Skip to content

Latest commit

 

History

History
564 lines (436 loc) · 35.8 KB

configuration.md

File metadata and controls

564 lines (436 loc) · 35.8 KB
Raw

Configuration options

Using a configuration file

k0s can be installed without a config file. In that case the default configuration will be used. You can, though, create and run your own non-default configuration (used by the k0s controller nodes).

k0s supports providing only partial configurations. In case of partial configuration is provided, k0s will use the defaults for any missing values.

  1. Generate a yaml config file that uses the default settings.

    mkdir -p /etc/k0s
k0s config create > /etc/k0s/k0s.yaml

  2. Modify the new yaml config file according to your needs, refer to Configuration file reference below. You can remove the default values if wanted as k0s supports partial configs too.

  3. Install k0s with your new config file.

    sudo k0s install controller -c /etc/k0s/k0s.yaml

  4. If you need to modify your existing configuration later on, you can change your config file also when k0s is running, but remember to restart k0s to apply your configuration changes.

    sudo k0s stop
sudo k0s start

Configuring k0s via k0sctl

k0sctl can deploy your configuration options at cluster creation time. Your options should be placed in the spec.k0s.config section of the k0sctl's configuration file. See the section on how to install k0s via k0sctl and the k0sctl README for more information.

Configuration file reference

CAUTION: As many of the available options affect items deep in the stack, you should fully understand the correlation between the configuration file components and your specific environment before making any changes.

A YAML config file follows, with defaults as generated by the k0s config create command:

apiVersion: k0s.k0sproject.io/v1beta1
kind: ClusterConfig
metadata:
  name: k0s
spec:
  api:
    address: 192.168.68.104
    k0sApiPort: 9443
    port: 6443
    sans:
    - 192.168.68.104
  controllerManager: {}
  extensions:
    helm:
      concurrencyLevel: 5
  installConfig:
    users:
      etcdUser: etcd
      kineUser: kube-apiserver
      konnectivityUser: konnectivity-server
      kubeAPIserverUser: kube-apiserver
      kubeSchedulerUser: kube-scheduler
  konnectivity:
    adminPort: 8133
    agentPort: 8132
  network:
    clusterDomain: cluster.local
    dualStack:
      enabled: false
    kubeProxy:
      iptables:
        minSyncPeriod: 0s
        syncPeriod: 0s
      ipvs:
        minSyncPeriod: 0s
        syncPeriod: 0s
        tcpFinTimeout: 0s
        tcpTimeout: 0s
        udpTimeout: 0s
      metricsBindAddress: 0.0.0.0:10249
      mode: iptables
    kuberouter:
      autoMTU: true
      hairpin: Enabled
      metricsPort: 8080
    nodeLocalLoadBalancing:
      enabled: false
      envoyProxy:
        apiServerBindPort: 7443
        konnectivityServerBindPort: 7132
      type: EnvoyProxy
    podCIDR: 10.244.0.0/16
    provider: kuberouter
    serviceCIDR: 10.96.0.0/12
  scheduler: {}
  storage:
    etcd:
      peerAddress: 192.168.68.104
    type: etcd
  telemetry:
    enabled: true

spec Key Detail

spec.api

Element Description
externalAddress The loadbalancer address (for k0s controllers running behind a loadbalancer). Configures all cluster components to connect to this address and also configures this address for use when joining new nodes to the cluster.
address Local address on which to bind an API. Also serves as one of the addresses pushed on the k0s create service certificate on the API. Defaults to first non-local address found on the node.
sans List of additional addresses to push to API servers serving the certificate.
extraArgs Map of key-values (strings) for any extra arguments to pass down to Kubernetes api-server process.
port¹ Custom port for kube-api server to listen on (default: 6443)
k0sApiPort¹ Custom port for k0s-api server to listen on (default: 9443)

¹ If port and k0sApiPort are used with the externalAddress element, the loadbalancer serving at externalAddress must listen on the same ports.

spec.storage

Element Description
type Type of the data store (valid values:etcd or kine). Note: Type etcd will cause k0s to create and manage an elastic etcd cluster within the controller nodes.
etcd.peerAddress Node address used for etcd cluster peering.
etcd.extraArgs Map of key-values (strings) for any extra arguments to pass down to etcd process.
kine.dataSource kine datasource URL.
etcd.externalCluster Configuration when etcd is externally managed, i.e. running on dedicated nodes. See spec.storage.etcd.externalCluster

spec.storage.etcd.externalCluster

k0s can also work with externally managed Etcd cluster. If this is configured, k0s will NOT set up etcd, it has to be managed manually.

