refactor(annotation): 简化任务选择逻辑并移除未使用的状态管理

- 移除了 resolveSegmentSummary 函数调用以简化完成状态判断
- 删除了未使用的 segmentStats 相关引用和缓存清理代码
- 简化了重置模式下的状态更新逻辑

Makefile.offline.mk

@@ -161,12 +161,7 @@ _offline-build-services: ensure-buildx
 	@echo "构建 datamate-database..."
 	@docker buildx build \
 		--cache-from type=local,src=$(CACHE_DIR)/buildkit/database-cache \
-		--network=none \
+		--pull=false \
-		-f scripts/images/database/Dockerfile \
-		-t datamate-database:$(OFFLINE_VERSION) \
-		--load . 2>/dev/null || echo "  Warning: database build may need network, retrying without --network=none..."
-	@docker buildx build \
-		--cache-from type=local,src=$(CACHE_DIR)/buildkit/database-cache \
 		-f scripts/images/database/Dockerfile \
 		-t datamate-database:$(OFFLINE_VERSION) \
 		--load . || echo "  Failed"
@@ -175,6 +170,7 @@ _offline-build-services: ensure-buildx
 	@echo "构建 datamate-gateway..."
 	@docker buildx build \
 		--cache-from type=local,src=$(CACHE_DIR)/buildkit/gateway-cache \
+		--pull=false \
 		-f scripts/images/gateway/Dockerfile \
 		-t datamate-gateway:$(OFFLINE_VERSION) \
 		--load . || echo "  Failed"
@@ -183,6 +179,7 @@ _offline-build-services: ensure-buildx
 	@echo "构建 datamate-backend..."
 	@docker buildx build \
 		--cache-from type=local,src=$(CACHE_DIR)/buildkit/backend-cache \
+		--pull=false \
 		-f scripts/images/backend/Dockerfile \
 		-t datamate-backend:$(OFFLINE_VERSION) \
 		--load . || echo "  Failed"
@@ -191,6 +188,7 @@ _offline-build-services: ensure-buildx
 	@echo "构建 datamate-frontend..."
 	@docker buildx build \
 		--cache-from type=local,src=$(CACHE_DIR)/buildkit/frontend-cache \
+		--pull=false \
 		-f scripts/images/frontend/Dockerfile \
 		-t datamate-frontend:$(OFFLINE_VERSION) \
 		--load . || echo "  Failed"
@@ -199,6 +197,7 @@ _offline-build-services: ensure-buildx
 	@echo "构建 datamate-runtime..."
 	@docker buildx build \
 		--cache-from type=local,src=$(CACHE_DIR)/buildkit/runtime-cache \
+		--pull=false \
 		--build-arg RESOURCES_DIR=$(CACHE_DIR)/resources \
 		-f scripts/images/runtime/Dockerfile \
 		-t datamate-runtime:$(OFFLINE_VERSION) \
@@ -208,6 +207,7 @@ _offline-build-services: ensure-buildx
 	@echo "构建 datamate-backend-python..."
 	@docker buildx build \
 		--cache-from type=local,src=$(CACHE_DIR)/buildkit/backend-python-cache \
+		--pull=false \
 		--build-arg RESOURCES_DIR=$(CACHE_DIR)/resources \
 		-f scripts/images/backend-python/Dockerfile \
 		-t datamate-backend-python:$(OFFLINE_VERSION) \
@@ -218,7 +218,7 @@ _offline-build-services: ensure-buildx
 	@echo "✓ 离线构建完成"
 	@echo "======================================"
 
-# 单个服务离线构建
+# 单个服务离线构建 (BuildKit)
 .PHONY: %-offline-build
 %-offline-build: offline-setup ensure-buildx
 	@echo "离线构建 $*..."
@@ -229,29 +229,76 @@ _offline-build-services: ensure-buildx
 	@$(eval IMAGE_NAME := $(if $(filter deer-flow%,$*),$*,datamate-$*))
 	@docker buildx build \
 		--cache-from type=local,src=$(CACHE_DIR)/buildkit/$*-cache \
+		--pull=false \
 		$(if $(filter runtime backend-python deer-flow%,$*),--build-arg RESOURCES_DIR=$(CACHE_DIR)/resources,) \
 		-f scripts/images/$*/Dockerfile \
 		-t $(IMAGE_NAME):$(OFFLINE_VERSION) \
 		--load .
 
+# 传统 Docker 构建（不使用 BuildKit，更稳定）
+.PHONY: offline-build-classic
+offline-build-classic: offline-setup
+	@echo "使用传统 docker build 进行离线构建..."
+	@bash scripts/offline/build-offline-classic.sh $(CACHE_DIR) $(OFFLINE_VERSION)
+
+# 诊断离线环境
+.PHONY: offline-diagnose
+offline-diagnose:
+	@bash scripts/offline/diagnose.sh $(CACHE_DIR)
+
+# 构建 APT 预装基础镜像（有网环境）
+.PHONY: offline-build-base-images
+offline-build-base-images:
+	@echo "构建 APT 预装基础镜像..."
+	@bash scripts/offline/build-base-images.sh $(CACHE_DIR)
+
+# 使用预装基础镜像进行离线构建（推荐）
+.PHONY: offline-build-final
+offline-build-final: offline-setup
+	@echo "使用预装 APT 包的基础镜像进行离线构建..."
+	@bash scripts/offline/build-offline-final.sh $(CACHE_DIR) $(OFFLINE_VERSION)
+
+# 完整离线导出（包含 APT 预装基础镜像）
+.PHONY: offline-export-full
+offline-export-full:
+	@echo "======================================"
+	@echo "完整离线缓存导出（含 APT 预装基础镜像）"
+	@echo "======================================"
+	@$(MAKE) offline-build-base-images
+	@$(MAKE) offline-export
+	@echo ""
+	@echo "导出完成！传输时请包含以下文件:"
+	@echo "  - build-cache/images/base-images-with-apt.tar"
+	@echo "  - build-cache-YYYYMMDD.tar.gz"
+
 # ========== 帮助 ==========
 
 .PHONY: help-offline
 help-offline:
 	@echo "离线构建命令:"
-	@echo "  make offline-export [CACHE_DIR=./build-cache]       - 在有网环境导出构建缓存"
+	@echo ""
+	@echo "【有网环境】"
+	@echo "  make offline-export [CACHE_DIR=./build-cache]       - 导出构建缓存"
+	@echo "  make offline-export-full                            - 导出完整缓存（含 APT 预装基础镜像）"
+	@echo "  make offline-build-base-images                      - 构建 APT 预装基础镜像"
+	@echo ""
+	@echo "【无网环境】"
 	@echo "  make offline-setup [CACHE_DIR=./build-cache]        - 解压并准备离线缓存"
-	@echo "  make offline-build [CACHE_DIR=./build-cache]        - 在无网环境构建所有服务"
+	@echo "  make offline-build-final                            - 使用预装基础镜像构建（推荐，解决 APT 问题）"
+	@echo "  make offline-build-classic                          - 使用传统 docker build"
+	@echo "  make offline-build                                  - 使用 BuildKit 构建"
+	@echo "  make offline-diagnose                               - 诊断离线构建环境"
 	@echo "  make <service>-offline-build                        - 离线构建单个服务"
-	@echo "                                                      (如: make backend-offline-build)"
 	@echo ""
-	@echo "完整工作流程:"
+	@echo "【完整工作流程（推荐）】"
-	@echo "  # 1. 有网环境导出缓存"
+	@echo "  # 1. 有网环境导出完整缓存"
-	@echo "  make offline-export"
+	@echo "  make offline-export-full"
 	@echo ""
-	@echo "  # 2. 传输缓存到无网环境"
+	@echo "  # 2. 传输到无网环境（需要传输两个文件）"
-	@echo "  scp build-cache-*.tar.gz user@offline-server:/path/to/project/"
+	@echo "  scp build-cache/images/base-images-with-apt.tar user@offline-server:/path/"
+	@echo "  scp build-cache-*.tar.gz user@offline-server:/path/"
 	@echo ""
 	@echo "  # 3. 无网环境构建"
 	@echo "  tar -xzf build-cache-*.tar.gz"
-	@echo "  make offline-build"
+	@echo "  docker load -i build-cache/images/base-images-with-apt.tar"
+	@echo "  make offline-build-final"

backend/openapi/specs/data-management.yaml

@@ -470,6 +470,23 @@ paths:
         '200':
           description: 上传成功
 
+  /data-management/datasets/upload/cancel-upload/{reqId}:
+    put:
+      tags: [ DatasetFile ]
+      operationId: cancelUpload
+      summary: 取消上传
+      description: 取消预上传请求并清理临时分片
+      parameters:
+        - name: reqId
+          in: path
+          required: true
+          schema:
+            type: string
+          description: 预上传请求ID
+      responses:
+        '200':
+          description: 取消成功
+
   /data-management/dataset-types:
     get:
       operationId: getDatasetTypes

backend/services/data-management-service/src/main/java/com/datamate/datamanagement/application/DatasetApplicationService.java

@@ -1,5 +1,6 @@
 package com.datamate.datamanagement.application;
 
+import com.baomidou.mybatisplus.core.conditions.update.LambdaUpdateWrapper;
 import com.baomidou.mybatisplus.core.metadata.IPage;
 import com.baomidou.mybatisplus.extension.plugins.pagination.Page;
 import com.datamate.common.domain.utils.ChunksSaver;
@@ -101,6 +102,7 @@ public class DatasetApplicationService {
     public Dataset updateDataset(String datasetId, UpdateDatasetRequest updateDatasetRequest) {
         Dataset dataset = datasetRepository.getById(datasetId);
         BusinessAssert.notNull(dataset, DataManagementErrorCode.DATASET_NOT_FOUND);
+        
         if (StringUtils.hasText(updateDatasetRequest.getName())) {
             dataset.setName(updateDatasetRequest.getName());
         }
@@ -113,13 +115,31 @@ public class DatasetApplicationService {
         if (Objects.nonNull(updateDatasetRequest.getStatus())) {
             dataset.setStatus(updateDatasetRequest.getStatus());
         }
-        if (updateDatasetRequest.getParentDatasetId() != null) {
+        if (updateDatasetRequest.isParentDatasetIdProvided()) {
+            // 保存原始的 parentDatasetId 值，用于比较是否发生了变化
+            String originalParentDatasetId = dataset.getParentDatasetId();
+
+            // 处理父数据集变更：仅当请求显式包含 parentDatasetId 时处理
+            // handleParentChange 内部通过 normalizeParentId 方法将空字符串和 null 都转换为 null
+            // 这样既支持设置新的父数据集，也支持清除关联
             handleParentChange(dataset, updateDatasetRequest.getParentDatasetId());
+
+            // 检查 parentDatasetId 是否发生了变化
+            if (!Objects.equals(originalParentDatasetId, dataset.getParentDatasetId())) {
+                // 使用 LambdaUpdateWrapper 显式地更新 parentDatasetId 字段
+                // 这样即使值为 null 也能被正确更新到数据库
+                datasetRepository.update(null, new LambdaUpdateWrapper<Dataset>()
+                    .eq(Dataset::getId, datasetId)
+                    .set(Dataset::getParentDatasetId, dataset.getParentDatasetId()));
+            }
         }
+        
         if (StringUtils.hasText(updateDatasetRequest.getDataSource())) {
             // 数据源id不为空，使用异步线程进行文件扫盘落库
             processDataSourceAsync(dataset.getId(), updateDatasetRequest.getDataSource());
         }
+        
+        // 更新其他字段（不包括 parentDatasetId，因为它已经在上面的代码中更新了）
         datasetRepository.updateById(dataset);
         return dataset;
     }

backend/services/data-management-service/src/main/java/com/datamate/datamanagement/application/DatasetFileApplicationService.java

@@ -505,6 +505,14 @@ public class DatasetFileApplicationService {
         saveFileInfoToDb(uploadResult, datasetId);
     }
 
+    /**
+     * 取消上传
+     */
+    @Transactional
+    public void cancelUpload(String reqId) {
+        fileService.cancelUpload(reqId);
+    }
+
     private void saveFileInfoToDb(FileUploadResult fileUploadResult, String datasetId) {
         if (Objects.isNull(fileUploadResult.getSavedFile())) {
             // 文件切片上传没有完成

backend/services/data-management-service/src/main/java/com/datamate/datamanagement/interfaces/dto/UpdateDatasetRequest.java

@@ -1,8 +1,10 @@
 package com.datamate.datamanagement.interfaces.dto;
 
 import com.datamate.datamanagement.common.enums.DatasetStatusType;
+import com.fasterxml.jackson.annotation.JsonIgnore;
 import jakarta.validation.constraints.NotBlank;
 import jakarta.validation.constraints.Size;
+import lombok.AccessLevel;
 import lombok.Getter;
 import lombok.Setter;
 
@@ -24,9 +26,18 @@ public class UpdateDatasetRequest {
     /** 归集任务id */
     private String dataSource;
     /** 父数据集ID */
+    @Setter(AccessLevel.NONE)
     private String parentDatasetId;
+    @JsonIgnore
+    @Setter(AccessLevel.NONE)
+    private boolean parentDatasetIdProvided;
     /** 标签列表 */
     private List<String> tags;
     /** 数据集状态 */
     private DatasetStatusType status;
+
+    public void setParentDatasetId(String parentDatasetId) {
+        this.parentDatasetIdProvided = true;
+        this.parentDatasetId = parentDatasetId;
+    }
 }

backend/services/data-management-service/src/main/java/com/datamate/datamanagement/interfaces/rest/DatasetUploadController.java

@@ -0,0 +1,33 @@
+package com.datamate.datamanagement.interfaces.rest;
+
+import com.datamate.datamanagement.application.DatasetFileApplicationService;
+import lombok.RequiredArgsConstructor;
+import lombok.extern.slf4j.Slf4j;
+import org.springframework.http.ResponseEntity;
+import org.springframework.web.bind.annotation.PathVariable;
+import org.springframework.web.bind.annotation.PutMapping;
+import org.springframework.web.bind.annotation.RequestMapping;
+import org.springframework.web.bind.annotation.RestController;
+
+/**
+ * 数据集上传控制器
+ */
+@Slf4j
+@RestController
+@RequiredArgsConstructor
+@RequestMapping("/data-management/datasets/upload")
+public class DatasetUploadController {
+
+    private final DatasetFileApplicationService datasetFileApplicationService;
+
+    /**
+     * 取消上传
+     *
+     * @param reqId 预上传请求ID
+     */
+    @PutMapping("/cancel-upload/{reqId}")
+    public ResponseEntity<Void> cancelUpload(@PathVariable("reqId") String reqId) {
+        datasetFileApplicationService.cancelUpload(reqId);
+        return ResponseEntity.ok().build();
+    }
+}

backend/shared/domain-common/src/main/java/com/datamate/common/domain/service/FileService.java

@@ -74,6 +74,26 @@ public class FileService {
             .build();
     }
 
+    /**
+     * 取消上传
+     */
+    @Transactional
+    public void cancelUpload(String reqId) {
+        if (reqId == null || reqId.isBlank()) {
+            throw BusinessException.of(CommonErrorCode.PARAM_ERROR);
+        }
+        ChunkUploadPreRequest preRequest = chunkUploadRequestMapper.findById(reqId);
+        if (preRequest == null) {
+            return;
+        }
+        String uploadPath = preRequest.getUploadPath();
+        if (uploadPath != null && !uploadPath.isBlank()) {
+            File tempDir = new File(uploadPath, String.format(ChunksSaver.TEMP_DIR_NAME_FORMAT, preRequest.getId()));
+            ChunksSaver.deleteFolder(tempDir.getPath());
+        }
+        chunkUploadRequestMapper.deleteById(reqId);
+    }
+
     private File uploadFile(ChunkUploadRequest fileUploadRequest, ChunkUploadPreRequest preRequest) {
         File savedFile = ChunksSaver.saveFile(fileUploadRequest, preRequest);
         preRequest.setTimeout(LocalDateTime.now().plusSeconds(DEFAULT_TIMEOUT));

frontend/src/hooks/useSliceUpload.tsx

@@ -1,5 +1,5 @@
 import { TaskItem } from "@/pages/DataManagement/dataset.model";
-import { calculateSHA256, checkIsFilesExist } from "@/utils/file.util";
+import { calculateSHA256, checkIsFilesExist, streamSplitAndUpload, StreamUploadResult } from "@/utils/file.util";
 import { App } from "antd";
 import { useRef, useState } from "react";
 
@@ -9,17 +9,18 @@ export function useFileSliceUpload(
     uploadChunk,
     cancelUpload,
   }: {
-    preUpload: (id: string, params: any) => Promise<{ data: number }>;
+    preUpload: (id: string, params: Record<string, unknown>) => Promise<{ data: number }>;
-    uploadChunk: (id: string, formData: FormData, config: any) => Promise<any>;
+    uploadChunk: (id: string, formData: FormData, config: Record<string, unknown>) => Promise<unknown>;
-    cancelUpload: ((reqId: number) => Promise<any>) | null;
+    cancelUpload: ((reqId: number) => Promise<unknown>) | null;
   },
-  showTaskCenter = true // 上传时是否显示任务中心
+  showTaskCenter = true, // 上传时是否显示任务中心
+  enableStreamUpload = true // 是否启用流式分割上传
 ) {
   const { message } = App.useApp();
   const [taskList, setTaskList] = useState<TaskItem[]>([]);
   const taskListRef = useRef<TaskItem[]>([]); // 用于固定任务顺序
 
-  const createTask = (detail: any = {}) => {
+  const createTask = (detail: Record<string, unknown> = {}) => {
     const { dataset } = detail;
     const title = `上传数据集: ${dataset.name} `;
     const controller = new AbortController();
@@ -37,6 +38,14 @@ export function useFileSliceUpload(
     taskListRef.current = [task, ...taskListRef.current];
 
     setTaskList(taskListRef.current);
+    
+    // 立即显示任务中心，让用户感知上传已开始
+    if (showTaskCenter) {
+      window.dispatchEvent(
+        new CustomEvent("show:task-popover", { detail: { show: true } })
+      );
+    }
+    
     return task;
   };
 
@@ -60,7 +69,7 @@ export function useFileSliceUpload(
       // 携带前缀信息，便于刷新后仍停留在当前目录
       window.dispatchEvent(
         new CustomEvent(task.updateEvent, {
-          detail: { prefix: (task as any).prefix },
+          detail: { prefix: task.prefix },
         })
       );
     }
@@ -71,7 +80,7 @@ export function useFileSliceUpload(
     }
   };
 
-  async function buildFormData({ file, reqId, i, j }) {
+  async function buildFormData({ file, reqId, i, j }: { file: { slices: Blob[]; name: string; size: number }; reqId: number; i: number; j: number }) {
     const formData = new FormData();
     const { slices, name, size } = file;
     const checkSum = await calculateSHA256(slices[j]);
@@ -86,12 +95,18 @@ export function useFileSliceUpload(
     return formData;
   }
 
-  async function uploadSlice(task: TaskItem, fileInfo) {
+  async function uploadSlice(task: TaskItem, fileInfo: { loaded: number; i: number; j: number; files: { slices: Blob[]; name: string; size: number }[]; totalSize: number }) {
     if (!task) {
       return;
     }
-    const { reqId, key } = task;
+    const { reqId, key, controller } = task;
     const { loaded, i, j, files, totalSize } = fileInfo;
+    
+    // 检查是否已取消
+    if (controller.signal.aborted) {
+      throw new Error("Upload cancelled");
+    }
+    
     const formData = await buildFormData({
       file: files[i],
       i,
@@ -101,6 +116,7 @@ export function useFileSliceUpload(
 
     let newTask = { ...task };
     await uploadChunk(key, formData, {
+      signal: controller.signal,
       onUploadProgress: (e) => {
         const loadedSize = loaded + e.loaded;
         const curPercent = Number((loadedSize / totalSize) * 100).toFixed(2);
@@ -116,7 +132,7 @@ export function useFileSliceUpload(
     });
   }
 
-  async function uploadFile({ task, files, totalSize }) {
+  async function uploadFile({ task, files, totalSize }: { task: TaskItem; files: { slices: Blob[]; name: string; size: number; originFile: Blob }[]; totalSize: number }) {
     console.log('[useSliceUpload] Calling preUpload with prefix:', task.prefix);
     const { data: reqId } = await preUpload(task.key, {
       totalFileNum: files.length,
@@ -132,24 +148,29 @@ export function useFileSliceUpload(
       reqId,
       isCancel: false,
       cancelFn: () => {
-        task.controller.abort();
+        // 使用 newTask 的 controller 确保一致性
+        newTask.controller.abort();
         cancelUpload?.(reqId);
-        if (task.updateEvent) window.dispatchEvent(new Event(task.updateEvent));
+        if (newTask.updateEvent) window.dispatchEvent(new Event(newTask.updateEvent));
       },
     };
     updateTaskList(newTask);
-    if (showTaskCenter) {
+    // 注意：show:task-popover 事件已在 createTask 中触发，此处不再重复触发
-      window.dispatchEvent(
-        new CustomEvent("show:task-popover", { detail: { show: true } })
-      );
-    }
     // // 更新数据状态
     if (task.updateEvent) window.dispatchEvent(new Event(task.updateEvent));
 
     let loaded = 0;
     for (let i = 0; i < files.length; i++) {
+      // 检查是否已取消
+      if (newTask.controller.signal.aborted) {
+        throw new Error("Upload cancelled");
+      }
       const { slices } = files[i];
       for (let j = 0; j < slices.length; j++) {
+        // 检查是否已取消
+        if (newTask.controller.signal.aborted) {
+          throw new Error("Upload cancelled");
+        }
         await uploadSlice(newTask, {
           loaded,
           i,
@@ -163,7 +184,7 @@ export function useFileSliceUpload(
     removeTask(newTask);
   }
 
