Re: [suse-oracle] 10.2.0.2l.0 + SLES 10 +SP2+ext3 +filesystemio_options=setall --> archiver error

Hi All:

This is just to update that this problem is fixed and fix is released in latest SLES10 kernel update (Novell Bug#413007):

SLES10 SP2: 2.6.16.60-0.29 (all but x86_64) and 2.6.16.60-0.30 (x86_64)
SLES10 SP1: 2.6.16.54-0.2.10 (all but x86_64) and 2.6.16.54-0.2.11 (for x86_64)

Thanks,
Arun

>>> On 7/25/2008 at  8:22 AM, "Arun Singh" <[email protected]> wrote:
> Hi Roni,
> 
> Appears to be bug in unique combination of "ext3 + directio" under I/O load.
> 
> Please file bug with Novell support with all possible details to get 
> analyzed and fixed.
> 
> Thanks,
> Arun
> 
>>>> Ronny Egner <[email protected]> 7/25/2008 4:08 AM >>>
> 
>> Some updates to this case:
>> 
>> Even with filesystemio_options=directio (i.e. *no* async
>> being used) the error can be reproduced with 10.2.0.2.0
>> on the lastest kernel (2.6.16.60-0.25-smp).
>> 
>> Using reiserfs with filesystemio_options=setall or
>> using no io-options with ext3 works fine.
>> 
>> I have contacted Arun Singh regarding this problem
>> and he told me with 10.2.0.3.0/10.2.0.4.0 this problem
>> does not exist.
>> 
>> I will check this now. However for my company this is not an option
>> because we are stuck at 10.2.0.2.0 at the moment.
> 
> Some furhter updates:
> 
> 
> I have updated the database to 10.2.0.4.0 (64-bit).
> 
> Even with 10.2.0.4.0 and the latest kernel i am able to reproduce
> the error. Database is Standard Edition (but i double oracle has
> different kernel für enterprise and standard) and file system
> is ext3 with mount options "rw,acl,user_xattr".
> 
> 
> Is anybody able to reproduce this error?
> 
> It is quite simple: Size your redo logs to 200 MB each
> and create a big table (2 GB in my testcase) and copy the
> table with "create table test as select * from <name>".
> 
> Important here is to enable "filesystemio_options=setall"
> or "filesystemio_options=directio".
> 
> After a few minutes the error should appear. It semms to be
> i/o-load related because on smaller systems the error can be
> triggered with one single table copy whereas bigger systems
> take 2 to 3 session to force this error.
> 
> 
> For those that are interesed the output from strace
> of the archiver process (10.2.0.4.0 this time!):
> 
> ******************************************************
> 
> pwrite(24, "\1\"\0\0\1\350\4\0-\0\0\0\250\200\347Fhshshshs,\0\1\f "..., 
> 1048576, 164626944) = -1 EIO (Input/output error)
> 
> ******************************************************
> 
> 
> -- 
> Mit freundlichen Grüßen
> 
> Ronny Egner
> Diplom-Ingenieur (BA)
> Systeme & Service
> Oracle Certified Professional (OCP)
> 
> Telefon: +49 381 2524-422
> Telefax: +49 381 2524-399
> 
> 
> SIV.AG - Service für Informationsverarbeitung AG
> Hauptsitz: Konrad-Zuse-Str. 1, 18184 Roggentin
> Handelsregister: Amtsgericht Rostock, HRB 8677, Ust.-IdNr.: DE 137477226
> Vorstand: Jörg Sinnig (Vorsitzender), Andreas Lehmann, Arno Weichbrodt
> Aufsichtsratsvorsitzender: Thomas Huth
> 
> *************************************************************************
> Aus Rechtsgründen ist die in dieser E-Mail gegebene Information nicht
> rechtsverbindlich. Eine rechtsverbindliche Bestätigung reichen wir
> Ihnen auf Anforderung in schriftlicher Form nach. Diese Nachricht ist
> ausschließlich für den Adressaten oder dessen Vertreter bestimmt.
> 
> The information contained in this email is not legally binding.
> At your request, we will provide you with a legally binding confirmation
> in written form. This message is intended solely for the addressee,
> entity to which the email is addressed or the authorised agent.
> 
> *************************************************************************
> 
> 
> 
> 
> 
> _______________________________________________
> suse-oracle mailing list
> [email protected] 
> http://listx.novell.com/mailman/listinfo/suse-oracle 


_______________________________________________
suse-oracle mailing list
[email protected]
http://listx.novell.com/mailman/listinfo/suse-oracle