Databases on Google Cloud

Database map

A fully managed relational database for MySQL, PostgreSQL, and SQL Server. It handles backups, high availability, connectivity, import and export, maintenance, monitoring, and logging.

• Each instance is a VM with the database engine and a persistent disk, with a static IP address in front.
• Connectivity over public IP or private IP (only from a VPC), and authorization with the Cloud SQL Auth Proxy or language connectors.
• It creates databases and users, but it isn't an administration tool: you use clients such as MySQL Workbench, pgAdmin, or SQL Server Management Studio.
• Pricing depends on storage, CPUs, memory, location, egress traffic, and IP addresses.

For MySQL 8.4 and 9.7 the default edition is Enterprise Plus.

Cloud SQL: high availability and replicas

Best practice: with HA on the primary, place replicas in another zone. A replica can be promoted to primary in a regional disaster or if the original becomes corrupted.

AlloyDB for PostgreSQL

A fully managed, PostgreSQL-compatible service with a Google-built engine and a disaggregated architecture: compute (nodes) and storage (distributed across zones, grows automatically) scale independently.

• Use cases: HTAP (analytics on live transactional data), low-latency AI with AlloyDB AI (vector search and model invocation), and mission-critical workloads.
• An in-memory columnar engine to speed up analytical queries, an index advisor, and adaptive autovacuum.
• On-demand and scheduled backups, and point-in-time recovery.
• Each instance has a static private IP in your VPC. You manage it with AlloyDB Studio in the console or standard clients such as psql.

A fully managed distributed database with horizontal read and write scaling, automatic sharding, and strong consistency. It supports the GoogleSQL and PostgreSQL dialects, plus key-value, graph (Spanner Graph), full-text search, and vector search. Because it is managed, there is no way to stop or restart an instance.

• Compute capacity: in nodes or processing units (1 node = 1,000 PUs). Below one node you must use PUs.
• You can move an instance from one configuration to another (for example, regional to multi-region) and create custom configurations with read-only replicas in another region.
• There is a 90-day free trial instance.

A fully managed, serverless NoSQL document database: no capacity provisioning, it scales automatically, and you pay per use. It offers ACID transactions, strongly consistent queries, and single-digit millisecond read latency.

Availability: 99.99% in regional locations and 99.999% in multi-region locations, with automatic replication across regions.

A sparsely populated table that scales to billions of rows and thousands of columns (terabytes or petabytes), ideal for large volumes of single-keyed data with low latency and high read and write throughput. Compatible with the Apache HBase API.

• Each table is a sorted key-value map. Only the row key is indexed; related columns are grouped into column families.
• Typical data: time series, marketing, financial, IoT, graphs. Each value is usually no larger than 10 MB.
• Performance scales in proportion to nodes; clusters resize without downtime and can autoscale on CPU.
• To replicate, add a second cluster to the instance and replication starts automatically.
• Enterprise and Enterprise Plus editions. A 10-day free trial, extendable to 90.

A managed in-memory service for caches with sub-millisecond access: sessions, frequent queries, game leaderboards, real-time analytics.

Multi-region redundancy

The exam guide asks for "maintaining multi-region redundancy across data solutions". Each service handles it differently:

Exam scenarios

Common exam traps