Extend the database definition, if feasible, to record which hosts are querying the clustersafety table.
For example, in PostgreSQL it could be done via a trigger on the table, possibly one which adds a row to a cluster_hosts table if the address isn't already present, and simply updates the timestamp if it's already there.

This wouldn't actually be used often, but could be invaluable in solving the mystery cases where a standalone system cluster-panics with no other known Confluence servers in the environment.