pty issue with ve 4 containers (converted from 3)

[PaulVM]

Problem still present for Centos 6.x (yum updates of initscripts package).

P.

 

[Ruben Waitz]

Please try this "semi optimal" work-around.

Add the following line to /etc/pve/lxc/XXXX.conf at your hostnode.

lxc.hook.stop: sh -c "pct mount ${LXC_NAME}; perl -pi -w -e 's/^(\/sbin\/start_udev)/# \$1/g' /var/lib/lxc/${LXC_NAME}/rootfs/etc/rc.d/rc.sysinit; pct unmount ${LXC_NAME}"

This hook will be executed after a VPS stop. The code mounts the VPS filesystem, does some regexp magic on rc.sysinit (commenting /sbin/start_udev) and unmounts the VPS filesystem.
I'm assuming the update of the initscripts only become active after a VPS is stopped and started from the hostnode (e.g. pct stop/start)
So entering the command "reboot" from within the VPS won't work. In my findings the VPS even doesn't come up.... I think "lxc.hook.stop" is used by Proxmox itself and we can't "enqueue" to it.

You can find the container-hooks here: https://linuxcontainers.org/lxc/manpages/man5/lxc.container.conf.5.html
Note that some hooks are executed on the hostnode (like 'pre-start') and some on the LXC (like 'start').

So if you use the latter one, the commands should be on the the LXC.
Example (non-working):

lxc.hook.start: sh -c "perl -pi -w -e 's/^(\/sbin\/start_udev)/# \$1/g' /etc/rc.d/rc.sysinit"

If someone succeeds, I'm glad to hear from you.

Regards,
Ruben

 

[Ruben Waitz]

A better solution:

Enter the container 101 (example) from the host command prompt:

root@pve01:~# pct enter 101

Inside the container enter:

[root@ct101 ~]# rpm -e --nodeps udev
[root@ct101 ~]# reboot

This will remove the udev package. Problem is solved but not sure what the side effects are.

Hope this helps!

 

[jazzl0ver]

The side effect is that /dev/shm is missing now!

Had to run the following commands inside the CT manually after creating it and disabling /sbin/start_udev:

# mknod -m 600 /dev/initctl p
# mkdir -m 1777 /dev/shm
# mount /dev/shm