https://community.cloudera.com/t5/Support-Questions/MapReduce-application-failed-with-OutOfMemoryError/m-p/55888#M36632 <P>is it fail at the Map phase or the reducer? MR jobs running with 6 GB for map and 12 G for reducer and fail, need a review.</P><P>&nbsp;</P><P>How much data the MR is running on? how many MR jobs do you have? 18K mappers also indicate that you have a lot of small files in your HDFS.</P><P>&nbsp;</P><P>your cluster consumption of ~300Vcores and 3T memory is meaning that each Vcore is running with an average of 10 G memory, how much nodes you have? from your stat seems your jobs are memory intenstive rather than vcores which this can reflects mainly on Spark jobs and you can manage this to get the optimal Vcores and memory for each Spark job.</P><P>&nbsp;</P><P>i suspect that you have a specific crazy job and prefer you if you are using Cloudera manager to search it in the application, or try to run the jobs under pools and then identify the problmatic ones.</P> Wed, 14 Jun 2017 22:20:55 GMT Fawze 2017-06-14T22:20:55Z