cancel
Showing results forย 
Search instead forย 
Did you mean:ย 
Warehousing & Analytics
Engage in discussions on data warehousing, analytics, and BI solutions within the Databricks Community. Share insights, tips, and best practices for leveraging data for informed decision-making.
cancel
Showing results forย 
Search instead forย 
Did you mean:ย 

How to remove/delete a Relationship Graph (Semantic Model) from an AI/BI Dashboard?

brickuser
New Contributor

Hi everyone,

I'm working on an AI/BI dashboard and I accidentally initialized a Relationship Graph (semantic model) by clicking "Get started with Relationships" in the Data tab. The model has no relationships and no measures -- it just auto-created entities from my existing datasets.

The issue: Since the relationship graph was created, I can no longer create new widgets normally through the UI. The field dropdown in new counter/chart widgets doesn't show my dataset columns. It seems the dashboard now forces everything into "entity-qualified" mode internally, which breaks the visual editor experience for simple single-dataset widgets.

What I've tried:
1. Navigated to Data, then Data Model, then Relationships -- I can see the graph with my entities but there's no option to delete/remove the model itself
2. The only workaround I've found is configuring widgets via the API using datasetName: "" with entity-qualified expressions like entityName.columnName

What I'm looking for: A way to completely remove/delete the relationship graph from my dashboard so the UI reverts to the normal behavior where I select a dataset and pick columns from the dropdown.

Environment:
- Databricks AI/BI Dashboard (Lakeview)
- Relationship Graph feature (Public Preview)

Has anyone encountered this or found a way to remove the semantic model once it's been initialized? Any help appreciated.

2 REPLIES 2

Armanillo_RoK
New Contributor III
Hi @brickuser,
 
  A (semi/)manual workaround (API/JSON editing): Edit the dashboard's serialized JSON (the .lvdash.json file) and remove the semanticModel object entirely. This requires accessing the workspace file and editing it directly, then reloading the dashboard. See the snip attached. Ask Genie for little help with editing the dashboard json instead of doing that manually (if you have it enabled in your account/workspace).

What Databricks should add: A "Delete Relationship Graph" or "Remove Semantic Model" button in the Data Model panel. Since the feature is in Public Preview, this is worth filing as feedback, apart from posting here you can submit product feedback through the UI (Help > Provide Feedback) to flag the missing "delete" action.

Best,
RoK
Commodore 64 - Ninja crush 'em all
https://youtu.be/RBvhfVGMsLQ?si=OfwbqWeygGUd_qU5

balajij8
Esteemed Contributor II

Hi brickuser,

You have hit a UI limitation with the Dashboard Relationships (Semantic Model) feature in Public Preview. Once you click Get started with Relationships, Databricks automatically initializes a semantic model from your existing datasets. While the documentation confirms this initialization step, UI mechanism to un initialize or delete the model is not available. This structural change alters the dashboard's internal state forcing entity-qualified mode across the board and breaking the standard field dropdown in the visual editor for single-dataset widgets.


You can revert it to the normal visual editor behavior in an export-edit-import workflow. Because the dashboard's .lvdash.json file captures the underlying definition including the semantic model - you can strip out the relationship payload directly from the raw JSON structure. You can follow below
  • Navigate to the dashboard in Draft mode, open the kebab menu (โ‹ฎ), go to File actions, and select Export.
  • Open the .lvdash.json file in a editor and search for the section defining the relationships. Locate the relationshipGraphs JSON structure and delete that block completely.
  • Save the file, navigate back to the workspace Dashboard listing and select Import dashboard from file to restore a clean dashboard without the semantic model layer.