今天有個比較老的系統啓動的時候報錯了：
[findb@oraprod]$ sqlplus "/ as sysdba"
SQL*Plus: Release 9.2.0.6.0 - Production on Wed Nov 28 23:15:22 2012
Copyright (c) 1982, 2002, Oracle Corporation. All rights reserved.
Connected to an idle instance.
SQL> startup
ORACLE instance started.
Total System Global Area 3887045192 bytes
Fixed Size 748104 bytes
Variable Size 2264924160 bytes
Database Buffers 1610612736 bytes
Redo Buffers 10760192 bytes
Database mounted.
ORA-27061: skgfospo: waiting for async I/Os failed
IBM AIX RISC System/6000 Error: 22: Invalid argument
ORA-27061: skgfospo: waiting for async I/Os failed
IBM AIX RISC System/6000 Error: 22: Invalid argument
ORA-27061: skgfospo: waiting for async I/Os failed
IBM AIX RISC System/6000 Error: 22: Invalid argument
ORA-27061: skgfospo: waiting for async I/Os failed
IBM AIX RISC System/6000 Error: 22: Invalid argument
從報錯內容來看這個是和AIX的異步io參數設置有關，Google了一下，發現者竟然還是AIX在5300-06上的一個Bug：IZ03260: LIO_LISTIO FAILS TO UPDATE AIO CONTROL BLOCKS ON ERROR APPLIES TO AIX 5300-06。
這個由於異步IO引起的問題，在AIX 5.3 TL3下，使用文件系統，打開Oracle的異步IO選項的時候，經常會出現。
當異步IO的請求（隊列）超出MAXREQS的時候，由於AIX上AIOWAIT的問題，會導致返回值是錯誤的，Oracle發現
該錯誤後，就會出現問題。如果正好是某些重要的後臺進程碰到這樣的錯誤，會導致數據庫宕掉。
解決方法（三選一，其中第二和第三可以一起使用）：
1、關閉數據庫的異步IO（最後沒辦法的時候採取的措施）
2、修改異步IO參數，加大MAXREQS到8192，甚至16384
chdev -l aio0 -a maxreqs=16384 --或者8192
要不啓動系統，動態修改請使用下面命令（臨時設置，重啓不起作用）
aioo -o maxreqs=16384 或者 8192
3、安裝IBM的IZ03260補丁包
【Oracle官文】
|
Database Crashes With ORA-27061 on AIX [ID 452866.1] |
|
修改時間:2012-7-20類型:PROBLEM狀態:PUBLISHED優先級:3 |
In this Document
Applies to:
Oracle Server -Enterprise Edition - Version 9.2.0.8 to 11.1.0.7 [Release 9.2 to 11.1]
IBM AIX on POWER Systems (64-bit)
***Checked for relevance on 20-Jul-2012***
-- Problem Statement:
Periodical database crashes on AIX with the following errors:
ORA-27061:skgfospo: waiting for async I/Os failed
IBM AIX RISC System/6000 Error: 22: Invalid argument
Oracle is waiting for an async. I/O tocomplete.
aiowait() is the operating System functionon AIX to handle these things and it did not return the requested data in anappropriate time to complete the I/O.
The problem is always encountered by the DBWR processes by AIX BUG for
* Systems running AIX 5.3 with the bos.rte.aio fileset at the 5.3.0.60or 5.3.0.61 level.
Please refer to
IZ03260: LIO_LISTIO FAILS TO UPDATE AIO CONTROL BLOCKS ONERROR ( http://www-1.ibm.com/support/docview.wss?uid=isg1IZ03260)
NOTE:
The issue may affect any Oracle version running on (and certified for) theactual AIX version (including TL and SP) that would include this bos.rte.aiofileset.
For certification, please refer to My Oracle Support.
For specific install requirements, please refer to Note 169706.1
Increase the max limit for AIO requests(maxreqs) using the chdev command.
The command to make the change permanent after reboot is:
chdev -l aio0 -amaxreqs=<value>
The command to make the change effective dynamically is:
aioo -omaxreqs=<value>
To check current Setting for maximum requests (maxreqs) :
lsattr -El aio0
Or Install Fix APAR IZ03260.
Prior availability of IZ03260, and interimfix is available from:
ftp://ftp.software.ibm.com/aix/efixes/iz03260/
As a workaround, DBWR IO slaves can beincreased to handle the IO workload. However, this is an issue that needto get addressed to the hardware vendor (IBM) to check the driver and I/Osystem.
-- To implement the workaround, pleaseexecute the following steps::
SQL> altersystem set DBWR_IO_SLAVES = 4 scope=spfile;
SQL> shutdown immediate
SQL> startup
【IBM官文】
IZ03260: LIO_LISTIO FAILS TO UPDATE AIOCONTROL BLOCKS ON ERROR APPLIES TO AIX 5300-06
窗體頂端
You can track all active APARs for thiscomponent.
窗體底端
When the maximum requests (maxreqs) isexceeded,
lio_listio
will update the control blocks but may notreturn an
error
to indicate that some IOs were not started.This results
in
an unexpected EINVAL from aio_nwait laterwhen these IOs
are waited for.
****************************************************************
* USERS AFFECTED:
* Systems running AIX 5.3 with thebos.rte.aio fileset at the
* 5.3.0.60 or 5.3.0.61 level.
* This problem is known to affect Oracle.
****************************************************************
* PROBLEM DESCRIPTION:
* When the maximum requests (maxreqs) isexceeded, lio_listio
* will update the control blocks, but maynot return an error
* to indicate that some I/Os were notstarted. This results
* in an unexpected EINVAL from aio_nwaitlater when these I/Os
* are waited for.
* Errors similar to the following may belogged by Oracle:
* ORA-27061: waiting for async I/Os failed
* IBM AIX RISC System/6000 Error: 22:Invalid argument
****************************************************************
* RECOMMENDATION:
* Increase the max limit for AIO requestsusing the chdev
* command.
* Install APAR IZ03260.
* Prior availability of IZ03260, andinterim fix is available
* from: ftp://ftp.software.ibm.com/aix/efixes/iz03260/
****************************************************************
lio_listio should return proper error codes
on failures.
*********
* HIPER *
*********
5300-06 - use AIX APAR IZ03260
5300-07 - use AIX APAR IZ03475
IZ03260
AIX 5.3
5765G0300
530
CLOSED PER
NoPE
YesHIPER
2007-08-17
2007-08-20
2007-12-06
APAR is sysrouted FROM one or more of the following:
APAR is sysrouted TO one or more of the following:
AIX 5.3
5765G0300
UP07/11/28 I 1000
U812745 bos.rte.aio5.3.0.62
