[ 
https://issues.apache.org/jira/browse/MAPREDUCE-5403?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13716503#comment-13716503
 ] 

Jason Lowe commented on MAPREDUCE-5403:
---------------------------------------

bq. I don't think yarn.resourcemanager.address should be an environment 
variable. I think there is a conceptual difference between a client knowing how 
to contact a service and a client knowing about details of the service's 
internals. Do you disagree?

I'm just pointing out that locating the YARN jars is not the only thing clients 
may need to do to communicate with YARN.  e.g.: a non-Java YARN client still 
needs to locate the ResourceManager address somehow, and currently it can do 
this via parsing yarn-site.xml.  To me, yarn-site.xml is a site-specific config 
regardless of whether it's a YARN server or YARN client processing it.  For a 
site that doesn't host any YARN daemons (e.g.: gateway or launcher node), that 
becomes essentially a client-side-only config.

I can see where you're coming from, and maybe for the classpath we should do 
something different (e.g.: environment variable and/or clients should use "yarn 
classpath" to get the classpath instead).  I wanted to point out there are 
other interesting tidbits of information in yarn-site.xml besides the classpath 
that clients may want to access, and I'm wondering what criteria qualifies a 
client-consumed property to graduate to an environment variable or some other 
mechanism for determining the value besides parsing yarn-site.xml.
                
> Get rid of yarn.application.classpath
> -------------------------------------
>
>                 Key: MAPREDUCE-5403
>                 URL: https://issues.apache.org/jira/browse/MAPREDUCE-5403
>             Project: Hadoop Map/Reduce
>          Issue Type: Improvement
>          Components: client
>    Affects Versions: 2.0.5-alpha
>            Reporter: Sandy Ryza
>            Assignee: Sandy Ryza
>
> yarn.application.classpath is a confusing property because it is used by 
> MapReduce and not YARN, and MapReduce already has 
> mapreduce.application.classpath, which provides the same functionality.

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira

Reply via email to