Hi Vignesh,
(just realized the Kind config is named kind-extra-config as opposed to say a config.yaml in the gist above)
Regarding the host path provisioner you could use the one from Rancher as long as you are pointing it to the path mounted onto the kind worker container from the underlying GKE persistent disk. Maybe you wouldn’t need to make the hostPath changes in Helm if you can get the local provisioner to work in such a configuration.
Dgraph should safely terminate soon enough unless the underlying node shuts down rather abruptly SIGKILLing the process and not allowing it to do the cleanup. A similar project for AWS does just that - sends SIGTERM to the pods so they can exit cleanly as and when it gets the 2 minute spot shutdown notification(grace period would ideally be under 120 seconds to work on spot in which case).