LQAdminPlatform/src/app/services/task_service.py

import logging
import json
import httpx
from datetime import datetime, date
from typing import List, Dict, Any, Tuple, Optional
from app.base.async_mysql_connection import get_db_connection
from app.base.minio_connection import get_minio_manager

logger = logging.getLogger(__name__)

class TaskService:
    """任务管理服务类"""
    
    _schema_verified = False  # 类级别变量，确保 DDL 逻辑只运行一次

    def __init__(self):
        self.minio_manager = get_minio_manager()

    async def _ensure_table_schema(self, cursor, conn):
        """确保表结构和索引正确 (DDL 操作)"""
        if TaskService._schema_verified:
            return
        
        try:
            # 1. 动态维护字段
            cursor.execute("SHOW COLUMNS FROM t_task_management LIKE 'tag'")
            if not cursor.fetchone():
                cursor.execute("ALTER TABLE t_task_management ADD COLUMN tag json NULL COMMENT '标签' AFTER type")
            
            cursor.execute("SHOW COLUMNS FROM t_task_management LIKE 'metadata'")
            if not cursor.fetchone():
                cursor.execute("ALTER TABLE t_task_management ADD COLUMN metadata json NULL COMMENT '业务元数据' AFTER tag")
            
            cursor.execute("SHOW COLUMNS FROM t_task_management LIKE 'project_name'")
            if not cursor.fetchone():
                cursor.execute("ALTER TABLE t_task_management ADD COLUMN project_name varchar(255) NULL COMMENT '项目显示名称' AFTER project_id")
            
            cursor.execute("SHOW COLUMNS FROM t_task_management LIKE 'external_completed_count'")
            if not cursor.fetchone():
                cursor.execute("ALTER TABLE t_task_management ADD COLUMN external_completed_count int NULL DEFAULT 0 COMMENT '外部平台完成数量' AFTER annotation_status")
            
            cursor.execute("SHOW COLUMNS FROM t_task_management LIKE 'external_total_count'")
            if not cursor.fetchone():
                cursor.execute("ALTER TABLE t_task_management ADD COLUMN external_total_count int NULL DEFAULT 0 COMMENT '外部平台总任务数量' AFTER external_completed_count")
            
            conn.commit()

            # 2. 处理索引冲突
            cursor.execute("SHOW COLUMNS FROM t_task_management LIKE 'file_url'")
            if not cursor.fetchone():
                cursor.execute("ALTER TABLE t_task_management ADD COLUMN file_url VARCHAR(512) DEFAULT NULL AFTER metadata")
                logger.info("Added column file_url to t_task_management")

            cursor.execute("SHOW INDEX FROM t_task_management WHERE Column_name = 'business_id'")
            indexes = cursor.fetchall()
            for idx in indexes:
                if not idx['Non_unique'] and idx['Key_name'] != 'PRIMARY' and idx['Seq_in_index'] == 1:
                    cursor.execute(f"SHOW INDEX FROM t_task_management WHERE Key_name = '{idx['Key_name']}'")
                    if len(cursor.fetchall()) == 1:
                        cursor.execute(f"DROP INDEX {idx['Key_name']} ON t_task_management")
                        logger.info(f"Dropped old unique index: {idx['Key_name']}")
            
            cursor.execute("SHOW INDEX FROM t_task_management WHERE Key_name = 'uk_business_project'")
            if not cursor.fetchone():
                cursor.execute("CREATE UNIQUE INDEX uk_business_project ON t_task_management (business_id, project_id)")
                logger.info("Created new composite unique index: uk_business_project")
            
            conn.commit()
            TaskService._schema_verified = True
        except Exception as e:
            logger.warning(f"表结构维护失败: {e}")
            conn.rollback()

    async def add_task(self, business_id: str, task_type: str, task_id: str = None, project_id: str = None, project_name: str = None, tag: str = None, metadata: str = None) -> Tuple[bool, str, Optional[int]]:
        """添加或更新任务记录 (适配单表结构)"""
        conn = get_db_connection()
        if not conn:
            return False, "数据库连接失败", None
        
        cursor = conn.cursor()
        try:
            # 确保表结构（仅在第一次调用时执行）
            await self._ensure_table_schema(cursor, conn)

