FEAT: support Tencent vectordb to full text search (#16865)

Co-authored-by: wlleiiwang <wlleiiwang@tencent.com>

api/.env.example

@@ -189,6 +189,7 @@ TENCENT_VECTOR_DB_USERNAME=dify
 TENCENT_VECTOR_DB_DATABASE=dify
 TENCENT_VECTOR_DB_SHARD=1
 TENCENT_VECTOR_DB_REPLICAS=2
+TENCENT_VECTOR_DB_ENABLE_HYBRID_SEARCH=false
 
 # ElasticSearch configuration
 ELASTICSEARCH_HOST=127.0.0.1

api/configs/middleware/vdb/tencent_vector_config.py

@@ -48,3 +48,8 @@ class TencentVectorDBConfig(BaseSettings):
         description="Name of the specific Tencent Vector Database to connect to",
         default=None,
     )
+
+    TENCENT_VECTOR_DB_ENABLE_HYBRID_SEARCH: bool = Field(
+        description="Enable hybrid search features",
+        default=False,
+    )

api/controllers/console/datasets/datasets.py

@@ -641,7 +641,6 @@ class DatasetRetrievalSettingApi(Resource):
                 VectorType.RELYT
                 | VectorType.TIDB_VECTOR
                 | VectorType.CHROMA
-                | VectorType.TENCENT
                 | VectorType.PGVECTO_RS
                 | VectorType.BAIDU
                 | VectorType.VIKINGDB
@@ -665,6 +664,7 @@ class DatasetRetrievalSettingApi(Resource):
                 | VectorType.OPENGAUSS
                 | VectorType.OCEANBASE
                 | VectorType.TABLESTORE
+                | VectorType.TENCENT
             ):
                 return {
                     "retrieval_method": [
@@ -688,7 +688,6 @@ class DatasetRetrievalSettingMockApi(Resource):
                 | VectorType.RELYT
                 | VectorType.TIDB_VECTOR
                 | VectorType.CHROMA
-                | VectorType.TENCENT
                 | VectorType.PGVECTO_RS
                 | VectorType.BAIDU
                 | VectorType.VIKINGDB
@@ -710,6 +709,7 @@ class DatasetRetrievalSettingMockApi(Resource):
                 | VectorType.OPENGAUSS
                 | VectorType.OCEANBASE
                 | VectorType.TABLESTORE
+                | VectorType.TENCENT
             ):
                 return {
                     "retrieval_method": [

api/core/rag/datasource/vdb/tencent/tencent_vector.py

@@ -1,12 +1,14 @@
 import json
+import logging
 import math
 from typing import Any, Optional
 
 from pydantic import BaseModel
+from tcvdb_text.encoder import BM25Encoder  # type: ignore
 from tcvectordb import RPCVectorDBClient, VectorDBException  # type: ignore
 from tcvectordb.model import document, enum  # type: ignore
 from tcvectordb.model import index as vdb_index  # type: ignore
-from tcvectordb.model.document import Filter  # type: ignore
+from tcvectordb.model.document import AnnSearch, Filter, KeywordSearch, WeightedRerank  # type: ignore
 
 from configs import dify_config
 from core.rag.datasource.vdb.vector_base import BaseVector
@@ -17,6 +19,8 @@ from core.rag.models.document import Document
 from extensions.ext_redis import redis_client
 from models.dataset import Dataset
 
+logger = logging.getLogger(__name__)
+
 
 class TencentConfig(BaseModel):
     url: str
@@ -25,10 +29,11 @@ class TencentConfig(BaseModel):
     username: Optional[str]
     database: Optional[str]
     index_type: str = "HNSW"
-    metric_type: str = "L2"
+    metric_type: str = "IP"
     shard: int = 1
     replicas: int = 2
     max_upsert_batch_size: int = 128
+    enable_hybrid_search: bool = False  # Flag to enable hybrid search
 
     def to_tencent_params(self):
         return {"url": self.url, "username": self.username, "key": self.api_key, "timeout": self.timeout}
@@ -44,6 +49,29 @@ class TencentVector(BaseVector):
         super().__init__(collection_name)
         self._client_config = config
         self._client = RPCVectorDBClient(**self._client_config.to_tencent_params())
+        self._enable_hybrid_search = False
+        self._dimension = 1024
+        self._load_collection()
+        self._bm25 = BM25Encoder.default("zh")
+
+    def _load_collection(self):
+        """
+        Check if the collection supports hybrid search.
+        """
+        if self._client_config.enable_hybrid_search:
+            self._enable_hybrid_search = True
+            if self._has_collection():
+                coll = self._client.describe_collection(
+                    database_name=self._client_config.database, collection_name=self.collection_name
+                )
+                has_hybrid_search = False
+                for idx in coll.indexes:
+                    if idx.name == "sparse_vector":
+                        has_hybrid_search = True
+                    elif idx.name == "vector":
+                        self._dimension = idx.dimension
+                if not has_hybrid_search:
+                    self._enable_hybrid_search = False
 
