test: unit test cases for rag.cleaner, rag.data_post_processor and rag.datasource (#32521)

api/core/rag/datasource/vdb/huawei/huawei_cloud_vector.py

@@ -124,13 +124,13 @@ class HuaweiCloudVector(BaseVector):
                 )
             )
 
+        score_threshold = float(kwargs.get("score_threshold") or 0.0)
         docs = []
         for doc, score in docs_and_scores:
-            score_threshold = float(kwargs.get("score_threshold") or 0.0)
             if score >= score_threshold:
                 if doc.metadata is not None:
                     doc.metadata["score"] = score
-            docs.append(doc)
+                    docs.append(doc)
 
         return docs
 

api/tests/unit_tests/core/rag/cleaner/test_clean_processor.py

@@ -211,3 +211,16 @@ class TestCleanProcessor:
         text = "[Text with (parens) and symbols](https://example.com)"
         expected = "[Text with (parens) and symbols](https://example.com)"
         assert CleanProcessor.clean(text, process_rule) == expected
+
+    def test_clean_remove_urls_emails_preserves_markdown_image_links(self):
+        """Remove plain URLs and emails while preserving markdown image links."""
+        process_rule = {"rules": {"pre_processing_rules": [{"id": "remove_urls_emails", "enabled": True}]}}
+        text = "Email test@example.com and remove https://remove.com but keep ![diagram](https://example.com/image.png)"
+        result = CleanProcessor.clean(text, process_rule)
+
+        assert result == "Email  and remove  but keep ![diagram](https://example.com/image.png)"
+
+    def test_filter_string_returns_input_text(self):
+        """Test filter_string passthrough behavior."""
+        processor = CleanProcessor()
+        assert processor.filter_string("raw text") == "raw text"

api/tests/unit_tests/core/rag/data_post_processor/test_data_post_processor.py

@@ -0,0 +1,249 @@
+from unittest.mock import MagicMock, patch
+
+import pytest
+
+from core.rag.data_post_processor.data_post_processor import DataPostProcessor
+from core.rag.data_post_processor.reorder import ReorderRunner
+from core.rag.index_processor.constant.query_type import QueryType
+from core.rag.models.document import Document
+from core.rag.rerank.rerank_type import RerankMode
+from dify_graph.model_runtime.entities.model_entities import ModelType
+from dify_graph.model_runtime.errors.invoke import InvokeAuthorizationError
+
+
+def _doc(content: str) -> Document:
+    return Document(page_content=content)
+
+
+class TestDataPostProcessor:
+    def test_init_sets_rerank_and_reorder_runners(self):
+        rerank_runner = object()
+        reorder_runner = object()
+
+        with patch.object(DataPostProcessor, "_get_rerank_runner", return_value=rerank_runner) as rerank_mock:
+            with patch.object(DataPostProcessor, "_get_reorder_runner", return_value=reorder_runner) as reorder_mock:
+                processor = DataPostProcessor(
+                    tenant_id="tenant-1",
+                    reranking_mode=RerankMode.WEIGHTED_SCORE,
+                    reranking_model={"config": "value"},
+                    weights={"weight": "value"},
+                    reorder_enabled=True,
+                )
+
+        assert processor.rerank_runner is rerank_runner
+        assert processor.reorder_runner is reorder_runner
+        rerank_mock.assert_called_once_with(
+            RerankMode.WEIGHTED_SCORE,
+            "tenant-1",
+            {"config": "value"},
+            {"weight": "value"},
+        )
+        reorder_mock.assert_called_once_with(True)
+
+    def test_invoke_applies_rerank_then_reorder(self):
+        original_documents = [_doc("doc-a")]
+        reranked_documents = [_doc("doc-b")]
+        reordered_documents = [_doc("doc-c")]
+
+        processor = DataPostProcessor.__new__(DataPostProcessor)
+        processor.rerank_runner = MagicMock()
+        processor.rerank_runner.run.return_value = reranked_documents
+        processor.reorder_runner = MagicMock()
+        processor.reorder_runner.run.return_value = reordered_documents
+
+        result = processor.invoke(
+            query="how to test",
+            documents=original_documents,
+            score_threshold=0.3,
+            top_n=2,
+            user="user-1",
+            query_type=QueryType.IMAGE_QUERY,
+        )
+
+        processor.rerank_runner.run.assert_called_once_with(
+            "how to test",
+            original_documents,
+            0.3,
+            2,
+            "user-1",
+            QueryType.IMAGE_QUERY,
+        )
+        processor.reorder_runner.run.assert_called_once_with(reranked_documents)
+        assert result == reordered_documents
+
+    def test_invoke_returns_original_documents_when_no_runner_is_configured(self):
+        documents = [_doc("doc-a"), _doc("doc-b")]
+
+        processor = DataPostProcessor.__new__(DataPostProcessor)
+        processor.rerank_runner = None
+        processor.reorder_runner = None
+
+        assert processor.invoke(query="query", documents=documents) == documents
+
+    def test_get_rerank_runner_for_weighted_score(self):
+        weights_config = {
+            "vector_setting": {
+                "vector_weight": 0.7,
+                "embedding_provider_name": "provider-x",
+                "embedding_model_name": "embedding-y",
+            },
+            "keyword_setting": {"keyword_weight": 0.3},
+        }
+        expected_runner = object()
+        processor = DataPostProcessor.__new__(DataPostProcessor)
+
+        with patch(
+            "core.rag.data_post_processor.data_post_processor.RerankRunnerFactory.create_rerank_runner",
+            return_value=expected_runner,
+        ) as factory_mock:
+            result = processor._get_rerank_runner(
+                reranking_mode=RerankMode.WEIGHTED_SCORE,
+                tenant_id="tenant-1",
+                reranking_model=None,
+                weights=weights_config,
+            )
+
+        assert result is expected_runner
+        kwargs = factory_mock.call_args.kwargs
+        assert kwargs["runner_type"] == RerankMode.WEIGHTED_SCORE
+        assert kwargs["tenant_id"] == "tenant-1"
+        assert kwargs["weights"].vector_setting.vector_weight == 0.7
+        assert kwargs["weights"].vector_setting.embedding_provider_name == "provider-x"
+        assert kwargs["weights"].vector_setting.embedding_model_name == "embedding-y"
+        assert kwargs["weights"].keyword_setting.keyword_weight == 0.3
+
+    def test_get_rerank_runner_for_reranking_model_returns_none_without_model_instance(self):
+        processor = DataPostProcessor.__new__(DataPostProcessor)
+        reranking_model = {
+            "reranking_provider_name": "provider-x",
+            "reranking_model_name": "model-y",
+        }
+
+        with patch.object(DataPostProcessor, "_get_rerank_model_instance", return_value=None) as model_mock:
+            with patch(
+                "core.rag.data_post_processor.data_post_processor.RerankRunnerFactory.create_rerank_runner"
+            ) as factory_mock:
+                result = processor._get_rerank_runner(
+                    reranking_mode=RerankMode.RERANKING_MODEL,
+                    tenant_id="tenant-1",
+                    reranking_model=reranking_model,
+                    weights=None,
+                )
+
+        assert result is None
+        model_mock.assert_called_once_with("tenant-1", reranking_model)
+        factory_mock.assert_not_called()
+
+    def test_get_rerank_runner_for_reranking_model_creates_runner_with_model_instance(self):
+        processor = DataPostProcessor.__new__(DataPostProcessor)
+        model_instance = object()
+        expected_runner = object()
+
+        with patch.object(DataPostProcessor, "_get_rerank_model_instance", return_value=model_instance):
+            with patch(
+                "core.rag.data_post_processor.data_post_processor.RerankRunnerFactory.create_rerank_runner",
+                return_value=expected_runner,
+            ) as factory_mock:
+                result = processor._get_rerank_runner(
+                    reranking_mode=RerankMode.RERANKING_MODEL,
+                    tenant_id="tenant-1",
+                    reranking_model={
+                        "reranking_provider_name": "provider-x",
+                        "reranking_model_name": "model-y",
+                    },
+                    weights=None,
+                )
+
+        assert result is expected_runner
+        factory_mock.assert_called_once_with(
+            runner_type=RerankMode.RERANKING_MODEL,
+            rerank_model_instance=model_instance,
+        )
+
+    def test_get_rerank_runner_returns_none_for_unsupported_mode(self):
+        processor = DataPostProcessor.__new__(DataPostProcessor)
+
+        assert processor._get_rerank_runner("unsupported", "tenant-1", None, None) is None
+        assert processor._get_rerank_runner(RerankMode.WEIGHTED_SCORE, "tenant-1", None, None) is None
+
+    def test_get_reorder_runner_by_flag(self):
+        processor = DataPostProcessor.__new__(DataPostProcessor)
+
+        assert isinstance(processor._get_reorder_runner(True), ReorderRunner)
+        assert processor._get_reorder_runner(False) is None
+
+    def test_get_rerank_model_instance_returns_none_when_config_is_missing(self):
+        processor = DataPostProcessor.__new__(DataPostProcessor)
+        assert processor._get_rerank_model_instance("tenant-1", None) is None
+
+    def test_get_rerank_model_instance_raises_key_error_for_incomplete_config(self):
+        processor = DataPostProcessor.__new__(DataPostProcessor)
+
+        with patch("core.rag.data_post_processor.data_post_processor.ModelManager") as manager_cls:
+            manager_instance = manager_cls.return_value
+            with pytest.raises(KeyError, match="reranking_model_name"):
+                processor._get_rerank_model_instance(
+                    tenant_id="tenant-1",
+                    reranking_model={"reranking_provider_name": "provider-x"},
+                )
+
+        manager_instance.get_model_instance.assert_not_called()
+
+    def test_get_rerank_model_instance_success(self):
+        processor = DataPostProcessor.__new__(DataPostProcessor)
+        model_instance = object()
+
+        with patch("core.rag.data_post_processor.data_post_processor.ModelManager") as manager_cls:
+            manager_instance = manager_cls.return_value
+            manager_instance.get_model_instance.return_value = model_instance
+
+            result = processor._get_rerank_model_instance(
+                tenant_id="tenant-1",
+                reranking_model={
+                    "reranking_provider_name": "provider-x",
+                    "reranking_model_name": "reranker-1",
+                },
+            )
+
+        assert result is model_instance
+        manager_instance.get_model_instance.assert_called_once_with(
+            tenant_id="tenant-1",
+            provider="provider-x",
+            model_type=ModelType.RERANK,
+            model="reranker-1",
+        )
+
+    def test_get_rerank_model_instance_handles_authorization_error(self):
+        processor = DataPostProcessor.__new__(DataPostProcessor)
+
+        with patch("core.rag.data_post_processor.data_post_processor.ModelManager") as manager_cls:
+            manager_instance = manager_cls.return_value
+            manager_instance.get_model_instance.side_effect = InvokeAuthorizationError("not authorized")
+
+            result = processor._get_rerank_model_instance(
+                tenant_id="tenant-1",
+                reranking_model={
+                    "reranking_provider_name": "provider-x",
+                    "reranking_model_name": "reranker-1",
+                },
+            )
+
+        assert result is None
+
+
+class TestReorderRunner:
+    def test_run_reorders_even_sized_document_list(self):
+        documents = [_doc("0"), _doc("1"), _doc("2"), _doc("3"), _doc("4"), _doc("5")]
+
+        reordered = ReorderRunner().run(documents)
+
+        assert [document.page_content for document in reordered] == ["0", "2", "4", "5", "3", "1"]
+
+    def test_run_handles_odd_sized_and_empty_document_lists(self):
+        odd_documents = [_doc("0"), _doc("1"), _doc("2"), _doc("3"), _doc("4")]
+        runner = ReorderRunner()
+
+        odd_reordered = runner.run(odd_documents)
+
+        assert [document.page_content for document in odd_reordered] == ["0", "2", "4", "3", "1"]
+        assert runner.run([]) == []

api/tests/unit_tests/core/rag/datasource/keyword/jieba/test_jieba.py

@@ -0,0 +1,414 @@
+import json
+from types import SimpleNamespace
+from unittest.mock import MagicMock
+
+import pytest
+
+import core.rag.datasource.keyword.jieba.jieba as jieba_module
+from core.rag.datasource.keyword.jieba.jieba import Jieba, dumps_with_sets, set_orjson_default
+from core.rag.models.document import Document
+
+
+class _DummyLock:
+    def __enter__(self):
+        return self
+
+    def __exit__(self, exc_type, exc, tb):
+        return False
+
+
+class _Field:
+    def __init__(self, name: str):
+        self._name = name
+
+    def __eq__(self, other):
+        return ("eq", self._name, other)
+
+    def in_(self, values):
+        return ("in", self._name, tuple(values))
+
+
+class _FakeQuery:
+    def __init__(self):
+        self.where_calls: list[tuple] = []
+
+    def where(self, *conditions):
+        self.where_calls.append(conditions)
+        return self
+
+
+class _FakeExecuteResult:
+    def __init__(self, segments: list[SimpleNamespace]):
+        self._segments = segments
+
+    def scalars(self):
+        return self
+
+    def all(self):
+        return self._segments
+
+
+class _FakeSelect:
+    def __init__(self):
+        self.where_conditions: tuple | None = None
+
+    def where(self, *conditions):
+        self.where_conditions = conditions
+        return self
+
+
+def _dataset_keyword_table(data_source_type: str = "database", keyword_table_dict: dict | None = None):
+    return SimpleNamespace(
+        data_source_type=data_source_type,
+        keyword_table_dict=keyword_table_dict,
+        keyword_table="",
+    )
+
+
+def _dataset(dataset_keyword_table=None, keyword_number=None):
+    return SimpleNamespace(
+        id="dataset-1",
+        tenant_id="tenant-1",
+        keyword_number=keyword_number,
+        dataset_keyword_table=dataset_keyword_table,
+    )
+
+
+@pytest.fixture
+def patched_runtime(monkeypatch):
+    session = MagicMock()
+    db = SimpleNamespace(session=session)
+    storage = MagicMock()
+    lock = MagicMock(return_value=_DummyLock())
+    redis_client = SimpleNamespace(lock=lock)
+
+    monkeypatch.setattr(jieba_module, "db", db)
+    monkeypatch.setattr(jieba_module, "storage", storage)
+    monkeypatch.setattr(jieba_module, "redis_client", redis_client)
+
+    return SimpleNamespace(session=session, storage=storage, lock=lock)
+
+
+def test_create_indexes_documents_and_returns_self(monkeypatch, patched_runtime):
+    dataset = _dataset(_dataset_keyword_table(), keyword_number=2)
+    keyword = Jieba(dataset)
+    handler = MagicMock()
+    handler.extract_keywords.return_value = {"kw1", "kw2"}
+
+    monkeypatch.setattr(jieba_module, "JiebaKeywordTableHandler", lambda: handler)
+    monkeypatch.setattr(keyword, "_get_dataset_keyword_table", MagicMock(return_value={}))
+    monkeypatch.setattr(keyword, "_update_segment_keywords", MagicMock())
+    monkeypatch.setattr(keyword, "_save_dataset_keyword_table", MagicMock())
+
+    result = keyword.create(
+        [
+            Document(page_content="alpha", metadata={"doc_id": "node-1"}),
+            SimpleNamespace(page_content="ignored", metadata=None),
+        ]
+    )
+
+    assert result is keyword
+    keyword._update_segment_keywords.assert_called_once()
+    call_args = keyword._update_segment_keywords.call_args.args
+    assert call_args[0] == "dataset-1"
+    assert call_args[1] == "node-1"
+    assert set(call_args[2]) == {"kw1", "kw2"}
+    saved_table = keyword._save_dataset_keyword_table.call_args.args[0]
+    assert saved_table["kw1"] == {"node-1"}
+    assert saved_table["kw2"] == {"node-1"}
+    patched_runtime.lock.assert_called_once_with("keyword_indexing_lock_dataset-1", timeout=600)
+
+
+def test_add_texts_supports_keywords_list_and_extract_fallback(monkeypatch, patched_runtime):
+    keyword = Jieba(_dataset(_dataset_keyword_table(), keyword_number=3))
+    handler = MagicMock()
+    handler.extract_keywords.return_value = {"auto"}
+
+    monkeypatch.setattr(jieba_module, "JiebaKeywordTableHandler", lambda: handler)
+    monkeypatch.setattr(keyword, "_get_dataset_keyword_table", MagicMock(return_value={}))
+    monkeypatch.setattr(keyword, "_update_segment_keywords", MagicMock())
+    monkeypatch.setattr(keyword, "_save_dataset_keyword_table", MagicMock())
+
+    texts = [
+        Document(page_content="extract-this", metadata={"doc_id": "node-1"}),
+        Document(page_content="use-manual", metadata={"doc_id": "node-2"}),
+    ]
+    keyword.add_texts(texts, keywords_list=[[], ["manual"]])
+
+    assert keyword._update_segment_keywords.call_count == 2
+    first_call = keyword._update_segment_keywords.call_args_list[0].args
+    second_call = keyword._update_segment_keywords.call_args_list[1].args
+    assert set(first_call[2]) == {"auto"}
+    assert second_call[2] == ["manual"]
+    keyword._save_dataset_keyword_table.assert_called_once()
+
+
+def test_add_texts_without_keywords_list_always_uses_extractor(monkeypatch, patched_runtime):
+    keyword = Jieba(_dataset(_dataset_keyword_table(), keyword_number=1))
+    handler = MagicMock()
+    handler.extract_keywords.return_value = {"from-extractor"}
+
+    monkeypatch.setattr(jieba_module, "JiebaKeywordTableHandler", lambda: handler)
+    monkeypatch.setattr(keyword, "_get_dataset_keyword_table", MagicMock(return_value={}))
+    monkeypatch.setattr(keyword, "_update_segment_keywords", MagicMock())
+    monkeypatch.setattr(keyword, "_save_dataset_keyword_table", MagicMock())
+
+    keyword.add_texts([Document(page_content="content", metadata={"doc_id": "node-1"})])
+
+    handler.extract_keywords.assert_called_once_with("content", 1)
+    assert set(keyword._update_segment_keywords.call_args.args[2]) == {"from-extractor"}
+
+
+def test_text_exists_handles_missing_and_existing_keyword_table(monkeypatch):
+    keyword = Jieba(_dataset(_dataset_keyword_table()))
+
+    monkeypatch.setattr(keyword, "_get_dataset_keyword_table", MagicMock(return_value=None))
+    assert keyword.text_exists("node-1") is False
+
+    monkeypatch.setattr(keyword, "_get_dataset_keyword_table", MagicMock(return_value={"k": {"node-1", "node-2"}}))
+    assert keyword.text_exists("node-2") is True
+    assert keyword.text_exists("node-x") is False
+
+
+def test_delete_by_ids_updates_table_when_present(monkeypatch, patched_runtime):
+    keyword = Jieba(_dataset(_dataset_keyword_table()))
+    monkeypatch.setattr(keyword, "_get_dataset_keyword_table", MagicMock(return_value={"k": {"node-1", "node-2"}}))
+    monkeypatch.setattr(keyword, "_delete_ids_from_keyword_table", MagicMock(return_value={"k": {"node-2"}}))
+    monkeypatch.setattr(keyword, "_save_dataset_keyword_table", MagicMock())
+
+    keyword.delete_by_ids(["node-1"])
+
+    keyword._delete_ids_from_keyword_table.assert_called_once_with({"k": {"node-1", "node-2"}}, ["node-1"])
+    keyword._save_dataset_keyword_table.assert_called_once_with({"k": {"node-2"}})
+
+
+def test_delete_by_ids_saves_none_when_keyword_table_is_missing(monkeypatch, patched_runtime):
+    keyword = Jieba(_dataset(_dataset_keyword_table()))
+    monkeypatch.setattr(keyword, "_get_dataset_keyword_table", MagicMock(return_value=None))
+    monkeypatch.setattr(keyword, "_delete_ids_from_keyword_table", MagicMock())
+    monkeypatch.setattr(keyword, "_save_dataset_keyword_table", MagicMock())
+
+    keyword.delete_by_ids(["node-1"])
+
+    keyword._delete_ids_from_keyword_table.assert_not_called()
+    keyword._save_dataset_keyword_table.assert_called_once_with(None)
+
+
+def test_search_returns_documents_in_rank_order_and_applies_filter(monkeypatch, patched_runtime):
+    class _FakeDocumentSegment:
+        dataset_id = _Field("dataset_id")
+        index_node_id = _Field("index_node_id")
+        document_id = _Field("document_id")
+
+    keyword = Jieba(_dataset(_dataset_keyword_table()))
+    query_stmt = _FakeQuery()
+    patched_runtime.session.query.return_value = query_stmt
+    patched_runtime.session.execute.return_value = _FakeExecuteResult(
+        [
+            SimpleNamespace(
+                index_node_id="node-2",
+                content="segment-content",
+                index_node_hash="hash-2",
+                document_id="doc-2",
+                dataset_id="dataset-1",
+            )
+        ]
+    )
+
+    monkeypatch.setattr(jieba_module, "DocumentSegment", _FakeDocumentSegment)
+    monkeypatch.setattr(keyword, "_get_dataset_keyword_table", MagicMock(return_value={"k": {"node-1", "node-2"}}))
+    monkeypatch.setattr(keyword, "_retrieve_ids_by_query", MagicMock(return_value=["node-1", "node-2"]))
+
+    documents = keyword.search("query", top_k=2, document_ids_filter=["doc-2"])
+
+    assert len(query_stmt.where_calls) == 2
+    assert len(documents) == 1
+    assert documents[0].page_content == "segment-content"
+    assert documents[0].metadata["doc_id"] == "node-2"
+    assert documents[0].metadata["doc_hash"] == "hash-2"
+
+
+def test_delete_removes_keyword_table_and_optional_file(monkeypatch, patched_runtime):
+    db_keyword = _dataset_keyword_table(data_source_type="database")
+    file_keyword = _dataset_keyword_table(data_source_type="object_storage")
+
+    keyword_db = Jieba(_dataset(db_keyword))
+    keyword_db.delete()
+    patched_runtime.storage.delete.assert_not_called()
+
+    keyword_file = Jieba(_dataset(file_keyword))
+    keyword_file.delete()
+
+    patched_runtime.storage.delete.assert_called_once_with("keyword_files/tenant-1/dataset-1.txt")
+    assert patched_runtime.session.delete.call_count == 2
+    assert patched_runtime.session.commit.call_count == 2
+
+
+def test_save_dataset_keyword_table_to_database(monkeypatch, patched_runtime):
+    dataset_keyword_table = _dataset_keyword_table(data_source_type="database")
+    keyword = Jieba(_dataset(dataset_keyword_table))
+
+    keyword._save_dataset_keyword_table({"kw": {"node-1"}})
+
+    assert '"__type__":"keyword_table"' in dataset_keyword_table.keyword_table
+    assert '"index_id":"dataset-1"' in dataset_keyword_table.keyword_table
+    patched_runtime.session.commit.assert_called_once()
+
+
+def test_save_dataset_keyword_table_to_file_storage(monkeypatch, patched_runtime):
+    dataset_keyword_table = _dataset_keyword_table(data_source_type="file")
+    keyword = Jieba(_dataset(dataset_keyword_table))
+    patched_runtime.storage.exists.return_value = True
+
+    keyword._save_dataset_keyword_table({"kw": {"node-1"}})
+
+    patched_runtime.storage.delete.assert_called_once_with("keyword_files/tenant-1/dataset-1.txt")
+    patched_runtime.storage.save.assert_called_once()
+    save_args = patched_runtime.storage.save.call_args.args
+    assert save_args[0] == "keyword_files/tenant-1/dataset-1.txt"
+    assert isinstance(save_args[1], bytes)
+
+
+def test_get_dataset_keyword_table_returns_existing_table_data(monkeypatch, patched_runtime):
+    existing = _dataset_keyword_table(
+        keyword_table_dict={"__type__": "keyword_table", "__data__": {"table": {"kw": ["node-1"]}}}
+    )
+    keyword = Jieba(_dataset(existing))
+    assert keyword._get_dataset_keyword_table() == {"kw": ["node-1"]}
+
+    missing_payload = _dataset_keyword_table(keyword_table_dict=None)
+    keyword_with_missing_payload = Jieba(_dataset(missing_payload))
+    assert keyword_with_missing_payload._get_dataset_keyword_table() == {}
+
+
+def test_get_dataset_keyword_table_creates_table_when_missing(monkeypatch, patched_runtime):
+    created_tables: list[SimpleNamespace] = []
+
+    def _fake_dataset_keyword_table(**kwargs):
+        kwargs.setdefault("keyword_table", "")
+        kwargs.setdefault("keyword_table_dict", None)
+        table = SimpleNamespace(**kwargs)
+        created_tables.append(table)
+        return table
+
+    keyword = Jieba(_dataset(dataset_keyword_table=None))
+    monkeypatch.setattr(jieba_module, "DatasetKeywordTable", _fake_dataset_keyword_table)
+    monkeypatch.setattr(jieba_module.dify_config, "KEYWORD_DATA_SOURCE_TYPE", "database")
+
+    result = keyword._get_dataset_keyword_table()
+
+    assert result == {}
+    assert len(created_tables) == 1
+    assert created_tables[0].dataset_id == "dataset-1"
+    assert created_tables[0].data_source_type == "database"
+    assert '"index_id":"dataset-1"' in created_tables[0].keyword_table
+    patched_runtime.session.add.assert_called_once_with(created_tables[0])
+    patched_runtime.session.commit.assert_called_once()
+
+
+def test_add_and_delete_ids_from_keyword_table_helpers():
+    keyword = Jieba(_dataset(_dataset_keyword_table()))
+    keyword_table = {"kw1": {"node-1"}, "kw2": {"node-1", "node-2"}}
+
+    updated = keyword._add_text_to_keyword_table(keyword_table, "node-3", ["kw1", "kw3"])
+    assert updated["kw1"] == {"node-1", "node-3"}
+    assert updated["kw3"] == {"node-3"}
+
+    deleted = keyword._delete_ids_from_keyword_table(updated, ["node-1", "node-3"])
+    assert "kw3" not in deleted
+    assert "kw1" not in deleted
+    assert deleted["kw2"] == {"node-2"}
+
+
+def test_retrieve_ids_by_query_ranks_by_keyword_frequency(monkeypatch):
+    keyword = Jieba(_dataset(_dataset_keyword_table()))
+    handler = MagicMock()
+    handler.extract_keywords.return_value = ["kw-a", "kw-b"]
+    monkeypatch.setattr(jieba_module, "JiebaKeywordTableHandler", lambda: handler)
+
+    ranked_ids = keyword._retrieve_ids_by_query(
+        {"kw-a": {"node-1", "node-2"}, "kw-b": {"node-2"}, "kw-c": {"node-3"}},
+        "query",
+        k=1,
+    )
+
+    assert ranked_ids == ["node-2"]
+
+
+def test_update_segment_keywords_updates_when_segment_exists(monkeypatch, patched_runtime):
+    class _FakeDocumentSegment:
+        dataset_id = _Field("dataset_id")
+        index_node_id = _Field("index_node_id")
+
+    monkeypatch.setattr(jieba_module, "DocumentSegment", _FakeDocumentSegment)
+    monkeypatch.setattr(jieba_module, "select", lambda *_: _FakeSelect())
+
+    keyword = Jieba(_dataset(_dataset_keyword_table()))
+    segment = SimpleNamespace(keywords=[])
+    patched_runtime.session.scalar.return_value = segment
+
+    keyword._update_segment_keywords("dataset-1", "node-1", ["kw1", "kw2"])
+
+    assert segment.keywords == ["kw1", "kw2"]
+    patched_runtime.session.add.assert_called_once_with(segment)
+    patched_runtime.session.commit.assert_called_once()
+
+    patched_runtime.session.reset_mock()
+    patched_runtime.session.scalar.return_value = None
+
+    keyword._update_segment_keywords("dataset-1", "node-missing", ["kw3"])
+
+    patched_runtime.session.add.assert_not_called()
+    patched_runtime.session.commit.assert_not_called()
+
+
+def test_create_segment_keywords_and_update_segment_keywords_index(monkeypatch):
+    keyword = Jieba(_dataset(_dataset_keyword_table()))
+    monkeypatch.setattr(keyword, "_get_dataset_keyword_table", MagicMock(return_value={}))
+    monkeypatch.setattr(keyword, "_update_segment_keywords", MagicMock())
+    monkeypatch.setattr(keyword, "_save_dataset_keyword_table", MagicMock())
+
+    keyword.create_segment_keywords("node-1", ["kw"])
+    keyword._update_segment_keywords.assert_called_once_with("dataset-1", "node-1", ["kw"])
+    keyword._save_dataset_keyword_table.assert_called_once()
+
+    keyword._save_dataset_keyword_table.reset_mock()
+    keyword.update_segment_keywords_index("node-2", ["kw2"])
+    keyword._save_dataset_keyword_table.assert_called_once()
+
+
+def test_multi_create_segment_keywords_uses_provided_and_extracted_keywords(monkeypatch):
+    keyword = Jieba(_dataset(_dataset_keyword_table(), keyword_number=2))
+    handler = MagicMock()
+    handler.extract_keywords.return_value = {"auto"}
+    monkeypatch.setattr(jieba_module, "JiebaKeywordTableHandler", lambda: handler)
+    monkeypatch.setattr(keyword, "_get_dataset_keyword_table", MagicMock(return_value={}))
+    monkeypatch.setattr(keyword, "_save_dataset_keyword_table", MagicMock())
+
+    first_segment = SimpleNamespace(index_node_id="node-1", content="first content", keywords=None)
+    second_segment = SimpleNamespace(index_node_id="node-2", content="second content", keywords=None)
+
+    keyword.multi_create_segment_keywords(
+        [
+            {"segment": first_segment, "keywords": ["manual"]},
+            {"segment": second_segment, "keywords": []},
+        ]
+    )
+
+    assert first_segment.keywords == ["manual"]
+    assert second_segment.keywords == ["auto"]
+    saved_table = keyword._save_dataset_keyword_table.call_args.args[0]
+    assert saved_table["manual"] == {"node-1"}
+    assert saved_table["auto"] == {"node-2"}
+
+
+def test_set_orjson_default_and_dumps_with_sets():
+    assert set(set_orjson_default({"a", "b"})) == {"a", "b"}
+
+    with pytest.raises(TypeError, match="is not JSON serializable"):
+        set_orjson_default(("not", "a", "set"))
+
+    payload = {"items": {"a", "b"}}
+    json_payload = dumps_with_sets(payload)
+    decoded = json.loads(json_payload)
+    assert set(decoded["items"]) == {"a", "b"}

api/tests/unit_tests/core/rag/datasource/keyword/jieba/test_jieba_keyword_table_handler.py

@@ -0,0 +1,142 @@
+import sys
+import types
+from types import SimpleNamespace
+
+from core.rag.datasource.keyword.jieba.jieba_keyword_table_handler import JiebaKeywordTableHandler
+from core.rag.datasource.keyword.jieba.stopwords import STOPWORDS
+
+
+class _DummyTFIDF:
+    def __init__(self):
+        self.stop_words = set()
+
+    @staticmethod
+    def extract_tags(sentence: str, top_k: int | None = 20, **kwargs):
+        return ["alpha_beta", "during", "gamma"]
+
+
+def _install_fake_jieba_modules(
+    monkeypatch,
+    analyse_module: types.ModuleType,
+    jieba_attrs: dict[str, object] | None = None,
+    tfidf_module: types.ModuleType | None = None,
+):
+    jieba_module = types.ModuleType("jieba")
+    jieba_module.__path__ = []
+    if jieba_attrs:
+        for key, value in jieba_attrs.items():
+            setattr(jieba_module, key, value)
+
+    jieba_module.analyse = analyse_module
+    analyse_module.__package__ = "jieba"
+
+    monkeypatch.setitem(sys.modules, "jieba", jieba_module)
+    monkeypatch.setitem(sys.modules, "jieba.analyse", analyse_module)
+    if tfidf_module is not None:
+        monkeypatch.setitem(sys.modules, "jieba.analyse.tfidf", tfidf_module)
+    else:
+        monkeypatch.delitem(sys.modules, "jieba.analyse.tfidf", raising=False)
+
+
+def test_init_uses_existing_default_tfidf(monkeypatch):
+    analyse_module = types.ModuleType("jieba.analyse")
+    default_tfidf = _DummyTFIDF()
+    analyse_module.default_tfidf = default_tfidf
+
+    _install_fake_jieba_modules(monkeypatch, analyse_module)
+
+    handler = JiebaKeywordTableHandler()
+
+    assert handler._tfidf is default_tfidf
+    assert handler._tfidf.stop_words == STOPWORDS
+
+
+def test_load_tfidf_extractor_uses_tfidf_class_and_caches_default(monkeypatch):
+    analyse_module = types.ModuleType("jieba.analyse")
+    analyse_module.default_tfidf = None
+
+    class _TFIDFFactory(_DummyTFIDF):
+        pass
+
+    analyse_module.TFIDF = _TFIDFFactory
+    _install_fake_jieba_modules(monkeypatch, analyse_module)
+
+    handler = JiebaKeywordTableHandler()
+
+    assert isinstance(handler._tfidf, _TFIDFFactory)
+    assert analyse_module.default_tfidf is handler._tfidf
+
+
+def test_load_tfidf_extractor_imports_from_tfidf_submodule(monkeypatch):
+    analyse_module = types.ModuleType("jieba.analyse")
+    analyse_module.default_tfidf = None
+
+    tfidf_module = types.ModuleType("jieba.analyse.tfidf")
+
+    class _ImportedTFIDF(_DummyTFIDF):
+        pass
+
+    tfidf_module.TFIDF = _ImportedTFIDF
+    _install_fake_jieba_modules(monkeypatch, analyse_module, tfidf_module=tfidf_module)
+
+    handler = JiebaKeywordTableHandler()
+
+    assert isinstance(handler._tfidf, _ImportedTFIDF)
+    assert analyse_module.default_tfidf is handler._tfidf
+
+
+def test_load_tfidf_extractor_falls_back_when_tfidf_unavailable(monkeypatch):
+    analyse_module = types.ModuleType("jieba.analyse")
+    analyse_module.default_tfidf = None
+    _install_fake_jieba_modules(monkeypatch, analyse_module)
+
+    handler = JiebaKeywordTableHandler()
+    fallback_keywords = handler._tfidf.extract_tags("one two two and three", topK=1)
+
+    assert fallback_keywords == ["two"]
+
+
+def test_build_fallback_tfidf_uses_lcut_when_available(monkeypatch):
+    analyse_module = types.ModuleType("jieba.analyse")
+    _install_fake_jieba_modules(monkeypatch, analyse_module, jieba_attrs={"lcut": lambda _: ["x", "x", "y"]})
+
+    tfidf = JiebaKeywordTableHandler._build_fallback_tfidf()
+
+    assert tfidf.extract_tags("ignored", topK=1) == ["x"]
+
+
+def test_build_fallback_tfidf_uses_cut_when_lcut_is_missing(monkeypatch):
+    analyse_module = types.ModuleType("jieba.analyse")
+    _install_fake_jieba_modules(
+        monkeypatch,
+        analyse_module,
+        jieba_attrs={"cut": lambda _: iter(["foo", "foo", "bar"])},
+    )
+
+    tfidf = JiebaKeywordTableHandler._build_fallback_tfidf()
+
+    assert tfidf.extract_tags("ignored", topK=1) == ["foo"]
+
+
+def test_extract_keywords_expands_subtokens():
+    handler = JiebaKeywordTableHandler.__new__(JiebaKeywordTableHandler)
+    handler._tfidf = SimpleNamespace(extract_tags=lambda *_args, **_kwargs: ["alpha-beta", "during", "gamma"])
+
+    keywords = handler.extract_keywords("input text", max_keywords_per_chunk=3)
+
+    assert "alpha-beta" in keywords
+    assert "alpha" in keywords
+    assert "beta" in keywords
+    assert "during" in keywords
+    assert "gamma" in keywords
+
+
+def test_expand_tokens_with_subtokens_filters_stopwords_from_subtokens():
+    handler = JiebaKeywordTableHandler.__new__(JiebaKeywordTableHandler)
+
+    expanded = handler._expand_tokens_with_subtokens({"alpha-during-beta"})
+
+    assert "alpha-during-beta" in expanded
+    assert "alpha" in expanded
+    assert "beta" in expanded
+    assert "during" not in expanded

api/tests/unit_tests/core/rag/datasource/keyword/jieba/test_stopwords.py

@@ -0,0 +1,6 @@
+from core.rag.datasource.keyword.jieba.stopwords import STOPWORDS
+
+
+def test_stopwords_loaded():
+    assert "during" in STOPWORDS
+    assert "the" in STOPWORDS

api/tests/unit_tests/core/rag/datasource/keyword/test_keyword_base.py

@@ -0,0 +1,97 @@
+from types import SimpleNamespace
+
+import pytest
+
+from core.rag.datasource.keyword.keyword_base import BaseKeyword
+from core.rag.models.document import Document
+
+
+class _KeywordThatRaises(BaseKeyword):
+    def create(self, texts: list[Document], **kwargs):
+        return super().create(texts, **kwargs)
+
+    def add_texts(self, texts: list[Document], **kwargs):
+        return super().add_texts(texts, **kwargs)
+
+    def text_exists(self, id: str) -> bool:
+        return super().text_exists(id)
+
+    def delete_by_ids(self, ids: list[str]):
+        return super().delete_by_ids(ids)
+
+    def delete(self):
+        return super().delete()
+
+    def search(self, query: str, **kwargs):
+        return super().search(query, **kwargs)
+
+
+class _KeywordForHelpers(BaseKeyword):
+    def __init__(self, dataset, existing_ids: set[str] | None = None):
+        super().__init__(dataset)
+        self._existing_ids = existing_ids or set()
+
+    def create(self, texts: list[Document], **kwargs):
+        return self
+
+    def add_texts(self, texts: list[Document], **kwargs):
+        return None
+
+    def text_exists(self, id: str) -> bool:
+        return id in self._existing_ids
+
+    def delete_by_ids(self, ids: list[str]):
+        return None
+
+    def delete(self):
+        return None
+
+    def search(self, query: str, **kwargs):
+        return []
+
+
+def test_abstract_methods_raise_not_implemented():
+    keyword = _KeywordThatRaises(SimpleNamespace(id="dataset-1"))
+
+    with pytest.raises(NotImplementedError):
+        keyword.create([])
+
+    with pytest.raises(NotImplementedError):
+        keyword.add_texts([])
+
+    with pytest.raises(NotImplementedError):
+        keyword.text_exists("doc-1")
+
+    with pytest.raises(NotImplementedError):
+        keyword.delete_by_ids(["doc-1"])
+
+    with pytest.raises(NotImplementedError):
+        keyword.delete()
+
+    with pytest.raises(NotImplementedError):
+        keyword.search("query")
+
+
+def test_filter_duplicate_texts_removes_existing_doc_ids():
+    keyword = _KeywordForHelpers(SimpleNamespace(id="dataset-1"), existing_ids={"duplicate"})
+    texts = [
+        Document(page_content="keep", metadata={"doc_id": "keep"}),
+        Document(page_content="duplicate", metadata={"doc_id": "duplicate"}),
+        SimpleNamespace(page_content="without-metadata", metadata=None),
+    ]
+
+    filtered = keyword._filter_duplicate_texts(texts)
+
+    assert [text.metadata["doc_id"] for text in filtered if text.metadata] == ["keep"]
+    assert any(text.metadata is None for text in filtered)
+
+
+def test_get_uuids_returns_only_docs_with_metadata():
+    keyword = _KeywordForHelpers(SimpleNamespace(id="dataset-1"))
+    texts = [
+        Document(page_content="doc-1", metadata={"doc_id": "doc-1"}),
+        Document(page_content="doc-2", metadata={"doc_id": "doc-2"}),
+        SimpleNamespace(page_content="doc-3", metadata=None),
+    ]
+
+    assert keyword._get_uuids(texts) == ["doc-1", "doc-2"]

api/tests/unit_tests/core/rag/datasource/keyword/test_keyword_factory.py

@@ -0,0 +1,84 @@
+import sys
+import types
+from types import SimpleNamespace
+from unittest.mock import MagicMock
+
+import pytest
+
+from core.rag.datasource.keyword.keyword_factory import Keyword
+from core.rag.datasource.keyword.keyword_type import KeyWordType
+from core.rag.models.document import Document
+
+
+def test_get_keyword_factory_returns_jieba_factory(monkeypatch):
+    fake_module = types.ModuleType("core.rag.datasource.keyword.jieba.jieba")
+
+    class FakeJieba:
+        pass
+
+    fake_module.Jieba = FakeJieba
+    monkeypatch.setitem(sys.modules, "core.rag.datasource.keyword.jieba.jieba", fake_module)
+
+    assert Keyword.get_keyword_factory(KeyWordType.JIEBA) is FakeJieba
+
+
+def test_get_keyword_factory_raises_for_unsupported_type():
+    with pytest.raises(ValueError, match="Keyword store unsupported is not supported"):
+        Keyword.get_keyword_factory("unsupported")
+
+
+def test_keyword_initialization_uses_configured_factory(monkeypatch):
+    dataset = SimpleNamespace(id="dataset-1")
+    fake_processor = MagicMock()
+
+    monkeypatch.setattr("core.rag.datasource.keyword.keyword_factory.dify_config.KEYWORD_STORE", KeyWordType.JIEBA)
+    monkeypatch.setattr(Keyword, "get_keyword_factory", staticmethod(lambda keyword_type: lambda _: fake_processor))
+
+    keyword = Keyword(dataset)
+
+    assert keyword._keyword_processor is fake_processor
+
+
+def test_keyword_methods_forward_to_processor():
+    processor = MagicMock()
+    processor.text_exists.return_value = True
+    processor.search.return_value = [Document(page_content="matched", metadata={"doc_id": "doc-1"})]
+
+    keyword = Keyword.__new__(Keyword)
+    keyword._keyword_processor = processor
+
+    docs = [Document(page_content="doc", metadata={"doc_id": "doc-1"})]
+    keyword.create(docs, foo="bar")
+    keyword.add_texts(docs, batch=True)
+    assert keyword.text_exists("doc-1") is True
+    keyword.delete_by_ids(["doc-1"])
+    keyword.delete()
+    assert keyword.search("query", top_k=1) == processor.search.return_value
+
+    processor.create.assert_called_once_with(docs, foo="bar")
+    processor.add_texts.assert_called_once_with(docs, batch=True)
+    processor.text_exists.assert_called_once_with("doc-1")
+    processor.delete_by_ids.assert_called_once_with(["doc-1"])
+    processor.delete.assert_called_once()
+    processor.search.assert_called_once_with("query", top_k=1)
+
+
+def test_keyword_getattr_returns_callable_and_raises_for_invalid_attributes():
+    class Processor:
+        value = 1
+
+        @staticmethod
+        def custom():
+            return "ok"
+
+    keyword = Keyword.__new__(Keyword)
+    keyword._keyword_processor = Processor()
+
+    assert keyword.custom() == "ok"
+
+    with pytest.raises(AttributeError):
+        _ = keyword.value
+
+    keyword._keyword_processor = None
+    with pytest.raises(AttributeError):
+        _ = keyword.missing_method

api/tests/unit_tests/core/rag/datasource/vdb/alibabacloud_mysql/test_alibabacloud_mysql_factory.py

@@ -0,0 +1,74 @@
+from types import SimpleNamespace
+from unittest.mock import MagicMock, patch
+
+import pytest
+
+import core.rag.datasource.vdb.alibabacloud_mysql.alibabacloud_mysql_vector as alibaba_module
+from core.rag.datasource.vdb.alibabacloud_mysql.alibabacloud_mysql_vector import AlibabaCloudMySQLVectorFactory
+
+
+def test_validate_distance_function_accepts_supported_values():
+    factory = AlibabaCloudMySQLVectorFactory()
+
+    assert factory._validate_distance_function("cosine") == "cosine"
+    assert factory._validate_distance_function("euclidean") == "euclidean"
+
+
+def test_validate_distance_function_rejects_unsupported_values():
+    factory = AlibabaCloudMySQLVectorFactory()
+
+    with pytest.raises(ValueError, match="Invalid distance function"):
+        factory._validate_distance_function("dot_product")
+
+
+def test_factory_init_vector_uses_existing_index_struct_class_prefix(monkeypatch):
+    factory = AlibabaCloudMySQLVectorFactory()
+    dataset = SimpleNamespace(
+        id="dataset-1",
+        index_struct_dict={"vector_store": {"class_prefix": "existing_collection"}},
+        index_struct=None,
+    )
+
+    monkeypatch.setattr(alibaba_module.dify_config, "ALIBABACLOUD_MYSQL_HOST", "host")
+    monkeypatch.setattr(alibaba_module.dify_config, "ALIBABACLOUD_MYSQL_PORT", 3306)
+    monkeypatch.setattr(alibaba_module.dify_config, "ALIBABACLOUD_MYSQL_USER", "user")
+    monkeypatch.setattr(alibaba_module.dify_config, "ALIBABACLOUD_MYSQL_PASSWORD", "password")
+    monkeypatch.setattr(alibaba_module.dify_config, "ALIBABACLOUD_MYSQL_DATABASE", "db")
+    monkeypatch.setattr(alibaba_module.dify_config, "ALIBABACLOUD_MYSQL_MAX_CONNECTION", 5)
+    monkeypatch.setattr(alibaba_module.dify_config, "ALIBABACLOUD_MYSQL_CHARSET", "utf8mb4")
+    monkeypatch.setattr(alibaba_module.dify_config, "ALIBABACLOUD_MYSQL_DISTANCE_FUNCTION", "cosine")
+    monkeypatch.setattr(alibaba_module.dify_config, "ALIBABACLOUD_MYSQL_HNSW_M", 6)
+
+    with patch.object(alibaba_module, "AlibabaCloudMySQLVector", return_value="vector") as vector_cls:
+        result = factory.init_vector(dataset, attributes=[], embeddings=MagicMock())
+
+    assert result == "vector"
+    assert vector_cls.call_args.kwargs["collection_name"] == "existing_collection"
+
+
+def test_factory_init_vector_generates_collection_name_when_index_struct_is_missing(monkeypatch):
+    factory = AlibabaCloudMySQLVectorFactory()
+    dataset = SimpleNamespace(
+        id="dataset-2",
+        index_struct_dict=None,
+        index_struct=None,
+    )
+
+    monkeypatch.setattr(alibaba_module.Dataset, "gen_collection_name_by_id", lambda dataset_id: f"COL_{dataset_id}")
+    monkeypatch.setattr(alibaba_module.dify_config, "ALIBABACLOUD_MYSQL_HOST", "host")
+    monkeypatch.setattr(alibaba_module.dify_config, "ALIBABACLOUD_MYSQL_PORT", 3306)
+    monkeypatch.setattr(alibaba_module.dify_config, "ALIBABACLOUD_MYSQL_USER", "user")
+    monkeypatch.setattr(alibaba_module.dify_config, "ALIBABACLOUD_MYSQL_PASSWORD", "password")
+    monkeypatch.setattr(alibaba_module.dify_config, "ALIBABACLOUD_MYSQL_DATABASE", "db")
+    monkeypatch.setattr(alibaba_module.dify_config, "ALIBABACLOUD_MYSQL_MAX_CONNECTION", 5)
+    monkeypatch.setattr(alibaba_module.dify_config, "ALIBABACLOUD_MYSQL_CHARSET", "utf8mb4")
+    monkeypatch.setattr(alibaba_module.dify_config, "ALIBABACLOUD_MYSQL_DISTANCE_FUNCTION", "euclidean")
+    monkeypatch.setattr(alibaba_module.dify_config, "ALIBABACLOUD_MYSQL_HNSW_M", 12)
+
+    with patch.object(alibaba_module, "AlibabaCloudMySQLVector", return_value="vector") as vector_cls:
+        result = factory.init_vector(dataset, attributes=[], embeddings=MagicMock())
+
+    assert result == "vector"
+    vector_cls.assert_called_once()
+    assert vector_cls.call_args.kwargs["collection_name"] == "COL_dataset-2"
+    assert dataset.index_struct is not None

api/tests/unit_tests/core/rag/datasource/vdb/analyticdb/test_analyticdb_vector.py

@@ -0,0 +1,133 @@
+from types import SimpleNamespace
+from unittest.mock import MagicMock, patch
+
+import pytest
+
+import core.rag.datasource.vdb.analyticdb.analyticdb_vector as analyticdb_module
+from core.rag.datasource.vdb.analyticdb.analyticdb_vector import AnalyticdbVector, AnalyticdbVectorFactory
+from core.rag.datasource.vdb.analyticdb.analyticdb_vector_openapi import AnalyticdbVectorOpenAPIConfig
+from core.rag.datasource.vdb.analyticdb.analyticdb_vector_sql import AnalyticdbVectorBySqlConfig
+from core.rag.models.document import Document
+
+
+def test_init_prefers_openapi_when_api_config_is_provided():
+    api_config = AnalyticdbVectorOpenAPIConfig(
+        access_key_id="ak",
+        access_key_secret="sk",
+        region_id="cn-hangzhou",
+        instance_id="instance-1",
+        account="account",
+        account_password="password",
+        namespace="dify",
+        namespace_password="ns-password",
+    )
+
+    with patch.object(analyticdb_module, "AnalyticdbVectorOpenAPI", return_value="openapi_runner") as openapi_cls:
+        vector = AnalyticdbVector("COLLECTION", api_config=api_config, sql_config=None)
+
+    assert vector.analyticdb_vector == "openapi_runner"
+    openapi_cls.assert_called_once_with("COLLECTION", api_config)
+
+
+def test_init_uses_sql_implementation_when_api_config_is_missing():
+    sql_config = AnalyticdbVectorBySqlConfig(
+        host="localhost",
+        port=5432,
+        account="account",
+        account_password="password",
+        min_connection=1,
+        max_connection=2,
+        namespace="dify",
+    )
+
+    with patch.object(analyticdb_module, "AnalyticdbVectorBySql", return_value="sql_runner") as sql_cls:
+        vector = AnalyticdbVector("COLLECTION", api_config=None, sql_config=sql_config)
+
+    assert vector.analyticdb_vector == "sql_runner"
+    sql_cls.assert_called_once_with("COLLECTION", sql_config)
+
+
+def test_init_raises_when_both_configs_are_missing():
+    with pytest.raises(ValueError, match="Either api_config or sql_config must be provided"):
+        AnalyticdbVector("COLLECTION", api_config=None, sql_config=None)
+
+
+def test_vector_methods_delegate_to_underlying_implementation():
+    runner = MagicMock()
+    runner.search_by_vector.return_value = [Document(page_content="v", metadata={"doc_id": "1"})]
+    runner.search_by_full_text.return_value = [Document(page_content="t", metadata={"doc_id": "2"})]
+    runner.text_exists.return_value = True
+
+    vector = AnalyticdbVector.__new__(AnalyticdbVector)
+    vector.analyticdb_vector = runner
+
+    texts = [Document(page_content="hello", metadata={"doc_id": "d1"})]
+    vector.create(texts=texts, embeddings=[[0.1, 0.2]])
+    vector.add_texts(documents=texts, embeddings=[[0.1, 0.2]])
+    assert vector.text_exists("d1") is True
+    vector.delete_by_ids(["d1"])
+    vector.delete_by_metadata_field("document_id", "doc-1")
+    assert vector.search_by_vector([0.1, 0.2], top_k=2) == runner.search_by_vector.return_value
+    assert vector.search_by_full_text("hello", top_k=2) == runner.search_by_full_text.return_value
+    vector.delete()
+
+    runner._create_collection_if_not_exists.assert_called_once_with(2)
+    runner.add_texts.assert_any_call(texts, [[0.1, 0.2]])
+    runner.delete_by_ids.assert_called_once_with(["d1"])
+    runner.delete_by_metadata_field.assert_called_once_with("document_id", "doc-1")
+    runner.delete.assert_called_once()
+
+
+def test_get_type_is_analyticdb():
+    vector = AnalyticdbVector.__new__(AnalyticdbVector)
+    assert vector.get_type() == "analyticdb"
+
+
+def test_factory_builds_openapi_config_when_host_is_missing(monkeypatch):
+    factory = AnalyticdbVectorFactory()
+    dataset = SimpleNamespace(id="dataset-1", index_struct_dict=None, index_struct=None)
+
+    monkeypatch.setattr(analyticdb_module.Dataset, "gen_collection_name_by_id", lambda _id: "AUTO_COLLECTION")
+    monkeypatch.setattr(analyticdb_module.dify_config, "ANALYTICDB_HOST", None)
+    monkeypatch.setattr(analyticdb_module.dify_config, "ANALYTICDB_KEY_ID", "ak")
+    monkeypatch.setattr(analyticdb_module.dify_config, "ANALYTICDB_KEY_SECRET", "sk")
+    monkeypatch.setattr(analyticdb_module.dify_config, "ANALYTICDB_REGION_ID", "cn-hz")
+    monkeypatch.setattr(analyticdb_module.dify_config, "ANALYTICDB_INSTANCE_ID", "instance")
+    monkeypatch.setattr(analyticdb_module.dify_config, "ANALYTICDB_ACCOUNT", "account")
+    monkeypatch.setattr(analyticdb_module.dify_config, "ANALYTICDB_PASSWORD", "password")
+    monkeypatch.setattr(analyticdb_module.dify_config, "ANALYTICDB_NAMESPACE", "dify")
+    monkeypatch.setattr(analyticdb_module.dify_config, "ANALYTICDB_NAMESPACE_PASSWORD", "ns-password")
+
+    with patch.object(analyticdb_module, "AnalyticdbVector", return_value="vector") as vector_cls:
+        result = factory.init_vector(dataset, attributes=[], embeddings=MagicMock())
+
+    assert result == "vector"
+    args = vector_cls.call_args.args
+    assert args[0] == "auto_collection"
+    assert isinstance(args[1], AnalyticdbVectorOpenAPIConfig)
+    assert args[2] is None
+    assert dataset.index_struct is not None
+
+
+def test_factory_builds_sql_config_when_host_is_present(monkeypatch):
+    factory = AnalyticdbVectorFactory()
+    dataset = SimpleNamespace(
+        id="dataset-2", index_struct_dict={"vector_store": {"class_prefix": "EXISTING"}}, index_struct=None
+    )
+
+    monkeypatch.setattr(analyticdb_module.dify_config, "ANALYTICDB_HOST", "127.0.0.1")
+    monkeypatch.setattr(analyticdb_module.dify_config, "ANALYTICDB_PORT", 5432)
+    monkeypatch.setattr(analyticdb_module.dify_config, "ANALYTICDB_ACCOUNT", "account")
+    monkeypatch.setattr(analyticdb_module.dify_config, "ANALYTICDB_PASSWORD", "password")
+    monkeypatch.setattr(analyticdb_module.dify_config, "ANALYTICDB_MIN_CONNECTION", 1)
+    monkeypatch.setattr(analyticdb_module.dify_config, "ANALYTICDB_MAX_CONNECTION", 3)
+    monkeypatch.setattr(analyticdb_module.dify_config, "ANALYTICDB_NAMESPACE", "dify")
+
+    with patch.object(analyticdb_module, "AnalyticdbVector", return_value="vector") as vector_cls:
+        result = factory.init_vector(dataset, attributes=[], embeddings=MagicMock())
+
+    assert result == "vector"
+    args = vector_cls.call_args.args
+    assert args[0] == "existing"
+    assert args[1] is None
+    assert isinstance(args[2], AnalyticdbVectorBySqlConfig)

api/tests/unit_tests/core/rag/datasource/vdb/analyticdb/test_analyticdb_vector_openapi.py

@@ -0,0 +1,384 @@
+import json
+import sys
+import types
+from types import SimpleNamespace
+from unittest.mock import MagicMock
+
+import pytest
+
+import core.rag.datasource.vdb.analyticdb.analyticdb_vector_openapi as openapi_module
+from core.rag.datasource.vdb.analyticdb.analyticdb_vector_openapi import (
+    AnalyticdbVectorOpenAPI,
+    AnalyticdbVectorOpenAPIConfig,
+)
+from core.rag.models.document import Document
+
+
+def _request_class(name: str):
+    class _Request:
+        def __init__(self, **kwargs):
+            for key, value in kwargs.items():
+                setattr(self, key, value)
+
+    _Request.__name__ = name
+    return _Request
+
+
+def _install_openapi_stubs(monkeypatch):
+    gpdb_package = types.ModuleType("alibabacloud_gpdb20160503")
+    gpdb_package.__path__ = []
+    gpdb_models = types.ModuleType("alibabacloud_gpdb20160503.models")
+    for class_name in [
+        "InitVectorDatabaseRequest",
+        "DescribeNamespaceRequest",
+        "CreateNamespaceRequest",
+        "DescribeCollectionRequest",
+        "CreateCollectionRequest",
+        "UpsertCollectionDataRequestRows",
+        "UpsertCollectionDataRequest",
+        "QueryCollectionDataRequest",
+        "DeleteCollectionDataRequest",
+        "DeleteCollectionRequest",
+    ]:
+        setattr(gpdb_models, class_name, _request_class(class_name))
+
+    class _Client:
+        def __init__(self, config):
+            self.config = config
+
+    gpdb_client = types.ModuleType("alibabacloud_gpdb20160503.client")
+    gpdb_client.Client = _Client
+    gpdb_package.models = gpdb_models
+
+    tea_openapi = types.ModuleType("alibabacloud_tea_openapi")
+    tea_openapi.__path__ = []
+    tea_openapi_models = types.ModuleType("alibabacloud_tea_openapi.models")
+
+    class OpenApiConfig:
+        def __init__(self, **kwargs):
+            for key, value in kwargs.items():
+                setattr(self, key, value)
+
+    tea_openapi_models.Config = OpenApiConfig
+    tea_openapi.models = tea_openapi_models
+
+    tea_package = types.ModuleType("Tea")
+    tea_package.__path__ = []
+    tea_exceptions = types.ModuleType("Tea.exceptions")
+
+    class TeaError(Exception):
+        def __init__(self, status_code=None, **kwargs):
+            super().__init__("TeaException")
+            status_code = kwargs.get("statusCode", status_code)
+            self.statusCode = status_code
+            self.status_code = status_code
+
+    tea_exceptions.TeaException = TeaError
+    tea_package.exceptions = tea_exceptions
+
+    monkeypatch.setitem(sys.modules, "alibabacloud_gpdb20160503", gpdb_package)
+    monkeypatch.setitem(sys.modules, "alibabacloud_gpdb20160503.models", gpdb_models)
+    monkeypatch.setitem(sys.modules, "alibabacloud_gpdb20160503.client", gpdb_client)
+    monkeypatch.setitem(sys.modules, "alibabacloud_tea_openapi", tea_openapi)
+    monkeypatch.setitem(sys.modules, "alibabacloud_tea_openapi.models", tea_openapi_models)
+    monkeypatch.setitem(sys.modules, "Tea", tea_package)
+    monkeypatch.setitem(sys.modules, "Tea.exceptions", tea_exceptions)
+
+    return SimpleNamespace(models=gpdb_models, TeaException=TeaError, OpenApiConfig=OpenApiConfig)
+
+
+def _config() -> AnalyticdbVectorOpenAPIConfig:
+    return AnalyticdbVectorOpenAPIConfig(
+        access_key_id="ak",
+        access_key_secret="sk",
+        region_id="cn-hangzhou",
+        instance_id="instance-1",
+        account="account",
+        account_password="password",
+        namespace="dify",
+        namespace_password="ns-password",
+    )
+
+
+@pytest.mark.parametrize(
+    ("field", "value", "error_message"),
+    [
+        ("access_key_id", "", "ANALYTICDB_KEY_ID"),
+        ("access_key_secret", "", "ANALYTICDB_KEY_SECRET"),
+        ("region_id", "", "ANALYTICDB_REGION_ID"),
+        ("instance_id", "", "ANALYTICDB_INSTANCE_ID"),
+        ("account", "", "ANALYTICDB_ACCOUNT"),
+        ("account_password", "", "ANALYTICDB_PASSWORD"),
+        ("namespace_password", "", "ANALYTICDB_NAMESPACE_PASSWORD"),
+    ],
+)
+def test_openapi_config_validation(field, value, error_message):
+    values = _config().model_dump()
+    values[field] = value
+
+    with pytest.raises(ValueError, match=error_message):
+        AnalyticdbVectorOpenAPIConfig.model_validate(values)
+
+
+def test_openapi_config_to_client_params():
+    config = _config()
+    params = config.to_analyticdb_client_params()
+
+    assert params["access_key_id"] == "ak"
+    assert params["access_key_secret"] == "sk"
+    assert params["region_id"] == "cn-hangzhou"
+    assert params["read_timeout"] == 60000
+
+
+def test_init_creates_openapi_client_and_runs_initialize(monkeypatch):
+    stubs = _install_openapi_stubs(monkeypatch)
+    initialize_mock = MagicMock()
+    monkeypatch.setattr(openapi_module.AnalyticdbVectorOpenAPI, "_initialize", initialize_mock)
+
+    vector = AnalyticdbVectorOpenAPI("COLLECTION_1", _config())
+
+    assert vector._collection_name == "collection_1"
+    assert isinstance(vector._client_config, stubs.OpenApiConfig)
+    assert vector._client_config.user_agent == "dify"
+    assert vector._client_config.access_key_id == "ak"
+    assert vector._client.config is vector._client_config
+    initialize_mock.assert_called_once_with()
+
+
+def test_initialize_skips_when_cached(monkeypatch):
+    lock = MagicMock()
+    lock.__enter__.return_value = None
+    lock.__exit__.return_value = None
+    monkeypatch.setattr(openapi_module.redis_client, "lock", MagicMock(return_value=lock))
+    monkeypatch.setattr(openapi_module.redis_client, "get", MagicMock(return_value=1))
+    monkeypatch.setattr(openapi_module.redis_client, "set", MagicMock())
+
+    vector = AnalyticdbVectorOpenAPI.__new__(AnalyticdbVectorOpenAPI)
+    vector.config = _config()
+    vector._initialize_vector_database = MagicMock()
+    vector._create_namespace_if_not_exists = MagicMock()
+
+    vector._initialize()
+
+    vector._initialize_vector_database.assert_not_called()
+    vector._create_namespace_if_not_exists.assert_not_called()
+
+
+def test_initialize_runs_when_cache_is_missing(monkeypatch):
+    lock = MagicMock()
+    lock.__enter__.return_value = None
+    lock.__exit__.return_value = None
+    monkeypatch.setattr(openapi_module.redis_client, "lock", MagicMock(return_value=lock))
+    monkeypatch.setattr(openapi_module.redis_client, "get", MagicMock(return_value=None))
+    monkeypatch.setattr(openapi_module.redis_client, "set", MagicMock())
+
+    vector = AnalyticdbVectorOpenAPI.__new__(AnalyticdbVectorOpenAPI)
+    vector.config = _config()
+    vector._initialize_vector_database = MagicMock()
+    vector._create_namespace_if_not_exists = MagicMock()
+
+    vector._initialize()
+
+    vector._initialize_vector_database.assert_called_once()
+    vector._create_namespace_if_not_exists.assert_called_once()
+    openapi_module.redis_client.set.assert_called_once()
+
+
+def test_initialize_vector_database_calls_openapi_client(monkeypatch):
+    _install_openapi_stubs(monkeypatch)
+    vector = AnalyticdbVectorOpenAPI.__new__(AnalyticdbVectorOpenAPI)
+    vector.config = _config()
+    vector._client = MagicMock()
+
+    vector._initialize_vector_database()
+
+    request = vector._client.init_vector_database.call_args.args[0]
+    assert request.dbinstance_id == "instance-1"
+    assert request.region_id == "cn-hangzhou"
+    assert request.manager_account == "account"
+    assert request.manager_account_password == "password"
+
+
+def test_create_namespace_creates_when_namespace_not_found(monkeypatch):
+    stubs = _install_openapi_stubs(monkeypatch)
+    vector = AnalyticdbVectorOpenAPI.__new__(AnalyticdbVectorOpenAPI)
+    vector.config = _config()
+    vector._client = MagicMock()
+    vector._client.describe_namespace.side_effect = stubs.TeaException(statusCode=404)
+
+    vector._create_namespace_if_not_exists()
+
+    vector._client.create_namespace.assert_called_once()
+
+
+def test_create_namespace_raises_on_unexpected_api_error(monkeypatch):
+    stubs = _install_openapi_stubs(monkeypatch)
+    vector = AnalyticdbVectorOpenAPI.__new__(AnalyticdbVectorOpenAPI)
+    vector.config = _config()
+    vector._client = MagicMock()
+    vector._client.describe_namespace.side_effect = stubs.TeaException(statusCode=500)
+
+    with pytest.raises(ValueError, match="failed to create namespace"):
+        vector._create_namespace_if_not_exists()
+
+
+def test_create_namespace_noop_when_namespace_exists(monkeypatch):
+    _install_openapi_stubs(monkeypatch)
+    vector = AnalyticdbVectorOpenAPI.__new__(AnalyticdbVectorOpenAPI)
+    vector.config = _config()
+    vector._client = MagicMock()
+
+    vector._create_namespace_if_not_exists()
+
+    vector._client.describe_namespace.assert_called_once()
+    vector._client.create_namespace.assert_not_called()
+
+
+def test_create_collection_if_not_exists_creates_when_missing(monkeypatch):
+    stubs = _install_openapi_stubs(monkeypatch)
+    lock = MagicMock()
+    lock.__enter__.return_value = None
+    lock.__exit__.return_value = None
+    monkeypatch.setattr(openapi_module.redis_client, "lock", MagicMock(return_value=lock))
+    monkeypatch.setattr(openapi_module.redis_client, "get", MagicMock(return_value=None))
+    monkeypatch.setattr(openapi_module.redis_client, "set", MagicMock())
+
+    vector = AnalyticdbVectorOpenAPI.__new__(AnalyticdbVectorOpenAPI)
+    vector._collection_name = "collection_1"
+    vector.config = _config()
+    vector._client = MagicMock()
+    vector._client.describe_collection.side_effect = stubs.TeaException(statusCode=404)
+
+    vector._create_collection_if_not_exists(embedding_dimension=1024)
+
+    vector._client.create_collection.assert_called_once()
+    openapi_module.redis_client.set.assert_called_once()
+
+
+def test_create_collection_if_not_exists_skips_when_cached(monkeypatch):
+    lock = MagicMock()
+    lock.__enter__.return_value = None
+    lock.__exit__.return_value = None
+    monkeypatch.setattr(openapi_module.redis_client, "lock", MagicMock(return_value=lock))
+    monkeypatch.setattr(openapi_module.redis_client, "get", MagicMock(return_value=1))
+    monkeypatch.setattr(openapi_module.redis_client, "set", MagicMock())
+
+    vector = AnalyticdbVectorOpenAPI.__new__(AnalyticdbVectorOpenAPI)
+    vector._collection_name = "collection_1"
+    vector.config = _config()
+    vector._client = MagicMock()
+
+    vector._create_collection_if_not_exists(embedding_dimension=1024)
+
+    vector._client.describe_collection.assert_not_called()
+    vector._client.create_collection.assert_not_called()
+
+
+def test_create_collection_if_not_exists_raises_on_non_404_errors(monkeypatch):
+    stubs = _install_openapi_stubs(monkeypatch)
+    lock = MagicMock()
+    lock.__enter__.return_value = None
+    lock.__exit__.return_value = None
+    monkeypatch.setattr(openapi_module.redis_client, "lock", MagicMock(return_value=lock))
+    monkeypatch.setattr(openapi_module.redis_client, "get", MagicMock(return_value=None))
+    monkeypatch.setattr(openapi_module.redis_client, "set", MagicMock())
+
+    vector = AnalyticdbVectorOpenAPI.__new__(AnalyticdbVectorOpenAPI)
+    vector._collection_name = "collection_1"
+    vector.config = _config()
+    vector._client = MagicMock()
+    vector._client.describe_collection.side_effect = stubs.TeaException(statusCode=500)
+
+    with pytest.raises(ValueError, match="failed to create collection collection_1"):
+        vector._create_collection_if_not_exists(embedding_dimension=512)
+
+
+def test_openapi_add_delete_and_search_methods(monkeypatch):
+    _install_openapi_stubs(monkeypatch)
+    vector = AnalyticdbVectorOpenAPI.__new__(AnalyticdbVectorOpenAPI)
+    vector._collection_name = "collection_1"
+    vector.config = _config()
+    vector._client = MagicMock()
+
+    documents = [
+        Document(page_content="doc 1", metadata={"doc_id": "d1", "document_id": "doc-1"}),
+        SimpleNamespace(page_content="doc 2", metadata=None),
+    ]
+    embeddings = [[0.1, 0.2], [0.2, 0.3]]
+    vector.add_texts(documents, embeddings)
+
+    upsert_request = vector._client.upsert_collection_data.call_args.args[0]
+    assert upsert_request.collection == "collection_1"
+    assert len(upsert_request.rows) == 1
+
+    vector._client.query_collection_data.return_value = SimpleNamespace(
+        body=SimpleNamespace(matches=SimpleNamespace(match=[SimpleNamespace()]))
+    )
+    assert vector.text_exists("d1") is True
+
+    vector.delete_by_ids(["d1", "d2"])
+    request = vector._client.delete_collection_data.call_args.args[0]
+    assert request.collection_data_filter == "ref_doc_id IN ('d1','d2')"
+
+    vector.delete_by_metadata_field("document_id", "doc-1")
+    request = vector._client.delete_collection_data.call_args.args[0]
+    assert request.collection_data_filter == "metadata_ ->> 'document_id' = 'doc-1'"
+
+    match_high = SimpleNamespace(
+        score=0.9,
+        metadata={"metadata_": json.dumps({"document_id": "doc-1"}), "page_content": "high"},
+        values=SimpleNamespace(value=[1.0, 2.0]),
+    )
+    match_low = SimpleNamespace(
+        score=0.1,
+        metadata={"metadata_": json.dumps({"document_id": "doc-2"}), "page_content": "low"},
+        values=SimpleNamespace(value=[3.0, 4.0]),
+    )
+    vector._client.query_collection_data.return_value = SimpleNamespace(
+        body=SimpleNamespace(matches=SimpleNamespace(match=[match_low, match_high]))
+    )
+
+    docs_by_vector = vector.search_by_vector([0.1, 0.2], top_k=2, score_threshold=0.5, document_ids_filter=["doc-1"])
+    assert len(docs_by_vector) == 1
+    assert docs_by_vector[0].page_content == "high"
+    assert docs_by_vector[0].metadata["score"] == 0.9
+
+    docs_by_text = vector.search_by_full_text("hello", top_k=2, score_threshold=0.2)
+    assert len(docs_by_text) == 1
+    assert docs_by_text[0].page_content == "high"
+
+
+def test_text_exists_returns_false_when_matches_empty(monkeypatch):
+    _install_openapi_stubs(monkeypatch)
+    vector = AnalyticdbVectorOpenAPI.__new__(AnalyticdbVectorOpenAPI)
+    vector._collection_name = "collection_1"
+    vector.config = _config()
+    vector._client = MagicMock()
+    vector._client.query_collection_data.return_value = SimpleNamespace(
+        body=SimpleNamespace(matches=SimpleNamespace(match=[]))
+    )
+
+    assert vector.text_exists("missing-id") is False
+
+
+def test_openapi_delete_success(monkeypatch):
+    _install_openapi_stubs(monkeypatch)
+    vector = AnalyticdbVectorOpenAPI.__new__(AnalyticdbVectorOpenAPI)
+    vector._collection_name = "collection_1"
+    vector.config = _config()
+    vector._client = MagicMock()
+
+    vector.delete()
+    vector._client.delete_collection.assert_called_once()
+
+
+def test_openapi_delete_propagates_errors(monkeypatch):
+    _install_openapi_stubs(monkeypatch)
+    vector = AnalyticdbVectorOpenAPI.__new__(AnalyticdbVectorOpenAPI)
+    vector._collection_name = "collection_1"
+    vector.config = _config()
+    vector._client = MagicMock()
+    vector._client.delete_collection.side_effect = RuntimeError("boom")
+
+    with pytest.raises(RuntimeError, match="boom"):
+        vector.delete()

api/tests/unit_tests/core/rag/datasource/vdb/analyticdb/test_analyticdb_vector_sql.py

