rag_server/routes/blueprint.py

import asyncio
import json
import os
import traceback
import uuid
from typing import Dict, Any

from quart import Blueprint, jsonify, request
from quart_cors import cors

from applications.api import get_basic_embedding
from applications.api import get_img_embedding
from applications.async_task import AutoRechunkTask, BuildGraph
from applications.async_task import ChunkEmbeddingTask, DeleteTask, ChunkBooksTask
from applications.config import (
    DEFAULT_MODEL,
    LOCAL_MODEL_CONFIG,
    BASE_MILVUS_SEARCH_PARAMS,
)
from applications.resource import get_resource_manager
from applications.search import HybridSearch
from applications.utils.chat import RAGChatAgent
from applications.utils.mysql import Dataset, Contents, ContentChunks, ChatResult, Books
from applications.api.qwen import QwenClient
from applications.utils.oss.oss_client import OSSClient
from applications.utils.task.async_task import handle_books, process_question, query_search

server_bp = Blueprint("api", __name__, url_prefix="/api")
server_bp = cors(server_bp, allow_origin="*")


@server_bp.route("/embed", methods=["POST"])
async def embed():
    body = await request.get_json()
    text = body.get("text")
    model_name = body.get("model", DEFAULT_MODEL)
    if not LOCAL_MODEL_CONFIG.get(model_name):
        return jsonify({"error": "error  model"})

    embedding = await get_basic_embedding(text, model_name)
    return jsonify({"embedding": embedding})


@server_bp.route("/img_embed", methods=["POST"])
async def img_embed():
    body = await request.get_json()
    url_list = body.get("url_list")
    if not url_list:
        return jsonify({"error": "error  url_list"})

    embedding = await get_img_embedding(url_list)
    return jsonify(embedding)


@server_bp.route("/delete", methods=["POST"])
async def delete():
    body = await request.get_json()
    level = body.get("level")
    params = body.get("params")
    if not level or not params:
        return jsonify({"error": "error  level or params"})
    resource = get_resource_manager()
    del_task = DeleteTask(resource)
    response = await del_task.deal(level, params)
    return jsonify(response)


@server_bp.route("/chunk", methods=["POST"])
async def chunk():
    body = await request.get_json()
    text = body.get("text", "")
    ori_doc_id = body.get("doc_id")
    text = text.strip()
    if not text:
        return jsonify({"error": "error  text"})
    resource = get_resource_manager()

    # generate doc id
    if ori_doc_id:
        body["re_chunk"] = True
        doc_id = ori_doc_id
    else:
        doc_id = f"doc-{uuid.uuid4()}"

    chunk_task = ChunkEmbeddingTask(doc_id=doc_id, resource=resource)
    doc_id = await chunk_task.deal(body)
    return jsonify({"doc_id": doc_id})


@server_bp.route("/chunk_book", methods=["POST"])
async def chunk_book():
    body = await request.get_json()
    resource = get_resource_manager()
    doc_id = f"doc-{uuid.uuid4()}"
    chunk_task = ChunkBooksTask(doc_id=doc_id, resource=resource)
    doc_id = await chunk_task.deal(body)
    return jsonify({"doc_id": doc_id})


@server_bp.route("/search", methods=["POST"])
async def search():
    """
    filters: Dict[str, Any], # 条件过滤
    query_vec: List[float], # query 的向量
    anns_field: str = "vector_text", # query指定的向量空间
    search_params: Optional[Dict[str, Any]] = None, # 向量距离方式
    query_text: str = None, #是否通过 topic 倒排
    _source=False, # 是否返回元数据
    es_size: int = 10000, #es 第一层过滤数量
    sort_by: str = None, # 排序
    milvus_size: int = 10 # milvus粗排返回数量
    :return:
    """
    body = await request.get_json()

    # 解析数据
    search_type: str = body.get("search_type")
    filters: Dict[str, Any] = body.get("filters", {})
    anns_field: str = body.get("anns_field", "vector_text")
    search_params: Dict[str, Any] = body.get("search_params", BASE_MILVUS_SEARCH_PARAMS)
    query_text: str = body.get("query_text")
    _source: bool = body.get("_source", False)
    es_size: int = body.get("es_size", 10000)
    sort_by: str = body.get("sort_by")
    milvus_size: int = body.get("milvus", 20)
    limit: int = body.get("limit", 10)
    path_between_chunks: dict = body.get("path_between_chunks", {})
    if not query_text:
        return jsonify({"error": "error  query_text"})

