Migrating billions of records: moving our active DNS database while it’s in use

Cloudflare describes how it migrated massive DNS datasets (e.g., 1.7B active records and 4.3B archived) from a monolithic Postgres cluster to a dedicated `dnsdb` using a custom Change Data Capture and Transfer Service with triggers, batching, auditing tables, and a gRPC API to control access. The approach enabled an incremental initial copy, sub-second change batching, bidirectional safety, partitioning by zone, and a migration manager that locked writes for just ~2 seconds during cutover. Post-migration, DNS Records API throughput more than doubled, CPU on both clusters stabilized below 10%, latency spikes reduced, and overall database-related incidents dropped, providing greater reliability and headroom for future growth.