-  const handleUpload = async ({ task, files }) => {
+  const handleUpload = async ({ task, files }: { task: TaskItem; files: { slices: Blob[]; name: string; size: number; originFile: Blob }[] }) => {
     const isErrorFile = await checkIsFilesExist(files);
     if (isErrorFile) {
       message.error("文件被修改或删除，请重新选择文件上传");
@@ -189,10 +210,174 @@ export function useFileSliceUpload(
     }
   };
 
+  /**
+   * 流式分割上传处理
+   * 用于大文件按行分割并立即上传的场景
+   */
+  const handleStreamUpload = async ({ task, files }: { task: TaskItem; files: File[] }) => {
+    try {
+      console.log('[useSliceUpload] Starting stream upload for', files.length, 'files');
+      
+      const totalSize = files.reduce((acc, file) => acc + file.size, 0);
+      
+      // 存储所有文件的 reqId，用于取消上传
+      const reqIds: number[] = [];
+      
+      const newTask: TaskItem = {
+        ...task,
+        reqId: -1,
+        isCancel: false,
+        cancelFn: () => {
+          // 使用 newTask 的 controller 确保一致性
+          newTask.controller.abort();
+          // 取消所有文件的预上传请求
+          reqIds.forEach(id => cancelUpload?.(id));
+          if (newTask.updateEvent) window.dispatchEvent(new Event(newTask.updateEvent));
+        },
+      };
+      updateTaskList(newTask);
+
+      let totalUploadedLines = 0;
+      let totalProcessedBytes = 0;
+      const results: StreamUploadResult[] = [];
+
+      // 逐个处理文件，每个文件单独调用 preUpload
+      for (let i = 0; i < files.length; i++) {
+        // 检查是否已取消
+        if (newTask.controller.signal.aborted) {
+          throw new Error("Upload cancelled");
+        }
+        
+        const file = files[i];
+        console.log(`[useSliceUpload] Processing file ${i + 1}/${files.length}: ${file.name}`);
+        
+        const result = await streamSplitAndUpload(
+          file,
+          (formData, config) => uploadChunk(task.key, formData, {
+            ...config,
+            signal: newTask.controller.signal,
+          }),
+          (currentBytes, totalBytes, uploadedLines) => {
+            // 检查是否已取消
+            if (newTask.controller.signal.aborted) {
+              return;
+            }
+            
+            // 更新进度
+            const overallBytes = totalProcessedBytes + currentBytes;
+            const curPercent = Number((overallBytes / totalSize) * 100).toFixed(2);
+            
+            const updatedTask: TaskItem = {
+              ...newTask,
+              ...taskListRef.current.find((item) => item.key === task.key),
+              size: overallBytes,
+              percent: curPercent >= 100 ? 99.99 : curPercent,
+              streamUploadInfo: {
+                currentFile: file.name,
+                fileIndex: i + 1,
+                totalFiles: files.length,
+                uploadedLines: totalUploadedLines + uploadedLines,
+              },
+            };
+            updateTaskList(updatedTask);
+          }, 
+          1024 * 1024, // 1MB chunk size
+          {
+            resolveReqId: async ({ totalFileNum, totalSize }) => {
+              const { data: reqId } = await preUpload(task.key, {
+                totalFileNum,
+                totalSize,
+                datasetId: task.key,
+                hasArchive: task.hasArchive,
+                prefix: task.prefix,
+              });
+              console.log(`[useSliceUpload] File ${file.name} preUpload response reqId:`, reqId);
+              reqIds.push(reqId);
+              return reqId;
+            },
+            hasArchive: newTask.hasArchive,
+            prefix: newTask.prefix,
+            signal: newTask.controller.signal,
+            maxConcurrency: 3,
+          }
+        );
+        
+        results.push(result);
+        totalUploadedLines += result.uploadedCount;
+        totalProcessedBytes += file.size;
+        
+        console.log(`[useSliceUpload] File ${file.name} processed, uploaded ${result.uploadedCount} lines`);
+      }
+
+      console.log('[useSliceUpload] Stream upload completed, total lines:', totalUploadedLines);
+      removeTask(newTask);
+      
+      message.success(`成功上传 ${totalUploadedLines} 个文件（按行分割）`);
+    } catch (err) {
+      console.error('[useSliceUpload] Stream upload error:', err);
+      if (err.message === "Upload cancelled") {
+        message.info("上传已取消");
+      } else {
+        message.error("文件上传失败，请稍后重试");
+      }
+      removeTask({
+        ...task,
+        isCancel: true,
+        ...taskListRef.current.find((item) => item.key === task.key),
+      });
+    }
+  };
+
+  /**
+   * 注册流式上传事件监听
+   * 返回注销函数
+   */
+  const registerStreamUploadListener = () => {
+    if (!enableStreamUpload) return () => {};
+
+    const streamUploadHandler = async (e: Event) => {
+      const customEvent = e as CustomEvent;
+      const { dataset, files, updateEvent, hasArchive, prefix } = customEvent.detail;
+      
+      const controller = new AbortController();
+      const task: TaskItem = {
+        key: dataset.id,
+        title: `上传数据集: ${dataset.name} (按行分割)`,
+        percent: 0,
+        reqId: -1,
+        controller,
+        size: 0,
+        updateEvent,
+        hasArchive,
+        prefix,
+      };
+      
+      taskListRef.current = [task, ...taskListRef.current];
+      setTaskList(taskListRef.current);
+      
+      // 显示任务中心
+      if (showTaskCenter) {
+        window.dispatchEvent(
+          new CustomEvent("show:task-popover", { detail: { show: true } })
+        );
+      }
+      
+      await handleStreamUpload({ task, files });
+    };
+
+    window.addEventListener("upload:dataset-stream", streamUploadHandler);
+    
+    return () => {
+      window.removeEventListener("upload:dataset-stream", streamUploadHandler);
+    };
+  };
+
   return {
     taskList,
     createTask,
     removeTask,
     handleUpload,
+    handleStreamUpload,
+    registerStreamUploadListener,
   };
 }

frontend/src/pages/DataAnnotation/Annotate/LabelStudioTextEditor.tsx

@@ -1,6 +1,6 @@
 import { useCallback, useEffect, useMemo, useRef, useState } from "react";
-import { App, Button, Card, List, Spin, Typography, Tag, Switch, Tree, Empty } from "antd";
+import { App, Button, Card, List, Spin, Typography, Tag, Empty } from "antd";
-import { LeftOutlined, ReloadOutlined, SaveOutlined, MenuFoldOutlined, MenuUnfoldOutlined, CheckOutlined } from "@ant-design/icons";
+import { LeftOutlined, ReloadOutlined, SaveOutlined, MenuFoldOutlined, MenuUnfoldOutlined } from "@ant-design/icons";
 import { useNavigate, useParams } from "react-router";
 
 import {
@@ -28,7 +28,6 @@ type EditorTaskListItem = {
   hasAnnotation: boolean;
   annotationUpdatedAt?: string | null;
   annotationStatus?: AnnotationResultStatus | null;
-  segmentStats?: SegmentStats;
 };
 
 type LsfMessage = {
@@ -36,21 +35,6 @@ type LsfMessage = {
   payload?: unknown;
 };
 
-type SegmentInfo = {
-  idx: number;
-  text: string;
-  start: number;
-  end: number;
-  hasAnnotation: boolean;
-  lineIndex: number;
-  chunkIndex: number;
-};
-
-type SegmentStats = {
-  done: number;
-  total: number;
-};
-
 type ApiResponse<T> = {
   code?: number;
   message?: string;
@@ -66,10 +50,11 @@ type EditorTaskPayload = {
 type EditorTaskResponse = {
   task?: EditorTaskPayload;
   segmented?: boolean;
-  segments?: SegmentInfo[];
+  totalSegments?: number;
   currentSegmentIndex?: number;
 };
 
+
 type EditorTaskListResponse = {
   content?: EditorTaskListItem[];
   totalElements?: number;
@@ -91,8 +76,6 @@ type ExportPayload = {
   requestId?: string | null;
 };
 
-type SwitchDecision = "save" | "discard" | "cancel";
-
 const LSF_IFRAME_SRC = "/lsf/lsf.html";
 const TASK_PAGE_START = 0;
 const TASK_PAGE_SIZE = 200;
@@ -154,16 +137,6 @@ const isAnnotationResultEmpty = (annotation?: Record<string, unknown>) => {
 };
 
 const resolveTaskStatusMeta = (item: EditorTaskListItem) => {
-  const segmentSummary = resolveSegmentSummary(item);
-  if (segmentSummary) {
-    if (segmentSummary.done >= segmentSummary.total) {
-      return { text: "已标注", type: "success" as const };
-    }
-    if (segmentSummary.done > 0) {
-      return { text: "标注中", type: "warning" as const };
-    }
-    return { text: "未标注", type: "secondary" as const };
-  }
   if (!item.hasAnnotation) {
     return { text: "未标注", type: "secondary" as const };
   }
@@ -216,25 +189,6 @@ const buildAnnotationSnapshot = (annotation?: Record<string, unknown>) => {
 const buildSnapshotKey = (fileId: string, segmentIndex?: number) =>
   `${fileId}::${segmentIndex ?? "full"}`;
 
-const buildSegmentStats = (segmentList?: SegmentInfo[] | null): SegmentStats | null => {
-  if (!Array.isArray(segmentList) || segmentList.length === 0) return null;
-  const total = segmentList.length;
-  const done = segmentList.reduce((count, seg) => count + (seg.hasAnnotation ? 1 : 0), 0);
-  return { done, total };
-};
-
-const normalizeSegmentStats = (stats?: SegmentStats | null): SegmentStats | null => {
-  if (!stats) return null;
-  const total = Number(stats.total);
-  const done = Number(stats.done);
-  if (!Number.isFinite(total) || total <= 0) return null;
-  const safeDone = Math.min(Math.max(done, 0), total);
-  return { done: safeDone, total };
-};
-
-const resolveSegmentSummary = (item: EditorTaskListItem) =>
-  normalizeSegmentStats(item.segmentStats);
-
 const mergeTaskItems = (base: EditorTaskListItem[], next: EditorTaskListItem[]) => {
   if (next.length === 0) return base;
   const seen = new Set(base.map((item) => item.fileId));
@@ -282,18 +236,13 @@ export default function LabelStudioTextEditor() {
     resolve: (payload?: ExportPayload) => void;
     timer?: number;
   } | null>(null);
-  const exportCheckSeqRef = useRef(0);
   const savedSnapshotsRef = useRef<Record<string, string>>({});
   const pendingAutoAdvanceRef = useRef(false);
-  const segmentStatsCacheRef = useRef<Record<string, SegmentStats>>({});
-  const segmentStatsSeqRef = useRef(0);
-  const segmentStatsLoadingRef = useRef<Set<string>>(new Set());
 
   const [loadingProject, setLoadingProject] = useState(true);
   const [loadingTasks, setLoadingTasks] = useState(false);
   const [loadingTaskDetail, setLoadingTaskDetail] = useState(false);
   const [saving, setSaving] = useState(false);
-  const [segmentSwitching, setSegmentSwitching] = useState(false);
 
   const [iframeReady, setIframeReady] = useState(false);
   const [lsReady, setLsReady] = useState(false);
@@ -306,12 +255,11 @@ export default function LabelStudioTextEditor() {
   const [prefetching, setPrefetching] = useState(false);
   const [selectedFileId, setSelectedFileId] = useState<string>("");
   const [sidebarCollapsed, setSidebarCollapsed] = useState(false);
-  const [autoSaveOnSwitch, setAutoSaveOnSwitch] = useState(false);
 
   // 分段相关状态
   const [segmented, setSegmented] = useState(false);
-  const [segments, setSegments] = useState<SegmentInfo[]>([]);
   const [currentSegmentIndex, setCurrentSegmentIndex] = useState(0);
+  const [segmentTotal, setSegmentTotal] = useState(0);
   const isTextProject = useMemo(
     () => (project?.datasetType || "").toUpperCase() === "TEXT",
     [project?.datasetType],
@@ -330,70 +278,6 @@ export default function LabelStudioTextEditor() {
     win.postMessage({ type, payload }, origin);
   }, [origin]);
 
-  const applySegmentStats = useCallback((fileId: string, stats: SegmentStats | null) => {
-    if (!fileId) return;
-    const normalized = normalizeSegmentStats(stats);
-    setTasks((prev) =>
-      prev.map((item) =>
-        item.fileId === fileId
-          ? { ...item, segmentStats: normalized || undefined }
-          : item
-      )
-    );
-  }, []);
-
-  const updateSegmentStatsCache = useCallback((fileId: string, stats: SegmentStats | null) => {
-    if (!fileId) return;
-    const normalized = normalizeSegmentStats(stats);
-    if (normalized) {
-      segmentStatsCacheRef.current[fileId] = normalized;
-    } else {
-      delete segmentStatsCacheRef.current[fileId];
-    }
-    applySegmentStats(fileId, normalized);
-  }, [applySegmentStats]);
-
-  const fetchSegmentStatsForFile = useCallback(async (fileId: string, seq: number) => {
-    if (!projectId || !fileId) return;
-    if (segmentStatsCacheRef.current[fileId] || segmentStatsLoadingRef.current.has(fileId)) return;
-    segmentStatsLoadingRef.current.add(fileId);
-    try {
-      const resp = (await getEditorTaskUsingGet(projectId, fileId, {
-        segmentIndex: 0,
-      })) as ApiResponse<EditorTaskResponse>;
-      if (segmentStatsSeqRef.current !== seq) return;
-      const data = resp?.data;
-      if (!data?.segmented) return;
-      const stats = buildSegmentStats(data.segments);
-      if (!stats) return;
-      segmentStatsCacheRef.current[fileId] = stats;
-      applySegmentStats(fileId, stats);
-    } catch (e) {
-      console.error(e);
-    } finally {
-      segmentStatsLoadingRef.current.delete(fileId);
-    }
-  }, [applySegmentStats, projectId]);
-
-  const prefetchSegmentStats = useCallback((items: EditorTaskListItem[]) => {
-    if (!projectId) return;
-    const fileIds = items
-      .map((item) => item.fileId)
-      .filter((fileId) => fileId && !segmentStatsCacheRef.current[fileId]);
-    if (fileIds.length === 0) return;
-    const seq = segmentStatsSeqRef.current;
-    let cursor = 0;
-    const workerCount = Math.min(3, fileIds.length);
-    const runWorker = async () => {
-      while (cursor < fileIds.length && segmentStatsSeqRef.current === seq) {
-        const fileId = fileIds[cursor];
-        cursor += 1;
-        await fetchSegmentStatsForFile(fileId, seq);
-      }
-    };
-    void Promise.all(Array.from({ length: workerCount }, () => runWorker()));
-  }, [fetchSegmentStatsForFile, projectId]);
-
   const confirmEmptyAnnotationStatus = useCallback(() => {
     return new Promise<AnnotationResultStatus | null>((resolve) => {
       let resolved = false;
@@ -446,8 +330,6 @@ export default function LabelStudioTextEditor() {
 
   const updateTaskSelection = useCallback((items: EditorTaskListItem[]) => {
     const isCompleted = (item: EditorTaskListItem) => {
-      const summary = resolveSegmentSummary(item);
-      if (summary) return summary.done >= summary.total;
       return item.hasAnnotation;
     };
     const defaultFileId =
@@ -508,9 +390,6 @@ export default function LabelStudioTextEditor() {
     if (mode === "reset") {
       prefetchSeqRef.current += 1;
       setPrefetching(false);
-      segmentStatsSeqRef.current += 1;
-      segmentStatsCacheRef.current = {};
-      segmentStatsLoadingRef.current = new Set();
     }
     if (mode === "append") {
       setLoadingMore(true);
@@ -591,20 +470,19 @@ export default function LabelStudioTextEditor() {
       if (seq !== initSeqRef.current) return;
 
       // 更新分段状态
-      const segmentIndex = data?.segmented
+      const isSegmented = !!data?.segmented;
+      const segmentIndex = isSegmented
         ? resolveSegmentIndex(data.currentSegmentIndex) ?? 0
         : undefined;
-      if (data?.segmented) {
+      if (isSegmented) {
-        const stats = buildSegmentStats(data.segments);
         setSegmented(true);
-        setSegments(data.segments || []);
         setCurrentSegmentIndex(segmentIndex ?? 0);
-        updateSegmentStatsCache(fileId, stats);
+        const totalSegments = Number(data?.totalSegments ?? 0);
+        setSegmentTotal(Number.isFinite(totalSegments) && totalSegments > 0 ? totalSegments : 0);
       } else {
         setSegmented(false);
-        setSegments([]);
         setCurrentSegmentIndex(0);
-        updateSegmentStatsCache(fileId, null);
+        setSegmentTotal(0);
       }
 
       const taskData = {
@@ -664,19 +542,14 @@ export default function LabelStudioTextEditor() {
     } finally {
       if (seq === initSeqRef.current) setLoadingTaskDetail(false);
     }
-  }, [iframeReady, message, postToIframe, project, projectId, updateSegmentStatsCache]);
+  }, [iframeReady, message, postToIframe, project, projectId]);
 
   const advanceAfterSave = useCallback(async (fileId: string, segmentIndex?: number) => {
     if (!fileId) return;
-    if (segmented && segments.length > 0) {
+    if (segmented && segmentTotal > 0) {
-      const sortedSegmentIndices = segments
+      const baseIndex = Math.max(segmentIndex ?? currentSegmentIndex, 0);
-        .map((seg) => seg.idx)
+      const nextSegmentIndex = baseIndex + 1;
-        .sort((a, b) => a - b);
+      if (nextSegmentIndex < segmentTotal) {
-      const baseIndex = segmentIndex ?? currentSegmentIndex;
-      const currentPos = sortedSegmentIndices.indexOf(baseIndex);
-      const nextSegmentIndex =
-        currentPos >= 0 ? sortedSegmentIndices[currentPos + 1] : sortedSegmentIndices[0];
-      if (nextSegmentIndex !== undefined) {
         await initEditorForFile(fileId, nextSegmentIndex);
         return;
       }
@@ -698,7 +571,7 @@ export default function LabelStudioTextEditor() {
     initEditorForFile,
     message,
     segmented,
-    segments,
+    segmentTotal,
     tasks,
   ]);
 
@@ -772,16 +645,6 @@ export default function LabelStudioTextEditor() {
       const snapshot = buildAnnotationSnapshot(isRecord(annotation) ? annotation : undefined);
       savedSnapshotsRef.current[snapshotKey] = snapshot;
 
-      // 分段模式下更新当前段落的标注状态
-      if (segmented && segmentIndex !== undefined) {
-        const nextSegments = segments.map((seg) =>
-          seg.idx === segmentIndex
-            ? { ...seg, hasAnnotation: true }
-            : seg
-        );
-        setSegments(nextSegments);
-        updateSegmentStatsCache(String(fileId), buildSegmentStats(nextSegments));
-      }
       if (options?.autoAdvance) {
         await advanceAfterSave(String(fileId), segmentIndex);
       }
@@ -800,69 +663,10 @@ export default function LabelStudioTextEditor() {
     message,
     projectId,
     segmented,
-    segments,
     selectedFileId,
     tasks,
-    updateSegmentStatsCache,
   ]);
 
