[7.16] Better error message for long keys in flattened fields (#80433) (#80487)

When a keyed ´flattened` field that is composed of both the fields key, a separator
and the actual value, exceeds Lucenes term limit, the user currently gets a confusing
IAE that among other things only mentions the fields `_keyed` subfield as the source
of the offending long term. Since it might be both key and value that might trip this
we can check earlier and throw a nices IAE that reports both key and value lengths
and the prefix of the offending key.

Closes #78248

Author: Christoph Büscher

server/src/main/java/org/elasticsearch/index/mapper/flattened/FlattenedFieldParser.java

@@ -11,6 +11,7 @@
 import org.apache.lucene.document.Field;
 import org.apache.lucene.document.SortedSetDocValuesField;
 import org.apache.lucene.document.StringField;
+import org.apache.lucene.index.IndexWriter;
 import org.apache.lucene.index.IndexableField;
 import org.apache.lucene.util.BytesRef;
 import org.elasticsearch.common.xcontent.XContentParserUtils;
@@ -131,15 +132,33 @@ private void addField(ContentPath path, String currentName, String value, List<I
             );
         }
         String keyedValue = createKeyedValue(key, value);
-
+        BytesRef bytesKeyedValue = new BytesRef(keyedValue);
+        // check the keyed value doesn't exceed the IndexWriter.MAX_TERM_LENGTH limit enforced by Lucene at index time
+        // in that case we can already throw a more user friendly exception here which includes the offending fields key and value lengths
+        if (bytesKeyedValue.length > IndexWriter.MAX_TERM_LENGTH) {
+            String msg = "Flattened field ["
+                + rootFieldName
+                + "] contains one immense field"
+                + " whose keyed encoding is longer than the allowed max length of "
+                + IndexWriter.MAX_TERM_LENGTH
+                + " bytes. Key length: "
+                + key.length()
+                + ", value length: "
+                + value.length()
+                + " for key starting with ["
+                + key.substring(0, Math.min(key.length(), 50))
+                + "]";
+            throw new IllegalArgumentException(msg);
+        }
+        BytesRef bytesValue = new BytesRef(value);
         if (fieldType.isSearchable()) {
-            fields.add(new StringField(rootFieldName, new BytesRef(value), Field.Store.NO));
-            fields.add(new StringField(keyedFieldName, new BytesRef(keyedValue), Field.Store.NO));
+            fields.add(new StringField(rootFieldName, bytesValue, Field.Store.NO));
+            fields.add(new StringField(keyedFieldName, bytesKeyedValue, Field.Store.NO));
         }
 
         if (fieldType.hasDocValues()) {
-            fields.add(new SortedSetDocValuesField(rootFieldName, new BytesRef(value)));
-            fields.add(new SortedSetDocValuesField(keyedFieldName, new BytesRef(keyedValue)));
+            fields.add(new SortedSetDocValuesField(rootFieldName, bytesValue));
+            fields.add(new SortedSetDocValuesField(keyedFieldName, bytesKeyedValue));
         }
     }
 

server/src/test/java/org/elasticsearch/index/mapper/flattened/FlattenedFieldMapperTests.java

@@ -262,7 +262,8 @@ public void testEagerGlobalOrdinals() throws IOException {
 
     public void testIgnoreAbove() throws IOException {
         // First verify the default behavior when ignore_above is not set.
-        DocumentMapper mapper = createDocumentMapper(fieldMapping(this::minimalMapping));
+        MapperService mapperService = createMapperService(fieldMapping(this::minimalMapping));
+        DocumentMapper mapper = mapperService.documentMapper();
 
         ParsedDocument parsedDoc = mapper.parse(source(b -> {
             b.startArray("field");
@@ -280,15 +281,66 @@ public void testIgnoreAbove() throws IOException {
             b.field("ignore_above", 10);
         }));
 
-        ParsedDocument newParsedDoc = newMapper.parse(source(b -> {
+        parsedDoc = newMapper.parse(source(b -> {
             b.startArray("field");
             {
                 b.startObject().field("key", "a longer then usual value").endObject();
             }
             b.endArray();
         }));
-        IndexableField[] newFields = newParsedDoc.rootDoc().getFields("field");
+        IndexableField[] newFields = parsedDoc.rootDoc().getFields("field");
         assertEquals(0, newFields.length);
+
+        // using a key bigger than ignore_above should not prevent the field from being indexed, although we store key:value pairs
+        parsedDoc = newMapper.parse(source(b -> {
+            b.startArray("field");
+            {
+                b.startObject().field("key_longer_than_10chars", "value").endObject();
+            }
+            b.endArray();
+        }));
+        newFields = parsedDoc.rootDoc().getFields("field");
+        assertEquals(2, fields.length);
+    }
+
+    /**
+     * using a key:value pair above the Lucene term length limit would throw an error on indexing
+     * that we pre-empt with a nices exception
+     */
+    public void testImmenseKeyedTermException() throws IOException {
+        DocumentMapper newMapper = createDocumentMapper(fieldMapping(b -> { b.field("type", "flattened"); }));
+
+        String longKey = new String(new char[32800]).replace('\0', 'x');
+        MapperParsingException ex = expectThrows(MapperParsingException.class, () -> newMapper.parse(source(b -> {
+            b.startArray("field");
+            {
+                b.startObject().field(longKey, "value").endObject();
+            }
+            b.endArray();
+        })));
+        assertEquals(
+            "Flattened field [field] contains one immense field whose keyed encoding is longer "
+                + "than the allowed max length of 32766 bytes. Key length: "
+                + longKey.length()
+                + ", value length: 5 for key starting with [xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx]",
+            ex.getCause().getMessage()
+        );
+
+        String value = new String(new char[32800]).replace('\0', 'x');
+        ex = expectThrows(MapperParsingException.class, () -> newMapper.parse(source(b -> {
+            b.startArray("field");
+            {
+                b.startObject().field("key", value).endObject();
+            }
+            b.endArray();
+        })));
+        assertEquals(
+            "Flattened field [field] contains one immense field whose keyed encoding is longer "
+                + "than the allowed max length of 32766 bytes. Key length: 3, value length: "
+                + value.length()
+                + " for key starting with [key]",
+            ex.getCause().getMessage()
+        );
     }
 
     public void testNullValues() throws Exception {