Re: Xfs Access to block zero exception and system crash

[Sagar Borikar <sagar_borikar@xxxxxxxxxxxxxx>]

Sagar Borikar wrote:
> Dave Chinner wrote:
> > On Sat, Jun 28, 2008 at 09:47:44AM -0700, Sagar Borikar wrote:
> >   Device Boot Start End Blocks Id System
> > > /dev/scsibd1             126         286       20608   83  Linux
> > > /dev/scsibd2             287        1023       94336   83  Linux
> > > /dev/scsibd3            1149        1309       20608   83  Linux
> > > /dev/scsibd4            1310        2046       94336   83  Linux
> > >     
> >
> > I'd have to assume thats a flash based root drive, right?
> >
> >   
> That's right,
> > > Disk /dev/md0: 251.0 GB, 251000160256 bytes
> > > 2 heads, 4 sectors/track, 61279336 cylinders
> > > Units = cylinders of 8 * 512 = 4096 bytes
> > >
> > > Disk /dev/md0 doesn't contain a valid partition table
> > >
> > > Disk /dev/dm-0: 107.3 GB, 107374182400 bytes
> > > 255 heads, 63 sectors/track, 13054 cylinders
> > > Units = cylinders of 16065 * 512 = 8225280 bytes
> > >     
> >
> > Neither of these tell me what /dev/RAIDA/vol is....
> > It is the device node to which /mnt/RAIDA/vol is mapped to. Its a 
> > JBOD with 233 GB size.
> >  
> > > But still the issue is why doesn't it happen every time and less 
> > > stress?
> > >
> > > I am surprised to see to let this happen immediately when the
> > > subdirectories increase more than 30. Else it decays slowly.
> > >     
> >
> > So it happens when you get more than 30 entries in a directory
> > under a certain load? That might be an extent->btree format
> > conversion bug or vice versa. I'd suggest setting up a test based
> > around this to try to narrow down the problem.
> >
> > Cheers,
> >
> > Dave.
> >   
> Thanks for all your help. Shall keep you posted with the progress on 
> debugging.
>
> Regards
> Sagar
Sorry if I was not clear.  As I mentioned the frequency of finding bad 
extents is much higher
when I increase simultaneous transactions to 30 ( say in 5 min ) but if 
I run only
two copies in infinite loop, the issue crops up in 2-3 hours roughly. 
And all the copies plus pdflush
are in uninterruptible sleep state continuously. And it is not 
uninterruptible sleep and waiting state ( DW )  but
just uninterruptible ( D ). 

Thanks
Sagar