We're thrilled to announce the release of Milvus 2.5, a significant step in our journey to build the world's most complete solution for all search workloads. While Milvus has been known for its powerful vector similarity search capabilities, our vision extends beyond vectors. We believe the future of search lies in seamlessly combining different search paradigms - semantic search, keyword search, and traditional filtering - all in one unified platform.
This release marks a significant milestone with the introduction of built-in full-text search, bringing us closer to this vision. Along with advanced query optimizations and improved data processing capabilities, Milvus 2.5 represents our commitment to making complex search operations simple and efficient for developers.
Built-in Full-Text Search: Powerful Text Processing Made Simple 🔍
The headline feature of Milvus 2.5 is our new built-in full-text search powered by Sparse-BM25. This addition is more than just a new feature - it's a foundational building block for next-generation search applications. By combining vector similarity search with powerful text processing capabilities, you can build more sophisticated search systems to better understand user intent. Whether you're working with technical documentation, domain-specific content, or general text data, you can leverage semantic understanding and keyword precision in a single system.
What makes this particularly powerful is its simplicity - you can now directly input raw text without generating vector embeddings ahead of time. Milvus handles the complexity internally, using built-in analyzers to process your text and updating BM25 statistical information in real-time as you add new data. This ensures optimal search accuracy while significantly reducing implementation complexity.
Enhanced Text and Data Processing 📊
Text Match
Our new text match capability leverages Tantivy's analyzers and indexing for precise natural language matching. This feature is particularly valuable when combining text matching with vector similarity search, enabling you to filter search results based on specific terms before performing vector similarity calculations.
Bitmap Index
The new Bitmap index accelerates searches on fields with a limited number of unique values (low-cardinality fields). For example, when searching through user data with fields like "gender" with only a few possible values, Bitmap indexes can significantly speed up query performance.
Nullable & Default Values
We've added support for nullable properties and default values in scalar fields, providing greater flexibility in data management. This feature simplifies data migration from other database systems and helps handle real-world use cases where field values might be unknown or require default settings.
New Beta Features ⚡
A standout beta feature in this release is our new Cluster Management WebUI, designed to tackle the complexity of managing large-scale Milvus deployments. This comprehensive dashboard automatically monitors system health, tracks performance metrics, and alerts you to potential issues before they impact your applications. When the system detects any unhealthy patterns, you'll have direct access to Milvus experts for quick resolution.
We're also testing Clustering Compaction, which helps optimize storage and retrieval in large collections by redistributing data based on specified scalar fields. These features are available for testing, and we welcome your feedback as we continue to refine them.
Additional Features and Improvements 🌟
This release packs numerous improvements to enhance your experience with Milvus. We've improved our HNSW implementation through Faiss integration, supporting multiple quantization methods (SQ, PQ, and PRQ) for better performance and recall.
Other notable improvements include IPv6 support, CSV bulk import capabilities, and a new streaming node for Write-Ahead Logging services. We've also optimized large-scale deletions, upgraded to ETCD 3.5.16 and Pulsar 3.0.7 LTS for better security, and introduced Local Storage V2 for improved loading and query efficiency.
Join Our Hands-on Workshop! 🎓
Ready to explore these new features? You can review our detailed release notes or join our technical workshop, where we'll demonstrate how to implement full-text search, set up efficient indexes, and leverage these new capabilities in your applications. Our engineers will guide you through practical examples and best practices for getting the most out of Milvus 2.5.
And there's more exciting news on the horizon - Milvus 2.5 is targeted for public preview in Zilliz Cloud in February 2025. Stay tuned to be among the first to experience these powerful features in a fully managed environment where you can focus on building your applications while we handle the infrastructure.
Top comments (0)