Disabling or delaying leader elections

[rombert]

I am running the nfs-client-provisioner in a simple self-hosted cluster. I am trying to narrow down excessive control plane disk writes, and traced lots of them back to etcd writing to /registry/services/endpoints/FOO/cluster.local-nfs-client-nfs-client-provisioner.

I got 896 writes to that key in 30 minutes, so it looks like once very two seconds, which is quite a lot ( actually half of the current writes, after disabling leader election for kube-scheduler and kube-controller-manager.

I tried to find a way of disabling or tweaking the leader elections timeout but did not find a way to do so. It would be great if this could be offered for simple deployments that don't need the redundancy of multiple replicas.

[rombert]

FWIW I would be willing to come up with a PR, given that we can agree on what options to expose and how. There are a wealth of options exposed by https://github.com/kubernetes-sigs/sig-storage-lib-external-provisioner/blob/203b2c9cdf9c44e504af7d8a15d9df6642cd9ea5/controller/controller.go#L623-L647 , which I think is what the nfs client provisioner users.

Thoughts?

[kmova]

@rombert - that would be great.

There was a similar functionality added to another provisioner via this PR. kubernetes-sigs/nfs-ganesha-server-and-external-provisioner#11

[rombert]

Thanks @kmova . I'll try and submit a similar PR with a single change - disabling leader election.