vars.md

Configurable Parameters in Kubespray

Generic Ansible variables

You can view facts gathered by Ansible automatically here.

Some variables of note include:

• ansible_user: user to connect to via SSH
• ansible_default_ipv4.address: IP address Ansible automatically chooses. Generated based on the output from the command ip -4 route get 8.8.8.8

Common vars that are used in Kubespray

• calico_version - Specify version of Calico to use
• calico_cni_version - Specify version of Calico CNI plugin to use
• docker_version - Specify version of Docker to use (should be quoted string). Must match one of the keys defined for docker_versioned_pkg in roles/container-engine/docker/vars/*.yml.
• containerd_version - Specify version of containerd to use when setting container_manager to containerd
• docker_containerd_version - Specify which version of containerd to use when setting container_manager to docker
• etcd_version - Specify version of ETCD to use
• calico_ipip_mode - Configures Calico ipip encapsulation - valid values are 'Never', 'Always' and 'CrossSubnet' (default 'Never')
• calico_vxlan_mode - Configures Calico vxlan encapsulation - valid values are 'Never', 'Always' and 'CrossSubnet' (default 'Always')
• calico_network_backend - Configures Calico network backend - valid values are 'none', 'bird' and 'vxlan' (default 'vxlan')
• kube_network_plugin - Sets k8s network plugin (default Calico)
• kube_proxy_mode - Changes k8s proxy mode to iptables mode
• kube_version - Specify a given Kubernetes version
• searchdomains - Array of DNS domains to search when looking up hostnames
• remove_default_searchdomains - Boolean that removes the default searchdomain
• nameservers - Array of nameservers to use for DNS lookup
• preinstall_selinux_state - Set selinux state, permitted values are permissive, enforcing and disabled.

Addressing variables

• ip - IP to use for binding services (host var). This would usually be the public ip.
• access_ip - IP for other hosts to use to connect to. Often required when deploying from a cloud, such as OpenStack or GCE and you have separate public/floating and private IPs. This would usually be the private ip.
• ansible_default_ipv4.address - Not Kubespray-specific, but it is used if ip and access_ip are undefined
• ip6 - IPv6 address to use for binding services. (host var) If enable_dual_stack_networks is set to true and ip6 is defined, kubelet's --node-ip and node's InternalIP will be the combination of ip and ip6.
• loadbalancer_apiserver - If defined, all hosts will connect to this address instead of localhost for kube_control_planes and kube_control_plane[0] for kube_nodes. See more details in the HA guide.
• loadbalancer_apiserver_localhost - makes all hosts to connect to the apiserver internally load balanced endpoint. Mutual exclusive to the loadbalancer_apiserver. See more details in the HA guide.

Cluster variables

Kubernetes needs some parameters in order to get deployed. These are the following default cluster parameters:

• cluster_name - Name of cluster (default is cluster.local)

• container_manager - Container Runtime to install in the nodes (default is containerd)

• image_command_tool - Tool used to pull images (default depends on container_manager and is nerdctl for containerd, crictl for crio, docker for docker)

• image_command_tool_on_localhost - Tool used to pull images on localhost (default is equal to image_command_tool)

• dns_domain - Name of cluster DNS domain (default is cluster.local)

• kube_network_plugin - Plugin to use for container networking

• kube_service_addresses - Subnet for cluster IPs (default is 10.233.0.0/18). Must not overlap with kube_pods_subnet

• kube_pods_subnet - Subnet for Pod IPs (default is 10.233.64.0/18). Must not overlap with kube_service_addresses.

• kube_network_node_prefix - Subnet allocated per-node for pod IPs. Remaining bits in kube_pods_subnet dictates how many kube_nodes can be in cluster. Setting this > 25 will raise an assertion in playbooks if the kubelet_max_pods var also isn't adjusted accordingly (assertion not applicable to calico which doesn't use this as a hard limit, see Calico IP block sizes).

• enable_dual_stack_networks - Setting this to true will provision both IPv4 and IPv6 networking for pods and services.

• kube_service_addresses_ipv6 - Subnet for cluster IPv6 IPs (default is fd85:ee78:d8a6:8607::1000/116). Must not overlap with kube_pods_subnet_ipv6.

