LXC & Cinder: AttributeError: 'NoneType' object has no attribute 'rpartition'

[Radosław Piliszek]

When using LXC with volumes from Cinder, nova-compute logs [1] every minute.

Also, the instance cannot be started after having been stopped and fails on [2] (same error, different stack).

Tested using master (Victoria) on Ubuntu 18.04 (bionic) using devstack.

Note this only happens with LXC *and* Cinder.
Cinder uses the lvm backend.

[1] http://paste.openstack.org/show/793940/
[2] http://paste.openstack.org/show/793941/

[Radosław Piliszek (yoctozepto)]

Another issue is it does not make instance appear in ERROR state - it appears just as if the instance shut itself down.

[Balazs Gibizer (balazs-gibizer)]

As far as I know there is no LXC CI jobs so I would not be surprised if nova's LXC support is degraded over time. Could you paste your devstack / nova config so I can try to reproduce the issue?

[Balazs Gibizer (balazs-gibizer)]

Putting this incomplete now. Please set it back to New once the requested config is provided.

[Radosław Piliszek (yoctozepto)]

I did not want to make it too hard for myself so it's very barebones really:

[[local|localrc]]

ADMIN_PASSWORD=secret
DATABASE_PASSWORD=$ADMIN_PASSWORD
RABBIT_PASSWORD=$ADMIN_PASSWORD
SERVICE_PASSWORD=$ADMIN_PASSWORD

LIBVIRT_TYPE=lxc

[Radosław Piliszek (yoctozepto)]

I was actually looking at reviving the job, I talked with Sean Mooney some time ago about that. I even have https://review.opendev.org/676024 bookmarked. Currently I run the Ubuntu Bionic Cloud LXC image which gets me farthest in the process (if done cinder-less that is). I still could not get to the console (except for the logs) nor ssh but the instance does get alive, gets ip address and is later pingable. SSH seems to be getting set up but does not respond yet.

[Balazs Gibizer (balazs-gibizer)]

I don't think openstack server stop works at all without even cinder involved. [1] It seems that stop puts the server to SHUTOFF state but the container running happily on the host according to ps. After the instance is started up again with opentack server start as second container is started and both containers starts running in parallel.

[1] http://paste.openstack.org/show/793987/

[Balazs Gibizer (balazs-gibizer)]

After openstack server delete only the second container is deleted the first is left on the host. :/

[Balazs Gibizer (balazs-gibizer)]

Also now I can reproduce the original problem you reported. I get the same stacktrace when a server is booted with additional cinder volumes. So marking this bug as Confirmed.

[Radosław Piliszek (yoctozepto)]

Thanks, Balazs, for confirming. Indeed the LXC support needs some love...