Most people spin up Lakebase and hit surprises they didn't see coming.
Here's everything I wish I'd known before shipping packed into 5 slides.
What's inside:
What is Lakebase?
Fully managed Postgres on Databricks. OLTP for the Lakehouse. No ETL pipelines,
no servers. Core capabilities: autoscaling, branching, HA, read replicas, instant
restore and Unity Catalog sync all built in.
Two versions now exist (since March 2026)
Lakebase Autoscaling is the new default. Scales to zero, supports branching and
30-day rollback. Lakebase Provisioned is still supported for existing workloads
but gets no new features. Know which one you're on.
The 7 roles Lakebase creates automatically
Most are internal. The one you actually use is your email role. databricks_superuser
sounds powerful but has NOLOGIN it cannot open a database connection. Never put
it in your connection string.
Lakebase Superuser โ Postgres Superuser
Same label, completely different behavior. The managed permission model breaks
vanilla Postgres assumptions silently. Four real-world gotchas covered with how
to avoid each one.
Lakebase syncs to Unity Catalog, but not the way you think
Four things from the official docs most people discover too late: 15-second minimum
sync delay, REPLICA IDENTITY FULL required on every table, no partitioned table
support, and only the databricks_postgres database syncs (beta restriction).
If you're building on Lakebase or planning to, save this before you ship.
#Databricks #Lakebase #DataEngineering #Architecture
