This document describes the maximum version skew supported between various Kubernetes components. Specific cluster deployment tools may place additional restrictions on version skew.
Kubernetes versions are expressed as x.y.z, where x is the major version, y is the minor version, and z is the patch version, following Semantic Versioning terminology. For more information, see Kubernetes Release Versioning.
The Kubernetes project maintains release branches for the most recent three minor releases.
Applicable fixes, including security fixes, may be backported to those three release branches, depending on severity and feasibility. Patch releases are cut from those branches at a regular cadence, or as needed. This decision is owned by the patch release team. The patch release team is part of release managers. For more information, see Kubernetes Patch releases.
Minor releases occur approximately every 3 months, so each minor release branch is maintained for approximately 9 months.
In highly-available (HA) clusters, the newest and oldest kube-apiserver
instances must be within one minor version.
Example:
kube-apiserver
is at 1.13kube-apiserver
instances are supported at 1.13 and 1.12kubelet
must not be newer than kube-apiserver
, and may be up to two minor versions older.
Example:
kube-apiserver
is at 1.13kubelet
is supported at 1.13, 1.12, and 1.11Note: If version skew exists betweenkube-apiserver
instances in an HA cluster, this narrows the allowedkubelet
versions.
Example:
kube-apiserver
instances are at 1.13 and 1.12kubelet
is supported at 1.12, and 1.11 (1.13 is not supported because that would be newer than the kube-apiserver
instance at version 1.12)kube-controller-manager
, kube-scheduler
, and cloud-controller-manager
must not be newer than the kube-apiserver
instances they communicate with. They are expected to match the kube-apiserver
minor version, but may be up to one minor version older (to allow live upgrades).
Example:
kube-apiserver
is at 1.13kube-controller-manager
, kube-scheduler
, and cloud-controller-manager
are supported at 1.13 and 1.12Note: If version skew exists betweenkube-apiserver
instances in an HA cluster, and these components can communicate with anykube-apiserver
instance in the cluster (for example, via a load balancer), this narrows the allowed versions of these components.
Example:
kube-apiserver
instances are at 1.13 and 1.12kube-controller-manager
, kube-scheduler
, and cloud-controller-manager
communicate with a load balancer that can route to any kube-apiserver
instancekube-controller-manager
, kube-scheduler
, and cloud-controller-manager
are supported at 1.12 (1.13 is not supported because that would be newer than the kube-apiserver
instance at version 1.12)kubectl
is supported within one minor version (older or newer) of kube-apiserver
.
Example:
kube-apiserver
is at 1.13kubectl
is supported at 1.14, 1.13, and 1.12Note: If version skew exists betweenkube-apiserver
instances in an HA cluster, this narrows the supportedkubectl
versions.
Example:
kube-apiserver
instances are at 1.13 and 1.12kubectl
is supported at 1.13 and 1.12 (other versions would be more than one minor version skewed from one of the kube-apiserver
components)The supported version skew between components has implications on the order in which components must be upgraded. This section describes the order in which components must be upgraded to transition an existing cluster from version 1.n to version 1.(n+1).
Pre-requisites:
kube-apiserver
instance is 1.nkube-apiserver
instances are at 1.n or 1.(n+1) (this ensures maximum skew of 1 minor version between the oldest and newest kube-apiserver
instance)kube-controller-manager
, kube-scheduler
, and cloud-controller-manager
instances that communicate with this server are at version 1.n (this ensures they are not newer than the existing API server version, and are within 1 minor version of the new API server version)kubelet
instances on all nodes are at version 1.n or 1.(n-1) (this ensures they are not newer than the existing API server version, and are within 2 minor versions of the new API server version)kube-apiserver
instance will send them:ValidatingWebhookConfiguration
and MutatingWebhookConfiguration
objects are updated to include any new versions of REST resources added in 1.(n+1) (or use the matchPolicy: Equivalent
option available in v1.15+)Upgrade kube-apiserver
to 1.(n+1)
Note: Project policies for API deprecation and API change guidelines requirekube-apiserver
to not skip minor versions when upgrading, even in single-instance clusters.
Pre-requisites:
kube-apiserver
instances these components communicate with are at 1.(n+1) (in HA clusters in which these control plane components can communicate with any kube-apiserver
instance in the cluster, all kube-apiserver
instances must be upgraded before upgrading these components)Upgrade kube-controller-manager
, kube-scheduler
, and cloud-controller-manager
to 1.(n+1)
Pre-requisites:
kube-apiserver
instances the kubelet
communicates with are at 1.(n+1)Optionally upgrade kubelet
instances to 1.(n+1) (or they can be left at 1.n or 1.(n-1))
Warning: Running a cluster withkubelet
instances that are persistently two minor versions behindkube-apiserver
is not recommended:
- they must be upgraded within one minor version of
kube-apiserver
before the control plane can be upgraded- it increases the likelihood of running
kubelet
versions older than the three maintained minor releases
Was this page helpful?
Thanks for the feedback. If you have a specific, answerable question about how to use Kubernetes, ask it on Stack Overflow. Open an issue in the GitHub repo if you want to report a problem or suggest an improvement.