HSEARCH-5064 Remove knn integration for Elasticsearch < 8.12

Author: marko-bekhta

backend/elasticsearch/src/main/java/org/hibernate/search/backend/elasticsearch/dialect/model/impl/Elasticsearch812ModelDialect.java

@@ -8,7 +8,7 @@
 
 import org.hibernate.search.backend.elasticsearch.types.dsl.provider.impl.Elasticsearch812IndexFieldTypeFactoryProvider;
 import org.hibernate.search.backend.elasticsearch.types.dsl.provider.impl.ElasticsearchIndexFieldTypeFactoryProvider;
-import org.hibernate.search.backend.elasticsearch.validation.impl.Elasticsearch8PropertyMappingValidatorProvider;
+import org.hibernate.search.backend.elasticsearch.validation.impl.Elasticsearch812PropertyMappingValidatorProvider;
 import org.hibernate.search.backend.elasticsearch.validation.impl.ElasticsearchPropertyMappingValidatorProvider;
 
 import com.google.gson.Gson;
@@ -25,6 +25,6 @@ public ElasticsearchIndexFieldTypeFactoryProvider createIndexTypeFieldFactoryPro
 
 	@Override
 	public ElasticsearchPropertyMappingValidatorProvider createElasticsearchPropertyMappingValidatorProvider() {
-		return new Elasticsearch8PropertyMappingValidatorProvider();
+		return new Elasticsearch812PropertyMappingValidatorProvider();
 	}
 }

backend/elasticsearch/src/main/java/org/hibernate/search/backend/elasticsearch/dialect/model/impl/Elasticsearch8ModelDialect.java

@@ -6,9 +6,9 @@
  */
 package org.hibernate.search.backend.elasticsearch.dialect.model.impl;
 
-import org.hibernate.search.backend.elasticsearch.types.dsl.provider.impl.Elasticsearch8IndexFieldTypeFactoryProvider;
+import org.hibernate.search.backend.elasticsearch.types.dsl.provider.impl.Elasticsearch7IndexFieldTypeFactoryProvider;
 import org.hibernate.search.backend.elasticsearch.types.dsl.provider.impl.ElasticsearchIndexFieldTypeFactoryProvider;
-import org.hibernate.search.backend.elasticsearch.validation.impl.Elasticsearch8PropertyMappingValidatorProvider;
+import org.hibernate.search.backend.elasticsearch.validation.impl.Elasticsearch7PropertyMappingValidatorProvider;
 import org.hibernate.search.backend.elasticsearch.validation.impl.ElasticsearchPropertyMappingValidatorProvider;
 
 import com.google.gson.Gson;
@@ -20,11 +20,11 @@ public class Elasticsearch8ModelDialect implements ElasticsearchModelDialect {
 
 	@Override
 	public ElasticsearchIndexFieldTypeFactoryProvider createIndexTypeFieldFactoryProvider(Gson userFacingGson) {
-		return new Elasticsearch8IndexFieldTypeFactoryProvider( userFacingGson );
+		return new Elasticsearch7IndexFieldTypeFactoryProvider( userFacingGson );
 	}
 
 	@Override
 	public ElasticsearchPropertyMappingValidatorProvider createElasticsearchPropertyMappingValidatorProvider() {
-		return new Elasticsearch8PropertyMappingValidatorProvider();
+		return new Elasticsearch7PropertyMappingValidatorProvider();
 	}
 }

backend/elasticsearch/src/main/java/org/hibernate/search/backend/elasticsearch/logging/impl/Log.java

@@ -865,26 +865,11 @@ SearchException vectorKnnMatchVectorTypeDiffersFromField(String absoluteFieldPat
 			+ " Use the array of the same size as the vector field.")
 	SearchException vectorKnnMatchVectorDimensionDiffersFromField(String absoluteFieldPath, int expected, int actual);
 
-	@Message(id = ID_OFFSET + 183, value = "A knn predicate cannot be added. "
-			+ "With Elasticsearch, a knn predicate can only be a top-level predicate or a should clause of a top-level bool predicate.")
-	SearchException cannotAddKnnClauseAtThisStep();
-
-	@Message(id = ID_OFFSET + 184,
-			value = "An OpenSearch distribution does not allow specifying the `number of candidates` option. "
-					+ "This option is only applicable to an Elastic distribution of an Elasticsearch backend.")
-	SearchException knnNumberOfCandidatesUnsupportedOption();
-
 	@Message(id = ID_OFFSET + 185,
 			value = "An %1$s distribution version in use is not compatible with the Hibernate Search integration of vector search. "
 					+ "Update your %1$s cluster to a %2$s series to get vector search integration enabled.")
 	SearchException searchBackendVersionIncompatibleWithVectorIntegration(String distribution, String version);
 
