Forum Discussion
BUG or FEATURE???::DirectLake mode not supported
And wouldn't you know it... you have to go traipsing through a labyrinthine thicket of uncommon sense and other assorted illogical UI paths. So here is the solution to connect to lakehouse tables in DirectQuery mode--the long and short of it, you need to figure out where the option to connect to the lakehouse SQL endpoint is, and once you do, select that and badabing-badaboom!, you are in business. Totally ridiculous UI/UX design if you ask me (not to mention that the previous attempts do not generate any visual feedback to let you know what the hell is going on), but there you've it:
Just a simple CREATE VIEW where I select a computed col in addition to all the cols from the source table. I'd be curious to learn what's the best approach though, in terms of efficiency and so on. What happens each time a report queries the lakehouse table (in DirectQuery or DirectLake mode)? Is the view recomputed from scratch every single time? Or does Fabric use materialized views like Oracle? The underlying data store being managed by Microsoft, I'd assume they configure it so that it is optimized for READ speed, but who knows, right?
I also heard you actually can WRITE to a lakehouse table, instead of just creating a virtual table, but only if using PySpark in a notebook. Haven't tried yet.DROP VIEW vTABLE_CALC_COL; GO CREATE VIEW vTABLE_CALC_COL AS SELECT *, ([count] /2 ) AS calcol FROM Data_15MThe new calculated col, calcol, then shows up in dataflows and models.
Ah-hah, I think I've figured it out!
Within fabric, select your lakehouse you want data from and in the explorer window, there's a top ribbon with "Home" and "Reporting" menus, select the "Reporting" menu and then "Manage default semantic model" (may work from "New semantic model" too although haven't got that to work yet)
Choose the tables you want and click "Confirm"
Then go back to the PowerBi report choose "Pick a published semantic model"
Choose your lakehouse and either "Auto-create report" or "Create a blank report" and it should now work. I don't see the option to connect relationships within your data so not entirely certain where that comes in to it just yet, possibly through foreign key definitions?
And in a follow up to that... my issue with "new semantic model" is that my session had timed out but browser was blocking the popup. You can set the relationships within a new semantic model
Kryt0n I don't understand. Aren't you just building a report from the PBI Service, ie Fabric, and not from PBI Desktop? I thought the issue was how to connect to Power BI semantic model from PBI Desktop and not from Power BI Service, wasn't it?
I was getting the errors within Fabric - what I assume you refer to as the Power BI service - I would click on "new report", choose my lakehouse and then it would fail with the error as per what you received
After choosing the "default semantic model" which initially had no tables selected (I note it says this is automatically retained but I found that wasn't the case), after selecting some and clicking new report, it all worked
As for where I see it, when I'm browsing my lakehouse and select a table, I get the "Reporting" menu on the top ribbon (this item doesn't show unless a table is selected) which then has "manage default semantic model".
If you're looking in the same place but still not seeing it, one other thing I had done (when I didn't have any idea what I was doing and clicking everything that may explain the error) is right click on the table names and choose "Add to default dataset" (I suspect you must do it for all you want in the dataset)
Nope... I checked both my lakehouse view and the SQL analytics view but the ribbon is different from yours. Are you sure you aren't in the warehouse view rather than the lakehouse view?
(Yes, Power BI Service is Power BI in the cloud. I can't bring myself to say Fabric as Fabric encompasses multiple services and not just PBI).
Also, where you exactly in Fabric when you see this ribbon? I don't have the menu option of 'Manage default semantic model.'