As a cutting-edge distributed intelligent system that seamlessly combines and collaborates between the information world and the physical world, cyber-physical systems (CPSs) possess vast practical applications. Given that CPSs are characterized by constant interaction with their surrounding environment, fluctuations in resource performance triggered by external changes present significant challenges for achieving optimal application scheduling. To harness the potential of heterogeneous and capacity-limited computing resources in CPSs, applications must be precisely modeled and allocated to resources to enhance their execution efficiency, even under uncertain conditions. Against this backdrop, this paper focuses on the online deadline-constrained scheduling problem of parallel applications and proposes a novel uncertainty-aware scheduling algorithm to schedule the applications in the changing environment. Firstly, parallel applications are expertly modeled as directed acyclic graphs (DAGs), and a novel ranking methodology is employed to generate task priorities based on their urgency of completion. Subsequently, the optimal resources are meticulously chosen for each task to expedite execution, supplemented by a discarding mechanism to amplify the likelihood of successful application. Finally, the priority of tasks and resource status is updated based on environmental feedback to expertly tackle uncertainties. Experiments with randomly-generated applications conclusively demonstrate that the proposed approach boasts superior scheduling performance compared to similar algorithms.