Semantic Model Breaks After DirectQuery to Direct Lake Migration Refresh

[Shreya_Barhate]

Hi,

We are migrating a semantic model from DirectQuery to Direct Lake in a Power BI / Fabric semantic model, aiming to retain the full model structure.

The migration completes successfully, and the model initially looks correct. However, after a refresh:

Joins/relationships are removed

KPIs and calculated columns are not migrated

This breaks the semantic model.

We want to understand:

Is this an expected limitation of DirectQuery → Direct Lake migration?

Are there any fixes or workarounds?

Do relationships need to be recreated after refresh?

Are we missing any required migration steps or settings?

Thanks,
Shreya

[Tamanchu]

Hi @Shreya_Barhate,

The fact that the model looks correct immediately after the migration is an important clue.

Based on the error message, I wonder if the relationships are actually a symptom rather than the root cause. The refresh error mentions that multiple source tables either do not exist or access was denied. If some Direct Lake tables become invalid during refresh, Fabric may no longer be able to maintain the relationships that depend on them.

A few things I would check :

•  Do all source tables still exist in the target Lakehouse/Warehouse?
•  Has anything been renamed (table name, schema, workspace, warehouse)?
•  Does the semantic model owner still have access to all underlying tables?
•  Are all relationships removed, or only those involving specific tables?
•  How was the migration performed (Desktop, XMLA, Tabular Editor, TMDL, etc.)?

The fact that everything looks fine before refresh but breaks afterwards makes me suspect a table binding, lineage, or permission issue rather than a relationship issue itself.

Hope this helps narrow it down

[Shreya_Barhate]

Hi,
We followed a similar approach for migration using the Semantic Link Labs notebook:
semantic-link-labs/notebooks/Migration to Direct Lake.ipynb at main · microsoft/semantic-link-labs ·...
The migration was executed via notebook, and all the source tables are available in the Lakehouse. Also, the required users have been granted Contributor access on the workspace.
Despite this setup, we’re still seeing the semantic model break post migration from DirectQuery to Direct Lake.
Would be helpful to understand if there are any additional configurations or limitations we might be missing—especially around permissions, table compatibility, or model refresh behavior in Direct Lake.


Thanks

[v-priyankata]

Hi @Shreya_Barhate

Thank you for reaching out to the Microsoft Fabric Forum Community.

@Tamanchu Thanks for the inputs.

Since the tables are available and permissions seem correct, the issue may be related to how the model reconnects to the Direct Lake tables after a refresh. It's also possible that KPIs and calculated columns aren't fully carried over during the migration process.

As a next step, can you please check whether the missing relationships are limited to specific tables and confirm if KPIs and calculated columns are supported in Direct Lake after migration.

If the issue still persists after these checks, the best approach would be to open a Microsoft support ticket Create a Fabric and Power BI Support Ticket - Power BI | Microsoft Learn, as a successful migration and refresh shouldn’t normally cause the model to break.

If there are any deviations from your expectation please let us know we are happy to address. 

Thanks.

[v-priyankata]

Hi @Shreya_Barhate 
I hope you’ve submitted the support ticket. If you’ve received a solution, please share it here so it can help others who encounter a similar issue.