Add support for nested queries for ivf indices (#128782)

This does a first pass at adding nested query support for bbq_ivf
indices. 

The support is pretty simple right now, basically, we keep exploring
until we at least get `k` results to cover the case when the nested docs
are all tightly clustered and the typical `nprobe` explores too few
clusters to actually get `k` docs.

I have some weird test failures I need to debug, so opening as draft for
now.

Author: benwtrent

rest-api-spec/src/yamlRestTest/resources/rest-api-spec/test/search.vectors/135_knn_query_nested_search_ivf.yml

@@ -0,0 +1,193 @@
+setup:
+  - requires:
+      cluster_features: "mapper.ivf_nested_support"
+      reason: 'ivf nested support required'
+  - do:
+      indices.create:
+        index: test
+        body:
+          settings:
+            index:
+              number_of_shards: 1
+          mappings:
+            properties:
+              name:
+                type: keyword
+              nested:
+                type: nested
+                properties:
+                  paragraph_id:
+                    type: keyword
+                  vector:
+                    type: dense_vector
+                    dims: 5
+                    index: true
+                    similarity: l2_norm
+                    index_options:
+                      type: bbq_ivf
+
+          aliases:
+            my_alias:
+              filter:
+                term:
+                  name: "rabbit.jpg"
+
+  - do:
+      index:
+        index: test
+        id: "1"
+        body:
+          name: cow.jpg
+          nested:
+          - paragraph_id: 0
+            vector: [230, 300.33, -34.8988, 15.555, -200]
+          - paragraph_id: 1
+            vector: [240, 300, -3, 1, -20]
+
+  - do:
+      index:
+        index: test
+        id: "2"
+        body:
+          name: moose.jpg
+          nested:
+          - paragraph_id: 0
+            vector: [-0.5, 100, -13, 14.8, -156]
+          - paragraph_id: 2
+            vector: [0, 100, 0, 14.8, -156]
+
+  - do:
+      index:
+        index: test
+        id: "3"
+        body:
+          name: rabbit.jpg
+          nested:
+            - paragraph_id: 0
+              vector: [0.5, 111.3, -13, 14.8, -156]
+  - do:
+      indices.forcemerge:
+        index: test
+        max_num_segments: 1
+
+  - do:
+      indices.refresh: {}
+
+---
+"nested kNN search that returns diverse parents docs":
+  - do:
+      search:
+        index: test
+        body:
+          fields: [ "name" ]
+          query:
+            nested:
+              path: nested
+              query:
+                knn:
+                  field: nested.vector
+                  query_vector: [-0.5, 90, -10, 14.8, -156]
+                  num_candidates: 3
+  - match: {hits.total.value: 3}
+
+  - match: {hits.hits.0._id: "2"}
+  - match: {hits.hits.0.fields.name.0: "moose.jpg"}
+
+  - match: {hits.hits.1._id: "3"}
+  - match: {hits.hits.1.fields.name.0: "rabbit.jpg"}
+
+  - do:
+      search:
+        index: test
+        body:
+          fields: [ "name" ]
+          query:
+            nested:
+              path: nested
+              query:
+                knn:
+                  field: nested.vector
+                  query_vector: [ -0.5, 90, -10, 14.8, -156 ]
+                  num_candidates: 3
+              inner_hits: { size: 1, "fields": [ "nested.paragraph_id" ], _source: false }
+
+  - match: {hits.total.value: 3}
+
+  - match: { hits.hits.0._id: "2" }
+  - match: { hits.hits.0.fields.name.0: "moose.jpg" }
+  - match: { hits.hits.0.inner_hits.nested.hits.hits.0.fields.nested.0.paragraph_id.0: "0" }
+
+  - match: { hits.hits.1._id: "3" }
+  - match: { hits.hits.1.fields.name.0: "rabbit.jpg" }
+  - match: { hits.hits.1.inner_hits.nested.hits.hits.0.fields.nested.0.paragraph_id.0: "0" }
+
+  - match: { hits.hits.2._id: "1" }
+  - match: { hits.hits.2.fields.name.0: "cow.jpg" }
+  - match: { hits.hits.2.inner_hits.nested.hits.hits.0.fields.nested.0.paragraph_id.0: "0" }
+
+---
+"nested kNN search pre-filtered on alias with filter on top level fields":
+  - do:
+      search:
+        index: my_alias # filter on name: "rabbit.jpg"
+        body:
+          fields: [ "name" ]
+          query:
+            nested:
+              path: nested
+              query:
+                knn:
+                  field: nested.vector
+                  query_vector: [ -0.5, 90.0, -10, 14.8, -156.0 ]
+                  num_candidates: 1
+              inner_hits: { size: 1, "fields": [ "nested.paragraph_id" ], _source: false }
+
+  - match: {hits.total.value: 1}  # as alias is passed as pre-filter, we get a single result
+  - match: {hits.hits.0._id: "3"}
+  - match: {hits.hits.0.fields.name.0: "rabbit.jpg"}
+  - match: { hits.hits.0.inner_hits.nested.hits.hits.0.fields.nested.0.paragraph_id.0: "0" }
+
+---
+"nested kNN search post-filtered on top level fields":
+  - do:
+      search:
+        index: test
+        body:
+          fields: [ "name" ]
+          query:
+            bool:
+              must:
+                - term:
+                    name: "rabbit.jpg"
+                - nested:
+                    path: nested
+                    query:
+                      knn:
+                        field: nested.vector
+                        query_vector: [ -0.5, 90.0, -10, 14.8, -156.0 ]
+                        num_candidates: 1
+  - match: { hits.total.value: 0 } # no hits because returned single vector did not pass post-filter
+
+  - do:
+      search:
+        index: test
+        body:
+          fields: [ "name" ]
+          query:
+            bool:
+              must:
+                - term:
+                    name: "rabbit.jpg"
+                - nested:
+                    path: nested
+                    query:
+                      knn:
+                        field: nested.vector
+                        query_vector: [ -0.5, 90.0, -10, 14.8, -156.0 ]
+                        num_candidates: 3
+                    inner_hits: { size: 1, fields: [ "nested.paragraph_id" ], _source: false }
+
+  - match: {hits.total.value: 1}
+  - match: {hits.hits.0._id: "3"}
+  - match: {hits.hits.0.fields.name.0: "rabbit.jpg"}
+  - match: { hits.hits.0.inner_hits.nested.hits.hits.0.fields.nested.0.paragraph_id.0: "0" }

