深入解析TRUNCATE TABLE – 手工修复和验证过程

  1. 云栖社区>
  2. 数据和云>
  3. 博客>
  4. 正文

深入解析TRUNCATE TABLE – 手工修复和验证过程

技术小能手 2018-08-06 13:12:11 浏览2092
展开阅读全文

众所周知,truncate table 是一种快速清空表内数据的一种方式,与 delete 方式不同,truncate 只产生非常少的 redo 和 undo,就实现了清空表数据并降低表 HWM 的功能。本文主要围绕 truncate table 的恢复来展开。

truncate 的恢复
无覆盖的 TRUNCATE 恢复
重要:假如确实失误 truncate 了表,需要马上停应用,最好将表空间设置为 offline 或者 read only,避免数据被覆盖。

通过之前对 table full scan、segment header 和前面 truncate 原理的分析,tfs 不会读取 L1,L2 块,所以恢复的时候 L1,L2 块和具体存放数据的块都不用管,只需尝试通过修改段头块和基表信息来恢复 truncate 的数据,bbed 修改段头块的具体 offset 含义见 segment header 章节。

修改段头块 dataobj#
由于表可能不止一次被 truncate,所以获取之前 dataobj# 最好的办法是通过 logminer 或者 redo dump,这里我使用的 redo dump
REDO RECORD - Thread:1 RBA: 0x000055.0000001b.0080 LEN: `js
0x00ac VLD: 0x01
SCN: 0x0000.003ed056 SUBSCN: 1 04/21/2018 12:25:34
CHANGE #1 TYP:0 CLS:34 AFN:6 DBA:0x018002ee OBJ:4294967295 SCN:0x0000.003ed054 SEQ:1 OP:5.1 ENC:0 RBL:0
ktudb redo: siz: 64 spc: 822 flg: 0x0022 seq: 0x017b rec: 0x33
xid: 0x0009.00f.00000353
ktubu redo: slt: 15 rci: 50 opc: 14.5 objn: 1 objd: 16840 tsn: 5
Undo type: Regular undo Undo type: Last buffer split: No
Tablespace Undo: Yes
0x00000000
kteopu undo - undo operation on extent map
segdba: 0x140076a class: 4 mapdba:0x140076a offset: 2
rbr extent - dba: 0x0 nbk: 0x0
kteop redo - redo operation on extent map
CDOBJ: new object number:16840
CHANGE #2 TYP:0 CLS:4 AFN:5 DBA:0x0140076a OBJ:16840 SCN:0x0000.003ed050 SEQ:2 OP:14.4 ENC:0 RBL:0
kteop redo - redo operation on extent map
CDOBJ: new object number:16860


从 redo dump 信息可以看到段头块 0x0140076a 的 dataobj#从16840 变成了16860,所以这里需要将段头块的 dataobj# 改回 16840,通过 bbed 进行修改:

BBED> set file 4 block 1898
FILE# 4
BLOCK# 1898
BBED> d offset 272 count 8
File: /u01/app/oracle/oradata/test/users01.dbf (4)

Block: 1898 Offsets: 272 to 279 Dba:0x0100076a

dc410000 00000010
<32 bytes per line>
BBED> m /x c841 offset 272
Warning: contents of previous BIFILE will be lost. Proceed? (Y/N) y
File: /u01/app/oracle/oradata/test/users01.dbf (4)

Block: 1898 Offsets: 272 to 279 Dba:0x0100076a

c8410000 00000010
<32 bytes per line>
BBED> sum apply
Check value for File 4, Block 1898:
current = 0xe90e, required = 0xe90e

修改基表
仅仅修改段头块的 dataobj# 再次查询表会报 ORA-08103: object no longer exists,原因是基表没有修改。查询的时候会通过表名去找到该表的 dataobj#

