Re: 9.5: Better memory accounting, towards memory-bounded HashAgg

On Wed, 2014-08-06 at 11:43 -0400, Robert Haas wrote:
> Comparing the median times, that's about a 3% regression. For this
> particular case, we might be able to recapture that by replacing the
> bespoke memory-tracking logic in tuplesort.c with use of this new
> facility. I'm not sure whether there are other cases that we might
> also want to test; I think stuff that runs all on the server side is
> likely to show up problems more clearly than pgbench. Maybe a
> PL/pgsql loop that does something allocation-intensive on each
> iteration, for example, like parsing a big JSON document.

I wasn't able to reproduce your results on my machine. At -s 300, with
maintenance_work_mem set high enough to do internal sort, it took about
40s and I heard some disk activity, so I didn't think it was a valid
result. I went down to -s 150, and it took around 5.3s on both master
and memory-accounting.

Either way, it's better to be conservative. Attached is a version of the
patch with opt-in memory usage tracking. Child contexts inherit the
setting from their parent.

Regards,
Jeff Davis