[jira] Updated: (HADOOP-6545) Cached FileSystem objects can lead to wrong token being used in setting up connections

["Devaraj Das (JIRA)" <ji...@apache.org>]

     [ https://issues.apache.org/jira/browse/HADOOP-6545?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Devaraj Das updated HADOOP-6545:
--------------------------------

    Attachment: 6545-1.patch

Looks like the correct solution is to have the UGI as part of the key in the FileSystem cache. Attaching a patch that has this change.

> Cached FileSystem objects can lead to wrong token being used in setting up connections
> --------------------------------------------------------------------------------------
>
>                 Key: HADOOP-6545
>                 URL: https://issues.apache.org/jira/browse/HADOOP-6545
>             Project: Hadoop Common
>          Issue Type: Bug
>          Components: security
>    Affects Versions: 0.22.0
>            Reporter: Devaraj Das
>             Fix For: 0.22.0
>
>         Attachments: 6545-1.patch
>
>
> The FileSystem class caches the filesystem objects that it creates for users. For some cases, e.g., if the filesystem object is actually a DistributedFileSystem, it also has an associated RPC client and hence an UGI for the respective user. This could lead to issues to do with using the right credentials when connecting with the namenode. The credentials in the UGI is never updated (even if the user in question now has new credentials) and in case the cached UGI's credentials have expired, this would lead to authentication error whenever there is a re-authentication (in the process of re-establishing connection to the namenode).

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