-  const requestExportForCheck = useCallback(() => {
-    if (!iframeReady || !lsReady) return Promise.resolve(undefined);
-    if (exportCheckRef.current) {
-      if (exportCheckRef.current.timer) {
-        window.clearTimeout(exportCheckRef.current.timer);
-      }
-      exportCheckRef.current.resolve(undefined);
-      exportCheckRef.current = null;
-    }
-    const requestId = `check_${Date.now()}_${++exportCheckSeqRef.current}`;
-    return new Promise<ExportPayload | undefined>((resolve) => {
-      const timer = window.setTimeout(() => {
-        if (exportCheckRef.current?.requestId === requestId) {
-          exportCheckRef.current = null;
-        }
-        resolve(undefined);
-      }, 3000);
-      exportCheckRef.current = {
-        requestId,
-        resolve,
-        timer,
-      };
-      postToIframe("LS_EXPORT_CHECK", { requestId });
-    });
-  }, [iframeReady, lsReady, postToIframe]);
-
-  const confirmSaveBeforeSwitch = useCallback(() => {
-    return new Promise<SwitchDecision>((resolve) => {
-      let resolved = false;
-      let modalInstance: { destroy: () => void } | null = null;
-      const settle = (decision: SwitchDecision) => {
-        if (resolved) return;
-        resolved = true;
-        resolve(decision);
-      };
-      const handleDiscard = () => {
-        if (modalInstance) modalInstance.destroy();
-        settle("discard");
-      };
-      modalInstance = modal.confirm({
-        title: "当前段落有未保存标注",
-        content: (
-          <div className="flex flex-col gap-2">
-            <Typography.Text>切换段落前请先保存当前标注。</Typography.Text>
-            <Button type="link" danger style={{ padding: 0, height: "auto" }} onClick={handleDiscard}>
-              放弃未保存并切换
-            </Button>
-          </div>
-        ),
-        okText: "保存并切换",
-        cancelText: "取消",
-        onOk: () => settle("save"),
-        onCancel: () => settle("cancel"),
-      });
-    });
-  }, [modal]);
-
   const requestExport = useCallback((autoAdvance: boolean) => {
     if (!selectedFileId) {
       message.warning("请先选择文件");
@@ -875,7 +679,7 @@ export default function LabelStudioTextEditor() {
   useEffect(() => {
     const handleSaveShortcut = (event: KeyboardEvent) => {
       if (!isSaveShortcut(event) || event.repeat) return;
-      if (saving || loadingTaskDetail || segmentSwitching) return;
+      if (saving || loadingTaskDetail) return;
       if (!iframeReady || !lsReady) return;
       event.preventDefault();
       event.stopPropagation();
@@ -883,83 +687,7 @@ export default function LabelStudioTextEditor() {
     };
     window.addEventListener("keydown", handleSaveShortcut);
     return () => window.removeEventListener("keydown", handleSaveShortcut);
-  }, [iframeReady, loadingTaskDetail, lsReady, requestExport, saving, segmentSwitching]);
+  }, [iframeReady, loadingTaskDetail, lsReady, requestExport, saving]);
-
-  // 段落切换处理
-  const handleSegmentChange = useCallback(async (newIndex: number) => {
-    if (newIndex === currentSegmentIndex) return;
-    if (segmentSwitching || saving || loadingTaskDetail) return;
-    if (!iframeReady || !lsReady) {
-      message.warning("编辑器未就绪，无法切换段落");
-      return;
-    }
-
-    setSegmentSwitching(true);
-    try {
-      const payload = await requestExportForCheck();
-      if (!payload) {
-        message.warning("无法读取当前标注，已取消切换");
-        return;
-      }
-
-      const payloadTaskId = payload.taskId;
-      if (expectedTaskIdRef.current && payloadTaskId) {
-        if (Number(payloadTaskId) !== expectedTaskIdRef.current) {
-          message.warning("已忽略过期的标注数据");
-          return;
-        }
-      }
-
-      const payloadFileId = payload.fileId || selectedFileId;
-      const payloadSegmentIndex = resolveSegmentIndex(payload.segmentIndex);
-      const resolvedSegmentIndex =
-        payloadSegmentIndex !== undefined
-          ? payloadSegmentIndex
-          : segmented
-            ? currentSegmentIndex
-            : undefined;
-      const annotation = isRecord(payload.annotation) ? payload.annotation : undefined;
-      const snapshotKey = payloadFileId
-        ? buildSnapshotKey(String(payloadFileId), resolvedSegmentIndex)
-        : undefined;
-      const latestSnapshot = buildAnnotationSnapshot(annotation);
-      const lastSnapshot = snapshotKey ? savedSnapshotsRef.current[snapshotKey] : undefined;
-      const hasUnsavedChange = snapshotKey !== undefined && lastSnapshot !== undefined && latestSnapshot !== lastSnapshot;
-
-      if (hasUnsavedChange) {
-        if (autoSaveOnSwitch) {
-          const saved = await saveFromExport(payload);
-          if (!saved) return;
-        } else {
-          const decision = await confirmSaveBeforeSwitch();
-          if (decision === "cancel") return;
-          if (decision === "save") {
-            const saved = await saveFromExport(payload);
-            if (!saved) return;
-          }
-        }
-      }
-
-      await initEditorForFile(selectedFileId, newIndex);
-    } finally {
-      setSegmentSwitching(false);
-    }
-  }, [
-    autoSaveOnSwitch,
-    confirmSaveBeforeSwitch,
-    currentSegmentIndex,
-    iframeReady,
-    initEditorForFile,
-    loadingTaskDetail,
-    lsReady,
-    message,
-    requestExportForCheck,
-    saveFromExport,
-    segmented,
-    selectedFileId,
-    segmentSwitching,
-    saving,
-  ]);
 
   useEffect(() => {
     setIframeReady(false);
@@ -977,12 +705,9 @@ export default function LabelStudioTextEditor() {
     expectedTaskIdRef.current = null;
     // 重置分段状态
     setSegmented(false);
-    setSegments([]);
     setCurrentSegmentIndex(0);
+    setSegmentTotal(0);
     savedSnapshotsRef.current = {};
-    segmentStatsSeqRef.current += 1;
-    segmentStatsCacheRef.current = {};
-    segmentStatsLoadingRef.current = new Set();
     if (exportCheckRef.current?.timer) {
       window.clearTimeout(exportCheckRef.current.timer);
     }
@@ -996,12 +721,6 @@ export default function LabelStudioTextEditor() {
     loadTasks({ mode: "reset" });
   }, [project?.supported, loadTasks]);
 
-  useEffect(() => {
-    if (!segmented) return;
-    if (tasks.length === 0) return;
-    prefetchSegmentStats(tasks);
-  }, [prefetchSegmentStats, segmented, tasks]);
-
   useEffect(() => {
     if (!selectedFileId) return;
     initEditorForFile(selectedFileId);
@@ -1026,60 +745,6 @@ export default function LabelStudioTextEditor() {
     return () => window.removeEventListener("focus", handleWindowFocus);
   }, [focusIframe, lsReady]);
 
-  const segmentTreeData = useMemo(() => {
-    if (!segmented || segments.length === 0) return [];
-    const lineMap = new Map<number, SegmentInfo[]>();
-    segments.forEach((seg) => {
-      const list = lineMap.get(seg.lineIndex) || [];
-      list.push(seg);
-      lineMap.set(seg.lineIndex, list);
-    });
-    return Array.from(lineMap.entries())
-      .sort((a, b) => a[0] - b[0])
-      .map(([lineIndex, lineSegments]) => ({
-        key: `line-${lineIndex}`,
-        title: `第${lineIndex + 1}行`,
-        selectable: false,
-        children: lineSegments
-          .sort((a, b) => a.chunkIndex - b.chunkIndex)
-          .map((seg) => ({
-            key: `seg-${seg.idx}`,
-            title: (
-              <span className="flex items-center gap-1">
-                <span>{`片${seg.chunkIndex + 1}`}</span>
-                {seg.hasAnnotation && (
-                  <CheckOutlined style={{ fontSize: 10, color: "#52c41a" }} />
-                )}
-              </span>
-            ),
-          })),
-      }));
-  }, [segmented, segments]);
-
-  const segmentLineKeys = useMemo(
-    () => segmentTreeData.map((item) => String(item.key)),
-    [segmentTreeData]
-  );
-
-  const inProgressSegmentedCount = useMemo(() => {
-    if (tasks.length === 0) return 0;
-    return tasks.reduce((count, item) => {
-      const summary = resolveSegmentSummary(item);
-      if (!summary) return count;
-      return summary.done < summary.total ? count + 1 : count;
-    }, 0);
-  }, [tasks]);
-
-  const handleSegmentSelect = useCallback((keys: Array<string | number>) => {
-    const [first] = keys;
-    if (first === undefined || first === null) return;
-    const key = String(first);
-    if (!key.startsWith("seg-")) return;
-    const nextIndex = Number(key.replace("seg-", ""));
-    if (!Number.isFinite(nextIndex)) return;
-    handleSegmentChange(nextIndex);
-  }, [handleSegmentChange]);
-
   useEffect(() => {
     const handler = (event: MessageEvent<LsfMessage>) => {
       if (event.origin !== origin) return;
@@ -1148,7 +813,7 @@ export default function LabelStudioTextEditor() {
 
   const canLoadMore = taskTotalPages > 0 && taskPage + 1 < taskTotalPages;
   const saveDisabled =
-    !iframeReady || !selectedFileId || saving || segmentSwitching || loadingTaskDetail;
+    !iframeReady || !selectedFileId || saving || loadingTaskDetail;
   const loadMoreNode = canLoadMore ? (
     <div className="p-2 text-center">
       <Button
@@ -1265,11 +930,6 @@ export default function LabelStudioTextEditor() {
         >
           <div className="px-3 py-2 border-b border-gray-200 bg-white font-medium text-sm flex items-center justify-between gap-2">
             <span>文件列表</span>
-            {segmented && (
-              <Tag color="orange" style={{ margin: 0 }}>
-                标注中 {inProgressSegmentedCount}
-              </Tag>
-            )}
           </div>
           <div className="flex-1 min-h-0 overflow-auto">
             <List
@@ -1278,7 +938,6 @@ export default function LabelStudioTextEditor() {
               dataSource={tasks}
               loadMore={loadMoreNode}
               renderItem={(item) => {
-                const segmentSummary = resolveSegmentSummary(item);
                 const statusMeta = resolveTaskStatusMeta(item);
                 return (
                   <List.Item
@@ -1300,11 +959,6 @@ export default function LabelStudioTextEditor() {
                             <Typography.Text type={statusMeta.type} style={{ fontSize: 11 }}>
                               {statusMeta.text}
                             </Typography.Text>
-                            {segmentSummary && (
-                              <Typography.Text type="secondary" style={{ fontSize: 10 }}>
-                                已标注 {segmentSummary.done}/{segmentSummary.total}
-                              </Typography.Text>
-                            )}
                           </div>
                           {item.annotationUpdatedAt && (
                             <Typography.Text type="secondary" style={{ fontSize: 10 }}>
@@ -1323,21 +977,16 @@ export default function LabelStudioTextEditor() {
               <div className="px-3 py-2 border-b border-gray-200 bg-gray-50 font-medium text-sm flex items-center justify-between">
                 <span>段落/分段</span>
                 <Tag color="blue" style={{ margin: 0 }}>
-                  {currentSegmentIndex + 1} / {segments.length}
+                  {segmentTotal > 0 ? currentSegmentIndex + 1 : 0} / {segmentTotal}
                 </Tag>
               </div>
               <div className="flex-1 min-h-0 overflow-auto px-2 py-2">
-                {segments.length > 0 ? (
+                {segmentTotal > 0 ? (
-                  <Tree
+                  <div className="py-6">
-                    showLine
+                    <Typography.Text type="secondary" style={{ fontSize: 12 }}>
-                    blockNode
+                      分段列表已关闭，请使用“保存并跳转到下一段/下一条”顺序标注。
-                    selectedKeys={
+                    </Typography.Text>
-                      segmented ? [`seg-${currentSegmentIndex}`] : []
+                  </div>
-                    }
-                    expandedKeys={segmentLineKeys}
-                    onSelect={handleSegmentSelect}
-                    treeData={segmentTreeData}
-                  />
                 ) : (
                   <div className="py-6">
                     <Empty
@@ -1347,17 +996,6 @@ export default function LabelStudioTextEditor() {
                   </div>
                 )}
               </div>
-              <div className="px-3 py-2 border-t border-gray-200 flex items-center justify-between">
-                <Typography.Text style={{ fontSize: 12 }}>
-                  切段自动保存
-                </Typography.Text>
-                <Switch
-                  size="small"
-                  checked={autoSaveOnSwitch}
-                  onChange={(checked) => setAutoSaveOnSwitch(checked)}
-                  disabled={segmentSwitching || saving || loadingTaskDetail || !lsReady}
-                />
-              </div>
             </div>
           )}
         </div>

frontend/src/pages/DataAnnotation/annotation.api.ts

@@ -3,16 +3,19 @@ import { get, post, put, del, download } from "@/utils/request";
 // 导出格式类型
 export type ExportFormat = "json" | "jsonl" | "csv" | "coco" | "yolo";
 
+type RequestParams = Record<string, unknown>;
+type RequestPayload = Record<string, unknown>;
+
 // 标注任务管理相关接口
-export function queryAnnotationTasksUsingGet(params?: any) {
+export function queryAnnotationTasksUsingGet(params?: RequestParams) {
   return get("/api/annotation/project", params);
 }
 
-export function createAnnotationTaskUsingPost(data: any) {
+export function createAnnotationTaskUsingPost(data: RequestPayload) {
   return post("/api/annotation/project", data);
 }
 
-export function syncAnnotationTaskUsingPost(data: any) {
+export function syncAnnotationTaskUsingPost(data: RequestPayload) {
   return post(`/api/annotation/task/sync`, data);
 }
 
@@ -25,7 +28,7 @@ export function getAnnotationTaskByIdUsingGet(taskId: string) {
   return get(`/api/annotation/project/${taskId}`);
 }
 
-export function updateAnnotationTaskByIdUsingPut(taskId: string, data: any) {
+export function updateAnnotationTaskByIdUsingPut(taskId: string, data: RequestPayload) {
   return put(`/api/annotation/project/${taskId}`, data);
 }
 
@@ -35,17 +38,17 @@ export function getTagConfigUsingGet() {
 }
 
 // 标注模板管理
-export function queryAnnotationTemplatesUsingGet(params?: any) {
+export function queryAnnotationTemplatesUsingGet(params?: RequestParams) {
   return get("/api/annotation/template", params);
 }
 
-export function createAnnotationTemplateUsingPost(data: any) {
+export function createAnnotationTemplateUsingPost(data: RequestPayload) {
   return post("/api/annotation/template", data);
 }
 
 export function updateAnnotationTemplateByIdUsingPut(
   templateId: string | number,
-  data: any
+  data: RequestPayload
 ) {
   return put(`/api/annotation/template/${templateId}`, data);
 }
@@ -65,7 +68,7 @@ export function getEditorProjectInfoUsingGet(projectId: string) {
   return get(`/api/annotation/editor/projects/${projectId}`);
 }
 
-export function listEditorTasksUsingGet(projectId: string, params?: any) {
+export function listEditorTasksUsingGet(projectId: string, params?: RequestParams) {
   return get(`/api/annotation/editor/projects/${projectId}/tasks`, params);
 }
 
@@ -77,11 +80,19 @@ export function getEditorTaskUsingGet(
   return get(`/api/annotation/editor/projects/${projectId}/tasks/${fileId}`, params);
 }
 
+export function getEditorTaskSegmentUsingGet(
+  projectId: string,
+  fileId: string,
+  params: { segmentIndex: number }
+) {
+  return get(`/api/annotation/editor/projects/${projectId}/tasks/${fileId}/segments`, params);
+}
+
 export function upsertEditorAnnotationUsingPut(
   projectId: string,
   fileId: string,
   data: {
-    annotation: any;
+    annotation: Record<string, unknown>;
     expectedUpdatedAt?: string;
     segmentIndex?: number;
   }

frontend/src/pages/DataManagement/Create/CreateDataset.tsx

@@ -4,14 +4,9 @@ import { ArrowLeft } from "lucide-react";
 import { Button, Form, App } from "antd";
 import { Link, useLocation, useNavigate } from "react-router";
 import { createDatasetUsingPost } from "../dataset.api";
-import { datasetTypes } from "../dataset.const";
 import { DatasetType } from "../dataset.model";
 import BasicInformation from "./components/BasicInformation";
 
-const textDatasetTypeOptions = datasetTypes.filter(
-  (type) => type.value === DatasetType.TEXT
-);
-
 export default function DatasetCreate() {
   const navigate = useNavigate();
   const location = useLocation();
@@ -87,7 +82,6 @@ export default function DatasetCreate() {
               data={newDataset}
               setData={setNewDataset}
               hidden={["dataSource"]}
-              datasetTypeOptions={textDatasetTypeOptions}
             />
           </Form>
         </div>

frontend/src/pages/DataManagement/Detail/components/ImportConfiguration.tsx

@@ -5,7 +5,7 @@ import { Dataset, DatasetType, DataSource } from "../../dataset.model";
 import { useCallback, useEffect, useMemo, useState } from "react";
 import { queryTasksUsingGet } from "@/pages/DataCollection/collection.apis";
 import { updateDatasetByIdUsingPut } from "../../dataset.api";
-import { sliceFile } from "@/utils/file.util";
+import { sliceFile, shouldStreamUpload } from "@/utils/file.util";
 import Dragger from "antd/es/upload/Dragger";
 
 const TEXT_FILE_MIME_PREFIX = "text/";
@@ -90,14 +90,16 @@ async function splitFileByLines(file: UploadFile): Promise<UploadFile[]> {
   const lines = text.split(/\r?\n/).filter((line: string) => line.trim() !== "");
   if (lines.length === 0) return [];
 
-  // 生成文件名：原文件名_序号.扩展名
+  // 生成文件名：原文件名_序号（不保留后缀）
   const nameParts = file.name.split(".");
-  const ext = nameParts.length > 1 ? "." + nameParts.pop() : "";
+  if (nameParts.length > 1) {
+    nameParts.pop();
+  }
   const baseName = nameParts.join(".");
   const padLength = String(lines.length).length;
 
   return lines.map((line: string, index: number) => {
-    const newFileName = `${baseName}_${String(index + 1).padStart(padLength, "0")}${ext}`;
+    const newFileName = `${baseName}_${String(index + 1).padStart(padLength, "0")}`;
     const blob = new Blob([line], { type: "text/plain" });
     const newFile = new File([blob], newFileName, { type: "text/plain" });
     return {
@@ -164,17 +166,75 @@ export default function ImportConfiguration({
   // 本地上传文件相关逻辑
 
   const handleUpload = async (dataset: Dataset) => {
-    let filesToUpload =
+    const filesToUpload =
       (form.getFieldValue("files") as UploadFile[] | undefined) || [];
 
-    // 如果启用分行分割，处理文件
+    // 如果启用分行分割，对大文件使用流式处理
     if (importConfig.splitByLine && !hasNonTextFile) {
-      const splitResults = await Promise.all(
+      // 检查是否有大文件需要流式分割上传
-        filesToUpload.map((file) => splitFileByLines(file))
+      const filesForStreamUpload: File[] = [];
-      );
+      const filesForNormalUpload: UploadFile[] = [];
-      filesToUpload = splitResults.flat();
+
+      for (const file of filesToUpload) {
+        const originFile = file.originFileObj ?? file;
+        if (originFile instanceof File && shouldStreamUpload(originFile)) {
+          filesForStreamUpload.push(originFile);
+        } else {
+          filesForNormalUpload.push(file);
+        }
+      }
+
+      // 大文件使用流式分割上传
+      if (filesForStreamUpload.length > 0) {
+        window.dispatchEvent(
+          new CustomEvent("upload:dataset-stream", {
+            detail: {
+              dataset,
+              files: filesForStreamUpload,
+              updateEvent,
+              hasArchive: importConfig.hasArchive,
+              prefix: currentPrefix,
+            },
+          })
+        );
+      }
+
+      // 小文件使用传统分割方式
+      if (filesForNormalUpload.length > 0) {
+        const splitResults = await Promise.all(
+          filesForNormalUpload.map((file) => splitFileByLines(file))
+        );
+        const smallFilesToUpload = splitResults.flat();
+        
+        // 计算分片列表
+        const sliceList = smallFilesToUpload.map((file) => {
+          const originFile = (file.originFileObj ?? file) as Blob;
+          const slices = sliceFile(originFile);
+          return {
+            originFile: originFile,
+            slices,
+            name: file.name,
+            size: originFile.size || 0,
+          };
+        });
+
+        console.log("[ImportConfiguration] Uploading small files with currentPrefix:", currentPrefix);
+        window.dispatchEvent(
+          new CustomEvent("upload:dataset", {
+            detail: {
+              dataset,
+              files: sliceList,
+              updateEvent,
+              hasArchive: importConfig.hasArchive,
+              prefix: currentPrefix,
+            },
+          })
+        );
+      }
+      return;
     }
 
+    // 未启用分行分割，使用普通上传
     // 计算分片列表
     const sliceList = filesToUpload.map((file) => {
       const originFile = (file.originFileObj ?? file) as Blob;
@@ -234,6 +294,10 @@ export default function ImportConfiguration({
     if (!data) return;
     console.log('[ImportConfiguration] handleImportData called, currentPrefix:', currentPrefix);
     if (importConfig.source === DataSource.UPLOAD) {
+      // 立即显示任务中心，让用户感知上传已开始（在文件分割等耗时操作之前）
+      window.dispatchEvent(
+        new CustomEvent("show:task-popover", { detail: { show: true } })
+      );
       await handleUpload(data);
     } else if (importConfig.source === DataSource.COLLECTION) {
       await updateDatasetByIdUsingPut(data.id, {

frontend/src/pages/DataManagement/dataset.model.ts

@@ -102,6 +102,13 @@ export interface DatasetTask {
   executionHistory?: { time: string; status: string }[];
 }
 
+export interface StreamUploadInfo {
+  currentFile: string;
+  fileIndex: number;
+  totalFiles: number;
+  uploadedLines: number;
+}
+
 export interface TaskItem {
   key: string;
   title: string;
@@ -113,4 +120,6 @@ export interface TaskItem {
   updateEvent?: string;
   size?: number;
   hasArchive?: boolean;
+  prefix?: string;
+  streamUploadInfo?: StreamUploadInfo;
 }

frontend/src/pages/Layout/TaskUpload.tsx

@@ -3,25 +3,28 @@ import {
   preUploadUsingPost,
   uploadFileChunkUsingPost,
 } from "@/pages/DataManagement/dataset.api";
-import { Button, Empty, Progress } from "antd";
+import { Button, Empty, Progress, Tag } from "antd";
-import { DeleteOutlined } from "@ant-design/icons";
+import { DeleteOutlined, FileTextOutlined } from "@ant-design/icons";
 import { useEffect } from "react";
 import { useFileSliceUpload } from "@/hooks/useSliceUpload";
 
 export default function TaskUpload() {
-  const { createTask, taskList, removeTask, handleUpload } = useFileSliceUpload(
+  const { createTask, taskList, removeTask, handleUpload, registerStreamUploadListener } = useFileSliceUpload(
     {
       preUpload: preUploadUsingPost,
       uploadChunk: uploadFileChunkUsingPost,
       cancelUpload: cancelUploadUsingPut,
-    }
+    },
+    true, // showTaskCenter
+    true  // enableStreamUpload
   );
 
   useEffect(() => {
-    const uploadHandler = (e: any) => {
+    const uploadHandler = (e: Event) => {
-      console.log('[TaskUpload] Received upload event detail:', e.detail);
+      const customEvent = e as CustomEvent;
-      const { files } = e.detail;
+      console.log('[TaskUpload] Received upload event detail:', customEvent.detail);
-      const task = createTask(e.detail);
+      const { files } = customEvent.detail;
+      const task = createTask(customEvent.detail);
       console.log('[TaskUpload] Created task with prefix:', task.prefix);
       handleUpload({ task, files });
     };
@@ -29,7 +32,13 @@ export default function TaskUpload() {
     return () => {
       window.removeEventListener("upload:dataset", uploadHandler);
     };
-  }, []);
+  }, [createTask, handleUpload]);
+
+  // 注册流式上传监听器
+  useEffect(() => {
+    const unregister = registerStreamUploadListener();
+    return unregister;
+  }, [registerStreamUploadListener]);
 
   return (
     <div
@@ -55,7 +64,22 @@ export default function TaskUpload() {
               ></Button>
             </div>
 
