import logging import platform import shutil import subprocess import tempfile import threading import uuid import warnings from datetime import timedelta, datetime, timezone from io import BytesIO from time import sleep from typing import Collection import PyPDF2 import chardet import cv2 import fitz import pandas as pd import redis import requests from PIL import Image from bs4 import BeautifulSoup from django.core.exceptions import ObjectDoesNotExist from django.core.paginator import Paginator, PageNotAnInteger, EmptyPage from django.db import transaction from django.shortcuts import get_object_or_404 from minio import Minio import re import os import markdown from paddleocr import PaddleOCR from pymilvus import connections, Collection, FieldSchema, CollectionSchema, DataType, utility from requests import RequestException from scipy import interpolate from tabulate import tabulate from DCbackend.settings import MILVUS_HOST, MILVUS_PORT, VECTOR_DIMENSION, MILVUS_USER, MILVUS_PASSWORD from base import logger os.environ['HF_ENDPOINT'] = 'https://hf-mirror.com' from scipy.spatial.distance import cosine from django.utils import timezone import traceback from DCbackend.utils.common import success, fail from backend.Service.MinioService import MinioService from backend.models import Knowledgebase, DocumentKbm, File2document, File, Task, TaskSublist, KbmDocumentType import pytesseract import numpy as np import json import time import pika from django.db.models import Count, Case, When, IntegerField, Q, Max from django.conf import settings class DocumentQueue: QUEUE_KEY = "document_process_queue" minio_client = Minio( settings.MINIO_ENDPOINT, access_key=settings.MINIO_ACCESS_KEY, secret_key=settings.MINIO_SECRET_KEY, secure=settings.MINIO_SECURE ) if os.name == 'nt': # Windows pytesseract.pytesseract.tesseract_cmd = r'D:\Program Files\OCR\tesseract.exe' else: # macOS 或 Linux pytesseract.pytesseract.tesseract_cmd = r'/usr/bin/tesseract' class KbmService: bert_model = None bert_tokenizer = None #大模型地址 API_URL = "http://127.0.0.1:11434/api/embeddings" #通用rabbitmq 放入队列 def send_to_rabbitmq(queue_name, message): """ 将消息发送到指定的RabbitMQ队列 """ try: connection = pika.BlockingConnection(pika.ConnectionParameters( host=settings.RABBITMQ_HOST, port=settings.RABBITMQ_PORT, credentials=pika.PlainCredentials( settings.RABBITMQ_USER, settings.RABBITMQ_PASSWORD ) )) channel = connection.channel() channel.queue_declare(queue=queue_name, durable=True) channel.basic_publish( exchange='', routing_key=queue_name, body=json.dumps(message), properties=pika.BasicProperties( delivery_mode=2, # 使消息持久化 ) ) connection.close() logger.info(f"消息已发送到队列 {queue_name}") return True except Exception as e: logger.error(f"发送消息到RabbitMQ时出错: {str(e)}") return False @staticmethod def selectBucketInfo(request): # user_id = request.POST.get("user_id") knowledgebases = Knowledgebase.objects.filter().exclude(status=4).order_by('-create_time').values('id', 'create_time', 'name', 'doc_num', 'description') result = [] for kb in knowledgebases: # 使用一次查询获取所有计数 counts = DocumentKbm.objects.filter(kb_id=kb['id']).aggregate( word_count=Count(Case(When(type__in=['doc', 'docx'], then=1), output_field=IntegerField())), pdf_count=Count(Case(When(type='pdf', then=1), output_field=IntegerField())), excel_count=Count(Case(When(type__in=['xls', 'xlsx'], then=1), output_field=IntegerField())) ) kb_data = kb.copy() # 创建 kb 的副本,以避免修改原始数据 kb_data.update(counts) result.append(kb_data) return success(result) @staticmethod def getFileInfo(request): try: bucket_id = request.POST.get("bucket_id") page = request.POST.get("page", 1) per_page = request.POST.get("pageSize", 10) object_name = request.POST.get("object_name", "") run = request.POST.get("run", "") type = request.POST.get("type", "") doc_type_id = request.POST.get("doc_type_id") if not bucket_id: return fail("bucket_id为空") # 确保 page 和 per_page 是整数 page = int(page) per_page = int(per_page) # # 查询文档并排序 # documents = DocumentKbm.objects.filter( # Q(kb_id=bucket_id) & # Q(name__icontains=object_name)& # Q(run__icontains=run)& # Q(type__icontains=type)& # Q(doc_type_id=doc_type_id)& # ~Q(status=4) # ).order_by('-create_time') # 构建基本查询 query = Q(kb_id=bucket_id) & Q(name__icontains=object_name) & Q(run__icontains=run) & Q(type__icontains=type) & ~Q(status=4) # 如果 doc_type_id 有值,则添加到查询条件 if doc_type_id: query &= Q(doc_type_id=doc_type_id) # 查询文档并排序 documents = DocumentKbm.objects.filter(query).order_by('-create_time') # 创建分页器 paginator = Paginator(documents, per_page) try: # 获取指定页的结果 documents_page = paginator.page(page) except PageNotAnInteger: # 如果页码不是整数,返回第一页 documents_page = paginator.page(1) except EmptyPage: # 如果页码超出范围,返回最后一页 documents_page = paginator.page(paginator.num_pages) # 将查询结果转换为列表 result = list(documents_page.object_list.values()) for info in result: document_id = info['id'] max_page = TaskSublist.objects.filter(doc_id=document_id).aggregate(Max('page_number'))['page_number__max'] info['max_page'] = max_page if max_page is not None else 0 pagination_info = { 'total_count': paginator.count, 'total_pages': paginator.num_pages, 'total_size': per_page, 