Use exact search in VectorSimilarityQuery when traversalSimilarity is -infinity (#15836)

Author: sgup432

lucene/CHANGES.txt

@@ -282,6 +282,9 @@ Bug Fixes
   native code at all (wrong platform); disable native code on 32 bit platforms.
   (Uwe Schindler, thanks to Christian Ortlepp for finding the issue)
 
+* GITHUB#15836: Use exact search in VectorSimilarityQuery when traversalSimilarity is -infinity to
+  guarantee completeness. (Sagar Upadhyaya)
+
 * GITHUB#12419, GITHUB#15119, GITHUB#15864: Fix circular dependency deadlock in
   TestSecrets initialization (Namgyu Kim, Uwe Schindler)
 

lucene/core/src/java/org/apache/lucene/search/AbstractVectorSimilarityQuery.java

@@ -122,7 +122,15 @@ public ScorerSupplier scorerSupplier(LeafReaderContext context) throws IOExcepti
 
         // If there is no filter
         if (filterWeight == null) {
-          // Return exhaustive results
+          if (traversalSimilarity == Float.NEGATIVE_INFINITY) {
+            // When traversalSimilarity is -∞, the intent is to find all vectors above
+            // resultSimilarity. The approximate graph search may miss nodes,
+            // so use exact search to guarantee completeness.
+            AcceptDocs acceptDocs = AcceptDocs.fromLiveDocs(liveDocs, leafReader.maxDoc());
+            return VectorSimilarityScorerSupplier.fromAcceptDocs(
+                boost, createVectorScorer(context), acceptDocs.iterator(), resultSimilarity);
+          }
+          // Return results via approximate graph search
           TopDocs results =
               approximateSearch(
                   context,
@@ -150,6 +158,13 @@ public ScorerSupplier scorerSupplier(LeafReaderContext context) throws IOExcepti
             return null;
           }
 
+          if (traversalSimilarity == Float.NEGATIVE_INFINITY) {
+            // When traversalSimilarity is -∞, skip approximate search and go straight
+            // to exact search over the filtered docs.
+            return VectorSimilarityScorerSupplier.fromAcceptDocs(
+                boost, createVectorScorer(context), acceptDocs.iterator(), resultSimilarity);
+          }
+
           // Perform an approximate search
           TopDocs results =
               approximateSearch(context, acceptDocs, cardinality, timeLimitingKnnCollectorManager);

lucene/core/src/test/org/apache/lucene/search/BaseVectorSimilarityQueryTestCase.java

@@ -456,6 +456,14 @@ public void testFallbackToExact() throws IOException {
 
       // Falls back to exact search
       expectThrows(UnsupportedOperationException.class, () -> searcher.count(query));
+
+      // When traversalSimilarity is -∞ without a filter, exact search should be used directly
+      Query exactQuery =
+          getThrowingVectorQuery(
+              vectorField, queryVector, Float.NEGATIVE_INFINITY, Float.NEGATIVE_INFINITY, null);
+
+      // Falls back to exact search because traversalSimilarity is -∞
+      expectThrows(UnsupportedOperationException.class, () -> searcher.count(exactQuery));
     }
   }
 
@@ -510,15 +518,13 @@ public void testTimeout() throws IOException {
       // This query is cacheable, explicitly prevent it
       searcher.setQueryCache(null);
 
+      // Use Math.nextUp(Float.NEGATIVE_INFINITY) to ensure approximate graph search is used,
+      // since Float.NEGATIVE_INFINITY now triggers exact search which bypasses graph traversal
+      float nearNegInf = Math.nextUp(Float.NEGATIVE_INFINITY);
+
       Query query =
-          new CountingQuery(
-              getVectorQuery(
-                  vectorField,
-                  queryVector,
-                  Float.NEGATIVE_INFINITY,
-                  Float.NEGATIVE_INFINITY,
-                  null));
-      assertEquals(numDocs, searcher.count(query)); // Expect some results without timeout
+          new CountingQuery(getVectorQuery(vectorField, queryVector, nearNegInf, nearNegInf, null));
+      assertEquals(numDocs, searcher.count(query)); // Expect all results without timeout
 
       searcher.setTimeout(() -> true); // Immediately timeout
       assertEquals(0, searcher.count(query)); // Expect no results with the timeout
@@ -534,12 +540,7 @@ public void testTimeout() throws IOException {
       Query filter = IntField.newSetQuery(idField, getFiltered(numFiltered));
       Query filteredQuery =
           new CountingQuery(
-              getVectorQuery(
-                  vectorField,
-                  queryVector,
-                  Float.NEGATIVE_INFINITY,
-                  Float.NEGATIVE_INFINITY,
-                  filter));
+              getVectorQuery(vectorField, queryVector, nearNegInf, nearNegInf, filter));
 
       searcher.setTimeout(() -> false); // Set a timeout which is never met
       assertEquals(numFiltered, searcher.count(filteredQuery));