Checkpoints[03]

In AlloyDB, WAL records are written to the storage layer synchronously. Then, the log processing service in the storage layer is responsible for replaying the logs continuously. AlloyDB supports checkpoints but there is no need for flushing the changes in the database layer to the storage layer.

Concurrency Control[04][05]

The concurrency control AlloyDB uses is the same as in PostgreSQL

Data Model

Foreign Keys[06]

AlloyDB supports foreign keys the same as PostgreSQL

Hardware Acceleration[07]

The system runs on Google Cloud instances but there is no specific hardware required, although two VM instances are required for an AlloyDB primary instance.

Indexes[08]

The concurrency control AlloyDB uses is the same as in PostgreSQL and PostgreSQL supports B-tree, Hash, GIST, SP-GIST, GIN and BRIN index types.

Isolation Levels[09]

Isolation level in AlloyDB can be set through database flag "default_transaction_isolation", where it can be set to either “read uncommitted”, “read committed”, “repeatable read”, or “serializable”.

Joins[10]

AlloyDB builds on PostgreSQL where PostgreSQL supports the following three join strategies: nested loop join and it is possible to utilize the values from the row of the left relation as keys for scanning the right relation's index given that the right relation can be processed with index scan, merge join and hash join.

Query Interface

Storage Architecture

Storage Model[11]

AlloyDB stores the data in the storage layer following the same format as PostgreSQL. However, a columnar engine is used where the data that gets loaded into memory can take on either the row-based format or the columnar format based on the workload, which is decided using machine learning models.

Storage Organization

System Architecture