Disallow dot_product and max_inner_product for int8_hnsw GPU type

docs/changelog/136881.yaml

@@ -0,0 +1,5 @@
+pr: 136881
+summary: Disallow `dot_product` and `max_inner_product` for int8_hnsw GPU type
+area: Search
+type: bug
+issues: []

server/src/main/java/org/elasticsearch/index/mapper/vectors/DenseVectorFieldMapper.java

@@ -2286,6 +2286,10 @@ public DenseVectorFieldType(
             this.isSyntheticSource = isSyntheticSource;
         }
 
+        public VectorSimilarity similarity() {
+            return similarity;
+        }
+
         @Override
         public String typeName() {
             return CONTENT_TYPE;
@@ -2890,7 +2894,7 @@ public KnnVectorsFormat getKnnVectorsFormatForField(KnnVectorsFormat defaultForm
             // if plugins provided alternative KnnVectorsFormat for this indexOptions, use it instead of standard
             KnnVectorsFormat extraKnnFormat = null;
             for (VectorsFormatProvider vectorsFormatProvider : extraVectorsFormatProviders) {
-                extraKnnFormat = vectorsFormatProvider.getKnnVectorsFormat(indexSettings, indexOptions);
+                extraKnnFormat = vectorsFormatProvider.getKnnVectorsFormat(indexSettings, indexOptions, fieldType().similarity());
                 if (extraKnnFormat != null) {
                     break;
                 }

server/src/main/java/org/elasticsearch/index/mapper/vectors/VectorsFormatProvider.java

@@ -24,7 +24,12 @@ public interface VectorsFormatProvider {
      *
      * @param indexSettings The index settings.
      * @param indexOptions The dense vector index options.
+     * @param similarity The vector similarity function.
      * @return A KnnVectorsFormat instance.
      */
-    KnnVectorsFormat getKnnVectorsFormat(IndexSettings indexSettings, DenseVectorFieldMapper.DenseVectorIndexOptions indexOptions);
+    KnnVectorsFormat getKnnVectorsFormat(
+        IndexSettings indexSettings,
+        DenseVectorFieldMapper.DenseVectorIndexOptions indexOptions,
+        DenseVectorFieldMapper.VectorSimilarity similarity
+    );
 }

server/src/test/java/org/elasticsearch/index/mapper/vectors/DenseVectorFieldTypeTests.java

@@ -96,6 +96,15 @@ public static DenseVectorFieldMapper.DenseVectorIndexOptions randomGpuSupportedI
         );
     }
 
+    public static DenseVectorFieldMapper.VectorSimilarity randomGPUSupportedSimilarity(
+        DenseVectorFieldMapper.DenseVectorIndexOptions indexOptions
+    ) {
+        if (indexOptions.getType() == DenseVectorFieldMapper.VectorIndexType.INT8_HNSW) {
+            return randomFrom(VectorSimilarity.L2_NORM, VectorSimilarity.COSINE);
+        }
+        return randomFrom(VectorSimilarity.values());
+    }
+
     public static DenseVectorFieldMapper.DenseVectorIndexOptions randomIndexOptionsAll() {
         List<DenseVectorFieldMapper.DenseVectorIndexOptions> options = new ArrayList<>(
             Arrays.asList(

x-pack/plugin/gpu/src/internalClusterTest/java/org/elasticsearch/plugin/gpu/GPUIndexIT.java

@@ -24,9 +24,11 @@
 import java.util.List;
 import java.util.Locale;
 
+import static org.apache.lucene.tests.index.BaseKnnVectorsFormatTestCase.randomNormalizedVector;
 import static org.elasticsearch.test.hamcrest.ElasticsearchAssertions.assertAcked;
 import static org.elasticsearch.test.hamcrest.ElasticsearchAssertions.assertNoFailures;
 import static org.elasticsearch.test.hamcrest.ElasticsearchAssertions.assertNoFailuresAndResponse;
+import static org.hamcrest.Matchers.containsString;
 
 @LuceneTestCase.SuppressCodecs("*") // use our custom codec
 public class GPUIndexIT extends ESIntegTestCase {
@@ -160,6 +162,44 @@ public void testSearchWithoutGPU() {
         assertSearch(indexName, randomFloatVector(dims), numDocs);
     }
 
