[
https://issues.apache.org/jira/browse/PHOENIX-6053?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17769062#comment-17769062
]
Istvan Toth commented on PHOENIX-6053:
--------------------------------------
Some general thoughts and comments, [~ameszaros] :
Do we want to add the new shaded artifacts (client-byo) in the same ticket, or
do want to split them into a different one ?
Can we figure out a better name than phoenix-coprocessors ? I only ever meant
it as a placeholder. Maybe phoenix-regionserver ?
Regarding the mapreduce support:
Keeping the Phoenix tools (like bulkload and indextool) in the
phoenix-coprocessors package is fine IMO. However, it would be really great if
we could separate out the generic mapreduce code so that it doesn't depend on
the hbase server classes. I am not sure if it is even possible, and how much
refactoring it would require, but it's worth looking into.
> Use shaded hadoop-client and hbase-shaded-client in phoenix-client
> ------------------------------------------------------------------
>
> Key: PHOENIX-6053
> URL: https://issues.apache.org/jira/browse/PHOENIX-6053
> Project: Phoenix
> Issue Type: Improvement
> Components: core
> Affects Versions: 5.1.0
> Reporter: Istvan Toth
> Assignee: Istvan Toth
> Priority: Major
> Attachments: Phoenix_separation.pdf
>
>
> The phoenix client is built with the unshaded Hadoop and HBase dependencies,
> and then relocates some (a lot) of those to declutter the client classpath.
> Try to use hadoop-client and hbase-shaded-client-byo-hadoop instead. (And
> keep relocating our direct dependecies)
> While this will likely further bloat the thick client JAR, it shouldÂ
> * clean up the classpath further
> * Would decouple most of the dependencies that Phoenix shares with
> Hbase/Hadoop from them.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
