5.32 Release Notes - Databricks

New Features

Serverless compute for Jupyter tools.

JupyterLab can now run on serverless compute, joining SQL Lab and Superset which already use serverless SQL warehouses. Your Space is ready sooner, and nothing changes about how you work once it is running.

For user guidance, see the "Serverless in Spaces" section within Spaces.

Serverless for Asset Create

EARLY ACCESS

Asset creation and updates can now run on Databricks serverless compute rather than a dedicated cluster.

• Available for Oracle, MySQL, SQL Server, and PostgreSQL connectors only.

• Requires both a platform-level and a connector-level opt-in, and is intended for non-production workloads at this stage.

For operator set-up, see Configuring Serverless Compute for Asset Creation.

New MCP Tools for Subscribing, Exporting and Data Sharing

EARLY ACCESS

Building on the Discovery MCP Tools introduced in 5.31, your AI assistant can now act on your behalf for several new workflows, all from within a normal chat conversation.

• Subscribe to a product on a self-serve plan, request access to an approval-gated plan and track its status, or activate a subscription that is on hold pending terms acceptance.

• Create a data share of a product or a standalone asset over Databricks, either an open, token-based share usable by any recipient, or a direct Databricks-to-Databricks share into the recipient's own Unity Catalog.

• Export a product, setting the destination, name, description, refresh schedule, and email notifications.

If you do not see the new tools right away, disconnect and reconnect your AI assistant to refresh the available tool list.

For user guidance, please see Using the Harbr MCP.

For operator set-up, see Configuring the MCP Server.

Improvements

Additional Connector Types Supported by Capability Checks

EARLY ACCESS

Connector capability checks now run for Google Cloud Storage, Azure, Amazon S3, DB2, SFTP, Power BI and Looker, alongside the database connectors covered since 5.31.

The checks test a connector from two places before you set it live: from Harbr, and from the processing environment where your jobs actually run. A connector that would have failed later is caught during setup instead.

This release extends that same checking to further connector types: Google Cloud Storage, Azure, Amazon S3, DB2, SFTP, Power BI, and Looker.

For user guidance, see the relevant section within Create a Connector.

For operator set-up, see Configuring Connector Capability Checks.

More Granular Control Over User Deletion

EARLY ACCESS

Ecosystem admins can now safely remove a user directly from the platform, with control over who can do so, whether that's a named list of admins or all ecosystem admins. Deletion is a soft deactivation rather than a hard delete: the person's access is revoked immediately, but everything they created is retained and re-attributed to "Name (Deactivated)", nothing they owned is transferred or lost.

• A user cannot be deleted while they have a running Space, a running export, a scheduled or in-flight task, or own a Space with active collaborators, the platform explains exactly what needs resolving first before the deletion can proceed.

• Deleting a user who was invited but never registered, or a service account, is out of scope for this release.

For operator set-up, see Delete a User.

AI Query Assets — Consistent Access Permissions

AI Query assets now apply the same access permission checks used across the rest of the platform, so only users with the correct permissions can call a Space.

Databricks Asset Update Retry and Timeout

Asset updates now recover from transient failures on their own, and can be given a timeout so a stalled update does not run for hours. Retries cover network failures, connection stalls and Databricks job timeouts. The timeout is calculated from the asset's own typical run time.

For user guidance, please see the relevant section in Databricks.

For operator set-up, see Configuring Databricks Asset Update Retry and Timeout.

Improvements to Connector Behaviour

• The SFTP connector is more resilient to transient network conditions.

• When browsing catalogs, schemas, and tables during Databricks connector set-up, the platform now only shows resources the connector's account can actually use to create an asset.

• Oracle connector listing shows only the objects your account can access, without needing elevated database privileges. For user guidance, please see the relevant section within Oracle.

Fixes

• Organisation creation no longer completes silently when the required Unity Catalog schema fails to provision.

• Complex data types from Databricks are now handled correctly.