            # 执行插入/更新
            sql = """
                INSERT INTO t_task_management (business_id, task_id, project_id, project_name, type, annotation_status, tag, metadata)
                VALUES (%s, %s, %s, %s, %s, %s, %s, %s)
                ON DUPLICATE KEY UPDATE 
                    task_id = IFNULL(VALUES(task_id), task_id),
                    project_id = IFNULL(VALUES(project_id), project_id),
                    project_name = IFNULL(VALUES(project_name), project_name),
                    annotation_status = IFNULL(VALUES(annotation_status), annotation_status),
                    tag = IFNULL(VALUES(tag), tag),
                    metadata = IFNULL(VALUES(metadata), metadata)
            """
            cursor.execute(sql, (business_id, task_id, project_id, project_name, task_type, 'pending', tag, metadata))
            record_id = cursor.lastrowid
            
            conn.commit()
            return True, "成功", record_id
        except Exception as e:
            conn.rollback()
            logger.exception(f"添加任务失败: {e}")
            return False, str(e), None
        finally:
            cursor.close()
            conn.close()

    async def delete_task_by_id(self, id: int) -> Tuple[bool, str]:
        """根据主键 id 删除任务记录"""
        conn = get_db_connection()
        if not conn:
            return False, "数据库连接失败"
        
        cursor = conn.cursor()
        try:
            sql = "DELETE FROM t_task_management WHERE id = %s"
            cursor.execute(sql, (id,))
            conn.commit()
            return True, "成功"
        except Exception as e:
            conn.rollback()
            logger.exception(f"删除任务失败: {e}")
            return False, str(e)
        finally:
            cursor.close()
            conn.close()

    def _serialize_datetime(self, obj: Any) -> Any:
        """递归遍历对象，将 datetime 转换为字符串"""
        if isinstance(obj, dict):
            return {k: self._serialize_datetime(v) for k, v in obj.items()}
        elif isinstance(obj, list):
            return [self._serialize_datetime(i) for i in obj]
        elif isinstance(obj, (datetime, date)):
            return obj.strftime('%Y-%m-%d %H:%M:%S') if isinstance(obj, datetime) else obj.strftime('%Y-%m-%d')
        return obj

    async def get_task_list(self, task_type: str) -> List[Dict[str, Any]]:
        """获取项目列表 (按 project_id 聚合)"""
        conn = get_db_connection()
        if not conn or task_type not in ['data', 'image']:
            return []
        
        cursor = conn.cursor()
        try:
            # 确保表结构
            await self._ensure_table_schema(cursor, conn)
            
            # 修改聚合逻辑：返回 project_id (UUID) 和 project_name (文字)
            sql = """
                SELECT 
                    project_id, 
                    MAX(project_name) as project_name,
                    MAX(task_id) as task_id,
                    MAX(tag) as tag,
                    MAX(id) as sort_id,
                    COALESCE(NULLIF(MAX(external_total_count), 0), COUNT(*)) as file_count,
                    COALESCE(
                        MAX(external_completed_count),
                        SUM(CASE WHEN annotation_status = 'completed' THEN 1 ELSE 0 END),
                        0
                    ) as completed_count
                FROM t_task_management 
                WHERE type = %s 
                GROUP BY project_id
                ORDER BY sort_id DESC
            """
            cursor.execute(sql, (task_type,))
            rows = cursor.fetchall()
            
            # 兼容旧数据：如果 project_name 为空，则用 project_id 代替
            for row in rows:
                if not row['project_name']:
                    row['project_name'] = row['project_id']
            
            return self._serialize_datetime(rows)
        except Exception as e:
            logger.error(f"获取任务列表失败: {e}")
            return []
        finally:
            cursor.close()
            conn.close()

    async def get_project_details(self, project_id: str, task_type: str) -> List[Dict[str, Any]]:
        """获取项目详情 (按 project_id 查询)"""
        conn = get_db_connection()
        if not conn:
            return []
        
        cursor = conn.cursor()
        try:
            # 确保表结构
            await self._ensure_table_schema(cursor, conn)
            
            # 修改查询逻辑：获取 project_name 并在结果中包含它
            sql = """
                SELECT 
                    id, business_id, task_id, project_id, project_name, 
                    type, annotation_status, tag, metadata
                FROM t_task_management
                WHERE project_id = %s AND type = %s
            """
            cursor.execute(sql, (project_id, task_type))
            rows = cursor.fetchall()
            
