Snowflake Postgres Enhances ML Feature Serving and Data Applications

Snowflake Postgres, a fully managed PostgreSQL database, is now generally available. It combines high transactional performance with native integration to Snowflake's analytical layer, offering enterprise reliability and security for demanding workloads.

The Snowflake Online Feature Store, built on Snowflake Postgres, is now in public preview. It's designed to serve low-latency ML features like risk scores and activity patterns to models at prediction time, enabling thousands of predictions per second with millisecond latency.

An open-source extension called pg_lake is included with Snowflake Postgres, enabling changes to be replicated from the transactional database to Snowflake's analytics layer in seconds without intermediate processes.

Snowflake Postgres offers 100% PostgreSQL compatibility, supports version 18, has a 64 TB storage ceiling, and includes native logical replication, facilitating lift-and-shift migrations with zero code changes.

The service provides advanced security controls such as PrivateLink and customer-managed keys, creating a unified security perimeter for both transactional and analytical data within Snowflake.

Internal benchmarks show Snowflake Postgres offers 2.5x lower latency and 7x higher QPS compared to Databricks Lakebase, with 10ms P50 latency at 1,500 QPS sustained, alongside a 99.95% uptime SLA and support for newer Postgres versions.