Sonic — Fast Lightweight Search Backend in Rust

Introduction

Sonic is a fast, lightweight, schema-less search backend that replaces heavy search engines when you need simple full-text search without the overhead. Built in Rust, it focuses on speed and minimal resource usage, making it ideal for small to medium applications.

What Sonic Does

• Accepts text via an ingestion channel and builds an internal inverted index
• Responds to search and auto-suggest queries in sub-millisecond time
• Supports multiple collections and buckets for data isolation
• Provides a simple Sonic Channel protocol over TCP for integration
• Runs with under 30 MB of RAM for typical workloads

Architecture Overview

Sonic uses a custom Sonic Channel text protocol (not HTTP) split into three modes: ingest for pushing data, search for querying, and control for administration. Internally it stores an FST-based index on disk using RocksDB, keeping only hot data in memory. This design allows Sonic to handle millions of indexed objects on commodity hardware.

Self-Hosting & Configuration

• Deploy via Docker or compile from source with Rust stable
• Configuration lives in a single config.cfg TOML file
• Set authentication passwords per channel mode for security
• Tune bucket_hash_count and kv_pool based on dataset size
• Supports log-level configuration and PID file for daemon management

Key Features

• Written in Rust for memory safety and zero-GC performance
• Sub-millisecond query latency on indexed collections
• Auto-suggest and auto-complete support built in
• Tiny memory footprint compared to Elasticsearch or Solr
• Multi-tenant via collections and buckets

Comparison with Similar Tools

• Elasticsearch — full-featured but resource-heavy; Sonic trades features for speed and simplicity
• Meilisearch — HTTP-based with rich features; Sonic is lower-level with less overhead
• Typesense — typo-tolerant HTTP search; Sonic uses a custom TCP protocol and is more minimal
• Tantivy — a Rust search library, not a standalone server; Sonic wraps its own index engine
• Zinc — lightweight Elasticsearch alternative with HTTP; Sonic is even simpler with channel protocol

FAQ

Q: Does Sonic support HTTP REST APIs? A: No. Sonic uses its own Sonic Channel text protocol over TCP. Client libraries are available for most languages.

Q: Can Sonic replace Elasticsearch for large-scale analytics? A: Sonic is designed for search and suggest on text, not aggregation or analytics. It works best for small-to-medium search needs.

Q: What storage engine does Sonic use? A: Sonic uses RocksDB internally for persistent key-value storage of its inverted index.

Q: Is there a managed cloud version? A: No. Sonic is self-hosted only. Deploy it alongside your application server.

Sources

• https://github.com/valeriansaliou/sonic
• https://github.com/valeriansaliou/sonic/blob/master/CONFIGURATION.md