Francesco, The mapreduce client will create a jar called job.jar and place it in HDFS in a staging directory. This is the jar that you specified to your job conf, or I believe that it tries to guess the jar based off of the Mapper class and the Reducer class but I am not sure of that. Once the job tracker has told a TaskTracker to run a given job the TaskTracker will download the jar, and then fork off a new JVM to execute the Mapper or Reducer. If you jar has dependencies then these usually have to be shipped with it as part of the cache archive interface.
--Bobby Evans On 5/27/11 9:16 AM, "Francesco De Luca" <[email protected]> wrote: Anyone knows the mechanism that hadoop use to load Map and Reduce class on the remote node where the JobTracker submit the tasks? In particular, how can hadoop retrieves the .class files ? Thanks
