Skip to content
Open
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
Original file line number Diff line number Diff line change
Expand Up @@ -150,14 +150,30 @@ A new experimental Admin UI is available alongside the existing Admin UI (SOLR-1
It can be accessed at the URL path `/solr/ui/`.
This UI is still in active development and provided as a preview; the existing Admin UI remains the default.

=== Vector Search Enhancements
=== Vector Search

==== Enhancements

* The `efSearchScaleFactor` parameter is now available for the KNN query parser (SOLR-17928). This parameter controls how many candidate vectors are explored during HNSW graph traversal, allowing users to independently tune search accuracy versus the number of results returned. Previously, improving accuracy required increasing `topK` (which returns more results), but `efSearchScaleFactor` enables exploring more candidates while still receiving exactly `topK` results. The `efSearch` value is calculated internally as `efSearchScaleFactor * topK`. Default value is `1.0`, which means `efSearch` defaults to `topK`.

* Scalar and binary quantized dense vectors are now supported for `DenseVectorField` (SOLR-17780, SOLR-17812). Quantization reduces memory consumption and can improve search performance at some cost to accuracy. See the reference guide for configuration details.

* GPU-accelerated approximate nearest neighbor search is now available via the cuVS-Lucene pluggable codec (SOLR-17892). This allows NVIDIA GPU hardware to be used for vector search workloads.

* Early termination strategy for KNN queries is now supported (SOLR-17814). `PatienceKnnVectorQuery` is a version of knn vector query that exits early the graph when HNSW queue saturates over a saturationThreshold for more than patience times.

* Lexically accelerated vector search is now supported (SOLR-17813). `SeededKnnVectorQuery` is a version of knn vector query that introduces a “seed” query, allowing the search to start from a predefined subset of documents and guide the vector similarity computation.

* The `filteredSearchThreshold` parameter is now available to regulate ACORN-based filtering in vector search (SOLR-17815). This approach addresses the performance limitations typically associated with pre-filtering and post-filtering strategies by modifying both the construction and search phases of the HNSW graph.

* Fixed incorrect behavior of `TextToVectorUpdateProcessor` during partial updates (SOLR-17843).

==== Renaming of HNSW Parameters
Attention:

* The `llm` module has been renamed to `language-models`.
* The HNSW parameters `hnswMaxConnections` and `hnswBeamWidth` have been renamed to `hnswM` and `hnswEfConstruction`, respectively, so they must be updated accordingly in the schema.xml file.

=== Deprecation Code Removals

* Several deprecated modules have been removed.
Expand Down Expand Up @@ -293,11 +309,9 @@ Due to Lucene 10 changes (https://github.com/apache/lucene/pull/12875), `PathHie
</fieldType>
----

=== Renaming in Vector Search
Attention:
=== Learning To Rank (LTR)

* The `llm` module has been renamed to `language-models`.
* The HNSW parameters `hnswMaxConnections` and `hnswBeamWidth` have been renamed to `hnswM` and `hnswEfConstruction`, respectively, so they must be updated accordingly in the schema.xml file.
A new feature vector cache was added that is used not only for feature logging but also for the reranking phase (SOLR-16667).

== Solr 10.1

Expand Down Expand Up @@ -335,3 +349,7 @@ If you require the ability to roll back, back up your indexes before upgrading.
=== Docker

The `gosu` binary is no longer installed in the Solr Docker image. See https://github.com/tianon/gosu[gosu github page] for alternatives, such as `runuser`, `setpriv` or `chroot`.

=== Query Changes

* Combined Query Feature is now available, enabling the execution of multiple queries of multiple kinds across multiple shards (SOLR-17319). Introduced `CombinedQuerySearchHandler` for hybrid search, using reciprocal rank fusion (RRF) by default and supporting custom ranking algorithms via plugins.
Loading