• kube_pods_subnet_ipv6 - Subnet for Pod IPv6 IPs (default is fd85:ee78:d8a6:8607::1:0000/112). Must not overlap with kube_service_addresses_ipv6.

• kube_network_node_prefix_ipv6 - Subnet allocated per-node for pod IPv6 IPs. Remaining bits in kube_pods_subnet_ipv6 dictates how many kube_nodes can be in cluster.

• skydns_server - Cluster IP for DNS (default is 10.233.0.3)

• skydns_server_secondary - Secondary Cluster IP for CoreDNS used with coredns_dual deployment (default is 10.233.0.4)

• enable_coredns_k8s_external - If enabled, it configures the k8s_external plugin on the CoreDNS service.

• coredns_k8s_external_zone - Zone that will be used when CoreDNS k8s_external plugin is enabled (default is k8s_external.local)

• enable_coredns_k8s_endpoint_pod_names - If enabled, it configures endpoint_pod_names option for kubernetes plugin. on the CoreDNS service.

• cloud_provider - Enable extra Kubelet option if operating inside GCE or OpenStack (default is unset)

• kube_feature_gates - A list of key=value pairs that describe feature gates for alpha/experimental Kubernetes features. (defaults is []). Additionally, you can use also the following variables to individually customize your kubernetes components installation (they works exactly like kube_feature_gates):

  • kube_apiserver_feature_gates
  • kube_controller_feature_gates
  • kube_scheduler_feature_gates
  • kube_proxy_feature_gates
  • kubelet_feature_gates

• kubeadm_feature_gates - A list of key=value pairs that describe feature gates for alpha/experimental Kubeadm features. (defaults is [])

• authorization_modes - A list of authorization mode that the cluster should be configured for. Defaults to ['Node', 'RBAC'] (Node and RBAC authorizers). Note: Node and RBAC are enabled by default. Previously deployed clusters can be converted to RBAC mode. However, your apps which rely on Kubernetes API will require a service account and cluster role bindings. You can override this setting by setting authorization_modes to [].

• kube_apiserver_admission_control_config_file - Enable configuration for kube-apiserver admission plugins. Currently this variable allow you to configure the EventRateLimit admission plugin.

  To configure the EventRateLimit plugin you have to define a data structure like this:

kube_apiserver_admission_event_rate_limits:
  limit_1:
    type: Namespace
    qps: 50
    burst: 100
    cache_size: 2000
  limit_2:
    type: User
    qps: 50
    burst: 100
  ...

• kube_apiserver_service_account_lookup - Enable validation service account before validating token. Default true.

Note, if cloud providers have any use of the 10.233.0.0/16, like instances' private addresses, make sure to pick another values for kube_service_addresses and kube_pods_subnet, for example from the 172.18.0.0/16.

Enabling Dual Stack (IPV4 + IPV6) networking

If enable_dual_stack_networks is set to true, Dual Stack networking will be enabled in the cluster. This will use the default IPv4 and IPv6 subnets specified in the defaults file in the kubespray-defaults role, unless overridden of course. The default config will give you room for up to 256 nodes with 126 pods per node, and up to 4096 services.

DNS variables

By default, hosts are set up with 8.8.8.8 as an upstream DNS server and all other settings from your existing /etc/resolv.conf are lost. Set the following variables to match your requirements.

• upstream_dns_servers - Array of upstream DNS servers configured on host in addition to Kubespray deployed DNS
• nameservers - Array of DNS servers configured for use by hosts
• searchdomains - Array of up to 4 search domains
• remove_default_searchdomains - Boolean. If enabled, searchdomains variable can hold 6 search domains.
• dns_etchosts - Content of hosts file for coredns and nodelocaldns
• dns_upstream_forward_extra_opts - Options to add in the forward section of coredns/nodelocaldns related to upstream DNS servers

For more information, see DNS Stack.

Other service variables

• docker_options - Commonly used to set --insecure-registry=myregistry.mydomain:5000

• docker_plugins - This list can be used to define Docker plugins to install.

• containerd_default_runtime - If defined, changes the default Containerd runtime used by the Kubernetes CRI plugin.

• containerd_additional_runtimes - Sets the additional Containerd runtimes used by the Kubernetes CRI plugin. Default config can be overridden in inventory vars.

