Do you mean, without scanning all the files line by line? I know little about implementation of hadoop, but as a programmer, I can presume that it's not possible without a complete scan.
But I can suggest a work-around: - compute number of records manually before putting a file to HDFS. - Append the computed number to the filename. - modify InputReader, so that reader appends that number to the key of every map. Hope this helps, Rasit 2009/1/27 Andy Liu <andyliu1...@gmail.com> > Is there a way to programatically get the number of records in a MapFile > without doing a complete scan? > -- M. Raşit ÖZDAŞ