Question

A retail analyst maintains a Direct Lake semantic model built on a Fabric lakehouse Delta table. An overnight Spark job appends the previous day's transactions to the Delta table, writing new Parquet files and updating the Delta log. The next morning, users report that reports still show yesterday's totals and none of the newly appended rows appear. The model is not configured for scheduled refresh. What is the correct explanation and remedy?

Accepted Answer

Direct Lake requires the model to be reframed; configure a refresh (scheduled or via API) so the model picks up the latest committed Delta files, since reframing does not happen automatically after every write.

Answer

Direct Lake always reflects the latest Delta data in real time, so the missing rows indicate the Spark job failed to commit the transaction to the Delta log.

Answer

The model has fallen back to DirectQuery because the guardrail was exceeded; switch the fallback setting to Direct Lake only to restore current data.

Answer

Import mode is required to see appended rows; convert the model from Direct Lake to Import and schedule an incremental refresh.