-            <Progress size="small" percent={task.percent} />
+            <Progress size="small" percent={Number(task.percent)} />
+            {task.streamUploadInfo && (
+              <div className="flex items-center gap-2 text-xs text-gray-500 mt-1">
+                <Tag icon={<FileTextOutlined />} size="small">
+                  按行分割
+                </Tag>
+                <span>
+                  已上传: {task.streamUploadInfo.uploadedLines} 行
+                </span>
+                {task.streamUploadInfo.totalFiles > 1 && (
+                  <span>
+                    ({task.streamUploadInfo.fileIndex}/{task.streamUploadInfo.totalFiles} 文件)
+                  </span>
+                )}
+              </div>
+            )}
           </div>
         ))}
       {taskList.length === 0 && (

frontend/src/utils/file.util.ts

@@ -1,79 +1,657 @@
 import { UploadFile } from "antd";
 import jsSHA from "jssha";
 
-const CHUNK_SIZE = 1024 * 1024 * 60;
+// 默认分片大小：5MB（适合大多数网络环境）
+export const DEFAULT_CHUNK_SIZE = 1024 * 1024 * 5;
+// 大文件阈值：10MB
+export const LARGE_FILE_THRESHOLD = 1024 * 1024 * 10;
+// 最大并发上传数
+export const MAX_CONCURRENT_UPLOADS = 3;
+// 文本文件读取块大小：20MB（用于计算 SHA256）
+const BUFFER_CHUNK_SIZE = 1024 * 1024 * 20;
 
-export function sliceFile(file, chunkSize = CHUNK_SIZE): Blob[] {
+/**
+ * 将文件分割为多个分片
+ * @param file 文件对象
+ * @param chunkSize 分片大小（字节），默认 5MB
+ * @returns 分片数组（Blob 列表）
+ */
+export function sliceFile(file: Blob, chunkSize = DEFAULT_CHUNK_SIZE): Blob[] {
   const totalSize = file.size;
+  const chunks: Blob[] = [];
+  
+  // 小文件不需要分片
+  if (totalSize <= chunkSize) {
+    return [file];
+  }
+  
   let start = 0;
-  let end = start + chunkSize;
-  const chunks = [];
   while (start < totalSize) {
+    const end = Math.min(start + chunkSize, totalSize);
     const blob = file.slice(start, end);
     chunks.push(blob);
-
     start = end;
-    end = start + chunkSize;
   }
+  
   return chunks;
 }
 
-export function calculateSHA256(file: Blob): Promise<string> {
+/**
-  let count = 0;
+ * 计算文件的 SHA256 哈希值
-  const hash = new jsSHA("SHA-256", "ARRAYBUFFER", { encoding: "UTF8" });
+ * @param file 文件 Blob
+ * @param onProgress 进度回调（可选）
+ * @returns SHA256 哈希字符串
+ */
+export function calculateSHA256(
+  file: Blob, 
+  onProgress?: (percent: number) => void
+): Promise<string> {
   return new Promise((resolve, reject) => {
+    const hash = new jsSHA("SHA-256", "ARRAYBUFFER", { encoding: "UTF8" });
     const reader = new FileReader();
+    let processedSize = 0;
 
     function readChunk(start: number, end: number) {
       const slice = file.slice(start, end);
       reader.readAsArrayBuffer(slice);
     }
 
-    const bufferChunkSize = 1024 * 1024 * 20;
-
     function processChunk(offset: number) {
       const start = offset;
-      const end = Math.min(start + bufferChunkSize, file.size);
+      const end = Math.min(start + BUFFER_CHUNK_SIZE, file.size);
-      count = end;
-
       readChunk(start, end);
     }
 
-    reader.onloadend = function () {
+    reader.onloadend = function (e) {
-      const arraybuffer = reader.result;
+      const arraybuffer = reader.result as ArrayBuffer;
+      if (!arraybuffer) {
+        reject(new Error("Failed to read file"));
+        return;
+      }
 
       hash.update(arraybuffer);
-      if (count < file.size) {
+      processedSize += (e.target as FileReader).result?.byteLength || 0;
-        processChunk(count);
+      
+      if (onProgress) {
+        const percent = Math.min(100, Math.round((processedSize / file.size) * 100));
+        onProgress(percent);
+      }
+      
+      if (processedSize < file.size) {
+        processChunk(processedSize);
       } else {
         resolve(hash.getHash("HEX", { outputLen: 256 }));
       }
     };
 
+    reader.onerror = () => reject(new Error("File reading failed"));
     processChunk(0);
   });
 }
 
+/**
+ * 批量计算多个文件的 SHA256
+ * @param files 文件列表
+ * @param onFileProgress 单个文件进度回调（可选）
+ * @returns 哈希值数组
+ */
+export async function calculateSHA256Batch(
+  files: Blob[],
+  onFileProgress?: (index: number, percent: number) => void
+): Promise<string[]> {
+  const results: string[] = [];
+  
+  for (let i = 0; i < files.length; i++) {
+    const hash = await calculateSHA256(files[i], (percent) => {
+      onFileProgress?.(i, percent);
+    });
+    results.push(hash);
+  }
+  
+  return results;
+}
+
+/**
+ * 检查文件是否存在（未被修改或删除）
+ * @param fileList 文件列表
+ * @returns 返回第一个不存在的文件，或 null（如果都存在）
+ */
 export function checkIsFilesExist(
-  fileList: UploadFile[]
+  fileList: Array<{ originFile?: Blob }>
-): Promise<UploadFile | null> {
+): Promise<{ originFile?: Blob } | null> {
   return new Promise((resolve) => {
-    const loadEndFn = (file: UploadFile, reachEnd: boolean, e) => {
+    if (!fileList.length) {
-      const fileNotExist = !e.target.result;
+      resolve(null);
+      return;
+    }
+
+    let checkedCount = 0;
+    const totalCount = fileList.length;
+
+    const loadEndFn = (file: { originFile?: Blob }, e: ProgressEvent<FileReader>) => {
+      checkedCount++;
+      const fileNotExist = !e.target?.result;
       if (fileNotExist) {
         resolve(file);
+        return;
       }
-      if (reachEnd) {
+      if (checkedCount >= totalCount) {
         resolve(null);
       }
     };
 
-    for (let i = 0; i < fileList.length; i++) {
+    for (const file of fileList) {
-      const { originFile: file } = fileList[i];
       const fileReader = new FileReader();
-      fileReader.readAsArrayBuffer(file);
+      const actualFile = file.originFile;
-      fileReader.onloadend = (e) =>
+      
-        loadEndFn(fileList[i], i === fileList.length - 1, e);
+      if (!actualFile) {
+        checkedCount++;
+        if (checkedCount >= totalCount) {
+          resolve(null);
+        }
+        continue;
+      }
+      
+      fileReader.readAsArrayBuffer(actualFile.slice(0, 1));
+      fileReader.onloadend = (e) => loadEndFn(file, e);
+      fileReader.onerror = () => {
+        checkedCount++;
+        resolve(file);
+      };
     }
   });
 }
+
+/**
+ * 判断文件是否为大文件
+ * @param size 文件大小（字节）
+ * @param threshold 阈值（字节），默认 10MB
+ */
+export function isLargeFile(size: number, threshold = LARGE_FILE_THRESHOLD): boolean {
+  return size > threshold;
+}
+
+/**
+ * 格式化文件大小为人类可读格式
+ * @param bytes 字节数
+ * @param decimals 小数位数
+ */
+export function formatFileSize(bytes: number, decimals = 2): string {
+  if (bytes === 0) return "0 B";
+  
+  const k = 1024;
+  const sizes = ["B", "KB", "MB", "GB", "TB", "PB"];
+  const i = Math.floor(Math.log(bytes) / Math.log(k));
+  
+  return `${parseFloat((bytes / Math.pow(k, i)).toFixed(decimals))} ${sizes[i]}`;
+}
+
+/**
+ * 并发执行异步任务
+ * @param tasks 任务函数数组
+ * @param maxConcurrency 最大并发数
+ * @param onTaskComplete 单个任务完成回调（可选）
+ */
+export async function runConcurrentTasks<T>(
+  tasks: (() => Promise<T>)[],
+  maxConcurrency: number,
+  onTaskComplete?: (index: number, result: T) => void
+): Promise<T[]> {
+  const results: T[] = new Array(tasks.length);
+  let index = 0;
+
+  async function runNext(): Promise<void> {
+    const currentIndex = index++;
+    if (currentIndex >= tasks.length) return;
+
+    const result = await tasks[currentIndex]();
+    results[currentIndex] = result;
+    onTaskComplete?.(currentIndex, result);
+    
+    await runNext();
+  }
+
+  const workers = Array(Math.min(maxConcurrency, tasks.length))
+    .fill(null)
+    .map(() => runNext());
+
+  await Promise.all(workers);
+  return results;
+}
+
+/**
+ * 按行分割文本文件内容
+ * @param text 文本内容
+ * @param skipEmptyLines 是否跳过空行，默认 true
+ * @returns 行数组
+ */
+export function splitTextByLines(text: string, skipEmptyLines = true): string[] {
+  const lines = text.split(/\r?\n/);
+  if (skipEmptyLines) {
+    return lines.filter((line) => line.trim() !== "");
+  }
+  return lines;
+}
+
+/**
+ * 创建分片信息对象
+ * @param file 原始文件
+ * @param chunkSize 分片大小
+ */
+export function createFileSliceInfo(
+  file: File | Blob,
+  chunkSize = DEFAULT_CHUNK_SIZE
+): {
+  originFile: Blob;
+  slices: Blob[];
+  name: string;
+  size: number;
+  totalChunks: number;
+} {
+  const slices = sliceFile(file, chunkSize);
+  return {
+    originFile: file,
+    slices,
+    name: (file as File).name || "unnamed",
+    size: file.size,
+    totalChunks: slices.length,
+  };
+}
+
+/**
+ * 支持的文本文件 MIME 类型前缀
+ */
+export const TEXT_FILE_MIME_PREFIX = "text/";
+
+/**
+ * 支持的文本文件 MIME 类型集合
+ */
+export const TEXT_FILE_MIME_TYPES = new Set([
+  "application/json",
+  "application/xml",
+  "application/csv",
+  "application/ndjson",
+  "application/x-ndjson",
+  "application/x-yaml",
+  "application/yaml",
+  "application/javascript",
+  "application/x-javascript",
+  "application/sql",
+  "application/rtf",
+  "application/xhtml+xml",
+  "application/svg+xml",
+]);
+
+/**
+ * 支持的文本文件扩展名集合
+ */
+export const TEXT_FILE_EXTENSIONS = new Set([
+  ".txt",
+  ".md",
+  ".markdown",
+  ".csv",
+  ".tsv",
+  ".json",
+  ".jsonl",
+  ".ndjson",
+  ".log",
+  ".xml",
+  ".yaml",
+  ".yml",
+  ".sql",
+  ".js",
+  ".ts",
+  ".jsx",
+  ".tsx",
+  ".html",
+  ".htm",
+  ".css",
+  ".scss",
+  ".less",
+  ".py",
+  ".java",
+  ".c",
+  ".cpp",
+  ".h",
+  ".hpp",
+  ".go",
+  ".rs",
+  ".rb",
+  ".php",
+  ".sh",
+  ".bash",
+  ".zsh",
+  ".ps1",
+  ".bat",
+  ".cmd",
+  ".svg",
+  ".rtf",
+]);
+
+/**
+ * 判断文件是否为文本文件（支持 UploadFile 类型）
+ * @param file UploadFile 对象
+ */
+export function isTextUploadFile(file: UploadFile): boolean {
+  const mimeType = (file.type || "").toLowerCase();
+  if (mimeType) {
+    if (mimeType.startsWith(TEXT_FILE_MIME_PREFIX)) return true;
+    if (TEXT_FILE_MIME_TYPES.has(mimeType)) return true;
+  }
+
+  const fileName = file.name || "";
+  const dotIndex = fileName.lastIndexOf(".");
+  if (dotIndex < 0) return false;
+  const ext = fileName.slice(dotIndex).toLowerCase();
+  return TEXT_FILE_EXTENSIONS.has(ext);
+}
+
+/**
+ * 判断文件名是否为文本文件
+ * @param fileName 文件名
+ */
+export function isTextFileByName(fileName: string): boolean {
+  const lowerName = fileName.toLowerCase();
+  
+  // 先检查 MIME 类型（如果有）
+  // 这里简化处理，主要通过扩展名判断
+  
+  const dotIndex = lowerName.lastIndexOf(".");
+  if (dotIndex < 0) return false;
+  const ext = lowerName.slice(dotIndex);
+  return TEXT_FILE_EXTENSIONS.has(ext);
+}
+
+/**
+ * 获取文件扩展名
+ * @param fileName 文件名
+ */
+export function getFileExtension(fileName: string): string {
+  const dotIndex = fileName.lastIndexOf(".");
+  if (dotIndex < 0) return "";
+  return fileName.slice(dotIndex).toLowerCase();
+}
+
+/**
+ * 安全地读取文件为文本
+ * @param file 文件对象
+ * @param encoding 编码，默认 UTF-8
+ */
+export function readFileAsText(
+  file: File | Blob,
+  encoding = "UTF-8"
+): Promise<string> {
+  return new Promise((resolve, reject) => {
+    const reader = new FileReader();
+    reader.onload = (e) => resolve(e.target?.result as string);
+    reader.onerror = () => reject(new Error("Failed to read file"));
+    reader.readAsText(file, encoding);
+  });
+}
+
+/**
+ * 流式分割文件并逐行上传
+ * 使用 Blob.slice 逐块读取，避免一次性加载大文件到内存
+ * @param file 文件对象
+ * @param datasetId 数据集ID
+ * @param uploadFn 上传函数，接收 FormData 和配置，返回 Promise
+ * @param onProgress 进度回调 (currentBytes, totalBytes, uploadedLines)
+ * @param chunkSize 每次读取的块大小，默认 1MB
+ * @param options 其他选项
+ * @returns 上传结果统计
+ */
+export interface StreamUploadOptions {
+  reqId?: number;
+  resolveReqId?: (params: { totalFileNum: number; totalSize: number }) => Promise<number>;
+  onReqIdResolved?: (reqId: number) => void;
+  fileNamePrefix?: string;
+  hasArchive?: boolean;
+  prefix?: string;
+  signal?: AbortSignal;
+  maxConcurrency?: number;
+}
+
+export interface StreamUploadResult {
+  uploadedCount: number;
+  totalBytes: number;
+  skippedEmptyCount: number;
+}
+
+async function processFileLines(
+  file: File,
+  chunkSize: number,
+  signal: AbortSignal | undefined,
+  onLine?: (line: string, index: number) => Promise<void> | void,
+  onProgress?: (currentBytes: number, totalBytes: number, processedLines: number) => void
+): Promise<{ lineCount: number; skippedEmptyCount: number }> {
+  const fileSize = file.size;
+  let offset = 0;
+  let buffer = "";
+  let skippedEmptyCount = 0;
+  let lineIndex = 0;
+
+  while (offset < fileSize) {
+    if (signal?.aborted) {
+      throw new Error("Upload cancelled");
+    }
+
+    const end = Math.min(offset + chunkSize, fileSize);
+    const chunk = file.slice(offset, end);
+    const text = await readFileAsText(chunk);
+    const combined = buffer + text;
+    const lines = combined.split(/\r?\n/);
+    buffer = lines.pop() || "";
+
+    for (const line of lines) {
+      if (signal?.aborted) {
+        throw new Error("Upload cancelled");
+      }
+      if (!line.trim()) {
+        skippedEmptyCount++;
+        continue;
+      }
+      const currentIndex = lineIndex;
+      lineIndex += 1;
+      if (onLine) {
+        await onLine(line, currentIndex);
+      }
+    }
+
+    offset = end;
+    onProgress?.(offset, fileSize, lineIndex);
+  }
+
+  if (buffer.trim()) {
+    const currentIndex = lineIndex;
+    lineIndex += 1;
+    if (onLine) {
+      await onLine(buffer, currentIndex);
+    }
+  } else if (buffer.length > 0) {
+    skippedEmptyCount++;
+  }
+
+  return { lineCount: lineIndex, skippedEmptyCount };
+}
+
+export async function streamSplitAndUpload(
+  file: File,
+  uploadFn: (formData: FormData, config?: { onUploadProgress?: (e: { loaded: number; total: number }) => void }) => Promise<unknown>,
+  onProgress?: (currentBytes: number, totalBytes: number, uploadedLines: number) => void,
+  chunkSize: number = 1024 * 1024, // 1MB
+  options: StreamUploadOptions
+): Promise<StreamUploadResult> {
+  const {
+    reqId: initialReqId,
+    resolveReqId,
+    onReqIdResolved,
+    fileNamePrefix,
+    prefix,
+    signal,
+    maxConcurrency = 3,
+  } = options;
+
+  const fileSize = file.size;
+  let uploadedCount = 0;
+  let skippedEmptyCount = 0;
+  
+  // 获取文件名基础部分和扩展名
+  const originalFileName = fileNamePrefix || file.name;
+  const lastDotIndex = originalFileName.lastIndexOf(".");
+  const baseName = lastDotIndex > 0 ? originalFileName.slice(0, lastDotIndex) : originalFileName;
+  const fileExtension = lastDotIndex > 0 ? originalFileName.slice(lastDotIndex) : "";
+
+  let resolvedReqId = initialReqId;
+  if (!resolvedReqId) {
+    const scanResult = await processFileLines(file, chunkSize, signal);
+    const totalFileNum = scanResult.lineCount;
+    skippedEmptyCount = scanResult.skippedEmptyCount;
+    if (totalFileNum === 0) {
+      return {
+        uploadedCount: 0,
+        totalBytes: fileSize,
+        skippedEmptyCount,
+      };
+    }
+    if (signal?.aborted) {
+      throw new Error("Upload cancelled");
+    }
+    if (!resolveReqId) {
+      throw new Error("Missing pre-upload request id");
+    }
+    resolvedReqId = await resolveReqId({ totalFileNum, totalSize: fileSize });
+    if (!resolvedReqId) {
+      throw new Error("Failed to resolve pre-upload request id");
+    }
+    onReqIdResolved?.(resolvedReqId);
+  }
+  if (!resolvedReqId) {
+    throw new Error("Missing pre-upload request id");
+  }
+
+  /**
+   * 上传单行内容
+   * 每行作为独立文件上传，fileNo 对应行序号，chunkNo 固定为 1
+   */
+  async function uploadLine(line: string, index: number): Promise<void> {
+    // 检查是否已取消
+    if (signal?.aborted) {
+      throw new Error("Upload cancelled");
+    }
+    
+    if (!line.trim()) {
+      skippedEmptyCount++;
+      return;
+    }
+
+    // 保留原始文件扩展名
+    const fileIndex = index + 1;
+    const newFileName = `${baseName}_${String(fileIndex).padStart(6, "0")}${fileExtension}`;
+    const blob = new Blob([line], { type: "text/plain" });
+    const lineFile = new File([blob], newFileName, { type: "text/plain" });
+    
+    // 计算分片（小文件通常只需要一个分片）
+    const slices = sliceFile(lineFile, DEFAULT_CHUNK_SIZE);
+    const checkSum = await calculateSHA256(slices[0]);
+    
+    // 检查是否已取消（计算哈希后）
+    if (signal?.aborted) {
+      throw new Error("Upload cancelled");
+    }
+    
+    const formData = new FormData();
+    formData.append("file", slices[0]);
+    formData.append("reqId", resolvedReqId.toString());
+    // 每行作为独立文件上传
+    formData.append("fileNo", fileIndex.toString());
+    formData.append("chunkNo", "1");
+    formData.append("fileName", newFileName);
+    formData.append("fileSize", lineFile.size.toString());
+    formData.append("totalChunkNum", "1");
+    formData.append("checkSumHex", checkSum);
+    if (prefix !== undefined) {
+      formData.append("prefix", prefix);
+    }
+
+    await uploadFn(formData, {
+      onUploadProgress: () => {
+        // 单行文件很小，进度主要用于追踪上传状态
+      },
+    });
+  }
+
+  const inFlight = new Set<Promise<void>>();
+  let uploadError: unknown = null;
+  const enqueueUpload = async (line: string, index: number) => {
+    if (signal?.aborted) {
+      throw new Error("Upload cancelled");
+    }
+    if (uploadError) {
+      throw uploadError;
+    }
+    const uploadPromise = uploadLine(line, index)
+      .then(() => {
+        uploadedCount++;
+      })
+      .catch((err) => {
+        uploadError = err;
+      });
+    inFlight.add(uploadPromise);
+    uploadPromise.finally(() => inFlight.delete(uploadPromise));
+    if (inFlight.size >= maxConcurrency) {
+      await Promise.race(inFlight);
+      if (uploadError) {
+        throw uploadError;
+      }
+    }
+  };
+
+  let uploadResult: { lineCount: number; skippedEmptyCount: number } | null = null;
+  try {
+    uploadResult = await processFileLines(
+      file,
+      chunkSize,
+      signal,
+      enqueueUpload,
+      (currentBytes, totalBytes) => {
+        onProgress?.(currentBytes, totalBytes, uploadedCount);
+      }
+    );
+    if (uploadError) {
+      throw uploadError;
+    }
+  } finally {
+    if (inFlight.size > 0) {
+      await Promise.allSettled(inFlight);
+    }
+  }
+
+  if (!uploadResult || (initialReqId && uploadResult.lineCount === 0)) {
+    return {
+      uploadedCount: 0,
+      totalBytes: fileSize,
+      skippedEmptyCount: uploadResult?.skippedEmptyCount ?? 0,
+    };
+  }
+
+  if (!initialReqId) {
+    skippedEmptyCount = skippedEmptyCount || uploadResult.skippedEmptyCount;
+  } else {
+    skippedEmptyCount = uploadResult.skippedEmptyCount;
+  }
+
+  return {
+    uploadedCount,
+    totalBytes: fileSize,
+    skippedEmptyCount,
+  };
+}
+
+/**
+ * 判断文件是否需要流式分割上传
+ * @param file 文件对象
+ * @param threshold 阈值，默认 5MB
+ */
+export function shouldStreamUpload(file: File, threshold: number = 5 * 1024 * 1024): boolean {
+  return file.size > threshold;
+}

frontend/src/utils/request.ts

@@ -92,6 +92,14 @@ class Request {
         });
       }
 
+      // 监听 AbortSignal 来中止请求
+      if (config.signal) {
+        config.signal.addEventListener("abort", () => {
+          xhr.abort();
+          reject(new Error("上传已取消"));
+        });
+      }
+
       // 监听上传进度
       xhr.upload.addEventListener("progress", function (event) {
         if (event.lengthComputable) {

runtime/datamate-python/app/module/annotation/interface/editor.py

@@ -19,6 +19,7 @@ from app.db.session import get_db
 from app.module.annotation.schema.editor import (
     EditorProjectInfo,
     EditorTaskListResponse,
+    EditorTaskSegmentResponse,
     EditorTaskResponse,
     UpsertAnnotationRequest,
     UpsertAnnotationResponse,
@@ -87,6 +88,21 @@ async def get_editor_task(
     return StandardResponse(code=200, message="success", data=task)
 
