add prelim fix

Author: sjrl

integrations/opensearch/src/haystack_integrations/document_stores/opensearch/document_store.py

@@ -327,6 +327,8 @@ async def _ensure_initialized_async(self) -> None:
 
     @staticmethod
     def _extract_nested_fields_from_mapping(mapping_properties: dict[str, Any]) -> set[str]:
+        if not isinstance(mapping_properties, dict):
+            return set()
         return {name for name, defn in mapping_properties.items() if defn.get("type") == "nested"}
 
     def _populate_nested_fields_from_mapping(self, mapping_properties: dict[str, Any]) -> None:

integrations/opensearch/tests/test_bm25_retriever.py

@@ -2,7 +2,7 @@
 #
 # SPDX-License-Identifier: Apache-2.0
 
-from unittest.mock import Mock, patch
+from unittest.mock import AsyncMock, MagicMock, Mock, patch
 
 import pytest
 from haystack.dataclasses import Document
@@ -538,3 +538,82 @@ async def test_bm25_retriever_async_runtime_document_store_switching(
 
     # Should get results from opensearch-1 (the initial store)
     assert len(results_1_again["documents"]) == 1
+
+
+@pytest.mark.asyncio
+async def test_run_async_in_async_pipeline():
+    """Test that OpenSearchBM25Retriever.run_async is called when used inside AsyncPipeline."""
+    from haystack import AsyncPipeline
+
+    mock_store = Mock(spec=OpenSearchDocumentStore)
+    mock_store._bm25_retrieval_async = AsyncMock(return_value=[Document(content="Test doc")])
+
+    retriever = OpenSearchBM25Retriever(document_store=mock_store)
+
+    pipeline = AsyncPipeline()
+    pipeline.add_component("retriever", retriever)
+
+    result = await pipeline.run_async({"retriever": {"query": "test query"}})
+
+    assert len(result["retriever"]["documents"]) == 1
+    assert result["retriever"]["documents"][0].content == "Test doc"
+    mock_store._bm25_retrieval_async.assert_called_once_with(
+        query="test query",
+        filters={},
+        fuzziness=0,
+        top_k=10,
+        scale_score=False,
+        all_terms_must_match=False,
+        custom_query=None,
+    )
+
+
+@pytest.mark.asyncio
+async def test_run_async_in_async_pipeline_bytes_error():
+    """
+    Reproduces: 'bytes' object has no attribute 'items' in AsyncPipeline.
+
+    In production, the error surfaces as:
+      Component name: 'BM25Retriever_...' Component type: 'OpenSearchBM25Retriever'
+      Error: 'bytes' object has no attribute 'items'
+
+    When run_async is called for the first time the document store lazily initialises
+    its async client via _ensure_initialized_async → _ensure_index_exists_async.
+    If the index already exists, the code retrieves its mapping and calls
+    _extract_nested_fields_from_mapping(properties).  That method does
+    ``properties.items()``.  If opensearch-py hands back bytes for that value
+    (observed with certain server/client version combinations) the call raises
+    AttributeError: 'bytes' object has no attribute 'items'.
+    """
+    from haystack import AsyncPipeline
+
+    with patch(
+        "haystack_integrations.document_stores.opensearch.document_store.AsyncOpenSearch"
+    ) as MockAsyncOS:
+        mock_client = MagicMock()
+        MockAsyncOS.return_value = mock_client
+
+        mock_client.indices.exists = AsyncMock(return_value=True)
+        mock_client.indices.get_mapping = AsyncMock(
+            return_value={
+                "test_index": {
+                    "mappings": {
+                        # bytes instead of dict — this is what triggers the bug
+                        "properties": b'{"content": {"type": "text"}}'
+                    }
+                }
+            }
+        )
+        mock_client.search = AsyncMock(return_value={"hits": {"hits": []}})
+
+        document_store = OpenSearchDocumentStore(hosts="fake_host", index="test_index")
+        retriever = OpenSearchBM25Retriever(document_store=document_store)
+
+        pipeline = AsyncPipeline()
+        pipeline.add_component("retriever", retriever)
+
+        # Currently raises (wrapped in PipelineRuntimeError):
+        #   AttributeError: 'bytes' object has no attribute 'items'
+        # Once the bug is fixed this should return empty documents without raising.
+        result = await pipeline.run_async({"retriever": {"query": "test query"}})
+        assert result["retriever"]["documents"] == []