Robert Treat <xzilla@users.sourceforge.net> writes:
> If the location doesn't exist will postgresql try to create it? istm it could
> do this and if it fails then you are no worse off, but if it were to succeed
> you're that much better off.
I think this would be fairly pointless. In most of the practical
tablespace scenarios I can think of, the tablespace directory probably
lives within a root-owned directory (eg, a filesystem root directory).
That's why CREATE TABLESPACE expects the directory to have been made
already.
In point of fact I think this discussion is much ado about nothing,
as there is already a workaround that is about as simple as anything
that we would likely be able to substitute. Suppose the dump contains
"CREATE TABLESPACE t1 LOCATION '/foo/bar'" and for some reason /foo/bar
is no longer an appropriate location. All the DBA need do is select
a location that *is* suitable and create tablespace t1 at that location.
Then run the restore. The create tablespace command will fail on
duplicate name, but the tablespace is there and all the subsequent
operations will be just fine.
Of course we need to document this procedure, but we'd have to document
any other approach as well...
regards, tom lane