SYS@TEST(test):1>select count(*) from test.truncate_table;
select count(*) from test.truncate_table
*
ERROR at line 1:
ORA-08103: object no longer exists
SYS@TEST(test):1>UPDATE OBJ$ SET DATAOBJ#=16840 WHERE OBJ#=16840;
1 row updated.
SYS@TEST(test):1>UPDATE TAB$ SET DATAOBJ#=16840 WHERE OBJ#=16840;
1 row updated.
SYS@TEST(test):1>COMMIT;
Commit complete.
SYS@TEST(test):1>alter system flush buffer_cache;
System altered.
SYS@TEST(test):1>alter system flush shared_pool;
System altered.
SYS@TEST(test):1>select count(*) from test.truncate_table;

COUNT(*)

0

恢复段头块 HWM
HWM 信息可以从 redo dump 中获取

REDO RECORD - Thread:1 RBA: 0x000055.00000016.0148 LEN: 0x01b8 VLD: 0x01
SCN: 0x0000.003ed050 SUBSCN: 5 04/21/2018 12:25:34
CHANGE #1 TYP:0 CLS:34 AFN:6 DBA:0x018002ee OBJ:4294967295 SCN:0x0000.003ed050 SEQ:3 OP:5.1 ENC:0 RBL:0
ktudb redo: siz: 164 spc: 1112 flg: 0x0022 seq: 0x017b rec: 0x30
xid: 0x0009.00f.00000353
ktubu redo: slt: 15 rci: 47 opc: 13.29 objn: 16840 objd: 16840 tsn: 5
Undo type: Regular undo Undo type: Last buffer split: No
Tablespace Undo: No
0x00000000
Segment Header Undo
Seghdr dba: 0x0140076a Mapblock dba: 0x00000000 Mapredo Offset: 4 scls: 4 mcls: 140733193388039
Both the HWMs
Low HWM
Highwater:: 0x01405b83 ext#: 27 blk#: 3 ext size: 128

blocks in seg. hdr's freelists: 0

blocks below: 1539

mapblk 0x00000000 offset: 27
lfdba: 0x01405b80
High HWM
Highwater:: 0x01405b83 ext#: 27 blk#: 3 ext size: 128

blocks in seg. hdr's freelists: 0

blocks below: 1539

mapblk 0x00000000 offset: 27
lfdba: 0x01405b80 hint dba: 0x01400769
Lasts in Header
LF: 20994945 LS: 20973417 LT: 0 FT: 0
CHANGE #2 TYP:0 CLS:4 AFN:5 DBA:0x0140076a OBJ:16840 SCN:0x0000.003ed050 SEQ:1 OP:13.28 ENC:0 RBL:0
Both the HWMs
Low HWM
Highwater:: 0x0140076b ext#: 0 blk#: 3 ext size: 8

blocks in seg. hdr's freelists: 0

blocks below: 0

mapblk 0x00000000 offset: 0
lfdba: 0x01400768
High HWM
Highwater:: 0x0140076b ext#: 0 blk#: 3 ext size: 8

blocks in seg. hdr's freelists: 0

blocks below: 0

mapblk 0x00000000 offset: 0
lfdba: 0x01400768 hint dba: 0x01400769
Lasts in Header
LF: 20973416 LS: 20973417 LT: 0 FT: 0
CHANGE #3 TYP:0 CLS:8 AFN:5 DBA:0x01400768 OBJ:16840 SCN:0x0000.003ed050 SEQ:2 OP:13.22 ENC:0 RBL:0
Redo on Level1 Bitmap Block
Redo to set hwm
Opcode: 32 Highwater:: 0x0140076b ext#: 0 blk#: 3 ext size: 8

blocks in seg. hdr's freelists: 0

blocks below: 0

mapblk 0x00000000 offset: 0

可以看到段头块 0x0140076a 的 LHWM 信息和 HHWM 是一样的,都是从
Highwater::  0x01405b83  ext#: 27     blk#: 3      ext size: 128
变成了
Highwater:: 0x0140076b  ext#:0      blk#:3      extsize:8
 
所以这里恢复 HWM 信息只需要根据 redo dump 改回去即可

m /x 1b offset 48
m /x 1b offset 92
m /x 03 offset 52
m /x 03 offset 96
m /x 80 offset 56
m /x 80 offset 100
m /x 835b offset 60
m /x 4001 offset 62
m /x 835b offset 104
m /x 4001 offset 106

