Deploy scalable and highly available databases in Google Cloud
This domain tests turning an architecture into a working deployment: network placement, HA settings, backups, maintenance, observability, and repeatable infrastructure. The exam expects concrete configuration choices, not just service selection.
Deploy with private networking and explicit HA
For Cloud SQL, set availability_type = REGIONAL, disable public IP, and attach a private network. Enable point-in-time recovery and automated backups, and set a maintenance window. Use the Cloud SQL Auth Proxy or language connectors with connection pooling so serverless or autoscaled clients do not exhaust connections. For Spanner, choose a regional or multi-region instance configuration and size with processing units; for Bigtable, create clusters in separate zones and enable replication for multi-cluster availability.
Make it repeatable
Use Terraform or Infrastructure Manager for instances, databases, users, flags, and backups. Store schema migrations in version control and run them as a controlled step, not ad hoc in the console. Export metrics and logs to Cloud Monitoring and Cloud Logging, and alert on replication lag, CPU, storage, and failover events.
Worked example
Deploy Cloud SQL Enterprise Plus for PostgreSQL with regional HA, private IP only, 7-day PITR, a Sunday maintenance window, and a cross-region read replica for DR reporting. Define the instance in Terraform with ipv4_enabled = false, private_network set to the VPC, and availability_type = REGIONAL. Add the replica in a different region and document that promotion is manual. Use the Auth Proxy from GKE with a pooled connection limit. This satisfies HA, recovery, and read scaling without exposing a public endpoint.
Traps that cost marks
Leaving public IP enabled; choosing zonal availability while claiming HA; omitting PITR; placing a DR replica in the same region and calling it regional resilience; skipping maintenance windows; and forgetting connection pooling for autoscaled clients. Also avoid console-only changes that drift from Terraform state.
Key rules
Private networking, explicit HA, tested backups, controlled maintenance, and observable deployments are the baseline.