It would be nice to have something similar to hadoop's counters, to keep track of how many times various things happen in the slaves. It seems like it would be most appropriate to tie the counters to the job object that goes into the run(self, job), but the map and reduce functions need access to them, so maybe they should go in mrs.MapReduce, so the map and reduce functions can get to them at self.counters, or something?

And it looks like I submitted this on my wife's account, so I'm commenting to put my email address in the notification list.