 
恢复段头块 extent map,Auxillary Map 以及 extent 个数
extent 信息和 aux map 信息同样可以从 redo dump 中获取
REDO RECORD - Thread:1 RBA: 0x000055.00000027.0180 LEN: ```js
0x0138 VLD: 0x01
SCN: 0x0000.003ed061 SUBSCN:  3 04/21/2018 12:25:35
CHANGE #1 TYP:0 CLS:17 AFN:6 DBA:0x01800120 OBJ:4294967295 SCN:0x0000.003ed061 SEQ:1 OP:5.2 ENC:0 RBL:0
ktudh redo: slt: 0x0014 sqn: 0x00000000 flg: 0x0002 siz: 112 fbi: 248
uba: 0x018006ea.018c.1f    pxid:  0x0000.000.00000000
CHANGE #2 TYP:0 CLS:18 AFN:6 DBA:0x018006ea OBJ:4294967295 SCN:0x0000.003ed061 SEQ:1 OP:5.1 ENC:0 RBL:0
ktudb redo: siz: 112 spc: 3358 flg: 0x0022 seq: 0x018c rec: 0x1f
xid:  0x0001.014.00000297
ktubu redo: slt: 20 rci: 0 opc: 14.5 objn: 1 objd: 16860 tsn: 5
Undo type:  Regular undo       Undo type:  Last buffer split:  No
Tablespace Undo:  Yes
0x00000000
kteopu undo - undo operation on extent map
segdba: 0x140076a  class: 4  mapdba:0x140076a  offset: 3
rbr extent - dba: 0x0  nbk: 0x0
kteop redo - redo operation on extent map

ADD: dba:0x1405b80 len:128 at offset:27 --truncate前的ext#为27 的extent条目信息

ADDAXT: offset:27 fdba:x01405b80 bdba:0x01405b82 --truncate
前```  
ext#为27的aux map条目信息

SETSTAT: exts:28 blks:1664 lastmap:0x0 mapcnt:0
CHANGE #3 TYP:0 CLS:4 AFN:5 DBA:0x0140076a OBJ:16860 SCN:0x0000.003ed061 SEQ:2 OP:14.4 ENC:0 RBL:0
kteop redo - redo operation on extent map
DELETE: entry:27
shift back: dba:0x0 len:0
SETSTAT: exts:27 blks:1536 lastmap:0x0 mapcnt:0


从 redo dump 可以发现 truncate 操作对于 extent map 和 aux map 是从最后一个 extent 开始逐一删除的,这里可以看到该表的 extent 总共有 28 个,ext#为27 是该表最后一个 extent,block_id 为 0x1405b80,该 extent size 为 128 个块,以此类推很容易可以通过简单的 grep 找出 extent map;同理 ext#为27 的 aux map 的 L1 dba 为 x01405b80,data dba 为 0x01405b82,以此类推很容易可以通过简单的 grep 找出 aux map
extent map:

