Vector similarity queries report isCacheable() == true although their per-segment result depends on live docs, the search timeout and the filter, so a shared query cache serves one reader's or one search's result to another
Finding LFV-106Area: search / vector similarity queries and the query cachePriority: highImpact: silent wrong resultsNovelty: new
AI-assisted finding, confirmed by automated checks, not yet triaged by a human Lucene committer. This site files nothing upstream.
What it shows
You search with a vector similarity (threshold) query, for example FloatVectorSimilarityQuery, in a non-scoring context such as IndexSearcher#count or ConstantScoreQuery, with a query cache that several readers or searchers share (the IndexSearcher default in 9.10.0-10.6.0, 10.6.0 by reading). You expect every reader and every search to get the hits it gets without the cache. Instead the cache hands one reader's or one search's per-segment result to another: an older point-in-time reader misses a document that only a newer reader has deleted (99 hits instead of 100), a search without a timeout gets the empty result of an earlier timed-out search (0 instead of 100), and with a filter on an updated doc-values field a newer reader gets a document that no longer matches the filter (100 instead of 99).
The test shows that FloatVectorSimilarityQuery says its per-segment result is cacheable, but the result depends on live docs, so a shared query cache gives an older reader the result of a newer reader that deleted a document (same segment core).
Expected vs actual
Expected count 2 on the older reader, where doc 0 is still live; actual 1 (expected:<2> but was:<1>), with -ea and -da.
Files
Run it
- Tested on Lucene main e1faeba08bde (2026-11-02): still reproduces.
- Tested on Lucene main 645d4fe14a20 (2026-10-28): reproduced.
- Recorded outcome of run 1 on Lucene main 645d4fe14a20: 1 test(s) run, 1 failed.
- How to read the outcome: a failing test is the finding reproducing (the failure message states expected against actual).
- Seeds: a rerun with the same seed replays the same random choices. Assertions off (-da -dsa -Dtests.asserts=false) is the production-like setting.
# Lucene checkout at the tested commit, built with: ./gradlew --no-daemon assemble (JDK 25)
LUCENE=/path/to/lucene; V=12.0.0-SNAPSHOT
CP=$LUCENE/lucene/core/build/libs/lucene-core-$V.jar
CP=$CP:$LUCENE/lucene/codecs/build/libs/lucene-codecs-$V.jar
CP=$CP:$LUCENE/lucene/test-framework/build/libs/lucene-test-framework-$V.jar
CP=$CP:randomizedtesting-runner-2.9.2.jar:junit-4.13.2.jar:hamcrest-3.0.jar # test-framework dependencies (Maven Central)
CP=$CP:junit-platform-console-standalone-6.1.3.jar # JUnit Platform console launcher (Maven Central)
javac -d classes -cp "$CP" $(find src -name '*.java')
# run 1: assertions on, seed DEADBEEF
java -ea -esa --add-modules jdk.incubator.vector --enable-native-access=ALL-UNNAMED -Dtests.seed=DEADBEEF \
-cp "classes:$CP" org.junit.platform.console.ConsoleLauncher \
execute --select-class org.apache.lucene.search.TestVectorSimilarityQueryCacheGradle alternative: copy the test into the matching module's src/test/ tree of the checkout and run ./gradlew -p lucene/<module> test --tests <class> -Ptests.seed=<seed> (add -Ptests.asserts=false for a run without assertions).
Formal models and scripts this finding's part of the hunt used: knn-query-layer
Feedback
Open an issue in lucene-ai-bug-hunt. Please do not open issues or pull requests in apache/lucene about this finding unless a Lucene committer agrees it should be filed.
Snapshot of 2026-11-04. AI-assisted; not yet triaged by a human Lucene committer. Feedback: open an issue in lucene-ai-bug-hunt.