GCP Professional Data Engineer Practice Question 6: CDC Replication to BigQuery
Google Cloud Professional Data Engineer · Test 1 · question 6 of 10
Analysts need an on-premises Oracle order-management database mirrored into BigQuery with data no more than 15 minutes stale, including row deletes. The database administrators refuse to allow repeated bulk extract queries against the production instance, and the data team has no capacity to write or operate custom pipeline code.
Which approach meets these requirements?
Show answer and explanation
Answer:
Datastream is a serverless change data capture and replication service that reads a source database's change log rather than querying its tables, so it imposes very little load on the production instance and captures inserts, updates and deletes alike. Oracle is a supported source and BigQuery is a supported destination for direct replication, so the whole path is configuration with nothing to code, deploy or scale.
Option B adds an unnecessary hop and misuses both tools. Database Migration Service exists to migrate databases into managed Google Cloud databases, not to feed an analytical warehouse, and routing every analytical query through a federated read of a Cloud SQL replica pushes the analytics load onto a transactional engine that is not built for it.
Option C misses the freshness requirement by an order of magnitude, a nightly export leaves the warehouse up to 24 hours stale, not 15 minutes.
Option D is precisely what the constraints forbid. High-water-mark polling issues a heavy query against production every 15 minutes, cannot see rows that were deleted (there is no later timestamp to find), and leaves the team owning custom pipeline code.
Reference: Datastream overview