The PageRank computation for each Web URL can be done using either MapReduce or Spark. In the presence of failures, what mechanism$/$property is used to ensure that MapReduce produces the same output as would have been produced by a non$-$faulting sequential execution of the entire program? How can MapReduce realize such a property?