root
8af82f8f7f
Some checks failed
构建并部署 AI Agent 服务 / deploy (push) Failing after 5m4s
- 实现了稠密+稀疏混合检索，使用 Qdrant 原生 RRF 融合 - 修复了 retriever.py 的 BaseRetriever 继承和稀疏向量包装问题 - 修复了 pipeline.py 的 Optional 导入问题 - 添加了稀疏 embedder 的缓存配置 - 简化了 vector_store.py，移除不必要的逻辑 - 修复了 start.sh 的 PROJECT_DIR 硬编码和端口配置问题 - 完善了 RAG 检索的测试文件
429 lines
15 KiB
Python
429 lines
15 KiB
Python
"""
|
||
Qdrant 混合检索器模块
|
||
|
||
提供基于 Qdrant 的混合检索（Dense + Sparse）功能，包括：
|
||
- 纯混合检索（无子父文档）
|
||
- 父子文档混合检索（先检索子文档，再返回父文档）
|
||
|
||
核心原理：
|
||
- 使用 Qdrant Universal Query API (query_points)
|
||
- 使用 Prefetch 并行检索多个源
|
||
- 使用 RRF 分数融合
|
||
"""
|
||
|
||
from typing import Dict, Any, Optional, List
|
||
from qdrant_client import QdrantClient, models
|
||
from qdrant_client.http.exceptions import UnexpectedResponse
|
||
from langchain_core.documents import Document
|
||
from langchain_core.embeddings import Embeddings
|
||
from langchain_core.retrievers import BaseRetriever, RetrieverOutput
|
||
from pydantic import Field, PrivateAttr
|
||
|
||
from rag_core import QdrantVectorStore, get_sparse_embedder, create_docstore
|
||
from rag_core.client import create_qdrant_client as create_core_qdrant_client
|
||
from app.model_services import get_embedding_service
|
||
from app.logger import info, warning, debug
|
||
|
||
|
||
# 模块级常量
|
||
DEFAULT_SEARCH_K = 20
|
||
DEFAULT_PARENT_SEARCH_K = 5
|
||
|
||
|
||
class HybridRetriever(BaseRetriever):
|
||
"""
|
||
混合检索器：稠密向量 + BM25 稀疏向量 RRF 分数融合
|
||
|
||
使用 Qdrant Universal Query API (query_points)
|
||
"""
|
||
collection_name: str = Field(description="Qdrant 集合名称")
|
||
search_k: int = Field(default=DEFAULT_SEARCH_K, description="检索返回结果数")
|
||
|
||
_vector_store: Any = PrivateAttr()
|
||
_client: Any = PrivateAttr()
|
||
_sparse_embedder: Any = PrivateAttr()
|
||
|
||
def __init__(
|
||
self,
|
||
collection_name: str,
|
||
vector_store: QdrantVectorStore,
|
||
search_k: int = DEFAULT_SEARCH_K,
|
||
):
|
||
"""
|
||
Args:
|
||
collection_name: Qdrant 集合名称
|
||
vector_store: QdrantVectorStore 实例
|
||
search_k: 检索返回结果数
|
||
"""
|
||
super().__init__(
|
||
collection_name=collection_name,
|
||
search_k=search_k
|
||
)
|
||
self._vector_store = vector_store
|
||
self._client = vector_store.get_qdrant_client()
|
||
self._sparse_embedder = get_sparse_embedder()
|
||
|
||
def _get_relevant_documents(
|
||
self, query: str, **kwargs
|
||
) -> List[Document]:
|
||
"""
|
||
同步检索相关文档
|
||
|
||
Args:
|
||
query: 查询字符串
|
||
|
||
Returns:
|
||
相关文档列表
|
||
"""
|
||
# 1. 生成双向量
|
||
dense_query = self._vector_store.embeddings.embed_query(query)
|
||
sparse_query = self._sparse_embedder.embed_query(query)
|
||
sparse_vec = models.SparseVector(
|
||
indices=sparse_query["indices"],
|
||
values=sparse_query["values"]
|
||
)
|
||
|
||
# 2. 使用官方的 query_points API（推荐方式）
|
||
response = self._client.query_points(
|
||
collection_name=self.collection_name,
|
||
prefetch=[ # 并行预取多个检索源
|
||
models.Prefetch(
|
||
query=dense_query,
|
||
using="dense", # 使用稠密向量进行语义搜索
|
||
limit=self.search_k
|
||
),
|
||
models.Prefetch(
|
||
query=sparse_vec,
