Implementing a control group manager that isn't a persistent daemon

Implementing a control group manager that isn't a persistent daemon

[Demi Marie Obenour]

[-- Attachment #1.1: Type: text/plain, Size: 815 bytes --]

Hello!

I'm currently implementing a control group manager that isn't a
persistent daemon.  Instead, it's a command-line tool.  The system
I'm working on uses s6 instead of systemd, and s6 doesn't have native
control group support.

The current plan is to offer a few options:

1. Create a cgroup if it doesn't exist.  If it does, wait for it to
   be empty.  Then execute a program in that cgroup.

2. Create a program in a cgroup, without waiting for the cgroup to
   be empty.
   
2. Kill all programs in a cgroup.

Unfortunately, I'm not able to wait for a cgroup to become empty.
I don't seem to be getting EPOLLPRI or EPOLLERR events.  What
is the proper way to get those events?  Is there a kernel version
with a bug in this area?
-- 
Sincerely,
Demi Marie Obenour (she/her/hers)



[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: Implementing a control group manager that isn't a persistent daemon

[Dhaval Giani]

[-- Attachment #1: Type: text/plain, Size: 1007 bytes --]

On Sun, Jul 12, 2026 at 3:51 PM Demi Marie Obenour <demiobenour@gmail.com>
wrote:

> Hello!
>
> I'm currently implementing a control group manager that isn't a
> persistent daemon.  Instead, it's a command-line tool.  The system
> I'm working on uses s6 instead of systemd, and s6 doesn't have native
> control group support.
>
> The current plan is to offer a few options:
>
> 1. Create a cgroup if it doesn't exist.  If it does, wait for it to
>    be empty.  Then execute a program in that cgroup.
>
> 2. Create a program in a cgroup, without waiting for the cgroup to
>    be empty.
>
> 2. Kill all programs in a cgroup.
>
>
Why can libcgroup not do the same for you?

Dhaval


> Unfortunately, I'm not able to wait for a cgroup to become empty.
> I don't seem to be getting EPOLLPRI or EPOLLERR events.  What
> is the proper way to get those events?  Is there a kernel version
> with a bug in this area?
> --
> Sincerely,
> Demi Marie Obenour (she/her/hers)
>
>
>

[-- Attachment #2: Type: text/html, Size: 1620 bytes --]