Open Stack

Hi,

I'm trying to figure out how things are supposed to be done.

Suppose I'm running compute nodes using shared instance storage (via 
NFS, for example).

Now suppose I have a network issue which isolates a compute node.  I 
evacuate the instances that were on that node, which causes them to 
start running somewhere else.

Now we fix the network issue.

As far as I can tell, we now have two separate kvm instances, on two 
different machines, both using the exact same virtual disk file.  This 
seems like a bad idea.

Am I missing something?  Or do you need a STONITH device for compute 
nodes to safely evacuate instances from a failed compute node?

Chris