Forum Discussion
How to Prevent Background Operation in Power bi Environment
I could suggest few things but everything has a trade-off to be honest.
Option 1) Switching to DirectLake mode instead using Import.
This would help you to eliminate those background CU's as it would not need to make those refresh operations but DirectLake mode is not as efficient as Import when it comes to report rendering. Thus, it would simply increase your CU's consumption on interactive part.
Option 2) Optimize your Semantic Models
- Remove unused columns and tables, every column consumes memory
- Use integer or boolean data types instead of strings where applicable (e.g., flag columns)
- Avoid high-cardinality text columns like free-form descriptions or GUIDs unless required
- Set up pre-aggregated tables for large fact tables and configure aggregation rules — DirectQuery hits drop dramatically
- Use composite models to serve summary queries from import mode while keeping detailed data in DirectQuery
- Disable "Auto date/time" in Power BI settings to prevent hidden date hierarchy tables from being generated for every date column
Background operations cover all non-user-triggered backend processes. The main contributors in a typical Fabric capacity are: semantic model scheduled refreshes, on-demand refreshes, Dataflow Gen1/Gen2 refreshes, Spark job runs, pipeline executions, and lakehouse table maintenance operations. All of these are classified as background regardless of whether they are triggered manually or by a schedule.
In the Capacity Metrics app matrix at the bottom of the compute page, sort by CU(s) descending, this immediately surfaces which items are consuming the most background CU. Hover over any row to see the operation-level breakdown in the tooltip. This would direct you where to look to lower your CU consumption.
Hi cengizhanarslan Thank you for sharing the above details I have checked under compute page which report consuming more CU under scheduled\on demand refresh also checked in details level when I hover over it. In our Power bi Environment most of the data sources from databricks, sharepoint and some are from excel files. I checked the report which are consuming more CU's it generally follwed import mode. Is there any steps can you suggest how to reduce this background CU's. Since semantic model size is increasing day by day as data volume is increasing in source.
I could suggest few things but everything has a trade-off to be honest.
Option 1) Switching to DirectLake mode instead using Import.
This would help you to eliminate those background CU's as it would not need to make those refresh operations but DirectLake mode is not as efficient as Import when it comes to report rendering. Thus, it would simply increase your CU's consumption on interactive part.
Option 2) Optimize your Semantic Models
- Remove unused columns and tables, every column consumes memory
- Use integer or boolean data types instead of strings where applicable (e.g., flag columns)
- Avoid high-cardinality text columns like free-form descriptions or GUIDs unless required
- Set up pre-aggregated tables for large fact tables and configure aggregation rules — DirectQuery hits drop dramatically
- Use composite models to serve summary queries from import mode while keeping detailed data in DirectQuery
- Disable "Auto date/time" in Power BI settings to prevent hidden date hierarchy tables from being generated for every date column
