转载地址:http://blog.itpub.net/17203031/viewspace-1473163/
3、使用10046诊断rebuild动作
10046诊断事件是我们经常用来使用跟踪事件,也是我们分析Oracle内部行为的最常用工具。下面笔者将用这个工具对rebuild动作进行检查。
首先获取一下数据表T和索引IDX_T_ID的分区结构。
--数据表T
SQL> select EXTENT_ID, FILE_ID, BLOCK_ID, BYTES, BLOCKS from dba_extents where owner='SYS' and segment_name='T';
EXTENT_ID FILE_ID BLOCK_ID BYTES BLOCKS
---------- ---------- ---------- ---------- ----------
0 1 31072 65536 8 –段首
1 1 31096 65536 8
2 1 99328 65536 8
3 1 99360 65536 8
4 1 99368 65536 8
5 1 99376 65536 8
6 1 99384 65536 8
7 1 99392 65536 8
8 1 99400 65536 8
9 1 99408 65536 8
10 1 99416 65536 8
11 1 99424 65536 8
12 1 99432 65536 8
13 1 99440 65536 8
14 1 99448 65536 8
15 1 100736 65536 8
16 1 99456 1048576 128
17 1 99584 1048576 128
18 1 99712 1048576 128
19 1 99840 1048576 128
EXTENT_ID FILE_ID BLOCK_ID BYTES BLOCKS
---------- ---------- ---------- ---------- ----------
20 1 99968 1048576 128
21 1 100096 1048576 128
22 1 100224 1048576 128
23 1 100352 1048576 128
24 1 100480 1048576 128
25 rows selected
--分区情况
SQL> select EXTENT_ID, FILE_ID, BLOCK_ID, BYTES, BLOCKS from dba_extents where owner='SYS' and segment_name='IDX_T_ID';
EXTENT_ID FILE_ID BLOCK_ID BYTES BLOCKS
---------- ---------- ---------- ---------- ----------
0 1 100760 65536 8 –段首
1 1 100768 65536 8
2 1 100776 65536 8
3 1 100784 65536 8
4 1 100792 65536 8
5 1 100800 65536 8
6 1 100808 65536 8
7 1 100816 65536 8
8 1 100824 65536 8
9 1 100832 65536 8
10 1 100840 65536 8
11 1 100848 65536 8
12 1 100856 65536 8
13 1 100608 65536 8
14 1 100616 65536 8
15 1 100624 65536 8
16 1 100864 1048576 128
17 rows selected
了解数据表和索引段结构之后,就可以从Trace文件中分析Oracle的读取写入动作。下面执行跟踪过程。
--清理内存
SQL> alter system flush shared_pool;
System altered
SQL> alter system flush buffer_cache;
System altered
跟踪过程:
SQL> select value from v$diag_info where name='Default Trace File';
VALUE
--------------------------------------------------------------------------------
/u01/app/diag/rdbms/ora11g/ora11g/trace/ora11g_ora_1962.trc
SQL> alter system flush shared_pool;
System altered
SQL> alter system flush buffer_cache;
System altered
SQL> alter session set events '10046 trace name context forever, level 12';
会话已更改。
SQL> alter index idx_t_id rebuild;
索引已更改。
SQL> alter session set events '10046 trace name context off';
会话已更改。
从Trace文件中,我们可以发现很多的SQL语句和执行过程。一个Oracle SQL语句的执行,往往伴随着很多的recursive call调用过程。详细研究可以帮助我们理解内部运行机理。篇幅有限,本次之研究与alert index … rebuild相关语句和游标记录。
首先找到了rebuild记录游标。
=====================
PARSING IN CURSOR #3075237312 len=28 dep=0 uid=0 oct=9 lid=0 tim=1427116687152197 hv=411325523 ad='35dc51e8' sqlid='1w5dx14c88p2m'
alter index idx_t_id rebuild
END OF STMT
PARSE #3075237312:c=147978,e=169400,p=26,cr=312,cu=0,mis=1,r=0,dep=0,og=1,plh=1483129259,tim=1427116687152192
