RE: Oracle 10g hangs intermittently waiting for I/O

From: Matthew Zito <mzito_at_gridapp.com>
Date: Fri, 15 May 2009 09:57:19 -0400
Message-ID: <C0A5E31718FC064A91E9FD7BE2F081B101D4B922_at_exchange.gridapp.com>

If you run a "dmesg" - do you see any errors in the kernel logs? If the devices stop responding to I/O for periods of time there should be SCSI timeouts in the logs, or at least some warnings from the multipathing driver.

Thanks,
Matt

--

Matthew Zito
Chief Scientist
GridApp Systems
P: 646-452-4090
mzito_at_gridapp.com
http://www.gridapp.com

-----Original Message-----

From: oracle-l-bounce_at_freelists.org on behalf of Pawel Kotlarz Sent: Fri 5/15/2009 9:48 AM
To: oracle-l_at_freelists.org
Subject: Oracle 10g hangs intermittently waiting for I/O  

Hello all.

I have oracle 10.2.0.3 data warehouse database on 11.1.0.7 ASM with asmlib. RHEL 4.7. Proliant DL585 G2 with MSA70 storage.

The problem I face is an 'I/O hiccup'. The database can work properly for a week or two and then suddenly keep stalling for no apparent reason. Users complain that their selects take 2x or 3x more time. vmstat shows I/O activity (bi, bo colums) for half a minute and for another half a minute shows no activity (bi and bo columns equal to 0) and a number of processes waiting for I/O (procs/b column). strace on an oracle process waiting for I/O shows it is waiting for a completion of 'read' call. The only thing that helps is rebooting the box.

I can isolate the problem to specific disks using iostat. These disks are the same on a day the problem occurs but they are different on another occurrance of the problem. Storage / Linux admins do not see any problem on their side.

I have several one-off patches recommended by Oracle support:

Bug 5452672: Hung database instance if linux kernel miss aio request
Bug 6656824: LNX-10204-TC6  SIGSEGV AT SKGFR_REAP64()+281, IN DBW0
Bug 6087207: WARNING:ORACLE PROCESS RUNNING OUT OF OS KERNEL I/O RESOURCES

Bug 6882513 - MERGE LABEL REQUEST ON TOP OF 10.2.0.3 FOR BUGS 6801535 5576584
Bug 5576584 (4880399): ASM PARALLEL READS PERFORMANCE NOT ACCEPTABLE

I plan to upgrade to 10.2.0.4 but need first to sort out some hash join bugs (yet unknown to Oracle) that break our large queries with ora-600 errors.

What would you recommend to do to narrow down the problem to Oracle / ASM / asmlib / Linux / storage fault?

Do you know of any other bugs that can show such a behaviour?

Thanks.

Pawel Kotlarz
--

http://www.freelists.org/webpage/oracle-l

--

http://www.freelists.org/webpage/oracle-l Received on Fri May 15 2009 - 08:57:19 CDT