@@ -0,0 +1,427 @@
+from contextlib import contextmanager
+from types import SimpleNamespace
+from unittest.mock import MagicMock
+
+import psycopg2.errors
+import pytest
+
+import core.rag.datasource.vdb.analyticdb.analyticdb_vector_sql as sql_module
+from core.rag.datasource.vdb.analyticdb.analyticdb_vector_sql import (
+    AnalyticdbVectorBySql,
+    AnalyticdbVectorBySqlConfig,
+)
+from core.rag.models.document import Document
+
+
+def _config_values() -> dict:
+    return {
+        "host": "localhost",
+        "port": 5432,
+        "account": "account",
+        "account_password": "password",
+        "min_connection": 1,
+        "max_connection": 2,
+        "namespace": "dify",
+    }
+
+
+@pytest.mark.parametrize(
+    ("field", "value", "error_message"),
+    [
+        ("host", "", "ANALYTICDB_HOST"),
+        ("port", 0, "ANALYTICDB_PORT"),
+        ("account", "", "ANALYTICDB_ACCOUNT"),
+        ("account_password", "", "ANALYTICDB_PASSWORD"),
+        ("min_connection", 0, "ANALYTICDB_MIN_CONNECTION"),
+        ("max_connection", 0, "ANALYTICDB_MAX_CONNECTION"),
+    ],
+)
+def test_sql_config_required_fields(field, value, error_message):
+    values = _config_values()
+    values[field] = value
+
+    with pytest.raises(ValueError, match=error_message):
+        AnalyticdbVectorBySqlConfig.model_validate(values)
+
+
+def test_sql_config_rejects_min_connection_greater_than_max_connection():
+    values = _config_values()
+    values["min_connection"] = 10
+    values["max_connection"] = 2
+
+    with pytest.raises(ValueError, match="ANALYTICDB_MIN_CONNECTION should less than ANALYTICDB_MAX_CONNECTION"):
+        AnalyticdbVectorBySqlConfig.model_validate(values)
+
+
+def test_initialize_skips_when_cache_exists(monkeypatch):
+    lock = MagicMock()
+    lock.__enter__.return_value = None
+    lock.__exit__.return_value = None
+    monkeypatch.setattr(sql_module.redis_client, "lock", MagicMock(return_value=lock))
+    monkeypatch.setattr(sql_module.redis_client, "get", MagicMock(return_value=1))
+    monkeypatch.setattr(sql_module.redis_client, "set", MagicMock())
+
+    vector = AnalyticdbVectorBySql.__new__(AnalyticdbVectorBySql)
+    vector.config = AnalyticdbVectorBySqlConfig(**_config_values())
+    vector._initialize_vector_database = MagicMock()
+
+    vector._initialize()
+
+    vector._initialize_vector_database.assert_not_called()
+
+
+def test_initialize_runs_when_cache_is_missing(monkeypatch):
+    lock = MagicMock()
+    lock.__enter__.return_value = None
+    lock.__exit__.return_value = None
+    monkeypatch.setattr(sql_module.redis_client, "lock", MagicMock(return_value=lock))
+    monkeypatch.setattr(sql_module.redis_client, "get", MagicMock(return_value=None))
+    monkeypatch.setattr(sql_module.redis_client, "set", MagicMock())
+
+    vector = AnalyticdbVectorBySql.__new__(AnalyticdbVectorBySql)
+    vector.config = AnalyticdbVectorBySqlConfig(**_config_values())
+    vector._initialize_vector_database = MagicMock()
+
+    vector._initialize()
+
+    vector._initialize_vector_database.assert_called_once()
+    sql_module.redis_client.set.assert_called_once()
+
+
+def test_create_connection_pool_uses_psycopg2_pool(monkeypatch):
+    vector = AnalyticdbVectorBySql.__new__(AnalyticdbVectorBySql)
+    vector.config = AnalyticdbVectorBySqlConfig(**_config_values())
+    vector.databaseName = "knowledgebase"
+
+    pool_instance = MagicMock()
+    monkeypatch.setattr(sql_module.psycopg2.pool, "SimpleConnectionPool", MagicMock(return_value=pool_instance))
+
+    pool = vector._create_connection_pool()
+
+    assert pool is pool_instance
+    sql_module.psycopg2.pool.SimpleConnectionPool.assert_called_once()
+
+
+def test_get_cursor_context_manager_handles_connection_lifecycle():
+    vector = AnalyticdbVectorBySql.__new__(AnalyticdbVectorBySql)
+    cursor = MagicMock()
+    connection = MagicMock()
+    connection.cursor.return_value = cursor
+    pool = MagicMock()
+    pool.getconn.return_value = connection
+    vector.pool = pool
+
+    with vector._get_cursor() as cur:
+        assert cur is cursor
+
+    cursor.close.assert_called_once()
+    connection.commit.assert_called_once()
+    pool.putconn.assert_called_once_with(connection)
+
+
+def test_add_texts_inserts_only_documents_with_metadata(monkeypatch):
+    vector = AnalyticdbVectorBySql.__new__(AnalyticdbVectorBySql)
+    vector.table_name = "dify.collection"
+
+    cursor = MagicMock()
+
+    @contextmanager
+    def _cursor_context():
+        yield cursor
+
+    vector._get_cursor = _cursor_context
+
+    monkeypatch.setattr(sql_module.uuid, "uuid4", lambda: "prefix-id")
+    monkeypatch.setattr(sql_module.psycopg2.extras, "execute_batch", MagicMock())
+
+    docs = [
+        Document(page_content="doc 1", metadata={"doc_id": "d1", "document_id": "doc-1"}),
+        SimpleNamespace(page_content="doc 2", metadata=None),
+    ]
+    vector.add_texts(docs, [[0.1, 0.2], [0.2, 0.3]])
+
+    execute_args = sql_module.psycopg2.extras.execute_batch.call_args.args
+    assert execute_args[0] is cursor
+    assert len(execute_args[2]) == 1
+
+
+def test_text_exists_returns_true_and_false_based_on_query_result():
+    vector = AnalyticdbVectorBySql.__new__(AnalyticdbVectorBySql)
+    vector.table_name = "dify.collection"
+    cursor = MagicMock()
+
+    @contextmanager
+    def _cursor_context():
+        yield cursor
+
+    vector._get_cursor = _cursor_context
+
+    cursor.fetchone.return_value = ("row",)
+    assert vector.text_exists("d1") is True
+
+    cursor.fetchone.return_value = None
+    assert vector.text_exists("d1") is False
+
+
+def test_delete_by_ids_handles_empty_input_and_missing_table_error():
+    vector = AnalyticdbVectorBySql.__new__(AnalyticdbVectorBySql)
+    vector.table_name = "dify.collection"
+    cursor = MagicMock()
+
+    @contextmanager
+    def _cursor_context():
+        yield cursor
+
+    vector._get_cursor = _cursor_context
+    vector.delete_by_ids([])
+    cursor.execute.assert_not_called()
+
+    cursor.execute.side_effect = psycopg2.errors.UndefinedTable("relation does not exist")
+    vector.delete_by_ids(["d1"])
+
+
+def test_delete_by_metadata_field_handles_missing_table_error():
+    vector = AnalyticdbVectorBySql.__new__(AnalyticdbVectorBySql)
+    vector.table_name = "dify.collection"
+    cursor = MagicMock()
+
+    @contextmanager
+    def _cursor_context():
+        yield cursor
+
+    vector._get_cursor = _cursor_context
+    cursor.execute.side_effect = psycopg2.errors.UndefinedTable("relation does not exist")
+    vector.delete_by_metadata_field("document_id", "doc-1")
+
+
+@pytest.mark.parametrize("invalid_top_k", [0, "x", -1])
+def test_search_by_vector_validates_top_k(invalid_top_k):
+    vector = AnalyticdbVectorBySql.__new__(AnalyticdbVectorBySql)
+    vector.table_name = "dify.collection"
+
+    with pytest.raises(ValueError, match="top_k must be a positive integer"):
+        vector.search_by_vector([0.1, 0.2], top_k=invalid_top_k)
+
+
+def test_search_by_vector_returns_documents_above_threshold():
+    vector = AnalyticdbVectorBySql.__new__(AnalyticdbVectorBySql)
+    vector.table_name = "dify.collection"
+    cursor = MagicMock()
+    cursor.__iter__.return_value = iter(
+        [
+            ("id1", [1.0], 0.8, "content 1", {"doc_id": "id1", "document_id": "doc-1"}),
+            ("id2", [2.0], 0.3, "content 2", {"doc_id": "id2", "document_id": "doc-2"}),
+        ]
+    )
+
+    @contextmanager
+    def _cursor_context():
+        yield cursor
+
+    vector._get_cursor = _cursor_context
+
+    docs = vector.search_by_vector([0.1, 0.2], top_k=2, score_threshold=0.5, document_ids_filter=["doc-1"])
+
+    assert len(docs) == 1
+    assert docs[0].page_content == "content 1"
+    assert docs[0].metadata["score"] == 0.8
+
+
+@pytest.mark.parametrize("invalid_top_k", [0, "x", -1])
+def test_search_by_full_text_validates_top_k(invalid_top_k):
+    vector = AnalyticdbVectorBySql.__new__(AnalyticdbVectorBySql)
+    vector.table_name = "dify.collection"
+
+    with pytest.raises(ValueError, match="top_k must be a positive integer"):
+        vector.search_by_full_text("query", top_k=invalid_top_k)
+
+
+def test_search_by_full_text_returns_documents():
+    vector = AnalyticdbVectorBySql.__new__(AnalyticdbVectorBySql)
+    vector.table_name = "dify.collection"
+    cursor = MagicMock()
+    cursor.__iter__.return_value = iter(
+        [
+            ("id1", [1.0], "content 1", {"doc_id": "id1", "document_id": "doc-1"}, 0.9),
+        ]
+    )
+
+    @contextmanager
+    def _cursor_context():
+        yield cursor
+
+    vector._get_cursor = _cursor_context
+    docs = vector.search_by_full_text("query", top_k=1, document_ids_filter=["doc-1"])
+
+    assert len(docs) == 1
+    assert docs[0].metadata["score"] == 0.9
+    assert docs[0].page_content == "content 1"
+
+
+def test_delete_drops_table():
+    vector = AnalyticdbVectorBySql.__new__(AnalyticdbVectorBySql)
+    vector.table_name = "dify.collection"
+    cursor = MagicMock()
+
+    @contextmanager
+    def _cursor_context():
+        yield cursor
+
+    vector._get_cursor = _cursor_context
+    vector.delete()
+
+    cursor.execute.assert_called_once()
+
+
+def test_init_normalizes_collection_name_and_creates_pool_when_missing(monkeypatch):
+    config = AnalyticdbVectorBySqlConfig(**_config_values())
+    created_pool = MagicMock()
+
+    monkeypatch.setattr(AnalyticdbVectorBySql, "_initialize", MagicMock())
+    monkeypatch.setattr(AnalyticdbVectorBySql, "_create_connection_pool", MagicMock(return_value=created_pool))
+
+    vector = AnalyticdbVectorBySql("My_Collection", config)
+
+    assert vector._collection_name == "my_collection"
+    assert vector.table_name == "dify.my_collection"
+    assert vector.databaseName == "knowledgebase"
+    assert vector.pool is created_pool
+
+
+def test_initialize_vector_database_handles_existing_database_and_search_config(monkeypatch):
+    vector = AnalyticdbVectorBySql.__new__(AnalyticdbVectorBySql)
+    vector.config = AnalyticdbVectorBySqlConfig(**_config_values())
+    vector.databaseName = "knowledgebase"
+
+    bootstrap_cursor = MagicMock()
+    bootstrap_connection = MagicMock()
+    bootstrap_connection.cursor.return_value = bootstrap_cursor
+    bootstrap_cursor.execute.side_effect = RuntimeError("database already exists")
+    monkeypatch.setattr(sql_module.psycopg2, "connect", MagicMock(return_value=bootstrap_connection))
+
+    worker_cursor = MagicMock()
+    worker_connection = MagicMock()
+    worker_cursor.connection = worker_connection
+
+    def _execute(sql, *args, **kwargs):
+        if "CREATE TEXT SEARCH CONFIGURATION zh_cn" in sql:
+            raise RuntimeError("already exists")
+
+    worker_cursor.execute.side_effect = _execute
+    pooled_connection = MagicMock()
+    pooled_connection.cursor.return_value = worker_cursor
+    pool = MagicMock()
+    pool.getconn.return_value = pooled_connection
+    vector._create_connection_pool = MagicMock(return_value=pool)
+
+    vector._initialize_vector_database()
+
+    bootstrap_cursor.close.assert_called_once()
+    bootstrap_connection.close.assert_called_once()
+    vector._create_connection_pool.assert_called_once()
+    assert any(
+        "CREATE OR REPLACE FUNCTION public.to_tsquery_from_text" in call.args[0]
+        for call in worker_cursor.execute.call_args_list
+    )
+    assert any("CREATE SCHEMA IF NOT EXISTS dify" in call.args[0] for call in worker_cursor.execute.call_args_list)
+
+
+def test_initialize_vector_database_raises_runtime_error_when_zhparser_fails(monkeypatch):
+    vector = AnalyticdbVectorBySql.__new__(AnalyticdbVectorBySql)
+    vector.config = AnalyticdbVectorBySqlConfig(**_config_values())
+    vector.databaseName = "knowledgebase"
+
+    bootstrap_cursor = MagicMock()
+    bootstrap_connection = MagicMock()
+    bootstrap_connection.cursor.return_value = bootstrap_cursor
+    monkeypatch.setattr(sql_module.psycopg2, "connect", MagicMock(return_value=bootstrap_connection))
+
+    worker_cursor = MagicMock()
+    worker_connection = MagicMock()
+    worker_cursor.connection = worker_connection
+    worker_cursor.execute.side_effect = RuntimeError("zhparser unavailable")
+
+    pooled_connection = MagicMock()
+    pooled_connection.cursor.return_value = worker_cursor
+    pool = MagicMock()
+    pool.getconn.return_value = pooled_connection
+    vector._create_connection_pool = MagicMock(return_value=pool)
+
+    with pytest.raises(RuntimeError, match="Failed to create zhparser extension"):
+        vector._initialize_vector_database()
+
+    worker_connection.rollback.assert_called_once()
+
+
+def test_create_collection_if_not_exists_creates_table_indexes_and_cache(monkeypatch):
+    vector = AnalyticdbVectorBySql.__new__(AnalyticdbVectorBySql)
+    vector.config = AnalyticdbVectorBySqlConfig(**_config_values())
+    vector._collection_name = "collection"
+    vector.table_name = "dify.collection"
+
+    lock = MagicMock()
+    lock.__enter__.return_value = None
+    lock.__exit__.return_value = None
+    monkeypatch.setattr(sql_module.redis_client, "lock", MagicMock(return_value=lock))
+    monkeypatch.setattr(sql_module.redis_client, "get", MagicMock(return_value=None))
+    monkeypatch.setattr(sql_module.redis_client, "set", MagicMock())
+
+    cursor = MagicMock()
+
+    @contextmanager
+    def _cursor_context():
+        yield cursor
+
+    vector._get_cursor = _cursor_context
+
+    vector._create_collection_if_not_exists(embedding_dimension=3)
+
+    assert any("CREATE TABLE IF NOT EXISTS dify.collection" in call.args[0] for call in cursor.execute.call_args_list)
+    assert any("CREATE INDEX collection_embedding_idx" in call.args[0] for call in cursor.execute.call_args_list)
+    sql_module.redis_client.set.assert_called_once()
+
+
+def test_create_collection_if_not_exists_raises_for_non_existing_error(monkeypatch):
+    vector = AnalyticdbVectorBySql.__new__(AnalyticdbVectorBySql)
+    vector.config = AnalyticdbVectorBySqlConfig(**_config_values())
+    vector._collection_name = "collection"
+    vector.table_name = "dify.collection"
+
+    lock = MagicMock()
+    lock.__enter__.return_value = None
+    lock.__exit__.return_value = None
+    monkeypatch.setattr(sql_module.redis_client, "lock", MagicMock(return_value=lock))
+    monkeypatch.setattr(sql_module.redis_client, "get", MagicMock(return_value=None))
+    monkeypatch.setattr(sql_module.redis_client, "set", MagicMock())
+
+    cursor = MagicMock()
+    cursor.execute.side_effect = RuntimeError("permission denied")
+
+    @contextmanager
+    def _cursor_context():
+        yield cursor
+
+    vector._get_cursor = _cursor_context
+
+    with pytest.raises(RuntimeError, match="permission denied"):
+        vector._create_collection_if_not_exists(embedding_dimension=3)
+
+
+def test_delete_methods_raise_when_error_is_not_missing_table():
+    vector = AnalyticdbVectorBySql.__new__(AnalyticdbVectorBySql)
+    vector.table_name = "dify.collection"
+    cursor = MagicMock()
+
+    @contextmanager
+    def _cursor_context():
+        yield cursor
+
+    vector._get_cursor = _cursor_context
+
+    cursor.execute.side_effect = RuntimeError("unexpected delete failure")
+    with pytest.raises(RuntimeError, match="unexpected delete failure"):
+        vector.delete_by_ids(["doc-1"])
+
+    cursor.execute.side_effect = RuntimeError("unexpected metadata failure")
+    with pytest.raises(RuntimeError, match="unexpected metadata failure"):
+        vector.delete_by_metadata_field("document_id", "doc-1")

api/tests/unit_tests/core/rag/datasource/vdb/baidu/test_baidu_vector.py

@@ -0,0 +1,542 @@
+import importlib
+import sys
+import types
+from types import SimpleNamespace
+from unittest.mock import MagicMock, patch
+
+import pytest
+
+from core.rag.models.document import Document
+
+
+def _build_fake_pymochow_modules():
+    pymochow = types.ModuleType("pymochow")
+    pymochow.__path__ = []
+    pymochow_auth = types.ModuleType("pymochow.auth")
+    pymochow_auth.__path__ = []
+    pymochow_credentials = types.ModuleType("pymochow.auth.bce_credentials")
+    pymochow_configuration = types.ModuleType("pymochow.configuration")
+    pymochow_exception = types.ModuleType("pymochow.exception")
+    pymochow_model = types.ModuleType("pymochow.model")
+    pymochow_model.__path__ = []
+    pymochow_model_database = types.ModuleType("pymochow.model.database")
+    pymochow_model_enum = types.ModuleType("pymochow.model.enum")
+    pymochow_model_schema = types.ModuleType("pymochow.model.schema")
+    pymochow_model_table = types.ModuleType("pymochow.model.table")
+
+    class _SimpleObject:
+        def __init__(self, *args, **kwargs):
+            self.args = args
+            for key, value in kwargs.items():
+                setattr(self, key, value)
+
+    class ServerError(Exception):
+        def __init__(self, code):
+            super().__init__(f"server error {code}")
+            self.code = code
+
+    class ServerErrCode:
+        TABLE_NOT_EXIST = 1001
+        DB_ALREADY_EXIST = 1002
+
+    class IndexType:
+        __members__ = {"HNSW": "HNSW"}
+
+    class MetricType:
+        __members__ = {"IP": "IP"}
+
+    class IndexState:
+        NORMAL = "NORMAL"
+
+    class TableState:
+        NORMAL = "NORMAL"
+
+    class InvertedIndexAnalyzer:
+        DEFAULT_ANALYZER = "DEFAULT_ANALYZER"
+
+    class InvertedIndexParseMode:
+        COARSE_MODE = "COARSE_MODE"
+
+    class InvertedIndexFieldAttribute:
+        ANALYZED = "ANALYZED"
+
+    class FieldType:
+        STRING = "STRING"
+        TEXT = "TEXT"
+        JSON = "JSON"
+        FLOAT_VECTOR = "FLOAT_VECTOR"
+
+    pymochow.MochowClient = _SimpleObject
+    pymochow_credentials.BceCredentials = _SimpleObject
+    pymochow_configuration.Configuration = _SimpleObject
+    pymochow_exception.ServerError = ServerError
+    pymochow_model_database.Database = _SimpleObject
+
+    pymochow_model_enum.FieldType = FieldType
+    pymochow_model_enum.IndexState = IndexState
+    pymochow_model_enum.IndexType = IndexType
+    pymochow_model_enum.MetricType = MetricType
+    pymochow_model_enum.ServerErrCode = ServerErrCode
+    pymochow_model_enum.TableState = TableState
+
+    for cls_name in [
+        "AutoBuildRowCountIncrement",
+        "Field",
+        "FilteringIndex",
+        "HNSWParams",
+        "InvertedIndex",
+        "InvertedIndexParams",
+        "Schema",
+        "VectorIndex",
+    ]:
+        setattr(pymochow_model_schema, cls_name, _SimpleObject)
+    pymochow_model_schema.InvertedIndexAnalyzer = InvertedIndexAnalyzer
+    pymochow_model_schema.InvertedIndexFieldAttribute = InvertedIndexFieldAttribute
+    pymochow_model_schema.InvertedIndexParseMode = InvertedIndexParseMode
+
+    for cls_name in ["AnnSearch", "BM25SearchRequest", "HNSWSearchParams", "Partition", "Row"]:
+        setattr(pymochow_model_table, cls_name, _SimpleObject)
+
+    pymochow.auth = pymochow_auth
+    pymochow.model = pymochow_model
+    pymochow_auth.bce_credentials = pymochow_credentials
+    pymochow_model.database = pymochow_model_database
+    pymochow_model.enum = pymochow_model_enum
+    pymochow_model.schema = pymochow_model_schema
+    pymochow_model.table = pymochow_model_table
+
+    modules = {
+        "pymochow": pymochow,
+        "pymochow.auth": pymochow_auth,
+        "pymochow.auth.bce_credentials": pymochow_credentials,
+        "pymochow.configuration": pymochow_configuration,
+        "pymochow.exception": pymochow_exception,
+        "pymochow.model": pymochow_model,
+        "pymochow.model.database": pymochow_model_database,
+        "pymochow.model.enum": pymochow_model_enum,
+        "pymochow.model.schema": pymochow_model_schema,
+        "pymochow.model.table": pymochow_model_table,
+    }
+    return modules
+
+
+@pytest.fixture
+def baidu_module(monkeypatch):
+    for name, module in _build_fake_pymochow_modules().items():
+        monkeypatch.setitem(sys.modules, name, module)
+    import core.rag.datasource.vdb.baidu.baidu_vector as module
+
+    return importlib.reload(module)
+
+
+def test_baidu_config_validation(baidu_module):
+    values = {
+        "endpoint": "https://example.com",
+        "account": "account",
+        "api_key": "key",
+        "database": "database",
+    }
+    config = baidu_module.BaiduConfig.model_validate(values)
+    assert config.endpoint == "https://example.com"
+
+    for key, error_message in [
+        ("endpoint", "BAIDU_VECTOR_DB_ENDPOINT"),
+        ("account", "BAIDU_VECTOR_DB_ACCOUNT"),
+        ("api_key", "BAIDU_VECTOR_DB_API_KEY"),
+        ("database", "BAIDU_VECTOR_DB_DATABASE"),
+    ]:
+        invalid = dict(values)
+        invalid[key] = ""
+        with pytest.raises(ValueError, match=error_message):
+            baidu_module.BaiduConfig.model_validate(invalid)
+
+
+def test_get_search_result_handles_metadata_and_threshold(baidu_module):
+    vector = baidu_module.BaiduVector.__new__(baidu_module.BaiduVector)
+    response = SimpleNamespace(
+        rows=[
+            {"row": {"page_content": "doc1", "metadata": '{"document_id":"d1"}'}, "score": 0.9},
+            {"row": {"page_content": "doc2", "metadata": {"document_id": "d2"}}, "score": 0.4},
+            {"row": {"page_content": "doc3", "metadata": 123}, "score": 0.95},
+        ]
+    )
+
+    docs = vector._get_search_res(response, score_threshold=0.8)
+
+    assert len(docs) == 2
+    assert docs[0].page_content == "doc1"
+    assert docs[0].metadata["score"] == 0.9
+    assert docs[1].page_content == "doc3"
+
+
+def test_delete_by_ids_and_delete_by_metadata_field(baidu_module):
+    vector = baidu_module.BaiduVector.__new__(baidu_module.BaiduVector)
+    table = MagicMock()
+    vector._db = MagicMock()
+    vector._db.table.return_value = table
+    vector._collection_name = "collection_1"
+
+    vector.delete_by_ids([])
+    table.delete.assert_not_called()
+
+    vector.delete_by_ids(["id1", "id2"])
+    table.delete.assert_called_once()
+
+    table.delete.reset_mock()
+    vector.delete_by_metadata_field("source", 'abc"def')
+    delete_filter = table.delete.call_args.kwargs["filter"]
+    assert delete_filter == 'metadata["source"] = "abc\\"def"'
+
+
+def test_delete_handles_table_not_exist_error_and_raises_for_other_codes(baidu_module):
+    vector = baidu_module.BaiduVector.__new__(baidu_module.BaiduVector)
+    vector._collection_name = "collection_1"
+    vector._db = MagicMock()
+
+    vector._db.drop_table.side_effect = baidu_module.ServerError(baidu_module.ServerErrCode.TABLE_NOT_EXIST)
+    vector.delete()
+
+    vector._db.drop_table.side_effect = baidu_module.ServerError(9999)
+    with pytest.raises(baidu_module.ServerError):
+        vector.delete()
+
+
+def test_init_database_uses_existing_or_creates_when_missing(baidu_module):
+    vector = baidu_module.BaiduVector.__new__(baidu_module.BaiduVector)
+    vector._client = MagicMock()
+    vector._client_config = SimpleNamespace(database="my_db")
+
+    vector._client.list_databases.return_value = [SimpleNamespace(database_name="my_db")]
+    vector._client.database.return_value = "existing_db"
+    assert vector._init_database() == "existing_db"
+
+    vector._client.list_databases.return_value = []
+    vector._client.database.return_value = "created_db"
+    vector._client.create_database.side_effect = None
+    assert vector._init_database() == "created_db"
+
+    vector._client.create_database.side_effect = baidu_module.ServerError(baidu_module.ServerErrCode.DB_ALREADY_EXIST)
+    assert vector._init_database() == "created_db"
+
+
+def test_table_existed_checks_table_access(baidu_module):
+    vector = baidu_module.BaiduVector.__new__(baidu_module.BaiduVector)
+    vector._collection_name = "collection_1"
+    vector._db = MagicMock()
+    vector._db.table.return_value = MagicMock()
+
+    assert vector._table_existed() is True
+
+    vector._db.table.side_effect = baidu_module.ServerError(baidu_module.ServerErrCode.TABLE_NOT_EXIST)
+    assert vector._table_existed() is False
+
+    vector._db.table.side_effect = baidu_module.ServerError(9999)
+    with pytest.raises(baidu_module.ServerError):
+        vector._table_existed()
+
+
+def test_search_methods_delegate_to_database_table(baidu_module):
+    vector = baidu_module.BaiduVector.__new__(baidu_module.BaiduVector)
+    vector._collection_name = "collection_1"
+    vector._db = MagicMock()
+    vector._get_search_res = MagicMock(return_value=[Document(page_content="doc", metadata={"doc_id": "1"})])
+
+    table = MagicMock()
+    vector._db.table.return_value = table
+    table.search.return_value = "vector_result"
+    table.bm25_search.return_value = "bm25_result"
+
+    result1 = vector.search_by_vector([0.1, 0.2], top_k=3, document_ids_filter=["doc-1"], score_threshold=0.2)
+    result2 = vector.search_by_full_text("query", top_k=3, document_ids_filter=["doc-1"], score_threshold=0.2)
+
+    assert result1 == vector._get_search_res.return_value
+    assert result2 == vector._get_search_res.return_value
+    assert vector._get_search_res.call_count == 2
+
+
+def test_factory_initializes_collection_name_and_index_struct(baidu_module, monkeypatch):
+    factory = baidu_module.BaiduVectorFactory()
+    dataset = SimpleNamespace(id="dataset-1", index_struct_dict=None, index_struct=None)
+    monkeypatch.setattr(baidu_module.Dataset, "gen_collection_name_by_id", lambda _id: "AUTO_COLLECTION")
+    monkeypatch.setattr(baidu_module.dify_config, "BAIDU_VECTOR_DB_ENDPOINT", "https://endpoint")
+    monkeypatch.setattr(baidu_module.dify_config, "BAIDU_VECTOR_DB_CONNECTION_TIMEOUT_MS", 1000)
+    monkeypatch.setattr(baidu_module.dify_config, "BAIDU_VECTOR_DB_ACCOUNT", "account")
+    monkeypatch.setattr(baidu_module.dify_config, "BAIDU_VECTOR_DB_API_KEY", "key")
+    monkeypatch.setattr(baidu_module.dify_config, "BAIDU_VECTOR_DB_DATABASE", "database")
+    monkeypatch.setattr(baidu_module.dify_config, "BAIDU_VECTOR_DB_SHARD", 1)
+    monkeypatch.setattr(baidu_module.dify_config, "BAIDU_VECTOR_DB_REPLICAS", 1)
+    monkeypatch.setattr(baidu_module.dify_config, "BAIDU_VECTOR_DB_INVERTED_INDEX_ANALYZER", "DEFAULT_ANALYZER")
+    monkeypatch.setattr(baidu_module.dify_config, "BAIDU_VECTOR_DB_INVERTED_INDEX_PARSER_MODE", "COARSE_MODE")
+    monkeypatch.setattr(baidu_module.dify_config, "BAIDU_VECTOR_DB_AUTO_BUILD_ROW_COUNT_INCREMENT", 500)
+    monkeypatch.setattr(baidu_module.dify_config, "BAIDU_VECTOR_DB_AUTO_BUILD_ROW_COUNT_INCREMENT_RATIO", 0.05)
+    monkeypatch.setattr(baidu_module.dify_config, "BAIDU_VECTOR_DB_REBUILD_INDEX_TIMEOUT_IN_SECONDS", 300)
+
+    with patch.object(baidu_module, "BaiduVector", return_value="vector") as vector_cls:
+        result = factory.init_vector(dataset, attributes=[], embeddings=MagicMock())
+
+    assert result == "vector"
+    assert vector_cls.call_args.kwargs["collection_name"] == "auto_collection"
+    assert dataset.index_struct is not None
+
+
+def test_init_get_type_to_index_struct_and_create_delegate(baidu_module, monkeypatch):
+    init_client = MagicMock(return_value="client")
+    init_database = MagicMock(return_value="database")
+    monkeypatch.setattr(baidu_module.BaiduVector, "_init_client", init_client)
+    monkeypatch.setattr(baidu_module.BaiduVector, "_init_database", init_database)
+
+    config = baidu_module.BaiduConfig(
+        endpoint="https://example.com",
+        account="account",
+        api_key="key",
+        database="db",
+    )
+    vector = baidu_module.BaiduVector(collection_name="my_collection", config=config)
+
+    assert vector.get_type() == baidu_module.VectorType.BAIDU
+    assert vector.to_index_struct()["vector_store"]["class_prefix"] == "my_collection"
+    assert vector._client == "client"
+    assert vector._db == "database"
+
+    vector._create_table = MagicMock()
+    vector.add_texts = MagicMock()
+    docs = [Document(page_content="p1", metadata={"doc_id": "d1"})]
+    vector.create(docs, [[0.1, 0.2]])
+    vector._create_table.assert_called_once_with(2)
+    vector.add_texts.assert_called_once_with(docs, [[0.1, 0.2]])
+
+
+def test_add_texts_batches_rows(baidu_module):
+    vector = baidu_module.BaiduVector.__new__(baidu_module.BaiduVector)
+    vector._collection_name = "collection_1"
+    table = MagicMock()
+    vector._db = MagicMock()
+    vector._db.table.return_value = table
+
+    docs = [
+        Document(page_content="doc-1", metadata={"doc_id": "id-1", "document_id": "doc-1"}),
+        Document(page_content="doc-2", metadata={"doc_id": "id-2", "document_id": "doc-2"}),
+    ]
+    vector.add_texts(docs, [[0.1, 0.2], [0.3, 0.4]])
+
+    assert table.upsert.call_count == 1
+    inserted_rows = table.upsert.call_args.kwargs["rows"]
+    assert len(inserted_rows) == 2
+
+
+def test_add_texts_batches_more_than_batch_size(baidu_module):
+    vector = baidu_module.BaiduVector.__new__(baidu_module.BaiduVector)
+    vector._collection_name = "collection_1"
+    table = MagicMock()
+    vector._db = MagicMock()
+    vector._db.table.return_value = table
+
+    docs = [
+        Document(page_content=f"doc-{idx}", metadata={"doc_id": f"id-{idx}", "document_id": f"doc-{idx}"})
+        for idx in range(1001)
+    ]
+    embeddings = [[0.1, 0.2] for _ in range(1001)]
+
+    vector.add_texts(docs, embeddings)
+
+    assert table.upsert.call_count == 2
+    assert len(table.upsert.call_args_list[0].kwargs["rows"]) == 1000
+    assert len(table.upsert.call_args_list[1].kwargs["rows"]) == 1
+
+
+def test_text_exists_returns_false_when_query_code_is_not_success(baidu_module):
+    vector = baidu_module.BaiduVector.__new__(baidu_module.BaiduVector)
+    vector._collection_name = "collection_1"
+    table = MagicMock()
+    vector._db = MagicMock()
+    vector._db.table.return_value = table
+
+    table.query.return_value = SimpleNamespace(code=0)
+    assert vector.text_exists("id-1") is True
+
+    table.query.return_value = SimpleNamespace(code=1)
+    assert vector.text_exists("id-1") is False
+
+    table.query.return_value = None
+    assert vector.text_exists("id-1") is False
+
+
+def test_get_search_result_handles_invalid_metadata_json(baidu_module):
+    vector = baidu_module.BaiduVector.__new__(baidu_module.BaiduVector)
+    response = SimpleNamespace(rows=[{"row": {"page_content": "doc1", "metadata": "{bad json"}, "score": 0.7}])
+
+    docs = vector._get_search_res(response, score_threshold=0.1)
+
+    assert len(docs) == 1
+    assert docs[0].metadata["score"] == 0.7
+    assert "document_id" not in docs[0].metadata
+
+
+def test_init_client_constructs_configuration_and_client(baidu_module, monkeypatch):
+    credentials = MagicMock(return_value="credentials")
+    configuration = MagicMock(return_value="configuration")
+    client_cls = MagicMock(return_value="client")
+    monkeypatch.setattr(baidu_module, "BceCredentials", credentials)
+    monkeypatch.setattr(baidu_module, "Configuration", configuration)
+    monkeypatch.setattr(baidu_module, "MochowClient", client_cls)
+
+    vector = baidu_module.BaiduVector.__new__(baidu_module.BaiduVector)
+    config = SimpleNamespace(account="account", api_key="key", endpoint="https://endpoint")
+
+    client = vector._init_client(config)
+
+    assert client == "client"
+    credentials.assert_called_once_with("account", "key")
+    configuration.assert_called_once_with(credentials="credentials", endpoint="https://endpoint")
+    client_cls.assert_called_once_with("configuration")
+
+
+def test_init_database_raises_for_unknown_create_database_error(baidu_module):
+    vector = baidu_module.BaiduVector.__new__(baidu_module.BaiduVector)
+    vector._client = MagicMock()
+    vector._client_config = SimpleNamespace(database="my_db")
+    vector._client.list_databases.return_value = []
+    vector._client.create_database.side_effect = baidu_module.ServerError(9999)
+
+    with pytest.raises(baidu_module.ServerError):
+        vector._init_database()
+
+
+def test_create_table_handles_cache_and_validation_paths(baidu_module, monkeypatch):
+    vector = baidu_module.BaiduVector.__new__(baidu_module.BaiduVector)
+    vector._collection_name = "collection_1"
+    vector._client_config = SimpleNamespace(
+        index_type="HNSW",
+        metric_type="IP",
+        inverted_index_analyzer="DEFAULT_ANALYZER",
+        inverted_index_parser_mode="COARSE_MODE",
+        auto_build_row_count_increment=500,
+        auto_build_row_count_increment_ratio=0.05,
+        rebuild_index_timeout_in_seconds=300,
+        replicas=1,
+        shard=1,
+    )
+    vector._db = MagicMock()
+    table = MagicMock()
+    table.state = baidu_module.TableState.NORMAL
+    vector._db.describe_table.return_value = table
+    vector._table_existed = MagicMock(return_value=False)
+    vector.delete = MagicMock()
+
+    lock = MagicMock()
+    lock.__enter__.return_value = None
+    lock.__exit__.return_value = None
+    monkeypatch.setattr(baidu_module.redis_client, "lock", MagicMock(return_value=lock))
+    monkeypatch.setattr(baidu_module.redis_client, "set", MagicMock())
+    monkeypatch.setattr(baidu_module.time, "sleep", lambda _s: None)
+    monkeypatch.setattr(vector, "_wait_for_index_ready", MagicMock())
+
+    # Cached table skips all work.
+    monkeypatch.setattr(baidu_module.redis_client, "get", MagicMock(return_value=1))
+    vector._create_table(3)
+    vector._db.create_table.assert_not_called()
+
+    # Existing table also skips creation.
+    monkeypatch.setattr(baidu_module.redis_client, "get", MagicMock(return_value=None))
+    vector._table_existed.return_value = True
+    vector._create_table(3)
+    vector._db.create_table.assert_not_called()
+
+    # Create table when cache is empty and table does not exist.
+    vector._table_existed.return_value = False
+    vector._create_table(3)
+    vector._db.create_table.assert_called_once()
+    baidu_module.redis_client.set.assert_called_once_with("vector_indexing_collection_1", 1, ex=3600)
+    table.rebuild_index.assert_called_once_with(vector.vector_index)
+    vector._wait_for_index_ready.assert_called_once_with(table, 3600)
+
+
+def test_create_table_raises_for_invalid_index_or_metric(baidu_module, monkeypatch):
+    vector = baidu_module.BaiduVector.__new__(baidu_module.BaiduVector)
+    vector._collection_name = "collection_1"
+    vector._db = MagicMock()
+    vector._table_existed = MagicMock(return_value=False)
+    vector.delete = MagicMock()
+    vector._client_config = SimpleNamespace(
+        index_type="INVALID",
+        metric_type="IP",
+        inverted_index_analyzer="DEFAULT_ANALYZER",
+        inverted_index_parser_mode="COARSE_MODE",
+        auto_build_row_count_increment=500,
+        auto_build_row_count_increment_ratio=0.05,
+        rebuild_index_timeout_in_seconds=300,
+        replicas=1,
+        shard=1,
+    )
+
+    lock = MagicMock()
+    lock.__enter__.return_value = None
+    lock.__exit__.return_value = None
+    monkeypatch.setattr(baidu_module.redis_client, "lock", MagicMock(return_value=lock))
+    monkeypatch.setattr(baidu_module.redis_client, "get", MagicMock(return_value=None))
+
+    with pytest.raises(ValueError, match="unsupported index_type"):
+        vector._create_table(3)
+
+    vector._client_config.index_type = "HNSW"
+    vector._client_config.metric_type = "INVALID"
+    with pytest.raises(ValueError, match="unsupported metric_type"):
+        vector._create_table(3)
+
+
+def test_create_table_raises_timeout_if_table_never_becomes_normal(baidu_module, monkeypatch):
+    vector = baidu_module.BaiduVector.__new__(baidu_module.BaiduVector)
+    vector._collection_name = "collection_1"
+    vector._client_config = SimpleNamespace(
+        index_type="HNSW",
+        metric_type="IP",
+        inverted_index_analyzer="DEFAULT_ANALYZER",
+        inverted_index_parser_mode="COARSE_MODE",
+        auto_build_row_count_increment=500,
+        auto_build_row_count_increment_ratio=0.05,
+        rebuild_index_timeout_in_seconds=300,
+        replicas=1,
+        shard=1,
+    )
+    vector._db = MagicMock()
+    vector._db.describe_table.return_value = SimpleNamespace(state="CREATING")
+    vector._table_existed = MagicMock(return_value=False)
+    vector.delete = MagicMock()
+
+    lock = MagicMock()
+    lock.__enter__.return_value = None
+    lock.__exit__.return_value = None
+    monkeypatch.setattr(baidu_module.redis_client, "lock", MagicMock(return_value=lock))
+    monkeypatch.setattr(baidu_module.redis_client, "get", MagicMock(return_value=None))
+    monkeypatch.setattr(baidu_module.time, "sleep", lambda _s: None)
+    monkeypatch.setattr(baidu_module.time, "time", MagicMock(side_effect=[0, 301]))
+
+    with pytest.raises(TimeoutError, match="Table creation timeout"):
+        vector._create_table(3)
+
+
+def test_factory_uses_existing_collection_prefix_when_index_struct_exists(baidu_module, monkeypatch):
+    factory = baidu_module.BaiduVectorFactory()
+    dataset = SimpleNamespace(
+        id="dataset-1",
+        index_struct_dict={"vector_store": {"class_prefix": "EXISTING_COLLECTION"}},
+        index_struct=None,
+    )
+    monkeypatch.setattr(baidu_module.dify_config, "BAIDU_VECTOR_DB_ENDPOINT", "https://endpoint")
+    monkeypatch.setattr(baidu_module.dify_config, "BAIDU_VECTOR_DB_CONNECTION_TIMEOUT_MS", 1000)
+    monkeypatch.setattr(baidu_module.dify_config, "BAIDU_VECTOR_DB_ACCOUNT", "account")
+    monkeypatch.setattr(baidu_module.dify_config, "BAIDU_VECTOR_DB_API_KEY", "key")
+    monkeypatch.setattr(baidu_module.dify_config, "BAIDU_VECTOR_DB_DATABASE", "database")
+    monkeypatch.setattr(baidu_module.dify_config, "BAIDU_VECTOR_DB_SHARD", 1)
+    monkeypatch.setattr(baidu_module.dify_config, "BAIDU_VECTOR_DB_REPLICAS", 1)
+    monkeypatch.setattr(baidu_module.dify_config, "BAIDU_VECTOR_DB_INVERTED_INDEX_ANALYZER", "DEFAULT_ANALYZER")
+    monkeypatch.setattr(baidu_module.dify_config, "BAIDU_VECTOR_DB_INVERTED_INDEX_PARSER_MODE", "COARSE_MODE")
+    monkeypatch.setattr(baidu_module.dify_config, "BAIDU_VECTOR_DB_AUTO_BUILD_ROW_COUNT_INCREMENT", 500)
+    monkeypatch.setattr(baidu_module.dify_config, "BAIDU_VECTOR_DB_AUTO_BUILD_ROW_COUNT_INCREMENT_RATIO", 0.05)
+    monkeypatch.setattr(baidu_module.dify_config, "BAIDU_VECTOR_DB_REBUILD_INDEX_TIMEOUT_IN_SECONDS", 300)
+
+    with patch.object(baidu_module, "BaiduVector", return_value="vector") as vector_cls:
+        result = factory.init_vector(dataset, attributes=[], embeddings=MagicMock())
+
+    assert result == "vector"
+    assert vector_cls.call_args.kwargs["collection_name"] == "existing_collection"

api/tests/unit_tests/core/rag/datasource/vdb/chroma/test_chroma_vector.py

@@ -0,0 +1,199 @@
+import importlib
+import sys
+import types
+from collections import UserDict
+from types import SimpleNamespace
+from unittest.mock import MagicMock, patch
+
+import pytest
+
+from core.rag.models.document import Document
+
+
+def _build_fake_chroma_modules():
+    chroma = types.ModuleType("chromadb")
+    chroma.DEFAULT_TENANT = "default_tenant"
+    chroma.DEFAULT_DATABASE = "default_database"
+
+    class Settings:
+        def __init__(self, **kwargs):
+            for key, value in kwargs.items():
+                setattr(self, key, value)
+
+    class QueryResult(UserDict):
+        pass
+
+    class _Collection:
+        def __init__(self):
+            self.upsert = MagicMock()
+            self.delete = MagicMock()
+            self.query = MagicMock()
+            self.get = MagicMock(return_value={})
+
+    class _Client:
+        def __init__(self, **kwargs):
+            self.kwargs = kwargs
+            self.collection = _Collection()
+            self.get_or_create_collection = MagicMock(return_value=self.collection)
+            self.delete_collection = MagicMock()
+
+    chroma.Settings = Settings
+    chroma.QueryResult = QueryResult
+    chroma.HttpClient = _Client
+    return chroma
+
+
+@pytest.fixture
+def chroma_module(monkeypatch):
+    fake_chroma = _build_fake_chroma_modules()
+    monkeypatch.setitem(sys.modules, "chromadb", fake_chroma)
+    import core.rag.datasource.vdb.chroma.chroma_vector as module
+
+    return importlib.reload(module)
+
+
+def test_chroma_config_to_params_builds_expected_payload(chroma_module):
+    config = chroma_module.ChromaConfig(
+        host="localhost",
+        port=8000,
+        tenant="tenant-1",
+        database="db-1",
+        auth_provider="provider",
+        auth_credentials="credentials",
+    )
+
+    params = config.to_chroma_params()
+
+    assert params["host"] == "localhost"
+    assert params["port"] == 8000
+    assert params["tenant"] == "tenant-1"
+    assert params["database"] == "db-1"
+    assert params["ssl"] is False
+    assert params["settings"].chroma_client_auth_provider == "provider"
+    assert params["settings"].chroma_client_auth_credentials == "credentials"
+
+
+def test_create_collection_uses_redis_lock_and_cache(chroma_module, monkeypatch):
+    lock = MagicMock()
+    lock.__enter__.return_value = None
+    lock.__exit__.return_value = None
+    monkeypatch.setattr(chroma_module.redis_client, "lock", MagicMock(return_value=lock))
+    monkeypatch.setattr(chroma_module.redis_client, "get", MagicMock(return_value=None))
+    monkeypatch.setattr(chroma_module.redis_client, "set", MagicMock())
+
+    vector = chroma_module.ChromaVector(
+        collection_name="collection_1",
+        config=chroma_module.ChromaConfig(host="localhost", port=8000, tenant="t", database="d"),
+    )
+    vector.create_collection("collection_1")
+
+    vector._client.get_or_create_collection.assert_called_once_with("collection_1")
+    chroma_module.redis_client.set.assert_called_once()
+
+
+def test_create_with_empty_texts_is_noop(chroma_module):
+    vector = chroma_module.ChromaVector(
+        collection_name="collection_1",
+        config=chroma_module.ChromaConfig(host="localhost", port=8000, tenant="t", database="d"),
+    )
+    vector.create([], [])
+    vector._client.get_or_create_collection.assert_not_called()
+
+
+def test_create_with_texts_creates_collection_and_upserts(chroma_module):
+    vector = chroma_module.ChromaVector(
+        collection_name="collection_1",
+        config=chroma_module.ChromaConfig(host="localhost", port=8000, tenant="t", database="d"),
+    )
+    docs = [Document(page_content="hello", metadata={"doc_id": "d1", "document_id": "doc-1"})]
+    vector.create(docs, [[0.1, 0.2]])
+
+    vector._client.get_or_create_collection.assert_called()
+    vector._client.collection.upsert.assert_called_once()
+
+
+def test_delete_methods_and_text_exists(chroma_module):
+    vector = chroma_module.ChromaVector(
+        collection_name="collection_1",
+        config=chroma_module.ChromaConfig(host="localhost", port=8000, tenant="t", database="d"),
+    )
+
+    vector.delete_by_ids([])
+    vector._client.collection.delete.assert_not_called()
+
+    vector.delete_by_ids(["id-1"])
+    vector._client.collection.delete.assert_called_with(ids=["id-1"])
+
+    vector.delete_by_metadata_field("document_id", "doc-1")
+    vector._client.collection.delete.assert_called_with(where={"document_id": {"$eq": "doc-1"}})
+
+    vector._client.collection.get.return_value = {"ids": ["id-1"]}
+    assert vector.text_exists("id-1") is True
+    vector._client.collection.get.return_value = {}
+    assert vector.text_exists("id-2") is False
+
+    vector.delete()
+    vector._client.delete_collection.assert_called_once_with("collection_1")
+
+
+def test_search_by_vector_handles_empty_results(chroma_module):
+    vector = chroma_module.ChromaVector(
+        collection_name="collection_1",
+        config=chroma_module.ChromaConfig(host="localhost", port=8000, tenant="t", database="d"),
+    )
+    vector._client.collection.query.return_value = {"ids": [], "documents": [], "metadatas": [], "distances": []}
+
+    assert vector.search_by_vector([0.1, 0.2], top_k=2) == []
+
+
+def test_search_by_vector_applies_score_threshold_and_sorting(chroma_module):
+    vector = chroma_module.ChromaVector(
+        collection_name="collection_1",
+        config=chroma_module.ChromaConfig(host="localhost", port=8000, tenant="t", database="d"),
+    )
+    vector._client.collection.query.return_value = {
+        "ids": [["id-1", "id-2"]],
+        "documents": [["doc high", "doc low"]],
+        "metadatas": [[{"doc_id": "id-1"}, {"doc_id": "id-2"}]],
+        "distances": [[0.1, 0.8]],
+    }
+
+    docs = vector.search_by_vector([0.1, 0.2], top_k=2, score_threshold=0.5, document_ids_filter=["doc-1"])
+
+    assert len(docs) == 1
+    assert docs[0].page_content == "doc high"
+    assert docs[0].metadata["score"] == 0.9
+
+
+def test_search_by_full_text_returns_empty_list(chroma_module):
+    vector = chroma_module.ChromaVector(
+        collection_name="collection_1",
+        config=chroma_module.ChromaConfig(host="localhost", port=8000, tenant="t", database="d"),
+    )
+    assert vector.search_by_full_text("query") == []
+
+
+def test_factory_init_vector_uses_existing_or_generated_collection(chroma_module, monkeypatch):
+    factory = chroma_module.ChromaVectorFactory()
+    dataset_with_index = SimpleNamespace(
+        id="dataset-1", index_struct_dict={"vector_store": {"class_prefix": "EXISTING"}}, index_struct=None
+    )
+    dataset_without_index = SimpleNamespace(id="dataset-2", index_struct_dict=None, index_struct=None)
+
+    monkeypatch.setattr(chroma_module.Dataset, "gen_collection_name_by_id", lambda _id: "AUTO_COLLECTION")
+    monkeypatch.setattr(chroma_module.dify_config, "CHROMA_HOST", "localhost")
+    monkeypatch.setattr(chroma_module.dify_config, "CHROMA_PORT", 8000)
+    monkeypatch.setattr(chroma_module.dify_config, "CHROMA_TENANT", None)
+    monkeypatch.setattr(chroma_module.dify_config, "CHROMA_DATABASE", None)
+    monkeypatch.setattr(chroma_module.dify_config, "CHROMA_AUTH_PROVIDER", None)
+    monkeypatch.setattr(chroma_module.dify_config, "CHROMA_AUTH_CREDENTIALS", None)
+
+    with patch.object(chroma_module, "ChromaVector", return_value="vector") as vector_cls:
+        result_1 = factory.init_vector(dataset_with_index, attributes=[], embeddings=MagicMock())
+        result_2 = factory.init_vector(dataset_without_index, attributes=[], embeddings=MagicMock())
+
+    assert result_1 == "vector"
+    assert result_2 == "vector"
+    assert vector_cls.call_args_list[0].kwargs["collection_name"] == "existing"
+    assert vector_cls.call_args_list[1].kwargs["collection_name"] == "auto_collection"
+    assert dataset_without_index.index_struct is not None

api/tests/unit_tests/core/rag/datasource/vdb/clickzetta/test_clickzetta_vector.py

@@ -0,0 +1,927 @@
+import importlib
+import queue
+import sys
+import types
+from contextlib import contextmanager
+from types import SimpleNamespace
+from unittest.mock import MagicMock, patch
+
+import pytest
+
+from core.rag.models.document import Document
+
+
+def _build_fake_clickzetta_module():
+    clickzetta = types.ModuleType("clickzetta")
+
+    class _FakeCursor:
+        def __init__(self):
+            self.execute = MagicMock()
+            self.executemany = MagicMock()
+            self.fetchall = MagicMock(return_value=[])
+            self.fetchone = MagicMock(return_value=(0,))
+
+        def __enter__(self):
+            return self
+
+        def __exit__(self, exc_type, exc, tb):
+            return False
+
+    class _FakeConnection:
+        def __init__(self):
+            self.cursor_obj = _FakeCursor()
+
+        def cursor(self):
+            return self.cursor_obj
+
+        def close(self):
+            return None
+
+    def connect(**_kwargs):
+        return _FakeConnection()
+
+    clickzetta.connect = connect
+    return clickzetta
+
+
+@pytest.fixture
+def clickzetta_module(monkeypatch):
+    monkeypatch.setitem(sys.modules, "clickzetta", _build_fake_clickzetta_module())
+    import core.rag.datasource.vdb.clickzetta.clickzetta_vector as module
+
+    return importlib.reload(module)
+
+
+def _config(module):
+    return module.ClickzettaConfig(
+        username="username",
+        password="password",
+        instance="instance",
+        service="service",
+        workspace="workspace",
+        vcluster="cluster",
+        schema_name="dify",
+    )
+
+
+@pytest.mark.parametrize(
+    ("field", "error_message"),
+    [
+        ("username", "CLICKZETTA_USERNAME"),
+        ("password", "CLICKZETTA_PASSWORD"),
+        ("instance", "CLICKZETTA_INSTANCE"),
+        ("service", "CLICKZETTA_SERVICE"),
+        ("workspace", "CLICKZETTA_WORKSPACE"),
+        ("vcluster", "CLICKZETTA_VCLUSTER"),
+        ("schema_name", "CLICKZETTA_SCHEMA"),
+    ],
+)
+def test_clickzetta_config_validation(clickzetta_module, field, error_message):
+    values = _config(clickzetta_module).model_dump()
+    values[field] = ""
+    with pytest.raises(ValueError, match=error_message):
+        clickzetta_module.ClickzettaConfig.model_validate(values)
+
+
+def test_parse_metadata_handles_valid_double_encoded_and_invalid_json(clickzetta_module):
+    vector = clickzetta_module.ClickzettaVector.__new__(clickzetta_module.ClickzettaVector)
+
+    parsed = vector._parse_metadata('{"document_id":"doc-1"}', "row-1")
+    assert parsed["doc_id"] == "row-1"
+    assert parsed["document_id"] == "doc-1"
+
+    parsed_double = vector._parse_metadata('"{\\"document_id\\": \\"doc-2\\"}"', "row-2")
+    assert parsed_double["doc_id"] == "row-2"
+    assert parsed_double["document_id"] == "doc-2"
+
+    parsed_fallback = vector._parse_metadata("not-json", "row-3")
+    assert parsed_fallback["doc_id"] == "row-3"
+    assert parsed_fallback["document_id"] == "row-3"
+
+
+def test_safe_doc_id_and_vector_format_helpers(clickzetta_module):
+    vector = clickzetta_module.ClickzettaVector.__new__(clickzetta_module.ClickzettaVector)
+
+    assert vector._format_vector_simple([0.1, 0.2, 0.3]) == "0.1,0.2,0.3"
+    assert vector._safe_doc_id("abc-123_DEF") == "abc-123_DEF"
+    assert vector._safe_doc_id("ab c;\n") == "abc"
+    assert len(vector._safe_doc_id("a" * 300)) == 255
+
+
+def test_table_exists_returns_false_for_not_found_and_other_exceptions(clickzetta_module):
+    vector = clickzetta_module.ClickzettaVector.__new__(clickzetta_module.ClickzettaVector)
+    vector._config = _config(clickzetta_module)
+    vector._table_name = "table_1"
+
+    @contextmanager
+    def _ctx_not_found():
+        connection = MagicMock()
+        cursor = MagicMock()
+        cursor.__enter__.return_value = cursor
+        cursor.__exit__.return_value = None
+        cursor.execute.side_effect = RuntimeError("CZLH-42000 table or view not found")
+        connection.cursor.return_value = cursor
+        yield connection
+
+    vector.get_connection_context = _ctx_not_found
+    assert vector._table_exists() is False
+
+    @contextmanager
+    def _ctx_other_error():
+        connection = MagicMock()
+        cursor = MagicMock()
+        cursor.__enter__.return_value = cursor
+        cursor.__exit__.return_value = None
+        cursor.execute.side_effect = RuntimeError("permission denied")
+        connection.cursor.return_value = cursor
+        yield connection
+
+    vector.get_connection_context = _ctx_other_error
+    assert vector._table_exists() is False
+
+
+def test_text_exists_handles_missing_table_and_existing_rows(clickzetta_module):
+    vector = clickzetta_module.ClickzettaVector.__new__(clickzetta_module.ClickzettaVector)
+    vector._config = _config(clickzetta_module)
+    vector._table_name = "table_1"
+
+    vector._table_exists = MagicMock(return_value=False)
+    assert vector.text_exists("doc-1") is False
+
+    vector._table_exists = MagicMock(return_value=True)
+
+    @contextmanager
+    def _ctx():
+        connection = MagicMock()
+        cursor = MagicMock()
+        cursor.__enter__.return_value = cursor
+        cursor.__exit__.return_value = None
+        cursor.fetchone.return_value = (1,)
+        connection.cursor.return_value = cursor
+        yield connection
+
+    vector.get_connection_context = _ctx
+    assert vector.text_exists("doc-1") is True
+
+
+def test_delete_by_ids_and_delete_by_metadata_field_short_circuit(clickzetta_module):
+    vector = clickzetta_module.ClickzettaVector.__new__(clickzetta_module.ClickzettaVector)
+    vector._config = _config(clickzetta_module)
+    vector._table_name = "table_1"
+    vector._execute_write = MagicMock()
+
+    vector.delete_by_ids([])
+    vector._execute_write.assert_not_called()
+
+    vector._table_exists = MagicMock(return_value=False)
+    vector.delete_by_ids(["doc-1"])
+    vector._execute_write.assert_not_called()
+
+    vector.delete_by_metadata_field("document_id", "doc-1")
+    vector._execute_write.assert_not_called()
+
+
+def test_search_short_circuit_behaviors(clickzetta_module):
+    vector = clickzetta_module.ClickzettaVector.__new__(clickzetta_module.ClickzettaVector)
+    vector._config = _config(clickzetta_module)
+    vector._table_name = "table_1"
+
+    vector._table_exists = MagicMock(return_value=False)
+    assert vector.search_by_vector([0.1, 0.2], top_k=2) == []
+
+    vector._config.enable_inverted_index = False
+    assert vector.search_by_full_text("query", top_k=2) == []
+
+
+def test_search_by_like_returns_documents_with_default_score(clickzetta_module):
+    vector = clickzetta_module.ClickzettaVector.__new__(clickzetta_module.ClickzettaVector)
+    vector._config = _config(clickzetta_module)
+    vector._table_name = "table_1"
+    vector._table_exists = MagicMock(return_value=True)
+    vector._parse_metadata = MagicMock(return_value={"document_id": "doc-1", "doc_id": "seg-1"})
+
+    @contextmanager
+    def _ctx():
+        connection = MagicMock()
+        cursor = MagicMock()
+        cursor.__enter__.return_value = cursor
+        cursor.__exit__.return_value = None
+        cursor.fetchall.return_value = [("seg-1", "content", '{"document_id":"doc-1"}')]
+        connection.cursor.return_value = cursor
+        yield connection
+
+    vector.get_connection_context = _ctx
+    docs = vector._search_by_like("query", top_k=3, document_ids_filter=["doc-1"])
+
+    assert len(docs) == 1
+    assert docs[0].page_content == "content"
+    assert docs[0].metadata["score"] == 0.5
+
+
+def test_factory_initializes_clickzetta_vector(clickzetta_module, monkeypatch):
+    factory = clickzetta_module.ClickzettaVectorFactory()
+    dataset = SimpleNamespace(id="dataset-1")
+
+    monkeypatch.setattr(clickzetta_module.Dataset, "gen_collection_name_by_id", lambda _id: "COLLECTION")
+    monkeypatch.setattr(clickzetta_module.dify_config, "CLICKZETTA_USERNAME", "username")
+    monkeypatch.setattr(clickzetta_module.dify_config, "CLICKZETTA_PASSWORD", "password")
+    monkeypatch.setattr(clickzetta_module.dify_config, "CLICKZETTA_INSTANCE", "instance")
+    monkeypatch.setattr(clickzetta_module.dify_config, "CLICKZETTA_SERVICE", "service")
+    monkeypatch.setattr(clickzetta_module.dify_config, "CLICKZETTA_WORKSPACE", "workspace")
+    monkeypatch.setattr(clickzetta_module.dify_config, "CLICKZETTA_VCLUSTER", "cluster")
+    monkeypatch.setattr(clickzetta_module.dify_config, "CLICKZETTA_SCHEMA", "dify")
+    monkeypatch.setattr(clickzetta_module.dify_config, "CLICKZETTA_BATCH_SIZE", 10)
+    monkeypatch.setattr(clickzetta_module.dify_config, "CLICKZETTA_ENABLE_INVERTED_INDEX", True)
+    monkeypatch.setattr(clickzetta_module.dify_config, "CLICKZETTA_ANALYZER_TYPE", "chinese")
+    monkeypatch.setattr(clickzetta_module.dify_config, "CLICKZETTA_ANALYZER_MODE", "smart")
+    monkeypatch.setattr(clickzetta_module.dify_config, "CLICKZETTA_VECTOR_DISTANCE_FUNCTION", "cosine_distance")
+
+    with patch.object(clickzetta_module, "ClickzettaVector", return_value="vector") as vector_cls:
+        result = factory.init_vector(dataset, attributes=[], embeddings=MagicMock())
+
+    assert result == "vector"
+    assert vector_cls.call_args.kwargs["collection_name"] == "collection"
+
+
+def test_connection_pool_singleton_and_config_key(clickzetta_module, monkeypatch):
+    clickzetta_module.ClickzettaConnectionPool._instance = None
+    monkeypatch.setattr(clickzetta_module.ClickzettaConnectionPool, "_start_cleanup_thread", MagicMock())
+
+    pool_1 = clickzetta_module.ClickzettaConnectionPool.get_instance()
+    pool_2 = clickzetta_module.ClickzettaConnectionPool.get_instance()
+    key = pool_1._get_config_key(_config(clickzetta_module))
+
+    assert pool_1 is pool_2
+    assert "username:instance:service:workspace:cluster:dify" in key
+
+
+def test_connection_pool_create_connection_retries_and_configures(clickzetta_module, monkeypatch):
+    monkeypatch.setattr(clickzetta_module.ClickzettaConnectionPool, "_start_cleanup_thread", MagicMock())
+    pool = clickzetta_module.ClickzettaConnectionPool()
+    config = _config(clickzetta_module)
+    connection = MagicMock()
+
+    monkeypatch.setattr(clickzetta_module.time, "sleep", lambda _s: None)
+    monkeypatch.setattr(
+        clickzetta_module.clickzetta, "connect", MagicMock(side_effect=[RuntimeError("boom"), connection])
+    )
+    pool._configure_connection = MagicMock()
+
+    created = pool._create_connection(config)
+
+    assert created is connection
+    assert clickzetta_module.clickzetta.connect.call_count == 2
+    pool._configure_connection.assert_called_once_with(connection)
+
+
+def test_connection_pool_create_connection_raises_after_retries(clickzetta_module, monkeypatch):
+    monkeypatch.setattr(clickzetta_module.ClickzettaConnectionPool, "_start_cleanup_thread", MagicMock())
+    pool = clickzetta_module.ClickzettaConnectionPool()
+    config = _config(clickzetta_module)
+
+    monkeypatch.setattr(clickzetta_module.time, "sleep", lambda _s: None)
+    monkeypatch.setattr(clickzetta_module.clickzetta, "connect", MagicMock(side_effect=RuntimeError("boom")))
+
+    with pytest.raises(RuntimeError, match="boom"):
+        pool._create_connection(config)
+
+
+def test_connection_pool_configure_and_validate_connection(clickzetta_module):
+    monkeypatch = pytest.MonkeyPatch()
+    monkeypatch.setattr(clickzetta_module.ClickzettaConnectionPool, "_start_cleanup_thread", MagicMock())
+    pool = clickzetta_module.ClickzettaConnectionPool()
+
+    cursor = MagicMock()
+    cursor.__enter__.return_value = cursor
+    cursor.__exit__.return_value = None
+    connection = MagicMock()
+    connection.cursor.return_value = cursor
+
+    pool._configure_connection(connection)
+    assert cursor.execute.call_count >= 2
+    assert pool._is_connection_valid(connection) is True
+
+    bad_connection = MagicMock()
+    bad_connection.cursor.side_effect = RuntimeError("bad connection")
+    assert pool._is_connection_valid(bad_connection) is False
+    monkeypatch.undo()
+
+
+def test_connection_pool_configure_connection_swallows_errors(clickzetta_module):
+    monkeypatch = pytest.MonkeyPatch()
+    monkeypatch.setattr(clickzetta_module.ClickzettaConnectionPool, "_start_cleanup_thread", MagicMock())
+    pool = clickzetta_module.ClickzettaConnectionPool()
+    connection = MagicMock()
+    connection.cursor.side_effect = RuntimeError("cannot configure")
+
+    pool._configure_connection(connection)
+    monkeypatch.undo()
+
+
+def test_connection_pool_get_return_cleanup_and_shutdown(clickzetta_module, monkeypatch):
+    monkeypatch.setattr(clickzetta_module.ClickzettaConnectionPool, "_start_cleanup_thread", MagicMock())
+    pool = clickzetta_module.ClickzettaConnectionPool()
+    config = _config(clickzetta_module)
+    key = pool._get_config_key(config)
+
+    created_connection = MagicMock()
+    pool._create_connection = MagicMock(return_value=created_connection)
+    first = pool.get_connection(config)
+    assert first is created_connection
+
+    reusable_connection = MagicMock()
+    pool._pools[key] = [(reusable_connection, clickzetta_module.time.time())]
+    pool._is_connection_valid = MagicMock(return_value=True)
+    reused = pool.get_connection(config)
+    assert reused is reusable_connection
+
+    expired_connection = MagicMock()
+    pool._pools[key] = [(expired_connection, 0.0)]
+    pool._is_connection_valid = MagicMock(return_value=False)
+    monkeypatch.setattr(clickzetta_module.time, "time", MagicMock(return_value=1000.0))
+    pool.get_connection(config)
+    expired_connection.close.assert_called_once()
+
+    random_connection = MagicMock()
+    pool._is_connection_valid = MagicMock(return_value=True)
+    pool.return_connection(config, random_connection)
+    assert len(pool._pools[key]) == 1
+
+    pool._pools[key] = [(MagicMock(), 0.0), (MagicMock(), 1000.0)]
+    pool._connection_timeout = 10
+    pool._cleanup_expired_connections()
+    assert len(pool._pools[key]) == 1
+
+    unknown_pool = MagicMock()
+    pool.return_connection(_config(clickzetta_module).model_copy(update={"workspace": "other"}), unknown_pool)
+    unknown_pool.close.assert_called_once()
+
+    pool.shutdown()
+    assert pool._shutdown is True
+
+
+def test_connection_pool_start_cleanup_thread_runs_worker_once(clickzetta_module, monkeypatch):
+    pool = clickzetta_module.ClickzettaConnectionPool.__new__(clickzetta_module.ClickzettaConnectionPool)
+    pool._shutdown = False
+    pool._cleanup_expired_connections = MagicMock(side_effect=lambda: setattr(pool, "_shutdown", True))
+
+    monkeypatch.setattr(clickzetta_module.time, "sleep", lambda _s: None)
+
+    class _Thread:
+        def __init__(self, target, daemon):
+            self._target = target
+            self.daemon = daemon
+            self.started = False
+
+        def start(self):
+            self.started = True
+            self._target()
+
+    monkeypatch.setattr(clickzetta_module.threading, "Thread", _Thread)
+    pool._start_cleanup_thread()
+
+    assert pool._cleanup_thread.started is True
+    pool._cleanup_expired_connections.assert_called_once()
+
+
+def test_vector_init_connection_context_and_helpers(clickzetta_module, monkeypatch):
+    pool = MagicMock()
+    pool.get_connection.return_value = "conn"
+    monkeypatch.setattr(clickzetta_module.ClickzettaConnectionPool, "get_instance", MagicMock(return_value=pool))
+    monkeypatch.setattr(clickzetta_module.ClickzettaVector, "_init_write_queue", MagicMock())
+
+    vector = clickzetta_module.ClickzettaVector("My-Collection", _config(clickzetta_module))
+    assert vector._table_name == "my_collection"
+
+    assert vector._get_connection() == "conn"
+    vector._return_connection("conn")
+    pool.return_connection.assert_called_with(vector._config, "conn")
+
+    with vector.get_connection_context() as conn:
+        assert conn == "conn"
+    assert pool.return_connection.call_count >= 2
+
+    assert vector.get_type() == "clickzetta"
+    assert vector._ensure_connection() == "conn"
+
+
+def test_write_queue_initialization_worker_and_execute_write(clickzetta_module, monkeypatch):
+    class _Thread:
+        def __init__(self, target, daemon):
+            self.target = target
+            self.daemon = daemon
+            self.started = 0
+
+        def start(self):
+            self.started += 1
+
+    monkeypatch.setattr(clickzetta_module.threading, "Thread", _Thread)
+    clickzetta_module.ClickzettaVector._write_queue = None
+    clickzetta_module.ClickzettaVector._write_thread = None
+    clickzetta_module.ClickzettaVector._shutdown = False
+    clickzetta_module.ClickzettaVector._init_write_queue()
+    clickzetta_module.ClickzettaVector._init_write_queue()
+    assert clickzetta_module.ClickzettaVector._write_thread.started == 1
+
+    result_queue_ok = queue.Queue()
+    result_queue_fail = queue.Queue()
+    clickzetta_module.ClickzettaVector._write_queue = queue.Queue()
+    clickzetta_module.ClickzettaVector._shutdown = False
+    clickzetta_module.ClickzettaVector._write_queue.put((lambda x: x + 1, (1,), {}, result_queue_ok))
+    clickzetta_module.ClickzettaVector._write_queue.put(
+        (lambda: (_ for _ in ()).throw(RuntimeError("worker error")), (), {}, result_queue_fail)
+    )
+    clickzetta_module.ClickzettaVector._write_queue.put(None)
+    clickzetta_module.ClickzettaVector._write_worker()
+
+    assert result_queue_ok.get() == (True, 2)
+    failed = result_queue_fail.get()
+    assert failed[0] is False
+    assert isinstance(failed[1], RuntimeError)
+
+    vector = clickzetta_module.ClickzettaVector.__new__(clickzetta_module.ClickzettaVector)
+    clickzetta_module.ClickzettaVector._write_queue = None
+    with pytest.raises(RuntimeError, match="Write queue not initialized"):
+        vector._execute_write(lambda: None)
+
+    class _ImmediateSuccessQueue:
+        def put(self, task):
+            func, args, kwargs, result_q = task
+            result_q.put((True, func(*args, **kwargs)))
+
+    clickzetta_module.ClickzettaVector._write_queue = _ImmediateSuccessQueue()
+    assert vector._execute_write(lambda x: x * 2, 3) == 6
+
+    class _ImmediateFailQueue:
+        def put(self, task):
+            _, _, _, result_q = task
+            result_q.put((False, ValueError("write failed")))
+
+    clickzetta_module.ClickzettaVector._write_queue = _ImmediateFailQueue()
+    with pytest.raises(ValueError, match="write failed"):
+        vector._execute_write(lambda: None)
+
+
+def test_table_exists_true_and_create_invokes_write_and_add_texts(clickzetta_module):
+    vector = clickzetta_module.ClickzettaVector.__new__(clickzetta_module.ClickzettaVector)
+    vector._config = _config(clickzetta_module)
+    vector._table_name = "table_1"
+
+    @contextmanager
+    def _ctx_exists():
+        connection = MagicMock()
+        cursor = MagicMock()
+        cursor.__enter__.return_value = cursor
+        cursor.__exit__.return_value = None
+        connection.cursor.return_value = cursor
+        yield connection
+
+    vector.get_connection_context = _ctx_exists
+    assert vector._table_exists() is True
+
+    vector._execute_write = MagicMock()
+    vector.add_texts = MagicMock()
+    docs = [Document(page_content="content", metadata={"doc_id": "d1"})]
+    vector.create(docs, [[0.1, 0.2]])
+    vector._execute_write.assert_called_once()
+    vector.add_texts.assert_called_once_with(docs, [[0.1, 0.2]])
+
+
+def test_create_table_and_indexes_paths(clickzetta_module):
+    vector = clickzetta_module.ClickzettaVector.__new__(clickzetta_module.ClickzettaVector)
+    vector._config = _config(clickzetta_module)
+    vector._table_name = "table_1"
+    vector._create_vector_index = MagicMock()
+    vector._create_inverted_index = MagicMock()
+
+    vector._table_exists = MagicMock(return_value=True)
+    vector._create_table_and_indexes([[0.1, 0.2]])
+    vector._create_vector_index.assert_not_called()
+
+    vector._table_exists = MagicMock(return_value=False)
+
+    @contextmanager
+    def _ctx():
+        connection = MagicMock()
+        cursor = MagicMock()
+        cursor.__enter__.return_value = cursor
+        cursor.__exit__.return_value = None
+        connection.cursor.return_value = cursor
+        yield connection
+
+    vector.get_connection_context = _ctx
+    vector._create_table_and_indexes([[0.1, 0.2, 0.3]])
+    vector._create_vector_index.assert_called_once()
+    vector._create_inverted_index.assert_called_once()
+
+    vector._config.enable_inverted_index = False
+    vector._create_vector_index.reset_mock()
+    vector._create_inverted_index.reset_mock()
+    vector._create_table_and_indexes([])
+    vector._create_vector_index.assert_called_once()
+    vector._create_inverted_index.assert_not_called()
+
+
+def test_create_vector_index_branches(clickzetta_module):
+    vector = clickzetta_module.ClickzettaVector.__new__(clickzetta_module.ClickzettaVector)
+    vector._config = _config(clickzetta_module)
+    vector._table_name = "table_1"
+    cursor = MagicMock()
+
+    cursor.fetchall.return_value = [("idx_table_vector", "embedding_vector")]
+    vector._create_vector_index(cursor)
+    assert cursor.execute.call_count == 1
+
+    cursor.reset_mock()
+    cursor.execute.side_effect = [RuntimeError("show index failed"), None]
+    vector._create_vector_index(cursor)
+    assert cursor.execute.call_count == 2
+
+    cursor.reset_mock()
+    cursor.execute.side_effect = [None, RuntimeError("already exists")]
+    cursor.fetchall.return_value = []
+    vector._create_vector_index(cursor)
+
+    cursor.reset_mock()
+    cursor.execute.side_effect = [None, RuntimeError("unexpected")]
+    cursor.fetchall.return_value = []
+    with pytest.raises(RuntimeError, match="unexpected"):
+        vector._create_vector_index(cursor)
+
+
+def test_create_inverted_index_branches(clickzetta_module):
+    vector = clickzetta_module.ClickzettaVector.__new__(clickzetta_module.ClickzettaVector)
+    vector._config = _config(clickzetta_module)
+    vector._table_name = "table_1"
+    cursor = MagicMock()
+
+    cursor.fetchall.return_value = [("idx_table_1_text", "INVERTED", "page_content")]
+    vector._create_inverted_index(cursor)
+    assert cursor.execute.call_count == 1
+
+    cursor.reset_mock()
+    cursor.execute.side_effect = [RuntimeError("show failed"), None]
+    vector._create_inverted_index(cursor)
+    assert cursor.execute.call_count == 2
+
+    cursor.reset_mock()
+    cursor.execute.side_effect = [
+        None,
+        RuntimeError("already has index"),
+        None,
+    ]
+    cursor.fetchall.return_value = [("idx_table_1_text", "INVERTED", "page_content")]
+    vector._create_inverted_index(cursor)
+
+    cursor.reset_mock()
+    cursor.execute.side_effect = [None, RuntimeError("other create failure")]
+    cursor.fetchall.return_value = []
+    vector._create_inverted_index(cursor)
+
+
+def test_add_texts_batches_and_insert_batch_behaviors(clickzetta_module, monkeypatch):
+    vector = clickzetta_module.ClickzettaVector.__new__(clickzetta_module.ClickzettaVector)
+    vector._config = _config(clickzetta_module)
+    vector._config.batch_size = 2
+    vector._table_name = "table_1"
+    vector._execute_write = MagicMock()
+    vector._safe_doc_id = MagicMock(side_effect=lambda doc_id: str(doc_id))
+
+    docs = [
+        Document(page_content="doc-1", metadata={"doc_id": "id-1"}),
+        Document(page_content="doc-2", metadata={"doc_id": "id-2"}),
+        Document(page_content="doc-3", metadata={"doc_id": "id-3"}),
+    ]
+    vectors = [[0.1], [0.2], [0.3]]
+
+    vector.add_texts([], [])
+    vector._execute_write.assert_not_called()
+
+    added_ids = vector.add_texts(docs, vectors)
+    assert added_ids == ["id-1", "id-2", "id-3"]
+    assert vector._execute_write.call_count == 2
+    assert vector._execute_write.call_args_list[0].args == (
+        vector._insert_batch,
+        docs[:2],
+        vectors[:2],
+        ["id-1", "id-2"],
+        0,
+        2,
+        2,
+    )
+    assert vector._execute_write.call_args_list[1].args == (
+        vector._insert_batch,
+        docs[2:],
+        vectors[2:],
+        ["id-3"],
+        2,
+        2,
+        2,
+    )
+
+    vector._insert_batch([], [], [], 0, 2, 1)
+    vector._insert_batch(docs[:1], vectors, ["id-1"], 0, 2, 1)
+
+    bad_doc = Document(page_content="doc-bad", metadata={"doc_id": "id-bad", "bad": {1}})
+    good_doc = Document(page_content="doc-good", metadata={"doc_id": "id-good"})
+
+    @contextmanager
+    def _ctx():
+        connection = MagicMock()
+        cursor = MagicMock()
+        cursor.__enter__.return_value = cursor
+        cursor.__exit__.return_value = None
+        connection.cursor.return_value = cursor
+        yield connection
+
+    vector.get_connection_context = _ctx
+    vector._insert_batch(
+        [bad_doc, good_doc],
+        [[0.1, 0.2], [0.3, 0.4]],
+        ["id-bad", "id-good"],
+        0,
+        2,
+        1,
+    )
+
+    @contextmanager
+    def _ctx_error():
+        connection = MagicMock()
+        cursor = MagicMock()
+        cursor.__enter__.return_value = cursor
+        cursor.__exit__.return_value = None
+        cursor.executemany.side_effect = RuntimeError("insert failed")
+        connection.cursor.return_value = cursor
+        yield connection
+
+    vector.get_connection_context = _ctx_error
+    with pytest.raises(RuntimeError, match="insert failed"):
+        vector._insert_batch([good_doc], [[0.1, 0.2]], ["id-good"], 0, 1, 1)
+
+    monkeypatch.setattr(clickzetta_module.uuid, "uuid4", lambda: "generated-id")
+    vector._safe_doc_id = clickzetta_module.ClickzettaVector._safe_doc_id.__get__(vector)
+    assert vector._safe_doc_id("") == "generated-id"
+    assert vector._safe_doc_id("!!!") == "generated-id"
+
+
+def test_delete_by_ids_and_metadata_impl_paths(clickzetta_module):
+    vector = clickzetta_module.ClickzettaVector.__new__(clickzetta_module.ClickzettaVector)
+    vector._config = _config(clickzetta_module)
+    vector._table_name = "table_1"
+    vector._execute_write = MagicMock()
+    vector._table_exists = MagicMock(return_value=True)
+
+    vector.delete_by_ids(["id-1", "id-2"])
+    vector._execute_write.assert_called_once()
+    assert vector._execute_write.call_args.args[0] == vector._delete_by_ids_impl
+
+    vector._execute_write.reset_mock()
+    vector.delete_by_metadata_field("document_id", "doc-1")
+    vector._execute_write.assert_called_once()
+    assert vector._execute_write.call_args.args[0] == vector._delete_by_metadata_field_impl
+
+    vector._safe_doc_id = MagicMock(side_effect=lambda x: x)
+
+    @contextmanager
+    def _ctx():
+        connection = MagicMock()
+        cursor = MagicMock()
+        cursor.__enter__.return_value = cursor
+        cursor.__exit__.return_value = None
+        connection.cursor.return_value = cursor
+        yield connection
+
+    vector.get_connection_context = _ctx
+    vector._delete_by_ids_impl(["id-1", "id-2"])
+    vector._delete_by_metadata_field_impl("document_id", "doc-1")
+
+
+def test_search_by_vector_covers_cosine_and_l2_paths(clickzetta_module):
+    vector = clickzetta_module.ClickzettaVector.__new__(clickzetta_module.ClickzettaVector)
+    vector._config = _config(clickzetta_module)
+    vector._config.vector_distance_function = "cosine_distance"
+    vector._table_name = "table_1"
+    vector._table_exists = MagicMock(return_value=True)
+    vector._parse_metadata = MagicMock(return_value={"document_id": "doc-1", "doc_id": "seg-1"})
+
+    @contextmanager
+    def _ctx():
+        connection = MagicMock()
+        cursor = MagicMock()
+        cursor.__enter__.return_value = cursor
+        cursor.__exit__.return_value = None
+        cursor.fetchall.return_value = [("seg-1", "content", '{"document_id":"doc-1"}', 0.2)]
+        connection.cursor.return_value = cursor
+        yield connection
+
+    vector.get_connection_context = _ctx
+    cosine_docs = vector.search_by_vector(
+        [0.1, 0.2], top_k=3, score_threshold=0.5, document_ids_filter=["doc-1"], filter={"k": "v"}
+    )
+    assert cosine_docs[0].metadata["score"] == pytest.approx(0.9)
+
+    vector._config.vector_distance_function = "l2_distance"
+    l2_docs = vector.search_by_vector([0.1, 0.2], top_k=3, score_threshold=0.5)
+    assert l2_docs[0].metadata["score"] == pytest.approx(1 / 1.2)
+
+
+def test_search_by_full_text_success_and_fallback(clickzetta_module):
+    vector = clickzetta_module.ClickzettaVector.__new__(clickzetta_module.ClickzettaVector)
+    vector._config = _config(clickzetta_module)
+    vector._table_name = "table_1"
+    vector._table_exists = MagicMock(return_value=True)
+
+    @contextmanager
+    def _ctx_success():
+        connection = MagicMock()
+        cursor = MagicMock()
+        cursor.__enter__.return_value = cursor
+        cursor.__exit__.return_value = None
+        cursor.fetchall.return_value = [
+            ("seg-1", "content-1", '"{\\"document_id\\":\\"doc-1\\"}"'),
+            ("seg-2", "content-2", "invalid-json"),
+        ]
+        connection.cursor.return_value = cursor
+        yield connection
+
+    vector.get_connection_context = _ctx_success
+    docs = vector.search_by_full_text("search'value", top_k=2, document_ids_filter=["doc-1"], filter={"a": 1})
+    assert len(docs) == 2
+    assert docs[0].metadata["score"] == 1.0
+    assert docs[1].metadata["doc_id"] == "seg-2"
+
+    @contextmanager
+    def _ctx_failure():
+        connection = MagicMock()
+        cursor = MagicMock()
+        cursor.__enter__.return_value = cursor
+        cursor.__exit__.return_value = None
+        cursor.execute.side_effect = RuntimeError("full text failed")
+        connection.cursor.return_value = cursor
+        yield connection
+
+    vector.get_connection_context = _ctx_failure
+    vector._search_by_like = MagicMock(return_value=[Document(page_content="fallback", metadata={"score": 0.5})])
+    fallback_docs = vector.search_by_full_text("query", top_k=1)
+    assert fallback_docs == vector._search_by_like.return_value
+
+
+def test_search_by_like_missing_table_and_delete_table(clickzetta_module):
+    vector = clickzetta_module.ClickzettaVector.__new__(clickzetta_module.ClickzettaVector)
+    vector._config = _config(clickzetta_module)
+    vector._table_name = "table_1"
+    vector._table_exists = MagicMock(return_value=False)
+    assert vector._search_by_like("query", top_k=1) == []
+
+    @contextmanager
+    def _ctx():
+        connection = MagicMock()
+        cursor = MagicMock()
+        cursor.__enter__.return_value = cursor
+        cursor.__exit__.return_value = None
+        connection.cursor.return_value = cursor
+        yield connection
+
+    vector.get_connection_context = _ctx
+    vector.delete()
+
+
+def test_clickzetta_pool_cleanup_and_shutdown_edge_paths(clickzetta_module):
+    pool = clickzetta_module.ClickzettaConnectionPool.__new__(clickzetta_module.ClickzettaConnectionPool)
+    pool._pools = {}
+    pool._pool_locks = {}
+    pool._max_pool_size = 1
+    pool._connection_timeout = 10
+    pool._lock = clickzetta_module.threading.Lock()
+    pool._shutdown = False
+
+    config = _config(clickzetta_module)
+    key = pool._get_config_key(config)
+    pool._pools[key] = [(MagicMock(), 1.0)]
+    pool._pool_locks[key] = clickzetta_module.threading.Lock()
+    pool._is_connection_valid = MagicMock(return_value=False)
+
+    conn = MagicMock()
+    pool.return_connection(config, conn)
+    conn.close.assert_called_once()
+
+    pool._pools["missing-lock-key"] = [(MagicMock(), 0.0)]
+    pool._cleanup_expired_connections()
+    pool.shutdown()
+    assert pool._shutdown is True
+
+
+def test_clickzetta_pool_cleanup_thread_and_worker_exception_paths(clickzetta_module, monkeypatch):
+    pool = clickzetta_module.ClickzettaConnectionPool.__new__(clickzetta_module.ClickzettaConnectionPool)
+    pool._shutdown = False
+
+    def _cleanup_then_fail():
+        pool._shutdown = True
+        raise RuntimeError("cleanup failed")
+
+    pool._cleanup_expired_connections = MagicMock(side_effect=_cleanup_then_fail)
+    monkeypatch.setattr(clickzetta_module.time, "sleep", lambda _s: None)
+
+    class _Thread:
+        def __init__(self, target, daemon):
+            self._target = target
+            self.daemon = daemon
+
+        def start(self):
+            self._target()
+
+    monkeypatch.setattr(clickzetta_module.threading, "Thread", _Thread)
+    pool._start_cleanup_thread()
+    pool._cleanup_expired_connections.assert_called_once()
+
+
+def test_clickzetta_parse_metadata_and_write_worker_additional_branches(clickzetta_module):
+    vector = clickzetta_module.ClickzettaVector.__new__(clickzetta_module.ClickzettaVector)
+
+    parsed_non_dict = vector._parse_metadata("[1,2,3]", "row-1")
+    assert parsed_non_dict["doc_id"] == "row-1"
+    assert parsed_non_dict["document_id"] == "row-1"
+
+    parsed_none = vector._parse_metadata(None, "row-2")
+    assert parsed_none["doc_id"] == "row-2"
+    assert parsed_none["document_id"] == "row-2"
+
+    clickzetta_module.ClickzettaVector._shutdown = False
+    clickzetta_module.ClickzettaVector._write_queue = None
+    clickzetta_module.ClickzettaVector._write_worker()
+
+    class _BadQueue:
+        def get(self, timeout):
+            clickzetta_module.ClickzettaVector._shutdown = True
+            raise RuntimeError("queue failed")
+
+    clickzetta_module.ClickzettaVector._shutdown = False
+    clickzetta_module.ClickzettaVector._write_queue = _BadQueue()
+    clickzetta_module.ClickzettaVector._write_worker()
+
+
+def test_clickzetta_inverted_index_existing_and_insert_non_dict_metadata(clickzetta_module):
+    vector = clickzetta_module.ClickzettaVector.__new__(clickzetta_module.ClickzettaVector)
+    vector._config = _config(clickzetta_module)
+    vector._table_name = "table_1"
+    cursor = MagicMock()
+    cursor.fetchall.return_value = [("idx_table_1_text", "INVERTED", "page_content")]
+    cursor.execute.side_effect = [
+        None,
+        RuntimeError("already has index with the same type cannot create inverted index"),
+        None,
+    ]
+
+    vector._create_inverted_index(cursor)
+
+    vector._safe_doc_id = MagicMock(side_effect=lambda value: str(value))
+
+    @contextmanager
+    def _ctx():
+        connection = MagicMock()
+        cursor_obj = MagicMock()
+        cursor_obj.__enter__.return_value = cursor_obj
+        cursor_obj.__exit__.return_value = None
+        connection.cursor.return_value = cursor_obj
+        yield connection
+
+    vector.get_connection_context = _ctx
+    vector._insert_batch(
+        [SimpleNamespace(page_content="content", metadata="not-a-dict")],
+        [[0.1, 0.2]],
+        ["doc-1"],
+        0,
+        1,
+        1,
+    )
+
+
+def test_clickzetta_full_text_table_missing_and_non_dict_metadata(clickzetta_module):
+    vector = clickzetta_module.ClickzettaVector.__new__(clickzetta_module.ClickzettaVector)
+    vector._config = _config(clickzetta_module)
+    vector._config.enable_inverted_index = True
+    vector._table_name = "table_1"
+
+    vector._table_exists = MagicMock(return_value=False)
+    assert vector.search_by_full_text("query") == []
+
+    vector._table_exists = MagicMock(return_value=True)
+
+    @contextmanager
+    def _ctx():
+        connection = MagicMock()
+        cursor = MagicMock()
+        cursor.__enter__.return_value = cursor
+        cursor.__exit__.return_value = None
+        cursor.fetchall.return_value = [
+            ("seg-1", "content-1", "[1,2,3]"),
+            ("seg-2", "content-2", None),
+        ]
+        connection.cursor.return_value = cursor
+        yield connection
+
+    vector.get_connection_context = _ctx
+    docs = vector.search_by_full_text("query")
+    assert len(docs) == 2
+    assert docs[0].metadata["doc_id"] == "seg-1"
+    assert docs[1].metadata["doc_id"] == "seg-2"

