Re: Nasty bug?

[Paul Furness <paul.furness@vil.ite.mee.com>]

Just to update everyone who was kind enough to make some suggestions
about how to trace this problem: It seems to be fixed now, and it looks
like it came down to the old chestnut of having the most recent versions
of everything important. Here's what I did (for those interested in
building a production data server):

I completely formatted the server and the raid disks, building from
scratch with a clean install of RedHat linux 7.3. I then applied all the
updates for the installed packages (most of which wouldn't have affected
this problem anyhow, to be fair). The raid is a hardware raid device, so
needs no configuration from within linux.

I then took a virgin copy of the source for 2.4.21 kernel, patched it
with the latest stable version of XFS (I'm not sure of the version
number - how do you work out what version of XFS you are using?) and
also the latest stable LVM, and compiled it.

Finally, I installed all the xfs packages and all the LVM packages.

I then build all the LVM partitions I wanted, and formatted them all
using xfs. No problems at all occurred during the build, and after
setting up all the other bits I needed (nfs shares and so on) and
rebooting the box, it all seems to be working absolutely flawlessly.

I can't come up with any particular reason why it gave me problems in
the first place; the build process I used this time was basically a more
streamlined version of what I did last time. On the other hand, it's a
newer kernel version than before, and this might well be what fixed it.

Thanks again to everyone involved with xfs for a really nice bit of
software!

Paul.


On Fri, 2003-08-01 at 17:24, Bogdan Costescu wrote:

> On 1 Aug 2003, Paul Furness wrote:
> 
> > First few were fine, then the same problem came back again.
> 
> This coupled with the fact that mkfs.ext2 worked fine (writes only in some 
> places on the partition) suggest to me that there might be problems with 
> the partition itself, either hardware or software (LVM). To check this I 
> would suggest making again one big partition and try to read it with 'dd'; 
> this will go through every (virtual) sector of the partition; 'badblocks' 
> might also be used...


[[HTML alternate version deleted]]