Question

You are building a Direct Lake semantic model over a Fabric lakehouse containing a fact table with over 5 billion rows. During testing, some queries unexpectedly fall back to DirectQuery and users report inconsistent performance. You want the model to keep as much data as possible resident in memory (VertiPaq) rather than falling back, and you need the model to support paging large column segments in and out of memory efficiently. Which configuration should you apply to the semantic model to best support this scale?

Accepted Answer

Enable the large semantic model storage format on the model so column segments can be paged and the model can exceed the default in-memory size limits

Answer

Switch the fact table's storage mode to Import so all rows are compressed and loaded at refresh time

Answer

Set the Direct Lake behavior to DirectQuery only so that no data is held in memory

Answer

Reduce the number of columns by hiding them in the model so VertiPaq loads fewer segments