Saturday, January 10, 2009

深入探讨Oracle数据库10g的Shrink机制

http://haoxiai.net/shujuku/Oracle/111104.html
从10g开始,oracle开始提供Shrink的命令,假如我们的表空间中支持自动段空间管理 (ASSM),就可以使用这个特性缩小段,即降低HWM。这里需要强调一点,10g的这个新特性,仅对ASSM表空间有效,否则会报 ORA-10635: Invalid segment or tablespace type。
在这里,我们来讨论如和对一个ASSM的segment回收浪费的空间。
同样,我们用系统视图all_objects来在tablespace ASSM上创建测试表my_objects,这一小节的内容,实验环境为oracle10.1.0.2:
SQL> select * from v$version;

BANNER
----------------------------------------------------------------
Oracle Database 10g Enterprise Edition Release 10.1.0.2.0 - Prod
PL/SQL Release 10.1.0.2.0 - Production
CORE 10.1.0.2.0 Production

TNS for 32-bit Windows: Version 10.1.0.2.0 - Production
NLSRTL Version 10.1.0.2.0 – Production

SQL> select TABLESPACE_NAME,BLOCK_SIZE,EXTENT_MANAGEMENT,
2 ALLOCATION_TYPE, SEGMENT_SPACE_MANAGEMENT
3 from dba_tablespaces where TABLESPACE_NAME = 'ASSM';

TABLESPACE_NAME BLOCK_SIZE EXTENT_MANAGEMENT ALLOCATION_TYPE SEGMENT_SPACE_MANAGEMENT
---------------- ---------- ----------------- --------------- ------------------------
ASSM 8192 LOCAL UNIFORM AUTO

SQL> create table my_objects tablespace assm
2 as select * from all_objects;
Table created


然后我们随机地从table MY_OBJECTS中删除一部分数据:
SQL> select count(*) from my_objects;
COUNT(*)
----------
47828
SQL> delete from my_objects where object_name like '%C%';
16950 rows deleted

SQL> delete from my_objects where object_name like '%U%';
4503 rows deleted

SQL> delete from my_objects where object_name like '%A%';
6739 rows deleted


现在我们使用show_space和show_space_assm来看看my_objects的数据存储状况:
SQL> exec show_space('MY_OBJECTS','DLINGER');
Total Blocks............................680
Total Bytes.............................5570560
Unused Blocks...........................1
Unused Bytes............................8192
Last Used Ext FileId....................6
Last Used Ext BlockId...................793
Last Used Block.........................4

PL/SQL 过程已成功完成。

SQL> exec show_space_assm('MY_OBJECTS','DLINGER');
free space 0-25% Blocks:................0
free space 25-50% Blocks:...............205
free space 50-75% Blocks:...............180
free space 75-100% Blocks:..............229
Full Blocks:............................45
Unformatted blocks:.....................0

PL/SQL 过程已成功完成。


这里,table my_objects的HWM下有679个block,其中,free space为25-50%的block有205个,free space为50-75%的block有180个,free space为75-100%的block有229个,full space的block只有45个,这种情况下,我们需要对这个table的现有数据行进行重组。
要使用assm上的shink,首先我们需要使该表支持行移动,可以用这样的命令来完成:
alter table my_objects enable row movement;
现在,就可以来降低my_objects的HWM,回收空间了,使用命令:
alter table bookings shrink space;
我们具体的看一下实验的结果:
SQL> alter table my_objects enable row movement;
表已更改。

SQL> alter table my_objects shrink space;
表已更改。

SQL> exec show_space('MY_OBJECTS','DLINGER');
Total Blocks............................265
Total Bytes.............................2170880
Unused Blocks...........................2
Unused Bytes............................16384
Last Used Ext FileId....................6
Last Used Ext BlockId...................308
Last Used Block.........................3

PL/SQL 过程已成功完成。

SQL> exec show_space_assm('MY_OBJECTS','DLINGER');
free space 0-25% Blocks:................0
free space 25-50% Blocks:...............1
free space 50-75% Blocks:...............0
free space 75-100% Blocks:..............0
Full Blocks:............................249
Unformatted blocks:.....................0

PL/SQL 过程已成功完成。


在执行玩shrink命令后,我们可以看到,table my_objects的HWM现在降到了264的位置,而且HWM下的block的空间使用状况,full space的block有249个,free space 为25-50% Block只有1个。

我们接下来讨论一下shrink的实现机制,我们同样使用讨论move机制的那个实验来观察。
SQL> create table TEST_HWM (id int ,name char(2000)) tablespace ASSM;

Table created


往table test_hwm中插入如下的数据:
insert into TEST_HWM values (1,'aa');
insert into TEST_HWM values (2,'bb');
insert into TEST_HWM values (2,'cc');
insert into TEST_HWM values (3,'dd');
insert into TEST_HWM values (4,'ds');
insert into TEST_HWM values (5,'dss');
insert into TEST_HWM values (6,'dss');
insert into TEST_HWM values (7,'ess');
insert into TEST_HWM values (8,'es');
insert into TEST_HWM values (9,'es');
insert into TEST_HWM values (10,'es');

