[ 
https://issues.apache.org/jira/browse/YARN-4985?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15860136#comment-15860136
 ] 

Sangjin Lee commented on YARN-4985:
-----------------------------------

Thanks for refreshing my memory on this Haibo. Yes, I recall that was the main 
issue. To tease out client and server correctly, we'd need a separate (common) 
module both client and server depends on. The only way we could avoid it is if 
all the fine-grained downstream dependencies can be isolated and moved into the 
hbase-server module. Do you know if it is feasible?

If that is not feasible, maybe this refactoring would be more problematic than 
beneficial. What would we lose if we didn't do this refactoring? Also, if we 
didn't do this refactoring and kept thins as is (just timelineservice-hbase), 
do we still have the dependency issues in terms of installing the coprocessor 
(we need to follow all transitive dependencies between classes)? If so, that 
needs to be addressed no matter what.

> Refactor the coprocessor code & other definition classes into independent 
> packages
> ----------------------------------------------------------------------------------
>
>                 Key: YARN-4985
>                 URL: https://issues.apache.org/jira/browse/YARN-4985
>             Project: Hadoop YARN
>          Issue Type: Sub-task
>          Components: timelineserver
>            Reporter: Vrushali C
>            Assignee: Haibo Chen
>              Labels: YARN-5355
>         Attachments: YARN-4985-YARN-5355.prelim.patch
>
>
> As part of the coprocessor deployment, we have realized that it will be much 
> cleaner to have the coprocessor code sit in a package which does not depend 
> on hadoop-yarn-server classes. It only needs hbase and other util classes.
> These util classes and tag definition related classes can be refactored into 
> their own independent "definition" class package so that making changes to 
> coprocessor code, upgrading hbase, deploying hbase on a different hadoop 
> version cluster etc all becomes operationally much easier and less error 
> prone to having different library jars etc.



--
This message was sent by Atlassian JIRA
(v6.3.15#6346)

---------------------------------------------------------------------
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org

Reply via email to