Databricks open-sources Lakebase, serverless Postgres with cloud-native storage separation

Databricks announced Lakebase, a serverless Postgres-compatible database that externalizes the write-ahead log (WAL) and data files into independent cloud services (SafeKeeper for durability, PageServer for storage), enabling stateless compute and eliminating the architectural pain points of monolithic databases. The design unlocks unlimited scalable storage, elastic compute, durable writes without misconfiguration risk, simplified high-availability setup, and instant branching—all without meaningful added latency compared to traditional Postgres.

Unlike traditional monolithic databases like MySQL and Postgres where the WAL and data files live on a single machine (creating data-loss risk, expensive read replicas, and friction scaling), Lakebase separates concerns: the compute engine remains Postgres-compatible, but no longer couples to storage, replication, or availability infrastructure. This separation eliminates the need for full physical database clones just to add replicas or standby nodes. Databricks also introduced LTAP (Lake Transactional and Analytics Processing), which stores operational data once in open columnar formats readable by both Postgres and Lakehouse engines, enabling analytics to run on the same fresh data transactions just wrote—eliminating CDC pipelines and data duplication.

For infrastructure teams evaluating data architecture, Lakebase addresses the 16-year-old complaint that OLTP databases are 'clunky, difficult to scale, and incredibly fragile.' The open-storage design is particularly compelling for cloud-native workloads where elasticity and durability matter. LTAP's unified storage layer (vs. traditional HTAP's single-engine approach) preserves the best engine for each workload—Postgres for OLTP, Lakehouse for analytics—while eliminating sync delays and infrastructure duplication.