[
https://issues.apache.org/jira/browse/HIVE-4505?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13651086#comment-13651086
]
Prasad Mujumdar commented on HIVE-4505:
---------------------------------------
[~owen.omalley] yes, totally agree that we need to take care of both CLI and
HiveServer2 cases.
Hive always supported user.name property in the config settings to that each
multiple CLI users in the same environment will have separate resource
directories. I believe the original HIVE-3431 patch was created to addresss
HiveServer/HiveServer2 issue where the you could have multiple sessions with
same user (eg when its not impersonating). In such case all the queries will be
sharing the same resource directories resulting into conflicts.
The proposed patch restores the pre - HIVE-3431 behavior. The CLI users can
continue to include user.name macro, there's already hive.session.id which can
also be include if needed. For example,
{noformat}
<name>hive.downloaded.resources.dir</name>
<value>/tmp/resource_dir/${user.name}/${hive.session.id}</value>
{noformat}
Besides restoring the old behavior, the patch also enabled the hiveserver2
session handle for such isolation. Unlike hive.session.id (which is a
timestamp), the hiveserver2 session handle is UUID which is a better option to
avoid conflicts. If you are using HiveServe2 without impersonation, then you
have an option to include this session handle in the directory path.
If you do use the session handles, then it will start creating a new directory
per session. Hence there's an option (which is disabled by default) to remove
that at the end of session.
> Hive can't load transforms added using 'ADD FILE'
> -------------------------------------------------
>
> Key: HIVE-4505
> URL: https://issues.apache.org/jira/browse/HIVE-4505
> Project: Hive
> Issue Type: Bug
> Components: Query Processor
> Affects Versions: 0.11.0
> Reporter: Prasad Mujumdar
> Assignee: Prasad Mujumdar
> Priority: Blocker
> Attachments: HIVE-4505-1.patch
>
>
> ADD FILE mangles name of the resource when copying to resource download
> directory. As a results following doesn't work:
> {code:sql}
> ADD FILE test.py;
> SELECT TRANSFORM (id) USING 'python test.py' AS b FROM tab1;
> {code}
> The resource gets added with a different name every time which makes it
> impossible to use transform in non-interactive mode.
> This seems to be due to HIVE-3431
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
