Object Management for Persistence and Recoverability

Abstract

As distribution becomes commonplace, there is a growing requirement for applications that behave reliably when node or network failures occur. To support reliability, operations on the components of a distributed application may be declared to occur within the scope of an atomic action. This thesis describes how atomic actions may be supported in an environment consisting of applications that operate on objects. To support the failure atomicity and permanence of effect properties of an atomic action, the objects accessed within the scope of an atomic action must be recoverable and persistent. This thesis describes how these properties may be added to the class of an object. The approach adopted is to provide a class that implements recovery and persistence mechanisms, and derive new classes from this base class. By refining inherited operations so that recovery and persistence is specific to that class, recoverable and persistent objects may be easily produced. This thesis also describes how an atomic action may be implemented as a class, so that instances of the class are atomic actions which manage the recoverable and persistent objects. Multiple instance declarations produce nested atomic actions, and the atomic action class also inherits persistence so that shortterm commit information may be saved in an object store which is used to maintain the passive state of persistent objects. Since the mechanisms and classes that support recovery, persistence, and atomic actions are constructed using the feature of an object-oriented language, they may be implemented in environments that provide suitable support for objects and object-oriented programming languages.