CoreDNS deployment to be mounted with /etc/resolv.conf file from the host

[FSociety07]

Currently, CoreDNS deployment is not mounted with the host's /etc/resolv.conf file. This prevents the CoreDNS from referring to the host's nameserver for DNS queries it cant resolve.

The corefile as defined in the coredns CM contains
forward . UPSTREAMNAMESERVER { max_concurrent 1000 }
https://github.com/coredns/deployment/blob/master/kubernetes/coredns.yaml.sed

The "UPSTREAMNAMESERVER " placeholder typically gets converted to /etc/resolv.conf during kubeadm init. But, it is no use since the file is empty and not mounted from the host file.

Would be better if the resolv.conf file could be mounted to the pod from the host via hostPath volume mount.

[neolit123]

my only concern is backwards compatibility.
if kubeadm starts doing that for users that don't want it, it would be a breaking change. so my vote for this use case goes for users deploying their own coredns and skipping the kubeadm default one.

@SataQiu @pacoxu @carlory @HirazawaUi
your thoughts?

[pacoxu]

IIRC, we use dnsPolicy: default for coredns deployment.

dnsPolicy: default means using the host configuration from /etc/resolv.conf by default, as this depends on the kubelet configuration:

  --resolv-conf string                                       Resolver configuration file used as the basis for the container DNS resolution configuration. (default "/etc/resolv.conf") (DEPRECATED: This parameter should be set via the config file specified by the Kubelet's --config flag. See https://kubernetes.io/docs/tasks/administer-cluster/kubelet-config-file/ for more information.)

Do you configurare a different path for kubelet --resolv-conf or resolvConf in /var/lib/kubelet/config.yaml?

[SataQiu]

In my local environment, the /etc/resolv.conf of coredns pod is not empty. Its content is exactly the content of the /etc/resolv.conf of the node on which it runs.

kubelet will automatically load the /etc/resolv.conf (/run/systemd/resolve/resolv.conf for systemd-resolved) of the node for pod when the pod dnsPolicy is Default.

https://github.com/kubernetes/kubernetes/blob/73014647f9bdd7186b5fe121326b961d6521c162/pkg/kubelet/network/dns/dns.go#L384-L448

[FSociety07]

Checked in a fresh k8s deployment.
the resolv.conf file of the coredns pod is same as the node's.
dnsPolicy: Default option is there in the coredns deployment.

Can check from a debugger container since coredns container does not have a shell.
kubectl debug -it <pod> -n kube-system --image=busybox:1.28 --target=coredns

Issue is specific to the node/environment. Not an issue with kubeadm.

thanks @neolit123 @pacoxu @SataQiu