Element Description
endpoints Array of Etcd endpoints to use.
etcdPrefix Prefix to use for this cluster. The same external Etcd cluster can be used for several k0s clusters, each prefixed with unique prefix to store data with.
caFile CaFile is the host path to a file with Etcd cluster CA certificate.
clientCertFile ClientCertFile is the host path to a file with TLS certificate for etcd client.
clientKeyFile ClientKeyFile is the host path to a file with TLS key for etcd client.

spec.network

Element Description
provider Network provider (valid values: calico, kuberouter, or custom). For custom, you can push any network provider (default: kuberouter). Be aware that it is your responsibility to configure all of the CNI-related setups, including the CNI provider itself and all necessary host levels setups (for example, CNI binaries). Note: Once you initialize the cluster with a network provider the only way to change providers is through a full cluster redeployment.
podCIDR Pod network CIDR to use in the cluster.
serviceCIDR Network CIDR to use for cluster VIP services.
clusterDomain Cluster Domain to be passed to the kubelet and the coredns configuration.

spec.network.calico

Element Description
mode vxlan (default), ipip or bird
overlay Overlay mode: Always (default), CrossSubnet or Never (requires mode=vxlan to disable calico overlay-network).
vxlanPort The UDP port for VXLAN (default: 4789).
vxlanVNI The virtual network ID for VXLAN (default: 4096).
mtu MTU for overlay network (default: 0, which causes Calico to detect optimal MTU during bootstrap).
wireguard Enable wireguard-based encryption (default: false). Your host system must be wireguard ready (refer to the Calico documentation for details).
flexVolumeDriverPath The host path for Calicos flex-volume-driver(default: /usr/libexec/k0s/kubelet-plugins/volume/exec/nodeagent~uds). Change this path only if the default path is unwriteable (refer to Project Calico Issue #2712 for details). Ideally, you will pair this option with a custom volumePluginDir in the profile you use for your worker nodes.
ipAutodetectionMethod Use to force Calico to pick up the interface for pod network inter-node routing (default: "", meaning not set, so that Calico will instead use its defaults). For more information, refer to the Calico documentation.
envVars Map of key-values (strings) for any calico-node environment variable.

spec.network.calico.envVars

Environment variable's value must be string, e.g.:

spec:
  network:
    provider: calico
    calico:
      envVars:
        TEST_BOOL_VAR: "true"
        TEST_INT_VAR: "42"
        TEST_STRING_VAR: test

K0s runs Calico with some predefined vars, which can be overwritten by setting new value in spec.network.calico.envVars:

CALICO_IPV4POOL_CIDR: "{{ spec.network.podCIDR }}"
CALICO_DISABLE_FILE_LOGGING: "true"
FELIX_DEFAULTENDPOINTTOHOSTACTION: "ACCEPT"
FELIX_LOGSEVERITYSCREEN: "info"
FELIX_HEALTHENABLED: "true"
FELIX_PROMETHEUSMETRICSENABLED: "true"
FELIX_FEATUREDETECTOVERRIDE: "ChecksumOffloadBroken=true"

FELIX_FEATUREDETECTOVERRIDE: ChecksumOffloadBroken=true disables VXLAN offloading because of projectcalico/calico#4727.

In single-stack mode, there are additional vars:

FELIX_IPV6SUPPORT: "false"

In dual-stack mode, there are additional vars:

CALICO_IPV6POOL_NAT_OUTGOING: "true"
FELIX_IPV6SUPPORT: "true"
IP6: "autodetect"
CALICO_IPV6POOL_CIDR: "{{ spec.network.dualStack.IPv6podCIDR }}"

spec.network.kuberouter

Element Description
autoMTU Autodetection of used MTU (default: true).
mtu Override MTU setting, if autoMTU must be set to false).
metricsPort Kube-router metrics server port. Set to 0 to disable metrics (default: 8080).
peerRouterIPs DEPRECATED: Use extraArgs with peerRouterIPs instead. Comma-separated list of global peer addresses.
peerRouterASNs DEPRECATED: Use extraArgs with peerRouterASNs instead. Comma-separated list of global peer ASNs.
hairpin Hairpin mode, supported modes Enabled: enabled cluster wide, Allowed: must be allowed per service using annotations, Disabled: doesn't work at all (default: Enabled)
hairpinMode Deprecated Use hairpin instead. If both hairpin and hairpinMode are defined, this is ignored. If only hairpinMode is configured explicitly activates hairpinMode (https://github.com/cloudnativelabs/kube-router/blob/master/docs/user-guide.md#hairpin-mode).
ipMasq IP masquerade for traffic originating from the pod network, and destined outside of it (default: false)
extraArgs Extra arguments to pass to kube-router. Can be also used to override any k0s managed args. For reference, see kube-router documentation. (default: empty)

Note: Kube-router allows many networking aspects to be configured per node, service, and pod (for more information, refer to the Kube-router user guide).

spec.network.kubeProxy

Element Description
disabled Disable kube-proxy altogether (default: false).
mode Kube proxy operating mode, supported modes iptables, ipvs, userspace (default: iptables)
iptables Kube proxy iptables settings
ipvs Kube proxy ipvs settings
nodePortAddresses Kube proxy nodePortAddresses

