Snowflake makes Postgres data mirroring generally available

What data mirroring does

Snowflake says Postgres data mirroring continuously replicates Postgres data into Snowflake. Once a mirror is created, Snowflake maintains target tables that reflect the current state of their source tables. Schema changes carry over on their own, and new tables added to mirrored schemas are picked up automatically.

A mirror can be set up in a few clicks in the Snowsight interface, with a single SQL command, or by asking Snowflake CoCo. Users can choose specific tables, a full schema or a database. Snowflake frames the feature as a way to end what it calls the "pipeline tax": CDC connectors that need babysitting, batch jobs that break and snapshots that drift.

The company argues that timeliness matters more in the age of AI. As an example, it says a fraud model cannot stop today's attack with last night's batch load.

Features and architecture

Snowflake lists several capabilities:

• No extra infrastructure: mirrors run inside Snowflake, so there are no additional services or vendors.
• Fresh reads: every mirrored table includes a $live view that combines already-applied data with in-flight changes. Readers see committed source changes within seconds.
• Transactional consistency: changes from a single source transaction appear on the target together, so foreign keys hold and joins stay accurate.
• Change history: each table gets a seven-day change feed ($changes) showing inserts, updates and deletes, queryable from both Snowflake and Postgres.
• High throughput: an optimized apply strategy skips full-table scans, so cost and performance stay stable as data grows.

Under the hood, mirroring uses pg_lake, Snowflake's open source Postgres extension. It captures changes through standard logical replication and writes them into Apache Iceberg tables. Users can mirror into standard Snowflake tables for simplicity, or into Iceberg tables on their own storage for full portability.

Snowflake describes the design as "push, don't pull". Many replication tools place a separate service between the database and Snowflake that pulls changes over the network. Here, a Postgres extension captures changes from inside the database and pushes them in transactional batches into open Iceberg change logs in object storage. Snowflake then applies those batches with serverless compute, also transactionally. Because the extension sees everything happening in Postgres, it coordinates snapshots, schema changes and recovery at the source. Inserts are appended rather than upserted, which Snowflake says keeps the largest, most insert-heavy tables fast and cost-efficient.

Use cases and customers

Snowflake points to several scenarios: replacing middleware pipelines with a single CREATE_MIRROR call, live operations dashboards for e-commerce, AI agents that answer from the current state of the business using Snowflake Cortex AI, fraud scoring on live payment events, and audit trails built from the $changes feed.

The company says Ericsson replaced its pipelines this way, consolidating four legacy databases, eliminating sync pipelines and reducing data processing time by 99%. Alan Buxton, Chief Technology Officer at Simfoni, is quoted in the post: "Transactional data mirrors to Snowflake automatically, so our customers can leverage AI over it without extra work."

Availability and what it means

Data mirroring is available now on AWS and Microsoft Azure in every region where Snowflake Postgres runs. To start, users spin up Snowflake Postgres or use an existing instance, create a mirror, and query the target tables in Snowflake. The example call in the post sets a refresh_interval of '1 minute'.

For teams that move application data into analytics, the release offers a route that avoids third-party replication vendors and extra infrastructure, with change data landing in open Apache Iceberg tables. Snowflake also hinted at a reverse direction, mirroring analytics data from Snowflake back to Postgres, saying only "stay tuned". The post carries a forward-looking statement noting that future offerings are not commitments.