Reregistering a worker node
On Equinix Metal (Packet) if a cluster is updated from
v0.6.0, it also installs Packet’s Cloud Controller Manager (CCM). The nodes added before
v0.6.0 were configured with BGP using Terraform. From
v0.6.0 onwards they are configured to use CCM for BGP setup. During the update, the older nodes won’t see BGP resource removal, but it is possible that someone can disable the BGP from the Equinix Metal console. To make the nodes resilient against such human errors, follow the next steps.
NOTE: SSH into a given node in a separate console upfront, before starting to follow steps.
This step ensures that you don’t see any abrupt changes. Any workloads running on this node are evicted and scheduled to other nodes. The node is marked as unschedulable after running this command.
export nodename="" kubectl drain --ignore-daemonsets $nodename
Delete the node object:
kubectl delete node $nodename
SSH into the node:
sudo systemctl restart kubelet