484 lines
13 KiB
Go
484 lines
13 KiB
Go
package service
|
||
|
||
import (
|
||
"context"
|
||
"fmt"
|
||
"rag/consts/document"
|
||
"rag/consts/public"
|
||
"rag/dao"
|
||
"rag/model/dto"
|
||
"rag/model/entity"
|
||
"strings"
|
||
"sync"
|
||
"time"
|
||
|
||
"gitea.com/red-future/common/beans"
|
||
"gitea.com/red-future/common/db/gfdb"
|
||
"gitea.com/red-future/common/full-text-search/meilisearch"
|
||
"gitea.com/red-future/common/http"
|
||
"gitea.com/red-future/common/rag/eino"
|
||
"gitea.com/red-future/common/rag/gse"
|
||
"gitea.com/red-future/common/utils"
|
||
gmq "github.com/bjang03/gmq/core/gmq"
|
||
"github.com/bjang03/gmq/mq"
|
||
"github.com/bjang03/gmq/types"
|
||
"github.com/cloudwego/eino/schema"
|
||
"github.com/gogf/gf/v2/container/gvar"
|
||
"github.com/gogf/gf/v2/crypto/gmd5"
|
||
"github.com/gogf/gf/v2/database/gdb"
|
||
"github.com/gogf/gf/v2/database/gredis"
|
||
"github.com/gogf/gf/v2/frame/g"
|
||
"github.com/gogf/gf/v2/util/gconv"
|
||
)
|
||
|
||
var Document = new(documentService)
|
||
|
||
type documentService struct{}
|
||
|
||
// Create 创建文件
|
||
func (s *documentService) Create(ctx context.Context, req *dto.CreateDocumentReq) (res *dto.CreateDocumentRes, err error) {
|
||
err = gfdb.DB(ctx).Transaction(ctx, func(ctx context.Context, tx gdb.TX) (err error) {
|
||
var id int64
|
||
id, err = dao.Document.Insert(ctx, req)
|
||
if err != nil {
|
||
return
|
||
}
|
||
datasetReq := &dto.UpdateDatasetReq{
|
||
Id: req.DatasetId,
|
||
DocumentCount: 1,
|
||
DocumentSize: req.FileSize,
|
||
}
|
||
_, err = dao.Dataset.Update(ctx, datasetReq)
|
||
if err != nil {
|
||
return
|
||
}
|
||
res = &dto.CreateDocumentRes{Id: id}
|
||
|
||
return
|
||
})
|
||
|
||
return
|
||
}
|
||
|
||
// Update 更新文件
|
||
func (s *documentService) Update(ctx context.Context, req *dto.UpdateDocumentReq) (err error) {
|
||
_, err = dao.Document.Update(ctx, req)
|
||
return
|
||
}
|
||
|
||
// Delete 删除文件
|
||
func (s *documentService) Delete(ctx context.Context, req *dto.DeleteDocumentReq) (err error) {
|
||
docs, err := dao.Document.GetByID(ctx, &dto.GetDocumentReq{Id: req.Id})
|
||
if err != nil {
|
||
return
|
||
}
|
||
err = gfdb.DB(ctx).Transaction(ctx, func(ctx context.Context, tx gdb.TX) (err error) {
|
||
datasetReq := &dto.UpdateDatasetReq{
|
||
Id: docs.DatasetId,
|
||
DocumentCount: -1,
|
||
DocumentSize: -docs.FileSize,
|
||
}
|
||
_, err = dao.Dataset.Update(ctx, datasetReq)
|
||
if err != nil {
|
||
return
|
||
}
|
||
_, err = dao.Document.Delete(ctx, req)
|
||
return
|
||
})
|
||
|
||
return
|
||
}
|
||
|
||
// Get 获取文件详情
|
||
func (s *documentService) Get(ctx context.Context, req *dto.GetDocumentReq) (res *dto.DocumentVO, err error) {
|
||
r, err := dao.Document.GetByID(ctx, req)
|
||
err = gconv.Struct(r, &res)
|
||
return
|
||
}
|
||
|
||
// List 文件列表
|
||
func (s *documentService) List(ctx context.Context, req *dto.ListDocumentReq) (res *dto.ListDocumentRes, err error) {
|
||
list, total, err := dao.Document.List(ctx, req)
|
