Bug#943981: Proposal: Switch to cgroupv2 by default

[Ryutaroh Matsumoto]

>> (2) lxc: autopkgtest of systemd in stretch fails with LXC backend in cgroup2 / unified hierarchy
>> https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=958508
> The networkd test is flaky sometimes. I've seen it fail occasionally.
> Can you reliably reproduce the failure with this setup?

Ahhh... You are right.
Deleting the container, running debci setup and autopkgtest again
made that test passed! I close 958508 by this email...

> probably can't wait for docker (I would document that in the release
> notes / NEWS entry instead what docker users can do)

Docker is cgroup2-ready, please check
https://github.com/moby/moby/pull/40174
On the other hand, its upstream developers seem to postpone its official release
until they sort out other issues, e.g., cgroup2-enabled CI for docker,
as https://github.com/moby/moby/issues/40360

On the other hand, I have no idea if the next docker  can be packaged in the Debian way.
It seems almost impossible to make a Debian package of podman,
an alternative to docker...

At least we can file a wishlist bug report against docker.io for
packaging the latest github source...

> What are your thoughts regarding doing the switch? Are we there yet
> (once lxc v4 enters testing)? For me lxc is the main blocker and we
> probably can't wait for docker

I think the default hierarchy can be switched to the unified
after lxc4 enters into the testing (ignoring docker).
I see little problem with switching to the cgroup2 / unified hierarchy.
Only blocker to this 943981 is docker, and lxc4 does not seem to have
a problem on unified or hybrid hierarchy. Lxd 4 (in snap) does
not have a problem on the unified hierarchy, as far as I see.

By the way, lxc.init.cmd = /sbin/init systemd.unified_cgroup_hierarchy=1
has no problem with a container having sysvinit (e.g. Gentoo and Devuan),
in the same way as lxc.mount.auto = cgroup:rw:force.

> P.S.: Thanks for your ongoing efforts regarding the cgroupv2 switch.
> This is very much appreciated.

You're welcome, it's my pleasure.

Best regards, Ryutaroh

From: Michael Biebl <biebl at debian.org>
Subject: Re: Bug#943981: Proposal: Switch to cgroupv2 by default
Date: Thu, 23 Apr 2020 12:52:48 +0200

> Am 23.04.2020 um 06:04 schrieb Ryutaroh Matsumoto:
>> Hi Michael,
>> 
>> I filed two unwelcome interactions between systemd and LXC on Debian:
>> 
>> (1) lxc: jessie LXC container does not start on hosts with cgroup2 / unified hierarchy
>> https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=958507
> 
> Hm, I probably wouldn't worry too much. Jessie is already really old and
> bullseye (for which the change is targetted) will not be released for
> yet another year.
> Then again, I don't know if it's an important use-case for admins to be
> able to run jessie containers on a bullseye host.
> 
>> (2) lxc: autopkgtest of systemd in stretch fails with LXC backend in cgroup2 / unified hierarchy
>> https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=958508
> 
> The networkd test is flaky sometimes. I've seen it fail occasionally.
> Can you reliably reproduce the failure with this setup?
> 
> Regards,
> Michael
> 
> P.S.: Thanks for your ongoing efforts regarding the cgroupv2 switch.
> This is very much appreciated.
> What are your thoughts regarding doing the switch? Are we there yet
> (once lxc v4 enters testing)? For me lxc is the main blocker and we
> probably can't wait for docker (I would document that in the release
> notes / NEWS entry instead what docker users can do)
> 
> -- 
> Why is it that all of the instruments seeking intelligent life in the
> universe are pointed away from Earth?
>