[
https://issues.apache.org/jira/browse/HIVE-1322?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12866408#action_12866408
]
Paul Yang commented on HIVE-1322:
---------------------------------
A unit test is just about impossible, so I was only able to run a manual test:
1. With a running DFS, I issued a create table command that would create a
FileSystem object.
2. The DFS was shut down, and a create table command was issued again. This
threw an IOException which would have cleared the cache.
3. The DFS was restarted, and I verified that a new create table command
succeeded.
Given that this is a straightforward change, the above should be sufficient to
demonstrate that this doesn't break anything.
> Cached FileSystem can lead to persistant IOExceptions
> -----------------------------------------------------
>
> Key: HIVE-1322
> URL: https://issues.apache.org/jira/browse/HIVE-1322
> Project: Hadoop Hive
> Issue Type: Bug
> Components: Metastore
> Reporter: Paul Yang
> Assignee: Paul Yang
> Priority: Minor
> Fix For: 0.6.0
>
> Attachments: HIVE-1322.1.patch, HIVE-1322.2.patch
>
>
> In the metastore, the FileSystem object is created through
> Path.getFileSytem(), which caches the created instance for performance. For
> DFS, the cached FileSystem seems to retain the original IP (resolved from the
> supplied hostname). If the hostname->IP mapping changes, then FS operations
> will throw IOExceptions. Because the FileSystem is cached, re-creating the
> object has no effect and will continue to result in IOExceptions.
> One solution is to call close on an IOException. That will remove the entry
> in the cache.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
