[
https://issues.apache.org/jira/browse/KNOX-1000?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16142588#comment-16142588
]
Jeffrey E Rodriguez commented on KNOX-1000:
--------------------------------------------
Thanks Larry and Rob.
"KNOX-1000" is really exciting.
As Larry said this JIRA would make use of the hdfs filesystem cli commands in
your cases where we can provide a webhdfs URL, and for other applications like
"notebooks" where we can access webhdfs through the Hadoop Filesystem API as an
alternative to webhdfs REST API through Knox, this would also allow Java apps
accessing the Hadoop Filesystem API.
Even though we have a webhdfs REST API which works well with Knox, many users
such as the one from our Knox user's mail list link and other user which have
manifested wanting to access the file system from a notebook (Jupyter or
Zeppelin), or an App using Hadoop File API.
I like the idea of a KIP for this issue which also involves changes to Hadoop
HDFS, investigation, and a solid use case.
Using distcp as a use case, with the ability to have Knox as the destination
and/or source of distcp, using the Hadoop Auth provider for the Kerberos case
in the case where Knox is the destination, if achievable, makes it a complete
end to end scenario.
> Hadoop Java Filesystem API support through Knox
> -----------------------------------------------
>
> Key: KNOX-1000
> URL: https://issues.apache.org/jira/browse/KNOX-1000
> Project: Apache Knox
> Issue Type: Bug
> Affects Versions: 0.9.1, 0.11.0, 0.12.0, 0.13.0
> Reporter: Jeffrey E Rodriguez
> Fix For: 0.14.0
>
>
> This Jira is to re-open a discussion on this issue from:
> https://mail-archives.apache.org/mod_mbox/knox-user/201511.mbox/%3CCACRbFyiy77fc5EHQ5c5oOQM=mz_upmevsoh7ermdreuoovn...@mail.gmail.com%3E
> 1. Hadoop FileSystem api needs to provide a "proxy" mechanism where we can
> add the user credentials. (same issue as any library or API try to access
> Knox). We may need to open a Hadoop Jira for this.
> 2. Knox Kerberos support (http client access to webhdfs) doesn't support
> "file access token". (whatever this means).
> 3. Use of Transfer-Encoding header in Knox.
> Ideally we should be able to use Knox URL through the Hadoop FileSystem api
> so we call from a Jupyter or Zeppelin notebook. Also it would help to
> programtically use the Hadoop Filesystem api through Knox.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
