HeadlinesBriefing favicon HeadlinesBriefing.com

Slater: Low-memory graph database

Hacker News •
×

Slater is a low-memory graph database designed to serve graphs that exceed available RAM, handling hundreds of millions of nodes and billions of edges using only hundreds of MB of RAM. It achieves this by paging the graph from an on-disk image, decoupling graph size from memory requirements. This allows for cost-effective scaling with numerous read replicas.

Slater supports standard Bolt protocol, making it compatible with existing neo4j drivers. It also features disk-native vector search alongside the graph data. The database supports live, durable writes through an opt-in log-structured-merge (LSM) layer that doesn't impact read performance. Writes are accumulated in a write-ahead log and in-memory table, eventually consolidated into a new immutable core.

The system offers features like bounded, predictable memory usage set by a cache budget, multi-tenant support, at-rest encryption, TLS, and a lightweight installation. Graphs can be updated via atomic file swaps, enabling zero-downtime deployments. Slater is built with safety and performance in mind, using Rust and avoiding unsafe code where possible, ensuring read operations are lock-free and do not contend with writes.