Muokkaa

Jaa

Azure Kubernetes Service patch and upgrade guidance

  • Artikkeli

This section of the Azure Kubernetes Service (AKS) day-2 operations guide describes patching and upgrading strategies for AKS worker nodes and Kubernetes versions. As a cluster operator, you need to have a plan for keeping your clusters up to date and monitoring Kubernetes API changes and deprecations over time.

Background and types of updates

There are three types of updates for AKS, each one building on the next:

Update type Frequency of upgrade Planned Maintenance supported Supported operation methods Target Link to documentation
Node OS security patches Nightly Yes Automatic (weekly), manual/unmanaged (nightly) Node Auto Upgrade Node Images
Node image version upgrades Linux: Weekly
Windows: Monthly		 Yes Automatic, manual Node pool AKS node image upgrade
Kubernetes version (cluster) upgrades Quarterly Yes Automatic, manual Cluster and node pool AKS cluster upgrade

Update types

  • Node OS security patches (Linux only). For Linux nodes, both Canonical Ubuntu and Azure Linux make operating system security patches available once per day. Microsoft tests and bundles these patches in the weekly updates to node images.

  • Weekly updates to node images. AKS provides weekly updates to node images. These updates include the latest OS and AKS security patches, bug fixes, and enhancements. Node updates don't change the Kubernetes version. Versions are formatted by date (for example, 202311.07.0) for Linux and by Windows Server OS build and date (for example, 20348.2113.231115) for Windows. For more information, see AKS Release Status.

  • Quarterly Kubernetes releases. AKS provides quarterly updates for Kubernetes releases. These updates allow AKS users to take advantage of the latest Kubernetes features and enhancements. They include security patches and node image updates. For more information, see Supported Kubernetes versions in AKS.

Pre-upgrade considerations

Overall cluster impact

  • In-place upgrades (both node and cluster) affect the performance of your Kubernetes environment while they're in progress. You can minimize this effect through proper configuration of pod disruption budgets, node surge configuration, and proper planning.
  • Using a blue/green update strategy instead of upgrading in place eliminates any impact to cluster performance but increases cost and complexity.
  • Regardless of your upgrade and patching strategy, you need to have a robust testing and validation process for your cluster. Patch and upgrade lower environments first, and perform a post-maintenance validation to check cluster, node, deployment, and application health.

AKS workload best practices

To ensure the smooth operation of your AKS cluster during maintenance, follow these best practices:

  • Define pod disruption budgets (PDBs). Setting up pod disruption budgets for your deployments is essential. PDBs enforce a minimum number of available application replicas to ensure continuous functionality during disruption events. PDBs help maintain the stability of your cluster during maintenance or node failures.

    Warning

    Misconfigured PDBs can block the upgrade process because the Kubernetes API prevents the necessary cordon and drain that occurs with a rolling node-image upgrade. Additionally, if too many pods are moved simultaneously, an application outage can occur. PDB configuration mitigates this risk.

  • Check available compute and network limits. Verify the available compute and network limits in your Azure subscription via the quota page in the Azure portal, or by using the az quota command. Check compute and network resources, especially VM vCPUs for your nodes, and the number of virtual machines and virtual machine scale sets. If you're close to a limit, request a quota increase before you upgrade.
  • Check available IP space in node subnets. During updates, extra surge nodes are created in your cluster and pods are moved to these nodes. It's important that you monitor the IP address space in your node subnets to ensure there's sufficient address space for these changes to occur. Different Kubernetes network configurations have different IP requirements. As a starting point, review these considerations:
    • During an upgrade, the number of node IPs increases according to your surge value. (The minimum surge value is 1.)
    • Clusters that are based on Azure CNI assign IP addresses to individual pods, so there needs to be sufficient IP space for pod movement.
    • Your cluster continues to operate during upgrades. Be sure that there's enough IP space left to allow node scaling (if it's enabled).
  • Set up multiple environments. We recommend that you set up separate environments, like development, staging, and production, to enable you to test and validate changes before you roll them out to production.
  • Tune surge upgrade values. By default, AKS has a surge value of 1, which means that one extra node is created at a time as part of the upgrade process. You can increase the speed of an AKS upgrade by increasing this value. 33% surge is the recommended maximum value for workloads that are sensitive to disruptions. For more information, see Customize node surge upgrade.
  • Tune node drain timeout. Node drain timeout specifies the maximum amount of time a cluster will wait while attempting to reschedule pods on a node that's updating. The default value for this is 30 minutes. For workloads that struggle to reschedule pods it can be helpful to adjust this default value.
  • Plan and schedule maintenance windows. Upgrade processes might affect the overall performance of your Kubernetes cluster. Schedule in-place upgrade processes outside of peak usage windows, and monitor cluster performance to ensure adequate sizing, especially during update processes.
  • Check other dependencies in your cluster. Kubernetes operators often deploy other tooling to the Kubernetes cluster as part of operations, like KEDA scalers, Dapr, and service meshes. When you plan your upgrade processes, check release notes for any components that you're using to ensure compatibility with the target version.

