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

On 16 October 2014 02:26, Jeff Davis <pgsql(at)j-davis(dot)com> wrote:

> The inheritance is awkward anyway, though. If you create a tracked
> context as a child of an already-tracked context, allocations in the
> newer one won't count against the original. I don't see a way around
> that without introducing even more performance problems.

This seems to have reached impasse, which is a shame. Let me throw out
a few questions to see if that might help.

Do I understand correctly that we are trying to account for exact
memory usage at palloc/pfree time? Why??

Surely we just want to keep track of the big chunks? Does this need to
be exact? How exact does it need to be?

Or alternatively, can't we just sample the allocations to reduce the overhead?

Are there some assumptions we can make about micro-contexts never
needing to be tracked at all? Jeff seems not too bothered by
inheritance, whereas Tomas thinks its essential. Perhaps there is a
middle ground, depending upon the role of the context?

--
Simon Riggs http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Training & Services