Pods may have requirements for the node (Longhorn needs open-iscsi, …). When I deploy nodes with a node driver, how can I fulfill them? Can I modify all nodes somehow (installing packages, configuring firewall, …)? Just on installation with cloudInit? Or are there other possibilities with Rancher?
my own hints/guesses/…:
Rancher develops https://k3os.io, which can be configured from kubernetes.
I guess, for other cases, there is no access from rancher. So probably use a configuratoin tool like ansible, salt, chef and maybe detect the inventoiry automatically.
I found a post doing configuration with ansible, and showing an alternative with scripts from cloud-init.
An other possibility could be in k8s itself:
Create a DeamonSet with a pod containing
- an init container with necessary privileges to adatp the host
- an app container running rancher/pause
The init container would run on creation (of node or DeaemonCreation) and on reboot. So it must be idempotent (run several times and keep the same state)
Idea from https://github.com/kubernetes/kubernetes/issues/64623
I forgot the link for "alternative with script from clould-init*: https://vitobotta.com/2020/01/17/secure-kubernetes-hetzner-cloud-rancher-node-driver/