    query_vector = await get_basic_embedding(text=query_text, model=DEFAULT_MODEL)
    resource = get_resource_manager()
    search_engine = HybridSearch(
        milvus_pool=resource.milvus_client,
        es_pool=resource.es_client,
        graph_pool=resource.graph_client,
    )
    try:
        match search_type:
            case "base":
                response = await search_engine.base_vector_search(
                    query_vec=query_vector,
                    anns_field=anns_field,
                    search_params=search_params,
                    limit=limit,
                )
                return jsonify(response), 200
            case "hybrid":
                response = await search_engine.hybrid_search(
                    filters=filters,
                    query_vec=query_vector,
                    anns_field=anns_field,
                    search_params=search_params,
                    es_size=es_size,
                    sort_by=sort_by,
                    milvus_size=milvus_size,
                )
                return jsonify(response), 200
            case "hybrid2":
                co_fields = {"Entity": filters["entities"][0]}
                response = await search_engine.hybrid_search_with_graph(
                    filters=filters,
                    query_vec=query_vector,
                    anns_field=anns_field,
                    search_params=search_params,
                    es_size=es_size,
                    sort_by=sort_by,
                    milvus_size=milvus_size,
                    co_occurrence_fields=co_fields,
                    shortest_path_fields=path_between_chunks,
                )
                return jsonify(response), 200
            case "strategy":
                return jsonify({"error": "strategy not implemented"}), 405
            case _:
                return jsonify({"error": "error  search_type"}), 200
    except Exception as e:
        return jsonify({"error": str(e), "traceback": traceback.format_exc()}), 500


@server_bp.route("/dataset/list", methods=["GET"])
async def dataset_list():
    resource = get_resource_manager()
    datasets = await Dataset(resource.mysql_client).select_dataset()

    # 创建所有任务
    tasks = [
        Contents(resource.mysql_client).select_count(dataset["id"])
        for dataset in datasets
    ]
    counts = await asyncio.gather(*tasks)

    # 组装数据
    data_list = [
        {
            "dataset_id": dataset["id"],
            "name": dataset["name"],
            "count": count,
            "created_at": dataset["created_at"].strftime("%Y-%m-%d"),
        }
        for dataset, count in zip(datasets, counts)
    ]

    return jsonify({"status_code": 200, "detail": "success", "data": data_list})


@server_bp.route("/dataset/add", methods=["POST"])
async def add_dataset():
    resource = get_resource_manager()
    dataset_mapper = Dataset(resource.mysql_client)
    # 从请求体里取参数
    body = await request.get_json()
    name = body.get("name")
    if not name:
        return jsonify({"status_code": 400, "detail": "name is required"})
    # 执行新增
    dataset = await dataset_mapper.select_dataset_by_name(name)
    if dataset:
        return jsonify({"status_code": 400, "detail": "name is exist"})
    await dataset_mapper.add_dataset(name)
    new_dataset = await dataset_mapper.select_dataset_by_name(name)
    return jsonify(
        {
            "status_code": 200,
            "detail": "success",
            "data": {"datasetId": new_dataset[0]["id"]},
        }
    )


@server_bp.route("/content/get", methods=["GET"])
async def get_content():
    resource = get_resource_manager()
    contents = Contents(resource.mysql_client)

    # 获取请求参数
    doc_id = request.args.get("docId")
    if not doc_id:
        return jsonify({"status_code": 400, "detail": "doc_id is required", "data": {}})

    # 查询内容
    rows = await contents.select_content_by_doc_id(doc_id)

    if not rows:
        return jsonify({"status_code": 404, "detail": "content not found", "data": {}})

    row = rows[0]

    return jsonify(
        {
            "status_code": 200,
            "detail": "success",
            "data": {
                "title": row.get("title", ""),
                "text": row.get("text", ""),
                "doc_id": row.get("doc_id", ""),
            },
        }
    )


@server_bp.route("/content/list", methods=["GET"])
async def content_list():
    resource = get_resource_manager()
    contents = Contents(resource.mysql_client)

    # 从 URL 查询参数获取分页和过滤参数
    page_num = int(request.args.get("page", 1))
    page_size = int(request.args.get("pageSize", 10))
    dataset_id = request.args.get("datasetId")
    doc_status = int(request.args.get("doc_status", 1))

    # order_by 可以用 JSON 字符串传递
    import json

    order_by_str = request.args.get("order_by", '{"id":"desc"}')
    try:
        order_by = json.loads(order_by_str)
    except Exception:
        order_by = {"id": "desc"}

