Dynamic Workflow Reconfigurations for Recovering from Faulty Cloud Services

The workflow paradigm is a well established approach to deal with application complexity by supporting the application development by composition of multiple activities. Furthermore workflows allow encapsulating parts of a problem inside an activity that can be reused in different workflow application scenarios for instance long-running experiments such as the ones involving data streaming. These workflows are characterized by multiple, eventually infinite, iterations processing datasets in multiple activities according to the workflow graph. Some of these activities can invoke Cloud services often unreliably or with limitations on quality of service provoking faults. After a fault the most common approach requires restarting of the entire workflow which can lead to a waste of execution time due to unnecessarily repeating of computations. This paper discuss how the AWARD (Autonomic Workflow Activities Reconfigurable and Dynamic) framework supports recovery from activity faults using dynamic reconfigurations. This is illustrated through an experimental scenario based on a long-running workflow where an activity fails when invoking a Cloud-hosted Web service with a variable level of availability. On detecting this, the AWARD framework allows the dynamic reconfiguration of the corresponding activity to access a new Web service, and avoiding restarting the complete workflow.