 
+@router.get(
+    "/projects/{project_id}/tasks/{file_id}/segments",
+    response_model=StandardResponse[EditorTaskSegmentResponse],
+)
+async def get_editor_task_segment(
+    project_id: str = Path(..., description="标注项目ID（t_dm_labeling_projects.id）"),
+    file_id: str = Path(..., description="文件ID（t_dm_dataset_files.id）"),
+    segment_index: int = Query(..., ge=0, alias="segmentIndex", description="段落索引（从0开始）"),
+    db: AsyncSession = Depends(get_db),
+):
+    service = AnnotationEditorService(db)
+    result = await service.get_task_segment(project_id, file_id, segment_index)
+    return StandardResponse(code=200, message="success", data=result)
+
+
 @router.put(
     "/projects/{project_id}/tasks/{file_id}/annotation",
     response_model=StandardResponse[UpsertAnnotationResponse],

runtime/datamate-python/app/module/annotation/interface/project.py

@@ -150,6 +150,18 @@ async def create_mapping(
             labeling_project, snapshot_file_ids
         )
 
+        # 如果启用了分段且为文本数据集，预生成切片结构
+        if dataset_type == TEXT_DATASET_TYPE and request.segmentation_enabled:
+            try:
+                from ..service.editor import AnnotationEditorService
+                editor_service = AnnotationEditorService(db)
+                # 异步预计算切片（不阻塞创建响应）
+                segmentation_result = await editor_service.precompute_segmentation_for_project(labeling_project.id)
+                logger.info(f"Precomputed segmentation for project {labeling_project.id}: {segmentation_result}")
+            except Exception as e:
+                logger.warning(f"Failed to precompute segmentation for project {labeling_project.id}: {e}")
+                # 不影响项目创建，只记录警告
+
         response_data = DatasetMappingCreateResponse(
             id=mapping.id,
             labeling_project_id=str(mapping.labeling_project_id),

runtime/datamate-python/app/module/annotation/schema/editor.py

@@ -79,12 +79,9 @@ class EditorTaskListResponse(BaseModel):
 
 
 class SegmentInfo(BaseModel):
-    """段落信息（用于文本分段标注）"""
+    """段落摘要（用于文本分段标注）"""
 
     idx: int = Field(..., description="段落索引")
-    text: str = Field(..., description="段落文本")
-    start: int = Field(..., description="在原文中的起始位置")
-    end: int = Field(..., description="在原文中的结束位置")
     has_annotation: bool = Field(False, alias="hasAnnotation", description="该段落是否已有标注")
     line_index: int = Field(0, alias="lineIndex", description="JSONL 行索引（从0开始）")
     chunk_index: int = Field(0, alias="chunkIndex", description="行内分片索引（从0开始）")
@@ -100,7 +97,29 @@ class EditorTaskResponse(BaseModel):
 
     # 分段相关字段
     segmented: bool = Field(False, description="是否启用分段模式")
-    segments: Optional[List[SegmentInfo]] = Field(None, description="段落列表")
+    total_segments: int = Field(0, alias="totalSegments", description="总段落数")
+    current_segment_index: int = Field(0, alias="currentSegmentIndex", description="当前段落索引")
+
+    model_config = ConfigDict(populate_by_name=True)
+
+
+class SegmentDetail(BaseModel):
+    """段落内容"""
+
+    idx: int = Field(..., description="段落索引")
+    text: str = Field(..., description="段落文本")
+    has_annotation: bool = Field(False, alias="hasAnnotation", description="该段落是否已有标注")
+    line_index: int = Field(0, alias="lineIndex", description="JSONL 行索引（从0开始）")
+    chunk_index: int = Field(0, alias="chunkIndex", description="行内分片索引（从0开始）")
+
+    model_config = ConfigDict(populate_by_name=True)
+
+
+class EditorTaskSegmentResponse(BaseModel):
+    """编辑器单段内容响应"""
+
+    segmented: bool = Field(False, description="是否启用分段模式")
+    segment: Optional[SegmentDetail] = Field(None, description="段落内容")
     total_segments: int = Field(0, alias="totalSegments", description="总段落数")
     current_segment_index: int = Field(0, alias="currentSegmentIndex", description="当前段落索引")
 

runtime/datamate-python/app/module/annotation/service/editor.py

@@ -36,7 +36,9 @@ from app.module.annotation.schema.editor import (
     EditorProjectInfo,
     EditorTaskListItem,
     EditorTaskListResponse,
+    EditorTaskSegmentResponse,
     EditorTaskResponse,
+    SegmentDetail,
     SegmentInfo,
     UpsertAnnotationRequest,
     UpsertAnnotationResponse,
@@ -538,6 +540,50 @@ class AnnotationEditorService:
                 return value
         return raw_text
 
+    def _build_segment_contexts(
+        self,
+        records: List[Tuple[Optional[Dict[str, Any]], str]],
+        record_texts: List[str],
+        segment_annotation_keys: set[str],
+    ) -> Tuple[List[SegmentInfo], List[Tuple[Optional[Dict[str, Any]], str, str, int, int]]]:
+        splitter = AnnotationTextSplitter(max_chars=self.SEGMENT_THRESHOLD)
+        segments: List[SegmentInfo] = []
+        segment_contexts: List[Tuple[Optional[Dict[str, Any]], str, str, int, int]] = []
+        segment_cursor = 0
+
+        for record_index, ((payload, raw_text), record_text) in enumerate(zip(records, record_texts)):
+            normalized_text = record_text or ""
+            if len(normalized_text) > self.SEGMENT_THRESHOLD:
+                raw_segments = splitter.split(normalized_text)
+                for chunk_index, seg in enumerate(raw_segments):
+                    segments.append(
+                        SegmentInfo(
+                            idx=segment_cursor,
+                            hasAnnotation=str(segment_cursor) in segment_annotation_keys,
+                            lineIndex=record_index,
+                            chunkIndex=chunk_index,
+                        )
+                    )
+                    segment_contexts.append((payload, raw_text, seg["text"], record_index, chunk_index))
+                    segment_cursor += 1
+            else:
+                segments.append(
+                    SegmentInfo(
+                        idx=segment_cursor,
+                        hasAnnotation=str(segment_cursor) in segment_annotation_keys,
+                        lineIndex=record_index,
+                        chunkIndex=0,
+                    )
+                )
+                segment_contexts.append((payload, raw_text, normalized_text, record_index, 0))
+                segment_cursor += 1
+
+        if not segments:
+            segments = [SegmentInfo(idx=0, hasAnnotation=False, lineIndex=0, chunkIndex=0)]
+            segment_contexts = [(None, "", "", 0, 0)]
+
+        return segments, segment_contexts
+
     async def get_project_info(self, project_id: str) -> EditorProjectInfo:
         project = await self._get_project_or_404(project_id)
 
@@ -668,6 +714,124 @@ class AnnotationEditorService:
 
         return await self._build_text_task(project, file_record, file_id, segment_index)
 
+    async def get_task_segment(
+        self,
+        project_id: str,
+        file_id: str,
+        segment_index: int,
+    ) -> EditorTaskSegmentResponse:
+        project = await self._get_project_or_404(project_id)
+
+        dataset_type = self._normalize_dataset_type(await self._get_dataset_type(project.dataset_id))
+        if dataset_type != DATASET_TYPE_TEXT:
+            raise HTTPException(
+                status_code=400,
+                detail="当前仅支持 TEXT 项目的段落内容",
+            )
+
+        file_result = await self.db.execute(
+            select(DatasetFiles).where(
+                DatasetFiles.id == file_id,
+                DatasetFiles.dataset_id == project.dataset_id,
+            )
+        )
+        file_record = file_result.scalar_one_or_none()
+        if not file_record:
+            raise HTTPException(status_code=404, detail=f"文件不存在或不属于该项目: {file_id}")
+
+        if not self._resolve_segmentation_enabled(project):
+            return EditorTaskSegmentResponse(
+                segmented=False,
+                segment=None,
+                totalSegments=0,
+                currentSegmentIndex=0,
+            )
+
+        text_content = await self._fetch_text_content_via_download_api(project.dataset_id, file_id)
+        assert isinstance(text_content, str)
+        label_config = await self._resolve_project_label_config(project)
+        primary_text_key = self._resolve_primary_text_key(label_config)
+        file_name = str(getattr(file_record, "file_name", "")).lower()
+
+        records: List[Tuple[Optional[Dict[str, Any]], str]] = []
+        if file_name.endswith(JSONL_EXTENSION):
+            records = self._parse_jsonl_records(text_content)
+        else:
+            parsed_payload = self._try_parse_json_payload(text_content)
+            if parsed_payload:
+                records = [(parsed_payload, text_content)]
+
+        if not records:
+            records = [(None, text_content)]
+
+        record_texts = [
+            self._resolve_primary_text_value(payload, raw_text, primary_text_key)
+            for payload, raw_text in records
+        ]
+        if not record_texts:
+            record_texts = [text_content]
+
+        needs_segmentation = len(records) > 1 or any(
+            len(text or "") > self.SEGMENT_THRESHOLD for text in record_texts
+        )
+        if not needs_segmentation:
+            return EditorTaskSegmentResponse(
+                segmented=False,
+                segment=None,
+                totalSegments=0,
+                currentSegmentIndex=0,
+            )
+
+        ann_result = await self.db.execute(
+            select(AnnotationResult).where(
+                AnnotationResult.project_id == project.id,
+                AnnotationResult.file_id == file_id,
+            )
+        )
+        ann = ann_result.scalar_one_or_none()
+        segment_annotations: Dict[str, Dict[str, Any]] = {}
+        if ann and isinstance(ann.annotation, dict):
+            segment_annotations = self._extract_segment_annotations(ann.annotation)
+        segment_annotation_keys = set(segment_annotations.keys())
+
+        segments, segment_contexts = self._build_segment_contexts(
+            records,
+            record_texts,
+            segment_annotation_keys,
+        )
+
+        total_segments = len(segment_contexts)
+        if total_segments == 0:
+            return EditorTaskSegmentResponse(
+                segmented=False,
+                segment=None,
+                totalSegments=0,
+                currentSegmentIndex=0,
+            )
+
+        if segment_index < 0 or segment_index >= total_segments:
+            raise HTTPException(
+                status_code=400,
+                detail=f"segmentIndex 超出范围: {segment_index}",
+            )
+
+        segment_info = segments[segment_index]
+        _, _, segment_text, line_index, chunk_index = segment_contexts[segment_index]
+        segment_detail = SegmentDetail(
+            idx=segment_info.idx,
+            text=segment_text,
+            hasAnnotation=segment_info.has_annotation,
+            lineIndex=line_index,
+            chunkIndex=chunk_index,
+        )
+
+        return EditorTaskSegmentResponse(
+            segmented=True,
+            segment=segment_detail,
+            totalSegments=total_segments,
+            currentSegmentIndex=segment_index,
+        )
+
     async def _build_text_task(
         self,
         project: LabelingProject,
@@ -723,7 +887,8 @@ class AnnotationEditorService:
         needs_segmentation = segmentation_enabled and (
             len(records) > 1 or any(len(text or "") > self.SEGMENT_THRESHOLD for text in record_texts)
         )
-        segments: Optional[List[SegmentInfo]] = None
+        segments: List[SegmentInfo] = []
+        segment_contexts: List[Tuple[Optional[Dict[str, Any]], str, str, int, int]] = []
         current_segment_index = 0
         display_text = record_texts[0] if record_texts else text_content
         selected_payload = records[0][0] if records else None
@@ -732,46 +897,13 @@ class AnnotationEditorService:
             display_text = "\n".join(record_texts) if record_texts else text_content
 
         if needs_segmentation:
-            splitter = AnnotationTextSplitter(max_chars=self.SEGMENT_THRESHOLD)
+            _, segment_contexts = self._build_segment_contexts(
-            segment_contexts: List[Tuple[Optional[Dict[str, Any]], str, str, int, int]] = []
+                records,
-            segments = []
+                record_texts,
-            segment_cursor = 0
+                segment_annotation_keys,
-
+            )
-            for record_index, ((payload, raw_text), record_text) in enumerate(zip(records, record_texts)):
-                normalized_text = record_text or ""
-                if len(normalized_text) > self.SEGMENT_THRESHOLD:
-                    raw_segments = splitter.split(normalized_text)
-                    for chunk_index, seg in enumerate(raw_segments):
-                        segments.append(SegmentInfo(
-                            idx=segment_cursor,
-                            text=seg["text"],
-                            start=seg["start"],
-                            end=seg["end"],
-                            hasAnnotation=str(segment_cursor) in segment_annotation_keys,
-                            lineIndex=record_index,
-                            chunkIndex=chunk_index,
-                        ))
-                        segment_contexts.append((payload, raw_text, seg["text"], record_index, chunk_index))
-                        segment_cursor += 1
-                else:
-                    segments.append(SegmentInfo(
-                        idx=segment_cursor,
-                        text=normalized_text,
-                        start=0,
-                        end=len(normalized_text),
-                        hasAnnotation=str(segment_cursor) in segment_annotation_keys,
-                        lineIndex=record_index,
-                        chunkIndex=0,
-                    ))
-                    segment_contexts.append((payload, raw_text, normalized_text, record_index, 0))
-                    segment_cursor += 1
-
-            if not segments:
-                segments = [SegmentInfo(idx=0, text="", start=0, end=0, hasAnnotation=False, lineIndex=0, chunkIndex=0)]
-                segment_contexts = [(None, "", "", 0, 0)]
-
             current_segment_index = segment_index if segment_index is not None else 0
-            if current_segment_index < 0 or current_segment_index >= len(segments):
+            if current_segment_index < 0 or current_segment_index >= len(segment_contexts):
                 current_segment_index = 0
 
             selected_payload, _, display_text, _, _ = segment_contexts[current_segment_index]
@@ -849,8 +981,7 @@ class AnnotationEditorService:
             task=task,
             annotationUpdatedAt=annotation_updated_at,
             segmented=needs_segmentation,
-            segments=segments,
+            totalSegments=len(segment_contexts) if needs_segmentation else 1,
-            totalSegments=len(segments) if segments else 1,
             currentSegmentIndex=current_segment_index,
         )
 
@@ -1185,3 +1316,195 @@ class AnnotationEditorService:
         except Exception as exc:
             logger.warning("标注同步知识管理失败：%s", exc)
 
+    async def precompute_segmentation_for_project(
+        self,
+        project_id: str,
+        max_retries: int = 3
+    ) -> Dict[str, Any]:
+        """
+        为指定项目的所有文本文件预计算切片结构并持久化到数据库
+
+        Args:
+            project_id: 标注项目ID
+            max_retries: 失败重试次数
+
+        Returns:
+            统计信息：{total_files, succeeded, failed}
+        """
+        project = await self._get_project_or_404(project_id)
+        dataset_type = self._normalize_dataset_type(await self._get_dataset_type(project.dataset_id))
+
+        # 只处理文本数据集
+        if dataset_type != DATASET_TYPE_TEXT:
+            logger.info(f"项目 {project_id} 不是文本数据集，跳过切片预生成")
+            return {"total_files": 0, "succeeded": 0, "failed": 0}
+
+        # 检查是否启用分段
+        if not self._resolve_segmentation_enabled(project):
+            logger.info(f"项目 {project_id} 未启用分段，跳过切片预生成")
+            return {"total_files": 0, "succeeded": 0, "failed": 0}
+
+        # 获取项目的所有文本文件（排除源文档）
+        files_result = await self.db.execute(
+            select(DatasetFiles)
+            .join(LabelingProjectFile, LabelingProjectFile.file_id == DatasetFiles.id)
+            .where(
+                LabelingProjectFile.project_id == project_id,
+                DatasetFiles.dataset_id == project.dataset_id,
+            )
+        )
+        file_records = files_result.scalars().all()
+
+        if not file_records:
+            logger.info(f"项目 {project_id} 没有文件，跳过切片预生成")
+            return {"total_files": 0, "succeeded": 0, "failed": 0}
+
+        # 过滤源文档文件
+        valid_files = []
+        for file_record in file_records:
+            file_type = str(getattr(file_record, "file_type", "") or "").lower()
+            file_name = str(getattr(file_record, "file_name", "")).lower()
+            is_source_document = (
+                file_type in SOURCE_DOCUMENT_TYPES or
+                any(file_name.endswith(ext) for ext in SOURCE_DOCUMENT_EXTENSIONS)
+            )
+            if not is_source_document:
+                valid_files.append(file_record)
+
+        total_files = len(valid_files)
+        succeeded = 0
+        failed = 0
+
+        label_config = await self._resolve_project_label_config(project)
+        primary_text_key = self._resolve_primary_text_key(label_config)
+
+        for file_record in valid_files:
+            file_id = str(file_record.id)  # type: ignore
+            file_name = str(getattr(file_record, "file_name", ""))
+
+            for retry in range(max_retries):
+                try:
+                    # 读取文本内容
+                    text_content = await self._fetch_text_content_via_download_api(project.dataset_id, file_id)
+                    if not isinstance(text_content, str):
+                        logger.warning(f"文件 {file_id} 内容不是字符串，跳过切片")
+                        failed += 1
+                        break
+
+                    # 解析文本记录
+                    records: List[Tuple[Optional[Dict[str, Any]], str]] = []
+                    if file_name.lower().endswith(JSONL_EXTENSION):
+                        records = self._parse_jsonl_records(text_content)
+                    else:
+                        parsed_payload = self._try_parse_json_payload(text_content)
+                        if parsed_payload:
+                            records = [(parsed_payload, text_content)]
+
+                    if not records:
+                        records = [(None, text_content)]
+
+                    record_texts = [
+                        self._resolve_primary_text_value(payload, raw_text, primary_text_key)
+                        for payload, raw_text in records
+                    ]
+                    if not record_texts:
+                        record_texts = [text_content]
+
+                    # 判断是否需要分段
+                    needs_segmentation = len(records) > 1 or any(
+                        len(text or "") > self.SEGMENT_THRESHOLD for text in record_texts
+                    )
+
+                    if not needs_segmentation:
+                        # 不需要分段的文件，跳过
+                        succeeded += 1
+                        break
+
+                    # 执行切片
+                    splitter = AnnotationTextSplitter(max_chars=self.SEGMENT_THRESHOLD)
+                    segment_cursor = 0
+                    segments = {}
+
+                    for record_index, ((payload, raw_text), record_text) in enumerate(zip(records, record_texts)):
+                        normalized_text = record_text or ""
+
+                        if len(normalized_text) > self.SEGMENT_THRESHOLD:
+                            raw_segments = splitter.split(normalized_text)
+                            for chunk_index, seg in enumerate(raw_segments):
+                                segments[str(segment_cursor)] = {
+                                    SEGMENT_RESULT_KEY: [],
+                                    SEGMENT_CREATED_AT_KEY: datetime.utcnow().isoformat() + "Z",
+                                    SEGMENT_UPDATED_AT_KEY: datetime.utcnow().isoformat() + "Z",
+                                }
+                                segment_cursor += 1
+                        else:
+                            segments[str(segment_cursor)] = {
+                                SEGMENT_RESULT_KEY: [],
+                                SEGMENT_CREATED_AT_KEY: datetime.utcnow().isoformat() + "Z",
+                                SEGMENT_UPDATED_AT_KEY: datetime.utcnow().isoformat() + "Z",
+                            }
+                            segment_cursor += 1
+
+                    if not segments:
+                        succeeded += 1
+                        break
+
+                    # 构造分段标注结构
+                    final_payload = {
+                        SEGMENTED_KEY: True,
+                        "version": 1,
+                        SEGMENTS_KEY: segments,
+                        SEGMENT_TOTAL_KEY: segment_cursor,
+                    }
+
+                    # 检查是否已存在标注
+                    existing_result = await self.db.execute(
+                        select(AnnotationResult).where(
+                            AnnotationResult.project_id == project_id,
+                            AnnotationResult.file_id == file_id,
+                        )
+                    )
+                    existing = existing_result.scalar_one_or_none()
+
+                    now = datetime.utcnow()
+
+                    if existing:
+                        # 更新现有标注
+                        existing.annotation = final_payload  # type: ignore[assignment]
+                        existing.annotation_status = ANNOTATION_STATUS_IN_PROGRESS  # type: ignore[assignment]
+                        existing.updated_at = now  # type: ignore[assignment]
+                    else:
+                        # 创建新标注记录
+                        record = AnnotationResult(
+                            id=str(uuid.uuid4()),
+                            project_id=project_id,
+                            file_id=file_id,
+                            annotation=final_payload,
+                            annotation_status=ANNOTATION_STATUS_IN_PROGRESS,
+                            created_at=now,
+                            updated_at=now,
+                        )
+                        self.db.add(record)
+
+                    await self.db.commit()
+                    succeeded += 1
+                    logger.info(f"成功为文件 {file_id} 预生成 {segment_cursor} 个切片")
+                    break
+
+                except Exception as e:
+                    logger.warning(
+                        f"为文件 {file_id} 预生成切片失败 (重试 {retry + 1}/{max_retries}): {e}"
+                    )
+                    if retry == max_retries - 1:
+                        failed += 1
+                        await self.db.rollback()
+
+        logger.info(
+            f"项目 {project_id} 切片预生成完成: 总计 {total_files}, 成功 {succeeded}, 失败 {failed}"
+        )
+        return {
+            "total_files": total_files,
+            "succeeded": succeeded,
+            "failed": failed,
+        }
+

