Kubernetesの基本を学ぶ

セットアップ

Kubernetes v1.15 のドキュメントは積極的にメンテナンスされていません。現在表示されているバージョンはスナップショットです。最新のドキュメントはこちらです: 最新バージョン

Edit This Page

kubeadmのトラブルシューティング

As with any program, you might run into an error installing or running kubeadm. This page lists some common failure scenarios and have provided steps that can help you understand and fix the problem.

If your problem is not listed below, please follow the following steps:

インストール中にebtablesもしくは他の似たような実行プログラムが見つからない

If you see the following warnings while running kubeadm init

[preflight] WARNING: ebtables not found in system path
[preflight] WARNING: ethtool not found in system path

Then you may be missing ebtables, ethtool or a similar executable on your node. You can install them with the following commands:

インストール中にkubeadmがコントロールプレーンを待ち続けて止まる

If you notice that kubeadm init hangs after printing out the following line:

[apiclient] Created API client, waiting for the control plane to become ready

This may be caused by a number of problems. The most common are:

There are two common ways to fix the cgroup driver problem:

  1. Install Docker again following instructions here.
  2. Change the kubelet config to match the Docker cgroup driver manually, you can refer to Configure cgroup driver used by kubelet on Master Node for detailed instructions.

管理コンテナを削除する時にkubeadmが止まる

The following could happen if Docker halts and does not remove any Kubernetes-managed containers:

sudo kubeadm reset
[preflight] Running pre-flight checks
[reset] Stopping the kubelet service
[reset] Unmounting mounted directories in "/var/lib/kubelet"
[reset] Removing kubernetes-managed containers
(block)

A possible solution is to restart the Docker service and then re-run kubeadm reset:

sudo systemctl restart docker.service
sudo kubeadm reset

Inspecting the logs for docker may also be useful:

journalctl -ul docker

Podの状態がRunContainerError、CrashLoopBackOff、またはError

Right after kubeadm init there should not be any pods in these states.

coredns(もしくはkube-dns)がPending状態でスタックする

This is expected and part of the design. kubeadm is network provider-agnostic, so the admin should install the pod network solution of choice. You have to install a Pod Network before CoreDNS may deployed fully. Hence the Pending state before the network is set up.

HostPortサービスが動かない

The HostPort and HostIP functionality is available depending on your Pod Network provider. Please contact the author of the Pod Network solution to find out whether HostPort and HostIP functionality are available.

Calico, Canal, and Flannel CNI providers are verified to support HostPort.

For more information, see the CNI portmap documentation.

If your network provider does not support the portmap CNI plugin, you may need to use the NodePort feature of services or use HostNetwork=true.

サービスIP経由でPodにアクセスすることができない

TLS証明書のエラー

The following error indicates a possible certificate mismatch.

# kubectl get pods
Unable to connect to the server: x509: certificate signed by unknown authority (possibly because of "crypto/rsa: verification error" while trying to verify candidate authority certificate "kubernetes")

Or set it to the default KUBECONFIG location:

  export KUBECONFIG=/etc/kubernetes/admin.conf

Vagrant内でPodネットワークとしてflannelを使用する時のデフォルトNIC

The following error might indicate that something was wrong in the pod network:

Error from server (NotFound): the server could not find the requested resource

Vagrant typically assigns two interfaces to all VMs. The first, for which all hosts are assigned the IP address 10.0.2.15, is for external traffic that gets NATed.

This may lead to problems with flannel, which defaults to the first interface on a host. This leads to all hosts thinking they have the same public IP address. To prevent this, pass the --iface eth1 flag to flannel so that the second interface is chosen.

公開されていないIPがコンテナに使われている

In some situations kubectl logs and kubectl run commands may return with the following errors in an otherwise functional cluster:

Error from server: Get https://10.19.0.41:10250/containerLogs/default/mysql-ddc65b868-glc5m/mysql: dial tcp 10.19.0.41:10250: getsockopt: no route to host

Use ip addr show to check for this scenario instead of ifconfig because ifconfig will not display the offending alias IP address. Alternatively an API endpoint specific to Digital Ocean allows to query for the anchor IP from the droplet:

  curl http://169.254.169.254/metadata/v1/interfaces/public/0/anchor_ipv4/address

The workaround is to tell kubelet which IP to use using --node-ip. When using Digital Ocean, it can be the public one (assigned to eth0) or the private one (assigned to eth1) should you want to use the optional private network. The KubeletExtraArgs section of the kubeadm NodeRegistrationOptions structure can be used for this.

Then restart kubelet:

  systemctl daemon-reload
  systemctl restart kubelet

corednsのPodがCrashLoopBackOffもしくはError状態になる

If you have nodes that are running SELinux with an older version of Docker you might experience a scenario where the coredns pods are not starting. To solve that you can try one of the following options:

Another cause for CoreDNS to have CrashLoopBackOff is when a CoreDNS Pod deployed in Kubernetes detects a loop. A number of workarounds are available to avoid Kubernetes trying to restart the CoreDNS Pod every time CoreDNS detects the loop and exits.

警告: Disabling SELinux or setting allowPrivilegeEscalation to true can compromise the security of your cluster.

etcdのpodが継続的に再起動する

If you encounter the following error:

rpc error: code = 2 desc = oci runtime error: exec failed: container_linux.go:247: starting container process caused "process_linux.go:110: decoding init error from pipe caused \"read parent: connection reset by peer\""

this issue appears if you run CentOS 7 with Docker 1.13.1.84. This version of Docker can prevent the kubelet from executing into the etcd container.

To work around the issue, choose one of these options:

フィードバック