Horton

Re: heavily contended lwlocks with long wait queues scale badly

Michael Paquier <michael@paquier.xyz>

View thread View on postgr.es
From: Michael Paquier <michael@paquier.xyz>
To: "Jonathan S. Katz" <jkatz@postgresql.org>
Cc: Nathan Bossart <nathandbossart@gmail.com>, Andres Freund <andres@anarazel.de>, Bharath Rupireddy <bharath.rupireddyforpostgres@gmail.com>, pgsql-hackers@postgresql.org, Thomas Munro <thomas.munro@gmail.com>, Yura Sokolov <y.sokolov@postgrespro.ru>
Date: 2024-01-17T06:19:16Z
Lists: pgsql-hackers 
On Tue, Jan 16, 2024 at 11:24:49PM -0500, Jonathan S. Katz wrote:
> On 1/16/24 1:11 AM, Michael Paquier wrote:
>> I'd like to apply that, just let me know if you have any comments
>> and/or objections.
> 
> Looking at the code, I understand an argument for not backpatching given we
> modify the struct, but this does seem low-risk/high-reward and should help
> PostgreSQL to run better on this higher throughput workloads.

Just to be clear here.  I have repeated tests on all the stable
branches yesterday, and the TPS falls off drastically around 256
concurrent sessions for all of them with patterns similar to what I've
posted for 12, getting back a lot of performance for the cases with
more than 1k connections.
--
Michael

Commits

  1. lwlock: Fix quadratic behavior with very long wait lists