Boosting OpenSearch Performance: Lucene Bulk Collection and gRPC Search in Action
summary
Lucene’s new bulk collection APIs (LongValues, stream.intoArray, collectRange) batch doc handling so JVM can inline and avoid virtual dispatch, producing measured aggregation speedups up to 40% faster and reducing per-call costs (demo shows mixed-object invocation dropping from 3.73 ms to 1.28 ms). gRPC Search in OpenSearch 3.4+ expands query support (match, fuzzy, match_phrase, prefix, constant_score), auto-detects multiple document formats and yields 20–30% KNN throughput gain in benchmarks while remaining experimental; binary transport still lands in JSON/CBOR parsing internally.
AI-written summary · about
chapters (9)
- 0:00Lucene bulk collection APIs
- 2:09Aggregation performance challenges
- 4:20JVM performance concepts
- 11:24Method inlining optimizations
- 15:38Lucene bulk collection changes
- 19:31gRPC in OpenSearch
- 21:15gRPC features and queries
- 24:50gRPC document format
- 27:46Performance improvements summary
related talks
40mLeveraging Lucene Skip Indexes for Faster Aggregations in OpenSearch
41mScaling Search and Powering Vector-Based Retrieval With OpenSea...
04mOpenSearch Performance and Monitoring
45mAccelerating OpenSearch With Streaming: Apache Arrow, Flight, Data...
26mOpenSearch and the Path to a Trillion-s...
18mSpeed Up Your Search: Concurrency and Merge Policies to the Rescue
05mKeynote: OpenSearch: From Code to Community
28mRunning OpenSearch at Scale in High-Traffic Gaming Systems