A team moved over 1 billion production records - no downtime, no drama. The stack: dual writes, Kafka retries, and idempotent inserts to keep it clean.
They ran shadow reads to sniff for errors, chunked the transfers with checksums, and held off indexing to keep inserts fast. Caches got warmed early to dodge cold start thrash.
What used to be a simple DB migration turned into a distributed systems playbook. Ops folks now think like SREs - patterning off high-reliability services instead of table moves.
