Open Source Postgres Hot/Cold Tiering with Apache Iceberg - ColdFront Deep Dive

Most of your Postgres data sits on expensive SSD and hasn't been queried in months. You can't delete it -- retention requirements keep it alive. So you keep paying SSD prices for data nobody's touching. ColdFront is pgEdge's open source transparent tiering system for Postgres. Cold data moves to Apache Iceberg on object storage (up to 90% cheaper than SSD) while hot data stays in native Postgres partitions. Your application sees the same table, the same connection string, the same SQL. Nothing in your code changes. In this session, Jimmy Angelakos (Staff Software Engineer and the engineer behind ColdFront) walks through the architecture and the decisions behind it -- including the hard parts, like getting multi-writer correctness right across Spock nodes. Paul Rothrock (Solutions Engineer) follows with a live demo from setup through reads, writes, and deletes against cold data. Live session with open Q&A. What's covered: -- Why proprietary tiering, ETL pipelines, and dump-to-files each have a structural problem -- How ColdFront routes queries transparently across hot Postgres partitions and cold Iceberg files, without a sidecar or RPC -- and how it handles DDL mirroring and rollback -- The writable cold tier: why read-only archival creates real problems for GDPR erasure, auditing, and corrections, and how ColdFront handles them with a plain DELETE or UPDATE -- Architecture: C extension, pg_duckdb (in-process), Archiver Go binary, Lakekeeper -- and why "each part does one thing well" drove every design choice -- Multi-writer correctness on pgEdge Spock nodes: a Lamport-style bakery protocol with Ricart-Agrawala deferred replies, model-checked in TLA+, plus a DuckDB iceberg patch that moves uploads outside the critical section for 2.5x better throughput -- Vector search on cold data: IVF-style clustering without building an index, 93% recall, 54x faster than a full file scan (10 million rows, 1,024 dimensions) 0:00 Introduction 1:14 Session overview 2:39 The cold data problem 5:19 Why existing approaches fall short 6:53 Architecture and design philosophy 9:28 How transparency works 11:28 Tiered mode vs. decoupled mode 13:01 The writable cold tier 14:55 The ColdFront extension 18:00 Distributed correctness 20:04 The bakery protocol 22:39 Async uploads and throughput 24:32 Formal verification with TLA+ 26:30 Vector search on cold data 29:19 Roadmap 31:09 Live demo 40:08 Q&A ColdFront is open source under the PostgreSQL license and currently in beta. Docs, guided walkthrough, and formal verification spec are all in the repo. Try it: https://github.com/pgEdge/ColdFront All pgEdge webinars: https://www.pgedge.com/webinars #PostgreSQL #Postgres #ApacheIceberg #OpenSource #Database