[root@prim1-11g ~]# grep -i "ADD: dba" /u01/app/oracle/diag/rdbms/test/test/trace/test_ora_7611.trc
ADD: dba:0x1405b80 len:128 at offset:27
ADD: dba:0x1405b00 len:128 at offset:26
ADD: dba:0x1405a80 len:128 at offset:25
ADD: dba:0x1405a00 len:128 at offset:24
ADD: dba:0x1405980 len:128 at offset:23
ADD: dba:0x1405900 len:128 at offset:22
ADD: dba:0x1405880 len:128 at offset:21
ADD: dba:0x1405800 len:128 at offset:20
ADD: dba:0x1404180 len:128 at offset:19
ADD: dba:0x1404080 len:128 at offset:18
ADD: dba:0x1404000 len:128 at offset:17
ADD: dba:0x1403f80 len:128 at offset:16
ADD: dba:0x1402eb8 len:8 at offset:15
ADD: dba:0x1402eb0 len:8 at offset:14
ADD: dba:0x1402ea8 len:8 at offset:13
ADD: dba:0x1402ea0 len:8 at offset:12
ADD: dba:0x1402e98 len:8 at offset:11
ADD: dba:0x1402e90 len:8 at offset:10
ADD: dba:0x1402e88 len:8 at offset:9
ADD: dba:0x1402e80 len:8 at offset:8
ADD: dba:0x1402f78 len:8 at offset:7
ADD: dba:0x1402f70 len:8 at offset:6
ADD: dba:0x1402f68 len:8 at offset:5
ADD: dba:0x1402f60 len:8 at offset:4
ADD: dba:0x1402f58 len:8 at offset:3
ADD: dba:0x1402f50 len:8 at offset:2
ADD: dba:0x1402f48 len:8 at offset:1
aux map:
[root@prim1-11g ~]# grep -i "ADDAXT:" /u01/app/oracle/diag/rdbms/test/test/trace/test_ora_7611.trc
ADDAXT: offset:27 fdba:x01405b80 bdba:0x01405b82
ADDAXT: offset:26 fdba:x01405b00 bdba:0x01405b02
ADDAXT: offset:25 fdba:x01405a80 bdba:0x01405a82
ADDAXT: offset:24 fdba:x01405a00 bdba:0x01405a02
ADDAXT: offset:23 fdba:x01405980 bdba:0x01405982
ADDAXT: offset:22 fdba:x01405900 bdba:0x01405902
ADDAXT: offset:21 fdba:x01405880 bdba:0x01405882
ADDAXT: offset:20 fdba:x01405800 bdba:0x01405802
ADDAXT: offset:19 fdba:x01404180 bdba:0x01404182
ADDAXT: offset:18 fdba:x01404080 bdba:0x01404082
ADDAXT: offset:17 fdba:x01404000 bdba:0x01404002
ADDAXT: offset:16 fdba:x01403f80 bdba:0x01403f82
ADDAXT: offset:15 fdba:x01402eb0 bdba:0x01402eb8
ADDAXT: offset:14 fdba:x01402eb0 bdba:0x01402eb1
ADDAXT: offset:13 fdba:x01402ea0 bdba:0x01402ea8
ADDAXT: offset:12 fdba:x01402ea0 bdba:0x01402ea1
ADDAXT: offset:11 fdba:x01402e90 bdba:0x01402e98
ADDAXT: offset:10 fdba:x01402e90 bdba:0x01402e91
ADDAXT: offset:9 fdba:x01402e80 bdba:0x01402e88
ADDAXT: offset:8 fdba:x01402e80 bdba:0x01402e81
ADDAXT: offset:7 fdba:x01402f70 bdba:0x01402f78
ADDAXT: offset:6 fdba:x01402f70 bdba:0x01402f71
ADDAXT: offset:5 fdba:x01402f60 bdba:0x01402f68
ADDAXT: offset:4 fdba:x01402f60 bdba:0x01402f61
ADDAXT: offset:3 fdba:x01402f50 bdba:0x01402f58
ADDAXT: offset:2 fdba:x01402f50 bdba:0x01402f51
ADDAXT: offset:1 fdba:x01400768 bdba:0x01402f48

这里可以写脚本根据 segment header 章节 offset 的含义来生成 bbed 命令,由于 bbed 命令较长省略一部分
修改exts信息:

m /x 1c offset 36
m /x 1c offset 264

将 grep 出的数据作为一整列全部导入表中用 sql 生成 bbed 命令
如创建表:create table aux_map(a varchar2(2000);create table ext_map(a varchar2(2000);插入数据后就可以用下列 sql 生成命令
生成 aux map 的 bbed 命令

with aa as (
select replace(regexp_substr(a,'1+',1,3),'fdba','') ext#,
(replace(regexp_substr(a,'1+',1,3),'fdba','')-1)*8+2744 offset,
trim(replace(regexp_substr(a,'1+',1,4),'bdba','')) l1,
trim(regexp_substr(a,'1+',1,5)) data
from aux_map)
select 'm /x '||a||' offset '||offset||chr(10)||'m /x '||b||' offset '||to_char(offset+2)||chr(10)||'m /x '||c||' offset '||to_char(offset+4)||chr(10)||'m /x '||d||' offset '||to_char(offset+6) from
(select ext#,to_number(offset) offset,substr(l1,-2,2)||substr(l1,-4,2) a,substr(l1,-6,2)||substr(l1,-8,2) b,substr(data,-2,2)||substr(data,-4,2) c,substr(data,-6,2)||substr(data,-8,2) d
from aa)

