debezium/dbz#2156 feat: implement unified metadata filtering API and …

tests/integration/test_chroma_pipeline.py

@@ -51,6 +51,24 @@ def test_chroma_as_retriever(chroma_adapter: ChromaAdapter) -> None:
     assert results[0].id == "test_id_retriever"
 
 
+def test_chroma_metadata_filtering(chroma_adapter: ChromaAdapter) -> None:
+    """Verify that ChromaAdapter retrieves only filtered metadata matching documents."""
+    doc_match = Document(
+        page_content="Electronics topic query", metadata={"category": "electronics", "tenant": "user1"}, id="doc_match"
+    )
+    doc_skip = Document(
+        page_content="Electronics topic query", metadata={"category": "clothing", "tenant": "user1"}, id="doc_skip"
+    )
+    chroma_adapter.upsert(doc_match)
+    chroma_adapter.upsert(doc_skip)
+
+    # Filter exactly category=electronics
+    retriever = chroma_adapter.as_retriever(metadata_filter={"category": "electronics"}, search_kwargs={"k": 2})
+    results = retriever.invoke("Electronics topic query")
+    assert len(results) == 1
+    assert results[0].id == "doc_match"
+
+
 def test_e2e_pipeline_chroma_hard_delete(
     chroma_adapter: ChromaAdapter,
     document_builder: DocumentBuilder,

tests/integration/test_milvus_adapter.py

@@ -102,3 +102,23 @@ def test_milvus_adapter_upsert_missing_id(milvus_adapter: MilvusAdapter) -> None
     with pytest.raises(ValueError) as exc_info:
         milvus_adapter.upsert(doc)
     assert "Document ID must be provided" in str(exc_info.value)
+
+
+def test_milvus_adapter_metadata_filtering(milvus_adapter: MilvusAdapter) -> None:
+    """Verify that MilvusAdapter filters documents matching metadata expression."""
+    doc_match = Document(
+        page_content="Matching product description", metadata={"category": "books", "in_stock": True}, id="doc_match"
+    )
+    doc_skip = Document(
+        page_content="Matching product description", metadata={"category": "clothing", "in_stock": True}, id="doc_skip"
+    )
+
+    milvus_adapter.upsert(doc_match)
+    milvus_adapter.upsert(doc_skip)
+
+    # Filter with category == books
+    retriever = milvus_adapter.as_retriever(metadata_filter={"category": "books"}, search_kwargs={"k": 2})
+    results = retriever.invoke("Matching product")
+    assert len(results) == 1
+    doc_id = results[0].id or results[0].metadata.get("pk")
+    assert doc_id == "doc_match"

tests/unit/test_pgvector_adapter.py

@@ -142,3 +142,16 @@ def test_pgvector_adapter_as_retriever(mock_pgvector: MagicMock) -> None:
 
     adapter.as_retriever(search_kwargs={"k": 3})
     adapter.store.as_retriever.assert_called_once_with(search_kwargs={"k": 3})
+
+
+def test_pgvector_adapter_metadata_filtering(mock_pgvector: MagicMock) -> None:
+    """Verify that PGVectorAdapter maps metadata_filter to search_kwargs filter dictionary."""
+    embeddings = FakeEmbeddings(size=128)
+    adapter = PGVectorAdapter(
+        connection_string="postgresql+psycopg://postgres:secret@localhost:5432/mydb",
+        collection_name="test_collection",
+        embeddings=embeddings,
+    )
+
+    adapter.as_retriever(metadata_filter={"category": "electronics"}, search_kwargs={"k": 3})
+    adapter.store.as_retriever.assert_called_once_with(search_kwargs={"k": 3, "filter": {"category": "electronics"}})