+    public void testInt8HnswDotProductFails() {
+        String indexName = "index_int8_dot_product_fails";
+        final int dims = randomIntBetween(4, 128);
+
+        Settings.Builder settingsBuilder = Settings.builder().put(indexSettings());
+        settingsBuilder.put("index.number_of_shards", 1);
+        settingsBuilder.put("index.vectors.indexing.use_gpu", true);
+
+        String mapping = String.format(Locale.ROOT, """
+            {
+              "properties": {
+                "my_vector": {
+                  "type": "dense_vector",
+                  "dims": %d,
+                  "similarity": "dot_product",
+                  "index_options": {
+                    "type": "int8_hnsw"
+                  }
+                }
+              }
+            }
+            """, dims);
+
+        // Index creation should succeed
+        assertAcked(prepareCreate(indexName).setSettings(settingsBuilder.build()).setMapping(mapping));
+        ensureGreen();
+
+        // Attempt to index a document and expect it to fail
+        IllegalArgumentException ex = expectThrows(
+            IllegalArgumentException.class,
+            () -> client().prepareIndex(indexName).setId("1").setSource("my_vector", randomNormalizedVector(dims)).get()
+        );
+        assertThat(
+            ex.getMessage(),
+            containsString("GPU vector indexing does not support [dot_product] similarity for [int8_hnsw] index type.")
+        );
+    }
+
     private void createIndex(String indexName, int dims, boolean sorted) {
         var settings = Settings.builder().put(indexSettings());
         settings.put("index.number_of_shards", 1);

x-pack/plugin/gpu/src/internalClusterTest/java/org/elasticsearch/plugin/gpu/GPUPluginInitializationIT.java

@@ -109,7 +109,7 @@ public void testFFOff() {
         GPUPlugin gpuPlugin = internalCluster().getInstance(GPUPlugin.class);
         VectorsFormatProvider vectorsFormatProvider = gpuPlugin.getVectorsFormatProvider();
 
-        var format = vectorsFormatProvider.getKnnVectorsFormat(null, null);
+        var format = vectorsFormatProvider.getKnnVectorsFormat(null, null, null);
         assertNull(format);
     }
 
@@ -136,7 +136,11 @@ public void testFFOffGPUFormatNull() {
         IndexSettings settings = getIndexSettings();
         final var indexOptions = DenseVectorFieldTypeTests.randomGpuSupportedIndexOptions();
 
-        var format = vectorsFormatProvider.getKnnVectorsFormat(settings, indexOptions);
+        var format = vectorsFormatProvider.getKnnVectorsFormat(
+            settings,
+            indexOptions,
+            DenseVectorFieldTypeTests.randomGPUSupportedSimilarity(indexOptions)
+        );
         assertNull(format);
     }
 
@@ -151,7 +155,11 @@ public void testIndexSettingOnIndexTypeSupportedGPUSupported() {
         IndexSettings settings = getIndexSettings();
         final var indexOptions = DenseVectorFieldTypeTests.randomGpuSupportedIndexOptions();
 
-        var format = vectorsFormatProvider.getKnnVectorsFormat(settings, indexOptions);
+        var format = vectorsFormatProvider.getKnnVectorsFormat(
+            settings,
+            indexOptions,
+            DenseVectorFieldTypeTests.randomGPUSupportedSimilarity(indexOptions)
+        );
         assertNotNull(format);
     }
 
@@ -166,7 +174,14 @@ public void testIndexSettingOnIndexTypeNotSupportedThrows() {
         IndexSettings settings = getIndexSettings();
         final var indexOptions = DenseVectorFieldTypeTests.randomFlatIndexOptions();
 
-        var ex = expectThrows(IllegalArgumentException.class, () -> vectorsFormatProvider.getKnnVectorsFormat(settings, indexOptions));
+        var ex = expectThrows(
+            IllegalArgumentException.class,
+            () -> vectorsFormatProvider.getKnnVectorsFormat(
+                settings,
+                indexOptions,
+                DenseVectorFieldTypeTests.randomGPUSupportedSimilarity(indexOptions)
+            )
+        );
         assertThat(ex.getMessage(), startsWith("[index.vectors.indexing.use_gpu] doesn't support [index_options.type] of"));
     }
 