            # 处理 tag 和 metadata 的 JSON 解析
            for row in rows:
                if row.get('tag'):
                    try: row['tag'] = json.loads(row['tag']) if isinstance(row['tag'], str) else row['tag']
                    except: pass
                if row.get('metadata'):
                    try: 
                        meta = json.loads(row['metadata']) if isinstance(row['metadata'], str) else row['metadata']
                        row['metadata'] = meta
                        # 提取名称供前端显示
                        if row['type'] == 'data':
                            row['name'] = meta.get('title') or meta.get('filename') or row['business_id']
                        elif row['type'] == 'image':
                            row['name'] = meta.get('image_name') or row['business_id']
                        else:
                            row['name'] = row['business_id']
                    except: 
                        row['name'] = row['business_id']
                else:
                    row['name'] = row['business_id']
            
            return self._serialize_datetime(rows)
        except Exception as e:
            logger.error(f"获取项目详情失败: {e}")
            return []
        finally:
            cursor.close()
            conn.close()

    async def delete_task(self, business_id: str) -> Tuple[bool, str]:
        """删除任务记录"""
        conn = get_db_connection()
        if not conn:
            return False, "数据库连接失败"
        
        cursor = conn.cursor()
        try:
            sql = "DELETE FROM t_task_management WHERE business_id = %s"
            cursor.execute(sql, (business_id,))
            conn.commit()
            return True, "删除成功"
        except Exception as e:
            logger.exception(f"删除任务记录失败: {e}")
            conn.rollback()
            return False, f"删除失败: {str(e)}"
        finally:
            cursor.close()
            conn.close()

    async def create_anno_project(self, data: Dict[str, Any]) -> Tuple[bool, str]:
        """创建标注项目并同步任务数据"""
        project_name = data.get('name')
        if not project_name:
            return False, "项目名称不能为空"
        
        # 0. 统一使用 UUID 方案获取或生成 project_id
        conn = get_db_connection()
        if not conn:
            return False, "数据库连接失败"
        
        cursor = conn.cursor()
        try:
            project_id = None
            cursor.execute("SELECT project_id FROM t_task_management WHERE project_name = %s LIMIT 1", (project_name,))
            existing_project = cursor.fetchone()
            if existing_project:
                project_id = existing_project['project_id']
            else:
                import uuid
                project_id = str(uuid.uuid4())

            task_type = data.get('task_type', 'data')
            # 映射回内部类型
            internal_type_map = {
                'text_classification': 'data',
                'image_classification': 'image'
            }
            internal_task_type = internal_type_map.get(task_type, task_type)
            
            tasks_data = data.get('data', [])
            if not tasks_data:
                return False, "任务数据不能为空"
            
            # 提取全局标签名列表
            global_tags = []
            if data.get('tags'):
                global_tags = [t['tag'] for t in data['tags'] if 'tag' in t]
            
            # 批量写入任务
            import json
            tag_str = json.dumps(global_tags, ensure_ascii=False) if global_tags else None
            
            for item in tasks_data:
                business_id = item.get('id')
                if not business_id: continue
                
                # 检查是否已存在 (使用联合主键逻辑)
                cursor.execute(
                    "SELECT id FROM t_task_management WHERE business_id = %s AND project_id = %s",
                    (business_id, project_id)
                )
                if cursor.fetchone():
                    continue
                
                metadata_str = json.dumps(item.get('metadata', {}), ensure_ascii=False)
                
                sql = """
                    INSERT INTO t_task_management 
                    (business_id, type, project_id, project_name, tag, metadata, annotation_status)
                    VALUES (%s, %s, %s, %s, %s, %s, 'pending')
                """
                cursor.execute(sql, (business_id, internal_task_type, project_id, project_name, tag_str, metadata_str))
            
            conn.commit()
            
