We do not test or support switching the CNI backend on a running cluster, in the same way we do not support switching CNIs or switching cluster/service CIDRs. I don't think the upstream CNI projects test or support this either, which is why they do not handle cleaning up the old interfaces when you switch.
You should have the CNI HelmChartConfig in place in the manifests directory on one of your nodes before starting the cluster for the first time so that things are deployed correctly the first time.
Originally posted by @brandond in #9338
Originally posted by @brandond in #9338