Faster by Doing Less: How Harper 5.2 Engineers Database Performance

Harper 5.2 introduces a broad set of database performance improvements aimed at reducing repeated work, isolating write-related delays, improving index use, and keeping storage behavior predictable under growth and pressure. Its central change is a Verification Table-backed record cache that lets multiple worker threads safely reuse decoded RocksDB records without re-reading the database for validation, delivering warm point reads roughly 5–8 times faster and extending similar benefits to HNSW vector search, where searches and index builds improved by about 2.6 and 2.4 times respectively. The release also adds predicate-aware vector traversal for more complete filtered results, moves database commits to dedicated per-database threads to prevent heavy writes from blocking Node.js’s shared worker pool, and makes an index-aware SQL engine the default when supported, substantially accelerating tested query types that previously relied on full scans. RocksDB compression, using LZ4 by default, reduces disk use and can improve point-read throughput by fitting more records into cache, while a shared adaptive memory pool balances read caching with write buffers across tables and databases. Finally, bounded open-file management helps prevent file-descriptor exhaustion during heavy ingestion and compaction, emphasizing predictable latency and reliability as well as higher throughput.

Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.