||
if err != nil {
|
||
return nil, err
|
||
}
|
||
res = &dto.ListDocumentRes{
|
||
Total: total,
|
||
}
|
||
err = gconv.Struct(list, &res.List)
|
||
|
||
//eino.TestIndexer()
|
||
//eino.TestRetriever()
|
||
|
||
return
|
||
}
|
||
|
||
// Process 处理文件(使用eino框架切分和向量化)
|
||
func (s *documentService) Process(ctx context.Context, req *dto.ProcessDocumentReq) (res *dto.ProcessDocumentRes, err error) {
|
||
startTime := time.Now()
|
||
|
||
// 1. 查询文件信息
|
||
documentReq := dto.GetDocumentReq{Id: req.Id}
|
||
doc, err := dao.Document.GetByID(ctx, &documentReq)
|
||
if err != nil {
|
||
return nil, err
|
||
}
|
||
|
||
// 2. 使用eino框架进行文件切分（并发执行）
|
||
var vectorDocsCount, chunks int64
|
||
// 用 gopool 或者简单的错误等待，绝对不用裸 goroutine
|
||
var err1, err2, err3 error
|
||
var wg sync.WaitGroup
|
||
wg.Add(3)
|
||
|
||
// 任务1
|
||
go func() {
|
||
defer wg.Done()
|
||
vectorDocsCount, chunks, err1 = s.sqlSplitDocument(ctx, doc)
|
||
}()
|
||
|
||
// 任务2
|
||
go func() {
|
||
defer wg.Done()
|
||
err2 = s.esSplitDocument(ctx, doc)
|
||
}()
|
||
|
||
// 任务3
|
||
go func() {
|
||
defer wg.Done()
|
||
err3 = s.extractDocument(ctx, doc)
|
||
}()
|
||
|
||
// 直接等待，不使用通道，避免泄漏
|
||
wg.Wait()
|
||
|
||
updateDocumentReq := new(dto.UpdateDocumentReq)
|
||
updateDocumentReq.Id = req.Id
|
||
|
||
// 统一判断错误
|
||
if err1 != nil || err2 != nil || err3 != nil {
|
||
// 更新文档状态
|
||
updateDocumentReq.VectorStatus = document.VectorStatusFailed.Code()
|
||
if _, err = dao.Document.Update(ctx, updateDocumentReq); err != nil {
|
||
return nil, err
|
||
}
|
||
if err1 != nil {
|
||
return nil, err1
|
||
}
|
||
if err2 != nil {
|
||
return nil, err2
|
||
}
|
||
return nil, err3
|
||
}
|
||
|
||
// 4. 更新文件状态为处理中和切分数量
|
||
if vectorDocsCount > 0 {
|
||
updateDocumentReq.VectorStatus = document.VectorStatusProcessing.Code()
|
||
} else {
|
||
updateDocumentReq.VectorStatus = document.VectorStatusCompleted.Code()
|
||
}
|
||
updateDocumentReq.ChunkCount = chunks
|
||
if _, err = dao.Document.Update(ctx, updateDocumentReq); err != nil {
|
||
return
|
||
}
|
||
|
||
costTime := time.Since(startTime).Milliseconds()
|
||
|
||
return &dto.ProcessDocumentRes{
|
||
ChunkCount: chunks,
|
||
CostTime: costTime,
|
||
}, nil
|
||
}
|
||
|
||
func (s *documentService) extractDocument(ctx context.Context, doc *entity.Document) (err error) {
|
||
// 1. 加载文件
|
||
docs, err := s.loadDocument(ctx, doc)
|
||
if err != nil {
|
||
return
|
||
}
|
||
|
||
var words []gse.Keyword
|
||
if len(docs[0].Content) < 500 {
|
||
words = gse.GseTool.Extract(docs[0].Content, 4)
|
||
} else if len(docs[0].Content) < 2000 {
|
||
words = gse.GseTool.Extract(docs[0].Content, 8)
|
||
} else if len(docs[0].Content) < 5000 {
|
||
words = gse.GseTool.Extract(docs[0].Content, 13)
|
||
} else {
|
||
var docsSplit []*schema.Document
|
||
docsSplit, err = eino.RecursiveSplitDocument(ctx, docs)
|
||
if err != nil {
|
||
return
|
||
}
|
||
for _, t := range docsSplit {
|
||
words = append(words, gse.GseTool.Extract(t.Content, 6)...)
