Re: BUG #19096: Hugepages configuration is probably not reserving the allocated number of pages - Mailing list pgsql-bugs

From Tomas Vondra
Subject Re: BUG #19096: Hugepages configuration is probably not reserving the allocated number of pages
Date
Msg-id 04c36177-74fc-4d4e-b492-e3d00099190e@vondra.me
Whole thread Raw
In response to BUG #19096: Hugepages configuration is probably not reserving the allocated number of pages  (PG Bug reporting form <noreply@postgresql.org>)
List pgsql-bugs

On 10/28/25 10:37, PG Bug reporting form wrote:
> The following bug has been logged on the website:
> 
> Bug reference:      19096
> Logged by:          dex terity
> Email address:      szmscy4nfxk@protonmail.com
> PostgreSQL version: 18.0
> Operating system:   Debian 12, 13
> Description:        
> 
> Thanks for Postgresql
> I have been exploring hugepages functionality in PG and other applications,
> found out there's actually a worthy performance improvement. The problem is
> that while other applications such as libvirt, php all play nice with HP, if
> I launch PG with HP then the next apps will show problems, even though the
> number of pages is in excess. Without PG it is possible to run any
> combination of apps and instances without problems. PG runs happy when alone
> or with one libvirt vm. One thing I noticed is PG does not reserve the full
> number of pages when started... maybe that's the problem.
> Regards
> 

I have no idea what "play nice" means for the other apps. Either there
are enough huge pages or not. And if the apps "show problems" after the
pages get used by Postgres, I'd not call that "play nice".

Postgres does not reserve huge pages (and the documentation makes this
fairly clear [1]). Because how would it know how many huge pages to
reserve? There can be other stuff running on the system, and we have no
idea how much memory / hugepages that may need.


[1]
https://www.postgresql.org/docs/current/kernel-resources.html#LINUX-HUGE-PAGES

-- 
Tomas Vondra




pgsql-bugs by date:

Previous
From: Paul A Jungwirth
Date:
Subject: Re: BUG #19098: Can't create unique gist index, where pg_indexes says that WITHOUT OVERLAPS does exacly that
Next
From: PG Bug reporting form
Date:
Subject: BUG #19099: Conditional DELETE from partitioned table with non-updatable partition raises internal error