Alter index …. Rebuild语句被解析parse为编号:3075237312。之后Trace文件中包括与这个编号有关的记录如下:
首先记录的是从对象中大量读取数据的过程:
WAIT #3075237312: nam='db file sequential read' ela= 148 file#=1 block#=100760 blocks=1 obj#=87690 tim=1427116687641730
WAIT #3075237312: nam='db file scattered read' ela= 155 file#=1 block#=100761 blocks=7 obj#=87690 tim=1427116687642009
WAIT #3075237312: nam='db file scattered read' ela= 339 file#=1 block#=100768 blocks=8 obj#=87690 tim=1427116687646571
WAIT #3075237312: nam='db file scattered read' ela= 176 file#=1 block#=100776 blocks=8 obj#=87690 tim=1427116687650926
WAIT #3075237312: nam='db file scattered read' ela= 204 file#=1 block#=100784 blocks=8 obj#=87690 tim=1427116687655912
WAIT #3075237312: nam='db file scattered read' ela= 197 file#=1 block#=100792 blocks=8 obj#=87690 tim=1427116687660075
WAIT #3075237312: nam='db file scattered read' ela= 207 file#=1 block#=100800 blocks=8 obj#=87690 tim=1427116687664669
WAIT #3075237312: nam='db file scattered read' ela= 203 file#=1 block#=100808 blocks=8 obj#=87690 tim=1427116687669300
WAIT #3075237312: nam='db file scattered read' ela= 220 file#=1 block#=100816 blocks=8 obj#=87690 tim=1427116687674227
WAIT #3075237312: nam='db file scattered read' ela= 162 file#=1 block#=100824 blocks=8 obj#=87690 tim=1427116687679009
WAIT #3075237312: nam='db file scattered read' ela= 210 file#=1 block#=100832 blocks=8 obj#=87690 tim=1427116687683670
*** 2015-03-23 21:18:07.688
WAIT #3075237312: nam='db file scattered read' ela= 196 file#=1 block#=100840 blocks=8 obj#=87690 tim=1427116687688942
WAIT #3075237312: nam='db file scattered read' ela= 456 file#=1 block#=100848 blocks=8 obj#=87690 tim=1427116687694629
WAIT #3075237312: nam='db file scattered read' ela= 248 file#=1 block#=100856 blocks=8 obj#=87690 tim=1427116687699340
WAIT #3075237312: nam='db file scattered read' ela= 318 file#=1 block#=100608 blocks=8 obj#=87690 tim=1427116687704678
WAIT #3075237312: nam='db file scattered read' ela= 336 file#=1 block#=100616 blocks=8 obj#=87690 tim=1427116687709104
WAIT #3075237312: nam='db file scattered read' ela= 216 file#=1 block#=100624 blocks=8 obj#=87690 tim=1427116687713798
WAIT #3075237312: nam='db file scattered read' ela= 3976 file#=1 block#=100864 blocks=32 obj#=87690 tim=1427116687724032
WAIT #3075237312: nam='db file scattered read' ela= 769 file#=1 block#=100896 blocks=32 obj#=87690 tim=1427116687749159
WAIT #3075237312: nam='db file sequential read' ela= 165 file#=1 block#=100928 blocks=1 obj#=87690 tim=1427116687771987
WAIT #3075237312: nam='Disk file operations I/O' ela= 41 FileOperation=2 fileno=3 filetype=2 obj#=0 tim=1427116687775771
WAIT #3075237312: nam='db file sequential read' ela= 170 file#=3 block#=3044 blocks=1 obj#=0 tim=1427116687776293
WAIT #3075237312: nam='db file sequential read' ela= 178 file#=1 block#=2 blocks=1 obj#=1 tim=1427116687778458
