If I'm understanding his question correctly, he's saying that none of the reduce tasks will start until the map tasks for all _jobs_ have finished.
This could just be a problem of saturation on the nodes... were all of their resources being consumed by the remaining map tasks? If so, then Hadoop is simply completing them in the most efficient way it can think of If not, then its definitely a bug. You could manually set the priority level of certain jobs if you wanted to guarantee that they finished first, but perhaps MapReduce should automatically raise the priority (slightly) of jobs that are closer to finishing. Thanks, Stu -----Original Message----- From: Toby DiPasquale
