SSSamarth Sinha

Note

Governing a Databricks lakehouse with Unity Catalog

One catalog for tables, files, models and permissions, and the decisions to make before teams start creating things.

Nov 2026 · 1 min read

  • Databricks
  • Unity Catalog
  • Governance
Unity Catalog: identity, hierarchy and controls Identity: Identity provider, Account groups, Service principals. Identity provider to Account groups. Account groups to Metastore. Service principals to Metastore. Metastore to Catalogs. Catalogs to Schemas. Schemas to Tables, views, models. Row filters and masks to Tables, views, models. Tables, views, models to Lineage and audit. Tables, views, models to Delta Sharing. Identity Identity provider users and groups Account groups grants go to groups Service principals jobs and pipelines Metastore one per region Catalogs per domain or stage Schemas per data product Tables, views, models plus volumes Row filters and masks fine-grained access Lineage and audit system tables Delta Sharing across clouds and orgs
IllustrativeUnity Catalog: identity, hierarchy and controlsScroll →

What Unity Catalog gives you

A single governance layer across workspaces: a three-level namespace (catalog, schema, object), privileges granted to groups, lineage captured automatically, audit data in system tables, and Delta Sharing for sharing data outside the platform.

Decisions to make first

  1. Catalog layout. Per environment, per domain, or domain catalogs bound to environment-specific workspaces. Pick one and write it down.
  2. Groups, not people. Grant privileges to account-level groups synced from your identity provider, and give pipelines their own service principals.
  3. Ownership. Every catalog and schema has an owning group, so access requests have somewhere to go.
  4. Managed or external storage. Managed tables keep the lifecycle in one place; external locations are for data that has to stay where it is.
  5. Fine-grained rules. Row filters and column masks for sensitive data, defined once on the table instead of copied into many views.

Running it

  • Check lineage before changing a table: it shows the jobs and dashboards downstream.
  • Query the audit system tables for unusual access, and review grants on a schedule.
  • Share across clouds or with partners through Delta Sharing rather than copying data.

Trade-offs

  • A strict layout slows the first week and saves the next year.
  • Row filters and masks add work to every query on large tables; test them on realistic volumes.