[ClusterLabs] Names of virtualdomain resources

[David Vossel]

----- Original Message -----
> 
> > On 4 Mar 2015, at 11:46 pm, Simon Lawrence <simon.lawrence at navaho.co.uk>
> > wrote:
> > 
> > Hi
> > 
> > In section 4.3.2 of this document...
> > 
> > http://clusterlabs.org/doc/en-US/Pacemaker/1.1/html-single/Pacemaker_Remote/
> > 
> > the following command is used to create a virtual domain resource...
> > 
> > pcs resource create vm-guest1 VirtualDomain hypervisor="qemu:///system"
> > config="/root/guest1.xml" meta remote-node=guest1
> > 
> > i.e. the resource is named 'vm-guest1'.
> > 
> > 
> > However, the KVM guest this resource is associated with has a KVM domain
> > name of 'guest1'.
> > 
> > My understanding is that resource name and KVM domain name should be the
> > same for the cluster to be able monitor that VM. Is that not the case?

Kind of. I'm confused by 'domain name'. In kvm the VMs are called domains.
In networking, a domain name is talking about a hostname.  If we're talking
about hostnames, then you are correct. If we're talking about the internal
kvm naming used for a VM, then no. 

When you are using the remote-node meta attribute, if you don't specify the
remote-addr attribute then pacemaker assumes the value, guest1, in
'remote-node=guest1' is a hostname pacemaker_remote can be contacted on. So,
guest1 must be a network reachable hostname that points to the virtual machine
associated with the remote node.

> I think its preferred (and makes fencing easier) but not required.
> David?

For the baremetal remote node use case, this could matter depending on the fencing
devices in use. For the container node use case (which is what is being discussed
here) this is not a concern. Container nodes are fenced, but the fencing action is
just destroying the VM using the VirtualDomain resource agent. No special
configuration is needed for this to happen. It just happens automatically.

-- David