scripts/offline/Dockerfile.backend-python.offline-v2

@@ -0,0 +1,82 @@
+# backend-python Dockerfile 离线版本 v2
+FROM maven:3-eclipse-temurin-8 AS datax-builder
+
+# 配置 Maven 阿里云镜像
+RUN mkdir -p /root/.m2 && \
+    echo '<?xml version="1.0" encoding="UTF-8"?>\n\
+<settings xmlns="http://maven.apache.org/SETTINGS/1.0.0"\n\
+          xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"\n\
+          xsi:schemaLocation="http://maven.apache.org/SETTINGS/1.0.0 http://maven.apache.org/xsd/settings-1.0.0.xsd">\n\
+  <mirrors>\n\
+    <mirror>\n\
+      <id>aliyunmaven</id>\n\
+      <mirrorOf>*</mirrorOf>\n\
+      <name>阿里云公共仓库</name>\n\
+      <url>https://maven.aliyun.com/repository/public</url>\n\
+    </mirror>\n\
+  </mirrors>\n\
+</settings>' > /root/.m2/settings.xml
+
+# 离线模式: 从构建参数获取本地 DataX 路径
+ARG RESOURCES_DIR=./build-cache/resources
+ARG DATAX_LOCAL_PATH=${RESOURCES_DIR}/DataX
+
+# 复制本地 DataX 源码
+COPY ${DATAX_LOCAL_PATH} /DataX
+
+COPY runtime/datax/ DataX/
+
+RUN cd DataX && \
+    sed -i "s/com.mysql.jdbc.Driver/com.mysql.cj.jdbc.Driver/g" \
+    plugin-rdbms-util/src/main/java/com/alibaba/datax/plugin/rdbms/util/DataBaseType.java && \
+    mvn -U clean package assembly:assembly -Dmaven.test.skip=true
+
+# 使用预装 APT 包的基础镜像
+FROM datamate-python-base:latest
+
+ENV PYTHONDONTWRITEBYTECODE=1 \
+    PYTHONUNBUFFERED=1 \
+    POETRY_VERSION=2.2.1 \
+    POETRY_NO_INTERACTION=1 \
+    POETRY_VIRTUALENVS_CREATE=false \
+    POETRY_CACHE_DIR=/tmp/poetry_cache
+
+ENV JAVA_HOME=/usr/lib/jvm/java-21-openjdk
+ENV PATH="/root/.local/bin:$JAVA_HOME/bin:$PATH"
+
+WORKDIR /app
+
+# 配置 pip 阿里云镜像并安装 Poetry
+RUN --mount=type=cache,target=/root/.cache/pip \
+    pip config set global.index-url https://mirrors.aliyun.com/pypi/simple/ && \
+    pip config set global.trusted-host mirrors.aliyun.com && \
+    pip install --upgrade --root-user-action=ignore pip \
+    && pip install --root-user-action=ignore pipx \
+    && pipx install "poetry==$POETRY_VERSION"
+
+COPY --from=datax-builder /DataX/target/datax/datax /opt/datax
+RUN cp /opt/datax/plugin/reader/mysqlreader/libs/mysql* /opt/datax/plugin/reader/starrocksreader/libs/
+
+# Copy only dependency files first
+COPY runtime/datamate-python/pyproject.toml runtime/datamate-python/poetry.lock* /app/
+
+# Install dependencies
+RUN --mount=type=cache,target=$POETRY_CACHE_DIR \
+    poetry install --no-root --only main
+
+# 离线模式: 使用本地 NLTK 数据
+ARG RESOURCES_DIR=./build-cache/resources
+ARG NLTK_DATA_LOCAL_PATH=${RESOURCES_DIR}/nltk_data
+COPY ${NLTK_DATA_LOCAL_PATH} /usr/local/nltk_data
+
+ENV NLTK_DATA=/usr/local/nltk_data
+
+# Copy the rest of the application
+COPY runtime/datamate-python /app
+
+COPY runtime/datamate-python/deploy/docker-entrypoint.sh /docker-entrypoint.sh
+RUN chmod +x /docker-entrypoint.sh || true
+
+EXPOSE 18000
+
+ENTRYPOINT ["/docker-entrypoint.sh"]

scripts/offline/Dockerfile.backend.offline

@@ -0,0 +1,71 @@
+# backend Dockerfile 离线版本
+# 使用预装 APT 包的基础镜像
+
+FROM maven:3-eclipse-temurin-21 AS builder
+
+# 配置 Maven 阿里云镜像
+RUN mkdir -p /root/.m2 && \
+    echo '<?xml version="1.0" encoding="UTF-8"?>\n\
+<settings xmlns="http://maven.apache.org/SETTINGS/1.0.0"\n\
+          xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"\n\
+          xsi:schemaLocation="http://maven.apache.org/SETTINGS/1.0.0 http://maven.apache.org/xsd/settings-1.0.0.xsd">\n\
+  <mirrors>\n\
+    <mirror>\n\
+      <id>aliyunmaven</id>\n\
+      <mirrorOf>*</mirrorOf>\n\
+      <name>阿里云公共仓库</name>\n\
+      <url>https://maven.aliyun.com/repository/public</url>\n\
+    </mirror>\n\
+  </mirrors>\n\
+</settings>' > /root/.m2/settings.xml
+
+WORKDIR /opt/backend
+
+# 先复制所有 pom.xml 文件
+COPY backend/pom.xml ./
+COPY backend/services/pom.xml ./services/
+COPY backend/shared/domain-common/pom.xml ./shared/domain-common/
+COPY backend/shared/security-common/pom.xml ./shared/security-common/
+COPY backend/services/data-annotation-service/pom.xml ./services/data-annotation-service/
+COPY backend/services/data-cleaning-service/pom.xml ./services/data-cleaning-service/
+COPY backend/services/data-evaluation-service/pom.xml ./services/data-evaluation-service/
+COPY backend/services/data-management-service/pom.xml ./services/data-management-service/
+COPY backend/services/data-synthesis-service/pom.xml ./services/data-synthesis-service/
+COPY backend/services/execution-engine-service/pom.xml ./services/execution-engine-service/
+COPY backend/services/main-application/pom.xml ./services/main-application/
+COPY backend/services/operator-market-service/pom.xml ./services/operator-market-service/
+COPY backend/services/pipeline-orchestration-service/pom.xml ./services/pipeline-orchestration-service/
+COPY backend/services/rag-indexer-service/pom.xml ./services/rag-indexer-service/
+COPY backend/services/rag-query-service/pom.xml ./services/rag-query-service/
+
+# 使用缓存卷下载依赖
+RUN --mount=type=cache,target=/root/.m2/repository \
+    cd /opt/backend/services && \
+    mvn dependency:go-offline -Dmaven.test.skip=true || true
+
+# 复制所有源代码
+COPY backend/ /opt/backend
+
+# 编译打包
+RUN --mount=type=cache,target=/root/.m2/repository \
+    cd /opt/backend/services && \
+    mvn clean package -Dmaven.test.skip=true
+
+# 使用预装 APT 包的基础镜像
+FROM datamate-java-base:latest
+
+# 不再执行 apt-get update，因为基础镜像已经预装了所有需要的包
+# 如果需要添加额外的包，可以在这里添加，但离线环境下会失败
+
+COPY --from=builder /opt/backend/services/main-application/target/datamate.jar /opt/backend/datamate.jar
+COPY scripts/images/backend/start.sh /opt/backend/start.sh
+COPY runtime/ops/examples/test_operator/test_operator.tar /opt/backend/test_operator.tar
+
+RUN dos2unix /opt/backend/start.sh \
+    && chmod +x /opt/backend/start.sh \
+    && ln -sf /usr/share/zoneinfo/Asia/Shanghai /etc/localtime
+
+EXPOSE 8080
+
+ENTRYPOINT ["/opt/backend/start.sh"]
+CMD ["java", "-Duser.timezone=Asia/Shanghai", "-jar", "/opt/backend/datamate.jar"]

scripts/offline/Dockerfile.base-images

@@ -0,0 +1,62 @@
+# 预安装 APT 包的基础镜像
+# 在有网环境构建这些镜像，在无网环境作为基础镜像使用
+
+# ==================== backend / gateway 基础镜像 ====================
+FROM eclipse-temurin:21-jdk AS datamate-java-base
+
+# 配置 apt 阿里云镜像源
+RUN if [ -f /etc/apt/sources.list.d/debian.sources ]; then \
+        sed -i 's/deb.debian.org/mirrors.aliyun.com/g' /etc/apt/sources.list.d/debian.sources; \
+    elif [ -f /etc/apt/sources.list.d/ubuntu.sources ]; then \
+        sed -i 's/archive.ubuntu.com/mirrors.aliyun.com/g; s/security.ubuntu.com/mirrors.aliyun.com/g' /etc/apt/sources.list.d/ubuntu.sources; \
+    elif [ -f /etc/apt/sources.list ]; then \
+        sed -i 's/deb.debian.org/mirrors.aliyun.com/g; s/archive.ubuntu.com/mirrors.aliyun.com/g; s/security.ubuntu.com/mirrors.aliyun.com/g' /etc/apt/sources.list; \
+    fi && \
+    apt-get update && \
+    apt-get install -y vim wget curl rsync python3 python3-pip python-is-python3 dos2unix libreoffice fonts-noto-cjk && \
+    apt-get clean && \
+    rm -rf /var/lib/apt/lists/*
+
+# ==================== backend-python 基础镜像 ====================
+FROM python:3.12-slim AS datamate-python-base
+
+RUN if [ -f /etc/apt/sources.list.d/debian.sources ]; then \
+        sed -i 's/deb.debian.org/mirrors.aliyun.com/g' /etc/apt/sources.list.d/debian.sources; \
+    elif [ -f /etc/apt/sources.list ]; then \
+        sed -i 's/deb.debian.org/mirrors.aliyun.com/g' /etc/apt/sources.list; \
+    fi && \
+    apt-get update && \
+    apt-get install -y --no-install-recommends vim openjdk-21-jre nfs-common glusterfs-client rsync && \
+    rm -rf /var/lib/apt/lists/*
+
+# ==================== runtime 基础镜像 ====================
+FROM ghcr.nju.edu.cn/astral-sh/uv:python3.11-bookworm AS datamate-runtime-base
+
+RUN if [ -f /etc/apt/sources.list.d/debian.sources ]; then \
+        sed -i 's/deb.debian.org/mirrors.aliyun.com/g' /etc/apt/sources.list.d/debian.sources; \
+    elif [ -f /etc/apt/sources.list ]; then \
+        sed -i 's/deb.debian.org/mirrors.aliyun.com/g' /etc/apt/sources.list; \
+    fi && \
+    apt update && \
+    apt install -y libgl1 libglib2.0-0 vim libmagic1 libreoffice dos2unix swig poppler-utils tesseract-ocr && \
+    rm -rf /var/lib/apt/lists/*
+
+# ==================== deer-flow-backend 基础镜像 ====================
+FROM ghcr.nju.edu.cn/astral-sh/uv:python3.12-bookworm AS deer-flow-backend-base
+
+RUN if [ -f /etc/apt/sources.list.d/debian.sources ]; then \
+        sed -i 's/deb.debian.org/mirrors.aliyun.com/g' /etc/apt/sources.list.d/debian.sources; \
+    elif [ -f /etc/apt/sources.list ]; then \
+        sed -i 's/deb.debian.org/mirrors.aliyun.com/g' /etc/apt/sources.list; \
+    fi && \
+    apt-get update && apt-get install -y libpq-dev git && \
+    rm -rf /var/lib/apt/lists/*
+
+# ==================== mineru 基础镜像 ====================
+FROM python:3.11-slim AS mineru-base
+
+RUN sed -i 's/deb.debian.org/mirrors.huaweicloud.com/g' /etc/apt/sources.list.d/debian.sources && \
+    apt-get update && \
+    apt-get install -y curl vim libgl1 libglx0 libopengl0 libglib2.0-0 procps && \
+    apt-get clean && \
+    rm -rf /var/lib/apt/lists/*

scripts/offline/Dockerfile.gateway.offline

@@ -0,0 +1,47 @@
+# gateway Dockerfile 离线版本
+FROM maven:3-eclipse-temurin-21 AS builder
+
+# 配置 Maven 阿里云镜像
+RUN mkdir -p /root/.m2 && \
+    echo '<?xml version="1.0" encoding="UTF-8"?>\n\
+<settings xmlns="http://maven.apache.org/SETTINGS/1.0.0"\n\
+          xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"\n\
+          xsi:schemaLocation="http://maven.apache.org/SETTINGS/1.0.0 http://maven.apache.org/xsd/settings-1.0.0.xsd">\n\
+  <mirrors>\n\
+    <mirror>\n\
+      <id>aliyunmaven</id>\n\
+      <mirrorOf>*</mirrorOf>\n\
+      <name>阿里云公共仓库</name>\n\
+      <url>https://maven.aliyun.com/repository/public</url>\n\
+    </mirror>\n\
+  </mirrors>\n\
+</settings>' > /root/.m2/settings.xml
+
+WORKDIR /opt/gateway
+
+COPY backend/pom.xml ./
+COPY backend/api-gateway/pom.xml ./api-gateway/
+
+RUN --mount=type=cache,target=/root/.m2/repository \
+    cd /opt/gateway/api-gateway && \
+    mvn dependency:go-offline -Dmaven.test.skip=true || true
+
+COPY backend/api-gateway /opt/gateway/api-gateway
+
+RUN --mount=type=cache,target=/root/.m2/repository \
+    cd /opt/gateway/api-gateway && \
+    mvn clean package -Dmaven.test.skip=true
+
+FROM datamate-java-base:latest
+
+COPY --from=builder /opt/gateway/api-gateway/target/gateway.jar /opt/gateway/gateway.jar
+COPY scripts/images/gateway/start.sh /opt/gateway/start.sh
+
+RUN dos2unix /opt/gateway/start.sh \
+    && chmod +x /opt/gateway/start.sh \
+    && ln -sf /usr/share/zoneinfo/Asia/Shanghai /etc/localtime
+
+EXPOSE 8080
+
+ENTRYPOINT ["/opt/gateway/start.sh"]
+CMD ["java", "-Duser.timezone=Asia/Shanghai", "-jar", "/opt/gateway/gateway.jar"]

scripts/offline/Dockerfile.runtime.offline-v2

@@ -0,0 +1,42 @@
+# runtime Dockerfile 离线版本 v2
+# 使用预装 APT 包的基础镜像
+
+FROM datamate-runtime-base:latest
+
+# 离线模式: 本地模型文件路径
+ARG RESOURCES_DIR=./build-cache/resources
+ARG MODELS_DIR=${RESOURCES_DIR}/models
+
+# 复制本地 PaddleOCR 模型
+RUN mkdir -p /home/models
+COPY ${MODELS_DIR}/ch_ppocr_mobile_v2.0_cls_infer.tar /home/models/
+RUN tar -xf /home/models/ch_ppocr_mobile_v2.0_cls_infer.tar -C /home/models
+
+COPY runtime/python-executor /opt/runtime
+COPY runtime/ops /opt/runtime/datamate/ops
+COPY runtime/ops/user /opt/runtime/user
+COPY scripts/images/runtime/start.sh /opt/runtime/start.sh
+
+ENV PYTHONPATH=/opt/runtime/datamate/
+ENV UV_EXTRA_INDEX_URL="https://download.pytorch.org/whl/cpu"
+ENV UV_INDEX_STRATEGY=unsafe-best-match
+ENV UV_INDEX_URL="https://mirrors.aliyun.com/pypi/simple/"
+
+WORKDIR /opt/runtime
+
+# 复制本地 spaCy 模型
+COPY ${MODELS_DIR}/zh_core_web_sm-3.8.0-py3-none-any.whl /tmp/
+
+RUN --mount=type=cache,target=/root/.cache/uv \
+    uv pip install -e .[all] --system \
+    && uv pip install -r /opt/runtime/datamate/ops/pyproject.toml --system \
+    && uv pip install /tmp/zh_core_web_sm-3.8.0-py3-none-any.whl --system \
+    && echo "/usr/local/lib/ops/site-packages" > /usr/local/lib/python3.11/site-packages/ops.pth
+
+RUN ln -sf /usr/share/zoneinfo/Asia/Shanghai /etc/localtime \
+    && chmod +x /opt/runtime/start.sh \
+    && dos2unix /opt/runtime/start.sh
+
+EXPOSE 8081
+
+ENTRYPOINT ["/opt/runtime/start.sh"]

scripts/offline/README.md

@@ -74,20 +74,29 @@ scp build-cache-20250202.tar.gz user@offline-server:/opt/datamate/
 # 或者使用 U 盘等物理介质
 ```
 
-#### 4. 无网环境构建
+#### 4. 无网环境构建（推荐使用传统方式）
 
 ```bash
 # 解压缓存
 tar -xzf build-cache-20250202.tar.gz
 
-# 设置环境并构建
+# 诊断环境（检查基础镜像等）
+make offline-diagnose
+
+# 方法 A：传统 docker build（推荐，更稳定）
+make offline-setup
+make offline-build-classic
+
+# 方法 B：BuildKit 构建（如果方法 A 失败）
 make offline-setup
 make offline-build
 
 # 或者指定版本号
-make offline-build OFFLINE_VERSION=v1.0.0
+make offline-build-classic OFFLINE_VERSION=v1.0.0
 ```
 
+**⚠️ 重要提示**：如果遇到镜像拉取问题，请使用 `make offline-build-classic` 而不是 `make offline-build`。
+
 ### 方法二：使用独立脚本
 
 #### 导出缓存
@@ -162,9 +171,133 @@ tar -czf build-cache-partial.tar.gz build-cache/buildkit/backend-cache
 make backend-offline-build
 ```
 
+## APT 缓存问题详解
+
+### 问题描述
+
+即使使用了 `--mount=type=cache,target=/var/cache/apt`，Dockerfile 中的 `apt-get update` 仍会尝试从网络获取包列表（list 数据），导致无网环境下构建失败：
+
+```
+Err:1 http://mirrors.aliyun.com/debian bookworm InRelease
+  Could not resolve 'mirrors.aliyun.com'
+Reading package lists...
+E: Failed to fetch http://mirrors.aliyun.com/debian/dists/bookworm/InRelease
+```
+
+### 根本原因
+
+- `--mount=type=cache,target=/var/cache/apt` 只缓存下载的 `.deb` 包
+- `apt-get update` 会尝试从配置的源获取最新的包索引（InRelease/Packages 文件）
+- `/var/lib/apt/lists/` 目录存储包索引，但通常不在缓存范围内
+
+### 解决方案
+
+#### 方案 1: 使用预装 APT 包的基础镜像（推荐）
+
+这是最有效的方法：
+
+**步骤 1**: 在有网环境构建预装所有依赖的基础镜像
+
+```bash
+# 构建并保存带 APT 预装包的基础镜像
+./scripts/offline/build-base-images.sh
+```
+
+这会创建以下预装基础镜像：
+- `datamate-java-base` - 用于 backend、gateway（预装 vim、python3、libreoffice 等）
+- `datamate-python-base` - 用于 backend-python（预装 openjdk、nfs-common 等）
+- `datamate-runtime-base` - 用于 runtime（预装 libgl1、tesseract-ocr 等）
+- `deer-flow-backend-base` - 用于 deer-flow-backend
+- `mineru-base` - 用于 mineru
+
+**步骤 2**: 在无网环境使用这些基础镜像构建
+
+```bash
+# 加载包含预装基础镜像的 tar 包
+docker load -i build-cache/images/base-images-with-apt.tar
+
+# 使用最终版构建脚本
+./scripts/offline/build-offline-final.sh
+```
+
+#### 方案 2: 修改 Dockerfile 跳过 apt update
+
+如果确定不需要安装新包，可以修改 Dockerfile：
+
+```dockerfile
+# 原代码
+RUN apt-get update && apt-get install -y xxx
+
+# 修改为（离线环境）
+# RUN apt-get update && \
+RUN apt-get install -y xxx || true
+```
+
+#### 方案 3: 挂载 apt lists 缓存
+
+在有网环境预先下载并保存 apt lists：
+
+```bash
+# 有网环境：保存 apt lists
+docker run --rm \
+    -v "$(pwd)/apt-lists:/var/lib/apt/lists" \
+    eclipse-temurin:21-jdk \
+    apt-get update
+
+# 无网环境：挂载保存的 lists
+docker build \
+    --mount=type=bind,source=$(pwd)/apt-lists,target=/var/lib/apt/lists,ro \
+    -f Dockerfile .
+```
+
+**注意**: BuildKit 的 `--mount=type=bind` 在 `docker build` 中不直接支持，需要在 Dockerfile 中使用。
+
+---
+
 ## 故障排查
 
