Deconstructing Lakebase: Kernel Mechanics for Dynamic Memory, Direct-to-Storage Writes, and Lakehouse Access
October 20–23
Disaggregating compute and storage requires re-engineering how the backend interacts with memory, the WAL, and persistence layers. Building on the Neon architecture and community work toward resizable shared buffers, Lakebase supports runtime buffer resizing and in-place analytical access to operational data directly from the lakehouse. This session explores: * Disaggregated Storage Foundations: Neon’s decoupled block operations, page versioning, and LSN-based reconstruction. * Dynamic Shared Buffers: We cover shared memory mapping, buffer reallocation, locking, and how Lakebase delivers a version of it today as part of a community effort to make shared_buffers dynamic. * Lakehouse Integration and Direct Storage Writes: Live analytical reads from the storage layer and direct writes of lakehouse data into Postgres.
