rust-lang / rust
github.com·1d ago
Developer Tools
Databricks Blog·18h ago·5 min read
Sizing Lakebase Search: lessons from a real app and 4 million passages
TL;DR
Databricks' Slide Hub app utilizes Lakebase Postgres to enable efficient search across 700 internal presentations, demonstrating the database's performance with real user data. Key findings reveal how to size compute resources effectively for search functionalities in applications.
✦ Why It Matters
Evaluate your compute size for search features in Lakebase Postgres before launching your application.
Key Takeaways
- Understand how to size compute resources for Lakebase Postgres.
- Utilize Lakebase Search extensions for integrated search functionality.
- Benchmark performance using real user data from applications.
- Plan for scaling from 1M to 4M rows in your database.
Full Summary
Slide Hub is an internal application at Databricks that allows field engineers to search for and reuse slides from approximately 700 presentations, including event talks and product decks. It employs a search bar that retrieves slides based on meaning and keywords, supported by a Model Context Protocol (MCP) server for AI agents.
The app's data, including slides and search telemetry, is stored in Lakebase Postgres, which serves as a benchmark for performance under real workloads. The post discusses the behavior of Lakebase Search as it scales from 1 million to 4 million rows, providing insights into its limits and costs.
Lakebase Search, launched in September, consists of Postgres extensions that operate alongside application tables, eliminating the need for a separate search system. This integration simplifies transaction management for developers.