID | 1 | 8 | | 2 (0)| 00:00:01 | 9 |00:00:00.01 | 3 |
-----------------------------------------------------------------------------------------------------------------------------------------------
Query Block Name / Object Alias (identified by operation id):
-------------------------------------------------------------
1 - SEL$1 / T@SEL$1
2 - SEL$1 / T@SEL$1
Outline Data
-------------
/*+
BEGIN_OUTLINE_DATA
IGNORE_OPTIM_EMBEDDED_HINTS
OPTIMIZER_FEATURES_ENABLE('12.1.0.1')
DB_VERSION('12.1.0.1')
ALL_ROWS
OUTLINE_LEAF(@"SEL$1")
INDEX_RS_ASC(@"SEL$1" "T"@"SEL$1" ("T"."OBJECT_ID"))
BATCH_TABLE_ACCESS_BY_ROWID(@"SEL$1" "T"@"SEL$1")
END_OUTLINE_DATA
*/
Predicate Information (identified by operation id):
---------------------------------------------------
2 - access("OBJECT_ID">=1 AND "OBJECT_ID"<=10)
--//可以发现这样使用提示并不生效,实际上还是TABLE ACCESS BY INDEX ROWID BATCHED.不过看Outline Data就知道提示的方法,要写成如下:
select /*+ NO_BATCH_TABLE_ACCESS_BY_ROWID(@"SEL$1" "T"@"SEL$1") */ rowid ,owner from t where object_id between 1 and 10;
select /*+ NO_BATCH_TABLE_ACCESS_BY_ROWID(t) */ rowid ,owner from t where object_id between 1 and 10;
SCOTT@test01p> @ dpc '' ''
PLAN_TABLE_OUTPUT
-------------------------------------
SQL_ID 7y2913wgp3t1b, child number 0
-------------------------------------
select /*+ NO_BATCH_TABLE_ACCESS_BY_ROWID(t) */ rowid ,owner from t
where object_id between 1 and 10
Plan hash value: 2683697726
---------------------------------------------------------------------------------------------------------------------------------------
| Id | Operation | Name | Starts | E-Rows |E-Bytes| Cost (%CPU)| E-Time | A-Rows | A-Time | Buffers |
---------------------------------------------------------------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1 | | | 10 (100)| | 9 |00:00:00.01 | 12 |
| 1 | TABLE ACCESS BY INDEX ROWID| T | 1 | 8 | 184 | 10 (0)| 00:00:01 | 9 |00:00:00.01 | 12 |
|* 2 | INDEX RANGE SCAN | I_T_OBJECT_ID | 1 | 8 | | 2 (0)| 00:00:01 | 9 |00:00:00.01 | 3 |
---------------------------------------------------------------------------------------------------------------------------------------
Query Block Name / Object Alias (identified by operation id):
-------------------------------------------------------------
1 - SEL$1 / T@SEL$1
2 - SEL$1 / T@SEL$1
Predicate Information (identified by operation id):
---------------------------------------------------
2 - access("OBJECT_ID">=1 AND "OBJECT_ID"<=10)
--//执行计划变为TABLE ACCESS BY INDEX ROWID.
--//顺便更正我以前一个错误的观点,一直以为执行计划选择TABLE ACCESS BY INDEX ROWID BATCHED逻辑读会减少,实际上不会变化,
--//因为输出的结果集顺序没有发生变化.可以看到最后的逻辑读都是一样的12.
--//我以前错误的理解确定读取扫描的数据块,然后输出满足条件的结果.实际上还是按照索引的顺序读取键值以及rowid,再读取数据块.
--//TABLE ACCESS BY INDEX ROWID BATCHED仅仅变成了db file parallel read,而且读取的数据块是按照顺序读取,并且是不连续的块,
--//最大127块(当然数据块不在缓存的情况下).
--//在看看如下例子:
SCOTT@test01p> select /*+ index(t) */ rowid ,owner,object_id from t where object_id between 1 and 520;
..
PLAN_TABLE_OUTPUT
-------------------------------------
SQL_ID 53r6yurs7jnk8, child number 0
-------------------------------------
select /*+ index(t) */ rowid ,owner,object_id from t where object_id between 1 and 520
Plan hash value: 2044526593
-----------------------------------------------------------------------------------------------------------------------------------------------
| Id | Operation | Name | Starts | E-Rows |E-Bytes| Cost (%CPU)| E-Time | A-Rows | A-Time | Buffers |
-----------------------------------------------------------------------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1 | | | 437 (100)| | 474 |00:00:00.01 | 479 |
| 1 | TABLE ACCESS BY INDEX ROWID BATCHED| T | 1 | 435 | 10005 | 437 (0)| 00:00:01 | 474 |00:00:00.01 | 479 |
|* 2 | INDEX RANGE SCAN | I_T_OBJECT_ID | 1 | 435 | | 2 (0)| 00:00:01 | 474 |00:00:00.01 | 5 |
-----------------------------------------------------------------------------------------------------------------------------------------------
SCOTT@test01p> select /*+ index(t) NO_BATCH_TABLE_ACCESS_BY_ROWID(t) */ rowid ,owner,object_id from t where object_id between 1 and 520;
...
