Kubernetes: Apiserver fails to start inside container in 1.6

[tyranron]

Is this a BUG REPORT or FEATURE REQUEST? (choose one): BUG REPORT

Kubernetes version (use kubectl version):
Client: v1.6.1
Server: v1.6.1+coreos.0

Environment:

• Cloud provider or hardware configuration: Bare metal, 1 node (also tested for 3 nodes = 1 master + 2 workers)
• OS: CoreOS 1298.7.0
• Kernel: 4.9.16-coreos-r1
• Install tools: quay.io/coreos/hyperkube:v1.6.1_coreos.0 image as systemd kubelet.service + manifests in /etc/kubernetes/manifests for core components

The kubelet.service is:

# /etc/systemd/system/kubelet.service
[Unit]
Description=Kubelet service of Kubernetes

[Service]
# Mounting /etc/hosts is required for kubelet to be able to resolve
# nodes via its aliases (which may be not DNS routable).
Environment="RKT_RUN_ARGS= \
  --volume=etc-hosts,kind=host,source=/etc/hosts --mount volume=etc-hosts,target=/etc/hosts \
  --volume=var-log,kind=host,source=/var/log --mount volume=var-log,target=/var/log \
  --volume=dns,kind=host,source=/etc/resolv.conf --mount volume=dns,target=/etc/resolv.conf \
  --uuid-file-save=/var/run/kubelet-pod.uuid"
Environment=KUBELET_IMAGE_URL=quay.io/coreos/hyperkube
Environment=KUBELET_IMAGE_TAG=v1.6.1_coreos.0
ExecStartPre=/usr/bin/mkdir -p /etc/kubernetes/manifests
ExecStartPre=/usr/bin/mkdir -p /var/log/containers
ExecStartPre=-/usr/bin/rkt rm --uuid-file=/var/run/kubelet-pod.uuid
# Specifying all node labels here is required for kubelet
# to let it know its node state for correct recovery/restarts
# before sychronization with kube-apiserver happens.
ExecStart=/usr/lib/coreos/kubelet-wrapper \
  --port=10250 \
  --tls-cert-file=/etc/kubernetes/ssl/kubelet.crt \
  --tls-private-key-file=/etc/kubernetes/ssl/kubelet.key \
  --kubeconfig=/etc/kubernetes/kubeconfig-kubelet.yaml \
  --require-kubeconfig \
  --register-node=true \
  --allow-privileged=true \
  --pod-manifest-path=/etc/kubernetes/manifests \
  --hostname-override=n1.myhost.net \
  --cluster-dns=10.3.0.10 \
  --cluster-domain=cluster.local \
  --node-labels=role/apiserver=true,role/application=true \
  --network-plugin= \
  --sync-frequency=30s
ExecStop=-/usr/bin/rkt stop --uuid-file=/var/run/kubelet-pod.uuid
Restart=always
RestartSec=10

[Install]
WantedBy=multi-user.target

The /etc/kubernetes/manifests/kube-apiserver.yml is:

kind: Pod
apiVersion: v1
metadata:
  name: kube-apiserver
  namespace: kube-system
spec:
  hostNetwork: true
  containers:
    - name: kube-apiserver
      image: quay.io/coreos/hyperkube:v1.6.1_coreos.0
      command:
        - /hyperkube
        - apiserver
        - --bind-address=0.0.0.0
        - --secure-port=8443
        - --insecure-port=8081
        - --advertise-address=<public ip>
        - --external-hostname=n1.myhost.net
        - --client-ca-file=/etc/kubernetes/ssl/ca.crt
        - --tls-cert-file=/etc/kubernetes/ssl/apiserver.crt
        - --tls-private-key-file=/etc/kubernetes/ssl/apiserver.key
        - --service-account-key-file=/etc/kubernetes/ssl/apiserver.key
        - --service-cluster-ip-range=10.3.0.0/24
        - --etcd-servers=http://n1.myhost.net:4001
        - --etcd-cafile=/etc/ssl/etcd/ca.crt
        - --etcd-certfile=/etc/ssl/etcd/client.crt
        - --etcd-keyfile=/etc/ssl/etcd/client.key
        - --storage-backend=etcd2
        - --storage-media-type=application/json
        - --kubelet-https=true
        - --kubelet-certificate-authority=/etc/kubernetes/ssl/ca.crt
        - --kubelet-client-certificate=/etc/kubernetes/ssl/apiserver.crt
        - --kubelet-client-key=/etc/kubernetes/ssl/apiserver.key
        - --allow-privileged=true
        - --anonymous-auth=false
        - --admission-control=NamespaceLifecycle,LimitRanger,ServiceAccount,ResourceQuota
        - --runtime-config=extensions/v1beta1=true,extensions/v1beta1/thirdpartyresources=true
      ports:
        - name: https
          containerPort: 8443
          hostPort: 8443
        - name: local
          containerPort: 8081
          hostPort: 8081
      volumeMounts:
        - name: ssl-certs-kubernetes
          mountPath: /etc/kubernetes/ssl
          readOnly: true
        - name: ssl-certs-etcd
          mountPath: /etc/ssl/etcd
          readOnly: true
        - name: ssl-certs-host
          mountPath: /etc/ssl/certs
          readOnly: true
  volumes:
    - name: ssl-certs-kubernetes
      hostPath:
        path: /etc/kubernetes/ssl
    - name: ssl-certs-etcd
      hostPath:
        path: /etc/ssl/etcd
    - name: ssl-certs-host
      hostPath:
        path: /usr/share/ca-certificates

What happened:

kube-apiserver Pod starts and immediately exits with error (from docker logs):

F0408 09:01:03.289906       1 controller.go:128] Unable to perform initial IP allocation check: unable to refresh the service IP block: Get https://localhost:8443/api/v1/services: dial tcp: lookup localhost on 8.8.8.8:53: no such host

Somehow apiserver tries to resolve localhost via external DNS.

What you expected to happen:

kube-apiserver should start normally as it does on 1.5 Kubernetes versions.

How to reproduce it (as minimally and precisely as possible):

By booting up Kubernetes cluster on CoreOS but for v1.6.1 Kubernetes version.

Anything else we need to know:

As i've investigated the ClusterFirstWithHostNet value was added to dnsPolicy parameter of PodSpec in 1.6. It says:

To have DNS options set along with hostNetwork, you have to specify DNS policy explicitly to 'ClusterFirstWithHostNet'.

I tried to specify dnsPolicy: ClusterFirstWithHostNet explicitly, and had a different error:

F0408 08:57:52.675808       1 controller.go:128] Unable to perform initial IP allocation check: unable to refresh the service IP block: Get https://localhost:8443/api/v1/services: dial tcp: lookup localhost on 10.3.0.10:53: write udp <public ip>:37836->10.3.0.10:53: write: operation not permitted

Also, looked at kube-apiserver CLI options and did not found something related to be tuned up.

The KubeDNS in my installation is deployed as usual Deployment in kube-system namespace, not as a cluster addon in /etc/kubernetes/manifests dir.

[tyranron]

Any suggestions why apiserver tries to resolve localhost on 8.8.8.8?

[lorenz]

Have a look at your host's /etc/hosts and see if it contains localhost. This was the problem on my setup with exactly the same configuration, my /etc/hostscontained only the hostname of my machine and an IP but no localhost. When I added 127.0.0.1 localhost and restarted the apiserver it worked.

[tyranron]

Engenius! 🤦🤦🤦🤦

That's exactly the issue.

@lorenz thanks a lot!