On Thu, Oct 4, 2012 at 10:54 AM, Hugo <Nabble> <hugo.tech@gmail.com> wrote:
> We use Debian linux 32-bits, so the addressable space available really seems
> to be in the 3.0Gb to 3.5Gb range. Last night I decreased the shared_buffers
> from 2Gb to 1Gb and tried the global ANALYZE again. It went out of memory
> after 3 hours 40 minutes. That database has 12,197 schemas with 22 tables
> each, which means 268,334 tables. Should I keep reducing the shared_buffers
> and trying again? We don't plan to run the ANALYZE every week, so we can
> keep the shared_buffers high most of the time and tweak it when needed. I
> will try again tonight with ~512Mb and see what happens. Please let me know
> if you have ideas.
Your analyzes must be taking more memory than mine did, probably your
tables had indexes and my dummy ones did not. But you said you had
shell scripts to analyze table by table, and that seems like the
solution (until you can upgrade to 64-bit) as it is both faster and
more memory efficient to do it that way.
Cheers,
Jeff