Default kube-proxy iptables settings:

iptables:
  masqueradeAll: false
  masqueradeBit: null
  minSyncPeriod: 0s
  syncPeriod: 0s

Default kube-proxy ipvs settings:

ipvs:
  excludeCIDRs: null
  minSyncPeriod: 0s
  scheduler: ""
  strictARP: false
  syncPeriod: 0s
  tcpFinTimeout: 0s
  tcpTimeout: 0s
  udpTimeout: 0s

spec.network.nodeLocalLoadBalancing

Configuration options related to k0s's node-local load balancing feature.

Element Description
enabled Indicates if node-local load balancing should be used to access Kubernetes API servers from worker nodes. Default: false.
type The type of the node-local load balancer to deploy on worker nodes. Default: EnvoyProxy. (This is the only option for now.)
envoyProxy Configuration options related to the "EnvoyProxy" type of load balancing.
spec.network.nodeLocalLoadBalancing.envoyProxy

Configuration options required for using Envoy as the backing implementation for node-local load balancing.

Note: This type of load balancing is not supported on ARMv7 workers.

Element Description
image The OCI image that's being used for the Envoy Pod.
imagePullPolicy The pull policy being used used for the Envoy Pod. Defaults to spec.images.default_pull_policy if omitted.
apiServerBindPort Port number on which to bind the Envoy load balancer for the Kubernetes API server to on a worker's loopback interface. Default: 7443.
konnectivityServerBindPort Port number on which to bind the Envoy load balancer for the konnectivity server to on a worker's loopback interface. Default: 7132.
spec.network.controlPlaneLoadBalancing

Configuration options related to k0s's control plane load balancing feature

Element Description
enabled Indicates if control plane load balancing should be enabled. Default: false.
type The type of the control plane load balancer to deploy on controller nodes. Currently, the only supported type is Keepalived.
keepalived Contains the keepalived configuration.
spec.network.controlPlaneLoadBalancing.Keepalived

Configuration options related to keepalived in control plane load balancing

Element Description
vrrpInstances Configuration options related to the VRRP. This is an array which allows to configure multiple virtual IPs.
virtualServers Configuration options related LoadBalancing. This is an array which allows to configure multiple LBs.
spec.network.controlPlaneLoadBalancing.keepalived.vrrpInstances

Configuration options required for using VRRP to configure VIPs in control plane load balancing.

Element Description
virtualIPs VirtualIPs is the list of virtual IP address used by the VRRP instance. Each virtual IP must be a CIDR as defined in RFC 4632 and RFC 4291.A list of the CIDRs handled by the VRRP instance.
interface The NIC used by the virtual router. If not specified, k0s will use the interface that owns the default route.
virtualRouterID The VRRP router ID. If not specified, k0s will automatically number the IDs for each VRRP instance, starting with 51. It must be in the range of 1-255, all the control plane nodes must use the same virtualRouterID. Other clusters in the same network must not use the same virtualRouterID.
advertIntervalSeconds Advertisement interval in seconds. Defaults to 1 second.
authPass The password for accessing VRRPD. This is not a security feature but a way to prevent accidental misconfigurations. It must be in the range of 1-8 characters
spec.network.controlPlaneLoadBalancing.keepalived.virtualServers

Configuration options required for using VRRP to configure VIPs in control plane load balancing.

Element Description
ipAddress The load balancer's listen address.
delayLoop Delay timer for check polling. DelayLoop accepts microsecond precision. Further precision will be truncated without warnings. Defaults to 1m.
lbAlgo Algorithm used by keepalived. Supported algorithms: rr, wrr, lc, wlc, lblc, dh, sh, sed, nq. Default: rr.
lbKind Kind of ipvs load balancer. Supported values: NAT, DR, TUN Default: DR.
persistenceTimeoutSeconds Timeout value for persistent connections in seconds. Must be in the range of 1-2678400 (31 days). If not specified, defaults to 360 (6 minutes).

spec.controllerManager

Element Description
extraArgs Map of key-values (strings) for any extra arguments you want to pass down to the Kubernetes controller manager process.

spec.scheduler

Element Description
extraArgs Map of key-values (strings) for any extra arguments you want to pass down to Kubernetes scheduler process.

spec.workerProfiles

Worker profiles are used to manage worker-specific configuration in a centralized manner. A ConfigMap is generated for each worker profile. Based on the --profile argument given to the k0s worker, the configuration in the corresponding ConfigMap is is picked up during startup.

The worker profiles are defined as an array. Each element has following properties:

Property Description
name String; name to use as profile selector for the worker process
values Object; Kubelet configuration overrides, see below for details

spec.workerProfiles[].values (Kubelet configuration overrides)

The Kubelet configuration overrides of a profile override the defaults defined by k0s.

