根据oracle入库数据进行告警,Oracle 启动故障案例之--ORA-600 [4193]错误

ORA-600[4193] 这个错误也是与UNDO 有关系，MOS 上有几篇相关的说明文章.

一.MOS说明

1.1 ORA-600 [4193] WhenTrying To Open The Database [ID 763566.1]

Symptoms

Copying database from one server to another server and getting an ORA-600 [4193] error when trying to open the database on the destination server.

--copy 数据库从一个server 到另一个server 后，尝试打开时报这个错误。

Cause

The online redo logs were copied when the source database was open, online redo logs should never be copied when the database is open.

--导致原因是因为在数据库open时把online redo logs 也一起copy 过去了。 在数据库open状态，online redo log 不应该copy。

Solution

In this instance the datafiles were being copied properly after the tablespaces were put in to backup mode, however, online redo logs should only be copied if the source database is shutdown first before copying the online redo logs.  The source database needed to remain open so, the datafiles were copied again (withthe tablespaces in backup mode) and then a number of archive logs were transferred over to the new server and after the last archivelog was applied the database could be opened with resetlogs and new online redo logs were created on the destination server.

--当表空间被设置为backup 模式之后，可以copy 数据文件，但是onlineredo log 只能是在数据库shutdown 之后才能copy，如果数据库一直是open 状态，那么只能把datafile copy 过去，然后把归档文件传送过去，最后用openresetlogs的方式打开数据库，在open时online redo log 会自动重建。

1.2 Ora-600 [4193] WhenOpening Or Shutting Down A Database [ID 452662.1]

1.2.1 Symptoms

Errors in alert.log:

Tue Jul 17 13:38:13 2007

Errors in file /home/Oracle/oracle/product/10.2.0/yms/rdbms/log/yms_smon_8337.trc:

ORA-00607: Internal error occurred while making a change to a data block

ORA-00600: internal error code, arguments: [4193], [3552], [3554], [], [], []

yms_smon_8337.trc:

SO: 0xdfaec728, type: 24, owner: 0xdf266580, flag: INIT/-/-/0x00

(buffer) PR: 0xdf1f1338 FLG: 0x1000

class bit: 0x80000

kcbbfbp: [BH: 0xded4bf40, LINK: 0xdfaec768]

kcbbfbx[0]: [BH: 0xdece41d8, LINK: 0xdfaec788]

where: ktuwh01: ktugus, why: 0

buffer tsn: 2 rdba: 0x00c00002 (3/2)

scn: 0x0000.03c95628 seq: 0x01 flg: 0x00 tail: 0x56280e01

frmt: 0x02 chkval: 0x0000 type: 0x0e=KTU UNDO HEADER W/UNLIMITED EXTENTS

BH (0xdece41d8) file#: 3 rdba: 0x00c003b6 (3/950) class: 20 ba: 0x11d6ba000

set: 6 blksize: 8192 bsi: 0 set-flg: 0 pwbcnt: 0

dbwrid: 0 obj: -1 objn: 0 tsn: 2 afn: 3

hash: [df870f70,df870f70] lru: [dece4488,dece4028]

obj-flags: object_ckpt_list

ckptq: [dedac4a0,ded47cb8] fileq: [dedac500,ded47cc8] objq: [ded47d78,db7bfd78]

use: [dfaec788,dfaec788] wait: [NULL]

st: XCURRENT md: EXCL tch: 0

flags: mod_started gotten_in_current_mode block_written_once

change state: ACTIVE

change count: 1

LRBA: [0xac3.4de07.0] HSCN: [0xffff.ffffffff] HSUB: [65535]

Using State Objects

----------------------------------------

SO: 0xdfaec728, type: 24, owner: 0xdf266580, flag: INIT/-/-/0x00

(buffer) PR: 0xdf1f1338 FLG: 0x1000

class bit: 0x80000

kcbbfbp: [BH: 0xded4bf40, LINK: 0xdfaec768]

kcbbfbx[0]: [BH: 0xdece41d8, LINK: 0xdfaec788]

where: ktuwh01: ktugus, why: 0

buffer tsn: 2 rdba: 0x00c003b6 (3/950)

scn: 0x0000.03be3c7d seq: 0x5a flg: 0x04 tail: 0x3c7d025a

frmt: 0x02 chkval: 0x0868 type: 0x02=KTU UNDO BLOCK

----------------------------------------

Error 607 in redo application callback

TYP:0 CLS:20 AFN:3 DBA:0x00c003b6 OBJ:4294967295 SCN:0x0000.03be3c7d SEQ: 90 OP:5.1

