Apache Paimon Vector Index: pure Rust IVF-PQ for data lake vector search.

Clone this repo:
  1. b5b4a77 fix: bound IVF-Flat serialization allocations (#77) by jerry · 11 hours ago main
  2. f3e031b [infra] Configure GitHub discussions notifications (#78) by jerry · 13 hours ago
  3. f41224a perf(ivfpq): speed up transposed PQ8 scans (#76) by jerry · 14 hours ago
  4. ee17bfd feat(logging): bridge native IVF-PQ diagnostics to SLF4J (#75) by jerry · 20 hours ago
  5. cf0506d ivf: incrementally scan expanded batch probe ranges (#73) by shyjsarah · 2 days ago

Apache Paimon Vector Index   Build Status

Apache Paimon Vector Index is a pure Rust vector indexing library designed for Apache Paimon and data lake storage such as S3, HDFS, and OSS. Its seek-based readers load only the index pages needed by each query.

The library supports IVF-FLAT, IVF-SQ, IVF-PQ, IVF-RQ, and DiskANN through shared Rust, C, C++, Java/JNI, and Python APIs. IVF-RQ stores deterministic rotated residuals as 1–8 bit planes (4 by default), uses a bounded sign-plane coarse pass before full refinement, and exposes the persisted width as rq_bits metadata. DiskANN combines a Vamana graph, resident PQ codes, paged adjacency/raw vectors, and F16 or F32 reranking for indexes queried from local SSD or object storage. Its parallel builder uses contiguous fixed-capacity adjacency, adaptive dense/sparse visited state, heap frontiers, and reusable prune scratch; query-local adjacency retention is bounded and the shared cold cache is sharded for concurrent hits.

Documentation

GitHub shows committed HTML files as source. To view the styled documentation, clone the repository and serve docs/ from the repository root:

python3 -m http.server --directory docs 8000

Then open http://localhost:8000/. All documentation assets are local, so no separate build step is required.

Source Layout

Public implementations live in core, ffi, include, jni, java, and python. See the development guide for responsibilities and verification commands.

Contributing

Getting Help

Submit an issue or ask a question in GitHub Discussions.

License

Licensed under the Apache License, Version 2.0.