'current_page': documents_page.number, 'has_next': documents_page.has_next(), 'has_previous': documents_page.has_previous() } data = { 'pagination': pagination_info, 'documents': result } return success(data) except Exception as e: return fail("获取信息失败") @staticmethod @transaction.atomic def updateName(request): try: new_name = request.POST.get("new_name") document_id = request.POST.get("document_id") if not new_name or not document_id: return fail("新名称和文件ID不能为空") # 获取 DocumentKbm 实例并更新 document = get_object_or_404(DocumentKbm, id=document_id) location = document.location # 获取原始文件的扩展名 _, original_extension = os.path.splitext(location) # 检查新名称是否包含扩展名,如果没有则添加原始扩展名 _, new_extension = os.path.splitext(new_name) if not new_extension: new_name = f"{new_name}{original_extension}" document.name = new_name document.save() # 获取关联的 File2document 和 File file2doc = File2document.objects.filter(document_id=document_id).first() if file2doc: file = get_object_or_404(File, id=file2doc.file_id) file.name = new_name file.save() else: # 记录一个警告,因为没有找到关联的 File logger.error(f"Warning: No associated File found for DocumentKbm with id {document_id}") return success("文件名更新成功") except ObjectDoesNotExist: return fail("指定的文件或关联文件不存在") except Exception as e: return fail(f"更新文件名失败: {str(e)}") @staticmethod @transaction.atomic def deleteDocument(request): document_id = request.POST.get("document_id") document = get_object_or_404(DocumentKbm, id=document_id) document.status = 4 document.save() file2doc = File2document.objects.filter(document_id=document_id).first() if file2doc: file = get_object_or_404(File, id=file2doc.file_id) file.status = 4 file.save() kb_id = document.kb_id new_count = DocumentKbm.objects.filter(kb_id=kb_id).exclude(status=4).count() Knowledgebase.objects.filter(id=kb_id).update(doc_num=new_count) try: # 清理milvus # 连接到 Milvus connections.connect( "default", host=MILVUS_HOST, port=MILVUS_PORT, user=MILVUS_USER, password=MILVUS_PASSWORD ) kmb = Knowledgebase.objects.filter(id=kb_id).first() collection = Collection(kmb.location) tasks = TaskSublist.objects.filter(doc_id=document.id) for task in tasks: expr = f'id in [{task.milvus_id}]' collection.delete(expr) return success("删除成功") except Exception as e: return fail(f"删除milvus集合时发生错误: {str(e)}") finally: # 断开 Milvus 连接 connections.disconnect("default") @staticmethod def getUrl(request): try: document_id = request.POST.get("document_id") if not document_id: return fail("文档ID不能为空") # 获取 DocumentKbm 对象 document = get_object_or_404(DocumentKbm, id=document_id) object_name = document.location # 获取对应的 Knowledgebase 对象 knowledgebase = get_object_or_404(Knowledgebase, id=document.kb_id) bucket_name = knowledgebase.location return MinioService.geturl(object_name, bucket_name) except ObjectDoesNotExist: return fail("指定的文档或知识库不存在") except Exception as e: return fail(f"获取URL失败: {str(e)}") #新rabbitmq队列 @staticmethod def analysis(request): document_id = request.POST.get("document_id") start_page = int(request.POST.get('start_page', 1)) end_page = int(request.POST.get('end_page', -1)) max_tokens = int(request.POST.get('max_tokens', 2048)) if max_tokens == 0: max_tokens = 2048 logger.info(f"开始处理文档 ID: {document_id}") try: document = DocumentKbm.objects.get(id=document_id) if int(document.run) in [1, 5]: # 1: 处理中, 5: 等待处理 logger.info(f"文档 {document_id} 已有队列") return success("文档正在处理中或已经处理完成") # 准备消息 message = { 'document_id': document_id, 'start_page': start_page, 'end_page': end_page, 'max_tokens': max_tokens } # 发送消息到队列 if KbmService.send_to_rabbitmq(settings.RABBITMQ_QUEUE_NAME, message): # 更新文档状态为等待处理 document.run = 5 # 5表示等待处理 document.save() logger.info(f"文档 {document_id} 状态已更新为等待处理") return success("文档已添加到处理队列") else: document.run = 4 document.save() return fail("添加文档到处理队列失败") except DocumentKbm.DoesNotExist: logger.error(f"文档 {document_id} 不存在") document.run = 4 document.save() return fail("文档不存在") except Exception as e: logger.error(f"处理文档 {document_id} 时出错: {str(e)}") document.run = 4 document.save() return fail("处理文档时出错") semaphore = threading.Semaphore(4) # @staticmethod # def process_queue(): # logger.info("开始监测RabbitMQ队列") # connection = pika.BlockingConnection(pika.ConnectionParameters( # host=settings.RABBITMQ_HOST, # port=settings.RABBITMQ_PORT, # credentials=pika.PlainCredentials( # settings.RABBITMQ_USER, # settings.RABBITMQ_PASSWORD # ) # )) # channel = connection.channel() # channel.queue_declare(queue=settings.RABBITMQ_QUEUE_NAME, durable=True) # # def callback(ch, method, properties, body): # with KbmService.semaphore: # try: # job = json.loads(body) # document_id = job['document_id'] # start_page = job['start_page'] # end_page = job['end_page'] # max_tokens = job['max_tokens'] # # logger.info(f"开始执行解析文档 {document_id}") # KbmService.async_analysis(document_id, start_page, end_page, max_tokens) # # # 处理成功,确认消息 # ch.basic_ack(delivery_tag=method.delivery_tag) # except Exception as e: # logger.error(f"处理队列消息时发生错误: {str(e)}") # # 处理失败,拒绝消息并重新入队 # ch.basic_nack(delivery_tag=method.delivery_tag, requeue=True) # # # 设置预取计数为4,与最大并发数相匹配 # channel.basic_qos(prefetch_count=4) # channel.basic_consume(queue=settings.RABBITMQ_QUEUE_NAME, on_message_callback=callback) # # logger.info('等待队列消息。