Rename DiskBBQ format to a versioned name

qa/vector/src/main/java/org/elasticsearch/test/knn/KnnIndexTester.java

@@ -30,7 +30,7 @@
 import org.elasticsearch.core.PathUtils;
 import org.elasticsearch.index.codec.vectors.ES813Int8FlatVectorFormat;
 import org.elasticsearch.index.codec.vectors.ES814HnswScalarQuantizedVectorsFormat;
-import org.elasticsearch.index.codec.vectors.IVFVectorsFormat;
+import org.elasticsearch.index.codec.vectors.diskbbq.ES920DiskBBQVectorsFormat;
 import org.elasticsearch.index.codec.vectors.es818.ES818BinaryQuantizedVectorsFormat;
 import org.elasticsearch.index.codec.vectors.es818.ES818HnswBinaryQuantizedVectorsFormat;
 import org.elasticsearch.logging.Level;
@@ -106,7 +106,7 @@ private static String formatIndexPath(CmdLineArgs args) {
     static Codec createCodec(CmdLineArgs args) {
         final KnnVectorsFormat format;
         if (args.indexType() == IndexType.IVF) {
-            format = new IVFVectorsFormat(args.ivfClusterSize(), IVFVectorsFormat.DEFAULT_CENTROIDS_PER_PARENT_CLUSTER);
+            format = new ES920DiskBBQVectorsFormat(args.ivfClusterSize(), ES920DiskBBQVectorsFormat.DEFAULT_CENTROIDS_PER_PARENT_CLUSTER);
         } else {
             if (args.quantizeBits() == 1) {
                 if (args.indexType() == IndexType.FLAT) {

server/src/main/java/module-info.java

@@ -7,6 +7,7 @@
  * License v3.0 only", or the "Server Side Public License, v 1".
  */
 
+import org.elasticsearch.index.codec.vectors.diskbbq.ES920DiskBBQVectorsFormat;
 import org.elasticsearch.plugins.internal.RestExtension;
 import org.elasticsearch.reservedstate.ReservedStateHandlerProvider;
 
@@ -461,7 +462,7 @@
             org.elasticsearch.index.codec.vectors.es816.ES816HnswBinaryQuantizedVectorsFormat,
             org.elasticsearch.index.codec.vectors.es818.ES818BinaryQuantizedVectorsFormat,
             org.elasticsearch.index.codec.vectors.es818.ES818HnswBinaryQuantizedVectorsFormat,
-            org.elasticsearch.index.codec.vectors.IVFVectorsFormat;
+            ES920DiskBBQVectorsFormat;
 
     provides org.apache.lucene.codecs.Codec
         with
@@ -487,4 +488,6 @@
     exports org.elasticsearch.index.codec.vectors to org.elasticsearch.test.knn;
     exports org.elasticsearch.index.codec.vectors.es818 to org.elasticsearch.test.knn;
     exports org.elasticsearch.inference.telemetry;
+    exports org.elasticsearch.index.codec.vectors.diskbbq to org.elasticsearch.test.knn;
+    exports org.elasticsearch.index.codec.vectors.cluster to org.elasticsearch.test.knn;
 }

server/src/main/java/org/elasticsearch/index/codec/vectors/cluster/KMeansLocal.java

@@ -13,7 +13,6 @@
 import org.apache.lucene.util.FixedBitSet;
 import org.apache.lucene.util.VectorUtil;
 import org.apache.lucene.util.hnsw.IntToIntFunction;
-import org.elasticsearch.index.codec.vectors.SampleReader;
 import org.elasticsearch.simdvec.ESVectorUtil;
 
 import java.io.IOException;

server/src/main/java/org/elasticsearch/index/codec/vectors/SampleReader.java → server/src/main/java/org/elasticsearch/index/codec/vectors/cluster/SampleReader.java

@@ -17,7 +17,7 @@
  *
  * Modifications copyright (C) 2025 Elasticsearch B.V.
  */
-package org.elasticsearch.index.codec.vectors;
+package org.elasticsearch.index.codec.vectors.cluster;
 
 import org.apache.lucene.codecs.lucene95.HasIndexSlice;
 import org.apache.lucene.index.FloatVectorValues;
@@ -29,7 +29,7 @@
 import java.util.Random;
 import java.util.function.IntUnaryOperator;
 
