[20191206]隱含參數_db_always_check_system_ts.txt

[20191206]隱含參數_db_always_check_system_ts.txt

--//今年年頭我作tab$刪除恢復時,遇到的問題,就是遇到延遲塊清除的問題.參考連接:
http://blog.itpub.net/267265/viewspace-2564716/
http://blog.itpub.net/267265/viewspace-2564717/

--//當時測試若是發生延遲塊清除,修復刪除記錄的塊,再讀取時若是是系統表空間時報錯,通常用戶的表空間是沒有問題.
--//一直想知道系統表空間有什麼隱含參數能夠繞過這個問題,畢竟修復數據庫若是錯誤太多,沒法一塊一塊來修復.
--//昨天才知道有一個隱含參數_db_always_check_system_ts能夠繞過這個錯誤.今天測試看看.

1.環境:
SCOTT@book> @ ver1
PORT_STRING                    VERSION        BANNER
------------------------------ -------------- --------------------------------------------------------------------------------
x86_64/Linux 2.4.xx            11.2.0.4.0     Oracle Database 11g Enterprise Edition Release 11.2.0.4.0 - 64bit Production

SYS@book> @ hide _db_always_check_system_ts
NAME                       DESCRIPTION                                                   DEFAULT_VALUE SESSION_VALUE SYSTEM_VALUE ISSES ISSYS_MOD
-------------------------- ------------------------------------------------------------- ------------- ------------- ------------ ----- ---------
_db_always_check_system_ts Always perform block check and checksum for System tablespace TRUE          TRUE          TRUE         FALSE IMMEDIATE
--//缺省_db_always_check_system_ts=true.

2.測試:
SYS@book> create table t tablespace system as select rownum id,'test' name from dual connect by level<=2;
Table created.

SYS@book> select rowid,t.* from t;
ROWID                      ID NAME
------------------ ---------- ----
AAAWEgAABAAAAl5AAA          1 test
AAAWEgAABAAAAl5AAB          2 test

SYS@book> @ rowid AAAWEgAABAAAAl5AAA
    OBJECT       FILE      BLOCK        ROW ROWID_DBA            DBA                  TEXT
---------- ---------- ---------- ---------- -------------------- -------------------- ----------------------------------------
     90400          1       2425          0   0x400979           1,2425               alter system dump datafile 1 block 2425
-//創建在system表空間.

SYS@book>  delete from t where id=1;
1 row deleted.

SYS@book>  alter system flush buffer_cache;
System altered.

SYS@book>  alter system flush buffer_cache;
System altered.

SYS@book> @ bh 1 2425
HLADDR              DBARFIL     DBABLK      CLASS CLASS_TYPE         STATE             TCH CR_SCN_BAS CR_SCN_WRP CR_UBA_FIL CR_UBA_BLK CR_UBA_SEQ BA               OBJECT_NAME
---------------- ---------- ---------- ---------- ------------------ ---------- ---------- ---------- ---------- ---------- ---------- ---------- ---------------- --------------------
0000000084DEACF8          1       2425          1 data block         free                0          0          0          0          0          0 00000000772C2000 T
0000000084DEACF8          1       2425          1 data block         free                0          0          0          0          0          0 00000000772C4000 T
--//肯定該塊不在數據庫緩存.

SYS@book> commit ;
Commit complete.
--//這個時候不會寫塊提交到塊中,由於數據塊已經不在數據緩存了.

3.使用bbed修復該記錄看看:
BBED> set dba   1,2425
        DBA             0x00400979 (4196729 1,2425)

BBED> x /rnc *kdbr[1]
rowdata[0]                                  @8166
----------
flag@8166: 0x2c (KDRHFL, KDRHFF, KDRHFH)
lock@8167: 0x00
cols@8168:    2

col    0[2] @8169: 2
col    1[4] @8172: test


BBED> x /rnc *kdbr[0]
rowdata[11]                                 @8177
-----------
flag@8177: 0x3c (KDRHFL, KDRHFF, KDRHFD, KDRHFH)
lock@8178: 0x02
cols@8179:    0

--//第1條記錄已經刪除,flag=0x3c.

BBED> assign offset 8177 =0x2c;
Warning: contents of previous BIFILE will be lost. Proceed? (Y/N) y
ub1 rowdata[0]                              @8177     0x2c

BBED> x /rnc *kdbr[0]
rowdata[11]                                 @8177
-----------
flag@8177: 0x2c (KDRHFL, KDRHFF, KDRHFH)
lock@8178: 0x02
cols@8179:    2

col    0[2] @8180: 1
col    1[4] @8183: test
--//ok,如今已經恢復.lock=0x02,使用itl槽1(從0開始)

BBED> sum apply
Check value for File 1, Block 2425:
current = 0xff20, required = 0xff20

BBED> verify
DBVERIFY - Verification starting
FILE = /mnt/ramdisk/book/system01.dbf
BLOCK = 2425

Block Checking: DBA = 4196729, Block Type = KTB-managed data block
data header at 0x7f344dbba274
kdbchk: the amount of space used is not equal to block size
        used=44 fsc=9 avsp=8028 dtl=8072
Block 2425 failed with check code 6110
--//注:這個報錯沒必要理會在select讀取時.

