[
https://issues.apache.org/jira/browse/HBASE-18640?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16134812#comment-16134812
]
Duo Zhang commented on HBASE-18640:
-----------------------------------
{{HBaseTestingUtility}} will start a mini cluster which means it must depend on
hbase-server, so I do not think it is possible that a user could use it without
depending on hbase-server.
And I think it is worth to introduce a hbase-mapreduce module that does not
depend on hbase-server. Usually user will only use the {{InputFormat}}s. And
for the tests, just put them in hbase-server. The most tests for hbase-client
are also placed in hbase-server. For pre commit usually we will introduce a
simple change in hbase-server to tell pre commit also run tests in hbase-server.
Thanks.
> Move mapreduce out of hbase-server into separate hbase-mapreduce moduel
> -----------------------------------------------------------------------
>
> Key: HBASE-18640
> URL: https://issues.apache.org/jira/browse/HBASE-18640
> Project: HBase
> Issue Type: Bug
> Reporter: Appy
> Assignee: Appy
> Attachments: HBASE-18640.master.001.patch,
> HBASE-18640.master.002.patch
>
>
> (Couldn't find another dedicated jira, so creating new one).
> Uploaded patch which is moving ~60 files to the new module. Few notes:
> - The classes remaining in hbase-server are the ones which are intensively
> coupled with visibility labels/wal/filesystem/hfile. These can not be
> migrated to new module until corresponding subcomponents are untangled out of
> hbase-server into their own separate modules.
> - Almost all mapreduce tests uses HBaseTestingUtil, so they can't be moved to
> hbase-mapreduce module. Given these dependency constraints, one way would be
> having a separate module for tests:
> hbase-mapreduce <---- hbase-server <------- hbase-mapreduce-tests
> Imo, this makes sense and looks fine.
> The only issue is - yetus' pre-commit. It won't run tests in
> hbase-mapreduce-tests module if something changed in just hbase-mapreduce.
> However, yetus' limitation shouldn't warrant against the idea.
> So i'd say that we should go that way, unless there are better suggestions.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