api/tests/unit_tests/core/rag/datasource/vdb/couchbase/test_couchbase_vector.py

@@ -0,0 +1,364 @@
+import importlib
+import sys
+import types
+from types import SimpleNamespace
+from unittest.mock import MagicMock, patch
+
+import pytest
+from pydantic import ValidationError
+
+from core.rag.models.document import Document
+
+
+def _build_fake_couchbase_modules():
+    couchbase = types.ModuleType("couchbase")
+    couchbase_auth = types.ModuleType("couchbase.auth")
+    couchbase_cluster = types.ModuleType("couchbase.cluster")
+    couchbase_management = types.ModuleType("couchbase.management")
+    couchbase_management_search = types.ModuleType("couchbase.management.search")
+    couchbase_options = types.ModuleType("couchbase.options")
+    couchbase_vector = types.ModuleType("couchbase.vector_search")
+    couchbase_search = types.ModuleType("couchbase.search")
+
+    class PasswordAuthenticator:
+        def __init__(self, user, password):
+            self.user = user
+            self.password = password
+
+    class ClusterOptions:
+        def __init__(self, auth):
+            self.auth = auth
+
+    class SearchOptions:
+        def __init__(self, **kwargs):
+            self.kwargs = kwargs
+
+    class VectorQuery:
+        def __init__(self, field, vector, top_k):
+            self.field = field
+            self.vector = vector
+            self.top_k = top_k
+
+    class VectorSearch:
+        @staticmethod
+        def from_vector_query(vector_query):
+            return {"vector_query": vector_query}
+
+    class QueryStringQuery:
+        def __init__(self, query):
+            self.query = query
+
+    class SearchRequest:
+        @staticmethod
+        def create(payload):
+            return {"payload": payload}
+
+    class SearchIndex:
+        def __init__(self, name, params, source_name):
+            self.name = name
+            self.params = params
+            self.source_name = source_name
+
+    class _QueryResult:
+        def __init__(self, rows=None):
+            self._rows = rows or []
+
+        def execute(self):
+            return self
+
+        def __iter__(self):
+            return iter(self._rows)
+
+    class _SearchIter:
+        def __init__(self, rows=None):
+            self._rows = rows or []
+
+        def rows(self):
+            return self._rows
+
+    class _Collection:
+        def __init__(self):
+            self.upsert = MagicMock(return_value=True)
+
+    class _SearchIndexManager:
+        def __init__(self):
+            self.upsert_index = MagicMock()
+
+    class _Scope:
+        def __init__(self):
+            self._collection = _Collection()
+            self._search_index_manager = _SearchIndexManager()
+            self.search = MagicMock(return_value=_SearchIter())
+
+        def collection(self, _name):
+            return self._collection
+
+        def search_indexes(self):
+            return self._search_index_manager
+
+    class _CollectionManager:
+        def __init__(self):
+            self.create_collection = MagicMock()
+            self.drop_collection = MagicMock()
+            self.get_all_scopes = MagicMock(return_value=[])
+
+    class _Bucket:
+        def __init__(self):
+            self._scope = _Scope()
+            self._collections = _CollectionManager()
+
+        def scope(self, _scope_name):
+            return self._scope
+
+        def collections(self):
+            return self._collections
+
+    class Cluster:
+        def __init__(self, connection_string, options):
+            self.connection_string = connection_string
+            self.options = options
+            self._bucket = _Bucket()
+            self.wait_until_ready = MagicMock()
+            self.query = MagicMock(return_value=_QueryResult())
+
+        def bucket(self, _name):
+            return self._bucket
+
+    couchbase_auth.PasswordAuthenticator = PasswordAuthenticator
+    couchbase_cluster.Cluster = Cluster
+    couchbase_management_search.SearchIndex = SearchIndex
+    couchbase_options.ClusterOptions = ClusterOptions
+    couchbase_options.SearchOptions = SearchOptions
+    couchbase_vector.VectorQuery = VectorQuery
+    couchbase_vector.VectorSearch = VectorSearch
+    couchbase_search.QueryStringQuery = QueryStringQuery
+    couchbase_search.SearchRequest = SearchRequest
+
+    couchbase.search = couchbase_search
+    couchbase.management = couchbase_management
+
+    return {
+        "couchbase": couchbase,
+        "couchbase.auth": couchbase_auth,
+        "couchbase.cluster": couchbase_cluster,
+        "couchbase.management": couchbase_management,
+        "couchbase.management.search": couchbase_management_search,
+        "couchbase.options": couchbase_options,
+        "couchbase.vector_search": couchbase_vector,
+        "couchbase.search": couchbase_search,
+    }
+
+
+@pytest.fixture
+def couchbase_module(monkeypatch):
+    for name, module in _build_fake_couchbase_modules().items():
+        monkeypatch.setitem(sys.modules, name, module)
+
+    import core.rag.datasource.vdb.couchbase.couchbase_vector as module
+
+    return importlib.reload(module)
+
+
+def _config(module):
+    return module.CouchbaseConfig(
+        connection_string="couchbase://localhost",
+        user="user",
+        password="pass",
+        bucket_name="bucket",
+        scope_name="scope",
+    )
+
+
+@pytest.mark.parametrize(
+    ("field", "value", "message"),
+    [
+        ("connection_string", "", "CONNECTION_STRING is required"),
+        ("user", "", "COUCHBASE_USER is required"),
+        ("password", "", "COUCHBASE_PASSWORD is required"),
+        ("bucket_name", "", "COUCHBASE_PASSWORD is required"),
+        ("scope_name", "", "COUCHBASE_SCOPE_NAME is required"),
+    ],
+)
+def test_couchbase_config_validation(couchbase_module, field, value, message):
+    values = _config(couchbase_module).model_dump()
+    values[field] = value
+    with pytest.raises(ValidationError, match=message):
+        couchbase_module.CouchbaseConfig.model_validate(values)
+
+
+def test_init_sets_cluster_handles(couchbase_module):
+    vector = couchbase_module.CouchbaseVector("collection_1", _config(couchbase_module))
+
+    assert vector._bucket_name == "bucket"
+    assert vector._scope_name == "scope"
+    vector._cluster.wait_until_ready.assert_called_once()
+
+
+def test_create_and_create_collection_branches(couchbase_module, monkeypatch):
+    vector = couchbase_module.CouchbaseVector.__new__(couchbase_module.CouchbaseVector)
+    vector._collection_name = "collection_1"
+    vector._client_config = _config(couchbase_module)
+    vector._scope_name = "scope"
+    vector._bucket_name = "bucket"
+    vector._bucket = MagicMock()
+    vector._scope = MagicMock()
+    vector._collection_exists = MagicMock(return_value=False)
+    vector.add_texts = MagicMock()
+
+    monkeypatch.setattr(couchbase_module.uuid, "uuid4", lambda: "a-b-c")
+    vector._create_collection = MagicMock()
+    docs = [Document(page_content="text", metadata={"doc_id": "id-1"})]
+    vector.create(docs, [[0.1, 0.2]])
+
+    vector._create_collection.assert_called_once_with(uuid="abc", vector_length=2)
+    vector.add_texts.assert_called_once_with(docs, [[0.1, 0.2]])
+
+    lock = MagicMock()
+    lock.__enter__.return_value = None
+    lock.__exit__.return_value = None
+    monkeypatch.setattr(couchbase_module.redis_client, "lock", MagicMock(return_value=lock))
+    monkeypatch.setattr(couchbase_module.redis_client, "set", MagicMock())
+
+    vector = couchbase_module.CouchbaseVector("collection_1", _config(couchbase_module))
+    monkeypatch.setattr(couchbase_module.redis_client, "get", MagicMock(return_value=1))
+    vector._create_collection(vector_length=2, uuid="uuid-1")
+    vector._bucket.collections().create_collection.assert_not_called()
+
+    monkeypatch.setattr(couchbase_module.redis_client, "get", MagicMock(return_value=None))
+    vector._collection_exists = MagicMock(return_value=True)
+    vector._create_collection(vector_length=2, uuid="uuid-2")
+    vector._bucket.collections().create_collection.assert_not_called()
+
+    vector._collection_exists = MagicMock(return_value=False)
+    vector._create_collection(vector_length=3, uuid="uuid-3")
+
+    vector._bucket.collections().create_collection.assert_called_once_with("scope", "collection_1")
+    vector._scope.search_indexes().upsert_index.assert_called_once()
+    search_index = vector._scope.search_indexes().upsert_index.call_args.args[0]
+    assert search_index.name == "collection_1_search"
+    assert (
+        search_index.params["mapping"]["types"]["scope.collection_1"]["properties"]["embedding"]["fields"][0]["dims"]
+        == 3
+    )
+    couchbase_module.redis_client.set.assert_called_once()
+
+
+def test_collection_exists_get_type_and_add_texts(couchbase_module):
+    vector = couchbase_module.CouchbaseVector("collection_1", _config(couchbase_module))
+
+    scope_obj = SimpleNamespace(name="scope", collections=[SimpleNamespace(name="collection_1")])
+    vector._bucket.collections().get_all_scopes.return_value = [scope_obj]
+    assert vector._collection_exists("collection_1") is True
+
+    scope_obj = SimpleNamespace(name="scope", collections=[SimpleNamespace(name="other")])
+    vector._bucket.collections().get_all_scopes.return_value = [scope_obj]
+    assert vector._collection_exists("collection_1") is False
+
+    vector._get_uuids = MagicMock(return_value=["id-1", "id-2"])
+    docs = [
+        Document(page_content="a", metadata={"doc_id": "id-1"}),
+        Document(page_content="b", metadata={"doc_id": "id-2"}),
+    ]
+    ids = vector.add_texts(docs, [[0.1], [0.2]])
+
+    assert ids == ["id-1", "id-2"]
+    assert vector._scope.collection("collection_1").upsert.call_count == 2
+    assert vector.get_type() == couchbase_module.VectorType.COUCHBASE
+
+
+def test_query_delete_helpers(couchbase_module):
+    vector = couchbase_module.CouchbaseVector("collection_1", _config(couchbase_module))
+
+    vector._cluster.query.return_value = SimpleNamespace(execute=lambda: iter([{"count": 2}]))
+    assert vector.text_exists("id-1") is True
+
+    vector._cluster.query.return_value = SimpleNamespace(execute=lambda: iter([]))
+    assert vector.text_exists("id-2") is False
+
+    query_result = MagicMock()
+    query_result.execute.return_value = None
+    vector._cluster.query.return_value = query_result
+
+    vector.delete_by_ids(["id-1", "id-2"])
+    vector.delete_by_document_id("id-1")
+    vector.delete_by_metadata_field("document_id", "doc-1")
+    assert vector._cluster.query.call_count >= 3
+
+    vector._cluster.query.side_effect = RuntimeError("delete failed")
+    vector.delete_by_ids(["id-3"])
+
+
+def test_search_methods_and_format_metadata(couchbase_module):
+    vector = couchbase_module.CouchbaseVector("collection_1", _config(couchbase_module))
+
+    row_1 = SimpleNamespace(fields={"text": "doc-a", "metadata.document_id": "d-1"}, score=0.9)
+    row_2 = SimpleNamespace(fields={"text": "doc-b", "metadata.document_id": "d-2"}, score=0.3)
+    vector._scope.search.return_value = SimpleNamespace(rows=lambda: [row_1, row_2])
+
+    docs = vector.search_by_vector([0.1, 0.2], top_k=2, score_threshold=0.5)
+    assert len(docs) == 1
+    assert docs[0].page_content == "doc-a"
+    assert docs[0].metadata["document_id"] == "d-1"
+    assert docs[0].metadata["score"] == pytest.approx(0.9)
+
+    vector._scope.search.side_effect = RuntimeError("search error")
+    with pytest.raises(ValueError, match="Search failed"):
+        vector.search_by_vector([0.1], top_k=1)
+
+    vector._scope.search.side_effect = None
+    row_3 = SimpleNamespace(fields={"text": "full-text", "metadata.doc_id": "x"}, score=0.7)
+    vector._scope.search.return_value = SimpleNamespace(rows=lambda: [row_3])
+    docs = vector.search_by_full_text("hello", top_k=1)
+    assert len(docs) == 1
+    assert docs[0].metadata["doc_id"] == "x"
+
+    vector._scope.search.side_effect = RuntimeError("full text failed")
+    with pytest.raises(ValueError, match="Search failed"):
+        vector.search_by_full_text("hello", top_k=1)
+
+    assert vector._format_metadata({"metadata.a": 1, "plain": 2}) == {"a": 1, "plain": 2}
+
+
+def test_delete_collection_and_factory(couchbase_module, monkeypatch):
+    vector = couchbase_module.CouchbaseVector("collection_1", _config(couchbase_module))
+    scopes = [
+        SimpleNamespace(collections=[SimpleNamespace(name="other")]),
+        SimpleNamespace(collections=[SimpleNamespace(name="collection_1")]),
+    ]
+    vector._bucket.collections().get_all_scopes.return_value = scopes
+
+    vector.delete()
+    vector._bucket.collections().drop_collection.assert_called_once_with("_default", "collection_1")
+
+    factory = couchbase_module.CouchbaseVectorFactory()
+    dataset_with_index = SimpleNamespace(
+        id="dataset-1",
+        index_struct_dict={"vector_store": {"class_prefix": "EXISTING_COLLECTION"}},
+        index_struct=None,
+    )
+    dataset_without_index = SimpleNamespace(id="dataset-2", index_struct_dict=None, index_struct=None)
+
+    monkeypatch.setattr(couchbase_module.Dataset, "gen_collection_name_by_id", lambda _id: "AUTO_COLLECTION")
+    monkeypatch.setattr(
+        couchbase_module,
+        "current_app",
+        SimpleNamespace(
+            config={
+                "COUCHBASE_CONNECTION_STRING": "couchbase://localhost",
+                "COUCHBASE_USER": "user",
+                "COUCHBASE_PASSWORD": "pass",
+                "COUCHBASE_BUCKET_NAME": "bucket",
+                "COUCHBASE_SCOPE_NAME": "scope",
+            }
+        ),
+    )
+
+    with patch.object(couchbase_module, "CouchbaseVector", return_value="vector") as vector_cls:
+        result_1 = factory.init_vector(dataset_with_index, attributes=[], embeddings=MagicMock())
+        result_2 = factory.init_vector(dataset_without_index, attributes=[], embeddings=MagicMock())
+
+    assert result_1 == "vector"
+    assert result_2 == "vector"
+    assert vector_cls.call_args_list[0].kwargs["collection_name"] == "EXISTING_COLLECTION"
+    assert vector_cls.call_args_list[1].kwargs["collection_name"] == "AUTO_COLLECTION"
+    assert dataset_without_index.index_struct is not None

api/tests/unit_tests/core/rag/datasource/vdb/elasticsearch/test_elasticsearch_ja_vector.py

@@ -0,0 +1,121 @@
+import importlib
+import sys
+import types
+from types import SimpleNamespace
+from unittest.mock import MagicMock, patch
+
+import pytest
+
+
+def _build_fake_elasticsearch_modules():
+    elasticsearch = types.ModuleType("elasticsearch")
+
+    class ConnectionError(Exception):
+        pass
+
+    class Elasticsearch:
+        def __init__(self, **kwargs):
+            self.kwargs = kwargs
+            self.ping = MagicMock(return_value=True)
+            self.info = MagicMock(return_value={"version": {"number": "8.12.0"}})
+            self.indices = SimpleNamespace(
+                refresh=MagicMock(), delete=MagicMock(), exists=MagicMock(return_value=False), create=MagicMock()
+            )
+
+    elasticsearch.Elasticsearch = Elasticsearch
+    elasticsearch.ConnectionError = ConnectionError
+    return {"elasticsearch": elasticsearch}
+
+
+@pytest.fixture
+def elasticsearch_ja_module(monkeypatch):
+    for name, module in _build_fake_elasticsearch_modules().items():
+        monkeypatch.setitem(sys.modules, name, module)
+
+    import core.rag.datasource.vdb.elasticsearch.elasticsearch_ja_vector as ja_module
+    import core.rag.datasource.vdb.elasticsearch.elasticsearch_vector as base_module
+
+    importlib.reload(base_module)
+    return importlib.reload(ja_module)
+
+
+def test_create_collection_cache_hit(elasticsearch_ja_module, monkeypatch):
+    lock = MagicMock()
+    lock.__enter__.return_value = None
+    lock.__exit__.return_value = None
+    monkeypatch.setattr(elasticsearch_ja_module.redis_client, "lock", MagicMock(return_value=lock))
+    monkeypatch.setattr(elasticsearch_ja_module.redis_client, "get", MagicMock(return_value=1))
+    monkeypatch.setattr(elasticsearch_ja_module.redis_client, "set", MagicMock())
+
+    vector = elasticsearch_ja_module.ElasticSearchJaVector.__new__(elasticsearch_ja_module.ElasticSearchJaVector)
+    vector._collection_name = "test"
+    vector._client = MagicMock()
+
+    vector.create_collection([[0.1, 0.2]], [{}])
+
+    vector._client.indices.create.assert_not_called()
+    elasticsearch_ja_module.redis_client.set.assert_not_called()
+
+
+def test_create_collection_create_and_exists_paths(elasticsearch_ja_module, monkeypatch):
+    lock = MagicMock()
+    lock.__enter__.return_value = None
+    lock.__exit__.return_value = None
+    monkeypatch.setattr(elasticsearch_ja_module.redis_client, "lock", MagicMock(return_value=lock))
+    monkeypatch.setattr(elasticsearch_ja_module.redis_client, "get", MagicMock(return_value=None))
+    monkeypatch.setattr(elasticsearch_ja_module.redis_client, "set", MagicMock())
+
+    vector = elasticsearch_ja_module.ElasticSearchJaVector.__new__(elasticsearch_ja_module.ElasticSearchJaVector)
+    vector._collection_name = "test"
+    vector._client = MagicMock()
+
+    vector._client.indices.exists.return_value = False
+    vector.create_collection([[0.1, 0.2, 0.3]], [{}])
+
+    vector._client.indices.create.assert_called_once()
+    kwargs = vector._client.indices.create.call_args.kwargs
+    assert kwargs["index"] == "test"
+    assert kwargs["mappings"]["properties"][elasticsearch_ja_module.Field.VECTOR]["dims"] == 3
+    elasticsearch_ja_module.redis_client.set.assert_called_once()
+
+    vector._client.indices.create.reset_mock()
+    elasticsearch_ja_module.redis_client.set.reset_mock()
+    vector._client.indices.exists.return_value = True
+    vector.create_collection([[0.1, 0.2]], [{}])
+
+    vector._client.indices.create.assert_not_called()
+    elasticsearch_ja_module.redis_client.set.assert_called_once()
+
+
+def test_ja_factory_uses_existing_or_generated_collection(elasticsearch_ja_module, monkeypatch):
+    factory = elasticsearch_ja_module.ElasticSearchJaVectorFactory()
+    dataset_with_index = SimpleNamespace(
+        id="dataset-1",
+        index_struct_dict={"vector_store": {"class_prefix": "EXISTING_COLLECTION"}},
+        index_struct=None,
+    )
+    dataset_without_index = SimpleNamespace(id="dataset-2", index_struct_dict=None, index_struct=None)
+
+    monkeypatch.setattr(elasticsearch_ja_module.Dataset, "gen_collection_name_by_id", lambda _id: "AUTO_COLLECTION")
+    monkeypatch.setattr(
+        elasticsearch_ja_module,
+        "current_app",
+        SimpleNamespace(
+            config={
+                "ELASTICSEARCH_HOST": "localhost",
+                "ELASTICSEARCH_PORT": 9200,
+                "ELASTICSEARCH_USERNAME": "elastic",
+                "ELASTICSEARCH_PASSWORD": "secret",
+            }
+        ),
+    )
+
+    with patch.object(elasticsearch_ja_module, "ElasticSearchJaVector", return_value="vector") as vector_cls:
+        result_1 = factory.init_vector(dataset_with_index, attributes=[], embeddings=MagicMock())
+        result_2 = factory.init_vector(dataset_without_index, attributes=[], embeddings=MagicMock())
+
+    assert result_1 == "vector"
+    assert result_2 == "vector"
+    assert vector_cls.call_args_list[0].kwargs["index_name"] == "EXISTING_COLLECTION"
+    assert vector_cls.call_args_list[1].kwargs["index_name"] == "AUTO_COLLECTION"
+    assert dataset_without_index.index_struct is not None

api/tests/unit_tests/core/rag/datasource/vdb/elasticsearch/test_elasticsearch_vector.py

@@ -0,0 +1,405 @@
+import importlib
+import sys
+import types
+from types import SimpleNamespace
+from unittest.mock import MagicMock, patch
+
+import pytest
+from pydantic import ValidationError
+
+from core.rag.models.document import Document
+
+
+def _build_fake_elasticsearch_modules():
+    elasticsearch = types.ModuleType("elasticsearch")
+
+    class ConnectionError(Exception):
+        pass
+
+    class Elasticsearch:
+        def __init__(self, **kwargs):
+            self.kwargs = kwargs
+            self.ping = MagicMock(return_value=True)
+            self.info = MagicMock(return_value={"version": {"number": "8.12.0-SNAPSHOT"}})
+            self.index = MagicMock()
+            self.exists = MagicMock(return_value=False)
+            self.delete = MagicMock()
+            self.search = MagicMock(return_value={"hits": {"hits": []}})
+            self.indices = SimpleNamespace(
+                refresh=MagicMock(),
+                delete=MagicMock(),
+                exists=MagicMock(return_value=False),
+                create=MagicMock(),
+            )
+
+    elasticsearch.Elasticsearch = Elasticsearch
+    elasticsearch.ConnectionError = ConnectionError
+    return {"elasticsearch": elasticsearch}
+
+
+@pytest.fixture
+def elasticsearch_module(monkeypatch):
+    for name, module in _build_fake_elasticsearch_modules().items():
+        monkeypatch.setitem(sys.modules, name, module)
+
+    import core.rag.datasource.vdb.elasticsearch.elasticsearch_vector as module
+
+    return importlib.reload(module)
+
+
+def _regular_config(module, **overrides):
+    values = {
+        "host": "localhost",
+        "port": 9200,
+        "username": "elastic",
+        "password": "secret",
+        "verify_certs": False,
+        "request_timeout": 10,
+        "retry_on_timeout": True,
+        "max_retries": 3,
+    }
+    values.update(overrides)
+    return module.ElasticSearchConfig.model_validate(values)
+
+
+def _cloud_config(module, **overrides):
+    values = {
+        "use_cloud": True,
+        "cloud_url": "https://cloud.example:9243",
+        "api_key": "api-key",
+        "verify_certs": True,
+        "ca_certs": "/tmp/ca.pem",
+        "request_timeout": 10,
+        "retry_on_timeout": True,
+        "max_retries": 3,
+    }
+    values.update(overrides)
+    return module.ElasticSearchConfig.model_validate(values)
+
+
+@pytest.mark.parametrize(
+    ("values", "message"),
+    [
+        ({"use_cloud": True, "cloud_url": None, "api_key": "x"}, "cloud_url is required"),
+        ({"use_cloud": True, "cloud_url": "https://cloud", "api_key": None}, "api_key is required"),
+        ({"host": None, "port": 9200, "username": "u", "password": "p"}, "HOST is required"),
+        ({"host": "h", "port": None, "username": "u", "password": "p"}, "PORT is required"),
+        ({"host": "h", "port": 9200, "username": None, "password": "p"}, "USERNAME is required"),
+        ({"host": "h", "port": 9200, "username": "u", "password": None}, "PASSWORD is required"),
+    ],
+)
+def test_elasticsearch_config_validation(elasticsearch_module, values, message):
+    with pytest.raises(ValidationError, match=message):
+        elasticsearch_module.ElasticSearchConfig.model_validate(values)
+
+
+def test_init_client_cloud_configuration(elasticsearch_module):
+    vector = elasticsearch_module.ElasticSearchVector.__new__(elasticsearch_module.ElasticSearchVector)
+    client = MagicMock()
+    client.ping.return_value = True
+
+    with patch.object(elasticsearch_module, "Elasticsearch", return_value=client) as es_cls:
+        result = vector._init_client(_cloud_config(elasticsearch_module))
+
+    assert result is client
+    kwargs = es_cls.call_args.kwargs
+    assert kwargs["hosts"] == ["https://cloud.example:9243"]
+    assert kwargs["api_key"] == "api-key"
+    assert kwargs["verify_certs"] is True
+    assert kwargs["ca_certs"] == "/tmp/ca.pem"
+
+
+def test_init_client_regular_https_and_http_fallback(elasticsearch_module):
+    vector = elasticsearch_module.ElasticSearchVector.__new__(elasticsearch_module.ElasticSearchVector)
+    client = MagicMock()
+    client.ping.return_value = True
+
+    with patch.object(elasticsearch_module, "Elasticsearch", return_value=client) as es_cls:
+        vector._init_client(
+            _regular_config(
+                elasticsearch_module,
+                host="https://es.example",
+                port=9443,
+                verify_certs=True,
+                ca_certs="/tmp/ca.pem",
+            )
+        )
+    kwargs = es_cls.call_args.kwargs
+    assert kwargs["hosts"] == ["https://es.example:9443"]
+    assert kwargs["verify_certs"] is True
+    assert kwargs["ca_certs"] == "/tmp/ca.pem"
+
+    with patch.object(elasticsearch_module, "Elasticsearch", return_value=client) as es_cls:
+        vector._init_client(_regular_config(elasticsearch_module, host="es.internal", port=9200))
+    kwargs = es_cls.call_args.kwargs
+    assert kwargs["hosts"] == ["http://es.internal:9200"]
+    assert "verify_certs" not in kwargs
+
+
+def test_init_client_connection_failures(elasticsearch_module):
+    vector = elasticsearch_module.ElasticSearchVector.__new__(elasticsearch_module.ElasticSearchVector)
+
+    client = MagicMock()
+    client.ping.return_value = False
+    with patch.object(elasticsearch_module, "Elasticsearch", return_value=client):
+        with pytest.raises(ConnectionError, match="Failed to connect"):
+            vector._init_client(_regular_config(elasticsearch_module))
+
+    with patch.object(
+        elasticsearch_module,
+        "Elasticsearch",
+        side_effect=elasticsearch_module.ElasticsearchConnectionError("boom"),
+    ):
+        with pytest.raises(ConnectionError, match="Vector database connection error"):
+            vector._init_client(_regular_config(elasticsearch_module))
+
+    with patch.object(elasticsearch_module, "Elasticsearch", side_effect=RuntimeError("oops")):
+        with pytest.raises(ConnectionError, match="initialization failed"):
+            vector._init_client(_regular_config(elasticsearch_module))
+
+
+def test_init_get_version_and_check_version(elasticsearch_module):
+    with (
+        patch.object(elasticsearch_module.ElasticSearchVector, "_init_client", return_value=MagicMock()) as init_client,
+        patch.object(elasticsearch_module.ElasticSearchVector, "_get_version", return_value="8.10.0") as get_version,
+        patch.object(elasticsearch_module.ElasticSearchVector, "_check_version") as check_version,
+    ):
+        vector = elasticsearch_module.ElasticSearchVector(
+            "collection_1", _regular_config(elasticsearch_module), attributes=["doc_id"]
+        )
+
+    init_client.assert_called_once()
+    get_version.assert_called_once()
+    check_version.assert_called_once()
+    assert vector._attributes == ["doc_id"]
+
+    vector = elasticsearch_module.ElasticSearchVector.__new__(elasticsearch_module.ElasticSearchVector)
+    vector._client = MagicMock()
+    vector._client.info.return_value = {"version": {"number": "8.13.2-SNAPSHOT"}}
+    assert vector._get_version() == "8.13.2"
+
+    vector._version = "7.17.0"
+    with pytest.raises(ValueError, match="greater than 8.0.0"):
+        vector._check_version()
+
+    vector._version = "8.0.0"
+    vector._check_version()
+
+
+def test_crud_methods_and_get_type(elasticsearch_module):
+    vector = elasticsearch_module.ElasticSearchVector.__new__(elasticsearch_module.ElasticSearchVector)
+    vector._collection_name = "collection_1"
+    vector._client = MagicMock()
+    vector._client.indices = SimpleNamespace(refresh=MagicMock(), delete=MagicMock())
+    vector._get_uuids = MagicMock(return_value=["id-1", "id-2"])
+
+    docs = [
+        Document(page_content="a", metadata={"doc_id": "id-1"}),
+        Document(page_content="b", metadata={"doc_id": "id-2"}),
+    ]
+
+    ids = vector.add_texts(docs, [[0.1], [0.2]])
+    assert ids == ["id-1", "id-2"]
+    assert vector._client.index.call_count == 2
+    vector._client.indices.refresh.assert_called_once_with(index="collection_1")
+
+    vector._client.exists.return_value = True
+    assert vector.text_exists("id-1") is True
+
+    vector.delete_by_ids([])
+    vector._client.delete.assert_not_called()
+    vector.delete_by_ids(["id-1", "id-2"])
+    assert vector._client.delete.call_count == 2
+
+    vector._client.search.return_value = {"hits": {"hits": [{"_id": "id-1"}]}}
+    vector.delete_by_ids = MagicMock()
+    vector.delete_by_metadata_field("doc_id", "d1")
+    vector.delete_by_ids.assert_called_once_with(["id-1"])
+
+    vector.delete_by_ids.reset_mock()
+    vector._client.search.return_value = {"hits": {"hits": []}}
+    vector.delete_by_metadata_field("doc_id", "d2")
+    vector.delete_by_ids.assert_not_called()
+
+    vector.delete()
+    vector._client.indices.delete.assert_called_once_with(index="collection_1")
+    assert vector.get_type() == elasticsearch_module.VectorType.ELASTICSEARCH
+
+
+def test_search_by_vector_and_full_text(elasticsearch_module):
+    vector = elasticsearch_module.ElasticSearchVector.__new__(elasticsearch_module.ElasticSearchVector)
+    vector._collection_name = "collection_1"
+    vector._client = MagicMock()
+
+    vector._client.search.return_value = {
+        "hits": {
+            "hits": [
+                {
+                    "_score": 0.8,
+                    "_source": {
+                        elasticsearch_module.Field.CONTENT_KEY: "doc-a",
+                        elasticsearch_module.Field.VECTOR: [0.1],
+                        elasticsearch_module.Field.METADATA_KEY: {"doc_id": "1", "document_id": "d-1"},
+                    },
+                },
+                {
+                    "_score": 0.2,
+                    "_source": {
+                        elasticsearch_module.Field.CONTENT_KEY: "doc-b",
+                        elasticsearch_module.Field.VECTOR: [0.2],
+                        elasticsearch_module.Field.METADATA_KEY: {"doc_id": "2", "document_id": "d-2"},
+                    },
+                },
+            ]
+        }
+    }
+
+    docs = vector.search_by_vector(
+        [0.1, 0.2],
+        top_k=2,
+        score_threshold=0.5,
+        document_ids_filter=["d-1", "d-2"],
+    )
+    assert len(docs) == 1
+    assert docs[0].metadata["score"] == pytest.approx(0.8)
+    knn = vector._client.search.call_args.kwargs["knn"]
+    assert knn["k"] == 2
+    assert knn["num_candidates"] == 3
+    assert "filter" in knn
+
+    vector._client.search.return_value = {
+        "hits": {
+            "hits": [
+                {
+                    "_source": {
+                        elasticsearch_module.Field.CONTENT_KEY: "text-hit",
+                        elasticsearch_module.Field.VECTOR: [0.3],
+                        elasticsearch_module.Field.METADATA_KEY: {"doc_id": "3"},
+                    }
+                }
+            ]
+        }
+    }
+    docs = vector.search_by_full_text("hello", top_k=3, document_ids_filter=["d-3"])
+    assert len(docs) == 1
+    assert docs[0].page_content == "text-hit"
+    query = vector._client.search.call_args.kwargs["query"]
+    assert "bool" in query
+
+
+def test_create_and_create_collection_paths(elasticsearch_module, monkeypatch):
+    lock = MagicMock()
+    lock.__enter__.return_value = None
+    lock.__exit__.return_value = None
+    monkeypatch.setattr(elasticsearch_module.redis_client, "lock", MagicMock(return_value=lock))
+    monkeypatch.setattr(elasticsearch_module.redis_client, "set", MagicMock())
+
+    vector = elasticsearch_module.ElasticSearchVector.__new__(elasticsearch_module.ElasticSearchVector)
+    vector._collection_name = "collection_1"
+    vector._client = MagicMock()
+    vector._client.indices = SimpleNamespace(exists=MagicMock(return_value=False), create=MagicMock())
+
+    vector.create_collection = MagicMock()
+    vector.add_texts = MagicMock()
+    docs = [Document(page_content="a", metadata={"doc_id": "1"})]
+    vector.create(docs, [[0.1]])
+    vector.create_collection.assert_called_once()
+    vector.add_texts.assert_called_once_with(docs, [[0.1]])
+
+    vector = elasticsearch_module.ElasticSearchVector.__new__(elasticsearch_module.ElasticSearchVector)
+    vector._collection_name = "collection_1"
+    vector._client = MagicMock()
+    vector._client.indices = SimpleNamespace(exists=MagicMock(return_value=False), create=MagicMock())
+
+    monkeypatch.setattr(elasticsearch_module.redis_client, "get", MagicMock(return_value=1))
+    vector.create_collection([[0.1, 0.2]], [{}])
+    vector._client.indices.create.assert_not_called()
+
+    monkeypatch.setattr(elasticsearch_module.redis_client, "get", MagicMock(return_value=None))
+    vector._client.indices.exists.return_value = False
+    vector.create_collection([[0.1, 0.2]], [{}])
+    vector._client.indices.create.assert_called_once()
+    mappings = vector._client.indices.create.call_args.kwargs["mappings"]
+    assert mappings["properties"][elasticsearch_module.Field.VECTOR]["dims"] == 2
+    elasticsearch_module.redis_client.set.assert_called_once()
+
+    vector._client.indices.create.reset_mock()
+    elasticsearch_module.redis_client.set.reset_mock()
+    vector._client.indices.exists.return_value = True
+    vector.create_collection([[0.1, 0.2]], [{}])
+    vector._client.indices.create.assert_not_called()
+    elasticsearch_module.redis_client.set.assert_called_once()
+
+
+def test_elasticsearch_factory_branches(elasticsearch_module, monkeypatch):
+    factory = elasticsearch_module.ElasticSearchVectorFactory()
+    dataset_with_index = SimpleNamespace(
+        id="dataset-1",
+        index_struct_dict={"vector_store": {"class_prefix": "EXISTING_COLLECTION"}},
+        index_struct=None,
+    )
+    dataset_without_index = SimpleNamespace(id="dataset-2", index_struct_dict=None, index_struct=None)
+
+    monkeypatch.setattr(elasticsearch_module.Dataset, "gen_collection_name_by_id", lambda _id: "AUTO_COLLECTION")
+
+    monkeypatch.setattr(
+        elasticsearch_module,
+        "current_app",
+        SimpleNamespace(
+            config={
+                "ELASTICSEARCH_USE_CLOUD": False,
+                "ELASTICSEARCH_HOST": "es-host",
+                "ELASTICSEARCH_PORT": 9200,
+                "ELASTICSEARCH_USERNAME": "elastic",
+                "ELASTICSEARCH_PASSWORD": "secret",
+                "ELASTICSEARCH_VERIFY_CERTS": False,
+            }
+        ),
+    )
+
+    with patch.object(elasticsearch_module, "ElasticSearchVector", return_value="vector") as vector_cls:
+        result_1 = factory.init_vector(dataset_with_index, attributes=[], embeddings=MagicMock())
+    assert result_1 == "vector"
+    cfg = vector_cls.call_args.kwargs["config"]
+    assert cfg.use_cloud is False
+    assert vector_cls.call_args.kwargs["index_name"] == "EXISTING_COLLECTION"
+
+    monkeypatch.setattr(
+        elasticsearch_module,
+        "current_app",
+        SimpleNamespace(
+            config={
+                "ELASTICSEARCH_USE_CLOUD": True,
+                "ELASTICSEARCH_CLOUD_URL": "https://cloud.elastic",
+                "ELASTICSEARCH_API_KEY": "api-key",
+                "ELASTICSEARCH_VERIFY_CERTS": True,
+            }
+        ),
+    )
+    with patch.object(elasticsearch_module, "ElasticSearchVector", return_value="vector") as vector_cls:
+        result_2 = factory.init_vector(dataset_without_index, attributes=[], embeddings=MagicMock())
+    assert result_2 == "vector"
+    cfg = vector_cls.call_args.kwargs["config"]
+    assert cfg.use_cloud is True
+    assert cfg.cloud_url == "https://cloud.elastic"
+    assert dataset_without_index.index_struct is not None
+
+    monkeypatch.setattr(
+        elasticsearch_module,
+        "current_app",
+        SimpleNamespace(
+            config={
+                "ELASTICSEARCH_USE_CLOUD": True,
+                "ELASTICSEARCH_CLOUD_URL": None,
+                "ELASTICSEARCH_HOST": "fallback-host",
+                "ELASTICSEARCH_PORT": 9201,
+                "ELASTICSEARCH_USERNAME": "elastic",
+                "ELASTICSEARCH_PASSWORD": "secret",
+            }
+        ),
+    )
+    with patch.object(elasticsearch_module, "ElasticSearchVector", return_value="vector") as vector_cls:
+        factory.init_vector(dataset_without_index, attributes=[], embeddings=MagicMock())
+    cfg = vector_cls.call_args.kwargs["config"]
+    assert cfg.use_cloud is False
+    assert cfg.host == "fallback-host"

api/tests/unit_tests/core/rag/datasource/vdb/hologres/test_hologres_vector.py

@@ -0,0 +1,371 @@
+import importlib
+import json
+import sys
+import types
+from types import SimpleNamespace
+from unittest.mock import MagicMock, patch
+
+import pytest
+from pydantic import ValidationError
+
+from core.rag.models.document import Document
+
+
+def _build_fake_hologres_modules():
+    holo_module = types.ModuleType("holo_search_sdk")
+    holo_types_module = types.ModuleType("holo_search_sdk.types")
+
+    holo_types_module.BaseQuantizationType = str
+    holo_types_module.DistanceType = str
+    holo_types_module.TokenizerType = str
+
+    def _connect(**kwargs):
+        client = MagicMock()
+        client.kwargs = kwargs
+        client.connect = MagicMock()
+        client.check_table_exist = MagicMock(return_value=False)
+        client.open_table = MagicMock(return_value=MagicMock())
+        client.execute = MagicMock(return_value=[])
+        client.drop_table = MagicMock()
+        return client
+
+    holo_module.connect = MagicMock(side_effect=_connect)
+
+    return {
+        "holo_search_sdk": holo_module,
+        "holo_search_sdk.types": holo_types_module,
+    }
+
+
+@pytest.fixture
+def hologres_module(monkeypatch):
+    for name, module in _build_fake_hologres_modules().items():
+        monkeypatch.setitem(sys.modules, name, module)
+
+    import core.rag.datasource.vdb.hologres.hologres_vector as module
+
+    return importlib.reload(module)
+
+
+def _valid_config(module):
+    return module.HologresVectorConfig(
+        host="localhost",
+        port=80,
+        database="dify",
+        access_key_id="ak",
+        access_key_secret="sk",
+        schema_name="public",
+        tokenizer="jieba",
+        distance_method="Cosine",
+        base_quantization_type="rabitq",
+        max_degree=64,
+        ef_construction=400,
+    )
+
+
+@pytest.mark.parametrize(
+    ("field", "value", "message"),
+    [
+        ("host", "", "config HOLOGRES_HOST is required"),
+        ("database", "", "config HOLOGRES_DATABASE is required"),
+        ("access_key_id", "", "config HOLOGRES_ACCESS_KEY_ID is required"),
+        ("access_key_secret", "", "config HOLOGRES_ACCESS_KEY_SECRET is required"),
+    ],
+)
+def test_hologres_config_validation(hologres_module, field, value, message):
+    values = _valid_config(hologres_module).model_dump()
+    values[field] = value
+
+    with pytest.raises(ValidationError, match=message):
+        hologres_module.HologresVectorConfig.model_validate(values)
+
+
+def test_init_client_and_get_type(hologres_module):
+    vector = hologres_module.HologresVector("Collection_One", _valid_config(hologres_module))
+
+    hologres_module.holo.connect.assert_called_once_with(
+        host="localhost",
+        port=80,
+        database="dify",
+        access_key_id="ak",
+        access_key_secret="sk",
+        schema="public",
+    )
+    vector._client.connect.assert_called_once()
+    assert vector.table_name == "embedding_collection_one"
+    assert vector.get_type() == hologres_module.VectorType.HOLOGRES
+
+
+def test_create_delegates_collection_creation_and_upsert(hologres_module):
+    vector = hologres_module.HologresVector("collection_one", _valid_config(hologres_module))
+    vector._create_collection = MagicMock()
+    vector.add_texts = MagicMock()
+    docs = [Document(page_content="hello", metadata={"doc_id": "seg-1"})]
+
+    result = vector.create(docs, [[0.1, 0.2]])
+
+    assert result is None
+    vector._create_collection.assert_called_once_with(2)
+    vector.add_texts.assert_called_once_with(docs, [[0.1, 0.2]])
+
+
+def test_add_texts_returns_empty_for_empty_documents(hologres_module):
+    vector = hologres_module.HologresVector("collection_one", _valid_config(hologres_module))
+
+    assert vector.add_texts([], []) == []
+    vector._client.open_table.assert_not_called()
+
+
+def test_add_texts_batches_and_serializes_metadata(hologres_module):
+    vector = hologres_module.HologresVector("collection_one", _valid_config(hologres_module))
+    table = vector._client.open_table.return_value
+    documents = [
+        Document(page_content=f"doc-{i}", metadata={"doc_id": f"id-{i}", "document_id": f"document-{i}"})
+        for i in range(100)
+    ]
+    documents.append(SimpleNamespace(page_content="doc-100", metadata=None))
+    embeddings = [[float(i)] for i in range(len(documents))]
+
+    ids = vector.add_texts(documents, embeddings)
+
+    assert ids[:2] == ["id-0", "id-1"]
+    assert ids[-1] == ""
+    assert len(ids) == 101
+    assert vector._client.open_table.call_count == 2
+    assert table.upsert_multi.call_count == 2
+    first_call = table.upsert_multi.call_args_list[0].kwargs
+    second_call = table.upsert_multi.call_args_list[1].kwargs
+    assert first_call["index_column"] == "id"
+    assert first_call["column_names"] == ["id", "text", "meta", "embedding"]
+    assert first_call["update_columns"] == ["text", "meta", "embedding"]
+    assert len(first_call["values"]) == 100
+    assert json.loads(first_call["values"][0][2]) == {"doc_id": "id-0", "document_id": "document-0"}
+    assert second_call["values"][0][0] == ""
+    assert second_call["values"][0][2] == "{}"
+
+
+def test_text_exists_handles_missing_and_present_tables(hologres_module):
+    vector = hologres_module.HologresVector("collection_one", _valid_config(hologres_module))
+    vector._client.check_table_exist.side_effect = [False, True]
+    vector._client.execute.return_value = [(1,)]
+
+    assert vector.text_exists("seg-1") is False
+    assert vector.text_exists("seg-1") is True
+    vector._client.execute.assert_called_once()
+
+
+def test_get_ids_by_metadata_field_returns_ids_or_none(hologres_module):
+    vector = hologres_module.HologresVector("collection_one", _valid_config(hologres_module))
+    vector._client.execute.side_effect = [[("id-1",), ("id-2",)], []]
+
+    assert vector.get_ids_by_metadata_field("document_id", "doc-1") == ["id-1", "id-2"]
+    assert vector.get_ids_by_metadata_field("document_id", "doc-1") is None
+
+
+def test_delete_by_ids_branches(hologres_module):
+    vector = hologres_module.HologresVector("collection_one", _valid_config(hologres_module))
+
+    vector.delete_by_ids([])
+    vector._client.check_table_exist.assert_not_called()
+
+    vector._client.check_table_exist.return_value = False
+    vector.delete_by_ids(["id-1"])
+    vector._client.execute.assert_not_called()
+
+    vector._client.check_table_exist.return_value = True
+    vector.delete_by_ids(["id-1", "id-2"])
+    vector._client.execute.assert_called_once()
+
+
+def test_delete_by_metadata_field_branches(hologres_module):
+    vector = hologres_module.HologresVector("collection_one", _valid_config(hologres_module))
+    vector._client.check_table_exist.return_value = False
+
+    vector.delete_by_metadata_field("document_id", "doc-1")
+    vector._client.execute.assert_not_called()
+
+    vector._client.check_table_exist.return_value = True
+    vector.delete_by_metadata_field("document_id", "doc-1")
+    vector._client.execute.assert_called_once()
+
+
+def test_search_by_vector_returns_empty_when_table_missing(hologres_module):
+    vector = hologres_module.HologresVector("collection_one", _valid_config(hologres_module))
+    vector._client.check_table_exist.return_value = False
+
+    assert vector.search_by_vector([0.1, 0.2]) == []
+
+
+def test_search_by_vector_applies_filter_and_processes_results(hologres_module):
+    vector = hologres_module.HologresVector("collection_one", _valid_config(hologres_module))
+    vector._client.check_table_exist.return_value = True
+    table = vector._client.open_table.return_value
+    query = MagicMock()
+    table.search_vector.return_value = query
+    query.select.return_value = query
+    query.limit.return_value = query
+    query.where.return_value = query
+    query.fetchall.return_value = [
+        (0.2, "seg-1", "doc-1", '{"doc_id":"seg-1","document_id":"doc-1"}'),
+        (0.9, "seg-2", "doc-2", {"doc_id": "seg-2", "document_id": "doc-2"}),
+    ]
+
+    docs = vector.search_by_vector(
+        [0.1, 0.2],
+        top_k=2,
+        score_threshold=0.5,
+        document_ids_filter=["doc-1"],
+    )
+
+    assert len(docs) == 1
+    assert docs[0].page_content == "doc-1"
+    assert docs[0].metadata["doc_id"] == "seg-1"
+    assert docs[0].metadata["score"] == pytest.approx(0.8)
+    table.search_vector.assert_called_once()
+    query.where.assert_called_once()
+
+
+def test_search_by_full_text_returns_empty_when_table_missing(hologres_module):
+    vector = hologres_module.HologresVector("collection_one", _valid_config(hologres_module))
+    vector._client.check_table_exist.return_value = False
+
+    assert vector.search_by_full_text("query") == []
+
+
+def test_search_by_full_text_applies_filter_and_processes_results(hologres_module):
+    vector = hologres_module.HologresVector("collection_one", _valid_config(hologres_module))
+    vector._client.check_table_exist.return_value = True
+    table = vector._client.open_table.return_value
+    search_query = MagicMock()
+    table.search_text.return_value = search_query
+    search_query.limit.return_value = search_query
+    search_query.where.return_value = search_query
+    search_query.fetchall.return_value = [
+        ("seg-1", "doc-1", '{"doc_id":"seg-1"}', [0.1], 0.95),
+        ("seg-2", "doc-2", {"doc_id": "seg-2"}, [0.2], 0.7),
+    ]
+
+    docs = vector.search_by_full_text("query", top_k=2, document_ids_filter=["doc-1"])
+
+    assert len(docs) == 2
+    assert docs[0].metadata["doc_id"] == "seg-1"
+    assert docs[0].metadata["score"] == pytest.approx(0.95)
+    assert docs[1].metadata["score"] == pytest.approx(0.7)
+    table.search_text.assert_called_once()
+    search_query.where.assert_called_once()
+
+
+def test_delete_handles_existing_and_missing_tables(hologres_module):
+    vector = hologres_module.HologresVector("collection_one", _valid_config(hologres_module))
+    vector._client.check_table_exist.side_effect = [False, True]
+
+    vector.delete()
+    vector._client.drop_table.assert_not_called()
+
+    vector.delete()
+    vector._client.drop_table.assert_called_once_with(vector.table_name)
+
+
+def test_create_collection_returns_early_when_cache_hits(hologres_module, monkeypatch):
+    lock = MagicMock()
+    lock.__enter__.return_value = None
+    lock.__exit__.return_value = False
+    monkeypatch.setattr(hologres_module.redis_client, "lock", MagicMock(return_value=lock))
+    monkeypatch.setattr(hologres_module.redis_client, "get", MagicMock(return_value=1))
+    monkeypatch.setattr(hologres_module.redis_client, "set", MagicMock())
+
+    vector = hologres_module.HologresVector("collection_one", _valid_config(hologres_module))
+    vector._create_collection(3)
+
+    vector._client.check_table_exist.assert_not_called()
+    hologres_module.redis_client.set.assert_not_called()
+
+
+def test_create_collection_creates_table_and_indexes(hologres_module, monkeypatch):
+    lock = MagicMock()
+    lock.__enter__.return_value = None
+    lock.__exit__.return_value = False
+    monkeypatch.setattr(hologres_module.redis_client, "lock", MagicMock(return_value=lock))
+    monkeypatch.setattr(hologres_module.redis_client, "get", MagicMock(return_value=None))
+    monkeypatch.setattr(hologres_module.redis_client, "set", MagicMock())
+    monkeypatch.setattr(hologres_module.time, "sleep", MagicMock())
+
+    vector = hologres_module.HologresVector("collection_one", _valid_config(hologres_module))
+    vector._client.check_table_exist.side_effect = [False, False, True]
+    table = vector._client.open_table.return_value
+
+    vector._create_collection(3)
+
+    vector._client.execute.assert_called_once()
+    table.set_vector_index.assert_called_once_with(
+        column="embedding",
+        distance_method="Cosine",
+        base_quantization_type="rabitq",
+        max_degree=64,
+        ef_construction=400,
+        use_reorder=True,
+    )
+    table.create_text_index.assert_called_once_with(
+        index_name="ft_idx_collection_one",
+        column="text",
+        tokenizer="jieba",
+    )
+    hologres_module.redis_client.set.assert_called_once()
+
+
+def test_create_collection_raises_when_table_never_becomes_ready(hologres_module, monkeypatch):
+    lock = MagicMock()
+    lock.__enter__.return_value = None
+    lock.__exit__.return_value = False
+    monkeypatch.setattr(hologres_module.redis_client, "lock", MagicMock(return_value=lock))
+    monkeypatch.setattr(hologres_module.redis_client, "get", MagicMock(return_value=None))
+    monkeypatch.setattr(hologres_module.redis_client, "set", MagicMock())
+    monkeypatch.setattr(hologres_module.time, "sleep", MagicMock())
+
+    vector = hologres_module.HologresVector("collection_one", _valid_config(hologres_module))
+    vector._client.check_table_exist.side_effect = [False] + [False] * 15
+
+    with pytest.raises(RuntimeError, match="was not ready after 30s"):
+        vector._create_collection(3)
+
+    hologres_module.redis_client.set.assert_not_called()
+
+
+def test_hologres_factory_uses_existing_or_generated_collection(hologres_module, monkeypatch):
+    factory = hologres_module.HologresVectorFactory()
+    dataset_with_index = SimpleNamespace(
+        id="dataset-1",
+        index_struct_dict={"vector_store": {"class_prefix": "existing_collection"}},
+        index_struct=None,
+    )
+    dataset_without_index = SimpleNamespace(id="dataset-2", index_struct_dict=None, index_struct=None)
+
+    monkeypatch.setattr(hologres_module.Dataset, "gen_collection_name_by_id", lambda _id: "generated_collection")
+    monkeypatch.setattr(hologres_module.dify_config, "HOLOGRES_HOST", "127.0.0.1")
+    monkeypatch.setattr(hologres_module.dify_config, "HOLOGRES_PORT", 80)
+    monkeypatch.setattr(hologres_module.dify_config, "HOLOGRES_DATABASE", "dify")
+    monkeypatch.setattr(hologres_module.dify_config, "HOLOGRES_ACCESS_KEY_ID", "ak")
+    monkeypatch.setattr(hologres_module.dify_config, "HOLOGRES_ACCESS_KEY_SECRET", "sk")
+    monkeypatch.setattr(hologres_module.dify_config, "HOLOGRES_SCHEMA", "public")
+    monkeypatch.setattr(hologres_module.dify_config, "HOLOGRES_TOKENIZER", "jieba")
+    monkeypatch.setattr(hologres_module.dify_config, "HOLOGRES_DISTANCE_METHOD", "Cosine")
+    monkeypatch.setattr(hologres_module.dify_config, "HOLOGRES_BASE_QUANTIZATION_TYPE", "rabitq")
+    monkeypatch.setattr(hologres_module.dify_config, "HOLOGRES_MAX_DEGREE", 64)
+    monkeypatch.setattr(hologres_module.dify_config, "HOLOGRES_EF_CONSTRUCTION", 400)
+
+    with patch.object(hologres_module, "HologresVector", return_value="vector") as vector_cls:
+        result_1 = factory.init_vector(dataset_with_index, attributes=[], embeddings=MagicMock())
+        result_2 = factory.init_vector(dataset_without_index, attributes=[], embeddings=MagicMock())
+
+    assert result_1 == "vector"
+    assert result_2 == "vector"
+    assert vector_cls.call_args_list[0].kwargs["collection_name"] == "existing_collection"
+    assert vector_cls.call_args_list[1].kwargs["collection_name"] == "generated_collection"
+    generated_config = vector_cls.call_args_list[1].kwargs["config"]
+    assert generated_config.host == "127.0.0.1"
+    assert generated_config.database == "dify"
+    assert generated_config.access_key_id == "ak"
+    assert json.loads(dataset_without_index.index_struct) == {
+        "type": hologres_module.VectorType.HOLOGRES,
+        "vector_store": {"class_prefix": "generated_collection"},
+    }