-	@LogMessage(level = Logger.Level.WARN)
-	@Message(id = ID_OFFSET + 186,
-			value = "The Elastic distribution of Elasticsearch does not allow to apply constant score to a knn predicate."
-					+ " Constant score will not be applied.")
-	void elasticsearchKnnIgnoresConstantScore();
-
 	@Message(id = ID_OFFSET + 187,
 			value = "An OpenSearch distribution does not allow specifying the `required minimum similarity` option. "
 					+ "This option is only applicable to an Elastic distribution of an Elasticsearch backend.")

backend/elasticsearch/src/main/java/org/hibernate/search/backend/elasticsearch/search/predicate/impl/ElasticsearchBooleanPredicate.java

@@ -65,16 +65,10 @@ private ElasticsearchBooleanPredicate(Builder builder) {
 
 	@Override
 	public void checkNestableWithin(PredicateNestingContext context) {
-		// For Elasticsearch backend only:
-		// If this method is called that means we are trying to pass a bool predicate as a clause/filter/etc to some other predicate
-		//  and that would mean that if our current bool predicate has a knn should clause -- it is not ok to continue,
-		//  since it will place a knn clause deeper than a should clause of a top level bool predicate, which is not acceptable.
-		//  Because of that we are making sure that the context we pass in to check the clauses is updated:
-		PredicateNestingContext updatedContext = context.rejectKnn();
-		checkAcceptableWithin( updatedContext, mustClauses );
-		checkAcceptableWithin( updatedContext, shouldClauses );
-		checkAcceptableWithin( updatedContext, filterClauses );
-		checkAcceptableWithin( updatedContext, mustNotClauses );
+		checkAcceptableWithin( context, mustClauses );
+		checkAcceptableWithin( context, shouldClauses );
+		checkAcceptableWithin( context, filterClauses );
+		checkAcceptableWithin( context, mustNotClauses );
 	}
 
 	@Override
@@ -116,15 +110,7 @@ private void contributeClauses(PredicateRequestContext context, JsonObject inner
 		}
 
 		for ( ElasticsearchSearchPredicate clause : clauses ) {
-			JsonObject jsonQuery = clause.toJsonQuery( context );
-			if ( jsonQuery != null ) {
-				// This is an exceptional case for a KNN search on Elasticsearch distribution.
-				//  A Knn predicate would contribute to a knn clause inside the request context itself,
-				//  and we do not want to add this json as a clause to the bool predicate.
-				//  Hence, when the predicate returns null as JSON query and we ignore it.
-
-				GsonUtils.setOrAppendToArray( innerObject, occurProperty, jsonQuery );
-			}
+			GsonUtils.setOrAppendToArray( innerObject, occurProperty, clause.toJsonQuery( context ) );
 		}
 	}
 
@@ -221,10 +207,8 @@ public void must(SearchPredicate clause) {
 				mustClauses = new ArrayList<>();
 			}
 			ElasticsearchSearchPredicate elasticsearchClause = ElasticsearchSearchPredicate.from( scope, clause );
-			elasticsearchClause.checkNestableWithin( PredicateNestingContext.doesNotAcceptKnn() );
+			elasticsearchClause.checkNestableWithin( PredicateNestingContext.simple() );
 			mustClauses.add( elasticsearchClause );
-
-			checkShouldClausesForKnnAcceptability();
 		}
 
 		@Override
@@ -233,23 +217,17 @@ public void mustNot(SearchPredicate clause) {
 				mustNotClauses = new ArrayList<>();
 			}
 			ElasticsearchSearchPredicate elasticsearchClause = ElasticsearchSearchPredicate.from( scope, clause );
-			elasticsearchClause.checkNestableWithin( PredicateNestingContext.doesNotAcceptKnn() );
+			elasticsearchClause.checkNestableWithin( PredicateNestingContext.simple() );
 			mustNotClauses.add( elasticsearchClause );
-
-			checkShouldClausesForKnnAcceptability();
 		}
 
 		@Override
 		public void should(SearchPredicate clause) {
 			if ( shouldClauses == null ) {
 				shouldClauses = new ArrayList<>();
 			}
-
 			ElasticsearchSearchPredicate elasticsearchClause = ElasticsearchSearchPredicate.from( scope, clause );
-			elasticsearchClause.checkNestableWithin(
-					!hasNonShouldClause()
-							? PredicateNestingContext.acceptsKnn()
-							: PredicateNestingContext.doesNotAcceptKnn() );
+			elasticsearchClause.checkNestableWithin( PredicateNestingContext.simple() );
 			shouldClauses.add( elasticsearchClause );
 		}
 