            # 自动推送至外部平台
            success, msg = await self.send_to_external_platform(project_id)
            if success:
                return True, project_id
            else:
                return False, f"任务已保存但推送失败: {msg}"
                
        except Exception as e:
            logger.exception(f"创建项目失败: {e}")
            conn.rollback()
            return False, str(e)
        finally:
            cursor.close()
            conn.close()

    async def get_project_progress(self, project_id: str) -> Dict[str, Any]:
        """获取项目进度统计 (单表化)"""
        conn = get_db_connection()
        if not conn:
            return {}
        
        cursor = conn.cursor()
        try:
            # 统计各状态数量
            sql = """
                SELECT 
                    annotation_status, 
                    COUNT(*) as count 
                FROM t_task_management 
                WHERE project_id = %s 
                GROUP BY annotation_status
            """
            cursor.execute(sql, (project_id,))
            stats = cursor.fetchall()
            
            if not stats:
                return {}

            total = sum(s['count'] for s in stats)
            completed = sum(s['count'] for s in stats if s['annotation_status'] == 'completed')
            
            return {
                "project_id": project_id,
                "total": total,
                "completed": completed,
                "progress": f"{round(completed/total*100, 2)}%" if total > 0 else "0%",
                "details": stats
            }
        except Exception as e:
            logger.error(f"查询进度失败: {e}")
            return {}
        finally:
            cursor.close()
            conn.close()

    def _get_milvus_content(self, task_id: str, kb_info: Dict[str, Any]) -> List[str]:
        """
        从 Milvus 获取文档分片内容
        """
        if not kb_info:
            return []

        # 优先使用数据库存储的子表名，父表名作为兜底
        collections = [c for c in [kb_info.get('collection_name_children'), kb_info.get('collection_name_parent')] if c]
        
        if not collections:
            return []

        from app.services.milvus_service import milvus_service
        contents = []
        
        # 简单的全局缓存，用于存储集合的字段探测结果，减少 describe_collection 调用
        if not hasattr(self, '_collection_schema_cache'):
            self._collection_schema_cache = {}

        for coll_name in collections:
            try:
                if not milvus_service.client.has_collection(coll_name):
                    continue
                
                # 获取或更新缓存的字段名
                if coll_name not in self._collection_schema_cache:
                    schema = milvus_service.client.describe_collection(coll_name)
                    field_names = [f['name'] for f in schema.get('fields', [])]
                    self._collection_schema_cache[coll_name] = {
                        "id": "document_id", # 统一使用 document_id
                        "content": "text" if "text" in field_names else "content"
                    }
                
                fields = self._collection_schema_cache[coll_name]
                id_field = fields["id"]
                content_field = fields["content"]
                
                res = milvus_service.client.query(
                    collection_name=coll_name,
                    filter=f'{id_field} == "{task_id}"',
                    output_fields=[content_field]
                )
                
                if res:
                    for s in res:
                        val = s.get(content_field)
                        if val: contents.append(val)
                    
                    if contents:
                        return contents
            except Exception as e:
                logger.error(f"查询 Milvus 集合 {coll_name} 异常: {e}")
                continue
        
        return []

    async def export_project_data(self, project_id: str, conn=None) -> Dict[str, Any]:
        """导出项目数据为标注平台要求的格式 (单表化)"""
        should_close = False
        if not conn:
            conn = get_db_connection()
            should_close = True
            
        if not conn:
            return {}
        
        cursor = conn.cursor()
        try:
            # 1. 获取任务记录
            sql_tasks = """
                SELECT business_id as id, type, task_id, project_name, tag, metadata
                FROM t_task_management
                WHERE project_id = %s
            """
            cursor.execute(sql_tasks, (project_id,))
            rows = cursor.fetchall()
            
            if not rows:
                return {}
            
            # 2. 解析基本信息
            first_row = rows[0]
            internal_task_type = first_row['type']
            project_name = first_row.get('project_name') or project_id
            remote_project_id = first_row.get('task_id') or project_id
            
            # 映射任务类型
            type_map = {'data': 'text_classification', 'image': 'image_classification'}
            external_task_type = type_map.get(internal_task_type, internal_task_type)

            # 3. 处理数据
            final_tasks = []
            all_project_tags = set()
            
            # 针对 'data' 类型的批量 Milvus 查询优化
            milvus_data_map = {}
            if internal_task_type == 'data':
                all_task_ids = [r['id'] for r in rows]
                sql_kb = """
                    SELECT kb.collection_name_parent, kb.collection_name_children
                    FROM t_samp_document_main d
                    LEFT JOIN t_samp_knowledge_base kb ON d.kb_id = kb.id
                    WHERE d.id COLLATE utf8mb4_unicode_ci = %s COLLATE utf8mb4_unicode_ci
                """
                cursor.execute(sql_kb, (all_task_ids[0],))
                kb_info = cursor.fetchone()
                if kb_info:
                    milvus_data_map = self._get_milvus_content_batch(all_task_ids, kb_info)

            for item in rows:
                task_id = item['id']
                
