Reference Documentation

Design docs, concept definitions, and references for APIs and CLIs.

Documentation for Kubernetes v1.8 is no longer actively maintained. The version you are currently viewing is a static snapshot. For up-to-date documentation, see the latest version.

Edit This Page

kubeadm Setup Tool Reference Guide

This document provides information on how to use kubeadm’s advanced options.

Running kubeadm init bootstraps a Kubernetes master node. This consists of the following steps:

  1. kubeadm runs a series of pre-flight checks to validate the system state before making changes. Some checks only trigger warnings, others are considered errors and will exit kubeadm until the problem is corrected or the user specifies --skip-preflight-checks.

  2. kubeadm generates a token that additional nodes can use to register themselves with the master in future. Optionally, the user can provide a token via --token, as described in the section on managing tokens below.

  3. kubeadm generates a self-signed CA to provision identities for each component (including nodes) in the cluster. It also generates client certificates to be used by various components. If the user has provided their own CA by dropping it in the cert directory configured via --cert-dir (/etc/kubernetes/pki by default) this step is skipped as described in the section on using custom certificates.

  4. kubeadm writes kubeconfig files in /etc/kubernetes/ for the kubelet, the controller-manager and the scheduler to use to connect to the API server, each one with their respective identities, as well as an additional kubeconfig file for administration.

  5. kubeadm generates static Pod manifests for the API server, controller manager and scheduler; in case an external etcd is not provided, an additional static Pod manifest will be generated for etcd.

    Static Pod manifests are written in /etc/kubernetes/manifests; the kubelet watches this directory for Pods to create on startup, as described in the section about kubelet drop-in.

    Once control plane Pods are up and running kubeadm init sequence can continue.

  6. kubeadm “labels” and “taints” the master node so that only control plane components will run there.

  7. kubeadm makes all the necessary configurations for allowing node joining with the Bootstrap Tokens and TLS Bootstrap mechanism:

    • Write a ConfigMap for making available all the information required for joining and set up related RBAC access rules.

    • Ensure access to the CSR signing API for bootstrap tokens.

    • Configure auto approval for new CSR requests.

    See Securing your installation for hardening.

  8. kubeadm installs add-on components via the API server. Right now this is the internal DNS server and the kube-proxy DaemonSet.

  9. If kubeadm init is invoked with the alpha self-hosting feature enabled, (--feature-gates=SelfHosting=true), the static Pod based control plane will be transformed into a self-hosted control plane.

Running kubeadm join on each node in the cluster consists of the following steps:

  1. kubeadm downloads necessary cluster information from the API server. By default, it uses the bootstrap token and the CA key hash to verify the authenticity of that data. The root CA can also be discovered directly via a file or URL.

  2. Once the cluster information are known, kubelet can start the TLS bootstrapping process (in v.1.7 this step was managed by kubeadm).

    The TLS bootstrap uses the shared token to temporarily authenticate with the Kubernetes Master to submit a certificate signing request (CSR); by default the control plane will sign this CSR request automatically.

  3. Finally, kubeadm will configure the local kubelet to connect to the API server with the definitive identity assigned to the node.


Fields that support multiple values do so either with comma separation, or by specifying the flag multiple times.

The kubeadm command line interface is currently in beta. We are aiming to not break any scripted use of the main kubeadm init and kubeadm join. Exceptions to this are documented below.

kubeadm init

It is usually sufficient to run kubeadm init without any flags, but in some cases you might like to override the default behaviour. Here we specify all the flags that can be used to customise the Kubernetes installation.

Options for kubeadm init:

kubeadm join

When joining a kubeadm initialized cluster, we need to establish bidirectional trust. This is split into discovery (having the Node trust the Kubernetes master) and TLS bootstrap (having the Kubernetes master trust the Node).

There are two main schemes for discovery:

Only one form can be used. If the discovery information is loaded from a URL, HTTPS must be used and the host installed CA bundle is used to verify the connection. For details on the security tradeoffs of these mechanisms, see the security model section below.

The TLS bootstrap mechanism is also driven via a shared token. This is used to temporarily authenticate with the Kubernetes master to submit a certificate signing request (CSR) for a locally created key pair. By default kubeadm will set up the Kubernetes master to automatically approve these signing requests. This token is passed in with the --tls-bootstrap-token abcdef.1234567890abcdef flag.

Often times the same token is used for both parts. In this case, the --token flag can be used instead of specifying the each token individually.

