The cracking furnace system is crucial for an olefin plant. Its operation needs to follow a predefined schedule to process various feeds continuously, meanwhile conducting a periodically decoking operation for each furnace when its performance apparently decreases. In practice, because the feed supply changes dynamically, the routine furnace scheduling is better performed in a dynamic and reactive way, through which the furnace operations can be smartly rescheduled with respect to any delivery of new coming feeds. Thus, the feeds from the new delivery and the leftover inventories can be timely, feasibly, and optimally allocated to different furnaces for processing to obtain the maximum average net profit per time. Facing this challenge, this paper develops a new MINLP-based reactive scheduling strategy, which can dynamically generate reschedules baged on the new feed deliveries, the leftover feeds, and current furnace operating conditions. It simultaneously addresses all the major scheduling issues of a cracking furnace system, such as semicontinuous operation, nonsimultaneous decoking, secondary ethane cracking, and seamless rescheduling. The efficacy of the study and its significant economic potential are demonstrated by a comprehensive case study.