Re: Strange memory error during import ...

From: joel garry <joel-garry_at_home.com>
Date: Tue, 12 May 2009 14:43:01 -0700 (PDT)
Message-ID: <7f7a9952-34c4-40d6-a90e-af4c6f8c6bc3_at_d25g2000prn.googlegroups.com>



On May 11, 5:39 pm, zigzag..._at_yahoo.com wrote:
> On May 11, 3:53 am, Mladen Gogala <gogala.mla..._at_bogus.email.com>
> wrote:
>
>
>
>
>
> > Na Sat, 09 May 2009 12:20:53 -0700, zigzagdna napisao:
>
> > > I am on Oracle 10.2.0.3 on HP UNXI 11. When doing import after all
> > > tables have been imported,
> > > I get following emmory errors. I have seen it in 2 different database
> > > instances. No clue, why this error
> > > Occurs.
> > > sh: 3127 Memory fault
> > > sh: 3300 Memory fault
> > > sh: 3441 Memory fault
> > > sh: 3471 Memory fault
> > > sh: 3528 Memory fault
> > > sh: 3531 Memory fault
> > > sh: 3535 Memory fault
>
> > Because your script is screwed up and your shell is dying. How bright
> > do you have to be to use the -x flag and debug a little?
>
> > --http://mgogala.freetia.com
>
> My script is so simple, it is a simple imp statement.., full import
> (full=Y). During imp, user1 is impoted (all tables,,, towards the end
> when it comes to view stored procedure..., this error occurs, but
> subsequenet users user2, user3  etc get imported successfuly. set -x
> will not provide any useful information.

What imp puts out can get truncated when there is an error, so the last thing outputted may be before the actual error. Wild speculation here, but watch your swapinfo -am and check what indexes, SP's actually have been created given this hypothesis:

user1 completes. User2 has a big table with many indices. When you use the imp defaults (as opposed to creating the indices later), the indices are created after the table is imported, which uses lots of memory due to sorts, until they hopefully spill over into temporary tables. So my guess is you have insufficient swap space, and too big sort memory to handle everything until the spillover.

So _in addition_ to checking all the memory settings against Oracle recommendations with a fine tooth comb as everyone else has been telling you, check your swap space configuration, usage, OS patches and any pga/sga settings. Remember, when you have two or more instances on a unix box, you have to add up their memory requirements! (Beyond that, some things like aborting instances can keep shared memory, be sure to check with the ipc utilities. Be especially careful about which hp-ux 11, too, there are some minor differences that can lead to major consequences - I've seen one machine that has no problems with a fiber disk array on 11.11 and 11.3x, but major problems on 11.23 IIRC, my guess due to OS patches.). And as several people have pointed out, it could be a hardware problem - later versions of hp-ux have the ability to use memory as swap (yes, it makes sense if you investigate it, and I've become convinced Oracle dbconsole, at least, doesn't account correctly for it), and you may be hitting bad memory or disk with the swap. Also, Oracle has had obscure issues with SGA size and selfabu ^H^H^Htuning, in disparate configurations, that only show up when something with a different performance profile is run. And Itanium is different.

Are you setting LD_LIBRARY_PATH or similar things anywhere, or had any other 32bit/64bit types of errors in the past? Using bash instead of ksh? core files left around? gazillions of trace files in udump?

jg

--
_at_home.com is bogus.
Wasn't there a star trek episode about planet eating aliens?
http://www.bloomberg.com/apps/news?pid=20601087&sid=agSuuvCA2EAU&refer=home
Received on Tue May 12 2009 - 16:43:01 CDT

Original text of this message