要退出请按 CTRL+C') # channel.start_consuming() connection = None channel = None should_stop = False @staticmethod def check_and_process_queue(): KbmService.should_stop = False while not KbmService.should_stop: try: if KbmService.queue_has_messages(): KbmService.process_queue() else: logger.info("队列为空,等待下一次检查...") time.sleep(60) # 等待60秒后再次检查 except Exception as e: logger.error(f"检查队列时发生错误: {str(e)}") time.sleep(60) # 发生错误时,等待60秒后重试 @staticmethod def queue_has_messages(): try: connection = KbmService.create_connection() channel = connection.channel() queue = channel.queue_declare(queue=settings.RABBITMQ_QUEUE_NAME, passive=True) message_count = queue.method.message_count connection.close() return message_count > 0 except Exception as e: logger.error(f"检查队列消息数量时发生错误: {str(e)}") return False @staticmethod def create_connection(): return pika.BlockingConnection(pika.ConnectionParameters( host=settings.RABBITMQ_HOST, port=settings.RABBITMQ_PORT, credentials=pika.PlainCredentials( settings.RABBITMQ_USER, settings.RABBITMQ_PASSWORD ) )) @staticmethod def process_queue(): logger.info("队列中有消息,开始处理...") KbmService.connection = KbmService.create_connection() KbmService.channel = KbmService.connection.channel() KbmService.channel.queue_declare(queue=settings.RABBITMQ_QUEUE_NAME, durable=True) KbmService.channel.basic_qos(prefetch_count=4) KbmService.channel.basic_consume(queue=settings.RABBITMQ_QUEUE_NAME, on_message_callback=KbmService.callback) try: KbmService.channel.start_consuming() except KeyboardInterrupt: KbmService.should_stop = True finally: KbmService.close_connection() @staticmethod def callback(ch, method, properties, body): with KbmService.semaphore: try: job = json.loads(body) document_id = job['document_id'] start_page = job['start_page'] end_page = job['end_page'] max_tokens = job['max_tokens'] logger.info(f"开始执行解析文档 {document_id}") KbmService.async_analysis(document_id, start_page, end_page, max_tokens) ch.basic_ack(delivery_tag=method.delivery_tag) except Exception as e: logger.error(f"处理队列消息时发生错误: {str(e)}") ch.basic_nack(delivery_tag=method.delivery_tag, requeue=True) # 检查是否还有更多消息 if not KbmService.queue_has_messages(): logger.info("队列处理完毕,停止消费...") ch.stop_consuming() @staticmethod def close_connection(): if KbmService.channel: try: KbmService.channel.close() except Exception: pass if KbmService.connection: try: KbmService.connection.close() except Exception: pass KbmService.channel = None KbmService.connection = None @staticmethod def stop_service(): KbmService.should_stop = True if KbmService.channel: KbmService.channel.stop_consuming() KbmService.close_connection() @staticmethod def get_embedding_excel(text, target_dim=768): """ 获取文本的嵌入向量 """ try: if not text or not text.strip(): logging.warning("Empty text provided for embedding. Returning zero vector.") return np.zeros(target_dim).tolist() # 确保文本被正确编码 encoded_text = text.encode('utf-8').decode('utf-8') payload = { "model": "nomic-embed-text:latest", "prompt": encoded_text } headers = {"Content-Type": "application/json"} response = requests.post(KbmService.API_URL, json=payload, headers=headers) logger.info(f"response::::{response}") response.raise_for_status() embedding_data = response.json() if 'embedding' not in embedding_data: raise ValueError(f"API 响应中没有找到嵌入向量. 响应内容: {embedding_data}") embedding = embedding_data['embedding'] original_embedding = np.array(embedding) if len(original_embedding) == target_dim: return original_embedding.tolist() # 如果原始维度不等于目标维度,进行插值 original_indices = np.arange(len(original_embedding)) new_indices = np.linspace(0, len(original_embedding) - 1, target_dim) f = interpolate.interp1d(original_indices, original_embedding) extended_embedding = f(new_indices) return extended_embedding.tolist() except requests.exceptions.RequestException as e: logging.error(f"API 请求错误: {str(e)}") raise except ValueError as e: logging.error(f"值错误: {str(e)}") raise except Exception as e: logging.error(f"获取文本嵌入时发生意外错误: {str(e)}") raise @classmethod def get_embedding_pdf(cls, text, target_dim=768, max_retries=3, backoff_factor=0.3): """ 获取文本的嵌入向量,并填充或截断到目标维度,包含重试机制 """ for attempt in range(max_retries): try: payload = { "model": "nomic-embed-text:latest", "prompt": text } headers = { "Content-Type": "application/json" } response = requests.post(cls.API_URL, json=payload, headers=headers, timeout=30) sleep(0.3) response.raise_for_status() result = response.json() embedding = result.get('embedding') if embedding is None: raise ValueError("API 响应中没有找到嵌入向量") embedding_array = np.array(embedding) current_dim = embedding_array.shape[0] if current_dim < target_dim: padded_embedding = np.pad(embedding_array, (0, target_dim - current_dim), 'constant') logging.info(f"向量已从 {current_dim} 维填充到 {target_dim} 维") return padded_embedding elif current_dim > target_dim: truncated_embedding = embedding_array[:target_dim] logging.info(f"向量已从 {current_dim} 维截断到 {target_dim} 维") return truncated_embedding else: return embedding_array except RequestException as e: logging.error(f"API 请求错误 (尝试 {attempt + 1}/{max_retries}): {e}") if