我们来看看这个table的rowid和block的ID和信息:
SQL> select rowid , id,name from TEST_HWM;

ROWID ID NAME
------------------ ---------- ----- ---------
AAANhqAAGAAAAFHAAA 1 aa
AAANhqAAGAAAAFHAAB 2 bb
AAANhqAAGAAAAFHAAC 2 cc
AAANhqAAGAAAAFIAAA 3 dd
AAANhqAAGAAAAFIAAB 4 ds
AAANhqAAGAAAAFIAAC 5 dss
AAANhqAAGAAAAFJAAA 6 dss
AAANhqAAGAAAAFJAAB 7 ess
AAANhqAAGAAAAFJAAC 8 es
AAANhqAAGAAAAFKAAA 9 es
AAANhqAAGAAAAFKAAB 10 es

11 rows selected

SQL> select EXTENT_ID,FILE_ID,RELATIVE_FNO,BLOCK_ID,BLOCKS
2 from dba_extents where segment_name='TEST_HWM' ;

EXTENT_ID FILE_ID RELATIVE_FNO BLOCK_ID BLOCKS
---------- ---------- ------------ ---------- ----------
0 6 6 324 5
1 6 6 329 5


然后从table test_hwm中删除一些数据:
delete from TEST_HWM where id = 2;
delete from TEST_HWM where id = 4;
delete from TEST_HWM where id = 3;
delete from TEST_HWM where id = 7;
delete from TEST_HWM where id = 8;

观察table test_hwm的rowid和blockid的信息:
SQL> select rowid , id,name from TEST_HWM;

ROWID ID NAME
------------------ ---------- ----- --------
AAANhqAAGAAAAFHAAA 1 aa
AAANhqAAGAAAAFIAAC 5 dss
AAANhqAAGAAAAFJAAA 6 dss
AAANhqAAGAAAAFKAAA 9 es
AAANhqAAGAAAAFKAAB 10 es

SQL> select EXTENT_ID,FILE_ID,RELATIVE_FNO,BLOCK_ID,BLOCKS
2 from dba_extents where segment_name='TEST_HWM' ;

EXTENT_ID FILE_ID RELATIVE_FNO BLOCK_ID BLOCKS
---------- ---------- ------------ ---------- ----------
0 6 6 324 5
1 6 6 329 5


从以上的信息,我们可以看到,在table test_hwm中,剩下的数据是分布在AAAAFH,AAAAFI,AAAAFJ,AAAAFK这样四个连续的block中。

SQL> exec show_space_assm('TEST_HWM','DLINGER');
free space 0-25% Blocks:................0
free space 25-50% Blocks:...............1
free space 50-75% Blocks:...............3
free space 75-100% Blocks:..............3
Full Blocks:............................0
Unformatted blocks:.....................0


通过show_space_assm我们可以看到目前这四个block的空间使用状况,AAAAFH,AAAAFI,AAAAFJ上各有一行数据,我们猜测free space为50-75%的3个block是这三个block,那么free space为25-50%的1个block就是AAAAFK了,剩下free space为 75-100% 的3个block,是HWM下已格式化的尚未使用的block。(关于assm下hwm的移动我们前面已经详细地讨论过了,在extent不大于于16个 block时,是以一个extent为单位来移动的)

然后,我们对table my_objects执行shtink的操作:
SQL> alter table test_hwm enable row movement;

Table altered

SQL> alter table test_hwm shrink space;

Table altered

SQL> select rowid ,id,name from TEST_HWM;

ROWID ID NAME
------------------ ---------- ------ -----------
AAANhqAAGAAAAFHAAA 1 aa
AAANhqAAGAAAAFHAAB 10 es
AAANhqAAGAAAAFHAAD 9 es
AAANhqAAGAAAAFIAAC 5 dss
AAANhqAAGAAAAFJAAA 6 dss

SQL> select EXTENT_ID,FILE_ID,RELATIVE_FNO,BLOCK_ID,BLOCKS
2 from dba_extents where segment_name='TEST_HWM' ;

EXTENT_ID FILE_ID RELATIVE_FNO BLOCK_ID BLOCKS
---------- ---------- ------------ ---------- ----------
0 6 6 324 5
1 6 6 329 5


当执行了shrink操作后,有意思的现象出现了。我们来看看oracle是如何移动行数据的,这里的情况和move已经不太一样了。我们知道,在 move操作的时候,所有行的rowid都发生了变化,table所位于的block的区域也发生了变化,但是所有行物理存储的顺序都没有发生变化,所以我们得到的结论是,oracle以block为单位,进行了block间的数据copy。那么shrink后,我们发现,部分行数据的rowid发生了变化,同时,部分行数据的物理存储的顺序也发生了变化,而table所位于的block的区域却没有变化,这就说明,shrink只移动了table其中一部分的行数据,来完成释放空间,而且,这个过程是在table当前所使用的block中完成的。

