Fix nested field missing sub embedding field

CHANGELOG.md

@@ -8,6 +8,7 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 ### Enhancements
 - Set neural-search plugin 3.0.0 baseline JDK version to JDK-2 ([#838](https://github.com/opensearch-project/neural-search/pull/838))
 ### Bug Fixes
+- Fix for nested field missing sub embedding field in text embedding processor ([#913](https://github.com/opensearch-project/neural-search/pull/913))
 ### Infrastructure
 ### Documentation
 ### Maintenance

src/main/java/org/opensearch/neuralsearch/processor/InferenceProcessor.java

@@ -10,6 +10,7 @@
 import java.util.Collections;
 import java.util.Comparator;
 import java.util.HashMap;
+import java.util.Iterator;
 import java.util.LinkedHashMap;
 import java.util.List;
 import java.util.Map;
@@ -285,7 +286,7 @@ private void createInferenceListForMapTypeInput(Object sourceValue, List<String>
         if (sourceValue instanceof Map) {
             ((Map<String, Object>) sourceValue).forEach((k, v) -> createInferenceListForMapTypeInput(v, texts));
         } else if (sourceValue instanceof List) {
-            texts.addAll(((List<String>) sourceValue));
+            ((List<String>) sourceValue).stream().filter(Objects::nonNull).forEach(texts::add);
         } else {
             if (sourceValue == null) return;
             texts.add(sourceValue.toString());
@@ -419,8 +420,12 @@ private void putNLPResultToSourceMapForMapType(
             for (Map.Entry<String, Object> inputNestedMapEntry : ((Map<String, Object>) sourceValue).entrySet()) {
                 if (sourceAndMetadataMap.get(processorKey) instanceof List) {
                     // build nlp output for list of nested objects
+                    Iterator<Object> inputNestedMapValueIt = ((List<Object>) inputNestedMapEntry.getValue()).iterator();
                     for (Map<String, Object> nestedElement : (List<Map<String, Object>>) sourceAndMetadataMap.get(processorKey)) {
-                        nestedElement.put(inputNestedMapEntry.getKey(), results.get(indexWrapper.index++));
+                        // Only fill in when value is not null
+                        if (inputNestedMapValueIt.hasNext() && inputNestedMapValueIt.next() != null) {
+                            nestedElement.put(inputNestedMapEntry.getKey(), results.get(indexWrapper.index++));
+                        }
                     }
                 } else {
                     Pair<String, Object> processedNestedKey = processNestedKey(inputNestedMapEntry);

src/test/java/org/opensearch/neuralsearch/processor/InferenceProcessorTests.java

@@ -66,7 +66,7 @@ public void test_batchExecute_emptyInput() {
         verify(clientAccessor, never()).inferenceSentences(anyString(), anyList(), any());
     }
 
-    public void test_batchExecute_allFailedValidation() {
+    public void test_batchExecuteWithEmpty_allFailedValidation() {
         final int docCount = 2;
         TestInferenceProcessor processor = new TestInferenceProcessor(createMockVectorResult(), BATCH_SIZE, null);
         List<IngestDocumentWrapper> wrapperList = createIngestDocumentWrappers(docCount);
@@ -79,6 +79,29 @@ public void test_batchExecute_allFailedValidation() {
         assertEquals(docCount, captor.getValue().size());
         for (int i = 0; i < docCount; ++i) {
             assertNotNull(captor.getValue().get(i).getException());
+            assertEquals(
+                "list type field [key1] has empty string, cannot process it",
+                captor.getValue().get(i).getException().getMessage()
+            );
+            assertEquals(wrapperList.get(i).getIngestDocument(), captor.getValue().get(i).getIngestDocument());
+        }
+        verify(clientAccessor, never()).inferenceSentences(anyString(), anyList(), any());
+    }
+
+    public void test_batchExecuteWithNull_allFailedValidation() {
+        final int docCount = 2;
+        TestInferenceProcessor processor = new TestInferenceProcessor(createMockVectorResult(), BATCH_SIZE, null);
+        List<IngestDocumentWrapper> wrapperList = createIngestDocumentWrappers(docCount);
+        wrapperList.get(0).getIngestDocument().setFieldValue("key1", Arrays.asList(null, "value1"));
+        wrapperList.get(1).getIngestDocument().setFieldValue("key1", Arrays.asList(null, "value1"));
+        Consumer resultHandler = mock(Consumer.class);
+        processor.batchExecute(wrapperList, resultHandler);
+        ArgumentCaptor<List<IngestDocumentWrapper>> captor = ArgumentCaptor.forClass(List.class);
+        verify(resultHandler).accept(captor.capture());
+        assertEquals(docCount, captor.getValue().size());
+        for (int i = 0; i < docCount; ++i) {
+            assertNotNull(captor.getValue().get(i).getException());
+            assertEquals("list type field [key1] has null, cannot process it", captor.getValue().get(i).getException().getMessage());
             assertEquals(wrapperList.get(i).getIngestDocument(), captor.getValue().get(i).getIngestDocument());
         }
         verify(clientAccessor, never()).inferenceSentences(anyString(), anyList(), any());

src/test/java/org/opensearch/neuralsearch/processor/TextEmbeddingProcessorIT.java