-### 问题 1: 缓存导入失败
+### 问题 1: 构建时仍然尝试拉取镜像（最常见）
+
+**现象**:
+```
+ERROR: failed to solve: pulling from host ...
+或
+ERROR: pull access denied, repository does not exist or may require authorization
+```
+
+**原因**: 
+- 基础镜像未正确加载
+- BuildKit 尝试验证远程镜像
+
+**解决方案**:
+
+1. **使用传统构建方式（推荐）**:
+```bash
+make offline-build-classic
+```
+
+2. **手动加载基础镜像**:
+```bash
+# 加载基础镜像
+docker load -i build-cache/images/base-images.tar
+
+# 验证镜像存在
+docker images | grep -E "(maven|eclipse-temurin|mysql|node|nginx)"
+```
+
+3. **使用 Docker 守护进程的离线模式**:
+```bash
+# 编辑 /etc/docker/daemon.json
+{
+  "registry-mirrors": [],
+  "insecure-registries": []
+}
+
+# 重启 Docker
+sudo systemctl restart docker
+```
+
+### 问题 2: 缓存导入失败
 
 ```
 ERROR: failed to solve: failed to read cache metadata
@@ -172,23 +305,26 @@ ERROR: failed to solve: failed to read cache metadata
 
 **解决**: 缓存目录可能损坏，重新在有网环境导出。
 
-### 问题 2: 基础镜像不存在
+### 问题 3: 基础镜像不存在
 
 ```
 ERROR: pull access denied
 ```
 
-**解决**: 先执行 `make offline-setup` 加载基础镜像。
+**解决**: 
+1. 先执行 `make offline-setup` 加载基础镜像
+2. 运行 `make offline-diagnose` 检查缺失的镜像
+3. 重新导出缓存时确保包含所有基础镜像
 
-### 问题 3: 网络连接错误（无网环境）
+### 问题 4: 网络连接错误（无网环境）
 
 ```
 ERROR: failed to do request: dial tcp: lookup ...
 ```
 
-**解决**: 检查 Dockerfile 中是否还有网络依赖，可能需要修改 Dockerfile 使用本地资源。
+**解决**: 检查 Dockerfile 中是否还有网络依赖（如 `git clone`、`wget`、`pip install` 等），可能需要修改 Dockerfile 使用本地资源。
 
-### 问题 4: 内存不足
+### 问题 5: 内存不足
 
 BuildKit 缓存可能占用大量内存，可以设置资源限制：
 
@@ -200,6 +336,39 @@ docker buildx create --name offline-builder \
   --use
 ```
 
+### 问题 6: BuildKit 构建器无法使用本地镜像
+
+**现象**: 即使镜像已加载，BuildKit 仍提示找不到镜像
+
+**解决**: BuildKit 的 `docker-container` 驱动无法直接访问本地镜像。使用以下方法之一：
+
+**方法 A**: 使用传统 Docker 构建（推荐）
+```bash
+make offline-build-classic
+```
+
+**方法 B**: 将镜像推送到本地 registry
+```bash
+# 启动本地 registry
+docker run -d -p 5000:5000 --name registry registry:2
+
+# 标记并推送镜像到本地 registry
+docker tag maven:3-eclipse-temurin-21 localhost:5000/maven:3-eclipse-temurin-21
+docker push localhost:5000/maven:3-eclipse-temurin-21
+
+# 修改 Dockerfile 使用本地 registry
+# FROM localhost:5000/maven:3-eclipse-temurin-21
+```
+
+**方法 C**: 使用 `docker` 驱动的 buildx 构建器（不需要推送镜像，但有其他限制）
+```bash
+# 创建使用 docker 驱动的构建器
+docker buildx create --name offline-builder --driver docker --use
+
+# 但这种方式无法使用 --cache-from type=local
+# 仅适用于简单的离线构建场景
+```
+
 ## 限制说明
 
