lance-format / lance-format/lance
perf: HNSW post-alignment follow-ups — benchmark evidence and #5183 regression test
Open
Nobody has claimed this yet.
performance
- Dominant language
- Rust
- Stars
- 7.1k
- Forks
- 852
- Avg merge
- 3d 18h
- Merged PRs (30d)
- 272
Description
Follow-up from #8036, whose construction deviations were fixed by #8188.
- Benchmark evidence: recall, latency, build time, and degree distribution before/after #8188, on SIFT and clustered data. The
keepPrunedConnectionsrefill saturates level-0 reciprocal lists at2Mmore often, so index size and build time are expected to move measurably and should be quantified. - Distance-computation regression test for #5183: the SEARCH-LAYER bound refresh is correct, but lacks a regression test to lock it in. The counter from #8142 makes this nearly free once it lands.
Related: #8237 tracks aligning the mem_wal memtable HNSW construction.
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
Start by reading #8036, #8188, #5183, and #8142 to understand the HNSW alignment change and available distance-computation counter. Run before/after benchmarks on SIFT and clustered data, measuring recall, latency, build time, degree distribution, index size, and build time. Add the regression test for the SEARCH-LAYER bound refresh; done means benchmark evidence is recorded and the test locks in the expected computation behavior.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- rust
- Domain
- performance, search, testing-qa
- Issue type
- Bug
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Quiet
- Clarity
- Mostly clear
- Newbie friendliness
- 48/100