I'm seeing runc run failed: unable to start container process: error during container init: error mounting "cgroup" to rootfs at "/sys/fs/cgroup": mount cgroup:/sys/fs/cgroup/elogind (via /proc/self/fd/6), flags: 0xf, data: elogind: invalid argument, which is not normal when running docker/podman containers
#Dagger oddly seems to be systemd dependent?
orchid inlet
Elogind is logind extracted from the systemd-logind project, it's part of guix
twilit kraken
Elogind is `logind` extracted from the `systemd-logind` project, it's part of gu...
When running locally, Dagger piggy backs on the docker CLI (docker engine, lima, podman, rancher, etc) to actually provision its engine. that's probably why you're seeing this.
what container runetime do you currently use?
I'd assume you're getting an error similar to this one: #1147312649810481204 message
orchid inlet
I have docker and podman installed and working
I'm not using systemd, so this is 99% some cgroups-adjacent annoyance
This is for sure the same issue as https://discord.com/channels/707636530424053791/1147312649810481204, but under different circumstances, they are having issues with lima/colima/rancher which likely use alpine and openrc
i'm using runit, where cgroups v2 appears to be working
error mounting "cgroup" to rootfs at "/sys/fs/cgroup": mount cgroup:/sys/fs/cgroup/elogind this is odd
as I already have a mount at /sys/fs/cgroup/elogind as cgroup /sys/fs/cgroup/elogind cgroup rw,relatime,name=elogind 0 0
This can be "fixed" on void linux / probably all runit distros by setting CGROUP_MODE=unified in /etc/rc.conf
I'm not sure why it would fail under hybrid and work under unified though
# Set the mode for cgroup mounts.
# hybrid: mount cgroup v1 under /sys/fs/cgroup and
# cgroup v2 under /sys/fs/cgroup/unified
# legacy: mount cgroup v1 /sys/fs/cgroup
# unified: mount cgroup v2 under /sys/fs/cgroup
is dagger trying to invoke docker rootless?
twilit kraken
is dagger trying to invoke docker rootless?
nope, not really
dagger just invokes docker locally and runs a privileged container that needs the correct mounts to run other containers
twilit kraken
glad you managed to get unblocked 🙌
twilit kraken
even more so, dagger on a fresh unified vm same distro works
iiiiiiiinteresting...
orchid inlet
it just does exec.Command right?
i might be able to attach to the dagger engine and work out which invocation is causing it, that is if i can get it to reproduce at all
twilit kraken
it just does `exec.Command` right?
yes, correct
twilit kraken
i might be able to attach to the dagger engine and work out which invocation is ...
this issue is not coming from the dagger binary itself. it's docker trying to spin containers
if you try to run docker-in-docker I'd assume you'll get the same error
if you look for that error in google, multiple users have gotten it with podman, docker, lima, etc in some specific kernels
orchid inlet
yeah i know its not dagger specific, i mean what docker command it's running to make it happen
twilit kraken
it's trying to run the dagger engine which depends on buildkit
if you run this docker run --rm --privileged registry.dagger.io/engine:v0.8.7 I'd assume you'll see it fail
orchid inlet
no thats the confusing part, after rebooting the issue has totally evaporated, even with hybrid cgroups
maybe I updated out of a bug or something, weird
twilit kraken
oh, I see... strange indeed
orchid inlet
annoying that there's no proper explanation and fix, but i guess its resolved
twilit kraken
annoying that there's no proper explanation and fix, but i guess its resolved
agree with u
resolving this thread 🙌
