en.Wedoany.com Reported - pgEdge has announced the launch of pgEdge ColdFront, a transparent data tiering solution for PostgreSQL. This solution supports direct UPDATE and DELETE operations on archived data, allowing applications to read and write cold-tier data using the same SQL they already use, without requiring code modifications or data reactivation. Old data is automatically migrated to Apache Iceberg in Parquet format and stored on any S3-compatible object storage, reducing storage costs by up to 90%. The complete dataset remains readable and writable through a single PostgreSQL table name, and cold-tier scanning leverages DuckDB's vectorized columnar engine to achieve analytical-level speed.
As production-grade PostgreSQL databases grow over time, historical data introduces increased storage costs and operational complexity, including backup overhead, vacuum overhead, and replica lag. Traditional approaches include deleting old data, archiving to flat files that break queries, or adopting proprietary solutions. pgEdge ColdFront eliminates these trade-offs by automatically moving cold data to inexpensive object storage while maintaining full read and write capabilities through the standard Postgres interface, with each tier stored in an open format.
An example use case is a GDPR deletion request targeting five-year-old archived data. With ColdFront, this can be executed with a single SQL statement, eliminating the cycle of restoring data to the hot tier, deleting it, re-archiving, and re-validating.
Phillip Merrick, Chief Product Officer at pgEdge, stated that the solution eliminates concerns about proprietary vendor lock-in. Applications retain the same SQL, DuckDB provides analytical speed for cold data in-process, the cold tier is writable, and it runs on standard, unpatched PostgreSQL.
Key features of ColdFront include: the only directly writable cold tier, supporting UPDATE and DELETE through the same PostgreSQL table name without reactivation or special paths; running DuckDB within the PostgreSQL process, delivering cold-tier scan performance on Parquet data that is 10 to 100 times faster than row-based storage; no code or schema changes required for applications; and all tiers using open formats, with cold data as standard Apache Iceberg (Parquet on S3), independently readable by tools such as Spark, Trino, and DuckDB.
The solution also features built-in partition lifecycle management, controlling the hot data working set via a single configuration parameter, hot_period, with an optional parameter, retention_period, to automatically delete cold data after a specified time. In a multi-master cluster environment, cold data is simultaneously readable and writable from each node. Through the Spock multi-master cluster, hot data is replicated by Spock, while cold data resides in a shared object store. The Bakery protocol (formally verified in TLA+) serializes Iceberg commits across nodes without 409 conflicts or application-level retries; in tests with 90 million rows across three small nodes, it achieved a performance of 756,000 rows per second.
Dave Page, Chief Technology Officer at pgEdge, noted that the solution automatically handles the data lifecycle, reducing storage costs by up to 90%. Its data infrastructure supports AI and ML pipelines, transforming PostgreSQL into a stateless compute frontend on Iceberg through a decoupled architecture, allowing new compute nodes to start within seconds without data synchronization.
pgEdge ColdFront offers three operational modes: Tiered Mode (hot + cold), Decoupled Mode (Iceberg only), and Partition-Only Mode (no cold tier). The product is now available as a production-grade beta, supporting PostgreSQL 16, 17, and 18 in single-instance and multi-master Spock grid topologies. ColdFront will be bundled with pgEdge Enterprise Postgres and is planned for integration into pgEdge Cloud in the second half of 2026. It is open-sourced under the PostgreSQL license, with documentation and installation instructions available on the designated website.
This article is compiled by Wedoany. All AI citations must indicate the source as "Wedoany". If there is any infringement or other issues, please notify us promptly, and we will modify or delete it accordingly. Email: news@wedoany.com
