← One-Pager Library
Databricks · One-Pager · OLTP

Lakebase Postgres

A fully managed, Postgres-compatible OLTP database built into Databricks — low-latency transactions and feature serving that sit right next to your lakehouse.

GA · AWS & Azure Separated storage & compute Postgres 16 / 17 Verified June 2026 docs

In one line

Lakebase is a serverless, Postgres-compatible OLTP database where storage and compute scale independently, with copy-on-write branching and two-way sync to Unity Catalog Delta tables.

Why it matters

The lakehouse is great at analytics but not at millisecond, row-level reads and writes. Lakebase gives you a real transactional Postgres for app backends, online feature stores, and agent state — without standing up and operating a separate database, and without ETL gymnastics to move data between your operational store and the lakehouse.

Core concepts

Managed Postgres OLTP

Standard Postgres wire protocol and SQL (PG 16/17) — connect with psql, psycopg, DBeaver, or any Postgres driver; OAuth token is the password over SSL.

Separated storage & compute

Compute (endpoints, sized in compute units) autoscales 0.5–112 CU and scales to zero; durable storage lives independently, so you pay for compute only while it serves.

Branching

Copy-on-write branches clone a database instantly for dev, testing, or CI/CD — including point-in-time branches for recovery — without copying the data.

Synced tables (lakehouse → Lakebase)

Sync a Unity Catalog Delta table into Postgres for low-latency serving, in snapshot, triggered, or continuous mode via a managed Lakeflow pipeline.

Reverse sync (Lakebase → lakehouse)

Capture Postgres changes back into Unity Catalog Delta tables (CDC) so operational writes land in the lakehouse for analytics — Public Preview.

App & serving use cases

Backs Databricks Apps, online feature stores for ML model serving, and AI agent state — transactional reads/writes co-located with governed lakehouse data.

How it works

Click a stage
1 · App / Agent 2 · Lakebase Postgres 3 · Synced tables 4 · Lakehouse (UC Delta)
App / Agent Lakebase PG Synced tables UC Delta
App / Agent — an application, ML serving endpoint, or AI agent issues low-latency SQL over the Postgres protocol (OAuth token + SSL).

Minimal example

SQL · standard Postgres against Lakebase-- ordinary Postgres DDL/DML — no Databricks dialect
CREATE TABLE orders (
  id        bigint PRIMARY KEY,
  customer  text NOT NULL,
  amount    numeric(12,2),
  created   timestamptz DEFAULT now()
);

INSERT INTO orders (id, customer, amount)
VALUES (1, 'acme', 42.00);

-- millisecond point lookup — the OLTP read pattern Lakebase is built for
SELECT * FROM orders WHERE id = 1;

Gotchas & limits

Interview soundbites

Say this out loud
  • "Lakebase is managed, Postgres-compatible OLTP inside Databricks — it adds the transactional, low-latency layer the lakehouse never had, governed by Unity Catalog."
  • "Storage and compute are separated: compute autoscales and scales to zero, and copy-on-write branching clones a database instantly without copying data."
  • "Synced tables push UC Delta tables into Postgres for sub-second serving; reverse sync brings operational writes back to Delta — one platform, no bespoke ETL."

References

Verified against docs.databricks.com · June 2026