WAIT #3075237312: nam='db file sequential read' ela= 134 file#=1 block#=3 blocks=1 obj#=1 tim=1427116687778761
第一句的sequential单块读动作,读取的是file#=1 block#=100760 blocks=1 obj#=87690。参考之前分区结构,恰好是索引IDX_T_ID段的段头块结构。
SQL> select EXTENT_ID, FILE_ID, BLOCK_ID, BYTES, BLOCKS from dba_extents where owner='SYS' and segment_name='IDX_T_ID';
EXTENT_ID FILE_ID BLOCK_ID BYTES BLOCKS
---------- ---------- ---------- ---------- ----------
0 1 100760 65536 8
……
之后一系列的“db file scattered read”多块读动作,持续的“吞”掉索引的结构块。这些特征完全符合index fast full scan工作特点。
之后,就将处理数据写入结构:
WAIT #3075237312: nam='direct path write' ela= 17189 file number=1 first dba=100636 block cnt=4 obj#=0 tim=1427116687900949
WAIT #3075237312: nam='direct path write' ela= 3769 file number=1 first dba=100640 block cnt=4 obj#=0 tim=1427116687907864
WAIT #3075237312: nam='direct path write' ela= 26675 file number=1 first dba=100644 block cnt=4 obj#=0 tim=1427116687939138
WAIT #3075237312: nam='direct path write' ela= 2119 file number=1 first dba=100648 block cnt=4 obj#=0 tim=1427116687944101
WAIT #3075237312: nam='direct path write' ela= 5425 file number=1 first dba=100652 block cnt=4 obj#=0 tim=1427116687953153
(篇幅原因,有省略……)
WAIT #3075237312: nam='direct path write' ela= 6109 file number=1 first dba=100692 block cnt=4 obj#=0 tim=1427116688030198
这部分动作是rebuild的后续动作,写入数据内容就是新索引IDX_T_ID段结构。这点从执行后新段结构信息可以证明。
SQL> select EXTENT_ID, FILE_ID, BLOCK_ID, BYTES, BLOCKS from dba_extents where owner='SYS' and segment_name='IDX_T_ID';
EXTENT_ID FILE_ID BLOCK_ID BYTES BLOCKS
---------- ---------- ---------- ---------- ----------
0 1 100632 65536 8
1 1 100640 65536 8
2 1 100648 65536 8
3 1 100656 65536 8
4 1 100664 65536 8
5 1 100672 65536 8
6 1 100680 65536 8
7 1 100688 65536 8
8 1 100696 65536 8
9 1 100704 65536 8
10 1 100712 65536 8
11 1 100720 65536 8
12 1 100728 65536 8
13 1 100992 65536 8
14 1 101000 65536 8
15 1 101008 65536 8
16 1 101120 1048576 128
17 rows selected
这系列也就证明了rebuild操作是基于原有索引结构数据,重新构建出索引段结构。
4、使用10046诊断rebuild online动作
下面测试一下rebuild online动作。为了保证实验质量,清理一下内存缓存。
SQL> alter system flush shared_pool;
System altered
SQL> alter system flush buffer_cache;
System altered
开启跟踪事件。
SQL> select value from v$diag_info where name='Default Trace File';
VALUE
-------------------------------------------------------------------------
/u01/app/diag/rdbms/ora11g/ora11g/trace/ora11g_ora_4149.trc
SQL> alter session set events '10046 trace name context forever, level 12';
会话已更改。
SQL> alter index idx_t_id rebuild online;
索引已更改。
SQL> alter session set events '10046 trace name context off';
会话已更改。
在跟踪文件中,找到对应的rebuild online游标信息。
=====================
PARSING IN CURSOR #3075319188 len=35 dep=0 uid=0 oct=9 lid=0 tim=1427116924266395 hv=572453287 ad='35d443ac' sqlid='6bvhyk0j1xwd7'
