Add Robust Backup System (Manual, Automatic, Integrity, and Restore)

lpm_kernel/api/__init__.py

@@ -11,6 +11,7 @@
 from .domains.kernel2.routes_talk import talk_bp
 from .domains.user_llm_config.routes import user_llm_config_bp
 from .domains.space.space_routes import space_bp
+from .domains.backup.routes import backup_bp # Add import for backup blueprint
 
 def init_routes(app: Flask):
     """Initialize all route blueprints"""
@@ -27,6 +28,7 @@ def init_routes(app: Flask):
     app.register_blueprint(space_bp)
     app.register_blueprint(talk_bp)
     app.register_blueprint(user_llm_config_bp)
+    app.register_blueprint(backup_bp) # Register backup blueprint
     app.config['SEND_FILE_MAX_AGE_DEFAULT'] = 0
     app.config['TEMPLATES_AUTO_RELOAD'] = True
     # Disable response buffering

lpm_kernel/api/domains/backup/__init__.py

@@ -0,0 +1 @@
+from .routes import *

lpm_kernel/api/domains/backup/routes.py

@@ -0,0 +1,63 @@
+from flask import Blueprint, request, jsonify, current_app
+from lpm_kernel.backup.backup_service import BackupService
+from lpm_kernel.configs.config import Config # Import Config 
+
+backup_bp = Blueprint('backup', __name__, url_prefix='/api/backups')
+
+# Instantiate BackupService using the application config
+# Note: This assumes Config is a singleton or can be instantiated here.
+# A better approach might involve dependency injection or accessing config via current_app.
+config = Config.from_env() # Attempt to load config
+backup_service = BackupService(config=config) # Pass the loaded config
+
+@backup_bp.route('/', methods=['POST'])
+def create_backup_route():
+    """API endpoint to manually trigger a backup."""
+    try:
+        data = request.get_json() or {}
+        description = data.get('description')
+        result = backup_service.create_backup(description=description)
+        if result:
+            return jsonify(result), 201
+        else:
+            return jsonify({"status": "error", "message": "Failed to create backup"}), 500
+    except Exception as e:
+        current_app.logger.error(f"Backup creation failed: {str(e)}")
+        return jsonify({"status": "error", "message": f"Backup creation failed: {str(e)}"}), 500
+
+@backup_bp.route('/', methods=['GET'])
+def list_backups_route():
+    """API endpoint to list available backups."""
+    try:
+        backups = backup_service.list_backups()
+        return jsonify({"status": "success", "backups": backups}), 200
+    except Exception as e:
+        current_app.logger.error(f"Listing backups failed: {str(e)}")
+        return jsonify({"status": "error", "message": f"Listing backups failed: {str(e)}"}), 500
+
+@backup_bp.route('/<string:backup_id>/restore', methods=['POST'])
+def restore_backup_route(backup_id):
+    """API endpoint to restore data from a specific backup."""
+    try:
+        result = backup_service.restore_backup(backup_id)
+        if result.get('status') == 'error':
+            return jsonify(result), 404 if 'not found' in result.get('message', '').lower() else 500
+        return jsonify(result), 200
+    except Exception as e:
+        return jsonify({"status": "error", "message": f"Restore failed: {str(e)}"}), 500
+
+@backup_bp.route('/<string:backup_id>', methods=['DELETE'])
+def delete_backup_route(backup_id):
+    """API endpoint to delete a specific backup."""
+    try:
+        result = backup_service.delete_backup(backup_id)
+        if result.get('status') == 'error':
+            return jsonify(result), 404 if 'not found' in result.get('message', '').lower() else 500
+        return jsonify(result), 200
+    except Exception as e:
+        current_app.logger.error(f"Backup deletion failed: {str(e)}")
+        return jsonify({"status": "error", "message": f"Backup deletion failed: {str(e)}"}), 500
+
+# TODO: Add error handling
+# TODO: Add more robust error handling (e.g., specific error codes)
+# TODO: Consider dependency injection for BackupService

lpm_kernel/api/domains/loads/load_service.py

@@ -423,28 +423,27 @@ def _clean_data_directories() -> None:
 
     @staticmethod
     def _reset_training_progress() -> None:
-        """Reset training progress objects in memory"""
+        """重置内存中的训练进度对象"""
         try:
-            # Get all possible training progress file patterns
+            from lpm_kernel.api.domains.trainprocess.trainprocess_service import TrainProcessService
+            import os
+            # 获取所有可能的训练进度文件模式
             base_dir = os.getenv('LOCAL_BASE_DIR', '.')
             progress_dir = os.path.join(base_dir, 'data', 'progress')
             if os.path.exists(progress_dir):
                 for file in os.listdir(progress_dir):
                     if file.startswith('trainprocess_progress_'):
-                        # Extract model name
+                        # 提取模型名称
                         model_name = file.replace('trainprocess_progress_', '').replace('.json', '')
-                        # Create a new service instance for each model and reset progress
+                        # 为每个模型创建新的服务实例并重置进度
                         train_service = TrainProcessService(current_model_name=model_name)
                         train_service.progress.reset_progress()
                         logger.info(f"Reset training progress for model: {model_name}")
-
-            # Reset default training progress
+            # 重置默认训练进度
             default_train_service = TrainProcessService.get_instance()
             if default_train_service is not None:
                 default_train_service.progress.reset_progress()
-
             logger.info("Reset default training progress")
-
         except Exception as e:
             logger.error(f"Failed to reset training progress objects: {str(e)}")