Fault-tolerance has long been a feature of database systems, with transactions supporting the structuring of applications so as to ensure continuation of updating applications in spite of machine failures. For read-only queries the perceived wisdom has been that support for fault-tolerance is too expensive to be worthwhile. Distributed query processing is coming to be seen as a promising way of implementing applications that combine structured data and analysis operations in dynamic distributed settings such as computational grids. Such a query may be long-running and having to redo the whole query after a failure may cause problems (e.g. if the result may trigger business or safety critical activities). This work describes and evaluates a new scheme for adding fault-tolerance to distributed query processing through a rollback-recovery mechanism. The high level expression of user requests in a physical algebra offers opportunities for tuning the fault-tolerance provision so as to reduce the cost, and give better performance than employment of generic fault-tolerance mechanisms at the lowest level of query processing. This paper outlines how the publicly-available OGSA-DQP computational grid-based distributed query processing system can be modified to include support for fault-tolerance and presents a performance evaluation which includes measurements of the cost of both protocol overheads and rollback-recovery, for a set of example distributed queries.