Deduplicate _field_names. (#26550)

This is a minor optimization that should save some utf8 conversions and indexing.

Author: jpountz

core/src/main/java/org/elasticsearch/index/mapper/FieldNamesFieldMapper.java

@@ -258,9 +258,19 @@ protected void parseCreateField(ParseContext context, List<IndexableField> field
             return;
         }
         for (ParseContext.Document document : context.docs()) {
-            final List<String> paths = new ArrayList<>();
+            final List<String> paths = new ArrayList<>(document.getFields().size());
+            String previousPath = ""; // used as a sentinel - field names can't be empty
             for (IndexableField field : document.getFields()) {
-                paths.add(field.name());
+                final String path = field.name();
+                if (path.equals(previousPath)) {
+                    // Sometimes mappers create multiple Lucene fields, eg. one for indexing,
+                    // one for doc values and one for storing. Deduplicating is not required
+                    // for correctness but this simple check helps save utf-8 conversions and
+                    // gives Lucene fewer values to deal with.
+                    continue;
+                }
+                paths.add(path);
+                previousPath = path;
             }
             for (String path : paths) {
                 for (String fieldName : extractFieldNames(path)) {

core/src/test/java/org/elasticsearch/index/mapper/FieldNamesFieldMapperTests.java

@@ -38,6 +38,7 @@
 import java.util.Collections;
 import java.util.List;
 import java.util.Map;
+import java.util.Set;
 import java.util.SortedSet;
 import java.util.TreeSet;
 import java.util.function.Supplier;
@@ -56,7 +57,7 @@ private static <T> SortedSet<T> set(T... values) {
         return new TreeSet<>(Arrays.asList(values));
     }
 
-    void assertFieldNames(SortedSet<String> expected, ParsedDocument doc) {
+    void assertFieldNames(Set<String> expected, ParsedDocument doc) {
         String[] got = doc.rootDoc().getValues("_field_names");
         assertEquals(expected, set(got));
     }
@@ -120,6 +121,26 @@ public void testExplicitEnabled() throws Exception {
         assertFieldNames(set("field", "field.keyword", "_id", "_version", "_seq_no", "_primary_term", "_source"), doc);
     }
 
+    public void testDedup() throws Exception {
+        String mapping = XContentFactory.jsonBuilder().startObject().startObject("type")
+            .startObject("_field_names").field("enabled", true).endObject()
+            .endObject().endObject().string();
+        DocumentMapper docMapper = createIndex("test").mapperService().documentMapperParser().parse("type", new CompressedXContent(mapping));
+        FieldNamesFieldMapper fieldNamesMapper = docMapper.metadataMapper(FieldNamesFieldMapper.class);
+        assertTrue(fieldNamesMapper.fieldType().isEnabled());
+
+        ParsedDocument doc = docMapper.parse(SourceToParse.source("test", "type", "1", XContentFactory.jsonBuilder()
+            .startObject()
+            .field("field", 3) // will create 2 lucene fields under the hood: index and doc values
+            .endObject()
+            .bytes(),
+            XContentType.JSON));
+
+        Set<String> fields = set("field", "_id", "_version", "_seq_no", "_primary_term", "_source");
+        assertFieldNames(fields, doc);
+        assertEquals(fields.size(), doc.rootDoc().getValues("_field_names").length);
+    }
+
     public void testDisabled() throws Exception {
         String mapping = XContentFactory.jsonBuilder().startObject().startObject("type")
             .startObject("_field_names").field("enabled", false).endObject()