new package: incus

[dkwo]

• I tested the changes in this PR: yes
• This new package conforms to the package requirements: yes

todo:

• void docs: mention incus groups and something like echo "root:1000000:1000000000" | doas tee -a /etc/subuid /etc/subgid

[ahesford]

I don't think we should drop lxd in this PR.

[dkwo]

when building incus, i need to update raft, which makes it harder to bump dqlite and lxd.
the current lxd is also outdated, and it hasn't been rebuilt with at least go 1.21 as its distfiles have been moved.

[sbromberger]

I managed to get incus running on void a couple days ago. here's my runit service if it helps:

run:

#!/bin/sh
exec 2>&1
_systemd_cgrp="/sys/fs/cgroup/systemd"
if [ ! -d ${_systemd_cgrp} ]; then
        mkdir ${_systemd_cgrp}
fi
if ! mountpoint -q "${_systemd_cgrp}"; then
        mount -t cgroup -o none,name=systemd cgroup ${_systemd_cgrp}
fi

[ -r conf ] && . ./conf
exec /usr/local/bin/incusd --group incus-admin --syslog ${OPTS:- --verbose}

check:

#!/bin/sh
incus config show >/dev/null 2>&1

[dkwo]

Thanks! Updated incus to 0.4, tried to follow upstream recommendations https://linuxcontainers.org/incus/docs/main/packaging/ which also suggests

The incusd and incus-user daemons should be kept outside of the user’s PATH. The same is true of incus-agent which needs to be available in the daemon’s PATH but not be visible to users.

how can i keep those daemons outside of path?

[sbromberger]

how can i keep those daemons outside of path?

Recommendation from folks on #lxc liberachat is to put them in /usr/sbin.

[sbromberger]

Would it be possible to make cowsql a separate PR and get that integrated sooner?

[ahesford]

Void is usr-merged and doesn't distinguish between root-visible and user-visible executables. If the executables are intended only for execution by other programs, a subdirectory of /usr/libexec is the proper place. Otherwise, just put them in /usr/bin.

[dkwo]

thanks, i'm now using libexec.
as for a separate cowsql pr: i had to update raft, and i'm afraid of touching lxd, but i can try.
some serious testing (in particular of runit services) is also welcome.

[dkwo]

is it ok to add export PATH=/usr/libexec/incus:$PATH to the services, so that they know where to find incus-agent?

[ahesford]

is it ok to add export PATH=/usr/libexec/incus:$PATH to the services, so that they know where to find incus-agent?

No. If the programs that use these utilities expect them to be in the system path, then they belong in the standard /usr/bin. Programs that use non-user-visible executables should expect to find these utilities in /usr/libexec without modifications to the search path. If the upstream suggestion is to put them in /usr/sbin to keep them out of the default path of ordinary users (/usr/sbin should always be in the default path of the super user), then the answer on Void is that we make no distinction.

[classabbyamp]

squash the two incus commits please

[dkwo]

squashed.

[sbromberger]

Now that this has been approved, what's the process for getting it merged? (Eagerly awaiting not having to have a custom install of incus!)

[sbromberger]

Erm. Why is lxc a dependency here? Incus doesn't require it (nor does it strictly require wget and dnsmasq...).

[dkwo]

where do you see wget?
as for lxc, it's surely needed at makedepends, isn't it? are you suggesting to remove lxc and dnsmasq from depends?
can you open pr?

[sbromberger]

I think wget may be a transitive dependency (possibly from lxc). I installed incus from scratch on a test void system with only cowsql and raft as dependencies (aside from the ones specified in https://linuxcontainers.org/incus/docs/main/installing/#installing). Specifically, one needs lxc-devel in makedepends but we don't need lxc.

I can try my hand at a PR but I'm unfamiliar with the config files for packaging so I might mess it up. Thanks.