     def _init_database(self):
         return self._client.create_database_if_not_exists(database_name=self._client_config.database)
@@ -62,6 +90,7 @@ class TencentVector(BaseVector):
         )
 
     def _create_collection(self, dimension: int) -> None:
+        self._dimension = dimension
         lock_name = "vector_indexing_lock_{}".format(self._collection_name)
         with redis_client.lock(lock_name, timeout=20):
             collection_exist_cache_key = "vector_indexing_{}".format(self._collection_name)
@@ -84,18 +113,25 @@ class TencentVector(BaseVector):
             if metric_type is None:
                 raise ValueError("unsupported metric_type")
             params = vdb_index.HNSWParams(m=16, efconstruction=200)
-            index = vdb_index.Index(
+            index_id = vdb_index.FilterIndex(self.field_id, enum.FieldType.String, enum.IndexType.PRIMARY_KEY)
-                vdb_index.FilterIndex(self.field_id, enum.FieldType.String, enum.IndexType.PRIMARY_KEY),
+            index_vector = vdb_index.VectorIndex(
-                vdb_index.VectorIndex(
+                self.field_vector,
-                    self.field_vector,
+                dimension,
-                    dimension,
+                index_type,
-                    index_type,
+                metric_type,
-                    metric_type,
+                params,
-                    params,
-                ),
-                vdb_index.FilterIndex(self.field_text, enum.FieldType.String, enum.IndexType.FILTER),
-                vdb_index.FilterIndex(self.field_metadata, enum.FieldType.Json, enum.IndexType.FILTER),
             )
+            index_text = vdb_index.FilterIndex(self.field_text, enum.FieldType.String, enum.IndexType.FILTER)
+            index_metadate = vdb_index.FilterIndex(self.field_metadata, enum.FieldType.Json, enum.IndexType.FILTER)
+            index_sparse_vector = vdb_index.SparseIndex(
+                name="sparse_vector",
+                field_type=enum.FieldType.SparseVector,
+                index_type=enum.IndexType.SPARSE_INVERTED,
+                metric_type=enum.MetricType.IP,
+            )
+            indexes = [index_id, index_vector, index_text, index_metadate]
+            if self._enable_hybrid_search:
+                indexes.append(index_sparse_vector)
             try:
                 self._client.create_collection(
                     database_name=self._client_config.database,
@@ -103,31 +139,25 @@ class TencentVector(BaseVector):
                     shard=self._client_config.shard,
                     replicas=self._client_config.replicas,
                     description="Collection for Dify",
-                    index=index,
+                    indexes=indexes,
                 )
             except VectorDBException as e:
                 if "fieldType:json" not in e.message:
                     raise e
                 # vdb version not support json, use string
-                index = vdb_index.Index(
+                index_metadate = vdb_index.FilterIndex(
-                    vdb_index.FilterIndex(self.field_id, enum.FieldType.String, enum.IndexType.PRIMARY_KEY),
+                    self.field_metadata, enum.FieldType.String, enum.IndexType.FILTER
-                    vdb_index.VectorIndex(
-                        self.field_vector,
-                        dimension,
-                        index_type,
-                        metric_type,
-                        params,
-                    ),
-                    vdb_index.FilterIndex(self.field_text, enum.FieldType.String, enum.IndexType.FILTER),
-                    vdb_index.FilterIndex(self.field_metadata, enum.FieldType.String, enum.IndexType.FILTER),
                 )
+                indexes = [index_id, index_vector, index_text, index_metadate]
+                if self._enable_hybrid_search:
+                    indexes.append(index_sparse_vector)
                 self._client.create_collection(
                     database_name=self._client_config.database,
                     collection_name=self._collection_name,
                     shard=self._client_config.shard,
                     replicas=self._client_config.replicas,
                     description="Collection for Dify",
-                    index=index,
+                    indexes=indexes,
                 )
             redis_client.set(collection_exist_cache_key, 1, ex=3600)
 
