Postgres Summit US2026

Disaggregating compute and storage requires re-engineering how the backend interacts with memory, write-ahead logs (WAL), and persistence layers. Building on the Neon architecture and on community work toward resizable shared buffers, Lakebase adds runtime buffer resizing and zero-ETL analytical reads against operational data. This session explores: Disaggregated Storage Foundations: how Neon decouples Postgres block operations, and how page versioning and LSN-based reconstruction work. Dynamic Shared Buffers: a community-initiated effort to make Postgres's static shared_buffers resizable at runtime, which we are helping develop upstream. We cover that design — how backends remap shared memory, handle buffer reallocation, and manage lock contention during pool expansion — and how Lakebase delivers a version of it today. Lakehouse Integration and Direct Storage Writes: how analytical engines read live Postgres state directly from the storage layer, and how direct writes land lakehouse data into Postgres, bypassing the compute node.