Fixes: community: fix LanceDB return no metadata (#27024)

- [ x ] Fix when lancedb return table without metadata column
- **Description:** Check the table schema, if not has metadata column,
init the Document with metadata argument equal to empty dict
    - **Issue:** https://github.com/langchain-ai/langchain/issues/27005

- [ x ] **Add tests and docs**

---------

Co-authored-by: ccurme <chester.curme@gmail.com>

libs/community/langchain_community/vectorstores/lancedb.py

@@ -151,12 +151,14 @@ class LanceDB(VectorStore):
             score_col = "_relevance_score"
         else:
             score_col = None
+        # Check if 'metadata' is in the columns
+        has_metadata = "metadata" in columns
 
         if score_col is None or not score:
             return [
                 Document(
                     page_content=results[self._text_key][idx].as_py(),
-                    metadata=results["metadata"][idx].as_py(),
+                    metadata=results["metadata"][idx].as_py() if has_metadata else {},
                 )
                 for idx in range(len(results))
             ]
@@ -165,7 +167,9 @@ class LanceDB(VectorStore):
                 (
                     Document(
                         page_content=results[self._text_key][idx].as_py(),
-                        metadata=results["metadata"][idx].as_py(),
+                        metadata=results["metadata"][idx].as_py()
+                        if has_metadata
+                        else {},
                     ),
                     results[score_col][idx].as_py(),
                 )

libs/community/tests/integration_tests/vectorstores/test_lancedb.py

@@ -114,3 +114,41 @@ def test_lancedb_all_searches() -> None:
     )
     assert len(result_3[0]) == 2  # type: ignore
     assert "text 1" in result_3[0][0].page_content  # type: ignore
+
+
+@pytest.mark.requires("lancedb")
+def test_lancedb_no_metadata() -> None:
+    lancedb = import_lancedb()
+    embeddings = FakeEmbeddings()
+    # Connect to a temporary LanceDB instance
+    db = lancedb.connect("/tmp/lancedb_no_metadata_test")
+    # Create data without the 'metadata' field
+    texts = ["text 1", "text 2", "item 3"]
+    data = []
+    for idx, text in enumerate(texts):
+        embedding = embeddings.embed_documents([text])[0]
+        data.append(
+            {
+                "vector": embedding,
+                "id": str(idx),
+                "text": text,
+                # Note: We're deliberately not including 'metadata' here
+            }
+        )
+    # Create the table without 'metadata' column
+    db.create_table("vectorstore_no_metadata", data=data)
+    # Initialize LanceDB with the existing connection and table name
+    store = LanceDB(
+        connection=db,
+        embedding=embeddings,
+        table_name="vectorstore_no_metadata",
+    )
+    # Perform a similarity search
+    result = store.similarity_search("text 1")
+    # Verify that the metadata in the Document objects is an empty dictionary
+    for doc in result:
+        assert (
+            doc.metadata == {}
+        ), "Expected empty metadata when 'metadata' column is missing"
+    # Clean up by deleting the table (optional)
+    db.drop_table("vectorstore_no_metadata")