I'm trying to run a TeraSort job to confirm that my cluster is set up correctly. The mappers perform fine, but in the reduce stage all the data is sent to a single node. My mapred.reduce.tasks parameter is set to an appropriate value greater than 1. I am launching multiple reducers, but only one of them is receiving input.
It looks like the TeraSort partition function is buggy, but there's no way that it would have a bug this obvious. I've looked for configuration errors on my part and found none. So now I'm asking if anyone else has seen this problem and can explain it. In the archives from February 27 of this year David Salle's post "TeraSort bug?<http://grokbase.com/t/hadoop.apache.org/common-user/2011/02/terasort-bug/27pzea46iowbfkbd4l5y566i4iv4>" describes what appears to be the same problem, but the only response I see is from David Salle the next day, apologizing and saying to ignore his previous post. Presumably he found some mistake on his end that he thought was trivial, but it doesn't look trivial to me.
