Make LXC works

[tieong]

Hey,

I did some progress on this which you can see here https://github.com/tieong/proxmox-nixos/tree/fix/make-lxc-works (not mergeable needs to be rebased to be cleaner).

I can create the container with the pct command line succesfully however I did not manage to start it.

The command I used to create is the following:

# Note that to be able to download the container templates I had to first do this to be able to download from the web ui/pveam cli
mkdir -p /usr/share/doc/pve-manager/
wget http://download.proxmox.com/debian/proxmox-release-bookworm.gpg
mv proxmox-release-bookworm.gpg /usr/share/doc/pve-manager/trustedkeys.gpg
wget https://github.com/turnkeylinux/turnkey-keyring/raw/master/turnkey-release-keyring.gpg
gpg --no-default-keyring --keyring /usr/share/doc/pve-manager/trustedkeys.gpg --import turnkey-release-keyring.gpg


pveam update
pveam list

pct create 99999   --description alpine-template  local:vztmpl/alpine-3.21-default_20241217_amd64.tar.xz   -ostype alpine   -net0 name=eth0   -storage local

It complains about pve-container service missing, I see that we have some code for that here https://github.com/SaumonNet/proxmox-nixos/blob/main/modules/proxmox-ve/container.nix but doesn't gets created for some reason?

Does it only gets created if the lxc service starts successfully? I did try to start it manually but it seems it is missing some coreutils?

Finally the lxc config that we generate is currently incorrect, it generates something like this:

lxc.include = /nix/store/q3qsr9wp5vc8w4h9vhp24mhz462yv0ky-pve-container-5.2.2/share/lxc/config/common.conf

while it should be

/nix/store/q3qsr9wp5vc8w4h9vhp24mhz462yv0ky-pve-container-5.2.2/share/lxc/config/common.conf.d/01-pve.conf

This stems from this line https://github.com/proxmox/pve-container/blob/master/src/PVE/LXC.pm#L673 and needs to be patched.

Also I found that the lxc-start is not very talkative when failing, It was much more useful when using the following options:

/nix/store/sqyzbxxi74c7g5h11bbxc5xbpv8nqpar-lxc-6.0.4/bin/lxc-start -F -n 99999 --logfile failure --logpriority TRACE

[tieong]

I did work a little on this today.

The "failed to connect to monitor socket" is fixed by having the lxc-monitord service running.

One problem that I had when I kept creating, trying to start/stop(just using ctrl-c when it hanged) containers via pct was that the underlying loop device would never umount cleanly, process was stuck in a interruptible sleep, so I had to reboot the machine to recover to a clean state.

Not sure what causes this there is no clear error message. I noticed that they like to shell out a lot, you find a lot of call that looks like this:

	eval {
	    PVE::Tools::run_command(['umount', '-d', $mount_path]);
	};

Commands in that eval block don't crash the whole program if they fail, but it doesn't seems it shows why a particular command fails.

And looking at that run_command definition https://github.com/proxmox/pve-common/blob/master/src/PVE/Tools.pm#L428 there are /bin/bash hardcoded everywhere so they need to get replaced.

I think I saw some umount hardcoded as well.

Also need to have proper systemd services in container.nix for lxc.service and lxc-monitord.service

[tieong]

The problem I had in previous comment was related to a kernel bug and ZFS, I just downgraded my kernel and it works now.

See #142

It Lives !

[JulienMalka]

Congratulations!! Will review the PR soon