@@ -310,5 +310,14 @@ private void ingestBatchDocumentWithBulk(String idPrefix, int docCount, Set<Inte
         );
         assertEquals(!failedIds.isEmpty(), map.get("errors"));
         assertEquals(docCount, ((List) map.get("items")).size());
+
+        int failedDocCount = 0;
+        for (Object item : ((List) map.get("items"))) {
+            Map<String, Map<String, Object>> itemMap = (Map<String, Map<String, Object>>) item;
+            if (itemMap.get("index").get("error") != null) {
+                failedDocCount++;
+            }
+        }
+        assertEquals(failedIds.size(), failedDocCount);
     }
 }

src/test/java/org/opensearch/neuralsearch/processor/TextEmbeddingProcessorTests.java

@@ -730,7 +730,7 @@ public void testBuildVectorOutput_withNestedList_successful() {
         IngestDocument ingestDocument = createNestedListIngestDocument();
         TextEmbeddingProcessor textEmbeddingProcessor = createInstanceWithNestedMapConfiguration(config);
         Map<String, Object> knnMap = textEmbeddingProcessor.buildMapWithTargetKeys(ingestDocument);
-        List<List<Float>> modelTensorList = createMockVectorResult();
+        List<List<Float>> modelTensorList = createRandomOneDimensionalMockVector(2, 2, 0.0f, 1.0f);
         textEmbeddingProcessor.buildNLPResult(knnMap, modelTensorList, ingestDocument.getSourceAndMetadata());
         List<Map<String, Object>> nestedObj = (List<Map<String, Object>>) ingestDocument.getSourceAndMetadata().get("nestedField");
         assertTrue(nestedObj.get(0).containsKey("vectorField"));
@@ -739,12 +739,27 @@ public void testBuildVectorOutput_withNestedList_successful() {
         assertNotNull(nestedObj.get(1).get("vectorField"));
     }
 
+    @SuppressWarnings("unchecked")
+    public void testBuildVectorOutput_withNestedListHasNotForEmbeddingField_successful() {
+        Map<String, Object> config = createNestedListConfiguration();
+        IngestDocument ingestDocument = createNestedListWithNotEmbeddingFieldIngestDocument();
+        TextEmbeddingProcessor textEmbeddingProcessor = createInstanceWithNestedMapConfiguration(config);
+        Map<String, Object> knnMap = textEmbeddingProcessor.buildMapWithTargetKeys(ingestDocument);
+        List<List<Float>> modelTensorList = createRandomOneDimensionalMockVector(1, 2, 0.0f, 1.0f);
+        textEmbeddingProcessor.buildNLPResult(knnMap, modelTensorList, ingestDocument.getSourceAndMetadata());
+        List<Map<String, Object>> nestedObj = (List<Map<String, Object>>) ingestDocument.getSourceAndMetadata().get("nestedField");
+        assertFalse(nestedObj.get(0).containsKey("vectorField"));
+        assertTrue(nestedObj.get(0).containsKey("textFieldNotForEmbedding"));
+        assertTrue(nestedObj.get(1).containsKey("vectorField"));
+        assertNotNull(nestedObj.get(1).get("vectorField"));
+    }
+
     public void testBuildVectorOutput_withNestedList_Level2_successful() {
         Map<String, Object> config = createNestedList2LevelConfiguration();
         IngestDocument ingestDocument = create2LevelNestedListIngestDocument();
         TextEmbeddingProcessor textEmbeddingProcessor = createInstanceWithNestedMapConfiguration(config);
         Map<String, Object> knnMap = textEmbeddingProcessor.buildMapWithTargetKeys(ingestDocument);
-        List<List<Float>> modelTensorList = createMockVectorResult();
+        List<List<Float>> modelTensorList = createRandomOneDimensionalMockVector(2, 2, 0.0f, 1.0f);
         textEmbeddingProcessor.buildNLPResult(knnMap, modelTensorList, ingestDocument.getSourceAndMetadata());
         Map<String, Object> nestedLevel1 = (Map<String, Object>) ingestDocument.getSourceAndMetadata().get("nestedField");
         List<Map<String, Object>> nestedObj = (List<Map<String, Object>>) nestedLevel1.get("nestedField");
@@ -754,6 +769,22 @@ public void testBuildVectorOutput_withNestedList_Level2_successful() {
         assertNotNull(nestedObj.get(1).get("vectorField"));
     }
 
