[
https://issues.apache.org/jira/browse/HIVE-10100?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15351929#comment-15351929
]
Prasanth Jayachandran commented on HIVE-10100:
----------------------------------------------
bq. Why is this snippet used in some files, but not others ?
Some scripts explicitly modifies the variables after hive script sets it up to
append service specific info. Example: beeline seems to be doing some ordering
of classpath and metastore appends its own java opts which might not be
required otherwise.
I can't find YARN_CLASSPATH in yarn script. I think after some version
YARN_CLASSPATH is renamed to YARN_USER_CLASSPATH. So I added HADOOP_CLASSPATH
to YARN_USER_CLASSPATH in the next patch.
Yes. Hive script is slow for 2 reasons
1) Automatic initialization of logging (when trying to run hadoop or hbase
commands entire classpath is searched to look for log4j2.properties file)
2) hbase mapredcp takes time
There is some info here HIVE-12497. Regardless we should bring down the time.
> Warning "yarn jar" instead of "hadoop jar" in hadoop 2.7.0
> ----------------------------------------------------------
>
> Key: HIVE-10100
> URL: https://issues.apache.org/jira/browse/HIVE-10100
> Project: Hive
> Issue Type: Bug
> Affects Versions: 1.3.0, 2.2.0
> Reporter: Gunther Hagleitner
> Assignee: Prasanth Jayachandran
> Priority: Critical
> Attachments: HIVE-10100.1.patch
>
>
> HADOOP-11257 adds a warning to stdout
> {noformat}
> WARNING: Use "yarn jar" to launch YARN applications.
> {noformat}
> which will cause issues if untreated with folks that programatically parse
> stdout for query results (i.e.: CLI, silent mode, etc).
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
