Snowflake announced a public preview of data mirroring, a feature that replicates Postgres tables into Snowflake using Apache Iceberg tables. The replication is performed by a Postgres extension that pushes change batches into per‑table logs and a meta log, then Snowflake applies the changes transactionally. According to Snowflake, the process aims for low latency, low cost, and transactional consistency.

Data replication between operational and analytical databases has been difficult because traditional change data capture tools must handle continuous updates, schema evolution, and failures. Snowflake’s approach decouples the producer and consumer by using object storage such as Amazon S3 as the intermediate destination.

The extension, called snowflake_cdc, captures logical row‑level changes from the Postgres write‑ahead log, groups them into batches, and records them in Iceberg change logs. A meta log tracks the sequence of operations, allowing Snowflake to execute the changes in a finite state machine. This design preserves transaction boundaries and foreign key relationships.

Because the replication is transactional, inserts are appended rather than upserted, which improves performance for high‑volume workloads. Live views combine unapplied changes with the base table, enabling queries with low lag without frequent refreshes.

The system scales to high throughput and avoids typical failure modes such as external connector lag, snapshot conflicts, or race conditions. Setup is a one‑time configuration, after which the replication runs continuously.

Snowflake plans to make the feature generally available after the preview period. The capability is intended to simplify data integration between Postgres and Snowflake for developers and enterprises.