hive.mapred.local.mem should only be used in case of local mode job submissions -------------------------------------------------------------------------------
Key: HIVE-1520 URL: https://issues.apache.org/jira/browse/HIVE-1520 Project: Hadoop Hive Issue Type: Bug Reporter: Joydeep Sen Sarma Currently - whenever we submit a map-reduce job via a child jvm process, hive sets HADOOP_HEAPSIZE to hive.mapred.local.mem (thereby limiting the max heap memory of the child jvm). the assumption being that we are submitting a job for local mode execution and different memory limits apply for that. however - one can submit jobs via a child jvm for non local mode execution as well. This is useful, for example, if hive wants to submit jobs via different hadoop clients (for sending jobs to different hadoop clusters). in such case, we can use the 'hive.exec.submitviachild' and 'hadoop.bin.path' to dispatch job via an alternate hadoop client install point. however in such case, we don't need to set HADOOP_HEAPSIZE. all we are using the child jvm is to run the small bit of hive code that submits the job (and not for local mode execution). in this case - we shouldn't be setting the child jvm's memory limit and should leave it to what the parent's value is. -- This message is automatically generated by JIRA. - You can reply to this email to add a comment to the issue online.