Add LLM token

backend/app/application/documents/services.py

@@ -526,10 +526,28 @@ class DocumentCommandService:
                     logger.warning("临时文件清理失败: {}", temp_path)
 
     def delete(self, doc_id: str) -> bool:
-        """Delete document record, binary file, and vector chunks."""
+        """Delete document record, binary file, and vector chunks.
+
+        Handles two cases:
+        - Normal docs: have a metadata record in the document repository.
+        - Milvus-only (synthetic) docs: visible in management-list because they
+          have Milvus vectors but no JSON/PG metadata record. We still clean up
+          the Milvus chunks so the document disappears from the list.
+        """
         document = self.document_repository.get(doc_id)
         if not document:
+            # No metadata record — might be a Milvus-only synthetic document.
+            # Attempt vector cleanup directly; treat as success if any chunks deleted.
+            try:
+                deleted_count = self.vector_index.delete_by_document(doc_id)
+                if deleted_count > 0:
+                    logger.info("Deleted Milvus-only doc (no metadata record): doc_id={} chunks={}", doc_id, deleted_count)
+                    return True
+            except Exception as exc:
+                logger.warning("Milvus-only delete failed for doc_id={}: {}", doc_id, exc)
             return False
+
+        # Normal doc: clean up binary, vectors, artifacts, processing records, metadata.
         try:
             self.binary_store.delete(document.object_name)
         except Exception:
@@ -627,13 +645,16 @@ class DocumentQueryService:
             result.append(doc)
 
         # Surface Milvus-only docs that have no metadata record at all.
+        # MinIO almost certainly has their binaries (they were uploaded), so
+        # set object_name to the sentinel "{doc_id}/" so the route marks
+        # has_file=True; the download endpoint will list MinIO to find the file.
         for doc_id, row in milvus_by_id.items():
             if doc_id not in meta_by_id:
                 synthetic = Document(
                     doc_id=doc_id,
                     doc_name=row.get("doc_title", doc_id),
                     file_name=row.get("doc_title", doc_id),
-                    object_name="",
+                    object_name=f"{doc_id}/",  # sentinel: MinIO prefix exists
                     content_type="",
                     size_bytes=0,
                     status=DocumentStatus.INDEXED,
@@ -646,9 +667,63 @@ class DocumentQueryService:
         result.sort(key=lambda d: d.updated_at, reverse=True)
         return result[:limit] if limit is not None else result
 
-    def download(self, doc_id: str) -> tuple[Document, bytes]:
-        """Handle download for the Document Query Service instance."""
+    def download(self, doc_id: str) -> tuple["Document", bytes]:
+        """Return the document record and its binary content from MinIO.
+
+        Fallback strategy for Milvus-only docs (no JSON/PG metadata record):
+          1. Try metadata repository first (normal path).
+          2. If metadata is missing, list MinIO objects with prefix ``{doc_id}/``
+             and synthesise a minimal Document from the first object found.
+             This handles documents whose metadata records were lost but whose
+             binary files are still in object storage.
+          3. If neither source has the file, raise FileNotFoundError.
+        """
+        from app.domain.documents import Document, DocumentStatus
+
         document = self.document_repository.get(doc_id)
-        if not document:
-            raise FileNotFoundError(f"文档不存在: {doc_id}")
+
+        if document and document.object_name and not document.object_name.endswith("/"):
+            # Normal doc with a concrete object_name — read directly.
+            return document, self.binary_store.read(document.object_name)
+
+        if document and not document.object_name:
+            raise FileNotFoundError(f"该文档无原始文件（仅含索引数据，无法下载）: {doc_id}")
+
+        if not document or document.object_name.endswith("/"):
+            # Metadata missing — try to find the file in MinIO by doc_id prefix.
+            try:
+                objects = self.binary_store.list_objects(prefix=f"{doc_id}/")
+                # Filter out artifact JSON files; prefer the source document.
+                candidates = [o for o in objects if not o.endswith(".json")]
+                if not candidates:
+                    candidates = objects  # fall back to all objects if only JSON found
+                if not candidates:
+                    raise FileNotFoundError(f"文档不存在（MinIO 和元数据均无记录）: {doc_id}")
+                object_name = candidates[0]
+                file_name = object_name.split("/", 1)[-1] if "/" in object_name else object_name
+                # Guess content type from extension.
+                ext = file_name.rsplit(".", 1)[-1].lower() if "." in file_name else ""
+                _ct_map = {
+                    "pdf": "application/pdf",
+                    "docx": "application/vnd.openxmlformats-officedocument.wordprocessingml.document",
+                    "doc": "application/msword",
+                    "txt": "text/plain",
+                }
+                content_type = _ct_map.get(ext, "application/octet-stream")
+                # Synthesise a minimal Document so the route can build the response.
+                document = Document(
+                    doc_id=doc_id,
+                    doc_name=file_name,
+                    file_name=file_name,
+                    object_name=object_name,
+                    content_type=content_type,
+                    size_bytes=0,
+                    status=DocumentStatus.INDEXED,
+                )
+                logger.info("MinIO fallback download: doc_id={} object={}", doc_id, object_name)
+            except FileNotFoundError:
+                raise
+            except Exception as exc:
+                raise FileNotFoundError(f"文档不存在: {doc_id}") from exc
+
         return document, self.binary_store.read(document.object_name)