[
https://issues.apache.org/jira/browse/HDFS-9179?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15188302#comment-15188302
]
Daniel Templeton commented on HDFS-9179:
----------------------------------------
The justification was that the NN service URI configuration is more complicated
than makes sense. First we look for a servicerpc-address. Failing at that, we
look for an rpc-address. Failing at that we fall back to the defaultFS. And
things just get worse with HA.
I also find it awkward that we set the server's address according to a
client-side configuration setting. This param gets used in the scenario where
the server can't figure out who it's supposed to be (no proper
servicerpc-address or rpc-address), so it falls back to whom it thinks the
client thinks it should be. That seems off to me.
I grant that those are fairly weak reasons to do something that's going to
cause lots of things to break, but it would make the configuration logic
cleaner. I did mark it as an incompatible change. :)
[~atm], anything you want to add?
> fs.defaultFS should not be used on the server side
> --------------------------------------------------
>
> Key: HDFS-9179
> URL: https://issues.apache.org/jira/browse/HDFS-9179
> Project: Hadoop HDFS
> Issue Type: Improvement
> Affects Versions: 2.7.1
> Reporter: Daniel Templeton
> Assignee: Daniel Templeton
>
> Currently the namenode will bind to the address given by defaultFS if no
> rpc-address is given. That behavior is an evolutionary artifact and should
> be removed. Instead, the rpc-address should be a required setting for the
> server side configuration.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
