Databricks explains how the Lakebase architecture (built on Neon) achieves resilience against cloud failures. Key design choices include stateless Postgres compute with zone-redundant remote storage, eliminating the need for costly hot standby replicas. The control plane is being split into a dedicated data-plane controller for hot-path operations (start/suspend) with minimal dependencies. A cell-based regional architecture limits blast radius — during an AWS us-east-1 incident in May 2026, only ~13% of databases were affected. Reliability is enforced through failure injection and chaos testing on every release, including whole-AZ-down simulations, with a target of no workload down for more than 30 seconds. Availability attainment data shows ~99.75–99.85% of databases meeting the 99.99% monthly SLO in early 2026.
