[
https://issues.apache.org/jira/browse/HIVE-1667?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12916649#action_12916649
]
Ashutosh Chauhan commented on HIVE-1667:
----------------------------------------
I am having second thought on this. What exactly is the usecase for storing
group name? HDFS follows BSD semantics for group resolution which means that
when table is created in Hive which results in mkdirs() call on HDFS, table dir
will be created with the group as group of parent dir (which will be warehouse
dir) Now its possible that user creating the table is not a part of parent
dir's group. By storing primary group of user in table's metadata in metastore,
you now are in inconsistent state. You have stored group name of the table as
user's primary group while on the other hand group of the dir on HDFS is
different and is the group of parent's dir.
In my opinion to avoid such divergences, HDFS should be considered as primary
source of truth w.r.t, permissions and group on both logical table as well as
physical dirs.
Whenever user asks Hive to set permissions or group we should directly try to
do chmod and chgrp on physical dirs and whenever Hive need user and group
information it should query filesystem. Thoughts?
> Store the group of the owner of the table in metastore
> ------------------------------------------------------
>
> Key: HIVE-1667
> URL: https://issues.apache.org/jira/browse/HIVE-1667
> Project: Hadoop Hive
> Issue Type: New Feature
> Reporter: Namit Jain
> Attachments: hive-1667.patch
>
>
> Currently, the group of the owner of the table is not stored in the metastore.
> Secondly, if you create a table, the table's owner group is set to the group
> for the parent. It is not read from the UGI passed in.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