    # 调用 select_contents，获取分页字典
    result = await contents.select_contents(
        page_num=page_num,
        page_size=page_size,
        dataset_id=dataset_id,
        doc_status=doc_status,
        order_by=order_by,
    )

    # 格式化 entities，只保留必要字段
    entities = [
        {
            "doc_id": row["doc_id"],
            "title": row.get("title") or "",
            "text": row.get("text") or "",
            "statusDesc": "可用" if row.get("status") == 2 else "不可用",
        }
        for row in result["entities"]
    ]

    return jsonify(
        {
            "status_code": 200,
            "detail": "success",
            "data": {
                "entities": entities,
                "total_count": result["total_count"],
                "page": result["page"],
                "page_size": result["page_size"],
                "total_pages": result["total_pages"],
            },
        }
    )


@server_bp.route("/query", methods=["GET"])
async def query():
    query_text = request.args.get("query")
    dataset_ids = request.args.get("datasetIds").split(",")
    search_type = request.args.get("search_type", "hybrid")
    query_results = await query_search(
        query_text=query_text,
        filters={"dataset_id": dataset_ids},
        search_type=search_type,
    )
    resource = get_resource_manager()
    dataset_mapper = Dataset(resource.mysql_client)
    for result in query_results:
        datasets = await dataset_mapper.select_dataset_by_id(result["datasetId"])
        if datasets:
            dataset_name = datasets[0]["name"]
            result["datasetName"] = dataset_name
    data = {"results": query_results}
    return jsonify({"status_code": 200, "detail": "success", "data": data})


@server_bp.route("/chat", methods=["GET"])
async def chat():
    query_text = request.args.get("query")
    dataset_id_strs = request.args.get("datasetIds")
    dataset_ids = dataset_id_strs.split(",")
    search_type = request.args.get("search_type", "hybrid")
    query_results = await query_search(
        query_text=query_text,
        filters={"dataset_id": dataset_ids},
        search_type=search_type,
    )
    resource = get_resource_manager()
    chat_result_mapper = ChatResult(resource.mysql_client)
    dataset_mapper = Dataset(resource.mysql_client)
    for result in query_results:
        datasets = await dataset_mapper.select_dataset_by_id(result["datasetId"])
        if datasets:
            dataset_name = datasets[0]["name"]
            result["datasetName"] = dataset_name

    rag_chat_agent = RAGChatAgent()
    qwen_client = QwenClient()
    chat_result = await rag_chat_agent.chat_with_deepseek(query_text, query_results)
    llm_search = qwen_client.search_and_chat(
        user_prompt=query_text, search_strategy="agent"
    )
    decision = await rag_chat_agent.make_decision(query_text, chat_result, llm_search)
    data = {
        "results": query_results,
        "chat_res": decision["result"],
        "rag_summary": chat_result["summary"],
        "llm_summary": llm_search["content"],
        # "used_tools": decision["used_tools"],
    }
    await chat_result_mapper.insert_chat_result(
        query_text,
        dataset_id_strs,
        json.dumps(query_results, ensure_ascii=False),
        chat_result["summary"],
        chat_result["relevance_score"],
        chat_result["status"],
        llm_search["content"],
        json.dumps(llm_search["search_results"], ensure_ascii=False),
        1,
        decision["result"],
        is_web=1,
    )
    return jsonify({"status_code": 200, "detail": "success", "data": data})


@server_bp.route("/chunk/list", methods=["GET"])
async def chunk_list():
    resource = get_resource_manager()
    content_chunk = ContentChunks(resource.mysql_client)

    # 从 URL 查询参数获取分页和过滤参数
    page_num = int(request.args.get("page", 1))
    page_size = int(request.args.get("pageSize", 10))
    doc_id = request.args.get("docId")
    if not doc_id:
        return jsonify({"status_code": 500, "detail": "docId not found", "data": {}})

    # 调用 select_contents，获取分页字典
    result = await content_chunk.select_chunk_contents(
        page_num=page_num, page_size=page_size, doc_id=doc_id
    )

    if not result:
        return jsonify({"status_code": 500, "detail": "chunk is empty", "data": {}})
    # 格式化 entities，只保留必要字段
    entities = [
        {
            "id": row["id"],
            "chunk_id": row["chunk_id"],
            "doc_id": row["doc_id"],
            "summary": row.get("summary") or "",
            "text": row.get("text") or "",
            "statusDesc": "可用" if row.get("chunk_status") == 2 else "不可用",
        }
        for row in result["entities"]
    ]

    return jsonify(
        {
            "status_code": 200,
            "detail": "success",
            "data": {
                "entities": entities,
                "total_count": result["total_count"],
                "page": result["page"],
                "page_size": result["page_size"],
                "total_pages": result["total_pages"],
            },
        }
    )


