Forum Discussion
Helper IIIDirect Lake Restrictions and Limitation
Hi All,
I am looking to Direct Lake with Fabric Lakehouse, I was really optimitics about this features but now I have concerns on two points :
1. You cannot apply RLS on Lakehouse otherwise Direct Lake will fall back to DirectQuery, a workaround has been proposed by Guy in the Cube : https://www.youtube.com/watch?v=xuEYxJ5gkGA&t=188s&ab_channel=GuyinaCube
2. Now comes a new issue with Direct Lake, you cannot use Apps without a workaround : https://www.youtube.com/watch?v=FFsWEqrTRHE&ab_channel=GuyinaCube
I would like first to thanks Guy in the Cube teams, Patrick and Marthe for the solution and my question is finally does it bring more complexity to apply Direct Lake than Benefit, what is the risk / Benefit ?
For the first issue if we apply RLS on Semantic level instead of the lakehouse, it could be manageable therefore and the second issue, we are moving from SSO to Service Principal and technical configuration that I am afraid could create additional maintenance or critical issue if the solution provided do not work due to continues improvement on Fabric
Do they are more suistanable solution on going ?
What is the work around this issue on Microsoft team ?
Should I keep to use Import Mode and not leverage Direct Lake ?
Hello Fantmas
The decision between Direct Lake and Import mode should be based on your specific data volume, refresh frequency requirements, technical capabilities, and tolerance for implementation complexity. For organizations with large datasets and technical resources to implement the necessary configurations, Direct Lake offers substantial benefits that may outweigh the additional complexity. For smaller implementations prioritizing simplicity and consistent performance, Import mode may remain the more pragmatic choice.
The root cause of App access issue , lies in how permissions propagate through the various layers of the architecture. When a user accesses a report through an App, they typically don't have direct access to the underlying Lakehouse data. However, Direct Lake's default behavior requires users to have appropriate permissions to the source data, leading to access denied errors.
The recommended solution would be to creating and using a service principal to mediate access between users and the Lakehouse data.
Microsoft has introduced controls that allow administrators to manage how Direct Lake handles fallback scenarios, providing more predictable behavior
https://fabric.guru/controlling-direct-lake-fallback-behaviorMicrosoft has expanded guidance on developing Direct Lake semantic models, including best practices for RLS implementation
https://learn.microsoft.com/en-us/fabric/fundamentals/direct-lake-develop
2 Replies
- nilendraFabric
Super User
Hello Fantmas
The decision between Direct Lake and Import mode should be based on your specific data volume, refresh frequency requirements, technical capabilities, and tolerance for implementation complexity. For organizations with large datasets and technical resources to implement the necessary configurations, Direct Lake offers substantial benefits that may outweigh the additional complexity. For smaller implementations prioritizing simplicity and consistent performance, Import mode may remain the more pragmatic choice.
The root cause of App access issue , lies in how permissions propagate through the various layers of the architecture. When a user accesses a report through an App, they typically don't have direct access to the underlying Lakehouse data. However, Direct Lake's default behavior requires users to have appropriate permissions to the source data, leading to access denied errors.
The recommended solution would be to creating and using a service principal to mediate access between users and the Lakehouse data.
Microsoft has introduced controls that allow administrators to manage how Direct Lake handles fallback scenarios, providing more predictable behavior
https://fabric.guru/controlling-direct-lake-fallback-behaviorMicrosoft has expanded guidance on developing Direct Lake semantic models, including best practices for RLS implementation
https://learn.microsoft.com/en-us/fabric/fundamentals/direct-lake-develop
powerbihelp87Helper V
what would you consider large? 100million rows?