Partitioned networks are a category of networks where nodes exhibit clustering properties, with numerous real-world counterparts such as routing networks and areas featuring distinct functional regions. In this paper, we examine security issues in partitioned networks, where the attacker attempts to target multiple regions/clusters and the defender plans patrol routes across different regions. We model this as a Stackelberg game, whereby the defender first announces her patrol routes, after which the attacker selects the optimal way to attack the regions. Our objective is to develop algorithms that can efficiently compute optimal patrol strategies for the defender. To this end, we propose a hierarchical patrol mechanism based on the characteristics of network structures and design an efficient algorithm to compute optimal patrol strategies. Experiment results demonstrate the superiority of our algorithm over other regular patrol strategies such as random walk and fixed schedules.