|
||
using="sparse", # 使用稀疏向量进行关键词搜索
|
||
limit=self.search_k
|
||
)
|
||
],
|
||
query=models.FusionQuery(fusion=models.Fusion.RRF), # 指定融合算法为 RRF
|
||
limit=self.search_k, # 最终返回的结果数量
|
||
with_payload=True
|
||
)
|
||
|
||
# 3. 转换结果
|
||
results = []
|
||
for point in response.points:
|
||
doc = Document(
|
||
page_content=point.payload.pop("page_content", point.payload.pop("text", "")),
|
||
metadata=point.payload
|
||
)
|
||
results.append(doc)
|
||
|
||
debug(f"混合检索返回 {len(results)} 个文档")
|
||
return results
|
||
|
||
async def _aget_relevant_documents(
|
||
self, query: str, **kwargs
|
||
) -> List[Document]:
|
||
"""异步检索（当前调用同步版本）"""
|
||
# Qdrant 客户端没有原生 async，这里用同步版本
|
||
return self._get_relevant_documents(query, **kwargs)
|
||
|
||
|
||
class ParentHybridRetriever(BaseRetriever):
|
||
"""
|
||
父子文档混合检索器：
|
||
|
||
1. 先用混合检索找到相关子文档
|
||
2. 根据子文档的 parent_id 找到对应的父文档
|
||
3. 去重并返回父文档
|
||
"""
|
||
|
||
collection_name: str = Field(description="Qdrant 集合名称")
|
||
search_k: int = Field(default=DEFAULT_PARENT_SEARCH_K, description="检索返回结果数")
|
||
|
||
_vector_store: Any = PrivateAttr()
|
||
_client: Any = PrivateAttr()
|
||
_sparse_embedder: Any = PrivateAttr()
|
||
_docstore: Any = PrivateAttr()
|
||
|
||
def __init__(
|
||
self,
|
||
collection_name: str,
|
||
vector_store: QdrantVectorStore,
|
||
search_k: int = DEFAULT_PARENT_SEARCH_K,
|
||
docstore: Optional[Any] = None,
|
||
):
|
||
"""
|
||
Args:
|
||
collection_name: Qdrant 集合名称
|
||
vector_store: QdrantVectorStore 实例
|
||
search_k: 最终返回的父文档数量
|
||
docstore: 文档存储（如果父文档在 PostgreSQL），可选
|
||
"""
|
||
super().__init__(
|
||
collection_name=collection_name,
|
||
search_k=search_k
|
||
)
|
||
self._vector_store = vector_store
|
||
self._client = vector_store.get_qdrant_client()
|
||
self._sparse_embedder = get_sparse_embedder()
|
||
self._docstore = docstore
|
||
|
||
def _get_relevant_documents(
|
||
self, query: str, **kwargs
|
||
) -> List[Document]:
|
||
"""
|
||
同步检索相关父文档
|
||
|
||
Args:
|
||
query: 查询字符串
|
||
|
||
Returns:
|
||
相关父文档列表
|
||
"""
|
||
# 1. 生成查询双向量
|
||
dense_query = self._vector_store.embeddings.embed_query(query)
|
||
sparse_query = self._sparse_embedder.embed_query(query)
|
||
sparse_vec = models.SparseVector(
|
||
indices=sparse_query["indices"],
|
||
values=sparse_query["values"]
|
||
)
|
||
|
||
# 2. 多取一些子文档，避免去重后数量不足
|
||
search_limit = self.search_k * 2
|
||
|
||
# 3. 使用 query_points API 进行混合检索
|
||
response = self._client.query_points(
|
||
collection_name=self.collection_name,
|
||
prefetch=[
|
||
models.Prefetch(
|
||
query=dense_query,
|
||
using="dense",
|
||
limit=search_limit
|
||
),
|
||
models.Prefetch(
|
||
query=sparse_vec,
|
||
using="sparse",
|
||
limit=search_limit
|
||
)
|
||
],
|
||
query=models.FusionQuery(fusion=models.Fusion.RRF),
|
||
limit=search_limit,
|
||
with_payload=True
|
||
)
|
||
|
||
if not response.points:
|
||
debug("混合检索未找到任何文档")
|
||
return []
|
||
|
||
# 4. 收集 parent_id 和对应最高得分
|
||
parent_score_map = {}
|
||
parent_ids = set()
|
||
child_point_map = {} # 保存子文档点用于降级
|
||
|
||