api/tests/unit_tests/core/rag/datasource/vdb/huawei/test_huawei_cloud_vector.py

@@ -0,0 +1,243 @@
+import importlib
+import sys
+import types
+from types import SimpleNamespace
+from unittest.mock import MagicMock, patch
+
+import pytest
+from pydantic import ValidationError
+
+from core.rag.models.document import Document
+
+
+def _build_fake_elasticsearch_modules():
+    elasticsearch = types.ModuleType("elasticsearch")
+
+    class Elasticsearch:
+        def __init__(self, **kwargs):
+            self.kwargs = kwargs
+            self.index = MagicMock()
+            self.exists = MagicMock(return_value=False)
+            self.delete = MagicMock()
+            self.search = MagicMock(return_value={"hits": {"hits": []}})
+            self.indices = SimpleNamespace(
+                refresh=MagicMock(), delete=MagicMock(), exists=MagicMock(return_value=False), create=MagicMock()
+            )
+
+    elasticsearch.Elasticsearch = Elasticsearch
+    return {"elasticsearch": elasticsearch}
+
+
+@pytest.fixture
+def huawei_module(monkeypatch):
+    for name, module in _build_fake_elasticsearch_modules().items():
+        monkeypatch.setitem(sys.modules, name, module)
+
+    import core.rag.datasource.vdb.huawei.huawei_cloud_vector as module
+
+    return importlib.reload(module)
+
+
+def _config(module):
+    return module.HuaweiCloudVectorConfig(hosts="http://localhost:9200", username="user", password="pass")
+
+
+def test_create_ssl_context(huawei_module):
+    ctx = huawei_module.create_ssl_context()
+    assert ctx.check_hostname is False
+    assert ctx.verify_mode == huawei_module.ssl.CERT_NONE
+
+
+def test_huawei_config_validation_and_params(huawei_module):
+    with pytest.raises(ValidationError, match="HOSTS is required"):
+        huawei_module.HuaweiCloudVectorConfig.model_validate({"hosts": ""})
+
+    config = _config(huawei_module)
+    params = config.to_elasticsearch_params()
+    assert params["hosts"] == ["http://localhost:9200"]
+    assert params["basic_auth"] == ("user", "pass")
+
+    config = huawei_module.HuaweiCloudVectorConfig(hosts="host1,host2", username=None, password=None)
+    params = config.to_elasticsearch_params()
+    assert "basic_auth" not in params
+
+
+def test_init_get_type_and_add_texts(huawei_module):
+    vector = huawei_module.HuaweiCloudVector("COLLECTION", _config(huawei_module))
+
+    assert vector._collection_name == "collection"
+    assert vector.get_type() == huawei_module.VectorType.HUAWEI_CLOUD
+
+    vector._get_uuids = MagicMock(return_value=["id-1", "id-2"])
+    docs = [
+        Document(page_content="a", metadata={"doc_id": "id-1"}),
+        Document(page_content="b", metadata={"doc_id": "id-2"}),
+    ]
+
+    ids = vector.add_texts(docs, [[0.1], [0.2]])
+    assert ids == ["id-1", "id-2"]
+    assert vector._client.index.call_count == 2
+    vector._client.indices.refresh.assert_called_once_with(index="collection")
+
+
+def test_crud_methods(huawei_module):
+    vector = huawei_module.HuaweiCloudVector("collection", _config(huawei_module))
+
+    vector._client.exists.return_value = True
+    assert vector.text_exists("id-1") is True
+
+    vector.delete_by_ids([])
+    vector._client.delete.assert_not_called()
+    vector.delete_by_ids(["id-1"])
+    vector._client.delete.assert_called_once_with(index="collection", id="id-1")
+
+    vector._client.search.return_value = {"hits": {"hits": [{"_id": "id-1"}]}}
+    vector.delete_by_ids = MagicMock()
+    vector.delete_by_metadata_field("doc_id", "x")
+    vector.delete_by_ids.assert_called_once_with(["id-1"])
+
+    vector.delete_by_ids.reset_mock()
+    vector._client.search.return_value = {"hits": {"hits": []}}
+    vector.delete_by_metadata_field("doc_id", "x")
+    vector.delete_by_ids.assert_not_called()
+
+    vector.delete()
+    vector._client.indices.delete.assert_called_once_with(index="collection")
+
+
+def test_search_by_vector_and_full_text(huawei_module):
+    vector = huawei_module.HuaweiCloudVector("collection", _config(huawei_module))
+    vector._client.search.return_value = {
+        "hits": {
+            "hits": [
+                {
+                    "_score": 0.9,
+                    "_source": {
+                        huawei_module.Field.CONTENT_KEY: "doc-a",
+                        huawei_module.Field.VECTOR: [0.1],
+                        huawei_module.Field.METADATA_KEY: {"doc_id": "1"},
+                    },
+                },
+                {
+                    "_score": 0.1,
+                    "_source": {
+                        huawei_module.Field.CONTENT_KEY: "doc-b",
+                        huawei_module.Field.VECTOR: [0.2],
+                        huawei_module.Field.METADATA_KEY: {"doc_id": "2"},
+                    },
+                },
+            ]
+        }
+    }
+
+    docs = vector.search_by_vector([0.1, 0.2], top_k=2, score_threshold=0.5)
+    assert len(docs) == 1
+    assert docs[0].metadata["score"] == pytest.approx(0.9)
+
+    query_body = vector._client.search.call_args.kwargs["body"]
+    assert query_body["query"]["vector"][huawei_module.Field.VECTOR]["topk"] == 2
+
+    vector._client.search.return_value = {
+        "hits": {
+            "hits": [
+                {
+                    "_source": {
+                        huawei_module.Field.CONTENT_KEY: "text-hit",
+                        huawei_module.Field.VECTOR: [0.3],
+                        huawei_module.Field.METADATA_KEY: {"doc_id": "3"},
+                    }
+                }
+            ]
+        }
+    }
+    docs = vector.search_by_full_text("hello", top_k=3)
+    assert len(docs) == 1
+    assert docs[0].page_content == "text-hit"
+
+
+def test_search_by_vector_skips_hits_without_metadata(huawei_module, monkeypatch):
+    class FakeDocument:
+        def __init__(self, page_content, vector, metadata):
+            self.page_content = page_content
+            self.vector = vector
+            self.metadata = None
+
+    monkeypatch.setattr(huawei_module, "Document", FakeDocument)
+
+    vector = huawei_module.HuaweiCloudVector("collection", _config(huawei_module))
+    vector._client.search.return_value = {
+        "hits": {
+            "hits": [
+                {
+                    "_score": 0.9,
+                    "_source": {
+                        huawei_module.Field.CONTENT_KEY: "doc-a",
+                        huawei_module.Field.VECTOR: [0.1],
+                        huawei_module.Field.METADATA_KEY: {"doc_id": "1"},
+                    },
+                }
+            ]
+        }
+    }
+
+    docs = vector.search_by_vector([0.1, 0.2], top_k=1, score_threshold=0.5)
+
+    assert docs == []
+
+
+def test_create_and_create_collection_paths(huawei_module, monkeypatch):
+    lock = MagicMock()
+    lock.__enter__.return_value = None
+    lock.__exit__.return_value = None
+    monkeypatch.setattr(huawei_module.redis_client, "lock", MagicMock(return_value=lock))
+    monkeypatch.setattr(huawei_module.redis_client, "set", MagicMock())
+
+    vector = huawei_module.HuaweiCloudVector("collection", _config(huawei_module))
+    vector.create_collection = MagicMock()
+    vector.add_texts = MagicMock()
+
+    docs = [Document(page_content="a", metadata={"doc_id": "1"})]
+    vector.create(docs, [[0.1]])
+    vector.create_collection.assert_called_once()
+    vector.add_texts.assert_called_once_with(docs, [[0.1]])
+
+    vector = huawei_module.HuaweiCloudVector("collection", _config(huawei_module))
+    monkeypatch.setattr(huawei_module.redis_client, "get", MagicMock(return_value=1))
+    vector.create_collection([[0.1, 0.2]], [{}])
+    vector._client.indices.create.assert_not_called()
+
+    monkeypatch.setattr(huawei_module.redis_client, "get", MagicMock(return_value=None))
+    vector._client.indices.exists.return_value = False
+    vector.create_collection([[0.1, 0.2]], [{}])
+    vector._client.indices.create.assert_called_once()
+
+    kwargs = vector._client.indices.create.call_args.kwargs
+    mappings = kwargs["mappings"]
+    assert mappings["properties"][huawei_module.Field.VECTOR]["dimension"] == 2
+    assert kwargs["settings"] == {"index.vector": True}
+    huawei_module.redis_client.set.assert_called_once()
+
+
+def test_huawei_factory_branches(huawei_module, monkeypatch):
+    factory = huawei_module.HuaweiCloudVectorFactory()
+    dataset_with_index = SimpleNamespace(
+        id="dataset-1",
+        index_struct_dict={"vector_store": {"class_prefix": "EXISTING_COLLECTION"}},
+        index_struct=None,
+    )
+    dataset_without_index = SimpleNamespace(id="dataset-2", index_struct_dict=None, index_struct=None)
+
+    monkeypatch.setattr(huawei_module.Dataset, "gen_collection_name_by_id", lambda _id: "AUTO_COLLECTION")
+    monkeypatch.setattr(huawei_module.dify_config, "HUAWEI_CLOUD_HOSTS", "http://huawei-es:9200")
+    monkeypatch.setattr(huawei_module.dify_config, "HUAWEI_CLOUD_USER", "user")
+    monkeypatch.setattr(huawei_module.dify_config, "HUAWEI_CLOUD_PASSWORD", "pass")
+
+    with patch.object(huawei_module, "HuaweiCloudVector", return_value="vector") as vector_cls:
+        result_1 = factory.init_vector(dataset_with_index, attributes=[], embeddings=MagicMock())
+        result_2 = factory.init_vector(dataset_without_index, attributes=[], embeddings=MagicMock())
+
+    assert result_1 == "vector"
+    assert result_2 == "vector"
+    assert vector_cls.call_args_list[0].kwargs["index_name"] == "existing_collection"
+    assert vector_cls.call_args_list[1].kwargs["index_name"] == "auto_collection"
+    assert dataset_without_index.index_struct is not None

api/tests/unit_tests/core/rag/datasource/vdb/iris/test_iris_vector.py

@@ -0,0 +1,412 @@
+import importlib
+import sys
+import types
+from contextlib import contextmanager
+from types import SimpleNamespace
+from unittest.mock import MagicMock, patch
+
+import pytest
+
+from core.rag.models.document import Document
+
+
+def _build_fake_iris_module():
+    iris = types.ModuleType("iris")
+
+    def connect(**_kwargs):
+        conn = MagicMock()
+        conn.cursor.return_value = MagicMock()
+        return conn
+
+    iris.connect = MagicMock(side_effect=connect)
+    return iris
+
+
+@pytest.fixture
+def iris_module(monkeypatch):
+    monkeypatch.setitem(sys.modules, "iris", _build_fake_iris_module())
+
+    import core.rag.datasource.vdb.iris.iris_vector as module
+
+    reloaded = importlib.reload(module)
+    reloaded._pool_instance = None
+    return reloaded
+
+
+def _config(module, **overrides):
+    values = {
+        "IRIS_HOST": "localhost",
+        "IRIS_SUPER_SERVER_PORT": 1972,
+        "IRIS_USER": "user",
+        "IRIS_PASSWORD": "pass",
+        "IRIS_DATABASE": "db",
+        "IRIS_SCHEMA": "schema",
+        "IRIS_CONNECTION_URL": "url",
+        "IRIS_MIN_CONNECTION": 1,
+        "IRIS_MAX_CONNECTION": 2,
+        "IRIS_TEXT_INDEX": True,
+        "IRIS_TEXT_INDEX_LANGUAGE": "en",
+    }
+    values.update(overrides)
+    return module.IrisVectorConfig.model_validate(values)
+
+
+def test_get_iris_pool_singleton(iris_module):
+    iris_module._pool_instance = None
+    cfg = _config(iris_module)
+
+    with patch.object(iris_module, "IrisConnectionPool", return_value="pool") as pool_cls:
+        pool_1 = iris_module.get_iris_pool(cfg)
+        pool_2 = iris_module.get_iris_pool(cfg)
+
+    assert pool_1 == "pool"
+    assert pool_2 == "pool"
+    pool_cls.assert_called_once_with(cfg)
+
+
+@pytest.fixture
+def pool_with_min_max(iris_module):
+    cfg = _config(iris_module, IRIS_MIN_CONNECTION=2, IRIS_MAX_CONNECTION=3)
+    with patch.object(iris_module.IrisConnectionPool, "_create_connection", return_value=MagicMock()) as create_conn:
+        pool = iris_module.IrisConnectionPool(cfg)
+        yield pool, create_conn
+
+
+def test_pool_initialization_respects_min_max(pool_with_min_max):
+    pool, create_conn = pool_with_min_max
+    assert len(pool._pool) == 2
+    assert create_conn.call_count == 2
+
+
+@pytest.fixture
+def pool_for_get_connection(iris_module):
+    cfg = _config(iris_module, IRIS_MIN_CONNECTION=2, IRIS_MAX_CONNECTION=3)
+    pool = iris_module.IrisConnectionPool(cfg)
+    return pool
+
+
+def test_get_connection_returns_existing_and_increments(pool_for_get_connection):
+    pool = pool_for_get_connection
+    conn = MagicMock()
+    pool._pool = [conn]
+    pool._in_use = 0
+    assert pool.get_connection() is conn
+    assert pool._in_use == 1
+
+
+def test_get_connection_creates_new_when_empty(pool_for_get_connection):
+    pool = pool_for_get_connection
+    pool._pool = []
+    pool._in_use = 0
+    pool._create_connection = MagicMock(return_value="new-conn")
+    assert pool.get_connection() == "new-conn"
+
+
+def test_get_connection_raises_when_exhausted(pool_for_get_connection):
+    pool = pool_for_get_connection
+    pool._pool = []
+    pool._in_use = pool._max_size
+    with pytest.raises(RuntimeError, match="exhausted"):
+        pool.get_connection()
+
+
+@pytest.fixture
+def pool_for_return_connection(iris_module):
+    cfg = _config(iris_module)
+    with patch.object(iris_module.IrisConnectionPool, "_initialize_pool", return_value=None):
+        pool = iris_module.IrisConnectionPool(cfg)
+    return pool
+
+
+def test_return_connection_adds_healthy(pool_for_return_connection):
+    pool = pool_for_return_connection
+    pool._in_use = 1
+    conn = MagicMock()
+    cursor = MagicMock()
+    conn.cursor.return_value = cursor
+    pool.return_connection(conn)
+    assert pool._pool[-1] is conn
+    assert pool._in_use == 0
+
+
+def test_return_connection_replaces_bad(pool_for_return_connection):
+    pool = pool_for_return_connection
+    pool._in_use = 1
+    bad_conn = MagicMock()
+    bad_cursor = MagicMock()
+    bad_cursor.execute.side_effect = OSError("bad")
+    bad_conn.cursor.return_value = bad_cursor
+    replacement = MagicMock()
+    pool._create_connection = MagicMock(return_value=replacement)
+    pool.return_connection(bad_conn)
+    bad_conn.close.assert_called_once()
+    assert pool._pool[-1] is replacement
+    assert pool._in_use == 0
+
+
+def test_return_connection_ignores_none(pool_for_return_connection):
+    pool = pool_for_return_connection
+    before = len(pool._pool)
+    pool.return_connection(None)
+    assert len(pool._pool) == before
+
+
+@pytest.fixture
+def pool_for_schema_and_close(iris_module):
+    cfg = _config(iris_module)
+    with patch.object(iris_module.IrisConnectionPool, "_initialize_pool", return_value=None):
+        pool = iris_module.IrisConnectionPool(cfg)
+    conn = MagicMock()
+    cursor = MagicMock()
+    conn.cursor.return_value = cursor
+    pool._pool = [conn]
+    return pool, conn, cursor
+
+
+def test_ensure_schema_exists_cached_noop(pool_for_schema_and_close):
+    pool, conn, cursor = pool_for_schema_and_close
+    pool._schemas_initialized = {"cached_schema"}
+    pool.ensure_schema_exists("cached_schema")
+    cursor.execute.assert_not_called()
+
+
+def test_ensure_schema_exists_creates_new(pool_for_schema_and_close):
+    pool, conn, cursor = pool_for_schema_and_close
+    pool._schemas_initialized = set()
+    cursor.fetchone.return_value = (0,)
+    pool.ensure_schema_exists("new_schema")
+    assert "new_schema" in pool._schemas_initialized
+    assert any("CREATE SCHEMA" in call.args[0] for call in cursor.execute.call_args_list)
+    conn.commit.assert_called_once()
+
+
+def test_ensure_schema_exists_existing_no_commit(pool_for_schema_and_close):
+    pool, conn, cursor = pool_for_schema_and_close
+    pool._schemas_initialized = set()
+    cursor.fetchone.return_value = (1,)
+    pool.ensure_schema_exists("existing_schema")
+    conn.commit.assert_not_called()
+
+
+def test_ensure_schema_exists_rollback_on_error(pool_for_schema_and_close):
+    pool, conn, cursor = pool_for_schema_and_close
+    pool._schemas_initialized = set()
+    cursor.execute.side_effect = RuntimeError("schema failure")
+    with pytest.raises(RuntimeError, match="schema failure"):
+        pool.ensure_schema_exists("broken_schema")
+    conn.rollback.assert_called()
+
+
+def test_close_all_closes_and_resets(iris_module):
+    cfg = _config(iris_module)
+    with patch.object(iris_module.IrisConnectionPool, "_initialize_pool", return_value=None):
+        pool = iris_module.IrisConnectionPool(cfg)
+    conn = MagicMock()
+    conn_2 = MagicMock()
+    conn_2.close.side_effect = OSError("close fail")
+    pool._pool = [conn, conn_2]
+    pool._schemas_initialized = {"x"}
+    pool.close_all()
+    assert pool._pool == []
+    assert pool._in_use == 0
+    assert pool._schemas_initialized == set()
+
+
+def test_iris_vector_init_get_cursor_and_create(iris_module):
+    pool = MagicMock()
+    pool.get_connection.return_value = MagicMock()
+
+    with patch.object(iris_module, "get_iris_pool", return_value=pool):
+        vector = iris_module.IrisVector("collection", _config(iris_module))
+
+    assert vector.table_name == "EMBEDDING_COLLECTION"
+    assert vector.schema == "schema"
+    assert vector.get_type() == iris_module.VectorType.IRIS
+
+    conn = MagicMock()
+    cursor = MagicMock()
+    conn.cursor.return_value = cursor
+    vector.pool.get_connection.return_value = conn
+
+    with vector._get_cursor() as got_cursor:
+        assert got_cursor is cursor
+    conn.commit.assert_called_once()
+    vector.pool.return_connection.assert_called_with(conn)
+
+    conn = MagicMock()
+    cursor = MagicMock()
+    conn.cursor.return_value = cursor
+    vector.pool.get_connection.return_value = conn
+    with pytest.raises(RuntimeError, match="boom"):
+        with vector._get_cursor():
+            raise RuntimeError("boom")
+    conn.rollback.assert_called_once()
+
+    vector._create_collection = MagicMock()
+    vector.add_texts = MagicMock(return_value=["id-1"])
+    docs = [Document(page_content="a", metadata={"doc_id": "id-1"})]
+    assert vector.create(docs, [[0.1, 0.2]]) == ["id-1"]
+    vector._create_collection.assert_called_once_with(2)
+
+
+def test_iris_vector_crud_and_vector_search(iris_module, monkeypatch):
+    with patch.object(iris_module, "get_iris_pool", return_value=MagicMock()):
+        vector = iris_module.IrisVector("collection", _config(iris_module))
+
+    cursor = MagicMock()
+
+    @contextmanager
+    def _cursor_ctx():
+        yield cursor
+
+    vector._get_cursor = _cursor_ctx
+    monkeypatch.setattr(iris_module.uuid, "uuid4", lambda: "generated-id")
+
+    docs = [
+        Document(page_content="a", metadata={"doc_id": "id-1"}),
+        SimpleNamespace(page_content="b", metadata=None),
+    ]
+    ids = vector.add_texts(docs, [[0.1], [0.2]])
+    assert ids == ["id-1", "generated-id"]
+    assert cursor.execute.call_count == 2
+
+    cursor.fetchone.return_value = (1,)
+    assert vector.text_exists("id-1") is True
+    cursor.fetchone.return_value = None
+    assert vector.text_exists("id-2") is False
+
+    vector._get_cursor = MagicMock(side_effect=RuntimeError("db down"))
+    assert vector.text_exists("id-3") is False
+
+    vector._get_cursor = _cursor_ctx
+    vector.delete_by_ids([])
+    before = cursor.execute.call_count
+    vector.delete_by_ids(["id-1", "id-2"])
+    assert cursor.execute.call_count == before + 1
+
+    vector.delete_by_metadata_field("document_id", "doc-1")
+    assert "meta LIKE" in cursor.execute.call_args.args[0]
+
+    cursor.fetchall.return_value = [
+        ("id-1", "text-1", '{"document_id":"d-1"}', 0.9),
+        ("id-2", "text-2", '{"document_id":"d-2"}', 0.2),
+        ("id-x",),
+    ]
+    docs = vector.search_by_vector([0.1, 0.2], top_k=3, score_threshold=0.5)
+    assert len(docs) == 1
+    assert docs[0].metadata["score"] == pytest.approx(0.9)
+
+
+def test_iris_vector_full_text_search_paths(iris_module, monkeypatch):
+    cfg = _config(iris_module, IRIS_TEXT_INDEX=True)
+    with patch.object(iris_module, "get_iris_pool", return_value=MagicMock()):
+        vector = iris_module.IrisVector("collection", cfg)
+
+    cursor = MagicMock()
+
+    @contextmanager
+    def _cursor_ctx():
+        yield cursor
+
+    vector._get_cursor = _cursor_ctx
+
+    cursor.execute.side_effect = None
+    cursor.fetchall.return_value = [
+        ("id-1", "text-1", '{"document_id":"d-1"}', 0.7),
+        ("id-2", "text-2", "{}", None),
+    ]
+    docs = vector.search_by_full_text("query", top_k=2, document_ids_filter=["d-1"])
+    assert len(docs) == 2
+    assert docs[0].metadata["score"] == pytest.approx(0.7)
+    assert docs[1].metadata["score"] == pytest.approx(0.0)
+
+    cursor.reset_mock()
+    cursor.execute.side_effect = [RuntimeError("rank failed"), None]
+    cursor.fetchall.return_value = [("id-3", "text-3", "{}", 0.5)]
+    docs = vector.search_by_full_text("query", top_k=1)
+    assert len(docs) == 1
+    assert cursor.execute.call_count == 2
+
+    cfg_like = _config(iris_module, IRIS_TEXT_INDEX=False)
+    with patch.object(iris_module, "get_iris_pool", return_value=MagicMock()):
+        vector_like = iris_module.IrisVector("collection", cfg_like)
+    vector_like._get_cursor = _cursor_ctx
+
+    fake_libs = types.ModuleType("libs")
+    fake_helper = types.ModuleType("libs.helper")
+    fake_helper.escape_like_pattern = lambda value: value.replace("%", "\\%")
+    monkeypatch.setitem(sys.modules, "libs", fake_libs)
+    monkeypatch.setitem(sys.modules, "libs.helper", fake_helper)
+
+    cursor.reset_mock()
+    cursor.execute.side_effect = None
+    cursor.fetchall.return_value = []
+    assert vector_like.search_by_full_text("100%", top_k=1) == []
+
+
+def test_iris_vector_delete_create_collection_and_factory(iris_module, monkeypatch):
+    with patch.object(iris_module, "get_iris_pool", return_value=MagicMock()):
+        vector = iris_module.IrisVector("collection", _config(iris_module, IRIS_TEXT_INDEX=True))
+
+    cursor = MagicMock()
+
+    @contextmanager
+    def _cursor_ctx():
+        yield cursor
+
+    vector._get_cursor = _cursor_ctx
+    vector.delete()
+    assert "DROP TABLE" in cursor.execute.call_args.args[0]
+
+    lock = MagicMock()
+    lock.__enter__.return_value = None
+    lock.__exit__.return_value = None
+    monkeypatch.setattr(iris_module.redis_client, "lock", MagicMock(return_value=lock))
+    monkeypatch.setattr(iris_module.redis_client, "set", MagicMock())
+
+    monkeypatch.setattr(iris_module.redis_client, "get", MagicMock(return_value=1))
+    vector._create_collection(2)
+    cursor.execute.assert_called_once()
+
+    cursor.reset_mock()
+    monkeypatch.setattr(iris_module.redis_client, "get", MagicMock(return_value=None))
+    vector.pool.ensure_schema_exists = MagicMock()
+    vector._create_collection(3)
+    assert cursor.execute.call_count == 3
+    iris_module.redis_client.set.assert_called_once()
+
+    cursor.reset_mock()
+    vector.config.IRIS_TEXT_INDEX = False
+    vector._create_collection(3)
+    assert cursor.execute.call_count == 2
+
+    factory = iris_module.IrisVectorFactory()
+    dataset_with_index = SimpleNamespace(
+        id="dataset-1",
+        index_struct_dict={"vector_store": {"class_prefix": "EXISTING_COLLECTION"}},
+        index_struct=None,
+    )
+    dataset_without_index = SimpleNamespace(id="dataset-2", index_struct_dict=None, index_struct=None)
+
+    monkeypatch.setattr(iris_module.Dataset, "gen_collection_name_by_id", lambda _id: "AUTO_COLLECTION")
+    monkeypatch.setattr(iris_module.dify_config, "IRIS_HOST", "localhost")
+    monkeypatch.setattr(iris_module.dify_config, "IRIS_SUPER_SERVER_PORT", 1972)
+    monkeypatch.setattr(iris_module.dify_config, "IRIS_USER", "user")
+    monkeypatch.setattr(iris_module.dify_config, "IRIS_PASSWORD", "pass")
+    monkeypatch.setattr(iris_module.dify_config, "IRIS_DATABASE", "db")
+    monkeypatch.setattr(iris_module.dify_config, "IRIS_SCHEMA", "schema")
+    monkeypatch.setattr(iris_module.dify_config, "IRIS_CONNECTION_URL", "url")
+    monkeypatch.setattr(iris_module.dify_config, "IRIS_MIN_CONNECTION", 1)
+    monkeypatch.setattr(iris_module.dify_config, "IRIS_MAX_CONNECTION", 2)
+    monkeypatch.setattr(iris_module.dify_config, "IRIS_TEXT_INDEX", True)
+    monkeypatch.setattr(iris_module.dify_config, "IRIS_TEXT_INDEX_LANGUAGE", "en")
+
+    with patch.object(iris_module, "IrisVector", return_value="vector") as vector_cls:
+        result_1 = factory.init_vector(dataset_with_index, attributes=[], embeddings=MagicMock())
+        result_2 = factory.init_vector(dataset_without_index, attributes=[], embeddings=MagicMock())
+
+    assert result_1 == "vector"
+    assert result_2 == "vector"
+    assert vector_cls.call_args_list[0].kwargs["collection_name"] == "EXISTING_COLLECTION"
+    assert vector_cls.call_args_list[1].kwargs["collection_name"] == "AUTO_COLLECTION"
+    assert dataset_without_index.index_struct is not None

api/tests/unit_tests/core/rag/datasource/vdb/lindorm/test_lindorm_vector.py

@@ -0,0 +1,394 @@
+import importlib
+import sys
+import types
+from types import SimpleNamespace
+from unittest.mock import MagicMock, patch
+
+import pytest
+from pydantic import ValidationError
+
+from core.rag.models.document import Document
+
+
+def _build_fake_opensearch_modules():
+    opensearchpy = types.ModuleType("opensearchpy")
+    opensearch_helpers = types.ModuleType("opensearchpy.helpers")
+
+    class BulkIndexError(Exception):
+        def __init__(self, errors):
+            super().__init__("bulk error")
+            self.errors = errors
+
+    class OpenSearch:
+        def __init__(self, **kwargs):
+            self.kwargs = kwargs
+            self.indices = SimpleNamespace(
+                refresh=MagicMock(),
+                exists=MagicMock(return_value=False),
+                delete=MagicMock(),
+                create=MagicMock(),
+            )
+            self.bulk = MagicMock(return_value={"errors": False, "items": []})
+            self.search = MagicMock(return_value={"hits": {"hits": []}})
+            self.delete_by_query = MagicMock()
+            self.get = MagicMock(return_value={"_id": "id"})
+            self.exists = MagicMock(return_value=True)
+
+    opensearch_helpers.BulkIndexError = BulkIndexError
+    opensearch_helpers.bulk = MagicMock()
+
+    opensearchpy.OpenSearch = OpenSearch
+    opensearchpy.helpers = opensearch_helpers
+
+    return {
+        "opensearchpy": opensearchpy,
+        "opensearchpy.helpers": opensearch_helpers,
+    }
+
+
+@pytest.fixture
+def lindorm_module(monkeypatch):
+    for name, module in _build_fake_opensearch_modules().items():
+        monkeypatch.setitem(sys.modules, name, module)
+
+    import core.rag.datasource.vdb.lindorm.lindorm_vector as module
+
+    return importlib.reload(module)
+
+
+def _config(module):
+    return module.LindormVectorStoreConfig(
+        hosts="http://localhost:9200",
+        username="user",
+        password="pass",
+        using_ugc=False,
+        request_timeout=3.0,
+    )
+
+
+@pytest.mark.parametrize(
+    ("field", "value", "message"),
+    [
+        ("hosts", None, "config URL is required"),
+        ("username", None, "config USERNAME is required"),
+        ("password", None, "config PASSWORD is required"),
+    ],
+)
+def test_lindorm_config_validation(lindorm_module, field, value, message):
+    values = _config(lindorm_module).model_dump()
+    values[field] = value
+
+    with pytest.raises(ValidationError, match=message):
+        lindorm_module.LindormVectorStoreConfig.model_validate(values)
+
+
+def test_to_opensearch_params_and_init(lindorm_module):
+    cfg = _config(lindorm_module)
+    params = cfg.to_opensearch_params()
+
+    assert params["hosts"] == "http://localhost:9200"
+    assert params["http_auth"] == ("user", "pass")
+
+    vector = lindorm_module.LindormVectorStore("Collection", cfg, using_ugc=False)
+    assert vector._collection_name == "collection"
+    assert vector.get_type() == lindorm_module.VectorType.LINDORM
+
+    with pytest.raises(ValueError, match="routing_value"):
+        lindorm_module.LindormVectorStore("c", cfg, using_ugc=True)
+
+    vector_ugc = lindorm_module.LindormVectorStore("c", cfg, using_ugc=True, routing_value="ROUTE")
+    assert vector_ugc._routing == "route"
+
+
+def test_create_refresh_and_add_texts_success(lindorm_module, monkeypatch):
+    vector = lindorm_module.LindormVectorStore(
+        "collection", _config(lindorm_module), using_ugc=True, routing_value="route"
+    )
+    vector.create_collection = MagicMock()
+    vector.add_texts = MagicMock()
+
+    docs = [Document(page_content="a", metadata={"doc_id": "id-1"})]
+    vector.create(docs, [[0.1]])
+    vector.create_collection.assert_called_once_with([[0.1]], [{"doc_id": "id-1"}])
+    vector.add_texts.assert_called_once_with(docs, [[0.1]])
+
+    vector = lindorm_module.LindormVectorStore(
+        "collection", _config(lindorm_module), using_ugc=True, routing_value="route"
+    )
+    monkeypatch.setattr(lindorm_module.time, "sleep", MagicMock())
+
+    docs = [
+        Document(page_content="a", metadata={"doc_id": "id-1"}),
+        Document(page_content="b", metadata={"doc_id": "id-2"}),
+        Document(page_content="c", metadata={"doc_id": "id-3"}),
+    ]
+    embeddings = [[0.1], [0.2], [0.3]]
+
+    vector.add_texts(docs, embeddings, batch_size=2, timeout=9)
+
+    assert vector._client.bulk.call_count == 2
+    actions = vector._client.bulk.call_args_list[0].args[0]
+    assert actions[0]["index"]["routing"] == "route"
+    assert actions[1][lindorm_module.ROUTING_FIELD] == "route"
+    vector.refresh()
+    vector._client.indices.refresh.assert_called_once_with(index="collection")
+
+
+def test_add_texts_error_paths(lindorm_module):
+    vector = lindorm_module.LindormVectorStore("collection", _config(lindorm_module), using_ugc=False)
+    vector._client.bulk.return_value = {"errors": True, "items": [{"index": {"error": "boom"}}]}
+
+    docs = [Document(page_content="a", metadata={"doc_id": "id-1"})]
+    with pytest.raises(Exception, match="RetryError"):
+        vector.add_texts(docs, [[0.1]], batch_size=1)
+
+    vector._client.bulk.side_effect = RuntimeError("bulk failed")
+    with pytest.raises(Exception, match="RetryError"):
+        vector.add_texts(docs, [[0.1]], batch_size=1)
+
+
+def test_metadata_lookup_and_delete_by_metadata(lindorm_module):
+    vector = lindorm_module.LindormVectorStore(
+        "collection", _config(lindorm_module), using_ugc=True, routing_value="route"
+    )
+    vector._client.search.return_value = {"hits": {"hits": [{"_id": "id-1"}, {"_id": "id-2"}]}}
+
+    ids = vector.get_ids_by_metadata_field("document_id", "doc-1")
+    assert ids == ["id-1", "id-2"]
+    query = vector._client.search.call_args.kwargs["body"]
+    must_conditions = query["query"]["bool"]["must"]
+    assert any("routing_field.keyword" in cond.get("term", {}) for cond in must_conditions)
+
+    vector.delete_by_ids = MagicMock()
+    vector.delete_by_metadata_field("document_id", "doc-1")
+    vector.delete_by_ids.assert_called_once_with(["id-1", "id-2"])
+
+    vector._client.search.return_value = {"hits": {"hits": []}}
+    vector.delete_by_ids.reset_mock()
+    vector.delete_by_metadata_field("document_id", "doc-2")
+    vector.delete_by_ids.assert_not_called()
+
+
+def test_delete_by_ids_paths(lindorm_module):
+    vector = lindorm_module.LindormVectorStore(
+        "collection", _config(lindorm_module), using_ugc=True, routing_value="route"
+    )
+
+    vector.delete_by_ids([])
+    vector._client.indices.exists.assert_not_called()
+
+    vector._client.indices.exists.return_value = False
+    vector.delete_by_ids(["id-1"])
+
+    vector._client.indices.exists.return_value = True
+    vector._client.exists.side_effect = [True, False]
+    lindorm_module.helpers.bulk.reset_mock()
+    vector.delete_by_ids(["id-1", "id-2"])
+    lindorm_module.helpers.bulk.assert_called_once()
+    actions = lindorm_module.helpers.bulk.call_args.args[1]
+    assert len(actions) == 1
+    assert actions[0]["routing"] == "route"
+
+    lindorm_module.helpers.bulk.reset_mock()
+    lindorm_module.helpers.bulk.side_effect = lindorm_module.BulkIndexError(
+        errors=[
+            {"delete": {"status": 404, "_id": "id-404"}},
+            {"delete": {"status": 500, "_id": "id-500"}},
+        ]
+    )
+    vector._client.exists.side_effect = [True]
+    vector.delete_by_ids(["id-1"])
+
+
+def test_delete_and_text_exists(lindorm_module):
+    vector = lindorm_module.LindormVectorStore(
+        "collection", _config(lindorm_module), using_ugc=True, routing_value="route"
+    )
+    vector.delete()
+    vector._client.delete_by_query.assert_called_once()
+    vector._client.indices.refresh.assert_called_once_with(index="collection")
+
+    vector = lindorm_module.LindormVectorStore("collection", _config(lindorm_module), using_ugc=False)
+    vector._client.indices.exists.return_value = True
+    vector.delete()
+    vector._client.indices.delete.assert_called_once_with(index="collection", params={"timeout": 60})
+
+    vector._client.indices.delete.reset_mock()
+    vector._client.indices.exists.return_value = False
+    vector.delete()
+    vector._client.indices.delete.assert_not_called()
+
+    assert vector.text_exists("id-1") is True
+    vector._client.get.side_effect = RuntimeError("missing")
+    assert vector.text_exists("id-1") is False
+
+
+def test_search_by_vector_validation_and_success(lindorm_module):
+    vector = lindorm_module.LindormVectorStore(
+        "collection", _config(lindorm_module), using_ugc=True, routing_value="route"
+    )
+
+    with pytest.raises(ValueError, match="should be a list"):
+        vector.search_by_vector("bad")
+
+    with pytest.raises(ValueError, match="should be floats"):
+        vector.search_by_vector([0.1, "bad"])
+
+    vector._client.search.return_value = {
+        "hits": {
+            "hits": [
+                {
+                    "_score": 0.9,
+                    "_source": {
+                        lindorm_module.Field.CONTENT_KEY: "doc-a",
+                        lindorm_module.Field.VECTOR: [0.1],
+                        lindorm_module.Field.METADATA_KEY: {"doc_id": "1", "document_id": "d-1"},
+                    },
+                },
+                {
+                    "_score": 0.2,
+                    "_source": {
+                        lindorm_module.Field.CONTENT_KEY: "doc-b",
+                        lindorm_module.Field.VECTOR: [0.2],
+                        lindorm_module.Field.METADATA_KEY: {"doc_id": "2", "document_id": "d-2"},
+                    },
+                },
+            ]
+        }
+    }
+    docs = vector.search_by_vector([0.1, 0.2], top_k=2, score_threshold=0.5, document_ids_filter=["d-1"])
+    assert len(docs) == 1
+    assert docs[0].metadata["score"] == pytest.approx(0.9)
+
+    call_kwargs = vector._client.search.call_args.kwargs
+    query = call_kwargs["body"]
+    assert "ext" in query
+    assert query["query"]["knn"][lindorm_module.Field.VECTOR]["filter"]["bool"]["must"]
+    assert call_kwargs["params"]["routing"] == "route"
+
+    vector._client.search.side_effect = RuntimeError("search failed")
+    with pytest.raises(RuntimeError, match="search failed"):
+        vector.search_by_vector([0.1])
+
+
+def test_search_by_full_text_success_and_error(lindorm_module):
+    vector = lindorm_module.LindormVectorStore(
+        "collection", _config(lindorm_module), using_ugc=True, routing_value="route"
+    )
+    vector._client.search.return_value = {
+        "hits": {
+            "hits": [
+                {
+                    "_source": {
+                        lindorm_module.Field.CONTENT_KEY: "doc-a",
+                        lindorm_module.Field.VECTOR: [0.1],
+                        lindorm_module.Field.METADATA_KEY: {"doc_id": "1"},
+                    }
+                }
+            ]
+        }
+    }
+
+    docs = vector.search_by_full_text("hello", top_k=2, document_ids_filter=["d-1"])
+    assert len(docs) == 1
+    assert docs[0].page_content == "doc-a"
+
+    query = vector._client.search.call_args.kwargs["body"]
+    assert query["query"]["bool"]["filter"]
+
+    vector._client.search.side_effect = RuntimeError("full text failed")
+    with pytest.raises(RuntimeError, match="full text failed"):
+        vector.search_by_full_text("hello")
+
+
+def test_create_collection_paths(lindorm_module, monkeypatch):
+    vector = lindorm_module.LindormVectorStore("collection", _config(lindorm_module), using_ugc=False)
+
+    with pytest.raises(ValueError, match="cannot be empty"):
+        vector.create_collection([])
+
+    lock = MagicMock()
+    lock.__enter__.return_value = None
+    lock.__exit__.return_value = None
+    monkeypatch.setattr(lindorm_module.redis_client, "lock", MagicMock(return_value=lock))
+    monkeypatch.setattr(lindorm_module.redis_client, "set", MagicMock())
+
+    monkeypatch.setattr(lindorm_module.redis_client, "get", MagicMock(return_value=1))
+    vector.create_collection([[0.1, 0.2]])
+    vector._client.indices.create.assert_not_called()
+
+    monkeypatch.setattr(lindorm_module.redis_client, "get", MagicMock(return_value=None))
+    vector._client.indices.exists.return_value = False
+    vector.create_collection([[0.1, 0.2]], index_params={"index_type": "ivf", "space_type": "cosine"})
+    vector._client.indices.create.assert_called_once()
+    body = vector._client.indices.create.call_args.kwargs["body"]
+    assert body["mappings"]["properties"][lindorm_module.Field.VECTOR]["method"]["name"] == "ivf"
+    assert body["mappings"]["properties"][lindorm_module.Field.VECTOR]["method"]["space_type"] == "cosine"
+
+    vector._client.indices.create.reset_mock()
+    vector._client.indices.exists.return_value = True
+    vector.create_collection([[0.1, 0.2]])
+    vector._client.indices.create.assert_not_called()
+
+
+def test_lindorm_factory_branches(lindorm_module, monkeypatch):
+    factory = lindorm_module.LindormVectorStoreFactory()
+
+    monkeypatch.setattr(lindorm_module.dify_config, "LINDORM_URL", "http://localhost:9200")
+    monkeypatch.setattr(lindorm_module.dify_config, "LINDORM_USERNAME", "user")
+    monkeypatch.setattr(lindorm_module.dify_config, "LINDORM_PASSWORD", "pass")
+    monkeypatch.setattr(lindorm_module.dify_config, "LINDORM_QUERY_TIMEOUT", 3.0)
+    monkeypatch.setattr(lindorm_module.dify_config, "LINDORM_INDEX_TYPE", "hnsw")
+    monkeypatch.setattr(lindorm_module.dify_config, "LINDORM_DISTANCE_TYPE", "l2")
+    monkeypatch.setattr(lindorm_module.Dataset, "gen_collection_name_by_id", lambda _id: "AUTO_COLLECTION")
+
+    dataset = SimpleNamespace(id="dataset-1", index_struct=None, index_struct_dict={})
+    embeddings = SimpleNamespace(embed_query=lambda _q: [0.1, 0.2, 0.3])
+
+    monkeypatch.setattr(lindorm_module.dify_config, "LINDORM_USING_UGC", None)
+    with pytest.raises(ValueError, match="LINDORM_USING_UGC is not set"):
+        factory.init_vector(dataset, attributes=[], embeddings=embeddings)
+
+    monkeypatch.setattr(lindorm_module.dify_config, "LINDORM_USING_UGC", False)
+
+    dataset_existing_plain = SimpleNamespace(
+        id="dataset-1",
+        index_struct="{}",
+        index_struct_dict={"vector_store": {"class_prefix": "EXISTING"}, "using_ugc": False},
+    )
+    with patch.object(lindorm_module, "LindormVectorStore", return_value="vector") as store_cls:
+        result = factory.init_vector(dataset_existing_plain, attributes=[], embeddings=embeddings)
+    assert result == "vector"
+    assert store_cls.call_args.args[0] == "existing"
+
+    dataset_existing_ugc = SimpleNamespace(
+        id="dataset-1",
+        index_struct="{}",
+        index_struct_dict={
+            "vector_store": {"class_prefix": "ROUTING"},
+            "using_ugc": True,
+            "dimension": 1536,
+            "index_type": "hnsw",
+            "distance_type": "l2",
+        },
+    )
+    with patch.object(lindorm_module, "LindormVectorStore", return_value="vector") as store_cls:
+        factory.init_vector(dataset_existing_ugc, attributes=[], embeddings=embeddings)
+    assert store_cls.call_args.args[0] == "ugc_index_1536_hnsw_l2"
+    assert store_cls.call_args.kwargs["routing_value"] == "ROUTING"
+
+    dataset_new = SimpleNamespace(id="dataset-2", index_struct=None, index_struct_dict={})
+
+    monkeypatch.setattr(lindorm_module.dify_config, "LINDORM_USING_UGC", True)
+    with patch.object(lindorm_module, "LindormVectorStore", return_value="vector") as store_cls:
+        factory.init_vector(dataset_new, attributes=[], embeddings=embeddings)
+    assert store_cls.call_args.args[0] == "ugc_index_3_hnsw_l2"
+    assert store_cls.call_args.kwargs["routing_value"] == "auto_collection"
+    assert dataset_new.index_struct is not None
+
+    dataset_new_plain = SimpleNamespace(id="dataset-3", index_struct=None, index_struct_dict={})
+    monkeypatch.setattr(lindorm_module.dify_config, "LINDORM_USING_UGC", False)
+    with patch.object(lindorm_module, "LindormVectorStore", return_value="vector") as store_cls:
+        factory.init_vector(dataset_new_plain, attributes=[], embeddings=embeddings)
+    assert store_cls.call_args.args[0] == "auto_collection"
+    assert store_cls.call_args.kwargs["routing_value"] is None

api/tests/unit_tests/core/rag/datasource/vdb/matrixone/test_matrixone_vector.py

@@ -0,0 +1,252 @@
+import importlib
+import sys
+import types
+from types import SimpleNamespace
+from unittest.mock import MagicMock, patch
+
+import pytest
+from pydantic import ValidationError
+
+from core.rag.models.document import Document
+
+
+def _build_fake_mo_vector_modules():
+    mo_vector = types.ModuleType("mo_vector")
+    mo_vector.__path__ = []
+    mo_vector_client = types.ModuleType("mo_vector.client")
+
+    class MoVectorClient:
+        def __init__(self, **kwargs):
+            self.kwargs = kwargs
+            self.create_full_text_index = MagicMock()
+            self.insert = MagicMock()
+            self.get = MagicMock(return_value=[])
+            self.delete = MagicMock()
+            self.query_by_metadata = MagicMock(return_value=[])
+            self.query = MagicMock(return_value=[])
+            self.full_text_query = MagicMock(return_value=[])
+
+    mo_vector_client.MoVectorClient = MoVectorClient
+    mo_vector.client = mo_vector_client
+    return {"mo_vector": mo_vector, "mo_vector.client": mo_vector_client}
+
+
+@pytest.fixture
+def matrixone_module(monkeypatch):
+    for name, module in _build_fake_mo_vector_modules().items():
+        monkeypatch.setitem(sys.modules, name, module)
+
+    import core.rag.datasource.vdb.matrixone.matrixone_vector as module
+
+    return importlib.reload(module)
+
+
+def _valid_config(module):
+    return module.MatrixoneConfig(
+        host="localhost",
+        port=6001,
+        user="dump",
+        password="111",
+        database="dify",
+        metric="l2",
+    )
+
+
+@pytest.mark.parametrize(
+    ("field", "value", "message"),
+    [
+        ("host", "", "config host is required"),
+        ("port", 0, "config port is required"),
+        ("user", "", "config user is required"),
+        ("password", "", "config password is required"),
+        ("database", "", "config database is required"),
+    ],
+)
+def test_matrixone_config_validation(matrixone_module, field, value, message):
+    values = _valid_config(matrixone_module).model_dump()
+    values[field] = value
+
+    with pytest.raises(ValidationError, match=message):
+        matrixone_module.MatrixoneConfig.model_validate(values)
+
+
+def test_get_client_creates_full_text_index_when_cache_misses(matrixone_module, monkeypatch):
+    lock = MagicMock()
+    lock.__enter__.return_value = None
+    lock.__exit__.return_value = None
+    monkeypatch.setattr(matrixone_module.redis_client, "lock", MagicMock(return_value=lock))
+    monkeypatch.setattr(matrixone_module.redis_client, "get", MagicMock(return_value=None))
+    monkeypatch.setattr(matrixone_module.redis_client, "set", MagicMock())
+
+    vector = matrixone_module.MatrixoneVector("Collection_1", _valid_config(matrixone_module))
+    client = vector._get_client(dimension=3, create_table=True)
+
+    assert client.kwargs["table_name"] == "collection_1"
+    client.create_full_text_index.assert_called_once()
+    matrixone_module.redis_client.set.assert_called_once()
+
+
+def test_get_client_skips_index_creation_when_cache_hits(matrixone_module, monkeypatch):
+    lock = MagicMock()
+    lock.__enter__.return_value = None
+    lock.__exit__.return_value = None
+    monkeypatch.setattr(matrixone_module.redis_client, "lock", MagicMock(return_value=lock))
+    monkeypatch.setattr(matrixone_module.redis_client, "get", MagicMock(return_value=1))
+    monkeypatch.setattr(matrixone_module.redis_client, "set", MagicMock())
+
+    vector = matrixone_module.MatrixoneVector("Collection_1", _valid_config(matrixone_module))
+    client = vector._get_client(dimension=3, create_table=True)
+
+    client.create_full_text_index.assert_not_called()
+    matrixone_module.redis_client.set.assert_not_called()
+
+
+def test_ensure_client_initializes_client_for_decorated_methods(matrixone_module):
+    vector = matrixone_module.MatrixoneVector("collection_1", _valid_config(matrixone_module))
+    vector.client = None
+    fake_client = MagicMock()
+    fake_client.get.return_value = [{"id": "seg-1"}]
+    vector._get_client = MagicMock(return_value=fake_client)
+
+    exists = vector.text_exists("seg-1")
+
+    assert exists is True
+    vector._get_client.assert_called_once_with(None, False)
+
+
+def test_search_by_full_text_parses_metadata_and_applies_threshold(matrixone_module):
+    vector = matrixone_module.MatrixoneVector("collection_1", _valid_config(matrixone_module))
+    vector.client = MagicMock()
+    vector.client.full_text_query.return_value = [
+        SimpleNamespace(document="doc-a", metadata='{"doc_id":"1"}', distance=0.1),
+        SimpleNamespace(document="doc-b", metadata={"doc_id": "2"}, distance=0.7),
+    ]
+
+    docs = vector.search_by_full_text("query", top_k=2, score_threshold=0.5, document_ids_filter=["doc-1"])
+
+    assert len(docs) == 1
+    assert docs[0].page_content == "doc-a"
+    assert docs[0].metadata["doc_id"] == "1"
+    assert docs[0].metadata["score"] == pytest.approx(0.9)
+    assert vector.client.full_text_query.call_args.kwargs["filter"] == {"document_id": {"$in": ["doc-1"]}}
+
+
+def test_get_type_and_create_delegate_to_add_texts(matrixone_module):
+    vector = matrixone_module.MatrixoneVector("collection_1", _valid_config(matrixone_module))
+    fake_client = MagicMock()
+    vector._get_client = MagicMock(return_value=fake_client)
+    vector.add_texts = MagicMock(return_value=["seg-1"])
+    docs = [Document(page_content="hello", metadata={"doc_id": "seg-1"})]
+
+    result = vector.create(docs, [[0.1, 0.2]])
+
+    assert vector.get_type() == "matrixone"
+    assert result == ["seg-1"]
+    vector._get_client.assert_called_once_with(2, True)
+    vector.add_texts.assert_called_once_with(docs, [[0.1, 0.2]])
+
+
+def test_get_client_handles_full_text_index_creation_error(matrixone_module, monkeypatch):
+    lock = MagicMock()
+    lock.__enter__.return_value = None
+    lock.__exit__.return_value = None
+    monkeypatch.setattr(matrixone_module.redis_client, "lock", MagicMock(return_value=lock))
+    monkeypatch.setattr(matrixone_module.redis_client, "get", MagicMock(return_value=None))
+    monkeypatch.setattr(matrixone_module.redis_client, "set", MagicMock())
+
+    failing_client = MagicMock()
+    failing_client.create_full_text_index.side_effect = RuntimeError("boom")
+    monkeypatch.setattr(matrixone_module, "MoVectorClient", MagicMock(return_value=failing_client))
+
+    vector = matrixone_module.MatrixoneVector("collection_1", _valid_config(matrixone_module))
+    client = vector._get_client(dimension=3, create_table=True)
+
+    assert client is failing_client
+    matrixone_module.redis_client.set.assert_not_called()
+
+
+def test_add_texts_generates_ids_and_inserts(matrixone_module, monkeypatch):
+    vector = matrixone_module.MatrixoneVector("collection_1", _valid_config(matrixone_module))
+    vector.client = MagicMock()
+    monkeypatch.setattr(matrixone_module.uuid, "uuid4", lambda: "generated-uuid")
+    docs = [
+        Document(page_content="a", metadata={"doc_id": "doc-a", "document_id": "d-1"}),
+        Document(page_content="b", metadata={"document_id": "d-2"}),
+        SimpleNamespace(page_content="c", metadata=None),
+    ]
+
+    ids = vector.add_texts(docs, [[0.1], [0.2], [0.3]])
+
+    # For current prod code, only docs with metadata get ids, so only two ids
+    assert ids == ["doc-a", "generated-uuid"]
+    vector.client.insert.assert_called_once()
+    insert_kwargs = vector.client.insert.call_args.kwargs
+    # All lists passed to insert should be the same length
+    texts = insert_kwargs["texts"]
+    embeddings = insert_kwargs["embeddings"]
+    metadatas = insert_kwargs["metadatas"]
+    ids_insert = insert_kwargs["ids"]
+    assert len(texts) == len(embeddings) == len(metadatas) == len(docs)
+    # ids may be shorter than docs for current prod code, but should match number of docs with metadata
+    assert ids_insert == ["doc-a", "generated-uuid"]
+
+
+def test_delete_and_metadata_methods(matrixone_module):
+    vector = matrixone_module.MatrixoneVector("collection_1", _valid_config(matrixone_module))
+    vector.client = MagicMock()
+    vector.client.query_by_metadata.return_value = [SimpleNamespace(id="seg-1"), SimpleNamespace(id="seg-2")]
+
+    vector.delete_by_ids([])
+    vector.client.delete.assert_not_called()
+
+    vector.delete_by_ids(["seg-1"])
+    vector.delete_by_metadata_field("document_id", "doc-1")
+    ids = vector.get_ids_by_metadata_field("document_id", "doc-1")
+    vector.delete()
+
+    assert ids == ["seg-1", "seg-2"]
+    assert vector.client.delete.call_count == 3
+
+
+def test_search_by_vector_builds_documents(matrixone_module):
+    vector = matrixone_module.MatrixoneVector("collection_1", _valid_config(matrixone_module))
+    vector.client = MagicMock()
+    vector.client.query.return_value = [
+        SimpleNamespace(document="doc-a", metadata={"doc_id": "1"}),
+        SimpleNamespace(document="doc-b", metadata={"doc_id": "2"}),
+    ]
+
+    docs = vector.search_by_vector([0.1, 0.2], top_k=2, document_ids_filter=["d-1"])
+
+    assert len(docs) == 2
+    assert docs[0].page_content == "doc-a"
+    assert docs[1].metadata["doc_id"] == "2"
+    assert vector.client.query.call_args.kwargs["filter"] == {"document_id": {"$in": ["d-1"]}}
+
+
+def test_matrixone_factory_uses_existing_or_generated_collection(matrixone_module, monkeypatch):
+    factory = matrixone_module.MatrixoneVectorFactory()
+    dataset_with_index = SimpleNamespace(
+        id="dataset-1",
+        index_struct_dict={"vector_store": {"class_prefix": "EXISTING_COLLECTION"}},
+        index_struct=None,
+    )
+    dataset_without_index = SimpleNamespace(id="dataset-2", index_struct_dict=None, index_struct=None)
+
+    monkeypatch.setattr(matrixone_module.Dataset, "gen_collection_name_by_id", lambda _id: "AUTO_COLLECTION")
+    monkeypatch.setattr(matrixone_module.dify_config, "MATRIXONE_HOST", "127.0.0.1")
+    monkeypatch.setattr(matrixone_module.dify_config, "MATRIXONE_PORT", 6001)
+    monkeypatch.setattr(matrixone_module.dify_config, "MATRIXONE_USER", "dump")
+    monkeypatch.setattr(matrixone_module.dify_config, "MATRIXONE_PASSWORD", "111")
+    monkeypatch.setattr(matrixone_module.dify_config, "MATRIXONE_DATABASE", "dify")
+    monkeypatch.setattr(matrixone_module.dify_config, "MATRIXONE_METRIC", "l2")
+
+    with patch.object(matrixone_module, "MatrixoneVector", return_value="vector") as vector_cls:
+        result_1 = factory.init_vector(dataset_with_index, attributes=[], embeddings=MagicMock())
+        result_2 = factory.init_vector(dataset_without_index, attributes=[], embeddings=MagicMock())
+
+    assert result_1 == "vector"
+    assert result_2 == "vector"
+    assert vector_cls.call_args_list[0].kwargs["collection_name"] == "EXISTING_COLLECTION"
+    assert vector_cls.call_args_list[1].kwargs["collection_name"] == "AUTO_COLLECTION"
+    assert dataset_without_index.index_struct is not None

api/tests/unit_tests/core/rag/datasource/vdb/milvus/test_milvus.py

@@ -1,18 +1,414 @@
+import importlib
+import sys
+import types
+from types import SimpleNamespace
+from unittest.mock import MagicMock, patch
+
 import pytest
 from pydantic import ValidationError
 
-from core.rag.datasource.vdb.milvus.milvus_vector import MilvusConfig
+from core.rag.models.document import Document
 
 
-def test_default_value():
+def _build_fake_pymilvus_modules():
+    pymilvus = types.ModuleType("pymilvus")
+    pymilvus.__path__ = []
+    pymilvus_milvus_client = types.ModuleType("pymilvus.milvus_client")
+    pymilvus_orm = types.ModuleType("pymilvus.orm")
+    pymilvus_orm.__path__ = []
+    pymilvus_orm_types = types.ModuleType("pymilvus.orm.types")
+
+    class MilvusError(Exception):
+        pass
+
+    class MilvusClient:
+        def __init__(self, **kwargs):
+            self.init_kwargs = kwargs
+            self.has_collection = MagicMock(return_value=False)
+            self.describe_collection = MagicMock(
+                return_value={"fields": [{"name": "id"}, {"name": "content"}, {"name": "metadata"}]}
+            )
+            self.get_server_version = MagicMock(return_value="2.5.0")
+            self.insert = MagicMock(return_value=[1])
+            self.query = MagicMock(return_value=[])
+            self.delete = MagicMock()
+            self.drop_collection = MagicMock()
+            self.search = MagicMock(return_value=[[]])
+            self.create_collection = MagicMock()
+
+    class IndexParams:
+        def __init__(self):
+            self.indexes = []
+
+        def add_index(self, **kwargs):
+            self.indexes.append(kwargs)
+
+    class DataType:
+        JSON = "JSON"
+        VARCHAR = "VARCHAR"
+        INT64 = "INT64"
+        SPARSE_FLOAT_VECTOR = "SPARSE_FLOAT_VECTOR"
+        FLOAT_VECTOR = "FLOAT_VECTOR"
+
+    class FieldSchema:
+        def __init__(self, name, dtype, **kwargs):
+            self.name = name
+            self.dtype = dtype
+            self.kwargs = kwargs
+
+    class CollectionSchema:
+        def __init__(self, fields):
+            self.fields = fields
+            self.functions = []
+
+        def add_function(self, func):
+            self.functions.append(func)
+
+    class FunctionType:
+        BM25 = "BM25"
+
+    class Function:
+        def __init__(self, **kwargs):
+            self.kwargs = kwargs
+
+    def infer_dtype_bydata(_value):
+        return DataType.FLOAT_VECTOR
+
+    pymilvus.MilvusException = MilvusError
+    pymilvus.MilvusClient = MilvusClient
+    pymilvus.IndexParams = IndexParams
+    pymilvus.CollectionSchema = CollectionSchema
+    pymilvus.DataType = DataType
+    pymilvus.FieldSchema = FieldSchema
+    pymilvus.Function = Function
+    pymilvus.FunctionType = FunctionType
+    pymilvus_milvus_client.IndexParams = IndexParams
+    pymilvus_orm.types = pymilvus_orm_types
+    pymilvus_orm_types.infer_dtype_bydata = infer_dtype_bydata
+
+    # Attach submodules for dotted imports
+    pymilvus.milvus_client = pymilvus_milvus_client
+    pymilvus.orm = pymilvus_orm
+
+    return {
+        "pymilvus": pymilvus,
+        "pymilvus.milvus_client": pymilvus_milvus_client,
+        "pymilvus.orm": pymilvus_orm,
+        "pymilvus.orm.types": pymilvus_orm_types,
+    }
+
+
+@pytest.fixture
+def milvus_module(monkeypatch):
+    for name, module in _build_fake_pymilvus_modules().items():
+        monkeypatch.setitem(sys.modules, name, module)
+
+    import core.rag.datasource.vdb.milvus.milvus_vector as module
+
+    return importlib.reload(module)
+
+
+def _config(module, **overrides):
+    values = {
+        "uri": "http://localhost:19530",
+        "user": "root",
+        "password": "Milvus",
+        "database": "default",
+        "enable_hybrid_search": False,
+        "analyzer_params": None,
+    }
+    values.update(overrides)
+    return module.MilvusConfig.model_validate(values)
+
+
+def test_config_validation_and_defaults(milvus_module):
     valid_config = {"uri": "http://localhost:19530", "user": "root", "password": "Milvus"}
 
     for key in valid_config:
         config = valid_config.copy()
         del config[key]
         with pytest.raises(ValidationError) as e:
-            MilvusConfig.model_validate(config)
+            milvus_module.MilvusConfig.model_validate(config)
         assert e.value.errors()[0]["msg"] == f"Value error, config MILVUS_{key.upper()} is required"
 
