Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: quoted-printable
Subject: Re: Safe vm.overcommit_ratio for Large Multi-Instance PostgreSQL
 Fleet
Mime-Version: 1.0 (1.0)
From: Frits Hoogland <frits.hoogland@gmail.com>
In-Reply-To: <6e36e010-c93c-4d9b-a7ed-db6da1416716@dalibo.com>
Cc: Joe Conway <mail@joeconway.com>, Priya V <mailme0216@gmail.com>,
 pgsql-performance@lists.postgresql.org
Date: Tue, 19 Aug 2025 17:37:10 +0200
Message-Id: <08FF1C8D-DDA7-400C-9AC7-71A92AC925E7@gmail.com>
References: <6e36e010-c93c-4d9b-a7ed-db6da1416716@dalibo.com>
To: =?utf-8?Q?Fr=C3=A9d=C3=A9ric_Yhuel?= <frederic.yhuel@dalibo.com>
Archived-At: 
 <https://www.postgresql.org/message-id/08FF1C8D-DDA7-400C-9AC7-71A92AC925E7%40gmail.com>
Precedence: bulk

Thank you for your message Frederic,

I am very much aware of that issue. It=E2=80=99s actually incorrect to say t=
hat is a bug: that is how cgroupsv1, which is bundled with rhel8, works. How=
ever, it is very counter intuitive. For that reason redhat created the force=
_cgroup_v2_swappiness parameter uniquely theirselves, it=E2=80=99s not a com=
mon Linux parameter.

The specific issue I see in certain cases leading to unreasonable swap usage=
 is Linux workingset detection kicking in, which can choose anonymous memory=
 despite having lots of file memory available, leading to swapping, which so=
metimes leads to a thrashing situation.

It is funny to see how emotional people react to removing swap, and how peop=
le go through great efforts to carefully trying wrap that in a technical rea=
son or point to people having said something that agrees with their emotion.=
 I should say that I understand the reluctance, it=E2=80=99s not weird to fe=
el anxious.

The kernel has no inherent swap requirement. Of course, removing swap cannot=
 be blindly applied, you have to carefully make it suit your environment, us=
age and intention. And there ARE cases where swap makes sense (if you have m=
emory usage that exceeds physical memory, and you add enough swap to sustain=
 that). But a database in general typically responds bad to swapping (or any=
thing that fluctuates latency), and when swap removal is sensibly done, it p=
revents anonymous (including lesser frequently used ;-)) memory from getting=
 swapped.=20

I will not convince everybody, but I hope I can make some people that unders=
tand the technology thinking about it and consider the arguments.

Friendly regards,

Frits


>=20
> Op 18 aug 2025 om 18:17 heeft Fr=C3=A9d=C3=A9ric Yhuel <frederic.yhuel@dal=
ibo.com> het volgende geschreven:
>=20
> =EF=BB=BF
>=20
>> On 8/8/25 10:21, Frits Hoogland wrote:
>> If swappiness is set to 0, but swap is available, some documentation sugg=
ests it will never use anonymous memory, however I found this not to be true=
, linux might still choose anonymous memory to reclaim.
>=20
>=20
> A bug in RHEL8 meant that swappiness was not taken into account unless cgr=
oupv2 was configured or vm.force_cgroup_v2_swappiness was set to 1. See refe=
rences [1] and [2]. Could this be the cause of your observation?
>=20
> [1] https://access.redhat.com/solutions/6785021
> [2] https://github.com/systemd/systemd/issues/9276