Windows applications constitute a large portion of the services and applications that run in many organizations. Windows containers provide a modern way to encapsulate processes and package dependencies, making it easier to use DevOps practices and follow cloud native patterns for Windows applications. Kubernetes has become the defacto standard container orchestrator, and the release of Kubernetes 1.14 includes production support for scheduling Windows containers on Windows nodes in a Kubernetes cluster, enabling a vast ecosystem of Windows applications to leverage the power of Kubernetes. Organizations with investments in Windows-based applications and Linux-based applications don’t have to look for separate orchestrators to manage their workloads, leading to increased operational efficiencies across their deployments, regardless of operating system.
To enable the orchestration of Windows containers in Kubernetes, simply include Windows nodes in your existing Linux cluster. Scheduling Windows containers in Pods on Kubernetes is as simple and easy as scheduling Linux-based containers.
In order to run Windows containers, your Kubernetes cluster must include multiple operating systems, with control plane nodes running Linux and workers running either Windows or Linux depending on your workload needs. Windows Server 2019 is the only Windows operating system supported, enabling Kubernetes Node on Windows (including kubelet, container runtime, and kube-proxy). For a detailed explanation of Windows distribution channels see the Microsoft documentation.
備考: The Kubernetes control plane, including the master components, continues to run on Linux. There are no plans to have a Windows-only Kubernetes cluster.
備考: In this document, when we talk about Windows containers we mean Windows containers with process isolation. Windows containers with Hyper-V isolation is planned for a future release.
From an API and kubectl perspective, Windows containers behave in much the same way as Linux-based containers. However, there are some notable differences in key functionality which are outlined in the limitation section.
Let’s start with the operating system version. Refer to the following table for Windows operating system support in Kubernetes. A single heterogeneous Kubernetes cluster can have both Windows and Linux worker nodes. Windows containers have to be scheduled on Windows nodes and Linux containers on Linux nodes.
|Kubernetes version||Host OS version (Kubernetes Node)|
|Windows Server 1709||Windows Server 1803||Windows Server 1809/Windows Server 2019|
|Kubernetes v1.14||Not Supported||Not Supported||Supported for Windows Server containers Builds 17763.* with Docker EE-basic 18.09|
備考: We don’t expect all Windows customers to update the operating system for their apps frequently. Upgrading your applications is what dictates and necessitates upgrading or introducing new nodes to the cluster. For the customers that chose to upgrade their operating system for containers running on Kubernetes, we will offer guidance and step-by-step instructions when we add support for a new operating system version. This guidance will include recommended upgrade procedures for upgrading user applications together with cluster nodes. Windows nodes adhere to Kubernetes version-skew policy (node to control plane versioning) the same way as Linux nodes do today.
備考: Windows containers with process isolation have strict compatibility rules, where the host OS version must match the container base image OS version. Once we support Windows containers with Hyper-V isolation in Kubernetes, the limitation and compatibility rules will change.
Key Kubernetes elements work the same way in Windows as they do in Linux. In this section, we talk about some of the key workload enablers and how they map to Windows.
A Pod is the basic building block of Kubernetes–the smallest and simplest unit in the Kubernetes object model that you create or deploy. The following Pod capabilities, properties and events are supported with Windows containers:
Kubernetes controllers handle the desired state of Pods. The following workload controllers are supported with Windows containers:
A Kubernetes Service is an abstraction which defines a logical set of Pods and a policy by which to access them - sometimes called a micro-service. You can use services for cross-operating system connectivity. In Windows, services can utilize the following types, properties and capabilities:
Pods, Controllers and Services are critical elements to managing Windows workloads on Kubernetes. However, on their own they are not enough to enable the proper lifecycle management of Windows workloads in a dynamic cloud native environment. We added support for the following features:
Docker EE-basic 18.09 is required on Windows Server 2019 / 1809 nodes for Kubernetes. This works with the dockershim code included in the kubelet. Additional runtimes such as CRI-ContainerD may be supported in later Kubernetes versions.