@server_bp.route("/auto_rechunk", methods=["GET"])
async def auto_rechunk():
    resource = get_resource_manager()
    auto_rechunk_task = AutoRechunkTask(mysql_client=resource.mysql_client)
    process_cnt = await auto_rechunk_task.deal()
    return jsonify({"status_code": 200, "detail": "success", "cnt": process_cnt})


@server_bp.route("/build_graph", methods=["POST"])
async def delete_task():
    body = await request.get_json()
    doc_id: str = body.get("doc_id")
    dataset_id: str = body.get("dataset_id", 12)
    batch: bool = body.get("batch_process", False)

    resource = get_resource_manager()
    build_graph_task = BuildGraph(
        neo4j=resource.graph_client,
        es_client=resource.es_client,
        mysql_client=resource.mysql_client,
    )
    if batch:
        await build_graph_task.deal_batch(dataset_id)
    else:
        await build_graph_task.deal(doc_id)
    return jsonify({"status_code": 200, "detail": "success", "data": {}})


@server_bp.route("/rag/search", methods=["POST"])
async def rag_search():
    body = await request.get_json()
    query_text = body.get("queryText")
    rag_chat_agent = RAGChatAgent()
    # spilt_query = await rag_chat_agent.split_query(query_text)
    # split_questions = spilt_query["split_questions"]
    # split_questions.append(query_text)
    split_questions = []
    split_questions.append(query_text)
    # 使用asyncio.gather并行处理每个问题
    tasks = [
        process_question(question, query_text, rag_chat_agent)
        for question in split_questions
    ]

    # 等待所有任务完成并收集结果
    data_list = await asyncio.gather(*tasks)
    return jsonify({"status_code": 200, "detail": "success", "data": data_list})



@server_bp.route("/chat/history", methods=["GET"])
async def chat_history():
    page_num = int(request.args.get("page", 1))
    page_size = int(request.args.get("pageSize", 10))
    resource = get_resource_manager()
    chat_result_mapper = ChatResult(resource.mysql_client)
    result = await chat_result_mapper.select_chat_results(page_num, page_size)
    return jsonify(
        {
            "status_code": 200,
            "detail": "success",
            "data": {
                "entities": result["entities"],
                "total_count": result["total_count"],
                "page": result["page"],
                "page_size": result["page_size"],
                "total_pages": result["total_pages"],
            },
        }
    )


@server_bp.route("/upload/file", methods=["POST"])
async def upload_pdf():
    # 获取前端上传的文件
    # 先等待 request.files 属性来确保文件已加载
    files = await request.files

    # 获取文件对象
    file = files.get("file")

    if file:
        # 检查文件扩展名是否是 .pdf
        if not file.filename.lower().endswith(".pdf"):
            return jsonify(
                {
                    "status": "error",
                    "message": "Invalid file format. Only PDF files are allowed.",
                }
            ), 400

        # 获取文件名
        filename = file.filename
        print(filename)

        book_id = f"book-{uuid.uuid4()}"
        # 检查文件的 MIME 类型是否是 application/pdf
        if file.content_type != "application/pdf":
            return jsonify(
                {
                    "status": "error",
                    "message": "Invalid MIME type. Only PDF files are allowed.",
                }
            ), 400

        # 保存到本地（可选，视需要）
        file_path = os.path.join("/tmp", book_id)  # 临时存储路径
        await file.save(file_path)
        resource = get_resource_manager()
        books = Books(resource.mysql_client)
        # 上传到 OSS
        try:
            oss_client = OSSClient()
            # 上传文件到 OSS
            oss_path = f"rag/pdfs/{book_id}"
            oss_client.upload_file(file_path, oss_path)
            await books.insert_book(book_id, filename, oss_path)
            # os.remove(file_path)
            return jsonify(
                {
                    "status": "success",
                    "message": f"File {filename} uploaded successfully to OSS!",
                }
            ), 200
        except Exception as e:
            return jsonify({"status": "error", "message": str(e)}), 500
    else:
        return jsonify({"status": "error", "message": "No file uploaded."}), 400


@server_bp.route("/process/book", methods=["GET"])
async def process_book():
    # 创建异步任务来后台处理书籍
    asyncio.create_task(handle_books())
    # 返回立即响应
    return jsonify({"status": "success", "message": "任务已提交后台处理"}), 200