@@ -259,10 +237,8 @@ public void filter(SearchPredicate clause) {
 				filterClauses = new ArrayList<>();
 			}
 			ElasticsearchSearchPredicate elasticsearchClause = ElasticsearchSearchPredicate.from( scope, clause );
-			elasticsearchClause.checkNestableWithin( PredicateNestingContext.doesNotAcceptKnn() );
+			elasticsearchClause.checkNestableWithin( PredicateNestingContext.simple() );
 			filterClauses.add( elasticsearchClause );
-
-			checkShouldClausesForKnnAcceptability();
 		}
 
 		@Override
@@ -351,20 +327,6 @@ private void optimizeClauseCollection(List<ElasticsearchSearchPredicate> collect
 			}
 		}
 
-		/*
-		 * For Elasticsearch backend only:
-		 * It may be that we've added knn should clauses and all was fine.
-		 * But now we are adding to such bool predicate a non-should clause.
-		 * We want to make sure that in such case there are no knn clauses in should or fail.
-		 * OpenSearch backend will not be affected by these checks.
-		 */
-		private void checkShouldClausesForKnnAcceptability() {
-			if ( shouldClauses != null ) {
-				shouldClauses.forEach(
-						should -> should.checkNestableWithin( PredicateNestingContext.doesNotAcceptKnn() ) );
-			}
-		}
-
 		private void checkAndClearClauseCollections() {
 			if ( mustClauses != null && mustClauses.isEmpty() ) {
 				mustClauses = null;
@@ -378,10 +340,6 @@ private boolean hasAtLeastOneMustOrFilterPredicate() {
 			return mustClauses != null || filterClauses != null;
 		}
 
-		private boolean hasNonShouldClause() {
-			return mustClauses != null || filterClauses != null || mustNotClauses != null;
-		}
-
 		private boolean hasOnlyOneMustClause() {
 			return mustClauses != null
 					&& mustClauses.size() == 1

backend/elasticsearch/src/main/java/org/hibernate/search/backend/elasticsearch/search/predicate/impl/ElasticsearchKnnPredicate.java

@@ -45,19 +45,6 @@ private ElasticsearchKnnPredicate(AbstractKnnBuilder<?> builder) {
 		builder.vector = null;
 	}
 
-	public static class ElasticsearchFactory<F>
-			extends AbstractElasticsearchCodecAwareSearchQueryElementFactory<KnnPredicateBuilder, F> {
-		public ElasticsearchFactory(ElasticsearchFieldCodec<F> codec) {
-			super( codec );
-		}
-
-		@Override
-		public KnnPredicateBuilder create(ElasticsearchSearchIndexScope<?> scope,
-				ElasticsearchSearchIndexValueFieldContext<F> field) {
-			return new ElasticsearchImpl.Builder<>( codec, scope, field );
-		}
-	}
-
 	public static class Elasticsearch812Factory<F>
 			extends AbstractElasticsearchCodecAwareSearchQueryElementFactory<KnnPredicateBuilder, F> {
 		public Elasticsearch812Factory(ElasticsearchFieldCodec<F> codec) {
@@ -132,7 +119,7 @@ public void vector(Object vector) {
 		@Override
 		public void filter(SearchPredicate filter) {
 			ElasticsearchSearchPredicate elasticsearchFilter = ElasticsearchSearchPredicate.from( scope, filter );
-			elasticsearchFilter.checkNestableWithin( PredicateNestingContext.doesNotAcceptKnn() );
+			elasticsearchFilter.checkNestableWithin( PredicateNestingContext.simple() );
 			this.filter = elasticsearchFilter;
 		}
 
@@ -153,81 +140,6 @@ private static JsonArray vectorToJsonArray(Object vector, Class<?> vectorElement
 		return array;
 	}
 