-public class SampleReader extends FloatVectorValues implements HasIndexSlice {
+class SampleReader extends FloatVectorValues implements HasIndexSlice {
     private final FloatVectorValues origin;
     private final int sampleSize;
     private final IntUnaryOperator sampleFunction;
@@ -81,7 +81,7 @@ public Bits getAcceptOrds(Bits acceptDocs) {
         throw new IllegalStateException("Not supported");
     }
 
-    public static SampleReader createSampleReader(FloatVectorValues origin, int k, long seed) {
+    static SampleReader createSampleReader(FloatVectorValues origin, int k, long seed) {
         // TODO can we do something algorithmically that aligns an ordinal with a unique integer between 0 and numVectors?
         if (k >= origin.size()) {
             new SampleReader(origin, origin.size(), i -> i);
@@ -101,7 +101,7 @@ public static SampleReader createSampleReader(FloatVectorValues origin, int k, l
      * @param seed random seed
      * @return array of k samples
      */
-    public static int[] reservoirSample(int n, int k, long seed) {
+    static int[] reservoirSample(int n, int k, long seed) {
         Random rnd = new Random(seed);
         int[] reservoir = new int[k];
         for (int i = 0; i < k; i++) {

server/src/main/java/org/elasticsearch/index/codec/vectors/CentroidAssignments.java → server/src/main/java/org/elasticsearch/index/codec/vectors/diskbbq/CentroidAssignments.java

@@ -7,7 +7,7 @@
  * License v3.0 only", or the "Server Side Public License, v 1".
  */
 
-package org.elasticsearch.index.codec.vectors;
+package org.elasticsearch.index.codec.vectors.diskbbq;
 
 record CentroidAssignments(int numCentroids, float[][] centroids, int[] assignments, int[] overspillAssignments) {
 

server/src/main/java/org/elasticsearch/index/codec/vectors/DiskBBQBulkWriter.java → server/src/main/java/org/elasticsearch/index/codec/vectors/diskbbq/DiskBBQBulkWriter.java

@@ -7,10 +7,11 @@
  * License v3.0 only", or the "Server Side Public License, v 1".
  */
 
-package org.elasticsearch.index.codec.vectors;
+package org.elasticsearch.index.codec.vectors.diskbbq;
 
 import org.apache.lucene.search.CheckedIntConsumer;
 import org.apache.lucene.store.IndexOutput;
+import org.elasticsearch.index.codec.vectors.OptimizedScalarQuantizer;
 
 import java.io.IOException;
 
@@ -28,7 +29,7 @@ protected DiskBBQBulkWriter(int bulkSize, IndexOutput out) {
         this.out = out;
     }
 
-    abstract void writeVectors(DefaultIVFVectorsWriter.QuantizedVectorValues qvv, CheckedIntConsumer<IOException> docsWriter)
+    abstract void writeVectors(ES920DiskBBQVectorsWriter.QuantizedVectorValues qvv, CheckedIntConsumer<IOException> docsWriter)
         throws IOException;
 
     static class OneBitDiskBBQBulkWriter extends DiskBBQBulkWriter {
@@ -40,7 +41,7 @@ static class OneBitDiskBBQBulkWriter extends DiskBBQBulkWriter {
         }
 
         @Override
-        void writeVectors(DefaultIVFVectorsWriter.QuantizedVectorValues qvv, CheckedIntConsumer<IOException> docsWriter)
+        void writeVectors(ES920DiskBBQVectorsWriter.QuantizedVectorValues qvv, CheckedIntConsumer<IOException> docsWriter)
             throws IOException {
             int limit = qvv.count() - bulkSize + 1;
             int i = 0;
@@ -103,7 +104,7 @@ static class SevenBitDiskBBQBulkWriter extends DiskBBQBulkWriter {
         }
 
         @Override
-        void writeVectors(DefaultIVFVectorsWriter.QuantizedVectorValues qvv, CheckedIntConsumer<IOException> docsWriter)
+        void writeVectors(ES920DiskBBQVectorsWriter.QuantizedVectorValues qvv, CheckedIntConsumer<IOException> docsWriter)
             throws IOException {
             int limit = qvv.count() - bulkSize + 1;
             int i = 0;

server/src/main/java/org/elasticsearch/index/codec/vectors/DocIdsWriter.java → server/src/main/java/org/elasticsearch/index/codec/vectors/diskbbq/DocIdsWriter.java