attempt == max_retries - 1: raise time.sleep(backoff_factor * (2 ** attempt)) except ValueError as e: logging.error(f"解析响应错误: {e}") raise except Exception as e: logging.error(f"获取文本嵌入时发生未知错误: {e}") raise raise Exception("达到最大重试次数,无法获取嵌入") @staticmethod def split_text_by_semantic(text, max_tokens, bucket_name, similarity_threshold=0.5, batch_size=1000): logging.info("开始分割文本并保存到向量数据库") try: #object1 object1为后续可能添加的字段 因为无法直接修改名称 备用 source = "知识库" object1 ="some_object1" object2 ="some_object2" # 连接到Milvus connections.connect("default", host=MILVUS_HOST, port=MILVUS_PORT,user=MILVUS_USER,password=MILVUS_PASSWORD) collection_name = f"{bucket_name}" collection = KbmService._get_or_create_collection(collection_name) sentences = KbmService._split_sentences(text) if not sentences: logging.warning("没有找到有效的句子,将文本按最大令牌数分割") sentences = [text[i:i + max_tokens] for i in range(0, len(text), max_tokens)] chunks = [] current_chunk = sentences[0] current_embedding = KbmService.get_embedding_pdf(current_chunk, target_dim=VECTOR_DIMENSION) batch_data = [] for sentence in sentences[1:]: sentence_embedding = KbmService.get_embedding_pdf(sentence, target_dim=VECTOR_DIMENSION) similarity = 1 - cosine(current_embedding, sentence_embedding) if len(current_chunk) + len(sentence) <= max_tokens and similarity >= similarity_threshold: current_chunk += sentence current_embedding = (current_embedding + sentence_embedding) / 2 else: batch_data.append((current_chunk, current_embedding)) if len(batch_data) >= batch_size: ids = KbmService._insert_batch(collection, batch_data,source,object1,object2) sleep(1) logger.info("减少milvus压力睡眠1秒") if ids is not None: chunks.extend([{'content': chunk, 'milvus_id': id} for (chunk, _), id in zip(batch_data, ids)]) else: logging.error("向 Milvus 插入批量数据失败,这批数据将被跳过") batch_data = [] current_chunk = sentence current_embedding = sentence_embedding # 处理最后一个chunk和剩余的batch数据 if current_chunk: batch_data.append((current_chunk, current_embedding)) if batch_data: ids = KbmService._insert_batch(collection, batch_data,source,object1,object2) sleep(1) logger.info("减少milvus压力睡眠1秒") if ids is not None: chunks.extend([{'content': chunk, 'milvus_id': id} for (chunk, _), id in zip(batch_data, ids)]) else: logging.error("向 Milvus 插入批量数据失败,这批数据将被跳过") KbmService._create_index_and_load(collection) logging.info(f"成功将{len(chunks)}个文本块分割并保存到Milvus") return chunks except Exception as e: logging.error(f"处理文本时发生错误: {str(e)}") raise finally: connections.disconnect("default") @staticmethod def _get_or_create_collection(collection_name): if not utility.has_collection(collection_name): fields = [ FieldSchema(name="id", dtype=DataType.INT64, is_primary=True, auto_id=True), FieldSchema(name="source", dtype=DataType.VARCHAR, max_length=65000), FieldSchema(name="object1", dtype=DataType.VARCHAR, max_length=65000), FieldSchema(name="object2", dtype=DataType.VARCHAR, max_length=65000), FieldSchema(name="content", dtype=DataType.VARCHAR, max_length=65000), FieldSchema(name="embedding", dtype=DataType.FLOAT_VECTOR, dim=VECTOR_DIMENSION) ] schema = CollectionSchema(fields, "Semantic text chunks collection") return Collection(name=collection_name, schema=schema) return Collection(name=collection_name) @staticmethod def _split_sentences(text): sentences = re.split('([。!?])', text) sentences = [''.join(i) for i in zip(sentences[0::2], sentences[1::2] + [''])] return [s.strip() for s in sentences if s.strip()] @staticmethod def _insert_batch(collection, batch_data, source, object1, object2): try: entities = [ [source] * len(batch_data), # source [object1]* len(batch_data), # object1 [object2]* len(batch_data), # object2 [chunk for chunk, _ in batch_data], [embedding.tolist() for _, embedding in batch_data] ] # 插入数据并获取插入操作的结果 insert_result = collection.insert(entities) # 获取插入的 ID inserted_ids = insert_result.primary_keys return inserted_ids logging.info(f"成功插入{len(batch_data)}个文本块到Milvus") except Exception as e: logging.error(f"批量插入数据时发生错误: {str(e)}") @staticmethod def _create_index_and_load(collection): index_params = { "index_type": "IVF_FLAT", "metric_type": "L2", "params": {"nlist": 768} } collection.create_index("embedding", index_params) collection.load() @staticmethod def async_analysis(document_id, start_page, end_page, max_tokens): start_time = time.time() excel_status = 1 try: logger.info(f"开始处理文档 {document_id}") # 更新文档状态为处理中 DocumentKbm.objects.filter(id=document_id).update(run=1) document = get_object_or_404(DocumentKbm, id=document_id) object_name = document.location file_extension = object_name.split('.')