On 2010-01-21, Fernando Schapachnik <fernando@mecon.gov.ar> wrote:
> Hi,
>
> I have a big database on FS1, now almost full. Have space on FS2,
> where I created a tablespace and moved every table and index to it.
> Still, lots of space used on FS1. The problem is not pg_xlog, but
> base:
>
> # du -hs base/105658651/* | fgrep G
> 1,0G base/105658651/106377323
> 1,0G base/105658651/106377323.1
> 1,0G base/105658651/106377323.2
> 1,0G base/105658651/106377323.3
> 1,0G base/105658651/106377323.4
> 1,0G base/105658651/125520217
> 1,0G base/105658651/127352052
> 1,0G base/105658651/127352052.1
> 1,0G base/105658651/127352052.2
> 1,0G base/105658651/127352052.3
> 1,0G base/105658651/127352052.4
> 1,0G base/105658651/127352052.5
>
> Unfortunately no volume management is available, so I can't move
> disks from FS2 to FS1.
>
> I could play soft links tricks, but I'm afraid of paying the
> FS-traversal penalty on each file access (is that right?).
>
> So, any way of instructing PG (8.1 if that matters) to place those
> files elsewhere without an initdb?
Linux has "mount -o bind" which works like a symlink, but at a lower level.
You appear to use BSD which appears to have an equivalent in nullfs.