Kubernetes Volumes enable complex applications with data persistence and Pod volume sharing requirements to be deployed on Kubernetes. Kubernetes on Windows supports the following types of volumes:
Networking for Windows containers is exposed through CNI plugins. Windows containers function similarly to virtual machines in regards to networking. Each container has a virtual network adapter (vNIC) which is connected to a Hyper-V virtual switch (vSwitch). The Host Networking Service (HNS) and the Host Compute Service (HCS) work together to create containers and attach container vNICs to networks. HCS is responsible for the management of containers whereas HNS is responsible for the management of networking resources such as:
The following service spec types are supported:
Windows supports five different networking drivers/modes: L2bridge, L2tunnel, Overlay, Transparent, and NAT. In a heterogeneous cluster with Windows and Linux worker nodes, you need to select a networking solution that is compatible on both Windows and Linux. The following out-of-tree plugins are supported on Windows, with recommendations on when to use each CNI:
|Network Driver||Description||Container Packet Modifications||Network Plugins||Network Plugin Characteristics|
|L2bridge||Containers are attached to an external vSwitch. Containers are attached to the underlay network, although the physical network doesn’t need to learn the container MACs because they are rewritten on ingress/egress. Inter-container traffic is bridged inside the container host.||MAC is rewritten to host MAC, IP remains the same.||win-bridge, Azure-CNI, Flannel host-gateway uses win-bridge||win-bridge uses L2bridge network mode, connects containers to the underlay of hosts, offering best performance. Requires L2 adjacency between container hosts|
|L2Tunnel||This is a special case of l2bridge, but only used on Azure. All packets are sent to the virtualization host where SDN policy is applied.||MAC rewritten, IP visible on the underlay network||Azure-CNI||Azure-CNI allows integration of containers with Azure vNET, and allows them to leverage the set of capabilities that Azure Virtual Network provides. For example, securely connect to Azure services or use Azure NSGs. See azure-cni for some examples|
|Overlay (Overlay networking for Windows in Kubernetes is in alpha stage)||Containers are given a vNIC connected to an external vSwitch. Each overlay network gets its own IP subnet, defined by a custom IP prefix.The overlay network driver uses VXLAN encapsulation.||Encapsulated with an outer header, inner packet remains the same.||Win-overlay, Flannel VXLAN (uses win-overlay)||win-overlay should be used when virtual container networks are desired to be isolated from underlay of hosts (e.g. for security reasons). Allows for IPs to be re-used for different overlay networks (which have different VNID tags) if you are restricted on IPs in your datacenter. This option may be used when the container hosts are not L2 adjacent but have L3 connectivity|
|Transparent (special use case for ovn-kubernetes)||Requires an external vSwitch. Containers are attached to an external vSwitch which enables intra-pod communication via logical networks (logical switches and routers).||Packet is encapsulated either via GENEVE or STT tunneling to reach pods which are not on the same host.|
Packets are forwarded or dropped via the tunnel metadata information supplied by the ovn network controller.
NAT is done for north-south communication.
|ovn-kubernetes||Deploy via ansible. Distributed ACLs can be applied via Kubernetes policies. IPAM support. Load-balancing can be achieved without kube-proxy. NATing is done without using iptables/netsh.|
|NAT (not used in Kubernetes)||Containers are given a vNIC connected to an internal vSwitch. DNS/DHCP is provided using an internal component called WinNAT||MAC and IP is rewritten to host MAC/IP.||nat||Included here for completeness|
As outlined above, the Flannel CNI meta plugin is also supported on Windows via the VXLAN network backend (alpha support ; delegates to win-overlay) and host-gateway network backend (stable support; delegates to win-bridge). This plugin supports delegating to one of the reference CNI plugins (win-overlay, win-bridge), to work in conjunction with Flannel daemon on Windows (Flanneld) for automatic node subnet lease assignment and HNS network creation. This plugin reads in its own configuration file (net-conf.json), and aggregates it with the environment variables from the FlannelD generated subnet.env file. It then delegates to one of the reference CNI plugins for network plumbing, and sends the correct configuration containing the node-assigned subnet to the IPAM plugin (e.g. host-local).