[-1].lower() knowledgebase = get_object_or_404(Knowledgebase, id=document.kb_id) bucket_name = knowledgebase.location KbmService.clearPreviousData(document_id, bucket_name) response = minio_client.get_object(bucket_name, object_name) file_content = BytesIO(response.read()) if file_extension in ['xls', 'xlsx']: result, excel_status = KbmService.process_excel(file_content, document_id, max_tokens, bucket_name) elif file_extension == 'pdf': result = KbmService.process_pdf(file_content, document_id, max_tokens,bucket_name) # elif file_extension == 'txt': # result = KbmService.process_txt(file_content, document_id, max_tokens, bucket_name) elif file_extension == 'md': result = KbmService.process_markdown(file_content, document_id, max_tokens, bucket_name) elif file_extension in ['doc', 'docx']: pdf_content = KbmService.convert_doc_to_pdf(file_content) result = KbmService.process_pdf(pdf_content, document_id, max_tokens, bucket_name) else: raise ValueError(f"Unsupported file type: {file_extension}") KbmService.saveTask(document_id, len(result)) end_time = time.time() execution_time = round(end_time - start_time, 2) KbmService.updateDocument(max_tokens, len(result), document_id, execution_time) # 更新文档状态 if excel_status == 6: DocumentKbm.objects.filter(id=document_id).update(run=6) # Excel 特殊情况 logger.info(f"文档 {document_id} 更新完成,状态设置为6(Excel特殊情况)") else: DocumentKbm.objects.filter(id=document_id).update(run=3) # 假设3表示成功状态 logger.info(f"文档 {document_id} 更新完成,状态设置为3(成功)") except Exception as e: logger.error(f"Analysis failed for document {document_id}: {str(e)}") logger.error("Exception traceback:") traceback.print_exc() # 更新文档状态为失败 DocumentKbm.objects.filter(id=document_id).update(run=4) logger.info(f"文档 {document_id} 处理完成") @staticmethod def convert_doc_to_pdf(file_content): try: # 创建临时文件 with tempfile.NamedTemporaryFile(delete=False, suffix='.docx') as temp_input: temp_input.write(file_content.getvalue()) temp_input_path = temp_input.name temp_output_dir = tempfile.mkdtemp() # 查找 LibreOffice 路径 libreoffice_path = KbmService.get_libreoffice_path() if not libreoffice_path: raise FileNotFoundError("找不到 LibreOffice 可执行文件") # 转换为 PDF pdf_path = KbmService.run_libreoffice_conversion(libreoffice_path, temp_input_path, temp_output_dir) # 读取 PDF 内容 with open(pdf_path, 'rb') as pdf_file: pdf_content = pdf_file.read() # 读取并返回 PDF 内容 with open(pdf_path, 'rb') as pdf_file: return BytesIO(pdf_file.read()) except Exception as e: logging.error(f"将文档转换为 PDF 时出错: {str(e)}", exc_info=True) return BytesIO() finally: KbmService.cleanup_temp_files(temp_input_path, temp_output_dir) @staticmethod def get_libreoffice_path(): system = platform.system() if system == "Windows": libreoffice_paths = [r"D:\Program Files\libreoffice\program\soffice.exe"] return next((path for path in libreoffice_paths if os.path.exists(path)), None) else: # Linux 或 macOS for path in ['/usr/bin/libreoffice', '/usr/bin/soffice', '/opt/libreoffice/program/soffice']: if os.path.exists(path): return path return shutil.which('libreoffice') or shutil.which('soffice') @staticmethod def run_libreoffice_conversion(libreoffice_path, input_path, output_dir): cmd = [ libreoffice_path, '--headless', '--convert-to', 'pdf:writer_pdf_Export:{"PageSize":{"Width":21000,"Height":29700}}', '--outdir', output_dir, input_path ] try: env = os.environ.copy() env['HOME'] = '/mnt/ql_api/tmp' # 设置一个临时的 HOME 目录 env['LC_ALL'] = 'C' # 设置一个标准的语言环境 sleep(0.3) result = subprocess.run(cmd, check=True, capture_output=True, text=True, timeout=60, env=env) logging.info(f"LibreOffice 转换输出: {result.stdout}") pdf_filename = os.path.splitext(os.path.basename(input_path))[0] + '.pdf' pdf_path = os.path.join(output_dir, pdf_filename) if not os.path.exists(pdf_path): raise FileNotFoundError(f"PDF 文件未生成。输出目录内容: {os.listdir(output_dir)}") # 使用 PyPDF2 检查页数 with open(pdf_path, 'rb') as pdf_file: pdf_reader = PyPDF2.PdfReader(pdf_file) page_count = len(pdf_reader.pages) logging.info(f"生成的 PDF 文件页数: {page_count}") pdf_size = os.path.getsize(pdf_path) if pdf_size < 1000: logging.warning(f"生成的 PDF 文件大小异常小: {pdf_size} bytes") return pdf_path except subprocess.TimeoutExpired: raise TimeoutError("LibreOffice 转换超时") except subprocess.CalledProcessError as e: raise RuntimeError(f"LibreOffice 转换失败: {e.output}") @staticmethod def cleanup_temp_files(temp_input_path, temp_output_dir): if os.path.exists(temp_input_path): os.remove(temp_input_path) if os.path.exists(temp_output_dir): shutil.rmtree(temp_output_dir) @staticmethod def process_excel(file_content, document_id, max_tokens, bucket_name): result = [] collection_name = f"{bucket_name}" status=1 # object1 object1为后续可能添加的字段 因为无法直接修改名称 备用 source = "知识库" object1 = "some_object1" object2 = "some_object2" def warning_catcher(message, category, filename, lineno, file=None, line=None): nonlocal status if category == UserWarning: if "File contains an invalid specification for 0" in str(message) or \ "Defined names for sheet index 0 cannot be located" in str(message): status = 6 logger.error(f"Warning: {message}") warnings.showwarning = warning_catcher try: excel_file = pd.ExcelFile(file_content) except Exception as e: logging.error(f"Error reading Excel file: {str(e)}") return result try: connections.connect("default", host=MILVUS_HOST, port=MILVUS_PORT,user=MILVUS_USER,password=MILVUS_PASSWORD) collection = KbmService._get_or_create_collection(collection_name) for sheet_name in excel_file.sheet_names: df = pd.read_excel(excel_file, sheet_name=sheet_name) if df.empty: logging.warning(f"Sheet '{sheet_name}' is empty. Skipping.") continue logging.info(f"Processing sheet '{sheet_name}' with shape {df.shape}") markdown_content = KbmService._excel_to_markdown(df, sheet_name) chunks = KbmService._split_markdown(markdown_content, max_tokens) for chunk_number, chunk_content in enumerate(chunks, start=1): try: if not chunk_content.strip(): logging.warning(f"Empty chunk {chunk_number} in sheet '{sheet_name}'. Skipping.") continue embedding = KbmService.get_embedding_excel(chunk_content, target_dim=VECTOR_DIMENSION) if isinstance(embedding, (list, np.ndarray)) and len(embedding) == VECTOR_DIMENSION: sleep(1) logger.info("减少milvus压力睡眠1秒") milvus_id = KbmService._insert_data(collection, chunk_content, embedding,source,object1,object2) else: logging.error(f"Invalid embedding format for chunk {chunk_number} of sheet {sheet_name}.") continue KbmService.saveTaskSublist( document_id=document_id, name=f"sheet_{sheet_name}", page_number=1, chunk_number=chunk_number, content=chunk_content, milvus_id=milvus_id ) result.append({ 'page_number': 1, 'chunk_number': chunk_number, }) except Exception as e: logging.error(f"Error processing chunk {chunk_number} of sheet {sheet_name}: {str(e)}") logging.error(f"Chunk content: {chunk_content}") logging.exception("Detailed error information:") except Exception as e: logger.error(f"Error processing Excel file: {str(e)}") logger.error("Detailed error information:") import traceback logger.error(traceback.format_exc()) status = 6 # 设置状态为6,表示处理出错 raise finally: connections.disconnect("default") return result, status #excel转markdown @staticmethod def _excel_to_markdown(df, sheet_name): if df.empty: return f"# {sheet_name}\n\n表格为空" headers = df.columns.tolist() data = df.values.tolist() # 将所有数据转换为字符串 data = [[str(cell) for cell in row] for row in data] markdown = f"# {sheet_name}\n\n" markdown += tabulate(data, headers=headers, tablefmt="pipe", showindex=False) return markdown #excel分割策略 @staticmethod def _split_json(json_str, max_tokens): # 简单的分割策略,可以根据需要优化 data = json.loads(json_str) chunks = [] current_chunk = [] current_size = 0 for item in data: item_str = json.dumps(item) item_size = len(item_str) if current_size + item_size > max_tokens and current_chunk: chunks.append(json.dumps(current_chunk)) current_chunk = [] current_size = 0 current_chunk.append(item) current_size += item_size if current_chunk: chunks.append(json.dumps(current_chunk)) return chunks @staticmethod def _split_markdown(markdown_content, max_tokens): chunks = [] current_chunk = "" lines = markdown_content.split('\n') for line in lines: if len(current_chunk) + len(line) + 1 > max_tokens: if current_chunk: chunks.append(current_chunk.strip()) current_chunk = line else: current_chunk += '\n' + line if current_chunk else line if current_chunk: chunks.append(current_chunk.strip()) return chunks #milvus excel插入格式 @staticmethod def _insert_data(collection, content, embedding,source,object1,object2): try: data = [ [source], # content field [object1], # content field [object2], # content field [content], # content field [embedding] # embedding field ] insert_result = collection.insert(data) logging.info(f"Inserted 1 record into Milvus") return insert_result.primary_keys[0] # 返回插入的 ID except Exception as e: logging.error(f"Error inserting data into Milvus: {str(e)}") raise #创建milvus索引 @staticmethod def _create_index_if_not_exists(collection): if not collection.has_index(): index_params = { "index_type": "IVF_FLAT", "metric_type": "L2", "params": {"nlist": 768} } collection.create_index("embedding", index_params) @staticmethod def process_pdf(file_content, document_id, max_tokens, bucket_name): logger.info(f"开始解析pdf") doc = fitz.open(stream=file_content, filetype="pdf") total_pages = len(doc) # 提取整个文档的文本 full_text = "" image_texts = [] page_images = [] for page_num in range(total_pages): page = doc[page_num] full_text += page.get_text() # OCR识别图片 for img in page.get_images(): xref = img[0] base_image = doc.extract_image(xref) image_data = base_image["image"] logger.info("开始ocr") image_text = KbmService.extract_text_from_image(image_data) sleep(1) logger.info("识别完成,防止调用频繁睡眠1秒") if image_text: image_texts.append(image_text) # 渲染页面为图像并保存 page_image = KbmService.render_page_to_image(page) image_name = KbmService.save_image_to_minio(page_image, bucket_name) page_images.append((page_num + 1, image_name)) # 将图片文本添加到全文中 if image_texts: full_text += " ".join(image_texts) # 对整个文本进行语义分割 text_chunks = KbmService.split_text_by_semantic(full_text, max_tokens, bucket_name) logger.info("分割完成") result = [] for i, chunk in enumerate(text_chunks): # 为每个chunk分配一个页面图像 page_number, image_name = page_images[min(i, len(page_images) - 1)] KbmService.saveTaskSublist( document_id=document_id, name=image_name, page_number=page_number, chunk_number=i + 1, content=chunk['content'], milvus_id=chunk['milvus_id'] ) result.append({ 'page_number': page_number, 'chunk_number': i + 1, }) doc.close() logger.info("解析结束") return result # @staticmethod # def process_pdf(file_content, document_id, start_page, end_page, max_tokens, bucket_name): # print(f"开始解析pdf:::: {str(file_content)}") # doc = fitz.open(stream=file_content, filetype="pdf") # total_pages = len(doc) # # start_page = max(1, start_page) - 1 # end_page = min(total_pages, end_page if end_page > 0 else