Note that there are several fields that cannot be overridden:

  • clusterDNS
  • clusterDomain
  • apiVersion
  • kind
  • staticPodURL

Configuration examples

Custom volumePluginDir
spec:
  workerProfiles:
    - name: custom-pluginDir
      values:
         volumePluginDir: /var/libexec/k0s/kubelet-plugins/volume/exec
Eviction Policy
spec:
  workerProfiles:
    - name: custom-eviction
      values:
        evictionHard:
          memory.available: "500Mi"
          nodefs.available: "1Gi"
          imagefs.available: "100Gi"
        evictionMinimumReclaim:
          memory.available: "0Mi"
          nodefs.available: "500Mi"
          imagefs.available: "2Gi"
Unsafe Sysctls
spec:
  workerProfiles:
    - name: custom-eviction
      values:
        allowedUnsafeSysctls:
          - fs.inotify.max_user_instances

spec.featureGates

Available components are:

  • kube-apiserver
  • kube-controller-manager
  • kubelet
  • kube-scheduler
  • kube-proxy

If components are omitted, propagates to all kube components.

Modifies extraArgs.

Examples

Generic feature gates example
spec:
    featureGates:
      - name: feature-gate-0
        enabled: true
        components: ["kube-apiserver", "kube-controller-manager", "kubelet", "kube-scheduler"]
      - name: feature-gate-1
        enabled: true
      - name: feature-gate-2
        enabled: false
Kubelet feature gates example

The below is an example of a k0s config with feature gates enabled:

spec:
    featureGates:
      - name: DevicePlugins
        enabled: true
        components: ["kubelet"]
      - name: Accelerators
        enabled: true
        components: ["kubelet"]
      - name: AllowExtTrafficLocalEndpoints
        enabled: false

spec.images

Nodes under the images key all have the same basic structure:

spec:
  images:
    coredns:
      image: quay.io/coredns/coredns
      version: v1.7.0

If you want the list of default images and their versions to be included, use k0s config create --include-images.

Available keys

  • spec.images.konnectivity
  • spec.images.metricsserver
  • spec.images.kubeproxy
  • spec.images.coredns
  • spec.images.pause
  • spec.images.calico.cni
  • spec.images.calico.flexvolume
  • spec.images.calico.node
  • spec.images.calico.kubecontrollers
  • spec.images.kuberouter.cni
  • spec.images.kuberouter.cniInstaller
  • spec.images.repository¹

¹ If spec.images.repository is set and not empty, every image will be pulled from images.repository

If spec.images.default_pull_policy is set and not empty, it will be used as a pull policy for each bundled image.

Image example

images:
  repository: "my.own.repo"
  konnectivity:
    image: calico/kube-controllers
    version: v3.16.2
  metricsserver:
    image: quay.io/k0sproject/metrics-server
    version: v0.7.1-0

In the runtime the image names are calculated as my.own.repo/calico/kube-controllers:v3.16.2 and my.own.repo/k0sproject/metrics-server:v0.7.1-0. This only affects the the imgages pull location, and thus omitting an image specification here will not disable component deployment.

spec.extensions.helm

spec.extensions.helm is the config file key in which you configure the list of Helm repositories and charts to deploy during cluster bootstrap (for more information, refer to Helm Charts).

spec.konnectivity

The spec.konnectivity key is the config file key in which you configure Konnectivity-related settings.

  • agentPort agent port to listen on (default 8132)
  • adminPort admin port to listen on (default 8133)

spec.telemetry

To improve the end-user experience k0s is configured by defaul to collect telemetry data from clusters and send it to the k0s development team. To disable the telemetry function, change the enabled setting to false.

The telemetry interval is ten minutes.

spec:
  telemetry:
    enabled: true

Disabling controller components

k0s allows to completely disable some of the system components. This allows users to build a minimal Kubernetes control plane and use what ever components they need to fulfill their need for the control plane. Disabling the system components happens through a command line flag for the controller process:

--disable-components strings                     disable components (valid items: applier-manager,autopilot,control-api,coredns,csr-approver,endpoint-reconciler,helm,konnectivity-server,kube-controller-manager,kube-proxy,kube-scheduler,metrics-server,network-provider,node-role,system-rbac,windows-node,worker-config)

Note: As of k0s 1.26, the kubelet-config component has been replaced by the worker-config component. k0s will issue a warning when the old component name is being used. It is scheduled for removal in k0s 1.27. Please update to the new component name.

If you use k0sctl, just add the flag when installing the cluster for the first controller at spec.hosts.installFlags in the config file like e.g.:

spec:
  hosts:
  - role: controller
    installFlags:
    - --disable-components=metrics-server

As seen from the component list, the only always-on component is the Kubernetes api-server, without that k0s serves no purpose.