For the node, pod, and service objects, the following network flows are supported for TCP/UDP traffic:
The following IPAM options are supported on Windows:
Windows is only supported as a worker node in the Kubernetes architecture and component matrix. This means that a Kubernetes cluster must always include Linux master nodes, zero or more Linux worker nodes, and zero or more Windows worker nodes.
Linux cgroups are used as a pod boundary for resource controls in Linux. Containers are created within that boundary for network, process and file system isolation. The cgroups APIs can be used to gather cpu/io/memory stats. In contrast, Windows uses a Job object per container with a system namespace filter to contain all processes in a container and provide logical isolation from the host. There is no way to run a Windows container without the namespace filtering in place. This means that system privileges cannot be asserted in the context of the host, and thus privileged containers are not available on Windows. Containers cannot assume an identity from the host because the Security Account Manager (SAM) is separate.
Windows has strict compatibility rules, where the host OS version must match the container base image OS version. Only Windows containers with a container operating system of Windows Server 2019 are supported. Hyper-V isolation of containers, enabling some backward compatibility of Windows container image versions, is planned for a future release.
Windows does not have an out-of-memory process killer as Linux does. Windows always treats all user-mode memory allocations as virtual, and pagefiles are mandatory. The net effect is that Windows won’t reach out of memory conditions the same way Linux does, and processes page to disk instead of being subject to out of memory (OOM) termination. If memory is over-provisioned and all physical memory is exhausted, then paging can slow down performance.
Keeping memory usage within reasonable bounds is possible with a two-step process. First, use the kubelet parameters
--system-reserve to account for memory usage on the node (outside of containers). This reduces NodeAllocatable). As you deploy workloads, use resource limits (must set only limits or limits must equal requests) on containers. This also subtracts from NodeAllocatable and prevents the scheduler from adding more pods once a node is full.
A best practice to avoid over-provisioning is to configure the kubelet with a system reserved memory of at least 2GB to account for Windows, Docker, and Kubernetes processes.
The behavior of the flags behave differently as described below:
--eviction-hardflags update Node Allocatable
--enforce-node-allocableis not implemented
--eviction-softare not implemented
--system-reservedo not set limits on kubelet or processes running on the host. This means kubelet or a process on the host could cause memory resource starvation outside the node-allocatable and scheduler
Windows has a layered filesystem driver to mount container layers and create a copy filesystem based on NTFS. All file paths in the container are resolved only within the context of that container.
As a result, the following storage functionality is not supported on Windows nodes
Windows Container Networking differs in some important ways from Linux networking. The Microsoft documentation for Windows Container Networking contains additional details and background.
The Windows host networking networking service and virtual switch implement namespacing and can create virtual NICs as needed for a pod or container. However, many configurations such as DNS, routes, and metrics are stored in the Windows registry database rather than /etc/… files as they are on Linux. The Windows registry for the container is separate from that of the host, so concepts like mapping /etc/resolv.conf from the host into a container don’t have the same effect they would on Linux. These must be configured using Windows APIs run in the context of that container. Therefore CNI implementations need to call the HNS instead of relying on file mappings to pass network details into the pod or container.
The following networking functionality is not supported on Windows nodes
curl <destination>to be able to debug connectivity to the outside world.
These features were added in Kubernetes v1.15:
Secrets are written in clear text on the node’s volume (as compared to tmpfs/in-memory on linux). This means customers have to do two things
RunAsUser is not currently supported on Windows. The workaround is to create local accounts before packaging the container. The RunAsUsername capability may be added in a future release.
Linux specific pod security context privileges such as SELinux, AppArmor, Seccomp, Capabilities (POSIX Capabilities), and others are not supported.
In addition, as mentioned already, privileged containers are not supported on Windows.
There are no differences in how most of the Kubernetes APIs work for Windows. The subtleties around what’s different come down to differences in the OS and container runtime. In certain situations, some properties on workload APIs such as Pod or Container were designed with an assumption that they are implemented on Linux, failing to run on Windows.
At a high level, these OS concepts are different:
/etc/passwdback to UID+GID. Windows uses a larger binary security identifier (SID) which is stored in the Windows Security Access Manager (SAM) database. This database is not shared between the host and containers, or between containers.