                # 记录原始数据状态
                logger.debug(f"正在处理导出任务 {task_id}, tag: {item.get('tag')}, metadata_keys: {list(json.loads(item['metadata']).keys()) if item.get('metadata') else 'None'}")
                
                # 提取并处理标签
                doc_tags = []
                if item.get('tag'):
                    try:
                        doc_tags = json.loads(item['tag']) if isinstance(item['tag'], str) else item['tag']
                        if doc_tags:
                            for t in doc_tags: all_project_tags.add(t)
                    except: pass
                
                # 解析数据库元数据
                db_metadata = {}
                if item.get('metadata'):
                    try:
                        db_metadata = json.loads(item['metadata']) if isinstance(item['metadata'], str) else item['metadata']
                    except: pass
                
                # 获取任务内容
                task_contents = []
                annotation_results = []
                if internal_task_type == 'data':
                    task_contents = milvus_data_map.get(task_id, [])
                    if not task_contents:
                        cursor.execute("SELECT title FROM t_samp_document_main WHERE id = %s", (task_id,))
                        res = cursor.fetchone()
                        if res: task_contents = [res['title']]
                elif internal_task_type == 'image':
                    cursor.execute("SELECT image_url FROM t_image_info WHERE id = %s", (task_id,))
                    res = cursor.fetchone()
                    if res:
                        img_url = res['image_url']
                        if img_url and not img_url.startswith('http'):
                            img_url = self.minio_manager.get_full_url(img_url)
                        task_contents = [img_url]

                # 构建最终任务列表
                for idx, content in enumerate(task_contents):
                    if not content: continue
                    
                    # 合并元数据：数据库数据 + 动态 ID
                    task_metadata = {
                        "original_id": task_id,
                        "chunk_index": idx
                    }
                    if db_metadata:
                        task_metadata.update(db_metadata)
                    
                    if doc_tags:
                        task_metadata['tags'] = [{"tag": tag} for tag in doc_tags]
                        
                    task_item = {
                        "id": f"{task_id}_{idx}" if len(task_contents) > 1 else task_id,
                        "content": content,
                        "metadata": task_metadata
                    }
                    
                    # 尝试从元数据中提取 annotation_result
                    if db_metadata and 'annotation_result' in db_metadata:
                        task_item['annotation_result'] = db_metadata['annotation_result']
                    
                    final_tasks.append(task_item)

            # 统一进行一次递归序列化处理
            return self._serialize_datetime({
                "name": project_name,
                "description": "",
                "task_type": external_task_type,
                "data": final_tasks,
                "external_id": remote_project_id,
                "tags": [{"tag": t} for t in sorted(list(all_project_tags))]
            })
        except Exception as e:
            logger.exception(f"导出项目数据异常: {e}")
            return {}
        finally:
            cursor.close()
            if should_close:
                conn.close()

    def _get_milvus_content_batch(self, task_ids: List[str], kb_info: Dict[str, Any]) -> Dict[str, List[str]]:
        """
        批量从 Milvus 获取文档分片内容
        """
        if not kb_info or not task_ids:
            return {}

        collections = [c for c in [kb_info.get('collection_name_children'), kb_info.get('collection_name_parent')] if c]
        if not collections:
            return {}

        from app.services.milvus_service import milvus_service
        result_map = {tid: [] for tid in task_ids}
        
        if not hasattr(self, '_collection_schema_cache'):
            self._collection_schema_cache = {}

        for coll_name in collections:
            try:
                if not milvus_service.client.has_collection(coll_name):
                    continue
                
                if coll_name not in self._collection_schema_cache:
                    schema = milvus_service.client.describe_collection(coll_name)
                    field_names = [f['name'] for f in schema.get('fields', [])]
                    self._collection_schema_cache[coll_name] = {
                        "id": "document_id",
                        "content": "text" if "text" in field_names else "content"
                    }
                
                fields = self._collection_schema_cache[coll_name]
                id_field = fields["id"]
                content_field = fields["content"]
                