|
||
}
|
||
}
|
||
|
||
var keywordReqs = make([]*dto.CreateKeywordReq, 0)
|
||
for _, word := range words {
|
||
keywordReqs = append(keywordReqs, &dto.CreateKeywordReq{
|
||
DatasetId: doc.DatasetId,
|
||
DocumentId: doc.Id,
|
||
Word: word.Word,
|
||
Weight: gconv.Int16(word.Score),
|
||
})
|
||
}
|
||
if len(keywordReqs) > 0 {
|
||
_, err = dao.Keyword.BatchSaveOrUpdate(ctx, keywordReqs)
|
||
if err != nil {
|
||
return
|
||
}
|
||
}
|
||
return
|
||
}
|
||
|
||
func (s *documentService) sqlSplitDocument(ctx context.Context, doc *entity.Document) (vectorDocsCount, docsSplitCount int64, err error) {
|
||
// 1. 加载文件
|
||
docs, err := s.loadDocument(ctx, doc)
|
||
if err != nil {
|
||
return
|
||
}
|
||
// 2. 语义切分文件
|
||
docsSplit, err := eino.SemanticSplitDocument(ctx, docs)
|
||
if err != nil {
|
||
return
|
||
}
|
||
docsSplitCount = gconv.Int64(len(docsSplit))
|
||
// 2. 获取历史数据
|
||
err = s.getHistoryData(ctx, doc, public.KnowledgeLockSqlKey, public.KnowledgeContentHashSqlKey)
|
||
if err != nil {
|
||
return
|
||
}
|
||
// 3. 组装向量文档
|
||
var vectorDocs = make([]dto.VectorDocumentChunkMsg, 0)
|
||
for i, t := range docsSplit {
|
||
contentHash := gmd5.MustEncryptString(t.Content)
|
||
// 检查是否重复
|
||
var success bool
|
||
success, err = s.checkRepeat(ctx, public.KnowledgeContentHashSqlKey, contentHash)
|
||
if err != nil {
|
||
return
|
||
}
|
||
if !success {
|
||
continue
|
||
}
|
||
vectorDocs = append(vectorDocs, dto.VectorDocumentChunkMsg{
|
||
TenantId: doc.TenantId,
|
||
Creator: doc.Creator,
|
||
DatasetId: doc.DatasetId,
|
||
DocumentId: doc.Id,
|
||
Content: t.Content,
|
||
ContentHash: contentHash,
|
||
ChunkIndex: gconv.Int64(i),
|
||
})
|
||
|
||
}
|
||
// 4. 发送消息到队列
|
||
if len(vectorDocs) > 0 {
|
||
err = gmq.GetGmq("primary").GmqPublish(ctx, &mq.RedisPubMessage{
|
||
PubMessage: types.PubMessage{
|
||
Topic: public.KnowledgeDocumentChunkTopic,
|
||
Data: vectorDocs,
|
||
},
|
||
})
|
||
}
|
||
vectorDocsCount = gconv.Int64(len(vectorDocs))
|
||
return
|
||
}
|
||
|
||
func (s *documentService) esSplitDocument(ctx context.Context, doc *entity.Document) (err error) {
|
||
// 1. 加载文件
|
||
docs, err := s.loadDocument(ctx, doc)
|
||
if err != nil {
|
||
return
|
||
}
|
||
// 2. 递归切分文件
|
||
docsSplit, err := eino.RecursiveSplitDocument(ctx, docs)
|
||
if err != nil {
|
||
return
|
||
}
|
||
// 2. 获取历史数据
|
||
err = s.getHistoryData(ctx, doc, public.KnowledgeLockEsKey, public.KnowledgeContentHashEsKey)
|
||
if err != nil {
|
||
return
|
||
}
|
||
// 3. 组装向量文档并同时构建meilisearch文档
|
||
var meiliDocs = make([]interface{}, 0)
|
||
for i, t := range docsSplit {