 1. **镜像版本**: 基础镜像版本必须与缓存导出时一致
@@ -239,6 +408,81 @@ docker buildx build \
   -f scripts/images/backend/Dockerfile .
 ```
 
+## 文件清单
+
+```
+scripts/offline/
+├── export-cache.sh                     # 有网环境导出缓存脚本
+├── build-base-images.sh                # 构建 APT 预装基础镜像
+├── build-offline.sh                    # 基础离线构建脚本（BuildKit）
+├── build-offline-v2.sh                 # 增强版离线构建脚本
+├── build-offline-classic.sh            # 传统 docker build 脚本
+├── build-offline-final.sh              # 最终版（使用预装基础镜像，推荐）
+├── diagnose.sh                         # 环境诊断脚本
+├── Dockerfile.base-images              # 预装 APT 包的基础镜像定义
+├── Dockerfile.backend.offline          # backend 离线 Dockerfile（使用预装基础镜像）
+├── Dockerfile.gateway.offline          # gateway 离线 Dockerfile（使用预装基础镜像）
+├── Dockerfile.backend-python.offline   # backend-python 离线 Dockerfile
+├── Dockerfile.backend-python.offline-v2 # backend-python 离线 Dockerfile v2（使用预装基础镜像）
+├── Dockerfile.runtime.offline          # runtime 离线 Dockerfile
+├── Dockerfile.runtime.offline-v2       # runtime 离线 Dockerfile v2（使用预装基础镜像）
+├── Dockerfile.deer-flow-backend.offline    # deer-flow-backend 离线 Dockerfile
+├── Dockerfile.deer-flow-frontend.offline   # deer-flow-frontend 离线 Dockerfile
+├── Makefile.offline                    # 独立离线构建 Makefile
+└── README.md                           # 本文档
+
+Makefile.offline.mk                     # Makefile 扩展（追加到主 Makefile）
+```
+
+## 推荐工作流（解决 APT 问题版）
+
+### 工作流 A: 使用预装 APT 包的基础镜像（彻底解决 APT 问题）
+
+```bash
+# ========== 有网环境 ==========
+
+# 1. 构建并保存带 APT 预装包的基础镜像
+./scripts/offline/build-base-images.sh
+# 输出: build-cache/images/base-images-with-apt.tar
+
+# 2. 导出其他缓存（BuildKit 缓存、外部资源）
+./scripts/offline/export-cache.sh
+
+# 3. 打包传输
+scp build-cache/images/base-images-with-apt.tar user@offline-server:/opt/datamate/build-cache/images/
+scp build-cache-*.tar.gz user@offline-server:/opt/datamate/
+
+# ========== 无网环境 ==========
+
+cd /opt/datamate
+
+# 4. 解压
+tar -xzf build-cache-*.tar.gz
+
+# 5. 加载预装基础镜像（关键！）
+docker load -i build-cache/images/base-images-with-apt.tar
+
+# 6. 使用最终版脚本构建
+./scripts/offline/build-offline-final.sh
+```
+
+### 工作流 B: 简单场景（使用传统构建）
+
+如果 APT 包需求简单，可以直接使用传统构建：
+
+```bash
+# 有网环境
+make offline-export
+
+# 传输到无网环境
+scp build-cache-*.tar.gz offline-server:/path/
+
+# 无网环境
+tar -xzf build-cache-*.tar.gz
+make offline-diagnose        # 检查环境
+make offline-build-classic   # 使用传统构建
+```
+
 ## 参考
 
 - [Docker BuildKit Documentation](https://docs.docker.com/build/buildkit/)

scripts/offline/build-base-images.sh

@@ -0,0 +1,87 @@
+#!/bin/bash
+# 构建带有预装 APT 包的基础镜像
+# Usage: ./build-base-images.sh [output-dir]
+
+set -e
+
+OUTPUT_DIR="${1:-./build-cache}"
+IMAGES_DIR="$OUTPUT_DIR/images"
+
+mkdir -p "$IMAGES_DIR"
+
+echo "======================================"
+echo "构建预装 APT 包的基础镜像"
+echo "======================================"
+
+# 构建各个基础镜像
+echo ""
+echo "1. 构建 datamate-java-base (用于 backend, gateway)..."
+docker build \
+    -t datamate-java-base:latest \
+    --target datamate-java-base \
+    -f scripts/offline/Dockerfile.base-images \
+    . || echo "Warning: datamate-java-base 构建失败"
+
+echo ""
+echo "2. 构建 datamate-python-base (用于 backend-python)..."
+docker build \
+    -t datamate-python-base:latest \
+    --target datamate-python-base \
+    -f scripts/offline/Dockerfile.base-images \
+    . || echo "Warning: datamate-python-base 构建失败"
+
+echo ""
+echo "3. 构建 datamate-runtime-base (用于 runtime)..."
+docker build \
+    -t datamate-runtime-base:latest \
+    --target datamate-runtime-base \
+    -f scripts/offline/Dockerfile.base-images \
+    . || echo "Warning: datamate-runtime-base 构建失败"
+
+echo ""
+echo "4. 构建 deer-flow-backend-base (用于 deer-flow-backend)..."
+docker build \
+    -t deer-flow-backend-base:latest \
+    --target deer-flow-backend-base \
+    -f scripts/offline/Dockerfile.base-images \
+    . || echo "Warning: deer-flow-backend-base 构建失败"
+
+echo ""
+echo "5. 构建 mineru-base (用于 mineru)..."
+docker build \
+    -t mineru-base:latest \
+    --target mineru-base \
+    -f scripts/offline/Dockerfile.base-images \
+    . || echo "Warning: mineru-base 构建失败"
+
+echo ""
+echo "======================================"
+echo "保存基础镜像集合"
+echo "======================================"
+
+docker save -o "$IMAGES_DIR/base-images-with-apt.tar" \
+    maven:3-eclipse-temurin-21 \
+    maven:3-eclipse-temurin-8 \
+    eclipse-temurin:21-jdk \
+    mysql:8 \
+    node:20-alpine \
+    nginx:1.29 \
+    ghcr.nju.edu.cn/astral-sh/uv:python3.11-bookworm \
+    ghcr.nju.edu.cn/astral-sh/uv:python3.12-bookworm \
+    ghcr.nju.edu.cn/astral-sh/uv:latest \
+    python:3.12-slim \
+    python:3.11-slim \
+    gcr.io/distroless/nodejs20-debian12 \
+    datamate-java-base:latest \
+    datamate-python-base:latest \
+    datamate-runtime-base:latest \
+    deer-flow-backend-base:latest \
+    mineru-base:latest \
+    2>/dev/null || echo "Warning: 部分镜像保存失败"
+
+echo ""
+echo "======================================"
+echo "✓ 基础镜像构建完成"
+echo "======================================"
+echo "镜像列表:"
+docker images | grep -E "(datamate-|deer-flow-|mineru-)base" || true

scripts/offline/build-offline-classic.sh

@@ -0,0 +1,206 @@
+#!/bin/bash
+# 传统 docker build 离线构建脚本（不使用 buildx）
+# 这种方式更稳定，兼容性更好
+# Usage: ./build-offline-classic.sh [cache-dir] [version]
+
+set -e
+
+CACHE_DIR="${1:-./build-cache}"
+VERSION="${2:-latest}"
+IMAGES_DIR="$CACHE_DIR/images"
+RESOURCES_DIR="$CACHE_DIR/resources"
+
+# 颜色输出
+RED='\033[0;31m'
+GREEN='\033[0;32m'
+YELLOW='\033[1;33m'
+BLUE='\033[0;34m'
+NC='\033[0m'
+
+log_info() { echo -e "${GREEN}[INFO]${NC} $1"; }
+log_warn() { echo -e "${YELLOW}[WARN]${NC} $1"; }
+log_error() { echo -e "${RED}[ERROR]${NC} $1"; }
+log_debug() { echo -e "${BLUE}[DEBUG]${NC} $1"; }
+
+# 检查缓存目录
+if [ ! -d "$CACHE_DIR" ]; then
+    log_error "缓存目录 $CACHE_DIR 不存在"
+    exit 1
+fi
+
+# 加载基础镜像
+load_base_images() {
+    log_info "加载基础镜像..."
+    
+    if [ ! -f "$IMAGES_DIR/base-images.tar" ]; then
+        log_warn "基础镜像 tar 包不存在，检查本地镜像..."
+        return
+    fi
+    
+    log_info "从 $IMAGES_DIR/base-images.tar 加载..."
+    docker load -i "$IMAGES_DIR/base-images.tar"
+    log_info "✓ 基础镜像加载完成"
+}
+
+# 检查镜像是否存在
+check_image() {
+    docker inspect "$1" > /dev/null 2>&1
+}
+
+# 构建函数
+build_service() {
+    local service_name=$1
+    local image_name=$2
+    local dockerfile=$3
+    
+    log_info "----------------------------------------"
+    log_info "构建 $service_name"
+    log_info "----------------------------------------"
+    
+    # 检查 Dockerfile 是否存在
+    if [ ! -f "$dockerfile" ]; then
+        log_error "Dockerfile 不存在: $dockerfile"
+        return 1
+    fi
+    
+    # 获取所需的基础镜像
+    local from_images
+    from_images=$(grep -E '^FROM' "$dockerfile" | sed 's/FROM //' | sed 's/ AS .*//' | sed 's/ as .*//' | awk '{print $1}' | sort -u)
+    
+    log_info "检查基础镜像..."
+    local all_exist=true
+    for img in $from_images; do
+        # 跳过多阶段构建的中间阶段引用
+        if [[ "$img" == --from=* ]]; then
+            continue
+        fi
+        
+        if check_image "$img"; then
+            log_info "  ✓ $img"
+        else
+            log_error "  ✗ $img (缺失)"
+            all_exist=false
+        fi
+    done
+    
+    if [ "$all_exist" = false ]; then
+        log_error "缺少必要的基础镜像，无法构建 $service_name"
+        return 1
+    fi
+    
+    # 准备构建参数
+    local build_args=()
+    
+    # 根据服务类型添加特殊处理
+    case "$service_name" in
+        runtime)
+            # runtime 需要模型文件
+            if [ -d "$RESOURCES_DIR/models" ]; then
+                log_info "使用本地模型文件"
+                build_args+=("--build-arg" "RESOURCES_DIR=$RESOURCES_DIR")
+            fi
+            ;;
+        backend-python)
+            if [ -d "$RESOURCES_DIR/DataX" ]; then
+                log_info "使用本地 DataX 源码"
+                build_args+=("--build-arg" "RESOURCES_DIR=$RESOURCES_DIR")
+                build_args+=("--build-arg" "DATAX_LOCAL_PATH=$RESOURCES_DIR/DataX")
+            fi
+            ;;
+        deer-flow-backend|deer-flow-frontend)
+            if [ -d "$RESOURCES_DIR/deer-flow" ]; then
+                log_info "使用本地 deer-flow 源码"
+                build_args+=("--build-arg" "RESOURCES_DIR=$RESOURCES_DIR")
+            fi
+            ;;
+    esac
+    
+    # 执行构建
+    log_info "开始构建..."
+    if docker build \
+        --pull=false \
+        "${build_args[@]}" \
+        -f "$dockerfile" \
+        -t "$image_name:$VERSION" \
+        . 2>&1; then
+        log_info "✓ $service_name 构建成功"
+        return 0
+    else
+        log_error "✗ $service_name 构建失败"
+        return 1
+    fi
+}
+
+# 主流程
+main() {
+    log_info "======================================"
+    log_info "传统 Docker 离线构建"
+    log_info "======================================"
+    
+    # 加载基础镜像
+    load_base_images
+    
+    # 定义要构建的服务
+    declare -A SERVICES=(
+        ["database"]="datamate-database:scripts/images/database/Dockerfile"
+        ["gateway"]="datamate-gateway:scripts/images/gateway/Dockerfile"
+        ["backend"]="datamate-backend:scripts/images/backend/Dockerfile"
+        ["frontend"]="datamate-frontend:scripts/images/frontend/Dockerfile"
+        ["runtime"]="datamate-runtime:scripts/images/runtime/Dockerfile"
+        ["backend-python"]="datamate-backend-python:scripts/images/backend-python/Dockerfile"
+    )
+    
+    # deer-flow 和 mineru 是可选的
+    OPTIONAL_SERVICES=(
+        "deer-flow-backend:deer-flow-backend:scripts/images/deer-flow-backend/Dockerfile"
+        "deer-flow-frontend:deer-flow-frontend:scripts/images/deer-flow-frontend/Dockerfile"
+        "mineru:datamate-mineru:scripts/images/mineru/Dockerfile"
+    )
+    
+    log_info ""
+    log_info "构建核心服务..."
+    local failed=()
+    local succeeded=()
+    
+    for service_name in "${!SERVICES[@]}"; do
+        IFS=':' read -r image_name dockerfile <<< "${SERVICES[$service_name]}"
+        if build_service "$service_name" "$image_name" "$dockerfile"; then
+            succeeded+=("$service_name")
+        else
+            failed+=("$service_name")
+        fi
+        echo ""
+    done
+    
+    # 尝试构建可选服务
+    log_info "构建可选服务..."
+    for service_config in "${OPTIONAL_SERVICES[@]}"; do
+        IFS=':' read -r service_name image_name dockerfile <<< "$service_config"
+        if build_service "$service_name" "$image_name" "$dockerfile"; then
+            succeeded+=("$service_name")
+        else
+            log_warn "$service_name 构建失败（可选服务，继续）"
+        fi
+        echo ""
+    done
+    
+    # 汇总
+    log_info "======================================"
+    log_info "构建结果"
+    log_info "======================================"
+    
+    if [ ${#succeeded[@]} -gt 0 ]; then
+        log_info "成功 (${#succeeded[@]}): ${succeeded[*]}"
+    fi
+    
+    if [ ${#failed[@]} -gt 0 ]; then
+        log_error "失败 (${#failed[@]}): ${failed[*]}"
+        exit 1
+    else
+        log_info "✓ 所有核心服务构建成功！"
+        echo ""
+        docker images --format "table {{.Repository}}:{{.Tag}}\t{{.Size}}" | grep -E "(datamate-|deer-flow-)" || true
+    fi
+}
+
+main "$@"

scripts/offline/build-offline-final.sh

@@ -0,0 +1,181 @@
+#!/bin/bash
+# 最终版离线构建脚本 - 使用预装 APT 包的基础镜像
+# Usage: ./build-offline-final.sh [cache-dir] [version]
+
+set -e
+
+CACHE_DIR="${1:-./build-cache}"
+VERSION="${2:-latest}"
+IMAGES_DIR="$CACHE_DIR/images"
+RESOURCES_DIR="$CACHE_DIR/resources"
+
+RED='\033[0;31m'
+GREEN='\033[0;32m'
+YELLOW='\033[1;33m'
+NC='\033[0m'
+
+log_info() { echo -e "${GREEN}[INFO]${NC} $1"; }
+log_warn() { echo -e "${YELLOW}[WARN]${NC} $1"; }
+log_error() { echo -e "${RED}[ERROR]${NC} $1"; }
+
+# 检查缓存目录
+if [ ! -d "$CACHE_DIR" ]; then
+    log_error "缓存目录 $CACHE_DIR 不存在"
+    exit 1
+fi
+
+# 加载基础镜像
+load_images() {
+    log_info "加载基础镜像..."
+    
+    # 优先加载带 APT 预装包的镜像集合
+    if [ -f "$IMAGES_DIR/base-images-with-apt.tar" ]; then
+        log_info "加载带 APT 预装包的基础镜像..."
+        docker load -i "$IMAGES_DIR/base-images-with-apt.tar"
+    elif [ -f "$IMAGES_DIR/base-images.tar" ]; then
+        log_warn "加载普通基础镜像（不含 APT 预装包）..."
+        docker load -i "$IMAGES_DIR/base-images.tar"
+    else
+        log_warn "基础镜像 tar 包不存在，检查本地镜像..."
+    fi
+    
+    log_info "✓ 镜像加载完成"
+}
+
+# 验证镜像是否存在
+verify_image() {
+    docker inspect "$1" > /dev/null 2>&1
+}
+
+# 构建函数
+build_service() {
+    local service_name=$1
+    local image_name=$2
+    local dockerfile=$3
+    local base_image=$4  # 必需的基础镜像
+    
+    log_info "----------------------------------------"
+    log_info "构建 $service_name"
+    log_info "----------------------------------------"
+    
+    if [ ! -f "$dockerfile" ]; then
+        log_error "Dockerfile 不存在: $dockerfile"
+        return 1
+    fi
+    
+    # 检查必需的基础镜像
+    if [ -n "$base_image" ]; then
+        if verify_image "$base_image"; then
+            log_info "✓ 基础镜像存在: $base_image"
+        else
+            log_error "✗ 缺少基础镜像: $base_image"
+            log_info "请确保已加载正确的 base-images-with-apt.tar"
+            return 1
+        fi
+    fi
+    
+    # 准备构建参数
+    local build_args=()
+    
+    # 添加资源目录参数
+    if [ -d "$RESOURCES_DIR" ]; then
+        build_args+=("--build-arg" "RESOURCES_DIR=$RESOURCES_DIR")
+    fi
+    
+    # 执行构建
+    log_info "开始构建..."
+    if docker build \
+        --pull=false \
+        "${build_args[@]}" \
+        -f "$dockerfile" \
+        -t "$image_name:$VERSION" \
+        . 2>&1; then
+        log_info "✓ $service_name 构建成功"
+        return 0
+    else
+        log_error "✗ $service_name 构建失败"
+        return 1
+    fi
+}
+
+# 主流程
+main() {
+    log_info "======================================"
+    log_info "最终版离线构建 (使用 APT 预装基础镜像)"
+    log_info "======================================"
+    
+    # 加载基础镜像
+    load_images
+    
+    # 验证关键基础镜像
+    log_info ""
+    log_info "验证预装基础镜像..."
+    REQUIRED_BASE_IMAGES=(
+        "datamate-java-base:latest"
+        "datamate-python-base:latest"
+        "datamate-runtime-base:latest"
+    )
+    
+    for img in "${REQUIRED_BASE_IMAGES[@]}"; do
+        if verify_image "$img"; then
+            log_info "  ✓ $img"
+        else
+            log_warn "  ✗ $img (缺失)"
+        fi
+    done
+    
+    # 定义服务配置
+    declare -A SERVICES=(
+        ["database"]="datamate-database:scripts/images/database/Dockerfile:"
+        ["gateway"]="datamate-gateway:scripts/offline/Dockerfile.gateway.offline:datamate-java-base:latest"
+        ["backend"]="datamate-backend:scripts/offline/Dockerfile.backend.offline:datamate-java-base:latest"
+        ["frontend"]="datamate-frontend:scripts/images/frontend/Dockerfile:"
+        ["runtime"]="datamate-runtime:scripts/offline/Dockerfile.runtime.offline-v2:datamate-runtime-base:latest"
+        ["backend-python"]="datamate-backend-python:scripts/offline/Dockerfile.backend-python.offline-v2:datamate-python-base:latest"
+    )
+    
+    log_info ""
+    log_info "======================================"
+    log_info "开始构建服务"
+    log_info "======================================"
+    
+    local failed=()
+    local succeeded=()
+    
+    for service_name in "${!SERVICES[@]}"; do
+        IFS=':' read -r image_name dockerfile base_image <<< "${SERVICES[$service_name]}"
+        if build_service "$service_name" "$image_name" "$dockerfile" "$base_image"; then
+            succeeded+=("$service_name")
+        else
+            failed+=("$service_name")
+        fi
+        echo ""
+    done
+    
+    # 汇总
+    log_info "======================================"
+    log_info "构建结果"
+    log_info "======================================"
+    
+    if [ ${#succeeded[@]} -gt 0 ]; then
+        log_info "成功 (${#succeeded[@]}): ${succeeded[*]}"
+    fi
+    
+    if [ ${#failed[@]} -gt 0 ]; then
+        log_error "失败 (${#failed[@]}): ${failed[*]}"
+        
+        log_info ""
+        log_info "提示: 如果失败是因为缺少预装基础镜像，请确保:"
+        log_info "  1. 在有网环境执行: ./scripts/offline/build-base-images.sh"
+        log_info "  2. 将生成的 base-images-with-apt.tar 传输到无网环境"
+        log_info "  3. 在无网环境加载: docker load -i base-images-with-apt.tar"
+        
+        exit 1
+    else
+        log_info "✓ 所有服务构建成功！"
+        echo ""
+        docker images --format "table {{.Repository}}:{{.Tag}}\t{{.Size}}" | grep -E "(datamate-|deer-flow-)" || true
+    fi
+}
+
+main "$@"

scripts/offline/build-offline-v2.sh

@@ -0,0 +1,249 @@
+#!/bin/bash
+# BuildKit 离线构建脚本 v2 - 增强版
+# Usage: ./build-offline-v2.sh [cache-dir] [version]
+
+set -e
+
+CACHE_DIR="${1:-./build-cache}"
+VERSION="${2:-latest}"
+BUILDKIT_CACHE_DIR="$CACHE_DIR/buildkit"
+IMAGES_DIR="$CACHE_DIR/images"
+RESOURCES_DIR="$CACHE_DIR/resources"
+
+# 颜色输出
+RED='\033[0;31m'
+GREEN='\033[0;32m'
+YELLOW='\033[1;33m'
+NC='\033[0m' # No Color
+
+log_info() {
+    echo -e "${GREEN}[INFO]${NC} $1"
+}
+
+log_warn() {
+    echo -e "${YELLOW}[WARN]${NC} $1"
+}
+
+log_error() {
+    echo -e "${RED}[ERROR]${NC} $1"
+}
+
+# 检查缓存目录
+if [ ! -d "$CACHE_DIR" ]; then
+    log_error "缓存目录 $CACHE_DIR 不存在"
+    log_info "请先解压缓存包: tar -xzf build-cache-*.tar.gz"
+    exit 1
+fi
+
+# 确保 buildx 构建器存在（使用 docker 驱动以支持本地镜像）
+setup_buildx() {
+    log_info "设置 BuildKit 构建器..."
+    
+    # 删除旧的构建器
+    if docker buildx inspect offline-builder > /dev/null 2>&1; then
+        docker buildx rm offline-builder 2>/dev/null || true
+    fi
+    
+    # 创建新的构建器，使用 docker 驱动（支持本地镜像，不需要推送到 registry）
+    docker buildx create --name offline-builder \
+        --driver docker-container \
+        --driver-opt image=moby/buildkit:buildx-stable-1 \
+        --use
+    
+    log_info "BuildKit 构建器创建完成"
+}
+
+# 加载基础镜像
+load_base_images() {
+    log_info "加载基础镜像..."
+    
+    if [ ! -f "$IMAGES_DIR/base-images.tar" ]; then
+        log_warn "基础镜像文件不存在: $IMAGES_DIR/base-images.tar"
+        log_info "检查本地是否存在所需镜像..."
+        
+        # 检查关键镜像是否存在
+        required_images=(
+            "maven:3-eclipse-temurin-21"
+            "eclipse-temurin:21-jdk"
+            "mysql:8"
+            "node:20-alpine"
+            "nginx:1.29"
+        )
+        
+        for img in "${required_images[@]}"; do
+            if ! docker inspect "$img" > /dev/null 2>&1; then
+                log_error "缺少基础镜像: $img"
+                log_info "请确保基础镜像已加载: docker load -i base-images.tar"
+                exit 1
+            fi
+        done
+        
+        log_info "本地基础镜像检查通过"
+        return
+    fi
+    
+    log_info "从 $IMAGES_DIR/base-images.tar 加载基础镜像..."
+    docker load -i "$IMAGES_DIR/base-images.tar"
+    log_info "基础镜像加载完成"
+}
+
+# 验证镜像是否存在
+verify_image() {
+    local image_name=$1
+    if docker inspect "$image_name" > /dev/null 2>&1; then
+        return 0
+    else
+        return 1
+    fi
+}
+
+# 离线构建函数
+offline_build() {
+    local service_name=$1
+    local image_name=$2
+    local dockerfile=$3
+    local cache_file="$BUILDKIT_CACHE_DIR/${service_name}-cache"
+    
+    log_info "----------------------------------------"
+    log_info "构建 [$service_name] -> $image_name:$VERSION"
+    log_info "----------------------------------------"
+    
+    if [ ! -d "$cache_file" ]; then
+        log_warn "$service_name 的缓存不存在，跳过..."
+        return 0
+    fi
+    
+    # 获取 Dockerfile 中的基础镜像
+    local base_images
+    base_images=$(grep -E '^FROM' "$dockerfile" | awk '{print $2}' | sort -u)
+    
+    log_info "检查基础镜像..."
+    for base_img in $base_images; do
+        # 跳过多阶段构建中的 AS 别名
+        base_img=$(echo "$base_img" | cut -d':' -f1-2 | sed 's/AS.*//i' | tr -d ' ')
+        
+        if [ -z "$base_img" ] || [[ "$base_img" == *"AS"* ]]; then
+            continue
+        fi
+        
+        if verify_image "$base_img"; then
+            log_info "  ✓ $base_img"
+        else
+            log_warn "  ✗ $base_img (未找到)"
+            # 尝试从 base-images.tar 中加载
+            if [ -f "$IMAGES_DIR/base-images.tar" ]; then
+                log_info "  尝试从 tar 包加载..."
+                docker load -i "$IMAGES_DIR/base-images.tar" 2>/dev/null || true
+            fi
+        fi
+    done
+    
+    # 执行离线构建
+    log_info "开始构建..."
+    
+    # 构建参数
+    local build_args=()
+    
+    # 为需要外部资源的服务添加 build-arg
+    case "$service_name" in
+        runtime|deer-flow-backend|deer-flow-frontend)
+            if [ -d "$RESOURCES_DIR" ]; then
+                build_args+=("--build-arg" "RESOURCES_DIR=$RESOURCES_DIR")
+            fi
+            ;;
+        backend-python)
+            if [ -d "$RESOURCES_DIR" ]; then
+                build_args+=("--build-arg" "RESOURCES_DIR=$RESOURCES_DIR")
+                build_args+=("--build-arg" "DATAX_LOCAL_PATH=$RESOURCES_DIR/DataX")
+            fi
+            ;;
+    esac
+    
+    # 执行构建
+    if docker buildx build \
+        --builder offline-builder \
+        --cache-from "type=local,src=$cache_file" \
+        --pull=false \
+        --output "type=docker" \
+        "${build_args[@]}" \
+        -f "$dockerfile" \
+        -t "$image_name:$VERSION" \
+        . 2>&1; then
+        
+        log_info "✓ $service_name 构建成功"
+        return 0
+    else
+        log_error "✗ $service_name 构建失败"
+        return 1
+    fi
+}
+
+# 主流程
+main() {
+    log_info "======================================"
+    log_info "BuildKit 离线构建"
+    log_info "======================================"
+    log_info "缓存目录: $CACHE_DIR"
+    log_info "版本: $VERSION"
+    
+    # 步骤 1: 设置构建器
+    setup_buildx
+    
+    # 步骤 2: 加载基础镜像
+    load_base_images
+    
+    # 步骤 3: 定义服务列表
+    declare -A SERVICES=(
+        ["database"]="datamate-database:scripts/images/database/Dockerfile"
+        ["gateway"]="datamate-gateway:scripts/images/gateway/Dockerfile"
+        ["backend"]="datamate-backend:scripts/images/backend/Dockerfile"
+        ["frontend"]="datamate-frontend:scripts/images/frontend/Dockerfile"
+        ["runtime"]="datamate-runtime:scripts/images/runtime/Dockerfile"
+        ["backend-python"]="datamate-backend-python:scripts/images/backend-python/Dockerfile"
+        ["deer-flow-backend"]="deer-flow-backend:scripts/images/deer-flow-backend/Dockerfile"
+        ["deer-flow-frontend"]="deer-flow-frontend:scripts/images/deer-flow-frontend/Dockerfile"
+        ["mineru"]="datamate-mineru:scripts/images/mineru/Dockerfile"
+    )
+    
+    # 步骤 4: 批量构建
+    log_info ""
+    log_info "======================================"
+    log_info "开始批量构建"
+    log_info "======================================"
+    
+    local failed=()
+    local succeeded=()
+    
+    for service_name in "${!SERVICES[@]}"; do
+        IFS=':' read -r image_name dockerfile <<< "${SERVICES[$service_name]}"
+        
+        if offline_build "$service_name" "$image_name" "$dockerfile"; then
+            succeeded+=("$service_name")
+        else
+            failed+=("$service_name")
+        fi
+        echo ""
+    done
+    
+    # 步骤 5: 汇总结果
+    log_info "======================================"
+    log_info "构建完成"
+    log_info "======================================"
+    
+    if [ ${#succeeded[@]} -gt 0 ]; then
+        log_info "成功 (${#succeeded[@]}): ${succeeded[*]}"
+    fi
+    
+    if [ ${#failed[@]} -gt 0 ]; then
+        log_error "失败 (${#failed[@]}): ${failed[*]}"
+        exit 1
+    else
+        log_info "✓ 所有服务构建成功！"
+        echo ""
+        log_info "镜像列表:"
+        docker images --format "table {{.Repository}}:{{.Tag}}\t{{.Size}}" | grep -E "(datamate-|deer-flow-)" || true
+    fi
+}
+
+# 执行主流程
+main "$@"

scripts/offline/build-offline.sh

@@ -77,22 +77,24 @@ for service_config in "${SERVICES[@]}"; do
     fi
     
     # 使用缓存进行离线构建
-    # --network=none 确保不访问网络
+    # --pull=false: 不尝试拉取镜像
+    # --network=none: 禁用网络访问
     docker buildx build \
         --cache-from "type=local,src=$cache_file" \
+        --pull=false \
         --network=none \
         -f "$dockerfile" \
         -t "$image_name:$VERSION" \
         --load \
-        . || {
+        . 2>&1 || {
-            echo "错误: $service_name 构建失败"
+            echo "警告: $service_name 离线构建遇到问题，尝试仅使用缓存..."
-            echo "尝试不使用 --network=none 重新构建..."
             docker buildx build \
                 --cache-from "type=local,src=$cache_file" \
+                --pull=false \
                 -f "$dockerfile" \
                 -t "$image_name:$VERSION" \
                 --load \
-                .
+                . 2>&1
         }
     
     echo "✓ $service_name 构建完成"

scripts/offline/diagnose.sh

@@ -0,0 +1,151 @@
+#!/bin/bash
+# 离线构建诊断脚本
+# Usage: ./diagnose.sh [cache-dir]
+
+set -e
+
+CACHE_DIR="${1:-./build-cache}"
+
+echo "======================================"
+echo "离线构建环境诊断"
+echo "======================================"
+echo ""
+
+# 1. 检查 Docker
+echo "1. Docker 版本:"
+docker version --format '{{.Server.Version}}' 2>/dev/null || echo "  无法获取版本"
+echo ""
+
+# 2. 检查 BuildKit
+echo "2. BuildKit 状态:"
+if docker buildx version > /dev/null 2>&1; then
+    docker buildx version
+    echo ""
+    echo "可用的构建器:"
+    docker buildx ls
+else
+    echo "  BuildKit 不可用"
+fi
+echo ""
+
+# 3. 检查缓存目录
+echo "3. 缓存目录检查 ($CACHE_DIR):"
+if [ -d "$CACHE_DIR" ]; then
+    echo "  ✓ 缓存目录存在"
+    
+    # 检查子目录
+    for subdir in buildkit images resources; do
+        if [ -d "$CACHE_DIR/$subdir" ]; then
+            echo "  ✓ $subdir/ 存在"
+            count=$(find "$CACHE_DIR/$subdir" -type d | wc -l)
+            echo "    子目录数量: $count"
+        else
+            echo "  ✗ $subdir/ 不存在"
+        fi
+    done
+else
+    echo "  ✗ 缓存目录不存在"
+fi
+echo ""
+
+# 4. 检查基础镜像
+echo "4. 基础镜像检查:"
+required_images=(
+    "maven:3-eclipse-temurin-21"
+    "maven:3-eclipse-temurin-8"
+    "eclipse-temurin:21-jdk"
+    "mysql:8"
+    "node:20-alpine"
+    "nginx:1.29"
+    "ghcr.nju.edu.cn/astral-sh/uv:python3.11-bookworm"
+    "python:3.12-slim"
+    "gcr.io/distroless/nodejs20-debian12"
+)
+
+missing_images=()
+for img in "${required_images[@]}"; do
+    if docker inspect "$img" > /dev/null 2>&1; then
+        size=$(docker images --format "{{.Size}}" "$img" | head -1)
+        echo "  ✓ $img ($size)"
+    else
+        echo "  ✗ $img (缺失)"
+        missing_images+=("$img")
+    fi
+done
+echo ""
+
+# 5. 检查 BuildKit 缓存
+echo "5. BuildKit 缓存检查:"
+if [ -d "$CACHE_DIR/buildkit" ]; then
+    for cache_dir in "$CACHE_DIR/buildkit"/*-cache; do
+        if [ -d "$cache_dir" ]; then
+            name=$(basename "$cache_dir")
+            size=$(du -sh "$cache_dir" 2>/dev/null | cut -f1)
+            echo "  ✓ $name ($size)"
+        fi
+    done
+else
+    echo "  ✗ 缓存目录不存在"
+fi
+echo ""
+
+# 6. 检查资源文件
+echo "6. 外部资源检查:"
+if [ -d "$CACHE_DIR/resources" ]; then
+    if [ -f "$CACHE_DIR/resources/models/ch_ppocr_mobile_v2.0_cls_infer.tar" ]; then
+        size=$(du -sh "$CACHE_DIR/resources/models/ch_ppocr_mobile_v2.0_cls_infer.tar" | cut -f1)
+        echo "  ✓ PaddleOCR 模型 ($size)"
+    else
+        echo "  ✗ PaddleOCR 模型缺失"
+    fi
+    
+    if [ -f "$CACHE_DIR/resources/models/zh_core_web_sm-3.8.0-py3-none-any.whl" ]; then
+        size=$(du -sh "$CACHE_DIR/resources/models/zh_core_web_sm-3.8.0-py3-none-any.whl" | cut -f1)
+        echo "  ✓ spaCy 模型 ($size)"
+    else
+        echo "  ✗ spaCy 模型缺失"
+    fi
+    
+    if [ -d "$CACHE_DIR/resources/DataX" ]; then
+        echo "  ✓ DataX 源码"
+    else
+        echo "  ✗ DataX 源码缺失"
+    fi
+    
+    if [ -d "$CACHE_DIR/resources/deer-flow" ]; then
+        echo "  ✓ deer-flow 源码"
+    else
+        echo "  ✗ deer-flow 源码缺失"
+    fi
+else
+    echo "  ✗ 资源目录不存在"
+fi
+echo ""
+
+# 7. 网络检查
+echo "7. 网络检查:"
+if ping -c 1 8.8.8.8 > /dev/null 2>&1; then
+    echo "  ⚠ 网络可用（离线构建环境通常不需要）"
+else
+    echo "  ✓ 网络不可达（符合离线环境）"
+fi
+echo ""
+
+# 8. 总结
+echo "======================================"
+echo "诊断总结"
+echo "======================================"
+
+if [ ${#missing_images[@]} -eq 0 ]; then
+    echo "✓ 所有基础镜像已就绪"
+else
+    echo "✗ 缺少 ${#missing_images[@]} 个基础镜像:"
+    printf '  - %s\n' "${missing_images[@]}"
+    echo ""
+    echo "修复方法:"
+    if [ -f "$CACHE_DIR/images/base-images.tar" ]; then
+        echo "  docker load -i $CACHE_DIR/images/base-images.tar"
+    else
+        echo "  请确保有网环境导出时包含所有基础镜像"
+    fi
+fi

scripts/offline/export-cache.sh

@@ -8,6 +8,7 @@ OUTPUT_DIR="${1:-./build-cache}"
 BUILDKIT_CACHE_DIR="$OUTPUT_DIR/buildkit"
 IMAGES_DIR="$OUTPUT_DIR/images"
 RESOURCES_DIR="$OUTPUT_DIR/resources"
+APT_CACHE_DIR="$OUTPUT_DIR/apt-cache"
 
 # 确保 buildx 构建器存在
 if ! docker buildx inspect offline-builder > /dev/null 2>&1; then
@@ -17,7 +18,7 @@ else
     docker buildx use offline-builder
 fi
 
-mkdir -p "$BUILDKIT_CACHE_DIR" "$IMAGES_DIR" "$RESOURCES_DIR"
+mkdir -p "$BUILDKIT_CACHE_DIR" "$IMAGES_DIR" "$RESOURCES_DIR" "$APT_CACHE_DIR"
 
 echo "======================================"
 echo "1. 导出基础镜像"
@@ -117,11 +118,42 @@ fi
 
 echo ""
 echo "======================================"
-echo "4. 打包缓存"
+echo "4. 导出 APT 缓存"
+echo "======================================"
+
+# 为需要 apt 的镜像预生成 apt 缓存
+echo "生成 APT list 缓存..."
+
+# eclipse-temurin:21-jdk 的 apt 缓存
+docker run --rm \
+    -v "$APT_CACHE_DIR/eclipse-temurin:/var/cache/apt/archives" \
+    -v "$APT_CACHE_DIR/eclipse-temurin-lists:/var/lib/apt/lists" \
+    eclipse-temurin:21-jdk \
+    bash -c "apt-get update && apt-get install -y --download-only vim wget curl rsync python3 python3-pip python-is-python3 dos2unix libreoffice fonts-noto-cjk 2>/dev/null || true" 2>/dev/null || echo "  Warning: eclipse-temurin apt 缓存导出失败"
+
+# python:3.12-slim 的 apt 缓存
+docker run --rm \
+    -v "$APT_CACHE_DIR/python312:/var/cache/apt/archives" \
+    -v "$APT_CACHE_DIR/python312-lists:/var/lib/apt/lists" \
+    python:3.12-slim \
+    bash -c "apt-get update && apt-get install -y --download-only vim openjdk-21-jre nfs-common glusterfs-client rsync 2>/dev/null || true" 2>/dev/null || echo "  Warning: python3.12 apt 缓存导出失败"
+
+# python:3.11-slim 的 apt 缓存
+docker run --rm \
+    -v "$APT_CACHE_DIR/python311:/var/cache/apt/archives" \
+    -v "$APT_CACHE_DIR/python311-lists:/var/lib/apt/lists" \
+    python:3.11-slim \
+    bash -c "apt-get update && apt-get install -y --download-only curl vim libgl1 libglx0 libopengl0 libglib2.0-0 procps 2>/dev/null || true" 2>/dev/null || echo "  Warning: python3.11 apt 缓存导出失败"
+
+echo "✓ APT 缓存导出完成"
+
+echo ""
+echo "======================================"
+echo "5. 打包缓存"
 echo "======================================"
 
 cd "$OUTPUT_DIR"
-tar -czf "build-cache-$(date +%Y%m%d).tar.gz" buildkit images resources
+tar -czf "build-cache-$(date +%Y%m%d).tar.gz" buildkit images resources apt-cache
 cd - > /dev/null
 
 echo ""
@@ -131,4 +163,10 @@ echo "======================================"
 echo "缓存位置: $OUTPUT_DIR"
 echo "传输文件: $OUTPUT_DIR/build-cache-$(date +%Y%m%d).tar.gz"
 echo ""
+echo "包含内容:"
+echo "  - 基础镜像 (images/)"
+echo "  - BuildKit 缓存 (buildkit/)"
+echo "  - 外部资源 (resources/)"
+echo "  - APT 缓存 (apt-cache/)"
+echo ""
 echo "请将此压缩包传输到无网环境后解压使用"