那么Oracle具体移动行数据的过程是怎样的呢?我们根据这样的实验结果,可以来猜测一下:
Oracle是以行为单位来移动数据的。Oracle从当前table存储的最后一行数据开始移动,从当前table最先使用的block 开始搜索空间,所以,shrink之前,rownum=10的那行数据(10,es),被移动到block AAAAFH上,写到(1,aa)这行数据的后面,所以(10,es)的rownum和rowid同时发生改变。然后是(9,es)这行数据,重复上述过程。这是oracle从后向前移动行数据的大致遵循的规则,那么具体移动行数据的的算法是比较复杂的,包括向ASSM的table中insert数据使用 block的顺序的算法也是比较复杂的,大家有兴趣的可以自己来研究,在这里我们不多做讨论。

我们还可以在shrink table的同时shrink这个table上的index:
alter table my_objects shrink space cascade;
同样地,这个操作只有当table上的index也是ASSM时,才能使用。

关于日志的问题,我们对比了同样数据量和分布状况的两张table,在move和shrink下生成的redo size(table上没有index的情况下):
SQL> select tablespace_name,SEGMENT_SPACE_MANAGEMENT from dba_tablespaces
2 where tablespace_name in('ASSM','HWM');

TABLESPACE_NAME SEGMENT_SPACE_MANAGEMENT
------------------------------ ------------------------
ASSM AUTO
HWM MANUAL
SQL> create table my_objects tablespace ASSM as select * from all_objects where rownum<20000;
Table created

SQL> create table my_objects1 tablespace HWM as select * from all_objects where rownum<20000;
Table created

SQL> select bytes/1024/1024 from user_segments where segment_name = 'MY_OBJECTS';
BYTES/1024/1024
---------------
2.1875

SQL> delete from my_objects where object_name like '%C%';
7278 rows deleted
SQL> delete from my_objects1 where object_name like '%C%';
7278 rows deleted
SQL> delete from my_objects where object_name like '%U%';
2732 rows deleted
SQL> delete from my_objects1 where object_name like '%U%';
2732 rows deleted
SQL> commit;
Commit complete
SQL> alter table my_objects enable row movement;
Table altered

SQL> select value from v$mystat, v$statname
2 where v$mystat.statistic# = v$statname.statistic#
3 and v$statname.name = 'redo size';
VALUE
----------
27808792

SQL> alter table my_objects shrink space;
Table altered

SQL> select value from v$mystat, v$statname
2 where v$mystat.statistic# = v$statname.statistic#
3 and v$statname.name = 'redo size';
VALUE
----------
32579712

SQL> alter table my_objects1 move;
Table altered

SQL> select value from v$mystat, v$statname
2 where v$mystat.statistic# = v$statname.statistic#
3 and v$statname.name = 'redo size';
VALUE
----------
32676784

对于table my_objects,进行shrink,产生了32579712 – 27808792=4770920,约4.5M的redo ;对table my_objects1进行move,产生了32676784-32579712= 97072,约95K的redo size。那么,与move比较起来,shrink的日志写要大得多。


Shrink的几点问题:
a. shrink后index是否需要rebuild:
因为shrink的操作也会改变行数据的rowid,那么,如果table上有index时,shrink table后index会不会变为UNUSABLE呢?我们来看这样的实验,同样构建my_objects的测试表:
create table my_objects tablespace ASSM as select * from all_objects where rownum<20000;
create index i_my_objects on my_objects (object_id);
delete from my_objects where object_name like '%C%';
delete from my_objects where object_name like '%U%';
现在我们来shrink table my_objects:
SQL> alter table my_objects enable row movement;

Table altered

SQL> alter table my_objects shrink space;

Table altered
SQL> select index_name,status from user_indexes where index_name='I_MY_OBJECTS';

INDEX_NAME STATUS
------------------------------ --------
I_MY_OBJECTS VALID

我们发现,table my_objects上的index的状态为VALID,估计shrink在移动行数据时,也一起维护了index上相应行的数据rowid的信息。我们认为,这是对于move操作后需要rebuild index的改进。但是如果一个table上的index数量较多,我们知道,维护index的成本是比较高的,shrink过程中用来维护index的成本也会比较高。

b. shrink时对table的lock
在对table进行shrink时,会对table进行怎样的锁定呢?当我们对table MY_OBJECTS进行shrink操作时,查询v$locked_objects视图可以发现,table MY_OBJECTS上加了row-X (SX) 的lock:
SQL>select OBJECT_ID, SESSION_ID,ORACLE_USERNAME,LOCKED_MODE from v$locked_objects;

OBJECT_ID SESSION_ID ORACLE_USERNAME LOCKED_MODE
---------- ---------- ------------------ -----------
55422 153 DLINGER 3
SQL> select object_id from user_objects where object_name = 'MY_OBJECTS';

OBJECT_ID
----------
55422

那么,当table在进行shrink时,我们对table是可以进行DML操作的。

c. shrink对空间的要求
我们在前面讨论了shrink的数据的移动机制,既然oracle是从后向前移动行数据,那么,shrink的操作就不会像move一样,shrink不需要使用额外的空闲空间。

No comments: