I would probably solve this in the following way:
- A load balancer must sit in front of the two or more real servers. The load balancer must have an API that can deactivate members from the pool.
- Use vRealize Operations to monitor the two VMs in question.
- Use the management pack for vRealize Orchestrator and connect vROps to vRO.
- Create a workflow that uses whatever method/plug-in is supported by the load balancer to deactivate a member. The input to this workflow would be the name of the VM that has reached 80% utilization memory.
- Create an alarm in vROps that uses this threshold as a threshold. Have it call said workflow.
- Now, when a VM in that group reaches the utilization threshold you want, vROps can call out to vRO and supply in its payload the name of the VM. The vRO workflow will in turn call out to your load balancer and deactivate that member in the pool.
- Caveat here is that you'd need automation logic that restores that member to the pool once the condition is no longer true. Also further logic that ensures at least one member in the pool is always available regardless of any triggering events.
As you can see, there are quite a lot of steps with some degree of complexity, but it's entirely possible.