[Sheepdog] General info

[MORITA Kazutaka]

At Thu, 19 May 2011 09:40:21 +0200,
Gandalf Corvotempesta wrote:
> hi all,
> i'm italian and I'm new to this really cool software
> I would like to ask some basic questions:
> 
> 1) as I've understood, sheepdog is a distributed storage system. So I can
> Install it on arbitrary number of storage server and see them as a single
> 'unit' from my virtual machines. Right ?

Yes, something like that.  You can create a single large storage pool
with multiple servers, and provide volumes from the pool to virtual
machines.

> 
> 2) is this high available? For example, if I'll create 3 server, i'll be
> sure that data will be replicated and one server failure will never make
> data loss?

Yes, it should behave so.

> 
> 3) example: i would like to create two KVM nodes, with 5 or 6 VM on each.
> Each node will be connected to the sheepdog cluster. VM1 is created on
> node1. In case of a huge splitbrain (storage1 down and node1 down) will be

Is node1 down?  Or is node1 still alive but cannot connect to the
second node?

In the former, VM1 can restart on the second node without data loss.
But automatic reconnection of the VM is not supported yet.

In the latter (split-brain), VM1 can alive on node1.  But after the
splitted nodes are merged, VM1 has a risk of reading an old data on
the second node.  In this case, it is safe to kill the second node
manually before merging.  It is one of the TODO items to handle
split-brain.

> possibile to restart the VM1 on the second node, taking datas from the
> second server in the cluster, with no data loss?
> 
> Do you have any configuration example to share?

Basically, there is no configuration for Sheepdog other than corosync
configuration.  See a wiki page for more details:
https://sourceforge.net/apps/trac/sheepdog/wiki/Getting%20Started


Thanks,

Kazutaka