total_pages) # result = [] # for page_num in range(start_page, end_page): # page = doc[page_num] # text = page.get_text() # # image_texts = [] # for img in page.get_images(): # xref = img[0] # base_image = doc.extract_image(xref) # image_data = base_image["image"] # #OCR识别图片 # print("开始ocr") # image_text = KbmService.extract_text_from_image(image_data) # sleep(1) # print("识别完成,防止调用频繁睡眠1秒") # if image_text: # image_texts.append(image_text) # if image_texts: # text += "".join(image_texts) # # # 使用语义分割替代简单的文本分割 # text_chunks = KbmService.split_text_by_semantic(text, max_tokens,bucket_name) # print("分割完成") # page_image = KbmService.render_page_to_image(page) # print("转化图片") # image_name = KbmService.save_image_to_minio(page_image, bucket_name) # print("上传minio") # for i, chunk in enumerate(text_chunks): # KbmService.saveTaskSublist( # document_id=document_id, # name=image_name, # page_number=page_num + 1, # chunk_number=i + 1, # content=chunk['content'], # milvus_id=chunk['milvus_id'] # ) # result.append({ # 'page_number': page_num + 1, # 'chunk_number': i + 1, # }) # # doc.close() # print("解析结束") # return result #解析markdown @staticmethod def process_markdown(file_content, document_id, max_tokens, bucket_name): logging.info(f"开始解析 Markdown,document_id: {document_id}, max_tokens: {max_tokens}") try: # 检测文件编码 raw_content = file_content.read() detected = chardet.detect(raw_content) encoding = detected['encoding'] logging.info(f"检测到的文件编码: {encoding}") # 解码文件内容 text = raw_content.decode(encoding) logging.info(f"Markdown 文件总字符数: {len(text)}") logging.debug(f"Markdown 文件前100个字符: {text[:100]}") # 将 Markdown 转换为 HTML html = markdown.markdown(text) # 使用 BeautifulSoup 提取纯文本 soup = BeautifulSoup(html, 'html.parser') plain_text = soup.get_text() logging.info(f"提取的纯文本总字符数: {len(plain_text)}") logging.debug(f"提取的纯文本前100个字符: {plain_text[:100]}") if not plain_text.strip(): logging.warning("Markdown 文件内容为空") return [] # 使用 split_text_by_semantic 方法分割文本 text_chunks = KbmService.split_text_by_semantic(plain_text, max_tokens, bucket_name) logging.info(f"分割后的文本块数: {len(text_chunks)}") result = [] for i, chunk in enumerate(text_chunks, 1): KbmService.saveTaskSublist( document_id=document_id, name="markdown_content", page_number=1, chunk_number=i, content=chunk['content'], milvus_id=chunk['milvus_id'] ) result.append({ 'page_number': 1, 'chunk_number': i, }) logging.info(f"Markdown 处理完成,总共生成 {len(result)} 个文本块") return result except Exception as e: logging.error(f"处理 Markdown 时发生错误: {str(e)}") logging.exception("详细错误信息:") return [] @staticmethod def split_text(text, max_tokens): words = text.split() chunks = [] current_chunk = [] current_token_count = 0 for word in words: word_tokens = KbmService.estimate_tokens(word) if current_token_count + word_tokens > max_tokens and current_chunk: chunks.append(' '.join(current_chunk)) current_chunk = [] current_token_count = 0 current_chunk.append(word) current_token_count += word_tokens if current_chunk: chunks.append(' '.join(current_chunk)) return chunks @staticmethod def estimate_tokens(text): return len(re.findall(r'\w+', text)) * 1.3 #OCR图片识别 ocr = PaddleOCR(use_angle_cls=True, lang="ch", use_gpu=False, det_db_thresh=0.3, det_db_box_thresh=0.3) @staticmethod def extract_text_from_image(image_data): try: if isinstance(image_data, BytesIO): image_data = image_data.getvalue() # 读取图像 nparr = np.frombuffer(image_data, np.uint8) image = cv2.imdecode(nparr, cv2.IMREAD_COLOR) # 图像预处理 preprocessed = KbmService.preprocess_image(image) # 使用PaddleOCR进行识别 result = KbmService.ocr.ocr(preprocessed, cls=True) # 提取文本并去除所有空格 if result and isinstance(result[0], list): text = ''.join( [line[1][0].replace(' ', '') for line in result[0] if line[1][1] > 0.5]) # 只保留置信度大于0.5的结果,并去除空格 else: text = "" # 后处理 text = KbmService.post_process_text(text) if text and len(text) > 2: # 假设有意义的文本至少有3个字符 logging.info(f"提取的文本长度: {len(text)}") return text else: logging.info("提取的内容似乎是图像,而不是文本") return "图片" except Exception as e: logging.error(f"从图像提取文本时出错: {str(e)}") return "图片" @staticmethod def preprocess_image(image): # 转换为灰度图像 gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY) # 自适应阈值处理 binary = cv2.adaptiveThreshold(gray, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C, cv2.THRESH_BINARY, 11, 2) # 对二值图像进行膨胀操作,使文字更粗 kernel = np.ones((2, 2), np.uint8) dilated = cv2.dilate(binary, kernel, iterations=1) return dilated def post_process_text(text): # 将连续的冒号或点替换为空格 text = re.sub(r'[:.]+', ' ', text) # 保留中文字符、英文字母、数字、常用标点 text = re.sub(r'[^\u4e00-\u9fff\u3000-\u303fa-zA-Z0-9.,!?;:()"\'\s]', '', text) # 删除连续的数字(3个或更多) text = re.sub(r'\d{3,}', '', text) # 处理多余的空白字符 text = re.sub(r'\s+', ' ', text).strip() # 删除单独的数字,但保留章节编号和有意义的数字 text = re.sub(r'\b(? 