Question

You maintain a Direct Lake semantic model over a large lakehouse fact table with 8 billion rows. Business users report that certain high-cardinality slicer visuals frequently trigger DirectQuery fallback because column memory guardrails are exceeded, causing slow performance. Your priority is to keep the model on Direct Lake for the bulk of queries while ensuring a small, frequently filtered dimension is always served from memory with no fallback. Which approach best meets this requirement?

Accepted Answer

Convert the model to a composite model, keeping the large fact table in Direct Lake and importing the small dimension table into Import storage mode.

Answer

Set the model's DirectLakeBehavior to DirectQueryOnly so all queries bypass the in-memory engine consistently.

Answer

Disable the SQL analytics endpoint on the lakehouse so Direct Lake can no longer fall back to DirectQuery.

Answer

Increase the capacity SKU so the guardrails are removed entirely and all columns stay in memory.