K8s hosts get addresses on the internal network
Closed this issue · 2 comments
uablrek commented
K8s hosts must get addresses on the "cluster" network. The internal network has no vm-vm connectivity when user-space networking is used.
This is a serious bug and causes operation with kube-router to fail for instance (no BGP peering).
uablrek commented
The bug seems to be limuted to the kube-router overlays. A parameter to kubelet was not updated.
But the /etc/host for k8s clusters was not updated to the vm names resolves to a internal address and this must bt fixed!
uablrek commented
All usage of the internal net (192.168.0.0/24) is removed from k8s.
Note however that for user-space networking the host (and internet) can only be accessed via the internal net, so on the router VMs the default route should be;
ip ro add default via 192.168.0.250