I have seen the map input bytes counter go negative temporarily on hadoop 1.x at the beginning of a job. It then corrects itself later in the job and seems to be accurate. Any ideas?

http://terrier.org/docs/v2.2.1/hadoop_indexing.html

I also saw this behavior in a job output listed at the above site:

INFO JobClient - Map output bytes=7724117
INFO JobClient - Map input bytes=-4167449
INFO JobClient - Combine input records=0

Reply via email to