Managing weekly updates to node images

Microsoft creates a new node image for AKS nodes approximately once per week. A node image contains up-to-date OS security patches, OS kernel updates, Kubernetes security updates, updated versions of binaries like kubelet, and component version updates that are listed in the release notes.

When a node image is updated, a cordon and drain action is triggered on the target node pool's nodes:

  • A node with the updated image is added to the node pool. The number of nodes added at the same time is governed by the surge value.
  • Depending on the surge value, a batch of existing nodes are cordoned and drained. Cordoning ensures that the node doesn't schedule pods. Draining removes its pods and schedules them to other nodes.
  • After these nodes are fully drained, they are removed from the node pool. The updated nodes added by the surge replace them.
  • This process is repeated for each remaining batch of nodes that needs to be updated in the node pool.

A similar process occurs during a cluster upgrade.

Automatic node image upgrades

Generally speaking, most clusters should use the NodeImage update channel. This channel provides an updated node image VHD on a weekly basis and is updated according to your cluster's maintenance window.

Available channels include the following:

  • None. No updates are automatically applied.
  • Unmanaged. Ubuntu and Azure Linux updates are applied by the OS on a nightly basis. Reboots must be managed separately. AKS is neither able to test this nor control the cadence of this.
  • SecurityPatch. OS security patches which are AKS-tested, fully managed, and applied with safe deployment practices. It does not contain any OS bug fixes just security updates.
  • NodeImage. AKS updates the nodes with a newly patched VHD containing security fixes and bug fixes on a weekly cadence. This is fully tested and deployed with safe deployment practices. For real-time information about currently deployed node images, see AKS node image updates.

To understand the default cadences without a maintenance window established, see update ownership and cadence.

If you choose the Unmanaged update channel, you need to manage the reboot process by using a tool like kured. Unmanaged does not come with AKS-provided safe deployment practices and will not work under maintenance windows. If you choose the SecurityPatch update channel, updates can be applied as frequently as weekly. This patch level requires the VHDs to be stored in your resource group, which incurs a nominal charge. Control when the SecurityPatch is applied by setting an appropriate aksManagedNodeOSUpgradeSchedule that aligns to a cadence that works best for your workload. For more information, see Creating a maintenance window. If you also need bug fixes that come typically with new node images (VHD), then you need to choose the NodeImage channel instead of SecurityPatch.

As a best practice, use the NodeImage update channel and configure an aksManagedNodeOSUpgradeSchedule maintenance window to a time when the cluster is outside of peak usage windows. See Creating a maintenance window for attributes that you can use to configure the cluster maintenance window. The key attributes are:

  • name. Use aksManagedNodeOSUpgradeSchedule for node OS updates.
  • utcOffset. Configure the time zone.
  • startTime. Set the start time of the maintenance window.
  • dayofWeek. Set the days of the week for the window. For example, Saturday.
  • schedule. Set the frequency of the window. For NodeImage updates, we recommend weekly.
  • durationHours. Set this attribute to at least four hours.

This example sets a weekly maintenance window to 8:00 PM Eastern Time on Saturdays:

az aks maintenanceconfiguration add -g <ResourceGroupName> --cluster-name <AKSClusterName> --name aksManagedNodeOSUpgradeSchedule --utc-offset=-05:00 --start-time 20:00 --day-of-week Saturday --schedule-type weekly --duration 4

For more examples, see Add a maintenance window configuration with Azure CLI.

This configuration would ideally be deployed as part of the infrastructure-as-code deployment of the cluster.

You can check for configured maintenance windows by using the Azure CLI:

az aks maintenanceconfiguration list -g <ResourceGroupName> --cluster-name <AKSClusterName>

You can also check the details of a specific maintenance window by using the CLI:

az aks maintenanceconfiguration show -g <ResourceGroupName> --cluster-name <AKSClusterName> --name aksManagedNodeOSUpgradeSchedule

If a cluster maintenance window isn't configured, node image updates occur biweekly. As much as possible, AKS maintenance occurs within the configured window, but the time of maintenance isn't guaranteed.

Important

If you have a node pool with a large number of nodes but it is not configured with node surge, the auto upgrade event might not trigger. Node images in a node pool will only be upgraded while the estimated total upgrade time is within 24 hours.

