Speed up your Lucene queries by avoiding searching
summary
Shows concrete tricks that turn slow Lucene queries into near-instant results: split hot filters into separate indices so those shards can run as term queries instead of expensive conjunctions (benchmarks show ~1,500 q/s per core for term lookups vs ~10–50 for complex queries), use index sorting (sorting by brand/price collapses millions of posting-list jumps into a few seeks), and index shingles for phrase queries (works but shingles increase index size and only helps exact-slap phrases). Also mentions constant_keyword and ~20% indexing overhead for sorting.
AI-written summary · about