| From: | Josh Berkus <josh(at)agliodbs(dot)com> | 
|---|---|
| To: | pgsql-hackers(at)postgresql(dot)org | 
| Subject: | Proposed TODO: CREATE .... WITH OWNER; | 
| Date: | 2004-10-23 23:59:38 | 
| Message-ID: | 200410231659.38093.josh@agliodbs.com | 
| Views: | Whole Thread | Raw Message | Download mbox | Resend email | 
| Thread: | |
| Lists: | pgsql-hackers | 
People:
Having today spent 3.5 hours correcting a pg_dump file with permissions 
problems, I've come to the inescapable realization that the "SESSION 
AUTHORIZATION" concept is WAY too fragile.
Instead, we should have a "CREATE .... WITH OWNER username" extension to all 
of our CREATE <object> statements.     Then any backup, or fragment of a 
backup, could be run by the superuser without fear that a bunch of objects 
could end up owned by a user with no permissions on them.   (And if you think 
such a fear does not exist, try using "CHANGE OWNER" on about 80 database 
objects, some of them with dependancies owned by other users, and then 
pg_dump and restore.  Fun, fun!).
CREATE followed by ALTER ... CHANGE OWNER would not be an adequate substitute.  
The orginal owner of the object (in the case of a restore, the superuser) 
retains all of their permissions on the object, which causes a lot of messy 
GRANT statements.
Hmmmm ... this would also require a GRANT .... AS USER name.  But those two 
changes should simplify dump and restore enormously.
-- 
--Josh
Josh Berkus
Aglio Database Solutions
San Francisco
| From | Date | Subject | |
|---|---|---|---|
| Next Message | Tom Lane | 2004-10-24 00:05:52 | Re: Proposed TODO: CREATE .... WITH OWNER; | 
| Previous Message | Gaetano Mendola | 2004-10-23 22:08:22 | Re: Question on the 8.0Beta Version |