Skip to content
CSA Loom — the Microsoft Fabric experience for Azure tenants where Fabric isn't yet available: lakehouses, warehouses, notebooks, semantic models, Activator rules, Data Agents, across Commercial, GCC, GCC-High, and DoD IL5

Loom Geo Dataset Editor — Azure-native parity spec

Reference: ADLS Gen2 + GeoJSON (RFC 7946) + Shapefile (ESRI) + GeoParquet 1.0 spec + Azure Maps Creator Dataset, captured 2026-05-26 by the catalog agent. There is no Fabric equivalent — geo-dataset is a Loom-native item that backs both geo-map rendering and geo-query spatial joins.

What's there in the Azure-native experience

Storage layouts

  • GeoJSON — single FeatureCollection per file; native to Azure Maps Web SDK DataSource.importDataFromUrl()
  • Shapefile — .shp + .shx + .dbf + .prj quad on ADLS; converted to GeoJSON on read via OGR / GeoPandas in Spark
  • GeoParquet 1.0 — columnar Parquet with geometry column (WKB) and geo metadata block; native read in Databricks 17.1+ ST functions and Synapse Spark via geopandas + pyarrow
  • KML / KMZ / GPX / GML / WKT / CSV — supported via atlas.io.read() (Spatial IO module) and Spark sedona package
  • Azure Maps Creator dataset — IMDF-converted indoor map, vendor-specific

Azure Maps Creator dataset blade

  • Dataset header: Microsoft.Maps/accounts/{name}/creators/{creator}/datasets/{datasetId} + alias
  • Conversion source — pointer to the uploaded DWG/IMDF zip
  • Feature collections — WFS-queryable: unit, level, facility, opening, directoryInfo, lineElement, areaElement
  • Tileset — derived vector tile output
  • Style + Map configuration — defaults applied when rendered

Inspector observed in raw GeoJSON/GeoParquet workflows

  • Feature count
  • Bounding box [west, south, east, north]
  • Geometry type histogram (Point / LineString / Polygon / Multi* / GeometryCollection)
  • Property schema (column name → type → sample value)
  • SRID (default 4326 / WGS84; reproject via st_transform)
  • Per-feature inspector: properties JSON + geometry preview on a mini-map

What Loom already has

  • ✅ GeoDatasetEditor stub captures an ADLS path
  • ✅ Honest sample T-SQL OPENROWSET over Parquet to inspect via existing /api/items/synapse-serverless-sql-pool/[id]/query route
  • ✅ Cosmos-backed state save through the generic item shell

What's missing for parity

  1. Format auto-detect — sniff the ADLS path (.geojson / .parquet + geo metadata / .shp quad / .kml / .kmz / .gpx) and label the dataset accordingly
  2. Feature count + bbox — backend computes feature count and bbox from a head-scan of the file (GeoJSON streaming parse / Parquet metadata)
  3. Geometry type histogram — surface "12,431 Point · 84 Polygon · 0 LineString" with badges
  4. Property schema panel — list properties with type inference + sample values (first 10 features)
  5. Sample feature inspector — paginated grid of N features with properties JSON + a mini Azure Maps preview of the geometry
  6. SRID display + reproject action — read the SRID from prj (Shapefile) or crs (GeoJSON) or geo block (GeoParquet); offer "Reproject to 4326" that runs geo-pipeline
  7. Conversion tool — convert Shapefile or KML → GeoJSON or GeoParquet via a one-click Synapse Spark job
  8. Validation report — flag invalid geometries (self-intersection, wrong ring order, NaN coords); offer "Fix with geo_polygon_simplify" action
  9. Link to geo-map — quick action "Open in map" creates or attaches to a sibling geo-map item
  10. Creator dataset mode — when the dataset is a Maps Creator IMDF dataset, swap the inspector for the facility/level/unit tree

Backend mapping

Loom surface Real Azure call
Sniff format ADLS REST HEAD /containers/{c}/path/{p} + GET ?range=0-65535 for magic bytes
Feature count + bbox (GeoJSON) Streaming parse via azure-storage-file-datalake Python SDK + ijson
Feature count + bbox (GeoParquet) pyarrow.parquet.read_metadata → geo field block + num_rows
Schema + sample Synapse Serverless OPENROWSET BULK '<adls-https>' FORMAT='PARQUET' TOP 10 (Parquet) or Spark read.json for GeoJSON
Convert Shapefile → GeoJSON Synapse Spark job using geopandas + fiona, output written back to ADLS
Reproject Spark geopandas.to_crs(4326) or Databricks st_transform(geom, 4326)
Creator dataset tree ARM GET .../creators/{c}/datasets/{d}/featurestatesets + WFS GET .../wfs/datasets/{d}/collections

Required Azure resources

  • ADLS Gen2 container at geo/datasets/<id>/ (one folder per dataset; reuses existing Loom lakehouse storage)
  • Synapse Serverless SQL pool (already deployed) for OPENROWSET preview
  • Synapse Spark pool (already deployed) for conversion + reprojection jobs
  • (optional) Microsoft.Maps/accounts/creators for Creator dataset mode
  • Role: managed identity needs Storage Blob Data Contributor on the container

Build plan

Phase Work
Backend New /api/items/geo-dataset/[id]/inspect → format sniff + feature count + bbox + geometry type histogram. New /api/items/geo-dataset/[id]/schema → property schema + sample 10 features. New /api/items/geo-dataset/[id]/convert (POST) → submits a Synapse Spark job to convert Shapefile/KML → GeoJSON/GeoParquet. New /api/items/geo-dataset/[id]/reproject (POST) → runs geo-pipeline instance to reproject.
Frontend Replace stub form with: format badge + feature count + bbox map preview at the top; tabs for Schema · Sample features · Validation · Convert/Reproject. Each row in the sample-features grid opens a side panel with the full property JSON and a mini Azure Maps preview.
Auto-pairing On geo-dataset create from a geo-map "create new source", auto-set the path under geo/maps/<mapId>/dataset/<datasetId>/ so the map can read it without extra config.

Estimated effort

1–2 focused sessions. Backend inspect + schema (session 1, mostly Python + Synapse), conversion/reprojection wiring + frontend grid (session 2).