生成 ext map 的 bbed 命令

with aa as (
select trim(replace(replace(regexp_substr(a,'1+',1,3),'len',''),'x','x0')) block_id,
(regexp_substr(a,'1+',1,5)-1)*8+288 offset,
lpad(trim((to_char((replace(regexp_substr(a,'1+',1,4),'at offset','')),'xxxx'))),2,0) blocks,
regexp_substr(a,'1+',1,5) ext#
from ext_map)
select 'm /x '||a||' offset '||offset||chr(10)||'m /x '||b||' offset '||to_char(offset+2)||chr(10)||'m /x '||blocks||' offset '||to_char(offset+4) from
(select ext#,to_number(offset) offset,substr(block_id,-2,2)||substr(block_id,-4,2) a,substr(block_id,-6,2)||substr(block_id,-8,2) b,blocks
from aa)

 
bbed 修改后 extent map 恢复成功:
B```js
BED> d /v offset 280
File: /u01/app/oracle/oradata/test/users01.dbf (4)
Block: 1898    Offsets:  280 to  579  Dba:0x0100076a
-------------------------------------------------------
68074001 08000000 482f4001 08000000 l h.@.....H/@.....
502f4001 08000000 582f4001 08000000 l P/@.....X/@.....
602f4001 08000000 682f4001 08000000 l `/@.....h/@.....
702f4001 08000000 782f4001 08000000 l p/@.....x/@.....
802e4001 08000000 882e4001 08000000 l ..@.......@.....
902e4001 08000000 982e4001 08000000 l ..@.......@.....
a02e4001 08000000 a82e4001 08000000 l ..@.......@.....
b02e4001 08000000 b82e4001 08000000 l ..@.......@.....
803f4001 80000000 00404001 80000000 l .?@......@@.....
80404001 80000000 80414001 80000000 l .@@......A@.....
00584001 80000000 80584001 80000000 l .X@......X@.....
00594001 80000000 80594001 80000000 l .Y@......Y@.....
005a4001 80000000 805a4001 80000000 l .Z@......Z@.....
005b4001 80000000 805b4001 80000000 l .[@......[@.....
00000000 00000000 00000000 00000000 l ................
00000000 00000000 00000000 00000000 l ................
00000000 00000000 00000000 00000000 l ................
00000000 00000000 00000000 00000000 l ................
00000000 00000000 00000000          l ............
<16 bytes per line>

aux map 恢复成功:

BBED> d /v offset 2736
File: /u01/app/oracle/oradata/test/users01.dbf (4)
Block: 1898    Offsets: 2736 to 3035  Dba:0x0100076a
-------------------------------------------------------
68074001 6b074001 68074001 482f4001 l h.@.k.@.h.@.H/@.
502f4001 512f4001 502f4001 582f4001 l P/@.Q/@.P/@.X/@.
602f4001 612f4001 602f4001 682f4001 l `/@.a/@.`/@.h/@.
702f4001 712f4001 702f4001 782f4001 l p/@.q/@.p/@.x/@.
802e4001 812e4001 802e4001 882e4001 l ..@...@...@...@.
902e4001 912e4001 902e4001 982e4001 l ..@...@...@...@.
a02e4001 a12e4001 a02e4001 a82e4001 l ..@...@...@...@.
b02e4001 b12e4001 b02e4001 b82e4001 l ..@...@...@...@.
803f4001 823f4001 00404001 02404001 l .?@..?@..@@..@@.
80404001 82404001 80414001 82414001 l .@@..@@..A@..A@.
00584001 02584001 80584001 82584001 l .X@..X@..X@..X@.
00594001 02594001 80594001 82594001 l .Y@..Y@..Y@..Y@.
005a4001 025a4001 805a4001 825a4001 l .Z@..Z@..Z@..Z@.
005b4001 025b4001 805b4001 825b4001 l .[@..[@..[@..[@.
00000000 00000000 00000000 00000000 l ................
00000000 00000000 00000000 00000000 l ................
00000000 00000000 00000000 00000000 l ................
00000000 00000000 00000000 00000000 l ................
00000000 00000000 00000000          l ............
<16 bytes per line>