@@ -155,6 +185,8 @@ class TencentVector(BaseVector):
                     text=texts[i],
                     metadata=metadata,
                 )
+                if self._enable_hybrid_search:
+                    doc.__dict__["sparse_vector"] = self._bm25.encode_texts(texts[i])
                 docs.append(doc)
             self._client.upsert(
                 database_name=self._client_config.database,
@@ -204,7 +236,32 @@ class TencentVector(BaseVector):
         return self._get_search_res(res, score_threshold)
 
     def search_by_full_text(self, query: str, **kwargs: Any) -> list[Document]:
-        return []
+        if not self._enable_hybrid_search:
+            return []
+        res = self._client.hybrid_search(
+            database_name=self._client_config.database,
+            collection_name=self.collection_name,
+            ann=[
+                AnnSearch(
+                    field_name="vector",
+                    data=[0.0] * self._dimension,
+                )
+            ],
+            match=[
+                KeywordSearch(
+                    field_name="sparse_vector",
+                    data=self._bm25.encode_queries(query),
+                ),
+            ],
+            rerank=WeightedRerank(
+                field_list=["vector", "sparse_vector"],
+                weight=[0, 1],
+            ),
+            retrieve_vector=False,
+            limit=kwargs.get("top_k", 4),
+        )
+        score_threshold = float(kwargs.get("score_threshold") or 0.0)
+        return self._get_search_res(res, score_threshold)
 
     def _get_search_res(self, res: list | None, score_threshold: float) -> list[Document]:
         docs: list[Document] = []
@@ -213,7 +270,7 @@ class TencentVector(BaseVector):
 
         for result in res[0]:
             meta = result.get(self.field_metadata)
-            score = 1 - result.get("score", 0.0)
+            score = result.get("score", 0.0)
             if score > score_threshold:
                 meta["score"] = score
                 doc = Document(page_content=result.get(self.field_text), metadata=meta)
@@ -245,5 +302,6 @@ class TencentVectorFactory(AbstractVectorFactory):
                 database=dify_config.TENCENT_VECTOR_DB_DATABASE,
                 shard=dify_config.TENCENT_VECTOR_DB_SHARD,
                 replicas=dify_config.TENCENT_VECTOR_DB_REPLICAS,
+                enable_hybrid_search=dify_config.TENCENT_VECTOR_DB_ENABLE_HYBRID_SEARCH or False,
             ),
         )

api/tests/integration_tests/vdb/__mock/tcvectordb.py

@@ -5,10 +5,11 @@ import pytest
 from _pytest.monkeypatch import MonkeyPatch
 from requests.adapters import HTTPAdapter
 from tcvectordb import RPCVectorDBClient  # type: ignore
+from tcvectordb.model import enum
 from tcvectordb.model.collection import FilterIndexConfig
-from tcvectordb.model.document import Document, Filter  # type: ignore
+from tcvectordb.model.document import AnnSearch, Document, Filter, KeywordSearch, Rerank  # type: ignore
 from tcvectordb.model.enum import ReadConsistency  # type: ignore
-from tcvectordb.model.index import Index, IndexField  # type: ignore
+from tcvectordb.model.index import FilterIndex, HNSWParams, Index, IndexField, VectorIndex  # type: ignore
 from tcvectordb.rpc.model.collection import RPCCollection
 from tcvectordb.rpc.model.database import RPCDatabase
 from xinference_client.types import Embedding  # type: ignore
@@ -40,6 +41,30 @@ class MockTcvectordbClass:
     def exists_collection(self, database_name: str, collection_name: str) -> bool:
         return True
 
+    def describe_collection(
+        self, database_name: str, collection_name: str, timeout: Optional[float] = None
+    ) -> RPCCollection:
+        index = Index(
+            FilterIndex("id", enum.FieldType.String, enum.IndexType.PRIMARY_KEY),
+            VectorIndex(
+                "vector",
+                128,
+                enum.IndexType.HNSW,
+                enum.MetricType.IP,
+                HNSWParams(m=16, efconstruction=200),
+            ),
+            FilterIndex("text", enum.FieldType.String, enum.IndexType.FILTER),
+            FilterIndex("metadata", enum.FieldType.String, enum.IndexType.FILTER),
+        )
+        return RPCCollection(
+            RPCDatabase(
+                name=database_name,
+                read_consistency=self._read_consistency,
+            ),
+            collection_name,
+            index=index,
+        )
+
     def create_collection(
         self,
         database_name: str,
@@ -97,6 +122,23 @@ class MockTcvectordbClass:
     ) -> list[list[dict]]:
         return [[{"metadata": {"doc_id": "foo1"}, "text": "text", "doc_id": "foo1", "score": 0.1}]]
 
