nixos-ha-kubernetes
nixos-ha-kubernetes copied to clipboard
DNS is not highly-available
Currently, we run one CoreDNS service per worker
node (good), but the containers' resolv.conf
points only to the host node itself as the name server (bad):
https://github.com/justinas/nixos-ha-kubernetes/blob/f9d01ef37174de4ac4e9537fa6f8508696a4f22c/modules/worker/coredns.nix#L39
If Kubelet is alive, but CoreDNS is dead on the node, containers running on that node will fail to resolve names.
Looks like no easy way to fix this because services.kubernetes.kubelet.clusterDns
only accepts string