↓ Skip to main content
Microsoft Fabric Deployment Options: A Successful Fabric Deployment Is Not a Working Solution
  1. Posts/

Microsoft Fabric Deployment Options: A Successful Fabric Deployment Is Not a Working Solution

·1188 words·6 mins
Table of Contents

The deployment succeeded. Everything is green. Then someone opens the report and finds yesterday’s data—or a connection to the development environment.

The tool did what it was asked to do. We asked too little of the release.

In the first post, I looked at ways to move changes through Microsoft Fabric and argued for choosing a process the team can maintain. There is a second question, just as important: what actually moves when you deploy?

For every item, ask what moves, what stays behind, and who checks that the result works. The answer changes with the item and the deployment method. Check the current Git integration and deployment pipeline support lists before designing a workflow around any particular item.

Variable Libraries: Values Still Need an Owner
#

Development, test, and production are not identical. Connections, paths, and settings differ. Fabric variable libraries can hold values in one item with a value set for each environment. Deployment rules in pipelines can also set certain stage-specific values.

The active value set is selected separately in each workspace. A newly created target starts with Default active until someone changes it. A deployment plan cannot switch the active set during deployment; its actions use whichever set is active in the target. Even the deployment pipeline comparison will not flag a different active value set, because that selection is workspace state rather than part of the deployed definition. Microsoft’s variable library guidance.

That sounds like a small operational detail. Point a production item at a development source, and it stops being small. Someone must own the values and verify the active set before a release.

SQL Databases and Warehouses: Schema Has Its Own Rules
#

For Fabric SQL databases and warehouses, the important change is often the schema. That schema can be versioned as a SQL project and applied as a differential update. Moving the item and changing its database objects are related, but they deserve separate review.

For a Fabric SQL database, an update from Git builds and validates the project before publishing changes. Pre- and post-deployment scripts can help maintain small static lookup tables. Database-level settings such as collation and compatibility level are outside source control and deployment pipelines at the time of writing.

Warehouses support Git integration and deployment pipelines too. Teams can also develop the SQL project locally and use DacFx or SqlPackage in external automation. Their development and deployment experience is currently in preview.

The maintainer needs to understand the proposed schema change and its effect on existing data. Unlike deploying a report definition, applying a database project can modify existing objects or data. A syntax error, unsupported feature, or potentially destructive change calls for a different response from a missing report reference. A green deployment status alone cannot establish that the database is safe to use.

Power BI Projects: Readable Files, Real Dependencies
#

A PBIX file is difficult to review in Git because it is a binary. A Power BI Project (PBIP) stores a report and semantic model as files: PBIR for the report and TMDL for the model. A pull request can show changes to pages, visuals, and model definitions. PBIP is a source format; Git integration, deployment pipelines, or a code-based tool still handles the deployment.

What I would show a client is the link between report and model. A report’s definition.pbir can refer to a model by relative path or by workspace connection. Deployment pipelines can rebind related items in corresponding stages, including across pipelines. Other workflows may need parameters or manual work.

After deploying to test, open the report and check the model it uses. Readable files make a change reviewable. They do not guarantee that the result points where you intended.

Lakehouses: The Item Moves, the Data Stays
#

A lakehouse makes the distinction especially clear. Git and deployment pipelines track its metadata and OneLake shortcut definitions, but they do not copy its tables, files, or data. Internal OneLake shortcuts can be remapped across stages, while external shortcuts normally retain their original targets unless explicitly configured otherwise. Microsoft’s lakehouse guidance.

The release therefore needs a plan for creating or loading the target data, checking shortcut targets, and proving that the downstream items can read what they expect.

Apps: Which Kind, and What Will Users See?
#

If readers consume reports through a classic Power BI app, promoting workspace content does not update that app’s content and settings. Someone must publish the updated experience to its audience.

Fabric org apps are workspace items with content and audience definitions. They can be versioned in Git and moved through deployment pipelines, but the audiences and access still deserve a check. Microsoft’s guidance currently says service principals cannot deploy org apps across pipeline stages, although org app item APIs support service principals for definition operations. That distinction matters for automation.

Fabric apps built with Rayfin are different again. They combine application code, database schema, and static content. Microsoft documents a GitHub Actions workflow that deploys them with rayfin up.

The names are similar; the release processes are not. Test the experience the user will actually open.

Connections and Runtime Settings: The Work Around the Item
#

An item’s data source settings or parameters may travel with its definition. Deployment rules or variable libraries can point the item at a different source in each stage. A Fabric connection is a separate resource. It needs to exist in the target environment with the right credentials and, where required, a gateway binding. Microsoft’s CI/CD planning guidance treats connections as part of environment setup.

For supported items, a connection reference variable can help select a connection, but its ID does not automatically change across stages. Verify the target value.

For semantic models, deployment pipelines do not copy role assignments, refresh schedules, or data source credentials. A newly deployed model may also need a refresh before anyone can use it. Workspace and item permissions need their own check.

“The deployment succeeded” and “the solution is ready” are not the same statement.

Let the Client Declare the Release Finished
#

Who decides that the release is complete? I would ask someone from the client team to check that the right users can open the result, the data is current, and every item points to the intended source. That is a more useful finish line than a green deployment status.

A handover should include a small change deployed to test, a failed run investigated together, and a new item added to the process. Can the future maintainer find the error and tell what already changed?

If production fails, who decides whether to fix forward or restore the previous version, and how? Make that decision before a failed release forces it on the team.

The consultant can build a sophisticated workflow. The client has to live with it. A successful delivery leaves them able to release, check, recover, and extend the solution themselves.


Join the Conversation
#

How do you think about deployment and handover? I am always curious to hear other perspectives and I’d love to hear about your approach. Find me on LinkedIn or BlueSky.


Photo by Haris Illahi on Unsplash