While upgrade the TKGm cluster if the TKG_CUSTOM_IMAGE_REPOSITORY is set to incorrect repository, upgrade will fail with the first CP node creation.
We can see recreation of 1st CP nodes continuously, this is due to the incorrect repository. We can validate this by describing the pods in CP node.
TKGm - management/workload clusters.
This is due to the incorrect repository path to the pod images.
Edit the KCP of the CP nodes and correct the repository name.
kubectl get KubeadmControlPlane -A
kubectl edit KubeadmControlPlane kcpname -n namespace
Add the correct repository path.