Here’s an example on how to use it:

kubeadm join --token=abcdef.1234567890abcdef --discovery-token-ca-cert-hash sha256:1234..cdef

Options for kubeadm join:

kubeadm completion

Output shell completion code for the specified shell (bash or zsh).

kubeadm config

Kubeadm v1.8.0+ automatically creates a ConfigMap with all the parameters used during kubeadm init.

If you initialized your cluster using kubeadm v1.7.x or lower, you must use the kubeadm config upload command to create this ConfigMap in order for kubeadm upgrade to be able to configure your upgraded cluster correctly.

kubeadm reset

Reverts any changes made to this host by kubeadm init or kubeadm join.

kubeadm token

Manage tokens on a running cluster. See managing tokens below for further details.

kubeadm alpha phases

WARNING: While kubeadm command line interface is in beta, commands under this entry is still considered alpha and may change in future versions.

kubeadm phase introduces a set of kubeadm CLI commands allowing to invoke individually each phase of the kubeadm init sequence; phases provide a reusable and composable API/toolbox for building your own automated cluster installer.

Options for kubeadm phases:

Each kubeadm phase exposes a subset of relevant options from kubeadm init.

Using kubeadm with a configuration file

WARNING: While kubeadm command line interface is in beta, the config file is still considered alpha and may change in future versions.

It’s possible to configure kubeadm with a configuration file instead of command line flags, and some more advanced features may only be available as configuration file options. This file is passed in to the --config option on both kubeadm init and kubeadm join.

Sample Master Configuration

kind: MasterConfiguration
  advertiseAddress: <address|string>
  bindPort: <int>
  - <endpoint1|string>
  - <endpoint2|string>
  caFile: <path|string>
  certFile: <path|string>
  keyFile: <path|string>
  dataDir: <path|string>
    <argument>: <value|string>
    <argument>: <value|string>
  image: <string>
  dnsDomain: <string>
  serviceSubnet: <cidr>
  podSubnet: <cidr>
kubernetesVersion: <string>
cloudProvider: <string>
nodeName: <string>
- <authorizationMode1|string>
- <authorizationMode2|string>
token: <string>
tokenTTL: <time duration>
selfHosted: <bool>
  <argument>: <value|string>
  <argument>: <value|string>
  <argument>: <value|string>
  <argument>: <value|string>
  <argument>: <value|string>
  <argument>: <value|string>
- <name1|string>
- <name2|string>
certificatesDir: <string>
imageRepository: <string>
unifiedControlPlaneImage: <string>
  <feature>: <bool>
  <feature>: <bool>

In addition, if authorizationMode is set to ABAC, you should write the config to /etc/kubernetes/abac_policy.json. However, if authorizationMode is set to Webhook, you should write the config to /etc/kubernetes/webhook_authz.conf.

Sample Node Configuration

kind: NodeConfiguration
caCertPath: <path|string>
discoveryFile: <path|string>
discoveryToken: <string>
- <address|string>
- <address|string>
nodeName: <string>
tlsBootstrapToken: <string>
token: <string>
- <SHA-256 hash|string>
- <SHA-256 hash|string>
discoveryTokenUnsafeSkipCAVerification: <bool>

Securing your installation even more

The defaults for kubeadm may not work for everyone. This section documents how to tighten up a kubeadm install at the cost of some usability.

Turning off auto-approval of Node Client Certificates

By default, there is a CSR auto-approver enabled that basically approves any client certificate request for a kubelet when a Bootstrap Token was used when authenticating. If you don’t want the cluster to automatically approve kubelet client certs, you can turn it off by executing this command:

$ kubectl delete clusterrole kubeadm:node-autoapprove-bootstrap

After that, kubeadm join will block until the admin has manually approved the CSR in flight:

$ kubectl get csr
NAME                                                   AGE       REQUESTOR                 CONDITION
node-csr-c69HXe7aYcqkS1bKmH4faEnHAWxn6i2bHZ2mD04jZyQ   18s       system:bootstrap:878f07   Pending

$ kubectl certificate approve node-csr-c69HXe7aYcqkS1bKmH4faEnHAWxn6i2bHZ2mD04jZyQ
certificatesigningrequest "node-csr-c69HXe7aYcqkS1bKmH4faEnHAWxn6i2bHZ2mD04jZyQ" approved

$ kubectl get csr
NAME                                                   AGE       REQUESTOR                 CONDITION
node-csr-c69HXe7aYcqkS1bKmH4faEnHAWxn6i2bHZ2mD04jZyQ   1m        system:bootstrap:878f07   Approved,Issued

Only after kubectl certificate approve has been run, kubeadm join can proceed.

