undo机制是oracle的巧妙之处,也是学习oracle的难点之一。
下面我们从内部来具体了解它。
会话1:scott用户,发起一个事务
hr@ORCL> conn scott/oracle Connected. scott@ORCL> UPDATE emp SET sal=4000 WHERE empno=7788; 1 row updated. scott@ORCL> SELECT empno,ename,job,mgr,sal FROM emp WHERE empno=7788; EMPNO ENAME JOB MGR SAL ---------- ---------- --------- ---------- ---------- 7788 SCOTT ANALYST 7566 4000
先不提交这个事务,会话2:SYS用户
获得事务信息
从v$transaction,可得:该事务在9号回滚段,位于26号事务槽
sys@ORCL> SELECT xidusn,xidslot FROM v$transaction; XIDUSN XIDSLOT ---------- ---------- 9 26
从v$rollstat,也可见该事务高就9号回滚段
sys@ORCL> SELECT usn,xacts FROM v$rollstat; USN XACTS ---------- ---------- 0 0 1 0 2 0 3 0 4 0 5 0 6 0 7 0 8 0 9 1 10 0
dump回滚段的段头信息
sys@ORCL> SELECT * FROM v$rollname a WHERE a.usn=9; USN NAME ---------- ------------------------------ 9 _SYSSMU9$ sys@ORCL> alter system dump undo header '_SYSSMU9$'; System altered. sys@ORCL> select spid from v$process where addr in (select paddr from v$session where sid in (select sid 2 from v$mystat where rownum=1)); SPID ------------ 10086
部分内容摘入如下:
******************************************************************************** Undo Segment: _SYSSMU9$ (9) ******************************************************************************** Extent Control Header ----------------------------------------------------------------- Extent Header:: spare1: 0 spare2: 0 #extents: 4 #blocks: 31 last map 0x00000000 #maps: 0 offset: 4080 Highwater:: 0x00800615 ext#: 2 blk#: 4 ext size: 8 #blocks in seg. hdr's freelists: 0 #blocks below: 0 mapblk 0x00000000 offset: 2 Unlocked Map Header:: next 0x00000000 #extents: 4 obj#: 0 flag: 0x40000000 Extent Map ----------------------------------------------------------------- 0x0080008a length: 7 0x008000d9 length: 8 0x00800611 length: 8 0x00800669 length: 8 Retention Table ----------------------------------------------------------- Extent Number:0 Commit Time: 1344269736 Extent Number:1 Commit Time: 1344276385 Extent Number:2 Commit Time: 1344264118 Extent Number:3 Commit Time: 1344264252 TRN CTL:: seq: 0x0105 chd: 0x0019 ctl: 0x0014 inc: 0x00000000 nfb: 0x0000 mgc: 0x8201 xts: 0x0068 flg: 0x0001 opt: 2147483646 (0x7ffffffe) uba: 0x00800615.0105.11 scn: 0x0000.000d2615 Version: 0x01 FREE BLOCK POOL:: uba: 0x00000000.0105.10 ext: 0x2 spc: 0x16f6 uba: 0x00000000.0105.03 ext: 0x2 spc: 0x1eac uba: 0x00000000.0101.0b ext: 0xb spc: 0x198c uba: 0x00000000.0000.00 ext: 0x0 spc: 0x0 uba: 0x00000000.0000.00 ext: 0x0 spc: 0x0 TRN TBL:: index state cflags wrap# uel scn dba parent-xid nub stmt_num cmt ------------------------------------------------------------------------------------------------ 0x00 9 0x00 0x0162 0x0029 0x0000.000d2ab9 0x00800613 0x0000.000.00000000 0x00000001 0x00000000 1344280244 ............................................ 0x1a 10 0x80 0x0162 0x0002 0x0000.00000000 0x00800615 0x0000.000.00000000 0x00000001 0x00000000 0 ............................................
