Session

Lakebase 101: core workflows and the Lakebase experience

Overview

Experience	In Person, Virtual
Track	Lakebase
Industry	Manufacturing, Communications, Media & Entertainment, Financial Services
Technologies	Lakebase
Skill Level	Beginner

New to Lakebase? This session walks you through the core workflows that define the Lakebase experience: provisioning a database, connecting your first application, using branching for safe development, applying point-in-time recovery, and controlling costs with autoscaling. Through live demos, you'll see how each feature maps to real-world scenarios and why running your operational database directly on the lakehouse changes what's possible for applications, analytics, and AI. No prior Lakebase experience required.

Session Speakers

Hatim Binani

/Solutions Engineer
Databricks

Youssef Mrini

/Sr DevRel
Databricks

Full Summary

Lakebase 101: Why Postgres is being reinvented for the AI era

Most operational databases powering modern apps were designed before serverless, elastic compute, and AI agents. Lakebase, a fully managed, serverless Postgres built into the Databricks Platform, rethinks OLTP for AI-era workloads by decoupling compute from storage, integrating with the lakehouse, and keeping the Postgres model and ecosystem developers already trust. Teams evaluating where transactional workloads fit within a modern data strategy will find useful context in the data lakehouse, which explains how OLTP, OLAP, and AI workloads converge on a single governed platform.

The core message is simple: applications have become highly dynamic and intelligent, while the transactional stores behind them remain rigid and expensive to scale. Lakebase aims to deliver Postgres that scales on demand, costs less when idle, and plugs natively into governance and analytics, without custom pipelines or database clones.

Why conventional managed Postgres constrains modern apps

Traditional OLTP services tightly couple compute and storage. Teams must pre-provision for peak traffic, pay for idle capacity, and clone entire databases for dev and staging. Point in time recovery is slow, history retention is short, and bridging OLTP and OLAP often means bespoke ETL, duplicated data, and parallel governance.

Lakebase keeps standard Postgres semantics and tooling, including ACID transactions, indices, triggers, procedures, and popular extensions like pgvector and pgcrypto. The shift is architectural, not a fork: compute and storage are separated, which unlocks elasticity, fast recovery, and instant, low-cost isolation for development.

How Lakebase's architecture unlocks elasticity

Compute runs on managed Kubernetes and activates in under 500 milliseconds. Writes flow to safekeepers, which replicate the write-ahead log across three availability zones. Page servers organize data into durable pages that live in object storage. Because compute and storage are independent, more CPU or memory can be added without downtime.

Key capabilities enabled by this design:

Autoscaling and scale to zero, with CPU-driven scale up or down in about a second and a configurable idle timeout so you only pay when the database is active.
Copy-on-write branches, so feature, agent, or dev branches spin up instantly and only the delta is billed.
Instant snapshots and point in time recovery within a configurable retention window, up to 30 days.
Independent read replicas that share the same storage, provision quickly, autoscale on their own, and scale to zero without data duplication.

High availability is built in. Safekeepers use quorum across availability zones, and active page servers eliminate catch-up delays typical of monolithic replicas. Security spans OAuth via Databricks or native Postgres credentials, TLS in transit, encryption at rest with optional customer managed keys, network isolation with private links and IP filters, and audit and lineage in system tables.

Unified with Unity Catalog and the lakehouse

Lakebase registers in Unity Catalog as read-only federated catalogs, so permissions, lineage, audit logs, and discovery are unified alongside lakehouse tables. Cross-source SQL lets teams mix OLTP and OLAP data in a single query, and the same Lakebase tables are available to AI/BI Dashboards, Databricks Notebooks, Databricks Apps, and Genie.

Two built-in sync patterns eliminate custom pipelines:

Sync tables move curated lakehouse data into Lakebase to serve applications.
Lakehouse Sync continuously replicates Postgres tables from Lakebase into Unity Catalog managed tables with low latency, without external compute, since Lakebase data already lives in object storage.

Genie can generate and execute code against Lakebase, reducing the barrier to interact with transactional data in natural language.

What the demos illustrated

Provisioning a Lakebase project took about ten seconds and produced a standard Postgres connection string. A simple customer lookup returned in milliseconds on Lakebase versus seconds on a cold Databricks SQL warehouse, underscoring the importance of choosing OLTP for request latency and OLAP for large aggregations. Creating a dev branch, adjusting schema, and diffing against the parent felt closer to a Git workflow than a heavyweight database clone.

Elastic OLTP that speaks Postgres and lives in the lakehouse

Lakebase delivers transactional speed, autoscaling, and developer agility while remaining standard Postgres and natively governed in Unity Catalog. It is generally available in paid workspaces and the Databricks free edition, making it easy to validate performance, cost, and workflows firsthand.

FAQ

Lakebase is standard Postgres, not a fork. Existing tools like DBeaver and pgAdmin work out of the box, core SQL features are unchanged, and extensions such as pgvector and pgcrypto are supported.

Compute is billed in compute units, where one CU equals 2 GB of RAM, and you pay only while compute is active thanks to scale to zero. Storage is billed for primary data, branches incur charges only for changed data, and read replicas add no extra storage cost.

No. Lakehouse Sync continuously replicates Lakebase tables into Unity Catalog managed tables without external compute or pipelines. Sync tables handle the reverse direction, pushing selected lakehouse data into Lakebase for serving, all under unified governance.

Highlighted patterns include serving analytical data at application latency, powering transactional apps and AI agents as a primary backend, and continuously bringing operational data into the lakehouse for long-horizon analysis.

Creating a Lakebase project in the UI took about ten seconds in the demo and immediately provided a standard Postgres connection string. Lakebase is GA in both paid workspaces and the free edition, with no credit card required to try it.