[
https://issues.apache.org/jira/browse/TEZ-1127?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14044849#comment-14044849
]
Rohini Palaniswamy commented on TEZ-1127:
-----------------------------------------
[~bikassaha],
We still want to do
dagAMConf.setIfUnset(TezConfiguration.TEZ_AM_LAUNCH_CMD_OPTS,
org.apache.tez.mapreduce.hadoop.MRHelpers.getMRAMJavaOpts(tezConf)); in Pig.
Many of our users have modifed AM command opts raising the AM memory to 2-2.5G
because of huge number of tasks. And our idea of deployment is to just have
user specifiy -x tez without having to change anything else. So we will just
change TEZ_AM_JAVA_OPTS -> TEZ_AM_LAUNCH_CMD_OPTS in Pig when this change is
checked in. We are compiling against the snapshot anyways and API
incompatibility is not a problem at least for now (May not be so in another 2
months).
> Add TEZ_TASK_JAVA_OPTS and TEZ_ENV configs to specify values from config
> ------------------------------------------------------------------------
>
> Key: TEZ-1127
> URL: https://issues.apache.org/jira/browse/TEZ-1127
> Project: Apache Tez
> Issue Type: Sub-task
> Affects Versions: 0.4.0
> Reporter: Pala M Muthaia
> Assignee: Bikas Saha
> Fix For: 0.5.0
>
> Attachments: PIG-TEZ-1127.patch, TEZ-1127.1.patch, TEZ-1127.2.patch
>
>
> There is no global setting available to specify JVM parameters for tez tasks
> (e.g: specifying native library location with java.library.path). Yarn allows
> specifying application.classpath, but not java opts. The only option right
> now is to set mapreduce.map.java.opts in the tez job config (or
> hive.tez.java.opts for Hive on Tez).
> Suggestion from Bikas, while discussing this in tez-user group, is tez should
> have "TEZ_TASK_JAVA_OPTS so that you don’t have to specify this via mapreduce
> configuration parameters"
--
This message was sent by Atlassian JIRA
(v6.2#6252)