-    config = MilvusConfig.model_validate(valid_config)
+    config = milvus_module.MilvusConfig.model_validate(valid_config)
     assert config.database == "default"
+
+    token_config = milvus_module.MilvusConfig.model_validate(
+        {"uri": "http://localhost:19530", "token": "token-value", "database": "db-1"}
+    )
+    assert token_config.token == "token-value"
+
+
+def test_config_to_milvus_params(milvus_module):
+    config = _config(milvus_module, analyzer_params='{"tokenizer":"standard"}')
+
+    params = config.to_milvus_params()
+
+    assert params["uri"] == "http://localhost:19530"
+    assert params["db_name"] == "default"
+    assert params["analyzer_params"] == '{"tokenizer":"standard"}'
+
+
+def test_init_client_supports_token_and_user_password(milvus_module):
+    vector = milvus_module.MilvusVector.__new__(milvus_module.MilvusVector)
+    token_client = vector._init_client(
+        milvus_module.MilvusConfig.model_validate({"uri": "http://localhost:19530", "token": "abc", "database": "db"})
+    )
+    assert token_client.init_kwargs == {"uri": "http://localhost:19530", "token": "abc", "db_name": "db"}
+
+    user_client = vector._init_client(_config(milvus_module))
+    assert user_client.init_kwargs["uri"] == "http://localhost:19530"
+    assert user_client.init_kwargs["user"] == "root"
+    assert user_client.init_kwargs["password"] == "Milvus"
+
+
+def test_init_loads_fields_when_collection_exists(milvus_module):
+    client = milvus_module.MilvusClient(uri="http://localhost:19530")
+    client.has_collection.return_value = True
+    client.describe_collection.return_value = {
+        "fields": [{"name": "id"}, {"name": "content"}, {"name": "metadata"}, {"name": "sparse_vector"}]
+    }
+
+    with patch.object(milvus_module.MilvusVector, "_init_client", return_value=client):
+        with patch.object(milvus_module.MilvusVector, "_check_hybrid_search_support", return_value=False):
+            vector = milvus_module.MilvusVector("collection_1", _config(milvus_module))
+
+    assert "id" not in vector._fields
+    assert "content" in vector._fields
+
+
+def test_load_collection_fields_from_argument_and_remote(milvus_module):
+    vector = milvus_module.MilvusVector.__new__(milvus_module.MilvusVector)
+    vector._client = MagicMock()
+    vector._collection_name = "collection_1"
+    vector._client.describe_collection.return_value = {"fields": [{"name": "id"}, {"name": "content"}]}
+
+    vector._load_collection_fields(["id", "metadata"])
+    assert vector._fields == ["metadata"]
+
+    vector._load_collection_fields()
+    assert vector._fields == ["content"]
+
+
+def test_check_hybrid_search_support_branches(milvus_module):
+    vector = milvus_module.MilvusVector.__new__(milvus_module.MilvusVector)
+    vector._client = MagicMock()
+
+    vector._client_config = SimpleNamespace(enable_hybrid_search=False)
+    assert vector._check_hybrid_search_support() is False
+
+    vector._client_config = SimpleNamespace(enable_hybrid_search=True)
+    vector._client.get_server_version.return_value = "Zilliz Cloud 2.4"
+    assert vector._check_hybrid_search_support() is True
+
+    vector._client.get_server_version.return_value = "2.5.1"
+    assert vector._check_hybrid_search_support() is True
+
+    vector._client.get_server_version.return_value = "2.4.9"
+    assert vector._check_hybrid_search_support() is False
+
+    vector._client.get_server_version.side_effect = RuntimeError("boom")
+    assert vector._check_hybrid_search_support() is False
+
+
+def test_get_type_and_create_delegate(milvus_module):
+    vector = milvus_module.MilvusVector.__new__(milvus_module.MilvusVector)
+    vector.create_collection = MagicMock()
+    vector.add_texts = MagicMock()
+    docs = [SimpleNamespace(page_content="hello", metadata=None)]
+
+    vector.create(docs, [[0.1, 0.2]])
+
+    assert vector.get_type() == "milvus"
+    vector.create_collection.assert_called_once()
+    create_args = vector.create_collection.call_args.args
+    assert create_args[0] == [[0.1, 0.2]]
+    assert create_args[1] == [{}]
+    vector.add_texts.assert_called_once_with(docs, [[0.1, 0.2]])
+
+
+def test_add_texts_batches_and_raises_milvus_exception(milvus_module):
+    vector = milvus_module.MilvusVector.__new__(milvus_module.MilvusVector)
+    vector._collection_name = "collection_1"
+    vector._client = MagicMock()
+    vector._client.insert.side_effect = [["id-1"], ["id-2"]]
+    docs = [Document(page_content=f"text-{i}", metadata={"doc_id": f"d-{i}"}) for i in range(1001)]
+    embeddings = [[0.1, 0.2] for _ in range(1001)]
+
+    ids = vector.add_texts(docs, embeddings)
+    assert ids == ["id-1", "id-2"]
+    assert vector._client.insert.call_count == 2
+
+    vector._client.insert.side_effect = milvus_module.MilvusException("insert failed")
+    with pytest.raises(milvus_module.MilvusException):
+        vector.add_texts([Document(page_content="x", metadata={})], [[0.1]])
+
+
+def test_get_ids_and_delete_methods(milvus_module):
+    vector = milvus_module.MilvusVector.__new__(milvus_module.MilvusVector)
+    vector._collection_name = "collection_1"
+    vector._client = MagicMock()
+    vector._client.query.return_value = [{"id": 1}, {"id": 2}]
+
+    assert vector.get_ids_by_metadata_field("document_id", "doc-1") == [1, 2]
+    vector._client.query.return_value = []
+    assert vector.get_ids_by_metadata_field("document_id", "doc-1") is None
+
+    vector._client.has_collection.return_value = True
+    vector.get_ids_by_metadata_field = MagicMock(return_value=[101, 102])
+    vector.delete_by_metadata_field("document_id", "doc-1")
+    vector._client.delete.assert_called_with(collection_name="collection_1", pks=[101, 102])
+
+    vector._client.delete.reset_mock()
+    vector._client.query.return_value = [{"id": 11}, {"id": 12}]
+    vector.delete_by_ids(["doc-a", "doc-b"])
+    vector._client.delete.assert_called_with(collection_name="collection_1", pks=[11, 12])
+
+    vector._client.has_collection.return_value = True
+    vector.delete()
+    vector._client.drop_collection.assert_called_once_with("collection_1", None)
+
+
+def test_text_exists_and_field_exists(milvus_module):
+    vector = milvus_module.MilvusVector.__new__(milvus_module.MilvusVector)
+    vector._collection_name = "collection_1"
+    vector._fields = ["content", "metadata"]
+    vector._client = MagicMock()
+    vector._client.has_collection.return_value = False
+    assert vector.text_exists("doc-1") is False
+
+    vector._client.has_collection.return_value = True
+    vector._client.query.return_value = [{"id": 1}]
+    assert vector.text_exists("doc-1") is True
+    vector._client.query.return_value = []
+    assert vector.text_exists("doc-1") is False
+    assert vector.field_exists("content") is True
+    assert vector.field_exists("unknown") is False
+
+
+def test_process_search_results_and_search_methods(milvus_module):
+    vector = milvus_module.MilvusVector.__new__(milvus_module.MilvusVector)
+    vector._collection_name = "collection_1"
+    vector._client = MagicMock()
+    vector._fields = ["content", "metadata", "sparse_vector"]
+
+    processed = vector._process_search_results(
+        [
+            [
+                {"entity": {"content": "doc-1", "metadata": {"doc_id": "1"}}, "distance": 0.9},
+                {"entity": {"content": "doc-2", "metadata": {"doc_id": "2"}}, "distance": 0.2},
+            ]
+        ],
+        [milvus_module.Field.CONTENT_KEY, milvus_module.Field.METADATA_KEY],
+        score_threshold=0.5,
+    )
+    assert len(processed) == 1
+    assert processed[0].metadata["score"] == 0.9
+
+    vector._client.search.return_value = [[{"entity": {"content": "doc"}, "distance": 0.8}]]
+    vector._process_search_results = MagicMock(return_value=["doc"])
+
+    docs = vector.search_by_vector([0.1, 0.2], top_k=3, document_ids_filter=["a", "b"], score_threshold=0.1)
+    assert docs == ["doc"]
+    assert vector._client.search.call_args.kwargs["filter"] == 'metadata["document_id"] in ["a", "b"]'
+
+    vector._hybrid_search_enabled = False
+    assert vector.search_by_full_text("query") == []
+
+    vector._hybrid_search_enabled = True
+    vector._fields = []
+    assert vector.search_by_full_text("query") == []
+
+    vector._fields = [milvus_module.Field.SPARSE_VECTOR]
+    vector._process_search_results = MagicMock(return_value=["full-text-doc"])
+    full_text_docs = vector.search_by_full_text("query", top_k=2, document_ids_filter=["d-1"], score_threshold=0.2)
+    assert full_text_docs == ["full-text-doc"]
+    assert "document_id" in vector._client.search.call_args.kwargs["filter"]
+
+
+def test_create_collection_cache_and_existing_collection(milvus_module, monkeypatch):
+    lock = MagicMock()
+    lock.__enter__.return_value = None
+    lock.__exit__.return_value = None
+    monkeypatch.setattr(milvus_module.redis_client, "lock", MagicMock(return_value=lock))
+    monkeypatch.setattr(milvus_module.redis_client, "set", MagicMock())
+
+    vector = milvus_module.MilvusVector.__new__(milvus_module.MilvusVector)
+    vector._collection_name = "collection_1"
+    vector._consistency_level = "Session"
+    vector._client_config = _config(milvus_module)
+    vector._hybrid_search_enabled = False
+    vector._client = MagicMock()
+
+    monkeypatch.setattr(milvus_module.redis_client, "get", MagicMock(return_value=1))
+    vector.create_collection([[0.1, 0.2]], metadatas=[{"doc_id": "1"}], index_params={"index_type": "HNSW"})
+    vector._client.create_collection.assert_not_called()
+
+    monkeypatch.setattr(milvus_module.redis_client, "get", MagicMock(return_value=None))
+    vector._client.has_collection.return_value = True
+    vector.create_collection([[0.1, 0.2]], metadatas=[{"doc_id": "1"}], index_params={"index_type": "HNSW"})
+    milvus_module.redis_client.set.assert_called()
+
+
+def test_create_collection_builds_schema_and_indexes(milvus_module, monkeypatch):
+    lock = MagicMock()
+    lock.__enter__.return_value = None
+    lock.__exit__.return_value = None
+    monkeypatch.setattr(milvus_module.redis_client, "lock", MagicMock(return_value=lock))
+    monkeypatch.setattr(milvus_module.redis_client, "get", MagicMock(return_value=None))
+    monkeypatch.setattr(milvus_module.redis_client, "set", MagicMock())
+
+    vector = milvus_module.MilvusVector.__new__(milvus_module.MilvusVector)
+    vector._collection_name = "collection_1"
+    vector._consistency_level = "Session"
+    vector._client = MagicMock()
+    vector._client.has_collection.return_value = False
+    vector._load_collection_fields = MagicMock()
+
+    vector._client_config = _config(milvus_module, analyzer_params='{"tokenizer":"standard"}')
+    vector._hybrid_search_enabled = True
+    vector.create_collection(
+        embeddings=[[0.1, 0.2]],
+        metadatas=[{"doc_id": "1"}],
+        index_params={"metric_type": "IP", "index_type": "HNSW", "params": {"M": 8}},
+    )
+
+    call_kwargs = vector._client.create_collection.call_args.kwargs
+    schema = call_kwargs["schema"]
+    index_params_obj = call_kwargs["index_params"]
+    field_names = [f.name for f in schema.fields]
+
+    assert milvus_module.Field.SPARSE_VECTOR in field_names
+    assert len(schema.functions) == 1
+    assert len(index_params_obj.indexes) == 2
+    assert call_kwargs["consistency_level"] == "Session"
+
+
+def test_factory_initializes_milvus_vector(milvus_module, monkeypatch):
+    factory = milvus_module.MilvusVectorFactory()
+    dataset_with_index = SimpleNamespace(
+        id="dataset-1",
+        index_struct_dict={"vector_store": {"class_prefix": "EXISTING_COLLECTION"}},
+        index_struct=None,
+    )
+    dataset_without_index = SimpleNamespace(id="dataset-2", index_struct_dict=None, index_struct=None)
+
+    monkeypatch.setattr(milvus_module.Dataset, "gen_collection_name_by_id", lambda _id: "AUTO_COLLECTION")
+    monkeypatch.setattr(milvus_module.dify_config, "MILVUS_URI", "http://localhost:19530")
+    monkeypatch.setattr(milvus_module.dify_config, "MILVUS_TOKEN", "")
+    monkeypatch.setattr(milvus_module.dify_config, "MILVUS_USER", "root")
+    monkeypatch.setattr(milvus_module.dify_config, "MILVUS_PASSWORD", "Milvus")
+    monkeypatch.setattr(milvus_module.dify_config, "MILVUS_DATABASE", "default")
+    monkeypatch.setattr(milvus_module.dify_config, "MILVUS_ENABLE_HYBRID_SEARCH", True)
+    monkeypatch.setattr(milvus_module.dify_config, "MILVUS_ANALYZER_PARAMS", '{"tokenizer":"standard"}')
+
+    with patch.object(milvus_module, "MilvusVector", return_value="vector") as vector_cls:
+        result_1 = factory.init_vector(dataset_with_index, attributes=[], embeddings=MagicMock())
+        result_2 = factory.init_vector(dataset_without_index, attributes=[], embeddings=MagicMock())
+
+    assert result_1 == "vector"
+    assert result_2 == "vector"
+    assert vector_cls.call_args_list[0].kwargs["collection_name"] == "EXISTING_COLLECTION"
+    assert vector_cls.call_args_list[1].kwargs["collection_name"] == "AUTO_COLLECTION"
+    assert dataset_without_index.index_struct is not None

api/tests/unit_tests/core/rag/datasource/vdb/myscale/test_myscale_vector.py

@@ -0,0 +1,230 @@
+import importlib
+import sys
+import types
+from types import SimpleNamespace
+from unittest.mock import MagicMock, patch
+
+import pytest
+
+from core.rag.models.document import Document
+
+
+def _build_fake_clickhouse_connect_module():
+    clickhouse_connect = types.ModuleType("clickhouse_connect")
+
+    class QueryResult:
+        def __init__(self, rows=None, named_rows=None):
+            self.row_count = len(rows or [])
+            self.result_rows = rows or []
+            self._named_rows = named_rows or []
+
+        def named_results(self):
+            return self._named_rows
+
+    class Client:
+        def __init__(self):
+            self.command = MagicMock()
+            self.query = MagicMock(return_value=QueryResult())
+
+    client = Client()
+
+    def get_client(**_kwargs):
+        return client
+
+    clickhouse_connect.get_client = get_client
+    clickhouse_connect.QueryResult = QueryResult
+    clickhouse_connect._fake_client = client
+    return clickhouse_connect
+
+
+@pytest.fixture
+def myscale_module(monkeypatch):
+    fake_module = _build_fake_clickhouse_connect_module()
+    monkeypatch.setitem(sys.modules, "clickhouse_connect", fake_module)
+
+    import core.rag.datasource.vdb.myscale.myscale_vector as module
+
+    return importlib.reload(module)
+
+
+def _config(module):
+    return module.MyScaleConfig(
+        host="localhost",
+        port=8123,
+        user="default",
+        password="",
+        database="dify",
+        fts_params="",
+    )
+
+
+def test_escape_str_replaces_backslash_and_quote(myscale_module):
+    escaped = myscale_module.MyScaleVector.escape_str(r"text\with'special")
+    assert escaped == "text with special"
+
+
+def test_search_raises_for_invalid_top_k(myscale_module):
+    vector = myscale_module.MyScaleVector("collection_1", _config(myscale_module))
+    with pytest.raises(ValueError, match="top_k must be a positive integer"):
+        vector._search("distance(vector, [0.1, 0.2])", myscale_module.SortOrder.ASC, top_k=0)
+
+
+def test_search_builds_where_clause_for_cosine_threshold(myscale_module):
+    vector = myscale_module.MyScaleVector("collection_1", _config(myscale_module))
+    vector._client.query.return_value = myscale_module.get_client().query.return_value.__class__(
+        named_rows=[{"text": "doc-1", "vector": [0.1, 0.2], "metadata": {"doc_id": "seg-1"}}]
+    )
+
+    docs = vector._search("distance(vector, [0.1, 0.2])", myscale_module.SortOrder.ASC, top_k=1, score_threshold=0.2)
+
+    assert len(docs) == 1
+    sql = vector._client.query.call_args.args[0]
+    assert "WHERE dist < 0.8" in sql
+
+
+def test_delete_by_ids_short_circuits_on_empty_list(myscale_module):
+    vector = myscale_module.MyScaleVector("collection_1", _config(myscale_module))
+    vector._client.command.reset_mock()
+
+    vector.delete_by_ids([])
+    vector._client.command.assert_not_called()
+
+
+def test_factory_initializes_lower_case_collection_name(myscale_module, monkeypatch):
+    factory = myscale_module.MyScaleVectorFactory()
+    dataset_with_index = SimpleNamespace(
+        id="dataset-1",
+        index_struct_dict={"vector_store": {"class_prefix": "EXISTING_COLLECTION"}},
+        index_struct=None,
+    )
+    dataset_without_index = SimpleNamespace(id="dataset-2", index_struct_dict=None, index_struct=None)
+
+    monkeypatch.setattr(myscale_module.Dataset, "gen_collection_name_by_id", lambda _id: "AUTO_COLLECTION")
+    monkeypatch.setattr(myscale_module.dify_config, "MYSCALE_HOST", "localhost")
+    monkeypatch.setattr(myscale_module.dify_config, "MYSCALE_PORT", 8123)
+    monkeypatch.setattr(myscale_module.dify_config, "MYSCALE_USER", "default")
+    monkeypatch.setattr(myscale_module.dify_config, "MYSCALE_PASSWORD", "")
+    monkeypatch.setattr(myscale_module.dify_config, "MYSCALE_DATABASE", "dify")
+    monkeypatch.setattr(myscale_module.dify_config, "MYSCALE_FTS_PARAMS", "")
+
+    with patch.object(myscale_module, "MyScaleVector", return_value="vector") as vector_cls:
+        result_1 = factory.init_vector(dataset_with_index, attributes=[], embeddings=MagicMock())
+        result_2 = factory.init_vector(dataset_without_index, attributes=[], embeddings=MagicMock())
+
+    assert result_1 == "vector"
+    assert result_2 == "vector"
+    assert vector_cls.call_args_list[0].kwargs["collection_name"] == "existing_collection"
+    assert vector_cls.call_args_list[1].kwargs["collection_name"] == "auto_collection"
+    assert dataset_without_index.index_struct is not None
+
+
+def test_init_and_get_type_set_expected_defaults(myscale_module):
+    vector = myscale_module.MyScaleVector("collection_1", _config(myscale_module))
+
+    assert vector.get_type() == "myscale"
+    assert vector._vec_order == myscale_module.SortOrder.ASC
+    vector._client.command.assert_called_with("SET allow_experimental_object_type=1")
+
+
+def test_create_calls_create_collection_and_add_texts(myscale_module):
+    vector = myscale_module.MyScaleVector("collection_1", _config(myscale_module))
+    vector._create_collection = MagicMock()
+    vector.add_texts = MagicMock(return_value=["seg-1"])
+    docs = [Document(page_content="hello", metadata={"doc_id": "seg-1"})]
+
+    result = vector.create(docs, [[0.1, 0.2]])
+
+    assert result == ["seg-1"]
+    vector._create_collection.assert_called_once_with(2)
+    vector.add_texts.assert_called_once()
+
+
+def test_create_collection_builds_expected_sql(myscale_module):
+    config = myscale_module.MyScaleConfig(
+        host="localhost",
+        port=8123,
+        user="default",
+        password="",
+        database="dify",
+        fts_params="tokenizer=unicode",
+    )
+    vector = myscale_module.MyScaleVector("collection_1", config)
+    vector._client.command.reset_mock()
+
+    vector._create_collection(3)
+
+    assert vector._client.command.call_count == 2
+    sql = vector._client.command.call_args_list[1].args[0]
+    assert "CREATE TABLE IF NOT EXISTS dify.collection_1" in sql
+    assert "CONSTRAINT cons_vec_len CHECK length(vector) = 3" in sql
+    assert "INDEX text_idx text TYPE fts('tokenizer=unicode')" in sql
+
+
+def test_add_texts_inserts_rows_and_returns_ids(myscale_module, monkeypatch):
+    vector = myscale_module.MyScaleVector("collection_1", _config(myscale_module))
+    monkeypatch.setattr(myscale_module.uuid, "uuid4", lambda: "generated-uuid")
+    docs = [
+        Document(page_content=r"te'xt\1", metadata={"doc_id": "doc-a", "document_id": "d-1"}),
+        Document(page_content="text-2", metadata={"document_id": "d-2"}),
+        SimpleNamespace(page_content="text-3", metadata=None),
+    ]
+
+    ids = vector.add_texts(docs, [[0.1], [0.2], [0.3]])
+
+    assert ids == ["doc-a", "generated-uuid"]
+    sql = vector._client.command.call_args.args[0]
+    assert "INSERT INTO dify.collection_1" in sql
+    assert "te xt 1" in sql
+
+
+def test_text_exists_and_metadata_operations(myscale_module):
+    vector = myscale_module.MyScaleVector("collection_1", _config(myscale_module))
+    vector._client.query.return_value = SimpleNamespace(row_count=1, result_rows=[("id-1",), ("id-2",)])
+
+    assert vector.text_exists("id-1") is True
+    assert vector.get_ids_by_metadata_field("document_id", "doc-1") == ["id-1", "id-2"]
+
+    vector.delete_by_ids(["id-1", "id-2"])
+    vector.delete_by_metadata_field("document_id", "doc-1")
+    assert vector._client.command.call_count >= 2
+
+
+def test_search_delegation_methods(myscale_module):
+    vector = myscale_module.MyScaleVector("collection_1", _config(myscale_module))
+    vector._search = MagicMock(return_value=["result"])
+
+    result_vector = vector.search_by_vector([0.1, 0.2], top_k=2)
+    result_text = vector.search_by_full_text("hello", top_k=2)
+
+    assert result_vector == ["result"]
+    assert result_text == ["result"]
+    assert vector._search.call_count == 2
+
+
+def test_search_with_document_filter_and_exception(myscale_module):
+    vector = myscale_module.MyScaleVector("collection_1", _config(myscale_module))
+    vector._client.query.return_value = SimpleNamespace(
+        named_results=lambda: [{"text": "doc", "vector": [0.1], "metadata": {"doc_id": "1"}}]
+    )
+
+    docs = vector._search(
+        "distance(vector, [0.1])",
+        myscale_module.SortOrder.ASC,
+        top_k=2,
+        document_ids_filter=["doc-1", "doc-2"],
+    )
+    assert len(docs) == 1
+    sql = vector._client.query.call_args.args[0]
+    assert "metadata['document_id'] in ('doc-1', 'doc-2')" in sql
+
+    vector._client.query.side_effect = RuntimeError("boom")
+    assert vector._search("distance(vector, [0.1])", myscale_module.SortOrder.ASC, top_k=1) == []
+
+
+def test_delete_drops_table(myscale_module):
+    vector = myscale_module.MyScaleVector("collection_1", _config(myscale_module))
+    vector._client.command.reset_mock()
+
+    vector.delete()
+
+    vector._client.command.assert_called_once_with("DROP TABLE IF EXISTS dify.collection_1")

api/tests/unit_tests/core/rag/datasource/vdb/oceanbase/test_oceanbase_vector.py

@@ -0,0 +1,553 @@
+import importlib
+import sys
+import types
+from types import SimpleNamespace
+from unittest.mock import MagicMock, patch
+
+import pytest
+from pydantic import ValidationError
+from sqlalchemy.exc import SQLAlchemyError
+
+from core.rag.models.document import Document
+
+
+def _build_fake_pyobvector_module():
+    pyobvector = types.ModuleType("pyobvector")
+
+    class VECTOR:
+        def __init__(self, dim):
+            self.dim = dim
+
+    def l2_distance(*_args, **_kwargs):
+        return "l2"
+
+    def cosine_distance(*_args, **_kwargs):
+        return "cosine"
+
+    def inner_product(*_args, **_kwargs):
+        return "inner_product"
+
+    class ObVecClient:
+        def __init__(self, **_kwargs):
+            self.metadata_obj = SimpleNamespace(tables={})
+            self.engine = MagicMock()
+            self.check_table_exists = MagicMock(return_value=False)
+            self.perform_raw_text_sql = MagicMock()
+            self.prepare_index_params = MagicMock()
+            self.create_table_with_index_params = MagicMock()
+            self.refresh_metadata = MagicMock()
+            self.insert = MagicMock()
+            self.refresh_index = MagicMock()
+            self.get = MagicMock()
+            self.delete = MagicMock()
+            self.set_ob_hnsw_ef_search = MagicMock()
+            self.ann_search = MagicMock(return_value=[])
+            self.drop_table_if_exist = MagicMock()
+
+    pyobvector.VECTOR = VECTOR
+    pyobvector.ObVecClient = ObVecClient
+    pyobvector.l2_distance = l2_distance
+    pyobvector.cosine_distance = cosine_distance
+    pyobvector.inner_product = inner_product
+    return pyobvector
+
+
+@pytest.fixture
+def oceanbase_module(monkeypatch):
+    monkeypatch.setitem(sys.modules, "pyobvector", _build_fake_pyobvector_module())
+
+    import core.rag.datasource.vdb.oceanbase.oceanbase_vector as module
+
+    return importlib.reload(module)
+
+
+def _config(module):
+    return module.OceanBaseVectorConfig(
+        host="127.0.0.1",
+        port=2881,
+        user="root",
+        password="secret",
+        database="test",
+        enable_hybrid_search=True,
+        batch_size=10,
+    )
+
+
+@pytest.mark.parametrize(
+    ("field", "value", "message"),
+    [
+        ("host", "", "config OCEANBASE_VECTOR_HOST is required"),
+        ("port", 0, "config OCEANBASE_VECTOR_PORT is required"),
+        ("user", "", "config OCEANBASE_VECTOR_USER is required"),
+        ("database", "", "config OCEANBASE_VECTOR_DATABASE is required"),
+    ],
+)
+def test_oceanbase_config_validation(oceanbase_module, field, value, message):
+    values = _config(oceanbase_module).model_dump()
+    values[field] = value
+
+    with pytest.raises(ValidationError, match=message):
+        oceanbase_module.OceanBaseVectorConfig.model_validate(values)
+
+
+def test_init_rejects_invalid_collection_name(oceanbase_module):
+    with pytest.raises(ValueError, match="Invalid collection name"):
+        oceanbase_module.OceanBaseVector("invalid-name", _config(oceanbase_module))
+
+
+def test_distance_to_score_for_supported_metrics(oceanbase_module):
+    vector = oceanbase_module.OceanBaseVector.__new__(oceanbase_module.OceanBaseVector)
+    vector._config = SimpleNamespace(metric_type="l2")
+    assert vector._distance_to_score(3.0) == pytest.approx(0.25)
+
+    vector._config = SimpleNamespace(metric_type="cosine")
+    assert vector._distance_to_score(0.2) == pytest.approx(0.8)
+
+    vector._config = SimpleNamespace(metric_type="inner_product")
+    assert vector._distance_to_score(-0.2) == pytest.approx(0.2)
+
+
+def test_get_distance_func_raises_for_unknown_metric(oceanbase_module):
+    vector = oceanbase_module.OceanBaseVector.__new__(oceanbase_module.OceanBaseVector)
+    vector._config = SimpleNamespace(metric_type="manhattan")
+
+    with pytest.raises(ValueError, match="Unsupported metric_type"):
+        vector._get_distance_func()
+
+
+def test_process_search_results_handles_json_and_score_threshold(oceanbase_module):
+    vector = oceanbase_module.OceanBaseVector.__new__(oceanbase_module.OceanBaseVector)
+    rows = [
+        ("doc-1", '{"doc_id":"1"}', 0.9),
+        ("doc-2", "not-json", 0.8),
+        ("doc-3", {"doc_id": "3"}, 0.3),
+    ]
+
+    docs = vector._process_search_results(rows, score_threshold=0.5, score_key="rank")
+
+    assert len(docs) == 2
+    assert docs[0].metadata["doc_id"] == "1"
+    assert docs[0].metadata["rank"] == 0.9
+    assert docs[1].metadata["rank"] == 0.8
+
+
+def test_search_by_vector_validates_document_id_format(oceanbase_module):
+    vector = oceanbase_module.OceanBaseVector.__new__(oceanbase_module.OceanBaseVector)
+    vector._collection_name = "collection_1"
+    vector._hnsw_ef_search = -1
+    vector._config = SimpleNamespace(metric_type="cosine")
+    vector._client = MagicMock()
+
+    with pytest.raises(ValueError, match="Invalid document ID format"):
+        vector.search_by_vector([0.1, 0.2], document_ids_filter=["bad id"])
+
+
+def test_search_by_full_text_returns_empty_when_disabled(oceanbase_module):
+    vector = oceanbase_module.OceanBaseVector.__new__(oceanbase_module.OceanBaseVector)
+    vector._hybrid_search_enabled = False
+    vector._collection_name = "collection_1"
+
+    assert vector.search_by_full_text("query") == []
+
+
+def test_check_hybrid_search_support_uses_version_comment(oceanbase_module):
+    vector = oceanbase_module.OceanBaseVector.__new__(oceanbase_module.OceanBaseVector)
+    vector._config = SimpleNamespace(enable_hybrid_search=True)
+    vector._client = MagicMock()
+    cursor = MagicMock()
+    cursor.fetchone.return_value = ("OceanBase_CE 4.3.5.1 (rxxxxxxxxx) (Built Mar 18 2025)",)
+    vector._client.perform_raw_text_sql.return_value = cursor
+
+    assert vector._check_hybrid_search_support() is True
+
+    cursor.fetchone.return_value = ("OceanBase_CE 4.3.4.0 (rxxxxxxxxx) (Built Mar 18 2025)",)
+    assert vector._check_hybrid_search_support() is False
+
+
+def test_init_get_type_and_field_loading(oceanbase_module):
+    config = _config(oceanbase_module)
+    config.enable_hybrid_search = False
+
+    table = SimpleNamespace(columns=[SimpleNamespace(name="id"), SimpleNamespace(name="text")])
+    fake_client = oceanbase_module.ObVecClient()
+    fake_client.check_table_exists.return_value = True
+    fake_client.metadata_obj.tables = {"collection_1": table}
+
+    with patch.object(oceanbase_module, "ObVecClient", return_value=fake_client):
+        vector = oceanbase_module.OceanBaseVector("collection_1", config)
+
+    assert vector.get_type() == "oceanbase"
+    assert vector.field_exists("text") is True
+
+
+def test_load_collection_fields_handles_missing_table_and_exception(oceanbase_module):
+    vector = oceanbase_module.OceanBaseVector.__new__(oceanbase_module.OceanBaseVector)
+    vector._collection_name = "collection_1"
+    vector._fields = []
+    vector._client = MagicMock()
+    vector._client.metadata_obj.tables = {}
+
+    vector._load_collection_fields()
+    assert vector._fields == []
+
+    vector._client.metadata_obj.tables = {"collection_1": MagicMock(columns=MagicMock(side_effect=RuntimeError("x")))}
+    vector._load_collection_fields()
+    assert vector._fields == []
+
+
+def test_create_delegates_to_collection_and_insert(oceanbase_module):
+    vector = oceanbase_module.OceanBaseVector.__new__(oceanbase_module.OceanBaseVector)
+    vector._create_collection = MagicMock()
+    vector.add_texts = MagicMock()
+    docs = [Document(page_content="text", metadata={"doc_id": "1"})]
+
+    vector.create(docs, [[0.1, 0.2]])
+
+    assert vector._vec_dim == 2
+    vector._create_collection.assert_called_once()
+    vector.add_texts.assert_called_once_with(docs, [[0.1, 0.2]])
+
+
+def test_create_collection_cache_and_existing_table_short_circuits(oceanbase_module, monkeypatch):
+    lock = MagicMock()
+    lock.__enter__.return_value = None
+    lock.__exit__.return_value = None
+    monkeypatch.setattr(oceanbase_module.redis_client, "lock", MagicMock(return_value=lock))
+    monkeypatch.setattr(oceanbase_module.redis_client, "set", MagicMock())
+
+    vector = oceanbase_module.OceanBaseVector.__new__(oceanbase_module.OceanBaseVector)
+    vector._collection_name = "collection_1"
+    vector._vec_dim = 2
+    vector._hybrid_search_enabled = False
+    vector._config = SimpleNamespace(metric_type="cosine", hnsw_m=16, hnsw_ef_construction=64)
+    vector._client = MagicMock()
+    vector.delete = MagicMock()
+    vector._load_collection_fields = MagicMock()
+
+    monkeypatch.setattr(oceanbase_module.redis_client, "get", MagicMock(return_value=1))
+    vector._create_collection()
+    vector._client.check_table_exists.assert_not_called()
+
+    monkeypatch.setattr(oceanbase_module.redis_client, "get", MagicMock(return_value=None))
+    vector._client.check_table_exists.return_value = True
+    vector._create_collection()
+    vector.delete.assert_not_called()
+
+
+def test_create_collection_happy_path_with_hybrid_and_index(oceanbase_module, monkeypatch):
+    lock = MagicMock()
+    lock.__enter__.return_value = None
+    lock.__exit__.return_value = None
+    monkeypatch.setattr(oceanbase_module.redis_client, "lock", MagicMock(return_value=lock))
+    monkeypatch.setattr(oceanbase_module.redis_client, "get", MagicMock(return_value=None))
+    monkeypatch.setattr(oceanbase_module.redis_client, "set", MagicMock())
+    monkeypatch.setattr(oceanbase_module.dify_config, "OCEANBASE_FULLTEXT_PARSER", "ik")
+    monkeypatch.setattr(oceanbase_module, "Column", lambda *args, **kwargs: SimpleNamespace(args=args, kwargs=kwargs))
+    monkeypatch.setattr(oceanbase_module, "VECTOR", lambda dim: SimpleNamespace(dim=dim))
+
+    vector = oceanbase_module.OceanBaseVector.__new__(oceanbase_module.OceanBaseVector)
+    vector._collection_name = "collection_1"
+    vector._vec_dim = 3
+    vector._hybrid_search_enabled = True
+    vector._config = SimpleNamespace(metric_type="cosine", hnsw_m=16, hnsw_ef_construction=64)
+    vector._client = MagicMock()
+    vector._client.check_table_exists.return_value = False
+    vector._client.perform_raw_text_sql.side_effect = [
+        [[None, None, None, None, None, None, "30"]],
+        None,
+        None,
+    ]
+    index_params = MagicMock()
+    vector._client.prepare_index_params.return_value = index_params
+    vector.delete = MagicMock()
+    vector._load_collection_fields = MagicMock()
+
+    vector._create_collection()
+
+    vector.delete.assert_called_once()
+    vector._client.create_table_with_index_params.assert_called_once()
+    index_params.add_index.assert_called_once()
+    vector._client.refresh_metadata.assert_called_once_with(["collection_1"])
+    oceanbase_module.redis_client.set.assert_called_once()
+
+
+def test_create_collection_error_paths(oceanbase_module, monkeypatch):
+    lock = MagicMock()
+    lock.__enter__.return_value = None
+    lock.__exit__.return_value = None
+    monkeypatch.setattr(oceanbase_module.redis_client, "lock", MagicMock(return_value=lock))
+    monkeypatch.setattr(oceanbase_module.redis_client, "get", MagicMock(return_value=None))
+    monkeypatch.setattr(oceanbase_module, "Column", lambda *args, **kwargs: SimpleNamespace(args=args, kwargs=kwargs))
+    monkeypatch.setattr(oceanbase_module, "VECTOR", lambda dim: SimpleNamespace(dim=dim))
+
+    vector = oceanbase_module.OceanBaseVector.__new__(oceanbase_module.OceanBaseVector)
+    vector._collection_name = "collection_1"
+    vector._vec_dim = 2
+    vector._hybrid_search_enabled = True
+    vector._config = SimpleNamespace(metric_type="cosine", hnsw_m=16, hnsw_ef_construction=64)
+    vector._client = MagicMock()
+    vector._client.check_table_exists.return_value = False
+    vector._client.prepare_index_params.return_value = MagicMock()
+    vector.delete = MagicMock()
+    vector._load_collection_fields = MagicMock()
+
+    vector._client.perform_raw_text_sql.return_value = []
+    with pytest.raises(ValueError, match="ob_vector_memory_limit_percentage not found"):
+        vector._create_collection()
+
+    vector._client.perform_raw_text_sql.side_effect = [
+        [[None, None, None, None, None, None, "0"]],
+        RuntimeError("no privilege"),
+    ]
+    with pytest.raises(Exception, match="Failed to set ob_vector_memory_limit_percentage"):
+        vector._create_collection()
+
+    vector._client.perform_raw_text_sql.side_effect = [[[None, None, None, None, None, None, "30"]]]
+    monkeypatch.setattr(oceanbase_module.dify_config, "OCEANBASE_FULLTEXT_PARSER", "not-valid")
+    with pytest.raises(ValueError, match="Invalid OceanBase full-text parser"):
+        vector._create_collection()
+
+
+def test_create_collection_fulltext_and_metadata_index_exceptions(oceanbase_module, monkeypatch):
+    lock = MagicMock()
+    lock.__enter__.return_value = None
+    lock.__exit__.return_value = None
+    monkeypatch.setattr(oceanbase_module.redis_client, "lock", MagicMock(return_value=lock))
+    monkeypatch.setattr(oceanbase_module.redis_client, "get", MagicMock(return_value=None))
+    monkeypatch.setattr(oceanbase_module.redis_client, "set", MagicMock())
+    monkeypatch.setattr(oceanbase_module.dify_config, "OCEANBASE_FULLTEXT_PARSER", "ik")
+    monkeypatch.setattr(oceanbase_module, "Column", lambda *args, **kwargs: SimpleNamespace(args=args, kwargs=kwargs))
+    monkeypatch.setattr(oceanbase_module, "VECTOR", lambda dim: SimpleNamespace(dim=dim))
+
+    vector = oceanbase_module.OceanBaseVector.__new__(oceanbase_module.OceanBaseVector)
+    vector._collection_name = "collection_1"
+    vector._vec_dim = 2
+    vector._hybrid_search_enabled = True
+    vector._config = SimpleNamespace(metric_type="cosine", hnsw_m=16, hnsw_ef_construction=64)
+    vector._client = MagicMock()
+    vector._client.check_table_exists.return_value = False
+    vector._client.prepare_index_params.return_value = MagicMock()
+    vector.delete = MagicMock()
+    vector._load_collection_fields = MagicMock()
+
+    vector._client.perform_raw_text_sql.side_effect = [
+        [[None, None, None, None, None, None, "30"]],
+        RuntimeError("fulltext failed"),
+    ]
+    with pytest.raises(Exception, match="Failed to add fulltext index"):
+        vector._create_collection()
+
+    vector._hybrid_search_enabled = False
+    vector._client.perform_raw_text_sql.side_effect = [
+        [[None, None, None, None, None, None, "30"]],
+        SQLAlchemyError("metadata index failed"),
+    ]
+    vector._create_collection()
+    vector._client.refresh_metadata.assert_called_once_with(["collection_1"])
+
+
+def test_check_hybrid_search_support_false_and_exception(oceanbase_module):
+    vector = oceanbase_module.OceanBaseVector.__new__(oceanbase_module.OceanBaseVector)
+    vector._config = SimpleNamespace(enable_hybrid_search=False)
+    vector._client = MagicMock()
+    assert vector._check_hybrid_search_support() is False
+
+    vector._config = SimpleNamespace(enable_hybrid_search=True)
+    vector._client.perform_raw_text_sql.side_effect = RuntimeError("boom")
+    assert vector._check_hybrid_search_support() is False
+
+
+def test_add_texts_batches_refresh_and_exceptions(oceanbase_module):
+    vector = oceanbase_module.OceanBaseVector.__new__(oceanbase_module.OceanBaseVector)
+    vector._collection_name = "collection_1"
+    vector._config = SimpleNamespace(batch_size=2, hnsw_refresh_threshold=2)
+    vector._client = MagicMock()
+    vector._get_uuids = MagicMock(return_value=["id-1", "id-2", "id-3"])
+    docs = [
+        Document(page_content="a", metadata={"doc_id": "id-1"}),
+        Document(page_content="b", metadata={"doc_id": "id-2"}),
+        Document(page_content="c", metadata={"doc_id": "id-3"}),
+    ]
+
+    vector.add_texts(docs, [[0.1], [0.2], [0.3]])
+    assert vector._client.insert.call_count == 2
+    vector._client.refresh_index.assert_called_once()
+
+    vector._client.insert.reset_mock()
+    vector._client.refresh_index.reset_mock()
+    vector._client.insert.side_effect = RuntimeError("insert failed")
+    with pytest.raises(Exception, match="Failed to insert batch"):
+        vector.add_texts([docs[0]], [[0.1]])
+
+    vector._client.insert.side_effect = None
+    vector._client.insert.return_value = None
+    vector._client.refresh_index.side_effect = SQLAlchemyError("refresh failed")
+    vector._config = SimpleNamespace(batch_size=10, hnsw_refresh_threshold=1)
+    vector._get_uuids.return_value = ["id-1"]
+    vector.add_texts([docs[0]], [[0.1]])
+
+
+def test_text_exists_and_delete_by_ids(oceanbase_module):
+    vector = oceanbase_module.OceanBaseVector.__new__(oceanbase_module.OceanBaseVector)
+    vector._collection_name = "collection_1"
+    vector._client = MagicMock()
+    vector._client.get.return_value = SimpleNamespace(rowcount=1)
+    assert vector.text_exists("id-1") is True
+
+    vector._client.get.side_effect = RuntimeError("boom")
+    with pytest.raises(Exception, match="Failed to check text existence"):
+        vector.text_exists("id-1")
+
+    vector.delete_by_ids([])
+    vector._client.delete.assert_not_called()
+
+    vector._client.delete.side_effect = None
+    vector.delete_by_ids(["id-1"])
+    vector._client.delete.assert_called_once()
+
+    vector._client.delete.side_effect = RuntimeError("boom")
+    with pytest.raises(Exception, match="Failed to delete documents"):
+        vector.delete_by_ids(["id-1"])
+
+
+def test_get_ids_and_delete_by_metadata_field(oceanbase_module):
+    vector = oceanbase_module.OceanBaseVector.__new__(oceanbase_module.OceanBaseVector)
+    vector._collection_name = "collection_1"
+    vector._client = MagicMock()
+    execute_result = [("id-1",), ("id-2",)]
+
+    conn = MagicMock()
+    conn.__enter__.return_value = conn
+    conn.__exit__.return_value = None
+    conn.execute.return_value = execute_result
+    vector._client.engine.connect.return_value = conn
+
+    ids = vector.get_ids_by_metadata_field("document_id", "doc-1")
+    assert ids == ["id-1", "id-2"]
+
+    with pytest.raises(Exception, match="Failed to query documents by metadata field"):
+        vector.get_ids_by_metadata_field("bad key!", "doc-1")
+
+    vector.get_ids_by_metadata_field = MagicMock(return_value=["id-1"])
+    vector.delete_by_ids = MagicMock()
+    vector.delete_by_metadata_field("document_id", "doc-1")
+    vector.delete_by_ids.assert_called_once_with(["id-1"])
+
+    vector.get_ids_by_metadata_field = MagicMock(return_value=[])
+    vector.delete_by_ids.reset_mock()
+    vector.delete_by_metadata_field("document_id", "doc-1")
+    vector.delete_by_ids.assert_not_called()
+
+
+def test_search_by_full_text_paths(oceanbase_module):
+    vector = oceanbase_module.OceanBaseVector.__new__(oceanbase_module.OceanBaseVector)
+    vector._collection_name = "collection_1"
+    vector._hybrid_search_enabled = True
+    vector.field_exists = MagicMock(return_value=False)
+
+    assert vector.search_by_full_text("query") == []
+
+    vector.field_exists.return_value = True
+    vector._client = MagicMock()
+    conn = MagicMock()
+    tx = MagicMock()
+    tx.__enter__.return_value = tx
+    tx.__exit__.return_value = None
+    conn.begin.return_value = tx
+    conn.__enter__.return_value = conn
+    conn.__exit__.return_value = None
+    conn.execute.return_value.fetchall.return_value = [("text-1", '{"doc_id":"1"}', 0.9)]
+    vector._client.engine.connect.return_value = conn
+
+    docs = vector.search_by_full_text("query", top_k=2, document_ids_filter=["d-1"], score_threshold=0.5)
+    assert len(docs) == 1
+    assert docs[0].metadata["score"] == 0.9
+
+    with pytest.raises(Exception, match="Full-text search failed"):
+        vector.search_by_full_text("query", top_k=0)
+
+
+def test_search_by_vector_paths(oceanbase_module):
+    vector = oceanbase_module.OceanBaseVector.__new__(oceanbase_module.OceanBaseVector)
+    vector._collection_name = "collection_1"
+    vector._hnsw_ef_search = -1
+    vector._config = SimpleNamespace(metric_type="cosine")
+    vector._client = MagicMock()
+    vector._client.ann_search.return_value = [("doc-1", '{"doc_id":"1"}', 0.2)]
+    vector._process_search_results = MagicMock(return_value=["doc"])
+
+    docs = vector.search_by_vector(
+        [0.1, 0.2],
+        ef_search=10,
+        top_k=3,
+        score_threshold=0.1,
+        document_ids_filter=["good_id"],
+    )
+    assert docs == ["doc"]
+    vector._client.set_ob_hnsw_ef_search.assert_called_once_with(10)
+
+    with pytest.raises(ValueError, match="Invalid score_threshold parameter"):
+        vector.search_by_vector([0.1], score_threshold="x")
+
+    vector._client.ann_search.side_effect = RuntimeError("boom")
+    with pytest.raises(Exception, match="Vector search failed"):
+        vector.search_by_vector([0.1], score_threshold=0.1)
+
+
+def test_get_distance_func_and_distance_to_score_errors(oceanbase_module):
+    vector = oceanbase_module.OceanBaseVector.__new__(oceanbase_module.OceanBaseVector)
+    vector._config = SimpleNamespace(metric_type="cosine")
+    assert vector._get_distance_func() is oceanbase_module.cosine_distance
+
+    vector._config = SimpleNamespace(metric_type="unknown")
+    with pytest.raises(ValueError, match="Unsupported metric_type"):
+        vector._distance_to_score(0.1)
+
+
+def test_delete_success_and_exception(oceanbase_module):
+    vector = oceanbase_module.OceanBaseVector.__new__(oceanbase_module.OceanBaseVector)
+    vector._collection_name = "collection_1"
+    vector._client = MagicMock()
+
+    vector.delete()
+    vector._client.drop_table_if_exist.assert_called_once_with("collection_1")
+
+    vector._client.drop_table_if_exist.side_effect = RuntimeError("boom")
+    with pytest.raises(Exception, match="Failed to delete collection"):
+        vector.delete()
+
+
+def test_oceanbase_factory_uses_existing_or_generated_collection(oceanbase_module, monkeypatch):
+    factory = oceanbase_module.OceanBaseVectorFactory()
+    dataset_with_index = SimpleNamespace(
+        id="dataset-1",
+        index_struct_dict={"vector_store": {"class_prefix": "EXISTING_COLLECTION"}},
+        index_struct=None,
+    )
+    dataset_without_index = SimpleNamespace(id="dataset-2", index_struct_dict=None, index_struct=None)
+
+    monkeypatch.setattr(oceanbase_module.Dataset, "gen_collection_name_by_id", lambda _id: "AUTO_COLLECTION")
+    monkeypatch.setattr(oceanbase_module.dify_config, "OCEANBASE_VECTOR_HOST", "127.0.0.1")
+    monkeypatch.setattr(oceanbase_module.dify_config, "OCEANBASE_VECTOR_PORT", 2881)
+    monkeypatch.setattr(oceanbase_module.dify_config, "OCEANBASE_VECTOR_USER", "root")
+    monkeypatch.setattr(oceanbase_module.dify_config, "OCEANBASE_VECTOR_PASSWORD", "password")
+    monkeypatch.setattr(oceanbase_module.dify_config, "OCEANBASE_VECTOR_DATABASE", "test")
+    monkeypatch.setattr(oceanbase_module.dify_config, "OCEANBASE_ENABLE_HYBRID_SEARCH", True)
+    monkeypatch.setattr(oceanbase_module.dify_config, "OCEANBASE_VECTOR_BATCH_SIZE", 10)
+    monkeypatch.setattr(oceanbase_module.dify_config, "OCEANBASE_VECTOR_METRIC_TYPE", "cosine")
+    monkeypatch.setattr(oceanbase_module.dify_config, "OCEANBASE_HNSW_M", 16)
+    monkeypatch.setattr(oceanbase_module.dify_config, "OCEANBASE_HNSW_EF_CONSTRUCTION", 64)
+    monkeypatch.setattr(oceanbase_module.dify_config, "OCEANBASE_HNSW_EF_SEARCH", -1)
+    monkeypatch.setattr(oceanbase_module.dify_config, "OCEANBASE_VECTOR_POOL_SIZE", 5)
+    monkeypatch.setattr(oceanbase_module.dify_config, "OCEANBASE_VECTOR_MAX_OVERFLOW", 10)
+    monkeypatch.setattr(oceanbase_module.dify_config, "OCEANBASE_HNSW_REFRESH_THRESHOLD", 1000)
+
+    with patch.object(oceanbase_module, "OceanBaseVector", return_value="vector") as vector_cls:
+        result_1 = factory.init_vector(dataset_with_index, attributes=[], embeddings=MagicMock())
+        result_2 = factory.init_vector(dataset_without_index, attributes=[], embeddings=MagicMock())
+
+    assert result_1 == "vector"
+    assert result_2 == "vector"
+    assert vector_cls.call_args_list[0].args[0] == "existing_collection"
+    assert vector_cls.call_args_list[1].args[0] == "auto_collection"
+    assert dataset_without_index.index_struct is not None

api/tests/unit_tests/core/rag/datasource/vdb/opengauss/test_opengauss.py

@@ -0,0 +1,400 @@
+import importlib
+import sys
+import types
+from contextlib import contextmanager
+from types import SimpleNamespace
+from unittest.mock import MagicMock, patch
+
+import pytest
+from pydantic import ValidationError
+
+from core.rag.models.document import Document
+
+
+def _build_fake_psycopg2_modules():
+    psycopg2 = types.ModuleType("psycopg2")
+    psycopg2.__path__ = []
+    psycopg2_extras = types.ModuleType("psycopg2.extras")
+    psycopg2_pool = types.ModuleType("psycopg2.pool")
+
+    class SimpleConnectionPool:
+        def __init__(self, *args, **kwargs):
+            self.args = args
+            self.kwargs = kwargs
+            self.getconn = MagicMock()
+            self.putconn = MagicMock()
+
+    psycopg2_pool.SimpleConnectionPool = SimpleConnectionPool
+    psycopg2_extras.execute_values = MagicMock()
+
+    psycopg2.pool = psycopg2_pool
+    psycopg2.extras = psycopg2_extras
+    return {
+        "psycopg2": psycopg2,
+        "psycopg2.pool": psycopg2_pool,
+        "psycopg2.extras": psycopg2_extras,
+    }
+
+
+@pytest.fixture
+def opengauss_module(monkeypatch):
+    for name, module in _build_fake_psycopg2_modules().items():
+        monkeypatch.setitem(sys.modules, name, module)
+
+    import core.rag.datasource.vdb.opengauss.opengauss as module
+
+    return importlib.reload(module)
+
+
+def _config(module, *, enable_pq=False):
+    return module.OpenGaussConfig(
+        host="localhost",
+        port=6600,
+        user="postgres",
+        password="password",
+        database="dify",
+        min_connection=1,
+        max_connection=5,
+        enable_pq=enable_pq,
+    )
+
+
+@pytest.mark.parametrize(
+    ("field", "value", "message"),
+    [
+        ("host", "", "config OPENGAUSS_HOST is required"),
+        ("port", 0, "config OPENGAUSS_PORT is required"),
+        ("user", "", "config OPENGAUSS_USER is required"),
+        ("password", "", "config OPENGAUSS_PASSWORD is required"),
+        ("database", "", "config OPENGAUSS_DATABASE is required"),
+        ("min_connection", 0, "config OPENGAUSS_MIN_CONNECTION is required"),
+        ("max_connection", 0, "config OPENGAUSS_MAX_CONNECTION is required"),
+    ],
+)
+def test_opengauss_config_validation(opengauss_module, field, value, message):
+    values = _config(opengauss_module).model_dump()
+    values[field] = value
+
+    with pytest.raises(ValidationError, match=message):
+        opengauss_module.OpenGaussConfig.model_validate(values)
+
+
+def test_opengauss_config_validation_rejects_min_greater_than_max(opengauss_module):
+    values = _config(opengauss_module).model_dump()
+    values["min_connection"] = 6
+    values["max_connection"] = 5
+
+    with pytest.raises(ValidationError, match="OPENGAUSS_MIN_CONNECTION should less than OPENGAUSS_MAX_CONNECTION"):
+        opengauss_module.OpenGaussConfig.model_validate(values)
+
+
+def test_init_sets_table_name_and_vector_type(opengauss_module, monkeypatch):
+    pool = MagicMock()
+    monkeypatch.setattr(opengauss_module.psycopg2.pool, "SimpleConnectionPool", MagicMock(return_value=pool))
+
+    vector = opengauss_module.OpenGauss("collection_1", _config(opengauss_module))
+
+    assert vector.table_name == "embedding_collection_1"
+    assert vector.get_type() == "opengauss"
+    assert vector.pool is pool
+
+
+def test_create_index_with_pq_executes_pq_sql(opengauss_module, monkeypatch):
+    pool = MagicMock()
+    monkeypatch.setattr(opengauss_module.psycopg2.pool, "SimpleConnectionPool", MagicMock(return_value=pool))
+
+    lock = MagicMock()
+    lock.__enter__.return_value = None
+    lock.__exit__.return_value = None
+    monkeypatch.setattr(opengauss_module.redis_client, "lock", MagicMock(return_value=lock))
+    monkeypatch.setattr(opengauss_module.redis_client, "get", MagicMock(return_value=None))
+    monkeypatch.setattr(opengauss_module.redis_client, "set", MagicMock())
+
+    vector = opengauss_module.OpenGauss("collection_1", _config(opengauss_module, enable_pq=True))
+    cursor = MagicMock()
+
+    @contextmanager
+    def _cursor_ctx():
+        yield cursor
+
+    vector._get_cursor = _cursor_ctx
+    vector._create_index(1536)
+
+    executed_sql = [call.args[0] for call in cursor.execute.call_args_list]
+    assert any("enable_pq=on" in sql for sql in executed_sql)
+    assert any("SET hnsw_earlystop_threshold = 320" in sql for sql in executed_sql)
+    opengauss_module.redis_client.set.assert_called_once()
+
+
+def test_create_index_skips_index_sql_for_large_dimension(opengauss_module, monkeypatch):
+    pool = MagicMock()
+    monkeypatch.setattr(opengauss_module.psycopg2.pool, "SimpleConnectionPool", MagicMock(return_value=pool))
+
+    lock = MagicMock()
+    lock.__enter__.return_value = None
+    lock.__exit__.return_value = None
+    monkeypatch.setattr(opengauss_module.redis_client, "lock", MagicMock(return_value=lock))
+    monkeypatch.setattr(opengauss_module.redis_client, "get", MagicMock(return_value=None))
+    monkeypatch.setattr(opengauss_module.redis_client, "set", MagicMock())
+
+    vector = opengauss_module.OpenGauss("collection_1", _config(opengauss_module, enable_pq=False))
+    cursor = MagicMock()
+
+    @contextmanager
+    def _cursor_ctx():
+        yield cursor
+
+    vector._get_cursor = _cursor_ctx
+    vector._create_index(3072)
+
+    cursor.execute.assert_not_called()
+    opengauss_module.redis_client.set.assert_called_once()
+
+
+def test_search_by_vector_validates_top_k(opengauss_module):
+    vector = opengauss_module.OpenGauss.__new__(opengauss_module.OpenGauss)
+
+    with pytest.raises(ValueError, match="top_k must be a positive integer"):
+        vector.search_by_vector([0.1, 0.2], top_k=0)
+
+
+def test_delete_by_ids_short_circuits_with_empty_input(opengauss_module, monkeypatch):
+    pool = MagicMock()
+    monkeypatch.setattr(opengauss_module.psycopg2.pool, "SimpleConnectionPool", MagicMock(return_value=pool))
+    vector = opengauss_module.OpenGauss("collection_1", _config(opengauss_module))
+    vector._get_cursor = MagicMock()
+
+    vector.delete_by_ids([])
+
+    vector._get_cursor.assert_not_called()
+
+
+def test_get_cursor_closes_commits_and_returns_connection(opengauss_module):
+    vector = opengauss_module.OpenGauss.__new__(opengauss_module.OpenGauss)
+    pool = MagicMock()
+    conn = MagicMock()
+    cur = MagicMock()
+    pool.getconn.return_value = conn
+    conn.cursor.return_value = cur
+    vector.pool = pool
+
+    with vector._get_cursor() as got_cur:
+        assert got_cur is cur
+
+    cur.close.assert_called_once()
+    conn.commit.assert_called_once()
+    pool.putconn.assert_called_once_with(conn)
+
+
+def test_create_calls_collection_insert_and_index(opengauss_module):
+    vector = opengauss_module.OpenGauss.__new__(opengauss_module.OpenGauss)
+    vector._create_collection = MagicMock()
+    vector.add_texts = MagicMock()
+    vector._create_index = MagicMock()
+    docs = [Document(page_content="text", metadata={"doc_id": "seg-1"})]
+
+    vector.create(docs, [[0.1, 0.2]])
+
+    vector._create_collection.assert_called_once_with(2)
+    vector.add_texts.assert_called_once_with(docs, [[0.1, 0.2]])
+    vector._create_index.assert_called_once_with(2)
+
+
+def test_create_index_returns_early_on_cache_hit(opengauss_module, monkeypatch):
+    pool = MagicMock()
+    monkeypatch.setattr(opengauss_module.psycopg2.pool, "SimpleConnectionPool", MagicMock(return_value=pool))
+
+    lock = MagicMock()
+    lock.__enter__.return_value = None
+    lock.__exit__.return_value = None
+    monkeypatch.setattr(opengauss_module.redis_client, "lock", MagicMock(return_value=lock))
+    monkeypatch.setattr(opengauss_module.redis_client, "get", MagicMock(return_value=1))
+    monkeypatch.setattr(opengauss_module.redis_client, "set", MagicMock())
+
+    vector = opengauss_module.OpenGauss("collection_1", _config(opengauss_module))
+    vector._get_cursor = MagicMock()
+
+    vector._create_index(1536)
+
+    vector._get_cursor.assert_not_called()
+    opengauss_module.redis_client.set.assert_not_called()
+
+
+def test_create_index_without_pq_executes_standard_index_sql(opengauss_module, monkeypatch):
+    pool = MagicMock()
+    monkeypatch.setattr(opengauss_module.psycopg2.pool, "SimpleConnectionPool", MagicMock(return_value=pool))
+
+    lock = MagicMock()
+    lock.__enter__.return_value = None
+    lock.__exit__.return_value = None
+    monkeypatch.setattr(opengauss_module.redis_client, "lock", MagicMock(return_value=lock))
+    monkeypatch.setattr(opengauss_module.redis_client, "get", MagicMock(return_value=None))
+    monkeypatch.setattr(opengauss_module.redis_client, "set", MagicMock())
+
+    vector = opengauss_module.OpenGauss("collection_1", _config(opengauss_module, enable_pq=False))
+    cursor = MagicMock()
+
+    @contextmanager
+    def _cursor_ctx():
+        yield cursor
+
+    vector._get_cursor = _cursor_ctx
+    vector._create_index(1536)
+
+    sql = [call.args[0] for call in cursor.execute.call_args_list]
+    assert any("embedding_cosine_embedding_collection_1_idx" in query for query in sql)
+
+
+def test_add_texts_uses_execute_values(opengauss_module, monkeypatch):
+    pool = MagicMock()
+    monkeypatch.setattr(opengauss_module.psycopg2.pool, "SimpleConnectionPool", MagicMock(return_value=pool))
+    vector = opengauss_module.OpenGauss("collection_1", _config(opengauss_module))
+    cursor = MagicMock()
+    opengauss_module.psycopg2.extras.execute_values.reset_mock()
+
+    @contextmanager
+    def _cursor_ctx():
+        yield cursor
+
+    vector._get_cursor = _cursor_ctx
+    docs = [
+        Document(page_content="text-1", metadata={"doc_id": "seg-1", "document_id": "d-1"}),
+        SimpleNamespace(page_content="text-2", metadata=None),
+    ]
+    monkeypatch.setattr(opengauss_module.uuid, "uuid4", lambda: "generated-uuid")
+
+    ids = vector.add_texts(docs, [[0.1], [0.2]])
+
+    assert ids == ["seg-1"]
+    opengauss_module.psycopg2.extras.execute_values.assert_called_once()
+
+
+def test_text_exists_and_get_by_ids(opengauss_module):
+    vector = opengauss_module.OpenGauss.__new__(opengauss_module.OpenGauss)
+    vector.table_name = "embedding_collection_1"
+    cursor = MagicMock()
+    cursor.fetchone.return_value = ("seg-1",)
+    cursor.__iter__.return_value = iter([({"doc_id": "1"}, "text-1"), ({"doc_id": "2"}, "text-2")])
+
+    @contextmanager
+    def _cursor_ctx():
+        yield cursor
+
+    vector._get_cursor = _cursor_ctx
+
+    assert vector.text_exists("seg-1") is True
+    docs = vector.get_by_ids(["seg-1", "seg-2"])
+    assert len(docs) == 2
+    assert docs[0].page_content == "text-1"
+
+
+def test_delete_and_metadata_field_queries(opengauss_module):
+    vector = opengauss_module.OpenGauss.__new__(opengauss_module.OpenGauss)
+    vector.table_name = "embedding_collection_1"
+    cursor = MagicMock()
+
+    @contextmanager
+    def _cursor_ctx():
+        yield cursor
+
+    vector._get_cursor = _cursor_ctx
+
+    vector.delete_by_ids(["seg-1", "seg-2"])
+    vector.delete_by_metadata_field("document_id", "doc-1")
+    vector.delete()
+
+    sql = [call.args[0] for call in cursor.execute.call_args_list]
+    assert any("DELETE FROM embedding_collection_1 WHERE id IN %s" in query for query in sql)
+    assert any("meta->>%s = %s" in query for query in sql)
+    assert any("DROP TABLE IF EXISTS embedding_collection_1" in query for query in sql)
+
+
+def test_search_by_vector_and_full_text(opengauss_module):
+    vector = opengauss_module.OpenGauss.__new__(opengauss_module.OpenGauss)
+    vector.table_name = "embedding_collection_1"
+    cursor = MagicMock()
+    cursor.__iter__.return_value = iter(
+        [
+            ({"doc_id": "1"}, "text-1", 0.1),
+            ({"doc_id": "2"}, "text-2", 0.6),
+        ]
+    )
+
+    @contextmanager
+    def _cursor_ctx():
+        yield cursor
+
+    vector._get_cursor = _cursor_ctx
+
+    docs = vector.search_by_vector([0.1, 0.2], top_k=2, score_threshold=0.5)
+    assert len(docs) == 1
+    assert docs[0].metadata["score"] == pytest.approx(0.9)
+
+    cursor.__iter__.return_value = iter([({"doc_id": "3"}, "full-text", 0.8)])
+    full_docs = vector.search_by_full_text("hello world", top_k=2)
+    assert len(full_docs) == 1
+    assert full_docs[0].page_content == "full-text"
+
+
+def test_search_by_full_text_validates_top_k(opengauss_module):
+    vector = opengauss_module.OpenGauss.__new__(opengauss_module.OpenGauss)
+    with pytest.raises(ValueError, match="top_k must be a positive integer"):
+        vector.search_by_full_text("query", top_k=0)
+
+
+def test_create_collection_cache_and_create_path(opengauss_module, monkeypatch):
+    pool = MagicMock()
+    monkeypatch.setattr(opengauss_module.psycopg2.pool, "SimpleConnectionPool", MagicMock(return_value=pool))
+    lock = MagicMock()
+    lock.__enter__.return_value = None
+    lock.__exit__.return_value = None
+    monkeypatch.setattr(opengauss_module.redis_client, "lock", MagicMock(return_value=lock))
+    monkeypatch.setattr(opengauss_module.redis_client, "set", MagicMock())
+
+    vector = opengauss_module.OpenGauss("collection_1", _config(opengauss_module))
+    cursor = MagicMock()
+
+    @contextmanager
+    def _cursor_ctx():
+        yield cursor
+
+    vector._get_cursor = _cursor_ctx
+
+    monkeypatch.setattr(opengauss_module.redis_client, "get", MagicMock(return_value=1))
+    vector._create_collection(1536)
+    cursor.execute.assert_not_called()
+
+    monkeypatch.setattr(opengauss_module.redis_client, "get", MagicMock(return_value=None))
+    vector._create_collection(1536)
+    cursor.execute.assert_called_once()
+    opengauss_module.redis_client.set.assert_called_once()
+
+
+def test_opengauss_factory_uses_existing_or_generated_collection(opengauss_module, monkeypatch):
+    factory = opengauss_module.OpenGaussFactory()
+    dataset_with_index = SimpleNamespace(
+        id="dataset-1",
+        index_struct_dict={"vector_store": {"class_prefix": "EXISTING_COLLECTION"}},
+        index_struct=None,
+    )
+    dataset_without_index = SimpleNamespace(id="dataset-2", index_struct_dict=None, index_struct=None)
+
+    monkeypatch.setattr(opengauss_module.Dataset, "gen_collection_name_by_id", lambda _id: "AUTO_COLLECTION")
+    monkeypatch.setattr(opengauss_module.dify_config, "OPENGAUSS_HOST", "localhost")
+    monkeypatch.setattr(opengauss_module.dify_config, "OPENGAUSS_PORT", 6600)
+    monkeypatch.setattr(opengauss_module.dify_config, "OPENGAUSS_USER", "postgres")
+    monkeypatch.setattr(opengauss_module.dify_config, "OPENGAUSS_PASSWORD", "password")
+    monkeypatch.setattr(opengauss_module.dify_config, "OPENGAUSS_DATABASE", "dify")
+    monkeypatch.setattr(opengauss_module.dify_config, "OPENGAUSS_MIN_CONNECTION", 1)
+    monkeypatch.setattr(opengauss_module.dify_config, "OPENGAUSS_MAX_CONNECTION", 5)
+    monkeypatch.setattr(opengauss_module.dify_config, "OPENGAUSS_ENABLE_PQ", False)
+
+    with patch.object(opengauss_module, "OpenGauss", return_value="vector") as vector_cls:
+        result_1 = factory.init_vector(dataset_with_index, attributes=[], embeddings=MagicMock())
+        result_2 = factory.init_vector(dataset_without_index, attributes=[], embeddings=MagicMock())
+
+    assert result_1 == "vector"
+    assert result_2 == "vector"
+    assert vector_cls.call_args_list[0].kwargs["collection_name"] == "EXISTING_COLLECTION"
+    assert vector_cls.call_args_list[1].kwargs["collection_name"] == "AUTO_COLLECTION"
+    assert dataset_without_index.index_struct is not None

api/tests/unit_tests/core/rag/datasource/vdb/opensearch/test_opensearch_vector.py

@@ -0,0 +1,360 @@
+import importlib
+import sys
+import types
+from types import SimpleNamespace
+from unittest.mock import MagicMock, patch
+
+import pytest
+from pydantic import ValidationError
+
+from core.rag.models.document import Document
+
+
+def _build_fake_opensearch_modules():
+    opensearchpy = types.ModuleType("opensearchpy")
+    opensearchpy_helpers = types.ModuleType("opensearchpy.helpers")
+
+    class BulkIndexError(Exception):
+        def __init__(self, errors):
+            super().__init__("bulk error")
+            self.errors = errors
+
+    class Urllib3AWSV4SignerAuth:
+        def __init__(self, credentials, region, service):
+            self.credentials = credentials
+            self.region = region
+            self.service = service
+
+    class Urllib3HttpConnection:
+        pass
+
+    class _IndicesClient:
+        def __init__(self):
+            self.exists = MagicMock(return_value=False)
+            self.create = MagicMock()
+            self.delete = MagicMock()
+
+    class OpenSearch:
+        def __init__(self, **kwargs):
+            self.kwargs = kwargs
+            self.indices = _IndicesClient()
+            self.search = MagicMock(return_value={"hits": {"hits": []}})
+            self.get = MagicMock()
+
+    helpers = SimpleNamespace(bulk=MagicMock())
+
+    opensearchpy.OpenSearch = OpenSearch
+    opensearchpy.Urllib3AWSV4SignerAuth = Urllib3AWSV4SignerAuth
+    opensearchpy.Urllib3HttpConnection = Urllib3HttpConnection
+    opensearchpy.helpers = helpers
+    opensearchpy_helpers.BulkIndexError = BulkIndexError
+
+    return {
+        "opensearchpy": opensearchpy,
+        "opensearchpy.helpers": opensearchpy_helpers,
+    }
+
+
+@pytest.fixture
+def opensearch_module(monkeypatch):
+    for name, module in _build_fake_opensearch_modules().items():
+        monkeypatch.setitem(sys.modules, name, module)
+
+    import core.rag.datasource.vdb.opensearch.opensearch_vector as module
+
+    return importlib.reload(module)
+
+
+def _config(module, **overrides):
+    values = {
+        "host": "localhost",
+        "port": 9200,
+        "secure": True,
+        "verify_certs": True,
+        "auth_method": "basic",
+        "user": "admin",
+        "password": "secret",
+    }
+    values.update(overrides)
+    return module.OpenSearchConfig.model_validate(values)
+
+
+@pytest.mark.parametrize(
+    ("field", "value", "message"),
+    [
+        ("host", "", "config OPENSEARCH_HOST is required"),
+        ("port", 0, "config OPENSEARCH_PORT is required"),
+    ],
+)
+def test_config_validation_required_fields(opensearch_module, field, value, message):
+    values = _config(opensearch_module).model_dump()
+    values[field] = value
+
+    with pytest.raises(ValidationError, match=message):
+        opensearch_module.OpenSearchConfig.model_validate(values)
+
+
+def test_config_validation_for_aws_auth_and_https_fields(opensearch_module):
+    values = {
+        "host": "localhost",
+        "port": 9200,
+        "secure": True,
+        "verify_certs": True,
+        "auth_method": "aws_managed_iam",
+        "user": "admin",
+        "password": "secret",
+    }
+    with pytest.raises(ValidationError, match="OPENSEARCH_AWS_REGION"):
+        opensearch_module.OpenSearchConfig.model_validate(values)
+
+    values = _config(opensearch_module).model_dump()
+    values["OPENSEARCH_SECURE"] = False
+    values["OPENSEARCH_VERIFY_CERTS"] = True
+    with pytest.raises(ValidationError, match="verify_certs=True requires secure"):
+        opensearch_module.OpenSearchConfig.model_validate(values)
+
+
+def test_create_aws_managed_iam_auth(opensearch_module, monkeypatch):
+    class _Session:
+        def get_credentials(self):
+            return "creds"
+
+    boto3 = types.ModuleType("boto3")
+    boto3.Session = _Session
+    monkeypatch.setitem(sys.modules, "boto3", boto3)
+
+    config = _config(
+        opensearch_module,
+        auth_method="aws_managed_iam",
+        aws_region="us-east-1",
+        aws_service="es",
+    )
+    auth = config.create_aws_managed_iam_auth()
+
+    assert auth.credentials == "creds"
+    assert auth.region == "us-east-1"
+    assert auth.service == "es"
+
+
+def test_to_opensearch_params_supports_basic_and_aws(opensearch_module):
+    basic_params = _config(opensearch_module).to_opensearch_params()
+    assert basic_params["http_auth"] == ("admin", "secret")
+
+    aws_config = _config(
+        opensearch_module,
+        auth_method="aws_managed_iam",
+        aws_region="us-west-2",
+        aws_service="es",
+    )
+    with patch.object(opensearch_module.OpenSearchConfig, "create_aws_managed_iam_auth", return_value="iam-auth"):
+        aws_params = aws_config.to_opensearch_params()
+
+    assert aws_params["http_auth"] == "iam-auth"
+
+
+def test_init_and_create_delegate_calls(opensearch_module):
+    vector = opensearch_module.OpenSearchVector("Collection_1", _config(opensearch_module))
+    vector.create_collection = MagicMock()
+    vector.add_texts = MagicMock()
+    docs = [Document(page_content="hello", metadata={"doc_id": "seg-1"})]
+
+    vector.create(docs, [[0.1, 0.2]])
+
+    assert vector.get_type() == "opensearch"
+    vector.create_collection.assert_called_once_with([[0.1, 0.2]], [{"doc_id": "seg-1"}])
+    vector.add_texts.assert_called_once_with(docs, [[0.1, 0.2]])
+
+
+def test_add_texts_supports_regular_and_aoss_clients(opensearch_module, monkeypatch):
+    vector = opensearch_module.OpenSearchVector("Collection_1", _config(opensearch_module, aws_service="es"))
+    docs = [
+        Document(page_content="a", metadata={"doc_id": "1"}),
+        Document(page_content="b", metadata={"doc_id": "2"}),
+    ]
+
+    monkeypatch.setattr(opensearch_module, "uuid4", lambda: SimpleNamespace(hex="generated-id"))
+    opensearch_module.helpers.bulk.reset_mock()
+    vector.add_texts(docs, [[0.1], [0.2]])
+    actions = opensearch_module.helpers.bulk.call_args.kwargs["actions"]
+    assert len(actions) == 2
+    assert all("_id" in action for action in actions)
+
+    vector._client_config.aws_service = "aoss"
+    opensearch_module.helpers.bulk.reset_mock()
+    vector.add_texts(docs, [[0.3], [0.4]])
+    aoss_actions = opensearch_module.helpers.bulk.call_args.kwargs["actions"]
+    assert all("_id" not in action for action in aoss_actions)
+
+
+def test_metadata_lookup_and_delete_by_metadata_field(opensearch_module):
+    vector = opensearch_module.OpenSearchVector("collection_1", _config(opensearch_module))
+    vector._client.search.return_value = {"hits": {"hits": [{"_id": "id-1"}, {"_id": "id-2"}]}}
+
+    assert vector.get_ids_by_metadata_field("document_id", "doc-1") == ["id-1", "id-2"]
+
+    vector._client.search.return_value = {"hits": {"hits": []}}
+    assert vector.get_ids_by_metadata_field("document_id", "doc-1") is None
+
+    vector.get_ids_by_metadata_field = MagicMock(return_value=["id-1"])
+    vector.delete_by_ids = MagicMock()
+    vector.delete_by_metadata_field("document_id", "doc-1")
+    vector.delete_by_ids.assert_called_once_with(["id-1"])
+
+
+def test_delete_by_ids_branches_and_bulk_error_handling(opensearch_module):
+    vector = opensearch_module.OpenSearchVector("collection_1", _config(opensearch_module))
+    opensearch_module.helpers.bulk.reset_mock()
+    vector._client.indices.exists.return_value = False
+    vector.delete_by_ids(["doc-1"])
+    opensearch_module.helpers.bulk.assert_not_called()
+
+    vector._client.indices.exists.return_value = True
+    vector.get_ids_by_metadata_field = MagicMock(side_effect=[["es-1"], None])
+    vector.delete_by_ids(["doc-1", "doc-2"])
+    opensearch_module.helpers.bulk.assert_called_once()
+
+    opensearch_module.helpers.bulk.reset_mock()
+    vector.get_ids_by_metadata_field = MagicMock(return_value=["es-404"])
+    opensearch_module.helpers.bulk.side_effect = opensearch_module.BulkIndexError(
+        [{"delete": {"status": 404, "_id": "es-404"}}]
+    )
+    vector.delete_by_ids(["doc-404"])
+    assert opensearch_module.helpers.bulk.call_count == 1
+
+    opensearch_module.helpers.bulk.side_effect = None
+
+
+def test_delete_and_text_exists(opensearch_module):
+    vector = opensearch_module.OpenSearchVector("collection_1", _config(opensearch_module))
+    vector.delete()
+    vector._client.indices.delete.assert_called_once_with(index="collection_1", ignore_unavailable=True)
+
+    vector._client.get.return_value = {"_id": "id-1"}
+    assert vector.text_exists("id-1") is True
+    vector._client.get.side_effect = RuntimeError("not found")
+    assert vector.text_exists("id-1") is False
+
+
+def test_search_by_vector_validates_and_builds_documents(opensearch_module):
+    vector = opensearch_module.OpenSearchVector("collection_1", _config(opensearch_module))
+
+    with pytest.raises(ValueError, match="query_vector should be a list"):
+        vector.search_by_vector("not-a-list")
+
+    with pytest.raises(ValueError, match="should be floats"):
+        vector.search_by_vector([0.1, 1])
+
+    vector._client.search.return_value = {
+        "hits": {
+            "hits": [
+                {
+                    "_source": {
+                        opensearch_module.Field.CONTENT_KEY: "doc-1",
+                        opensearch_module.Field.METADATA_KEY: None,
+                    },
+                    "_score": 0.9,
+                },
+                {
+                    "_source": {
+                        opensearch_module.Field.CONTENT_KEY: "doc-2",
+                        opensearch_module.Field.METADATA_KEY: {"doc_id": "2"},
+                    },
+                    "_score": 0.1,
+                },
+            ]
+        }
+    }
+    docs = vector.search_by_vector([0.1, 0.2], top_k=2, score_threshold=0.5)
+    assert len(docs) == 1
+    assert docs[0].page_content == "doc-1"
+    assert docs[0].metadata["score"] == pytest.approx(0.9)
+
+    vector.search_by_vector([0.1, 0.2], top_k=3, document_ids_filter=["doc-a", "doc-b"])
+    query = vector._client.search.call_args.kwargs["body"]
+    assert "script_score" in query["query"]
+
+
+def test_search_by_vector_reraises_client_error(opensearch_module):
+    vector = opensearch_module.OpenSearchVector("collection_1", _config(opensearch_module))
+    vector._client.search.side_effect = RuntimeError("boom")
+
+    with pytest.raises(RuntimeError, match="boom"):
+        vector.search_by_vector([0.1, 0.2])
+
+
+def test_search_by_full_text_and_filters(opensearch_module):
+    vector = opensearch_module.OpenSearchVector("collection_1", _config(opensearch_module))
+    vector._client.search.return_value = {
+        "hits": {
+            "hits": [
+                {
+                    "_source": {
+                        opensearch_module.Field.METADATA_KEY: {"doc_id": "1"},
+                        opensearch_module.Field.VECTOR: [0.1],
+                        opensearch_module.Field.CONTENT_KEY: "matched text",
+                    }
+                },
+            ]
+        }
+    }
+
+    docs = vector.search_by_full_text("hello", document_ids_filter=["d-1"])
+
+    assert len(docs) == 1
+    assert docs[0].page_content == "matched text"
+    query = vector._client.search.call_args.kwargs["body"]
+    assert query["query"]["bool"]["filter"] == [{"terms": {"metadata.document_id": ["d-1"]}}]
+
+
+def test_create_collection_cache_and_create_path(opensearch_module, monkeypatch):
+    lock = MagicMock()
+    lock.__enter__.return_value = None
+    lock.__exit__.return_value = None
+    monkeypatch.setattr(opensearch_module.redis_client, "lock", MagicMock(return_value=lock))
+    monkeypatch.setattr(opensearch_module.redis_client, "set", MagicMock())
+
+    vector = opensearch_module.OpenSearchVector("Collection_1", _config(opensearch_module))
+
+    monkeypatch.setattr(opensearch_module.redis_client, "get", MagicMock(return_value=1))
+    vector._client.indices.create.reset_mock()
+    vector.create_collection([[0.1, 0.2]])
+    vector._client.indices.create.assert_not_called()
+
+    monkeypatch.setattr(opensearch_module.redis_client, "get", MagicMock(return_value=None))
+    vector._client.indices.exists.return_value = False
+    vector.create_collection([[0.1, 0.2]])
+    vector._client.indices.create.assert_called_once()
+    index_body = vector._client.indices.create.call_args.kwargs["body"]
+    assert index_body["mappings"]["properties"]["vector"]["dimension"] == 2
+    opensearch_module.redis_client.set.assert_called()
+
+
+def test_opensearch_factory_initializes_expected_collection_name(opensearch_module, monkeypatch):
+    factory = opensearch_module.OpenSearchVectorFactory()
+    dataset_with_index = SimpleNamespace(
+        id="dataset-1",
+        index_struct_dict={"vector_store": {"class_prefix": "EXISTING_COLLECTION"}},
+        index_struct=None,
+    )
+    dataset_without_index = SimpleNamespace(id="dataset-2", index_struct_dict=None, index_struct=None)
+
+    monkeypatch.setattr(opensearch_module.Dataset, "gen_collection_name_by_id", lambda _id: "AUTO_COLLECTION")
+    monkeypatch.setattr(opensearch_module.dify_config, "OPENSEARCH_HOST", "localhost")
+    monkeypatch.setattr(opensearch_module.dify_config, "OPENSEARCH_PORT", 9200)
+    monkeypatch.setattr(opensearch_module.dify_config, "OPENSEARCH_SECURE", True)
+    monkeypatch.setattr(opensearch_module.dify_config, "OPENSEARCH_VERIFY_CERTS", True)
+    monkeypatch.setattr(opensearch_module.dify_config, "OPENSEARCH_AUTH_METHOD", "basic")
+    monkeypatch.setattr(opensearch_module.dify_config, "OPENSEARCH_USER", "admin")
+    monkeypatch.setattr(opensearch_module.dify_config, "OPENSEARCH_PASSWORD", "secret")
+    monkeypatch.setattr(opensearch_module.dify_config, "OPENSEARCH_AWS_REGION", None)
+    monkeypatch.setattr(opensearch_module.dify_config, "OPENSEARCH_AWS_SERVICE", None)
+
+    with patch.object(opensearch_module, "OpenSearchVector", return_value="vector") as vector_cls:
+        result_1 = factory.init_vector(dataset_with_index, attributes=[], embeddings=MagicMock())
+        result_2 = factory.init_vector(dataset_without_index, attributes=[], embeddings=MagicMock())
+
+    assert result_1 == "vector"
+    assert result_2 == "vector"
+    assert vector_cls.call_args_list[0].kwargs["collection_name"] == "existing_collection"
+    assert vector_cls.call_args_list[1].kwargs["collection_name"] == "auto_collection"
+    assert dataset_without_index.index_struct is not None

