5.1.4. Gateways and Scheduled Refresh
💡 First Principle: The Power BI service lives in the cloud and can only refresh data it can reach. When a source sits behind a corporate firewall, an on-premises data gateway installed inside the network acts as a secure relay, making outbound connections to the cloud so the service can request data without opening inbound ports.
When a gateway is required:
- On-premises databases (SQL Server, Oracle, SAP), files on local or network drives (
\\fileserver\share\sales.xlsx), on-premises SharePoint, and other sources in private networks. - Sources in an Azure virtual network without public access can use a VNet data gateway (a Microsoft-managed option) instead of a VM-hosted gateway.
- Not required for cloud sources reachable from the internet: Azure SQL Database (with public access), SharePoint Online and OneDrive, Dataverse, Fabric lakehouses, Salesforce, and similar.
Gateway modes:
| Mode | Who uses it | Supports | Notes |
|---|---|---|---|
| Standard (on-premises data gateway) | Multiple users, centrally managed | Import refresh, DirectQuery, live connections, dataflows, many sources | Admins add data sources and users; can be clustered for high availability |
| Personal mode | One user only | Import refresh only | No DirectQuery or live connection, can't be shared; suitable for individual use |
Configure scheduled refresh on a semantic model's settings page: map the model to a gateway connection (if needed) and set data source credentials; turn on Keep your data up to date; choose frequency (daily or weekly), time zone, and time slots. Limits: up to 8 scheduled refreshes per day in shared capacity (Pro) and up to 48 per day in Premium, PPU, and Fabric capacity. Configure refresh failure notifications to the owner and additional contacts. The service disables a schedule after repeated consecutive failures and pauses refresh for models that haven't been used for an extended period, so notifications matter.
Incremental refresh (configured on a table in Desktop, using the RangeStart and RangeEnd parameters from 2.1.4) partitions large tables by date so each refresh reloads only recent data ("store 5 years, refresh the last 10 days"), optionally with detect data changes and, on Premium/Fabric, a real-time DirectQuery partition (hybrid table). DirectQuery and Direct Lake models don't need scheduled data refresh in the same way, although Direct Lake reframing can be scheduled or automatic and DirectQuery models still refresh metadata and caches.
⚠️ Exam Trap: A personal-mode gateway can't support DirectQuery or be shared with colleagues. When a stem says "multiple report authors must use the same on-premises SQL Server source in DirectQuery mode," the answer is a standard on-premises data gateway with the data source added and users granted access.
Reflection Question: A model combines an Excel file on a network share with an Azure SQL Database. Which source needs a gateway, and what happens to the refresh if the gateway's server is offline?
