Re: Tablespaces

Simon Riggs <simon@2ndquadrant.com> writes:
> I've reviewed your patch by eye, but can't see anything in your patch
> about relocating the pg_xlog directory.

I see no reasonable way to move pg_xlog while the postmaster is up.
So this is something that will always require outside-the-database
activity.  We could maybe offer a script to do it, but I can't see
any fundamental advantage there over telling people how to do it by
hand.

> There was/is a TODO item:
> - Allow xlog directory location to be specified during initdb, perhaps
> using symlinks

Note this says "during initdb", which is a whole different ballgame.
We could certainly add a "symlink pg_xlog to foo" switch to initdb,
but again, just how much functionality is that really adding?
Not a whole lot.  (I'll not stand in the way if someone wants to write
and document such a switch in the next ten days; but really the bang
for buck ratio seems low, compared to working on things that you can't
do nohow in current PG.)

> AFAICS, pg_xlog could be treated as the name of a tablespace, rather
> than as a fixed directory beneath PGDATA.

Definitely not.  pg_xlog is *not* a tablespace.  xlog activity happens
below the level at which it is reasonable to pay attention to tablespaces.

            regards, tom lane