Update Kubernetes version
In a Managed Kubernetes cluster, you can:
- upgrade minor release;
- enable auto-update for patch releases — Kubernetes patch releases contain bug fixes and security vulnerability patches for a Kubernetes minor version. They are compatible with each other within the same minor version.
You can choose a Kubernetes version upgrade strategy for a node group when creating a cluster or adding a node group.
During a Kubernetes version update, modifying node configuration, changing the number of nodes, and autoscaling are not available.
Upgrade minor version
If, when creating a cluster on cloud servers or adding a node group, you selected the node upgrade method Reinstall, all nodes will be reinstalled when upgrading the Managed Kubernetes cluster on cloud servers. All data on the boot disk will be deleted.
In Managed Kubernetes clusters on dedicated servers, master nodes will be rebooted one by one first, followed by worker nodes. The upgrade can take from 30 minutes to 12 hours, depending on the cluster size. Once started, the upgrade cannot be stopped.
-
Prepare for the version upgrade and ensure that:
- each Pod Disruption Budget (PDB) allows at least one pod to be evicted: the value in the
ALLOWED DISRUPTIONScolumn is greater than0. Read more about PodDisruptionBudget in the Pod disruption budgets section of the Kubernetes documentation; - applications are run via Deployment, StatefulSet, or ReplicaSet controllers;
- data is stored in persistent volumes rather than in
emptyDiror local node disks; - the version upgrade is scheduled for a time when the application is not in use. If you have applications running on a single replica, the application will be unavailable until the pod restarts on another node.
- each Pod Disruption Budget (PDB) allows at least one pod to be evicted: the value in the
-
In the control panel, on the top menu, click Products and select Managed Kubernetes.
-
In the Clusters section, open the cluster page → Settings tab.
-
In the Cluster settings block, click Show available versions.
-
Select the version to upgrade to.
-
Optional: to view the version changelog in the official Kubernetes documentation, click Version changelog.
-
Click Upgrade. During the version upgrade, the cluster will be in the status
PENDING_UPGRADE_MINOR_VERSION.
Enable patch version auto-upgrade
Patch version auto-upgrade is not available for:
- basic clusters;
- fault-tolerant clusters with fewer than two worker nodes.
If, when creating a cluster on cloud servers or adding a node group, you selected the node upgrade method Reinstall, all nodes will be reinstalled when upgrading the Managed Kubernetes cluster on cloud servers. All data on the boot disk will be deleted.
In Managed Kubernetes clusters on dedicated servers, master nodes will be rebooted one by one first, followed by worker nodes. The upgrade can take from 30 minutes to 12 hours, depending on the cluster size. Once started, the upgrade cannot be stopped.
The cluster will be upgraded to the latest available patch version during the next cluster maintenance.
You can enable patch version auto-upgrade when creating a cluster or on an existing cluster.
To enable automatic patch version updates:
-
In the control panel, on the top menu, click Products and select Managed Kubernetes.
-
In the Clusters section, open the cluster page → Settings tab.
-
In the Automation block, turn on the Auto-update patch versions toggle.
-
Prepare for the version update and ensure that:
- each Pod disruption budget (PDB) allows evicting at least one pod: the value in the
ALLOWED DISRUPTIONScolumn is greater than0. Read more about PodDisruptionBudget in the Pod disruption budgets section of the Kubernetes documentation; - applications are run via Deployment, StatefulSet, or ReplicaSet controllers;
- data is stored in persistent volumes rather than in
emptyDiror local node disks; - the version update is scheduled for a time when the application is not in use. If you have applications running on a single replica, the application will be unavailable until the pod restarts on another node.
- each Pod disruption budget (PDB) allows evicting at least one pod: the value in the