api/tests/unit_tests/core/rag/datasource/vdb/oracle/test_oraclevector.py

@@ -0,0 +1,375 @@
+import array
+import importlib
+import sys
+import types
+from types import SimpleNamespace
+from unittest.mock import MagicMock, patch
+
+import numpy
+import pytest
+from pydantic import ValidationError
+
+from core.rag.models.document import Document
+
+
+def _build_fake_oracle_modules():
+    jieba = types.ModuleType("jieba")
+    jieba_posseg = types.ModuleType("jieba.posseg")
+    jieba_posseg.cut = MagicMock(return_value=[])
+    jieba.posseg = jieba_posseg
+
+    oracledb = types.ModuleType("oracledb")
+    oracledb_connection = types.ModuleType("oracledb.connection")
+
+    class Connection:
+        pass
+
+    oracledb_connection.Connection = Connection
+    oracledb.defaults = SimpleNamespace(fetch_lobs=True)
+    oracledb.DB_TYPE_VECTOR = object()
+    oracledb.create_pool = MagicMock(return_value=MagicMock(release=MagicMock()))
+    oracledb.connect = MagicMock()
+
+    return {
+        "jieba": jieba,
+        "jieba.posseg": jieba_posseg,
+        "oracledb": oracledb,
+        "oracledb.connection": oracledb_connection,
+    }
+
+
+def _connection_with_cursor(cursor):
+    cursor_ctx = MagicMock()
+    cursor_ctx.__enter__.return_value = cursor
+    cursor_ctx.__exit__.return_value = None
+
+    connection = MagicMock()
+    connection.__enter__.return_value = connection
+    connection.__exit__.return_value = None
+    connection.cursor.return_value = cursor_ctx
+    return connection
+
+
+@pytest.fixture
+def oracle_module(monkeypatch):
+    for name, module in _build_fake_oracle_modules().items():
+        monkeypatch.setitem(sys.modules, name, module)
+
+    import core.rag.datasource.vdb.oracle.oraclevector as module
+
+    return importlib.reload(module)
+
+
+def _config(module, **overrides):
+    values = {
+        "user": "system",
+        "password": "oracle",
+        "dsn": "oracle:1521/freepdb1",
+        "is_autonomous": False,
+    }
+    values.update(overrides)
+    return module.OracleVectorConfig.model_validate(values)
+
+
+@pytest.mark.parametrize(
+    ("field", "value", "message"),
+    [
+        ("user", "", "config ORACLE_USER is required"),
+        ("password", "", "config ORACLE_PASSWORD is required"),
+        ("dsn", "", "config ORACLE_DSN is required"),
+    ],
+)
+def test_oracle_config_validation_required_fields(oracle_module, field, value, message):
+    values = _config(oracle_module).model_dump()
+    values[field] = value
+
+    with pytest.raises(ValidationError, match=message):
+        oracle_module.OracleVectorConfig.model_validate(values)
+
+
+def test_oracle_config_validation_autonomous_requirements(oracle_module):
+    with pytest.raises(ValidationError, match="config_dir is required"):
+        oracle_module.OracleVectorConfig.model_validate(
+            {"user": "u", "password": "p", "dsn": "d", "is_autonomous": True}
+        )
+
+
+def test_init_and_get_type(oracle_module, monkeypatch):
+    pool = MagicMock()
+    monkeypatch.setattr(oracle_module.oracledb, "create_pool", MagicMock(return_value=pool))
+    vector = oracle_module.OracleVector("collection_1", _config(oracle_module))
+
+    assert vector.get_type() == "oracle"
+    assert vector.table_name == "embedding_collection_1"
+    assert vector.pool is pool
+
+
+def test_numpy_converters_and_type_handlers(oracle_module):
+    vector = oracle_module.OracleVector.__new__(oracle_module.OracleVector)
+
+    in_float64 = vector.numpy_converter_in(numpy.array([0.1], dtype=numpy.float64))
+    in_float32 = vector.numpy_converter_in(numpy.array([0.1], dtype=numpy.float32))
+    in_int8 = vector.numpy_converter_in(numpy.array([1], dtype=numpy.int8))
+    assert in_float64.typecode == "d"
+    assert in_float32.typecode == "f"
+    assert in_int8.typecode == "b"
+
+    cursor = MagicMock()
+    vector.input_type_handler(cursor, numpy.array([0.1], dtype=numpy.float32), 2)
+    cursor.var.assert_called_with(
+        oracle_module.oracledb.DB_TYPE_VECTOR,
+        arraysize=2,
+        inconverter=vector.numpy_converter_in,
+    )
+
+    metadata = SimpleNamespace(type_code=oracle_module.oracledb.DB_TYPE_VECTOR)
+    cursor.arraysize = 3
+    vector.output_type_handler(cursor, metadata)
+    cursor.var.assert_called_with(
+        metadata.type_code,
+        arraysize=3,
+        outconverter=vector.numpy_converter_out,
+    )
+
+    out_int8 = vector.numpy_converter_out(array.array("b", [1]))
+    assert out_int8.dtype == numpy.int8
+    out_float32 = vector.numpy_converter_out(array.array("f", [1.0]))
+    assert out_float32.dtype == numpy.float32
+    out_float64 = vector.numpy_converter_out(array.array("d", [1.0]))
+    assert out_float64.dtype == numpy.float64
+
+
+def test_get_connection_supports_standard_and_autonomous_paths(oracle_module, monkeypatch):
+    connect = MagicMock(return_value="connection")
+    monkeypatch.setattr(oracle_module.oracledb, "connect", connect)
+
+    vector = oracle_module.OracleVector.__new__(oracle_module.OracleVector)
+    vector.config = _config(oracle_module)
+    assert vector._get_connection() == "connection"
+    connect.assert_called_with(user="system", password="oracle", dsn="oracle:1521/freepdb1")
+
+    vector.config = _config(
+        oracle_module,
+        is_autonomous=True,
+        config_dir="/wallet",
+        wallet_location="/wallet",
+        wallet_password="pw",
+    )
+    vector._get_connection()
+    assert connect.call_args.kwargs["config_dir"] == "/wallet"
+    assert connect.call_args.kwargs["wallet_location"] == "/wallet"
+
+
+def test_create_delegates_collection_and_insert(oracle_module):
+    vector = oracle_module.OracleVector.__new__(oracle_module.OracleVector)
+    vector._create_collection = MagicMock()
+    vector.add_texts = MagicMock(return_value=["seg-1"])
+    docs = [Document(page_content="doc", metadata={"doc_id": "seg-1"})]
+
+    result = vector.create(docs, [[0.1, 0.2]])
+
+    assert result == ["seg-1"]
+    vector._create_collection.assert_called_once_with(2)
+    vector.add_texts.assert_called_once_with(docs, [[0.1, 0.2]])
+
+
+def test_add_texts_inserts_and_logs_on_failures(oracle_module, monkeypatch):
+    vector = oracle_module.OracleVector.__new__(oracle_module.OracleVector)
+    vector.table_name = "embedding_collection_1"
+    vector.input_type_handler = MagicMock()
+    vector.output_type_handler = MagicMock()
+
+    cursor = MagicMock()
+    cursor.execute.side_effect = [None, RuntimeError("insert failed")]
+    connection = _connection_with_cursor(cursor)
+    vector._get_connection = MagicMock(return_value=connection)
+
+    monkeypatch.setattr(oracle_module.uuid, "uuid4", lambda: "generated-uuid")
+    docs = [
+        Document(page_content="a", metadata={"doc_id": "doc-a"}),
+        Document(page_content="b", metadata={"document_id": "doc-b"}),
+        SimpleNamespace(page_content="c", metadata=None),
+    ]
+
+    ids = vector.add_texts(docs, [[0.1], [0.2], [0.3]])
+
+    assert ids == ["doc-a", "generated-uuid"]
+    assert cursor.execute.call_count == 2
+    assert connection.commit.call_count >= 1
+    connection.close.assert_called()
+
+
+def test_text_exists_and_get_by_ids(oracle_module):
+    vector = oracle_module.OracleVector.__new__(oracle_module.OracleVector)
+    vector.table_name = "embedding_collection_1"
+    vector.pool = MagicMock()
+
+    cursor = MagicMock()
+    cursor.fetchone.return_value = ("id-1",)
+    cursor.__iter__.return_value = iter([({"doc_id": "1"}, "text-1"), ({"doc_id": "2"}, "text-2")])
+    vector._get_connection = MagicMock(return_value=_connection_with_cursor(cursor))
+
+    assert vector.text_exists("id-1") is True
+    docs = vector.get_by_ids(["id-1", "id-2"])
+    assert len(docs) == 2
+    assert docs[0].page_content == "text-1"
+    vector.pool.release.assert_called_once()
+    assert vector.get_by_ids([]) == []
+
+
+def test_delete_methods(oracle_module):
+    vector = oracle_module.OracleVector.__new__(oracle_module.OracleVector)
+    vector.table_name = "embedding_collection_1"
+
+    cursor = MagicMock()
+    vector._get_connection = MagicMock(return_value=_connection_with_cursor(cursor))
+
+    vector.delete_by_ids([])
+    vector._get_connection.assert_not_called()
+
+    vector.delete_by_ids(["id-1", "id-2"])
+    vector.delete_by_metadata_field("document_id", "doc-1")
+    vector.delete()
+
+    executed_sql = [call.args[0] for call in cursor.execute.call_args_list]
+    assert any("DELETE FROM embedding_collection_1 WHERE id IN" in sql for sql in executed_sql)
+    assert any("JSON_VALUE(meta" in sql for sql in executed_sql)
+    assert any("DROP TABLE IF EXISTS embedding_collection_1" in sql for sql in executed_sql)
+
+
+def test_search_by_vector_with_threshold_and_filter(oracle_module):
+    vector = oracle_module.OracleVector.__new__(oracle_module.OracleVector)
+    vector.table_name = "embedding_collection_1"
+    vector.input_type_handler = MagicMock()
+    vector.output_type_handler = MagicMock()
+
+    cursor = MagicMock()
+    cursor.__iter__.return_value = iter([({"doc_id": "1"}, "doc-1", 0.1), ({"doc_id": "2"}, "doc-2", 0.8)])
+    connection = _connection_with_cursor(cursor)
+    vector._get_connection = MagicMock(return_value=connection)
+
+    docs = vector.search_by_vector(
+        [0.1, 0.2],
+        top_k=0,
+        score_threshold=0.5,
+        document_ids_filter=["d-1", "d-2"],
+    )
+
+    assert len(docs) == 1
+    assert docs[0].metadata["score"] == pytest.approx(0.9)
+    sql = cursor.execute.call_args.args[0]
+    assert "fetch first 4 rows only" in sql
+    assert "JSON_VALUE(meta, '$.document_id') IN (:2, :3)" in sql
+
+
+def _fake_nltk_module(*, missing_data=False):
+    nltk = types.ModuleType("nltk")
+    nltk_corpus = types.ModuleType("nltk.corpus")
+
+    class _Data:
+        @staticmethod
+        def find(_path):
+            if missing_data:
+                raise LookupError("missing")
+            return True
+
+    nltk.data = _Data()
+    nltk.word_tokenize = lambda text: text.split()
+    nltk_corpus.stopwords = SimpleNamespace(words=lambda _lang: ["and", "the"])
+    return nltk, nltk_corpus
+
+
+def test_search_by_full_text_chinese_and_english_paths(oracle_module, monkeypatch):
+    vector = oracle_module.OracleVector.__new__(oracle_module.OracleVector)
+    vector.table_name = "embedding_collection_1"
+
+    cursor = MagicMock()
+    cursor.__iter__.return_value = iter([({"doc_id": "1"}, "text-1", [0.1, 0.2])])
+    vector._get_connection = MagicMock(return_value=_connection_with_cursor(cursor))
+
+    monkeypatch.setattr(oracle_module.pseg, "cut", MagicMock(return_value=[("张", "nr"), ("三", "nr"), ("。", "x")]))
+    zh_docs = vector.search_by_full_text("张三", top_k=2)
+    assert len(zh_docs) == 1
+    zh_params = cursor.execute.call_args.args[1]
+    assert zh_params["kk"] == "张三"
+
+    nltk, nltk_corpus = _fake_nltk_module(missing_data=False)
+    monkeypatch.setitem(sys.modules, "nltk", nltk)
+    monkeypatch.setitem(sys.modules, "nltk.corpus", nltk_corpus)
+    cursor.__iter__.return_value = iter([({"doc_id": "2"}, "text-2", [0.3, 0.4])])
+    en_docs = vector.search_by_full_text("alice and bob", top_k=-1, document_ids_filter=["d-1"])
+    assert len(en_docs) == 1
+    en_sql = cursor.execute.call_args.args[0]
+    en_params = cursor.execute.call_args.args[1]
+    assert "fetch first 5 rows only" in en_sql
+    assert "doc_id_0" in en_params
+
+
+def test_search_by_full_text_empty_query_and_missing_nltk(oracle_module, monkeypatch):
+    vector = oracle_module.OracleVector.__new__(oracle_module.OracleVector)
+    vector.table_name = "embedding_collection_1"
+    vector._get_connection = MagicMock()
+
+    empty_result = vector.search_by_full_text("")
+    assert empty_result[0].page_content == ""
+
+    nltk, nltk_corpus = _fake_nltk_module(missing_data=True)
+    monkeypatch.setitem(sys.modules, "nltk", nltk)
+    monkeypatch.setitem(sys.modules, "nltk.corpus", nltk_corpus)
+    with pytest.raises(LookupError, match="required NLTK data package"):
+        vector.search_by_full_text("english query")
+
+
+def test_create_collection_cache_and_execute_path(oracle_module, monkeypatch):
+    lock = MagicMock()
+    lock.__enter__.return_value = None
+    lock.__exit__.return_value = None
+    monkeypatch.setattr(oracle_module.redis_client, "lock", MagicMock(return_value=lock))
+    monkeypatch.setattr(oracle_module.redis_client, "set", MagicMock())
+
+    vector = oracle_module.OracleVector.__new__(oracle_module.OracleVector)
+    vector._collection_name = "collection_1"
+    vector.table_name = "embedding_collection_1"
+
+    cursor = MagicMock()
+    vector._get_connection = MagicMock(return_value=_connection_with_cursor(cursor))
+
+    monkeypatch.setattr(oracle_module.redis_client, "get", MagicMock(return_value=1))
+    vector._create_collection(2)
+    cursor.execute.assert_not_called()
+
+    monkeypatch.setattr(oracle_module.redis_client, "get", MagicMock(return_value=None))
+    vector._create_collection(2)
+    executed_sql = [call.args[0] for call in cursor.execute.call_args_list]
+    assert any("CREATE TABLE IF NOT EXISTS embedding_collection_1" in sql for sql in executed_sql)
+    assert any("CREATE INDEX IF NOT EXISTS idx_docs_embedding_collection_1" in sql for sql in executed_sql)
+    oracle_module.redis_client.set.assert_called_once()
+
+
+def test_oracle_factory_init_vector_uses_existing_or_generated_collection(oracle_module, monkeypatch):
+    factory = oracle_module.OracleVectorFactory()
+    dataset_with_index = SimpleNamespace(
+        id="dataset-1",
+        index_struct_dict={"vector_store": {"class_prefix": "EXISTING_COLLECTION"}},
+        index_struct=None,
+    )
+    dataset_without_index = SimpleNamespace(id="dataset-2", index_struct_dict=None, index_struct=None)
+
+    monkeypatch.setattr(oracle_module.Dataset, "gen_collection_name_by_id", lambda _id: "AUTO_COLLECTION")
+    monkeypatch.setattr(oracle_module.dify_config, "ORACLE_USER", "system")
+    monkeypatch.setattr(oracle_module.dify_config, "ORACLE_PASSWORD", "oracle")
+    monkeypatch.setattr(oracle_module.dify_config, "ORACLE_DSN", "oracle:1521/freepdb1")
+    monkeypatch.setattr(oracle_module.dify_config, "ORACLE_CONFIG_DIR", None)
+    monkeypatch.setattr(oracle_module.dify_config, "ORACLE_WALLET_LOCATION", None)
+    monkeypatch.setattr(oracle_module.dify_config, "ORACLE_WALLET_PASSWORD", None)
+    monkeypatch.setattr(oracle_module.dify_config, "ORACLE_IS_AUTONOMOUS", False)
+
+    with patch.object(oracle_module, "OracleVector", return_value="vector") as vector_cls:
+        result_1 = factory.init_vector(dataset_with_index, attributes=[], embeddings=MagicMock())
+        result_2 = factory.init_vector(dataset_without_index, attributes=[], embeddings=MagicMock())
+
+    assert result_1 == "vector"
+    assert result_2 == "vector"
+    assert vector_cls.call_args_list[0].kwargs["collection_name"] == "EXISTING_COLLECTION"
+    assert vector_cls.call_args_list[1].kwargs["collection_name"] == "AUTO_COLLECTION"
+    assert dataset_without_index.index_struct is not None

api/tests/unit_tests/core/rag/datasource/vdb/pgvecto_rs/test_pgvecto_rs.py

@@ -0,0 +1,317 @@
+import importlib
+import sys
+import types
+from types import SimpleNamespace
+from unittest.mock import MagicMock, patch
+
+import pytest
+from pydantic import ValidationError
+from sqlalchemy.types import UserDefinedType
+
+from core.rag.models.document import Document
+
+
+def _build_fake_pgvecto_modules():
+    pgvecto_rs = types.ModuleType("pgvecto_rs")
+    pgvecto_rs_sqlalchemy = types.ModuleType("pgvecto_rs.sqlalchemy")
+
+    class VECTOR(UserDefinedType):
+        def __init__(self, dim):
+            self.dim = dim
+
+    pgvecto_rs_sqlalchemy.VECTOR = VECTOR
+    return {
+        "pgvecto_rs": pgvecto_rs,
+        "pgvecto_rs.sqlalchemy": pgvecto_rs_sqlalchemy,
+    }
+
+
+class _FakeSessionContext:
+    def __init__(self, calls, execute_results=None):
+        self.calls = calls
+        self.execute_results = execute_results or []
+        self.execute = MagicMock(side_effect=self._execute_side_effect)
+        self.commit = MagicMock()
+
+    def _execute_side_effect(self, *args, **kwargs):
+        self.calls.append((args, kwargs))
+        if self.execute_results:
+            return self.execute_results.pop(0)
+        return MagicMock()
+
+    def __enter__(self):
+        return self
+
+    def __exit__(self, exc_type, exc, tb):
+        return None
+
+
+def _session_factory(calls, execute_results=None):
+    def _session(_client):
+        return _FakeSessionContext(calls=calls, execute_results=execute_results)
+
+    return _session
+
+
+@pytest.fixture
+def pgvecto_module(monkeypatch):
+    for name, module in _build_fake_pgvecto_modules().items():
+        monkeypatch.setitem(sys.modules, name, module)
+
+    import core.rag.datasource.vdb.pgvecto_rs.collection as collection_module
+    import core.rag.datasource.vdb.pgvecto_rs.pgvecto_rs as module
+
+    return importlib.reload(module), importlib.reload(collection_module)
+
+
+def _config(module, **overrides):
+    values = {
+        "host": "localhost",
+        "port": 5432,
+        "user": "postgres",
+        "password": "secret",
+        "database": "postgres",
+    }
+    values.update(overrides)
+    return module.PgvectoRSConfig.model_validate(values)
+
+
+@pytest.mark.parametrize(
+    ("field", "value", "message"),
+    [
+        ("host", "", "config PGVECTO_RS_HOST is required"),
+        ("port", 0, "config PGVECTO_RS_PORT is required"),
+        ("user", "", "config PGVECTO_RS_USER is required"),
+        ("password", "", "config PGVECTO_RS_PASSWORD is required"),
+        ("database", "", "config PGVECTO_RS_DATABASE is required"),
+    ],
+)
+def test_pgvecto_config_validation(pgvecto_module, field, value, message):
+    module, _ = pgvecto_module
+    values = _config(module).model_dump()
+    values[field] = value
+
+    with pytest.raises(ValidationError, match=message):
+        module.PgvectoRSConfig.model_validate(values)
+
+
+def test_collection_base_has_expected_annotations(pgvecto_module):
+    _, collection_module = pgvecto_module
+    annotations = collection_module.CollectionORM.__annotations__
+    assert {"id", "text", "meta", "vector"} <= set(annotations)
+
+
+def test_init_get_type_and_create_delegate(pgvecto_module, monkeypatch):
+    module, _ = pgvecto_module
+    session_calls = []
+    monkeypatch.setattr(module, "create_engine", MagicMock(return_value="engine"))
+    monkeypatch.setattr(module, "Session", _session_factory(session_calls))
+
+    vector = module.PGVectoRS("collection_1", _config(module), dim=3)
+    vector.create_collection = MagicMock()
+    vector.add_texts = MagicMock()
+    docs = [Document(page_content="hello", metadata={"doc_id": "1"})]
+    vector.create(docs, [[0.1, 0.2]])
+
+    assert vector.get_type() == module.VectorType.PGVECTO_RS
+    module.create_engine.assert_called_once_with("postgresql+psycopg2://postgres:secret@localhost:5432/postgres")
+    assert any("CREATE EXTENSION IF NOT EXISTS vectors" in str(args[0]) for args, _ in session_calls)
+    vector.create_collection.assert_called_once_with(2)
+    vector.add_texts.assert_called_once_with(docs, [[0.1, 0.2]])
+
+
+def test_create_collection_cache_and_sql_execution(pgvecto_module, monkeypatch):
+    module, _ = pgvecto_module
+    session_calls = []
+    monkeypatch.setattr(module, "create_engine", MagicMock(return_value="engine"))
+    monkeypatch.setattr(module, "Session", _session_factory(session_calls))
+
+    lock = MagicMock()
+    lock.__enter__.return_value = None
+    lock.__exit__.return_value = None
+    monkeypatch.setattr(module.redis_client, "lock", MagicMock(return_value=lock))
+    monkeypatch.setattr(module.redis_client, "set", MagicMock())
+
+    vector = module.PGVectoRS("collection_1", _config(module), dim=3)
+    monkeypatch.setattr(module.redis_client, "get", MagicMock(return_value=1))
+    vector.create_collection(3)
+    assert not any("CREATE TABLE IF NOT EXISTS collection_1" in str(args[0]) for args, _ in session_calls)
+
+    monkeypatch.setattr(module.redis_client, "get", MagicMock(return_value=None))
+    vector.create_collection(3)
+    assert any("CREATE TABLE IF NOT EXISTS collection_1" in str(args[0]) for args, _ in session_calls)
+    assert any("CREATE INDEX IF NOT EXISTS collection_1_embedding_index" in str(args[0]) for args, _ in session_calls)
+    module.redis_client.set.assert_called()
+
+
+def test_add_texts_get_ids_and_delete_methods(pgvecto_module, monkeypatch):
+    module, _ = pgvecto_module
+    init_calls = []
+    runtime_calls = []
+    execute_results = [SimpleNamespace(fetchall=lambda: [("id-1",), ("id-2",)]), SimpleNamespace(fetchall=lambda: [])]
+
+    monkeypatch.setattr(module, "create_engine", MagicMock(return_value="engine"))
+    monkeypatch.setattr(module, "Session", _session_factory(init_calls))
+    vector = module.PGVectoRS("collection_1", _config(module), dim=3)
+
+    monkeypatch.setattr(module, "Session", _session_factory(runtime_calls, execute_results=list(execute_results)))
+
+    class _InsertBuilder:
+        def __init__(self, table):
+            self.table = table
+
+        def values(self, **kwargs):
+            return ("insert", kwargs)
+
+    monkeypatch.setattr(module, "insert", lambda table: _InsertBuilder(table))
+    monkeypatch.setattr(module, "uuid4", MagicMock(side_effect=["uuid-1", "uuid-2"]))
+    docs = [
+        Document(page_content="a", metadata={"doc_id": "1"}),
+        Document(page_content="b", metadata={"doc_id": "2"}),
+    ]
+
+    ids = vector.add_texts(docs, [[0.1], [0.2]])
+    assert ids == ["uuid-1", "uuid-2"]
+    assert any(call[0][0][0] == "insert" for call in runtime_calls if call[0])
+
+    monkeypatch.setattr(
+        module,
+        "Session",
+        _session_factory(runtime_calls, execute_results=[SimpleNamespace(fetchall=lambda: [("id-1",), ("id-2",)])]),
+    )
+    assert vector.get_ids_by_metadata_field("document_id", "doc-1") == ["id-1", "id-2"]
+
+    monkeypatch.setattr(
+        module,
+        "Session",
+        _session_factory(runtime_calls, execute_results=[SimpleNamespace(fetchall=lambda: [])]),
+    )
+    assert vector.get_ids_by_metadata_field("document_id", "doc-1") is None
+
+    vector.get_ids_by_metadata_field = MagicMock(return_value=["id-1"])
+    vector.delete_by_metadata_field("document_id", "doc-1")
+    assert any("DELETE FROM collection_1 WHERE id = ANY(:ids)" in str(args[0]) for args, _ in runtime_calls)
+
+    runtime_calls.clear()
+    monkeypatch.setattr(
+        module,
+        "Session",
+        _session_factory(
+            runtime_calls,
+            execute_results=[
+                SimpleNamespace(fetchall=lambda: [("row-id-1",)]),
+                MagicMock(),
+            ],
+        ),
+    )
+    vector.delete_by_ids(["doc-1"])
+    assert any("meta->>'doc_id' = ANY (:doc_ids)" in str(args[0]) for args, _ in runtime_calls)
+    assert any("DELETE FROM collection_1 WHERE id = ANY(:ids)" in str(args[0]) for args, _ in runtime_calls)
+
+    runtime_calls.clear()
+    monkeypatch.setattr(module, "Session", _session_factory(runtime_calls, execute_results=[MagicMock()]))
+    vector.delete()
+    assert any("DROP TABLE IF EXISTS collection_1" in str(args[0]) for args, _ in runtime_calls)
+
+
+def test_text_exists_search_and_full_text(pgvecto_module, monkeypatch):
+    module, _ = pgvecto_module
+    init_calls = []
+    monkeypatch.setattr(module, "create_engine", MagicMock(return_value="engine"))
+    monkeypatch.setattr(module, "Session", _session_factory(init_calls))
+    vector = module.PGVectoRS("collection_1", _config(module), dim=3)
+
+    runtime_calls = []
+    monkeypatch.setattr(
+        module,
+        "Session",
+        _session_factory(
+            runtime_calls,
+            execute_results=[
+                SimpleNamespace(fetchall=lambda: [("id-1",)]),
+                SimpleNamespace(fetchall=lambda: []),
+            ],
+        ),
+    )
+    assert vector.text_exists("doc-1") is True
+    assert vector.text_exists("doc-1") is False
+
+    class _DistanceExpr:
+        def label(self, _name):
+            return self
+
+    class _VectorColumn:
+        def op(self, _operator, return_type=None):
+            def _call(_query_vector):
+                return _DistanceExpr()
+
+            return _call
+
+    class _MetaFilter:
+        def in_(self, values):
+            return ("in", values)
+
+    class _MetaColumn:
+        def __getitem__(self, _item):
+            return _MetaFilter()
+
+    class _Stmt:
+        def __init__(self):
+            self.where_called = False
+
+        def limit(self, _value):
+            return self
+
+        def order_by(self, _value):
+            return self
+
+        def where(self, _value):
+            self.where_called = True
+            return self
+
+    stmt = _Stmt()
+    monkeypatch.setattr(module, "select", lambda *_args: stmt)
+
+    vector._table = SimpleNamespace(vector=_VectorColumn(), meta=_MetaColumn())
+    rows = [
+        (SimpleNamespace(meta={"doc_id": "1"}, text="text-1"), 0.1),
+        (SimpleNamespace(meta={"doc_id": "2"}, text="text-2"), 0.8),
+    ]
+    monkeypatch.setattr(module, "Session", _session_factory(runtime_calls, execute_results=[rows]))
+
+    docs = vector.search_by_vector([0.1, 0.2], top_k=2, score_threshold=0.5, document_ids_filter=["d-1"])
+    assert len(docs) == 1
+    assert docs[0].metadata["score"] == pytest.approx(0.9)
+    assert stmt.where_called is True
+    assert vector.search_by_full_text("hello") == []
+
+
+def test_factory_uses_existing_or_generated_collection(pgvecto_module, monkeypatch):
+    module, _ = pgvecto_module
+    factory = module.PGVectoRSFactory()
+    dataset_with_index = SimpleNamespace(
+        id="dataset-1",
+        index_struct_dict={"vector_store": {"class_prefix": "EXISTING_COLLECTION"}},
+        index_struct=None,
+    )
+    dataset_without_index = SimpleNamespace(id="dataset-2", index_struct_dict=None, index_struct=None)
+
+    monkeypatch.setattr(module.Dataset, "gen_collection_name_by_id", lambda _id: "AUTO_COLLECTION")
+    monkeypatch.setattr(module.dify_config, "PGVECTO_RS_HOST", "localhost")
+    monkeypatch.setattr(module.dify_config, "PGVECTO_RS_PORT", 5432)
+    monkeypatch.setattr(module.dify_config, "PGVECTO_RS_USER", "postgres")
+    monkeypatch.setattr(module.dify_config, "PGVECTO_RS_PASSWORD", "secret")
+    monkeypatch.setattr(module.dify_config, "PGVECTO_RS_DATABASE", "postgres")
+
+    embeddings = MagicMock()
+    embeddings.embed_query.return_value = [0.1, 0.2, 0.3]
+
+    with patch.object(module, "PGVectoRS", return_value="vector") as vector_cls:
+        result_1 = factory.init_vector(dataset_with_index, attributes=[], embeddings=embeddings)
+        result_2 = factory.init_vector(dataset_without_index, attributes=[], embeddings=embeddings)
+
+    assert result_1 == "vector"
+    assert result_2 == "vector"
+    assert vector_cls.call_args_list[0].kwargs["collection_name"] == "existing_collection"
+    assert vector_cls.call_args_list[1].kwargs["collection_name"] == "auto_collection"
+    assert dataset_without_index.index_struct is not None

api/tests/unit_tests/core/rag/datasource/vdb/pgvector/test_pgvector.py

@@ -1,16 +1,19 @@
-import unittest
+from contextlib import contextmanager
+from types import SimpleNamespace
 from unittest.mock import MagicMock, patch
 
 import pytest
 
+import core.rag.datasource.vdb.pgvector.pgvector as pgvector_module
 from core.rag.datasource.vdb.pgvector.pgvector import (
     PGVector,
     PGVectorConfig,
 )
+from core.rag.models.document import Document
 
 
-class TestPGVector(unittest.TestCase):
+class TestPGVector:
-    def setUp(self):
+    def setup_method(self, method):
         self.config = PGVectorConfig(
             host="localhost",
             port=5432,
@@ -323,5 +326,172 @@ def test_config_validation_parametrized(invalid_config_override):
         PGVectorConfig(**config)
 
 
-if __name__ == "__main__":
+def test_create_delegates_collection_creation_and_insert():
-    unittest.main()
+    vector = PGVector.__new__(PGVector)
+    vector._create_collection = MagicMock()
+    vector.add_texts = MagicMock(return_value=["doc-a"])
+    docs = [Document(page_content="hello", metadata={"doc_id": "doc-a"})]
+
+    result = vector.create(docs, [[0.1, 0.2]])
+
+    assert result == ["doc-a"]
+    vector._create_collection.assert_called_once_with(2)
+    vector.add_texts.assert_called_once_with(docs, [[0.1, 0.2]])
+
+
+def test_add_texts_uses_execute_values_and_returns_ids(monkeypatch):
+    vector = PGVector.__new__(PGVector)
+    vector.table_name = "embedding_collection_1"
+
+    cursor = MagicMock()
+
+    @contextmanager
+    def _cursor_ctx():
+        yield cursor
+
+    vector._get_cursor = _cursor_ctx
+    monkeypatch.setattr(pgvector_module.uuid, "uuid4", lambda: "generated-uuid")
+    execute_values = MagicMock()
+    monkeypatch.setattr(pgvector_module.psycopg2.extras, "execute_values", execute_values)
+
+    docs = [
+        Document(page_content="a", metadata={"doc_id": "doc-a"}),
+        Document(page_content="b", metadata={"document_id": "doc-b"}),
+        SimpleNamespace(page_content="c", metadata=None),
+    ]
+    ids = vector.add_texts(docs, [[0.1], [0.2], [0.3]])
+
+    assert ids == ["doc-a", "generated-uuid"]
+    execute_values.assert_called_once()
+
+
+def test_text_get_and_delete_methods():
+    vector = PGVector.__new__(PGVector)
+    vector.table_name = "embedding_collection_1"
+    cursor = MagicMock()
+    cursor.fetchone.return_value = ("id-1",)
+    cursor.__iter__.return_value = iter([({"doc_id": "1"}, "text-1"), ({"doc_id": "2"}, "text-2")])
+
+    @contextmanager
+    def _cursor_ctx():
+        yield cursor
+
+    vector._get_cursor = _cursor_ctx
+
+    assert vector.text_exists("id-1") is True
+    docs = vector.get_by_ids(["id-1", "id-2"])
+    assert len(docs) == 2
+    assert docs[0].page_content == "text-1"
+
+    vector.delete_by_metadata_field("document_id", "doc-1")
+    vector.delete()
+    executed_sql = [call.args[0] for call in cursor.execute.call_args_list]
+    assert any("meta->>%s = %s" in sql for sql in executed_sql)
+    assert any("DROP TABLE IF EXISTS embedding_collection_1" in sql for sql in executed_sql)
+
+
+def test_delete_by_ids_handles_empty_undefined_table_and_generic_exception(monkeypatch):
+    vector = PGVector.__new__(PGVector)
+    vector.table_name = "embedding_collection_1"
+    cursor = MagicMock()
+
+    @contextmanager
+    def _cursor_ctx():
+        yield cursor
+
+    vector._get_cursor = _cursor_ctx
+    vector.delete_by_ids([])
+    cursor.execute.assert_not_called()
+
+    class _UndefinedTableError(Exception):
+        pass
+
+    monkeypatch.setattr(pgvector_module.psycopg2.errors, "UndefinedTable", _UndefinedTableError)
+    cursor.execute.side_effect = _UndefinedTableError("missing")
+    vector.delete_by_ids(["doc-1"])
+
+    cursor.execute.side_effect = RuntimeError("boom")
+    with pytest.raises(RuntimeError, match="boom"):
+        vector.delete_by_ids(["doc-1"])
+
+
+def test_search_by_vector_supports_filter_and_threshold():
+    vector = PGVector.__new__(PGVector)
+    vector.table_name = "embedding_collection_1"
+    cursor = MagicMock()
+    cursor.__iter__.return_value = iter([({"doc_id": "1"}, "text-1", 0.1), ({"doc_id": "2"}, "text-2", 0.8)])
+
+    @contextmanager
+    def _cursor_ctx():
+        yield cursor
+
+    vector._get_cursor = _cursor_ctx
+
+    with pytest.raises(ValueError, match="top_k must be a positive integer"):
+        vector.search_by_vector([0.1], top_k=0)
+
+    docs = vector.search_by_vector([0.1, 0.2], top_k=2, score_threshold=0.5, document_ids_filter=["d-1"])
+    assert len(docs) == 1
+    assert docs[0].metadata["score"] == pytest.approx(0.9)
+    sql = cursor.execute.call_args.args[0]
+    assert "meta->>'document_id' in ('d-1')" in sql
+
+
+def test_search_by_full_text_branches_for_bigm_and_standard():
+    vector = PGVector.__new__(PGVector)
+    vector.table_name = "embedding_collection_1"
+    cursor = MagicMock()
+    cursor.__iter__.return_value = iter([({"doc_id": "1"}, "text-1", 0.7)])
+
+    @contextmanager
+    def _cursor_ctx():
+        yield cursor
+
+    vector._get_cursor = _cursor_ctx
+
+    with pytest.raises(ValueError, match="top_k must be a positive integer"):
+        vector.search_by_full_text("hello", top_k=0)
+
+    vector.pg_bigm = False
+    docs = vector.search_by_full_text("hello world", top_k=2, document_ids_filter=["d-1"])
+    assert len(docs) == 1
+    assert docs[0].metadata["score"] == pytest.approx(0.7)
+    standard_sql = cursor.execute.call_args.args[0]
+    assert "to_tsvector(text) @@ plainto_tsquery(%s)" in standard_sql
+
+    cursor.execute.reset_mock()
+    cursor.__iter__.return_value = iter([({"doc_id": "2"}, "text-2", 0.6)])
+    vector.pg_bigm = True
+    vector.search_by_full_text("hello world", top_k=2, document_ids_filter=["d-2"])
+    assert "SET pg_bigm.similarity_limit TO 0.000001" in cursor.execute.call_args_list[0].args[0]
+    assert "bigm_similarity" in cursor.execute.call_args_list[1].args[0]
+
+
+def test_pgvector_factory_initializes_expected_collection_name(monkeypatch):
+    factory = pgvector_module.PGVectorFactory()
+    dataset_with_index = SimpleNamespace(
+        id="dataset-1",
+        index_struct_dict={"vector_store": {"class_prefix": "EXISTING_COLLECTION"}},
+        index_struct=None,
+    )
+    dataset_without_index = SimpleNamespace(id="dataset-2", index_struct_dict=None, index_struct=None)
+
+    monkeypatch.setattr(pgvector_module.Dataset, "gen_collection_name_by_id", lambda _id: "AUTO_COLLECTION")
+    monkeypatch.setattr(pgvector_module.dify_config, "PGVECTOR_HOST", "localhost")
+    monkeypatch.setattr(pgvector_module.dify_config, "PGVECTOR_PORT", 5432)
+    monkeypatch.setattr(pgvector_module.dify_config, "PGVECTOR_USER", "postgres")
+    monkeypatch.setattr(pgvector_module.dify_config, "PGVECTOR_PASSWORD", "secret")
+    monkeypatch.setattr(pgvector_module.dify_config, "PGVECTOR_DATABASE", "postgres")
+    monkeypatch.setattr(pgvector_module.dify_config, "PGVECTOR_MIN_CONNECTION", 1)
+    monkeypatch.setattr(pgvector_module.dify_config, "PGVECTOR_MAX_CONNECTION", 5)
+    monkeypatch.setattr(pgvector_module.dify_config, "PGVECTOR_PG_BIGM", False)
+
+    with patch.object(pgvector_module, "PGVector", return_value="vector") as vector_cls:
+        result_1 = factory.init_vector(dataset_with_index, attributes=[], embeddings=MagicMock())
+        result_2 = factory.init_vector(dataset_without_index, attributes=[], embeddings=MagicMock())
+
+    assert result_1 == "vector"
+    assert result_2 == "vector"
+    assert vector_cls.call_args_list[0].kwargs["collection_name"] == "EXISTING_COLLECTION"
+    assert vector_cls.call_args_list[1].kwargs["collection_name"] == "AUTO_COLLECTION"
+    assert dataset_without_index.index_struct is not None

api/tests/unit_tests/core/rag/datasource/vdb/pyvastbase/test_vastbase_vector.py

@@ -0,0 +1,269 @@
+import importlib
+import sys
+import types
+from contextlib import contextmanager
+from types import SimpleNamespace
+from unittest.mock import MagicMock, patch
+
+import pytest
+from pydantic import ValidationError
+
+from core.rag.models.document import Document
+
+
+def _build_fake_psycopg2_modules():
+    psycopg2 = types.ModuleType("psycopg2")
+    psycopg2.__path__ = []
+    psycopg2_extras = types.ModuleType("psycopg2.extras")
+    psycopg2_pool = types.ModuleType("psycopg2.pool")
+
+    class SimpleConnectionPool:
+        def __init__(self, *args, **kwargs):
+            self.args = args
+            self.kwargs = kwargs
+            self.getconn = MagicMock()
+            self.putconn = MagicMock()
+
+    psycopg2_pool.SimpleConnectionPool = SimpleConnectionPool
+    psycopg2_extras.execute_values = MagicMock()
+    psycopg2.pool = psycopg2_pool
+    psycopg2.extras = psycopg2_extras
+
+    return {
+        "psycopg2": psycopg2,
+        "psycopg2.pool": psycopg2_pool,
+        "psycopg2.extras": psycopg2_extras,
+    }
+
+
+@pytest.fixture
+def vastbase_module(monkeypatch):
+    for name, module in _build_fake_psycopg2_modules().items():
+        monkeypatch.setitem(sys.modules, name, module)
+
+    import core.rag.datasource.vdb.pyvastbase.vastbase_vector as module
+
+    return importlib.reload(module)
+
+
+def _config(module):
+    return module.VastbaseVectorConfig(
+        host="localhost",
+        port=5432,
+        user="dify",
+        password="secret",
+        database="dify",
+        min_connection=1,
+        max_connection=5,
+    )
+
+
+@pytest.mark.parametrize(
+    ("field", "value", "message"),
+    [
+        ("host", "", "config VASTBASE_HOST is required"),
+        ("port", 0, "config VASTBASE_PORT is required"),
+        ("user", "", "config VASTBASE_USER is required"),
+        ("password", "", "config VASTBASE_PASSWORD is required"),
+        ("database", "", "config VASTBASE_DATABASE is required"),
+        ("min_connection", 0, "config VASTBASE_MIN_CONNECTION is required"),
+        ("max_connection", 0, "config VASTBASE_MAX_CONNECTION is required"),
+    ],
+)
+def test_vastbase_config_validation(vastbase_module, field, value, message):
+    values = _config(vastbase_module).model_dump()
+    values[field] = value
+
+    with pytest.raises(ValidationError, match=message):
+        vastbase_module.VastbaseVectorConfig.model_validate(values)
+
+
+def test_vastbase_config_rejects_invalid_connection_window(vastbase_module):
+    with pytest.raises(ValidationError, match="VASTBASE_MIN_CONNECTION should less than VASTBASE_MAX_CONNECTION"):
+        vastbase_module.VastbaseVectorConfig.model_validate(
+            {
+                "host": "localhost",
+                "port": 5432,
+                "user": "dify",
+                "password": "secret",
+                "database": "dify",
+                "min_connection": 6,
+                "max_connection": 5,
+            }
+        )
+
+
+def test_init_and_get_cursor_context_manager(vastbase_module, monkeypatch):
+    pool = MagicMock()
+    monkeypatch.setattr(vastbase_module.psycopg2.pool, "SimpleConnectionPool", MagicMock(return_value=pool))
+
+    conn = MagicMock()
+    cur = MagicMock()
+    pool.getconn.return_value = conn
+    conn.cursor.return_value = cur
+
+    vector = vastbase_module.VastbaseVector("collection_1", _config(vastbase_module))
+    assert vector.get_type() == "vastbase"
+    assert vector.table_name == "embedding_collection_1"
+
+    with vector._get_cursor() as got_cur:
+        assert got_cur is cur
+
+    cur.close.assert_called_once()
+    conn.commit.assert_called_once()
+    pool.putconn.assert_called_once_with(conn)
+
+
+def test_create_and_add_texts(vastbase_module, monkeypatch):
+    vector = vastbase_module.VastbaseVector.__new__(vastbase_module.VastbaseVector)
+    vector.table_name = "embedding_collection_1"
+    vector._create_collection = MagicMock()
+
+    cursor = MagicMock()
+
+    @contextmanager
+    def _cursor_ctx():
+        yield cursor
+
+    vector._get_cursor = _cursor_ctx
+    monkeypatch.setattr(vastbase_module.uuid, "uuid4", lambda: "generated-uuid")
+
+    docs = [
+        Document(page_content="a", metadata={"doc_id": "doc-a"}),
+        Document(page_content="b", metadata={"document_id": "doc-b"}),
+        SimpleNamespace(page_content="c", metadata=None),
+    ]
+
+    ids = vector.add_texts(docs, [[0.1], [0.2], [0.3]])
+    assert ids == ["doc-a", "generated-uuid"]
+    vastbase_module.psycopg2.extras.execute_values.assert_called_once()
+
+    vector.add_texts = MagicMock(return_value=["doc-a"])
+    result = vector.create(docs, [[0.1], [0.2], [0.3]])
+    vector._create_collection.assert_called_once_with(1)
+    assert result == ["doc-a"]
+
+
+def test_text_get_delete_and_metadata_methods(vastbase_module):
+    vector = vastbase_module.VastbaseVector.__new__(vastbase_module.VastbaseVector)
+    vector.table_name = "embedding_collection_1"
+    cursor = MagicMock()
+    cursor.fetchone.return_value = ("id-1",)
+    cursor.__iter__.return_value = iter([({"doc_id": "1"}, "text-1"), ({"doc_id": "2"}, "text-2")])
+
+    @contextmanager
+    def _cursor_ctx():
+        yield cursor
+
+    vector._get_cursor = _cursor_ctx
+
+    assert vector.text_exists("id-1") is True
+    docs = vector.get_by_ids(["id-1", "id-2"])
+    assert len(docs) == 2
+    assert docs[0].page_content == "text-1"
+
+    vector.delete_by_ids([])
+    vector.delete_by_ids(["id-1"])
+    vector.delete_by_metadata_field("document_id", "doc-1")
+    vector.delete()
+    executed_sql = [call.args[0] for call in cursor.execute.call_args_list]
+    assert any("DELETE FROM embedding_collection_1 WHERE id IN %s" in sql for sql in executed_sql)
+    assert any("meta->>%s = %s" in sql for sql in executed_sql)
+    assert any("DROP TABLE IF EXISTS embedding_collection_1" in sql for sql in executed_sql)
+
+
+def test_search_by_vector_and_full_text(vastbase_module):
+    vector = vastbase_module.VastbaseVector.__new__(vastbase_module.VastbaseVector)
+    vector.table_name = "embedding_collection_1"
+    cursor = MagicMock()
+    cursor.__iter__.return_value = iter(
+        [
+            ({"doc_id": "1"}, "text-1", 0.1),
+            ({"doc_id": "2"}, "text-2", 0.8),
+        ]
+    )
+
+    @contextmanager
+    def _cursor_ctx():
+        yield cursor
+
+    vector._get_cursor = _cursor_ctx
+
+    with pytest.raises(ValueError, match="top_k must be a positive integer"):
+        vector.search_by_vector([0.1, 0.2], top_k=0)
+
+    docs = vector.search_by_vector([0.1, 0.2], top_k=2, score_threshold=0.5)
+    assert len(docs) == 1
+    assert docs[0].metadata["score"] == pytest.approx(0.9)
+
+    with pytest.raises(ValueError, match="top_k must be a positive integer"):
+        vector.search_by_full_text("hello", top_k=0)
+
+    cursor.__iter__.return_value = iter([({"doc_id": "3"}, "full-text", 0.7)])
+    full_docs = vector.search_by_full_text("hello world", top_k=2)
+    assert len(full_docs) == 1
+    assert full_docs[0].page_content == "full-text"
+
+
+def test_create_collection_cache_and_dimension_branches(vastbase_module, monkeypatch):
+    lock = MagicMock()
+    lock.__enter__.return_value = None
+    lock.__exit__.return_value = None
+    monkeypatch.setattr(vastbase_module.redis_client, "lock", MagicMock(return_value=lock))
+    monkeypatch.setattr(vastbase_module.redis_client, "set", MagicMock())
+
+    vector = vastbase_module.VastbaseVector.__new__(vastbase_module.VastbaseVector)
+    vector._collection_name = "collection_1"
+    vector.table_name = "embedding_collection_1"
+    cursor = MagicMock()
+
+    @contextmanager
+    def _cursor_ctx():
+        yield cursor
+
+    vector._get_cursor = _cursor_ctx
+
+    monkeypatch.setattr(vastbase_module.redis_client, "get", MagicMock(return_value=1))
+    vector._create_collection(3)
+    cursor.execute.assert_not_called()
+
+    monkeypatch.setattr(vastbase_module.redis_client, "get", MagicMock(return_value=None))
+    vector._create_collection(17000)
+    executed_sql = [call.args[0] for call in cursor.execute.call_args_list]
+    assert any("CREATE TABLE IF NOT EXISTS embedding_collection_1" in sql for sql in executed_sql)
+    assert all("embedding_cosine_v1_idx" not in sql for sql in executed_sql)
+
+    cursor.execute.reset_mock()
+    vector._create_collection(3)
+    executed_sql = [call.args[0] for call in cursor.execute.call_args_list]
+    assert any("embedding_cosine_v1_idx" in sql for sql in executed_sql)
+    vastbase_module.redis_client.set.assert_called()
+
+
+def test_vastbase_factory_uses_existing_or_generated_collection(vastbase_module, monkeypatch):
+    factory = vastbase_module.VastbaseVectorFactory()
+    dataset_with_index = SimpleNamespace(
+        id="dataset-1",
+        index_struct_dict={"vector_store": {"class_prefix": "EXISTING_COLLECTION"}},
+        index_struct=None,
+    )
+    dataset_without_index = SimpleNamespace(id="dataset-2", index_struct_dict=None, index_struct=None)
+
+    monkeypatch.setattr(vastbase_module.Dataset, "gen_collection_name_by_id", lambda _id: "AUTO_COLLECTION")
+    monkeypatch.setattr(vastbase_module.dify_config, "VASTBASE_HOST", "localhost")
+    monkeypatch.setattr(vastbase_module.dify_config, "VASTBASE_PORT", 5432)
+    monkeypatch.setattr(vastbase_module.dify_config, "VASTBASE_USER", "dify")
+    monkeypatch.setattr(vastbase_module.dify_config, "VASTBASE_PASSWORD", "secret")
+    monkeypatch.setattr(vastbase_module.dify_config, "VASTBASE_DATABASE", "dify")
+    monkeypatch.setattr(vastbase_module.dify_config, "VASTBASE_MIN_CONNECTION", 1)
+    monkeypatch.setattr(vastbase_module.dify_config, "VASTBASE_MAX_CONNECTION", 5)
+
+    with patch.object(vastbase_module, "VastbaseVector", return_value="vector") as vector_cls:
+        result_1 = factory.init_vector(dataset_with_index, attributes=[], embeddings=MagicMock())
+        result_2 = factory.init_vector(dataset_without_index, attributes=[], embeddings=MagicMock())
+
+    assert result_1 == "vector"
+    assert result_2 == "vector"
+    assert vector_cls.call_args_list[0].kwargs["collection_name"] == "EXISTING_COLLECTION"
+    assert vector_cls.call_args_list[1].kwargs["collection_name"] == "AUTO_COLLECTION"
+    assert dataset_without_index.index_struct is not None

api/tests/unit_tests/core/rag/datasource/vdb/qdrant/test_qdrant_vector.py

@@ -0,0 +1,328 @@
+import importlib
+import os
+import sys
+import types
+from collections import UserDict
+from types import SimpleNamespace
+from unittest.mock import MagicMock, patch
+
+import pytest
+
+from core.rag.models.document import Document
+
+
+def _build_fake_qdrant_modules():
+    qdrant_client = types.ModuleType("qdrant_client")
+    qdrant_http = types.ModuleType("qdrant_client.http")
+    qdrant_http_models = types.ModuleType("qdrant_client.http.models")
+    qdrant_http_exceptions = types.ModuleType("qdrant_client.http.exceptions")
+    qdrant_local_pkg = types.ModuleType("qdrant_client.local")
+    qdrant_local_mod = types.ModuleType("qdrant_client.local.qdrant_local")
+
+    class UnexpectedResponseError(Exception):
+        def __init__(self, status_code):
+            super().__init__(f"status={status_code}")
+            self.status_code = status_code
+
+    class FilterSelector:
+        def __init__(self, filter):
+            self.filter = filter
+
+    class HnswConfigDiff:
+        def __init__(self, **kwargs):
+            self.kwargs = kwargs
+
+    class TextIndexParams:
+        def __init__(self, **kwargs):
+            self.kwargs = kwargs
+
+    class VectorParams:
+        def __init__(self, **kwargs):
+            self.kwargs = kwargs
+
+    class PointStruct:
+        def __init__(self, **kwargs):
+            self.id = kwargs["id"]
+            self.vector = kwargs["vector"]
+            self.payload = kwargs["payload"]
+
+    class Filter:
+        def __init__(self, must=None):
+            self.must = must or []
+
+    class FieldCondition:
+        def __init__(self, key, match):
+            self.key = key
+            self.match = match
+
+    class MatchValue:
+        def __init__(self, value):
+            self.value = value
+
+    class MatchAny:
+        def __init__(self, any):
+            self.any = any
+
+    class MatchText:
+        def __init__(self, text):
+            self.text = text
+
+    class _Distance(UserDict):
+        def __getitem__(self, key):
+            return key
+
+    class QdrantClient:
+        def __init__(self, **kwargs):
+            self.kwargs = kwargs
+            self.get_collections = MagicMock(return_value=SimpleNamespace(collections=[]))
+            self.create_collection = MagicMock()
+            self.create_payload_index = MagicMock()
+            self.upsert = MagicMock()
+            self.delete = MagicMock()
+            self.delete_collection = MagicMock()
+            self.retrieve = MagicMock(return_value=[])
+            self.search = MagicMock(return_value=[])
+            self.scroll = MagicMock(return_value=([], None))
+
+    class QdrantLocal(QdrantClient):
+        def __init__(self, **kwargs):
+            super().__init__(**kwargs)
+            self._load = MagicMock()
+
+    qdrant_client.QdrantClient = QdrantClient
+    qdrant_http_models.FilterSelector = FilterSelector
+    qdrant_http_models.HnswConfigDiff = HnswConfigDiff
+    qdrant_http_models.PayloadSchemaType = SimpleNamespace(KEYWORD="KEYWORD")
+    qdrant_http_models.TextIndexParams = TextIndexParams
+    qdrant_http_models.TextIndexType = SimpleNamespace(TEXT="TEXT")
+    qdrant_http_models.TokenizerType = SimpleNamespace(MULTILINGUAL="MULTILINGUAL")
+    qdrant_http_models.VectorParams = VectorParams
+    qdrant_http_models.Distance = _Distance()
+    qdrant_http_models.PointStruct = PointStruct
+    qdrant_http_models.Filter = Filter
+    qdrant_http_models.FieldCondition = FieldCondition
+    qdrant_http_models.MatchValue = MatchValue
+    qdrant_http_models.MatchAny = MatchAny
+    qdrant_http_models.MatchText = MatchText
+    qdrant_http_exceptions.UnexpectedResponse = UnexpectedResponseError
+
+    qdrant_http.models = qdrant_http_models
+    qdrant_local_mod.QdrantLocal = QdrantLocal
+    qdrant_local_pkg.qdrant_local = qdrant_local_mod
+
+    return {
+        "qdrant_client": qdrant_client,
+        "qdrant_client.http": qdrant_http,
+        "qdrant_client.http.models": qdrant_http_models,
+        "qdrant_client.http.exceptions": qdrant_http_exceptions,
+        "qdrant_client.local": qdrant_local_pkg,
+        "qdrant_client.local.qdrant_local": qdrant_local_mod,
+    }
+
+
+@pytest.fixture
+def qdrant_module(monkeypatch):
+    for name, module in _build_fake_qdrant_modules().items():
+        monkeypatch.setitem(sys.modules, name, module)
+
+    import core.rag.datasource.vdb.qdrant.qdrant_vector as module
+
+    return importlib.reload(module)
+
+
+def _config(module, **overrides):
+    values = {
+        "endpoint": "http://localhost:6333",
+        "api_key": "api-key",
+        "timeout": 20,
+        "root_path": "/tmp",
+        "grpc_port": 6334,
+        "prefer_grpc": False,
+        "replication_factor": 1,
+        "write_consistency_factor": 1,
+    }
+    values.update(overrides)
+    return module.QdrantConfig.model_validate(values)
+
+
+def test_qdrant_config_to_params(qdrant_module):
+    url_params = _config(qdrant_module).to_qdrant_params().model_dump()
+    assert url_params["url"] == "http://localhost:6333"
+    assert url_params["verify"] is False
+
+    path_config = _config(qdrant_module, endpoint="path:storage")
+    assert path_config.to_qdrant_params().path == os.path.join("/tmp", "storage")
+
+    with pytest.raises(ValueError, match="Root path is not set"):
+        _config(qdrant_module, endpoint="path:storage", root_path=None).to_qdrant_params()
+
+
+def test_init_and_basic_behaviour(qdrant_module):
+    vector = qdrant_module.QdrantVector("collection_1", "group-1", _config(qdrant_module))
+    assert vector.get_type() == qdrant_module.VectorType.QDRANT
+    assert vector.to_index_struct()["vector_store"]["class_prefix"] == "collection_1"
+
+    docs = [Document(page_content="a", metadata={"doc_id": "a"})]
+    vector.create_collection = MagicMock()
+    vector.add_texts = MagicMock()
+    vector.create(docs, [[0.1]])
+    vector.create_collection.assert_called_once_with("collection_1", 1)
+    vector.add_texts.assert_called_once()
+
+
+def test_create_collection_and_add_texts(qdrant_module, monkeypatch):
+    vector = qdrant_module.QdrantVector("collection_1", "group-1", _config(qdrant_module))
+    lock = MagicMock()
+    lock.__enter__.return_value = None
+    lock.__exit__.return_value = None
+    monkeypatch.setattr(qdrant_module.redis_client, "lock", MagicMock(return_value=lock))
+    monkeypatch.setattr(qdrant_module.redis_client, "set", MagicMock())
+
+    monkeypatch.setattr(qdrant_module.redis_client, "get", MagicMock(return_value=1))
+    vector.create_collection("collection_1", 3)
+    vector._client.create_collection.assert_not_called()
+
+    monkeypatch.setattr(qdrant_module.redis_client, "get", MagicMock(return_value=None))
+    vector._client.get_collections.return_value = SimpleNamespace(collections=[])
+    vector.create_collection("collection_1", 3)
+    vector._client.create_collection.assert_called_once()
+    assert vector._client.create_payload_index.call_count == 4
+    qdrant_module.redis_client.set.assert_called_once()
+
+    # add_texts and generated batches
+    docs = [
+        Document(page_content="a", metadata={"doc_id": "id-1"}),
+        Document(page_content="b", metadata={"doc_id": "id-2"}),
+    ]
+    ids = vector.add_texts(docs, [[0.1], [0.2]])
+    assert ids == ["id-1", "id-2"]
+    assert vector._client.upsert.call_count == 1
+
+    payloads = qdrant_module.QdrantVector._build_payloads(
+        ["a"], [{"doc_id": "id-1"}], "content", "metadata", "g1", "group_id"
+    )
+    assert payloads[0]["group_id"] == "g1"
+    with pytest.raises(ValueError, match="At least one of the texts is None"):
+        qdrant_module.QdrantVector._build_payloads(
+            [None], [{"doc_id": "id-1"}], "content", "metadata", "g1", "group_id"
+        )
+
+
+def test_delete_and_exists_paths(qdrant_module):
+    vector = qdrant_module.QdrantVector("collection_1", "group-1", _config(qdrant_module))
+    unexpected = sys.modules["qdrant_client.http.exceptions"].UnexpectedResponse
+
+    vector._client.delete.side_effect = unexpected(404)
+    vector.delete_by_metadata_field("document_id", "doc-1")
+    vector._client.delete.side_effect = None
+
+    vector._client.delete.side_effect = unexpected(500)
+    with pytest.raises(unexpected):
+        vector.delete_by_metadata_field("document_id", "doc-1")
+    vector._client.delete.side_effect = None
+
+    vector._client.delete.side_effect = unexpected(404)
+    vector.delete()
+    vector._client.delete.side_effect = unexpected(500)
+    with pytest.raises(unexpected):
+        vector.delete()
+    vector._client.delete.side_effect = None
+
+    vector._client.delete.side_effect = unexpected(404)
+    vector.delete_by_ids(["doc-1"])
+    vector._client.delete.side_effect = unexpected(500)
+    with pytest.raises(unexpected):
+        vector.delete_by_ids(["doc-1"])
+    vector._client.delete.side_effect = None
+
+    vector._client.get_collections.return_value = SimpleNamespace(collections=[SimpleNamespace(name="other")])
+    assert vector.text_exists("id-1") is False
+    vector._client.get_collections.return_value = SimpleNamespace(collections=[SimpleNamespace(name="collection_1")])
+    vector._client.retrieve.return_value = [{"id": "id-1"}]
+    assert vector.text_exists("id-1") is True
+
+
+def test_search_and_helper_methods(qdrant_module):
+    vector = qdrant_module.QdrantVector("collection_1", "group-1", _config(qdrant_module))
+    assert vector.search_by_vector([0.1], score_threshold=1.0) == []
+
+    vector._client.search.return_value = [
+        SimpleNamespace(payload=None, score=0.9, vector=[0.1]),
+        SimpleNamespace(payload={"metadata": {"doc_id": "1"}, "page_content": "doc-a"}, score=0.8, vector=[0.1]),
+    ]
+    docs = vector.search_by_vector([0.1], top_k=2, score_threshold=0.5, document_ids_filter=["d-1"])
+    assert len(docs) == 1
+    assert docs[0].metadata["score"] == pytest.approx(0.8)
+
+    # full text search: keyword split, dedup and top_k limit
+    scroll_results = [
+        (
+            [
+                SimpleNamespace(id="p1", payload={"page_content": "doc-1", "metadata": {"doc_id": "1"}}, vector=[0.1]),
+                SimpleNamespace(id="p2", payload={"page_content": "doc-2", "metadata": {"doc_id": "2"}}, vector=[0.2]),
+            ],
+            None,
+        ),
+        (
+            [
+                SimpleNamespace(id="p2", payload={"page_content": "doc-2", "metadata": {"doc_id": "2"}}, vector=[0.2]),
+            ],
+            None,
+        ),
+    ]
+    vector._client.scroll.side_effect = scroll_results
+    docs = vector.search_by_full_text("hello world", top_k=2, document_ids_filter=["d-1"])
+    assert len(docs) == 2
+    assert vector.search_by_full_text("   ", top_k=2) == []
+
+    local_client = qdrant_module.QdrantLocal()
+    vector._client = local_client
+    vector._reload_if_needed()
+    local_client._load.assert_called_once()
+
+    doc = vector._document_from_scored_point(
+        SimpleNamespace(payload={"page_content": "doc", "metadata": {"doc_id": "1"}}, vector=[0.1]),
+        "page_content",
+        "metadata",
+    )
+    assert doc.page_content == "doc"
+
+
+def test_qdrant_factory_paths(qdrant_module, monkeypatch):
+    factory = qdrant_module.QdrantVectorFactory()
+    dataset = SimpleNamespace(
+        id="dataset-1",
+        tenant_id="tenant-1",
+        collection_binding_id=None,
+        index_struct_dict=None,
+        index_struct=None,
+    )
+    monkeypatch.setattr(qdrant_module.Dataset, "gen_collection_name_by_id", lambda _id: "AUTO_COLLECTION")
+    monkeypatch.setattr(qdrant_module, "current_app", SimpleNamespace(config=SimpleNamespace(root_path="/root")))
+    monkeypatch.setattr(qdrant_module.dify_config, "QDRANT_URL", "http://localhost:6333")
+    monkeypatch.setattr(qdrant_module.dify_config, "QDRANT_API_KEY", "api-key")
+    monkeypatch.setattr(qdrant_module.dify_config, "QDRANT_CLIENT_TIMEOUT", 20)
+    monkeypatch.setattr(qdrant_module.dify_config, "QDRANT_GRPC_PORT", 6334)
+    monkeypatch.setattr(qdrant_module.dify_config, "QDRANT_GRPC_ENABLED", False)
+    monkeypatch.setattr(qdrant_module.dify_config, "QDRANT_REPLICATION_FACTOR", 1)
+
+    with patch.object(qdrant_module, "QdrantVector", return_value="vector") as vector_cls:
+        result = factory.init_vector(dataset, attributes=[], embeddings=MagicMock())
+    assert result == "vector"
+    assert vector_cls.call_args.kwargs["collection_name"] == "AUTO_COLLECTION"
+    assert dataset.index_struct is not None
+
+    # collection binding lookup path
+    dataset.collection_binding_id = "binding-1"
+    dataset.index_struct_dict = {"vector_store": {"class_prefix": "existing"}}
+    monkeypatch.setattr(qdrant_module, "select", lambda _model: SimpleNamespace(where=lambda *_args: "stmt"))
+    qdrant_module.db.session.scalars = MagicMock(
+        return_value=SimpleNamespace(one_or_none=lambda: SimpleNamespace(collection_name="BOUND_COLLECTION"))
+    )
+    with patch.object(qdrant_module, "QdrantVector", return_value="vector") as vector_cls:
+        factory.init_vector(dataset, attributes=[], embeddings=MagicMock())
+    assert vector_cls.call_args.kwargs["collection_name"] == "BOUND_COLLECTION"
+
+    qdrant_module.db.session.scalars = MagicMock(return_value=SimpleNamespace(one_or_none=lambda: None))
+    with pytest.raises(ValueError, match="Dataset Collection Bindings does not exist"):
+        factory.init_vector(dataset, attributes=[], embeddings=MagicMock())

api/tests/unit_tests/core/rag/datasource/vdb/relyt/test_relyt_vector.py

@@ -0,0 +1,303 @@
+import importlib
+import sys
+import types
+from types import SimpleNamespace
+from unittest.mock import MagicMock, patch
+
+import pytest
+from pydantic import ValidationError
+from sqlalchemy.types import UserDefinedType
+
+from core.rag.models.document import Document
+
+
+def _build_fake_relyt_modules():
+    pgvecto_rs = types.ModuleType("pgvecto_rs")
+    pgvecto_rs_sqlalchemy = types.ModuleType("pgvecto_rs.sqlalchemy")
+
+    class VECTOR(UserDefinedType):
+        def __init__(self, dim):
+            self.dim = dim
+
+    pgvecto_rs_sqlalchemy.VECTOR = VECTOR
+    return {
+        "pgvecto_rs": pgvecto_rs,
+        "pgvecto_rs.sqlalchemy": pgvecto_rs_sqlalchemy,
+    }
+
+
+class _FakeSession:
+    def __init__(self, execute_result=None):
+        self.execute_result = execute_result or MagicMock(fetchall=lambda: [])
+        self.execute = MagicMock(return_value=self.execute_result)
+        self.commit = MagicMock()
+
+    def __enter__(self):
+        return self
+
+    def __exit__(self, exc_type, exc, tb):
+        return None
+
+
+@pytest.fixture
+def relyt_module(monkeypatch):
+    for name, module in _build_fake_relyt_modules().items():
+        monkeypatch.setitem(sys.modules, name, module)
+
+    import core.rag.datasource.vdb.relyt.relyt_vector as module
+
+    return importlib.reload(module)
+
+
+def _config(module, **overrides):
+    values = {
+        "host": "localhost",
+        "port": 5432,
+        "user": "postgres",
+        "password": "secret",
+        "database": "relyt",
+    }
+    values.update(overrides)
+    return module.RelytConfig.model_validate(values)
+
+
+@pytest.mark.parametrize(
+    ("field", "value", "message"),
+    [
+        ("host", "", "config RELYT_HOST is required"),
+        ("port", 0, "config RELYT_PORT is required"),
+        ("user", "", "config RELYT_USER is required"),
+        ("password", "", "config RELYT_PASSWORD is required"),
+        ("database", "", "config RELYT_DATABASE is required"),
+    ],
+)
+def test_relyt_config_validation(relyt_module, field, value, message):
+    values = _config(relyt_module).model_dump()
+    values[field] = value
+    with pytest.raises(ValidationError, match=message):
+        relyt_module.RelytConfig.model_validate(values)
+
+
+def test_init_get_type_and_create_delegate(relyt_module, monkeypatch):
+    engine = MagicMock()
+    monkeypatch.setattr(relyt_module, "create_engine", MagicMock(return_value=engine))
+    vector = relyt_module.RelytVector("collection_1", _config(relyt_module), group_id="group-1")
+    vector.create_collection = MagicMock()
+    vector.add_texts = MagicMock()
+    docs = [Document(page_content="hello", metadata={"doc_id": "seg-1"})]
+
+    vector.create(docs, [[0.1, 0.2]])
+
+    assert vector.get_type() == relyt_module.VectorType.RELYT
+    assert vector._url == "postgresql+psycopg2://postgres:secret@localhost:5432/relyt"
+    assert vector.embedding_dimension == 2
+    vector.create_collection.assert_called_once_with(2)
+    vector.add_texts.assert_called_once_with(docs, [[0.1, 0.2]])
+
+
+def test_create_collection_cache_and_sql_execution(relyt_module, monkeypatch):
+    lock = MagicMock()
+    lock.__enter__.return_value = None
+    lock.__exit__.return_value = None
+    monkeypatch.setattr(relyt_module.redis_client, "lock", MagicMock(return_value=lock))
+    monkeypatch.setattr(relyt_module.redis_client, "set", MagicMock())
+
+    vector = relyt_module.RelytVector.__new__(relyt_module.RelytVector)
+    vector._collection_name = "collection_1"
+    vector.client = MagicMock()
+
+    monkeypatch.setattr(relyt_module.redis_client, "get", MagicMock(return_value=1))
+    session = _FakeSession()
+    monkeypatch.setattr(relyt_module, "Session", lambda _client: session)
+    vector.create_collection(3)
+    session.execute.assert_not_called()
+
+    monkeypatch.setattr(relyt_module.redis_client, "get", MagicMock(return_value=None))
+    session = _FakeSession()
+    monkeypatch.setattr(relyt_module, "Session", lambda _client: session)
+    vector.create_collection(3)
+    executed_sql = [str(call.args[0]) for call in session.execute.call_args_list]
+    assert any("DROP TABLE IF EXISTS" in sql for sql in executed_sql)
+    assert any("CREATE TABLE IF NOT EXISTS" in sql for sql in executed_sql)
+    assert any("CREATE INDEX" in sql for sql in executed_sql)
+    relyt_module.redis_client.set.assert_called_once()
+
+
+def test_add_texts_and_metadata_queries(relyt_module, monkeypatch):
+    vector = relyt_module.RelytVector.__new__(relyt_module.RelytVector)
+    vector._collection_name = "collection_1"
+    vector._group_id = "group-1"
+    vector.client = MagicMock()
+
+    begin_ctx = MagicMock()
+    begin_ctx.__enter__.return_value = None
+    begin_ctx.__exit__.return_value = None
+    conn = MagicMock()
+    conn.__enter__.return_value = conn
+    conn.__exit__.return_value = None
+    conn.begin.return_value = begin_ctx
+    vector.client.connect.return_value = conn
+
+    monkeypatch.setattr(relyt_module.uuid, "uuid1", MagicMock(side_effect=["id-1", "id-2"]))
+    docs = [
+        Document(page_content="a", metadata={"doc_id": "d-1"}),
+        Document(page_content="b", metadata={"doc_id": "d-2"}),
+    ]
+    ids = vector.add_texts(docs, [[0.1], [0.2]])
+
+    assert ids == ["id-1", "id-2"]
+    assert conn.execute.call_count >= 1
+    first_insert_values = conn.execute.call_args.args[0].compile().params
+    assert "group_id" in str(first_insert_values)
+
+    session = _FakeSession(execute_result=MagicMock(fetchall=lambda: [("id-a",), ("id-b",)]))
+    monkeypatch.setattr(relyt_module, "Session", lambda _client: session)
+    assert vector.get_ids_by_metadata_field("document_id", "doc-1") == ["id-a", "id-b"]
+
+    session = _FakeSession(execute_result=MagicMock(fetchall=lambda: []))
+    monkeypatch.setattr(relyt_module, "Session", lambda _client: session)
+    assert vector.get_ids_by_metadata_field("document_id", "doc-1") is None
+
+
+# 1. delete_by_uuids: success and connect error
+def test_delete_by_uuids_success_and_connect_error(relyt_module):
+    vector = relyt_module.RelytVector.__new__(relyt_module.RelytVector)
+    vector._collection_name = "collection_1"
+    vector.client = MagicMock()
+    vector.embedding_dimension = 3
+    with pytest.raises(ValueError, match="No ids provided"):
+        vector.delete_by_uuids(None)
+    conn = MagicMock()
+    conn.__enter__.return_value = conn
+    conn.__exit__.return_value = None
+    begin_ctx = MagicMock()
+    begin_ctx.__enter__.return_value = None
+    begin_ctx.__exit__.return_value = None
+    conn.begin.return_value = begin_ctx
+    vector.client.connect.return_value = conn
+    assert vector.delete_by_uuids(["id-1"]) is True
+    vector.client.connect.side_effect = RuntimeError("boom")
+    assert vector.delete_by_uuids(["id-1"]) is False
+
+
+# 2. delete_by_metadata_field calls delete_by_uuids
+def test_delete_by_metadata_field_calls_delete_by_uuids(relyt_module):
+    vector = relyt_module.RelytVector.__new__(relyt_module.RelytVector)
+    vector._collection_name = "collection_1"
+    vector.client = MagicMock()
+    vector.embedding_dimension = 3
+    vector.get_ids_by_metadata_field = MagicMock(return_value=["id-1"])
+    vector.delete_by_uuids = MagicMock(return_value=True)
+    vector.delete_by_metadata_field("document_id", "doc-1")
+    vector.delete_by_uuids.assert_called_once_with(["id-1"])
+
+
+# 3. delete_by_ids translates to uuids
+def test_delete_by_ids_translates_to_uuids(relyt_module, monkeypatch):
+    vector = relyt_module.RelytVector.__new__(relyt_module.RelytVector)
+    vector._collection_name = "collection_1"
+    vector.client = MagicMock()
+    vector.embedding_dimension = 3
+    session = _FakeSession(execute_result=MagicMock(fetchall=lambda: [("uuid-1",), ("uuid-2",)]))
+    monkeypatch.setattr(relyt_module, "Session", lambda _client: session)
+    vector.delete_by_uuids = MagicMock(return_value=True)
+    vector.delete_by_ids(["doc-1", "doc-2"])
+    vector.delete_by_uuids.assert_called_once_with(["uuid-1", "uuid-2"])
+
+
+# 4. text_exists True
+def test_text_exists_true(relyt_module, monkeypatch):
+    vector = relyt_module.RelytVector.__new__(relyt_module.RelytVector)
+    vector._collection_name = "collection_1"
+    vector.client = MagicMock()
+    vector.embedding_dimension = 3
+    session = _FakeSession(execute_result=MagicMock(fetchall=lambda: [("id-1",)]))
+    monkeypatch.setattr(relyt_module, "Session", lambda _client: session)
+    assert vector.text_exists("doc-1") is True
+
+
+# 5. text_exists False
+def test_text_exists_false(relyt_module, monkeypatch):
+    vector = relyt_module.RelytVector.__new__(relyt_module.RelytVector)
+    vector._collection_name = "collection_1"
+    vector.client = MagicMock()
+    vector.embedding_dimension = 3
+    session = _FakeSession(execute_result=MagicMock(fetchall=lambda: []))
+    monkeypatch.setattr(relyt_module, "Session", lambda _client: session)
+    assert vector.text_exists("doc-1") is False
+
+
+# 6. similarity_search_with_score_by_vector returns Documents and scores
+def test_similarity_search_with_score_by_vector(relyt_module):
+    vector = relyt_module.RelytVector.__new__(relyt_module.RelytVector)
+    vector._collection_name = "collection_1"
+    vector.client = MagicMock()
+    vector.embedding_dimension = 3
+    result_rows = [
+        SimpleNamespace(document="doc-a", metadata={"doc_id": "1"}, distance=0.1),
+        SimpleNamespace(document="doc-b", metadata={"doc_id": "2"}, distance=0.8),
+    ]
+    conn = MagicMock()
+    conn.__enter__.return_value = conn
+    conn.__exit__.return_value = None
+    conn.execute.return_value.fetchall.return_value = result_rows
+    vector.client.connect.return_value = conn
+    similarities = vector.similarity_search_with_score_by_vector([0.1, 0.2], k=2, filter={"document_id": ["d-1"]})
+    assert len(similarities) == 2
+    assert similarities[0][0].page_content == "doc-a"
+
+
+# 7. search_by_vector filters by score and ids
+def test_search_by_vector_filters_by_score_and_ids(relyt_module):
+    vector = relyt_module.RelytVector.__new__(relyt_module.RelytVector)
+    vector._collection_name = "collection_1"
+    vector.client = MagicMock()
+    vector.embedding_dimension = 3
+    vector.similarity_search_with_score_by_vector = MagicMock(
+        return_value=[
+            (Document(page_content="a", metadata={"doc_id": "1"}), 0.1),
+            (Document(page_content="b", metadata={}), 0.9),
+        ]
+    )
+    docs = vector.search_by_vector([0.1], top_k=2, score_threshold=0.5, document_ids_filter=["d-1"])
+    assert len(docs) == 1
+    assert vector.search_by_full_text("query") == []
+
+
+# 8. delete commits session
+def test_delete_commits_session(relyt_module, monkeypatch):
+    vector = relyt_module.RelytVector.__new__(relyt_module.RelytVector)
+    vector._collection_name = "collection_1"
+    vector.client = MagicMock()
+    vector.embedding_dimension = 3
+    session = _FakeSession()
+    monkeypatch.setattr(relyt_module, "Session", lambda _client: session)
+    vector.delete()
+    session.commit.assert_called_once()
+
+
+def test_relyt_factory_existing_and_generated_collection(relyt_module, monkeypatch):
+    factory = relyt_module.RelytVectorFactory()
+    dataset_with_index = SimpleNamespace(
+        id="dataset-1",
+        index_struct_dict={"vector_store": {"class_prefix": "EXISTING_COLLECTION"}},
+        index_struct=None,
+    )
+    dataset_without_index = SimpleNamespace(id="dataset-2", index_struct_dict=None, index_struct=None)
+
+    monkeypatch.setattr(relyt_module.Dataset, "gen_collection_name_by_id", lambda _id: "AUTO_COLLECTION")
+    monkeypatch.setattr(relyt_module.dify_config, "RELYT_HOST", "localhost")
+    monkeypatch.setattr(relyt_module.dify_config, "RELYT_PORT", 5432)
+    monkeypatch.setattr(relyt_module.dify_config, "RELYT_USER", "postgres")
+    monkeypatch.setattr(relyt_module.dify_config, "RELYT_PASSWORD", "secret")
+    monkeypatch.setattr(relyt_module.dify_config, "RELYT_DATABASE", "relyt")
+
+    with patch.object(relyt_module, "RelytVector", return_value="vector") as vector_cls:
+        result_1 = factory.init_vector(dataset_with_index, attributes=[], embeddings=MagicMock())
+        result_2 = factory.init_vector(dataset_without_index, attributes=[], embeddings=MagicMock())
+
+    assert result_1 == "vector"
+    assert result_2 == "vector"
+    assert vector_cls.call_args_list[0].kwargs["collection_name"] == "EXISTING_COLLECTION"
+    assert vector_cls.call_args_list[1].kwargs["collection_name"] == "AUTO_COLLECTION"
+    assert dataset_without_index.index_struct is not None

api/tests/unit_tests/core/rag/datasource/vdb/tablestore/test_tablestore_vector.py

