question Re: Yarn Distributed Shell - MapReduce job in Archives of Support Questions (Read Only)

Yarn Distributed Shell - MapReduce job

sunile_manjee — Wed, 30 Mar 2016 11:29:54 GMT

I am not able to specify a nodel label when I submit my mapreduce job. Only yarn distributed shell job are allowed for node labels. How to run run the mapreduce job as yarn distributed shell job?

Re: Yarn Distributed Shell - MapReduce job

mjohansson — Wed, 30 Mar 2016 16:52:07 GMT

True Node Label support for Map Reduce will first be available in Hadoop Core 2.8 according to https://issues.apache.org/jira/browse/MAPREDUCE-6304.

A trick to use Node Labels with MR is to submit the MR job to a queue that has a default node label expression and the default node label will be applied to the MR job.

However using default node label expressions tends to constrain portions of the cluster, which at some point starts to become counter-productive for jobs -- such as MR jobs -- that benefit from the advantages offered by distributed parallel processing.

Also YARN distributed shell is an example YARN application on how to develop YARN jobs without Map Reduce.

Re: Yarn Distributed Shell - MapReduce job

sunile_manjee — Thu, 31 Mar 2016 01:25:24 GMT

@Mats Johansson Understood. However core of services like hive/pig use map reduce. Does that have the same constraints for node labeling? it seems node labeling is only applicable to storm/spark/kafka/hbase/etc. Services which do not use map reduce as its engine.