When invoking "hadoop jar /usr/local/hadoop/contrib/streaming/hadoop-0.20.2-streaming.jar -file <files>",
hadoop will package the files <files>, but it will forget about their timestamps.
After the files are unpacked in <tmp_dir>/mapred/local/taskTracker/jobcache/job_$job/jars, all files will have the timestamps of when the files were unpacked.
The problem is that this way meaningful information is lost.
For example in my case i ship some files along with my job, and I need to compare the age (mtime) of 2 files and rebuild one of them if it's too old,
but because of this hadoop behavior, my logic breaks.