[21u] Backport intention of 8322420: [Linux] cgroup v2: Limits in parent nested control groups are not detected

[Severin Gehwolf]

Hi,

This is a rather invasive patch set. Is this something that Azul ships
in the downstream distribution?

On Fri, 2025-02-07 at 21:13 +0800, Jan Kratochvil wrote:
> On Mon, 20 Jan 2025 17:58:01 +0800, Severin Gehwolf wrote:
> > You didn't say *why* you'd like to backport it
> 
> https://bugs.openjdk.org/browse/JDK-8322420
> The following text is based on:
> 	https://mail.openjdk.org/pipermail/jdk-updates-dev/2022-February/012398.html
> 
> cgroups v2 support hierarchical groups. Limits from outer groups apply also
> for the inner groups. To calculate an effective limit for the current nested
> group one needs to calculate a minimum of all the limits in the current group
> and its parent groups up to the root one.
> 
> The risk of NOT backporting those patches is that the JVM would wrongly detect
> resource limits. Container detection code would just give up early and would
> use the host systems' resource limits instead. Random OOM kills and other
> strange behaviour can happen.

Surely this only applies to systems that have such a setup (of
hierarchical cgroups), not cgroup limits in general. Common container
frameworks (kubernetes) aren't affected AFAIK (I might be wrong,
though). The risk is a function of such affected systems. Would you
have any estimation of the affected systems out in the wild? That would
help assess this.

> As the code is the same in 21u as in 24u and all the backports are clean I do
> not expect regressions.

JDK 24 hasn't been officially released yet, so the proposed patches
haven't seen much real-world usage yet.

> If pre-approved I need to submit those patches as separate backports.

OK.

> I find the April release as the suitable one.

I don't think that's a good time-frame. For one, JDK 24 hasn't been
released. For two, we are already fairly late in the April CPU
development cycle.

Let's reconsider this for a later release.

Thanks,
Severin