[8.19] add default inference endpoint for Elastic Inference Service rerank (elastic#129681) (elastic#129985)

timgrein · brendan-jugan-elastic · web-flow · commit f922b8ee7432 · 2025-06-26T03:36:41.000-04:00
* add default inference endpoint for Elastic Inference Service rerank (elastic#129681) * add Elastic Inference Service rerank default inference endpoint * [CI] Auto commit changes from spotless * fix integ tests * update mock Elastic Inference Service authorization response * fix rerank service test --------- Co-authored-by: elasticsearchmachine <infra-root+elasticsearchmachine@elastic.co> (cherry picked from commit cef717c) # Conflicts: # x-pack/plugin/inference/qa/inference-service-tests/src/javaRestTest/java/org/elasticsearch/xpack/inference/InferenceGetServicesIT.java # x-pack/plugin/inference/src/main/java/org/elasticsearch/xpack/inference/services/elastic/ElasticInferenceService.java * Fix InferenceGetServicesIT rerank integration test * Fix InferenceGetServicesIT rerank integration test * spotlessApply * Fix integration test --------- Co-authored-by: Brendan Jugan <brendan.jugan@elastic.co>
diff --git a/x-pack/plugin/inference/qa/inference-service-tests/src/javaRestTest/java/org/elasticsearch/xpack/inference/InferenceGetServicesIT.java b/x-pack/plugin/inference/qa/inference-service-tests/src/javaRestTest/java/org/elasticsearch/xpack/inference/InferenceGetServicesIT.java
@@ -112,21 +112,26 @@ private Iterable<String> providersFor(TaskType taskType) throws IOException {
     }
 
     public void testGetServicesWithRerankTaskType() throws IOException {
+        List<Object> services = getServices(TaskType.RERANK);
+        assertThat(services.size(), equalTo(11));
+
+        var providers = providers(services);
+
         assertThat(
             providersFor(TaskType.RERANK),
             containsInAnyOrder(
                 List.of(
                     "alibabacloud-ai-search",
+                    "amazon_sagemaker",
                     "cohere",
                     "custom",
                     "elastic",
                     "elasticsearch",
                     "googlevertexai",
+                    "hugging_face",
                     "jinaai",
                     "test_reranking_service",
-                    "voyageai",
-                    "hugging_face",
-                    "amazon_sagemaker"
+                    "voyageai"
                 ).toArray()
             )
         );
diff --git a/x-pack/plugin/inference/qa/inference-service-tests/src/javaRestTest/java/org/elasticsearch/xpack/inference/MockElasticInferenceServiceAuthorizationServer.java b/x-pack/plugin/inference/qa/inference-service-tests/src/javaRestTest/java/org/elasticsearch/xpack/inference/MockElasticInferenceServiceAuthorizationServer.java
@@ -46,7 +46,7 @@ public void enqueueAuthorizeAllModelsResponse() {
                       "model_name": "multilingual-embed-v1",
                       "task_types": ["embed/text/dense"]
                     },
-                  {
+                    {
                       "model_name": "rerank-v1",
                       "task_types": ["rerank/text/text-similarity"]
                     }
diff --git a/x-pack/plugin/inference/src/internalClusterTest/java/org/elasticsearch/xpack/inference/integration/InferenceRevokeDefaultEndpointsIT.java b/x-pack/plugin/inference/src/internalClusterTest/java/org/elasticsearch/xpack/inference/integration/InferenceRevokeDefaultEndpointsIT.java
@@ -283,6 +283,7 @@ public void testRemoves_DefaultChatCompletion_V1_WhenAuthorizationDoesNotReturnA
                         {
                           "model_name": "rerank-v1",
                           "task_types": ["rerank/text/text-similarity"]
+
                         },
                         {
                           "model_name": "multilingual-embed-v1",
@@ -299,28 +300,29 @@ public void testRemoves_DefaultChatCompletion_V1_WhenAuthorizationDoesNotReturnA
 
