SQL Database Hyperscale Architecture

​Azure SQL Database Hyperscale is a scale-out database deployment option that separates the database engine, log services, and storage. Key to Hyperscale are three layers:

Compute Nodes

The compute node layer consists of multiple SQL Server instances, each containing the SQL Server database engine with a predefined number of vCores and memory. There is a single primary node (top left) that processes all read-write operations and is the only node that can modify data. The primary can have up to four high-availability (HA) replicas with the same configuration and cost as the primary.

Hyperscale differs from a typical SQL deployment in the option to have up to 30 read-only compute nodes (top right) that can be used to create a scale-out deployment for multiple read-only applications. Microsoft calls them secondaries (and sometimes secondary replicas). However, this is a misnomer because they are not replicas of the primary node. Each read-only compute node is configured separately, with an individual name, connection string, security settings, and size. Perhaps most importantly, the read-only compute nodes access the same back-end data as the primary node, which is significantly different from a traditional secondary replica that has its own copy of the database.