[Sheepdog] No redundancy with sheepdog

MORITA Kazutaka morita.kazutaka at lab.ntt.co.jp
Wed Aug 25 23:36:33 CEST 2010


At Tue, 24 Aug 2010 16:32:45 +0100 (BST),
James Ray wrote:
> 
> All,
> 
> Started playing with sheepdog today, I have following the
> instructions at:
> http://www.osrg.net/sheepdog/usage.html
> Pulled the latest sheepdog git (last commit tag is
> 4d4d2a928fbc098ab1cf6afdd01bcad540afe7b4)
> 
> I dont seem to get any redundancy in my setup, killing sheep on any one of
> the boxes causes all IO to die. My full build details are lower down the
> email. But here is what I do to get a failiure:
> 
> Started converting an debian qcow file to sheepdog like this:
> [root at pavlov sheepdog]# qemu-img convert \
> debian_lenny_amd64_small.qcow2 sheepdog:debian
> 
> While it is converting I kill sheep on a machine with kill -9 (not very
> nice, but should be handled) and I get the following error from qemu-img:
> 
> I/O error
> qemu-img: error while writing
> 
> The same happends if I am running a VM. On restarting sheep things don't
> seem to get any better with all the 'collie vdi show' commands reporting:
> 
>    name        id    size    used  shared    creation time   vdi id
> ------------------------------------------------------------------
> error 11661488, 0
> 
> Any ideas?


Thanks for your report!  It seems that Sheepdog doesn't work properly
if the node membership changes during I/O operations are performed.  I
want to fix these critical bugs as soon as possible.

Thanks,

Kazutaka



More information about the sheepdog mailing list