RHEVM : A question of fencing

Hello

My understanding is that although the hypervisor fencing is triggered via RHEV-M , the actual fencing is handled by the hypervisors. For instance, a restart of a hypervisor A is requested via RHEV-M webadmin but the fencing of the hypervisor A node is carried out by hypervisor B.

Is that correct ?     

If so, why not fence hypervisors directly from the Manager ?   I'm sure there's an obvious answer.

Thank You