@@ -0,0 +1,316 @@
+import importlib
+import json
+import sys
+import types
+from types import SimpleNamespace
+from unittest.mock import MagicMock, patch
+
+import pytest
+from pydantic import ValidationError
+
+from core.rag.models.document import Document
+
+
+def _build_fake_tablestore_module():
+    tablestore = types.ModuleType("tablestore")
+
+    class _BatchGetRowRequest:
+        def __init__(self):
+            self.items = []
+
+        def add(self, item):
+            self.items.append(item)
+
+    class _TableInBatchGetRowItem:
+        def __init__(self, table_name, rows_to_get, columns_to_get, _unused, _ver):
+            self.table_name = table_name
+            self.rows_to_get = rows_to_get
+            self.columns_to_get = columns_to_get
+
+    class _Row:
+        def __init__(self, primary_key, attribute_columns=None):
+            self.primary_key = primary_key
+            self.attribute_columns = attribute_columns or []
+
+    class _Client:
+        def __init__(self, *_args):
+            self.list_table = MagicMock(return_value=[])
+            self.create_table = MagicMock()
+            self.list_search_index = MagicMock(return_value=[])
+            self.create_search_index = MagicMock()
+            self.delete_search_index = MagicMock()
+            self.delete_table = MagicMock()
+            self.put_row = MagicMock()
+            self.delete_row = MagicMock()
+            self.get_row = MagicMock(return_value=(None, None, None))
+            self.batch_get_row = MagicMock()
+            self.search = MagicMock()
+
+    tablestore.OTSClient = _Client
+    tablestore.BatchGetRowRequest = _BatchGetRowRequest
+    tablestore.TableInBatchGetRowItem = _TableInBatchGetRowItem
+    tablestore.Row = _Row
+    tablestore.TableMeta = lambda name, schema: ("table_meta", name, schema)
+    tablestore.TableOptions = lambda: ("table_options",)
+    tablestore.CapacityUnit = lambda read, write: ("capacity", read, write)
+    tablestore.ReservedThroughput = lambda cap: ("reserved", cap)
+    tablestore.FieldSchema = lambda *args, **kwargs: ("field", args, kwargs)
+    tablestore.VectorOptions = lambda **kwargs: ("vector_options", kwargs)
+    tablestore.SearchIndexMeta = lambda field_schemas: ("search_index_meta", field_schemas)
+    tablestore.SearchQuery = lambda query, **kwargs: SimpleNamespace(query=query, **kwargs)
+    tablestore.TermQuery = lambda key, value: ("term_query", key, value)
+    tablestore.ColumnsToGet = lambda **kwargs: ("columns_to_get", kwargs)
+    tablestore.KnnVectorQuery = lambda **kwargs: SimpleNamespace(**kwargs)
+    tablestore.TermsQuery = lambda key, values: ("terms_query", key, values)
+    tablestore.Sort = lambda **kwargs: ("sort", kwargs)
+    tablestore.ScoreSort = lambda **kwargs: ("score_sort", kwargs)
+    tablestore.BoolQuery = lambda **kwargs: SimpleNamespace(**kwargs)
+    tablestore.MatchQuery = lambda **kwargs: ("match_query", kwargs)
+
+    tablestore.FieldType = SimpleNamespace(TEXT="TEXT", VECTOR="VECTOR", KEYWORD="KEYWORD")
+    tablestore.AnalyzerType = SimpleNamespace(MAXWORD="MAXWORD")
+    tablestore.VectorDataType = SimpleNamespace(VD_FLOAT_32="VD_FLOAT_32")
+    tablestore.VectorMetricType = SimpleNamespace(VM_COSINE="VM_COSINE")
+    tablestore.ColumnReturnType = SimpleNamespace(SPECIFIED="SPECIFIED", ALL_FROM_INDEX="ALL_FROM_INDEX")
+    tablestore.SortOrder = SimpleNamespace(DESC="DESC")
+    return tablestore
+
+
+@pytest.fixture
+def tablestore_module(monkeypatch):
+    fake_module = _build_fake_tablestore_module()
+    monkeypatch.setitem(sys.modules, "tablestore", fake_module)
+
+    import core.rag.datasource.vdb.tablestore.tablestore_vector as module
+
+    return importlib.reload(module)
+
+
+def _config(module, **overrides):
+    values = {
+        "access_key_id": "ak",
+        "access_key_secret": "sk",
+        "instance_name": "instance",
+        "endpoint": "endpoint",
+        "normalize_full_text_bm25_score": False,
+    }
+    values.update(overrides)
+    return module.TableStoreConfig.model_validate(values)
+
+
+@pytest.mark.parametrize(
+    ("field", "value", "message"),
+    [
+        ("access_key_id", "", "config ACCESS_KEY_ID is required"),
+        ("access_key_secret", "", "config ACCESS_KEY_SECRET is required"),
+        ("instance_name", "", "config INSTANCE_NAME is required"),
+        ("endpoint", "", "config ENDPOINT is required"),
+    ],
+)
+def test_tablestore_config_validation(tablestore_module, field, value, message):
+    values = _config(tablestore_module).model_dump()
+    values[field] = value
+    with pytest.raises(ValidationError, match=message):
+        tablestore_module.TableStoreConfig.model_validate(values)
+
+
+def test_init_and_basic_delegation(tablestore_module):
+    vector = tablestore_module.TableStoreVector("collection_1", _config(tablestore_module))
+    assert vector.get_type() == tablestore_module.VectorType.TABLESTORE
+    assert vector._table_name == "collection_1"
+    assert vector._index_name == "collection_1_idx"
+
+    vector._create_collection = MagicMock()
+    vector.add_texts = MagicMock()
+    docs = [Document(page_content="hello", metadata={"doc_id": "d-1"})]
+    vector.create(docs, [[0.1, 0.2]])
+    vector._create_collection.assert_called_once_with(2)
+    vector.add_texts.assert_called_once_with(documents=docs, embeddings=[[0.1, 0.2]])
+
+    vector.create_collection([[0.1, 0.2]])
+    assert vector._create_collection.call_count == 2
+
+
+def test_get_by_ids_text_exists_delete_and_wrappers(tablestore_module):
+    vector = tablestore_module.TableStoreVector("collection_1", _config(tablestore_module))
+
+    # get_by_ids
+    ok_item = SimpleNamespace(
+        is_ok=True,
+        row=SimpleNamespace(
+            attribute_columns=[("metadata", json.dumps({"doc_id": "1"}), None), ("page_content", "text-1", None)]
+        ),
+    )
+    fail_item = SimpleNamespace(is_ok=False, row=None)
+    batch_resp = SimpleNamespace(get_result_by_table=lambda _table: [ok_item, fail_item])
+    vector._tablestore_client.batch_get_row.return_value = batch_resp
+    docs = vector.get_by_ids(["id-1"])
+    assert len(docs) == 1
+    assert docs[0].page_content == "text-1"
+
+    # text_exists
+    vector._tablestore_client.get_row.return_value = (None, object(), None)
+    assert vector.text_exists("id-1") is True
+    vector._tablestore_client.get_row.return_value = (None, None, None)
+    assert vector.text_exists("id-1") is False
+
+    # delete wrappers
+    vector._delete_row = MagicMock()
+    vector.delete_by_ids([])
+    vector._delete_row.assert_not_called()
+    vector.delete_by_ids(["id-1", "id-2"])
+    assert vector._delete_row.call_count == 2
+
+    vector._search_by_metadata = MagicMock(return_value=["id-a"])
+    assert vector.get_ids_by_metadata_field("document_id", "doc-1") == ["id-a"]
+    vector.delete_by_ids = MagicMock()
+    vector.delete_by_metadata_field("document_id", "doc-1")
+    vector.delete_by_ids.assert_called_once_with(["id-a"])
+
+    vector._search_by_vector = MagicMock(return_value=["vec-doc"])
+    vector._search_by_full_text = MagicMock(return_value=["fts-doc"])
+    assert vector.search_by_vector([0.1], top_k=2, score_threshold=0.5, document_ids_filter=["d-1"]) == ["vec-doc"]
+    assert vector.search_by_full_text("query", top_k=2, score_threshold=0.3, document_ids_filter=["d-1"]) == ["fts-doc"]
+
+    vector._delete_table_if_exist = MagicMock()
+    vector.delete()
+    vector._delete_table_if_exist.assert_called_once()
+
+
+def test_create_collection_and_table_index_lifecycle(tablestore_module, monkeypatch):
+    vector = tablestore_module.TableStoreVector("collection_1", _config(tablestore_module))
+    lock = MagicMock()
+    lock.__enter__.return_value = None
+    lock.__exit__.return_value = None
+    monkeypatch.setattr(tablestore_module.redis_client, "lock", MagicMock(return_value=lock))
+    monkeypatch.setattr(tablestore_module.redis_client, "set", MagicMock())
+
+    monkeypatch.setattr(tablestore_module.redis_client, "get", MagicMock(return_value=1))
+    vector._create_table_if_not_exist = MagicMock()
+    vector._create_search_index_if_not_exist = MagicMock()
+    vector._create_collection(3)
+    vector._create_table_if_not_exist.assert_not_called()
+
+    monkeypatch.setattr(tablestore_module.redis_client, "get", MagicMock(return_value=None))
+    vector._create_collection(3)
+    vector._create_table_if_not_exist.assert_called_once()
+    vector._create_search_index_if_not_exist.assert_called_once_with(3)
+    tablestore_module.redis_client.set.assert_called_once()
+
+    vector = tablestore_module.TableStoreVector("collection_2", _config(tablestore_module))
+    vector._tablestore_client.list_table.return_value = ["collection_2"]
+    assert vector._create_table_if_not_exist() is None
+    vector._tablestore_client.list_table.return_value = []
+    vector._create_table_if_not_exist()
+    vector._tablestore_client.create_table.assert_called_once()
+
+    vector._tablestore_client.list_search_index.return_value = [("collection_2", "collection_2_idx")]
+    assert vector._create_search_index_if_not_exist(3) is None
+    vector._tablestore_client.list_search_index.return_value = []
+    vector._create_search_index_if_not_exist(3)
+    vector._tablestore_client.create_search_index.assert_called_once()
+
+    vector._tablestore_client.list_search_index.return_value = [("collection_2", "idx_a"), ("collection_2", "idx_b")]
+    vector._delete_table_if_exist()
+    assert vector._tablestore_client.delete_search_index.call_count == 2
+    vector._tablestore_client.delete_table.assert_called_once_with("collection_2")
+
+    vector._delete_search_index()
+    vector._tablestore_client.delete_search_index.assert_called_with("collection_2", "collection_2_idx")
+
+
+def test_write_row_and_search_helpers(tablestore_module):
+    vector = tablestore_module.TableStoreVector("collection_1", _config(tablestore_module))
+
+    vector._write_row(
+        "id-1",
+        {
+            "page_content": "hello",
+            "vector": [0.1, 0.2],
+            "metadata": {"doc_id": "d-1", "document_id": "doc-1"},
+        },
+    )
+    put_row_call = vector._tablestore_client.put_row.call_args
+    assert put_row_call.args[0] == "collection_1"
+    attrs = put_row_call.args[1].attribute_columns
+    assert any(item[0] == "metadata_tags" for item in attrs)
+
+    vector._delete_row("id-1")
+    vector._tablestore_client.delete_row.assert_called_once()
+
+    # metadata search pagination
+    first_page = SimpleNamespace(rows=[[(("id", "row-1"),)]], next_token=b"next")
+    second_page = SimpleNamespace(rows=[[(("id", "row-2"),)]], next_token=b"")
+    vector._tablestore_client.search.side_effect = [first_page, second_page]
+    ids = vector._search_by_metadata("document_id", "doc-1")
+    assert ids == ["row-1", "row-2"]
+    vector._tablestore_client.search.side_effect = None
+
+    # vector search
+    hit1 = SimpleNamespace(
+        score=0.9,
+        row=(
+            None,
+            [("page_content", "doc-a"), ("metadata", json.dumps({"doc_id": "1"})), ("vector", json.dumps([0.1]))],
+        ),
+    )
+    hit2 = SimpleNamespace(
+        score=0.2,
+        row=(
+            None,
+            [("page_content", "doc-b"), ("metadata", json.dumps({"doc_id": "2"})), ("vector", json.dumps([0.2]))],
+        ),
+    )
+    vector._tablestore_client.search.return_value = SimpleNamespace(search_hits=[hit1, hit2])
+    docs = vector._search_by_vector([0.1], document_ids_filter=["document_id=doc-1"], top_k=2, score_threshold=0.5)
+    assert len(docs) == 1
+    assert docs[0].metadata["score"] == pytest.approx(0.9)
+
+    assert tablestore_module.TableStoreVector._normalize_score_exp_decay(0) == pytest.approx(0.0)
+    assert tablestore_module.TableStoreVector._normalize_score_exp_decay(100) <= 1.0
+
+    # full text search with and without normalized score filter
+    vector._normalize_full_text_bm25_score = True
+    hit3 = SimpleNamespace(
+        score=10.0, row=(None, [("page_content", "doc-c"), ("metadata", json.dumps({"doc_id": "3"}))])
+    )
+    hit4 = SimpleNamespace(
+        score=0.1, row=(None, [("page_content", "doc-d"), ("metadata", json.dumps({"doc_id": "4"}))])
+    )
+    vector._tablestore_client.search.return_value = SimpleNamespace(search_hits=[hit3, hit4])
+    docs = vector._search_by_full_text("query", document_ids_filter=["document_id=doc-1"], top_k=2, score_threshold=0.2)
+    assert len(docs) == 1
+    assert "score" in docs[0].metadata
+
+    vector._normalize_full_text_bm25_score = False
+    vector._tablestore_client.search.return_value = SimpleNamespace(search_hits=[hit3])
+    docs = vector._search_by_full_text("query", document_ids_filter=None, top_k=2, score_threshold=0.0)
+    assert len(docs) == 1
+    assert "score" not in docs[0].metadata
+
+
+def test_tablestore_factory_uses_existing_or_generated_collection(tablestore_module, monkeypatch):
+    factory = tablestore_module.TableStoreVectorFactory()
+    dataset_with_index = SimpleNamespace(
+        id="dataset-1",
+        index_struct_dict={"vector_store": {"class_prefix": "EXISTING_COLLECTION"}},
+        index_struct=None,
+    )
+    dataset_without_index = SimpleNamespace(id="dataset-2", index_struct_dict=None, index_struct=None)
+
+    monkeypatch.setattr(tablestore_module.Dataset, "gen_collection_name_by_id", lambda _id: "AUTO_COLLECTION")
+    monkeypatch.setattr(tablestore_module.dify_config, "TABLESTORE_ENDPOINT", "endpoint")
+    monkeypatch.setattr(tablestore_module.dify_config, "TABLESTORE_INSTANCE_NAME", "instance")
+    monkeypatch.setattr(tablestore_module.dify_config, "TABLESTORE_ACCESS_KEY_ID", "ak")
+    monkeypatch.setattr(tablestore_module.dify_config, "TABLESTORE_ACCESS_KEY_SECRET", "sk")
+    monkeypatch.setattr(tablestore_module.dify_config, "TABLESTORE_NORMALIZE_FULLTEXT_BM25_SCORE", True)
+
+    with patch.object(tablestore_module, "TableStoreVector", return_value="vector") as vector_cls:
+        result_1 = factory.init_vector(dataset_with_index, attributes=[], embeddings=MagicMock())
+        result_2 = factory.init_vector(dataset_without_index, attributes=[], embeddings=MagicMock())
+
+    assert result_1 == "vector"
+    assert result_2 == "vector"
+    assert vector_cls.call_args_list[0].kwargs["collection_name"] == "EXISTING_COLLECTION"
+    assert vector_cls.call_args_list[1].kwargs["collection_name"] == "AUTO_COLLECTION"
+    assert dataset_without_index.index_struct is not None

api/tests/unit_tests/core/rag/datasource/vdb/tencent/test_tencent_vector.py

@@ -0,0 +1,309 @@
+import importlib
+import sys
+import types
+from types import SimpleNamespace
+from unittest.mock import MagicMock, patch
+
+import pytest
+
+from core.rag.models.document import Document
+
+
+def _build_fake_tencent_modules():
+    tcvdb_text = types.ModuleType("tcvdb_text")
+    tcvdb_text_encoder = types.ModuleType("tcvdb_text.encoder")
+    tcvectordb = types.ModuleType("tcvectordb")
+    tcvectordb_model = types.ModuleType("tcvectordb.model")
+    tcvectordb_document = types.ModuleType("tcvectordb.model.document")
+    tcvectordb_index = types.ModuleType("tcvectordb.model.index")
+    tcvectordb_enum = types.ModuleType("tcvectordb.model.enum")
+
+    class _BM25Encoder:
+        def encode_texts(self, text):
+            return {"encoded_text": text}
+
+        def encode_queries(self, query):
+            return {"encoded_query": query}
+
+        @classmethod
+        def default(cls, _lang):
+            return cls()
+
+    class VectorDBError(Exception):
+        def __init__(self, message):
+            super().__init__(message)
+            self.message = message
+
+    class RPCVectorDBClient:
+        def __init__(self, **kwargs):
+            self.kwargs = kwargs
+            self.create_database_if_not_exists = MagicMock()
+            self.exists_collection = MagicMock(return_value=False)
+            self.describe_collection = MagicMock(return_value=SimpleNamespace(indexes=[]))
+            self.create_collection = MagicMock()
+            self.upsert = MagicMock()
+            self.query = MagicMock(return_value=[])
+            self.delete = MagicMock()
+            self.search = MagicMock(return_value=[])
+            self.hybrid_search = MagicMock(return_value=[])
+            self.drop_collection = MagicMock()
+
+    class _Document:
+        def __init__(self, **kwargs):
+            self.__dict__.update(kwargs)
+
+    class _HNSWSearchParams:
+        def __init__(self, ef):
+            self.ef = ef
+
+    class _AnnSearch:
+        def __init__(self, **kwargs):
+            self.kwargs = kwargs
+
+    class _KeywordSearch:
+        def __init__(self, **kwargs):
+            self.kwargs = kwargs
+
+    class _WeightedRerank:
+        def __init__(self, **kwargs):
+            self.kwargs = kwargs
+
+    class _Filter:
+        @staticmethod
+        def in_(field, values):
+            return ("in", field, values)
+
+        def __init__(self, condition):
+            self.condition = condition
+
+    _Filter.In = staticmethod(_Filter.in_)
+
+    class _HNSWParams:
+        def __init__(self, **kwargs):
+            self.kwargs = kwargs
+
+    class _FilterIndex:
+        def __init__(self, *args):
+            self.args = args
+
+    class _VectorIndex:
+        def __init__(self, *args):
+            self.args = args
+
+    class _SparseIndex:
+        def __init__(self, **kwargs):
+            self.kwargs = kwargs
+
+    tcvectordb_enum.IndexType = SimpleNamespace(
+        __members__={"HNSW": "HNSW", "PRIMARY_KEY": "PRIMARY_KEY", "FILTER": "FILTER", "SPARSE_INVERTED": "SPARSE"},
+        PRIMARY_KEY="PRIMARY_KEY",
+        FILTER="FILTER",
+        SPARSE_INVERTED="SPARSE",
+    )
+    tcvectordb_enum.MetricType = SimpleNamespace(__members__={"IP": "IP"}, IP="IP")
+    tcvectordb_enum.FieldType = SimpleNamespace(String="String", Json="Json", SparseVector="SparseVector")
+
+    tcvectordb_document.Document = _Document
+    tcvectordb_document.HNSWSearchParams = _HNSWSearchParams
+    tcvectordb_document.AnnSearch = _AnnSearch
+    tcvectordb_document.Filter = _Filter
+    tcvectordb_document.KeywordSearch = _KeywordSearch
+    tcvectordb_document.WeightedRerank = _WeightedRerank
+
+    tcvectordb_index.HNSWParams = _HNSWParams
+    tcvectordb_index.FilterIndex = _FilterIndex
+    tcvectordb_index.VectorIndex = _VectorIndex
+    tcvectordb_index.SparseIndex = _SparseIndex
+
+    tcvdb_text_encoder.BM25Encoder = _BM25Encoder
+
+    tcvectordb_model.document = tcvectordb_document
+    tcvectordb_model.enum = tcvectordb_enum
+    tcvectordb_model.index = tcvectordb_index
+
+    tcvectordb.RPCVectorDBClient = RPCVectorDBClient
+    tcvectordb.VectorDBException = VectorDBError
+
+    return {
+        "tcvdb_text": tcvdb_text,
+        "tcvdb_text.encoder": tcvdb_text_encoder,
+        "tcvectordb": tcvectordb,
+        "tcvectordb.model": tcvectordb_model,
+        "tcvectordb.model.document": tcvectordb_document,
+        "tcvectordb.model.index": tcvectordb_index,
+        "tcvectordb.model.enum": tcvectordb_enum,
+    }
+
+
+@pytest.fixture
+def tencent_module(monkeypatch):
+    for name, module in _build_fake_tencent_modules().items():
+        monkeypatch.setitem(sys.modules, name, module)
+
+    import core.rag.datasource.vdb.tencent.tencent_vector as module
+
+    return importlib.reload(module)
+
+
+def _config(module, **overrides):
+    values = {
+        "url": "http://vdb.local",
+        "api_key": "api-key",
+        "timeout": 30,
+        "username": "user",
+        "database": "db",
+        "index_type": "HNSW",
+        "metric_type": "IP",
+        "shard": 1,
+        "replicas": 2,
+        "max_upsert_batch_size": 2,
+        "enable_hybrid_search": False,
+    }
+    values.update(overrides)
+    return module.TencentConfig.model_validate(values)
+
+
+def test_config_and_init_paths(tencent_module):
+    config = _config(tencent_module)
+    assert config.to_tencent_params()["url"] == "http://vdb.local"
+
+    vector = tencent_module.TencentVector("collection_1", config)
+    assert vector.get_type() == tencent_module.VectorType.TENCENT
+    assert vector._client.kwargs["key"] == "api-key"
+
+    vector._client.exists_collection.return_value = True
+    vector._client.describe_collection.return_value = SimpleNamespace(
+        indexes=[SimpleNamespace(name="vector", dimension=768), SimpleNamespace(name="sparse_vector", dimension=0)]
+    )
+    vector._client_config.enable_hybrid_search = True
+    vector._load_collection()
+    assert vector._enable_hybrid_search is True
+    assert vector._dimension == 768
+
+    vector._client.describe_collection.return_value = SimpleNamespace(
+        indexes=[SimpleNamespace(name="vector", dimension=512)]
+    )
+    vector._load_collection()
+    assert vector._enable_hybrid_search is False
+
+
+def test_create_collection_branches(tencent_module, monkeypatch):
+    vector = tencent_module.TencentVector("collection_1", _config(tencent_module))
+
+    lock = MagicMock()
+    lock.__enter__.return_value = None
+    lock.__exit__.return_value = None
+    monkeypatch.setattr(tencent_module.redis_client, "lock", MagicMock(return_value=lock))
+    monkeypatch.setattr(tencent_module.redis_client, "set", MagicMock())
+
+    monkeypatch.setattr(tencent_module.redis_client, "get", MagicMock(return_value=1))
+    vector._create_collection(3)
+    vector._client.create_collection.assert_not_called()
+
+    monkeypatch.setattr(tencent_module.redis_client, "get", MagicMock(return_value=None))
+    vector._client.exists_collection.return_value = True
+    vector._create_collection(3)
+    vector._client.create_collection.assert_not_called()
+
+    vector._client.exists_collection.return_value = False
+    vector._client_config.index_type = "UNKNOWN"
+    with pytest.raises(ValueError, match="unsupported index_type"):
+        vector._create_collection(3)
+
+    vector._client_config.index_type = "HNSW"
+    vector._client_config.metric_type = "UNKNOWN"
+    with pytest.raises(ValueError, match="unsupported metric_type"):
+        vector._create_collection(3)
+
+    vector._client_config.metric_type = "IP"
+    vector._client.create_collection.side_effect = [
+        tencent_module.VectorDBException("fieldType:json unsupported"),
+        None,
+    ]
+    vector._enable_hybrid_search = True
+    vector._create_collection(3)
+    assert vector._client.create_collection.call_count == 2
+    tencent_module.redis_client.set.assert_called_once()
+    vector._client.create_collection.side_effect = None
+
+
+def test_create_add_delete_and_search_behaviour(tencent_module):
+    vector = tencent_module.TencentVector("collection_1", _config(tencent_module, enable_hybrid_search=True))
+    vector._create_collection = MagicMock()
+    docs = [
+        Document(page_content="text-a", metadata={"doc_id": "a", "document_id": "doc-a"}),
+        Document(page_content="text-b", metadata={"doc_id": "b", "document_id": "doc-b"}),
+        Document(page_content="text-c", metadata={"doc_id": "c", "document_id": "doc-c"}),
+    ]
+    embeddings = [[0.1], [0.2], [0.3]]
+    vector.create(docs, embeddings)
+    vector._create_collection.assert_called_once_with(1)
+
+    vector._client.upsert.reset_mock()
+    vector.add_texts(docs, embeddings)
+    assert vector._client.upsert.call_count == 2
+    first_docs = vector._client.upsert.call_args_list[0].kwargs["documents"]
+    assert "sparse_vector" in first_docs[0].__dict__
+
+    vector._client.query.return_value = [{"id": "a"}]
+    assert vector.text_exists("a") is True
+    vector._client.query.return_value = []
+    assert vector.text_exists("a") is False
+
+    vector.delete_by_ids([])
+    vector._client.delete.assert_not_called()
+    vector.delete_by_ids(["a", "b", "c"])
+    assert vector._client.delete.call_count == 2
+    vector.delete_by_metadata_field("document_id", "doc-a")
+    assert vector._client.delete.call_count >= 3
+
+    vector._client.search.return_value = [[{"metadata": {"doc_id": "1"}, "text": "vec-doc", "score": 0.9}]]
+    vec_docs = vector.search_by_vector([0.1], top_k=2, score_threshold=0.5, document_ids_filter=["doc-a"])
+    assert len(vec_docs) == 1
+    assert vec_docs[0].metadata["score"] == pytest.approx(0.9)
+
+    vector._enable_hybrid_search = False
+    assert vector.search_by_full_text("query") == []
+    vector._enable_hybrid_search = True
+    vector._client.hybrid_search.return_value = [[{"metadata": {"doc_id": "2"}, "text": "fts-doc", "score": 0.8}]]
+    fts_docs = vector.search_by_full_text("query", top_k=2, score_threshold=0.5, document_ids_filter=["doc-a"])
+    assert len(fts_docs) == 1
+
+    # _get_search_res handles old string metadata format
+    compat_docs = vector._get_search_res([[{"metadata": '{"doc_id": "3"}', "text": "compat", "score": 0.2}]], 0.5)
+    assert len(compat_docs) == 1
+    assert compat_docs[0].metadata["score"] == pytest.approx(0.8)
+
+    vector._has_collection = MagicMock(return_value=True)
+    vector.delete()
+    vector._client.drop_collection.assert_called_once()
+
+
+def test_tencent_factory_existing_and_generated_collection(tencent_module, monkeypatch):
+    factory = tencent_module.TencentVectorFactory()
+    dataset_with_index = SimpleNamespace(
+        id="dataset-1",
+        index_struct_dict={"vector_store": {"class_prefix": "EXISTING_COLLECTION"}},
+        index_struct=None,
+    )
+    dataset_without_index = SimpleNamespace(id="dataset-2", index_struct_dict=None, index_struct=None)
+
+    monkeypatch.setattr(tencent_module.Dataset, "gen_collection_name_by_id", lambda _id: "AUTO_COLLECTION")
+    monkeypatch.setattr(tencent_module.dify_config, "TENCENT_VECTOR_DB_URL", "http://vdb.local")
+    monkeypatch.setattr(tencent_module.dify_config, "TENCENT_VECTOR_DB_API_KEY", "api-key")
+    monkeypatch.setattr(tencent_module.dify_config, "TENCENT_VECTOR_DB_TIMEOUT", 30)
+    monkeypatch.setattr(tencent_module.dify_config, "TENCENT_VECTOR_DB_USERNAME", "user")
+    monkeypatch.setattr(tencent_module.dify_config, "TENCENT_VECTOR_DB_DATABASE", "db")
+    monkeypatch.setattr(tencent_module.dify_config, "TENCENT_VECTOR_DB_SHARD", 1)
+    monkeypatch.setattr(tencent_module.dify_config, "TENCENT_VECTOR_DB_REPLICAS", 2)
+    monkeypatch.setattr(tencent_module.dify_config, "TENCENT_VECTOR_DB_ENABLE_HYBRID_SEARCH", True)
+
+    with patch.object(tencent_module, "TencentVector", return_value="vector") as vector_cls:
+        result_1 = factory.init_vector(dataset_with_index, attributes=[], embeddings=MagicMock())
+        result_2 = factory.init_vector(dataset_without_index, attributes=[], embeddings=MagicMock())
+
+    assert result_1 == "vector"
+    assert result_2 == "vector"
+    assert vector_cls.call_args_list[0].kwargs["collection_name"] == "existing_collection"
+    assert vector_cls.call_args_list[1].kwargs["collection_name"] == "auto_collection"
+    assert dataset_without_index.index_struct is not None

api/tests/unit_tests/core/rag/datasource/vdb/test_vector_base.py

@@ -0,0 +1,88 @@
+from types import SimpleNamespace
+
+import pytest
+
+from core.rag.datasource.vdb.vector_base import BaseVector
+from core.rag.models.document import Document
+
+
+class _DummyVector(BaseVector):
+    def __init__(self, collection_name: str, existing_ids: set[str] | None = None):
+        super().__init__(collection_name)
+        self._existing_ids = existing_ids or set()
+
+    def get_type(self) -> str:
+        return "dummy"
+
+    def create(self, texts: list[Document], embeddings: list[list[float]], **kwargs):
+        return None
+
+    def add_texts(self, documents: list[Document], embeddings: list[list[float]], **kwargs):
+        return None
+
+    def text_exists(self, id: str) -> bool:
+        return id in self._existing_ids
+
+    def delete_by_ids(self, ids: list[str]):
+        return None
+
+    def delete_by_metadata_field(self, key: str, value: str):
+        return None
+
+    def search_by_vector(self, query_vector: list[float], **kwargs):
+        return []
+
+    def search_by_full_text(self, query: str, **kwargs):
+        return []
+
+    def delete(self):
+        return None
+
+
+@pytest.mark.parametrize(
+    ("base_method", "args"),
+    [
+        (BaseVector.get_type, ()),
+        (BaseVector.create, ([], [])),
+        (BaseVector.add_texts, ([], [])),
+        (BaseVector.text_exists, ("doc-1",)),
+        (BaseVector.delete_by_ids, ([],)),
+        (BaseVector.get_ids_by_metadata_field, ("doc_id", "doc-1")),
+        (BaseVector.delete_by_metadata_field, ("doc_id", "doc-1")),
+        (BaseVector.search_by_vector, ([0.1],)),
+        (BaseVector.search_by_full_text, ("query",)),
+        (BaseVector.delete, ()),
+    ],
+)
+def test_base_vector_default_methods_raise_not_implemented(base_method, args):
+    vector = _DummyVector("collection_1")
+
+    with pytest.raises(NotImplementedError):
+        base_method(vector, *args)
+
+
+def test_filter_duplicate_texts_removes_existing_docs():
+    vector = _DummyVector("collection_1", existing_ids={"dup"})
+    docs = [
+        SimpleNamespace(page_content="keep-no-meta", metadata=None),
+        Document(page_content="keep-no-doc-id", metadata={"document_id": "d1"}),
+        Document(page_content="remove-dup", metadata={"doc_id": "dup"}),
+        Document(page_content="keep-unique", metadata={"doc_id": "unique"}),
+    ]
+
+    filtered = vector._filter_duplicate_texts(docs)
+
+    assert [d.page_content for d in filtered] == ["keep-no-meta", "keep-no-doc-id", "keep-unique"]
+
+
+def test_get_uuids_and_collection_name_property():
+    vector = _DummyVector("collection_1")
+    docs = [
+        Document(page_content="a", metadata={"doc_id": "id-1"}),
+        SimpleNamespace(page_content="b", metadata=None),
+        Document(page_content="c", metadata={"document_id": "d-1"}),
+        Document(page_content="d", metadata={"doc_id": "id-2"}),
+    ]
+
+    assert vector._get_uuids(docs) == ["id-1", "id-2"]
+    assert vector.collection_name == "collection_1"

api/tests/unit_tests/core/rag/datasource/vdb/test_vector_factory.py

@@ -0,0 +1,434 @@
+import base64
+import sys
+import types
+from types import SimpleNamespace
+from unittest.mock import MagicMock, patch
+
+import pytest
+
+from core.rag.models.document import Document
+
+
+def _register_fake_factory_module(monkeypatch, module_path: str, class_name: str):
+    fake_module = types.ModuleType(module_path)
+    fake_cls = type(class_name, (), {})
+    setattr(fake_module, class_name, fake_cls)
+    monkeypatch.setitem(sys.modules, module_path, fake_module)
+    return fake_cls
+
+
+@pytest.fixture
+def vector_factory_module():
+    import importlib
+
+    import core.rag.datasource.vdb.vector_factory as module
+
+    return importlib.reload(module)
+
+
+def test_gen_index_struct_dict(vector_factory_module):
+    result = vector_factory_module.AbstractVectorFactory.gen_index_struct_dict(
+        vector_factory_module.VectorType.WEAVIATE,
+        "collection_1",
+    )
+
+    assert result == {
+        "type": vector_factory_module.VectorType.WEAVIATE,
+        "vector_store": {"class_prefix": "collection_1"},
+    }
+
+
+@pytest.mark.parametrize(
+    ("vector_type", "module_path", "class_name"),
+    [
+        ("CHROMA", "core.rag.datasource.vdb.chroma.chroma_vector", "ChromaVectorFactory"),
+        ("MILVUS", "core.rag.datasource.vdb.milvus.milvus_vector", "MilvusVectorFactory"),
+        (
+            "ALIBABACLOUD_MYSQL",
+            "core.rag.datasource.vdb.alibabacloud_mysql.alibabacloud_mysql_vector",
+            "AlibabaCloudMySQLVectorFactory",
+        ),
+        ("MYSCALE", "core.rag.datasource.vdb.myscale.myscale_vector", "MyScaleVectorFactory"),
+        ("PGVECTOR", "core.rag.datasource.vdb.pgvector.pgvector", "PGVectorFactory"),
+        ("VASTBASE", "core.rag.datasource.vdb.pyvastbase.vastbase_vector", "VastbaseVectorFactory"),
+        ("PGVECTO_RS", "core.rag.datasource.vdb.pgvecto_rs.pgvecto_rs", "PGVectoRSFactory"),
+        ("QDRANT", "core.rag.datasource.vdb.qdrant.qdrant_vector", "QdrantVectorFactory"),
+        ("RELYT", "core.rag.datasource.vdb.relyt.relyt_vector", "RelytVectorFactory"),
+        (
+            "ELASTICSEARCH",
+            "core.rag.datasource.vdb.elasticsearch.elasticsearch_vector",
+            "ElasticSearchVectorFactory",
+        ),
+        (
+            "ELASTICSEARCH_JA",
+            "core.rag.datasource.vdb.elasticsearch.elasticsearch_ja_vector",
+            "ElasticSearchJaVectorFactory",
+        ),
+        ("TIDB_VECTOR", "core.rag.datasource.vdb.tidb_vector.tidb_vector", "TiDBVectorFactory"),
+        ("WEAVIATE", "core.rag.datasource.vdb.weaviate.weaviate_vector", "WeaviateVectorFactory"),
+        ("TENCENT", "core.rag.datasource.vdb.tencent.tencent_vector", "TencentVectorFactory"),
+        ("ORACLE", "core.rag.datasource.vdb.oracle.oraclevector", "OracleVectorFactory"),
+        (
+            "OPENSEARCH",
+            "core.rag.datasource.vdb.opensearch.opensearch_vector",
+            "OpenSearchVectorFactory",
+        ),
+        ("ANALYTICDB", "core.rag.datasource.vdb.analyticdb.analyticdb_vector", "AnalyticdbVectorFactory"),
+        ("COUCHBASE", "core.rag.datasource.vdb.couchbase.couchbase_vector", "CouchbaseVectorFactory"),
+        ("BAIDU", "core.rag.datasource.vdb.baidu.baidu_vector", "BaiduVectorFactory"),
+        ("VIKINGDB", "core.rag.datasource.vdb.vikingdb.vikingdb_vector", "VikingDBVectorFactory"),
+        ("UPSTASH", "core.rag.datasource.vdb.upstash.upstash_vector", "UpstashVectorFactory"),
+        (
+            "TIDB_ON_QDRANT",
+            "core.rag.datasource.vdb.tidb_on_qdrant.tidb_on_qdrant_vector",
+            "TidbOnQdrantVectorFactory",
+        ),
+        ("LINDORM", "core.rag.datasource.vdb.lindorm.lindorm_vector", "LindormVectorStoreFactory"),
+        ("OCEANBASE", "core.rag.datasource.vdb.oceanbase.oceanbase_vector", "OceanBaseVectorFactory"),
+        ("SEEKDB", "core.rag.datasource.vdb.oceanbase.oceanbase_vector", "OceanBaseVectorFactory"),
+        ("OPENGAUSS", "core.rag.datasource.vdb.opengauss.opengauss", "OpenGaussFactory"),
+        ("TABLESTORE", "core.rag.datasource.vdb.tablestore.tablestore_vector", "TableStoreVectorFactory"),
+        (
+            "HUAWEI_CLOUD",
+            "core.rag.datasource.vdb.huawei.huawei_cloud_vector",
+            "HuaweiCloudVectorFactory",
+        ),
+        ("MATRIXONE", "core.rag.datasource.vdb.matrixone.matrixone_vector", "MatrixoneVectorFactory"),
+        ("CLICKZETTA", "core.rag.datasource.vdb.clickzetta.clickzetta_vector", "ClickzettaVectorFactory"),
+        ("IRIS", "core.rag.datasource.vdb.iris.iris_vector", "IrisVectorFactory"),
+    ],
+)
+def test_get_vector_factory_supported(vector_factory_module, monkeypatch, vector_type, module_path, class_name):
+    expected_cls = _register_fake_factory_module(monkeypatch, module_path, class_name)
+
+    result_cls = vector_factory_module.Vector.get_vector_factory(getattr(vector_factory_module.VectorType, vector_type))
+
+    assert result_cls is expected_cls
+
+
+def test_get_vector_factory_unsupported(vector_factory_module):
+    with pytest.raises(ValueError, match="not supported"):
+        vector_factory_module.Vector.get_vector_factory("unknown")
+
+
+def test_vector_init_uses_default_and_custom_attributes(vector_factory_module):
+    dataset = SimpleNamespace(id="dataset-1")
+
+    with (
+        patch.object(vector_factory_module.Vector, "_get_embeddings", return_value="embeddings"),
+        patch.object(vector_factory_module.Vector, "_init_vector", return_value="processor"),
+    ):
+        default_vector = vector_factory_module.Vector(dataset)
+        custom_vector = vector_factory_module.Vector(dataset, attributes=["doc_id"])
+
+    assert default_vector._attributes == ["doc_id", "dataset_id", "document_id", "doc_hash", "doc_type"]
+    assert custom_vector._attributes == ["doc_id"]
+    assert default_vector._embeddings == "embeddings"
+    assert default_vector._vector_processor == "processor"
+
+
+def test_init_vector_prefers_dataset_index_struct(vector_factory_module, monkeypatch):
+    calls = {"vector_type": None, "init_args": None}
+
+    class _Factory:
+        def init_vector(self, dataset, attributes, embeddings):
+            calls["init_args"] = (dataset, attributes, embeddings)
+            return "vector-processor"
+
+    monkeypatch.setattr(
+        vector_factory_module.Vector,
+        "get_vector_factory",
+        staticmethod(lambda vector_type: calls.update(vector_type=vector_type) or _Factory),
+    )
+
+    vector = vector_factory_module.Vector.__new__(vector_factory_module.Vector)
+    vector._dataset = SimpleNamespace(
+        index_struct_dict={"type": vector_factory_module.VectorType.UPSTASH}, tenant_id="tenant-1"
+    )
+    vector._attributes = ["doc_id"]
+    vector._embeddings = "embeddings"
+
+    result = vector._init_vector()
+
+    assert result == "vector-processor"
+    assert calls["vector_type"] == vector_factory_module.VectorType.UPSTASH
+    assert calls["init_args"] == (vector._dataset, ["doc_id"], "embeddings")
+
+
+def test_init_vector_uses_whitelist_override(vector_factory_module, monkeypatch):
+    class _Expr:
+        def __eq__(self, _other):
+            return "expr"
+
+    calls = {"vector_type": None}
+
+    class _Factory:
+        def init_vector(self, dataset, attributes, embeddings):
+            return "vector-processor"
+
+    monkeypatch.setattr(vector_factory_module, "Whitelist", SimpleNamespace(tenant_id=_Expr(), category=_Expr()))
+    monkeypatch.setattr(vector_factory_module, "select", lambda _model: SimpleNamespace(where=lambda *_args: "stmt"))
+    monkeypatch.setattr(
+        vector_factory_module,
+        "db",
+        SimpleNamespace(session=SimpleNamespace(scalars=lambda _stmt: SimpleNamespace(one_or_none=lambda: object()))),
+    )
+    monkeypatch.setattr(vector_factory_module.dify_config, "VECTOR_STORE", vector_factory_module.VectorType.CHROMA)
+    monkeypatch.setattr(vector_factory_module.dify_config, "VECTOR_STORE_WHITELIST_ENABLE", True)
+    monkeypatch.setattr(
+        vector_factory_module.Vector,
+        "get_vector_factory",
+        staticmethod(lambda vector_type: calls.update(vector_type=vector_type) or _Factory),
+    )
+
+    vector = vector_factory_module.Vector.__new__(vector_factory_module.Vector)
+    vector._dataset = SimpleNamespace(index_struct_dict=None, tenant_id="tenant-1")
+    vector._attributes = ["doc_id"]
+    vector._embeddings = "embeddings"
+
+    result = vector._init_vector()
+
+    assert result == "vector-processor"
+    assert calls["vector_type"] == vector_factory_module.VectorType.TIDB_ON_QDRANT
+
+
+def test_init_vector_raises_when_vector_store_missing(vector_factory_module, monkeypatch):
+    monkeypatch.setattr(vector_factory_module.dify_config, "VECTOR_STORE", None)
+    monkeypatch.setattr(vector_factory_module.dify_config, "VECTOR_STORE_WHITELIST_ENABLE", False)
+
+    vector = vector_factory_module.Vector.__new__(vector_factory_module.Vector)
+    vector._dataset = SimpleNamespace(index_struct_dict=None, tenant_id="tenant-1")
+    vector._attributes = []
+    vector._embeddings = "embeddings"
+
+    with pytest.raises(ValueError, match="Vector store must be specified"):
+        vector._init_vector()
+
+
+def test_create_batches_texts_and_skips_empty_input(vector_factory_module):
+    vector = vector_factory_module.Vector.__new__(vector_factory_module.Vector)
+    vector._embeddings = MagicMock()
+    vector._vector_processor = MagicMock()
+
+    docs = [Document(page_content=f"doc-{i}", metadata={"doc_id": f"id-{i}"}) for i in range(1001)]
+    vector._embeddings.embed_documents.side_effect = [
+        [[0.1] for _ in range(1000)],
+        [[0.2]],
+    ]
+
+    vector.create(texts=docs, trace_id="trace-1")
+
+    assert vector._embeddings.embed_documents.call_count == 2
+    assert vector._vector_processor.create.call_count == 2
+    assert vector._vector_processor.create.call_args_list[0].kwargs["trace_id"] == "trace-1"
+
+    vector._embeddings.embed_documents.reset_mock()
+    vector._vector_processor.create.reset_mock()
+    vector.create(texts=None)
+    vector._embeddings.embed_documents.assert_not_called()
+    vector._vector_processor.create.assert_not_called()
+
+
+def test_create_multimodal_filters_missing_uploads(vector_factory_module, monkeypatch):
+    class _Field:
+        def in_(self, value):
+            return value
+
+        def __eq__(self, value):
+            return value
+
+    vector = vector_factory_module.Vector.__new__(vector_factory_module.Vector)
+    vector._embeddings = MagicMock()
+    vector._embeddings.embed_multimodal_documents.return_value = [[0.1, 0.2]]
+    vector._vector_processor = MagicMock()
+
+    monkeypatch.setattr(vector_factory_module, "UploadFile", SimpleNamespace(id=_Field()))
+    monkeypatch.setattr(vector_factory_module, "select", lambda _model: SimpleNamespace(where=lambda *_args: "stmt"))
+    monkeypatch.setattr(
+        vector_factory_module,
+        "db",
+        SimpleNamespace(
+            session=SimpleNamespace(
+                scalars=lambda _stmt: SimpleNamespace(all=lambda: [SimpleNamespace(id="f-1", key="k-1")])
+            )
+        ),
+    )
+    monkeypatch.setattr(vector_factory_module.storage, "load_once", MagicMock(return_value=b"abc"))
+
+    docs = [
+        Document(page_content="file-1", metadata={"doc_id": "f-1", "doc_type": "image"}),
+        Document(page_content="file-2", metadata={"doc_id": "f-2", "doc_type": "image"}),
+    ]
+
+    vector.create_multimodal(file_documents=docs, request_id="r-1")
+
+    file_base64 = base64.b64encode(b"abc").decode()
+    vector._embeddings.embed_multimodal_documents.assert_called_once_with(
+        [{"content": file_base64, "content_type": "image", "file_id": "f-1"}]
+    )
+    vector._vector_processor.create.assert_called_once_with(
+        texts=[docs[0]],
+        embeddings=[[0.1, 0.2]],
+        request_id="r-1",
+    )
+
+    vector._embeddings.embed_multimodal_documents.reset_mock()
+    vector._vector_processor.create.reset_mock()
+    vector.create_multimodal(file_documents=None)
+    vector._embeddings.embed_multimodal_documents.assert_not_called()
+    vector._vector_processor.create.assert_not_called()
+
+
+def test_add_texts_with_optional_duplicate_check(vector_factory_module):
+    vector = vector_factory_module.Vector.__new__(vector_factory_module.Vector)
+    vector._embeddings = MagicMock()
+    vector._vector_processor = MagicMock()
+    vector._filter_duplicate_texts = MagicMock()
+
+    docs = [
+        Document(page_content="a", metadata={"doc_id": "id-1"}),
+        Document(page_content="b", metadata={"doc_id": "id-2"}),
+    ]
+    vector._filter_duplicate_texts.return_value = [docs[0]]
+    vector._embeddings.embed_documents.return_value = [[0.1]]
+
+    vector.add_texts(docs, duplicate_check=True, flag=True)
+
+    vector._filter_duplicate_texts.assert_called_once_with(docs)
+    vector._vector_processor.create.assert_called_once_with(
+        texts=[docs[0]], embeddings=[[0.1]], duplicate_check=True, flag=True
+    )
+
+    vector._filter_duplicate_texts.reset_mock()
+    vector._vector_processor.create.reset_mock()
+    vector._embeddings.embed_documents.return_value = [[0.2], [0.3]]
+
+    vector.add_texts(docs, duplicate_check=False)
+
+    vector._filter_duplicate_texts.assert_not_called()
+    vector._vector_processor.create.assert_called_once()
+
+
+def test_vector_delegation_methods(vector_factory_module):
+    vector = vector_factory_module.Vector.__new__(vector_factory_module.Vector)
+    vector._embeddings = MagicMock()
+    vector._embeddings.embed_query.return_value = [0.1, 0.2]
+    vector._vector_processor = MagicMock()
+    vector._vector_processor.text_exists.return_value = True
+    vector._vector_processor.search_by_vector.return_value = ["vector-doc"]
+    vector._vector_processor.search_by_full_text.return_value = ["text-doc"]
+
+    assert vector.text_exists("doc-1") is True
+    vector.delete_by_ids(["doc-1"])
+    vector.delete_by_metadata_field("doc_id", "doc-1")
+    assert vector.search_by_vector("hello", top_k=3) == ["vector-doc"]
+    assert vector.search_by_full_text("hello", top_k=3) == ["text-doc"]
+
+    vector._vector_processor.delete_by_ids.assert_called_once_with(["doc-1"])
+    vector._vector_processor.delete_by_metadata_field.assert_called_once_with("doc_id", "doc-1")
+
+
+def test_search_by_file_handles_missing_and_existing_upload(vector_factory_module, monkeypatch):
+    class _Field:
+        def __eq__(self, value):
+            return value
+
+    upload_query = MagicMock()
+    upload_query.where.return_value = upload_query
+
+    vector = vector_factory_module.Vector.__new__(vector_factory_module.Vector)
+    vector._embeddings = MagicMock()
+    vector._vector_processor = MagicMock()
+
+    monkeypatch.setattr(vector_factory_module, "UploadFile", SimpleNamespace(id=_Field()))
+    monkeypatch.setattr(
+        vector_factory_module, "db", SimpleNamespace(session=SimpleNamespace(query=lambda _model: upload_query))
+    )
+
+    upload_query.first.return_value = None
+    assert vector.search_by_file("file-1") == []
+
+    upload_query.first.return_value = SimpleNamespace(key="blob-key")
+    monkeypatch.setattr(vector_factory_module.storage, "load_once", MagicMock(return_value=b"file-bytes"))
+    vector._embeddings.embed_multimodal_query.return_value = [0.3, 0.4]
+    vector._vector_processor.search_by_vector.return_value = ["hit"]
+
+    result = vector.search_by_file("file-2", top_k=2)
+
+    assert result == ["hit"]
+    payload = vector._embeddings.embed_multimodal_query.call_args.args[0]
+    assert payload["content_type"] == vector_factory_module.DocType.IMAGE
+    assert payload["file_id"] == "file-2"
+
+
+def test_delete_clears_redis_cache_when_collection_exists(vector_factory_module, monkeypatch):
+    delete_mock = MagicMock()
+    redis_delete = MagicMock()
+    monkeypatch.setattr(vector_factory_module.redis_client, "delete", redis_delete)
+
+    vector = vector_factory_module.Vector.__new__(vector_factory_module.Vector)
+    vector._vector_processor = SimpleNamespace(delete=delete_mock, collection_name="collection_1")
+
+    vector.delete()
+
+    delete_mock.assert_called_once()
+    redis_delete.assert_called_once_with("vector_indexing_collection_1")
+
+    vector._vector_processor = SimpleNamespace(delete=delete_mock, collection_name="")
+    redis_delete.reset_mock()
+    vector.delete()
+    redis_delete.assert_not_called()
+
+
+def test_get_embeddings_builds_cache_embedding(vector_factory_module, monkeypatch):
+    model_manager = MagicMock()
+    model_manager.get_model_instance.return_value = "model-instance"
+
+    monkeypatch.setattr(vector_factory_module, "ModelManager", MagicMock(return_value=model_manager))
+    monkeypatch.setattr(vector_factory_module, "CacheEmbedding", MagicMock(return_value="cached-embedding"))
+
+    vector = vector_factory_module.Vector.__new__(vector_factory_module.Vector)
+    vector._dataset = SimpleNamespace(
+        tenant_id="tenant-1",
+        embedding_model_provider="openai",
+        embedding_model="text-embedding-3-small",
+    )
+
+    result = vector._get_embeddings()
+
+    assert result == "cached-embedding"
+    model_manager.get_model_instance.assert_called_once_with(
+        tenant_id="tenant-1",
+        provider="openai",
+        model_type=vector_factory_module.ModelType.TEXT_EMBEDDING,
+        model="text-embedding-3-small",
+    )
+
+
+def test_filter_duplicate_texts_and_getattr(vector_factory_module):
+    vector = vector_factory_module.Vector.__new__(vector_factory_module.Vector)
+    vector.text_exists = MagicMock(side_effect=lambda doc_id: doc_id == "dup")
+
+    docs = [
+        SimpleNamespace(page_content="no-meta", metadata=None),
+        Document(page_content="empty-doc-id", metadata={"doc_id": ""}),
+        Document(page_content="duplicate", metadata={"doc_id": "dup"}),
+        Document(page_content="unique", metadata={"doc_id": "ok"}),
+    ]
+
+    filtered = vector._filter_duplicate_texts(docs)
+    assert [doc.page_content for doc in filtered] == ["no-meta", "empty-doc-id", "unique"]
+
+    class _Processor:
+        def ping(self):
+            return "pong"
+
+    vector._vector_processor = _Processor()
+    assert vector.ping() == "pong"
+
+    with pytest.raises(AttributeError):
+        _ = vector.unknown_method
+
+    vector._vector_processor = None
+    with pytest.raises(AttributeError, match="vector_processor"):
+        _ = vector.another_missing

api/tests/unit_tests/core/rag/datasource/vdb/tidb_vector/test_tidb_vector.py

@@ -0,0 +1,443 @@
+import importlib
+import sys
+import types
+from types import SimpleNamespace
+from unittest.mock import MagicMock, patch
+
+import pytest
+from pydantic import ValidationError
+
+from core.rag.models.document import Document
+
+
+@pytest.fixture
+def tidb_module():
+    import core.rag.datasource.vdb.tidb_vector.tidb_vector as module
+
+    return importlib.reload(module)
+
+
+def _config(tidb_module):
+    return tidb_module.TiDBVectorConfig(
+        host="localhost",
+        port=4000,
+        user="root",
+        password="secret",
+        database="dify",
+        program_name="dify-app",
+    )
+
+
+@pytest.mark.parametrize(
+    ("field", "value", "message"),
+    [
+        ("host", "", "config TIDB_VECTOR_HOST is required"),
+        ("port", 0, "config TIDB_VECTOR_PORT is required"),
+        ("user", "", "config TIDB_VECTOR_USER is required"),
+        ("database", "", "config TIDB_VECTOR_DATABASE is required"),
+        ("program_name", "", "config APPLICATION_NAME is required"),
+    ],
+)
+def test_tidb_config_validation(tidb_module, field, value, message):
+    values = _config(tidb_module).model_dump()
+    values[field] = value
+
+    with pytest.raises(ValidationError, match=message):
+        tidb_module.TiDBVectorConfig.model_validate(values)
+
+
+def test_init_get_type_and_distance_func(tidb_module, monkeypatch):
+    monkeypatch.setattr(tidb_module, "create_engine", MagicMock(return_value="engine"))
+
+    vector = tidb_module.TiDBVector("collection_1", _config(tidb_module), distance_func="L2")
+
+    assert vector.get_type() == tidb_module.VectorType.TIDB_VECTOR
+    assert vector._url.startswith("mysql+pymysql://root:secret@localhost:4000/dify")
+    assert vector._dimension == 1536
+    assert vector._get_distance_func() == "VEC_L2_DISTANCE"
+
+    vector._distance_func = "cosine"
+    assert vector._get_distance_func() == "VEC_COSINE_DISTANCE"
+
+    vector._distance_func = "other"
+    assert vector._get_distance_func() == "VEC_COSINE_DISTANCE"
+
+
+def test_table_builds_columns_with_tidb_vector_type(tidb_module, monkeypatch):
+    fake_tidb_vector = types.ModuleType("tidb_vector")
+    fake_tidb_sqlalchemy = types.ModuleType("tidb_vector.sqlalchemy")
+
+    class _VectorType:
+        def __init__(self, dim):
+            self.dim = dim
+
+    fake_tidb_sqlalchemy.VectorType = _VectorType
+
+    monkeypatch.setitem(sys.modules, "tidb_vector", fake_tidb_vector)
+    monkeypatch.setitem(sys.modules, "tidb_vector.sqlalchemy", fake_tidb_sqlalchemy)
+    monkeypatch.setattr(tidb_module, "create_engine", MagicMock(return_value=MagicMock()))
+    monkeypatch.setattr(tidb_module, "Column", lambda *args, **kwargs: SimpleNamespace(args=args, kwargs=kwargs))
+    monkeypatch.setattr(
+        tidb_module,
+        "Table",
+        lambda name, _metadata, *columns, **_kwargs: SimpleNamespace(name=name, columns=columns),
+    )
+
+    vector = tidb_module.TiDBVector("collection_1", _config(tidb_module))
+    table = vector._table(3)
+
+    assert table.name == "collection_1"
+    column_names = [column.args[0] for column in table.columns]
+    assert tidb_module.Field.PRIMARY_KEY in column_names
+    assert tidb_module.Field.VECTOR in column_names
+    assert tidb_module.Field.TEXT_KEY in column_names
+
+
+def test_create_calls_collection_and_add_texts(tidb_module):
+    vector = tidb_module.TiDBVector.__new__(tidb_module.TiDBVector)
+    vector._collection_name = "collection_1"
+    vector._create_collection = MagicMock()
+    vector.add_texts = MagicMock()
+
+    docs = [Document(page_content="a", metadata={"doc_id": "id-1"})]
+    vector.create(docs, [[0.1, 0.2]])
+
+    vector._create_collection.assert_called_once_with(2)
+    vector.add_texts.assert_called_once_with(docs, [[0.1, 0.2]])
+    assert vector._dimension == 2
+
+
+def test_create_collection_skips_when_cache_hit(tidb_module, monkeypatch):
+    lock = MagicMock()
+    lock.__enter__.return_value = None
+    lock.__exit__.return_value = None
+    monkeypatch.setattr(tidb_module.redis_client, "lock", MagicMock(return_value=lock))
+    monkeypatch.setattr(tidb_module.redis_client, "get", MagicMock(return_value=1))
+    monkeypatch.setattr(tidb_module.redis_client, "set", MagicMock())
+
+    vector = tidb_module.TiDBVector.__new__(tidb_module.TiDBVector)
+    vector._collection_name = "collection_1"
+    vector._engine = MagicMock()
+
+    tidb_module.Session = MagicMock()
+
+    vector._create_collection(3)
+
+    tidb_module.Session.assert_not_called()
+    tidb_module.redis_client.set.assert_not_called()
+
+
+def test_create_collection_executes_create_sql_and_sets_cache(tidb_module, monkeypatch):
+    lock = MagicMock()
+    lock.__enter__.return_value = None
+    lock.__exit__.return_value = None
+    monkeypatch.setattr(tidb_module.redis_client, "lock", MagicMock(return_value=lock))
+    monkeypatch.setattr(tidb_module.redis_client, "get", MagicMock(return_value=None))
+    monkeypatch.setattr(tidb_module.redis_client, "set", MagicMock())
+
+    session = MagicMock()
+
+    class _SessionCtx:
+        def __enter__(self):
+            return session
+
+        def __exit__(self, exc_type, exc, tb):
+            return False
+
+    monkeypatch.setattr(tidb_module, "Session", lambda _engine: _SessionCtx())
+
+    vector = tidb_module.TiDBVector.__new__(tidb_module.TiDBVector)
+    vector._collection_name = "collection_1"
+    vector._engine = MagicMock()
+    vector._distance_func = "l2"
+
+    vector._create_collection(3)
+
+    session.begin.assert_called_once()
+    sql = str(session.execute.call_args.args[0])
+    assert "VECTOR<FLOAT>(3)" in sql
+    assert "VEC_L2_DISTANCE" in sql
+    session.commit.assert_called_once()
+    tidb_module.redis_client.set.assert_called_once()
+
+
+def test_add_texts_batches_inserts_and_returns_ids(tidb_module, monkeypatch):
+    class _InsertStmt:
+        def __init__(self, table):
+            self.table = table
+
+        def values(self, rows):
+            return {"table": self.table, "rows": rows}
+
+    monkeypatch.setattr(tidb_module, "insert", lambda table: _InsertStmt(table))
+
+    conn = MagicMock()
+    transaction = MagicMock()
+    transaction.__enter__.return_value = None
+    transaction.__exit__.return_value = None
+    conn.begin.return_value = transaction
+
+    connection_ctx = MagicMock()
+    connection_ctx.__enter__.return_value = conn
+    connection_ctx.__exit__.return_value = None
+
+    engine = MagicMock()
+    engine.connect.return_value = connection_ctx
+
+    vector = tidb_module.TiDBVector.__new__(tidb_module.TiDBVector)
+    vector._engine = engine
+    vector._table = MagicMock(return_value="table")
+
+    docs = [Document(page_content=f"text-{i}", metadata={"doc_id": f"id-{i}"}) for i in range(501)]
+    embeddings = [[float(i)] for i in range(501)]
+
+    ids = vector.add_texts(docs, embeddings)
+
+    assert ids[0] == "id-0"
+    assert len(ids) == 501
+    assert conn.execute.call_count == 2
+
+
+@pytest.fixture
+def tidb_vector_with_session(tidb_module, monkeypatch):
+    vector = tidb_module.TiDBVector.__new__(tidb_module.TiDBVector)
+    vector._collection_name = "collection_1"
+    vector._engine = MagicMock()
+    session = MagicMock()
+
+    class _SessionCtx:
+        def __enter__(self):
+            return session
+
+        def __exit__(self, exc_type, exc, tb):
+            return False
+
+    monkeypatch.setattr(tidb_module, "Session", lambda _engine: _SessionCtx())
+    return vector, session, tidb_module
+
+
+# 1. search_by_full_text returns empty
+def test_search_by_full_text_returns_empty(tidb_vector_with_session):
+    vector, _, _ = tidb_vector_with_session
+    assert vector.search_by_full_text("query") == []
+
+
+# 2. text_exists returns True when ids found
+def test_text_exists_returns_true_when_ids_found(tidb_vector_with_session):
+    vector, _, _ = tidb_vector_with_session
+    vector.get_ids_by_metadata_field = MagicMock(return_value=["id-1"])
+    assert vector.text_exists("doc-1") is True
+
+
+# 3. text_exists returns False when no ids
+def test_text_exists_returns_false_when_no_ids(tidb_vector_with_session):
+    vector, _, _ = tidb_vector_with_session
+    vector.get_ids_by_metadata_field = MagicMock(return_value=None)
+    assert vector.text_exists("doc-1") is False
+
+
+# 4. delete_by_ids delegates to _delete_by_ids when ids found
+def test_delete_by_ids_delegates_to_internal_delete(tidb_vector_with_session):
+    vector, session, tidb_module = tidb_vector_with_session
+    session.execute.return_value.fetchall.return_value = [("id-a",), ("id-b",)]
+    vector._delete_by_ids = MagicMock()
+    # Use real get_ids_by_metadata_field
+    vector.get_ids_by_metadata_field = tidb_module.TiDBVector.get_ids_by_metadata_field.__get__(
+        vector, tidb_module.TiDBVector
+    )
+    vector.delete_by_ids(["doc-a", "doc-b"])
+    vector._delete_by_ids.assert_called_once_with(["id-a", "id-b"])
+
+
+# 5. delete_by_ids skips when no ids found
+def test_delete_by_ids_skips_when_no_ids_found(tidb_vector_with_session):
+    vector, session, tidb_module = tidb_vector_with_session
+    session.execute.return_value.fetchall.return_value = []
+    vector._delete_by_ids = MagicMock()
+    # Use real get_ids_by_metadata_field
+    vector.get_ids_by_metadata_field = tidb_module.TiDBVector.get_ids_by_metadata_field.__get__(
+        vector, tidb_module.TiDBVector
+    )
+    vector.delete_by_ids(["doc-c"])
+    vector._delete_by_ids.assert_not_called()
+
+
+# 6. get_ids_by_metadata_field returns ids and returns None
+def test_get_ids_by_metadata_field_returns_ids_and_returns_none(tidb_vector_with_session):
+    vector, session, tidb_module = tidb_vector_with_session
+    # Returns ids
+    session.execute.return_value.fetchall.return_value = [("id-1",)]
+    assert vector.get_ids_by_metadata_field("doc_id", "doc-1") == ["id-1"]
+    # Returns None
+    session.execute.return_value.fetchall.return_value = []
+    assert vector.get_ids_by_metadata_field("doc_id", "doc-1") is None
+
+
+# 1. _delete_by_ids raises on None
+def test__delete_by_ids_raises_on_none(tidb_module):
+    vector = tidb_module.TiDBVector.__new__(tidb_module.TiDBVector)
+    with pytest.raises(ValueError, match="No ids provided"):
+        vector._delete_by_ids(None)
+
+
+# 2. _delete_by_ids returns True and calls execute
+def test__delete_by_ids_returns_true_and_calls_execute(tidb_module):
+    class _IDColumn:
+        def in_(self, ids):
+            return ids
+
+    class _Delete:
+        def where(self, condition):
+            return condition
+
+    table = SimpleNamespace(c=SimpleNamespace(id=_IDColumn()), delete=lambda: _Delete())
+    conn = MagicMock()
+    tx = MagicMock()
+    tx.__enter__.return_value = None
+    tx.__exit__.return_value = None
+    conn.begin.return_value = tx
+    conn_ctx = MagicMock()
+    conn_ctx.__enter__.return_value = conn
+    conn_ctx.__exit__.return_value = None
+    vector = tidb_module.TiDBVector.__new__(tidb_module.TiDBVector)
+    vector._collection_name = "collection_1"
+    vector._dimension = 2
+    vector._engine = SimpleNamespace(connect=MagicMock(return_value=conn_ctx))
+    vector._table = MagicMock(return_value=table)
+    assert vector._delete_by_ids(["id-1"]) is True
+    conn.execute.assert_called_once()
+
+
+# 3. _delete_by_ids returns False on RuntimeError
+def test__delete_by_ids_returns_false_on_runtime_error(tidb_module):
+    class _IDColumn:
+        def in_(self, ids):
+            return ids
+
+    class _Delete:
+        def where(self, condition):
+            return condition
+
+    table = SimpleNamespace(c=SimpleNamespace(id=_IDColumn()), delete=lambda: _Delete())
+    conn = MagicMock()
+    tx = MagicMock()
+    tx.__enter__.return_value = None
+    tx.__exit__.return_value = None
+    conn.begin.return_value = tx
+    conn_ctx = MagicMock()
+    conn_ctx.__enter__.return_value = conn
+    conn_ctx.__exit__.return_value = None
+    conn.execute.side_effect = RuntimeError("delete failed")
+    vector = tidb_module.TiDBVector.__new__(tidb_module.TiDBVector)
+    vector._collection_name = "collection_1"
+    vector._dimension = 2
+    vector._engine = SimpleNamespace(connect=MagicMock(return_value=conn_ctx))
+    vector._table = MagicMock(return_value=table)
+    assert vector._delete_by_ids(["id-2"]) is False
+
+
+# 4. delete_by_metadata_field calls _delete_by_ids when ids found
+def test_delete_by_metadata_field_calls__delete_by_ids_when_ids_found(tidb_module):
+    vector = tidb_module.TiDBVector.__new__(tidb_module.TiDBVector)
+    vector.get_ids_by_metadata_field = MagicMock(return_value=["id-3"])
+    vector._delete_by_ids = MagicMock()
+    vector.delete_by_metadata_field("doc_id", "doc-3")
+    vector._delete_by_ids.assert_called_once_with(["id-3"])
+
+
+# 5. delete_by_metadata_field does nothing when no ids
+def test_delete_by_metadata_field_does_nothing_when_no_ids(tidb_module):
+    vector = tidb_module.TiDBVector.__new__(tidb_module.TiDBVector)
+    vector.get_ids_by_metadata_field = MagicMock(return_value=[])
+    vector._delete_by_ids = MagicMock()
+    vector.delete_by_metadata_field("doc_id", "doc-4")
+    vector._delete_by_ids.assert_not_called()
+
+
+# Test search_by_vector filters and scores
+def test_search_by_vector_filters_and_scores(tidb_module, monkeypatch):
+    session = MagicMock()
+    session.execute.return_value = [
+        ('{"doc_id":"id-1","document_id":"d-1"}', "text-1", 0.2),
+        ('{"doc_id":"id-2","document_id":"d-2"}', "text-2", 0.4),
+    ]
+    session.commit = MagicMock()
+
+    class _SessionCtx:
+        def __enter__(self):
+            return session
+
+        def __exit__(self, exc_type, exc, tb):
+            return False
+
+    monkeypatch.setattr(tidb_module, "Session", lambda _engine: _SessionCtx())
+    vector = tidb_module.TiDBVector.__new__(tidb_module.TiDBVector)
+    vector._collection_name = "collection_1"
+    vector._engine = MagicMock()
+    vector._distance_func = "cosine"
+    docs = vector.search_by_vector(
+        [0.1, 0.2],
+        top_k=2,
+        score_threshold=0.5,
+        document_ids_filter=["d-1", "d-2"],
+    )
+    assert len(docs) == 2
+    assert docs[0].metadata["score"] == pytest.approx(0.8)
+    assert docs[1].metadata["score"] == pytest.approx(0.6)
+    sql = str(session.execute.call_args.args[0])
+    params = session.execute.call_args.kwargs["params"]
+    assert "meta->>'$.document_id' in ('d-1', 'd-2')" in sql
+    assert params["distance"] == pytest.approx(0.5)
+    assert params["top_k"] == 2
+    session.commit.assert_not_called()
+
+
+# Test delete drops table
+def test_delete_drops_table(tidb_module, monkeypatch):
+    session = MagicMock()
+    session.execute.return_value = None
+    session.commit = MagicMock()
+
+    class _SessionCtx:
+        def __enter__(self):
+            return session
+
+        def __exit__(self, exc_type, exc, tb):
+            return False
+
+    monkeypatch.setattr(tidb_module, "Session", lambda _engine: _SessionCtx())
+    vector = tidb_module.TiDBVector.__new__(tidb_module.TiDBVector)
+    vector._collection_name = "collection_1"
+    vector._engine = MagicMock()
+    vector.delete()
+    drop_sql = str(session.execute.call_args.args[0])
+    assert "DROP TABLE IF EXISTS collection_1" in drop_sql
+    session.commit.assert_called_once()
+
+
+def test_tidb_factory_uses_existing_or_generated_collection(tidb_module, monkeypatch):
+    factory = tidb_module.TiDBVectorFactory()
+    dataset_with_index = SimpleNamespace(
+        id="dataset-1",
+        index_struct_dict={"vector_store": {"class_prefix": "EXISTING_COLLECTION"}},
+        index_struct=None,
+    )
+    dataset_without_index = SimpleNamespace(id="dataset-2", index_struct_dict=None, index_struct=None)
+
+    monkeypatch.setattr(tidb_module.Dataset, "gen_collection_name_by_id", lambda _id: "AUTO_COLLECTION")
+    monkeypatch.setattr(tidb_module.dify_config, "TIDB_VECTOR_HOST", "localhost")
+    monkeypatch.setattr(tidb_module.dify_config, "TIDB_VECTOR_PORT", 4000)
+    monkeypatch.setattr(tidb_module.dify_config, "TIDB_VECTOR_USER", "root")
+    monkeypatch.setattr(tidb_module.dify_config, "TIDB_VECTOR_PASSWORD", "secret")
+    monkeypatch.setattr(tidb_module.dify_config, "TIDB_VECTOR_DATABASE", "dify")
+    monkeypatch.setattr(tidb_module.dify_config, "APPLICATION_NAME", "dify-app")
+
+    with patch.object(tidb_module, "TiDBVector", return_value="vector") as vector_cls:
+        result_1 = factory.init_vector(dataset_with_index, attributes=[], embeddings=MagicMock())
+        result_2 = factory.init_vector(dataset_without_index, attributes=[], embeddings=MagicMock())
+
+    assert result_1 == "vector"
+    assert result_2 == "vector"
+    assert vector_cls.call_args_list[0].kwargs["collection_name"] == "existing_collection"
+    assert vector_cls.call_args_list[1].kwargs["collection_name"] == "auto_collection"
+    assert dataset_without_index.index_struct is not None

