Unfortunately, I havent had the time to think about a framework for debugging.
HAdoop has something related to debug scripts that can be executed on
the offending 'split'.
I could (in future) look towards using that. It might be of help.
but yes, currently, debugging can be, to say the least, adventurous.
My priority now is to add a custom multiple file output format - i.e.
based on the key value write to different
output directories.
Regards
J