Allow missing semantic text field in bulk updates (elastic#116478) (e…

…lastic#116500) This update enables bulk update operations to succeed even if the semantic text field is absent in the partial update. For the simple case where the field isn’t referenced by a copy_to operation from another source, the inference can be safely bypassed, allowing the update to proceed without errors.
mark-vieira · Nov 8, 2024 · 1a6e1ba · 1a6e1ba
1 parent ef4489a
commit 1a6e1ba
Show file tree

Hide file tree

Showing 5 changed files with 69 additions and 2 deletions.
diff --git a/docs/changelog/116478.yaml b/docs/changelog/116478.yaml
@@ -0,0 +1,5 @@
+pr: 116478
+summary: Semantic text simple partial update
+area: Search
+type: bug
+issues: []
diff --git a/...k/plugin/inference/src/main/java/org/elasticsearch/xpack/inference/InferenceFeatures.java b/...k/plugin/inference/src/main/java/org/elasticsearch/xpack/inference/InferenceFeatures.java
@@ -36,6 +36,9 @@ public Set<NodeFeature> getFeatures() {
 
     @Override
     public Set<NodeFeature> getTestFeatures() {
-        return Set.of(SemanticTextFieldMapper.SEMANTIC_TEXT_IN_OBJECT_FIELD_FIX);
+        return Set.of(
+            SemanticTextFieldMapper.SEMANTIC_TEXT_IN_OBJECT_FIELD_FIX,
+            SemanticTextFieldMapper.SEMANTIC_TEXT_SINGLE_FIELD_UPDATE_FIX
+        );
     }
 }
diff --git a/.../java/org/elasticsearch/xpack/inference/action/filter/ShardBulkInferenceActionFilter.java b/.../java/org/elasticsearch/xpack/inference/action/filter/ShardBulkInferenceActionFilter.java
@@ -446,7 +446,8 @@ private Map<String, List<FieldInferenceRequest>> createFieldInferenceRequests(Bu
                     String field = entry.getName();
                     String inferenceId = entry.getInferenceId();
                     var originalFieldValue = XContentMapValues.extractValue(field, docMap);
-                    if (originalFieldValue instanceof Map) {
+                    if (originalFieldValue instanceof Map || (originalFieldValue == null && entry.getSourceFields().length == 1)) {
+                        // Inference has already been computed, or there is no inference required.
                         continue;
                     }
                     int order = 0;

diff --git a/...rence/src/main/java/org/elasticsearch/xpack/inference/mapper/SemanticTextFieldMapper.java b/...rence/src/main/java/org/elasticsearch/xpack/inference/mapper/SemanticTextFieldMapper.java
@@ -89,6 +89,8 @@ public class SemanticTextFieldMapper extends FieldMapper implements InferenceFie
     public static final NodeFeature SEMANTIC_TEXT_DEFAULT_ELSER_2 = new NodeFeature("semantic_text.default_elser_2");
     public static final NodeFeature SEMANTIC_TEXT_IN_OBJECT_FIELD_FIX = new NodeFeature("semantic_text.in_object_field_fix");
 
+    public static final NodeFeature SEMANTIC_TEXT_SINGLE_FIELD_UPDATE_FIX = new NodeFeature("semantic_text.single_field_update_fix");
+
     public static final String CONTENT_TYPE = "semantic_text";
     public static final String DEFAULT_ELSER_2_INFERENCE_ID = DEFAULT_ELSER_ID;
 

diff --git a/...yamlRestTest/resources/rest-api-spec/test/inference/60_semantic_text_inference_update.yml b/...yamlRestTest/resources/rest-api-spec/test/inference/60_semantic_text_inference_update.yml
@@ -610,3 +610,59 @@ setup:
   - exists: _source.dense_field.inference.chunks.0.embeddings
   - match: { _source.dense_field.inference.chunks.0.text: "another updated inference test" }
   - match: { _source.non_inference_field: "updated non inference test" }
+
+---
+"Bypass inference on bulk update operation":
+  - requires:
+      cluster_features: semantic_text.single_field_update_fix
+      reason: Standalone semantic text fields are now optional in a bulk update operation
+
+  # Update as upsert
+  - do:
+      bulk:
+        body:
+          - '{"update": {"_index": "test-index", "_id": "doc_1"}}'
+          - '{"doc": { "sparse_field": "inference test", "dense_field": "another inference test", "non_inference_field": "non inference test" }, "doc_as_upsert": true}'
+
+  - match: { errors: false }
+  - match: { items.0.update.result: "created" }
+
+  - do:
+      bulk:
+        body:
+          - '{"update": {"_index": "test-index", "_id": "doc_1"}}'
+          - '{"doc": { "non_inference_field": "another value" }, "doc_as_upsert": true}'
+
+  - match: { errors: false }
+  - match: { items.0.update.result: "updated" }
+
+  - do:
+      get:
+        index: test-index
+        id: doc_1
+
+  - match: { _source.sparse_field.text: "inference test" }
+  - exists: _source.sparse_field.inference.chunks.0.embeddings
+  - match: { _source.sparse_field.inference.chunks.0.text: "inference test" }
+  - match: { _source.dense_field.text: "another inference test" }
+  - exists: _source.dense_field.inference.chunks.0.embeddings
+  - match: { _source.dense_field.inference.chunks.0.text: "another inference test" }
+  - match: { _source.non_inference_field: "another value" }
+
+  - do:
+      bulk:
+        body:
+          - '{"update": {"_index": "test-index", "_id": "doc_1"}}'
+          - '{"doc": { "sparse_field": null, "dense_field": null, "non_inference_field": "updated value" }, "doc_as_upsert": true}'
+
+  - match: { errors: false }
+  - match: { items.0.update.result: "updated" }
+
+  - do:
+      get:
+        index: test-index
+        id: doc_1
+
+  - match: { _source.sparse_field: null }
+  - match: { _source.dense_field: null }
+  - match: { _source.non_inference_field: "updated value" }