                 assertThat(service.supportedStreamingTasks(), is(EnumSet.noneOf(TaskType.class)));
                 assertThat(
-                    service.defaultConfigIds(),
-                    containsInAnyOrder(
-                        new InferenceService.DefaultConfigId(
-                            ".elser-v2-elastic",
-                            MinimalServiceSettings.sparseEmbedding(ElasticInferenceService.NAME),
-                            service
-                        ),
-                        new InferenceService.DefaultConfigId(
-                            ".multilingual-embed-v1-elastic",
-                            MinimalServiceSettings.textEmbedding(
-                                ElasticInferenceService.NAME,
-                                ElasticInferenceService.DENSE_TEXT_EMBEDDINGS_DIMENSIONS,
-                                ElasticInferenceService.defaultDenseTextEmbeddingsSimilarity(),
-                                DenseVectorFieldMapper.ElementType.FLOAT
-                            ),
-                            service
+                    service.supportedTaskTypes(),
+                    is(EnumSet.of(TaskType.TEXT_EMBEDDING, TaskType.SPARSE_EMBEDDING, TaskType.RERANK))
+                );
+                containsInAnyOrder(
+                    new InferenceService.DefaultConfigId(
+                        ".elser-v2-elastic",
+                        MinimalServiceSettings.sparseEmbedding(ElasticInferenceService.NAME),
+                        service
+                    ),
+                    new InferenceService.DefaultConfigId(
+                        ".multilingual-embed-v1-elastic",
+                        MinimalServiceSettings.textEmbedding(
+                            ElasticInferenceService.NAME,
+                            ElasticInferenceService.DENSE_TEXT_EMBEDDINGS_DIMENSIONS,
+                            ElasticInferenceService.defaultDenseTextEmbeddingsSimilarity(),
+                            DenseVectorFieldMapper.ElementType.FLOAT
                         ),
-                        new InferenceService.DefaultConfigId(
-                            ".rerank-v1-elastic",
-                            MinimalServiceSettings.rerank(ElasticInferenceService.NAME),
-                            service
-                        )
+                        service
+                    ),
+                    new InferenceService.DefaultConfigId(
+                        ".rerank-v1-elastic",
+                        MinimalServiceSettings.rerank(ElasticInferenceService.NAME),
+                        service
                     )
                 );
                 assertThat(
diff --git a/x-pack/plugin/inference/src/main/java/org/elasticsearch/xpack/inference/services/elastic/ElasticInferenceService.java b/x-pack/plugin/inference/src/main/java/org/elasticsearch/xpack/inference/services/elastic/ElasticInferenceService.java
@@ -210,7 +210,6 @@ private static Map<String, DefaultModelConfig> initDefaultEndpoints(
                     DenseVectorFieldMapper.ElementType.FLOAT
                 )
             ),
-
             DEFAULT_RERANK_MODEL_ID_V1,
             new DefaultModelConfig(
                 new ElasticInferenceServiceRerankModel(
diff --git a/x-pack/plugin/inference/src/main/java/org/elasticsearch/xpack/inference/services/elastic/rerank/ElasticInferenceServiceRerankModel.java b/x-pack/plugin/inference/src/main/java/org/elasticsearch/xpack/inference/services/elastic/rerank/ElasticInferenceServiceRerankModel.java
@@ -87,7 +87,7 @@ public URI uri() {
     private URI createUri() throws ElasticsearchStatusException {
         try {
             // TODO, consider transforming the base URL into a URI for better error handling.
-            return new URI(elasticInferenceServiceComponents().elasticInferenceServiceUrl() + "/api/v1/rerank");
+            return new URI(elasticInferenceServiceComponents().elasticInferenceServiceUrl() + "/api/v1/rerank/text/text-similarity");
         } catch (URISyntaxException e) {
             throw new ElasticsearchStatusException(
                 "Failed to create URI for service ["
diff --git a/x-pack/plugin/inference/src/test/java/org/elasticsearch/xpack/inference/services/elastic/ElasticInferenceServiceTests.java b/x-pack/plugin/inference/src/test/java/org/elasticsearch/xpack/inference/services/elastic/ElasticInferenceServiceTests.java
@@ -1315,6 +1315,7 @@ public void testDefaultConfigs_Returns_DefaultEndpoints_WhenTaskTypeIsCorrect()
             PlainActionFuture<List<Model>> listener = new PlainActionFuture<>();
             service.defaultConfigs(listener);
             var models = listener.actionGet(TIMEOUT);
+
             assertThat(models.size(), is(4));
             assertThat(models.get(0).getConfigurations().getInferenceEntityId(), is(".elser-v2-elastic"));
             assertThat(models.get(1).getConfigurations().getInferenceEntityId(), is(".multilingual-embed-v1-elastic"));

Original file line number	Diff line number	Diff line change
`@@ -46,7 +46,7 @@ public void enqueueAuthorizeAllModelsResponse() {`
`46`	`46`	`"model_name": "multilingual-embed-v1",`
`47`	`47`	`"task_types": ["embed/text/dense"]`
`48`	`48`	`},`
`49`		`- {`
	`49`	`+ {`
`50`	`50`	`"model_name": "rerank-v1",`
`51`	`51`	`"task_types": ["rerank/text/text-similarity"]`
`52`	`52`	`}`
Original file line number	Diff line number	Diff line change
`@@ -210,7 +210,6 @@ private static Map<String, DefaultModelConfig> initDefaultEndpoints(`
`210`	`210`	`DenseVectorFieldMapper.ElementType.FLOAT`
`211`	`211`	`)`
`212`	`212`	`),`
`213`		`-`
`214`	`213`	`DEFAULT_RERANK_MODEL_ID_V1,`
`215`	`214`	`new DefaultModelConfig(`
`216`	`215`	`new ElasticInferenceServiceRerankModel(`