Turning off public access to the cluster-info ConfigMap

In order to achieve the joining flow using the token as the only piece of validation information, a public ConfigMap with some data needed for validation of the master’s identity is exposed publicly by default. While there is no private data in this ConfigMap, some users are sensitive and wish to turn it off regardless. Doing so will disable the ability to use the --discovery-token flag of the kubeadm join flow. Here are the steps to do so:

Fetch the cluster-info file from the API Server:

$ kubectl -n kube-public get cm cluster-info -o yaml | grep "kubeconfig:" -A11 | grep "apiVersion" -A10 | sed "s/    //" | tee cluster-info.yaml
apiVersion: v1
- cluster:
    certificate-authority-data: <ca-cert>
    server: https://<ip>:<port>
  name: ""
contexts: []
current-context: ""
kind: Config
preferences: {}
users: []

You can then use the cluster-info.yaml file as an argument to kubeadm join --discovery-file.

Turning off public access to the cluster-info ConfigMap:

$ kubectl -n kube-public delete rolebinding kubeadm:bootstrap-signer-clusterinfo

These commands should be run after kubeadm init but before kubeadm join.

Managing Tokens

You can use the kubeadm tool to manage tokens on a running cluster. It will automatically grab the default admin credentials on a master from a kubeadm created cluster (/etc/kubernetes/admin.conf). You can specify an alternate kubeconfig file for credentials with the --kubeconfig to the following commands.

For the gory details on how the tokens are implemented (including managing them outside of kubeadm) see the Bootstrap Token docs.

Automating kubeadm

Rather than copying the token you obtained from kubeadm init to each node, as in the basic kubeadm tutorial, you can parallelize the token distribution for easier automation. To implement this automation, you must know the IP address that the master will have after it is started.

  1. Generate a token. This token must have the form <6 character string>.<16 character string>. More formally, it must match the regex: [a-z0-9]{6}\.[a-z0-9]{16}.

    kubeadm can generate a token for you:

    kubeadm token generate
  2. Start both the master node and the worker nodes concurrently with this token. As they come up they should find each other and form the cluster. The same --token argument can be used on both kubeadm init and kubeadm join.

Once the cluster is up, you can grab the admin credentials from the master node at /etc/kubernetes/admin.conf and use that to talk to the cluster.

Note that this style of bootstrap has some relaxed security guarantees because it does not allow the root CA hash to be validated with --discovery-token-ca-cert-hash (since it’s not generated when the nodes are provisioned). For details, see the security model.

Security model

The kubeadm discovery system has several options, each with security tradeoffs. The right method for your environment depends on how you provision nodes and the security expectations you have about your network and node lifecycles.

Token-based discovery with CA pinning

This is the default mode in Kubernetes 1.8. In this mode, kubeadm downloads the cluster configuration (including root CA) and validates it using the token as well as validating that the root CA public key matches the provided hash and that the API server certificate is valid under the root CA.

Example kubeadm join command:



Token-based discovery without CA pinning

This was the default in Kubernetes 1.7 and earlier, but comes with some important caveats. This mode relies only on the symmetric token to sign (HMAC-SHA256) the discovery information that establishes the root of trust for the master. It’s still possible in Kubernetes 1.8 and above using the --discovery-token-unsafe-skip-ca-verification flag, but you should consider using one of the other modes if possible.

Example kubeadm join command:



File or HTTPS-based discovery

This provides an out-of-band way to establish a root of trust between the master and bootstrapping nodes. Consider using this mode if you are building automated provisioning using kubeadm.

Example kubeadm join commands:



Use kubeadm with other CRI runtimes

Since Kubernetes 1.6 release, Kubernetes container runtimes have been transferred to using CRI by default. Currently, the built-in container runtime is Docker which is enabled by built-in dockershim in kubelet.

Using other CRI based runtimes with kubeadm is very simple, and currently supported runtimes are:

After you have successfully installed kubeadm and kubelet, please follow these two steps:

  1. Install runtime shim on every node. You will need to follow the installation document in the runtime shim project listing above.

  2. Configure kubelet to use remote CRI runtime. Please remember to change RUNTIME_ENDPOINT to your own value like /var/run/{your_runtime}.sock:

$ cat > /etc/systemd/system/kubelet.service.d/20-cri.conf <<EOF
Environment="KUBELET_EXTRA_ARGS=--container-runtime=remote --container-runtime-endpoint=$RUNTIME_ENDPOINT --feature-gates=AllAlpha=true"
$ systemctl daemon-reload

Now kubelet is ready to use the specified CRI runtime, and you can continue with kubeadm init and kubeadm join workflow to deploy Kubernetes cluster.

