Get startedGet started for free

Keeping semantic models up to date with refresh

1. Keeping semantic models up to date with refresh

A semantic model shows data as of its last successful refresh, so new source rows don't reach reports until a refresh completes. In this lesson, you'll decide how much each refresh processes, when it runs, and how to confirm it worked.

2. Choosing how much data to process

Your first decision is how much data each refresh reloads. Reloading less saves time and capacity, but only when you know which rows can change. A full refresh reloads an entire Import table. For a 20-row currency lookup that changes quarterly, that's simple and cheap. For an Orders table with 1.2 million rows where only the last seven days can change, a full refresh mostly reloads unchanged rows. Incremental refresh keeps older rows and reloads only that recent window. Direct Lake models read Delta tables in OneLake directly, so refresh copies no data. Instead, it reframes the model to point at the latest table versions. By default, reframing happens automatically when the Delta tables change.

3. Choosing when refresh starts

With the amount settled, the next question is when a refresh starts. A scheduled refresh runs automatically at set times, based on how often the source changes and how current reports must be. DimCurrency rarely changes, so a weekly full refresh is enough. An on-demand refresh is started manually, perhaps after correcting DimCurrency's source data. A pipeline refresh runs within a larger workflow, so it can wait until its upstream load succeeds. Orders is loaded nightly by a pipeline, so its incremental refresh runs right after that load.

4. Incremental refresh separates history from change

Incremental refresh lets a model keep years of data while each refresh processes only recent rows that can still change. RangeStart and RangeEnd are Date and Time parameters that filter the source query by date. The historical period sets how far back the model keeps data, for example five years. The refresh period sets the recent window reloaded every time, for example seven days. Rows between the two stay unchanged, and rows older than the historical period are removed. The first refresh loads the full history and splits the table into partitions: slices of rows by time period, such as one month each. Later refreshes reprocess only partitions inside the refresh period.

5. Refresh after the source load succeeds

If a model refreshes while its source load is still running, it captures incomplete or old data and shows it until the next refresh. A pipeline makes the required order explicit. The source-loading activity, often a Script activity running T-SQL, writes the latest data. Use its green arrow to add a Semantic model refresh activity. The On success dependency triggers the refresh only after the load succeeds.

6. Configuring semantic model refresh

The activity's Connection, Workspace, and Semantic model settings identify the model. The Tables and Select partitions settings can limit the refresh to selected tables or partitions, such as recent Orders partitions.

7. Semantic model refresh history

Refresh history supports auditing and troubleshooting. To open it, expand Queries on the ribbon, expand Refresh, and select Refresh history.

8. Reading refresh history fields

Entries are split across Scheduled, OneDrive, Direct Lake, and OneLake Integration tabs. Type identifies the refresh route. Start and End show when it ran, and Status shows whether it Completed or Failed. Message explains the result, such as the error behind a failure.

9. Viewing refresh details

Selecting Show in the Details column expands an entry, displaying a Request ID that distinguishes the operation and a Time for the recorded event.

10. Let's practice!

Now let's put refresh into practice!

Create Your Free Account

or

By continuing, you accept our Terms of Use, our Privacy Policy and that your data is stored in the USA.