api/tests/unit_tests/core/rag/datasource/vdb/upstash/test_upstash_vector.py

@@ -0,0 +1,186 @@
+import importlib
+import sys
+import types
+from types import SimpleNamespace
+from unittest.mock import MagicMock, patch
+
+import pytest
+from pydantic import ValidationError
+
+from core.rag.models.document import Document
+
+
+def _build_fake_upstash_module():
+    upstash_module = types.ModuleType("upstash_vector")
+
+    class Vector:
+        def __init__(self, id, vector, metadata, data):
+            self.id = id
+            self.vector = vector
+            self.metadata = metadata
+            self.data = data
+
+    class Index:
+        def __init__(self, url, token):
+            self.url = url
+            self.token = token
+            self.info = MagicMock(return_value=SimpleNamespace(dimension=8))
+            self.upsert = MagicMock()
+            self.query = MagicMock(return_value=[])
+            self.delete = MagicMock()
+            self.reset = MagicMock()
+
+    upstash_module.Vector = Vector
+    upstash_module.Index = Index
+    return upstash_module
+
+
+@pytest.fixture
+def upstash_module(monkeypatch):
+    # Remove patched modules if present
+    for modname in ["upstash_vector", "core.rag.datasource.vdb.upstash.upstash_vector"]:
+        if modname in sys.modules:
+            monkeypatch.delitem(sys.modules, modname, raising=False)
+    monkeypatch.setitem(sys.modules, "upstash_vector", _build_fake_upstash_module())
+    module = importlib.import_module("core.rag.datasource.vdb.upstash.upstash_vector")
+    return module
+
+
+def _config(module):
+    return module.UpstashVectorConfig(url="https://upstash.example", token="token-123")
+
+
+@pytest.mark.parametrize(
+    ("field", "value", "message"),
+    [
+        ("url", "", "Upstash URL is required"),
+        ("token", "", "Upstash Token is required"),
+    ],
+)
+def test_upstash_config_validation(upstash_module, field, value, message):
+    values = _config(upstash_module).model_dump()
+    values[field] = value
+
+    with pytest.raises(ValidationError, match=message):
+        upstash_module.UpstashVectorConfig.model_validate(values)
+
+
+def test_init_get_type_and_dimension(upstash_module, monkeypatch):
+    vector = upstash_module.UpstashVector("collection_1", _config(upstash_module))
+
+    assert vector.get_type() == upstash_module.VectorType.UPSTASH
+    assert vector._table_name == "collection_1"
+    assert vector._get_index_dimension() == 8
+
+    vector.index.info.return_value = SimpleNamespace(dimension=None)
+    assert vector._get_index_dimension() == 1536
+
+    vector.index.info.return_value = None
+    assert vector._get_index_dimension() == 1536
+
+    monkeypatch.setattr(upstash_module, "uuid4", lambda: "generated-uuid")
+    docs = [Document(page_content="hello", metadata={"doc_id": "id-1"})]
+    vector.add_texts(docs, [[0.1, 0.2]])
+
+    vector.index.upsert.assert_called_once()
+    upsert_vectors = vector.index.upsert.call_args.kwargs["vectors"]
+    assert upsert_vectors[0].id == "generated-uuid"
+
+
+def test_create_text_exists_and_delete_by_ids(upstash_module):
+    vector = upstash_module.UpstashVector("collection_1", _config(upstash_module))
+    vector.add_texts = MagicMock()
+
+    docs = [Document(page_content="hello", metadata={"doc_id": "id-1"})]
+    vector.create(docs, [[0.1]])
+    vector.add_texts.assert_called_once_with(docs, [[0.1]])
+
+    vector.get_ids_by_metadata_field = MagicMock(return_value=["id-1"])
+    assert vector.text_exists("doc-1") is True
+    vector.get_ids_by_metadata_field.return_value = []
+    assert vector.text_exists("doc-1") is False
+
+    vector.get_ids_by_metadata_field = MagicMock(side_effect=[["item-1"], [], ["item-2"]])
+    vector._delete_by_ids = MagicMock()
+    vector.delete_by_ids(["doc-1", "doc-2", "doc-3"])
+    vector._delete_by_ids.assert_called_once_with(ids=["item-1", "item-2"])
+
+
+def test_delete_helpers_and_search(upstash_module):
+    vector = upstash_module.UpstashVector("collection_1", _config(upstash_module))
+
+    vector._delete_by_ids([])
+    vector.index.delete.assert_not_called()
+    vector._delete_by_ids(["a", "b"])
+    vector.index.delete.assert_called_once_with(ids=["a", "b"])
+
+    vector.index.query.return_value = [SimpleNamespace(id="x-1"), SimpleNamespace(id="x-2")]
+    ids = vector.get_ids_by_metadata_field("doc_id", "doc-1")
+    assert ids == ["x-1", "x-2"]
+    query_kwargs = vector.index.query.call_args.kwargs
+    assert query_kwargs["top_k"] == 1000
+    assert query_kwargs["filter"] == "doc_id = 'doc-1'"
+
+    vector._delete_by_ids = MagicMock()
+    vector.get_ids_by_metadata_field = MagicMock(return_value=["x-1"])
+    vector.delete_by_metadata_field("doc_id", "doc-1")
+    vector._delete_by_ids.assert_called_once_with(["x-1"])
+
+    vector._delete_by_ids.reset_mock()
+    vector.get_ids_by_metadata_field.return_value = []
+    vector.delete_by_metadata_field("doc_id", "doc-2")
+    vector._delete_by_ids.assert_not_called()
+
+
+def test_search_by_vector_filter_threshold_and_delete(upstash_module):
+    vector = upstash_module.UpstashVector("collection_1", _config(upstash_module))
+    vector.index.query.return_value = [
+        SimpleNamespace(metadata={"document_id": "d-1"}, data="text-1", score=0.9),
+        SimpleNamespace(metadata={"document_id": "d-2"}, data="text-2", score=0.3),
+        SimpleNamespace(metadata=None, data="text-3", score=0.99),
+        SimpleNamespace(metadata={"document_id": "d-4"}, data=None, score=0.99),
+    ]
+
+    docs = vector.search_by_vector(
+        [0.1, 0.2],
+        top_k=3,
+        score_threshold=0.5,
+        document_ids_filter=["d-1", "d-2"],
+    )
+
+    assert len(docs) == 1
+    assert docs[0].page_content == "text-1"
+    assert docs[0].metadata["score"] == pytest.approx(0.9)
+
+    search_kwargs = vector.index.query.call_args.kwargs
+    assert search_kwargs["top_k"] == 3
+    assert search_kwargs["filter"] == "document_id in ('d-1', 'd-2')"
+
+    assert vector.search_by_full_text("query") == []
+
+    vector.delete()
+    vector.index.reset.assert_called_once()
+
+
+def test_upstash_factory_uses_existing_or_generated_collection(upstash_module, monkeypatch):
+    factory = upstash_module.UpstashVectorFactory()
+    dataset_with_index = SimpleNamespace(
+        id="dataset-1",
+        index_struct_dict={"vector_store": {"class_prefix": "EXISTING_COLLECTION"}},
+        index_struct=None,
+    )
+    dataset_without_index = SimpleNamespace(id="dataset-2", index_struct_dict=None, index_struct=None)
+
+    monkeypatch.setattr(upstash_module.Dataset, "gen_collection_name_by_id", lambda _id: "AUTO_COLLECTION")
+    monkeypatch.setattr(upstash_module.dify_config, "UPSTASH_VECTOR_URL", "https://upstash.example")
+    monkeypatch.setattr(upstash_module.dify_config, "UPSTASH_VECTOR_TOKEN", "token-123")
+
+    with patch.object(upstash_module, "UpstashVector", return_value="vector") as vector_cls:
+        result_1 = factory.init_vector(dataset_with_index, attributes=[], embeddings=MagicMock())
+        result_2 = factory.init_vector(dataset_without_index, attributes=[], embeddings=MagicMock())
+
+    assert result_1 == "vector"
+    assert result_2 == "vector"
+    assert vector_cls.call_args_list[0].kwargs["collection_name"] == "existing_collection"
+    assert vector_cls.call_args_list[1].kwargs["collection_name"] == "auto_collection"
+    assert dataset_without_index.index_struct is not None

api/tests/unit_tests/core/rag/datasource/vdb/vikingdb/test_vikingdb_vector.py

@@ -0,0 +1,310 @@
+import importlib
+import json
+import sys
+import types
+from collections import UserDict
+from types import SimpleNamespace
+from unittest.mock import MagicMock, patch
+
+import pytest
+
+from core.rag.models.document import Document
+
+
+def _build_fake_vikingdb_modules():
+    volcengine = types.ModuleType("volcengine")
+    volcengine.__path__ = []
+    viking_db = types.ModuleType("volcengine.viking_db")
+
+    class Data(UserDict):
+        def __init__(self, payload):
+            super().__init__(payload)
+            self.fields = payload
+
+    class DistanceType:
+        L2 = "L2"
+
+    class IndexType:
+        HNSW = "HNSW"
+
+    class QuantType:
+        Float = "Float"
+
+    class FieldType:
+        String = "string"
+        Text = "text"
+        Vector = "vector"
+
+    class Field:
+        def __init__(self, **kwargs):
+            self.kwargs = kwargs
+
+    class VectorIndexParams:
+        def __init__(self, **kwargs):
+            self.kwargs = kwargs
+
+    class _Collection:
+        def __init__(self):
+            self.upsert_data = MagicMock()
+            self.fetch_data = MagicMock(return_value=None)
+            self.delete_data = MagicMock()
+
+    class _Index:
+        def __init__(self):
+            self.search = MagicMock(return_value=[])
+            self.search_by_vector = MagicMock(return_value=[])
+
+    class VikingDBService:
+        def __init__(self, **kwargs):
+            self.kwargs = kwargs
+            self.create_collection = MagicMock()
+            self.create_index = MagicMock()
+            self.drop_index = MagicMock()
+            self.drop_collection = MagicMock()
+            self._collection = _Collection()
+            self._index = _Index()
+            self.get_collection = MagicMock(return_value=self._collection)
+            self.get_index = MagicMock(return_value=self._index)
+
+    viking_db.Data = Data
+    viking_db.DistanceType = DistanceType
+    viking_db.Field = Field
+    viking_db.FieldType = FieldType
+    viking_db.IndexType = IndexType
+    viking_db.QuantType = QuantType
+    viking_db.VectorIndexParams = VectorIndexParams
+    viking_db.VikingDBService = VikingDBService
+
+    return {"volcengine": volcengine, "volcengine.viking_db": viking_db}
+
+
+@pytest.fixture
+def vikingdb_module(monkeypatch):
+    for name, module in _build_fake_vikingdb_modules().items():
+        monkeypatch.setitem(sys.modules, name, module)
+
+    import core.rag.datasource.vdb.vikingdb.vikingdb_vector as module
+
+    return importlib.reload(module)
+
+
+def _config(module):
+    return module.VikingDBConfig(
+        access_key="ak",
+        secret_key="sk",
+        host="host",
+        region="region",
+        scheme="https",
+        connection_timeout=10,
+        socket_timeout=20,
+    )
+
+
+def test_init_get_type_and_has_checks(vikingdb_module):
+    vector = vikingdb_module.VikingDBVector("collection_1", "group-1", _config(vikingdb_module))
+
+    assert vector.get_type() == vikingdb_module.VectorType.VIKINGDB
+    assert vector._index_name == "collection_1_idx"
+
+    assert vector._has_collection() is True
+    assert vector._has_index() is True
+
+    vector._client.get_collection.side_effect = RuntimeError("missing")
+    assert vector._has_collection() is False
+    vector._client.get_collection.side_effect = None
+
+    vector._client.get_index.side_effect = RuntimeError("missing")
+    assert vector._has_index() is False
+
+
+def test_create_collection_cache_and_creation_paths(vikingdb_module, monkeypatch):
+    lock = MagicMock()
+    lock.__enter__.return_value = None
+    lock.__exit__.return_value = None
+    monkeypatch.setattr(vikingdb_module.redis_client, "lock", MagicMock(return_value=lock))
+    monkeypatch.setattr(vikingdb_module.redis_client, "set", MagicMock())
+
+    vector = vikingdb_module.VikingDBVector("collection_1", "group-1", _config(vikingdb_module))
+
+    monkeypatch.setattr(vikingdb_module.redis_client, "get", MagicMock(return_value=1))
+    vector._create_collection(3)
+    vector._client.create_collection.assert_not_called()
+    vector._client.create_index.assert_not_called()
+
+    monkeypatch.setattr(vikingdb_module.redis_client, "get", MagicMock(return_value=None))
+    vector._has_collection = MagicMock(return_value=False)
+    vector._has_index = MagicMock(return_value=False)
+    vector._create_collection(4)
+
+    vector._client.create_collection.assert_called_once()
+    vector._client.create_index.assert_called_once()
+    vikingdb_module.redis_client.set.assert_called_once()
+
+
+def test_create_and_add_texts(vikingdb_module):
+    vector = vikingdb_module.VikingDBVector("collection_1", "group-1", _config(vikingdb_module))
+    vector._create_collection = MagicMock()
+    vector.add_texts = MagicMock()
+
+    docs = [Document(page_content="hello", metadata={"doc_id": "id-1"})]
+    vector.create(docs, [[0.1, 0.2]])
+
+    vector._create_collection.assert_called_once_with(2)
+    vector.add_texts.assert_called_once_with(docs, [[0.1, 0.2]])
+
+    vector = vikingdb_module.VikingDBVector("collection_2", "group-2", _config(vikingdb_module))
+    docs = [
+        Document(page_content="a", metadata={"doc_id": "id-a", "document_id": "d-1"}),
+        Document(page_content="b", metadata={"doc_id": "id-b", "document_id": "d-2"}),
+    ]
+    vector.add_texts(docs, [[0.1], [0.2]])
+
+    vector._client.get_collection.assert_called()
+    upsert_docs = vector._client.get_collection.return_value.upsert_data.call_args.args[0]
+    assert upsert_docs[0][vikingdb_module.vdb_Field.PRIMARY_KEY] == "id-a"
+    assert upsert_docs[0][vikingdb_module.vdb_Field.GROUP_KEY] == "group-2"
+
+
+def test_text_exists_and_delete_operations(vikingdb_module):
+    vector = vikingdb_module.VikingDBVector("collection_1", "group-1", _config(vikingdb_module))
+
+    vector._client.get_collection.return_value.fetch_data.return_value = SimpleNamespace(fields={"message": "ok"})
+    assert vector.text_exists("id-1") is True
+
+    vector._client.get_collection.return_value.fetch_data.return_value = SimpleNamespace(
+        fields={"message": "data does not exist"}
+    )
+    assert vector.text_exists("id-1") is False
+
+    vector._client.get_collection.return_value.fetch_data.return_value = None
+    assert vector.text_exists("id-1") is False
+
+    vector.delete_by_ids(["id-1"])
+    vector._client.get_collection.return_value.delete_data.assert_called_once_with(["id-1"])
+
+    vector.get_ids_by_metadata_field = MagicMock(return_value=["id-2"])
+    vector.delete_by_ids = MagicMock()
+    vector.delete_by_metadata_field("doc_id", "doc-1")
+    vector.delete_by_ids.assert_called_once_with(["id-2"])
+
+
+def test_get_ids_and_search_helpers(vikingdb_module):
+    vector = vikingdb_module.VikingDBVector("collection_1", "group-1", _config(vikingdb_module))
+
+    vector._client.get_index.return_value.search.return_value = []
+    assert vector.get_ids_by_metadata_field("doc_id", "x") == []
+
+    vector._client.get_index.return_value.search.return_value = [
+        SimpleNamespace(id="a", fields={vikingdb_module.vdb_Field.METADATA_KEY: json.dumps({"doc_id": "x"})}),
+        SimpleNamespace(id="b", fields={vikingdb_module.vdb_Field.METADATA_KEY: json.dumps({"doc_id": "y"})}),
+        SimpleNamespace(id="c", fields={}),
+    ]
+    assert vector.get_ids_by_metadata_field("doc_id", "x") == ["a"]
+
+    empty_docs = vector._get_search_res([], score_threshold=0.1)
+    assert empty_docs == []
+
+    results = [
+        SimpleNamespace(
+            id="a",
+            score=0.3,
+            fields={
+                vikingdb_module.vdb_Field.CONTENT_KEY: "doc-a",
+                vikingdb_module.vdb_Field.METADATA_KEY: json.dumps({"document_id": "d-1"}),
+            },
+        ),
+        SimpleNamespace(
+            id="b",
+            score=0.9,
+            fields={
+                vikingdb_module.vdb_Field.CONTENT_KEY: "doc-b",
+                vikingdb_module.vdb_Field.METADATA_KEY: json.dumps({"document_id": "d-2"}),
+            },
+        ),
+    ]
+
+    docs = vector._get_search_res(results, score_threshold=0.2)
+    assert [doc.page_content for doc in docs] == ["doc-b", "doc-a"]
+
+    vector._client.get_index.return_value.search_by_vector.return_value = results
+    filtered_docs = vector.search_by_vector([0.1], top_k=2, score_threshold=0.2, document_ids_filter=["d-2"])
+    assert len(filtered_docs) == 1
+    assert filtered_docs[0].page_content == "doc-b"
+    assert vector.search_by_full_text("query") == []
+
+
+def test_delete_drops_index_and_collection_when_present(vikingdb_module):
+    vector = vikingdb_module.VikingDBVector("collection_1", "group-1", _config(vikingdb_module))
+    vector._has_index = MagicMock(return_value=True)
+    vector._has_collection = MagicMock(return_value=True)
+
+    vector.delete()
+
+    vector._client.drop_index.assert_called_once_with("collection_1", "collection_1_idx")
+    vector._client.drop_collection.assert_called_once_with("collection_1")
+
+    vector._client.drop_index.reset_mock()
+    vector._client.drop_collection.reset_mock()
+    vector._has_index.return_value = False
+    vector._has_collection.return_value = False
+    vector.delete()
+
+    vector._client.drop_index.assert_not_called()
+    vector._client.drop_collection.assert_not_called()
+
+
+def test_vikingdb_factory_validates_config_and_builds_vector(vikingdb_module, monkeypatch):
+    factory = vikingdb_module.VikingDBVectorFactory()
+    dataset_with_index = SimpleNamespace(
+        id="dataset-1",
+        index_struct_dict={"vector_store": {"class_prefix": "EXISTING_COLLECTION"}},
+        index_struct=None,
+    )
+    dataset_without_index = SimpleNamespace(id="dataset-2", index_struct_dict=None, index_struct=None)
+
+    monkeypatch.setattr(vikingdb_module.Dataset, "gen_collection_name_by_id", lambda _id: "AUTO_COLLECTION")
+
+    with patch.object(vikingdb_module, "VikingDBVector", return_value="vector") as vector_cls:
+        monkeypatch.setattr(vikingdb_module.dify_config, "VIKINGDB_ACCESS_KEY", "ak")
+        monkeypatch.setattr(vikingdb_module.dify_config, "VIKINGDB_SECRET_KEY", "sk")
+        monkeypatch.setattr(vikingdb_module.dify_config, "VIKINGDB_HOST", "host")
+        monkeypatch.setattr(vikingdb_module.dify_config, "VIKINGDB_REGION", "region")
+        monkeypatch.setattr(vikingdb_module.dify_config, "VIKINGDB_SCHEME", "https")
+        monkeypatch.setattr(vikingdb_module.dify_config, "VIKINGDB_CONNECTION_TIMEOUT", 10)
+        monkeypatch.setattr(vikingdb_module.dify_config, "VIKINGDB_SOCKET_TIMEOUT", 20)
+
+        result_1 = factory.init_vector(dataset_with_index, attributes=[], embeddings=MagicMock())
+        result_2 = factory.init_vector(dataset_without_index, attributes=[], embeddings=MagicMock())
+
+    assert result_1 == "vector"
+    assert result_2 == "vector"
+    assert vector_cls.call_args_list[0].kwargs["collection_name"] == "existing_collection"
+    assert vector_cls.call_args_list[1].kwargs["collection_name"] == "auto_collection"
+    assert dataset_without_index.index_struct is not None
+
+
+@pytest.mark.parametrize(
+    ("field", "message"),
+    [
+        ("VIKINGDB_ACCESS_KEY", "VIKINGDB_ACCESS_KEY should not be None"),
+        ("VIKINGDB_SECRET_KEY", "VIKINGDB_SECRET_KEY should not be None"),
+        ("VIKINGDB_HOST", "VIKINGDB_HOST should not be None"),
+        ("VIKINGDB_REGION", "VIKINGDB_REGION should not be None"),
+        ("VIKINGDB_SCHEME", "VIKINGDB_SCHEME should not be None"),
+    ],
+)
+def test_vikingdb_factory_raises_when_required_config_missing(vikingdb_module, monkeypatch, field, message):
+    factory = vikingdb_module.VikingDBVectorFactory()
+    dataset = SimpleNamespace(
+        id="dataset-1", index_struct_dict={"vector_store": {"class_prefix": "existing"}}, index_struct=None
+    )
+
+    monkeypatch.setattr(vikingdb_module.dify_config, "VIKINGDB_ACCESS_KEY", "ak")
+    monkeypatch.setattr(vikingdb_module.dify_config, "VIKINGDB_SECRET_KEY", "sk")
+    monkeypatch.setattr(vikingdb_module.dify_config, "VIKINGDB_HOST", "host")
+    monkeypatch.setattr(vikingdb_module.dify_config, "VIKINGDB_REGION", "region")
+    monkeypatch.setattr(vikingdb_module.dify_config, "VIKINGDB_SCHEME", "https")
+    monkeypatch.setattr(vikingdb_module.dify_config, field, None)
+
+    with pytest.raises(ValueError, match=message):
+        factory.init_vector(dataset, attributes=[], embeddings=MagicMock())

api/tests/unit_tests/core/rag/datasource/vdb/weaviate/test_weaviate_vector.py

@@ -7,10 +7,14 @@ Focuses on verifying that doc_type is properly handled in:
 - Full-text search result metadata (search_by_full_text)
 """
 
+import datetime
+import json
 import unittest
 from types import SimpleNamespace
 from unittest.mock import MagicMock, patch
 
+import pytest
+
 from core.rag.datasource.vdb.weaviate import weaviate_vector as weaviate_vector_module
 from core.rag.datasource.vdb.weaviate.weaviate_vector import WeaviateConfig, WeaviateVector
 from core.rag.models.document import Document
@@ -32,6 +36,10 @@ class TestWeaviateVector(unittest.TestCase):
     def tearDown(self):
         weaviate_vector_module._weaviate_client = None
 
+    def test_config_requires_endpoint(self):
+        with pytest.raises(ValueError, match="config WEAVIATE_ENDPOINT is required"):
+            WeaviateConfig(endpoint="")
+
     @patch("core.rag.datasource.vdb.weaviate.weaviate_vector.weaviate")
     def _create_weaviate_vector(self, mock_weaviate_module):
         """Helper to create a WeaviateVector instance with mocked client."""
@@ -46,6 +54,85 @@ class TestWeaviateVector(unittest.TestCase):
         )
         return wv, mock_client
 
+    def test_shutdown_client_logs_debug_when_close_fails(self):
+        mock_client = MagicMock()
+        mock_client.close.side_effect = RuntimeError("close failed")
+        weaviate_vector_module._weaviate_client = mock_client
+
+        with patch.object(weaviate_vector_module.logger, "debug") as mock_debug:
+            weaviate_vector_module._shutdown_weaviate_client()
+
+        assert weaviate_vector_module._weaviate_client is None
+        mock_client.close.assert_called_once()
+        mock_debug.assert_called_once()
+
+    @patch("core.rag.datasource.vdb.weaviate.weaviate_vector.weaviate.connect_to_custom")
+    def test_init_client_reuses_cached_client_without_reconnect(self, mock_connect):
+        cached_client = MagicMock()
+        cached_client.is_ready.return_value = True
+        weaviate_vector_module._weaviate_client = cached_client
+
+        wv = WeaviateVector.__new__(WeaviateVector)
+
+        client = wv._init_client(self.config)
+
+        assert client is cached_client
+        mock_connect.assert_not_called()
+
+    @patch("core.rag.datasource.vdb.weaviate.weaviate_vector.weaviate.connect_to_custom")
+    def test_init_client_reuses_cached_client_after_lock_recheck(self, mock_connect):
+        cached_client = MagicMock()
+        cached_client.is_ready.side_effect = [False, True]
+        weaviate_vector_module._weaviate_client = cached_client
+
+        wv = WeaviateVector.__new__(WeaviateVector)
+
+        client = wv._init_client(self.config)
+
+        assert client is cached_client
+        mock_connect.assert_not_called()
+
+    @patch("core.rag.datasource.vdb.weaviate.weaviate_vector.Auth.api_key", return_value="auth-token")
+    @patch("core.rag.datasource.vdb.weaviate.weaviate_vector.weaviate.connect_to_custom")
+    def test_init_client_parses_custom_grpc_endpoint_without_scheme(self, mock_connect, mock_api_key):
+        mock_client = MagicMock()
+        mock_client.is_ready.return_value = True
+        mock_connect.return_value = mock_client
+
+        wv = WeaviateVector.__new__(WeaviateVector)
+        config = WeaviateConfig(
+            endpoint="https://weaviate.example.com",
+            grpc_endpoint="grpc.example.com:6000",
+            api_key="test-key",
+            batch_size=50,
+        )
+
+        client = wv._init_client(config)
+
+        assert client is mock_client
+        assert mock_connect.call_args.kwargs == {
+            "http_host": "weaviate.example.com",
+            "http_port": 443,
+            "http_secure": True,
+            "grpc_host": "grpc.example.com",
+            "grpc_port": 6000,
+            "grpc_secure": False,
+            "auth_credentials": "auth-token",
+            "skip_init_checks": True,
+        }
+        mock_api_key.assert_called_once_with("test-key")
+
+    @patch("core.rag.datasource.vdb.weaviate.weaviate_vector.weaviate.connect_to_custom")
+    def test_init_client_raises_when_database_not_ready(self, mock_connect):
+        mock_client = MagicMock()
+        mock_client.is_ready.return_value = False
+        mock_connect.return_value = mock_client
+
+        wv = WeaviateVector.__new__(WeaviateVector)
+
+        with pytest.raises(ConnectionError, match="Vector database is not ready"):
+            wv._init_client(self.config)
+
     @patch("core.rag.datasource.vdb.weaviate.weaviate_vector.weaviate")
     def test_init(self, mock_weaviate_module):
         """Test WeaviateVector initialization stores attributes including doc_type."""
@@ -62,6 +149,40 @@ class TestWeaviateVector(unittest.TestCase):
         assert wv._collection_name == self.collection_name
         assert "doc_type" in wv._attributes
 
+    def test_get_type_and_to_index_struct(self):
+        wv = WeaviateVector.__new__(WeaviateVector)
+        wv._collection_name = self.collection_name
+
+        assert wv.get_type() == weaviate_vector_module.VectorType.WEAVIATE
+        assert wv.to_index_struct() == {
+            "type": weaviate_vector_module.VectorType.WEAVIATE,
+            "vector_store": {"class_prefix": self.collection_name},
+        }
+
+    def test_get_collection_name_uses_existing_class_prefix_and_appends_suffix(self):
+        dataset = SimpleNamespace(index_struct_dict={"vector_store": {"class_prefix": "ExistingCollection"}}, id="ds-1")
+        wv = WeaviateVector.__new__(WeaviateVector)
+
+        assert wv.get_collection_name(dataset) == "ExistingCollection_Node"
+
+    def test_get_collection_name_generates_name_from_dataset_id(self):
+        dataset = SimpleNamespace(index_struct_dict=None, id="ds-2")
+        wv = WeaviateVector.__new__(WeaviateVector)
+
+        with patch.object(weaviate_vector_module.Dataset, "gen_collection_name_by_id", return_value="Generated_Node"):
+            assert wv.get_collection_name(dataset) == "Generated_Node"
+
+    def test_create_calls_collection_setup_then_add_texts(self):
+        doc = Document(page_content="hello", metadata={})
+        wv = WeaviateVector.__new__(WeaviateVector)
+        wv._create_collection = MagicMock()
+        wv.add_texts = MagicMock()
+
+        wv.create([doc], [[0.1, 0.2]])
+
+        wv._create_collection.assert_called_once()
+        wv.add_texts.assert_called_once_with([doc], [[0.1, 0.2]])
+
     @patch("core.rag.datasource.vdb.weaviate.weaviate_vector.redis_client")
     @patch("core.rag.datasource.vdb.weaviate.weaviate_vector.dify_config")
     @patch("core.rag.datasource.vdb.weaviate.weaviate_vector.weaviate")
@@ -111,6 +232,44 @@ class TestWeaviateVector(unittest.TestCase):
             f"doc_type should be in collection schema properties, got: {property_names}"
         )
 
+    @patch("core.rag.datasource.vdb.weaviate.weaviate_vector.redis_client")
+    def test_create_collection_returns_early_when_cache_key_exists(self, mock_redis):
+        mock_lock = MagicMock()
+        mock_lock.__enter__ = MagicMock()
+        mock_lock.__exit__ = MagicMock()
+        mock_redis.lock.return_value = mock_lock
+        mock_redis.get.return_value = 1
+
+        wv = WeaviateVector.__new__(WeaviateVector)
+        wv._collection_name = self.collection_name
+        wv._client = MagicMock()
+        wv._ensure_properties = MagicMock()
+
+        wv._create_collection()
+
+        wv._client.collections.exists.assert_not_called()
+        wv._ensure_properties.assert_not_called()
+        mock_redis.set.assert_not_called()
+
+    @patch("core.rag.datasource.vdb.weaviate.weaviate_vector.redis_client")
+    def test_create_collection_logs_and_reraises_errors(self, mock_redis):
+        mock_lock = MagicMock()
+        mock_lock.__enter__ = MagicMock()
+        mock_lock.__exit__ = MagicMock(return_value=False)
+        mock_redis.lock.return_value = mock_lock
+        mock_redis.get.return_value = None
+
+        wv = WeaviateVector.__new__(WeaviateVector)
+        wv._collection_name = self.collection_name
+        wv._client = MagicMock()
+        wv._client.collections.exists.side_effect = RuntimeError("create failed")
+
+        with patch.object(weaviate_vector_module.logger, "exception") as mock_exception:
+            with pytest.raises(RuntimeError, match="create failed"):
+                wv._create_collection()
+
+        mock_exception.assert_called_once()
+
     @patch("core.rag.datasource.vdb.weaviate.weaviate_vector.weaviate")
     def test_ensure_properties_adds_missing_doc_type(self, mock_weaviate_module):
         """Test that _ensure_properties adds doc_type when it's missing from existing schema."""
@@ -146,6 +305,29 @@ class TestWeaviateVector(unittest.TestCase):
         added_names = [call.args[0].name for call in add_calls]
         assert "doc_type" in added_names, f"doc_type should be added to existing collection, added: {added_names}"
 
+    @patch("core.rag.datasource.vdb.weaviate.weaviate_vector.weaviate")
+    def test_ensure_properties_adds_all_missing_core_properties(self, mock_weaviate_module):
+        mock_client = MagicMock()
+        mock_client.is_ready.return_value = True
+        mock_weaviate_module.connect_to_custom.return_value = mock_client
+        mock_client.collections.exists.return_value = True
+        mock_col = MagicMock()
+        mock_client.collections.use.return_value = mock_col
+        mock_cfg = MagicMock()
+        mock_cfg.properties = [SimpleNamespace(name="text")]
+        mock_col.config.get.return_value = mock_cfg
+
+        wv = WeaviateVector(
+            collection_name=self.collection_name,
+            config=self.config,
+            attributes=self.attributes,
+        )
+        wv._ensure_properties()
+
+        add_calls = mock_col.config.add_property.call_args_list
+        added_names = [call.args[0].name for call in add_calls]
+        assert added_names == ["document_id", "doc_id", "doc_type", "chunk_index"]
+
     @patch("core.rag.datasource.vdb.weaviate.weaviate_vector.weaviate")
     def test_ensure_properties_skips_existing_doc_type(self, mock_weaviate_module):
         """Test that _ensure_properties does not add doc_type when it already exists."""
@@ -179,6 +361,30 @@ class TestWeaviateVector(unittest.TestCase):
         # No properties should be added
         mock_col.config.add_property.assert_not_called()
 
+    @patch("core.rag.datasource.vdb.weaviate.weaviate_vector.weaviate")
+    def test_ensure_properties_logs_warning_when_property_addition_fails(self, mock_weaviate_module):
+        mock_client = MagicMock()
+        mock_client.is_ready.return_value = True
+        mock_weaviate_module.connect_to_custom.return_value = mock_client
+        mock_client.collections.exists.return_value = True
+        mock_col = MagicMock()
+        mock_client.collections.use.return_value = mock_col
+        mock_cfg = MagicMock()
+        mock_cfg.properties = []
+        mock_col.config.get.return_value = mock_cfg
+        mock_col.config.add_property.side_effect = RuntimeError("cannot add")
+
+        wv = WeaviateVector(
+            collection_name=self.collection_name,
+            config=self.config,
+            attributes=self.attributes,
+        )
+
+        with patch.object(weaviate_vector_module.logger, "warning") as mock_warning:
+            wv._ensure_properties()
+
+        assert mock_warning.call_count == 4
+
     @patch("core.rag.datasource.vdb.weaviate.weaviate_vector.weaviate")
     def test_search_by_vector_returns_doc_type_in_metadata(self, mock_weaviate_module):
         """Test that search_by_vector returns doc_type in document metadata.
@@ -226,6 +432,58 @@ class TestWeaviateVector(unittest.TestCase):
         assert len(docs) == 1
         assert docs[0].metadata.get("doc_type") == "image"
 
+    @patch("core.rag.datasource.vdb.weaviate.weaviate_vector.weaviate")
+    def test_search_by_vector_uses_document_filter_and_default_distance(self, mock_weaviate_module):
+        mock_client = MagicMock()
+        mock_client.is_ready.return_value = True
+        mock_weaviate_module.connect_to_custom.return_value = mock_client
+        mock_client.collections.exists.return_value = True
+        mock_col = MagicMock()
+        mock_client.collections.use.return_value = mock_col
+
+        mock_obj = MagicMock()
+        mock_obj.properties = {
+            "text": "fallback distance result",
+            "document_id": "doc-1",
+            "doc_id": "segment-1",
+        }
+        mock_obj.metadata = None
+
+        mock_result = MagicMock()
+        mock_result.objects = [mock_obj]
+        mock_col.query.near_vector.return_value = mock_result
+
+        wv = WeaviateVector(
+            collection_name=self.collection_name,
+            config=self.config,
+            attributes=self.attributes,
+        )
+        docs = wv.search_by_vector(
+            query_vector=[0.2] * 3,
+            document_ids_filter=["doc-1"],
+            top_k=2,
+            score_threshold=-1,
+        )
+
+        assert len(docs) == 1
+        assert docs[0].metadata["score"] == 0.0
+        assert mock_col.query.near_vector.call_args.kwargs["filters"] is not None
+
+    @patch("core.rag.datasource.vdb.weaviate.weaviate_vector.weaviate")
+    def test_search_by_vector_returns_empty_when_collection_is_missing(self, mock_weaviate_module):
+        mock_client = MagicMock()
+        mock_client.is_ready.return_value = True
+        mock_weaviate_module.connect_to_custom.return_value = mock_client
+        mock_client.collections.exists.return_value = False
+
+        wv = WeaviateVector(
+            collection_name=self.collection_name,
+            config=self.config,
+            attributes=self.attributes,
+        )
+
+        assert wv.search_by_vector(query_vector=[0.1] * 3) == []
+
     @patch("core.rag.datasource.vdb.weaviate.weaviate_vector.weaviate")
     def test_search_by_full_text_returns_doc_type_in_metadata(self, mock_weaviate_module):
         """Test that search_by_full_text also returns doc_type in document metadata."""
@@ -268,6 +526,49 @@ class TestWeaviateVector(unittest.TestCase):
         assert len(docs) == 1
         assert docs[0].metadata.get("doc_type") == "image"
 
+    @patch("core.rag.datasource.vdb.weaviate.weaviate_vector.weaviate")
+    def test_search_by_full_text_uses_document_filter(self, mock_weaviate_module):
+        mock_client = MagicMock()
+        mock_client.is_ready.return_value = True
+        mock_weaviate_module.connect_to_custom.return_value = mock_client
+        mock_client.collections.exists.return_value = True
+        mock_col = MagicMock()
+        mock_client.collections.use.return_value = mock_col
+
+        mock_obj = MagicMock()
+        mock_obj.properties = {"text": "bm25 result", "doc_id": "segment-1"}
+        mock_obj.vector = [0.3, 0.4]
+
+        mock_result = MagicMock()
+        mock_result.objects = [mock_obj]
+        mock_col.query.bm25.return_value = mock_result
+
+        wv = WeaviateVector(
+            collection_name=self.collection_name,
+            config=self.config,
+            attributes=self.attributes,
+        )
+        docs = wv.search_by_full_text(query="bm25", document_ids_filter=["doc-1"])
+
+        assert len(docs) == 1
+        assert docs[0].vector == [0.3, 0.4]
+        assert mock_col.query.bm25.call_args.kwargs["filters"] is not None
+
+    @patch("core.rag.datasource.vdb.weaviate.weaviate_vector.weaviate")
+    def test_search_by_full_text_returns_empty_when_collection_is_missing(self, mock_weaviate_module):
+        mock_client = MagicMock()
+        mock_client.is_ready.return_value = True
+        mock_weaviate_module.connect_to_custom.return_value = mock_client
+        mock_client.collections.exists.return_value = False
+
+        wv = WeaviateVector(
+            collection_name=self.collection_name,
+            config=self.config,
+            attributes=self.attributes,
+        )
+
+        assert wv.search_by_full_text(query="missing") == []
+
     @patch("core.rag.datasource.vdb.weaviate.weaviate_vector.weaviate")
     def test_add_texts_stores_doc_type_in_properties(self, mock_weaviate_module):
         """Test that add_texts includes doc_type from document metadata in stored properties."""
@@ -310,6 +611,135 @@ class TestWeaviateVector(unittest.TestCase):
         stored_props = call_kwargs.kwargs.get("properties")
         assert stored_props.get("doc_type") == "image", f"doc_type should be stored in properties, got: {stored_props}"
 
+    @patch("core.rag.datasource.vdb.weaviate.weaviate_vector.weaviate")
+    def test_add_texts_falls_back_to_random_uuid_and_serializes_datetime_metadata(self, mock_weaviate_module):
+        mock_client = MagicMock()
+        mock_client.is_ready.return_value = True
+        mock_weaviate_module.connect_to_custom.return_value = mock_client
+        mock_col = MagicMock()
+        mock_client.collections.use.return_value = mock_col
+
+        mock_batch = MagicMock()
+        mock_batch.__enter__ = MagicMock(return_value=mock_batch)
+        mock_batch.__exit__ = MagicMock(return_value=False)
+        mock_col.batch.dynamic.return_value = mock_batch
+
+        created_at = datetime.datetime(2024, 1, 2, 3, 4, 5, tzinfo=datetime.UTC)
+        doc = Document(page_content="text", metadata={"created_at": created_at})
+
+        wv = WeaviateVector(
+            collection_name=self.collection_name,
+            config=self.config,
+            attributes=self.attributes,
+        )
+
+        with (
+            patch.object(wv, "_get_uuids", return_value=["not-a-uuid"]),
+            patch("core.rag.datasource.vdb.weaviate.weaviate_vector._uuid.uuid4", return_value="fallback-uuid"),
+        ):
+            ids = wv.add_texts(documents=[doc], embeddings=[[]])
+
+        assert ids == ["fallback-uuid"]
+        call_kwargs = mock_batch.add_object.call_args
+        assert call_kwargs.kwargs["uuid"] == "fallback-uuid"
+        assert call_kwargs.kwargs["vector"] is None
+        assert call_kwargs.kwargs["properties"]["created_at"] == created_at.isoformat()
+
+    def test_is_uuid_handles_invalid_values(self):
+        wv = WeaviateVector.__new__(WeaviateVector)
+
+        assert wv._is_uuid("123e4567-e89b-12d3-a456-426614174000") is True
+        assert wv._is_uuid("not-a-uuid") is False
+
+    def test_delete_by_metadata_field_returns_when_collection_is_missing(self):
+        wv = WeaviateVector.__new__(WeaviateVector)
+        wv._collection_name = self.collection_name
+        wv._client = MagicMock()
+        wv._client.collections.exists.return_value = False
+
+        wv.delete_by_metadata_field("doc_id", "segment-1")
+
+        wv._client.collections.use.assert_not_called()
+
+    def test_delete_by_metadata_field_deletes_matching_objects(self):
+        wv = WeaviateVector.__new__(WeaviateVector)
+        wv._collection_name = self.collection_name
+        wv._client = MagicMock()
+        wv._client.collections.exists.return_value = True
+        mock_col = MagicMock()
+        wv._client.collections.use.return_value = mock_col
+
+        wv.delete_by_metadata_field("doc_id", "segment-1")
+
+        mock_col.data.delete_many.assert_called_once()
+
+    def test_delete_removes_collection_when_present(self):
+        wv = WeaviateVector.__new__(WeaviateVector)
+        wv._collection_name = self.collection_name
+        wv._client = MagicMock()
+        wv._client.collections.exists.return_value = True
+
+        wv.delete()
+
+        wv._client.collections.delete.assert_called_once_with(self.collection_name)
+
+    def test_text_exists_handles_missing_and_present_documents(self):
+        wv = WeaviateVector.__new__(WeaviateVector)
+        wv._collection_name = self.collection_name
+        wv._client = MagicMock()
+        wv._client.collections.exists.side_effect = [False, True]
+        mock_col = MagicMock()
+        wv._client.collections.use.return_value = mock_col
+        mock_col.query.fetch_objects.return_value = SimpleNamespace(objects=[SimpleNamespace()])
+
+        assert wv.text_exists("segment-1") is False
+        assert wv.text_exists("segment-1") is True
+
+    def test_delete_by_ids_handles_missing_collections_and_404s(self):
+        class FakeUnexpectedStatusCodeError(Exception):
+            def __init__(self, status_code):
+                super().__init__(f"status={status_code}")
+                self.status_code = status_code
+
+        wv = WeaviateVector.__new__(WeaviateVector)
+        wv._collection_name = self.collection_name
+        wv._client = MagicMock()
+        wv._client.collections.exists.side_effect = [False, True]
+        mock_col = MagicMock()
+        wv._client.collections.use.return_value = mock_col
+        mock_col.data.delete_by_id.side_effect = [FakeUnexpectedStatusCodeError(404), None]
+
+        with patch.object(weaviate_vector_module, "UnexpectedStatusCodeError", FakeUnexpectedStatusCodeError):
+            wv.delete_by_ids(["ignored"])
+            wv.delete_by_ids(["missing-id", "ok-id"])
+
+        assert mock_col.data.delete_by_id.call_count == 2
+
+    def test_delete_by_ids_reraises_non_404_errors(self):
+        class FakeUnexpectedStatusCodeError(Exception):
+            def __init__(self, status_code):
+                super().__init__(f"status={status_code}")
+                self.status_code = status_code
+
+        wv = WeaviateVector.__new__(WeaviateVector)
+        wv._collection_name = self.collection_name
+        wv._client = MagicMock()
+        wv._client.collections.exists.return_value = True
+        mock_col = MagicMock()
+        wv._client.collections.use.return_value = mock_col
+        mock_col.data.delete_by_id.side_effect = FakeUnexpectedStatusCodeError(500)
+
+        with patch.object(weaviate_vector_module, "UnexpectedStatusCodeError", FakeUnexpectedStatusCodeError):
+            with pytest.raises(FakeUnexpectedStatusCodeError, match="status=500"):
+                wv.delete_by_ids(["bad-id"])
+
+    def test_json_serializable_converts_datetime(self):
+        wv = WeaviateVector.__new__(WeaviateVector)
+        created_at = datetime.datetime(2024, 1, 2, 3, 4, 5, tzinfo=datetime.UTC)
+
+        assert wv._json_serializable(created_at) == created_at.isoformat()
+        assert wv._json_serializable("plain") == "plain"
+
 
 class TestVectorDefaultAttributes(unittest.TestCase):
     """Tests for Vector class default attributes list."""
@@ -331,5 +761,65 @@ class TestVectorDefaultAttributes(unittest.TestCase):
         assert "doc_type" in vector._attributes, f"doc_type should be in default attributes, got: {vector._attributes}"
 
 
+class TestWeaviateVectorFactory(unittest.TestCase):
+    def test_init_vector_uses_existing_dataset_index_struct(self):
+        dataset = SimpleNamespace(
+            id="dataset-1",
+            index_struct_dict={"vector_store": {"class_prefix": "ExistingCollection_Node"}},
+            index_struct=None,
+        )
+        attributes = ["doc_id"]
+
+        with (
+            patch.object(weaviate_vector_module.dify_config, "WEAVIATE_ENDPOINT", "http://localhost:8080"),
+            patch.object(weaviate_vector_module.dify_config, "WEAVIATE_GRPC_ENDPOINT", "localhost:50051"),
+            patch.object(weaviate_vector_module.dify_config, "WEAVIATE_API_KEY", "api-key"),
+            patch.object(weaviate_vector_module.dify_config, "WEAVIATE_BATCH_SIZE", 88),
+            patch(
+                "core.rag.datasource.vdb.weaviate.weaviate_vector.WeaviateVector", return_value="vector"
+            ) as mock_vector,
+        ):
+            factory = weaviate_vector_module.WeaviateVectorFactory()
+            result = factory.init_vector(dataset, attributes, MagicMock())
+
+        assert result == "vector"
+        config = mock_vector.call_args.kwargs["config"]
+        assert mock_vector.call_args.kwargs["collection_name"] == "ExistingCollection_Node"
+        assert mock_vector.call_args.kwargs["attributes"] == attributes
+        assert config.endpoint == "http://localhost:8080"
+        assert config.grpc_endpoint == "localhost:50051"
+        assert config.api_key == "api-key"
+        assert config.batch_size == 88
+        assert dataset.index_struct is None
+
+    def test_init_vector_generates_collection_and_updates_index_struct(self):
+        dataset = SimpleNamespace(id="dataset-2", index_struct_dict=None, index_struct=None)
+        attributes = ["doc_id", "doc_type"]
+
+        with (
+            patch.object(weaviate_vector_module.dify_config, "WEAVIATE_ENDPOINT", "http://localhost:8080"),
+            patch.object(weaviate_vector_module.dify_config, "WEAVIATE_GRPC_ENDPOINT", ""),
+            patch.object(weaviate_vector_module.dify_config, "WEAVIATE_API_KEY", None),
+            patch.object(weaviate_vector_module.dify_config, "WEAVIATE_BATCH_SIZE", 100),
+            patch.object(
+                weaviate_vector_module.Dataset,
+                "gen_collection_name_by_id",
+                return_value="GeneratedCollection_Node",
+            ),
+            patch(
+                "core.rag.datasource.vdb.weaviate.weaviate_vector.WeaviateVector", return_value="vector"
+            ) as mock_vector,
+        ):
+            factory = weaviate_vector_module.WeaviateVectorFactory()
+            result = factory.init_vector(dataset, attributes, MagicMock())
+
+        assert result == "vector"
+        assert mock_vector.call_args.kwargs["collection_name"] == "GeneratedCollection_Node"
+        assert json.loads(dataset.index_struct) == {
+            "type": weaviate_vector_module.VectorType.WEAVIATE,
+            "vector_store": {"class_prefix": "GeneratedCollection_Node"},
+        }
+
+
 if __name__ == "__main__":
     unittest.main()

api/tests/unit_tests/models/test_app_models.py

@@ -1164,7 +1164,7 @@ class TestConversationStatusCount:
         conversation.id = str(uuid4())
 
         # Mock the database query to return no messages
-        with patch("models.model.db.session.scalars", autospec=True) as mock_scalars:
+        with patch("models.model.db.session.scalars") as mock_scalars:
             mock_scalars.return_value.all.return_value = []
 
             # Act
@@ -1189,7 +1189,7 @@ class TestConversationStatusCount:
         conversation.id = conversation_id
 
         # Mock the database query to return no messages with workflow_run_id
-        with patch("models.model.db.session.scalars", autospec=True) as mock_scalars:
+        with patch("models.model.db.session.scalars") as mock_scalars:
             mock_scalars.return_value.all.return_value = []
 
             # Act
@@ -1274,7 +1274,7 @@ class TestConversationStatusCount:
             return mock_result
 
         # Act & Assert
-        with patch("models.model.db.session.scalars", side_effect=mock_scalars, autospec=True):
+        with patch("models.model.db.session.scalars", side_effect=mock_scalars):
             result = conversation.status_count
 
             # Verify only 2 database queries were made (not N+1)
@@ -1337,7 +1337,7 @@ class TestConversationStatusCount:
             return mock_result
 
         # Act
-        with patch("models.model.db.session.scalars", side_effect=mock_scalars, autospec=True):
+        with patch("models.model.db.session.scalars", side_effect=mock_scalars):
             result = conversation.status_count
 
             # Assert - query should include app_id filter
@@ -1382,7 +1382,7 @@ class TestConversationStatusCount:
             ),
         ]
 
-        with patch("models.model.db.session.scalars", autospec=True) as mock_scalars:
+        with patch("models.model.db.session.scalars") as mock_scalars:
             # Mock the messages query
             def mock_scalars_side_effect(query):
                 mock_result = MagicMock()
@@ -1438,7 +1438,7 @@ class TestConversationStatusCount:
             ),
         ]
 
-        with patch("models.model.db.session.scalars", autospec=True) as mock_scalars:
+        with patch("models.model.db.session.scalars") as mock_scalars:
 
             def mock_scalars_side_effect(query):
                 mock_result = MagicMock()

api/tests/unit_tests/services/plugin/test_oauth_service.py

@@ -13,6 +13,10 @@ import pytest
 from services.plugin.oauth_service import OAuthProxyService
 
 
+def _oauth_proxy_setex_calls(redis_client) -> list:
+    return [call for call in redis_client.setex.call_args_list if call.args[0].startswith("oauth_proxy_context:")]
+
+
 class TestCreateProxyContext:
     def test_stores_context_in_redis_with_ttl(self):
         context_id = OAuthProxyService.create_proxy_context(
@@ -22,8 +26,9 @@ class TestCreateProxyContext:
         assert context_id  # non-empty UUID string
         from extensions.ext_redis import redis_client
 
-        redis_client.setex.assert_called_once()
+        oauth_calls = _oauth_proxy_setex_calls(redis_client)
-        call_args = redis_client.setex.call_args
+        assert len(oauth_calls) == 1
+        call_args = oauth_calls[0]
         key = call_args[0][0]
         ttl = call_args[0][1]
         stored_data = json.loads(call_args[0][2])

api/tests/unit_tests/services/test_app_dsl_service.py

@@ -211,6 +211,7 @@ def test_import_app_overwrite_only_allows_workflow_and_advanced_chat(monkeypatch
 
 def test_import_app_pending_stores_import_info_in_redis():
     service = AppDslService(MagicMock())
+    app_dsl_service.redis_client.setex.reset_mock()
     result = service.import_app(
         account=_account_mock(),
         import_mode=ImportMode.YAML_CONTENT,
@@ -375,10 +376,13 @@ def test_confirm_import_success_deletes_redis_key(monkeypatch):
     created_app = SimpleNamespace(id="confirmed-app", mode=AppMode.WORKFLOW.value, tenant_id="tenant-1")
     monkeypatch.setattr(AppDslService, "_create_or_update_app", lambda *_args, **_kwargs: created_app)
 
+    app_dsl_service.redis_client.delete.reset_mock()
     result = service.confirm_import(import_id="import-1", account=_account_mock())
     assert result.status == ImportStatus.COMPLETED
     assert result.app_id == "confirmed-app"
-    app_dsl_service.redis_client.delete.assert_called_once()
+    app_dsl_service.redis_client.delete.assert_called_once_with(
+        f"{app_dsl_service.IMPORT_INFO_REDIS_KEY_PREFIX}import-1"
+    )
 
 
 def test_confirm_import_exception_returns_failed(monkeypatch):

api/tests/unit_tests/services/test_audio_service.py

@@ -405,7 +405,7 @@ class TestAudioServiceTTS:
             voice="en-US-Neural",
         )
 
-    @patch("services.audio_service.db.session", autospec=True)
+    @patch("services.audio_service.db.session")
     @patch("services.audio_service.ModelManager", autospec=True)
     def test_transcript_tts_with_message_id_success(self, mock_model_manager_class, mock_db_session, factory):
         """Test successful TTS with message ID."""
@@ -549,7 +549,7 @@ class TestAudioServiceTTS:
         with pytest.raises(ValueError, match="Text is required"):
             AudioService.transcript_tts(app_model=app, text=None)
 
-    @patch("services.audio_service.db.session", autospec=True)
+    @patch("services.audio_service.db.session")
     def test_transcript_tts_returns_none_for_invalid_message_id(self, mock_db_session, factory):
         """Test that TTS returns None for invalid message ID format."""
         # Arrange
@@ -564,7 +564,7 @@ class TestAudioServiceTTS:
         # Assert
         assert result is None
 
-    @patch("services.audio_service.db.session", autospec=True)
+    @patch("services.audio_service.db.session")
     def test_transcript_tts_returns_none_for_nonexistent_message(self, mock_db_session, factory):
         """Test that TTS returns None when message doesn't exist."""
         # Arrange
@@ -585,7 +585,7 @@ class TestAudioServiceTTS:
         # Assert
         assert result is None
 
-    @patch("services.audio_service.db.session", autospec=True)
+    @patch("services.audio_service.db.session")
     def test_transcript_tts_returns_none_for_empty_message_answer(self, mock_db_session, factory):
         """Test that TTS returns None when message answer is empty."""
         # Arrange

api/tests/unit_tests/services/test_human_input_delivery_test_service.py

@@ -313,7 +313,8 @@ class TestEmailDeliveryTestHandler:
             recipients=[DeliveryTestEmailRecipient(email="test@example.com", form_token="token123")],
         )
 
-        subs = EmailDeliveryTestHandler._build_substitutions(context=context, recipient_email="test@example.com")
+        with patch.object(dify_config, "APP_WEB_URL", "http://example.com"):
+            subs = EmailDeliveryTestHandler._build_substitutions(context=context, recipient_email="test@example.com")
 
         assert subs["node_title"] == "title"
         assert subs["form_content"] == "content"

api/tests/unit_tests/services/test_tag_service.py

@@ -316,7 +316,7 @@ class TestTagServiceRetrieval:
     - get_tags_by_target_id: Get all tags bound to a specific target
     """
 
-    @patch("services.tag_service.db.session", autospec=True)
+    @patch("services.tag_service.db.session")
     def test_get_tags_with_binding_counts(self, mock_db_session, factory):
         """
         Test retrieving tags with their binding counts.
@@ -373,7 +373,7 @@ class TestTagServiceRetrieval:
         # Verify database query was called
         mock_db_session.query.assert_called_once()
 
-    @patch("services.tag_service.db.session", autospec=True)
+    @patch("services.tag_service.db.session")
     def test_get_tags_with_keyword_filter(self, mock_db_session, factory):
         """
         Test retrieving tags filtered by keyword (case-insensitive).
@@ -427,7 +427,7 @@ class TestTagServiceRetrieval:
         # 2. Additional WHERE clause for keyword filtering
         assert mock_query.where.call_count >= 2, "Keyword filter should add WHERE clause"
 
-    @patch("services.tag_service.db.session", autospec=True)
+    @patch("services.tag_service.db.session")
     def test_get_target_ids_by_tag_ids(self, mock_db_session, factory):
         """
         Test retrieving target IDs by tag IDs.
@@ -483,7 +483,7 @@ class TestTagServiceRetrieval:
         # Verify both queries were executed
         assert mock_db_session.scalars.call_count == 2, "Should execute tag query and binding query"
 
-    @patch("services.tag_service.db.session", autospec=True)
+    @patch("services.tag_service.db.session")
     def test_get_target_ids_with_empty_tag_ids(self, mock_db_session, factory):
         """
         Test that empty tag_ids returns empty list.
@@ -511,7 +511,7 @@ class TestTagServiceRetrieval:
         assert results == [], "Should return empty list for empty input"
         mock_db_session.scalars.assert_not_called(), "Should not query database for empty input"
 
-    @patch("services.tag_service.db.session", autospec=True)
+    @patch("services.tag_service.db.session")
     def test_get_tag_by_tag_name(self, mock_db_session, factory):
         """
         Test retrieving tags by name.
@@ -553,7 +553,7 @@ class TestTagServiceRetrieval:
         assert len(results) == 1, "Should find exactly one tag"
         assert results[0].name == tag_name, "Tag name should match"
 
-    @patch("services.tag_service.db.session", autospec=True)
+    @patch("services.tag_service.db.session")
     def test_get_tag_by_tag_name_returns_empty_for_missing_params(self, mock_db_session, factory):
         """
         Test that missing tag_type or tag_name returns empty list.
@@ -581,7 +581,7 @@ class TestTagServiceRetrieval:
         # Verify no database queries were executed
         mock_db_session.scalars.assert_not_called(), "Should not query database for invalid input"
 
-    @patch("services.tag_service.db.session", autospec=True)
+    @patch("services.tag_service.db.session")
     def test_get_tags_by_target_id(self, mock_db_session, factory):
         """
         Test retrieving tags associated with a specific target.
@@ -654,7 +654,7 @@ class TestTagServiceCRUD:
 
     @patch("services.tag_service.current_user", autospec=True)
     @patch("services.tag_service.TagService.get_tag_by_tag_name", autospec=True)
-    @patch("services.tag_service.db.session", autospec=True)
+    @patch("services.tag_service.db.session")
     @patch("services.tag_service.uuid.uuid4", autospec=True)
     def test_save_tags(self, mock_uuid, mock_db_session, mock_get_tag_by_name, mock_current_user, factory):
         """
@@ -743,7 +743,7 @@ class TestTagServiceCRUD:
 
     @patch("services.tag_service.current_user", autospec=True)
     @patch("services.tag_service.TagService.get_tag_by_tag_name", autospec=True)
-    @patch("services.tag_service.db.session", autospec=True)
+    @patch("services.tag_service.db.session")
     def test_update_tags(self, mock_db_session, mock_get_tag_by_name, mock_current_user, factory):
         """
         Test updating a tag name.
@@ -795,7 +795,7 @@ class TestTagServiceCRUD:
 
     @patch("services.tag_service.current_user", autospec=True)
     @patch("services.tag_service.TagService.get_tag_by_tag_name", autospec=True)
-    @patch("services.tag_service.db.session", autospec=True)
+    @patch("services.tag_service.db.session")
     def test_update_tags_raises_error_for_duplicate_name(
         self, mock_db_session, mock_get_tag_by_name, mock_current_user, factory
     ):
@@ -827,7 +827,7 @@ class TestTagServiceCRUD:
         with pytest.raises(ValueError, match="Tag name already exists"):
             TagService.update_tags(args, tag_id="tag-123")
 
-    @patch("services.tag_service.db.session", autospec=True)
+    @patch("services.tag_service.db.session")
     def test_update_tags_raises_not_found_for_missing_tag(self, mock_db_session, factory):
         """
         Test that updating a non-existent tag raises NotFound.
@@ -859,7 +859,7 @@ class TestTagServiceCRUD:
                 with pytest.raises(NotFound, match="Tag not found"):
                     TagService.update_tags(args, tag_id="nonexistent")
 
-    @patch("services.tag_service.db.session", autospec=True)
+    @patch("services.tag_service.db.session")
     def test_get_tag_binding_count(self, mock_db_session, factory):
         """
         Test getting the count of bindings for a tag.
@@ -895,7 +895,7 @@ class TestTagServiceCRUD:
         # Verify count matches expectation
         assert result == expected_count, "Binding count should match"
 
-    @patch("services.tag_service.db.session", autospec=True)
+    @patch("services.tag_service.db.session")
     def test_delete_tag(self, mock_db_session, factory):
         """
         Test deleting a tag and its bindings.
@@ -951,7 +951,7 @@ class TestTagServiceCRUD:
         # Verify transaction was committed
         mock_db_session.commit.assert_called_once(), "Should commit transaction"
 
-    @patch("services.tag_service.db.session", autospec=True)
+    @patch("services.tag_service.db.session")
     def test_delete_tag_raises_not_found(self, mock_db_session, factory):
         """
         Test that deleting a non-existent tag raises NotFound.
@@ -999,7 +999,7 @@ class TestTagServiceBindings:
 
     @patch("services.tag_service.current_user", autospec=True)
     @patch("services.tag_service.TagService.check_target_exists", autospec=True)
-    @patch("services.tag_service.db.session", autospec=True)
+    @patch("services.tag_service.db.session")
     def test_save_tag_binding(self, mock_db_session, mock_check_target, mock_current_user, factory):
         """
         Test creating tag bindings.
@@ -1050,7 +1050,7 @@ class TestTagServiceBindings:
 
     @patch("services.tag_service.current_user", autospec=True)
     @patch("services.tag_service.TagService.check_target_exists", autospec=True)
-    @patch("services.tag_service.db.session", autospec=True)
+    @patch("services.tag_service.db.session")
     def test_save_tag_binding_is_idempotent(self, mock_db_session, mock_check_target, mock_current_user, factory):
         """
         Test that saving duplicate bindings is idempotent.
@@ -1090,7 +1090,7 @@ class TestTagServiceBindings:
         mock_db_session.add.assert_not_called(), "Should not create duplicate binding"
 
     @patch("services.tag_service.TagService.check_target_exists", autospec=True)
-    @patch("services.tag_service.db.session", autospec=True)
+    @patch("services.tag_service.db.session")
     def test_delete_tag_binding(self, mock_db_session, mock_check_target, factory):
         """
         Test deleting a tag binding.
@@ -1138,7 +1138,7 @@ class TestTagServiceBindings:
         mock_db_session.commit.assert_called_once(), "Should commit transaction"
 
     @patch("services.tag_service.TagService.check_target_exists", autospec=True)
-    @patch("services.tag_service.db.session", autospec=True)
+    @patch("services.tag_service.db.session")
     def test_delete_tag_binding_does_nothing_if_not_exists(self, mock_db_session, mock_check_target, factory):
         """
         Test that deleting a non-existent binding is a no-op.
@@ -1175,7 +1175,7 @@ class TestTagServiceBindings:
         mock_db_session.commit.assert_not_called(), "Should not commit if nothing to delete"
 
     @patch("services.tag_service.current_user", autospec=True)
-    @patch("services.tag_service.db.session", autospec=True)
+    @patch("services.tag_service.db.session")
     def test_check_target_exists_for_dataset(self, mock_db_session, mock_current_user, factory):
         """
         Test validating that a dataset target exists.
@@ -1216,7 +1216,7 @@ class TestTagServiceBindings:
         mock_db_session.query.assert_called_once(), "Should query database for dataset"
 
     @patch("services.tag_service.current_user", autospec=True)
-    @patch("services.tag_service.db.session", autospec=True)
+    @patch("services.tag_service.db.session")
     def test_check_target_exists_for_app(self, mock_db_session, mock_current_user, factory):
         """
         Test validating that an app target exists.
@@ -1257,7 +1257,7 @@ class TestTagServiceBindings:
         mock_db_session.query.assert_called_once(), "Should query database for app"
 
     @patch("services.tag_service.current_user", autospec=True)
-    @patch("services.tag_service.db.session", autospec=True)
+    @patch("services.tag_service.db.session")
     def test_check_target_exists_raises_not_found_for_missing_dataset(
         self, mock_db_session, mock_current_user, factory
     ):
@@ -1289,7 +1289,7 @@ class TestTagServiceBindings:
             TagService.check_target_exists("knowledge", "nonexistent")
 
     @patch("services.tag_service.current_user", autospec=True)
-    @patch("services.tag_service.db.session", autospec=True)
+    @patch("services.tag_service.db.session")
     def test_check_target_exists_raises_not_found_for_missing_app(self, mock_db_session, mock_current_user, factory):
         """
         Test that missing app raises NotFound.

api/tests/unit_tests/tasks/test_dataset_indexing_task.py

@@ -59,6 +59,11 @@ def mock_redis():
     # Redis is already mocked globally in conftest.py
     # Reset it for each test
     redis_client.reset_mock()
+    redis_client.get.reset_mock()
+    redis_client.setex.reset_mock()
+    redis_client.delete.reset_mock()
+    redis_client.lpush.reset_mock()
+    redis_client.rpop.reset_mock()
     redis_client.get.return_value = None
     redis_client.setex.return_value = True
     redis_client.delete.return_value = True