[sheepdog-users] Live migration of VM disk from one sheepdog node to another

[Bastian Scholz]

You could use a gateway sheep on each server and
let the VM connect to the local sheep process every
time.

Cheers

Bastian

Am 2014-03-07 20:05, schrieb Struan Bartlett:
> Hi
> 
> I am experimenting with different options for distributed storage of
> KVM/QEMU virtual machine disk images. I like the simplicity of setup
> of sheepdog+corosync, and the homogeneous nature of this installation
> (each server is the same). However, given our current distribution of
> raw metal servers, I'm not sure it makes sense to us that every VM
> host should be running sheepdog. This is because the raw metal servers
> we have with the greatest RAM and spare processing capacity, also have
> the strongest disk storage which is heavily utilised. Conversely, we
> have other raw metal servers that are shorter on RAM and spare CPU
> capacity, but have plenty of free disk space. Our VMs are used for
> development and testing purposes, and don't need to be high powered.
> It therefore seems to make sense to run sheepdog on the machines with
> plenty of free disk space, and the VMs on the machines with plenty of
> CPU and RAM.
> 
> I know this is possible with sheepdog, since the qemu driver supports
> talking to sheepdog on a remote host, and I've managed to get that
> working. What I don't know is, whether there are any plans to make it
> possible to migrate VMs backed by such storage from one sheepdog node
> to another. This is important for when a sheepdog server needs to be
> taken down for maintenance. For example, is it planned to either (a)
> allow the choice of remote sheepdog server backing the guest disk to
> be changed on-the-fly? (b) Allow a set of remote sheepdog servers to
> be specified for the guest disk, allowing the qemu driver to
> automatically failover from one to another in the event the sheepdog
> node becomes inaccessible? or (c) use a protocol such as rbd (like
> Ceph) that supports talking to multiple storage servers natively?
> 
> Kind regards
> 
> Struan