This is a cache of https://docs.okd.io/4.6/windows_containers/understanding-windows-container-workloads.html. It is a snapshot of the page at 2024-11-29T03:00:50.692+0000.
Understanding Windows container workloads | Windows Container Support for OpenShift | OKD 4.6
×

Red Hat OpenShift support for Windows Containers provides built-in support for running Microsoft Windows Server containers on OKD. For those that administer heterogeneous environments with a mix of Linux and Windows workloads, OKD allows you to deploy Windows workloads running on Windows Server containers while also providing traditional Linux workloads hosted on Fedora CoreOS (FCOS) or Fedora.

Windows container workloads are supported for clusters running on the following cloud providers:

  • Amazon Web Services (AWS)

  • Microsoft Azure

The following Windows Server operating systems are supported for OKD 4.6:

  • Windows Server Long-Term Servicing Channel (LTSC): Windows Server 2019

For more information, see Microsoft’s documentation on Windows Server channels.

Multi-tenancy for clusters that have Windows nodes is not supported. Hostile multi-tenant usage introduces security concerns in all Kubernetes environments. Additional security features like pod security policies, or more fine-grained role-based access control (RBAC) for nodes, make exploits more difficult. However, if you choose to run hostile multi-tenant workloads, a hypervisor is the only security option you should use. The security domain for Kubernetes encompasses the entire cluster, not an individual node. For these types of hostile multi-tenant workloads, you should use physically isolated clusters.

Windows Server Containers provide resource isolation using a shared kernel but are not intended to be used in hostile multitenancy scenarios. Scenarios that involve hostile multitenancy should use Hyper-V Isolated Containers to strongly isolate tenants.

Windows workload management

To run Windows workloads in your cluster, you must first install the Windows Machine Config Operator (WMCO). The WMCO is a Linux-based Operator that runs on Linux-based control plane and compute nodes. The WMCO orchestrates the process of deploying and managing Windows workloads on a cluster.

WMCO workflow
Figure 1. WMCO design

Before deploying Windows workloads, you must create a Windows compute node and have it join the cluster. The Windows node hosts the Windows workloads in a cluster, and can run alongside other Linux-based compute nodes. You can create a Windows compute node by creating a Windows machine set to host Windows Server compute machines. You must apply a Windows-specific label to the machine set that specifies a Windows OS image that has the Docker-formatted container runtime add-on enabled.

Currently, the Docker-formatted container runtime is used in Windows nodes. Kubernetes is deprecating Docker as a container runtime; you can reference the Kubernetes documentation for more information in Docker deprecation. Containerd will be the new supported container runtime for Windows nodes in a future release of Kubernetes.

The WMCO watches for machines with the Windows label. After a Windows machine set is detected and its respective machines are provisioned, the WMCO configures the underlying Windows virtual machine (VM) so that it can join the cluster as a compute node.

Mixed Windows and Linux workloads
Figure 2. Mixed Windows and Linux workloads

The WMCO expects a predetermined secret in its namespace containing a private key that is used to interact with the Windows instance. WMCO checks for this secret during boot up time and creates a user data secret which you must reference in the Windows MachineSet object that you created. Then the WMCO populates the user data secret with a public key that corresponds to the private key. With this data in place, the cluster can connect to the Windows VM using an SSH connection.

After the cluster establishes a connection with the Windows VM, you can manage the Windows node using similar practices as you would a Linux-based node.

The OKD web console does not provide node graphs and workload graphs for Windows nodes. No metrics are available for Windows nodes at this time.

Scheduling Windows workloads to a Windows node can be done with typical pod scheduling practices like taints, tolerations, and node selectors; alternatively, you can differentiate your Windows workloads from Linux workloads and other Windows-versioned workloads by using a RuntimeClass object.

Windows node services

The following Windows-specific services are installed on each Windows node:

Service Description

kubelet

Registers the Windows node and manages its status.

Container Network Interface (CNI) plug-ins

Exposes networking for Windows nodes.

Windows Machine Config Bootstrapper (WMCB)

Configures the kubelet and CNI plug-ins.

hybrid-overlay

Creates the OKD Host Network Service (HNS).

kube-proxy

Maintains network rules on nodes allowing outside communication.

Known limitations

Note the following limitations when working with Windows nodes managed by the WMCO (Windows nodes):

  • The following OKD features are not supported on Windows nodes:

    • Red Hat OpenShift Developer CLI (odo)

    • Image builds

    • OpenShift Pipelines

    • OpenShift Service Mesh

    • OpenShift monitoring of user-defined projects

    • OpenShift Serverless

    • Horizontal Pod Autoscaling

    • Vertical Pod Autoscaling

  • The following Red Hat features are not supported on Windows nodes:

  • Windows nodes do not support pulling container images from private registries. You can use images from public registries or pre-pull the images.

  • Windows nodes do not support workloads created by using deployment configs. You can use a deployment or other method to deploy workloads.

  • Windows nodes are not supported in clusters that use a cluster-wide proxy. This is because the WMCO is not able to route traffic through the proxy connection for the workloads.

  • Windows nodes are not supported in clusters that are in a disconnected environment.

  • Red Hat OpenShift support for Windows Containers supports only in-tree storage drivers for all cloud providers.

  • Kubernetes has identified the following node feature limitations :

    • Huge pages are not supported for Windows containers.

    • Privileged containers are not supported for Windows containers.

    • Pod termination grace periods require the containerd container runtime to be installed on the Windows node.

  • Kubernetes has identified several API compatibility issues.