diff --git a/PROGRESS.md b/PROGRESS.md index 70d2c16..f24191d 100644 --- a/PROGRESS.md +++ b/PROGRESS.md @@ -100,7 +100,7 @@ - [ ] 6 个提交待 push(`5d76652` → `79f9912`,需在普通终端 `git push origin main`) - [ ] PDF 导出 Wails 真机验证(不行则回退后端内嵌 CJK 字体出 PDF) - [x] 报告生成并发健壮性(每次 LLM 调用 60s 超时上限,挂死自释放;规划/分章/撰写均套) -- [ ] MinIO 大文档改名/删除的孤儿对象 GC +- [x] MinIO 孤儿 GC:重名覆盖后旧对象(转内联/换键)从 MinIO 删除(SaveDoc 返回旧键,runIngest 清理) - [x] `make test` 目标(test-go / test-web / test-py 一键跑) --- diff --git a/sundynix-gateway/internal/handler/kb.go b/sundynix-gateway/internal/handler/kb.go index 4764d7e..e1d4455 100644 --- a/sundynix-gateway/internal/handler/kb.go +++ b/sundynix-gateway/internal/handler/kb.go @@ -257,10 +257,15 @@ func (h *Handler) runIngest(job, owner, kbName, scoped, forceDoc, filename strin log.Printf("[gateway] 大文档转 MinIO 失败,回退内联: %v", err) } } - docID, err := h.db.SaveDoc(ctx, owner, kbName, docName, ext, md5hex, inline, objectKey, size, head(text, 500)) + docID, oldKey, err := h.db.SaveDoc(ctx, owner, kbName, docName, ext, md5hex, inline, objectKey, size, head(text, 500)) if err != nil { log.Printf("[gateway] 文件入库失败: %v", err) } else if docID != "" { + // 孤儿 GC:重名覆盖后旧对象键若已不用(转内联或换键),从 MinIO 删除,避免泄漏。 + if oldKey != "" && oldKey != objectKey && h.blob.Ready() { + h.blob.Delete(ctx, oldKey) + log.Printf("[gateway] 清理被覆盖的 MinIO 孤儿对象: %s", oldKey) + } _ = h.db.ReplaceDocLinks(ctx, owner, kbName, docID, wikiLinks(text)) // 以本文件 ID 维护出链 _ = h.db.ResolveInboundLinks(ctx, owner, kbName, docName, docID) // 回填指向本文件的悬空链接 } diff --git a/sundynix-gateway/internal/store/model.go b/sundynix-gateway/internal/store/model.go index 70ad0f2..cbb4063 100644 --- a/sundynix-gateway/internal/store/model.go +++ b/sundynix-gateway/internal/store/model.go @@ -115,29 +115,31 @@ type DocLink struct { func (DocLink) TableName() string { return "sundynix_doc_link" } -// SaveDoc 写入/更新一份文件(owner+kb+name 唯一,重名覆盖),返回该文件的雪花 ID(供关联用)。 +// SaveDoc 写入/更新一份文件(owner+kb+name 唯一,重名覆盖),返回 (文件雪花 ID, 被覆盖的旧对象键)。 +// oldObjectKey 是重名覆盖前该文档的 ObjectKey(新建则为空)—— 供调用方清理 MinIO 孤儿对象。 // content 为内联正文(大文档转 MinIO 时传空 + objectKey);ext/md5/preview/size 由调用方按全文给出。 -func (p *Postgres) SaveDoc(ctx context.Context, owner, kb, name, ext, md5, content, objectKey string, size int, preview string) (string, error) { +func (p *Postgres) SaveDoc(ctx context.Context, owner, kb, name, ext, md5, content, objectKey string, size int, preview string) (id, oldObjectKey string, err error) { if p.db == nil { - return "", nil + return "", "", nil } var d Doc - err := p.db.WithContext(ctx).Where("owner = ? AND kb = ? AND name = ?", owner, kb, name).First(&d).Error - if errors.Is(err, gorm.ErrRecordNotFound) { + qerr := p.db.WithContext(ctx).Where("owner = ? AND kb = ? AND name = ?", owner, kb, name).First(&d).Error + if errors.Is(qerr, gorm.ErrRecordNotFound) { d = Doc{Owner: owner, KB: kb, Name: name, Ext: ext, MD5: md5, Content: content, ObjectKey: objectKey, Size: size, Preview: preview} if err := p.db.WithContext(ctx).Create(&d).Error; err != nil { - return "", err + return "", "", err } - return d.ID, nil + return d.ID, "", nil } - if err != nil { - return "", err + if qerr != nil { + return "", "", qerr } + oldObjectKey = d.ObjectKey // 覆盖前的旧对象键 d.Ext, d.MD5, d.Content, d.ObjectKey, d.Size, d.Preview = ext, md5, content, objectKey, size, preview if err := p.db.WithContext(ctx).Save(&d).Error; err != nil { - return "", err + return "", "", err } - return d.ID, nil + return d.ID, oldObjectKey, nil } // ListVault 返回文库列表(仅元数据 + 预览,不含全文),避免一次拉回整库正文。