for point in response.points:
|
||
# 先复制 payload，避免修改原始对象
|
||
payload_copy = point.payload.copy()
|
||
parent_id = payload_copy.get("parent_id", point.id)
|
||
score = point.score
|
||
|
||
# 同一个 parent_id 只保留最高得分
|
||
if parent_id not in parent_score_map or score > parent_score_map[parent_id]:
|
||
parent_score_map[parent_id] = score
|
||
parent_ids.add(parent_id)
|
||
child_point_map[parent_id] = point
|
||
|
||
# 5. 批量查询父文档
|
||
# 首先尝试从 Qdrant 直接查询（因为父文档可能也在 Qdrant 中）
|
||
parent_docs = []
|
||
found_parent_ids = set()
|
||
|
||
try:
|
||
parent_points = self._client.retrieve(
|
||
collection_name=self.collection_name,
|
||
ids=list(parent_ids),
|
||
with_payload=True
|
||
)
|
||
|
||
# 处理找到的父文档
|
||
for point in parent_points:
|
||
payload_copy = point.payload.copy()
|
||
doc = Document(
|
||
page_content=payload_copy.pop("page_content", payload_copy.pop("text", "")),
|
||
metadata=payload_copy
|
||
)
|
||
parent_docs.append(doc)
|
||
found_parent_ids.add(point.id)
|
||
|
||
except Exception as e:
|
||
warning(f"从 Qdrant 查询父文档失败: {e}")
|
||
|
||
# 6. 如果有 docstore，尝试从 docstore 查询剩余的父文档
|
||
if self._docstore and len(found_parent_ids) < len(parent_ids):
|
||
missing_parent_ids = parent_ids - found_parent_ids
|
||
try:
|
||
docstore_docs = self._docstore.mget(missing_parent_ids)
|
||
for doc_id, doc in zip(missing_parent_ids, docstore_docs):
|
||
if doc is not None:
|
||
parent_docs.append(doc)
|
||
found_parent_ids.add(doc_id)
|
||
except Exception as e:
|
||
warning(f"从 docstore 查询父文档失败: {e}")
|
||
|
||
# 7. 降级：对于仍未找到的父文档，用子文档本身代替
|
||
missing_parent_ids = parent_ids - found_parent_ids
|
||
if missing_parent_ids:
|
||
warning(f"以下 parent_id 未找到对应的父文档，将返回子文档本身: {missing_parent_ids}")
|
||
for parent_id in missing_parent_ids:
|
||
child_point = child_point_map.get(parent_id)
|
||
if child_point:
|
||
payload_copy = child_point.payload.copy()
|
||
doc = Document(
|
||
page_content=payload_copy.pop("page_content", payload_copy.pop("text", "")),
|
||
metadata=payload_copy
|
||
)
|
||
parent_docs.append(doc)
|
||
|
||
# 8. 按照得分降序排序，返回前 k 个
|
||
parent_docs_with_scores = [
|
||
(doc, parent_score_map.get(doc.metadata.get("id", doc.id), 0.0))
|
||
for doc in parent_docs
|
||
]
|
||
parent_docs_with_scores.sort(key=lambda x: x[1], reverse=True)
|
||
|
||
final_docs = [doc for doc, _ in parent_docs_with_scores[:self.search_k]]
|
||
debug(f"父子文档混合检索返回 {len(final_docs)} 个父文档")
|
||
|
||
return final_docs
|
||
|
||
async def _aget_relevant_documents(
|
||
self, query: str, **kwargs
|
||
) -> List[Document]:
|
||
"""异步检索（当前调用同步版本）"""
|
||
return self._get_relevant_documents(query, **kwargs)
|
||
|
||
|
||
def create_hybrid_retriever(
|
||
collection_name: str,
|
||
search_k: int = DEFAULT_SEARCH_K,
|
||
embeddings: Optional[Embeddings] = None,
|
||
) -> BaseRetriever:
|
||
"""
|
||
创建混合检索器（稠密向量 + BM25 稀疏向量）。
|
||
|
||
这是默认推荐的检索方式，效果最优。
|
||
|
||
Args:
|
||
collection_name: Qdrant 集合名称
|
||
search_k: 检索返回结果数
|
||
embeddings: 可选的嵌入模型实例。若未提供，将自动获取统一嵌入服务。
|
||
|
||
Returns:
|
||
HybridRetriever 实例
|
||
"""
|
||
# 默认使用统一嵌入服务
|
