Re: using a lot of maintenance_work_mem

* Greg Stark (gsstark(at)mit(dot)edu) wrote:
> On Sat, Apr 9, 2011 at 6:20 PM, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
> > BTW, it sounded like your argument had to do with whether it would use
> > HashAgg or not -- that is *not* dependent on the per-palloc limit, and
> > never has been.
>
> His point was he wanted to be allowed to set work_mem > 1GB. This is
> going to become a bigger and bigger problem with 72-128GB and larger
> machines already becoming quite standard.

Actually, Tom has a point in that work_mem can be set above 1GB (which
is where I had it set previously..). I didn't think it'd actually do
anything given the MaxAlloc limit, but suprisingly, it does (at least,
under 8.4). I'm currently trying to see if we've got anything that's
going to *break* with work_mem set up that high; right now I have a
hashagg plan running across this data set which has 2.4G allocted to
it so far.

I'll update this thread with whatever I find out. I'm trying to
remember the other issues that I ran in to with this limit (beyond the
whole sort limit, which I do think would be helped by allowing a larger
value, but it's not as big a deal).

Thanks,

Stephen