W/o HIVE-493 behavior and having shared, partitioned base table data such schema metadata duplication gets uglier -- while we had multiple schemas in general most users fell into a single schema and it is definitely preferred that there could be permissioned views in the single schema metadata even to reduce clutter of user tables in eg 'show tables'. We had disabled general availability of DROP (paranoid code tweak) and you have to be using HDFS permissions to protect against insert overwrites.
Also aside from metadata from an HDFS resource mgmt perspective it would have been nice if user created tables defaulted to a different root directory than schema root e.g. if shared tables in /analytics user tables in /analytics_usr/userx -- the external table support allows for such now but the use of 'external' is a user behavior that is not inherently enforced. I suppose in your per-user schema model u could technically be specifying the root as /analytics_usr/userx and created the shared tables as external but such config is unwieldy. Actually, w/o doing this HDFS scoping but using Hive schema scoping don't you have the possibility of blind conflicts I.e. two users can create the same table w/o realizing it and may/maynot pollute each other solely on HDFS permissions? On Jul 13, 2009 10:18 PM, "Min Zhou" <[email protected]> wrote: Hmm, HIVE-493 is okay since authentication subsystem isn't existing. I was considering using something like crontab to automatically add paritions for our tables. Thanks, Prasad! On Tue, Jul 14, 2009 at 1:05 PM, Prasad Chakka <[email protected]> wrote: > > You could fix htt...
