[ 
https://issues.apache.org/jira/browse/HADOOP-3479?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12604745#action_12604745
 ] 

Hemanth Yamijala commented on HADOOP-3479:
------------------------------------------

One of the alternative designs that we considered is to see if the 
Configuration class can be reused.

The basic difference, as explained in comments above, is in the nested nature 
of the configuration - sets of properties are grouped under a queue 
configuration element, for e.g. We could possibly enhance Configuration to 
support parsing and exposing such nested configuration also, without changing 
the basic API. One approach could be to support nesting that occurs in the 
following manner:
{code:xml}
<configuration>
  <level1 key="value">
    <level2>
      <property>
        <!-- usual key-value elements here -->
      </property>
    </level2>
  </level1>
</configuration>
{code}
We could then build a Map of an XPath type expression to a Properties object, 
where the XPatch expression takes us till the parent of a list of *property* 
elements. The current configuration is then basically a Properties object 
mapped for the element */configuration*. Queue related properties could be 
looked up like /configuration/queues/[EMAIL PROTECTED]"default"]. We could even 
modify current configuration to be stored as a hierarchy:
{code:xml}
<configuration>
  <hadoop>
    <mapred>
       <!-- mapred configuration -->
    </mapred>
    <hdfs>
       <!-- hdfs configuration -->
    </hdfs>
    <resource-manager>
       <!-- resource-manager configuration -->
    </resource-manager>
  </hadoop>
</configuration>
{code}
... and so on.

However, given this is the first attempt for building resource manager 
functionality, it seems prudent to keep things simple, if at the cost of 
duplicating a little code between Configuration and ResourceManagerConf, and 
refactor if things are working out fine. Also, Configuration is a religious 
topic and it would take a while to converge of how and what to do. *smile*. 
Only if people feel  very strongly about this approach, should we consider it 
for Hadoop 0.19 though.

> Implement configuration items useful for Hadoop resource manager (v1)
> ---------------------------------------------------------------------
>
>                 Key: HADOOP-3479
>                 URL: https://issues.apache.org/jira/browse/HADOOP-3479
>             Project: Hadoop Core
>          Issue Type: New Feature
>            Reporter: Hemanth Yamijala
>            Assignee: Hemanth Yamijala
>         Attachments: 3479.patch
>
>
> HADOOP-3421 lists requirements for a new resource manager for Hadoop. 
> Implementation for these will require support for new configuration items in 
> Hadoop. This JIRA is to define such configuration, and track it's 
> implementation.

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.

Reply via email to