refactor: run kubelet and etcd in system containerd

[dsseng]

This change allows us to ensure their contexts are protected by SELinux policy, as well as making CRI containerd only host containers managed by Kubernetes.

Signed-off-by: Dmitry Sharshakov [email protected]

[smira]

I'm not sure if we can do that, as system containerd runs from tmpfs, so all container state and images should be in memory.

[dsseng]

Well, etcd data is on a mount to where it belongs to be

[frezbo]

Well, etcd data is on a mount to where it belongs to be

but the uncompressed layers and other metadata will fillup tmpfs, just wasting memory

[dsseng]

yes, perhaps this solution is not really efficient since those images aren't tiny

[dsseng]

Idea: we could create a namespace belonging to the system containerd, but with snapshotter configured to store data on the /var partition? This should bring together best of both concepts.

[smira]

Idea: we could create a namespace belonging to the system containerd, but with snapshotter configured to store data on the /var partition? This should bring together best of both concepts.

that sounds interesting, if we can actually pull it through, but still it's kubelet/etcd only

[github-actions]

This PR is stale because it has been open 45 days with no activity.