Apache Kvrocks

Overview

Apache Kvrocks is a distributed key-value NoSQL database that uses RocksDB as its storage engine and is compatible with the Redis protocol. Vector similarity search support is currently under active development.

Vector Search Implementation (Experimental)

Kvrocks is integrating vector indexing capabilities to support vector similarity searches with real-time processing and efficient large-scale vector data management.

Current Status

• Experimental Feature: Vector search via Kvrocks Search module (KQIR)
• HNSW Algorithm: Initial implementation focuses on on-disk HNSW index
• Field Type: VECTOR field type added to existing TAG and NUMERIC types
• Production Status: Under active development, not yet production-ready

Vector Search Features

Supported Query Types

• Vector Range Queries: @vec_field:[VECTOR_RANGE range $vec]
• KNN Queries: * => [KNN n @vec_field $vec] (without prefiltering)

Vector Field Configuration

• Data Type: FLOAT64 vectors
• Dimensionality: Configurable vector dimensions
• Distance Metrics: L2 (Euclidean), IP (Inner Product), COSINE

HNSW Parameters

• Initial capacity (default: 500,000)
• M parameter for maximum edges per node (default: 16)
• ef_construction (default: 200)
• ef_runtime (default: 10)
• Epsilon for approximate search (default: 0.01)
• Number of levels in the graph

Core Kvrocks Features

• Redis Compatibility: Compatible with Redis protocol and commands
• RocksDB Backend: Persistent storage with RocksDB
• Distributed Architecture: Built for horizontal scaling
• Disk-Based: Lower memory requirements than in-memory databases
• Kvrocks Search (KQIR): Query engine supporting SQL and RediSearch queries

Use Cases

• RAG Applications: LangChain can utilize Kvrocks as a vector database
• Large-Scale Vector Storage: Cost-effective disk-based vector storage
• Redis Migration: Drop-in replacement for Redis with vector capabilities
• Hybrid Workloads: Combine key-value operations with vector search

Architecture

Kvrocks Search encoding design efficiently maps HNSW index structures to RocksDB key-values, enabling persistent, scalable vector search on disk.

Integration

• Redis client libraries (any language with Redis support)
• LangChain for RAG applications
• RediSearch-compatible tools
• SQL query interface via KQIR

Roadmap

• Complete HNSW implementation
• Additional index types based on performance requirements
• Enhanced filtering capabilities
• Production-ready vector search module

Pricing

Free and open-source under the Apache 2.0 license.