[
https://issues.apache.org/jira/browse/HDFS-10208?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15237483#comment-15237483
]
Sangjin Lee commented on HDFS-10208:
------------------------------------
Thanks [~mingma] for the patch!
The default value for the new config is false, meaning we want to disable the
feature for HDFS-9579 by default. Is that desirable? I would think that we
should be able to enable it by default unless people want to turn it off. I'd
like to hear your thought on the default...
I see that the lookup map in {{ClientContext}} was removed as part of this
patch. However, if {{getNetworkDistance()}} is called repeatedly on the same
datanodes, there is definitely a benefit in caching the result of the
computation. Is the concern for removing the {{nodeToDistance}} map about its
size?
> Addendum for HDFS-9579: to handle the case when client machine can't resolve
> network path
> -----------------------------------------------------------------------------------------
>
> Key: HDFS-10208
> URL: https://issues.apache.org/jira/browse/HDFS-10208
> Project: Hadoop HDFS
> Issue Type: Bug
> Reporter: Ming Ma
> Assignee: Ming Ma
> Attachments: HDFS-10208-2.patch, HDFS-10208-3.patch,
> HDFS-10208-4.patch, HDFS-10208.patch
>
>
> If DFSClient runs on a machine that can't resolve network path,
> {{DNSToSwitchMapping}} will return {{DEFAULT_RACK}}. In addition, if somehow
> {{dnsToSwitchMapping.resolve}} returns null, that will cause exception when
> it tries to create {{clientNode}}. In either case, there is no need to create
> {{clientNode}} and we should treat its network distance with any datanode as
> Integer.MAX_VALUE.
> {noformat}
> clientNode = new NodeBase(clientHostName,
> dnsToSwitchMapping.resolve(nodes).get(0));
> {noformat}
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