0 else -1 # 更新 DocumentKbm 对象 updated = DocumentKbm.objects.filter(id=document_id).update( token_num=max_tokens, chunk_num=total_chunks, progress=progress_status, process_begin_at=timezone.now(), process_duation= execution_time ) if updated: return True, f"Document {document_id} updated successfully." else: return False, f"Document {document_id} not found." except Exception as e: # 如果发生任何错误,事务会自动回滚 return False, f"Error updating document: {str(e)}" #异步调用 @staticmethod def searchTaskInfo(request): document_id = request.POST.get("document_id") page = request.POST.get('page', 1) page_size = request.POST.get('page_size', 10) # 每页显示的项目数,默认为10 taskSublists = TaskSublist.objects.filter(doc_id=document_id).order_by('id') document = get_object_or_404(DocumentKbm, id=document_id) location = document.location knowledgebase = get_object_or_404(Knowledgebase, id=document.kb_id) bucket_name = knowledgebase.location documentUrl = minio_client.presigned_get_object( bucket_name=bucket_name, object_name=location, expires=timedelta(days=1) # URL有效期为1天 ) # 创建分页器 paginator = Paginator(taskSublists, page_size) try: tasks_page = paginator.page(page) except PageNotAnInteger: # 如果页码不是整数,返回第一页 tasks_page = paginator.page(1) except EmptyPage: # 如果页码超出范围,返回最后一页 tasks_page = paginator.page(paginator.num_pages) task_results = [] for task in tasks_page: try: # 生成MinIO对象的预签名URL url = minio_client.presigned_get_object( bucket_name=bucket_name, object_name=task.name, expires=timedelta(days=1) # URL有效期为1天 ) task_results.append({ 'id': task.id, 'doc_id': task.doc_id, 'name': task.name, 'page_number': task.page_number, 'chunk_number': task.chunk_number, 'content': task.content, 'url': url }) except Exception as e: logger.error(f"Error generating URL for object {task.name}: {str(e)}") # 如果生成URL失败,我们仍然添加其他信息,但URL为None task_results.append({ 'id': task.id, 'doc_id': task.doc_id, 'name': task.name, 'page_number': task.page_number, 'chunk_number': task.chunk_number, 'content': task.content, 'url': None }) # 创建包含 documentUrl 和分页信息的最终结果 result = { 'documentUrl': documentUrl, 'tasks': task_results, 'pagination': { 'current_page': tasks_page.number, 'num_pages': paginator.num_pages, 'per_page': page_size, 'total_count': paginator.count, 'has_next': tasks_page.has_next(), 'has_previous': tasks_page.has_previous(), } } return success(result) @staticmethod @transaction.atomic def deleteBucket(request): bucket_id = request.POST.get("bucket_id") if not bucket_id: return fail("Bucket ID 为空") try: # 检查是否存在未删除的文档 active_docs_count = DocumentKbm.objects.filter(kb_id=bucket_id).exclude(status=4).count() if active_docs_count > 0: return fail(f"无法删除知识库,还有 {active_docs_count} 个未删除的文档") # 如果没有未删除的文档,则更新知识库状态 updated_count = Knowledgebase.objects.filter(id=bucket_id).update(status=4,name=bucket_id, location=bucket_id) if updated_count == 0: return fail("指定的知识库不存在") return success("知识库已成功删除") except Exception as e: return fail(f"删除知识库时发生错误: {str(e)}") @staticmethod def getRunStatus(request): document_id = request.POST.get("document_id") run = DocumentKbm.objects.filter(id = document_id).values("run").first() return success(run) @staticmethod def batchAnalysis(request): ids_str = request.POST.get("ids") start_page = int(request.POST.get('start_page', 1)) end_page = int(request.POST.get('end_page', -1)) max_tokens = int(request.POST.get('max_tokens', 2048)) try: # 尝试将字符串解析为 JSON 列表 ids = json.loads(ids_str) if not isinstance(ids, list): return fail("无效输入:'ids'应该是一个列表") results = [] for document_id in ids: sleep(0.1) logger.info("缓解压力沉睡0.1秒") # 为每个 document_id 创建一个新的请求对象 analysis_request = type('AnalysisRequest', (), {})() analysis_request.POST = { 'document_id': document_id, 'start_page': start_page, 'end_page': end_page, 'max_tokens': max_tokens } # 调用 analysis 方法 response = KbmService.analysis(analysis_request) message = response.get('message') return success(message,"已添加到队列") except Exception as e: return fail(f"An error occurred: {str(e)}") # 假设这是您支持的文件后缀名列表 SUPPORTED_SUFFIXES = [ 'txt', 'pdf', 'doc', 'docx', 'xls', 'xlsx', 'md' ] @staticmethod def getSuffixName(request): try: # 获取数据库中的所有不重复的 type 值 db_types = DocumentKbm.objects.values_list('type', flat=True).distinct() # 将数据库中的类型转换为集合 db_types_set = set(db_types) # 将 SUPPORTED_SUFFIXES 转换为集合 supported_set = set(KbmService.SUPPORTED_SUFFIXES) # 合并两个集合,自动去除重复项 combined_set = supported_set.union(db_types_set) # 将结果转换回列表 combined_suffixes = list(combined_set) # 对结果进行排序(可选) combined_suffixes.sort() return success(combined_suffixes) except Exception as e: return fail(f"获取文件后缀名时发生错误: {str(e)}") @staticmethod @transaction.atomic def batchMove(request): ids = json.loads(request.POST.get("ids")) doc_type_id = request.POST.get("doc_type_id") if not doc_type_id: return fail("分类id为空") if not ids: return fail("未传出文件id") type = KbmDocumentType.objects.filter(id=doc_type_id).exclude(status=4).first() if not type: return fail("当前分类不存在") DocumentKbm.objects.filter(id__in=ids).update(doc_type_id=doc_type_id) return success("批量移动成功") @staticmethod @transaction.atomic def moveDocument(doc_id, doc_type_id): # 这个方法可以保持不变,因为它已经是单次更新操作 return DocumentKbm.objects.filter(id=doc_id).update(doc_type_id=doc_type_id) @staticmethod @transaction.atomic def updateKbm(request): id = request.POST.get("id") if not id: return fail("id为空") kmb = Knowledgebase.objects.filter(id=id).first() name = request.POST.get("name") if not name: return fail("名称不能为空") if kmb: kmb.name = name kmb.description = request.POST.get("description","") kmb.save() return success("修改成功") else: return fail("修改失败")