Horton

Re: Default setting for enable_hashagg_disk

Peter Geoghegan <pg@bowt.ie>

View thread View on postgr.es
From: Peter Geoghegan <pg@bowt.ie>
To: Tom Lane <tgl@sss.pgh.pa.us>
Cc: Stephen Frost <sfrost@snowman.net>, Jeff Davis <pgsql@j-davis.com>, Alvaro Herrera <alvherre@2ndquadrant.com>, Andres Freund <andres@anarazel.de>, Bruce Momjian <bruce@momjian.us>, Robert Haas <robertmhaas@gmail.com>, David Rowley <dgrowleyml@gmail.com>, Justin Pryzby <pryzby@telsasoft.com>, Melanie Plageman <melanieplageman@gmail.com>, Tomas Vondra <tomas.vondra@2ndquadrant.com>, "pgsql-hackers@postgresql.org" <pgsql-hackers@postgresql.org>
Date: 2020-07-10T21:00:00Z
Lists: pgsql-hackers, pgsql-docs 
On Fri, Jul 10, 2020 at 10:47 AM Tom Lane <tgl@sss.pgh.pa.us> wrote:
> I looked over Peter's patch in [1], and it seems generally pretty
> sane to me, though I concur with the idea that it'd be better to
> define the GUC as a multiplier for work_mem.  (For one thing, we could
> then easily limit it to be at least 1.0, ensuring sanity; also, if
> work_mem does eventually become more dynamic than it is now, we might
> still be able to salvage this knob as something useful.  Or if not,
> we just rip it out.)  So my vote is for moving in that direction.

Cool. I agree that it makes sense to constrain the effective value to
be at least work_mem in all cases.

With that in mind, I propose that this new GUC have the following
characteristics:

* It should be named "hash_mem_multiplier", a floating point GUC
(somewhat like bgwriter_lru_multiplier).

* The default value is 2.0.

* The minimum allowable value is 1.0, to protect users from
accidentally giving less memory to hash-based nodes.

* The maximum allowable value is 100.0, to protect users from
accidentally setting hash_mem_multiplier to a value intended to work
like a work_mem-style KB value (you can't provide an absolute value
like that directly). This maximum is absurdly high.

I think that it's possible that a small number of users will find it
useful to set the value of hash_mem_multiplier as high as 5.0. That is
a very aggressive value, but one that could still make sense with
certain workloads.

Thoughts?
-- 
Peter Geoghegan

Commits

  1. Add hash_mem_multiplier GUC.

  2. HashAgg: use better cardinality estimate for recursive spilling.

  3. Remove hashagg_avoid_disk_plan GUC.

  4. Doc fixup for hashagg_avoid_disk_plan GUC.

  5. Rework HashAgg GUCs.

  6. Disk-based Hash Aggregation.

  7. Implement partition-wise grouping/aggregation.

  8. Defer creation of partially-grouped relation until it's needed.