A Workflow-Based Failure Recovery in Web Services Composition

Abstract

In previous researches in the field of supporting reliability and fault tolerance in web service composition, only low level programming constructs such as exception handling (for example in WSBPEL) were considered. However we believe that the reliability and fault tolerance for composite services must be handled at a higher level of abstraction, i.e. at the workflow level. Therefore a language and technology independent method for fault-tolerant composition of web services is needed. To do this, a fault tolerant workflow is built in which the execution order of the services is determined such that upon a service failure a recovery process with the lowest cost is started. The cost of a service failure includes the cost of failed service and the total costs of roll-baking the previously executed services which are dependent on the failed service. In this article a FSP language is applied to formally specify the workflow

Dynamic recovery decision during composite web services execution

“…Centralized approaches consider a coordinator managing the whole execution process [17,25]. In distributed approaches, the execution process proceeds with collaboration of several participants without a central coordinator [1,4]. On the other hand, the execution control could be attached to the WS [9] or independent of its implementation [7].…”

“…Some execution engines are capable of managing failures during the execution. Ones are based on exception handling [14,18], others are based on transactional properties [4,6,8], others use a combination of both approaches [9], while some works base fault tolerance on replication techniques [1,27].…”

Extended Fault Taxonomy of SOA-Based Systems

“…Service composition process may fail at intermediate state due to an incomplete description of goal service requirements or due to the fact that the user is unaware or uninformed of the functionality provided by the existing participant services [44], [45]. For example: if the "Online Booking" service is lacking all other committed services such as "Item Availability Check" the service should be roll-backed.…”

FaCETa: Backward and Forward Recovery for Execution of Transactional Composite WS

“…We can distinct two kinds of distributed coordination approach. In the first one, nodes interact directly based on a peer-to-peer application architecture and collaborate, in order to execute a CWS with every node executing a part of it [2,5,16,18,28]. In the second one, they use a shared space for coordination [4,12,19].…”

“…Only results of the first replica finished are accepted, other executions are halted or ignored. As our work, in [5] a rollback workflow is automatically created considering the service dependencies. Those frameworks support users and developers to construct CWS based on WS-BPEL technologies, then they are not transparent to users and developers.…”