这就是神秘的回滚段头的信息,其中包括事务表信息。其中,0x1a是个16进制数,转成10进制:
scott@ORCL> select to_number('1a','xx') from dual; TO_NUMBER('1A','XX') -------------------- 26
由此,可知:0x1a为26.回顾前面的事务信息,该事务正好占用的是第26号事务槽,状态(state)为10代表的是活动事务。
dump旧值信息
段头信息的第七列:dba(data block address),指的就是回滚块的地址0x00800615
下面把dba翻译一下
sys@ORCL> select to_number('00800615','xxxxxxxx') from dual; TO_NUMBER('00800615','XXXXXXXX') -------------------------------- 8390165 sys@ORCL> select dbms_utility.data_block_address_file(8390165) from dual; DBMS_UTILITY.DATA_BLOCK_ADDRESS_FILE(8390165) --------------------------------------------- 2 sys@ORCL> select dbms_utility.data_block_address_block(8390165) from dual; DBMS_UTILITY.DATA_BLOCK_ADDRESS_BLOCK(8390165) ---------------------------------------------- 1557
由此,可知:旧值在2号文件的第1557个回滚块上
这与从v$transaction查询得到的数据完全一致:
sys@ORCL> select ubafil,ubablk from v$transaction; UBAFIL UBABLK ---------- ---------- 2 1557
回到会话1,继续DML事务:scott用户
scott@ORCL> update emp set sal=4000 where empno=7788; 1 row updated. scott@ORCL> update emp set sal=4000 where empno=7782; 1 row updated. scott@ORCL> update emp set sal=40000 where empno=7698; 1 row updated.
会话2:sys用户
把回滚块1557dump出来
sys@ORCL> select spid from v$process where addr in (select paddr from v$session where sid in (select sid from v$mystat where rownum=1)); SPID ------------ 11146 sys@ORCL> alter system dump datafile 2 block 1557; System altered.
回滚块1557里面开头的信息部分如下:
*** 2012-08-07 16:31:42.449 *** SERVICE NAME:(SYS$USERS) 2012-08-07 16:31:42.355 *** SESSION ID:(135.102) 2012-08-07 16:31:42.355 Start dump data blocks tsn: 1 file#: 2 minblk 1557 maxblk 1557 buffer tsn: 1 rdba: 0x00800615 (2/1557) scn: 0x0000.000d36d2 seq: 0x01 flg: 0x04 tail: 0x36d20201 frmt: 0x02 chkval: 0xbeca type: 0x02=KTU UNDO BLOCK Hex dump of block: st=0, typ_found=1 Dump of memory from 0x0E7E4400 to 0x0E7E6400 ...................... ******************************************************************************** UNDO BLK: xid: 0x0009.01a.00000162 seq: 0x105 cnt: 0x14 irb: 0x14 icl: 0x0 flg: 0x0000 Rec Offset Rec Offset Rec Offset Rec Offset Rec Offset --------------------------------------------------------------------------- 0x01 0x1f08 0x02 0x1e58 0x03 0x1df4 0x04 0x1da0 0x05 0x1d1c 0x06 0x1ca0 0x07 0x1c54 0x08 0x1b80 0x09 0x1afc 0x0a 0x1a70 0x0b 0x198c 0x0c 0x1924 0x0d 0x18cc 0x0e 0x17e8 0x0f 0x1780 0x10 0x1728 0x11 0x16a8 0x12 0x1650 0x13 0x15f4 0x14 0x1598 *-----------------------------
注意,这部分信息有个参数irb: 0x14。irb指的是回滚段中记录的最近所有未提交的事务的开始之处,也就是最近一次被修改的地方,如果要回滚,这便是回滚的起点。
由上,可知:(irb: 0x14)的偏移地址为0x1598
