Defective Disk not reported as Dead

Defective Disk not reported as Dead

[Oktay Akbal]

Hello !

We do have some strange problem here.
A Server with some qlogic qla2002f-Adapter (2-channel)is connected to 2
external Raid-Arrays via multipathing and raid1 on top of it. But the
multipathing and raid should not be the problem here.

The Raid-Arrays itself are Fibre-to-Ide and present themselfs as 1 disks
each. Now for the problem: Due to a firmware bug the raid-boxes sometimes
seems to loose the ability to write (and read i think) to their internal
disks. The effect is, that the cache fills up but does not get flushed to
disks. When full, the box is in a strange state. It gets detected when
reloading the kernel-modules but linux can no longer access the disks.
when accessing the partition all processes on that disk hang (like ls or
ps -efa etc.)

The disk is never recognized as defective and never thrown out of raid.
So the processes do not continue to work.

Is this fault simply not detectable, or could this be a problem with the
qlogic-driver ?

The Kernel used is some 2.4.18 Version by SuSE.

Thanks for help

Oktay Akbal

Re: Defective Disk not reported as Dead

[Oktay Akbal]

Hi Denis,

you should understand, that I really try not to reproduce the Problem,
since this is a VERY productive Database- and Fileserver.

So am seeking hints on known Bugs or just some hints.

For the Kernel-traces I assume that I would need some messages.
But for the kernel everything seems to be normal. At least there are
no messages in syslog.

Oktay

On Fri, 23 May 2003, Denis Vlasenko wrote:

> On 23 May 2003 10:23, Oktay Akbal wrote:
> > Hello !
> >
> > We do have some strange problem here.
> > A Server with some qlogic qla2002f-Adapter (2-channel)is connected to
> > 2 external Raid-Arrays via multipathing and raid1 on top of it. But
> > the multipathing and raid should not be the problem here.
> >
> > The Raid-Arrays itself are Fibre-to-Ide and present themselfs as 1
> > disks each. Now for the problem: Due to a firmware bug the raid-boxes
> > sometimes seems to loose the ability to write (and read i think) to
> > their internal disks. The effect is, that the cache fills up but does
> > not get flushed to disks. When full, the box is in a strange state.
> > It gets detected when reloading the kernel-modules but linux can no
> > longer access the disks. when accessing the partition all processes
> > on that disk hang (like ls or ps -efa etc.)
>
> This sounds like bug. Can you determine where exactly those processes
> are nailed? I bet you'll see them in D state, but folks will need
> more details, more precisely kernel stack backtraces.