Data corruption issue was found in etcd v3.5.0 release that was shipped with 1.22 Kubernetes release. Please read up-to-date production recommendations for etcd. Affects etcd 3.5.0 to 3.5.2. We'll go straight to 3.5.4, so we should be fine.
We cannot update CNI to the latest version, because flannel has been extracted to its own library: https://github.com/flannel-io/cni-plugin. Maybe we should extract both of them and make sure the binaries are in place?
Also upgrades:
1.6.12
1.1.4
3.5.6
0.9.1
Known dangers:
TODO: