Allow updating inference_id of semantic_text fields

docs/changelog/136120.yaml

@@ -0,0 +1,5 @@
+pr: 136120
+summary: Allow updating `inference_id` of `semantic_text` fields
+area: "Mapping"
+type: enhancement
+issues: []

docs/reference/elasticsearch/mapping-reference/semantic-text.md

@@ -138,12 +138,31 @@ While we do encourage experimentation, we do not recommend implementing producti
 
 `inference_id`
 :   (Optional, string) {{infer-cap}} endpoint that will be used to generate
-embeddings for the field. By default, `.elser-2-elasticsearch` is used. This
-parameter cannot be updated. Use
-the [Create {{infer}} API](https://www.elastic.co/docs/api/doc/elasticsearch/operation/operation-inference-put)
+embeddings for the field. By default, `.elser-2-elasticsearch` is used.
+Use the [Create {{infer}} API](https://www.elastic.co/docs/api/doc/elasticsearch/operation/operation-inference-put)
 to create the endpoint. If `search_inference_id` is specified, the {{infer}}
 endpoint will only be used at index time.
 
+::::{applies-switch}
+
+:::{applies-item} { "stack": "ga 9.0" }
+This parameter cannot be updated.
+:::
+
+:::{applies-item} { "stack": "ga 9.3" }
+
+You can update this parameter by using
+the [Update mapping API](https://www.elastic.co/docs/api/doc/elasticsearch/operation/operation-indices-put-mapping).
+You can update the inference endpoint if no values have been indexed or if the new endpoint is compatible with the current one.
+
+::::{warning}
+When updating an `inference_id` it is important to ensure the new {{infer}} endpoint produces the correct embeddings for your use case. This typically means using the same underlying model.
+::::
+
+:::
+
+::::
+
 `search_inference_id`
 :   (Optional, string) {{infer-cap}} endpoint that will be used to generate
 embeddings at query time. You can update this parameter by using

x-pack/plugin/inference/qa/inference-service-tests/src/javaRestTest/java/org/elasticsearch/xpack/inference/InferenceGetServicesIT.java

@@ -74,6 +74,7 @@ public void testGetServicesWithoutTaskType() throws IOException {
                     "completion_test_service",
                     "test_reranking_service",
                     "test_service",
+                    "alternate_sparse_embedding_test_service",
                     "text_embedding_test_service",
                     "voyageai",
                     "watsonxai",
@@ -209,6 +210,7 @@ public void testGetServicesWithSparseEmbeddingTaskType() throws IOException {
                     "hugging_face",
                     "streaming_completion_test_service",
                     "test_service",
+                    "alternate_sparse_embedding_test_service",
                     "amazon_sagemaker"
                 ).toArray()
             )

x-pack/plugin/inference/qa/test-service-plugin/src/main/java/org/elasticsearch/xpack/inference/mock/TestSparseInferenceServiceExtension.java

@@ -13,7 +13,6 @@
 import org.elasticsearch.common.ValidationException;
 import org.elasticsearch.common.io.stream.StreamInput;
 import org.elasticsearch.common.io.stream.StreamOutput;
-import org.elasticsearch.common.util.LazyInitializable;
 import org.elasticsearch.core.Nullable;
 import org.elasticsearch.core.TimeValue;
 import org.elasticsearch.inference.ChunkInferenceInput;
@@ -43,12 +42,13 @@
 import java.util.HashMap;
 import java.util.List;
 import java.util.Map;
+import java.util.Objects;
 
 public class TestSparseInferenceServiceExtension implements InferenceServiceExtension {
 
     @Override
     public List<Factory> getInferenceServiceFactories() {
-        return List.of(TestInferenceService::new);
+        return List.of(TestInferenceService::new, TestAlternateSparseInferenceService::new);
     }
 
     public static class TestSparseModel extends Model {
@@ -60,16 +60,40 @@ public TestSparseModel(String inferenceEntityId, TestServiceSettings serviceSett
         }
     }
 
-    public static class TestInferenceService extends AbstractTestInferenceService {
+    public static class TestInferenceService extends AbstractSparseTestInferenceService {
         public static final String NAME = "test_service";
 
+        public TestInferenceService(InferenceServiceFactoryContext inferenceServiceFactoryContext) {}
+
+        @Override
+        protected String testServiceName() {
+            return NAME;
+        }
+    }
+
+    /**
+     * A second sparse service allows testing updates from one service to another.
+     */
+    public static class TestAlternateSparseInferenceService extends AbstractSparseTestInferenceService {
+        public static final String NAME = "alternate_sparse_embedding_test_service";
+
+        public TestAlternateSparseInferenceService(InferenceServiceFactoryContext inferenceServiceFactoryContext) {}
+
+        @Override
+        protected String testServiceName() {
+            return NAME;
+        }
+    }
+
+    abstract static class AbstractSparseTestInferenceService extends AbstractTestInferenceService {
+
         private static final EnumSet<TaskType> supportedTaskTypes = EnumSet.of(TaskType.SPARSE_EMBEDDING);
 
-        public TestInferenceService(InferenceServiceExtension.InferenceServiceFactoryContext context) {}
+        protected abstract String testServiceName();
 
         @Override
         public String name() {
-            return NAME;
+            return testServiceName();
         }
 
         @Override
@@ -92,7 +116,7 @@ public void parseRequestConfig(
 
         @Override
         public InferenceServiceConfiguration getConfiguration() {
-            return Configuration.get();
+            return new Configuration(testServiceName()).get();
         }
 
         @Override
@@ -195,41 +219,43 @@ private static float generateEmbedding(String input, int position) {
         }
 
         public static class Configuration {
-            public static InferenceServiceConfiguration get() {
-                return configuration.getOrCompute();
+
+            private final String serviceName;
+
+            Configuration(String serviceName) {
+                this.serviceName = Objects.requireNonNull(serviceName);
+            }
+
+            InferenceServiceConfiguration get() {
+                var configurationMap = new HashMap<String, SettingsConfiguration>();
+
+                configurationMap.put(
+                    "model",
+                    new SettingsConfiguration.Builder(EnumSet.of(TaskType.SPARSE_EMBEDDING)).setDescription("")
+                        .setLabel("Model")
+                        .setRequired(true)
+                        .setSensitive(false)
+                        .setType(SettingsConfigurationFieldType.STRING)
+                        .build()
+                );
+
+                configurationMap.put(
+                    "hidden_field",
+                    new SettingsConfiguration.Builder(EnumSet.of(TaskType.SPARSE_EMBEDDING)).setDescription("")
+                        .setLabel("Hidden Field")
+                        .setRequired(true)
+                        .setSensitive(false)
+                        .setType(SettingsConfigurationFieldType.STRING)
+                        .build()
+                );
+
+                return new InferenceServiceConfiguration.Builder().setService(serviceName)
+                    .setName(serviceName)
+                    .setTaskTypes(supportedTaskTypes)
+                    .setConfigurations(configurationMap)
+                    .build();
             }
 
-            private static final LazyInitializable<InferenceServiceConfiguration, RuntimeException> configuration = new LazyInitializable<>(
-                () -> {
-                    var configurationMap = new HashMap<String, SettingsConfiguration>();
-
-                    configurationMap.put(
-                        "model",
-                        new SettingsConfiguration.Builder(EnumSet.of(TaskType.SPARSE_EMBEDDING)).setDescription("")
-                            .setLabel("Model")
-                            .setRequired(true)
-                            .setSensitive(false)
-                            .setType(SettingsConfigurationFieldType.STRING)
-                            .build()
-                    );
-
-                    configurationMap.put(
-                        "hidden_field",
-                        new SettingsConfiguration.Builder(EnumSet.of(TaskType.SPARSE_EMBEDDING)).setDescription("")
-                            .setLabel("Hidden Field")
-                            .setRequired(true)
-                            .setSensitive(false)
-                            .setType(SettingsConfigurationFieldType.STRING)
-                            .build()
-                    );
-
-                    return new InferenceServiceConfiguration.Builder().setService(NAME)
-                        .setName(NAME)
-                        .setTaskTypes(supportedTaskTypes)
-                        .setConfigurations(configurationMap)
-                        .build();
-                }
-            );
         }
     }
 

x-pack/plugin/inference/src/main/java/org/elasticsearch/xpack/inference/InferenceFeatures.java

@@ -24,6 +24,7 @@
 import static org.elasticsearch.xpack.inference.mapper.SemanticTextFieldMapper.SEMANTIC_TEXT_INDEX_OPTIONS_WITH_DEFAULTS;
 import static org.elasticsearch.xpack.inference.mapper.SemanticTextFieldMapper.SEMANTIC_TEXT_SPARSE_VECTOR_INDEX_OPTIONS;
 import static org.elasticsearch.xpack.inference.mapper.SemanticTextFieldMapper.SEMANTIC_TEXT_SUPPORT_CHUNKING_CONFIG;
+import static org.elasticsearch.xpack.inference.mapper.SemanticTextFieldMapper.SEMANTIC_TEXT_UPDATABLE_INFERENCE_ID;
 import static org.elasticsearch.xpack.inference.queries.LegacySemanticKnnVectorQueryRewriteInterceptor.SEMANTIC_KNN_FILTER_FIX;
 import static org.elasticsearch.xpack.inference.queries.LegacySemanticKnnVectorQueryRewriteInterceptor.SEMANTIC_KNN_VECTOR_QUERY_REWRITE_INTERCEPTION_SUPPORTED;
 import static org.elasticsearch.xpack.inference.queries.LegacySemanticMatchQueryRewriteInterceptor.SEMANTIC_MATCH_QUERY_REWRITE_INTERCEPTION_SUPPORTED;
@@ -93,6 +94,7 @@ public Set<NodeFeature> getTestFeatures() {
                 SEMANTIC_TEXT_HIGHLIGHTING_FLAT,
                 SEMANTIC_TEXT_SPARSE_VECTOR_INDEX_OPTIONS,
                 SEMANTIC_TEXT_FIELDS_CHUNKS_FORMAT,
+                SEMANTIC_TEXT_UPDATABLE_INFERENCE_ID,
                 SemanticQueryBuilder.SEMANTIC_QUERY_MULTIPLE_INFERENCE_IDS,
                 SemanticQueryBuilder.SEMANTIC_QUERY_FILTER_FIELD_CAPS_FIX,
                 InterceptedInferenceQueryBuilder.NEW_SEMANTIC_QUERY_INTERCEPTORS,

x-pack/plugin/inference/src/main/java/org/elasticsearch/xpack/inference/mapper/SemanticTextFieldMapper.java

@@ -150,6 +150,7 @@ public class SemanticTextFieldMapper extends FieldMapper implements InferenceFie
     public static final NodeFeature SEMANTIC_TEXT_SPARSE_VECTOR_INDEX_OPTIONS = new NodeFeature(
         "semantic_text.sparse_vector_index_options"
     );
+    public static final NodeFeature SEMANTIC_TEXT_UPDATABLE_INFERENCE_ID = new NodeFeature("semantic_text.updatable_inference_id");
 
     public static final String CONTENT_TYPE = "semantic_text";
     public static final String DEFAULT_ELSER_2_INFERENCE_ID = DEFAULT_ELSER_ID;
@@ -243,7 +244,7 @@ public Builder(
 
             this.inferenceId = Parameter.stringParam(
                 INFERENCE_ID_FIELD,
-                false,
+                true,
                 mapper -> ((SemanticTextFieldType) mapper.fieldType()).inferenceId,
                 DEFAULT_ELSER_2_INFERENCE_ID
             ).addValidator(v -> {
@@ -326,9 +327,65 @@ protected Parameter<?>[] getParameters() {
         @Override
         protected void merge(FieldMapper mergeWith, Conflicts conflicts, MapperMergeContext mapperMergeContext) {
             SemanticTextFieldMapper semanticMergeWith = (SemanticTextFieldMapper) mergeWith;
-            semanticMergeWith = copySettings(semanticMergeWith, mapperMergeContext);
 
-            // We make sure to merge the inference field first to catch any model conflicts
+            final boolean isInferenceIdUpdate = semanticMergeWith.fieldType().inferenceId.equals(inferenceId.get()) == false;
+            final boolean hasExplicitModelSettings = modelSettings.get() != null;
+
+            MinimalServiceSettings updatedModelSettings = modelSettings.get();
+            if (isInferenceIdUpdate && hasExplicitModelSettings) {
+                validateModelsAreCompatibleWhenInferenceIdIsUpdated(semanticMergeWith.fieldType().inferenceId, conflicts);
+                // As the mapper previously had explicit model settings, we need to apply to the new merged mapper
+                // the resolved model settings if not explicitly set.
+                updatedModelSettings = modelRegistry.getMinimalServiceSettings(semanticMergeWith.fieldType().inferenceId);
+            }
+
+            semanticMergeWith = copyWithNewModelSettingsIfNotSet(semanticMergeWith, updatedModelSettings, mapperMergeContext);
+
+            // We make sure to merge the inference field first to catch any model conflicts.
+            // If inference_id is updated and there are no explicit model settings, we should be
+            // able to switch to the new inference field without the need to check for conflicts.
+            if (isInferenceIdUpdate == false || hasExplicitModelSettings) {
+                mergeInferenceField(mapperMergeContext, semanticMergeWith);
+            }
+
+            super.merge(semanticMergeWith, conflicts, mapperMergeContext);
+            conflicts.check();
+        }
+
+        private void validateModelsAreCompatibleWhenInferenceIdIsUpdated(String newInferenceId, Conflicts conflicts) {
+            MinimalServiceSettings currentModelSettings = modelSettings.get();
+            MinimalServiceSettings updatedModelSettings = modelRegistry.getMinimalServiceSettings(newInferenceId);
+            if (currentModelSettings != null && updatedModelSettings == null) {
+                throw new IllegalArgumentException(
+                    "Cannot update ["
+                        + CONTENT_TYPE
+                        + "] field ["
+                        + leafName()
+                        + "] because inference endpoint ["
+                        + newInferenceId
+                        + "] does not exist."
+                );
+            }
+            if (canMergeModelSettings(currentModelSettings, updatedModelSettings, conflicts) == false) {
+                throw new IllegalArgumentException(
+                    "Cannot update ["
+                        + CONTENT_TYPE
+                        + "] field ["
+                        + leafName()
+                        + "] because inference endpoint ["
+                        + inferenceId.get()
+                        + "] with model settings ["
+                        + currentModelSettings
+                        + "] is not compatible with new inference endpoint ["
+                        + newInferenceId
+                        + "] with model settings ["
+                        + updatedModelSettings
+                        + "]."
+                );
+            }
+        }
+
+        private void mergeInferenceField(MapperMergeContext mapperMergeContext, SemanticTextFieldMapper semanticMergeWith) {
             try {
                 var context = mapperMergeContext.createChildContext(semanticMergeWith.leafName(), ObjectMapper.Dynamic.FALSE);
                 var inferenceField = inferenceFieldBuilder.apply(context.getMapperBuilderContext());
@@ -341,9 +398,6 @@ protected void merge(FieldMapper mergeWith, Conflicts conflicts, MapperMergeCont
                     : "";
                 throw new IllegalArgumentException(errorMessage, e);
             }
-
-            super.merge(semanticMergeWith, conflicts, mapperMergeContext);
-            conflicts.check();
         }
 
         /**
@@ -499,18 +553,23 @@ private void validateIndexOptions(SemanticTextIndexOptions indexOptions, String
         }
 
         /**
-         * As necessary, copy settings from this builder to the passed-in mapper.
-         * Used to preserve {@link MinimalServiceSettings} when updating a semantic text mapping to one where the model settings
-         * are not specified.
+         * Creates a new mapper with the new model settings if model settings are not set on the mapper.
+         * If the mapper already has model settings or the new model settings are null, the mapper is
+         * returned unchanged.
          *
-         * @param mapper The mapper
+         * @param mapper        The mapper
+         * @param modelSettings the new model settings. If null the mapper will be returned unchanged.
          * @return A mapper with the copied settings applied
          */
-        private SemanticTextFieldMapper copySettings(SemanticTextFieldMapper mapper, MapperMergeContext mapperMergeContext) {
+        private SemanticTextFieldMapper copyWithNewModelSettingsIfNotSet(
+            SemanticTextFieldMapper mapper,
+            @Nullable MinimalServiceSettings modelSettings,
+            MapperMergeContext mapperMergeContext
+        ) {
             SemanticTextFieldMapper returnedMapper = mapper;
             if (mapper.fieldType().getModelSettings() == null) {
                 Builder builder = from(mapper);
-                builder.setModelSettings(modelSettings.getValue());
+                builder.setModelSettings(modelSettings);
                 returnedMapper = builder.build(mapperMergeContext.getMapperBuilderContext());
             }