gpu_driver=none should clear GPU related blocked status

[Chris Johnston]

I have deployed containerd with gpu_driver set to auto. Due to a firewall I'm unable to download the nvidia apt keys. This placed the unit into a blocked state. Because I don't need the gpu currently, I set gpu_driver to none, expecting to leave the blocked state. The unit remained in a blocked state even though it no longer needs the keys.

1.26/stable charm

[George Kraft (cynerva)]

Thanks for the report. It looks like when the download failed, the charm set a containerd.nvidia.fetch_keys_failed flag[1], which causes the charm to enter blocked status. After you changed the GPU driver, the charm never cleared the flag.

As a workaround, you can manually clear the flag and run the update-status hook:

juju run --application containerd -- charms.reactive clear_flag containerd.nvidia.fetch_keys_failed
juju run --application containerd -- hooks/update-status

[1]: https://github.com/charmed-kubernetes/charm-containerd/blob/47c508ca2add6000649761364df322b6831e0d0d/reactive/containerd.py#L620