In real-time systems it is of paramount importance that time constraints of tasks are enforced. A tremendous amount of research has been carried out on scheduling problems associated with such systems, primarily focusing on priority assignment policies in non-overloaded systems. While static real-time systems, by definition, do not suffer from overloads, they offer limited or no flexibility and ability to adapt to new situations, often making them a poor choice for complex real-time applications. While dynamic real-time systems often meet these demands, they are prone to transient overloads. In this paper we introduce a novel scheduling architecture with a new algorithm for dynamically resolving transient overloads, that is executed when a new transaction cannot be admitted to the system due to scarce resources. The resolver algorithm generates a cost effective overload resolution plan which, in order to admit the new transaction, finds the required time by de-allocating time among the previously admitted but not yet completed transactions. Considering the cost efficiency of executing the plan and the importance of the new transaction, a decision is made whether to execute the plan and admit the new transaction, or to reject it. the new transaction. We consider a multi-class transaction workload consisting of hard critical and firm transactions, where critical transactions have contingency transactions that can be invoked during overloads. We present a thorough performance analysis showing to what degree the overload resolver enforces predictability and ensures the timeliness of critical transactions when handling extreme overload scenarios in real-time database systems.