Forum Discussion
Issues with Power BI CI/CD using PBIP format, Direct Lake, Git Integration, and Deployment Pipelines
Hi IvanGennaro ,
Thank you for reaching out to Microsoft Fabric Community.
I recommend the following approach for implementing a reliable CI/CD flow for PBIP reports in Microsoft Fabric:
Place your Direct Lake model in a shared workspace. Reference that shared model from all report workspaces (Sandbox, Dev, Prod). This avoids broken bindings when the model doesn’t exist in downstream workspaces.
Maintain Git branches like main, dev, and prod.
Each Fabric workspace syncs to its respective branch (e.g., Dev WS <-> dev branch).
Use Git pull requests (PRs) to promote code and control approvals.
Use Git as the source for reports. sync from Git in every workspace. Deployment Pipelines can be used for model promotion or basic stagingPBIP reports should include a connections.json referencing the datasetId and workspaceId (GUIDs), not just names.
This ensures the report knows where to find the dataset, regardless of workspace context.You can refer to the following Microsoft documentation for more details:
https://learn.microsoft.com/en-us/fabric/cicd/git-integration/manage-branches?tabs=azure-devops
https://learn.microsoft.com/en-us/fabric/cicd/best-practices-cicd
https://learn.microsoft.com/en-us/power-bi/developer/projects/projects-git
If this post helps, then please consider Accepting as solution to help the other members find it more quickly, don't forget to give a "Kudos" – I’d truly appreciate it!
Thank you!!
Community SupportHi IvanGennaro ,
Thank you for reaching out to Microsoft Fabric Community.
I recommend the following approach for implementing a reliable CI/CD flow for PBIP reports in Microsoft Fabric:
Place your Direct Lake model in a shared workspace. Reference that shared model from all report workspaces (Sandbox, Dev, Prod). This avoids broken bindings when the model doesn’t exist in downstream workspaces.
Maintain Git branches like main, dev, and prod.
Each Fabric workspace syncs to its respective branch (e.g., Dev WS <-> dev branch).
Use Git pull requests (PRs) to promote code and control approvals.
Use Git as the source for reports. sync from Git in every workspace. Deployment Pipelines can be used for model promotion or basic staging
PBIP reports should include a connections.json referencing the datasetId and workspaceId (GUIDs), not just names.
This ensures the report knows where to find the dataset, regardless of workspace context.
You can refer to the following Microsoft documentation for more details:
https://learn.microsoft.com/en-us/fabric/cicd/git-integration/manage-branches?tabs=azure-devops
https://learn.microsoft.com/en-us/fabric/cicd/best-practices-cicd
https://learn.microsoft.com/en-us/power-bi/developer/projects/projects-git
If this post helps, then please consider Accepting as solution to help the other members find it more quickly, don't forget to give a "Kudos" – I’d truly appreciate it!
Thank you!!
Hey! Thanks v-sathmakuri for your suggestions! I've tested it and it worked. It is definitelly possible to use git to move the reports across Fabric and Deployment Pipelines for Models while maintaining a Direct-Lake connection.
Even though it worked, we still went for a more simplistic approach because the full-fledge git ci/cd approach was adding too much complexity for our use case. We moved to a Fabric-centric approach, using the "Publish report" method and Deployment pipelines to move the reports and models across nonprod and prod workspaces. The only part that we want to rely on pure Git is for report distribution across functional workspaces, we would have our Report-Production workspace with all our reports, but we need to distribute copies of those in functional/department workspaces for business users to consume. All of them will be pointing to the production model in report-production ws.
There is also a bit more information on my reddit post regarding this topic: https://www.reddit.com/r/MicrosoftFabric/comments/1kmhx09/issues_with_power_bi_cicd_using_pbip_format/
This is how our final approach looks: