Well if your application is not a service, that other clients can send requests to, then a loadbalancer is just not needed. You just have to make sure, that your service is up and healthy(scripting) and if not to fence(forcefully remote initiated shutdown/disconnect) the failed server/vm/container.
Common fencing methods are for example:
- shutdown by baseboard power control(IPMI,AMT,iDrac,iLo,Remote Controllable Rack Power Distributor,..)
- shutdown switch port of a managed switch
- executing an arbitrary command(ssh vmhost kill vm, ...)
Further there is a need for clusters to have an unequal number of nodes within the cluster, so if there are different views of the system state on the different nodes, the cluster can reliably decide which node is failing. So you need at least 3 instances.
My practical experience of the pacemaker stack is a bit limited. I have read some about it and created several test setups, but I did not operate those directly managed systems. I have some few prebuilt solutions like proxmox(preferred, Open Source Variant available) or XCP NG(Open Source, Clone of Citrix XenServer aka Citrix Hypervisor), that offer HA-functionality for virtual machines.