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

Cloudflare details how it migrated billions of DNS records from a shared Postgres cluster to a dedicated, partitioned dnsdb using a gRPC-based DNS Records API, a trigger-driven change data capture pipeline with audited batch application, and a migration manager to lock and switch writes for a near-zero-downtime cutover and bidirectional fallback. The approach combined chunked initial copy, continuous incremental syncing every few seconds, and strict auditing, resulting in significantly reduced load on the original cluster, higher API throughput, lower CPU usage, fewer incidents, and improved replica-read consistency controls.