NutsDB — Fast Embeddable Key-Value Store in Pure Go

Introduction

NutsDB is an embeddable, persistent key/value store written in pure Go. It supports multiple data structures including strings, lists, sets, and sorted sets, all within a single-file database with ACID transaction guarantees. It is designed for applications that need a fast embedded database without running a separate server process.

What NutsDB Does

• Stores key/value pairs with optional TTL expiration
• Supports list, set, and sorted set data structures within buckets
• Provides fully serializable ACID transactions
• Persists data using an append-only log with automatic compaction
• Offers prefix and range scan operations for ordered iteration

Architecture Overview

NutsDB uses a bitcask-inspired append-only write model combined with an in-memory index (B+ tree or hash map) for fast lookups. All writes go to a write-ahead entry log on disk, while the in-memory index maps keys to their on-disk positions. Background merge processes compact old log segments to reclaim space. Transactions use a copy-on-write approach with commit-time conflict detection.

Self-Hosting & Configuration

• Import as a Go module: go get github.com/nutsdb/nutsdb
• Open a database by specifying a data directory path
• Configure segment size, sync strategy, and index type via options
• Choose between B+ tree index (range scans) or hash map index (point lookups)
• Set entry-level TTL for automatic expiration of stale data

Key Features

• Pure Go implementation with zero CGo dependencies
• Multiple data structures (KV, list, set, sorted set) in a unified API
• B+ tree index mode enabling efficient prefix and range scans
• Configurable TTL per entry for cache-like expiration behavior
• Automatic log segment merging for space reclamation

Comparison with Similar Tools

• BoltDB/bbolt — B+ tree with page-level MVCC; NutsDB uses a bitcask log with richer data structures (lists, sets)
• BadgerDB — LSM-tree optimized for SSDs; NutsDB uses an append-only log with simpler compaction
• LevelDB/RocksDB — C/C++ engines; NutsDB is pure Go with no CGo overhead
• Redis — in-memory server with similar data structures; NutsDB is an embedded library with disk persistence

FAQ

Q: Does NutsDB support concurrent access? A: Yes, NutsDB supports concurrent reads and serialized writes within a single process using its transaction system.

Q: How does TTL work? A: Each entry can be stored with a TTL in seconds. Expired entries are cleaned up during reads and background merge operations.

Q: What index modes are available? A: NutsDB offers B+ tree index mode for range scans and prefix queries, and hash map index mode for faster point lookups.

Q: Is NutsDB suitable for production use? A: NutsDB is used in production by several projects. It works well for embedded storage scenarios where a separate database server is not desired.

Sources

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