Kubernetes version 1.5 introduces support for Windows Server Containers. In version 1.5, the Kubernetes control plane (API Server, Scheduler, Controller Manager, etc) continue to run on Linux, while the kubelet and kube-proxy can be run on Windows Server.
**Note:** Windows Server Containers on Kubernetes is an Alpha feature in Kubernetes 1.5.
Network is achieved using L3 routing. Because third-party networking plugins (e.g. flannel, calico, etc) don’t natively work on Windows Server, existing technology that is built into the Windows and Linux operating systems is relied on. In this L3 networking approach, a /16 subnet is chosen for the cluster nodes, and a /24 subnet is assigned to each worker node. All pods on a given worker node will be connected to the /24 subnet. This allows pods on the same node to communicate with each other. In order to enable networking between pods running on different nodes, routing features that are built into Windows Server 2016 and Linux are used.
The above networking approach is already supported on Linux using a bridge interface, which essentially creates a private network local to the node. Similar to the Windows side, routes to all other pod CIDRs must be created in order to send packets via the “public” NIC.
1. Two NICs (virtual networking adapters) are required on each Windows Server node - The two Windows container networking modes of interest (transparent and L2 bridge) use an external Hyper-V virtual switch. This means that one of the NICs is entirely allocated to the bridge, creating the need for the second NIC.
2. Transparent container network created - This is a manual configuration step and is shown in **_Route Setup_** section below
3. RRAS (Routing) Windows feature enabled - Allows routing between NICs on the box, and also “captures” packets that have the destination IP of a POD running on the node. To enable, open “Server Manager”. Click on “Roles”, “Add Roles”. Click “Next”. Select “Network Policy and Access Services”. Click on “Routing and Remote Access Service” and the underlying checkboxes
4. Routes defined pointing to the other pod CIDRs via the “public” NIC - These routes are added to the built-in routing table as shown in **_Route Setup_** section below
The following diagram illustrates the Windows Server networking setup for Kubernetes Setup
## Setting up Windows Server Containers on Kubernetes
To run Windows Server Containers on Kubernetes, you'll need to set up both your host machines and the Kubernetes node components for Windows and setup Routes for Pod communication on different nodes
1. Windows Server container host running Windows Server 2016 and Docker v1.12. Follow the setup instructions outlined by this blog post: https://msdn.microsoft.com/en-us/virtualization/windowscontainers/quick_start/quick_start_windows_server
2. DNS support for Windows recently got merged to docker master and is currently not supported in a stable docker release. To use DNS build docker from master or download the binary from [Docker master](https://master.dockerproject.org/)
3. Pull the `apprenda/pause` image from `https://hub.docker.com/r/apprenda/pause`
* Important notes and other dependencies are listed [here](https://github.com/kubernetes/kubernetes/blob/master/docs/devel/development.md#building-kubernetes-on-a-local-osshell-environment)
Run the following in a PowerShell window. Be aware that if the node reboots or the process exits, you will have to rerun the commands below to restart the kubelet
1. Set environment variable *CONTAINER_NETWORK* value to the docker container network to use
`$env:CONTAINER_NETWORK = "<docker network>"`
2. Run *kubelet* executable using the below command
`kubelet.exe --hostname-override=<ip address/hostname of the windows node> --pod-infra-container-image="apprenda/pause" --resolv-conf="" --api_servers=<api server location>`
Run the following in a PowerShell window with administrative privileges. Be aware that if the node reboots or the process exits, you will have to rerun the commands below to restart the kube-proxy.
1. Set environment variable *INTERFACE_TO_ADD_SERVICE_IP* value to a node only network interface. The interface created when docker is installed should work
2. Run *kube-proxy* executable using the below command
`.\proxy.exe --v=3 --proxy-mode=userspace --hostname-override=<ip address/hostname of the windows node> --master=<api server location> --bind-address=<ip address of the windows node>`
Because your cluster has both Linux and Windows nodes, you must explictly set the nodeSelector constraint to be able to schedule Pods to Windows nodes. You must set nodeSelector with the label beta.kubernetes.io/os to the value windows; see the following example: