[
https://issues.apache.org/jira/browse/HDFS-2336?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13104921#comment-13104921
]
Daryn Sharp commented on HDFS-2336:
-----------------------------------
The description no longer accurately reflects the primary interest of this
jira: Hftp appears unable to support remote filesystems other than hdfs. Hftp
now unconditionally changes the token type to hdfs w/o regard to the orginal
token's kind before it was flipped to hftp. Would you please kindly change the
description back?
The fetchdt issue is orthogonal to this issue. It's only related in the sense
that it calls the same methods that hftp calls to get the tokens.
> fetchdt's --webservice option only works for hftp tokens
> --------------------------------------------------------
>
> Key: HDFS-2336
> URL: https://issues.apache.org/jira/browse/HDFS-2336
> Project: Hadoop HDFS
> Issue Type: Improvement
> Affects Versions: 0.20.205.0, 0.23.0
> Reporter: Daryn Sharp
> Fix For: 0.23.0
>
>
> The new token renewal implementation appears to introduce invalid assumptions
> regarding token kinds.
> Any token acquired over http is assumed to be hftp, so the kind is
> unconditionally changed to hftp. This precludes the acquisition of any other
> token types over http. This new limitation was added to a generic method in
> a public class. It should have been encapsulated in the hftp class, not the
> generic http token fetching methods.
> Furthermore, hftp will unconditionally change a hftp token's kind to hdfs. I
> believe this assumption means that hftp is now broken if the remote cluster's
> default filesystem is not hdfs.
--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira
