Geometry Field and RTREE Index for Geospatial Filtering

2 posts · May 2025 – Mar 2026

What this trail covers

Milvus 2.6 introduces the Geometry Field and RTREE index, enabling the combination of geospatial filtering with vector similarity search. This allows for unified storage and querying of spatial data (Points, LineStrings, Polygons) using Well-Known Text (WKT) format and efficient spatial operators like st_contains, st_within, and st_dwithin. The RTREE index accelerates spatial filtering through Minimum Bounding Rectangles (MBRs) and a hierarchical pruning approach. This integration supports real-time hybrid search scenarios.

All Milvus trails

Mar 2026How to Use Hybrid Spatial and Vector Search with MilvusThis post details the implementation of the GEOMETRY data type and RTREE index in Milvus 2.6.4. It explains the limitations of traditional two-database approaches for hybrid spatial-vector search, the internal workings of GEOMETRY (following OGC Simple Features Access standard, using WKT/WKB) and RTREE indexing (MBRs, layered construction, O(log n) complexity), and the two-phase query execution (rough filter via R-Tree, fine filter via GEOS). It also provides performance insights on query latency, accuracy, and hybrid search throughput, along with a Python SDK example for defining schemas, inserting WKT data, and running hybrid queries using ST_DWITHIN.
Dec 2025Bringing Geospatial Filtering and Vector Search Together with Geometry Fields and RTREE in Milvus 2.6This post details the introduction of the Geometry Field and RTREE index in Milvus 2.6. It explains the Geometry Field's schema definition (`DataType.GEOMETRY`) for storing Points, LineStrings, and Polygons using WKT. It also elaborates on the RTREE index's two-phase filtering process (coarse filtering with RTREE, fine filtering with exact geometry checks), its construction via MBRs, and its performance benefits derived from Boost.Geometry. The post also lists supported geometry operators and illustrates how to combine geolocation and vector indexes for joint retrieval, providing use cases in delivery services and autonomous driving.
May 2025Vector Search in the Real World: How to Filter Efficiently Without Killing RecallThis post details advanced filtering techniques for vector search, focusing on the challenges of combining metadata filtering with ANN search. It introduces and elaborates on several strategies: 1. Graph Index Optimization: Discusses the problem of 'broken connectivity' when filters remove nodes from graph indexes and explores solutions like the Alpha Strategy and ACORN Method. It highlights Milvus's dynamic strategy switching (falling back to brute-force for highly selective filters) and Cardinal's dynamic combination of 'include-all' and 'exclude-all' traversal. 2. Metadata-Aware Indexing: Explains the concept of building specialized 'subgraphs' for different metadata values (column graphs) to accelerate filtered searches, contrasting this with a base graph. 3. Iterative Filtering: Describes a batch-based approach to filtering where batches of results are retrieved and filtered sequentially until enough results are obtained, inspired by VBase. 4. External Filtering: Introduces an SDK-level solution that reverses the traditional workflow by performing vector search first and then applying custom filter functions to retrieved batches on the client side, reducing network traffic and processing overhead. The post uses examples and benchmarks to illustrate the effectiveness of these techniques.
Advertisement