In this situation, you can consider one of the following:

  • splitting nodes into different node pools if your vCPU quota is almost full and you cannot increase the vCPU quota
  • configuring node surge to decrease the estimated upgrade time if your vCPU quota is enough

You can check the status of upgrade events through your Azure activity logs, or by reviewing the resource logs on your cluster.

You can Subscribe to Azure Kubernetes Service (AKS) events with Azure Event Grid which includes AKS upgrade events. These events can alert you when new version of Kubernetes is available and help to track node status changes during upgrade processes.

You can also manage the weekly update process by using GitHub Actions. This method provides more granular control of the update process.

Manual node update process

You can use the kubectl describe nodes command to determine the OS kernel version and the OS image version of the nodes in your cluster:

kubectl describe nodes <NodeName>

Example output (truncated):

System Info:
  Machine ID:                 bb2e85e682ae475289f2e2ca4ed6c579
  System UUID:                6f80de9d-91ba-490c-8e14-9e68b7b82a76
  Boot ID:                    3aed0fd5-5d1d-4e43-b7d6-4e840c8ee3cf
  Kernel Version:             5.15.0-1041-azure
  OS Image:                   Ubuntu 22.04.2 LTS
  Operating System:           linux
  Architecture:               arm64
  Container Runtime Version:  containerd://1.7.1+azure-1
  Kubelet Version:            v1.26.6
  Kube-Proxy Version:         v1.26.6

Use the Azure CLI az aks nodepool list command to determine the node image versions of the nodes in a cluster:

az aks nodepool list \
   --resource-group <ResourceGroupName> --cluster-name <AKSClusterName> \
   --query "[].{Name:name,NodeImageVersion:nodeImageVersion}" --output table

Example output:

Name       NodeImageVersion
---------  ---------------------------------------------
systempool  AKSUbuntu-2204gen2containerd-202307.12.0
usernodepool  AKSUbuntu-2204gen2arm64containerd-202307.12.0

Use az aks nodepool get-upgrades to determine the latest available node image version for a specific node pool:

az aks nodepool get-upgrades \
   --resource-group <ResourceGroupName> \
   --cluster-name <AKSClusterName> \
   --nodepool-name <NodePoolName> --output table

Example output:

KubernetesVersion    LatestNodeImageVersion                         Name     OsType
-------------------  ---------------------------------------------  -------  --------
1.26.6               AKSUbuntu-2204gen2arm64containerd-202308.10.0  default  Linux

Cluster upgrades

The Kubernetes community releases minor versions of Kubernetes approximately every three months. To keep you informed about new AKS versions and releases, the AKS release notes page is updated regularly. You can also subscribe to the GitHub AKS RSS feed, which provides real-time updates about changes and enhancements.

AKS follows an N - 2 support policy, which means that full support is provided for the latest release (N) and two previous minor versions. Limited platform support is offered for the third prior minor version. For more information, see AKS support policy.

To ensure that your AKS clusters remain supported, you need to establish a continuous cluster upgrade process. This process involves testing new versions in lower environments and planning the upgrade to production before the new version becomes the default. This approach can maintain predictability in your upgrade process and minimize disruptions to applications. For more information, see Upgrade an AKS cluster.

If your cluster requires a longer upgrade cycle, use AKS versions that support the Long Term Support (LTS) option. If you enable the LTS option, Microsoft provides extended support for Kubernetes versions for two years, which enables a more prolonged and controlled upgrade cycle. For more information, see Supported Kubernetes versions in AKS.

A cluster upgrade includes a node upgrade and uses a similar cordon and drain process.

Before you upgrade

As a best practice, you should always upgrade and test in lower environments to minimize the risk of disruption in production. Cluster upgrades require extra testing because they involve API changes, which can affect Kubernetes deployments. The following resources can assist you in the upgrade process:

  • AKS workbook for deprecated APIs. From the cluster overview page in the Azure portal, you can select Diagnose and solve problems, go to the Create, Upgrade, Delete and Scale category, and then select Kubernetes API deprecations. Doing so runs a workbook that checks for deprecated API versions that are being used in your cluster. For more information, see Remove usage of deprecated APIs.
  • AKS release notes page. This page provides comprehensive information about new AKS versions and releases. Review these notes to stay informed about the latest updates and changes.
  • Kubernetes release notes page. This page provides detailed insights into the latest Kubernetes versions. Pay special attention to urgent upgrade notes, which highlight critical information that might affect your AKS cluster.
  • AKS components breaking changes by version. This table provides a comprehensive overview of breaking changes in AKS components, by version. By referring to this guide, you can proactively address any potential compatibility issues before the upgrade process.

In addition to these Microsoft resources, consider using open-source tools to optimize your cluster upgrade process. One such tool is Fairwinds pluto, which can scan your deployments and Helm charts for deprecated Kubernetes APIs. These tools can help you ensure that your applications remain compatible with the latest Kubernetes versions.

