Recovery From Node Failure in Distributed Query Processing

While distributed query processing has many advantages, the use of many independent, physically widespread computers almost universally leads to reliability issues. Several techniques have been developed to provide redundancy and the ability to recover from node failure during query processing. In this survey, we examine three techniques--upstream backup, active standby, and passive standby--that have been used in both distributed stream data processing and the distributed processing of static data. We also compare several recent systems that use these techniques, and explore which recovery techniques work well under various conditions.