[
https://issues.apache.org/jira/browse/BIGTOP-1339?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14056882#comment-14056882
]
Mark Grover commented on BIGTOP-1339:
-------------------------------------
Thanks for the explanation [~rguo], however, in my opinion, it's a good thing
that /etc/<component>/conf is a symlink and not a package delivered directory.
The reason is we want to enable our users to change configurations easily.
Almost always, default configuration is not good enough for clusters of more
than one node. For example, hadoop namenode name needs to be set other than
something trivial like {{localhost:80020}}.
If the users have the change configurations, one of the easiest ways to do so,
would be to create a new configuration directory like
/etc/<component>/conf.personal.mr2 and then symlink /etc/<component>/conf ->
/etc/component/conf.personal.mr2 via alternatives. They could have a similar
configuration for mr1, say, or anything else for that matter, and all it would
take to make the other configuration default would be an update-alternative
command.
So, while I understand your intent, I think having separate convenience
packages like pseudo-conf for those requiring quick "out of the box" experience
while delivering pretty basic and generic default configuration (or none, for
that matter) via base packages is still a valid approach.
Does that make sense?
> hadoop pseudo configuration can be default configuration
> --------------------------------------------------------
>
> Key: BIGTOP-1339
> URL: https://issues.apache.org/jira/browse/BIGTOP-1339
> Project: Bigtop
> Issue Type: Improvement
> Components: RPM
> Reporter: Guo Ruijing
>
> in existing implementation, hadoop-conf-pseudo-2.2.0.x86_64.rpm is to provide
> hadoop pseudo configuration. we may take hadoop pseudo configuration as
> hadoop default configuration and move hadoop pseudo configuration to
> core/hdfs/yarn rpm.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
