I've encountered same issue recently, there's a good chance that this is an data skew issue, as certain output keys of mapper contains huge values, corresponding reducer might not able to aggregate the values correctly with out of memory error. it can be solved by define the key precisely to avoid skew.... Can you get the failed map/reduce process log via jobtracker?
-- View this message in context: http://lucene.472066.n3.nabble.com/PFP-Growth-tp1520639p4047629.html Sent from the Mahout User List mailing list archive at Nabble.com.