@@ -16,7 +16,7 @@
  * limitations under the License.
  * Modifications copyright (C) 2025 Elasticsearch B.V.
  */
-package org.elasticsearch.index.codec.vectors;
+package org.elasticsearch.index.codec.vectors.diskbbq;
 
 import org.apache.lucene.index.PointValues.IntersectVisitor;
 import org.apache.lucene.store.DataOutput;

server/src/main/java/org/elasticsearch/index/codec/vectors/IVFVectorsFormat.java → server/src/main/java/org/elasticsearch/index/codec/vectors/diskbbq/ES920DiskBBQVectorsFormat.java

@@ -7,7 +7,7 @@
  * License v3.0 only", or the "Server Side Public License, v 1".
  */
 
-package org.elasticsearch.index.codec.vectors;
+package org.elasticsearch.index.codec.vectors.diskbbq;
 
 import org.apache.lucene.codecs.KnnVectorsFormat;
 import org.apache.lucene.codecs.KnnVectorsReader;
@@ -17,6 +17,7 @@
 import org.apache.lucene.codecs.lucene99.Lucene99FlatVectorsFormat;
 import org.apache.lucene.index.SegmentReadState;
 import org.apache.lucene.index.SegmentWriteState;
+import org.elasticsearch.index.codec.vectors.OptimizedScalarQuantizer;
 
 import java.io.IOException;
 
@@ -42,9 +43,9 @@
  * <p> Stores metadata including the number of centroids and their offsets in the clivf file</p>
  *
  */
-public class IVFVectorsFormat extends KnnVectorsFormat {
+public class ES920DiskBBQVectorsFormat extends KnnVectorsFormat {
 
-    public static final String NAME = "IVFVectorsFormat";
+    public static final String NAME = "ES920DiskBBQVectorsFormat";
     // centroid ordinals -> centroid values, offsets
     public static final String CENTROID_EXTENSION = "cenivf";
     // offsets contained in cen_ivf, [vector ordinals, actually just docIds](long varint), quantized
@@ -72,7 +73,7 @@ public class IVFVectorsFormat extends KnnVectorsFormat {
     private final int vectorPerCluster;
     private final int centroidsPerParentCluster;
 
-    public IVFVectorsFormat(int vectorPerCluster, int centroidsPerParentCluster) {
+    public ES920DiskBBQVectorsFormat(int vectorPerCluster, int centroidsPerParentCluster) {
         super(NAME);
         if (vectorPerCluster < MIN_VECTORS_PER_CLUSTER || vectorPerCluster > MAX_VECTORS_PER_CLUSTER) {
             throw new IllegalArgumentException(
@@ -99,18 +100,18 @@ public IVFVectorsFormat(int vectorPerCluster, int centroidsPerParentCluster) {
     }
 
     /** Constructs a format using the given graph construction parameters and scalar quantization. */
-    public IVFVectorsFormat() {
+    public ES920DiskBBQVectorsFormat() {
         this(DEFAULT_VECTORS_PER_CLUSTER, DEFAULT_CENTROIDS_PER_PARENT_CLUSTER);
     }
 
     @Override
     public KnnVectorsWriter fieldsWriter(SegmentWriteState state) throws IOException {
-        return new DefaultIVFVectorsWriter(state, rawVectorFormat.fieldsWriter(state), vectorPerCluster, centroidsPerParentCluster);
+        return new ES920DiskBBQVectorsWriter(state, rawVectorFormat.fieldsWriter(state), vectorPerCluster, centroidsPerParentCluster);
     }
 
     @Override
     public KnnVectorsReader fieldsReader(SegmentReadState state) throws IOException {
-        return new DefaultIVFVectorsReader(state, rawVectorFormat.fieldsReader(state));
+        return new ES920DiskBBQVectorsReader(state, rawVectorFormat.fieldsReader(state));
     }
 
