Re: MultiXact\SLRU buffers configuration

Thomas Munro <thomas.munro@gmail.com>

From: Thomas Munro <thomas.munro@gmail.com>

To: Andrey Borodin <x4mmm@yandex-team.ru>

Cc: Gilles Darold <gilles@darold.net>, Tomas Vondra <tomas.vondra@enterprisedb.com>, Tomas Vondra <tomas.vondra@2ndquadrant.com>, Alexander Korotkov <aekorotkov@gmail.com>, Anastasia Lubennikova <a.lubennikova@postgrespro.ru>, Daniel Gustafsson <daniel@yesql.se>, Kyotaro Horiguchi <horikyota.ntt@gmail.com>, pgsql-hackers <pgsql-hackers@postgresql.org>

Date: 2021-03-24T21:31:56Z

Lists: pgsql-hackers

Attachments

v10-0001-Make-all-SLRU-buffer-sizes-configurable.patch (text/x-patch) patch v10-0001
v10-0002-Add-buffer-mapping-table-for-SLRUs.patch (text/x-patch) patch v10-0002

Hi Andrey,

On Sat, Mar 13, 2021 at 1:44 AM Andrey Borodin <x4mmm@yandex-team.ru> wrote:
> [v10]

+int            multixact_offsets_slru_buffers = 8;
+int            multixact_members_slru_buffers = 16;
+int            subtrans_slru_buffers = 32;
+int            notify_slru_buffers = 8;
+int            serial_slru_buffers = 16;
+int            clog_slru_buffers = 0;
+int            commit_ts_slru_buffers = 0;

I don't think we should put "slru" (the name of the buffer replacement
algorithm, implementation detail) in the GUC names.

+        It defaults to 0, in this case CLOG size is taken as
<varname>shared_buffers</varname> / 512.

We already know that increasing the number of CLOG buffers above the
current number hurts as the linear search begins to dominate
(according to the commit message for 5364b357), and it doesn't seem
great to ship a new feature that melts your CPU when you turn it up.
Perhaps, to ship this, we need to introduce a buffer mapping table?  I
have attached a "one coffee" attempt at that, on top of your v10 patch
(unmodified), for discussion.  It survives basic testing but I don't
know how it performs.

Commits

Rework new SLRU test with injection points
- 94a3373ac5c3 18.0 landed
injection_point: Add injection_points.stats
- 2e35c67f9568 18.0 landed
injection_points: Add initialization of shmem state when loading module
- b2b023aa3706 18.0 landed
Add injection-point test for new multixact CV usage
- 768a9fd5535f 18.0 landed
Use conditional variable to wait for next MultiXact offset
- a0e0fb1ba56f 17.0 landed
Improve performance of subsystems on top of SLRU
- 53c2a97a9266 17.0 landed
Adjust VACUUM hastup LP_REDIRECT comments.
- 325bc54eed4e 16.0 cited
Test replay of regression tests, attempt II.
- f47ed79cc8a0 15.0 cited