hadoop-common-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Steve Loughran (Commented) (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HADOOP-7030) new topology mapping implementations
Date Wed, 22 Feb 2012 17:37:49 GMT

    [ https://issues.apache.org/jira/browse/HADOOP-7030?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13213779#comment-13213779
] 

Steve Loughran commented on HADOOP-7030:
----------------------------------------

With the changes I've been doing to the topology classes (with an optional abstract base class),
these classes would be best if 
{{RefreshableDNSToSwitchMapping}} was a subclass of the new {{AbstractDNSToSwitchMapping}}.
 

On barrier to any dynamic/refreshable topology class is that the values get cached. In particular

{{org.apache.hadoop.yarn.util.RackResolver}} checks to see if the mapping is an instance of
{{CachedDNSToSwitchMapping}} -and if not, wraps it in one. the DNManager doesn't do anything
so dramatic - it does do an instance check, but only to trigger a pre-emptive lookup of all
known hosts.

Unless we add flags to {{AbstractDNSToSwitchMapping}} to provide more info about the resolver
(e.g. caching, dynamic, etc), it's probably safest to subclass {{CachedDNSToSwitchMapping}}
 and overwrite its resolve method (and anything else that is useful). 


                
> new topology mapping implementations
> ------------------------------------
>
>                 Key: HADOOP-7030
>                 URL: https://issues.apache.org/jira/browse/HADOOP-7030
>             Project: Hadoop Common
>          Issue Type: New Feature
>    Affects Versions: 0.20.1, 0.20.2, 0.21.0
>            Reporter: Patrick Angeles
>            Assignee: Patrick Angeles
>         Attachments: HADOOP-7030-2.patch, HADOOP-7030.patch, topology.patch
>
>
> The default ScriptBasedMapping implementation of DNSToSwitchMapping for determining cluster
topology has some drawbacks. Principally, it forks to an OS-specific script.
> This issue proposes two new Java implementations of DNSToSwitchMapping. TableMapping
reads a two column text file that maps an IP or hostname to a rack ID. Ip4RangeMapping reads
a three column text file where each line represents a start and end IP range plus a rack ID.

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators: https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira

        

Mime
View raw message