Re: Peculier raidframe problems

[Greg Oster <oster%netbsd.org@localhost>]

On Fri, 19 Dec 2014 21:24:06 +0000
Dave Tyson <dtyson%anduin.org.uk@localhost> wrote:

> On 12/19/14 15:10, Greg Oster wrote:
> > On Fri, 19 Dec 2014 15:03:48 +0000
> > Dave Tyson <dtyson%anduin.org.uk@localhost> wrote:
> >
> >> On 12/19/14 00:49, Greg Troxel wrote:
> >>   > Does dmesg show any disk errors?  Can you dd from the disk
> >>   > to /dev/null without errors?
> >>   >
> >>
> >> No disk errors reported. I started a dd of /dev/rwd1d
> >> to /dev/nulland this was over halfway through when the UPS threw a
> >> wobbler and caused a reboot. When the system came back raid0
> >> showed both disks optimal.
> >>
> >> No idea why the problem arose or how it fixed itself - but I'll
> >> take fs dump to tape which should flush out any problems...
> >>
> >> Sorry for the noise!
> > Hmmmmm.... it shouldn't just 'fix' itself like that..... I'd do a
> > 'rebuild-in-place' ('raidctl -R')  on whatever component was last to
> > join... otherwise you run the risk of at some point reading in
> > corrupt data...
> >
> > Later...
> >
> > Greg Oster
> 
> Hmm. I did a rebuild in place for /dev/wd1a which was OK, but after a 
> shutdown and reboot its marked as failed again :-( A full dump of the 
> filesystems on raid0 was fine and the block counts are about what is 
> expected.
> 
> I think it might be easier to restore the dump to a spare disk, boot
> off that having marked raid0 as non-autoroot and destroy/recreate it
> and then restore the fiesystems. Unless there is anything else worth
> trying...

This is starting to smell like your drives arn't flushing their caches
or something.... That is, the kernel is telling the drives to write out
bits, but those bits never make it to the platters... :( 

Later...

Greg Oster