An NRT reader opened during a merge reuses the SegmentReaders the merge opened with a MERGE IOContext, so its searches read through merge-only inputs (direct I/O with DirectIODirectory, on-disk instead of RAM postings with DirectPostingsFormat, sequential read advice with ADVISE_BY_CONTEXT)
Finding LFV-099Area: index / IndexWriter NRT reader pool (readers shared with merges)Priority: lowImpact: minorNovelty: known upstream
AI-assisted finding, confirmed by automated checks, not yet triaged by a human Lucene committer. This site files nothing upstream.
What it shows
You let a merge start on segments that no NRT reader has opened yet (for example segments committed after the last refresh) and open an NRT reader with DirectoryReader.open(writer) while that merge runs. You expect that reader's searches to read the files through inputs opened for searching, as DirectIODirectory ('For all other cases (searching, writing) we delegate to the provided Directory instance') and DirectPostingsFormat ('at read time loads and stores all terms and postings directly in RAM') document. Instead the NRT reader gets the very SegmentReaders the merge opened with the MERGE IOContext, so with DirectIODirectory its searches read through the merge's direct-I/O inputs, with DirectPostingsFormat it searches the on-disk terms instead of the RAM postings, and with ADVISE_BY_CONTEXT its stored fields, vectors and HNSW graph are mapped with SEQUENTIAL advice; the hits stay correct.
Expected vs actual
- Expected: You expect that reader's searches to read the files through inputs opened for searching, as DirectIODirectory ('For all other cases (searching, writing) we delegate to the provided Directory instance') and DirectPostingsFormat ('at read time loads and stores all terms and postings directly in RAM') document.
- Actual: The NRT reader gets the very SegmentReaders the merge opened with the MERGE IOContext, so with DirectIODirectory its searches read through the merge's direct-I/O inputs, with DirectPostingsFormat it searches the on-disk terms instead of the RAM postings, and with ADVISE_BY_CONTEXT its stored fields, vectors and HNSW graph are mapped with SEQUENTIAL advice; the hits stay correct.
Files
Run it
- Tested on Lucene main e1faeba08bde (2026-11-02): still reproduces.
- Tested on Lucene main ea57360912ed (2026-10-24): reproduced.
- 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.index.TestNrtReaderSharesMergeOpenedInputsGradle 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).
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 2026-11-03. AI-assisted; not yet triaged by a human Lucene committer. Feedback: open an issue in lucene-ai-bug-hunt.