When IndexWriter#addBatch rejects a vector part-way through a DENSE vector column, the core HNSW and scalar-quantized vectors writers fall out of step with a flat field writer that keeps the default addDenseValues (new in #16734): the next commit closes the IndexWriter, or with assertions off commits a corrupt HNSW graph or misaligned COSINE quantization state

Finding LFV-579Area: codecs / vectors writers (Lucene99HnswVectorsWriter and Lucene104ScalarQuantizedVectorsWriter field writers over a flat field writer; columnar addBatch DENSE vector bulk path)Priority: lowImpact: data corruptionNovelty: new

Finding page for LFV-579

AI-assisted finding, confirmed by automated checks, not yet triaged by a human Lucene committer. This site files nothing upstream.

What it shows

You index vectors with IndexWriter#addBatch and a DENSE VectorColumn into an HNSW or scalar-quantized format whose per-field flat storage is your own FlatFieldVectorsWriter (plugged in through the public Lucene99FlatVectorsWriter strategyFactory constructor, without overriding addDenseValues), and Lucene rejects the third vector of a 3-document batch (NaN, or a zero vector with COSINE). You expect what the addDenseValues javadoc and Lucene's own recovery test promise: the batch's documents are deleted, the writer stays open and consistent, and the next commit writes a readable, searchable segment. Instead the flat writer has kept the batch's first two vectors while the HNSW wrapper added no graph nodes for them (and the scalar-quantized wrapper no centroid sums or COSINE magnitudes), so with assertions on the next commit fails ('Level 0 expects to have all nodes', or an assertion in the quantizer) and closes the writer, losing every uncommitted document, and with assertions off it commits an HNSW graph that cannot be read back (EOFException, CheckIndex at level 2 or more fails, a kNN query for a document's own vector returns another document), or the COSINE quantized writer throws ArrayIndexOutOfBoundsException at commit or commits NaN codes so that kNN finds no document of the segment.

after IndexWriter#addBatch rejects the third vector of a DENSE vector batch, the stock HNSW writer is out of step with a flat field writer that keeps the default addDenseValues, so the next commit fails or writes a broken graph.

Expected vs actual

Expected: commit succeeds and CheckIndex is clean; actual: AssertionError: Level 0 expects to have all nodes at commit (-ea), EOFException reading the HNSW graph in CheckIndex (-da).

Files

• src/org/apache/lucene/codecs/lucene99/TestDenseBatchHnswOverDefaultDelegate.java (JUnit test)

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.codecs.lucene99.TestDenseBatchHnswOverDefaultDelegate

Gradle 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: upstream-delta-columnar-dense-vector-batch

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.