[
https://issues.apache.org/jira/browse/HADOOP-13949?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Steve Loughran resolved HADOOP-13949.
-------------------------------------
Resolution: Won't Fix
Closing as a WONTFIX, as this has been discussed in the past, and rejected then.
Why? Kerberos
Kerberos likes hostnames, not IP addresses, and likes to do reverse DNS too.
If you want security, you need Kerberos, so you need hostnames. And if you
aren't running Kerberos, your Hadoop cluster and all data on it is available to
everyone on the network. Which is generally considered a bad thing.
Generally, unless you run a DNS service, you can edit /etc/hosts for sharing
the host tables.
* If you are creating virtual clusters with short livespans, your config
tooling (chef, puppet, ansible) can do this.
* if you have physical clusters, embrace DNS
We do recognise that sometimes it would seem useful to use IPAddrs, but its not
something anyone should be doing in production. By not supporting it, we avoid
people starting off with deployments which work during development/setup, but
which can't go live.
Sorry
> Hadoop could provide IP instead of hostname to represent the node, which can
> reduce the DNS hostname resolution consumption
> ---------------------------------------------------------------------------------------------------------------------------
>
> Key: HADOOP-13949
> URL: https://issues.apache.org/jira/browse/HADOOP-13949
> Project: Hadoop Common
> Issue Type: Improvement
> Components: common
> Reporter: DENG FEI
>
> Normally,Hadoop use hostname represent the node,such as
> datanode,nodemanager,will lookup nameserver at rack resolve or create rpc
> connnect .
> May be could provide a chance use ip walk around dns resolution.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