Using custom images

By default, kubeadm will pull images from, unless requested kubernetes version is a ci version; in this case will be used.

This behaviour can be overridden by using kubeadm with a configuration file. Allowed customization are:

Running kubeadm without an internet connection

All of the control plane components run in Pods started by the kubelet and the following images are required for the cluster works will be automatically pulled by the kubelet if they don’t exist locally while kubeadm init is initializing your master:

Image Name v1.7 release branch version v1.8 release branch version${ARCH} v1.7.x v1.8.x${ARCH} v1.7.x v1.8.x${ARCH} v1.7.x v1.8.x${ARCH} v1.7.x v1.8.x${ARCH} 3.0.17 3.0.17${ARCH} 3.0 3.0${ARCH} 1.14.4 1.14.4${ARCH} 1.14.4 1.14.4${ARCH} 1.14.4 1.14.4

Here v1.7.x means the “latest patch release of the v1.7 branch”.

${ARCH} can be one of: amd64, arm, arm64, ppc64le or s390x.

Managing the kubeadm drop-in file for the kubelet

The kubeadm deb package ships with configuration for how the kubelet should be run. Note that the kubeadm CLI command will never touch this drop-in file. This drop-in file belongs to the kubeadm deb/rpm package.

This is what it looks like in v1.7:

Environment="KUBELET_KUBECONFIG_ARGS=--kubeconfig=/etc/kubernetes/kubelet.conf --require-kubeconfig=true"
Environment="KUBELET_SYSTEM_PODS_ARGS=--pod-manifest-path=/etc/kubernetes/manifests --allow-privileged=true"
Environment="KUBELET_NETWORK_ARGS=--network-plugin=cni --cni-conf-dir=/etc/cni/net.d --cni-bin-dir=/opt/cni/bin"
Environment="KUBELET_DNS_ARGS=--cluster-dns= --cluster-domain=cluster.local"
Environment="KUBELET_AUTHZ_ARGS=--authorization-mode=Webhook --client-ca-file=/etc/kubernetes/pki/ca.crt"

A breakdown of what/why:

Cloud provider integrations (experimental)

Enabling specific cloud providers is a common request. This currently requires manual configuration and is therefore not yet fully supported. If you wish to do so, edit the kubeadm drop-in for the kubelet service (/etc/systemd/system/kubelet.service.d/10-kubeadm.conf) on all nodes, including the master. If your cloud provider requires any extra packages installed on the host, for example for volume mounting/unmounting, install those packages.

Specify the --cloud-provider flag for the kubelet and set it to the cloud of your choice. If your cloud provider requires a configuration file, create the file /etc/kubernetes/cloud-config on every node. The exact format and content of that file depends on the requirements imposed by your cloud provider. If you use the /etc/kubernetes/cloud-config file, you must append it to the kubelet arguments as follows: --cloud-config=/etc/kubernetes/cloud-config

Note that there is most likely other per-provider configuration that may be needed (IAM roles for AWS) that is currently underdocumented.

Next, specify the cloud provider in the kubeadm config file. Create a file called kubeadm.conf with the following contents:

kind: MasterConfiguration
cloudProvider: <cloud provider>

Lastly, run kubeadm init --config=kubeadm.conf to bootstrap your cluster with the cloud provider.

This workflow is not yet fully supported, however we hope to make it extremely easy to spin up clusters with cloud providers in the future. (See this proposal for more information) The Kubelet Dynamic Settings feature may also help to fully automate this process in the future.

Environment variables

Note: These environment variables are deprecated and will stop functioning in v1.8!

There are some environment variables that modify the way that kubeadm works. Most users will have no need to set these. These environment variables are a short-term solution, eventually they will be integrated in the kubeadm configuration file.

Variable Default Description
KUBE_KUBERNETES_DIR /etc/kubernetes Where most configuration files are written to and read from
KUBE_HYPERKUBE_IMAGE   If set, use a single hyperkube image with this name. If not set, individual images per server component will be used.
KUBE_ETCD_IMAGE<arch>:3.0.17 The etcd container image to use.
KUBE_REPO_PREFIX The image prefix for all images that are used.

If KUBE_KUBERNETES_DIR is specified, you may need to rewrite the arguments of the kubelet. (e.g. –kubeconfig, –pod-manifest-path)

If KUBE_REPO_PREFIX is specified, you may need to set the kubelet flag --pod-infra-container-image which specifies which pause image to use.