     @Override
@@ -120,7 +121,7 @@ public int getMaxDimensions(String fieldName) {
 
     @Override
     public String toString() {
-        return "IVFVectorsFormat(" + "vectorPerCluster=" + vectorPerCluster + ')';
+        return "ES920DiskBBQVectorsFormat(" + "vectorPerCluster=" + vectorPerCluster + ')';
     }
 
 }

server/src/main/java/org/elasticsearch/index/codec/vectors/DefaultIVFVectorsReader.java → server/src/main/java/org/elasticsearch/index/codec/vectors/diskbbq/ES920DiskBBQVectorsReader.java

@@ -7,7 +7,7 @@
  * License v3.0 only", or the "Server Side Public License, v 1".
  */
 
-package org.elasticsearch.index.codec.vectors;
+package org.elasticsearch.index.codec.vectors.diskbbq;
 
 import org.apache.lucene.codecs.hnsw.FlatVectorsReader;
 import org.apache.lucene.index.FieldInfo;
@@ -18,6 +18,7 @@
 import org.apache.lucene.util.ArrayUtil;
 import org.apache.lucene.util.Bits;
 import org.apache.lucene.util.VectorUtil;
+import org.elasticsearch.index.codec.vectors.OptimizedScalarQuantizer;
 import org.elasticsearch.index.codec.vectors.cluster.NeighborQueue;
 import org.elasticsearch.index.codec.vectors.reflect.OffHeapStats;
 import org.elasticsearch.simdvec.ES91OSQVectorsScorer;
@@ -38,9 +39,9 @@
  * Default implementation of {@link IVFVectorsReader}. It scores the posting lists centroids using
  * brute force and then scores the top ones using the posting list.
  */
-public class DefaultIVFVectorsReader extends IVFVectorsReader implements OffHeapStats {
+public class ES920DiskBBQVectorsReader extends IVFVectorsReader implements OffHeapStats {
 
-    public DefaultIVFVectorsReader(SegmentReadState state, FlatVectorsReader rawVectorsReader) throws IOException {
+    public ES920DiskBBQVectorsReader(SegmentReadState state, FlatVectorsReader rawVectorsReader) throws IOException {
         super(state, rawVectorsReader);
     }
 

server/src/main/java/org/elasticsearch/index/codec/vectors/DefaultIVFVectorsWriter.java → server/src/main/java/org/elasticsearch/index/codec/vectors/diskbbq/ES920DiskBBQVectorsWriter.java

@@ -7,7 +7,7 @@
  * License v3.0 only", or the "Server Side Public License, v 1".
  */
 
-package org.elasticsearch.index.codec.vectors;
+package org.elasticsearch.index.codec.vectors.diskbbq;
 
 import org.apache.lucene.codecs.hnsw.FlatVectorsWriter;
 import org.apache.lucene.index.FieldInfo;
@@ -23,6 +23,8 @@
 import org.apache.lucene.util.packed.PackedInts;
 import org.apache.lucene.util.packed.PackedLongValues;
 import org.elasticsearch.core.SuppressForbidden;
+import org.elasticsearch.index.codec.vectors.BQVectorUtils;
+import org.elasticsearch.index.codec.vectors.OptimizedScalarQuantizer;
 import org.elasticsearch.index.codec.vectors.cluster.HierarchicalKMeans;
 import org.elasticsearch.index.codec.vectors.cluster.KMeansResult;
 import org.elasticsearch.logging.LogManager;
@@ -42,13 +44,13 @@
  * partition the vector space, and then stores the centroids and posting list in a sequential
  * fashion.
  */
-public class DefaultIVFVectorsWriter extends IVFVectorsWriter {
-    private static final Logger logger = LogManager.getLogger(DefaultIVFVectorsWriter.class);
+public class ES920DiskBBQVectorsWriter extends IVFVectorsWriter {
+    private static final Logger logger = LogManager.getLogger(ES920DiskBBQVectorsWriter.class);
 
     private final int vectorPerCluster;
     private final int centroidsPerParentCluster;
 
-    public DefaultIVFVectorsWriter(
+    public ES920DiskBBQVectorsWriter(
         SegmentWriteState state,
         FlatVectorsWriter rawVectorDelegate,
         int vectorPerCluster,

server/src/main/java/org/elasticsearch/index/codec/vectors/IVFVectorsReader.java → server/src/main/java/org/elasticsearch/index/codec/vectors/diskbbq/IVFVectorsReader.java

