Automatic Undo for Cloud Management via AI Planning

The facility to rollback a collection of changes, i.e., reverting to a previous acceptable state, a checkpoint , is widely recognised as valuable support for dependability [3, 5, 10]. This paper considers the particular needs of users of cloud computing resources, wishing to manage the resources. Cloud computing provides infrastructure programmatically managed through a fixed set of simple system administration commands. For instance, creating and configuring a virtualized Web server on Amazon Web services (AWS) can be done with a few calls to operations that are offered through the AWS management API. This improves the efficiency of system operations; but having simple powerful system operations may increase the chances of human-induced faults, which play a large role in overall dependability [24, 25]. Catastrophic errors, like deleting a disk volume in a production environment, can happen easily with a few wrong API calls.