Handle NRT FixedBits live docs in the delegation handle

[aasom143]

fillLiveDocsWords cast every non-null liveDocs to LiveDocs. That holds only for readers opened from a commit, where the codec returns DenseLiveDocs or SparseLiveDocs. An NRT reader whose segment has deletes since the last flush gets PendingDeletes' read-only FixedBits view, which is a Bits but not a LiveDocs, so the first match-all query after a delete failed with ClassCastException until the node was restarted.

Match-all path: route every non-LiveDocs shape through Bits#applyMask on an all-alive slice; FixedBits and FixedBitSet implement it word-level. The codec LiveDocs branches are unchanged.

Predicate path: liveIntersection previously led the conjunction only for DenseLiveDocs, so an NRT segment fell back to the raw scorer with a per-doc liveDocs.get regardless of how many docs were deleted. It now also leads for a non-LiveDocs Bits when the leaf's deletion rate is above the codec's own sparse/dense threshold (delCount / maxDoc > 0.01), materialising the live set once per collector with FixedBitSet.copyOf, a word clone for FixedBits. Below the threshold the raw-scorer path is kept, matching what SparseLiveDocs gets.

Tests: buildIndex gains an nrt flag that leaves deletes uncommitted and opens the reader on the writer; NRT cases cover dense (25%), sparse (0.3%) and exactly-1% deletes for both match-all and predicate. A document-level-security style reader wrapper covers live docs replaced by a FixedBitSet and by a SparseFixedBitSet on both paths.

Description

[Describe what this change achieves]

Related Issues

Resolves #[Issue number to be closed when this PR is merged]

Check List

• Functionality includes testing.
• API changes companion pull request created, if applicable.
• Public documentation issue/PR created, if applicable.

By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license.
For more information on following Developer Certificate of Origin and signing off your commits, please check here.

[github-actions]

PR Reviewer Guide 🔍

Here are some key observations to aid the review process:

🧪 PR contains tests

🔒 No security concerns identified

✅ No TODO sections

🔀 No multiple PR themes

⚡ Recommended focus areas for review Potential IndexOutOfBoundsException In fillLiveDocsWords, when liveDocs is a non-LiveDocs Bits, the code computes Math.min(span, liveDocs.length() - minDoc) to bound the set call. If liveDocs.length() < minDoc (e.g. a wrapper whose Bits reports a shorter length than maxDoc), this yields a negative value and live.set(0, negative) throws. Additionally, applyMask's contract typically requires that liveDocs cover [minDoc, minDoc+span); passing a bitset at offset minDoc without verifying liveDocs.length() >= minDoc + span may fail or silently include deleted docs. Consider validating / clamping length and asserting the invariant. if (liveDocs instanceof LiveDocs == false) { // Not a codec LiveDocs: start all-alive and let the Bits clear the deleted docs. FixedBitSet live = new FixedBitSet(span); live.set(0, Math.min(span, liveDocs.length() - minDoc)); liveDocs.applyMask(live, minDoc); MemorySegment.copy(live.getBits(), 0, out, ValueLayout.JAVA_LONG, 0L, wordCount); return; }

[github-actions]

PR Code Suggestions ✨

Explore these optional code suggestions:

Category	Suggestion	Impact
Possible issue	Clamp live-docs slice range to Bits length If minDoc >= liveDocs.length(), liveDocs.length() - minDoc becomes non-positive, which FixedBitSet.set(0, end) rejects with an exception. Also, when minDoc + span extends past liveDocs.length(), docs beyond the Bits length should remain unset (treated as not-live) rather than be set live. Clamp the range and guard against a non-positive end. sandbox/plugins/analytics-backend-lucene/src/main/java/org/opensearch/be/lucene/LuceneFilterDelegationHandle.java [400-407] if (liveDocs instanceof LiveDocs == false) { // Not a codec LiveDocs: start all-alive and let the Bits clear the deleted docs. FixedBitSet live = new FixedBitSet(span); - live.set(0, Math.min(span, liveDocs.length() - minDoc)); - liveDocs.applyMask(live, minDoc); + int end = Math.min(span, Math.max(0, liveDocs.length() - minDoc)); + if (end > 0) { + live.set(0, end); + liveDocs.applyMask(live, minDoc); + } MemorySegment.copy(live.getBits(), 0, out, ValueLayout.JAVA_LONG, 0L, wordCount); return; } Suggestion importance[1-10]: 5 __ Why: The concern about liveDocs.length() - minDoc becoming non-positive is valid in theory, though in practice the caller likely ensures minDoc < maxDoc. The clamping adds defensive safety but may not address a real bug given the invariants.	Low
General	Guard against divide-by-zero on empty leaves maxDoc() could theoretically be 0 for an empty leaf, producing a divide-by-zero NaN comparison. Guard against zero maxDoc to make the threshold check robust. sandbox/plugins/analytics-backend-lucene/src/main/java/org/opensearch/be/lucene/LuceneFilterDelegationHandle.java [272-274] -} else if ((double) leaf.reader().numDeletedDocs() / leaf.reader().maxDoc() > SPARSE_DENSE_THRESHOLD) { - liveDocsBitSet = FixedBitSet.copyOf(liveDocs); +} else { + int maxDoc = leaf.reader().maxDoc(); + if (maxDoc > 0 && (double) leaf.reader().numDeletedDocs() / maxDoc > SPARSE_DENSE_THRESHOLD) { + liveDocsBitSet = FixedBitSet.copyOf(liveDocs); + } } Suggestion importance[1-10]: 2 __ Why: An empty leaf with maxDoc == 0 is extremely unlikely in practice (Lucene doesn't produce empty segments typically), and NaN comparison would just evaluate to false, which is actually the desired behavior. The guard is a very minor defensive improvement.	Low

[github-actions]

✅ Gradle check result for c5ddf62: SUCCESS

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 71.84%. Comparing base (3934026) to head (c5ddf62).
⚠️ Report is 1 commits behind head on main.

Additional details and impacted files

@@             Coverage Diff              @@
##               main   #23215      +/-   ##
============================================
+ Coverage     71.79%   71.84%   +0.05%     
- Complexity    77987    78026      +39     
============================================
  Files          6185     6185              
  Lines        361374   361374              
  Branches      52599    52599              
============================================
+ Hits         259458   259647     +189     
+ Misses        81284    81179     -105     
+ Partials      20632    20548      -84     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.