Manage a solution that can span multiple database solutions
This domain tests operating a heterogeneous data estate: moving data between databases, keeping schemas and identity consistent, monitoring across engines, and governing backups and access. The exam expects you to know which service owns which job rather than forcing one database to do everything.
Choose the integration pattern deliberately
- CDC replication: Datastream captures changes from MySQL, PostgreSQL, or Oracle and delivers to BigQuery or Cloud Storage for analytics and history.
- Federated queries: BigQuery can query Cloud SQL and Spanner through federation, but this is for analytical or operational reporting, not high-QPS application traffic.
- External tables: BigQuery external tables query data in Cloud Storage without loading it; useful for staging and lakehouse patterns.
- Operational fleet view: Cloud Monitoring, Cloud Logging, and Database Center give inventory, health, and security signals across database services.
Keep identity, schema, and backups consistent
Use IAM at the project and database level, and prefer IAM database authentication where supported so application identity is not a shared password. Treat schema changes as versioned migrations and apply them in dependency order across engines. Centralize backups with service-native PITR where available, and document restore procedures per engine because backup formats are not interchangeable.
Worked example
An order platform runs Cloud SQL for transactions, Spanner for global inventory, and BigQuery for reporting. The reporting team needs near-real-time order facts without affecting the OLTP primary. Stream Cloud SQL changes with Datastream into BigQuery, and use BigQuery federated queries only to enrich dimensions held in Spanner. Monitor replication lag in Cloud Monitoring and alert on it. Do not point dashboards directly at Cloud SQL with high-concurrency queries; that competes with application traffic and bypasses the analytical layer.
Traps that cost marks
Assuming BigQuery federation is a replacement for operational reads; using Datastream without handling schema evolution; granting broad project-level IAM instead of database-scoped roles; and assuming one backup tool restores all engines. Also watch for cross-service latency and egress costs when joining data across regions.
Key rules
Use CDC for movement, federation for selective joins, and native backups for recovery. Monitor lag and schema drift as first-class signals.