nvidia-container-runtime package versions cause install failure

[Peter Jose De Sousa]

When deploying a CDK on a GPU enabled machine, upon removing containerd, and adding Docker the install hook will fail due to left-over packages.

Error here: https://pastebin.canonical.com/p/BjDntBxGXh/

In addtion containerd leaves "/etc/apt/sources.list.d/nvidia-container-runtime.list" over after removal.

Steps to reproduce:

1. Deploy CDK using latest stable on GPU enabled machine, upgrade containerd to edge
2. Let deployment settle.
3. Remove containerd
4. Allow to settle
5. Add docker runtime, install should fail with downgrade error

Expected Behaviour:

Docker installs, downgrading package; or containerd removes all related packages upon containerd `relation-removed`.

Actual Behaviour:

Docker fails to install due to package downgrade warning.

Cheers,

Peter

[Peter Jose De Sousa (pjds)]

Current workaround I have for this is to run: `sudo apt remove nvidia-container*`

[Joseph Borg (joeborg)]

https://github.com/charmed-kubernetes/charm-containerd/pull/16

[Kevin W Monroe (kwmonroe)]

Fixes cherry picked to stable:

https://github.com/charmed-kubernetes/charm-containerd/commit/40423626f535e8f8298531ee6be3698a0cae5146

No stable fixes required for charm-docker as this issue occurred because containerd was not cleaning itself properly.