Two Architectures: “Postgres-on-a-Disk” vs Cloud-Native Disaggregated Storage

The one thing to understand first

Every managed PostgreSQL offering on the market falls into one of two architectural families, and almost everything that distinguishes one product from another flows from which family it belongs to. Get this distinction and the rest of the pathway becomes a series of variations on a theme.

Family one: Postgres-on-a-disk. A normal PostgreSQL process runs on a compute instance with a network-attached block volume underneath it (Amazon EBS, Azure managed disk, Google Persistent Disk). The heap and the WAL live on that volume, just like a server you would build yourself. This is AWS RDS, Google Cloud SQL, and Azure Database for PostgreSQL Flexible Server.

Family two: disaggregated, log-is-the-database. The storage layer is rebuilt as a distributed, multi-node service that PostgreSQL talks to over the network. The database tier no longer writes data pages to a disk; it ships its WAL records to a smart storage fleet that materializes pages on demand. This is Amazon Aurora, Google AlloyDB, Neon, Databricks Lakebase, and Azure HorizonDB.