Her fix? Not shares. Not limits. Reservations. She right-clicked the VMs, went to Resources, and locked down 4 GHz of dedicated CPU per node. Then she did the same for memory—all 8GB, reserved and pinned. No ballooning. No swapping. It was ugly from a cluster efficiency standpoint, but it was safe .
It was working.
CUCM's virtualized heartbeat timers are notoriously sensitive. In a physical world, a 200ms delay is a shrug. In a hypervisor, if the ESXi host gets busy, that same delay can trigger a "node isolation" event. The cluster would split-brain faster than you could say "call manager group." cucm virtualization