Lucene: Status=Open

47.3 minutes ago 52 comments 0 votes 0 watches benwtrent, epotyom, jpountz, mikemccand, msokolov, rmuir, scampi, shubhamsrkdev, stefanvodita, uschindler

Description We are seeing regressions on lucene nightly benchmarks on 29/01 The set of commits ... We should find the root cause of the regression. Version and environment details_No response_

msokolov 47.3 minutes ago:	Exciting! I clicked on one of youre "here, here, here, ..." links above hoping to see the pictures of the ...
benwtrent 54.8 minutes ago:	Great stuff @mikemccand !!! For vectors, it might be a real regression? Or the JDK downgrade harmed something.

#15742 PR: Optimize int4 vector computations by avoiding conversions

2.1 hours ago 16 comments 0 votes 0 watches github-actions[bot], kaivalnp, mccullocht

Spinoff from 15736, where @mccullocht identified vector conversions as a potentially slow area ( ... This PR loads and operates on SIMD registers of the preferred bit size to avoid intermediate ... This was sparked from 15697 where we observed a performance drop in JMH benchmarks of some 4-bit ...

kaivalnp 2.1 hours ago:	Thanks @mccullocht! Looks like performance is equivalent or better in all functions (except small dip in ... I'll merge + backport this to 10.x in a day or so!
mccullocht 3.4 hours ago:	As of ac60bae35de6b521c60c161619ebe184820baaa8 ` Benchmark ...

#12440: Make HNSW merges faster

2.9 hours ago 20 comments 0 votes 0 watches benwtrent, hpvd, jbellis, jmazanec15, mbrette, mccullocht, tteofili, zhaih

Description It is well known that when merging segments, HNSW is not merged in linear time. We lose a fair bit of useful information as all the NSWs are effectively thrown away (except for a ... This means we have useful distributed work from each segment that is ignored.

mccullocht 2.9 hours ago:	Looking at the two graph merge in the paper it seems like the idea is to seed the searches ... This is necessary but doesn't seem sufficient -- you avoid navigation (you could probably start ... Even if the initial candidate list by seeding is the correct top N we still have to visit each of ...
benwtrent 1.1 days ago:	New paper just dropped, seems like it could really help with merging! https://arxiv.org/pdf/2602.17099

#15722 PR: Add interface to do prefetch on KnnVectorValues and an example implementation to use prefetch in ...

2.9 hours ago 18 comments 0 votes 0 watches benwtrent, github-actions[bot], mccullocht, mikemccand, navneet1v, vigyasharma

Description This change add the ability to do prefetch for KnnVectorValues. Along with the change, there is a sample implementation of how to add create a PrefetchableFlatVecto ... Places where this prefetch can be useful: 1.

navneet1v 1.5 days ago:	@mccullocht , @benwtrent can I get a review on this PR? Since I have moved out prefetching logic outside. There should be no impact on current HNSW algorithm
navneet1v 5.7 days ago:	Hi @mccullocht and @benwtrent I took another pass at the code where I cleaned up the prefetch ... I have also added an example implementation of how to do prefetch (PrefetchableFlatVectorsScorer). This should now keep things clean and separated.

#15727 PR: Fix incorrect handling of missing values in ToParentBlockJoinSortField (#15548)

7.2 hours ago 30 comments 0 votes 0 watches github-actions[bot], gmarouli, mccullocht, romseygeek

Description **Problem** ToParentBlockJoinSortField did not handle missing values for child ... When a parent had a mix of children with a value and children without, the children without values ... This led to incorrect sort ordering -- the missing children should participate in the comparison ...

gmarouli 1.3 days ago:	I see, so we could use the same approach based on the child sortMissingLast. If parent and child have different reverse or sortMissingLast will be able to differentiate between ... Thank you!
romseygeek 1.3 days ago:	TermOrdValComparator has this snippet: ` if (sortMissingLast) { missingOrd = ...

#15768: Add support for BM25's k3 parameter to query parsers

9.1 hours ago 0 comments 0 votes 0 watches jpountz

Description Currently, duplicate terms in query strings see their scores boosted by their number ... It would likely be better to saturate the contribution of such repeated terms, as suggested in the ... Terrier: https://github.com/terrier-org/terrier-core/blob/54e6faf802bb6275b53ad6de14a93a53389719b8/ ...

#15767 PR: Ensure that index is actually there in LegacyBaseDocValuesFormatTestCase

9.3 hours ago 1 comments 0 votes 0 watches romseygeek

In rare cases, testSortedNumericsSingleValuedMissingVsStoredFields in LegacyBaseDocValuesFormatTest ... This adds an explicit commit() call so that we avoid spurious failures, in line with other ...

romseygeek 9.3 hours ago:

This could conceivably go into a 10.4.1 if we release one? I'll wait until 10.4.0 is actually finished before adding a Changes entry.