Gurus,
I hope I can make this clear somehow... Anyway... This all involves PG
8.1.4 on a 64-bit FC5 box.
Select version() says "PostgreSQL 8.1.4 on x86_64-redhat-linux-gnu,
compiled by GCC x86_64-redhat-linux-gcc (GCC) 4.1.0 20060304 (Red Hat
4.1.0-3)".
I guess the best question I can see is, under what circumstances is the
directory name in pg_tablespace actually used?
I have a scenario where I want to restore from a PITR backup, into an
alternate location on the same machine it came from, while the original
database is still up and running. I have one alternate table space.
It goes like this. First I expand the base archive into an alternate
location, then expand the table space archive(s) into alternate
location(s). Then I recreate the links under pg_tblspc. I then fiddle
a little bit with config files and run postgres on the new alternate
location. Everthing goes fine, the database rolls forward, and then
postgres quits (because I give it an SQL file for stdin). Great.
But now I have a problem. What if I move objects from the main
tablespace to the alternate one, such as indexes, between the time of
the backup and the restore? During the restore/recovery, the
pg_tablespace table is out of date. If the tablespace directory listed
there was used in copying files, I'd have a big fat mess involving a
badly broken production database.
Hopefully that all makes sense...
-Glen
