I've seen partitions in Postgres clusters where a node gets overloaded to the point it's not responding to health checks and the cluster management software per

I've seen partitions in Postgres clusters where a node gets overloaded to the point it's not responding to health checks and the cluster management software performs a failover. Since the previous master can't be reached but is still running, it accepts writes even after a new replica is promoted. I've seen similar things happen in other software under load as well (which can easily cause cascading failures, too)