On Mon, Jun 13, 2016 at 2:39 AM, hubert depesz lubaczewski
<depesz@depesz.com> wrote:
> On Sun, Jun 12, 2016 at 01:50:48PM -0700, Jeff Janes wrote:
>
> Line #2 shows output of ps nh uww -p <backend_pid> before start of work.
> There are, in total, 74002 tables, and then I iterate over list of them,
> and for each, I do the select I mentioned.
>
> Every 1000 tables, I get stats - ps output, and (in parent) sum of
> "Anonymous:" lines from /proc/<backend_pid>/smaps.
>
> As you can see - we're getting ~ 32kB of cache per table.
What hardware and OS are you using? I only get a bit over 8kB per
table, and that amount doesn't change much between 9.3 and 9.6.
Can you share a representative table definition, including constraints?
Cheers,
Jeff