+    @SuppressWarnings("unchecked")
+    public void testBuildVectorOutput_withNestedListHasNotForEmbeddingField_Level2_successful() {
+        Map<String, Object> config = createNestedList2LevelConfiguration();
+        IngestDocument ingestDocument = create2LevelNestedListWithNotEmbeddingFieldIngestDocument();
+        TextEmbeddingProcessor textEmbeddingProcessor = createInstanceWithNestedMapConfiguration(config);
+        Map<String, Object> knnMap = textEmbeddingProcessor.buildMapWithTargetKeys(ingestDocument);
+        List<List<Float>> modelTensorList = createRandomOneDimensionalMockVector(1, 2, 0.0f, 1.0f);
+        textEmbeddingProcessor.buildNLPResult(knnMap, modelTensorList, ingestDocument.getSourceAndMetadata());
+        Map<String, Object> nestedLevel1 = (Map<String, Object>) ingestDocument.getSourceAndMetadata().get("nestedField");
+        List<Map<String, Object>> nestedObj = (List<Map<String, Object>>) nestedLevel1.get("nestedField");
+        assertFalse(nestedObj.get(0).containsKey("vectorField"));
+        assertTrue(nestedObj.get(0).containsKey("textFieldNotForEmbedding"));
+        assertTrue(nestedObj.get(1).containsKey("vectorField"));
+        assertNotNull(nestedObj.get(1).get("vectorField"));
+    }
+
     public void test_updateDocument_appendVectorFieldsToDocument_successful() {
         Map<String, Object> config = createPlainStringConfiguration();
         IngestDocument ingestDocument = createPlainIngestDocument();
@@ -1039,6 +1070,16 @@ private IngestDocument createNestedListIngestDocument() {
         return new IngestDocument(nestedList, new HashMap<>());
     }
 
+    private IngestDocument createNestedListWithNotEmbeddingFieldIngestDocument() {
+        HashMap<String, Object> nestedObj1 = new HashMap<>();
+        nestedObj1.put("textFieldNotForEmbedding", "This is a text field");
+        HashMap<String, Object> nestedObj2 = new HashMap<>();
+        nestedObj2.put("textField", "This is another text field");
+        HashMap<String, Object> nestedList = new HashMap<>();
+        nestedList.put("nestedField", Arrays.asList(nestedObj1, nestedObj2));
+        return new IngestDocument(nestedList, new HashMap<>());
+    }
+
     private IngestDocument create2LevelNestedListIngestDocument() {
         HashMap<String, Object> nestedObj1 = new HashMap<>();
         nestedObj1.put("textField", "This is a text field");
@@ -1050,4 +1091,16 @@ private IngestDocument create2LevelNestedListIngestDocument() {
         nestedList1.put("nestedField", nestedList);
         return new IngestDocument(nestedList1, new HashMap<>());
     }
+
+    private IngestDocument create2LevelNestedListWithNotEmbeddingFieldIngestDocument() {
+        HashMap<String, Object> nestedObj1 = new HashMap<>();
+        nestedObj1.put("textFieldNotForEmbedding", "This is a text field");
+        HashMap<String, Object> nestedObj2 = new HashMap<>();
+        nestedObj2.put("textField", "This is another text field");
+        HashMap<String, Object> nestedList = new HashMap<>();
+        nestedList.put("nestedField", Arrays.asList(nestedObj1, nestedObj2));
+        HashMap<String, Object> nestedList1 = new HashMap<>();
+        nestedList1.put("nestedField", nestedList);
+        return new IngestDocument(nestedList1, new HashMap<>());
+    }
 }

src/test/resources/processor/IndexMappings.json

@@ -90,6 +90,9 @@
           "text": {
             "type": "text"
           },
+          "text_not_for_embedding": {
+            "type": "text"
+          },
           "embedding": {
             "type": "knn_vector",
             "dimension": 768,

src/test/resources/processor/ingest_doc1.json

@@ -12,6 +12,9 @@
     "movie": null
   },
   "nested_passages": [
+    {
+      "text_not_for_embedding": "test"
+    },
     {
       "text": "hello"
     },

src/test/resources/processor/ingest_doc2.json

@@ -10,6 +10,9 @@
     "movie": null
   },
   "nested_passages": [
+    {
+      "text_not_for_embedding": "test"
+    },
     {
       "text": "apple"
     },

src/testFixtures/java/org/opensearch/neuralsearch/BaseNeuralSearchIT.java

@@ -203,6 +203,7 @@ protected void loadModel(final String modelId) throws Exception {
             isComplete = checkComplete(taskQueryResult);
             Thread.sleep(DEFAULT_TASK_RESULT_QUERY_INTERVAL_IN_MILLISECOND);
         }
+        assertTrue(isComplete);
     }
 
     /**