Since this is about DO managed Postgres: if you're using it with replicas, they use async replication and RPO can be greater than 15 minutes. Since failover is triggered during upgrades, there ends up being a lot of periods where you can lose multiple minutes of committed data.