最后恢复成功
这里由于L1,L2没有恢复,所以insert会有问题,但是可以通过CTAS重建表完全恢复。

SYS@TEST(test):1>select count(*) from test.truncate_table;
COUNT(*)
----------
113426
SYS@TEST(test):1>insert into test.truncate_table select * from dba_objects;
insert into test.truncate_table select * from dba_objects
*
ERROR at line 1:
ORA-00600: internal error code, arguments: [ktspgfblk3:kcbz_objdchk], [0], [0], [1], [], [], [], [], [], [], [], []
SYS@TEST(test):1>CREATE TABLE TEST.RECOVER_TABLE AS SELECT * FROM TEST.TRUNCATE_TABLE;
Table created.
SYS@TEST(test):1>INSERT INTO TEST.RECOVER_TABLE select * from dba_objects WHERE ROWNUM=1;
1 row created.
SYS@TEST(test):1>COMMIT;
Commit complete.

但是这种恢复方式只适用于truncate的数据没有被覆盖,也就是没有其他对象占用了truncate所释放的空间,所以在操作时需要查询dba_extents对比redo dump出来的extent map,如果没有覆盖的情况,则上述方法可行。那么被覆盖的情况该如何处理呢?

被覆盖的 truncate 恢复
建立测试环境:
创建测试表 truncate_table_2,truncate 后创建 aaa 表覆盖 truncate_table_2 表 truncate 后释放的空间。

SYS@TEST(test):1>create table test.truncate_table_2 tablespace test as select * from dba_objects where rownum<=6000;
Table created.
SYS@TEST(test):1>select extent_id,file_id,block_id,blocks from dba_extents where segment_name='TRUNCATE_TABLE_2';
EXTENT_ID    FILE_ID   BLOCK_ID     BLOCKS
---------- ---------- ---------- ----------
0                         4             136                  8
1                         4             144                  8
2                         4             152                  8
3                         4             160                  8
4                         4             168                  8
5                         4             176                  8
6                         4             184                  8
7                         4             192                  8
8                         4             200                  8
9                         4             208                  8
10                       4             216                  8
11 rows selected.
SYS@TEST(test):1>select count(*) from test.truncate_table_2;
COUNT(*)
----------
6000
SYS@TEST(test):1>truncate table test.truncate_table_2;
Table truncated.
SYS@TEST(test):1>alter system switch logfile;
System altered.
SYS@TEST(test):1>oradebug setmypid;
Statement processed.
SYS@TEST(test):1>alter system dump logfile '/u01/app/oracle/oradata/test/redo01.log';
System altered.
SYS@TEST(test):1>oradebug tracefile_name
/u01/app/oracle/diag/rdbms/test/test/trace/test_ora_17034.trc
SYS@TEST(test):1>create table test.aaa tablespace test as select * from dba_objects where rownum<=500;
Table created.
SYS@TEST(test):1>select extent_id,file_id,block_id,blocks from dba_extents where segment_name='AAA';
EXTENT_ID    FILE_ID   BLOCK_ID     BLOCKS
---------- ---------- ---------- ----------
0                         4                 144                 8
1                         4                 152                 8

在这里就可以清楚的发现表 truncate_table_2 的数据被表 aaa 覆盖了一部分。

将占用truncate释放的空间的对象move到其他表空间
move 之前先查询该对象占用的 extent 信息,并且 dump segment header 以便后续的处理

SYS@:>select extent_id,file_id,block_id,blocks from dba_extents where segment_name='AAA';
EXTENT_ID    FILE_ID   BLOCK_ID     BLOCKS
---------- ---------- ---------- ----------
0          4        144          8
1          4        152          8
SYS@:>alter table test.aaa move tablespace users;
Table altered.