找到irb:0x14信息,如下:
*----------------------------- * Rec #0x14 slt: 0x1a objn: 51148(0x0000c7cc) objd: 51148 tblspc: 4(0x00000004) * Layer: 11 (Row) opc: 1 rci 0x13 Undo type: Regular undo Last buffer split: No Temp Object: No Tablespace Undo: No rdba: 0x00000000 *----------------------------- KDO undo record: KTB Redo op: 0x02 ver: 0x01 op: C uba: 0x00800615.0105.13 KDO Op code: URP row dependencies Disabled xtype: XA flags: 0x00000000 bdba: 0x01000020 hdba: 0x0100001b itli: 2 ispac: 0 maxfr: 4858 tabn: 0 slot: 5(0x5) flag: 0x2c lock: 0 ckix: 191 ncol: 8 nnew: 1 size: 1 col 5: [ 3] c2 1d 33
c2 1d 33转换为10进制就是2850.这是最后被更新的记录的旧值。也就是update emp set sal=4000 where empno=7698。
注意,这里有几个参数需要关注:
1)参数rci 0x13,该参数代表的就是undo chain(同一个事务中的多次修改,根据chain链接关联),此处的rci 0x13指向第二次修改update emp set sal=4000 where empno=7782.此时,如果有其他进程查询scott.emp表,则oracle需要构造一致性读来把数据展现给用户。
2)参数bdba: 0x01000020,该参数代表旧值对应的数据块的地址,下面把这个参数翻译出来:
scott@ORCL> select to_number('01000020','xxxxxxxx') from dual; TO_NUMBER('01000020','XXXXXXXX') -------------------------------- 16777248 scott@ORCL> select dbms_utility.data_block_address_file(16777248) from dual; DBMS_UTILITY.DATA_BLOCK_ADDRESS_FILE(16777248) ---------------------------------------------- 4 scott@ORCL> select dbms_utility.data_block_address_block(16777248) from dual; DBMS_UTILITY.DATA_BLOCK_ADDRESS_BLOCK(16777248) ----------------------------------------------- 32
再将4号文件的第32个块dump出来:
sys@ORCL> select spid from v$process where addr in (select paddr from v$session 2 where sid in (select sid from v$mystat where rownum=1)); SPID ------------ 13929 sys@ORCL> alter system dump datafile 4 block 32; System altered.
部分内容摘入如下:
Start dump data blocks tsn: 4 file#: 4 minblk 32 maxblk 32 buffer tsn: 4 rdba: 0x01000020 (4/32) scn: 0x0000.000d36d2 seq: 0x01 flg: 0x04 tail: 0x36d20601 frmt: 0x02 chkval: 0xa9ea type: 0x06=trans data Block header dump: 0x01000020 Object id on Block? Y seg/obj: 0xc7cc csc: 0x00.b9cf3 itc: 2 flg: E typ: 1 - DATA brn: 0 bdba: 0x1000019 ver: 0x01 opc: 0 inc: 0 exflg: 0 Itl Xid Uba Flag Lck Scn/Fsc 0x01 0x0003.011.000000f2 0x00805794.00c8.49 C--- 0 scn 0x0000.0006bfdb 0x02 0x0009.01a.00000162 0x00800615.0105.14 ---- 3 fsc 0x0002.00000000
这里存在ITL事务槽信息,事务必须得到一个ITL事务槽才能够进行数据修改。ITL内容主要包括xid(指向回滚段头的ITL)、uba(指向回滚块)、Lck(行头锁的个数,此处修改三次,故有3个锁)。在以上输出中,可知:itl2(0x02)上存在活动事务。
将xid(0x0009.01a.00000162)分解一下:0009表明该事务指向9号回滚段,01a表明该事务指向26号事务槽,00000162表明该事务第354次被覆盖(因为回滚段是循环使用的)。
将uba(0x00800615.0105.14)分解一下:00800615表明第1557个回滚块,14是irb信息(回滚的起点)。
至此,undo段和数据块里面的事务信息被完全解析出来,二者完全一致!