Setting up replication and failover for services like Cloud SQL and Redis

Google Cloud offers built-in replication and failover for managed databases like Cloud SQL and Memorystore for Redis, but the mechanisms differ between high-availability (HA) configurations and read replicas. Data engineers must know which replication topology is synchronous and automatic versus asynchronous and manual to design resilient pipelines.

Must-know

Cloud SQL HA (regional) configuration uses synchronous replication to a standby instance in a different zone and performs automatic failover with no data loss, typically completing in under a minute.
Cloud SQL read replicas (same-region or cross-region) use asynchronous replication and are for read scaling, not automatic failover; promoting a replica to a standalone primary is a manual, one-way, irreversible action that breaks replication.
Memorystore for Redis Cluster provides built-in intra-cluster replication across shards and automatic failover to replica nodes without manual intervention, improving on the older standalone Memorystore for Redis offering.
Memorystore for Redis (non-cluster) Standard Tier offers cross-zone replication with automatic failover to a replica, while Basic Tier has no replication and any node failure causes data loss.
Cross-region Cloud SQL replicas can lag behind the primary due to network latency and asynchronous replication, so RPO is non-zero and must be accounted for in failover planning.
After a manual promotion or failover event, applications must update connection endpoints (or rely on the instance's stable IP/DNS in HA cases) since replica promotion changes the instance's role and sometimes its connection details.

Drill this objective Take a weighted mock Try free practice questions

Coverage checked against the published exam guide on Aug 11, 2026.

These are independent practice questions, written against this certification's published exam guide. They are not the certification vendor's own questions, and not the real exam.