pgEdge has released ColdFront v1.0.0-beta1, an open-source transparent data tiering extension for PostgreSQL that unifies OLTP, analytics, and AI workloads. Its headline feature is a fully writable cold tier — archived data can be updated or deleted via standard SQL without restoring it to hot storage first. ColdFront runs on stock upstream PostgreSQL 17/18 with no proprietary forks, uses Apache Iceberg (Parquet on S3-compatible storage) for cold data, and embeds DuckDB in-process for columnar analytical performance. Applications interact with a single table name regardless of which tier data lives in, requiring no code changes. A distributed bakery protocol (model-checked in TLA+) enables concurrent cold writes across pgEdge Spock mesh nodes. The tool also handles full partition lifecycle management, supports 2-level tenant-by-time sub-partitioning, and is positioned for AI governance workloads requiring append-heavy writes with mandatory retention and occasional corrections.

9m read timeFrom postgr.es
Post cover image
Table of contents
The trade-off nobody wants to makeThe options so far all ask you to give something upColdFront doesn't askThe data layer underneath AI governanceTry it
313 Impressions