BBED> p ktbbh.ktbbhitl[1]
struct ktbbhitl[1], 24 bytes                @68
   struct ktbitxid, 8 bytes                 @68
      ub2 kxidusn                           @68       0x000a
      ub2 kxidslt                           @70       0x0013
      ub4 kxidsqn                           @72       0x00004d92
   struct ktbituba, 8 bytes                 @76
      ub4 kubadba                           @76       0x00c00288
      ub2 kubaseq                           @80       0x0f0a
      ub1 kubarec                           @82       0x0e
   ub2 ktbitflg                             @84       0x0002 (NONE)
   union _ktbitun, 2 bytes                  @86
      sb2 _ktbitfsc                         @86       9
      ub2 _ktbitwrp                         @86       0x0009
   ub4 ktbitbas                             @88       0x00000000

--//能夠發現ktbitflg=0x0002,表示沒有提交.有點奇怪爲何是0x0002,應該是0x0001(由於我僅僅刪除1條記錄)
--//注:關於這點我在之前blog提到參考連接http://blog.itpub.net/267265/viewspace-2564779/,視乎使用表空間類型是mssm就能看到
--//這樣的狀況.
--//ktbitbas=0x00000000,也就是沒有scn相關信息寫入.

--//若是我這時讀取該塊就會遇到連接測試遇到的狀況:連接http://blog.itpub.net/267265/viewspace-2564717/
ORA-00607: Internal error occurred while making a change to a data block
ORA-00600: internal error code, arguments: [kdBlkCheckError], [1], [2537], [6110], [], [], [], [], [], [], [], []
--//注意錯誤號6110,與bbed的錯誤號一致.

--//由於延遲塊清除,在讀取該塊時要寫入itl槽scn號.設置提交標識.這樣對於system表空間這樣的塊就會報錯(bbed verify沒有經過)
--//如今修改參數:
SYS@book> alter system set "_db_always_check_system_ts"=false scope=memory ;
System altered.

SYS@book> @ hide "_db_always_check_system_ts"
NAME                       DESCRIPTION                                                   DEFAULT_VALUE SESSION_VALUE SYSTEM_VALUE ISSES ISSYS_MOD
-------------------------- ------------------------------------------------------------- ------------- ------------- ------------ ----- ---------
_db_always_check_system_ts Always perform block check and checksum for System tablespace TRUE          FALSE         FALSE        FALSE IMMEDIATE

--//按照介紹該參數是當即生效ISSYS_MOD=IMMEDIATE.保險起見仍是退出會話在登陸看看.

SYS@book> select rowid,t.* from t;
ROWID               ID NAME
------------------ --- -----
AAAWEgAABAAAAl5AAA   1 test
AAAWEgAABAAAAl5AAB   2 test

--//OK,如今讀取就沒有問題.再次經過bbed觀察:

BBED> set dba   1,2425
        DBA             0x00400979 (4196729 1,2425)

BBED> p ktbbh.ktbbhitl[1]
struct ktbbhitl[1], 24 bytes                @68
   struct ktbitxid, 8 bytes                 @68
      ub2 kxidusn                           @68       0x000a
      ub2 kxidslt                           @70       0x0013
      ub4 kxidsqn                           @72       0x00004d92
   struct ktbituba, 8 bytes                 @76
      ub4 kubadba                           @76       0x00c00288
      ub2 kubaseq                           @80       0x0f0a
      ub1 kubarec                           @82       0x0e
   ub2 ktbitflg                             @84       0xa000 (KTBFUPB, KTBFCOM)`
   union _ktbitun, 2 bytes                  @86
      sb2 _ktbitfsc                         @86       3
      ub2 _ktbitwrp                         @86       0x0003
   ub4 ktbitbas                             @88       0x17600426

--//能夠發現ktbitflg=0xa000(KTBFUPB, KTBFCOM),表示提交.
--//ktbitbas=0x17600426,也就是scn相關信息已經寫入.

BBED> verify
DBVERIFY - Verification starting
FILE = /mnt/ramdisk/book/system01.dbf
BLOCK = 2425

Block Checking: DBA = 4196729, Block Type = KTB-managed data block
data header at 0x1f09e74
kdbchk: the amount of space used is not equal to block size
        used=44 fsc=0 avsp=8037 dtl=8072
Block 2425 failed with check code 6110

--//仍是報6110錯誤.可是該塊的讀取是沒有問題的.

