After a physical hardware migration of some backend servers for XRPLCluster.com, IP address changes were not committed for the JSON POST RPC config, after being successfully changed for WebSocket connections.
The mistake wasn’t noticed because the cluster is configured to be redundant on several levels. So all traffic that didn't hit the changed IPs was redirected to other, still available backends.
When the infra serving all fallback requests went down formaintenance, the uncommitted IP change resulted in zero failover capacity.
JSON POST RPC upstream health has now obviously been added to monitoring. We already meticulously monitored WebSocket upstream health, on connectivity and even data sanity.