Indexing, or How We Made Indexes Shared and Fast
summary
Indexing cost drops by prebuilding and sharing index data: JetBrains converts indexes (about 99 in Community, ~208 in Ultimate) into a trigram index and a hash-to-file mapping, builds them with a headless generator on TeamCity, uploads ~700 GB to an HTTP/CDN, and IDEs download ~30 MB JDK or library bundles. Project tests show ~50–65% faster cold indexing (some cases halve codebase time); public Maven coverage includes 123 libraries and the feature is live.
AI-written summary · about
chapters (11)
- 0:00Introduction to Eugene
- 4:00Evolution of indexing
- 7:29Performance of code search
- 10:56Trigram index mechanics
- 14:19The necessity of indexing
- 19:25Concept of shared indexes
- 22:55Implementing shared indexes
- 30:50Real-world performance data
- 37:08Measuring and enabling indexes
- 41:14Availability and final thoughts
- 47:16Q&A and closing
related talks
31mLet's focus on IDE Indexing
42mLook Ma – No Indexing!
54mHow IntelliJ IDEA Performance Improved: Past, Present, and Future
07mIndexed Find in Files
46mDogfooding Developer Productivity: The Development Process at JetBrains
36mLessons from building GitHub code search
46mHow we’re improving performance of IntelliJ IDEA Kotlin plugin
48mJava's Hidden Gems: Tools and Libraries