Because of the unreliable and dynamic characteristics, cloud service failures are inevitable. It has an adverse effect on task execution and scheduling. To improve cloud service reliability, we first analyze the fault recovery mechanism, and then, cloud service failures considered in this paper are classified into unrecoverable failures and recoverable failures. By extending the traditional dynamic level scheduling (DLS) algorithm, a novel scheduling algorithm based on fault recovery mechanism named fault recovery-based DLS algorithm is proposed to reduce the failure probability of task scheduling. The experimental results confirm that fault recovery mechanism can meet the reliability requirements of cloud computing infrastructures and the proposed algorithm can effectively ensure trustworthy execution of tasks. Copyright © 2014 John Wiley & Sons, Ltd.