ktudb redo: siz: 132 spc: 4462 flg: 0x0012seq: 0x0de2 rec: 0x09

UNDO BLK:

xid: 0x0002.045.00006c61seq:0xde0 cnt: 0x60 irb: 0x60 icl: 0x0 flg: 0x0000

1.2.2 Cause

When we try toapply redo to an undo block (forward changes are made by  the applicationof redo to a block) we check that the seq# in the undo  record matches the seq# in the redo record.

--数据库在启动时需要进行一个前滚的操作，在前滚时会应用redo 到undo block上，操作时会检查undorecord里的seq#和 redo record里的seq#.

These seq# should be the  same because when we apply a redo record we must apply itto the correct version of the block.

--正常情况下，这2者的seq# 应该是一致的。

We can only apply a redo record to a  block that contains the same seq# as in the redo record.

--在一致的情况下，我们才应用redo record 到undo record。

If the seq# do not match then ORA-600[4193][a].[b] is raised. .

Arg [a] Undorecord seq number --> seq: 0xde0 = 3552

Arg [b] Redo record seq number --> seq: 0x0de2   = 3554

--如果不一致就会出现ORA-600[4193][a][b]的错误。其中a 是undo 里的seq#记录，b是redo 里的seq# 值。 这里的值都是十六进程，我们可以通过to_number() 这个函数来转换一下：

SYS@anqing1(rac1)>  Select to_number('de0','xxxx') from dual;

TO_NUMBER('DE0','XXXX')

-----------------------

3552

This implies some kind of block corruptionin either the redo or the undo block.

--当redo record 和 undo record 不一致时，就会抛出ORA-600[4193]的错误。

相关的文章参考：

1.2.3 Solution

1.2.3.1 If Database is opened:

--在db open 状态下，解决的方法如下：

1) Find out the rollback segment, based onthe first part of the xid: 0x0002.045.00006c61

usn=2 is the segment_id

select segment_name,status from dba_rollback_segs where segment_id=2;

RS_DATA1   ONLINE

2) Dump the transaction table of the rollbacksegment to see if all TX are commited:

alter system dump undoheader RS_DATA1;

3) check the trace file created underuser_dump_dest

In the trace file search for the Keyword "TRN TBL"

TRN TBL::

index state cflags wrap#   uel   scn            dba

-----------------------------------------------------------------------------

0x00   9     0x00 0x21eb1 0x0023 0x0000.d28c43e9 0x00000000 ......

state=9 means transaction is committed

4) offline the rollback segment:

alter rollback segment rs_data1 offline;

select status from dba_rollback_segs where segment_id=2;

5)   if STATUS=OFFLINE

drop rollback segment RS_DATA1;

1.2.3.2 If Database doesn't open:

--如果数据库不是open状态，处理方法如下：

1.   a) If using rollback segments,remove the rollback_segments line from init.ora, and open database

b) If using undo segments setundo_management = manualin init.ora/spfile, and try to opendatabase.

2. If database opens means all transactions are committed, and you can drop the rollback segment or the undo tablespace

1.3 bug 导致的ORA-600[4193]

MOS:

ORA-600 [4193] "seq# mismatch while adding undo record" [ID 39282.1]

Bug 8240762 - Undo corruptions with ORA-600[4193]/ORA-600 [4194] or ORA-600 [4137] [ID 8240762.8]

Undo corruptionmay be caused after a shrink and the same undo block may be used for two different transactions causing several internal errors like:

ORA-600 [4193] / ORA-600 [4194] for new transactions

ORA-600 [4137] for a transaction rollback

Undo segment shrink is internally done by Oracle.

--undo shrink 导致的undo corruptions

Workaround

Drop the undo segment.

Affects:

Product (Component)

Oracle Server (Rdbms)

Range of versions believed to be affected

Versions >= 10.2 but BELOW 11.2

Versions confirmed as being affected

Platforms affected

Generic (all / most platforms affected)

Fixed:

在Oracle 10.2 以上到11.2 的DB 会受Bug 8240762的影响导致undo 的corruption。在10.2.0.5 中已经修复了这个bug。如果出现这种问题，drop 对应的undo segment 即可。