@@ -181,7 +196,14 @@ public void testIndexSettingOnGPUNotSupportedThrows() {
         IndexSettings settings = getIndexSettings();
         final var indexOptions = DenseVectorFieldTypeTests.randomGpuSupportedIndexOptions();
 
-        var ex = expectThrows(IllegalArgumentException.class, () -> vectorsFormatProvider.getKnnVectorsFormat(settings, indexOptions));
+        var ex = expectThrows(
+            IllegalArgumentException.class,
+            () -> vectorsFormatProvider.getKnnVectorsFormat(
+                settings,
+                indexOptions,
+                DenseVectorFieldTypeTests.randomGPUSupportedSimilarity(indexOptions)
+            )
+        );
         assertThat(
             ex.getMessage(),
             equalTo("[index.vectors.indexing.use_gpu] was set to [true], but GPU resources are not accessible on the node.")
@@ -200,7 +222,14 @@ public void testIndexSettingOnGPUSupportThrowsRethrows() {
         IndexSettings settings = getIndexSettings();
         final var indexOptions = DenseVectorFieldTypeTests.randomGpuSupportedIndexOptions();
 
-        var ex = expectThrows(IllegalArgumentException.class, () -> vectorsFormatProvider.getKnnVectorsFormat(settings, indexOptions));
+        var ex = expectThrows(
+            IllegalArgumentException.class,
+            () -> vectorsFormatProvider.getKnnVectorsFormat(
+                settings,
+                indexOptions,
+                DenseVectorFieldTypeTests.randomGPUSupportedSimilarity(indexOptions)
+            )
+        );
         assertThat(
             ex.getMessage(),
             equalTo("[index.vectors.indexing.use_gpu] was set to [true], but GPU resources are not accessible on the node.")
@@ -218,7 +247,11 @@ public void testIndexSettingAutoIndexTypeSupportedGPUSupported() {
         IndexSettings settings = getIndexSettings();
         final var indexOptions = DenseVectorFieldTypeTests.randomGpuSupportedIndexOptions();
 
-        var format = vectorsFormatProvider.getKnnVectorsFormat(settings, indexOptions);
+        var format = vectorsFormatProvider.getKnnVectorsFormat(
+            settings,
+            indexOptions,
+            DenseVectorFieldTypeTests.randomGPUSupportedSimilarity(indexOptions)
+        );
         assertNotNull(format);
     }
 
@@ -233,7 +266,11 @@ public void testIndexSettingAutoGPUNotSupported() {
         IndexSettings settings = getIndexSettings();
         final var indexOptions = DenseVectorFieldTypeTests.randomGpuSupportedIndexOptions();
 
-        var format = vectorsFormatProvider.getKnnVectorsFormat(settings, indexOptions);
+        var format = vectorsFormatProvider.getKnnVectorsFormat(
+            settings,
+            indexOptions,
+            DenseVectorFieldTypeTests.randomGPUSupportedSimilarity(indexOptions)
+        );
         assertNull(format);
     }
 
@@ -248,7 +285,11 @@ public void testIndexSettingAutoIndexTypeNotSupported() {
         IndexSettings settings = getIndexSettings();
         final var indexOptions = DenseVectorFieldTypeTests.randomFlatIndexOptions();
 
-        var format = vectorsFormatProvider.getKnnVectorsFormat(settings, indexOptions);
+        var format = vectorsFormatProvider.getKnnVectorsFormat(
+            settings,
+            indexOptions,
+            DenseVectorFieldTypeTests.randomGPUSupportedSimilarity(indexOptions)
+        );
         assertNull(format);
     }
 
@@ -263,7 +304,11 @@ public void testIndexSettingOff() {
         IndexSettings settings = getIndexSettings();
         final var indexOptions = DenseVectorFieldTypeTests.randomGpuSupportedIndexOptions();
 
