Hi Fulvio,

On 20/11/2014 18:08, fulvio ciriaco wrote:
> $ dpkg -S /sbin/init
> sysvinit-core: /sbin/init

Ok, so systemd-shim and cgmanager are actually required on your system.
No surprise from a dependency chain standpoint.

> All right, you are in the driver seat.

Nope, I'm just a user as you are. I'm just interested in the well-being
of the systemd-shim stack :)

> Choosing systemd rather than cgmanager does not solve anything.

I wasn't implying that you'd have to switch, I was just saying that
under systemd cgmanager is inoperative because cgroup transitions are
handled internally by systemd itself.

Bottomline: I was trying to get out of the picture a possible regression
of cgmanager on systems with systemd-sysv.

> So, is this good practice? Should users learn to dig mount files in /proc?
> Where is this documented?

As Ted Ts'o stated above, the reliance of cgmanager on a private mount
namespace is outlined in its README. The behavior of df et al. is also
expected because isolating mount states is what mount namespaces do by
design.

Perhaps accessing namespace data from the "root" namespace without
digging into /proc could be streamlined a bit, but as it stands IMHO
it's not a cgmanager issue.

The actual problem (cgmanager using a private mount namespace) is
debatable, but it's hardly a bug per-se (given that it's documented -
perhaps it should go into the NEWS file as well so users are warned upon
first installation?)

> In the meanwhile, (before reading your advice), I had disabled cgmanager,
> and have not yet had any problem. I will leave it so, just out of
> curiosity of which bomb it will trigger.

TL;DR: everything that relies upon policykit-1 and by extension
libpam-systemd (or on logind directly) could fail in unexpected ways.
You need cgmanager running.

> Thank you all the same,
> best wishes

Regards,
-- 
Matteo Panella

Attachment: signature.asc
Description: OpenPGP digital signature

Reply via email to