Query object storage and enterprise databases, generate vector tiles from current data, and expose cloud-hosted imagery through a client-compatible ImageServer surface—without forcing every source through a proprietary publish-and-copy cycle.
The DuckDB spatial engine creates Mapbox Vector Tiles with ST_AsMVT() for the requested z/x/y view. This avoids treating the tile pyramid as a second authoritative dataset.
DuckDB httpfs, spatial, Azure and Delta extensions provide a zero-copy query path with projection and predicate pushdown where the underlying format supports it.
| Source | Connection model | Data paths | What the engine can do |
|---|---|---|---|
| Amazon S3 / MinIO | Access key/secret, endpoint and region; anonymous public buckets supported | s3://, S3-compatible endpoints, HTTPS | Browse objects, inspect schemas, query Parquet/GeoParquet/CSV/GeoJSON, register spatial result collections |
| Google Cloud Storage | HMAC interoperability keys through the S3-compatible API | gs:// and configured endpoint access | Browse, query and register cloud datasets through the same SQL workflow |
| Azure Blob Storage | Storage account/key or SAS-oriented deployment configuration | az://, abfs://, HTTPS | Browse containers, query supported columnar/spatial files and retain source-aware connection profiles |
| Microsoft Fabric OneLake | Azure storage semantics plus DuckDB Delta extension | OneLake ABFS endpoints | Read Delta Lake table directories with delta_scan and join to spatial sources |
| HTTP/HTTPS | Public or signed URL | Remote Parquet, CSV, GeoJSON, COG and PMTiles | Range-read cloud-optimized assets or query supported tabular formats |
| Enterprise databases | Saved, credential-redacted connection profiles | PostgreSQL/PostGIS, MySQL, SQL Server, Oracle, Snowflake, ClickHouse, BigQuery, SQLite, DuckDB | Discover tables, execute SQL, map business data to geometry and register tables as service collections |
A Cloud Optimized GeoTIFF in a web-addressable S3, Azure Blob, GCS or compatible object-store endpoint can be registered by HTTPS URL. A public object or time-limited signed URL is the straightforward secure pattern in the current range reader.
The server checks the remote object and reads TIFF metadata without downloading the complete raster.
HTTP Range requests fetch IFD metadata and the image blocks required for the requested map extent and resolution.
ArcGIS ImageryLayer-compatible metadata, exportImage, identify, samples, histograms, statistics and appropriate tile endpoints.
The object remains a standards-based GeoTIFF that GDAL, QGIS, rasterio and cloud-native pipelines can also use.
Virtual collections are the core abstraction: a queryable dataset is registered once and presented through the interface each consumer expects.
The comparison below describes the 3DGeoServEngine design relative to common publish-and-cache GIS deployments. It is not a claim that no other platform can be customized to achieve similar patterns.
| Concern | Conventional publish/cache pattern | 3DGeoServEngine pattern |
|---|---|---|
| Vector visualization | Generate and maintain a separate cached tile product. | Generate MVT dynamically from the current spatial query, or use a stored tile archive when immutability is preferable. |
| Cloud imagery | Copy imagery into a managed raster store or build a platform-specific mosaic/service definition. | Register a remote COG URL and expose it as ImageServer while retaining the original cloud-native file. |
| Client compatibility | Publish a separate service for each client/protocol. | Expose multiple service contracts over the same virtual collection. |
| Analytical access | ETL data into the GIS datastore before spatial query. | Use DuckDB to query supported object-store formats in place and materialize only when performance, governance or DDIL requires it. |
| Offline/edge | Depend on the central portal and prepare offline packages separately. | Use the same service binary with staged GeoPackage, MBTiles, PMTiles and COG data at the edge. |