installing cluster with kubeadm about digitalocean-cloud-controller-manager HOT 12 CLOSED

Ahh, I misunderstood. I thought it would try to match the node's internal ip with the droplets private ip (for example), not check if the name is the same as the private ip :)

Btw, my error came from line 269:
https://github.com/digitalocean/digitalocean-cloud-controller-manager/blob/master/do/loadbalancers.go#L269

So it didn't get to the section to match addresses at all.

from digitalocean-cloud-controller-manager.

After updating kubelet config you need to

sudo systemctl daemon-reload
sudo systemctl kubelet restart

Can you try that and let me know if that works?

from digitalocean-cloud-controller-manager.

Hi, yeah, I tried that a few times, no success...

(also I could see in systemctl status kubelet that --cloud-provider external was in use...

from digitalocean-cloud-controller-manager.

The problem here is that the name of your droplets don't match the name of your worker nodes. This is a requirement as mentioned here

from digitalocean-cloud-controller-manager.

Okay, I'll give it a try...

It is important that the node name on Kubernetes matches either the droplet name, private ipv4 ip or the public ipv4 ip, otherwise cloud controller manager cannot find the corresponding droplet to nodes.

But maybe the readme should be updated then?

Also, can it be installed "after" the fact (after kubernetes is in use), or do I need to set this up before starting the kubelet the first time?

from digitalocean-cloud-controller-manager.

Good question, the flags need to be set on the initial registration. You can do it on an existing cluster but you need to delete the node object before starting the kubelet back up.

from digitalocean-cloud-controller-manager.

Ok, it works :)

But the readme is still wrong... Or there is a bug:

Kubernetes node names must match the droplet name, private ipv4 ip or public ipv4 ip
7190ad6

from digitalocean-cloud-controller-manager.

Sorry, I don't follow. What is wrong about it?

from digitalocean-cloud-controller-manager.

the readme says the droplet name doesn't need to match the node name, it sais, name, public ip or private ip.

To me that means that ccm can match the droplet based on public/private ip, and the name is not important anymore?

from digitalocean-cloud-controller-manager.

Kubernetes node names must match the droplet name, private ipv4 ip or public ipv4 ip

To me, that implies the Kubernetes node name (usually the hostname of the VM) has to match the actual name of the droplet OR the IPv4 address OR the IPv6 address. For example, if the droplet name is worker01 with private IP 10.11.12.13 and public IP 1.2.3.4 then you can name the Kubernetes node either worker01, 10.11.12.13 or 1.2.3.4.

from digitalocean-cloud-controller-manager.

Yeah, that's what I thought to, in other words, as long as private/public ip can match, it doesn't matter what the node name is?

But that doesn't work for the load balancer, it still requires the node name to match the droplet name...

from digitalocean-cloud-controller-manager.

hmmm... that should work based on the code here. Can you show me a cluster that uses private/public IPs for the node name and the LoadBalancer is not adding backends?

from digitalocean-cloud-controller-manager.