Re: LGWR: terminating instance due to error 27062

Ushkalo Igor <igorus_at_protek.ru> wrote:
>Hi, All!
>
>I have: Oracle 8.0.5.1 + Unixware 7.1.1 (with latest patches)
>Compaq Proliant8500 (8 x Xeon / 4GbRAM / a lot of HDD's)
>Database: OLTP + DSS - total about 35Gb datafiles, up to 350
 connects.
>There are RAW-devices: Redo-logs, Rollback & Temporary.
>
>Third time i got crash:
>
>!Thread 1 advanced to log sequence 67434
>! Current log# 4 seq# 67434 mem# 0:

 /d_ora_sys/ORC/RAW/log04-1ORC.ora
>! Current log# 4 seq# 67434 mem# 1:

 /d_ora_sys/ORC/RAW/log04-2ORC.ora
>!LGWR: terminating instance due to error 27062
>!Instance terminated by LGWR, pid = 19410
>!
>!ORA-27062: skgfospo: could not find pending async I/Os
>!Cause: This is an internal error. There should have been some
>!async I/Os in the system, but a blocking aiowait indicates
>!that there are no more I/Os.
>!Action: Look for trace file, and contact Oracle Support.
>
>After that I can startup instance and everithing seems to be Ok.
>First time crash was on another server, but DB was the same
>(a few monthes ago). Second and third times this error appeared
>during last two weeks...
>Yes, I'll call to support on Monday, but I guess these Oracle
 guys won't
>help me...
>
>People, what should I do with configuration of async I/Os?
>Help, please.....
>
>-- Best regards, Igor Ushkalo (igorus!) - igorus(at)mail.ru,
> ICQ #19972198
>
>

The ORA-27062 error means that the process (LGWR in this case) issued an aiowait() system call to wait for any outstanding aiowrite() request to complete, but none completed within some timeout value (600 seconds, I think). This is quite a long time for a write request to go unfulfilled. Either you were having very serious I/O issues at the time, or there is a bug with some OS driver. I've seen this same problem on a different OS while using a popular volume manager and its corresponding filesystem product. An update to the filesystem product fixed the problem.

The difficulty in diagnosing this problem is that the database usually crashes before you realize there is a problem (even though the problem had been occurring for at least 10 minutes). Once the database crashes, all processes are gone and any chance of debugging the problem further is lost. Oracle may be able to provide a debug patch to prevent the database from crashing, instead just reissuing the aiowait() call and printing a message to the alert.log or dropping a trace file.

Hope this helps, Doug.

Got questions? Get answers over the phone at Keen.com. Up to 100 minutes free!
http://www.keen.com Received on Sun Aug 13 2000 - 00:00:00 CDT