[systemd-devel] Antw: [EXT] I/O error on "systemctl kill -s HUP rsyslog.service"

[Lennart Poettering]

On Do, 13.08.20 10:05, Andrei Borzenkov (arvidjaar at gmail.com) wrote:

> 13.08.2020 09:54, Harald Dunkel пишет:
> > On 8/12/20 2:16 PM, Andrei Borzenkov wrote:
> >> 12.08.2020 14:03, Harald Dunkel пишет:
> >>> See attachment. Hope this helps
> >>> Harri
> >>
> >>
> >>> 1     openat(AT_FDCWD,
> >>> "/sys/fs/cgroup/unified/system.slice/rsyslog.service/cgroup.procs",
> >>> O_RDONLY|O_CLOEXEC) = 24
> >>> 1     read(24, "0\n1544456\n", 4096)    = 10
> >>
> >>
> >> kernel returns "0" as process number in this cgroup which results in EIO
> >> returned by systemd.
> >
>
> systemd should really clearly log this (invalid PID and and in which
> cgroup it was). Returning generic error message without any indication
> what caused this error is not useful at all.

I think it's necessary that we log, but I am not sure we have to have
a clear error message ready for every kind of impossible error. I'd
just leave this as is. The setup that causes this is seriously broken,
and I am not convinced we need to handle that gracefully or help the
user with a clear message.

You cannot run multiple containers in the same cgroup hierarchy,
that's against the delegation model of cgroups. Containers need a
clearly delegated subtree of cgroup hierarchy, and running everything
in the same top-level tree is just entirely bogus. This is very
clearly documented.

It's an LXC bug, that's all. And yes, it causes weird error messages
in systemd, but that's because the setup is just so broken, and as
long as you do get *some* error messages I think we are good.

Lennart

--
Lennart Poettering, Berlin