alter index idx_t_id rebuild online
END OF STMT
PARSE #3075319188:c=92986,e=109127,p=30,cr=236,cu=0,mis=1,r=0,dep=0,og=1,plh=1193657316,tim=1427116924266390
对rebuild online语句,游标编号:#3075319188。在文件中找到对应游标的记录。
WAIT #3075319188: nam='db file sequential read' ela= 199 file#=3 block#=192 blocks=1 obj#=0 tim=1427116926331398
WAIT #3075319188: nam='db file sequential read' ela= 146 file#=3 block#=6548 blocks=1 obj#=0 tim=1427116926331706
WAIT #3075319188: nam='db file sequential read' ela= 147 file#=1 block#=31072 blocks=1 obj#=87689 tim=1427116926332503
WAIT #3075319188: nam='db file sequential read' ela= 166 file#=1 block#=31072 blocks=1 obj#=87689 tim=1427116926332741
WAIT #3075319188: nam='db file scattered read' ela= 315 file#=1 block#=31073 blocks=7 obj#=87689 tim=1427116926333268
(篇幅原因,有省略……)
WAIT #3075319188: nam='db file scattered read' ela= 961 file#=1 block#=100511 blocks=32 obj#=87689 tim=1427116926441946
WAIT #3075319188: nam='db file scattered read' ela= 392 file#=1 block#=100543 blocks=15 obj#=87689 tim=1427116926444785
首先我们看一下比较奇怪的对file#=3的读动作,单块读动作可以找到对应段的情况。
SQL> select * from dba_extents where FILE_ID=3 and block_id<=192 and block_id+blocks-1>=192;
SEGMENT_NAME SEGMENT_TYPE TABLESPACE_NAME EXTENT_ID FILE_ID BLOCK_ID BYTES BLOCKS RELATIVE_FNO
-------------------------------------------------------------------------------- ------------------------------ ------------------ -------------
_SYSSMU5_3994777876$ TYPE2 UNDO UNDOTBS1 0 3 192 65536 8 3
(结果有删减)
SQL> select * from dba_extents where FILE_ID=3 and block_id<=6548 and block_id+blocks-1>=6548;
SEGMENT_NAME SEGMENT_TYPE TABLESPACE_NAME EXTENT_ID FILE_ID BLOCK_ID BYTES BLOCKS RELATIVE_FNO
------------------------ ------------------ ------------------------------ ---------- ---------- ---------- ---------- ---------- ------------
_SYSSMU5_3994777876$ TYPE2 UNDO UNDOTBS1 10 3 6528 1048576 128 3
这两个部分内容是Undo段中数据。说明在读过程,有访问Undo前镜像的情景。其他的数据内容为数据表段读取:
SQL> select EXTENT_ID, FILE_ID, BLOCK_ID, BYTES, BLOCKS from dba_extents where owner='SYS' and segment_name='T';
EXTENT_ID FILE_ID BLOCK_ID BYTES BLOCKS
---------- ---------- ---------- ---------- ----------
0 1 31072 65536 8
1 1 31096 65536 8
2 1 99328 65536 8
之后是写入动作:
WAIT #3075319188: nam='direct path write' ela= 104461 file number=1 first dba=100618 block cnt=2 obj#=87689 tim=1427116926569710
WAIT #3075319188: nam='direct path write' ela= 13504 file number=1 first dba=100620 block cnt=4 obj#=87689 tim=1427116926604072
(篇幅原因,有省略……)
WAIT #3075319188: nam='direct path write' ela= 2319 file number=1 first dba=100928 block cnt=1 obj#=87689 tim=1427116928413458
从Trace文件结果看,rebuild online过程是直接对数据表的访问,将数据读取后进行索引化过程。
5、结论
索引rebuild是一个我们经常遇到的操作过程,详细理解rebuild和rebuild online可以帮助在实际工作中强化分析能力,更好解决问题。