MatrixOne v25.2.1.0 Release Note
We are thrilled to announce the official release of MatrixOne v25.2.1.0 on April 06, 2025!
What is MatrixOne?
MatrixOne is an AI-driven cloud-native hyper-converged database that adopts a storage-compute separation architecture and fully leverages cloud infrastructure. It is MySQL-compatible and supports hybrid workload scenarios. By combining vector data types and full-text search capabilities, MatrixOne efficiently handles multi-modal data querying and management for generative AI applications.
Feature Overview
The new version of MatrixOne comprehensively supports full-text search and BM25 relevance retrieval, significantly improving the accuracy and efficiency of text data queries. With the optimized LOAD DATA command, it enables efficient loading of large-scale data from the HDFS file system, further enhancing data processing performance. The newly added inter-cluster CDC (Change Data Capture) feature ensures real-time data synchronization and consistency across multiple clusters. Additionally, as an experimental feature, it introduces vector indexing technology based on the HNSW algorithm, providing a new solution for high-dimensional data processing and vector similarity search. Furthermore, the newly supported Python SDK for full-text and vector search lowers the development barrier, accelerating the construction of intelligent applications.
Use Cases
MatrixOne is suitable for the following application scenarios. We warmly welcome users with the following business pain points and needs to contact us for trial testing.
- Generative AI Scenarios: MatrixOne's hyper-converged database provides robust multi-modal data support, real-time retrieval, and intelligent data processing capabilities for generative AI, forming the core infrastructure for generative AI applications. In multi-modal scenarios such as text generation and image generation, MatrixOne ensures rapid responses and high-quality generative results on large-scale datasets through efficient data management, vector and hybrid search, Python UDF-supported data cleaning and preprocessing, and GPU-accelerated real-time inference. Whether handling large-scale data access and storage or online inference and dynamic feedback, MatrixOne delivers stable, low-latency support for generative AI applications, helping enterprises quickly deploy, iterate, and optimize generative AI solutions.
- Time-Series Data Applications: In modern IoT applications, billions of devices and sensors continuously collect and transmit data, including industrial production lines, smart grids, smart city infrastructure, and autonomous vehicles, generating terabytes of real-time data daily. MatrixOne's hyper-converged database provides efficient real-time data processing capabilities for IoT scenarios, supporting millisecond-level high-concurrency writes and fast retrieval while offering superior scalability to handle peak loads. Its real-time analytics capabilities enable businesses to quickly derive critical insights from massive IoT data. Seamless integration with machine learning models allows real-time data streams to feed directly into models for prediction and anomaly detection, making it ideal for industrial predictive maintenance, energy efficiency optimization, and intelligent monitoring applications, fully meeting IoT needs for high throughput, low latency, and intelligent data management.
- Hybrid Workload Support: In enterprise OA, ERP, and CRM systems, traditional single-machine databases often struggle to meet performance demands during peak periods as data volume and business complexity grow. Critical timeframes, such as month-end or quarter-end, typically require high-frequency analysis and real-time statistical reporting for decision-making. Many enterprises resort to standalone analytical databases or sharding to alleviate query loads on primary databases. MatrixOne's hybrid workload support eliminates the need for additional systems by enabling both operational and analytical needs within a single database, ensuring rapid responses under high concurrency through real-time data analytics. Its scalability allows seamless expansion as business grows, maintaining efficient real-time queries and statistics even with large-scale data growth, ensuring real-time, continuous, and efficient data-driven decision-making while enhancing flexibility in data management.
- Enterprise SaaS Scenarios: With the rapid growth of enterprise SaaS applications, SaaS development must address multi-tenant model requirements. Traditional approaches often force a choice between shared database instances for multiple tenants or dedicated instances per tenant, creating trade-offs between management costs and tenant isolation. MatrixOne natively supports multi-tenancy, providing workload isolation and independent scalability between tenants while offering unified management. This architecture effectively reduces management costs, ensures data isolation, and improves operational efficiency, fully meeting SaaS needs for cost control, ease of management, and isolation, making it an ideal database choice for SaaS applications.
New Features
Key New Features
- Full-Text Search and BM25 Relevance Retrieval
Full-text search is now fully supported, leveraging the BM25 algorithm for precise relevance retrieval, significantly improving query efficiency and accuracy for text data. - HDFS Data Loading
The optimizedLOAD DATAcommand enables users to directly load large-scale data from the HDFS file system, meeting efficient data import needs for big data scenarios. - Inter-Cluster CDC Support
The newly added inter-cluster CDC feature enables real-time data synchronization between MatrixOne clusters, ensuring data consistency and high availability for enterprise-grade real-time data management. - HNSW-Based Vector Indexing (Beta)
Introduces vector indexing technology based on the HNSW algorithm, providing a new solution for high-dimensional data processing and vector similarity search, expanding the database's applications in AI and machine learning. - Python SDK for Hybrid Search
The MatrixOne Python SDK offers developers convenient interfaces for efficient vector and full-text search, enabling vector, full-text, and hybrid retrieval while significantly reducing development complexity for rapid RAG application construction.
Other New Features
- Support select xx from xx{as of timestamp 'YYYY-MM-DD HH:MM:SS'}
- Optimized PITR-related syntax
- Optimized snapshot-related syntax
Known Issues
- CDC tasks currently only support table-level synchronization.
- Creating a CDC task requires first creating a PITR that includes the synchronization scope.
- Snapshot and PITR backups cannot restore data from deleted tenants.
- HNSW-type vector indexes currently do not support real-time index updates.