[
https://issues.apache.org/jira/browse/HADOOP-11257?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14194856#comment-14194856
]
Allen Wittenauer commented on HADOOP-11257:
-------------------------------------------
We've had more than a few people confused that hadoop jar and yarn jar work
differently due to the fact that hadoop-env.sh and yarn-env.sh are treated as
separate by the two scripts. This, in turn, also increases the testing matrix
because one needs to verify both work as expected. The alternative here is to
eliminate yarn-env.sh, $YARN_OPTS, and related. Needless to say, that can't
happen until trunk which means we'll be stuck with the confusion for another
handful of years.
Even though trunk currently has yarn read the hadoop-env.sh file first,
YARN_OPTS takes precedence, thus still providing some conflicting handling. On
the plus side, removing hadoop jar gives a clear path to users as to how we
expect jobs to be launched.
bq. we should not be recommending its use for MR jobs and so forth
The irony is that the mapred command (luckily?) doesn't have a way to launch
jobs so mapred-env.sh is somewhat irrelevant. Well, unless you care that
distcp, pipes, and the sampler don't use $YARN_OPTS either...
> Deprecate 'hadoop jar'
> ----------------------
>
> Key: HADOOP-11257
> URL: https://issues.apache.org/jira/browse/HADOOP-11257
> Project: Hadoop Common
> Issue Type: Bug
> Reporter: Allen Wittenauer
>
> Given that 'hadoop jar' and 'yarn jar' work differently, we should mark
> 'hadoop jar' as deprecated in 2.7 and remove it in trunk.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