@@ -7,7 +7,7 @@
  * License v3.0 only", or the "Server Side Public License, v 1".
  */
 
-package org.elasticsearch.index.codec.vectors;
+package org.elasticsearch.index.codec.vectors.diskbbq;
 
 import org.apache.lucene.codecs.CodecUtil;
 import org.apache.lucene.codecs.KnnVectorsReader;
@@ -35,7 +35,7 @@
 import java.io.IOException;
 
 import static org.apache.lucene.codecs.lucene99.Lucene99HnswVectorsReader.SIMILARITY_FUNCTIONS;
-import static org.elasticsearch.index.codec.vectors.IVFVectorsFormat.DYNAMIC_VISIT_RATIO;
+import static org.elasticsearch.index.codec.vectors.diskbbq.ES920DiskBBQVectorsFormat.DYNAMIC_VISIT_RATIO;
 
 /**
  * Reader for IVF vectors. This reader is used to read the IVF vectors from the index.
@@ -54,7 +54,11 @@ protected IVFVectorsReader(SegmentReadState state, FlatVectorsReader rawVectorsR
         this.fieldInfos = state.fieldInfos;
         this.rawVectorsReader = rawVectorsReader;
         this.fields = new IntObjectHashMap<>();
-        String meta = IndexFileNames.segmentFileName(state.segmentInfo.name, state.segmentSuffix, IVFVectorsFormat.IVF_META_EXTENSION);
+        String meta = IndexFileNames.segmentFileName(
+            state.segmentInfo.name,
+            state.segmentSuffix,
+            ES920DiskBBQVectorsFormat.IVF_META_EXTENSION
+        );
 
         int versionMeta = -1;
         boolean success = false;
@@ -63,9 +67,9 @@ protected IVFVectorsReader(SegmentReadState state, FlatVectorsReader rawVectorsR
             try {
                 versionMeta = CodecUtil.checkIndexHeader(
                     ivfMeta,
-                    IVFVectorsFormat.NAME,
-                    IVFVectorsFormat.VERSION_START,
-                    IVFVectorsFormat.VERSION_CURRENT,
+                    ES920DiskBBQVectorsFormat.NAME,
+                    ES920DiskBBQVectorsFormat.VERSION_START,
+                    ES920DiskBBQVectorsFormat.VERSION_CURRENT,
                     state.segmentInfo.getId(),
                     state.segmentSuffix
                 );
@@ -75,8 +79,20 @@ protected IVFVectorsReader(SegmentReadState state, FlatVectorsReader rawVectorsR
             } finally {
                 CodecUtil.checkFooter(ivfMeta, priorE);
             }
-            ivfCentroids = openDataInput(state, versionMeta, IVFVectorsFormat.CENTROID_EXTENSION, IVFVectorsFormat.NAME, state.context);
-            ivfClusters = openDataInput(state, versionMeta, IVFVectorsFormat.CLUSTER_EXTENSION, IVFVectorsFormat.NAME, state.context);
+            ivfCentroids = openDataInput(
+                state,
+                versionMeta,
+                ES920DiskBBQVectorsFormat.CENTROID_EXTENSION,
+                ES920DiskBBQVectorsFormat.NAME,
+                state.context
+            );
+            ivfClusters = openDataInput(
+                state,
+                versionMeta,
+                ES920DiskBBQVectorsFormat.CLUSTER_EXTENSION,
+                ES920DiskBBQVectorsFormat.NAME,
+                state.context
+            );
             success = true;
         } finally {
             if (success == false) {
@@ -108,8 +124,8 @@ private static IndexInput openDataInput(
             final int versionVectorData = CodecUtil.checkIndexHeader(
                 in,
                 codecName,
-                IVFVectorsFormat.VERSION_START,
-                IVFVectorsFormat.VERSION_CURRENT,
+                ES920DiskBBQVectorsFormat.VERSION_START,
+                ES920DiskBBQVectorsFormat.VERSION_CURRENT,
                 state.segmentInfo.getId(),
                 state.segmentSuffix
             );