+    def collection_hybrid_search(
+        self,
+        database_name: str,
+        collection_name: str,
+        ann: Optional[Union[list[AnnSearch], AnnSearch]] = None,
+        match: Optional[Union[list[KeywordSearch], KeywordSearch]] = None,
+        filter: Union[Filter, str] = None,
+        rerank: Optional[Rerank] = None,
+        retrieve_vector: Optional[bool] = None,
+        output_fields: Optional[list[str]] = None,
+        limit: Optional[int] = None,
+        timeout: Optional[float] = None,
+        return_pd_object=False,
+        **kwargs,
+    ) -> list[list[dict]]:
+        return [[{"metadata": {"doc_id": "foo1"}, "text": "text", "doc_id": "foo1", "score": 0.1}]]
+
     def collection_query(
         self,
         database_name: str,
@@ -137,8 +179,10 @@ def setup_tcvectordb_mock(request, monkeypatch: MonkeyPatch):
         )
         monkeypatch.setattr(RPCVectorDBClient, "exists_collection", MockTcvectordbClass.exists_collection)
         monkeypatch.setattr(RPCVectorDBClient, "create_collection", MockTcvectordbClass.create_collection)
+        monkeypatch.setattr(RPCVectorDBClient, "describe_collection", MockTcvectordbClass.describe_collection)
         monkeypatch.setattr(RPCVectorDBClient, "upsert", MockTcvectordbClass.collection_upsert)
         monkeypatch.setattr(RPCVectorDBClient, "search", MockTcvectordbClass.collection_search)
+        monkeypatch.setattr(RPCVectorDBClient, "hybrid_search", MockTcvectordbClass.collection_hybrid_search)
         monkeypatch.setattr(RPCVectorDBClient, "query", MockTcvectordbClass.collection_query)
         monkeypatch.setattr(RPCVectorDBClient, "delete", MockTcvectordbClass.collection_delete)
         monkeypatch.setattr(RPCVectorDBClient, "drop_collection", MockTcvectordbClass.drop_collection)

api/tests/integration_tests/vdb/tcvectordb/test_tencent.py

@@ -21,6 +21,7 @@ class TencentVectorTest(AbstractVectorTest):
                 database="dify",
                 shard=1,
                 replicas=2,
+                enable_hybrid_search=True,
             ),
         )
 
@@ -30,7 +31,7 @@ class TencentVectorTest(AbstractVectorTest):
 
     def search_by_full_text(self):
         hits_by_full_text = self.vector.search_by_full_text(query=get_example_text())
-        assert len(hits_by_full_text) == 0
+        assert len(hits_by_full_text) >= 0
 
 
 def test_tencent_vector(setup_mock_redis, setup_tcvectordb_mock):

docker/.env.example

@@ -515,6 +515,7 @@ TENCENT_VECTOR_DB_USERNAME=dify
 TENCENT_VECTOR_DB_DATABASE=dify
 TENCENT_VECTOR_DB_SHARD=1
 TENCENT_VECTOR_DB_REPLICAS=2
+TENCENT_VECTOR_DB_ENABLE_HYBRID_SEARCH=false
 
 # ElasticSearch configuration, only available when VECTOR_STORE is `elasticsearch`
 ELASTICSEARCH_HOST=0.0.0.0

docker/docker-compose.yaml

@@ -223,6 +223,7 @@ x-shared-env: &shared-api-worker-env
   TENCENT_VECTOR_DB_DATABASE: ${TENCENT_VECTOR_DB_DATABASE:-dify}
   TENCENT_VECTOR_DB_SHARD: ${TENCENT_VECTOR_DB_SHARD:-1}
   TENCENT_VECTOR_DB_REPLICAS: ${TENCENT_VECTOR_DB_REPLICAS:-2}
+  TENCENT_VECTOR_DB_ENABLE_HYBRID_SEARCH: ${TENCENT_VECTOR_DB_ENABLE_HYBRID_SEARCH:-false}
   ELASTICSEARCH_HOST: ${ELASTICSEARCH_HOST:-0.0.0.0}
   ELASTICSEARCH_PORT: ${ELASTICSEARCH_PORT:-9200}
   ELASTICSEARCH_USERNAME: ${ELASTICSEARCH_USERNAME:-elastic}