rest-api-spec/src/yamlRestTest/resources/rest-api-spec/test/search.vectors/46_knn_search_bbq_ivf.yml

@@ -198,47 +198,6 @@ setup:
                 index_options:
                   type: bbq_ivf
 ---
-"Test few dimensions fail indexing":
-  - do:
-      catch: bad_request
-      indices.create:
-        index: bad_bbq_ivf
-        body:
-          mappings:
-            properties:
-              vector:
-                type: dense_vector
-                dims: 42
-                index: true
-                index_options:
-                  type: bbq_ivf
-
-  - do:
-      indices.create:
-        index: dynamic_dim_bbq_ivf
-        body:
-          mappings:
-            properties:
-              vector:
-                type: dense_vector
-                index: true
-                similarity: l2_norm
-                index_options:
-                  type: bbq_ivf
-
-  - do:
-      catch: bad_request
-      index:
-        index: dynamic_dim_bbq_ivf
-        body:
-          vector: [1.0, 2.0, 3.0, 4.0, 5.0]
-
-  - do:
-      index:
-        index: dynamic_dim_bbq_ivf
-        body:
-          vector: [1.0, 2.0, 3.0, 4.0, 1.0, 2.0, 3.0, 4.0, 1.0, 2.0, 3.0, 4.0, 1.0, 2.0, 3.0, 4.0, 1.0, 2.0, 3.0, 4.0, 1.0, 2.0, 3.0, 4.0, 1.0, 2.0, 3.0, 4.0, 1.0, 2.0, 3.0, 4.0, 1.0, 2.0, 3.0, 4.0, 1.0, 2.0, 3.0, 4.0, 1.0, 2.0, 3.0, 4.0, 1.0, 2.0, 3.0, 4.0, 1.0, 2.0, 3.0, 4.0, 1.0, 2.0, 3.0, 4.0, 1.0, 2.0, 3.0, 4.0, 1.0, 2.0, 3.0, 4.0]
----
 "Test index configured rescore vector":
   - skip:
       features: "headers"