|
||
contentHash := gmd5.MustEncryptString(t.Content)
|
||
// 检查是否重复
|
||
var success bool
|
||
success, err = s.checkRepeat(ctx, public.KnowledgeContentHashEsKey, contentHash)
|
||
if err != nil {
|
||
return
|
||
}
|
||
if !success {
|
||
continue
|
||
}
|
||
// 构建Meilisearch文档
|
||
meiliDocs = append(meiliDocs, map[string]interface{}{
|
||
"id": contentHash,
|
||
"datasetId": doc.DatasetId,
|
||
"documentId": doc.Id,
|
||
"content": t.Content,
|
||
"contentHash": contentHash,
|
||
"chunkIndex": i,
|
||
})
|
||
}
|
||
// 4. 写入到meilisearch数据库中
|
||
if len(meiliDocs) > 0 {
|
||
if _, err = meilisearch.DB().InsertMany(ctx, meiliDocs, public.IndexNameDocumentChunk); err != nil {
|
||
g.Log().Errorf(ctx, "写入meilisearch失败: %v", err)
|
||
return
|
||
}
|
||
}
|
||
return
|
||
}
|
||
|
||
// loadDocument 加载文件
|
||
func (s *documentService) loadDocument(ctx context.Context, doc *entity.Document) (docs []*schema.Document, err error) {
|
||
return eino.LoadDocument(ctx, doc.FilePath, doc.Format)
|
||
}
|
||
|
||
// getHistoryData 获取历史数据
|
||
func (s *documentService) getHistoryData(ctx context.Context, doc *entity.Document, lockKey, contentKey string) (err error) {
|
||
docsLockKey := fmt.Sprintf(lockKey, doc.DatasetId)
|
||
success, err := utils.Lock(ctx, docsLockKey, int64(60), func(ctx context.Context) error {
|
||
// 1. 扫描 Redis 中所有 前缀为 rag:knowledge:xxx:contentHash 的 key
|
||
pattern := fmt.Sprintf(contentKey, "*")
|
||
keys, err := g.Redis().Keys(ctx, pattern)
|
||
if err != nil {
|
||
return err
|
||
}
|
||
|
||
// 2. Redis 有数据：只刷新过期时间，不查库
|
||
if len(keys) > 0 {
|
||
// 批量刷新过期时间为 60s
|
||
for _, key := range keys {
|
||
_, err = g.Redis().Expire(ctx, key, 600)
|
||
if err != nil {
|
||
return err
|
||
}
|
||
}
|
||
return nil
|
||
}
|
||
|
||
// 3. Redis 无数据：根据 contentKey 类型选择查询方式
|
||
var dictData = make([]*dto.DocumentChunkRPC, 0)
|
||
if public.KnowledgeContentHashSqlKey == contentKey {
|
||
// SQL 方式：调用 HTTP 接口查询
|
||
dictData, err = s.getHistoryDataFromHttp(ctx, doc)
|
||
} else {
|
||
// ES 方式：查询 meilisearch
|
||
dictData, err = s.getHistoryDataFromMeilisearch(ctx, doc)
|
||
}
|
||
if err != nil {
|
||
return err
|
||
}
|
||
|
||
// 4. 把查询到的数据写入 Redis（600s过期）
|
||
for _, item := range dictData {
|
||
// 去除可能的 JSON 引号
|
||
contentHash := strings.Trim(item.ContentHash, `"`)
|
||
key := fmt.Sprintf(contentKey, contentHash)
|
||
_, err = g.Redis().Set(ctx, key, true, gredis.SetOption{
|
||
TTLOption: gredis.TTLOption{
|
||
EX: gconv.PtrInt64(600),
|
||
},
|
||
NX: true,
|
||
})
|
||
if err != nil {
|
||
return err
|
||
}
|
||
}
|
||
|
||
