Re: Oracle & linprocfs (was: Re: linux kernel-land)

On Wed, Jan 05, 2005 at 02:21:27PM +0000, Ceri Davies wrote:
> On Wed, Jan 05, 2005 at 11:24:22AM +0000, Ceri Davies wrote:
> > On Wed, Jan 05, 2005 at 12:11:10PM +0100, Alexander Leidinger wrote:
> > > On Tue, 04 Jan 2005 19:11:26 +0200
> > > Andriy Gapon <avg_at_icyb.net.ua> wrote:
> > > 
> > > [Full quote for the benefit of the people in the CC]
> > > 
> > > > BTW, I've got another linux kernel-land thingy: I have Oracle
> > > > installation made with previous default linux base (7), now after
> > > > upgrade to linux_base-8 (and linux_devtools-8) I can not start oracle
> > > > server (actually database instance) with the following error:
> > > > 
> > > > oracle$ ./dbora.sh start
> > > > 
> > > > SQL*Plus: Release 9.2.0.4.0 - Production on Tue Jan 4 18:49:51 2005
> > > > 
> > > > Copyright (c) 1982, 2002, Oracle Corporation.  All rights reserved.
> > > > 
> > > > SQL> Connected to an idle instance.
> > > > SQL> ORA-00600: internal error code, arguments: [kcbbnwa_1], [0], [0],
> > > > [], [], [], [], []
> > > > SQL> Disconnected
> > > > 
> > > > I see that there was at least another person who had the similar problem
> > > > with linux_base-8 a while ago:
> > > > http://lists.freebsd.org/pipermail/freebsd-database/2004-February/000149.html
> > > 
> > > Summary: don't mount linprocfs if you want to run Oracle with
> > > linux_base-8.
> > > 
> > > > Using ktrace I tracked it down to reading "/proc/stat" and after playing
> > > > around a little bit I found out that Oracle (or some redhat 8 lib that
> > > > Oracle used) probably missed one line there:
> > > > 
> > > > cpu 2010442 17367 694544 14507692
> > > > + cpu0 2010442 17367 694544 14507692
> > > > page 131869 24567
> > > > swap 1570 51873
> > > > intr 772076592
> > > > ctxt 1512209526
> > > > btime 1104346761
> > > > 
> > > > I see that "real linux" (Fedora Core 2) has this line too (yes, even in
> > > > the single CPU case).
> > > > 
> > > > After I fixed sys/compat/linprocfs/linprocfs.c to always produce cpu%d
> > > > line(s) my oracle started to behave again.
> > > > 
> > > > I am not sure if this is actually linux_base or linux_devtools related
> > > > or if this affects fresh Oracle installations.
> > > 
> > > I think this is a serious issue. Please send-pr the patch and tell us at
> > > current_at_ (CCed) the PR number. A src-committer should have a look at it.
> > > 
> > > I don't think it's linux_devtools related (it's easy to test it, right
> > > :-) ).
> > > 
> > > Do you have a 4.x system (or maybe ceri_at_ can test it on a 4.x system)? I
> > > like to know if we have the problem there too now (re_at_ CCed because they
> > > may decide to take this issue into account regarding the upcoming 4.11
> > > release... at least there should be a note somewhere to not use
> > > linprocfs and Oracle at the same time).
> > 
> > I don't have one running Oracle at the moment, and ten minutes ago I just
> > offered to work this evening.  I'll try to get out of this but otherwise
> > it will have be tomorrow evening (~36 hours time).
> 
> It looks like there is actually more time than I thought before the
> release, so I will definitely have time to check this out on a 4.x
> system.  Won't be today, as mentioned.

There doesn't seem to be a significant regression for Oracle on 4.x.
I've spent 4 hours on this this evening and the installer *still*
doesn't work properly on 4.x, with or without an "equivalent" patch to
linprocfs_misc.c.  This isn't new with the linux8 patch, and it's widely
believed that RELENG_5 or greater is a basic requirement for Oracle to
work under linux emulation (for 9i at least - I don't have earlier
versions to hand - marcel wrote the handbook section on this and may be
able to test Oracle 8.x).

Ceri
-- 
Only two things are infinite, the universe and human stupidity, and I'm
not sure about the former.			  -- Einstein (attrib.)