Lakehouse tables and files

A Lakehouse source reads data straight from a Lakehouse in your OneLake — the most direct way to map data you already hold in Fabric.

Tables

Connect to a Delta table in a Lakehouse. Icon Map reads the table in place; use Detect columns to pull in the schema, then choose the columns you need and set a location (longitude/latitude columns or a geometry column).

Files

Connect to a file in the Lakehouse's Files area:

  • Parquet (including GeoParquet)
  • CSV
  • GeoJSON

Files are useful for reference datasets and extracts that aren't managed as tables.

GeoParquet: automatic setup and viewport streaming

Pick a GeoParquet file and Icon Map configures it for you: the geometry column, its encoding, and the file's coordinate system are read from the file's own metadata - no manual location setup.

Large GeoParquet files can also stream by viewport. When the file is streaming-ready - spatially sorted, with a bounding-box (bbox covering) column - Icon Map turns on Only load data in the current map view automatically and then fetches only the parts of the file that intersect the view as you pan and zoom, using the file's own statistics. A country-scale file behaves like a tile service, but with exact geometries and every attribute column intact. The data source panel shows whether a file is streaming-ready and how many row groups it has.

To make a streaming-ready file, export with a spatial sort and a bbox column - for example with GDAL (SORT_BY_BBOX=YES, WRITE_COVERING_BBOX=YES) or DuckDB spatial (ORDER BY ST_Hilbert(...)). Datasets ingested by the Organizational Catalog are written streaming-ready automatically. Files without the bbox column still work - they simply load in full.

Good to know

  • Reads happen as you, using your Fabric permissions — you only see data you can already access.
  • Use a row limit and, for large tables, Only load data in the current map view to keep the map responsive.
  • Viewport streaming works with any spatial layer over the source — and charts or data tables on the same source read just the columns they need, so one GeoParquet file can drive the map and its analytics together.

Next steps