-	private static class ElasticsearchImpl extends ElasticsearchKnnPredicate {
-
-		private static final Log log = LoggerFactory.make( Log.class, MethodHandles.lookup() );
-
-		private static final JsonAccessor<String> FIELD_ACCESSOR = JsonAccessor.root().property( "field" ).asString();
-		private static final JsonArrayAccessor QUERY_VECTOR_ACCESSOR = JsonAccessor.root().property( "query_vector" ).asArray();
-		private static final JsonAccessor<Integer> K_ACCESSOR = JsonAccessor.root().property( "k" ).asInteger();
-
-		private static final JsonObjectAccessor FILTER_ACCESSOR = JsonAccessor.root().property( "filter" ).asObject();
-		private static final JsonAccessor<Integer> NUM_CANDIDATES_ACCESSOR =
-				JsonAccessor.root().property( "num_candidates" ).asInteger();
-		private static final JsonAccessor<Float> SIMILARITY_ACCESSOR = JsonAccessor.root().property( "similarity" ).asFloat();
-
-
-		private ElasticsearchImpl(Builder<?> builder) {
-			super( builder );
-		}
-
-		@Override
-		public JsonObject toJsonQuery(PredicateRequestContext context) {
-			// we want the query to get created and passed to the request context
-			context.contributeKnnClause( ( super.toJsonQuery( context ) ) );
-			// but we don't want it to be an actual query so we return `null`:
-			return null;
-		}
-
-		@Override
-		protected JsonObject doToJsonQuery(PredicateRequestContext context, JsonObject outerObject, JsonObject innerObject) {
-			FIELD_ACCESSOR.set( innerObject, absoluteFieldPath );
-			K_ACCESSOR.set( innerObject, k );
-			if ( filter != null ) {
-				JsonObject query = filter.toJsonQuery( context );
-				// we shouldn't get a null query here, since that's only possible if a filter was a knn predicate,
-				//   and in that case we are failing much faster for am Elasticsearch distribution...
-				FILTER_ACCESSOR.set( innerObject, query );
-			}
-			NUM_CANDIDATES_ACCESSOR.set( innerObject, k );
-			QUERY_VECTOR_ACCESSOR.set( innerObject, vector );
-			if ( similarity != null ) {
-				SIMILARITY_ACCESSOR.set( innerObject, similarity );
-			}
-			return innerObject;
-		}
-
-		@Override
-		public void checkNestableWithin(PredicateNestingContext context) {
-			if ( context.getNestedPath() != null || !context.acceptsKnnClause() ) {
-				throw log.cannotAddKnnClauseAtThisStep();
-			}
-		}
-
-		private static class Builder<F> extends AbstractKnnBuilder<F> {
-
-			private Builder(ElasticsearchFieldCodec<F> codec, ElasticsearchSearchIndexScope<?> scope,
-					ElasticsearchSearchIndexValueFieldContext<F> field) {
-				super( codec, scope, field );
-			}
-
-			@Override
-			public void constantScore() {
-				log.elasticsearchKnnIgnoresConstantScore();
-			}
-
-			@Override
-			public void requiredMinimumSimilarity(float similarity) {
-				this.similarity = similarity;
-			}
-
-			@Override
-			public SearchPredicate build() {
-				return new ElasticsearchImpl( this );
-			}
-		}
-	}
-
 	private static class Elasticsearch812Impl extends ElasticsearchKnnPredicate {
 
 		private static final JsonObjectAccessor KNN_ACCESSOR = JsonAccessor.root().property( "knn" ).asObject();

backend/elasticsearch/src/main/java/org/hibernate/search/backend/elasticsearch/search/predicate/impl/PredicateNestingContext.java

@@ -7,51 +7,26 @@
 package org.hibernate.search.backend.elasticsearch.search.predicate.impl;
 
 public class PredicateNestingContext {
-	private static final PredicateNestingContext ACCEPTS_KNN = new PredicateNestingContext( true );
-	private static final PredicateNestingContext DOES_NOT_ACCEPT_KNN = new PredicateNestingContext( false );
+	private static final PredicateNestingContext EMPTY = new PredicateNestingContext();
 	private final String nestedPath;
-	private final boolean acceptsKnnClause;
 
-	public static PredicateNestingContext acceptsKnn() {
-		return ACCEPTS_KNN;
-	}
-
-	public static PredicateNestingContext doesNotAcceptKnn() {
-		return DOES_NOT_ACCEPT_KNN;
+	public static PredicateNestingContext simple() {
+		return EMPTY;
 	}
 
 	public static PredicateNestingContext nested(String nestedPath) {
 		return new PredicateNestingContext( nestedPath );
 	}
 
-	private PredicateNestingContext(String nestedPath, boolean acceptsKnnClause) {
-		this.nestedPath = nestedPath;
-		this.acceptsKnnClause = acceptsKnnClause;
-	}
-
 	private PredicateNestingContext(String nestedPath) {
-		this( nestedPath, false );
+		this.nestedPath = nestedPath;
 	}
 
-	private PredicateNestingContext(boolean acceptsKnnClause) {
-		this( null, acceptsKnnClause );
+	private PredicateNestingContext() {
+		this( null );
 	}
 
 	public String getNestedPath() {
 		return nestedPath;
 	}
-
-	public boolean acceptsKnnClause() {
-		return acceptsKnnClause;
-	}
-
-	public PredicateNestingContext rejectKnn() {
-		if ( !acceptsKnnClause ) {
-			return this;
-		}
-		if ( nestedPath == null ) {
-			return DOES_NOT_ACCEPT_KNN;
-		}
-		return new PredicateNestingContext( nestedPath, false );
-	}
 }