                # 使用 in 表达式进行批量查询
                # 注意：如果 task_ids 非常多，可能需要分批（如每批 100 个）
                id_list_str = ", ".join([f'"{tid}"' for tid in task_ids])
                res = milvus_service.client.query(
                    collection_name=coll_name,
                    filter=f'{id_field} in [{id_list_str}]',
                    output_fields=[id_field, content_field]
                )
                
                if res:
                    for s in res:
                        tid = s.get(id_field)
                        val = s.get(content_field)
                        if tid in result_map and val:
                            result_map[tid].append(val)
                    
                    # 如果当前集合已经查到了内容，就不再查兜底集合（逻辑同单条查询）
                    if any(result_map.values()):
                        return result_map
            except Exception as e:
                logger.error(f"批量查询 Milvus 集合 {coll_name} 异常: {e}")
                continue
        
        return result_map

    async def get_project_progress(self, project_id: str) -> Dict[str, Any]:
        """获取外部标注项目的进度"""
        conn = get_db_connection()
        if not conn:
            return {"error": "数据库连接失败"}
            
        try:
            # 1. 查询 remote_project_id (task_id)
            cursor = conn.cursor()
            cursor.execute("SELECT task_id FROM t_task_management WHERE project_id = %s AND task_id IS NOT NULL LIMIT 1", (project_id,))
            row = cursor.fetchone()
            cursor.close()
            
            if not row or not row['task_id']:
                return {"error": "未找到已推送的外部项目ID"}
                
            remote_project_id = row['task_id']
            
            # 2. 获取配置
            from app.core.config import config_handler
            api_base_url = config_handler.get('external_api', 'project_api_url', 'http://192.168.92.61:9003/api/external/projects').rstrip('/')
            progress_url = f"{api_base_url}/{remote_project_id}/progress"
            token = config_handler.get('external_api', 'admin_token', '')

            # 3. 发送请求
            async with httpx.AsyncClient(timeout=10.0) as client:
                headers = {"Authorization": f"Bearer {token}"}
                response = await client.get(progress_url, headers=headers)
                
                if response.status_code == 200:
                    data = response.json()
                    # 同步更新本地缓存的完成数量和总数
                    completed_count = data.get('completed_tasks', 0)
                    total_count = data.get('total_tasks', 0)
                    try:
                        conn_update = get_db_connection()
                        if conn_update:
                            cursor_update = conn_update.cursor()
                            cursor_update.execute(
                                """
                                UPDATE t_task_management 
                                SET external_completed_count = %s, external_total_count = %s 
                                WHERE project_id = %s
                                """,
                                (completed_count, total_count, project_id)
                            )
                            conn_update.commit()
                            cursor_update.close()
                            conn_update.close()
                    except Exception as ex:
                        logger.warning(f"更新本地进度缓存失败: {ex}")
                        
                    return data
                else:
                    logger.error(f"查询进度失败: {response.status_code} - {response.text}")
                    return {"error": f"外部平台返回错误 ({response.status_code})"}
                    
        except Exception as e:
            logger.exception(f"查询进度异常: {e}")
            return {"error": str(e)}
        finally:
            conn.close()

    async def export_labeled_data(self, project_id: str, export_format: str = 'json', completed_only: bool = True) -> Dict[str, Any]:
        """触发外部标注项目的数据导出"""
        conn = get_db_connection()
        if not conn:
            return {"error": "数据库连接失败"}
            
        try:
            # 1. 查询 remote_project_id (task_id)
            cursor = conn.cursor()
            cursor.execute("SELECT task_id FROM t_task_management WHERE project_id = %s AND task_id IS NOT NULL LIMIT 1", (project_id,))
            row = cursor.fetchone()
            cursor.close()
            
            if not row or not row['task_id']:
                return {"error": "未找到已推送的外部项目ID"}
                
            remote_project_id = row['task_id']
            
            # 2. 获取配置
            from app.core.config import config_handler
            api_base_url = config_handler.get('external_api', 'project_api_url', 'http://192.168.92.61:9003/api/external/projects').rstrip('/')
            export_url = f"{api_base_url}/{remote_project_id}/export"
            token = config_handler.get('external_api', 'admin_token', '')

            # 3. 发送请求
            async with httpx.AsyncClient(timeout=30.0) as client:
                headers = {
                    "Authorization": f"Bearer {token}",
                    "Content-Type": "application/json"
                }
                payload = {
                    "format": export_format,
                    "completed_only": completed_only
                }
                response = await client.post(export_url, json=payload, headers=headers)
                
                if response.status_code in (200, 201):
                    res_data = response.json()
                    logger.info(f"外部平台导出响应数据类型: {type(res_data)}, 键名: {list(res_data.keys()) if isinstance(res_data, dict) else 'None'}")
                    