/. The Go IO libraries typically accept both and just make it work, but when you’re setting a path or command line that’s interpreted inside a container,
\may be needed.
Exit Codes follow the same convention where 0 is success, nonzero is failure. The specific error codes may differ across Windows and Linux. However, exit codes passed from the Kubernetes components (kubelet, kube-proxy) are unchanged.
None of the PodSecurityContext fields work on Windows. They’re listed here for reference.
Your main source of help for troubleshooting your Kubernetes cluster should start with this section. Some additional, Windows-specific troubleshooting help is included in this section. Logs are an important element of troubleshooting issues in Kubernetes. Make sure to include them any time you seek troubleshooting assistance from other contributors. Follow the instructions in the SIG-Windows contributing guide on gathering logs.
How do I know start.ps1 completed successfully?
You should see kubelet, kube-proxy, and (if you chose Flannel as your networking solution) flanneld host-agent processes running on your node, with running logs being displayed in separate PowerShell windows. In addition to this, your Windows node should be listed as “Ready” in your Kubernetes cluster.
Can I configure the Kubernetes node processes to run in the background as services?
Kubelet and kube-proxy are already configured to run as native Windows Services, offering resiliency by re-starting the services automatically in the event of failure (for example a process crash). You have two options for configuring these node components as services.
As native Windows Services
Kubelet & kube-proxy can be run as native Windows Services using
# Create the services for kubelet and kube-proxy in two separate commands sc.exe create <component_name> binPath= "<path_to_binary> --service <other_args>" # Please note that if the arguments contain spaces, they must be escaped. sc.exe create kubelet binPath= "C:\kubelet.exe --service --hostname-override 'minion' <other_args>" # Start the services Start-Service kubelet Start-Service kube-proxy # Stop the service Stop-Service kubelet (-Force) Stop-Service kube-proxy (-Force) # Query the service status Get-Service kubelet Get-Service kube-proxy
You can also always use alternative service managers like nssm.exe to run these processes (flanneld, kubelet & kube-proxy) in the background for you. You can use this sample script, leveraging nssm.exe to register kubelet, kube-proxy, and flanneld.exe to run as Windows services in the background.
register-svc.ps1 -NetworkMode <Network mode> -ManagementIP <Windows Node IP> -ClusterCIDR <Cluster subnet> -KubeDnsServiceIP <Kube-dns Service IP> -LogDir <Directory to place logs> # NetworkMode = The network mode l2bridge (flannel host-gw, also the default value) or overlay (flannel vxlan) chosen as a network solution # ManagementIP = The IP address assigned to the Windows node. You can use ipconfig to find this # ClusterCIDR = The cluster subnet range. (Default value 10.244.0.0/16) # KubeDnsServiceIP = The Kubernetes DNS service IP (Default value 10.96.0.10) # LogDir = The directory where kubelet and kube-proxy logs are redirected into their respective output files (Default value C:\k)
If the above referenced script is not suitable, you can manually configure nssm.exe using the following examples.
