rebuild() of bulk operations (IndexOperation, CreateOperation, UpdateOperation) drops most fields #1353

Java API client version

9.5.4 (also current main)

Java version

25

Elasticsearch Version

9.5.4

Problem description

(full disclosure: analysis by copilot)

rebuild() should return a builder "initialized with field values of this instance". For the bulk operation classes it doesn't: most fields are lost.

Expected: op.rebuild().build() is equal to op.

Actual:

• IndexOperation and CreateOperation: the private Builder(IndexOperation<TDocument> instance) / Builder(CreateOperation<TDocument> instance) constructors are empty. Nothing is copied:

  • fields inherited from BulkOperationBase: id, index, routing, ifSeqNo, ifPrimaryTerm, version, versionType;
  • fields inherited from WriteOperation: pipeline, requireAlias, dynamicTemplates;
  • their own fields: document and tDocumentSerializer.

  Because document is required, rebuild().build() throws a MissingRequiredPropertyException. If you set a new document first (for example .document(binaryData)), it builds without errors, but without _id, _index, routing and so on.

• UpdateOperation: only requireAlias and retryOnConflict are copied. The inherited BulkOperationBase fields, action, binaryAction and the serializers are lost.

This looks like a code generator problem. Fields inherited from a generated base class are never copied, and neither are fields of the generic type parameter. Request classes without a base class, such as IndexRequest, copy their fields correctly.

Reproducer:

IndexOperation<Map<String, Object>> op = IndexOperation.of(b -> b
    .index("my-index")
    .id("1")
    .routing("r1")
    .ifSeqNo(5L)
    .ifPrimaryTerm(1L)
    .pipeline("my-pipeline")
    .document(Map.of("field", "value")));

// throws MissingRequiredPropertyException: Missing required property 'IndexOperation.document'
op.rebuild().build();

// builds without errors, but all values are lost
IndexOperation<?> copy = op.rebuild().document(Map.of("field", "value")).build();
System.out.println(copy.index());   // null, expected "my-index"
System.out.println(copy.id());      // null, expected "1"
System.out.println(copy.routing()); // null, expected "r1"
System.out.println(copy.pipeline()); // null, expected "my-pipeline"

CreateOperation behaves the same. For UpdateOperation, rebuild().build() does not throw, but only requireAlias and retryOnConflict are kept:

UpdateOperation<Object, Object> op = UpdateOperation.of(b -> b
    .index("my-index")
    .id("1")
    .routing("r1")
    .retryOnConflict(3)
    .action(a -> a.docAsUpsert(true)));

UpdateOperation<?, ?> copy = op.rebuild().build();
System.out.println(copy.index());           // null, expected "my-index"
System.out.println(copy.id());              // null, expected "1"
System.out.println(copy.routing());         // null, expected "r1"
System.out.println(copy.retryOnConflict()); // 3
System.out.println(copy.action());          // null

Context: I hit this in Spring Data Elasticsearch. Like the client's own BulkIngester, it needs to replace the document of an operation with a pre-serialized BinaryData so that it can measure its size. The obvious way is op.rebuild().document(binaryData).build(), but that silently drops _id and _index. So we copy all properties by hand, as IngesterOperation.copyIndexProperties() / copyBaseProperties() do.
see spring-projects/spring-data-elasticsearch#3357

Note: rebuild() also returns the raw type Builder instead of Builder<TDocument>, so callers get unchecked warnings.