• crio_criu_support_enabled - When set to true, enables the container checkpoint/restore in CRI-O. It's required to install CRIU on the host when dumping/restoring checkpoints. And it's recommended to enable the feature gate ContainerCheckpoint so that the kubelet get a higher level API to simplify the operations (Note: It's still in experimental stage, just for container analytics so far). You can follow the documentation.

• http_proxy/https_proxy/no_proxy/no_proxy_exclude_workers/additional_no_proxy - Proxy variables for deploying behind a proxy. Note that no_proxy defaults to all internal cluster IPs and hostnames that correspond to each node.

• kubelet_cgroup_driver - Allows manual override of the cgroup-driver option for Kubelet. By default autodetection is used to match container manager configuration. systemd is the preferred driver for containerd though it can have issues with cgroups v1 and kata-containers in which case you may want to change to cgroupfs.

• kubelet_rotate_certificates - Auto rotate the kubelet client certificates by requesting new certificates from the kube-apiserver when the certificate expiration approaches.

• kubelet_rotate_server_certificates - Auto rotate the kubelet server certificates by requesting new certificates from the kube-apiserver when the certificate expiration approaches. Note that enabling this also activates kubelet_csr_approver which approves automatically the CSRs. To customize its behavior, you can override the Helm values via kubelet_csr_approver_values. See kubelet-csr-approver for more information.

• kubelet_streaming_connection_idle_timeout - Set the maximum time a streaming connection can be idle before the connection is automatically closed.

• kubelet_image_gc_high_threshold - Set the percent of disk usage after which image garbage collection is always run. The percent is calculated by dividing this field value by 100, so this field must be between 0 and 100, inclusive. When specified, the value must be greater than imageGCLowThresholdPercent. Default: 85

• kubelet_image_gc_low_threshold - Set the percent of disk usage before which image garbage collection is never run. Lowest disk usage to garbage collect to. The percent is calculated by dividing this field value by 100, so the field value must be between 0 and 100, inclusive. When specified, the value must be less than imageGCHighThresholdPercent. Default: 80

• kubelet_max_parallel_image_pulls - Sets the maximum number of image pulls in parallel. The value is 1 by default which means the default is serial image pulling, set it to a integer great than 1 to enable image pulling in parallel.

• kubelet_make_iptables_util_chains - If true, causes the kubelet ensures a set of iptables rules are present on host.

• kubelet_cpu_manager_policy - If set to static, allows pods with certain resource characteristics to be granted increased CPU affinity and exclusivity on the node. And it should be set with kube_reserved or system-reserved, enable this with the following guide:Control CPU Management Policies on the Node

• kubelet_cpu_manager_policy_options - A dictionary of cpuManagerPolicyOptions to enable. Keep in mind to enable the corresponding feature gates and make sure to pass the booleans as string (i.e. don't forget the quotes)!

kubelet_cpu_manager_policy_options:
    distribute-cpus-across-numa: "true"
    full-pcpus-only: "true"

• kubelet_topology_manager_policy - Control the behavior of the allocation of CPU and Memory from different NUMA Nodes. Enable this with the following guide: Control Topology Management Policies on a node.

• kubelet_topology_manager_scope - The Topology Manager can deal with the alignment of resources in a couple of distinct scopes: container and pod. See Topology Manager Scopes.

• kubelet_systemd_hardening - If true, provides kubelet systemd service with security features for isolation.

  N.B. To enable this feature, ensure you are using the cgroup v2 on your system. Check it out with command: sudo ls -l /sys/fs/cgroup/*.slice. If directory does not exist, enable this with the following guide: enable cgroup v2.

  • kubelet_secure_addresses - By default kubelet_systemd_hardening set the control plane ansible_host IPs as the kubelet_secure_addresses. In case you have multiple interfaces in your control plane nodes and the kube-apiserver is not bound to the default interface, you can override them with this variable. Example:

    The control plane node may have 2 interfaces with the following IP addresses: eth0:10.0.0.110, eth1:192.168.1.110.

    By default the kubelet_secure_addresses is set with the 10.0.0.110 the ansible control host uses eth0 to connect to the machine. In case you want to use eth1 as the outgoing interface on which kube-apiserver connects to the kubelets, you should override the variable in this way: kubelet_secure_addresses: "192.168.1.110".

• kubelet_systemd_wants_dependencies - List of kubelet service dependencies, other than container runtime.

