Recovering from Cloud Application Deployment Failures Through Re-execution

In this paper we study the problem of automated cloud application deployment and configuration. Transient failures are commonly found in current cloud infrastructures, attributed to the complexity of the software and hardware stacks utilized. These errors affect cloud application deployment, forcing the users to manually check and intervene in the deployment process. To address this challenge, we propose a simple yet powerful deployment methodology with error recovery features that bases its functionality on identifying the script dependencies and re-executing the appropriate configuration scripts. To guarantee the idempotent script execution, we adopt a filesystem snapshot mechanism that enables our approach to revert to a healthy filesystem state in case of failed script executions. Our experimental analysis indicates that our approach can resolve any transient deployment failure appearing during the deployment phase, even in highly unpredictable cloud environments.