# Register flanneld.exe nssm install flanneld C:\flannel\flanneld.exe nssm set flanneld AppParameters --kubeconfig-file=c:\k\config --iface=<ManagementIP> --ip-masq=1 --kube-subnet-mgr=1 nssm set flanneld AppEnvironmentExtra NODE_NAME=<hostname> nssm set flanneld AppDirectory C:\flannel nssm start flanneld # Register kubelet.exe # Microsoft releases the pause infrastructure container at mcr.microsoft.com/k8s/core/pause:1.2.0 # For more info search for "pause" in the "Guide for adding Windows Nodes in Kubernetes" nssm install kubelet C:\k\kubelet.exe nssm set kubelet AppParameters --hostname-override=<hostname> --v=6 --pod-infra-container-image=mcr.microsoft.com/k8s/core/pause:1.2.0 --resolv-conf="" --allow-privileged=true --enable-debugging-handlers --cluster-dns=<DNS-service-IP> --cluster-domain=cluster.local --kubeconfig=c:\k\config --hairpin-mode=promiscuous-bridge --image-pull-progress-deadline=20m --cgroups-per-qos=false --log-dir=<log directory> --logtostderr=false --enforce-node-allocatable="" --network-plugin=cni --cni-bin-dir=c:\k\cni --cni-conf-dir=c:\k\cni\config nssm set kubelet AppDirectory C:\k nssm start kubelet # Register kube-proxy.exe (l2bridge / host-gw) nssm install kube-proxy C:\k\kube-proxy.exe nssm set kube-proxy AppDirectory c:\k nssm set kube-proxy AppParameters --v=4 --proxy-mode=kernelspace --hostname-override=<hostname>--kubeconfig=c:\k\config --enable-dsr=false --log-dir=<log directory> --logtostderr=false nssm.exe set kube-proxy AppEnvironmentExtra KUBE_NETWORK=cbr0 nssm set kube-proxy DependOnService kubelet nssm start kube-proxy # Register kube-proxy.exe (overlay / vxlan) nssm install kube-proxy C:\k\kube-proxy.exe nssm set kube-proxy AppDirectory c:\k nssm set kube-proxy AppParameters --v=4 --proxy-mode=kernelspace --feature-gates="WinOverlay=true" --hostname-override=<hostname> --kubeconfig=c:\k\config --network-name=vxlan0 --source-vip=<source-vip> --enable-dsr=false --log-dir=<log directory> --logtostderr=false nssm set kube-proxy DependOnService kubelet nssm start kube-proxy
For initial troubleshooting, you can use the following flags in nssm.exe to redirect stdout and stderr to a output file:
nssm set <Service Name> AppStdout C:\k\mysvc.log nssm set <Service Name> AppStderr C:\k\mysvc.log
For additional details, see official nssm usage docs.
My Windows Pods do not have network connectivity
If you are using virtual machines, ensure that MAC spoofing is enabled on all the VM network adapter(s).
My Windows Pods cannot ping external resources
Windows Pods do not have outbound rules programmed for the ICMP protocol today. However, TCP/UDP is supported. When trying to demonstrate connectivity to resources outside of the cluster, please substitute
ping <IP> with corresponding
curl <IP> commands.
If you are still facing problems, most likely your network configuration in cni.conf deserves some extra attention. You can always edit this static file. The configuration update will apply to any newly created Kubernetes resources.
One of the Kubernetes networking requirements (see Kubernetes model) is for cluster communication to occur without NAT internally. To honor this requirement, there is an ExceptionList for all the communication where we do not want outbound NAT to occur. However, this also means that you need to exclude the external IP you are trying to query from the ExceptionList. Only then will the traffic originating from your Windows pods be SNAT’ed correctly to receive a response from the outside world. In this regard, your ExceptionList in
cni.conf should look as follows:
"ExceptionList": [ "10.244.0.0/16", # Cluster subnet "10.96.0.0/12", # Service subnet "10.127.130.0/24" # Management (host) subnet ]
My Windows node cannot access NodePort service
Local NodePort access from the node itself fails. This is a known limitation. NodePort access works from other nodes or external clients.
vNICs and HNS endpoints of containers are being deleted
This issue can be caused when the
hostname-override parameter is not passed to kube-proxy. To resolve it, users need to pass the hostname to kube-proxy as follows:
With flannel my nodes are having issues after rejoining a cluster
Whenever a previously deleted node is being re-joined to the cluster, flannelD tries to assign a new pod subnet to the node. Users should remove the old pod subnet configuration files in the following paths:
Remove-Item C:\k\SourceVip.json Remove-Item C:\k\SourceVipRequest.json
start.ps1, flanneld is stuck in “Waiting for the Network to be created”
There are numerous reports of this issue which are being investigated; most likely it is a timing issue for when the management IP of the flannel network is set. A workaround is to simply relaunch start.ps1 or relaunch it manually as follows:
PS C:> [Environment]::SetEnvironmentVariable("NODE_NAME", "<Windows_Worker_Hostname>") PS C:> C:\flannel\flanneld.exe --kubeconfig-file=c:\k\config --iface=<Windows_Worker_Node_IP> --ip-masq=1 --kube-subnet-mgr=1
My Windows Pods cannot launch because of missing
This indicates that Flannel didn’t launch correctly. You can either try to restart flanneld.exe or you can copy the files over manually from
/run/flannel/subnet.env on the Kubernetes master to
C:\run\flannel\subnet.env on the Windows worker node and modify the
FLANNEL_SUBNET row to a different number. For example, if node subnet 10.244.4.1/24 is desired:
FLANNEL_NETWORK=10.244.0.0/16 FLANNEL_SUBNET=10.244.4.1/24 FLANNEL_MTU=1500 FLANNEL_IPMASQ=true
My Windows node cannot access my services using the service IP
This is a known limitation of the current networking stack on Windows. Windows Pods are able to access the service IP however.
