Which Cluster Access Modes Work with Unity Catalog?

My “Databricks Ninja Moose” series is an unofficial, cartoonish way to share data engineering tips. Today, we’re taking a quick look at which Databricks cluster access modes actually work with Unity Catalog — and why some familiar methods, like sc.parallelize, are off-limits in the new “Shared” mode.

1. From Single User to Shared — What Changed?

Databricks originally offered:

• Single User — a cluster owned and used by one specific person.
• Shared (old) — a multi-user cluster without tight isolation, meaning multiple users could somewhat overlap resources.

With the introduction of Unity Catalog, user isolation and security became critical. Enter:

• Shared (new, with user isolation) — each user session is isolated, so multiple people can share the same cluster without risking cross-access to each other’s data.

2. Which Modes Are Officially Supported with Unity Catalog?

Per the Databricks documentation:

• Single User
• Shared (with user isolation, i.e., new Shared)

The old Shared mode (without isolation) is not supported by Unity Catalog.