Cloud Cost Sense
Cloud SQL HA, Backup, and Read Replica Cost Guide
Estimate Cloud SQL cost beyond the primary instance by modeling high availability, backup retention, read replicas, and cross-region traffic.
Separate the primary instance from high availability
Start with the primary instance edition, machine resources, storage type, provisioned capacity, region, and hours running. Cloud SQL is normally a persistent database baseline rather than request-only compute, so a low-traffic application can still pay for the selected capacity throughout the month.
High availability changes a zonal instance into a regional configuration with a standby in another zone. Official pricing applies HA rates to regional instances, including applicable compute and storage resources. Compare the HA premium with the downtime and recovery process the product can tolerate; a read replica alone is not the same automatic failover configuration.
Model backups by retained data, not database size once
Backup cost follows backup storage used, so do not multiply the current database size by the retention count without checking change rate and the backup system in use. Record automated-backup retention, on-demand backups, point-in-time recovery logs, deleted-instance backup retention, and exports kept separately in Cloud Storage.
Estimate a normal month and a high-write month because churn can make retained backup data grow differently from the live database. Treat exports as Cloud Storage objects with their own storage, operation, and transfer lifecycle. Regularly test restore procedures before shortening retention solely to reduce cost.
Price each replica and its network path
A read replica is billed like a separate Cloud SQL instance, with its own compute and storage configuration. Count every replica by region and purpose: read scaling, reporting, migration, disaster recovery, or export isolation. Frequent subset exports may justify a replica, but compare its always-on cost with serverless export charges and operational impact.
Cross-region replication and application traffic can add network transfer, while same-region placement can avoid some paths. Include replica lag monitoring, backups, logging, private connectivity, extended support, and downstream query traffic where applicable. Compare three explicit scenarios—single-zone primary, regional HA, and HA plus replicas—using current regional prices before committing.