Enable caching of all filters in knn queries
#134458
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This change makes all filters in the `knn` query eligible for query caching. By default, Lucene considers some simple filters (e.g., term queries) too cheap to cache. In the context of vector search, these filters are eagerly materialized as bitsets, which makes them significantly more expensive to evaluate on every request. Forcing them to be cacheable avoids repeated recomputation. This is a stop-gap change to support simple use cases such as a single term query used as a filter in `knn`. The long-term solution is to move this decision logic into the Lucene `knn` codec itself, but that will require more time. ### Benchmark Results Dataset: **20M 128D vectors**, term filter matching \~80% of documents. **With this change:** ``` Precision QPS P50 (ms) P95 (ms) 0.91 632.8 5.763 9.900 ``` **Without this change:** ``` Precision QPS P50 (ms) P95 (ms) 0.91 68.2 82.52 193.92 ```
Collaborator
|
Pinging @elastic/es-search-relevance (Team:Search Relevance) |
Collaborator
|
Hi @jimczi, I've created a changelog YAML for you. |
…_query_filter_cache
benwtrent
reviewed
Sep 10, 2025
...n/java/org/elasticsearch/index/cache/query/ElasticsearchUsageTrackingQueryCachingPolicy.java
Outdated
Show resolved
Hide resolved
benwtrent
approved these changes
Sep 10, 2025
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This change makes all filters in the
knnquery eligible for query caching. By default, Lucene considers some simple filters (e.g., term queries) too cheap to cache. In the context of vector search, these filters are eagerly materialized as bitsets, which makes them significantly more expensive to evaluate on every request. Forcing them to be cacheable avoids repeated recomputation.This is a stop-gap change to support simple use cases such as a single term query used as a filter in
knn. The long-term solution is to move this decision logic into the Luceneknncodec itself, but that will require more time.Benchmark Results
Dataset: 20M 128D vectors, term filter matching ~80% of documents.
With this change:
Without this change: