Neon's instant Point-in-Time Recovery (PITR) works by leveraging its unique architecture: compute and storage are separated, with WAL safekeepers, pageservers, and cloud object storage working together. Every data change is assigned a Log Sequence Number (LSN), and pageservers use a non-overwriting copy-on-write model to preserve all historical versions. When a restore is triggered, Neon simply creates a new branch at the target LSN using GetPage@LSN — no data is copied. This makes restores complete in under a second regardless of database size, compared to hours for traditional systems like AWS RDS that must spin up a new instance and replay WAL logs.
Table of contents
A refresher on the Neon architectureMechanics of PITR in NeonHow Neon’s PITR differs from traditional restore processes (e.g., AWS RDS)Conclusion1 Impression