A transaction in a Database Management System (DBMS) is a sequence of database operations that is executed as a single logical unit of work. Transactions ensure that data remains accurate and reliable even when multiple users or applications access the database simultaneously. The correctness of transactions is guaranteed through the ACID properties: Atomicity, Consistency, Isolation, and Durability. These properties are especially important in modern data science pipelines, where large datasets, feature stores, and machine learning model metadata are continuously updated. Atomicity ensures that a transaction is completed entirely or not executed at all. If any operation within a transaction fails, all previous operations are rolled back, leaving the database unchanged. Consider a feature store update in a machine learning pipeline. Suppose new customer transaction data is processed to generate features such as average purchase amount, purchase frequency, and customer lifetime value. These features must all be updated together. If the update of one feature fails due to a system error while others