[Sheepdog] How to use Sheepdog as network storage?
morita.kazutaka at lab.ntt.co.jp
Thu Jul 21 09:10:17 CEST 2011
At Wed, 20 Jul 2011 12:52:30 -0700 (PDT),
luc_rom at yahoo.com wrote:
> I'm new to this list and I'm sorry if this question was already answered, but I'm testing Sheepdog as a backend storage and I didn't find a way yet to use it as network storage.
> Our plan is to have compute nodes running KVM that would use the Sheepdog cluster for space, basically the same setup as presented on page 22 of this presentation: http://www.osrg.net/sheepdog/_static/kvmforum2010.pdf
> In my test lab I have three Sheepdog nodes configured and I was able to create a qemu image that was distributed across all three nodes (using the "qemu-img create sheepdog:imagenew -o
> preallocation=data 100G"), Then I used the patched tgt to present it to a KVM client as iSCSI target, but the problem is that if the Sheepdog node that I use a iSCSI target crashes, the KVM client has to rediscover the target from another Sheepdog node, which is a lengthy manual process.
It is an experimental feature to use Sheepdog as something like a
network storage. The automatic connection failover between storage
server and VM is not supporeted.
If you want to do failover automatically, you need to export the same
LUN on the multiple Sheepdog nodes and the VM needs to access the
iSCSI targets with a multipath configuration.
More information about the sheepdog