Summary

Datalevin is a simple, fast, and versatile open-source Datalog database for developers preferring declarative querying over SQL. It offers high performance, ACID compliance, and broad functionality.

How It Works

Datalevin implements a Datomic-style Datalog query language with a novel cost-based query optimizer, achieving performance competitive with SQL and graph databases. It's built on a fork of LMDB for durable storage and high read throughput, supporting ACID transactions, WAL, and asynchronous transactions for write-intensive workloads. This approach prioritizes simplicity and familiar semantics.

Quick Start & Requirements

Primarily a Clojure library, Datalevin offers embedded JVM, client/server (port 8898), and babashka pod options. Requires a Clojure environment; Java consumers need Maven Central. See doc/install.md and https://cljdoc.org/d/datalevin/datalevin for details.

Highlighted Details

• Performance: Benchmarks show significant gains over PostgreSQL/SQLite on complex queries (e.g., 2.4x faster than PostgreSQL on JOB-bench) and superior write transaction speeds (over 20x faster than SQLite async). Outperforms Datomic/Datascript on complex reads.
• Versatility: Functions as a Datalog DB, high-performance EDN key-value store, document DB (JSON/EDN/Markdown with path indexing), and integrates vector DB capabilities with SIMD search.
• Deployment: Supports embedded library, client/server with RBAC, and babashka pod integration.
• Advanced Features: Novel full-text search engine and an advanced, high-performance rule engine.

Maintenance & Community

Actively developed with a clear roadmap to v3.0.0, including HA and JSON APIs. Used in production by Juji. Community engagement via #datalevin on Clojurians Slack.

Licensing & Compatibility

Licensed under the permissive Eclipse Public License (EPL), suitable for commercial and closed-source applications.

Limitations & Caveats

Intentionally omits Datomic's temporal features for simplicity. Features like read-only replicas, high availability, and a JSON API are planned for v1.0.0 and are not yet implemented.