[ https://issues.apache.org/jira/browse/MAPREDUCE-6423?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14707436#comment-14707436 ]
Chris Douglas commented on MAPREDUCE-6423: ------------------------------------------ Thanks for taking a look at this. That the sampler only works on input data was always a weakness for jobs requiring their output be totally ordered. Could you generate a patch? The contribution wiki is [here|http://wiki.apache.org/hadoop/HowToContribute]. It might be easier for others to use if the Mapper was integrated with the InputSampler, but a separate tool is still an improvement. > MapOutput Sampler > ----------------- > > Key: MAPREDUCE-6423 > URL: https://issues.apache.org/jira/browse/MAPREDUCE-6423 > Project: Hadoop Map/Reduce > Issue Type: Improvement > Reporter: Ram Manohar Bheemana > Assignee: Ram Manohar Bheemana > Priority: Minor > Attachments: MapOutputSampler.java > > > Need a sampler based on the MapOutput Keys. Current InputSampler > implementation has a major drawback which is input and output of a mapper > should be same, generally this isn't the case. > approach: > 1. Create a Sampler which samples the data based on the input. > 2. Run a small map reduce in uber task mode using the original job mapper and > identity reducer to generate required MapOutputSample keys > 3. Optionally, we can input the input file to be sample. For example inputs > files A, B; we should be able to specify to use only file A for sampling. -- This message was sent by Atlassian JIRA (v6.3.4#6332)