Upgrade process

To check when your cluster requires an upgrade, use az aks get-upgrades to get a list of available upgrade versions for your AKS cluster. Determine the target version for your cluster from the results.

Here's an example:

az aks get-upgrades \
   --resource-group <ResourceGroupName> --name <AKSClusterName> --output table

Example output:

MasterVersion  Upgrades
-------------  ---------------------------------
1.26.6         1.27.1, 1.27.3

Check the Kubernetes versions of the nodes in your node pools to determine the pools that need to be upgraded:

az aks nodepool list \
   --resource-group <ResourceGroupName> --cluster-name <AKSClusterName> \
   --query "[].{Name:name,k8version:orchestratorVersion}" --output table

Example output:

Name          K8version
------------  ------------
systempool    1.26.6
usernodepool  1.26.6

Manually upgrading

To minimize disruptions and help ensure a smooth upgrade for your AKS cluster, follow this upgrade approach:

  1. Upgrade the AKS control plane. Start by upgrading the AKS control plane. This step involves upgrading the control plane components that are responsible for managing and orchestrating your cluster. Upgrading the control plane first helps ensure compatibility and stability before you upgrade the individual node pools.
  2. Upgrade your system node pool. After you upgrade the control plane, upgrade the system node pool in your AKS cluster. Node pools consist of the virtual machine instances that run your application workloads. Upgrading the node pools separately enables a controlled and systematic upgrade of the underlying infrastructure that supports your applications.
  3. Upgrade user node pools. After you upgrade the system node pool, upgrade any user node pools in your AKS cluster.

By following this approach, you can minimize disruptions during the upgrade process and maintain the availability of your applications. These are the detailed steps:

  1. Run the az aks upgrade command with the --control-plane-only flag to upgrade only the cluster control plane and not the cluster's node pools:

    az aks upgrade \
   --resource-group <ResourceGroupName> --name <AKSClusterName> \
   --control-plane-only \
   --kubernetes-version <KubernetesVersion>

  2. Run az aks nodepool upgrade to upgrade node pools to the target version:

    az aks nodepool upgrade \
   --resource-group <ResourceGroupName> --cluster-name <AKSClusterName> --name <NodePoolName> \
   --no-wait --kubernetes-version <KubernetesVersion>

    During the node pool upgrade, AKS creates a surge node, cordons and drains pods in the node that's being upgraded, reimages the node, and then uncordons the pods. This process then repeats for any other nodes in the node pool.

You can check the status of the upgrade process by running kubectl get events. For information about troubleshooting cluster upgrade problems, see AKS troubleshooting documentation.

Enroll clusters in auto-upgrade release channels

AKS also offers an automatic cluster upgrade solution to keep your cluster up to date. If you use this solution, you should pair it with a maintenance window to control the timing of upgrades. The upgrade window must be four hours or more. When you enroll a cluster in a release channel, Microsoft automatically manages the version and upgrade cadence for the cluster and its node pools.

The cluster auto-upgrade offers different release channel options. Here's a recommended environment and release channel configuration:

Environment Upgrade channel Description
Production stable For stability and version maturity, use the stable or regular channel for production workloads.
Staging, testing, development Same as production To ensure that your tests are indicative of the version that you'll upgrade your production environment to, use the same release channel as production.
Canary rapid To test the latest Kubernetes releases and new AKS features or APIs, use the rapid channel. You can improve your time to market when the version in rapid is promoted to the channel you're using for production.

Considerations

The following table describes the characteristics of various AKS upgrade and patching scenarios:

Scenario User initiated Kubernetes upgrade OS kernel upgrade Node image upgrade
Security patching No No Yes, after reboot Yes
Cluster creation Yes Maybe Yes, if an updated node image uses an updated kernel Yes, relative to an existing cluster if a new release is available
Control plane Kubernetes upgrade Yes Yes No No
Node pool Kubernetes upgrade Yes Yes Yes, if an updated node image uses an updated kernel Yes, if a new release is available
Node pool scale-up Yes No No No
Node image upgrade Yes No Yes, if an updated node image uses an updated kernel Yes
Cluster auto-upgrade No Yes Yes, if an updated node image uses an updated kernel Yes, if a new release is available
  • An OS security patch that's applied as part of a node image upgrade might install a later version of the kernel than creation of a new cluster would install.
  • Node pool scale-up uses the model that's currently associated with the virtual machine scale set. The OS kernels are upgraded when security patches are applied, and the nodes reboot.

Contributors

This article is maintained by Microsoft. It was originally written by the following contributors.

Principal author:

Other contributors:

To see nonpublic LinkedIn profiles, sign in to LinkedIn.

Next steps