Re: Tablespaces

Tom Lane wrote:
> Gavin Sherry <swm(at)linuxworld(dot)com(dot)au> writes:
> >> Speaking of locking, can we do anything to prevent people from shooting
> >> themselves in the foot by changing active tablespaces? Are we even
> >> going to have a DROP TABLESPACE command, and if so what would it do?
>
> > Ahh. I forgot to detail my ideas on this. It seems to me that we cannot
> > drop a table space until the directory is empty.
>
> How would it get to be empty? Are you thinking of some sort of "connect
> database to tablespace" and "disconnect database from tablespace"
> commands that would respectively create and delete the per-database
> subdirectory? That seems moderately reasonable to me. We could then
> invent a locking protocol that requires backends to lock a tablespace
> before they can execute either of these operations (or delete the
> tablespace of course).

One crude solution would be to remove the tablespace oid directory only
when the database is dropped, and require an empty tablespace directory
to drop the tablespace. This allows a lock only on tablespace creation,
and not a lock on object creation in each tablespace.

--
Bruce Momjian | http://candle.pha.pa.us
pgman(at)candle(dot)pha(dot)pa(dot)us | (610) 359-1001
+ If your life is a hard drive, | 13 Roberts Road
+ Christ can be your backup. | Newtown Square, Pennsylvania 19073