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

@@ -22,6 +22,7 @@
 import org.apache.lucene.index.VectorEncoding;
 import org.apache.lucene.index.VectorSimilarityFunction;
 import org.apache.lucene.internal.hppc.IntObjectHashMap;
+import org.apache.lucene.search.AbstractKnnCollector;
 import org.apache.lucene.search.KnnCollector;
 import org.apache.lucene.store.ChecksumIndexInput;
 import org.apache.lucene.store.DataInput;
@@ -252,7 +253,10 @@ public final void search(String field, float[] target, KnnCollector knnCollector
             }
             return visitedDocs.getAndSet(docId) == false;
         };
+        assert knnCollector instanceof AbstractKnnCollector;
+        AbstractKnnCollector knnCollectorImpl = (AbstractKnnCollector) knnCollector;
         int nProbe = DYNAMIC_NPROBE;
+        // Search strategy may be null if this is being called from checkIndex (e.g. from a test)
         if (knnCollector.getSearchStrategy() instanceof IVFKnnSearchStrategy ivfSearchStrategy) {
             nProbe = ivfSearchStrategy.getNProbe();
         }
@@ -280,7 +284,10 @@ public final void search(String field, float[] target, KnnCollector knnCollector
         long expectedDocs = 0;
         long actualDocs = 0;
         // initially we visit only the "centroids to search"
-        while (centroidQueue.size() > 0 && centroidsVisited < nProbe && actualDocs < knnCollector.k()) {
+        // Note, numCollected is doing the bare minimum here.
+        // TODO do we need to handle nested doc counts similarly to how we handle
+        // filtering? E.g. keep exploring until we hit an expected number of parent documents vs. child vectors?
+        while (centroidQueue.size() > 0 && (centroidsVisited < nProbe || knnCollectorImpl.numCollected() < knnCollector.k())) {
             ++centroidsVisited;
             // todo do we actually need to know the score???
             int centroidOrdinal = centroidQueue.pop();

server/src/main/java/org/elasticsearch/index/mapper/MapperFeatures.java

@@ -42,6 +42,7 @@ public class MapperFeatures implements FeatureSpecification {
     );
     static final NodeFeature NPE_ON_DIMS_UPDATE_FIX = new NodeFeature("mapper.npe_on_dims_update_fix");
     static final NodeFeature IVF_FORMAT_CLUSTER_FEATURE = new NodeFeature("mapper.ivf_format_cluster_feature");
+    static final NodeFeature IVF_NESTED_SUPPORT = new NodeFeature("mapper.ivf_nested_support");
 
     @Override
     public Set<NodeFeature> getTestFeatures() {
@@ -70,7 +71,8 @@ public Set<NodeFeature> getTestFeatures() {
             NPE_ON_DIMS_UPDATE_FIX,
             RESCORE_ZERO_VECTOR_QUANTIZED_VECTOR_MAPPING,
             USE_DEFAULT_OVERSAMPLE_VALUE_FOR_BBQ,
-            IVF_FORMAT_CLUSTER_FEATURE
+            IVF_FORMAT_CLUSTER_FEATURE,
+            IVF_NESTED_SUPPORT
         );
     }
 }

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

@@ -64,7 +64,6 @@
 import org.elasticsearch.index.mapper.Mapper;
 import org.elasticsearch.index.mapper.MapperBuilderContext;
 import org.elasticsearch.index.mapper.MapperParsingException;
-import org.elasticsearch.index.mapper.MappingLookup;
 import org.elasticsearch.index.mapper.MappingParser;
 import org.elasticsearch.index.mapper.NumberFieldMapper;
 import org.elasticsearch.index.mapper.SimpleMappedFieldType;
@@ -77,6 +76,7 @@
 import org.elasticsearch.search.aggregations.support.CoreValuesSourceType;
 import org.elasticsearch.search.lookup.Source;
 import org.elasticsearch.search.vectors.DenseVectorQuery;
+import org.elasticsearch.search.vectors.DiversifyingChildrenIVFKnnFloatVectorQuery;
 import org.elasticsearch.search.vectors.ESDiversifyingChildrenByteKnnVectorQuery;
 import org.elasticsearch.search.vectors.ESDiversifyingChildrenFloatKnnVectorQuery;
 import org.elasticsearch.search.vectors.ESKnnByteVectorQuery;
@@ -1650,7 +1650,7 @@ public boolean supportsElementType(ElementType elementType) {
 
