hmm. Suppose in case second node is 100 KMs away with less than 500ms on LLT heartbeats, we can implement local HA.
But a Heartbeat network disconnection can be a concern. How can we cater this to avoid split brain scenario ? any suggestion.
You also need to consider what is the design of your cluster especially the storage, it's unlikely that you can build a reasonably performing shared storage with 500ms RTT, hence you may have a replicated or shared-nothing cluster, the former usually built to tolerate longer RTOs so not a problem and the latter is redundant and usually comes with a load balancing mechanism that solves the problem.
The standard solutions for split-brain arbitration are CPS and SCSI-3 PGR fencing, the latter guarantees both data integrity and membership arbitration, with CPS you may have issues with data integrity on shared storage but again, I doubt that a storage that runs on 500ms RTT can deliver satisfactory performance.