Time travel with Neon

on YouTube ↗45m

" summary

Neon replaces local disks with a networked storage system that keeps every page version and reconstructs pages on demand from WAL-like reshuffled delta files and S3, enabling page-level recovery and instant branching. A demo shows a serverless Postgres instance starting in about one second to run queries at arbitrary LSNs (time travel works), backed by safekeepers for recent WAL and page servers for reads. Implementation notes: LSM-style compaction, local SSD cache, shared_buffers set to 128MB, current storage amplification ~2–3x, largest deployments ~1–5 TB.

AI-written summary · about

" chapters (26)

1. 0:00Introduction to Neon
2. 1:09What is Neon and its storage system
3. 3:14Traditional PostgreSQL setup vs Neon
4. 6:45Neon storage architecture overview
5. 9:59Time travel and page reconstruction
6. 12:04Point-in-time recovery with time travel
7. 16:17Live demo of Neon time travel
8. 22:42Branching in Neon
9. 24:29Q&A on hot pages and edge cases
10. 26:17Managing branches at scale
11. 28:22Scaling large databases
12. 29:50Open source and on-premises options
13. 31:16Caching and performance optimizations
14. 32:39Optimizations and user data changes
15. 33:21On-the-fly time travel limitations
16. 34:04PostgreSQL version upgrades
17. 35:29Storage size and pricing model
18. 37:57Pricing model challenges
19. 38:18Image compression and patches
20. 39:23Retention policies and branching
21. 39:45Streaming replication with Neon
22. 40:48PostgreSQL backups with Neon
23. 41:52Shared buffers and caching
24. 42:36PostgreSQL extensions support
25. 43:18PostgreSQL version releases
26. 44:01Fast VM spin-up for branches

" related talks

• 41mNeon tech - Separation of storage and compute for PostgresBojan Serafimov · Percona Live 2022
• 26mNeon: How we made PostgreSQL serverlessHeikki Linnakangas · POSETTE: An Event for Postgres 2025
• 45mPostgres: ServerLESS is more?Nikhil Sontakke · PGConf India 2025
• 31mReimagining the Future: Where the Market is EvolvingSrini Srinivasan, Don Haderle · Aerospike Digital Summit 2021: Reimagine Real-time
• 46mEdge Performance with In memory NoSQLLiviu Costea · Voxxed Days Vienna 2017
• 39mTimescaleDB: Re-engineering PostgreSQL as a time-series databaseErik Nordström · Øredev 2018
• 32mLow Latency Serving of Offline Data: Efficient, Safe, and Reliable Data...William Schor · SREcon26 Americas
• 20mOpen Source Database Operations in the Age of AIPeter Zaitsev · Percona Live Amsterdam 2026