                    # 1. 统一获取下载地址并回写 (兼容 download_url 和 file_url)
                    download_url = res_data.get('download_url') or res_data.get('file_url')
                    if isinstance(res_data, dict) and download_url:
                        try:
                            cursor_update = conn.cursor()
                            affected = cursor_update.execute(
                                "UPDATE t_task_management SET file_url = %s WHERE project_id = %s",
                                (download_url, project_id)
                            )
                            conn.commit()
                            cursor_update.close()
                            logger.info(f"导出时同步更新 file_url: {download_url}, 受影响行数: {affected}")
                        except Exception as ex:
                            logger.warning(f"导出同步回写 file_url 失败: {ex}")
                    
                    # 2. 同步回写 annotation_result 到 metadata
                    # 情况 A: 接口直接返回了任务列表数据
                    export_items = []
                    if isinstance(res_data, dict) and 'data' in res_data and isinstance(res_data['data'], list):
                        export_items = res_data['data']
                    
                    # 情况 B: 接口返回了文件链接，且格式为 JSON，尝试下载并解析以获取标注结果
                    elif isinstance(res_data, dict) and download_url and res_data.get('format') == 'json':
                        try:
                            # 补全 URL 协议 (如果外部平台返回的是相对路径)
                            full_download_url = download_url
                            if not download_url.startswith('http'):
                                from app.core.config import config_handler
                                # 尝试从配置获取 base_url，如果没有则从 api_url 中提取
                                base_url = config_handler.get('external_api', 'download_base_url', '')
                                if not base_url:
                                    # 兜底：从 project_api_url 中提取域名部分
                                    api_base = config_handler.get('external_api', 'project_api_url', 'http://192.168.92.61:9003')
                                    from urllib.parse import urlparse
                                    parsed = urlparse(api_base)
                                    base_url = f"{parsed.scheme}://{parsed.netloc}"
                                
                                full_download_url = f"{base_url.rstrip('/')}/{download_url.lstrip('/')}"
                                
                            logger.info(f"正在从导出链接获取详细标注数据以同步数据库: {full_download_url}")
                            # 注意：这里需要带上 token
                            file_res = await client.get(full_download_url, headers=headers)
                            if file_res.status_code == 200:
                                file_json = file_res.json()
                                # 外部平台导出的 JSON 结构通常是 { "data": [...] } 或直接是 [...]
                                if isinstance(file_json, dict) and 'data' in file_json:
                                    export_items = file_json['data']
                                elif isinstance(file_json, list):
                                    export_items = file_json
                                
                                if export_items:
                                    logger.info(f"成功获取导出项，共 {len(export_items)} 条。")
                                else:
                                    logger.warning("获取到的导出列表为空")
                        except Exception as ex:
                            logger.warning(f"从导出文件同步标注数据失败: {ex}")

                    if export_items:
                        updated_count = 0
                        try:
                            cursor_meta = conn.cursor()
                            for ext_item in export_items:
                                # 根据实际 JSON 结构提取数据
                                # 1. 提取 original_id
                                original_data = ext_item.get('original_data', {})
                                meta = original_data.get('metadata', {}) if original_data else ext_item.get('metadata', {})
                                # 增加兼容性：如果 metadata 里没有，尝试直接从 ext_item 找，或者从 external_id 提取
                                original_id = meta.get('original_id')
                                if not original_id:
                                    ext_id = ext_item.get('external_id', '')
                                    if '_' in ext_id: # 比如 "uuid_4" 这种结构，提取前面的 uuid
                                        original_id = ext_id.rsplit('_', 1)[0]
                                    else:
                                        original_id = ext_id
                                
                                # 2. 提取 annotation_result
                                annotations = ext_item.get('annotations', [])
                                if annotations and isinstance(annotations, list):
                                    annotation_res = annotations[0].get('result')
                                else:
                                    annotation_res = ext_item.get('annotation_result')
                                
