DiceDB — Reactive Key-Value Store with Query Subscriptions

Introduction

DiceDB is a fork of Valkey (the Redis successor) that adds reactive query subscriptions to the key-value model. Clients can subscribe to query results and receive automatic push notifications when the underlying data changes, eliminating the need for polling.

What DiceDB Does

• Provides a Redis-compatible in-memory key-value store with sub-millisecond latency
• Supports query subscriptions that push updates when watched data changes
• Implements multi-threaded I/O for higher throughput than single-threaded Redis
• Offers hierarchical storage tiers for managing hot and cold data
• Maintains wire-protocol compatibility with Redis and Valkey clients

Architecture Overview

DiceDB is written in C and extends the Valkey codebase with a reactive subscription engine. The query subscription system tracks which keys affect each registered query and triggers re-evaluation when those keys are modified. A multi-threaded I/O layer distributes client connections across CPU cores for improved concurrency compared to the traditional single-threaded Redis event loop.

Self-Hosting & Configuration

• Deploy with Docker using the official DiceDB image on port 7379
• Build from source on Linux or macOS with standard C build tools
• Configure memory limits, persistence, and eviction policies via a config file
• Connect using any Redis-compatible client library in any language
• Monitor performance with built-in INFO and SLOWLOG commands

Key Features

• Reactive subscriptions push real-time updates to clients without polling
• Redis wire-protocol compatibility means existing tools and libraries work unchanged
• Multi-threaded architecture for better utilization of modern multi-core hardware
• Tiered storage moves cold data to disk while keeping hot data in memory
• Low operational overhead with a single binary and minimal configuration

Comparison with Similar Tools

• Redis / Valkey — The baseline key-value store; DiceDB adds reactive subscriptions and multi-threaded I/O on top
• DragonflyDB — Multi-threaded Redis alternative focused on raw throughput, but no query subscription feature
• KeyDB — Multi-threaded Redis fork with active replication, but not actively maintained
• Garnet — Microsoft Research cache store in C#, optimized for .NET workloads
• Apache Ignite — Distributed in-memory computing platform, much heavier operational footprint

FAQ

Q: Is DiceDB compatible with existing Redis clients? A: Yes. DiceDB uses the Redis wire protocol, so existing client libraries (redis-py, ioredis, Jedis, etc.) work without changes.

Q: What are query subscriptions? A: Query subscriptions let a client register a query (like GET or filtered key patterns) and receive automatic notifications when the result changes, similar to database triggers but at the client level.

Q: How does DiceDB compare to Redis in performance? A: DiceDB aims for comparable or better throughput than Redis for standard operations, with the multi-threaded I/O layer providing an advantage on multi-core systems.

Q: What is the license? A: DiceDB is licensed under BSD 3-Clause, the same as the original Valkey/Redis license.

Sources

• https://github.com/dicedb/dicedb
• https://dicedb.io