No network adapter is found when starting kubelet
The Windows networking stack needs a virtual adapter for Kubernetes networking to work. If the following commands return no results (in an admin shell), virtual network creation — a necessary prerequisite for Kubelet to work — has failed:
Get-HnsNetwork | ? Name -ieq "cbr0" Get-NetAdapter | ? Name -Like "vEthernet (Ethernet*"
Often it is worthwhile to modify the InterfaceName parameter of the start.ps1 script, in cases where the host’s network adapter isn’t “Ethernet”. Otherwise, consult the output of the
start-kubelet.ps1 script to see if there are errors during virtual network creation.
My Pods are stuck at “Container Creating” or restarting over and over
Check that your pause image is compatible with your OS version. The instructions assume that both the OS and the containers are version 1803. If you have a later version of Windows, such as an Insider build, you need to adjust the images accordingly. Please refer to the Microsoft’s Docker repository for images. Regardless, both the pause image Dockerfile and the sample service expect the image to be tagged as :latest.
Starting with Kubernetes v1.14, Microsoft releases the pause infrastructure container at
mcr.microsoft.com/k8s/core/pause:1.2.0. For more information search for “pause” in the Guide for adding Windows Nodes in Kubernetes.
DNS resolution is not properly working
Check the DNS limitations for Windows in this section.
kubectl port-forward fails with “unable to do port forwarding: wincat not found”
This was implemented in Kubernetes 1.15, and the pause infrastructure container
mcr.microsoft.com/k8s/core/pause:1.2.0. Be sure to use these versions or newer ones.
If you would like to build your own pause infrastructure container, be sure to include wincat
If these steps don’t resolve your problem, you can get help running Windows containers on Windows nodes in Kubernetes through:
If you have what looks like a bug, or you would like to make a feature request, please use the GitHub issue tracking system. You can open issues on GitHub and assign them to SIG-Windows. You should first search the list of issues in case it was reported previously and comment with your experience on the issue and add additional logs. SIG-Windows Slack is also a great avenue to get some initial support and troubleshooting ideas prior to creating a ticket.
If filing a bug, please include detailed information about how to reproduce the problem, such as:
/sig windowsto bring it to a SIG-Windows member’s attention
containerdA container runtime with an emphasis on simplicity, robustness and portability is another OCI-compliant runtime that recently graduated as a CNCFCloud Native Computing Foundation project. It’s currently tested on Linux, but 1.3 will bring support for Windows and Hyper-V. [reference]
The CRI-ContainerD interface will be able to manage sandboxes based on Hyper-V. This provides a foundation where RuntimeClass could be implemented for new use cases including:
The existing Hyper-V isolation support, an experimental feature as of v1.10, will be deprecated in the future in favor of the CRI-ContainerD and RuntimeClass features mentioned above. To use the current features and create a Hyper-V isolated container, the kubelet should be started with feature gates
HyperVContainer=true and the Pod should include the annotation
experimental.windows.kubernetes.io/isolation-type=hyperv. In the experiemental release, this feature is limited to 1 container per Pod.
apiVersion: apps/v1 kind: Deployment metadata: name: iis spec: selector: matchLabels: app: iis replicas: 3 template: metadata: labels: app: iis annotations: experimental.windows.kubernetes.io/isolation-type: hyperv spec: containers: - name: iis image: microsoft/iis ports: - containerPort: 80
Kubeadm is becoming the de facto standard for users to deploy a Kubernetes cluster. Windows node support in kubeadm will come in a future release. We are also making investments in cluster API to ensure Windows nodes are properly provisioned.