  If you use nfs dynamically mounted volumes, sometimes rpc-statd does not start within the kubelet. You can fix it with this parameter : kubelet_systemd_wants_dependencies: ["rpc-statd.service"] This will add Wants=rpc-statd.service in [Unit] section of /etc/systemd/system/kubelet.service

• node_labels - Labels applied to nodes via kubectl label node. For example, labels can be set in the inventory as variables or more widely in group_vars. node_labels can only be defined as a dict:

node_labels:
  label1_name: label1_value
  label2_name: label2_value

• node_taints - Taints applied to nodes via kubectl taint node. For example, taints can be set in the inventory as variables or more widely in group_vars. node_taints has to be defined as a list of strings in format key=value:effect, e.g.:

node_taints:
  - "node.example.com/external=true:NoSchedule"

• kubernetes_audit - When set to true, enables Auditing. The auditing parameters can be tuned via the following variables (which default values are shown below):

  • audit_log_path: /var/log/audit/kube-apiserver-audit.log
  • audit_log_maxage: 30
  • audit_log_maxbackups: 10
  • audit_log_maxsize: 100
  • audit_policy_file: "{{ kube_config_dir }}/audit-policy/apiserver-audit-policy.yaml"

  By default, the audit_policy_file contains default rules that can be overridden with the audit_policy_custom_rules variable.

• kubernetes_audit_webhook - When set to true, enables the webhook audit backend. The webhook parameters can be tuned via the following variables (which default values are shown below):

  • audit_webhook_config_file: "{{ kube_config_dir }}/audit-policy/apiserver-audit-webhook-config.yaml"
  • audit_webhook_server_url: "https://audit.app"
  • audit_webhook_server_extra_args: {}
  • audit_webhook_mode: batch
  • audit_webhook_batch_max_size: 100
  • audit_webhook_batch_max_wait: 1s

• kubectl_alias - Bash alias of kubectl to interact with Kubernetes cluster much easier.

• remove_anonymous_access - When set to true, removes the kubeadm:bootstrap-signer-clusterinfo rolebinding created by kubeadm. By default, kubeadm creates a rolebinding in the kube-public namespace which grants permissions to anonymous users. This rolebinding allows kubeadm to discover and validate cluster information during the join phase. In a nutshell, this option removes the rolebinding after the init phase of the first control plane node and then configures kubeadm to use file discovery for the join phase of other nodes. This option does not remove the anonymous authentication feature of the API server.

Custom flags for Kube Components

For all kube components, custom flags can be passed in. This allows for edge cases where users need changes to the default deployment that may not be applicable to all deployments.

Extra flags for the kubelet can be specified using these variables, in the form of dicts of key-value pairs of configuration parameters that will be inserted into the kubelet YAML config file. The kubelet_node_config_extra_args apply kubelet settings only to nodes and not control planes. Example:

kubelet_config_extra_args:
  evictionHard:
    memory.available: "100Mi"
  evictionSoftGracePeriod:
    memory.available: "30s"
  evictionSoft:
    memory.available: "300Mi"

The possible vars are:

• kubelet_config_extra_args
• kubelet_node_config_extra_args

Previously, the same parameters could be passed as flags to kubelet binary with the following vars:

• kubelet_custom_flags
• kubelet_node_custom_flags

The kubelet_node_custom_flags apply kubelet settings only to nodes and not control planes. Example:

kubelet_custom_flags:
  - "--eviction-hard=memory.available<100Mi"
  - "--eviction-soft-grace-period=memory.available=30s"
  - "--eviction-soft=memory.available<300Mi"

This alternative is deprecated and will remain until the flags are completely removed from kubelet

Extra flags for the API server, controller, and scheduler components can be specified using these variables, in the form of dicts of key-value pairs of configuration parameters that will be inserted into the kubeadm YAML config file:

• kube_kubeadm_apiserver_extra_args
• kube_kubeadm_controller_extra_args
• kube_kubeadm_scheduler_extra_args

Kubeadm patches

When extra flags are not sufficient and there is a need to further customize kubernetes components, kubeadm patches can be used. You should use the kubeadm_patches variable for that purpose.

App variables

• helm_version - Only supports v3.x. Existing v2 installs (with Tiller) will not be modified and need to be removed manually.