||
if embeddings is None:
|
||
embeddings = get_embedding_service()
|
||
info("✅ 使用统一嵌入服务（本地 llama.cpp → 智谱 API 自动降级）")
|
||
|
||
# 创建向量存储
|
||
vector_store = QdrantVectorStore(collection_name=collection_name, embeddings=embeddings)
|
||
|
||
# 验证集合是否存在
|
||
try:
|
||
vector_store.get_client().get_collection(collection_name)
|
||
except UnexpectedResponse as e:
|
||
if e.status_code == 404:
|
||
warning(f"⚠️ Qdrant 集合 '{collection_name}' 不存在，请先创建并索引文档")
|
||
raise ValueError(f"Qdrant 集合 '{collection_name}' 不存在")
|
||
raise
|
||
|
||
info(f"✅ Qdrant 混合检索器初始化成功（search_k={search_k}）")
|
||
return HybridRetriever(
|
||
collection_name=collection_name,
|
||
vector_store=vector_store,
|
||
search_k=search_k
|
||
)
|
||
|
||
|
||
def create_parent_hybrid_retriever(
|
||
collection_name: str,
|
||
search_k: int = DEFAULT_PARENT_SEARCH_K,
|
||
embeddings: Optional[Embeddings] = None,
|
||
use_docstore: bool = True,
|
||
) -> BaseRetriever:
|
||
"""
|
||
创建父子文档混合检索器（默认推荐）。
|
||
|
||
检索流程：
|
||
1. 混合检索找到相关子文档
|
||
2. 根据 parent_id 找到对应的父文档
|
||
3. 去重并返回父文档
|
||
|
||
Args:
|
||
collection_name: Qdrant 集合名称
|
||
search_k: 最终返回的父文档数量
|
||
embeddings: 可选的嵌入模型实例
|
||
use_docstore: 是否使用 PostgreSQL docstore 存储父文档
|
||
|
||
Returns:
|
||
ParentHybridRetriever 实例
|
||
"""
|
||
# 默认使用统一嵌入服务
|
||
if embeddings is None:
|
||
embeddings = get_embedding_service()
|
||
info("✅ 使用统一嵌入服务（本地 llama.cpp → 智谱 API 自动降级）")
|
||
|
||
# 创建向量存储
|
||
vector_store = QdrantVectorStore(collection_name=collection_name, embeddings=embeddings)
|
||
|
||
# 验证集合是否存在
|
||
try:
|
||
vector_store.get_client().get_collection(collection_name)
|
||
except UnexpectedResponse as e:
|
||
if e.status_code == 404:
|
||
warning(f"⚠️ Qdrant 集合 '{collection_name}' 不存在，请先创建并索引文档")
|
||
raise ValueError(f"Qdrant 集合 '{collection_name}' 不存在")
|
||
raise
|
||
|
||
# 创建 docstore（如果需要）
|
||
docstore = None
|
||
if use_docstore:
|
||
try:
|
||
docstore, _ = create_docstore()
|
||
info("✅ 文档存储初始化成功（PostgreSQL）")
|
||
except Exception as e:
|
||
warning(f"⚠️ 文档存储初始化失败，将不使用 docstore: {e}")
|
||
|
||
info(f"✅ Qdrant 父子文档混合检索器初始化成功（search_k={search_k}）")
|
||
return ParentHybridRetriever(
|
||
collection_name=collection_name,
|
||
vector_store=vector_store,
|
||
search_k=search_k,
|
||
docstore=docstore
|
||
)
|
||
|
||
|
||
def create_base_retriever(
|
||
collection_name: str,
|
||
search_k: int = DEFAULT_SEARCH_K,
|
||
embeddings: Optional[Embeddings] = None,
|
||
) -> BaseRetriever:
|
||
"""
|
||
创建基础稠密检索器（向后兼容）。
|
||
|
||
Args:
|
||
collection_name: Qdrant 集合名称
|
||
search_k: 检索返回结果数
|
||
embeddings: 可选的嵌入模型实例
|
||
|
||
Returns:
|
||
LangChain 的 BaseRetriever 实例
|
||
"""
|
||
# 默认使用统一嵌入服务
|
||
if embeddings is None:
|
||
embeddings = get_embedding_service()
|
||
|
||
vector_store = QdrantVectorStore(collection_name=collection_name, embeddings=embeddings)
|
||
|
||
info(f"✅ Qdrant 基础稠密检索器初始化成功（search_k={search_k}）")
|
||
return vector_store.as_langchain_vectorstore().as_retriever(k=search_k)
|
||
|
||
|
||
# 别名：默认就是父子文档混合检索
|
||
create_retriever = create_parent_hybrid_retriever
|