PLAN_TABLE_OUTPUT
-------------------------------------
SQL_ID 3kf824wz4rdk5, child number 0
-------------------------------------
select /*+ index(t) NO_BATCH_TABLE_ACCESS_BY_ROWID(t) */ rowid ,owner,object_id from t where object_id between 1 and 520
Plan hash value: 2683697726
---------------------------------------------------------------------------------------------------------------------------------------
| Id | Operation | Name | Starts | E-Rows |E-Bytes| Cost (%CPU)| E-Time | A-Rows | A-Time | Buffers |
---------------------------------------------------------------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1 | | | 437 (100)| | 474 |00:00:00.01 | 479 |
| 1 | TABLE ACCESS BY INDEX ROWID| T | 1 | 435 | 10005 | 437 (0)| 00:00:01 | 474 |00:00:00.01 | 479 |
|* 2 | INDEX RANGE SCAN | I_T_OBJECT_ID | 1 | 435 | | 2 (0)| 00:00:01 | 474 |00:00:00.01 | 5 |
---------------------------------------------------------------------------------------------------------------------------------------
--//可以发现逻辑读479.两种情况都是一样的.只有参数array以及索引的群集因子才可能改变逻辑读的数量.这里的数据太离散,改变array效果不会太大.
SCOTT@test01p> set array 300
SCOTT@test01p> select /*+ index(t) */ rowid ,owner,object_id from t where object_id between 1 and 520;
..
Plan hash value: 2044526593
-----------------------------------------------------------------------------------------------------------------------------------------------
| Id | Operation | Name | Starts | E-Rows |E-Bytes| Cost (%CPU)| E-Time | A-Rows | A-Time | Buffers |
-----------------------------------------------------------------------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1 | | | 437 (100)| | 474 |00:00:00.01 | 478 |
| 1 | TABLE ACCESS BY INDEX ROWID BATCHED| T | 1 | 435 | 10005 | 437 (0)| 00:00:01 | 474 |00:00:00.01 | 478 |
|* 2 | INDEX RANGE SCAN | I_T_OBJECT_ID | 1 | 435 | | 2 (0)| 00:00:01 | 474 |00:00:00.01 | 4 |
-----------------------------------------------------------------------------------------------------------------------------------------------
--//仅仅减少1个.而且在INDEX RANGE SCAN操作,减少到4个.
--//重新建立表按照object_id导入:
SCOTT@test01p> create table tx as select * from t order by object_id;
Table created.
SCOTT@test01p> create index i_tx_object_id on tx(object_id);
Index created.
--//分析略.
SCOTT@test01p> set array 200
SCOTT@test01p> show array
arraysize 200
SCOTT@test01p> select /*+ index(t) */ rowid ,owner,object_id from tx where object_id between 1 and 520;
Plan hash value: 475430699
------------------------------------------------------------------------------------------------------------------------------------------------
| Id | Operation | Name | Starts | E-Rows |E-Bytes| Cost (%CPU)| E-Time | A-Rows | A-Time | Buffers |
------------------------------------------------------------------------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1 | | | 10 (100)| | 474 |00:00:00.01 | 15 |
| 1 | TABLE ACCESS BY INDEX ROWID BATCHED| TX | 1 | 435 | 10005 | 10 (0)| 00:00:01 | 474 |00:00:00.01 | 15 |
|* 2 | INDEX RANGE SCAN | I_TX_OBJECT_ID | 1 | 435 | | 2 (0)| 00:00:01 | 474 |00:00:00.01 | 5 |
------------------------------------------------------------------------------------------------------------------------------------------------
标签:
版权申明:本站文章部分自网络,如有侵权,请联系:west999com@outlook.com
特别注意:本站所有转载文章言论不代表本站观点,本站所提供的摄影照片,插画,设计作品,如需使用,请与原作者联系,版权归原作者所有
上一篇:Redis安装使用
下一篇:GBase数据库——常用命令
- Python编程Message: CGI script is not executable ( 2019-03-01
- SQLAlchemy Table(表)类方式 - Table类和Column类 2018-10-26
- DataTable与List<T>相互转换 2018-06-17
- DataSet & DataTable &DataRow 深入浅出 2018-06-17
- datatable 转excel 2018-06-17
IDC资讯: 主机资讯 注册资讯 托管资讯 vps资讯 网站建设
网站运营: 建站经验 策划盈利 搜索优化 网站推广 免费资源
网络编程: Asp.Net编程 Asp编程 Php编程 Xml编程 Access Mssql Mysql 其它
服务器技术: Web服务器 Ftp服务器 Mail服务器 Dns服务器 安全防护
软件技巧: 其它软件 Word Excel Powerpoint Ghost Vista QQ空间 QQ FlashGet 迅雷
网页制作: FrontPages Dreamweaver Javascript css photoshop fireworks Flash