             @Override
             public boolean supportsDimension(int dims) {
-                return dims >= BBQ_MIN_DIMS;
+                return true;
             }
         };
 
@@ -2521,12 +2521,19 @@ && isNotUnitVector(squaredMagnitude)) {
                 adjustedK = Math.min((int) Math.ceil(k * oversample), OVERSAMPLE_LIMIT);
                 numCands = Math.max(adjustedK, numCands);
             }
-            if (parentFilter != null && indexOptions instanceof BBQIVFIndexOptions) {
-                throw new IllegalArgumentException("IVF index does not support nested queries");
-            }
             Query knnQuery;
             if (indexOptions instanceof BBQIVFIndexOptions bbqIndexOptions) {
-                knnQuery = new IVFKnnFloatVectorQuery(name(), queryVector, adjustedK, numCands, filter, bbqIndexOptions.defaultNProbe);
+                knnQuery = parentFilter != null
+                    ? new DiversifyingChildrenIVFKnnFloatVectorQuery(
+                        name(),
+                        queryVector,
+                        adjustedK,
+                        numCands,
+                        filter,
+                        parentFilter,
+                        bbqIndexOptions.defaultNProbe
+                    )
+                    : new IVFKnnFloatVectorQuery(name(), queryVector, adjustedK, numCands, filter, bbqIndexOptions.defaultNProbe);
             } else {
                 knnQuery = parentFilter != null
                     ? new ESDiversifyingChildrenFloatKnnVectorQuery(
@@ -2769,19 +2776,6 @@ public FieldMapper.Builder getMergeBuilder() {
         return new Builder(leafName(), indexCreatedVersion).init(this);
     }
 
-    @Override
-    public void doValidate(MappingLookup mappers) {
-        if (indexOptions instanceof BBQIVFIndexOptions && mappers.nestedLookup().getNestedParent(fullPath()) != null) {
-            throw new IllegalArgumentException(
-                "["
-                    + CONTENT_TYPE
-                    + "] fields with index type ["
-                    + indexOptions.type
-                    + "] cannot be indexed if they're within [nested] mappings"
-            );
-        }
-    }
-
     private static IndexOptions parseIndexOptions(String fieldName, Object propNode, IndexVersion indexVersion) {
         @SuppressWarnings("unchecked")
         Map<String, ?> indexOptionsMap = (Map<String, ?>) propNode;

server/src/main/java/org/elasticsearch/search/vectors/AbstractIVFKnnVectorQuery.java

@@ -169,7 +169,7 @@ abstract TopDocs approximateSearch(
     ) throws IOException;
 
     protected KnnCollectorManager getKnnCollectorManager(int k, IndexSearcher searcher) {
-        return new IVFCollectorManager(k, nProbe);
+        return new IVFCollectorManager(k);
     }
 
     @Override
@@ -195,16 +195,14 @@ protected boolean match(int doc) {
 
     static class IVFCollectorManager implements KnnCollectorManager {
         private final int k;
-        private final int nprobe;
 
-        IVFCollectorManager(int k, int nprobe) {
+        IVFCollectorManager(int k) {
             this.k = k;
-            this.nprobe = nprobe;
         }
 
         @Override
         public KnnCollector newCollector(int visitedLimit, KnnSearchStrategy searchStrategy, LeafReaderContext context) throws IOException {
-            return new TopKnnCollector(k, visitedLimit, new IVFKnnSearchStrategy(nprobe));
+            return new TopKnnCollector(k, visitedLimit, searchStrategy);
         }
     }
 }