                                if original_id and annotation_res is not None:
                                    # 注意：这里需要根据 business_id 或 metadata 里的 original_id 来匹配
                                    # 样本中心 t_task_management 表的 business_id 存的是原始数据的唯一标识
                                    cursor_meta.execute(
                                        "SELECT id, metadata FROM t_task_management WHERE business_id = %s OR id = %s", 
                                        (original_id, original_id)
                                    )
                                    row = cursor_meta.fetchone()
                                    if row:
                                        db_id = row['id']
                                        current_meta = json.loads(row['metadata']) if row['metadata'] else {}
                                        current_meta['annotation_result'] = annotation_res
                                        
                                        cursor_meta.execute(
                                            "UPDATE t_task_management SET metadata = %s WHERE id = %s",
                                            (json.dumps(current_meta, ensure_ascii=False), db_id)
                                        )
                                        updated_count += 1
                                    else:
                                        logger.debug(f"未在数据库中找到对应的任务: {original_id}")
                            
                            conn.commit()
                            cursor_meta.close()
                            logger.info(f"已从导出数据同步回写 {updated_count} 条任务的 annotation_result")
                        except Exception as ex:
                            logger.warning(f"同步回写 annotation_result 异常: {ex}")
                            
                    return res_data
                else:
                    logger.error(f"导出数据失败: {response.status_code} - {response.text}")
                    return {"error": f"外部平台返回错误 ({response.status_code})"}
                    
        except Exception as e:
            logger.exception(f"导出数据异常: {e}")
            return {"error": str(e)}
        finally:
            conn.close()

    async def send_to_external_platform(self, project_id: str) -> Tuple[bool, str]:
        """将项目数据推送至外部标注平台 (单表化)"""
        conn = get_db_connection()
        if not conn:
            return False, "数据库连接失败"
            
        try:
            # 1. 准备数据 (复用数据库连接)
            payload = await self.export_project_data(project_id=project_id, conn=conn)
            
            if not payload:
                return False, "项目导出失败，请检查项目ID是否正确"
                
            if not payload.get('data'):
                return False, f"项目数据为空 (查询到0条有效任务)，无法推送"
            
            # 2. 获取配置
            from app.core.config import config_handler
            api_base_url = config_handler.get('external_api', 'project_api_url', 'http://192.168.92.61:9003/api/external/projects').rstrip('/')
            api_url = f"{api_base_url}/init"
            token = config_handler.get('external_api', 'admin_token', '')

            # 3. 发送请求
            async with httpx.AsyncClient(timeout=60.0) as client:
                headers = {
                    "Authorization": f"Bearer {token}",
                    "Content-Type": "application/json"
                }
                logger.info(f"正在推送项目 {project_id} 至外部平台: {api_url}, 数据条数: {len(payload['data'])}")
                response = await client.post(api_url, json=payload, headers=headers)
                
                if response.status_code in (200, 201):
                    res_data = response.json()
                    logger.info(f"外部平台推送成功响应: {res_data}")
                    remote_project_id = res_data.get('project_id')
                    
                    # 获取下载地址并回写 (兼容多种返回格式)
                    download_url = res_data.get('download_url') or res_data.get('file_url')
                    if download_url:
                        # 4. 回写外部项目 ID 和下载地址 (复用当前连接)
                        cursor = conn.cursor()
                        # 先检查受影响行数
                        affected = cursor.execute(
                            "UPDATE t_task_management SET task_id = %s, file_url = %s WHERE project_id = %s", 
                            (remote_project_id, download_url, project_id)
                        )
                        conn.commit()
                        cursor.close()
                        logger.info(f"已回写 task_id: {remote_project_id} 和 file_url: {download_url}, 受影响行数: {affected}")
                    elif remote_project_id:
                        # 仅回写外部项目 ID
                        cursor = conn.cursor()
                        affected = cursor.execute(
                            "UPDATE t_task_management SET task_id = %s WHERE project_id = %s", 
                            (remote_project_id, project_id)
                        )
                        conn.commit()
                        cursor.close()
                        logger.info(f"仅回写 task_id: {remote_project_id}, 受影响行数: {affected}")
                    
                    return True, f"推送成功！外部项目ID: {remote_project_id or '未知'}"
                else:
                    error_msg = response.text
                    logger.error(f"推送失败: {response.status_code} - {error_msg}")
                    return False, f"外部平台返回错误 ({response.status_code})"
                    
        except Exception as e:
            logger.exception(f"推送至外部平台异常: {e}")
            return False, f"推送异常: {str(e)}"
        finally:
            conn.close()

task_service = TaskService()