Postgres: ServerLESS is more?
summary
Proposes separating Postgres into thin compute nodes and two storage services: shared WAL storage and an LSN-versioned page server that stores multiple page versions so checkpoints and full-page-writes move off primaries. Claims: 8KB page/PDLSN mechanics remain, WAL volume drops, replicas start and recover seconds faster, clones and PITR become near-instant using object-store blobs, compute can scale to zero and support multi-tenancy and sharding. Presents a blueprint with early POCs and research rather than a finished, production roll-out.
AI-written summary · about
chapters (9)
- 0:00Introduction to serverless
- 1:31Why serverless in cloud
- 9:51Serverless database requirements
- 13:57Evolving Postgres architecture
- 22:53Shared relation page storage
- 31:15Object storage integration
- 36:16Multi-tenancy and sharding
- 40:53Future of Postgres serverless
- 42:38Q&A
related talks
41mNeon tech - Separation of storage and compute for Postgres
14mMaking Postgres serverless
26mNeon: How we made PostgreSQL serverless
42mPostgreSQL Distributed: Architectures & Best practices
51mScaling Postgres to the next level at OpenAI
51mPostgres on OpenStack
54mPostgres: From Cloud to Hybrid and On Prem Again
36mPostgres at any Scale