return nil
|
||
})
|
||
if err != nil && !success {
|
||
return
|
||
}
|
||
return
|
||
}
|
||
|
||
// getHistoryDataFromHttp 通过 HTTP 接口查询历史数据
|
||
func (s *documentService) getHistoryDataFromHttp(ctx context.Context, doc *entity.Document) (dictData []*dto.DocumentChunkRPC, err error) {
|
||
headers := make(map[string]string)
|
||
if r := g.RequestFromCtx(ctx); r != nil {
|
||
for k, v := range r.Request.Header {
|
||
if len(v) > 0 {
|
||
headers[k] = v[0]
|
||
}
|
||
}
|
||
}
|
||
|
||
// 调用接口获取数据
|
||
d := &dto.ListDocumentChunkRPC{}
|
||
if err = http.Get(ctx, "rag-vector/document/chunk/listDocumentChunk", headers, &d,
|
||
"datasetId", gconv.String(doc.DatasetId),
|
||
"status", 1); err != nil {
|
||
return
|
||
}
|
||
dictData = d.List
|
||
return
|
||
}
|
||
|
||
// getHistoryDataFromMeilisearch 通过 meilisearch 查询历史数据
|
||
func (s *documentService) getHistoryDataFromMeilisearch(ctx context.Context, doc *entity.Document) (dictData []*dto.DocumentChunkRPC, err error) {
|
||
// 构建 meilisearch 查询参数
|
||
searchParams := &meilisearch.SearchParams{
|
||
Filter: fmt.Sprintf("datasetId = %d", doc.DatasetId),
|
||
Limit: 10000,
|
||
}
|
||
|
||
// 执行搜索
|
||
var hits []map[string]interface{}
|
||
_, err = meilisearch.DB().Search(ctx, searchParams, public.IndexNameDocumentChunk, &hits)
|
||
if err != nil {
|
||
return
|
||
}
|
||
|
||
// 转换查询结果
|
||
dictData = make([]*dto.DocumentChunkRPC, 0)
|
||
for _, hit := range hits {
|
||
item := &dto.DocumentChunkRPC{}
|
||
if err = gconv.Struct(hit, item); err != nil {
|
||
return
|
||
}
|
||
dictData = append(dictData, item)
|
||
}
|
||
return
|
||
}
|
||
|
||
// checkRepeat 检查是否重复
|
||
func (s *documentService) checkRepeat(ctx context.Context, contentKey, contentHash string) (success bool, err error) {
|
||
var val *gvar.Var
|
||
if val, err = g.Redis().Set(ctx, fmt.Sprintf(contentKey, contentHash), true, gredis.SetOption{
|
||
TTLOption: gredis.TTLOption{
|
||
EX: gconv.PtrInt64(600),
|
||
},
|
||
NX: true,
|
||
}); err != nil {
|
||
return
|
||
}
|
||
success = val.Bool()
|
||
return
|
||
}
|
||
|
||
func (s *documentService) DocsVectorStatusMsg(ctx context.Context, msg any) (err error) {
|
||
var req = new(dto.KnowledgeDocumentMsg)
|
||
if err = gconv.Struct(msg, &req); err != nil {
|
||
g.Log().Error(ctx, "DocsVectorStatusMsg err:", err)
|
||
return
|
||
}
|
||
ctx = context.WithValue(ctx, "user", &beans.User{
|
||
TenantId: req.TenantId,
|
||
UserName: req.Creator,
|
||
})
|
||
_, err = dao.Document.Update(ctx, &dto.UpdateDocumentReq{
|
||
Id: req.Id,
|
||
VectorStatus: req.VectorStatus,
|
||
})
|
||
return
|
||
}
|