Defaults to${ARCH}:3.0 where ${ARCH} can be one of amd64, arm, arm64, ppc64le or s390x.

cat > /etc/systemd/system/kubelet.service.d/20-pod-infra-image.conf <<EOF
systemctl daemon-reload
systemctl restart kubelet

If you want to use kubeadm with an http proxy, you may need to configure it to support http_proxy, https_proxy, or no_proxy.

For example, if your kube master node IP address is and you have a proxy which supports both http/https on port 8080, you can use the following command:

export PROXY_PORT=8080
export PROXY_IP=
export http_proxy=http://$PROXY_IP:$PROXY_PORT
export HTTP_PROXY=$http_proxy
export https_proxy=$http_proxy
export HTTPS_PROXY=$http_proxy
export no_proxy="localhost,,localaddress,,,"

Remember to change proxy_ip and add a kube master node IP address to no_proxy.

Using custom certificates

By default kubeadm will generate all the certificates needed for a cluster to run. You can override this behaviour by providing your own certificates.

To do so, you must place them in whatever directory is specified by the --cert-dir flag or CertificatesDir configuration file key. By default this is /etc/kubernetes/pki.

If a given certificate and private key pair both exist, kubeadm will skip the generation step and those files will be validated and used for the prescribed use-case.

This means you can, for example, prepopulate /etc/kubernetes/pki/ca.crt and /etc/kubernetes/pki/ca.key with an existing CA, which then will be used for signing the rest of the certs.

Only for the CA, it is possible to provide the ca.crt file but not the ca.key file; if all other certificates and kubeconfig files already are in place kubeadm recognise this condition and activates the so called “ExternalCA” mode, which also implies the csrsignercontroller in controller-manager won’t be started.

Self-hosting the Kubernetes control plane

As of 1.8, kubeadm can experimentally create a self-hosted Kubernetes control plane. This means that key components such as the API server, controller manager, and scheduler run as DaemonSet pods configured via the Kubernetes API instead of static pods configured in the kubelet via static files.

Self-hosting is alpha in kubeadm 1.8 but is expected to become the default in a future version. To create a self-hosted cluster, pass the --feature-gates=SelfHosting=true flag to kubeadm init.


Kubeadm self-hosting in 1.8 has some important limitations. In particular, a self-hosted cluster cannot currently recover from a reboot of the master node without manual intervention. This and other limitations are expected to be resolved before self-hosting graduates from alpha.

By default, self-hosted control plane pods rely on credentials loaded from hostPath volumes. Except for initial creation, these credentials are not managed by kubeadm. You can use --feature-gates=StoreCertsInSecrets=true to enable an experimental mode where control plane credentials are loaded from Secrets instead. This requires very careful control over the authentication and authorization configuration for your cluster, and may not be appropriate for your environment.

In 1.8, the self-hosted portion of the control plane does not include etcd, which still runs as a static pod.


The self-hosting bootstrap process is documented in the kubeadm 1.8 design document. In summary, kubeadm init --feature-gates=SelfHosting=true works as follows:

  1. As usual, kubeadm creates static pod YAML files in /etc/kubernetes/manifests/.

  2. Kubelet loads these files and launches the initial static control plane. Kubeadm waits for this initial static control plane to be running and healthy. This is identical to the kubeadm init process without self-hosting.

  3. Kubeadm uses the static control plane pod manifests to construct a set of DaemonSet manifests that will run the self-hosted control plane.

  4. Kubeadm creates DaemonSets in the kube-system namespace and waits for the resulting pods to be running.

  5. Once the new control plane is running (but not yet active), kubeadm deletes the static pod YAML files. This triggers kubelet to stop those static pods.

  6. When the original static control plane stops, the new self-hosted control plane is able to bind to listening ports and become active.

This process (steps 3-6) can also be triggered with kubeadm phase selfhosting convert-from-staticpods.

Customising the control plane with custom arguments

If you would like to override or extend the behaviour of a control plane component, you can provide extra arguments to kubeadm. When the component is deployed, it will use these additional arguments in its pod command.

For example, to add flag --feature-gates=APIResponseCompression=true to kube-apiserver, your configuration file will need to look like this:

kind: MasterConfiguration
   feature-gates: APIResponseCompression=true

To customise the scheduler or controller-manager, use schedulerExtraArgs and controllerManagerExtraArgs respectively.

More information on custom arguments can be found here:

Releases and release notes

If you already have kubeadm installed and want to upgrade, run apt-get update && apt-get upgrade or yum update to get the latest version of kubeadm.

Refer to the for more information.


Create an Issue Edit this Page