Designing An ML-Driven Framework for Automatic Generation of Rollback Statements for Database Commands

Abstract

System administrators need automated rollback feature support to protect data quality while handling database system errors. Human-based database workflows and script rules do not work well for building error-free solutions with modern application complexity. Our research develops an ML system with Transformer Seq2Seq architecture to automatically generate rollback statements for different SQL operations. Researchers tested the proposed system using sets of 1,000, 10,000 and 100,000 SQL queries to evaluate its performance. The model correctly generated rollback statements for 98.2% of all SQL operations while achieving 99% success for INSERT queries and 97.4% and 98.2% success rates for UPDATE and DELETE queries respectively. Our system reacted quickly within 60 milliseconds when processing large amounts of data which shows its real-time performance. Our system used resources moderately while working with data sets because it reached 56% CPU use and needed 710MB memory for the largest dataset. The proposed framework surpassed traditional databases because it recorded better results with 5.7% improved accuracy while reducing resource needs by 10% and rollback times by 40%. These test results validate that the framework produces reliable rollback statements for high-quality database performance.