恢复段头块以及数据字典信息
这里步骤和没有覆盖情况一致,不做详述。

BBED> m /x 3e offset 272 --修改段头dataobj#
/*修改数据字典信息
SYS@TEST(test):1>update obj$ set dataobj#=16958 where obj#=16958;
1 row updated.
SYS@TEST(test):1>update tab$ set dataobj#=16958 where obj#=16958;
1 row updated.
SYS@TEST(test):1>commit;
Commit complete.
*/
/*从redo dump里获取HWM信息并修改段头块
m /x 0a offset 48
m /x 0a offset 92
m /x 05 offset 52
m /x 05 offset 96
m /x 08 offset 56
m /x 08 offset 100
m /x dd00 offset 60
m /x 0001 offset 62
m /x dd00 offset 104
m /x 0001 offset 106
*/
/*修改段头ext map、aux map以及extent个数
略
*/
尝试查询报错
SYS@TEST(test):1>select count(*) from test.truncate_table_2;
select count(*) from test.truncate_table_2
*
ERROR at line 1:
ORA-08103: object no longer exists

确认具体哪些块被覆盖
剩下的问题,其实就是解决 8103 的问题了,实际上 8103 的错误原因就是被覆盖的数据块的 dataobj# 和 truncate_table_2 的 dataobj# 不一致导致的

Corrupt block relative dba: 0x01000090 (file 4, block 144)
Fractured block found during multiblock buffer read
Data in bad block:
type: 6 format: 2 rdba: 0x01000090
last change scn: 0x0000.003f9e3d seq: 0x3 flg: 0x04
spare1: 0x0 spare2: 0x0 spare3: 0x0
consistency value in tail: 0x9e3d2003
check value in block header: 0x5bde
computed block checksum: 0x0
Reading datafile '/u01/app/oracle/oradata/test/TEST/datafile/o1_mf_test_fg6gjt1d_.dbf' for corruption at rdba: 0x01000090 (file 4, block 144)
Reread (file 4, block 144) found same corrupt data (no logical check)
*** SESSION ID:(178.85) 2018-04-28 00:24:12.968
OBJD MISMATCH typ=33, seg.obj=16958, diskobj=16960, dsflg=100001, dsobj=16958, tid=16958, cls=1

明确指出在全表扫描的时候发现file 4 block 144的dataobj#为16960,但是seg的dataobj#为16958。
表AAA在move之前,需要先查出extent信息和dump segment header,目的就在这里(假如没有dump 段头仍然可以从redo dump里找到HWM信息)
extent 信息

SYS@:>select extent_id,file_id,block_id,blocks from dba_extents where segment_name='AAA';
EXTENT_ID    FILE_ID   BLOCK_ID     BLOCKS
---------- ---------- ---------- ----------
0          4        144          8
1          4        152          8

可以确定4号文件144号块开始被覆盖,但是不一定ext#为1的最后一个块也被覆盖了,因为LHWM和HHWM之间的块不一定被覆盖
HWM信息

Highwater::  0x0100009a  ext#: 1      blk#: 2      ext size: 8  --file 4 block 154

解决 ORA-8103
构造 rowid 绕过 dataobj# 不匹配的块

SYS@TEST(test):1>select count(*) from
2  (select * from test.truncate_table_2 where rowid<dbms_rowid.rowid_create(1, 16958, 4, 144, 0)
3  union all
4  select * from test.truncate_table_2 where rowid>=dbms_rowid.rowid_create(1, 16958, 4, 154, 0));
COUNT(*)
----------
5303

成功挽救 5303 条数据,少了 697 条,也可以用 dbms_repair 标记这些块为坏块,然后跳过这些块。

总结

对于 truncate 的恢复,在无备份的情况下,如果需要恢复的表比较多,或者分区表的分区比较多,修复元数据的方式较为麻烦,还是建议使用 odu 来进行恢复。

原文发布时间为:2018-08-03
本文作者:李翔宇
本文来自云栖社区合作伙伴“数据和云”,了解相关信息可以关注“数据和云”。


  1. :

网友评论

登录后评论
0/500
评论
技术小能手
+ 关注
所属团队号: 数据和云