Currently, the DC refreshes its remote peer cache before every scheduler invocation by requesting the CIB and parsing it for remote node config/state. However relying on the CIB is a bit brittle and likely subject to corner cases.

A better approach would be for remote connection hosts to broadcast controller cluster messages for remote node join/leave events and maintain the remote peer cache from that (no longer needing CIB). This would allow moving remote node CIB state updates from the connection host to the DC and in general, Pacemaker Remote nodes would be handled more similarly to cluster nodes.

The main challenge is what to do about cluster nodes that don't receive the message (due to not being up at the time, being in a different partition, etc.). Perhaps when a node is first elected DC, it could broadcast a request for active remote node connections and all remote connection hosts could respond, or remote connections could be reported as part of the join process, but there would still need to be special handling when partitions are healed.