Question

You maintain a Direct Lake semantic model on a Fabric F64 capacity. Business users report that a specific measure occasionally runs much slower than usual, and Performance Analyzer shows the underlying queries intermittently executing in DirectQuery mode against the SQL analytics endpoint. Investigation reveals that the fact table exceeds the maximum number of rows per table allowed for Direct Lake on F64 before falling back. Which single characteristic of Direct Lake behavior best explains why some queries succeed in Direct Lake mode while others fall back?

Accepted Answer

When a query would require exceeding a capacity guardrail (such as rows-per-table), Direct Lake transparently falls back to DirectQuery for that query while other queries within limits stay in Direct Lake mode.

Answer

Direct Lake caches the entire model in memory only during scheduled refresh windows, so queries outside those windows always use DirectQuery.

Answer

Direct Lake permanently switches the whole model to DirectQuery the first time any guardrail is exceeded, and it stays there until the model is reframed.

Answer

The SQL analytics endpoint randomly load-balances queries between import and DirectQuery to distribute capacity pressure.