SYS@book> alter system set "_db_always_check_system_ts"=true scope=memory ;
System altered.

SYS@book> select rowid,t.* from t;
ROWID                      ID NAME
------------------ ---------- ----
AAAWEgAABAAAAl5AAA          1 test
AAAWEgAABAAAAl5AAB          2 test

--//讀取沒有問題,若是修改該記錄如今就會報錯(注意"_db_always_check_system_ts"=true),驗證看看.

SYS@book> update t set name='TEST' where id=2;
update t set name='TEST' where id=2
       *
ERROR at line 1:
ORA-00600: internal error code, arguments: [kdBlkCheckError], [1], [2425], [6110], [], [], [], [], [], [], [], []

SYS@book> alter system set "_db_always_check_system_ts"=false scope=memory ;
System altered.

SYS@book> update t set name='TEST' where id=2;
update t set name='TEST' where id=2
       *
ERROR at line 1:
ORA-01578: ORACLE data block corrupted (file # 1, block # 2425)
ORA-01110: data file 1: '/mnt/ramdisk/book/system01.dbf'

SYS@book> alter system flush buffer_cache;
System altered.

SYS@book> update t set name='TEST' where id=2;
1 row updated.
--//ok如今沒有問題.

SYS@book> commit ;
Commit complete.

SYS@book> select rowid,t.* from t;
ROWID               ID NAME
------------------ --- -----
AAAWEgAABAAAAl5AAA   1 test
AAAWEgAABAAAAl5AAB   2 TEST

--//實際上這個時候使用bbed verify檢查仍是報錯.
BBED> set dba   1,2425
        DBA             0x00400979 (4196729 1,2425)

BBED> verify
DBVERIFY - Verification starting
FILE = /mnt/ramdisk/book/system01.dbf
BLOCK = 2425

Block Checking: DBA = 4196729, Block Type = KTB-managed data block
data header at 0x7f756ceee274
kdbchk: the amount of space used is not equal to block size
        used=44 fsc=0 avsp=8037 dtl=8072
Block 2425 failed with check code 6110

4.總結:
--//設置_db_always_check_system_ts=false,能夠繞過一些數據庫塊錯誤,實際上遇到這樣的狀況,最佳的方式設置read only.
--//採用exp或者expdp方式儘快取出數據從新建庫.

5.補充如何修復該塊:
BBED> verify dba 1,2425
DBVERIFY - Verification starting
FILE = /mnt/ramdisk/book/system01.dbf
BLOCK = 2425

Block Checking: DBA = 4196729, Block Type = KTB-managed data block
data header at 0x15b9e74
kdbchk: the amount of space used is not equal to block size
        used=44 fsc=0 avsp=8037 dtl=8072
Block 2425 failed with check code 6110

--//avsp= dtl-used-fsc = 8072-0 -44 = 8028

BBED> assign kdbh.kdbhavsp=8028
Warning: contents of previous BIFILE will be lost. Proceed? (Y/N) y
sb2 kdbhavsp                                @126      8116

BBED> assign kdbh.kdbhavsp=8028
sb2 kdbhavsp                                @126      8028

BBED> sum apply
Check value for File 1, Block 2425:
current = 0x8014, required = 0x8014

BBED> verify dba 1,2425
DBVERIFY - Verification starting
FILE = /mnt/ramdisk/book/system01.dbf
BLOCK = 2425

Block Checking: DBA = 4196729, Block Type = KTB-managed data block
data header at 0x16d5e74
kdbchk: space available on commit is incorrect
        tosp=8039 fsc=0 stb=0 avsp=8028
Block 2425 failed with check code 6111

--// tosp = avsp+stb+fsc= 8028+0+0 = 8028
BBED> assign kdbh.kdbhtosp=8028
sb2 kdbhtosp                                @128      8028

BBED> sum apply
Check value for File 1, Block 2425:
current = 0x802f, required = 0x802f

BBED> verify dba 1,2425
DBVERIFY - Verification starting
FILE = /mnt/ramdisk/book/system01.dbf
BLOCK = 2425

--//如今設置"_db_always_check_system_ts"=true,在執行dml就不錯報錯了.

SYS@book> alter system set "_db_always_check_system_ts"=true scope=memory ;
System altered.

SYS@book> update t set name='Tttt' where id=2;
1 row updated.

SYS@book> commit ;
Commit complete.

SYS@book> select rowid,t.* from t;
ROWID                      ID NAME
------------------ ---------- ----
AAAWEgAABAAAAl5AAA          1 test
AAAWEgAABAAAAl5AAB          2 Tttt