A deep dive into the challenges of recovering a 100 TB PostgreSQL database on AWS RDS, covering two primary failure modes: logical errors (requiring PITR) and infrastructure/AZ failures (requiring HA). AWS RDS snapshot-based PITR is slow at scale—restoring a 100 TB snapshot can take hours—while Multi-AZ standbys don't solve PITR and can suffer replication lag at large sizes, plus double costs. Neon's storage-compute separation and log-structured storage engine are presented as an alternative, enabling instant PITR via database branching at any historical timestamp and built-in multi-AZ HA without dedicated standby instances.