-        var format = vectorsFormatProvider.getKnnVectorsFormat(settings, indexOptions);
+        var format = vectorsFormatProvider.getKnnVectorsFormat(
+            settings,
+            indexOptions,
+            DenseVectorFieldTypeTests.randomGPUSupportedSimilarity(indexOptions)
+        );
         assertNull(format);
     }
 

x-pack/plugin/gpu/src/main/java/org/elasticsearch/xpack/gpu/GPUPlugin.java

@@ -62,7 +62,7 @@ public List<Setting<?>> getSettings() {
 
     @Override
     public VectorsFormatProvider getVectorsFormatProvider() {
-        return (indexSettings, indexOptions) -> {
+        return (indexSettings, indexOptions, similarity) -> {
             if (GPU_FORMAT.isEnabled()) {
                 GpuMode gpuMode = indexSettings.getValue(VECTORS_INDEXING_USE_GPU_SETTING);
                 if (gpuMode == GpuMode.TRUE) {
@@ -76,10 +76,10 @@ public VectorsFormatProvider getVectorsFormatProvider() {
                             "[index.vectors.indexing.use_gpu] was set to [true], but GPU resources are not accessible on the node."
                         );
                     }
-                    return getVectorsFormat(indexOptions);
+                    return getVectorsFormat(indexOptions, similarity);
                 }
                 if (gpuMode == GpuMode.AUTO && vectorIndexTypeSupported(indexOptions.getType()) && isGpuSupported) {
-                    return getVectorsFormat(indexOptions);
+                    return getVectorsFormat(indexOptions, similarity);
                 }
             }
             return null;
@@ -90,7 +90,10 @@ private boolean vectorIndexTypeSupported(DenseVectorFieldMapper.VectorIndexType
         return type == DenseVectorFieldMapper.VectorIndexType.HNSW || type == DenseVectorFieldMapper.VectorIndexType.INT8_HNSW;
     }
 
-    private static KnnVectorsFormat getVectorsFormat(DenseVectorFieldMapper.DenseVectorIndexOptions indexOptions) {
+    private static KnnVectorsFormat getVectorsFormat(
+        DenseVectorFieldMapper.DenseVectorIndexOptions indexOptions,
+        DenseVectorFieldMapper.VectorSimilarity similarity
+    ) {
         if (indexOptions.getType() == DenseVectorFieldMapper.VectorIndexType.HNSW) {
             DenseVectorFieldMapper.HnswIndexOptions hnswIndexOptions = (DenseVectorFieldMapper.HnswIndexOptions) indexOptions;
             int efConstruction = hnswIndexOptions.efConstruction();
@@ -99,6 +102,18 @@ private static KnnVectorsFormat getVectorsFormat(DenseVectorFieldMapper.DenseVec
             }
             return new ES92GpuHnswVectorsFormat(hnswIndexOptions.m(), efConstruction);
         } else if (indexOptions.getType() == DenseVectorFieldMapper.VectorIndexType.INT8_HNSW) {
+            if (similarity == DenseVectorFieldMapper.VectorSimilarity.DOT_PRODUCT
+                || similarity == DenseVectorFieldMapper.VectorSimilarity.MAX_INNER_PRODUCT) {
+                throw new IllegalArgumentException(
+                    "GPU vector indexing does not support ["
+                        + similarity
+                        + "] similarity for [int8_hnsw] index type. "
+                        + "Instead, consider using ["
+                        + DenseVectorFieldMapper.VectorSimilarity.COSINE
+                        + "] or "
+                        + " [hnsw] index type."
+                );
+            }
             DenseVectorFieldMapper.Int8HnswIndexOptions int8HnswIndexOptions = (DenseVectorFieldMapper.Int8HnswIndexOptions) indexOptions;
             int efConstruction = int8HnswIndexOptions.efConstruction();
             if (efConstruction == HnswGraphBuilder.DEFAULT_BEAM_WIDTH) {

x-pack/plugin/gpu/src/test/java/org/elasticsearch/xpack/gpu/codec/GPUDenseVectorFieldMapperTests.java

@@ -63,7 +63,7 @@ private KnnVectorsFormat getKnnVectorsFormat(String indexOptionsType) throws IOE
             b.field("type", "dense_vector");
             b.field("dims", dims);
             b.field("index", true);
-            b.field("similarity", "dot_product");
+            b.field("similarity", "cosine");
             b.startObject("index_options");
             b.field("type", indexOptionsType);
             b.endObject();