The current ha-proxy keepalive implementation using cephadm uses all available RGWs to be configured as backend , But by design RGW daemons can be part multiple zones , or multiple realms, all of which cannot be covered by the same haproxy daemon.

This design is very limiting in that it can be used only in cases where  all RGW daemons are part of the same zone.

/]# ceph -s
  cluster:
    id:     b8f37602-8c98-11eb-b5c1-fa163eb9cc01
    health: HEALTH_WARN
            7 failed cephadm daemon(s)
 
  services:
    mon: 5 daemons, quorum ceph-tejas-1616579452828-node1-mon-mgr-installer-node-exporter,ceph-tejas-1616579452828-node9-client,ceph-tejas-1616579452828-node6-mon-rgw-node-exporter-crash,ceph-tejas-1616579452828-node5-osd-node-exporter-crash,ceph-tejas-1616579452828-node3-osd-node-exporter-crash (age 4h)
    mgr: ceph-tejas-1616579452828-node1-mon-mgr-installer-node-exporter.rdfiei(active, since 5d), standbys: ceph-tejas-1616579452828-node9-client.spjimk
    osd: 12 osds: 12 up (since 14h), 12 in (since 5d)
    rgw: 7 daemons active (3 hosts, 3 zones)    <--- 3 zones across 2 realms